source stringlengths 3 92 | c stringlengths 26 2.25M |
|---|---|
GB_binop__ne_uint32.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__ne_uint32)
// A.*B function (eWiseMult): GB (_AemultB_08__ne_uint32)
// A.*B function (eWiseMult): GB (_AemultB_02__ne_uint32)
// A.*B function (eWiseMult): GB (_AemultB_04__ne_uint32)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__ne_uint32)
// A*D function (colscale): GB (_AxD__ne_uint32)
// D*A function (rowscale): GB (_DxB__ne_uint32)
// C+=B function (dense accum): GB (_Cdense_accumB__ne_uint32)
// C+=b function (dense accum): GB (_Cdense_accumb__ne_uint32)
// C+=A+B function (dense ewise3): GB ((none))
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__ne_uint32)
// C=scalar+B GB (_bind1st__ne_uint32)
// C=scalar+B' GB (_bind1st_tran__ne_uint32)
// C=A+scalar GB (_bind2nd__ne_uint32)
// C=A'+scalar GB (_bind2nd_tran__ne_uint32)
// C type: bool
// A type: uint32_t
// B,b type: uint32_t
// BinaryOp: cij = (aij != bij)
#define GB_ATYPE \
uint32_t
#define GB_BTYPE \
uint32_t
#define GB_CTYPE \
bool
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
0
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
0
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
uint32_t aij = GBX (Ax, pA, A_iso)
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
uint32_t bij = GBX (Bx, pB, B_iso)
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
bool t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = (x != y) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_NE || GxB_NO_UINT32 || GxB_NO_NE_UINT32)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_ewise3_noaccum__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if 0
{
#include "GB_dense_subassign_23_template.c"
}
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__ne_uint32)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if 0
{
// get the scalar b for C += b, of type uint32_t
uint32_t bwork = (*((uint32_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *restrict Cx = (bool *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *restrict Cx = (bool *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__ne_uint32)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
#include "GB_add_template.c"
GB_FREE_WORK ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__ne_uint32)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__ne_uint32)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__ne_uint32)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *Cx = (bool *) Cx_output ;
uint32_t x = (*((uint32_t *) x_input)) ;
uint32_t *Bx = (uint32_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
uint32_t bij = GBX (Bx, p, false) ;
Cx [p] = (x != bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__ne_uint32)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
bool *Cx = (bool *) Cx_output ;
uint32_t *Ax = (uint32_t *) Ax_input ;
uint32_t y = (*((uint32_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
uint32_t aij = GBX (Ax, p, false) ;
Cx [p] = (aij != y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
uint32_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (x != aij) ; \
}
GrB_Info GB (_bind1st_tran__ne_uint32)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
uint32_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint32_t x = (*((const uint32_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
uint32_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
uint32_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (aij != y) ; \
}
GrB_Info GB (_bind2nd_tran__ne_uint32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint32_t y = (*((const uint32_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
displacement_residual_contact_criteria.h | // KRATOS ___| | | |
// \___ \ __| __| | | __| __| | | __| _` | |
// | | | | | ( | | | | ( | |
// _____/ \__|_| \__,_|\___|\__|\__,_|_| \__,_|_| MECHANICS
//
// License: BSD License
// license: StructuralMechanicsApplication/license.txt
//
// Main authors: Vicente Mataix Ferrandiz
//
#if !defined(KRATOS_DISPLACEMENT_RESIDUAL_CONTACT_CRITERIA_H)
#define KRATOS_DISPLACEMENT_RESIDUAL_CONTACT_CRITERIA_H
/* System includes */
/* External includes */
/* Project includes */
#include "utilities/table_stream_utility.h"
#include "solving_strategies/convergencecriterias/convergence_criteria.h"
#include "utilities/color_utilities.h"
namespace Kratos
{
///@addtogroup ContactStructuralMechanicsApplication
///@{
///@name Kratos Globals
///@{
///@}
///@name Type Definitions
///@{
///@}
///@name Enum's
///@{
///@}
///@name Functions
///@{
///@name Kratos Classes
///@{
/**
* @class DisplacementResidualContactCriteria
* @ingroup ContactStructuralMechanicsApplication
* @brief Convergence criteria for contact problems
* This class implements a convergence control based on nodal displacement (for penalty contact)
* @author Vicente Mataix Ferrandiz
*/
template< class TSparseSpace,
class TDenseSpace >
class DisplacementResidualContactCriteria
: public ConvergenceCriteria< TSparseSpace, TDenseSpace >
{
public:
///@name Type Definitions
///@{
/// Pointer definition of DisplacementResidualContactCriteria
KRATOS_CLASS_POINTER_DEFINITION( DisplacementResidualContactCriteria );
/// Local Flags
KRATOS_DEFINE_LOCAL_FLAG( PRINTING_OUTPUT );
KRATOS_DEFINE_LOCAL_FLAG( TABLE_IS_INITIALIZED );
KRATOS_DEFINE_LOCAL_FLAG( ROTATION_DOF_IS_CONSIDERED );
KRATOS_DEFINE_LOCAL_FLAG( INITIAL_RESIDUAL_IS_SET );
/// The base class definition (and it subclasses)
typedef ConvergenceCriteria< TSparseSpace, TDenseSpace > BaseType;
typedef typename BaseType::TDataType TDataType;
typedef typename BaseType::DofsArrayType DofsArrayType;
typedef typename BaseType::TSystemMatrixType TSystemMatrixType;
typedef typename BaseType::TSystemVectorType TSystemVectorType;
/// The sparse space used
typedef TSparseSpace SparseSpaceType;
/// The r_table stream definition TODO: Replace by logger
typedef TableStreamUtility::Pointer TablePrinterPointerType;
/// The index type definition
typedef std::size_t IndexType;
/// The key type definition
typedef std::size_t KeyType;
///@}
///@name Life Cycle
///@{
/**
* @brief Default constructor
* @param DispRatioTolerance Relative tolerance for displacement residual error
* @param DispAbsTolerance Absolute tolerance for displacement residual error
* @param RotRatioTolerance Relative tolerance for rotation residual error
* @param RotAbsTolerance Absolute tolerance for rotation residual error
* @param pTable The pointer to the output r_table
* @param PrintingOutput If the output is going to be printed in a txt file
*/
explicit DisplacementResidualContactCriteria(
const TDataType DispRatioTolerance,
const TDataType DispAbsTolerance,
const TDataType RotRatioTolerance,
const TDataType RotAbsTolerance,
const bool PrintingOutput = false
)
: BaseType()
{
// Set local flags
mOptions.Set(DisplacementResidualContactCriteria::PRINTING_OUTPUT, PrintingOutput);
mOptions.Set(DisplacementResidualContactCriteria::TABLE_IS_INITIALIZED, false);
mOptions.Set(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED, false);
mOptions.Set(DisplacementResidualContactCriteria::INITIAL_RESIDUAL_IS_SET, false);
// The displacement residual
mDispRatioTolerance = DispRatioTolerance;
mDispAbsTolerance = DispAbsTolerance;
// The rotation residual
mRotRatioTolerance = RotRatioTolerance;
mRotAbsTolerance = RotAbsTolerance;
}
/**
* @brief Default constructor (parameters)
* @param ThisParameters The configuration parameters
*/
explicit DisplacementResidualContactCriteria( Parameters ThisParameters = Parameters(R"({})"))
: BaseType()
{
// Validate and assign defaults
ThisParameters = this->ValidateAndAssignParameters(ThisParameters, this->GetDefaultParameters());
this->AssignSettings(ThisParameters);
}
//* Copy constructor.
DisplacementResidualContactCriteria( DisplacementResidualContactCriteria const& rOther )
:BaseType(rOther)
,mOptions(rOther.mOptions)
,mDispRatioTolerance(rOther.mDispRatioTolerance)
,mDispAbsTolerance(rOther.mDispAbsTolerance)
,mDispInitialResidualNorm(rOther.mDispInitialResidualNorm)
,mDispCurrentResidualNorm(rOther.mDispCurrentResidualNorm)
,mRotRatioTolerance(rOther.mRotRatioTolerance)
,mRotAbsTolerance(rOther.mRotAbsTolerance)
,mRotInitialResidualNorm(rOther.mRotInitialResidualNorm)
,mRotCurrentResidualNorm(rOther.mRotCurrentResidualNorm)
{
}
/// Destructor.
~DisplacementResidualContactCriteria() override = default;
///@}
///@name Operators
///@{
/**
* @brief Compute relative and absolute error.
* @param rModelPart Reference to the ModelPart containing the contact problem.
* @param rDofSet Reference to the container of the problem's degrees of freedom (stored by the BuilderAndSolver)
* @param rA System matrix (unused)
* @param rDx Vector of results (variations on nodal variables)
* @param rb RHS vector (residual)
* @return true if convergence is achieved, false otherwise
*/
bool PostCriteria(
ModelPart& rModelPart,
DofsArrayType& rDofSet,
const TSystemMatrixType& rA,
const TSystemVectorType& rDx,
const TSystemVectorType& rb
) override
{
if (SparseSpaceType::Size(rb) != 0) { //if we are solving for something
// Initialize
TDataType disp_residual_solution_norm = 0.0;
IndexType disp_dof_num(0);
TDataType rot_residual_solution_norm = 0.0;
IndexType rot_dof_num(0);
// First iterator
const auto it_dof_begin = rDofSet.begin();
// Auxiliar values
std::size_t dof_id = 0;
TDataType residual_dof_value = 0.0;
// Auxiliar displacement DoF check
const std::function<bool(const VariableData&)> check_without_rot =
[](const VariableData& rCurrVar) -> bool {return true;};
const std::function<bool(const VariableData&)> check_with_rot =
[](const VariableData& rCurrVar) -> bool {return ((rCurrVar == DISPLACEMENT_X) || (rCurrVar == DISPLACEMENT_Y) || (rCurrVar == DISPLACEMENT_Z));};
const auto* p_check_disp = (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) ? &check_with_rot : &check_without_rot;
// Loop over Dofs
#pragma omp parallel for reduction(+:disp_residual_solution_norm,disp_dof_num,rot_residual_solution_norm,rot_dof_num,dof_id,residual_dof_value)
for (int i = 0; i < static_cast<int>(rDofSet.size()); i++) {
auto it_dof = it_dof_begin + i;
if (it_dof->IsFree()) {
dof_id = it_dof->EquationId();
residual_dof_value = rb[dof_id];
const auto& r_curr_var = it_dof->GetVariable();
if ((*p_check_disp)(r_curr_var)) {
disp_residual_solution_norm += std::pow(residual_dof_value, 2);
++disp_dof_num;
} else { // We will assume is rotation dof
KRATOS_DEBUG_ERROR_IF_NOT((r_curr_var == ROTATION_X) || (r_curr_var == ROTATION_Y) || (r_curr_var == ROTATION_Z)) << "Variable must be a ROTATION and it is: " << r_curr_var.Name() << std::endl;
rot_residual_solution_norm += std::pow(residual_dof_value, 2);
++rot_dof_num;
}
}
}
mDispCurrentResidualNorm = disp_residual_solution_norm;
mRotCurrentResidualNorm = rot_residual_solution_norm;
TDataType residual_disp_ratio = 1.0;
TDataType residual_rot_ratio = 1.0;
// We initialize the solution
if (mOptions.IsNot(DisplacementResidualContactCriteria::INITIAL_RESIDUAL_IS_SET)) {
mDispInitialResidualNorm = (disp_residual_solution_norm == 0.0) ? 1.0 : disp_residual_solution_norm;
residual_disp_ratio = 1.0;
if (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) {
mRotInitialResidualNorm = (rot_residual_solution_norm == 0.0) ? 1.0 : rot_residual_solution_norm;
residual_rot_ratio = 1.0;
}
mOptions.Set(DisplacementResidualContactCriteria::INITIAL_RESIDUAL_IS_SET, true);
}
// We calculate the ratio of the displacements
residual_disp_ratio = mDispCurrentResidualNorm/mDispInitialResidualNorm;
residual_rot_ratio = mRotCurrentResidualNorm/mRotInitialResidualNorm;
// We calculate the absolute norms
const TDataType residual_disp_abs = mDispCurrentResidualNorm/disp_dof_num;
const TDataType residual_rot_abs = mRotCurrentResidualNorm/rot_dof_num;
// The process info of the model part
ProcessInfo& r_process_info = rModelPart.GetProcessInfo();
// We print the results // TODO: Replace for the new log
if (rModelPart.GetCommunicator().MyPID() == 0 && this->GetEchoLevel() > 0) {
if (r_process_info.Has(TABLE_UTILITY)) {
std::cout.precision(4);
TablePrinterPointerType p_table = r_process_info[TABLE_UTILITY];
auto& r_table = p_table->GetTable();
if (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) {
r_table << residual_disp_ratio << mDispRatioTolerance << residual_disp_abs << mDispAbsTolerance << residual_rot_ratio << mRotRatioTolerance << residual_rot_abs << mRotAbsTolerance;
} else {
r_table << residual_disp_ratio << mDispRatioTolerance << residual_disp_abs << mDispAbsTolerance;
}
} else {
std::cout.precision(4);
if (mOptions.IsNot(DisplacementResidualContactCriteria::PRINTING_OUTPUT)) {
KRATOS_INFO("DisplacementResidualContactCriteria") << BOLDFONT("RESIDUAL CONVERGENCE CHECK") << "\tSTEP: " << r_process_info[STEP] << "\tNL ITERATION: " << r_process_info[NL_ITERATION_NUMBER] << std::endl << std::scientific;
KRATOS_INFO("DisplacementResidualContactCriteria") << BOLDFONT("\tDISPLACEMENT: RATIO = ") << residual_disp_ratio << BOLDFONT(" EXP.RATIO = ") << mDispRatioTolerance << BOLDFONT(" ABS = ") << residual_disp_abs << BOLDFONT(" EXP.ABS = ") << mDispAbsTolerance << std::endl;
if (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) {
KRATOS_INFO("DisplacementResidualContactCriteria") << BOLDFONT("\tDISPLACEMENT: RATIO = ") << residual_rot_ratio << BOLDFONT(" EXP.RATIO = ") << mRotRatioTolerance << BOLDFONT(" ABS = ") << residual_rot_abs << BOLDFONT(" EXP.ABS = ") << mRotAbsTolerance << std::endl;
}
} else {
KRATOS_INFO("DisplacementResidualContactCriteria") << "RESIDUAL CONVERGENCE CHECK" << "\tSTEP: " << r_process_info[STEP] << "\tNL ITERATION: " << r_process_info[NL_ITERATION_NUMBER] << std::endl << std::scientific;
KRATOS_INFO("DisplacementResidualContactCriteria") << "\tDISPLACEMENT: RATIO = " << residual_disp_ratio << " EXP.RATIO = " << mDispRatioTolerance << " ABS = " << residual_disp_abs << " EXP.ABS = " << mDispAbsTolerance << std::endl;
if (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) {
KRATOS_INFO("DisplacementResidualContactCriteria") << "\tDISPLACEMENT: RATIO = " << residual_rot_ratio << " EXP.RATIO = " << mRotRatioTolerance << " ABS = " << residual_rot_abs << " EXP.ABS = " << mRotAbsTolerance << std::endl;
}
}
}
}
r_process_info[CONVERGENCE_RATIO] = residual_disp_ratio;
r_process_info[RESIDUAL_NORM] = residual_disp_abs;
// We check if converged
const bool disp_converged = (residual_disp_ratio <= mDispRatioTolerance || residual_disp_abs <= mDispAbsTolerance);
const bool rot_converged = (mOptions.Is(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) ? (residual_rot_ratio <= mRotRatioTolerance || residual_rot_abs <= mRotAbsTolerance) : true;
if (disp_converged && rot_converged) {
if (rModelPart.GetCommunicator().MyPID() == 0 && this->GetEchoLevel() > 0) {
if (r_process_info.Has(TABLE_UTILITY)) {
TablePrinterPointerType p_table = r_process_info[TABLE_UTILITY];
auto& r_table = p_table->GetTable();
if (mOptions.IsNot(DisplacementResidualContactCriteria::PRINTING_OUTPUT))
r_table << BOLDFONT(FGRN(" Achieved"));
else
r_table << "Achieved";
} else {
if (mOptions.IsNot(DisplacementResidualContactCriteria::PRINTING_OUTPUT))
KRATOS_INFO("DisplacementResidualContactCriteria") << BOLDFONT("\tResidual") << " convergence is " << BOLDFONT(FGRN("achieved")) << std::endl;
else
KRATOS_INFO("DisplacementResidualContactCriteria") << "\tResidual convergence is achieved" << std::endl;
}
}
return true;
} else {
if (rModelPart.GetCommunicator().MyPID() == 0 && this->GetEchoLevel() > 0) {
if (r_process_info.Has(TABLE_UTILITY)) {
TablePrinterPointerType p_table = r_process_info[TABLE_UTILITY];
auto& r_table = p_table->GetTable();
if (mOptions.IsNot(DisplacementResidualContactCriteria::PRINTING_OUTPUT))
r_table << BOLDFONT(FRED(" Not achieved"));
else
r_table << "Not achieved";
} else {
if (mOptions.IsNot(DisplacementResidualContactCriteria::PRINTING_OUTPUT))
KRATOS_INFO("DisplacementResidualContactCriteria") << BOLDFONT("\tResidual") << " convergence is " << BOLDFONT(FRED(" not achieved")) << std::endl;
else
KRATOS_INFO("DisplacementResidualContactCriteria") << "\tResidual convergence is not achieved" << std::endl;
}
}
return false;
}
} else // In this case all the displacements are imposed!
return true;
}
/**
* @brief This function initialize the convergence criteria
* @param rModelPart Reference to the ModelPart containing the contact problem. (unused)
*/
void Initialize( ModelPart& rModelPart) override
{
BaseType::mConvergenceCriteriaIsInitialized = true;
ProcessInfo& r_process_info = rModelPart.GetProcessInfo();
if (r_process_info.Has(TABLE_UTILITY) && mOptions.IsNot(DisplacementResidualContactCriteria::TABLE_IS_INITIALIZED)) {
TablePrinterPointerType p_table = r_process_info[TABLE_UTILITY];
auto& r_table = p_table->GetTable();
r_table.AddColumn("DP RATIO", 10);
r_table.AddColumn("EXP. RAT", 10);
r_table.AddColumn("ABS", 10);
r_table.AddColumn("EXP. ABS", 10);
if (mOptions.IsNot(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED)) {
r_table.AddColumn("RT RATIO", 10);
r_table.AddColumn("EXP. RAT", 10);
r_table.AddColumn("ABS", 10);
r_table.AddColumn("EXP. ABS", 10);
}
r_table.AddColumn("CONVERGENCE", 15);
mOptions.Set(DisplacementResidualContactCriteria::TABLE_IS_INITIALIZED, true);
}
// Check rotation dof
mOptions.Set(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED, ContactUtilities::CheckModelPartHasRotationDoF(rModelPart));
}
/**
* @brief This function initializes the solution step
* @param rModelPart Reference to the ModelPart containing the contact problem.
* @param rDofSet Reference to the container of the problem's degrees of freedom (stored by the BuilderAndSolver)
* @param rA System matrix (unused)
* @param rDx Vector of results (variations on nodal variables)
* @param rb RHS vector (residual)
*/
void InitializeSolutionStep(
ModelPart& rModelPart,
DofsArrayType& rDofSet,
const TSystemMatrixType& rA,
const TSystemVectorType& rDx,
const TSystemVectorType& rb
) override
{
mOptions.Set(DisplacementResidualContactCriteria::INITIAL_RESIDUAL_IS_SET, false);
}
/**
* @brief This method provides the defaults parameters to avoid conflicts between the different constructors
* @return The default parameters
*/
Parameters GetDefaultParameters() const override
{
Parameters default_parameters = Parameters(R"(
{
"name" : "displacement_residual_contact_criteria",
"ensure_contact" : false,
"print_convergence_criterion" : false,
"residual_relative_tolerance" : 1.0e-4,
"residual_absolute_tolerance" : 1.0e-9,
"rotation_residual_relative_tolerance" : 1.0e-4,
"rotation_residual_absolute_tolerance" : 1.0e-9
})");
// Getting base class default parameters
const Parameters base_default_parameters = BaseType::GetDefaultParameters();
default_parameters.RecursivelyAddMissingParameters(base_default_parameters);
return default_parameters;
}
/**
* @brief Returns the name of the class as used in the settings (snake_case format)
* @return The name of the class
*/
static std::string Name()
{
return "displacement_residual_contact_criteria";
}
///@}
///@name Operations
///@{
///@}
///@name Acces
///@{
///@}
///@name Inquiry
///@{
///@}
///@name Friends
///@{
protected:
///@name Protected static Member Variables
///@{
///@}
///@name Protected member Variables
///@{
///@}
///@name Protected Operators
///@{
///@}
///@name Protected Operations
///@{
/**
* @brief This method assigns settings to member variables
* @param ThisParameters Parameters that are assigned to the member variables
*/
void AssignSettings(const Parameters ThisParameters) override
{
BaseType::AssignSettings(ThisParameters);
// The displacement residual
mDispRatioTolerance = ThisParameters["residual_relative_tolerance"].GetDouble();
mDispAbsTolerance = ThisParameters["residual_absolute_tolerance"].GetDouble();
// The rotation residual
mRotRatioTolerance = ThisParameters["rotation_residual_relative_tolerance"].GetDouble();
mRotAbsTolerance = ThisParameters["rotation_residual_absolute_tolerance"].GetDouble();
// Set local flags
mOptions.Set(DisplacementResidualContactCriteria::PRINTING_OUTPUT, ThisParameters["print_convergence_criterion"].GetBool());
mOptions.Set(DisplacementResidualContactCriteria::TABLE_IS_INITIALIZED, false);
mOptions.Set(DisplacementResidualContactCriteria::ROTATION_DOF_IS_CONSIDERED, false);
mOptions.Set(DisplacementResidualContactCriteria::INITIAL_RESIDUAL_IS_SET, false);
}
///@}
///@name Protected Access
///@{
///@}
///@name Protected Inquiry
///@{
///@}
///@name Protected LifeCycle
///@{
///@}
private:
///@name Static Member Variables
///@{
///@}
///@name Member Variables
///@{
Flags mOptions; /// Local flags
TDataType mDispRatioTolerance; /// The ratio threshold for the norm of the displacement residual
TDataType mDispAbsTolerance; /// The absolute value threshold for the norm of the displacement residual
TDataType mDispInitialResidualNorm; /// The reference norm of the displacement residual
TDataType mDispCurrentResidualNorm; /// The current norm of the displacement residual
TDataType mRotRatioTolerance; /// The ratio threshold for the norm of the rotation residual
TDataType mRotAbsTolerance; /// The absolute value threshold for the norm of the rotation residual
TDataType mRotInitialResidualNorm; /// The reference norm of the rotation residual
TDataType mRotCurrentResidualNorm; /// The current norm of the rotation residual
///@}
///@name Private Operators
///@{
///@}
///@name Private Operations
///@{
///@}
///@name Private Access
///@{
///@}
///@}
///@name Serialization
///@{
///@name Private Inquiry
///@{
///@}
///@name Unaccessible methods
///@{
///@}
}; // Kratos DisplacementResidualContactCriteria
///@name Local flags creation
///@{
/// Local Flags
template<class TSparseSpace, class TDenseSpace>
const Kratos::Flags DisplacementResidualContactCriteria<TSparseSpace, TDenseSpace>::PRINTING_OUTPUT(Kratos::Flags::Create(1));
template<class TSparseSpace, class TDenseSpace>
const Kratos::Flags DisplacementResidualContactCriteria<TSparseSpace, TDenseSpace>::TABLE_IS_INITIALIZED(Kratos::Flags::Create(2));
template<class TSparseSpace, class TDenseSpace>
const Kratos::Flags DisplacementResidualContactCriteria<TSparseSpace, TDenseSpace>::ROTATION_DOF_IS_CONSIDERED(Kratos::Flags::Create(3));
template<class TSparseSpace, class TDenseSpace>
const Kratos::Flags DisplacementResidualContactCriteria<TSparseSpace, TDenseSpace>::INITIAL_RESIDUAL_IS_SET(Kratos::Flags::Create(4));
}
#endif /* KRATOS_DISPLACEMENT_RESIDUAL_CONTACT_CRITERIA_H */
|
bits-insns.c | #include <math.h>
#define N 12
int main()
{
unsigned int arguments[N] = {0u, 1u, 2u, 3u, 111u, 333u, 444u, 0x80000000u, 0x0000ffffu, 0xf0000000u, 0xff000000u, 0xffffffffu};
int clrsb[N] = {};
int clz[N] = {};
int ctz[N] = {};
int ffs[N] = {};
int parity[N] = {};
int popcount[N] = {};
int ref_clrsb[N] = {};
int ref_clz[N] = {};
int ref_ctz[N] = {};
int ref_ffs[N] = {};
int ref_parity[N] = {};
int ref_popcount[N] = {};
for (unsigned i = 0; i < N; i++)
{
ref_clrsb[i] = __builtin_clrsb (arguments[i]);
ref_clz[i] = __builtin_clz (arguments[i]);
ref_ctz[i] = __builtin_ctz (arguments[i]);
ref_ffs[i] = __builtin_ffs (arguments[i]);
ref_parity[i] = __builtin_parity (arguments[i]);
ref_popcount[i] = __builtin_popcount (arguments[i]);
}
#pragma omp target map(from:clz, ctz, ffs, parity, popcount)
{
for (unsigned i = 0; i < N; i++)
{
clrsb[i] = __builtin_clrsb (arguments[i]);
clz[i] = __builtin_clz (arguments[i]);
ctz[i] = __builtin_ctz (arguments[i]);
ffs[i] = __builtin_ffs (arguments[i]);
parity[i] = __builtin_parity (arguments[i]);
popcount[i] = __builtin_popcount (arguments[i]);
}
}
for (unsigned i = 0; i < N; i++)
if (ref_clrsb[i] != clrsb[i])
__builtin_abort ();
/* CLZ of zero is undefined for zero. */
for (unsigned i = 1; i < N; i++)
if (ref_clz[i] != clz[i])
__builtin_abort ();
/* Likewise for ctz */
for (unsigned i = 1; i < N; i++)
if (ref_ctz[i] != ctz[i])
__builtin_abort ();
for (unsigned i = 0; i < N; i++)
if (ref_ffs[i] != ffs[i])
__builtin_abort ();
for (unsigned i = 0; i < N; i++)
if (ref_parity[i] != parity[i])
__builtin_abort ();
for (unsigned i = 0; i < N; i++)
if (ref_popcount[i] != popcount[i])
__builtin_abort ();
return 0;
}
|
cancel_parallel.c | // RUN: %libomp-compile && env OMP_CANCELLATION=true %libomp-run | %sort-threads | FileCheck %s
// REQUIRES: ompt
// Current GOMP interface implementation does not support cancellation
// XFAIL: gcc
#include "callback.h"
#include "omp.h"
int main() {
#pragma omp parallel num_threads(2)
{
if (omp_get_thread_num() == 0) {
print_fuzzy_address_blocks(get_ompt_label_address(1));
#pragma omp cancel parallel
define_ompt_label(1);
// We cannot print at this location because the parallel region is cancelled!
} else {
delay(100);
print_fuzzy_address_blocks(get_ompt_label_address(2));
#pragma omp cancellation point parallel
define_ompt_label(2);
// We cannot print at this location because the parallel region is cancelled!
}
}
// Check if libomp supports the callbacks for this test.
// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_task_create'
// CHECK-NOT: {{^}}0: Could not register callback 'ompt_callback_cancel'
// CHECK: {{^}}0: NULL_POINTER=[[NULL:.*$]]
// CHECK: {{^}}[[MASTER_ID:[0-9]+]]: ompt_event_task_create: parent_task_id=[[PARENT_TASK_ID:[0-9]+]], parent_task_frame.exit=[[NULL]], parent_task_frame.reenter=[[NULL]], new_task_id=[[TASK_ID:[0-9]+]], codeptr_ra=[[NULL]], task_type=ompt_task_initial=1, has_dependences=no
// CHECK-DAG: {{^}}[[MASTER_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_parallel|ompt_cancel_activated=17, codeptr_ra=[[RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
// CHECK-DAG: {{^}}[[MASTER_ID]]: fuzzy_address={{.*}}[[RETURN_ADDRESS]]
// CHECK: {{^}}[[THREAD_ID:[0-9]+]]: ompt_event_implicit_task_begin
// CHECK-DAG: {{^}}[[THREAD_ID]]: ompt_event_cancel: task_data=[[TASK_ID:[0-9]+]], flags=ompt_cancel_parallel|ompt_cancel_detected=33, codeptr_ra=[[OTHER_RETURN_ADDRESS:0x[0-f]+]]{{[0-f][0-f]}}
// CHECK-DAG: {{^}}[[THREAD_ID]]: fuzzy_address={{.*}}[[OTHER_RETURN_ADDRESS]]
return 0;
}
|
sum.h | #pragma once
#include <vector>
#include <unordered_map>
#include <algorithm>
#include <memory>
#include <omp.h>
#include "_cuda.h"
using std::vector;
using std::unique_ptr;
using std::max;
template <class T>
T sum(T *x, int N) {
T a = T();
for (int i=0; i<N; i++)
a += x[i];
return a;
}
template <class T>
T sum(vector<T>& x) {
return sum(x.data(), x.size());
}
template <class K, class T>
T sum(unordered_map<K, T>& x) {
T a = T();
for (auto& p : x)
a += p.second;
return a;
}
template <class T, class C>
T sumAt(T *x, C&& is) {
T a = T();
for (int i : is)
a += x[i];
return a;
}
template <class T, class C>
T sumAt(vector<T>& x, C&& is) {
return sumAt(x.data(), is);
}
template <class K, class T, class C>
T sumAt(unordered_map<K, T>& x, C&& ks) {
T a = T();
for (auto&& k : ks)
a += x[k];
return a;
}
template <class T>
T sumOmp(T *x, int N) {
T a = T();
#pragma omp parallel for reduction (+:a)
for (int i=0; i<N; i++)
a += x[i];
return a;
}
template <class T>
T sumOmp(vector<T>& x) {
return sumOmp(x.data(), x.size());
}
template <class T>
__device__ void sumKernelReduce(T* a, int N, int i) {
__syncthreads();
for (N=N/2; N>0; N/=2) {
if (i < N) a[i] += a[N+i];
__syncthreads();
}
}
template <class T>
__device__ T sumKernelLoop(T *x, int N, int i, int DI) {
T a = T();
for (; i<N; i+=DI)
a += x[i];
return a;
}
template <class T>
__global__ void sumKernel(T *a, T *x, int N) {
DEFINE(t, b, B, G);
__shared__ T cache[_THREADS];
cache[t] = sumKernelLoop(x, N, B*b+t, G*B);
sumKernelReduce(cache, B, t);
if (t == 0) a[b] = cache[0];
}
template <class T>
T sumCuda(T *x, int N) {
int threads = _THREADS;
int blocks = min(ceilDiv(N, threads), _BLOCKS);
size_t X1 = N * sizeof(T);
size_t A1 = blocks * sizeof(T);
unique_ptr<T> a(new T[A1]);
T *xD, *aD;
TRY( cudaMalloc(&xD, X1) );
TRY( cudaMalloc(&aD, A1) );
TRY( cudaMemcpy(xD, x, X1, cudaMemcpyHostToDevice) );
sumKernel<<<blocks, threads>>>(aD, xD, N);
TRY( cudaMemcpy(a.get(), aD, A1, cudaMemcpyDeviceToHost) );
TRY( cudaFree(xD) );
TRY( cudaFree(aD) );
return sum(a.get(), blocks);
}
template <class T>
T sumCuda(vector<T>& x) {
return sumCuda(x.data(), x.size());
}
template <class T>
__device__ T sumAtKernelLoop(T *x, int *is, int N, int i, int DI) {
T a = T();
for (; i<N; i+=DI)
a += x[is[i]];
return a;
}
template <class T>
__global__ void sumAtKernel(T *a, T *x, T *is, int N) {
DEFINE(t, b, B, G);
__shared__ T cache[_THREADS];
cache[t] = sumAtKernelLoop(x, is, N, B*b+t, G*B);
sumKernelReduce(cache, B, t);
if (t == 0) a[b] = cache[0];
}
template <class T, class C>
__device__ T sumIfNotKernelLoop(T *x, C *cs, int N, int i, int DI) {
T a = T();
for (; i<N; i+=DI)
if (cs[i] == 0) a += x[i];
return a;
}
template <class T, class C>
__global__ void sumIfNotKernel(T *a, T *x, C *cs, int N) {
DEFINE(t, b, B, G);
__shared__ T cache[_THREADS];
cache[t] = sumIfNotKernelLoop(x, cs, N, B*b+t, G*B);
sumKernelReduce(cache, B, t);
if (t == 0) a[b] = cache[0];
}
|
ParallelHashMap.h | /**
* @file ParallelHashMap.h
* @brief A thread-safe hash map supporting insertion and lookup operations.
* @details The parallel hash map is built on top of a fixed-sized hash map
* object and features OpenMP concurrency structures. The underlying
* fixed-sized hash map handles collisions with chaining.
* @date June 6, 2015
* @author Geoffrey Gunow, MIT, Course 22 (geogunow@mit.edu)
*/
#ifndef __PARALLEL_HASH_MAP__
#define __PARALLEL_HASH_MAP__
#include<iostream>
#include<stdexcept>
#include<functional>
#include<omp.h>
#include "log.h"
/**
* @class FixedHashMap ParallelHashMap.h "src/ParallelHashMap.h"
* @brief A fixed-size hash map supporting insertion and lookup operations.
* @details The FixedHashMap class supports insertion and lookup operations
* but not deletion as deletion is not needed in the OpenMOC application.
* This hash table uses chaining for collisions and does not incorporate
* concurrency objects except for tracking the number of entries in the
* table for which an atomic increment is used. This hash table is not
* thread safe but is used as a building block for the ParallelHashMap
* class. This table guarantees O(1) insertions and lookups on average.
*/
template <class K, class V>
class FixedHashMap {
struct node {
node(K k_in, V v_in) : next(NULL), key(k_in), value(v_in) {}
K key;
V value;
node *next;
};
private:
size_t _M; /* table size */
size_t _N; /* number of elements present in table */
node ** _buckets; /* buckets of values stored in nodes */
public:
FixedHashMap(size_t M = 64);
virtual ~FixedHashMap();
bool contains(K& key);
V& at(K& key);
void insert(K key, V value);
long insert_and_get_count(K key, V value);
size_t size();
size_t bucket_count();
K* keys();
V* values();
void clear();
void print_buckets();
};
/**
* @class ParallelHashMap ParallelHashMap.h "src/ParallelHashMap.h"
* @brief A thread-safe hash map supporting insertion and lookup operations.
* @details The ParallelHashMap class is built on top of the FixedHashMap
* class, supporting insertion and lookup operations but not deletion as
* deletion is not needed in the OpenMOC application. This hash table uses
* chaining for collisions, as defined in FixedHashMap. It offers lock
* free lookups in O(1) time on average and fine-grained locking for
* insertions in O(1) time on average as well. Resizing is conducted
* periodically during inserts, although the starting table size can be
* chosen to limit the number of resizing operations.
*/
template <class K, class V>
class ParallelHashMap {
/* padded pointer to hash table to avoid false sharing */
struct paddedPointer {
volatile long pad_L1;
volatile long pad_L2;
volatile long pad_L3;
volatile long pad_L4;
volatile long pad_L5;
volatile long pad_L6;
volatile long pad_L7;
FixedHashMap<K,V> volatile* value;
volatile long pad_R1;
volatile long pad_R2;
volatile long pad_R3;
volatile long pad_R4;
volatile long pad_R5;
volatile long pad_R6;
volatile long pad_R7;
volatile long pad_R8;
};
private:
FixedHashMap<K,V> *_table;
paddedPointer *_announce;
size_t _num_threads;
size_t _N;
omp_lock_t * _locks;
size_t _num_locks;
bool _fixed_size;
void resize();
public:
ParallelHashMap(size_t M = 64, size_t L = 64);
virtual ~ParallelHashMap();
bool contains(K& key);
V at(K& key);
void update(K& key, V value);
void insert(K key, V value);
long insert_and_get_count(K key, V value);
size_t size();
size_t bucket_count();
size_t num_locks();
K* keys();
V* values();
void clear();
void setFixedSize();
void print_buckets();
void realloc(size_t M);
};
/**
* @brief Constructor initializes fixed-size table of buckets filled with empty
* linked lists.
* @details The constructor initializes a fixed-size hash map with the size
* as an input parameter. If no size is given the default size (64)
* is used. Buckets are filled with empty linked lists presented as
* NULL pointers.
* @param M size of fixed hash map
*/
template <class K, class V>
FixedHashMap<K,V>::FixedHashMap(size_t M) {
/* ensure M is a power of 2 */
if ((M & (M-1)) != 0) {
/* if not, round up to nearest power of 2 */
M--;
for (size_t i = 1; i < 8 * sizeof(size_t); i*=2)
M |= M >> i;
M++;
}
/* allocate table */
_M = M;
_N = 0;
_buckets = new node*[_M]();
}
/**
* @brief Destructor deletes all nodes in the linked lists associated with each
* bucket in the fixed-size table and their pointers.
*/
template <class K, class V>
FixedHashMap<K,V>::~FixedHashMap() {
/* for each bucket, scan through linked list and delete all nodes */
for (size_t i=0; i<_M; i++) {
node *iter_node = _buckets[i];
while (iter_node != NULL) {
node *next_node = iter_node->next;
delete iter_node;
iter_node = next_node;
}
}
/* delete all buckets (now pointers to empty linked lists) */
delete [] _buckets;
}
/**
* @brief Determine whether the fixed-size table contains a given key.
* @details The linked list in the bucket associated with the key is searched
* to determine whether the key is present.
* @param key key to be searched
* @return boolean value referring to whether the key is contained in the map
*/
template <class K, class V>
bool FixedHashMap<K,V>::contains(K& key) {
/* get hash into table assuming M is a power of 2, using fast modulus */
size_t key_hash = std::hash<K>()(key) & (_M-1);
/* search corresponding bucket for key */
node *iter_node = _buckets[key_hash];
while (iter_node != NULL) {
if (iter_node->key == key)
return true;
else
iter_node = iter_node->next;
}
return false;
}
/**
* @brief Determine the value associated with a given key in the fixed-size
* table.
* @details The linked list in the bucket associated with the key is searched
* and once the key is found, the corresponding value is returned.
* An exception is thrown if the key is not present in the map.
* @param key key whose corresponding value is desired
* @return value associated with the given key
*/
template <class K, class V>
V& FixedHashMap<K,V>::at(K& key) {
/* get hash into table assuming M is a power of 2, using fast modulus */
size_t key_hash = std::hash<K>()(key) & (_M-1);
/* search bucket for key and return the corresponding value if found */
node *iter_node = _buckets[key_hash];
while (iter_node != NULL) {
if (iter_node->key == key)
return iter_node->value;
else
iter_node = iter_node->next;
}
/* after the bucket has been completely searched without finding the key,
print an error message */
log_printf(ERROR, "Key not present in map");
}
/**
* @brief Inserts a key/value pair into the fixed-size table.
* @details The specified key value pair is inserted into the fixed-size table.
* If the key already exists in the table, the pair is not inserted
* and the function returns.
* @param key key of the key/value pair to be inserted
* @param value value of the key/value pair to be inserted
*/
template <class K, class V>
void FixedHashMap<K,V>::insert(K key, V value) {
/* get hash into table using fast modulus */
size_t key_hash = std::hash<K>()(key) & (_M-1);
/* check to see if key already exists in map */
if (contains(key))
return;
/* create new node */
node *new_node = new node(key, value);
/* find where to place element in linked list */
node **iter_node = &_buckets[key_hash];
while (*iter_node != NULL)
iter_node = &(*iter_node)->next;
/* place element in linked list */
*iter_node = new_node;
/* increment counter */
#pragma omp atomic
_N++;
}
/**
* @brief Inserts a key/value pair into the fixed-size table and returns the
* order number with which it was inserted.
* @details The specified key value pair is inserted into the fixed-size table.
* If the key already exists in the table, the pair is not inserted
* and the function returns -1.
* @param key key of the key/value pair to be inserted
* @param value value of the key/value pair to be inserted
* @return order number in which key/value pair was inserted, -1 is returned if
* key was already present in map.
*/
template <class K, class V>
long FixedHashMap<K,V>::insert_and_get_count(K key, V value) {
/* get hash into table using fast modulus */
size_t key_hash = std::hash<K>()(key) & (_M-1);
/* check to see if key already exists in map */
if (contains(key))
return -1;
/* create new node */
node *new_node = new node(key, value);
/* find where to place element in linked list */
node **iter_node = &_buckets[key_hash];
while (*iter_node != NULL)
iter_node = &(*iter_node)->next;
/* place element in linked list */
*iter_node = new_node;
/* increment counter and return number */
size_t N;
#pragma omp atomic capture
N = _N++;
return (long) N;
}
/**
* @brief Returns the number of key/value pairs in the fixed-size table.
* @return number of key/value pairs in the map
*/
template <class K, class V>
size_t FixedHashMap<K,V>::size() {
return _N;
}
/**
* @brief Returns the number of buckets in the fixed-size table.
* @return number of buckets in the map
*/
template <class K, class V>
size_t FixedHashMap<K,V>::bucket_count() {
return _M;
}
/**
* @brief Returns an array of the keys in the fixed-size table.
* @details All buckets are scanned in order to form a list of all keys
* present in the table and then the list is returned. WARNING: The user
* is responsible for freeing the allocated memory once the array is no
* longer needed.
* @return an array of keys in the map whose length is the number of key/value
* pairs in the table.
*/
template <class K, class V>
K* FixedHashMap<K,V>::keys() {
/* allocate array of keys */
K *key_list = new K[_N];
/* fill array with keys */
size_t ind = 0;
for (size_t i=0; i<_M; i++) {
node *iter_node = _buckets[i];
while (iter_node != NULL) {
key_list[ind] = iter_node->key;
iter_node = iter_node->next;
ind++;
}
}
return key_list;
}
/**
* @brief Returns an array of the values in the fixed-size table.
* @details All buckets are scanned in order to form a list of all values
* present in the table and then the list is returned. WARNING: The user
* is responsible for freeing the allocated memory once the array is no
* longer needed.
* @return an array of values in the map whose length is the number of
* key/value pairs in the table.
*/
template <class K, class V>
V* FixedHashMap<K,V>::values() {
/* allocate array of values */
V *values = new V[_N];
/* fill array with values */
size_t ind = 0;
for (size_t i=0; i<_M; i++) {
node *iter_node = _buckets[i];
while (iter_node != NULL) {
values[ind] = iter_node->value;
iter_node = iter_node->next;
ind++;
}
}
return values;
}
/**
* @brief Clears all key/value pairs form the hash table.
*/
template <class K, class V>
void FixedHashMap<K,V>::clear() {
/* for each bucket, scan through linked list and delete all nodes */
for (size_t i=0; i<_M; i++) {
node *iter_node = _buckets[i];
while (iter_node != NULL) {
node *next_node = iter_node->next;
delete iter_node;
iter_node = next_node;
}
}
/* reset each bucket to null */
for (size_t i=0; i<_M; i++)
_buckets[i] = NULL;
/* reset the number of entries to zero */
_N = 0;
}
/**
* @brief Prints the contents of each bucket to the screen.
* @details All buckets are scanned and the contents of the buckets are
* printed, which are pointers to linked lists. If the pointer is NULL
* suggesting that the linked list is empty, NULL is printed to the
* screen.
*/
template <class K, class V>
void FixedHashMap<K,V>::print_buckets() {
log_printf(NORMAL, "Printing all buckets in the hash map...");
for (size_t i=0; i<_M; i++) {
if (_buckets[i] == NULL)
log_printf(NORMAL, "Bucket %d -> NULL", i);
else
log_printf(NORMAL, "Bucket %d -> %p", i, _buckets[i]);
}
}
/**
* @brief Constructor generates initial underlying table as a fixed-sized
* hash map and initializes concurrency structures.
*/
template <class K, class V>
ParallelHashMap<K,V>::ParallelHashMap(size_t M, size_t L) {
/* allocate table */
_table = new FixedHashMap<K,V>(M);
_fixed_size = false;
/* get number of threads and create concurrency structures */
_num_threads = 1;
_num_threads = omp_get_max_threads();
_num_locks = L;
_locks = new omp_lock_t[_num_locks];
for (size_t i=0; i<_num_locks; i++)
omp_init_lock(&_locks[i]);
_announce = new paddedPointer[_num_threads];
for (size_t t=0; t<_num_threads; t++)
_announce[t].value = NULL;
}
/**
* @brief Destructor frees memory associated with fixed-sized hash map and
* concurrency structures.
*/
template <class K, class V>
ParallelHashMap<K,V>::~ParallelHashMap() {
delete _table;
delete [] _locks;
delete [] _announce;
}
/**
* @brief Determine whether the parallel hash map contains a given key.
* @details First the thread accessing the table announces its presence and
* which table it is reading. Then the linked list in the bucket
* associated with the key is searched without setting any locks
* to determine whether the key is present. When the thread has
* finished accessing the table, the announcement is reset to NULL.
* The announcement ensures that the data in the map is not freed
* during a resize until all threads have finished accessing the map.
* @param key key to be searched
* @return boolean value referring to whether the key is contained in the map
*/
template <class K, class V>
bool ParallelHashMap<K,V>::contains(K& key) {
/* get thread ID */
size_t tid = 0;
tid = omp_get_thread_num();
/* get pointer to table, announce it will be searched,
and ensure consistency */
FixedHashMap<K,V> *table_ptr;
do {
table_ptr = _table;
_announce[tid].value = table_ptr;
#pragma omp flush
} while (table_ptr != _table);
/* see if current table contains the thread */
bool present = table_ptr->contains(key);
/* reset table announcement to not searching */
_announce[tid].value = NULL;
return present;
}
/**
* @brief Determine the value associated with a given key.
* @details This function follows the same algorithm as "contains" except that
* the value associated with the searched key is returned.
* First the thread accessing the table acquires the lock corresponding
* with the associated bucket based on the key. Then the linked list
* in the bucket is searched for the key. An exception is thrown if the
* key is not found. When the thread has finished accessing the table,
* it releases the lock.
* @param key key to be searched
* @return value associated with the key
*/
template <class K, class V>
V ParallelHashMap<K,V>::at(K& key) {
/* If the size is fixed, simply return the value from the fixed hash map */
if (_fixed_size)
return _table->at(key);
/* get thread ID */
size_t tid = 0;
tid = omp_get_thread_num();
/* get pointer to table, announce it will be searched */
FixedHashMap<K,V> *table_ptr;
do {
table_ptr = _table;
_announce[tid].value = table_ptr;
#pragma omp flush
} while (table_ptr != _table);
/* get value associated with the key in the underlying table */
V value = table_ptr->at(key);
/* reset table announcement to not searching */
_announce[tid].value = NULL;
return value;
}
/**
* @brief Insert a given key/value pair into the parallel hash map.
* @details First the underlying table is checked to determine if a resize
* should be conducted. Then, the table is checked to see if it
* already contains the key. If so, the key/value pair is not inserted
* and the function returns. Otherwise, the lock of the associated
* bucket is acquired and the key/value pair is added to the bucket.
* @param key key of the key/value pair to be inserted
* @param value value of the key/value pair to be inserted
*/
template <class K, class V>
void ParallelHashMap<K,V>::insert(K key, V value) {
/* check if resize needed */
if (2*_table->size() > _table->bucket_count())
resize();
/* check to see if key is already contained in the table */
if (contains(key))
return;
/* get lock hash */
size_t lock_hash = (std::hash<K>()(key) & (_table->bucket_count() - 1))
% _num_locks;
/* acquire lock */
omp_set_lock(&_locks[lock_hash]);
/* insert value */
_table->insert(key, value);
/* release lock */
omp_unset_lock(&_locks[lock_hash]);
}
/**
* @brief Updates the value associated with a key in the parallel hash map.
* @details The thread first acquires the lock for the bucket associated with
* the key is acquired, then the linked list in the bucket is searched
* for the key. If the key is not found, an exception is returned. When
* the key is found, the value is updated and the lock is released.
* @param key the key of the key/value pair to be updated
* @param value the new value for the key/value pair
*/
template <class K, class V>
void ParallelHashMap<K,V>::update(K& key, V value) {
/* get lock hash */
size_t lock_hash = (std::hash<K>()(key) & (_table->bucket_count() - 1))
% _num_locks;
/* acquire lock */
omp_set_lock(&_locks[lock_hash]);
/* insert value */
_table->at(key) = value;
/* release lock */
omp_unset_lock(&_locks[lock_hash]);
}
/**
* @brief Insert a given key/value pair into the parallel hash map and return
the order number.
* @details First the underlying table is checked to determine if a resize
* should be conducted. Then, the table is checked to see if it
* already contains the key. If so, the key/value pair is not inserted
* and the function returns. Otherwise, the lock of the associated
* bucket is acquired and the key/value pair is added to the bucket.
* @param key key of the key/value pair to be inserted
* @param value value of the key/value pair to be inserted
* @return order number in which the key/value pair was inserted, -1 if it
* already exists
*/
template <class K, class V>
long ParallelHashMap<K,V>::insert_and_get_count(K key, V value) {
/* check if resize needed */
if (2*_table->size() > _table->bucket_count())
resize();
/* check to see if key is already contained in the table */
if (contains(key))
return -1;
/* get lock hash */
size_t lock_hash = (std::hash<K>()(key) & (_table->bucket_count() - 1))
% _num_locks;
/* acquire lock */
omp_set_lock(&_locks[lock_hash]);
/* insert value */
long N =_table->insert_and_get_count(key, value);
/* release lock */
omp_unset_lock(&_locks[lock_hash]);
return N;
}
/**
* @brief Resizes the underlying table to twice its current capacity.
* @details In a thread-safe manner, this procedure resizes the underlying
* FixedHashMap table to twice its current capacity using locks and the
* announce array. First, all locks are set in order to block inserts and
* prevent deadlock. A new table is allocated of twice the size and all
* key/value pairs from the old table, then the pointer is switched to the
* new table and locks are released. Finally the memory needs to be freed.
* To prevent threads currently reading the table from encountering
* segmentation faults, the resizing threads waits for the announce array
* to be free of references to the old table before freeing the memory.
*/
template <class K, class V>
void ParallelHashMap<K,V>::resize() {
/* do not resize if fixed size */
if (_fixed_size)
return;
/* acquire all locks in order */
for (size_t i=0; i<_num_locks; i++)
omp_set_lock(&_locks[i]);
/* recheck if resize needed */
if (2*_table->size() < _table->bucket_count()) {
/* release locks */
for (size_t i=0; i<_num_locks; i++)
omp_unset_lock(&_locks[i]);
return;
}
/* allocate new hash map of double the size */
FixedHashMap<K,V> *new_map =
new FixedHashMap<K,V>(2*_table->bucket_count());
/* get keys, values, and number of elements */
K *key_list = _table->keys();
V *value_list = _table->values();
/* insert key/value pairs into new hash map */
for (size_t i=0; i<_table->size(); i++)
new_map->insert(key_list[i], value_list[i]);
/* save pointer of old table */
FixedHashMap<K,V> *old_table = _table;
/* reassign pointer */
_table = new_map;
#pragma omp flush
/* release all locks */
for (size_t i=0; i<_num_locks; i++)
omp_unset_lock(&_locks[i]);
/* delete key and value list */
delete [] key_list;
delete [] value_list;
/* wait for all threads to stop reading from the old table */
for (size_t i=0; i<_num_threads; i++) {
while (_announce[i].value == old_table) {
#pragma omp flush
continue;
}
}
/* free memory associated with old table */
delete old_table;
}
/**
* @brief Returns the number of key/value pairs in the underlying table.
* @return number of key/value pairs in the map
*/
template <class K, class V>
size_t ParallelHashMap<K,V>::size() {
return _table->size();
}
/**
* @brief Returns the number of buckets in the underlying table.
* @return number of buckets in the map
*/
template <class K, class V>
size_t ParallelHashMap<K,V>::bucket_count() {
return _table->bucket_count();
}
/**
* @brief Returns the number of locks in the parallel hash map.
* @return number of locks in the map
*/
template <class K, class V>
size_t ParallelHashMap<K,V>::num_locks() {
return _num_locks;
}
/**
* @brief Returns an array of the keys in the underlying table.
* @details All buckets are scanned in order to form a list of all keys
* present in the table and then the list is returned. Threads
* announce their presence to ensure table memory is not freed
* during access. WARNING: The user is responsible for freeing the
* allocated memory once the array is no longer needed.
* @return an array of keys in the map whose length is the number of key/value
* pairs in the table.
*/
template <class K, class V>
K* ParallelHashMap<K,V>::keys() {
/* get thread ID */
size_t tid = 0;
tid = omp_get_thread_num();
/* get pointer to table, announce it will be searched */
FixedHashMap<K,V> *table_ptr;
do {
table_ptr = _table;
_announce[tid].value = table_ptr;
#pragma omp flush
} while (table_ptr != _table);
/* get key list */
K* key_list = table_ptr->keys();
/* reset table announcement to not searching */
_announce[tid].value = NULL;
return key_list;
}
/**
* @brief Returns an array of the values in the underlying table.
* @details All buckets are scanned in order to form a list of all values
* present in the table and then the list is returned. Threads
* announce their presence to ensure table memory is not freed
* during access. WARNING: The user is responsible for freeing the
* allocated memory once the array is no longer needed.
* @return an array of values in the map whose length is the number of key/value
* pairs in the table.
*/
template <class K, class V>
V* ParallelHashMap<K,V>::values() {
/* get thread ID */
size_t tid = 0;
tid = omp_get_thread_num();
/* get pointer to table, announce it will be searched */
FixedHashMap<K,V> *table_ptr;
do {
table_ptr = _table;
_announce[tid].value = table_ptr;
#pragma omp flush
} while (table_ptr != _table);
/* get value list */
V* value_list = table_ptr->values();
/* reset table announcement to not searching */
_announce[tid].value = NULL;
return value_list;
}
/**
* @brief Prevents the parallel hash map from further resizing.
*/
template <class K, class V>
void ParallelHashMap<K,V>::setFixedSize() {
_fixed_size = true;
}
/**
* @brief Clears all key/value pairs form the hash table.
*/
template <class K, class V>
void ParallelHashMap<K,V>::clear() {
/* acquire all locks in order */
for (size_t i=0; i<_num_locks; i++)
omp_set_lock(&_locks[i]);
/* clear underlying fixed table */
_table->clear();
/* release all locks in order */
for (size_t i=0; i<_num_locks; i++)
omp_unset_lock(&_locks[i]);
}
/**
* @brief Prints the contents of each bucket to the screen.
* @details All buckets are scanned and the contents of the buckets are
* printed, which are pointers to linked lists. If the pointer is NULL
* suggesting that the linked list is empty, NULL is printed to the
* screen. Threads announce their presence to ensure table memory is
* not freed during access.
*/
template <class K, class V>
void ParallelHashMap<K,V>::print_buckets() {
/* get thread ID */
size_t tid = 0;
tid = omp_get_thread_num();
/* get pointer to table, announce it will be searched */
FixedHashMap<K,V> *table_ptr;
do {
table_ptr = _table;
_announce[tid].value = table_ptr;
#pragma omp flush
} while (table_ptr != _table);
/* print buckets */
table_ptr->print_buckets();
/* reset table announcement to not searching */
_announce[tid].value = NULL;
}
/**
* @brief Reallocates the underlying table to the desired size.
* @param M The requested new size
*/
template <class K, class V>
void ParallelHashMap<K,V>::realloc(size_t M) {
/* delete old table */
delete _table;
/* allocate new table */
_table = new FixedHashMap<K,V>(M);
}
#endif
|
setbv.c | //-------------------------------------------------------------------------//
// //
// This benchmark is an OpenMP C version of the NPB LU code. This OpenMP //
// C version is developed by the Center for Manycore Programming at Seoul //
// National University and derived from the OpenMP Fortran versions in //
// "NPB3.3-OMP" developed by NAS. //
// //
// Permission to use, copy, distribute and modify this software for any //
// purpose with or without fee is hereby granted. This software is //
// provided "as is" without express or implied warranty. //
// //
// Information on NPB 3.3, including the technical report, the original //
// specifications, source code, results and information on how to submit //
// new results, is available at: //
// //
// http://www.nas.nasa.gov/Software/NPB/ //
// //
// Send comments or suggestions for this OpenMP C version to //
// cmp@aces.snu.ac.kr //
// //
// Center for Manycore Programming //
// School of Computer Science and Engineering //
// Seoul National University //
// Seoul 151-744, Korea //
// //
// E-mail: cmp@aces.snu.ac.kr //
// //
//-------------------------------------------------------------------------//
//-------------------------------------------------------------------------//
// Authors: Sangmin Seo, Jungwon Kim, Jun Lee, Jeongho Nah, Gangwon Jo, //
// and Jaejin Lee //
//-------------------------------------------------------------------------//
#include "applu.incl"
//---------------------------------------------------------------------
// set the boundary values of dependent variables
//---------------------------------------------------------------------
void setbv()
{
//---------------------------------------------------------------------
// local variables
//---------------------------------------------------------------------
int i, j, k, m;
double temp1[5], temp2[5];
//---------------------------------------------------------------------
// set the dependent variable values along the top and bottom faces
//---------------------------------------------------------------------
#pragma omp parallel default(shared) private(i,j,k,m,temp1,temp2) \
shared(nx,ny,nz)
{
#pragma omp for schedule(static)
for (j = 0; j < ny; j++) {
for (i = 0; i < nx; i++) {
exact( i, j, 0, temp1 );
exact( i, j, nz-1, temp2 );
for (m = 0; m < 5; m++) {
u[0][j][i][m] = temp1[m];
u[nz-1][j][i][m] = temp2[m];
}
}
}
//---------------------------------------------------------------------
// set the dependent variable values along north and south faces
//---------------------------------------------------------------------
#pragma omp for schedule(static) nowait
for (k = 0; k < nz; k++) {
for (i = 0; i < nx; i++) {
exact( i, 0, k, temp1 );
exact( i, ny-1, k, temp2 );
for (m = 0; m < 5; m++) {
u[k][0][i][m] = temp1[m];
u[k][ny-1][i][m] = temp2[m];
}
}
}
//---------------------------------------------------------------------
// set the dependent variable values along east and west faces
//---------------------------------------------------------------------
#pragma omp for schedule(static) nowait
for (k = 0; k < nz; k++) {
for (j = 0; j < ny; j++) {
exact( 0, j, k, temp1 );
exact( nx-1, j, k, temp2 );
for (m = 0; m < 5; m++) {
u[k][j][0][m] = temp1[m];
u[k][j][nx-1][m] = temp2[m];
}
}
}
} //end parallel
}
|
macro-4.c | /* PR preprocessor/27746 */
/* { dg-do compile } */
/* { dg-options "-fopenmp -Wunknown-pragmas" } */
#define p _Pragma ("omp parallel")
#define omp_p _Pragma ("omp p")
void bar (void);
void
foo (void)
{
#pragma omp p /* { dg-warning "ignoring #pragma omp _Pragma" } */
bar ();
omp_p /* { dg-warning "ignoring #pragma omp _Pragma" } */
bar ();
}
#define parallel serial
#define omp_parallel _Pragma ("omp parallel")
void
baz (void)
{
#pragma omp parallel /* { dg-warning "ignoring #pragma omp serial" } */
bar ();
omp_parallel /* { dg-warning "ignoring #pragma omp serial" } */
bar ();
}
|
mscash1_fmt_plug.c | /* MSCASH patch for john (performance improvement)
*
* Modified for utf-8 support by magnum in 2011, same terms as below
*
* Written by Alain Espinosa <alainesp at gmail.com> in 2007. No copyright
* is claimed, and the software is hereby placed in the public domain.
* In case this attempt to disclaim copyright and place the software in the
* public domain is deemed null and void, then the software is
* Copyright (c) 2007 Alain Espinosa and it is hereby released to the
* general public under the following terms:
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted.
*
* There's ABSOLUTELY NO WARRANTY, express or implied.
*
* (This is a heavily cut-down "BSD license".)
*/
#if FMT_EXTERNS_H
extern struct fmt_main fmt_mscash;
#elif FMT_REGISTERS_H
john_register_one(&fmt_mscash);
#else
#include <string.h>
#include "arch.h"
#include "misc.h"
#include "memory.h"
#include "common.h"
#include "formats.h"
#include "unicode.h"
#include "options.h"
#include "loader.h"
#include "johnswap.h"
#include "mscash_common.h"
#ifdef _OPENMP
#include <omp.h>
#ifndef OMP_SCALE
#define OMP_SCALE 192
#endif
#endif
#include "memdbg.h"
#define FORMAT_LABEL "mscash"
#define FORMAT_NAME "MS Cache Hash (DCC)"
#define ALGORITHM_NAME "MD4 32/" ARCH_BITS_STR
#define PLAINTEXT_LENGTH 27
#define SALT_SIZE (11*4)
#define OK_NUM_KEYS 64
#define BEST_NUM_KEYS 512
#ifdef _OPENMP
#define MS_NUM_KEYS OK_NUM_KEYS
#else
#define MS_NUM_KEYS BEST_NUM_KEYS
#endif
#define MIN_KEYS_PER_CRYPT OK_NUM_KEYS
#define MAX_KEYS_PER_CRYPT MS_NUM_KEYS
static unsigned int *ms_buffer1x;
static unsigned int *output1x;
static unsigned int *crypt_out;
static unsigned int *last;
static unsigned int *last_i;
static unsigned int *salt_buffer;
static unsigned int new_key;
//Init values
#define INIT_A 0x67452301
#define INIT_B 0xefcdab89
#define INIT_C 0x98badcfe
#define INIT_D 0x10325476
#define SQRT_2 0x5a827999
#define SQRT_3 0x6ed9eba1
static void set_key_utf8(char *_key, int index);
static void set_key_encoding(char *_key, int index);
struct fmt_main fmt_mscash;
#if !ARCH_LITTLE_ENDIAN
inline static void swap(unsigned int *x, int count)
{
while (count--) {
*x = JOHNSWAP(*x);
x++;
}
}
#endif
static void init(struct fmt_main *self)
{
#ifdef _OPENMP
int omp_t = omp_get_max_threads();
self->params.min_keys_per_crypt *= omp_t;
omp_t *= OMP_SCALE;
fmt_mscash.params.max_keys_per_crypt *= omp_t;
#endif
ms_buffer1x = mem_calloc(sizeof(ms_buffer1x[0]), 16*fmt_mscash.params.max_keys_per_crypt);
output1x = mem_calloc(sizeof(output1x[0]) , 4*fmt_mscash.params.max_keys_per_crypt);
crypt_out = mem_calloc(sizeof(crypt_out[0]) , 4*fmt_mscash.params.max_keys_per_crypt);
last = mem_calloc(sizeof(last[0]) , 4*fmt_mscash.params.max_keys_per_crypt);
last_i = mem_calloc(sizeof(last_i[0]) , fmt_mscash.params.max_keys_per_crypt);
new_key=1;
mscash1_adjust_tests(self, options.target_enc, PLAINTEXT_LENGTH,
set_key_utf8, set_key_encoding);
}
static void done(void)
{
MEM_FREE(last_i);
MEM_FREE(last);
MEM_FREE(crypt_out);
MEM_FREE(output1x);
MEM_FREE(ms_buffer1x);
}
static void set_salt(void *salt)
{
salt_buffer=salt;
}
static void *get_salt(char *ciphertext)
{
unsigned char input[19*3+1];
int i, utf16len;
char *lasth = strrchr(ciphertext, '#');
union {
UTF16 u16[22];
uint32_t w[11];
} static out;
memset(out.u16, 0, sizeof(out.u16));
ciphertext += FORMAT_TAG_LEN;
for (i = 0; &ciphertext[i] < lasth; i++)
input[i] = ciphertext[i];
input[i] = 0;
utf16len = enc_to_utf16(out.u16, 19, input, i);
if (utf16len < 0)
utf16len = strlen16(out.u16);
#if ARCH_LITTLE_ENDIAN
out.u16[utf16len] = 0x80;
#else
out.u16[utf16len] = 0x8000;
swap(out.w, (i>>1)+1);
#endif
out.w[10] = (8 + utf16len) << 4;
// dump_stuff(out.w, 44);
return out.u16;
}
static void *get_binary(char *ciphertext)
{
static unsigned int out[BINARY_SIZE/sizeof(unsigned int)];
unsigned int i=0;
unsigned int temp;
unsigned int *salt=fmt_mscash.methods.salt(ciphertext);
/* We need to allow salt containing '#' so we search backwards */
ciphertext = strrchr(ciphertext, '#') + 1;
for (; i<4 ;i++)
{
temp = ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+0])]))<<4;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+1])]));
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+2])]))<<12;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+3])]))<<8;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+4])]))<<20;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+5])]))<<16;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+6])]))<<28;
temp |= ((unsigned int)(atoi16[ARCH_INDEX(ciphertext[i*8+7])]))<<24;
out[i]=temp;
}
out[0] -= INIT_A;
out[1] -= INIT_B;
out[2] -= INIT_C;
out[3] -= INIT_D;
// Reversed b += (c ^ d ^ a) + salt_buffer[11] + SQRT_3; b = (b << 15) | (b >> 17);
out[1] = (out[1] >> 15) | (out[1] << 17);
out[1] -= SQRT_3 + (out[2] ^ out[3] ^ out[0]);
// Reversed c += (d ^ a ^ b) + salt_buffer[3] + SQRT_3; c = (c << 11) | (c >> 21);
out[2] = (out[2] << 21) | (out[2] >> 11);
out[2]-= SQRT_3 + (out[3] ^ out[0] ^ out[1]) + salt[3];
// Reversed d += (a ^ b ^ c) + salt_buffer[7] + SQRT_3; d = (d << 9 ) | (d >> 23);
out[3] = (out[3] << 23) | (out[3] >> 9);
out[3] -= SQRT_3 + (out[0] ^ out[1] ^ out[2]) + salt[7];
//+ SQRT_3; d = (d << 9 ) | (d >> 23);
out[3]=(out[3] << 23 ) | (out[3] >> 9);
out[3]-=SQRT_3;
return out;
}
static int binary_hash_0(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_0; }
static int binary_hash_1(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_1; }
static int binary_hash_2(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_2; }
static int binary_hash_3(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_3; }
static int binary_hash_4(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_4; }
static int binary_hash_5(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_5; }
static int binary_hash_6(void *binary) { return ((unsigned int*)binary)[3] & PH_MASK_6; }
static int get_hash_0(int index) { return output1x[4*index+3] & PH_MASK_0; }
static int get_hash_1(int index) { return output1x[4*index+3] & PH_MASK_1; }
static int get_hash_2(int index) { return output1x[4*index+3] & PH_MASK_2; }
static int get_hash_3(int index) { return output1x[4*index+3] & PH_MASK_3; }
static int get_hash_4(int index) { return output1x[4*index+3] & PH_MASK_4; }
static int get_hash_5(int index) { return output1x[4*index+3] & PH_MASK_5; }
static int get_hash_6(int index) { return output1x[4*index+3] & PH_MASK_6; }
static void nt_hash(int count)
{
int i;
#if MS_NUM_KEYS > 1 && defined(_OPENMP)
#pragma omp parallel for default(none) private(i) shared(count, ms_buffer1x, crypt_out, last)
#endif
for (i = 0; i < count; i++)
{
unsigned int a;
unsigned int b;
unsigned int c;
unsigned int d;
/* Round 1 */
a = 0xFFFFFFFF + ms_buffer1x[16*i+0];a = (a << 3 ) | (a >> 29);
d = INIT_D + (INIT_C ^ (a & 0x77777777)) + ms_buffer1x[16*i+1];d = (d << 7 ) | (d >> 25);
c = INIT_C + (INIT_B ^ (d & (a ^ INIT_B)))+ ms_buffer1x[16*i+2];c = (c << 11) | (c >> 21);
b = INIT_B + (a ^ (c & (d ^ a))) + ms_buffer1x[16*i+3];b = (b << 19) | (b >> 13);
a += (d ^ (b & (c ^ d))) + ms_buffer1x[16*i+4] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + ms_buffer1x[16*i+5] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + ms_buffer1x[16*i+6] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a))) + ms_buffer1x[16*i+7] ;b = (b << 19) | (b >> 13);
a += (d ^ (b & (c ^ d))) + ms_buffer1x[16*i+8] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + ms_buffer1x[16*i+9] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + ms_buffer1x[16*i+10] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a))) + ms_buffer1x[16*i+11] ;b = (b << 19) | (b >> 13);
a += (d ^ (b & (c ^ d))) + ms_buffer1x[16*i+12] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + ms_buffer1x[16*i+13] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + ms_buffer1x[16*i+14] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a)))/*+ms_buffer1x[16*i+15]*/;b = (b << 19) | (b >> 13);
/* Round 2 */
a += ((b & (c | d)) | (c & d)) + ms_buffer1x[16*i+0] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + ms_buffer1x[16*i+4] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + ms_buffer1x[16*i+8] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + ms_buffer1x[16*i+12] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + ms_buffer1x[16*i+1] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + ms_buffer1x[16*i+5] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + ms_buffer1x[16*i+9] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + ms_buffer1x[16*i+13] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + ms_buffer1x[16*i+2] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + ms_buffer1x[16*i+6] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + ms_buffer1x[16*i+10] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + ms_buffer1x[16*i+14] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + ms_buffer1x[16*i+3] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + ms_buffer1x[16*i+7] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + ms_buffer1x[16*i+11] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a))/*+ms_buffer1x[16*i+15]*/+SQRT_2; b = (b << 13) | (b >> 19);
/* Round 3 */
a += (b ^ c ^ d) + ms_buffer1x[16*i+0] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + ms_buffer1x[16*i+8] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + ms_buffer1x[16*i+4] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) + ms_buffer1x[16*i+12] + SQRT_3; b = (b << 15) | (b >> 17);
a += (b ^ c ^ d) + ms_buffer1x[16*i+2] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + ms_buffer1x[16*i+10] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + ms_buffer1x[16*i+6] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) + ms_buffer1x[16*i+14] + SQRT_3; b = (b << 15) | (b >> 17);
a += (b ^ c ^ d) + ms_buffer1x[16*i+1] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + ms_buffer1x[16*i+9] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + ms_buffer1x[16*i+5] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) + ms_buffer1x[16*i+13] + SQRT_3; b = (b << 15) | (b >> 17);
a += (b ^ c ^ d) + ms_buffer1x[16*i+3] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + ms_buffer1x[16*i+11] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + ms_buffer1x[16*i+7] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) /*+ ms_buffer1x[16*i+15] */+ SQRT_3; b = (b << 15) | (b >> 17);
crypt_out[4*i+0] = a + INIT_A;
crypt_out[4*i+1] = b + INIT_B;
crypt_out[4*i+2] = c + INIT_C;
crypt_out[4*i+3] = d + INIT_D;
//Another MD4_crypt for the salt
/* Round 1 */
a= 0xFFFFFFFF +crypt_out[4*i+0]; a=(a<<3 )|(a>>29);
d=INIT_D + ( INIT_C ^ ( a & 0x77777777)) +crypt_out[4*i+1]; d=(d<<7 )|(d>>25);
c=INIT_C + ( INIT_B ^ ( d & ( a ^ INIT_B))) +crypt_out[4*i+2]; c=(c<<11)|(c>>21);
b=INIT_B + ( a ^ ( c & ( d ^ a ))) +crypt_out[4*i+3]; b=(b<<19)|(b>>13);
last[4*i+0]=a;
last[4*i+1]=b;
last[4*i+2]=c;
last[4*i+3]=d;
}
}
static int crypt_all(int *pcount, struct db_salt *salt)
{
int count = *pcount;
int i;
if (new_key)
{
new_key=0;
nt_hash(count);
}
#if MS_NUM_KEYS > 1 && defined(_OPENMP)
#pragma omp parallel for default(none) private(i) shared(count, last, crypt_out, salt_buffer, output1x)
#endif
for (i = 0; i < count; i++)
{
unsigned int a;
unsigned int b;
unsigned int c;
unsigned int d;
a = last[4*i+0];
b = last[4*i+1];
c = last[4*i+2];
d = last[4*i+3];
a += (d ^ (b & (c ^ d))) + salt_buffer[0] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + salt_buffer[1] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + salt_buffer[2] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a))) + salt_buffer[3] ;b = (b << 19) | (b >> 13);
a += (d ^ (b & (c ^ d))) + salt_buffer[4] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + salt_buffer[5] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + salt_buffer[6] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a))) + salt_buffer[7] ;b = (b << 19) | (b >> 13);
a += (d ^ (b & (c ^ d))) + salt_buffer[8] ;a = (a << 3 ) | (a >> 29);
d += (c ^ (a & (b ^ c))) + salt_buffer[9] ;d = (d << 7 ) | (d >> 25);
c += (b ^ (d & (a ^ b))) + salt_buffer[10] ;c = (c << 11) | (c >> 21);
b += (a ^ (c & (d ^ a)))/*+salt_buffer[11]*/;b = (b << 19) | (b >> 13);
/* Round 2 */
a += ((b & (c | d)) | (c & d)) + crypt_out[4*i+0] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + salt_buffer[0] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + salt_buffer[4] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + salt_buffer[8] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + crypt_out[4*i+1] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + salt_buffer[1] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + salt_buffer[5] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + salt_buffer[9] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + crypt_out[4*i+2] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + salt_buffer[2] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + salt_buffer[6] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a)) + salt_buffer[10] + SQRT_2; b = (b << 13) | (b >> 19);
a += ((b & (c | d)) | (c & d)) + crypt_out[4*i+3] + SQRT_2; a = (a << 3 ) | (a >> 29);
d += ((a & (b | c)) | (b & c)) + salt_buffer[3] + SQRT_2; d = (d << 5 ) | (d >> 27);
c += ((d & (a | b)) | (a & b)) + salt_buffer[7] + SQRT_2; c = (c << 9 ) | (c >> 23);
b += ((c & (d | a)) | (d & a))/*+ salt_buffer[11]*/+ SQRT_2; b = (b << 13) | (b >> 19);
/* Round 3 */
a += (b ^ c ^ d) + crypt_out[4*i+0] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + salt_buffer[4] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + salt_buffer[0] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) + salt_buffer[8] + SQRT_3; b = (b << 15) | (b >> 17);
a += (b ^ c ^ d) + crypt_out[4*i+2] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + salt_buffer[6] + SQRT_3; d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + salt_buffer[2] + SQRT_3; c = (c << 11) | (c >> 21);
b += (c ^ d ^ a) + salt_buffer[10] + SQRT_3; b = (b << 15) | (b >> 17);
a += (b ^ c ^ d) + crypt_out[4*i+1] + SQRT_3; a = (a << 3 ) | (a >> 29);
d += (a ^ b ^ c) + salt_buffer[5];
output1x[4*i+0]=a;
output1x[4*i+1]=b;
output1x[4*i+2]=c;
output1x[4*i+3]=d;
}
return count;
}
static int cmp_all(void *binary, int count)
{
unsigned int i=0;
unsigned int d=((unsigned int *)binary)[3];
for (;i<count;i++)
if (d==output1x[i*4+3])
return 1;
return 0;
}
static int cmp_one(void * binary, int index)
{
unsigned int *t=(unsigned int *)binary;
unsigned int a=output1x[4*index+0];
unsigned int b=output1x[4*index+1];
unsigned int c=output1x[4*index+2];
unsigned int d=output1x[4*index+3];
if (d!=t[3])
return 0;
d+=SQRT_3;d = (d << 9 ) | (d >> 23);
c += (d ^ a ^ b) + salt_buffer[1] + SQRT_3; c = (c << 11) | (c >> 21);
if (c!=t[2])
return 0;
b += (c ^ d ^ a) + salt_buffer[9] + SQRT_3; b = (b << 15) | (b >> 17);
if (b!=t[1])
return 0;
a += (b ^ c ^ d) + crypt_out[4*index+3]+ SQRT_3; a = (a << 3 ) | (a >> 29);
return (a==t[0]);
}
static int cmp_exact(char *source, int index)
{
// This check is for the unreal case of collisions.
// It verifies that the salts are the same.
unsigned int *salt=fmt_mscash.methods.salt(source);
unsigned int i=0;
for (;i<11;i++)
if (salt[i]!=salt_buffer[i])
return 0;
return 1;
}
// This is common code for the SSE/MMX/generic variants of non-UTF8 set_key
inline static void set_key_helper(unsigned int * keybuffer,
unsigned int xBuf,
const unsigned char * key,
unsigned int lenStoreOffset,
unsigned int *last_length)
{
unsigned int i=0;
unsigned int md4_size=0;
for (; key[md4_size] && md4_size < PLAINTEXT_LENGTH; i += xBuf, md4_size++)
{
unsigned int temp;
if ((temp = key[++md4_size]))
{
keybuffer[i] = key[md4_size-1] | (temp << 16);
}
else
{
keybuffer[i] = key[md4_size-1] | 0x800000;
goto key_cleaning;
}
}
keybuffer[i] = 0x80;
key_cleaning:
i += xBuf;
for (;i <= *last_length; i += xBuf)
keybuffer[i] = 0;
*last_length = (md4_size >> 1)+1;
keybuffer[lenStoreOffset] = md4_size << 4;
}
static void set_key(char *_key, int index)
{
set_key_helper(&ms_buffer1x[index << 4], 1, (unsigned char *)_key, 14,
&last_i[index]);
//new password_candidate
new_key=1;
}
// UTF-8 conversion right into key buffer
// This is common code for the SSE/MMX/generic variants
inline static void set_key_helper_utf8(unsigned int * keybuffer, unsigned int xBuf,
const UTF8 * source, unsigned int lenStoreOffset, unsigned int *lastlen)
{
unsigned int *target = keybuffer;
UTF32 chl, chh = 0x80;
unsigned int outlen = 0;
while (*source) {
chl = *source;
if (chl >= 0xC0) {
unsigned int extraBytesToRead = opt_trailingBytesUTF8[chl & 0x3f];
switch (extraBytesToRead) {
case 3:
++source;
if (*source) {
chl <<= 6;
chl += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 2:
++source;
if (*source) {
chl <<= 6;
chl += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 1:
++source;
if (*source) {
chl <<= 6;
chl += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 0:
break;
default:
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
chl -= offsetsFromUTF8[extraBytesToRead];
}
source++;
outlen++;
if (chl > UNI_MAX_BMP) {
if (outlen == PLAINTEXT_LENGTH) {
chh = 0x80;
*target = (chh << 16) | chl;
target += xBuf;
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
break;
}
#define halfBase 0x0010000UL
#define halfShift 10
#define halfMask 0x3FFUL
#define UNI_SUR_HIGH_START (UTF32)0xD800
#define UNI_SUR_LOW_START (UTF32)0xDC00
chl -= halfBase;
chh = (UTF16)((chl & halfMask) + UNI_SUR_LOW_START);;
chl = (UTF16)((chl >> halfShift) + UNI_SUR_HIGH_START);
outlen++;
} else if (*source && outlen < PLAINTEXT_LENGTH) {
chh = *source;
if (chh >= 0xC0) {
unsigned int extraBytesToRead =
opt_trailingBytesUTF8[chh & 0x3f];
switch (extraBytesToRead) {
case 3:
++source;
if (*source) {
chl <<= 6;
chl += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 2:
++source;
if (*source) {
chh <<= 6;
chh += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 1:
++source;
if (*source) {
chh <<= 6;
chh += *source;
} else {
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
case 0:
break;
default:
*lastlen = ((PLAINTEXT_LENGTH >> 1) + 1) * xBuf;
return;
}
chh -= offsetsFromUTF8[extraBytesToRead];
}
source++;
outlen++;
} else {
chh = 0x80;
*target = chh << 16 | chl;
target += xBuf;
break;
}
*target = chh << 16 | chl;
target += xBuf;
}
if (chh != 0x80 || outlen == 0) {
*target = 0x80;
target += xBuf;
}
while(target < &keybuffer[*lastlen]) {
*target = 0;
target += xBuf;
}
*lastlen = ((outlen >> 1) + 1) * xBuf;
keybuffer[lenStoreOffset] = outlen << 4;
}
static void set_key_utf8(char *_key, int index)
{
set_key_helper_utf8(&ms_buffer1x[index << 4], 1, (UTF8 *)_key, 14,
&last_i[index]);
//new password_candidate
new_key=1;
}
// This is common code for the SSE/MMX/generic variants of non-UTF8 non-ISO-8859-1 set_key
inline static void set_key_helper_encoding(unsigned int * keybuffer,
unsigned int xBuf,
const unsigned char * key,
unsigned int lenStoreOffset,
unsigned int *last_length)
{
unsigned int i=0;
int md4_size;
md4_size = enc_to_utf16( (UTF16 *)keybuffer, PLAINTEXT_LENGTH, (UTF8 *) key, strlen((char*)key));
if (md4_size < 0)
md4_size = strlen16((UTF16 *)keybuffer);
#if ARCH_LITTLE_ENDIAN
((UTF16*)keybuffer)[md4_size] = 0x80;
#else
((UTF16*)keybuffer)[md4_size] = 0x8000;
#endif
((UTF16*)keybuffer)[md4_size+1] = 0;
#if !ARCH_LITTLE_ENDIAN
((UTF16*)keybuffer)[md4_size+2] = 0;
#endif
i = md4_size>>1;
i += xBuf;
for (;i <= *last_length; i += xBuf)
keybuffer[i] = 0;
#if !ARCH_LITTLE_ENDIAN
swap(keybuffer, (md4_size>>1)+1);
#endif
*last_length = (md4_size >> 1) + 1;
keybuffer[lenStoreOffset] = md4_size << 4;
}
static void set_key_encoding(char *_key, int index)
{
set_key_helper_encoding(&ms_buffer1x[index << 4], 1, (unsigned char *)_key, 14,
&last_i[index]);
//new password_candidate
new_key=1;
}
// Get the key back from the key buffer, from UCS-2 LE
static char *get_key(int index)
{
static union {
UTF16 u16[PLAINTEXT_LENGTH + 1];
unsigned int u32[(PLAINTEXT_LENGTH + 1 + 1) / 2];
} key;
unsigned int * keybuffer = &ms_buffer1x[index << 4];
unsigned int md4_size;
unsigned int i=0;
int len = keybuffer[14] >> 4;
for (md4_size = 0; md4_size < len; i++, md4_size += 2)
{
#if ARCH_LITTLE_ENDIAN
key.u16[md4_size] = keybuffer[i];
key.u16[md4_size+1] = keybuffer[i] >> 16;
#else
key.u16[md4_size] = keybuffer[i] >> 16;
key.u16[md4_size+1] = keybuffer[i];
#endif
}
#if !ARCH_LITTLE_ENDIAN
swap(key.u32, md4_size >> 1);
#endif
key.u16[len] = 0x00;
return (char *)utf16_to_enc(key.u16);
}
// Public domain hash function by DJ Bernstein (salt is a username)
static int salt_hash(void *salt)
{
UTF16 *s = salt;
unsigned int hash = 5381;
while (*s != 0x80)
hash = ((hash << 5) + hash) ^ *s++;
return hash & (SALT_HASH_SIZE - 1);
}
struct fmt_main fmt_mscash = {
{
FORMAT_LABEL,
FORMAT_NAME,
ALGORITHM_NAME,
BENCHMARK_COMMENT,
BENCHMARK_LENGTH,
0,
PLAINTEXT_LENGTH,
BINARY_SIZE,
BINARY_ALIGN,
SALT_SIZE,
SALT_ALIGN,
MIN_KEYS_PER_CRYPT,
MAX_KEYS_PER_CRYPT,
FMT_CASE | FMT_8_BIT | FMT_SPLIT_UNIFIES_CASE | FMT_OMP | FMT_UNICODE | FMT_UTF8,
{ NULL },
{ FORMAT_TAG },
mscash1_common_tests
}, {
init,
done,
fmt_default_reset,
mscash1_common_prepare,
mscash1_common_valid,
mscash1_common_split,
get_binary, // NOTE, not using the 'common' binary function.
get_salt,
{ NULL },
fmt_default_source,
{
binary_hash_0,
binary_hash_1,
binary_hash_2,
binary_hash_3,
binary_hash_4,
binary_hash_5,
binary_hash_6
},
salt_hash,
NULL,
set_salt,
set_key,
get_key,
fmt_default_clear_keys,
crypt_all,
{
get_hash_0,
get_hash_1,
get_hash_2,
get_hash_3,
get_hash_4,
get_hash_5,
get_hash_6
},
cmp_all,
cmp_one,
cmp_exact
}
};
#endif /* plugin stanza */
|
libartificial.h | /*
* libartificial - Small header-only C library for Artificial Neural Networks
*
* Copyright (c) 2018 Jim Karoukis
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
*/
#ifndef libartificial_h__
#define libartificial_h__
#ifdef __WIN32__
#if defined(COMPILING_DLL)
#define PUBLIC_API __declspec(dllexport)
#else
#define PUBLIC_API __declspec(dllimport)
#endif
#else
#define PUBLIC_API
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <time.h>
#define KRED "\x1B[31m"
#define KGRN "\x1B[32m"
#define RESET "\033[0m"
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Forward declarations
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// UTILITIES ///////////////////////////////////////////////////////////////////////////////////////////////////////////
// For randomization
static inline void swap(double *restrict a, double *restrict b);
// For normalization
static inline double mean(const int *restrict rows, const double *restrict col);
static inline double stdev(const int *restrict rows, const double *restrict col, const double *restrict mean);
// To convert array of names into array of ints for fast gradient and activations
static inline int *name2int(const int *restrict layers, char funcs[(*layers) + 1][30]);
// Activations
static inline void activate(double *restrict Y, const double *restrict X,
const int *restrict r, const int *restrict c,
const int *restrict f);
// Gradients
static inline void gradient(double *restrict Y, const double *restrict X,
const int *restrict r, const int *restrict c,
const int *restrict f);
// Losses
static inline double rmse(const int *restrict r, const int *restrict c,
const double *restrict Y, const double *restrict Z);
static inline double xentropy(const int *restrict r, const int *restrict c,
const double *restrict Y, const double *restrict Z);
// Store already trained weights (used by cpu_gd_train)
static inline void save_w(double **restrict weights, const int *restrict layers, const int *restrict nodes,
const int *restrict cols_Y, const int *restrict cols_X);
// Convolution utilities
static inline int ***imgpad(int ***restrict images, const int *restrict no_of_images,
const int *restrict img_width,
const int *restrict img_height,
const int *restrict img_channels,
const int *restrict padding, // Zeros around
const int *restrict delete_originals); // 0 = no, 1 = yes (keep only vector in memory)
static inline int **im2col(int ***restrict images,const int *restrict no_of_images,
const int *restrict img_width,
const int *restrict img_height,
const int *restrict img_channels,
const int *restrict spatial, // width and height of weights
const int *restrict stride, // (img_width - spatial + 2 * padding)/stride should be int
const int *restrict padding, // Zeros around
const int *restrict delete_originals); // 0 = no, 1 = yes (keep only vectorized in memory)
// Freedom
static inline void delete_Z(const int *restrict layers, double ***restrict Z);
static inline void delete_img_vector(const int *restrict no_of_images, int **restrict images);
// End of UTILITIES ////////////////////////////////////////////////////////////////////////////////////////////////////
// Training Utilities
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Feedforward pass
static inline void cpu_mm_notrans(const double *restrict A, const double *restrict B, double *restrict C,
const int *restrict rows, const int *restrict cols, const int *restrict coms);
static inline void cpu_feedforward_update(const int *restrict r, const int *restrict cY,
const int *restrict cX, const int *restrict layers,
double ***restrict Z,
const double *restrict X, double **restrict w,
const int *restrict n, const int *restrict f);
static inline double ***cpu_feedforward_cache(const int *restrict r, const int *restrict cY,
const int *restrict cX, const int *restrict layers,
const double *restrict X, double **restrict w,
const int *restrict n, const int *restrict f);
// Deltas
static inline void cpu_mm_notrans_trans(const double *restrict A, const double *restrict B, double *restrict C,
const int *restrict rows, const int *restrict cols, const int *restrict coms);
static inline void cpu_gd_delta(double **restrict d, double **restrict h1, double **restrict h2,
const int *restrict r, const int *restrict c, const int *restrict layers,
const double *restrict Y, double ***restrict Z, double **restrict w,
const int *restrict n, const int *restrict f);
// returns new wb
static inline void cpu_threaded_update(const double *restrict X, const double *restrict d,
double *restrict gw,
double *restrict w,
const int *restrict m,
const int *restrict n,
const int *restrict k,
const double *restrict c);
// End of forward declarations
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// PUBLIC API
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
static inline void PUBLIC_API randomize(double *restrict X, const int *restrict rows, const int *restrict cols)
{
// Use a different seed value so that we don't get same
// result each time we run this program
srand(time(NULL));
// Start from the last element and swap one by one. We don't
// need to run for the first element that's why --i > 0
int i = (*rows) * (*cols) - 1;
do {
// Pick a random index from 0 to i
int j = rand() % (i+1);
// Swap X[i] with the element at random index
swap(&X[i], &X[j]);
} while (--i > 0);
}
static inline void PUBLIC_API normalize(double *restrict X, const int *restrict rows, const int *restrict cols)
{
int i, j;
double m = 0.0, sd = 0.0, col[(*rows)];
for (j = 0; j < (*cols); j++) {
if (j != 0) {
for (i = 0; i < (*rows); i++) {
col[i] = X[i * (*cols) + j];
}
m = mean(rows, col);
sd = stdev(rows, col, &m);
i = (*rows) - 1;
for (i = 0; i < (*rows); i++) {
X[i * (*cols) + j] = (X[i * (*cols) + j] - m)/sd;
}
m = 0.0;
sd = 0.0;
}
}
}
static inline double PUBLIC_API rand_normal(const double mu, const double sigma)
{
static double n2 = 0.0;
static double n2_cached = 0.0;
if (!n2_cached) {
double x, y, r;
do
{
x = 2.0 * (double)rand()/RAND_MAX - 1;
y = 2.0 * (double)rand()/RAND_MAX - 1;
r = x*x + y*y;
} while (r == 0.0 || r > 1.0);
double d = sqrt(-2.0 * log(r)/r);
double n1 = x * d;
n2 = y*d;
double result = n1 * sigma + mu;
n2_cached = 1.0;
return result;
} else {
n2_cached = 0.0;
return n2 * sigma + mu;
}
}
// Variance is needed since depending on the data, tanh/relu may give nans.
// Variance < 1 and close to 0.01 if data range too large
static inline PUBLIC_API double **init_w(const double *restrict variance, const int *restrict layers,
const int *restrict nodes, char funcs[(*layers)][30],
const int *restrict cols_Y, const int *restrict cols_X)
{
// l layers
int l = (*layers), prod;
// wb[0] is weights;
// wb[1] is biases;
double **weights = malloc((l + 1) * sizeof(double *));
// For the heuristics of weight initialization
double correction;
do {
int isRelu = strcmp(funcs[l], "relu") + strcmp(funcs[l], "lrelu");
int isTanh = strcmp(funcs[l], "tanh");
switch (l > 0 && l < (*layers)) {
// The statement is false
case 0:
switch (l == 0) {
// The statement is true
case 1:
prod = (*cols_X) * nodes[l];
weights[l] = malloc(prod * sizeof(double));
switch (isRelu == 2 && isTanh == 1) {
// One of the two is false
case 0:
switch (isRelu == 1) {
// Either relu or lrelu
case 1:
// He et al.
correction = sqrt(2.0/(double)(*cols_X));
break;
default:
// Xavier
correction = sqrt(1.0/(double)(*cols_X));
break;
}
break;
default:
correction = sqrt(2.0/(double)(prod));
break;
}
--prod;
do {
weights[l][prod] = correction * rand_normal(0.0, (*variance));
} while (--prod >= 0);
break;
// l = layers
default:
prod = nodes[l-1] * (*cols_Y);
weights[l] = malloc(prod * sizeof(double));
switch (isRelu == 2 && isTanh == 1) {
case 0:
switch (isRelu == 1) {
// Either relu or lrelu
case 1:
// He et al.
correction = sqrt(2.0/(double)nodes[l-1]);
break;
default:
// Xavier
correction = sqrt(1.0/(double)nodes[l-1]);
break;
}
break;
default:
correction = sqrt(2.0/(double)(prod));
break;
}
--prod;
do {
weights[l][prod] = correction * rand_normal(0.0, (*variance));
} while (--prod >= 0);
break;
}
break;
// We are in between input and output
default:
prod = nodes[l-1] * nodes[l];
weights[l] = malloc(prod * sizeof(double));
switch (isRelu == 2 && isTanh == 1) {
case 0:
switch (isRelu == 1) {
// Either relu or lrelu
case 1:
// He et al.
correction = sqrt(2.0/(double)nodes[l-1]);
break;
default:
// Xavier
correction = sqrt(1.0/(double)nodes[l-1]);
break;
}
break;
default:
correction = sqrt(2.0/(double)(prod));
break;
}
--prod;
do {
weights[l][prod] = correction * rand_normal(0.0, (*variance));
} while (--prod >= 0);
break;
}
} while (--l >= 0);
printf(KGRN "\nWeights and biases initialized successfully!\n" RESET);
return weights;
}
// Load pretrained wb files
static inline double PUBLIC_API **load_w(const int *restrict layers, const int *restrict nodes,
const int *restrict cols_Y, const int *restrict cols_X)
{
int l;
char path[1024];
getcwd(path, sizeof(path));
char w_path[1036];
strcpy(w_path, path);
strcat(w_path, "/weights");
double **w = malloc(((*layers) + 1) * sizeof(double *));
w[0] = malloc((*cols_X) * nodes[0] * sizeof(double));
switch ((*layers) > 1) {
case 1:
for (l = 1; l < (*layers); l++) {
w[l] = malloc(nodes[l-1] * nodes[l] * sizeof(double));
}
break;
default:
break;
}
w[(*layers)] = malloc(nodes[(*layers) - 1] * (*cols_Y) * sizeof(double));
FILE *ptr_fp;
for (l = 0; l < (*layers) + 1; l++) {
// This needs to change every time
char w_path_filename[1050];
char number[100];
char filename[15] = "layer_";
sprintf(number, "%d", l);
strcat(filename, number);
strcat(filename, ".bin");
strcpy(w_path_filename, w_path);
strcat(w_path_filename, "/");
strcat(w_path_filename, filename);
if((ptr_fp = fopen(w_path_filename, "rb")) == NULL) {
printf("Unable to open file!\n");
exit(1);
}
if (l == 0) {
if(fread(w[l], (*cols_X) * nodes[l] * sizeof(double), 1, ptr_fp) != 1) {
printf("Read error!\n");
exit(1);
}
fclose(ptr_fp);
} else if (l == (*layers)) {
if(fread(w[l], nodes[l-1] * (*cols_Y) * sizeof(double), 1, ptr_fp) != 1) {
printf("Read error!\n");
exit(1);
}
fclose(ptr_fp);
} else {
if(fread(w[l], nodes[l-1] * nodes[l] * sizeof(double), 1, ptr_fp) != 1) {
printf("Read error!\n");
exit(1);
}
fclose(ptr_fp);
}
}
return w;
}
// Free wb
static inline void PUBLIC_API delete_w(const int *restrict layers, double **restrict w)
{
int l;
for (l = 0; l < (*layers) + 1; l++) free(w[l]);
free(w);
}
// Actual perceptron
static inline double PUBLIC_API *cpu_feedforward_predict(const int *restrict rows, const int *restrict cols_Y,
const int *restrict cols_X, const int *restrict layers,
const double *restrict X, double **restrict w,
const int *restrict nodes, char funcs[(*layers) + 1][30])
{
// l is for layers
// i is for each row * column of X, Y
int l = (*layers);
int r_times_col = (*rows) * (*cols_Y);
int *f;
double *Z_pred = malloc(r_times_col * sizeof(double));
if (Z_pred) {
memset(Z_pred, 0.0, r_times_col * sizeof(double));
} else {
printf(KRED "\nCould not allocate Z_prediction. Aborting...\n" RESET);
abort();
}
// feeds at every layer
double ***Z = malloc(2 * sizeof(double **));
if (Z) {
Z[0] = malloc(((*layers) + 1) * sizeof(double *));
Z[1] = malloc(((*layers) + 1) * sizeof(double *));
if (Z[0] && Z[1]) {
Z[0][l] = malloc(r_times_col * sizeof(double));
Z[1][l] = malloc(r_times_col * sizeof(double));
if (Z[0][l] && Z[1][l]) {
for (l = 0; l < (*layers); l++) {
r_times_col = (*rows) * nodes[l];
Z[0][l] = malloc(r_times_col * sizeof(double));
Z[1][l] = malloc(r_times_col * sizeof(double));
if (Z[0][l] && Z[1][l]) {
memset(Z[0][l], 0.0, r_times_col * sizeof(double));
memset(Z[1][l], 0.0, r_times_col * sizeof(double));
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z[0]);
free(Z[1]);
free(Z);
free(Z_pred);
abort();
}
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z[0]);
free(Z[1]);
free(Z);
free(Z_pred);
abort();
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z);
free(Z_pred);
abort();
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z_pred);
abort();
}
f = name2int(layers, funcs);
// Directly manipulates Z
cpu_feedforward_update(rows, cols_Y, cols_X, layers, Z, X, w, nodes, f);
memcpy(Z_pred, Z[1][(*layers)], (*rows) * (*cols_Y) * sizeof(double));
free(f);
delete_Z(layers, Z);
return Z_pred;
}
static inline void PUBLIC_API cpu_gd_train(const int *restrict rows,
const int *restrict cols_Y,
const int *restrict cols_X,
const int *restrict batch,
const int *restrict layers,
const int *restrict nodes,
const double *restrict Y,
const double *restrict X,
double **restrict w,
char funcs[(*layers) + 1][30],
const double *restrict learning_rate,
const int *restrict epochs)
{
// l for layers
// i for rows
// e for epochs
int l, i, for_helper_w, for_helper_batch, e = (*epochs), r_over_b = (*rows)/(*batch);
register int *f;
f = name2int(layers, funcs);
double loss = 0.0;
// For the averaging of deltas in batch/mini-batch
double correction = 1.0;
register double **deltas = malloc(((*layers) + 1) * sizeof(double *));
// The values to be subtracted from weights
register double **grad_w = malloc(((*layers) + 1) * sizeof(double *));
//////////////////////////////////////////////////////////////////
// Gradient of layer's unactivated output
double **help_1 = malloc((*layers) * sizeof(double *));
// Product of next layer's transposed weights and deltas
double **help_2 = malloc((*layers) * sizeof(double *));
// We do not need them at the output layer
//////////////////////////////////////////////////////////////////
if (deltas && grad_w && help_1 && help_2) {
// Allocations
for (l = (*layers) + 1; l--; ) {
if (l == 0) {
for_helper_batch = (*batch) * nodes[l];
for_helper_w = (*cols_X) * nodes[l];
help_1[l] = malloc((*batch) * nodes[l] * sizeof(double));
help_2[l] = malloc((*batch) * nodes[l] * sizeof(double));
memset(help_1[l], 0.0, (*batch) * nodes[l] * sizeof(double));
memset(help_2[l], 0.0, (*batch) * nodes[l] * sizeof(double));
} else if (l == (*layers)) {
for_helper_batch = (*batch) * (*cols_Y);
for_helper_w = nodes[l-1] * (*cols_Y);
} else {
for_helper_batch = (*batch) * nodes[l];
for_helper_w = nodes[l-1] * nodes[l];
help_1[l] = malloc((*batch) * nodes[l] * sizeof(double));
help_2[l] = malloc((*batch) * nodes[l] * sizeof(double));
memset(help_1[l], 0.0, (*batch) * nodes[l] * sizeof(double));
memset(help_2[l], 0.0, (*batch) * nodes[l] * sizeof(double));
}
deltas[l] = malloc(for_helper_batch * sizeof(double));
grad_w[l] = malloc(for_helper_w * sizeof(double));
if (deltas[l] && grad_w[l]) {
memset(deltas[l], 0.0, for_helper_batch * sizeof(double));
memset(grad_w[l], 0.0, for_helper_w * sizeof(double));
} else {
printf(KRED "\nFailed to allocate deltas, gradients or helpers. Aborting...\n" RESET);
free(deltas);
free(grad_w);
free(help_2);
free(help_1);
return;
}
}
} else {
printf(KRED "\nFailed to allocate deltas, gradients or helpers. Aborting...\n" RESET);
return;
}
register double ***Z = cpu_feedforward_cache(rows, cols_Y, cols_X, layers, X, w, nodes, f);
// Big switch in case we have pure batch (do not allocate mini-batches)
switch ((*batch) == (*rows)) {
// True
case 1:
correction = (*learning_rate) * 1.0/(double)(*rows);
// Training
do {
// Find deltas
cpu_gd_delta(deltas, help_1, help_2, rows, cols_Y, layers, Y, Z, w, nodes, f);
// Now we update the weights and biases
// #pragma omp parallel for
for (l = (*layers) + 1; l--; ) {
switch (l == 0) {
case 1:
cpu_threaded_update(X, deltas[l], grad_w[l], w[l], cols_X, &nodes[l], rows, &correction);
continue;
default:
break;
}
switch (l > 0 && l < (*layers)) {
case 1:
cpu_threaded_update(Z[1][l-1], deltas[l], grad_w[l], w[l], &nodes[l-1], &nodes[l], rows, &correction);
continue;
default:
break;
}
switch (l == (*layers)) {
case 1:
cpu_threaded_update(Z[1][l-1], deltas[l], grad_w[l], w[l], &nodes[l-1], cols_Y, rows, &correction);
break;
default:
break;
}
}
// Update Zs with the new wb's
cpu_feedforward_update(rows, cols_Y, cols_X, layers, Z, X, w, nodes, f);
switch (f[(*layers)]) {
// If softmax then cross entropy
case 4:
loss = xentropy(rows, cols_Y, Y, Z[1][(*layers)]);
break;
default:
loss = rmse(rows, cols_Y, Y, Z[1][(*layers)]);
break;
}
if (loss != loss) {
printf(KRED "\nWe got NaN values during training. Aborting...\n" RESET);
for (l = 0; l < (*layers) + 1; l++) {
free(deltas[l]);
free(grad_w[l]);
if (l < (*layers)) {
free(help_2[l]);
free(help_1[l]);
}
}
free(deltas);
free(grad_w);
free(funcs);
free(help_2);
free(help_1);
delete_Z(layers, Z);
return;
}
printf("\nLoss = %.10lf at epoch = %d\n", loss, (*epochs) - e);
} while (--e >= 0);
break;
default:
correction = (*learning_rate) * 1.0/(double)(*batch);
// They need to be allocated once
double **X_batch = malloc(r_over_b * sizeof(double *));
double **Y_batch = malloc(r_over_b * sizeof(double *));
if (X_batch && Y_batch) {
for (i = r_over_b; i--; ) {
X_batch[i] = malloc((*batch) * (*cols_X) * sizeof(double));
Y_batch[i] = malloc((*batch) * (*cols_Y) * sizeof(double));
if (X_batch[i] && Y_batch[i]) {
memset(X_batch[i], 0.0, (*batch) * (*cols_X) * sizeof(double));
memset(Y_batch[i], 0.0, (*batch) * (*cols_Y) * sizeof(double));
} else {
printf(KRED "\nFailed to allocate X or Y batches. Aborting...\n" RESET);
free(X_batch);
free(Y_batch);
return;
}
}
} else {
printf(KRED "\nFailed to allocate X or Y batches. Aborting...\n" RESET);
return;
}
i = r_over_b - 1;
// Fill X_batch, Y_batch
do {
memcpy(X_batch[i], X + i * (*batch) * (*cols_X), (*batch) * (*cols_X) * sizeof(double));
memcpy(Y_batch[i], Y + i * (*batch) * (*cols_Y), (*batch) * (*cols_Y) * sizeof(double));
} while (--i >= 0);
// The perceptrons' outputs
double ***Z_batch = malloc(2 * sizeof(double **));
if (Z_batch) {
// Unactivated
Z_batch[0] = malloc(((*layers) + 1) * sizeof(double *));
// Activated
Z_batch[1] = malloc(((*layers) + 1) * sizeof(double *));
if (Z_batch[0] && Z_batch[1]) {
for (l = (*layers) + 1; l--; ) {
switch (l == (*layers)) {
// Last layer
case 1:
Z_batch[0][l] = malloc((*batch) * (*cols_Y) * sizeof(double));
Z_batch[1][l] = malloc((*batch) * (*cols_Y) * sizeof(double));
if (Z_batch[0][l] && Z_batch[1][l]) {
memset(Z_batch[0][l], 0.0, (*batch) * (*cols_Y) * sizeof(double));
memset(Z_batch[1][l], 0.0, (*batch) * (*cols_Y) * sizeof(double));
} else {
printf(KRED "\nFailed to allocate Z batches. Aborting...\n" RESET);
free(Z_batch[0]);
free(Z_batch[1]);
return;
}
continue;
default:
Z_batch[0][l] = malloc((*batch) * nodes[l] * sizeof(double));
Z_batch[1][l] = malloc((*batch) * nodes[l] * sizeof(double));
if (Z_batch[0][l] && Z_batch[1][l]) {
memset(Z_batch[0][l], 0.0, (*batch) * nodes[l] * sizeof(double));
memset(Z_batch[1][l], 0.0, (*batch) * nodes[l] * sizeof(double));
} else {
printf(KRED "\nFailed to allocate Z batches. Aborting...\n" RESET);
free(Z_batch[0]);
free(Z_batch[1]);
return;
}
continue;
}
}
} else {
printf(KRED "\nFailed to allocate Z batches. Aborting...\n" RESET);
free(Z_batch);
return;
}
} else {
printf(KRED "\nFailed to allocate Z batches. Aborting...\n" RESET);
return;
}
do {
i = r_over_b - 1;
do {
printf(" \t#%d/%d\n", i + 1, r_over_b);
// Fill Z_batch with new Zs
for (l = (*layers); l >= 0; l--) {
switch (l == (*layers)) {
case 1:
memcpy(Z_batch[0][l], Z[0][l] + i * (*batch) * (*cols_Y), (*batch) * (*cols_Y) * sizeof(double));
memcpy(Z_batch[1][l], Z[1][l] + i * (*batch) * (*cols_Y), (*batch) * (*cols_Y) * sizeof(double));
break;
default:
memcpy(Z_batch[0][l], Z[0][l] + i * (*batch) * nodes[l], (*batch) * nodes[l] * sizeof(double));
memcpy(Z_batch[1][l], Z[1][l] + i * (*batch) * nodes[l], (*batch) * nodes[l] * sizeof(double));
break;
}
}
// Fill the deltas
cpu_gd_delta(deltas, help_1, help_2, batch, cols_Y, layers, Y_batch[i], Z_batch, w, nodes, f);
// Now we update the weights and biases
// #pragma omp parallel for
for (l = (*layers) + 1; l--; ) {
switch (l == 0) {
case 1:
cpu_threaded_update(X, deltas[l], grad_w[l], w[l], cols_X, &nodes[l], batch, &correction);
continue;
default:
break;
}
switch (l > 0 && l < (*layers)) {
case 1:
cpu_threaded_update(Z[1][l-1], deltas[l], grad_w[l], w[l], &nodes[l-1], &nodes[l], batch, &correction);
continue;
default:
break;
}
switch (l == (*layers)) {
case 1:
cpu_threaded_update(Z[1][l-1], deltas[l], grad_w[l], w[l], &nodes[l-1], cols_Y, batch, &correction);
break;
default:
break;
}
}
// Do an update of Z's with the new wb's
cpu_feedforward_update(rows, cols_Y, cols_X, layers, Z, X, w, nodes, f);
} while (--i >= 0);
switch (f[(*layers)]) {
// If softmax then cross entropy
case 4:
loss = xentropy(rows, cols_Y, Y, Z[1][(*layers)]);
break;
default:
loss = rmse(rows, cols_Y, Y, Z[1][(*layers)]);
break;
}
if (loss != loss) {
printf(KRED "\nWe got NaN values during training. Aborting...\n" RESET);
// Free before quitting
for (l = 0; l < (*layers) + 1; l++) {
free(Z_batch[0][l]);
free(Z_batch[1][l]);
free(deltas[l]);
free(grad_w[l]);
if (l < (*layers)) {
free(help_2[l]);
free(help_1[l]);
}
}
free(Z_batch[0]);
free(Z_batch[1]);
free(Z_batch);
free(help_2);
free(help_1);
free(deltas);
free(grad_w);
for (i = 0; i < r_over_b; i++) {
free(X_batch[i]);
free(Y_batch[i]);
}
free(Y_batch);
free(X_batch);
free(funcs);
delete_Z(layers, Z);
return;
}
printf("\nLoss = %.10lf at epoch = %d\n", loss, (*epochs) - e);
} while (--e >= 0);
// End of batching
for (l = 0; l < (*layers) + 1; l++) {
free(Z_batch[0][l]);
free(Z_batch[1][l]);
}
free(Z_batch[0]);
free(Z_batch[1]);
free(Z_batch);
for (i = 0; i < r_over_b; i++) {
free(X_batch[i]);
free(Y_batch[i]);
}
free(Y_batch);
free(X_batch);
break;
}
// Save weights and free memory
save_w(w, layers, nodes, cols_Y, cols_X);
// Free the rest
for (l = 0; l < (*layers) + 1; l++) {
free(deltas[l]);
free(grad_w[l]);
if (l < (*layers)) {
free(help_2[l]);
free(help_1[l]);
}
}
free(deltas);
free(grad_w);
free(f);
free(help_2);
free(help_1);
delete_Z(layers, Z);
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// End of PUBLIC API ///////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Definitions of forward declarated functions
static inline void swap(double *restrict a, double *restrict b)
{
int temp = *a;
*a = *b;
*b = temp;
}
static inline double mean(const int *restrict rows, const double *restrict col)
{
double sum = 0.0;
int i = (*rows) - 1;
do {
sum += col[i];
} while (--i >= 0);
sum /= (double)(*rows);
return sum;
}
static inline double stdev(const int *restrict rows, const double *restrict col, const double *restrict mean)
{
double sumsq = 0.0, subtr;
int i = (*rows) - 1;
do {
subtr = col[i] - (*mean);
sumsq += subtr * subtr;
} while (--i >= 0);
return sqrt(sumsq/(double)((*rows) - 1));
}
static inline int *name2int(const int *restrict layers, char funcs[(*layers) + 1][30])
{
int *names2ints = malloc(((*layers) + 1) * sizeof(int));
int l = (*layers);
do {
switch (strcmp(funcs[l], "relu")) {
case 0:
names2ints[l] = 0;
continue;
default:
break;
}
switch (strcmp(funcs[l], "logistic")) {
case 0:
names2ints[l] = 1;
continue;
default:
break;
}
switch (strcmp(funcs[l], "linear")) {
case 0:
names2ints[l] = 2;
continue;
default:
break;
}
switch (strcmp(funcs[l], "tanh")) {
case 0:
names2ints[l] = 3;
continue;
default:
break;
}
switch (strcmp(funcs[l], "softmax")) {
case 0:
names2ints[l] = 4;
continue;
default:
break;
}
// Leaky relu
switch (strcmp(funcs[l], "lrelu")) {
case 0:
names2ints[l] = 5;
continue;
default:
break;
}
switch (strcmp(funcs[l], "softplus")) {
case 0:
names2ints[l] = 6;
continue;
default:
break;
}
switch (strcmp(funcs[l], "softsign")) {
case 0:
names2ints[l] = 7;
continue;
default:
break;
}
switch (strcmp(funcs[l], "arctan")) {
case 0:
names2ints[l] = 8;
continue;
default:
break;
}
//Inverse square root with a = 1
switch (strcmp(funcs[l], "isru")) {
case 0:
names2ints[l] = 9;
continue;
default:
break;
}
//Inverse sqrt linear unit \w a=1
switch (strcmp(funcs[l], "isrlu")) {
case 0:
names2ints[l] = 10;
continue;
default:
break;
}
switch (strcmp(funcs[l], "bent")) {
case 0:
names2ints[l] = 11;
continue;
default:
break;
}
switch (strcmp(funcs[l], "sinus")) {
case 0:
names2ints[l] = 12;
continue;
default:
break;
}
switch (strcmp(funcs[l], "sinusc")) {
case 0:
names2ints[l] = 13;
continue;
default:
// Gaussian if nothing else
names2ints[l] = 14;
break;
}
} while (--l >= 0);
return names2ints;
}
static inline void activate(double *restrict Y, const double *restrict X,
const int *restrict r, const int *restrict c,
const int *restrict f)
{
int i = (*r) * (*c) - 1;
switch ((*f)) {
// Relu
case 0:
do {
switch (X[i] < 0.0) {
case 1:
Y[i] = 0.0;
continue;
default:
Y[i] = X[i];
continue;
}
} while (--i >= 0);
return;
// Logistic
case 1:
do {
Y[i] = 1/(1 + exp(-X[i]));
} while (--i >= 0);
return;
// Linear
case 2:
memcpy(Y, X, (i + 1) * sizeof(double));
return;
// Tanh
case 3:
do {
Y[i] = tanh(X[i]);
} while (--i >= 0);
return;
// Softmax
case 4: {
double *e = malloc((*c) * sizeof(double));
double e_X;
int row = (*r) - 1, col;
do {
e_X = 0.0;
col = (*c) - 1;
do {
e[col] = exp(X[row * (*c) + col]);
e_X += e[col];
} while (--col >= 0);
col = (*c) - 1;
do {
Y[row * (*c) + col] = e[col]/e_X;
} while (--col >= 0);
} while (--row >= 0);
free(e);
return;
}
// Lrelu
case 5:
do {
switch (X[i] < 0.0) {
case 1:
Y[i] = 0.01 * X[i];
continue;
default:
Y[i] = X[i];
continue;
}
} while (--i >= 0);
return;
// Softplus
case 6:
do {
Y[i] = log(1 + exp(X[i]));
} while (--i >= 0);
return;
// Softsign
case 7:
do {
Y[i] = X[i]/(1 + fabs(X[i]));
} while (--i >= 0);
return;
// Arctan
case 8:
do {
Y[i] = atan(X[i]);
} while (--i >= 0);
return;
// Isru
case 9:
do {
Y[i] = X[i]/sqrt(1 + X[i] * X[i]);
} while (--i >= 0);
return;
// Isrlu
case 10:
do {
switch (X[i] < 0.0) {
case 1:
Y[i] = X[i]/sqrt(1 + X[i] * X[i]);
continue;
default:
Y[i] = X[i];
continue;
}
} while (--i >= 0);
return;
// Bent
case 11:
do {
Y[i] = (sqrt(X[i] * X[i] + 1.0) - 1.0)/2.0 + X[i];
} while (--i >= 0);
return;
// Sinus
case 12:
do {
Y[i] = sin(X[i]);
} while (--i >= 0);
return;
// Sinusc
case 13:
do {
switch (X[i] == 0.0) {
case 1:
Y[i] = 1.0;
continue;
default:
Y[i] = sin(X[i])/X[i];
continue;
}
} while (--i >= 0);
return;
// Gauss
default:
do {
Y[i] = exp(-(X[i] * X[i]));
} while (--i >= 0);
return;
}
}
static inline void gradient(double *restrict Y, const double *restrict X,
const int *restrict r, const int *restrict c,
const int *restrict f)
{
int i = (*r) * (*c) - 1;
switch ((*f)) {
// Relu
case 0:
do {
switch (X[i] < 0.0) {
case 1:
Y[i] = 0.0;
continue;
default:
Y[i] = 1.0;
continue;
}
} while (--i >= 0);
return;
// Logistic
case 1: {
double y;
do {
y = 1/(1 + exp(-X[i]));
Y[i] = y * (1 - y);
} while (--i >= 0);
return;
}
// Linear
case 2:
memset(Y, 1.0, (i + 1) * sizeof(double));
return;
// Tanh
case 3: {
double e_X, e_mX, y;
do {
e_X = exp(X[i]);
e_mX = exp(-X[i]);
y = (e_X - e_mX)/(e_X + e_mX);
Y[i] = 1 - y * y;
} while (--i >= 0);
return;
}
// Softmax
case 4: {
double *e = malloc((*c) * sizeof(double));
int row = (*r) - 1, col;
double e_X, e_mX;
do {
e_X = 0.0;
col = (*c) - 1;
do {
e[col] = exp(X[row * (*c) + col]);
e_X += e[col];
} while (--col >= 0);
col = (*c) - 1;
do {
e_mX = e[col]/e_X;
switch (row == col) {
case 1:
Y[row * (*c) + col] = e_mX * (1 - e_mX);
break;
default:
Y[row * (*c) + col] = - e_mX * e_mX;
break;
}
} while (--col >= 0);
} while (--row >= 0);
free(e);
return;
}
// Lrelu
case 5:
do {
switch (X[i] < 0.0) {
case 1:
Y[i] = 0.01;
continue;
default:
Y[i] = 1.0;
continue;
}
} while (--i >= 0);
return;
// Softplus
case 6:
do {
Y[i] = 1/(1 + exp(-X[i]));
} while (--i >= 0);
return;
// Softsign
case 7: {
double y;
do {
y = 1 + fabs(X[i]);
Y[i] = 1/(y * y);
} while (--i >= 0);
return;
}
// Arctan
case 8:
do {
Y[i] = 1/(X[i] * X[i] + 1);
} while (--i >= 0);
return;
// Isru
case 9: {
double sq, y;
do {
sq = sqrt(1 + X[i] * X[i]);
y = X[i]/sq;
Y[i] = y * y * y;
} while (--i >= 0);
return;
}
// Isrlu
case 10: {
double sq, y;
do {
switch (X[i] < 0.0) {
case 1:
sq = sqrt(1 + X[i] * X[i]);
y = X[i]/sq;
Y[i] = y * y * y;
continue;
default:
Y[i] = 1.0;
continue;
}
} while (--i >= 0);
return;
}
// Bent
case 11: {
double y, add;
do {
add = X[i] + 1;
y = sqrt(add * add);
Y[i] = X[i]/(2 * y) + 1;
} while (--i >= 0);
return;
}
// Sinus
case 12:
do {
Y[i] = cos(X[i]);
} while (--i >= 0);
return;
// Sinusc
case 13:
do {
switch (X[i] == 0.0) {
case 1:
Y[i] = 0.0;
continue;
default:
Y[i] = cos(X[i])/X[i] - sin(X[i])/(X[i] * X[i]);
continue;
}
} while (--i >= 0);
return;
// Gauss
default:
do {
Y[i] = -2.0 * X[i] * exp(-(X[i] * X[i]));
} while (--i >= 0);
return;
}
}
static inline double rmse(const int *restrict r, const int *restrict c,
const double *restrict Y, const double *restrict Z)
{
int i = (*r) * (*c) - 1;
double loss = 0.0;
double d;
do {
d = Z[i] - Y[i];
loss += (d * d)/(double)(*c);
} while (--i >= 0);
loss = loss/(double)(*r);
return sqrt(loss);
}
static inline double xentropy(const int *restrict r, const int *restrict c,
const double *restrict Y, const double *restrict Z)
{
int i = (*r) * (*c) - 1;
double loss = 0.0;
do {
loss += Y[i] * log(Z[i])/(double)(*c);
} while (--i >= 0);
return -loss/(double)(*r);
}
static inline void save_w(double **restrict w, const int *restrict layers, const int *restrict nodes,
const int *restrict cols_Y, const int *restrict cols_X)
{
int l;
char path[1024];
char w_path[1036];
getcwd(path, sizeof(path));
strcpy(w_path, path);
struct stat st = {0};
if (stat(strcat(w_path, "/weights"), &st) == -1) {
mkdir(w_path, 0700);
printf("\nCreated ./wb/weights directory\n");
}
FILE *ptr_fp;
for (l = 0; l < (*layers) + 1; l++) {
// This needs to change every time
char w_path_filename[1050];
char number[100];
char filename[15] = "layer_";
sprintf(number, "%d", l);
strcat(filename, number);
strcat(filename, ".bin");
strcpy(w_path_filename, w_path);
strcat(w_path_filename, "/");
strcat(w_path_filename, filename);
if((ptr_fp = fopen(w_path_filename, "wb")) == NULL) {
printf("Unable to create file!\n");
exit(1);
}
if (l == 0) {
if(fwrite(w[l], (*cols_X) * nodes[l] * sizeof(double), 1, ptr_fp) != 1) {
printf("Write error!\n");
exit(1);
}
fclose(ptr_fp);
} else if (l == (*layers)) {
if(fwrite(w[l], nodes[l-1] * (*cols_Y) * sizeof(double), 1, ptr_fp) != 1) {
printf("Write error!\n");
exit(1);
}
fclose(ptr_fp);
} else {
if(fwrite(w[l], nodes[l-1] * nodes[l] * sizeof(double), 1, ptr_fp) != 1) {
printf("Write error!\n");
exit(1);
}
fclose(ptr_fp);
}
}
}
static inline int ***imgpad(int ***restrict images, const int *restrict no_of_images,
const int *restrict img_width,
const int *restrict img_height,
const int *restrict img_channels,
const int *restrict padding, // Zeros around
const int *restrict delete_originals) // 0 = no, 1 = yes (keep only vector in memory)
{
if ((*padding) == 0) {
return images;
} else {
int image, i, j, rgb;
int multiplication = ((*img_width) + 2 * (*padding)) * ((*img_height) + 2 * (*padding));
int ***images_padded = malloc((*no_of_images) * sizeof(int **));
for (image = (*no_of_images); image--; ) {
images_padded[image] = malloc(multiplication * sizeof(int *));
for (i = multiplication; i--; ) {
images_padded[image][i] = malloc((*img_channels) * sizeof(int));
for (rgb = (*img_channels); rgb--; ) {
images_padded[image][i][rgb] = 0;
}
}
for (i = (*img_height); i--; ) {
for (j = (*img_width); j--; ) {
for (rgb = (*img_channels); rgb--; ) {
images_padded[image][(i + (*padding)) * ((*img_width) + 2 * (*padding)) + j + (*padding)][rgb] =
images[image][i * (*img_width) + j][rgb];
}
}
}
}
if ((*delete_originals) == 1) {
multiplication = (*img_width) * (*img_height);
for (i = 0; i < (*no_of_images); i++) {
for (j = 0; j < multiplication; j++) {
free(images[i][j]);
}
free(images[i]);
}
free(images);
}
return images_padded;
}
}
static inline int **im2col(int ***restrict images,const int *restrict no_of_images,
const int *restrict img_width,
const int *restrict img_height,
const int *restrict img_channels,
const int *restrict spatial, // width and height of weights
const int *restrict stride, // (img_width - spatial + 2 * padding)/stride should be int
const int *restrict padding, // Zeros around
const int *restrict delete_originals) // 0 = no, 1 = yes (keep only vectorized in memory)
{
int image, i, i_prime, j, pixel_x, pixel_y, multiplication;
size_t rgb;
// How many boxes horizontally
int locations_width = ((*img_width) - (*spatial) + 2 * (*padding))/(*stride) + 1;
// How many boxes vertically
int locations_height = ((*img_height) - (*spatial) + 2 * (*padding))/(*stride) + 1;
// Rows of vectorized image
int receptive_fields = locations_width * locations_height;
multiplication = (*spatial) * (*spatial) * (*img_channels);
int ***images_w_pad = imgpad(images, no_of_images, img_width, img_height, img_channels, padding, delete_originals);
int **images_as_matrices = malloc((*no_of_images) * sizeof(int *));
for (image = (*no_of_images); image--; ) {
pixel_x = 0;
pixel_y = 0;
i_prime = 0;
// dim(receptive_fields X (spatial * spatial * img_channels))
images_as_matrices[image] = malloc(receptive_fields * multiplication * sizeof(int));
for (i = 0; i < receptive_fields; i++) {
rgb = 0;
if (i % locations_height == 0 && i > 0) {
i_prime += 1;
pixel_x = 0;
} else if (i % locations_height != 0 && i > 0){
pixel_x += (*stride);
} else {
pixel_x = 0;
}
pixel_y = i_prime * (*stride);
for (j = 0; j < multiplication; j++) {
// The channel change happens in every row for every conv box
if (j > 0 && j % ((*spatial) * (*spatial)) == 0) {
rgb += 1;
if (rgb == (*img_channels)) {
rgb = 0;
}
}
// Width and height pixels
if (j % (*spatial) == 0 && j > 0) {
pixel_y += 1;
if (j % ((*spatial) * (*spatial)) == 0) {
pixel_y = i_prime * (*stride);
}
pixel_x -= (*spatial) - 1;
if (i % locations_height == 0 && i > 0) {
pixel_x = 0;
}
} else if (j % (*spatial) != 0 && j > 0) {
pixel_x += 1;
}
/////////////////////////////////////////////////////////////
// The operation
images_as_matrices[image][i * multiplication + j] =
images_w_pad[image][pixel_x * ((*img_width) + 2 * (*padding)) + pixel_y][rgb];
if (j == multiplication - 1) {
pixel_x -= (*spatial) - 1;
}
}
}
}
if ((*padding) != 0) {
multiplication = ((*img_width) + 2 * (*padding)) * ((*img_height) + 2 * (*padding));
for (i = 0; i < (*no_of_images); i++) {
for (j = 0; j < multiplication; j++) {
free(images_w_pad[i][j]);
}
free(images_w_pad[i]);
}
free(images_w_pad);
}
if ((*delete_originals) == 1 && (*padding) == 0) {
for (i = 0; i < (int)(*no_of_images); i++) {
for (j = 0; j < (int)((*img_width) * (*img_height)); j++) {
free(images[i][j]);
}
free(images[i]);
}
free(images);
}
return images_as_matrices;
}
static inline void delete_Z(const int *restrict layers, double ***restrict Z)
{
int l;
for (l = 0; l < (*layers) + 1; l++) {
free(Z[1][l]);
free(Z[0][l]);
}
free(Z[1]);
free(Z[0]);
free(Z);
}
static inline void delete_img_vector(const int *restrict no_of_images, int **restrict images)
{
int image;
for (image = 0; image < (*no_of_images); image++) free(images[image]);
free(images);
}
static inline void cpu_mm_notrans(const double *restrict A, const double *restrict B, double *restrict C,
const int *restrict rows, const int *restrict cols, const int *restrict coms)
{
memset(C, 0.0, (*rows) * (*cols) * sizeof(double));
double A_i;
for (int i = (*rows); i--; ) {
for (int k = (*coms); k--; ) {
A_i = A[i * (*coms) + k];
for (int j = (*cols); j--; ) {
C[i * (*cols) + j] += A_i * B[k * (*cols) + j];
}
}
}
}
static inline void cpu_feedforward_update(const int *restrict r, const int *restrict cY,
const int *restrict cX, const int *restrict layers,
double ***restrict Z,
const double *restrict X, double **restrict w,
const int *restrict n, const int *restrict f)
{
// l is for layers
int l = 0;
do {
switch (l == 0) {
case 1:
cpu_mm_notrans(X, w[l], Z[0][l], r, &n[l], cX);
activate(Z[1][l], Z[0][l], r, &n[l], &f[l]);
++l;
continue;
default:
break;
}
switch (l > 0 && l < (*layers)) {
case 1:
cpu_mm_notrans(Z[1][l-1], w[l], Z[0][l], r, &n[l], &n[l-1]);
activate(Z[1][l], Z[0][l], r, &n[l], &f[l]);
++l;
continue;
default:
break;
}
switch (l == (*layers)) {
case 1:
cpu_mm_notrans(Z[1][l-1], w[l], Z[0][l], r, cY, &n[l-1]);
activate(Z[1][l], Z[0][l], r, cY, &f[l]);
return;
}
} while (l <= (*layers));
}
static inline double ***cpu_feedforward_cache(const int *restrict r, const int *restrict cY,
const int *restrict cX, const int *restrict layers,
const double *restrict X, double **restrict w,
const int *restrict n, const int *restrict f)
{
// l is for layers
// i is for each row * column of X, Y
int l = (*layers);
int i = (*r) * (*cY);
// feeds at every layer
double ***Z = malloc(2 * sizeof(double **));
if (Z) {
Z[0] = malloc(((*layers) + 1) * sizeof(double *));
Z[1] = malloc(((*layers) + 1) * sizeof(double *));
if (Z[0] && Z[1]) {
Z[0][l] = malloc(i * sizeof(double));
Z[1][l] = malloc(i * sizeof(double));
if (Z[0][l] && Z[1][l]) {
for (l = 0; l < (*layers); l++) {
i = (*r) * n[l];
Z[0][l] = malloc(i * sizeof(double));
Z[1][l] = malloc(i * sizeof(double));
if (Z[0][l] && Z[1][l]) {
memset(Z[0][l], 0.0, i * sizeof(double));
memset(Z[1][l], 0.0, i * sizeof(double));
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z[0]);
free(Z[1]);
free(Z);
abort();
}
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z[0]);
free(Z[1]);
free(Z);
abort();
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
free(Z);
abort();
}
} else {
printf(KRED "\nCould not allocate Zs. Aborting...\n" RESET);
abort();
}
// Directly manipulates Z
cpu_feedforward_update(r, cY, cX, layers, Z, X, w, n, f);
return Z;
}
static inline void cpu_mm_notrans_trans(const double *restrict A, const double *restrict B, double *restrict C,
const int *restrict rows, const int *restrict cols, const int *restrict coms)
{
memset(C, 0.0, (*rows) * (*cols) * sizeof(double));
for (int i = (*rows); i--; ) {
for (int j = (*cols); j--; ) {
for (int k = (*coms); k--; ) {
C[i * (*cols) + j] += A[i * (*coms) + k] * B[j * (*coms) + k];
}
}
}
}
static inline void cpu_gd_delta(double **restrict d, double **restrict h1, double **restrict h2,
const int *restrict r, const int *restrict c, const int *restrict layers,
const double *restrict Y, double ***restrict Z, double **restrict w,
const int *restrict n, const int *restrict f)
{
int l, i = (*r) * (*c) - 1;
// Last layer
switch (f[(*layers)]) {
// Linear case
case 2:
do {
d[(*layers)][i] = Z[1][(*layers)][i] - Y[i];
} while (--i >= 0);
break;
// Softmax Crossentropy
case 4:
do {
d[(*layers)][i] = Z[1][(*layers)][i] - Y[i];
} while (--i >= 0);
break;
default:
gradient(d[(*layers)], Z[0][(*layers)], r, c, &f[(*layers)]);
do {
d[(*layers)][i] *= Z[1][(*layers)][i] - Y[i];
} while (--i >= 0);
break;
}
// Before last
l = (*layers) - 1;
i = (*r) * n[l] - 1;
gradient(h1[l], Z[0][l], r, &n[l], &f[l]);
cpu_mm_notrans_trans(d[l+1], w[l+1], h2[l], r, &n[l], c);
// Hadamard product
do {
d[l][i] = h1[l][i] * h2[l][i];
} while (--i >= 0);
switch ((*layers) == 1) {
case 1:
return;
default:
// All other layers
l = (*layers) - 2;
do {
i = (*r) * n[l] - 1;
gradient(h1[l], Z[0][l], r, &n[l], &f[l]);
cpu_mm_notrans_trans(d[l+1], w[l+1], h2[l], r, &n[l], &n[l+1]);
// Hadamard product
do {
d[l][i] = h1[l][i] * h2[l][i];
} while (--i >= 0);
} while (--l >= 0);
return;
}
}
// returns new wb
static inline void cpu_threaded_update(const double *restrict X, const double *restrict d,
double *restrict gw,
double *restrict w,
const int *restrict m,
const int *restrict n,
const int *restrict k,
const double *restrict c)
{
int i = (*m) * (*n) - 1;
memset(gw, 0.0, (*m) * (*n) * sizeof(double));
double A_i;
for (int com = (*k); com--; ) {
for (int row = (*m); row--; ) {
A_i = X[com * (*m) + row];
for (int col = (*n); col--; ) {
gw[row * (*n) + col] += A_i * d[com * (*n) + col];
}
}
}
do {
w[i] -= (*c) * gw[i];
} while (--i >= 0);
}
#endif // libartificial_h__
|
generated-funcs-regex.c | // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -fopenmp %s -emit-llvm -o - | FileCheck %s
void __test_offloading_42_abcdef_bar_l123();
void use(int);
void foo(int a)
{
#pragma omp target
use(a);
__test_offloading_42_abcdef_bar_l123();
}
|
GB_binop__plus_int16.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB_AaddB__plus_int16
// A.*B function (eWiseMult): GB_AemultB__plus_int16
// A*D function (colscale): GB_AxD__plus_int16
// D*A function (rowscale): GB_DxB__plus_int16
// C+=B function (dense accum): GB_Cdense_accumB__plus_int16
// C+=b function (dense accum): GB_Cdense_accumb__plus_int16
// C+=A+B function (dense ewise3): GB_Cdense_ewise3_accum__plus_int16
// C=A+B function (dense ewise3): GB_Cdense_ewise3_noaccum__plus_int16
// C=scalar+B GB_bind1st__plus_int16
// C=scalar+B' GB_bind1st_tran__plus_int16
// C=A+scalar GB_bind2nd__plus_int16
// C=A'+scalar GB_bind2nd_tran__plus_int16
// C type: int16_t
// A type: int16_t
// B,b type: int16_t
// BinaryOp: cij = (aij + bij)
#define GB_ATYPE \
int16_t
#define GB_BTYPE \
int16_t
#define GB_CTYPE \
int16_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int16_t aij = Ax [pA]
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB) \
int16_t bij = Bx [pB]
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int16_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA) \
cij = Ax [pA]
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB) \
cij = Bx [pB]
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z, x, y, i, j) \
z = (x + y) ;
// op is second
#define GB_OP_IS_SECOND \
0
// op is plus_fp32 or plus_fp64
#define GB_OP_IS_PLUS_REAL \
0
// op is minus_fp32 or minus_fp64
#define GB_OP_IS_MINUS_REAL \
0
// GB_cblas_*axpy gateway routine, if it exists for this operator and type:
#define GB_CBLAS_AXPY \
(none)
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_PLUS || GxB_NO_INT16 || GxB_NO_PLUS_INT16)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB_Cdense_ewise3_accum__plus_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_ewise3_noaccum__plus_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumB__plus_int16
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumb__plus_int16
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int16_t
int16_t bwork = (*((int16_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_AxD__plus_int16
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_colscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_DxB__plus_int16
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_rowscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_ek_slice_free (&pstart_Mslice, &kfirst_Mslice, &klast_Mslice) ; \
GB_ek_slice_free (&pstart_Aslice, &kfirst_Aslice, &klast_Aslice) ; \
GB_ek_slice_free (&pstart_Bslice, &kfirst_Bslice, &klast_Bslice) ; \
}
GrB_Info GB_AaddB__plus_int16
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *pstart_Mslice = NULL, *kfirst_Mslice = NULL, *klast_Mslice = NULL ;
int64_t *pstart_Aslice = NULL, *kfirst_Aslice = NULL, *klast_Aslice = NULL ;
int64_t *pstart_Bslice = NULL, *kfirst_Bslice = NULL, *klast_Bslice = NULL ;
#include "GB_add_template.c"
GB_FREE_ALL ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB_AemultB__plus_int16
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *pstart_Mslice = NULL, *kfirst_Mslice = NULL, *klast_Mslice = NULL ;
int64_t *pstart_Aslice = NULL, *kfirst_Aslice = NULL, *klast_Aslice = NULL ;
int64_t *pstart_Bslice = NULL, *kfirst_Bslice = NULL, *klast_Bslice = NULL ;
#include "GB_emult_template.c"
GB_FREE_ALL ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB_bind1st__plus_int16
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *GB_RESTRICT Bb,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *Cx = (int16_t *) Cx_output ;
int16_t x = (*((int16_t *) x_input)) ;
int16_t *Bx = (int16_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Bb, p)) continue ;
int16_t bij = Bx [p] ;
Cx [p] = (x + bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB_bind2nd__plus_int16
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *GB_RESTRICT Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int16_t *Cx = (int16_t *) Cx_output ;
int16_t *Ax = (int16_t *) Ax_input ;
int16_t y = (*((int16_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int16_t aij = Ax [p] ;
Cx [p] = (aij + y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (x + aij) ; \
}
GrB_Info GB_bind1st_tran__plus_int16
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *GB_RESTRICT *Workspaces,
const int64_t *GB_RESTRICT A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t x = (*((const int16_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (aij + y) ; \
}
GrB_Info GB_bind2nd_tran__plus_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *GB_RESTRICT *Workspaces,
const int64_t *GB_RESTRICT A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t y = (*((const int16_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
GB_binop__lor_int8.c |
//------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCUDA_DEV
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__lor_int8)
// A.*B function (eWiseMult): GB (_AemultB_08__lor_int8)
// A.*B function (eWiseMult): GB (_AemultB_02__lor_int8)
// A.*B function (eWiseMult): GB (_AemultB_04__lor_int8)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__lor_int8)
// A*D function (colscale): GB (_AxD__lor_int8)
// D*A function (rowscale): GB (_DxB__lor_int8)
// C+=B function (dense accum): GB (_Cdense_accumB__lor_int8)
// C+=b function (dense accum): GB (_Cdense_accumb__lor_int8)
// C+=A+B function (dense ewise3): GB ((none))
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__lor_int8)
// C=scalar+B GB (_bind1st__lor_int8)
// C=scalar+B' GB (_bind1st_tran__lor_int8)
// C=A+scalar GB (_bind2nd__lor_int8)
// C=A'+scalar GB (_bind2nd_tran__lor_int8)
// C type: int8_t
// A type: int8_t
// A pattern? 0
// B type: int8_t
// B pattern? 0
// BinaryOp: cij = ((aij != 0) || (bij != 0))
#define GB_ATYPE \
int8_t
#define GB_BTYPE \
int8_t
#define GB_CTYPE \
int8_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
int8_t aij = GBX (Ax, pA, A_iso)
// true if values of A are not used
#define GB_A_IS_PATTERN \
0 \
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
int8_t bij = GBX (Bx, pB, B_iso)
// true if values of B are not used
#define GB_B_IS_PATTERN \
0 \
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int8_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = ((x != 0) || (y != 0)) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_LOR || GxB_NO_INT8 || GxB_NO_LOR_INT8)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
void GB (_Cdense_ewise3_noaccum__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_noaccum_template.c"
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__lor_int8)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int8_t
int8_t bwork = (*((int8_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix D,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *restrict Cx = (int8_t *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix D,
const GrB_Matrix B,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *restrict Cx = (int8_t *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__lor_int8)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool is_eWiseUnion,
const GB_void *alpha_scalar_in,
const GB_void *beta_scalar_in,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
int8_t alpha_scalar ;
int8_t beta_scalar ;
if (is_eWiseUnion)
{
alpha_scalar = (*((int8_t *) alpha_scalar_in)) ;
beta_scalar = (*((int8_t *) beta_scalar_in )) ;
}
#include "GB_add_template.c"
GB_FREE_WORKSPACE ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__lor_int8)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__lor_int8)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__lor_int8)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *Cx = (int8_t *) Cx_output ;
int8_t x = (*((int8_t *) x_input)) ;
int8_t *Bx = (int8_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
int8_t bij = GBX (Bx, p, false) ;
Cx [p] = ((x != 0) || (bij != 0)) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__lor_int8)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int8_t *Cx = (int8_t *) Cx_output ;
int8_t *Ax = (int8_t *) Ax_input ;
int8_t y = (*((int8_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int8_t aij = GBX (Ax, p, false) ;
Cx [p] = ((aij != 0) || (y != 0)) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int8_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = ((x != 0) || (aij != 0)) ; \
}
GrB_Info GB (_bind1st_tran__lor_int8)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int8_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t x = (*((const int8_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int8_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int8_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = ((aij != 0) || (y != 0)) ; \
}
GrB_Info GB (_bind2nd_tran__lor_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t y = (*((const int8_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
2.c | #include <math.h>
#include <omp.h>
#include <stdio.h>
// task size 10 = 12.8 ms ± 0.8 ms
// task size 1000 = 8.3 ms ± 0.8 ms
// task size 2000 = 7.9 ms ± 1.2 ms
// task size 4000 = 8.5 ms ± 0.9 ms
// task size 5000 = 9.1 ms ± 0.5 ms
//
// static = 9.0 ms ± 1.7 ms
int main() {
int result[1] = {0};
#pragma omp parallel for schedule(dynamic, 2000)
for (int i = 0; i < 1000000; i++) {
result[0] += sin(i);
}
}
|
joseph3d_back_tof_lm_2.c | /**
* @file joseph3d_back_tof_lm_2.c
*/
#include<stdio.h>
#include<stdlib.h>
#include<stdint.h>
#include<math.h>
#include<omp.h>
#include "tof_utils.h"
#include "ray_cube_intersection.h"
/** @brief 3D listmode tof joseph back projector
*
* All threads back project in one image using openmp's atomic add.
*
* @param xstart array of shape [3*nlors] with the coordinates of the start points of the LORs.
* The start coordinates of the n-th LOR are at xstart[n*3 + i] with i = 0,1,2
* @param xend array of shape [3*nlors] with the coordinates of the end points of the LORs.
* The start coordinates of the n-th LOR are at xstart[n*3 + i] with i = 0,1,2
* @param img array of shape [n0*n1*n2] containing the 3D image used for back projection (output).
* The pixel [i,j,k] ist stored at [n1*n2*i + n2*j + k].
* @param img_origin array [x0_0,x0_1,x0_2] of coordinates of the center of the [0,0,0] voxel
* @param voxsize array [vs0, vs1, vs2] of the voxel sizes
* @param p array of length nlors with the values to be back projected
* @param nlors number of geometrical LORs
* @param img_dim array with dimensions of image [n0,n1,n2]
* @param tofbin_width width of the TOF bins in spatial units (units of xstart and xend)
* @param sigma_tof array of length nlors with the TOF resolution (sigma) for each LOR in
* spatial units (units of xstart and xend)
* @param tofcenter_offset array of length nlors with the offset of the central TOF bin from the
* midpoint of each LOR in spatial units (units of xstart and xend)
* @param n_sigmas number of sigmas to consider for calculation of TOF kernel
* @param tof_bin array containing the TOF bin of each event
*/
void joseph3d_back_tof_lm_2(const float *xstart,
const float *xend,
float *img,
const float *img_origin,
const float *voxsize,
const float *p,
long long nlors,
const int *img_dim,
float tofbin_width,
const float *sigma_tof,
const float *tofcenter_offset,
float n_sigmas,
const short *tof_bin)
{
long long i;
int n0 = img_dim[0];
int n1 = img_dim[1];
int n2 = img_dim[2];
# pragma omp parallel for schedule(static)
for(i = 0; i < nlors; i++)
{
float d0, d1, d2, d0_sq, d1_sq, d2_sq;
float cs0, cs1, cs2, cf;
float lsq, cos0_sq, cos1_sq, cos2_sq;
unsigned short direction;
int i0, i1, i2;
int i0_floor, i1_floor, i2_floor;
int i0_ceil, i1_ceil, i2_ceil;
float x_pr0, x_pr1, x_pr2;
float tmp_0, tmp_1, tmp_2;
float u0, u1, u2, d_norm;
float x_m0, x_m1, x_m2;
float x_v0, x_v1, x_v2;
short it = tof_bin[i];
float dtof, tw;
float sig_tof = sigma_tof[i];
float tc_offset = tofcenter_offset[i];
float xstart0 = xstart[i*3 + 0];
float xstart1 = xstart[i*3 + 1];
float xstart2 = xstart[i*3 + 2];
float xend0 = xend[i*3 + 0];
float xend1 = xend[i*3 + 1];
float xend2 = xend[i*3 + 2];
float voxsize0 = voxsize[0];
float voxsize1 = voxsize[1];
float voxsize2 = voxsize[2];
float img_origin0 = img_origin[0];
float img_origin1 = img_origin[1];
float img_origin2 = img_origin[2];
unsigned char intersec;
float t1, t2;
float istart_f, iend_f, tmp;
int istart, iend;
float istart_tof_f, iend_tof_f;
int istart_tof, iend_tof;
// test whether the ray between the two detectors is most parallel
// with the 0, 1, or 2 axis
d0 = xend0 - xstart0;
d1 = xend1 - xstart1;
d2 = xend2 - xstart2;
//-----------
//--- test whether ray and cube intersect
intersec = ray_cube_intersection(xstart0, xstart1, xstart2,
img_origin0 - 1*voxsize0, img_origin1 - 1*voxsize1, img_origin2 - 1*voxsize2,
img_origin0 + n0*voxsize0, img_origin1 + n1*voxsize1, img_origin2 + n2*voxsize2,
d0, d1, d2, &t1, &t2);
if (intersec == 1)
{
d0_sq = d0*d0;
d1_sq = d1*d1;
d2_sq = d2*d2;
lsq = d0_sq + d1_sq + d2_sq;
cos0_sq = d0_sq / lsq;
cos1_sq = d1_sq / lsq;
cos2_sq = d2_sq / lsq;
cs0 = sqrtf(cos0_sq);
cs1 = sqrtf(cos1_sq);
cs2 = sqrtf(cos2_sq);
direction = 0;
if ((cos1_sq >= cos0_sq) && (cos1_sq >= cos2_sq))
{
direction = 1;
}
if ((cos2_sq >= cos0_sq) && (cos2_sq >= cos1_sq))
{
direction = 2;
}
//---------------------------------------------------------
//--- calculate TOF related quantities
// unit vector (u0,u1,u2) that points from xstart to end
d_norm = sqrtf(lsq);
u0 = d0 / d_norm;
u1 = d1 / d_norm;
u2 = d2 / d_norm;
// calculate mid point of LOR
x_m0 = 0.5f*(xstart0 + xend0);
x_m1 = 0.5f*(xstart1 + xend1);
x_m2 = 0.5f*(xstart2 + xend2);
//---------------------------------------------------------
if(direction == 0)
{
// case where ray is most parallel to the 0 axis
// we step through the volume along the 0 direction
// factor for correctiong voxel size and |cos(theta)|
cf = voxsize0/cs0;
//--- check where ray enters / leaves cube
istart_f = (xstart0 + t1*d0 - img_origin0) / voxsize0;
iend_f = (xstart0 + t2*d0 - img_origin0) / voxsize0;
if (istart_f > iend_f){
tmp = iend_f;
iend_f = istart_f;
istart_f = tmp;
}
istart = (int)floor(istart_f);
iend = (int)ceil(iend_f);
//-- check where we should start and stop according to the TOF kernel
//-- the tof weights outside +- 3 sigma will be close to 0 so we can
//-- ignore them
istart_tof_f = (x_m0 + (it*tofbin_width - n_sigmas*sig_tof)*u0 - img_origin0) / voxsize0;
iend_tof_f = (x_m0 + (it*tofbin_width + n_sigmas*sig_tof)*u0 - img_origin0) / voxsize0;
if (istart_tof_f > iend_tof_f){
tmp = iend_tof_f;
iend_tof_f = istart_tof_f;
istart_tof_f = tmp;
}
istart_tof = (int)floor(istart_tof_f);
iend_tof = (int)ceil(iend_tof_f);
if(istart_tof > istart){istart = istart_tof;}
if(iend_tof < iend){iend = iend_tof;}
//-----------
if (istart < 0){istart = 0;}
if (iend >= n0){iend = n0;}
//---
for(i0 = istart; i0 < iend; i0++)
{
// get the indices where the ray intersects the image plane
x_pr1 = xstart1 + (img_origin0 + i0*voxsize0 - xstart0)*d1 / d0;
x_pr2 = xstart2 + (img_origin0 + i0*voxsize0 - xstart0)*d2 / d0;
i1_floor = (int)floor((x_pr1 - img_origin1)/voxsize1);
i1_ceil = i1_floor + 1;
i2_floor = (int)floor((x_pr2 - img_origin2)/voxsize2);
i2_ceil = i2_floor + 1;
// calculate the distances to the floor normalized to [0,1]
// for the bilinear interpolation
tmp_1 = (x_pr1 - (i1_floor*voxsize1 + img_origin1)) / voxsize1;
tmp_2 = (x_pr2 - (i2_floor*voxsize2 + img_origin2)) / voxsize2;
//--------- TOF related quantities
// calculate the voxel center needed for TOF weights
x_v0 = img_origin0 + i0*voxsize0;
x_v1 = x_pr1;
x_v2 = x_pr2;
if(p[i] != 0){
// calculate distance of voxel to tof bin center
dtof = sqrtf(powf((x_m0 + (it*tofbin_width + tc_offset)*u0 - x_v0), 2) +
powf((x_m1 + (it*tofbin_width + tc_offset)*u1 - x_v1), 2) +
powf((x_m2 + (it*tofbin_width + tc_offset)*u2 - x_v2), 2));
//calculate the TOF weight
tw = 0.5f*(erff_as((dtof + 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)) -
erff_as((dtof - 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)));
if ((i1_floor >= 0) && (i1_floor < n1) && (i2_floor >= 0) && (i2_floor < n2))
{
#pragma omp atomic
img[n1*n2*i0 + n2*i1_floor + i2_floor] +=
(tw * p[i] * (1 - tmp_1) * (1 - tmp_2) * cf);
}
if ((i1_ceil >= 0) && (i1_ceil < n1) && (i2_floor >= 0) && (i2_floor < n2))
{
#pragma omp atomic
img[n1*n2*i0 + n2*i1_ceil + i2_floor] +=
(tw * p[i] * tmp_1 * (1 - tmp_2) * cf);
}
if ((i1_floor >= 0) && (i1_floor < n1) && (i2_ceil >= 0) && (i2_ceil < n2))
{
#pragma omp atomic
img[n1*n2*i0 + n2*i1_floor + i2_ceil] +=
(tw * p[i] * (1 - tmp_1) * tmp_2*cf);
}
if ((i1_ceil >= 0) && (i1_ceil < n1) && (i2_ceil >= 0) && (i2_ceil < n2))
{
#pragma omp atomic
img[n1*n2*i0 + n2*i1_ceil + i2_ceil] +=
(tw * p[i] * tmp_1 * tmp_2 * cf);
}
}
}
}
// ---------------------------------------------------------------------------------
if(direction == 1)
{
// case where ray is most parallel to the 1 axis
// we step through the volume along the 1 direction
// factor for correctiong voxel size and |cos(theta)|
cf = voxsize1/cs1;
//--- check where ray enters / leaves cube
istart_f = (xstart1 + t1*d1 - img_origin1) / voxsize1;
iend_f = (xstart1 + t2*d1 - img_origin1) / voxsize1;
if (istart_f > iend_f){
tmp = iend_f;
iend_f = istart_f;
istart_f = tmp;
}
istart = (int)floor(istart_f);
iend = (int)ceil(iend_f);
//-- check where we should start and stop according to the TOF kernel
//-- the tof weights outside +- 3 sigma will be close to 0 so we can
//-- ignore them
istart_tof_f = (x_m1 + (it*tofbin_width - n_sigmas*sig_tof)*u1 - img_origin1) / voxsize1;
iend_tof_f = (x_m1 + (it*tofbin_width + n_sigmas*sig_tof)*u1 - img_origin1) / voxsize1;
if (istart_tof_f > iend_tof_f){
tmp = iend_tof_f;
iend_tof_f = istart_tof_f;
istart_tof_f = tmp;
}
istart_tof = (int)floor(istart_tof_f);
iend_tof = (int)ceil(iend_tof_f);
if(istart_tof > istart){istart = istart_tof;}
if(iend_tof < iend){iend = iend_tof;}
//-----------
if (istart < 0){istart = 0;}
if (iend >= n1){iend = n1;}
//---
for(i1 = istart; i1 < iend; i1++)
{
// get the indices where the ray intersects the image plane
x_pr0 = xstart0 + (img_origin1 + i1*voxsize1 - xstart1)*d0 / d1;
x_pr2 = xstart2 + (img_origin1 + i1*voxsize1 - xstart1)*d2 / d1;
i0_floor = (int)floor((x_pr0 - img_origin0)/voxsize0);
i0_ceil = i0_floor + 1;
i2_floor = (int)floor((x_pr2 - img_origin2)/voxsize2);
i2_ceil = i2_floor + 1;
// calculate the distances to the floor normalized to [0,1]
// for the bilinear interpolation
tmp_0 = (x_pr0 - (i0_floor*voxsize0 + img_origin0)) / voxsize0;
tmp_2 = (x_pr2 - (i2_floor*voxsize2 + img_origin2)) / voxsize2;
//--------- TOF related quantities
// calculate the voxel center needed for TOF weights
x_v0 = x_pr0;
x_v1 = img_origin1 + i1*voxsize1;
x_v2 = x_pr2;
if(p[i] != 0){
// calculate distance of voxel to tof bin center
dtof = sqrtf(powf((x_m0 + (it*tofbin_width + tc_offset)*u0 - x_v0), 2) +
powf((x_m1 + (it*tofbin_width + tc_offset)*u1 - x_v1), 2) +
powf((x_m2 + (it*tofbin_width + tc_offset)*u2 - x_v2), 2));
//calculate the TOF weight
tw = 0.5f*(erff_as((dtof + 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)) -
erff_as((dtof - 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)));
if ((i0_floor >= 0) && (i0_floor < n0) && (i2_floor >= 0) && (i2_floor < n2))
{
#pragma omp atomic
img[n1*n2*i0_floor + n2*i1 + i2_floor] +=
(tw * p[i] * (1 - tmp_0) * (1 - tmp_2) * cf);
}
if ((i0_ceil >= 0) && (i0_ceil < n0) && (i2_floor >= 0) && (i2_floor < n2))
{
#pragma omp atomic
img[n1*n2*i0_ceil + n2*i1 + i2_floor] +=
(tw * p[i] * tmp_0 * (1 - tmp_2) * cf);
}
if ((i0_floor >= 0) && (i0_floor < n0) && (i2_ceil >= 0) && (i2_ceil < n2))
{
#pragma omp atomic
img[n1*n2*i0_floor + n2*i1 + i2_ceil] +=
(tw * p[i] * (1 - tmp_0) * tmp_2 * cf);
}
if((i0_ceil >= 0) && (i0_ceil < n0) && (i2_ceil >= 0) && (i2_ceil < n2))
{
#pragma omp atomic
img[n1*n2*i0_ceil + n2*i1 + i2_ceil] +=
(tw * p[i] * tmp_0 * tmp_2 * cf);
}
}
}
}
//---------------------------------------------------------------------------------
if (direction == 2)
{
// case where ray is most parallel to the 2 axis
// we step through the volume along the 2 direction
// factor for correctiong voxel size and |cos(theta)|
cf = voxsize2/cs2;
//--- check where ray enters / leaves cube
istart_f = (xstart2 + t1*d2 - img_origin2) / voxsize2;
iend_f = (xstart2 + t2*d2 - img_origin2) / voxsize2;
if (istart_f > iend_f){
tmp = iend_f;
iend_f = istart_f;
istart_f = tmp;
}
istart = (int)floor(istart_f);
iend = (int)ceil(iend_f);
//-- check where we should start and stop according to the TOF kernel
//-- the tof weights outside +- 3 sigma will be close to 0 so we can
//-- ignore them
istart_tof_f = (x_m2 + (it*tofbin_width - n_sigmas*sig_tof)*u2 - img_origin2) / voxsize2;
iend_tof_f = (x_m2 + (it*tofbin_width + n_sigmas*sig_tof)*u2 - img_origin2) / voxsize2;
if (istart_tof_f > iend_tof_f){
tmp = iend_tof_f;
iend_tof_f = istart_tof_f;
istart_tof_f = tmp;
}
istart_tof = (int)floor(istart_tof_f);
iend_tof = (int)ceil(iend_tof_f);
if(istart_tof > istart){istart = istart_tof;}
if(iend_tof < iend){iend = iend_tof;}
//-----------
if (istart < 0){istart = 0;}
if (iend >= n2){iend = n2;}
//---
for(i2 = istart; i2 < iend; i2++)
{
// get the indices where the ray intersects the image plane
x_pr0 = xstart0 + (img_origin2 + i2*voxsize2 - xstart2)*d0 / d2;
x_pr1 = xstart1 + (img_origin2 + i2*voxsize2 - xstart2)*d1 / d2;
i0_floor = (int)floor((x_pr0 - img_origin0)/voxsize0);
i0_ceil = i0_floor + 1;
i1_floor = (int)floor((x_pr1 - img_origin1)/voxsize1);
i1_ceil = i1_floor + 1;
// calculate the distances to the floor normalized to [0,1]
// for the bilinear interpolation
tmp_0 = (x_pr0 - (i0_floor*voxsize0 + img_origin0)) / voxsize0;
tmp_1 = (x_pr1 - (i1_floor*voxsize1 + img_origin1)) / voxsize1;
//--------- TOF related quantities
// calculate the voxel center needed for TOF weights
x_v0 = x_pr0;
x_v1 = x_pr1;
x_v2 = img_origin2 + i2*voxsize2;
if(p[i] != 0){
// calculate distance of voxel to tof bin center
dtof = sqrtf(powf((x_m0 + (it*tofbin_width + tc_offset)*u0 - x_v0), 2) +
powf((x_m1 + (it*tofbin_width + tc_offset)*u1 - x_v1), 2) +
powf((x_m2 + (it*tofbin_width + tc_offset)*u2 - x_v2), 2));
//calculate the TOF weight
tw = 0.5f*(erff_as((dtof + 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)) -
erff_as((dtof - 0.5f*tofbin_width)/(sqrtf(2)*sig_tof)));
if ((i0_floor >= 0) && (i0_floor < n0) && (i1_floor >= 0) && (i1_floor < n1))
{
#pragma omp atomic
img[n1*n2*i0_floor + n2*i1_floor + i2] +=
(tw * p[i] * (1 - tmp_0) * (1 - tmp_1) * cf);
}
if ((i0_ceil >= 0) && (i0_ceil < n0) && (i1_floor >= 0) && (i1_floor < n1))
{
#pragma omp atomic
img[n1*n2*i0_ceil + n2*i1_floor + i2] +=
(tw * p[i] * tmp_0 * (1 - tmp_1) * cf);
}
if ((i0_floor >= 0) && (i0_floor < n0) && (i1_ceil >= 0) && (i1_ceil < n1))
{
#pragma omp atomic
img[n1*n2*i0_floor + n2*i1_ceil + i2] +=
(tw * p[i] * (1 - tmp_0) * tmp_1 * cf);
}
if ((i0_ceil >= 0) && (i0_ceil < n0) && (i1_ceil >= 0) && (i1_ceil < n1))
{
#pragma omp atomic
img[n1*n2*i0_ceil + n2*i1_ceil + i2] +=
(tw * p[i] * tmp_0 * tmp_1 * cf);
}
}
}
}
}
}
}
|
Par-01-SimpleOmpParallelFor.c |
int main(int argc, char **argv) {
int a[4] = {1,2,3,4};
#pragma omp parallel for
for (int i = 0; i < 4; ++i) {
a[i] = 3*a[i];
}
return 0;
}
|
swap.h | /*
* Author: Salvatore Mandra (salvatore.mandra@nasa.gov)
*
* Copyright © 2021, United States Government, as represented by the
* Administrator of the National Aeronautics and Space Administration. All
* rights reserved.
*
* The HybridQ: A Hybrid Simulator for Quantum Circuits platform is licensed
* under the Apache License, Version 2.0 (the "License"); you may not use this
* file except in compliance with the License. You may obtain a copy of the
* License at http://www.apache.org/licenses/LICENSE-2.0.
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* License for the specific language governing permissions and limitations under
* the License.
*/
#ifndef HYBRIDQ__SWAP_H
#define HYBRIDQ__SWAP_H
#include "pack.h"
#include "utils.h"
namespace hybridq::swap {
template <typename Positions, std::size_t size = array_size_v<Positions>>
constexpr inline auto swap(std::size_t x, Positions&& pos) {
std::size_t y{0};
for (std::size_t i = 0; i < size; ++i) y ^= ((x >> i) & 1uL) << pos[i];
return y;
}
template <typename pack_type, typename Positions, std::size_t... I>
constexpr inline void _swap(pack_type&& pack, Positions&& pos,
std::index_sequence<I...>) {
pack = remove_pcvr_t<pack_type>{pack[swap(I, pos)]...};
}
template <typename pack_type, typename Positions,
std::size_t size = pack_size_v<pack_type>>
constexpr inline void swap(pack_type&& pack, Positions&& pos) {
_swap(pack, pos, std::make_index_sequence<size>{});
}
template <typename float_type, typename Positions,
std::size_t swap_size = array_size_v<Positions>>
int swap_array(float_type* array, Positions&& pos, const std::size_t size) {
// Reinterpret
auto* _array = reinterpret_cast<
typename hybridq::__pack__<float_type, 1uL << swap_size>::value_type*>(
array);
#pragma omp parallel for
for (std::size_t i = 0; i < (size >> swap_size); ++i) swap(_array[i], pos);
return 0;
}
template <typename float_type, typename index_type>
int swap_array(float_type* array, index_type* pos, const std::size_t size,
const std::size_t n_pos) {
// Get U_Size
const std::size_t swap_size = 1uL << n_pos;
// Compute offset
std::size_t offset[n_pos];
for (std::size_t i = 0; i < n_pos; ++i) {
offset[i] = 0;
for (std::size_t j = i + 1; j < n_pos; ++j) offset[i] += (pos[j] < pos[i]);
}
// Allocate buffers
float_type _array[swap_size];
std::size_t _swap_pos[swap_size];
for (std::size_t x = 0; x < swap_size; ++x) {
std::size_t y{0};
for (std::size_t i = 0; i < n_pos; ++i) y ^= ((x >> i) & 1uL) << pos[i];
_swap_pos[x] = y;
}
#pragma omp parallel for private(_array)
for (std::size_t i = 0; i < size >> n_pos; ++i) {
// Load buffer
for (std::size_t j = 0; j < swap_size; ++j)
_array[j] = array[_swap_pos[j] ^ (i << n_pos)];
// Dump buffer
for (std::size_t j = 0; j < swap_size; ++j)
array[j ^ (i << n_pos)] = _array[j];
}
return 0;
}
} // namespace hybridq::swap
#endif
|
simpson_omp.c | #include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <omp.h>
#include <math.h>
#include <gsl/gsl_integration.h>
#include <sys/time.h>
// ***************************************************************************************
// * Pre-Port Entwicklung und Testen eines OMP-parallisierten, adaptiven Simpson-integrators für
// * die fiesen doppelintgrale in imd_colrad.c
// ***************************************************************************************
int num_threads;
int simpson_error;
int funcevals;
int iter;
const double tolmax=1e-30;
#define INITIAL_STACK_SIZE 128; //128 /* initial size of new stacks */
const double alpha=0.816496580927726;
const double beta=0.447213595499958;
static const double xgkq[12] =
{
0.0,
-0.942882415695480,
-0.816496580927726,
-0.641853342345781,
-0.447213595499958,
-0.236383199662150,
0.0,
0.236383199662150,
0.447213595499958,
0.641853342345781,
0.816496580927726,
0.942882415695480
};
/* the stack structure */
struct stack_s{
int el_count; /* count of elements on stack */
int el_size; /* size of an element */
int mem_reserve; /* allocated memory for stack */
void* elements; /* pointer to begin of stack */
};
struct my_f_params { double ne; double T;double mu; double E;double DeltaE; int allowed;};
typedef struct _work_t{
double a;
double b;
double tol;
double S;
double fa;
double fb;
double fm;
double rec;
int iter;
struct my_f_params * p; //pointer auf params
} work_t;
typedef struct stack_s* stack_t;
typedef struct _work_t_gkq{
double a;
double b;
double toler;
double I_13;
double fa;
double fb;
struct my_f_params * p; //pointer auf params
double (*f)(double, struct my_f_params*);
stack_t stack_inner;
short int is_parent;
short int subtask_nr; //inner task nr
short int task_nr; //outer task nr
short int subtasks_left; //erst pop'n wenn aller inner tasks (intergals) vollständig!
double inner_integrals[5];
} work_gkq;
// **************************************** FUNCTION DEFS *************************************************
void create_stack(stack_t* stack, int element_size);
int empty_stack(stack_t stack);
void push_stack(stack_t stack, void* element);
void pop_stack(stack_t stack, void* element);
double
integral2( double (*f)(double, struct my_f_params*), /* function to integrate */
stack_t stack);
double gkq_adapt_double(stack_t stack);
double gkq_adapt_single(stack_t stack); //for single integral, first integral
double gkq_double(double (*f)(double, struct my_f_params*), double (*finner)(double, struct my_f_params*),
double a, double b,
double TOL, struct my_f_params* p,stack_t stack);
double gkq_adapt_serial(double (*f)(double, struct my_f_params*), double a, double b, double fa,double fb, double toler,double I_13, struct my_f_params* p);
work_gkq gkq_create_inner_task(double (*f)(double, struct my_f_params*), double a, double b, double TOL, struct my_f_params* p);
double gkq_single(double (*f)(double, struct my_f_params*), double a, double b,
double TOL, struct my_f_params* p,stack_t stack);
int terminate_serial;
int terminate_gkq;
// static double myfun(double x,struct my_f_params* p)
// {
// double T=p->T;
// double fermi=1/(1+exp(-x/T));
// double fermi2=1- 1/(1+exp(-x/T));
// double sigma=x/T*log(x/T*2)/pow(1/x,2.0); //einfach nur ärger machen
// // return exp(-x*x)/p->T+log(x*pow(p->T,2.0))/p->ne;
// return fermi*fermi2*sigma;
// }
// static double myfun2(double x,void* pv)
// {
// struct my_f_params *p= (struct my_f_params*) pv;
// double T=p->T;
// double fermi=1/(1+exp(-x/T));
// double fermi2=1- 1/(1+exp(-x/T));
// double sigma=x/T*log(x/T*2)/pow(1/x,2.0); //einfach nur ärger machen
// // return exp(-x*x)/p->T+log(x*pow(p->T,2.0))/p->ne;
// return fermi*fermi2*sigma;
// }
static double inner_integrand(double x,void *pv)
{
struct my_f_params *p= (struct my_f_params*) pv;
double T=p->T;
double fermi=1/(1+exp(-x/T));
double fermi2=1- 1/(1+exp(-x/T));
return exp(-x*x); //fermi*fermi2;
}
static double inner_integrand2(double x,struct my_f_params* p)
{
double T=p->T;
double fermi=1/(1+exp(-x/T));
double fermi2=1- 1/(1+exp(-x/T));
return fermi*fermi2;
}
void get_integ_bounds_inner(double *integ_bnd, double x, struct my_f_params *p)
{
integ_bnd[0]=-4;
integ_bnd[1]=4;
}
gsl_integration_workspace * gsinner;
gsl_integration_workspace * gsinner2;
static double myfun(double x,struct my_f_params* p)
{
double T=p->T;
double fermi=1/(1+exp(-x/T));
double fermi2=1- 1/(1+exp(-x/T));
double sigma=x/T*log(x/T*2)/pow(1/x,2.0); //einfach nur ärger machen
// double result, err;
// gsl_function gslfun;
// gslfun.function=&inner_integrand;
// gslfun.params=p;
// gsl_integration_qag(&gslfun, x, 300, 0.0, 1e-3, 1000,1,
// gsinner2, &result, &err);
// gsl_integration_workspace_free(gsinner2);
// stack_t stack2;
// create_stack(&stack2, sizeof(work_gkq));
//double result=gkq(inner_integrand2, x, 600, 1e-3, p);//, stack);
// free(stack2->elements);
return sigma;
//return exp(-x*x);
};
static double myfun_gsl(double x,void* pv)
{
struct my_f_params *p= (struct my_f_params*) pv;
double T=p->T;
double fermi=1/(1+exp(-x/T));
double fermi2=1- 1/(1+exp(-x/T));
double sigma=x/T*log(x/T*2)/pow(1/x,2.0); //einfach nur ärger machen
gsl_function gslfun;
gslfun.function=&inner_integrand;
gslfun.params=pv;
double result, err;
gsl_integration_qag(&gslfun, x, 600, 0.0, 1e-3, 1000,1,
gsinner, &result, &err);
return sigma*result;
}
// static double myfun2(double x,void* pv)
// {
// struct my_f_params *p= (struct my_f_params*) pv;
// static float sinfc(float x) { return sinf(x); }
// static double frand48c(double x) { funcevals++; return (double) drand48(); }
// *************************************************************************************************************
int main(int argc,char** argv)
{
iter=0;
num_threads=omp_get_num_threads();
funcevals=0;
simpson_error=0;
struct timeval start, end;
double gsltime, simpsontime;
// ************************************************
gsinner= gsl_integration_workspace_alloc (1000);
// double gslinteg=0.0;
// double integ_err;
// gsl_function gslfun;
// gslfun.function=&myfun;
// **********************************************
double pi=3.141592653589793;
double xmin, xmax;
double answer = 0.0;
int i;
xmin=2;
xmax=100;
// ***********************************************
double T0=1;
double ne0=1e5;
struct my_f_params ptest;
ptest.T=T0;
ptest.ne=ne0;
// *************************************************
int method=GSL_INTEG_GAUSS61;
gsl_function gslfun;
gslfun.function=&myfun_gsl;
gslfun.params=&ptest;
gsl_integration_workspace * gswork=NULL;
gswork= gsl_integration_workspace_alloc (1000);
double gslinteg=0.0;
double integ_err;
gettimeofday(&start, NULL);
gsl_integration_qag(&gslfun, xmin, xmax, 0.0, 1e-8, 1000,1,
gswork, &gslinteg, &integ_err);
gettimeofday(&end, NULL);
gsltime = ((end.tv_sec - start.tv_sec) * 1000000u +
end.tv_usec - start.tv_usec) /1.e6;
printf("gslres:%.12e, gsltime:%.4e\n", gslinteg, gsltime);
// *************************************************
stack_t stack; //global stack
create_stack(&stack, sizeof(work_gkq));
gettimeofday(&start, NULL);
double integ=gkq_double(myfun,inner_integrand2, xmin, xmax, 1e-2, &ptest,stack);
gettimeofday(&end, NULL);
free(stack->elements);
free(stack);
double partime = ((end.tv_sec - start.tv_sec) * 1000000u +
end.tv_usec - start.tv_usec) /1.e6;
printf("my:%.12e, partime:%.4e\n", integ,partime);
return 0;
}
/******************************************
* create new stack
******************************************/
void create_stack(
stack_t* stack, /* stack to create */
int element_size) /* size of a stack element */
{
int initial_size = INITIAL_STACK_SIZE;
/* allocate memory for new stack struct */
(*stack) = (stack_t) malloc(sizeof(struct stack_s));
if (!(*stack)){
fprintf(stderr, "error: could not allocate memory for stack.. Abort.\n");
exit(1);
}
/* allocate memory for stack elements */
(*stack)->elements = (void*) malloc(element_size * initial_size);
(*stack)->mem_reserve = initial_size;
if (!(*stack)->elements){
fprintf(stderr, "error: could not allocate memory for stack.. Abort.\n");
exit(1);
}
(*stack)->el_size = element_size;
(*stack)->el_count = 0;
}
/*****************************************
* check if the stack is empty
*****************************************/
int empty_stack(stack_t stack)
{
//MYMOD
if(stack==NULL)
return 1;
//ENDOF MYMOD
return stack->el_count <= 0;
}
/*****************************************
* push a element on stack
*****************************************/
void push_stack(
stack_t stack, /* target stack */
void* element) /* element to push */
{
int i, new_reserve;
int log2_count;
/* check if we need more memory for stack */
if (stack->el_count >= stack->mem_reserve)
{
/* calculate new size for the stack
it should be a power of two */
// for (i = stack->el_count, log2_count = 0; i > 0; i>>1, log2_count++);
// new_reserve = 1 << log2_count;
log2_count=0;
for(i=stack->el_count;i>0; i=i*0.5) //i>>1)
{
log2_count++;
}
new_reserve= 1 << log2_count;
/* reallocate memory for phase thread tables
and nullify new values */
stack->elements = (void *) realloc(stack->elements,
stack->el_size * new_reserve);
if (!stack->elements){
fprintf(stderr, "error: can't reallocate stack.. Aborting\n");
exit(1);
}
stack->mem_reserve = new_reserve;
}
/* now push the element on top of the stack */
memcpy((char*)stack->elements + stack->el_count*stack->el_size,
element, stack->el_size);
stack->el_count++;
}
/*****************************************
* pop an element from stack
*****************************************/
void pop_stack(
stack_t stack, /* target stack */
void* element) /* where poped el. should be stored */
{
if (stack->el_count <= 0){
fprintf(stderr, "error: trying to pop from empty stack.\n");
exit(2);
}
stack->el_count--;
memcpy(element,
(char*)stack->elements + stack->el_count*stack->el_size,
stack->el_size);
}
// ***************************************************************************
// * Gauss-kronard quadrature
// ***************************************************************************
work_gkq gkq_create_inner_task(double (*f)(double, struct my_f_params*), double a, double b, double TOL, struct my_f_params* p)
{
work_gkq work;
work.f=f;
struct my_f_params* pinner;
pinner=(struct my_f_params*) malloc(sizeof(struct my_f_params)); //ACHTUNG: Muss wieder gefreed werden
pinner->mu= p->mu;
pinner->T=p->T;
pinner->ne=p->ne;
work.p=pinner;
work.a=a;
work.b=b;
work.subtasks_left=0;//beliebig!
double m=0.5*(a+b);
double h=0.5*(b-a);
double y[13];
double fa=y[0]=f(a,p);
double fb=y[12]=f(b,p);
int i;
for(i=1;i<12;i++)
y[i]=f(m+xgkq[i]*h,p);
double I_4= (h/6.0)*(y[0]+y[12]+5.0*(y[4]+y[8])); // 4-point gauss-lobatto
double I_7= (h/1470.0)*(77.0*(y[0]+y[12])+432.0*(y[2]+y[10])+ // 7-point kronrod
625.0*(y[4]+y[8])+672.0*y[6]);
double I_13= h*(0.0158271919734802*(y[0]+y[12])+0.0942738402188500*(y[1]+y[11])+0.155071987336585*(y[2]+y[10])+
0.188821573960182*(y[3]+y[9])+0.199773405226859*(y[4]+y[8])+0.224926465333340*(y[5]+y[7])+
0.242611071901408*y[6]); //13-point Kronrod
double Err1=fabs(I_7-I_13);
double Err2=fabs(I_4-I_13);
double r=(Err2 != 0.0) ? Err1/Err2 : 1.0;
double toler=(r > 0.0 && r < 1.0) ? TOL/r : TOL;
if(I_13 == 0)
I_13=b-a;
I_13=fabs(I_13);
// printf("create inner task: a:%f, b:%f, I4:%f,I7:%f,I13:%f\n",a,b,I_4,I_7,I_13);
work.toler = toler;
work.I_13=I_13;
work.fa=fa;
work.fb=fb;
work.is_parent=0;
for(i=0;i<5;i++)
work.inner_integrals[i]=0.0;
return work;
}
double gkq_double(double (*f)(double, struct my_f_params*), double (*finner)(double, struct my_f_params*),
double a, double b,
double TOL, struct my_f_params* p,stack_t stack)
{
//1st integration
//create parent task
double result=0.0;
// *********************************************
double m=0.5*(a+b);
double h=0.5*(b-a);
int i;
double y[13];
stack_t st;
create_stack(&st, sizeof(work_gkq));
double integ_bnd[2];
get_integ_bounds_inner(integ_bnd, a, p);
y[0]=f(a,p)* gkq_single(finner,integ_bnd[0],integ_bnd[1],1e-3, p, st);
for(i=1;i<12;i++)
{
get_integ_bounds_inner(integ_bnd, m+xgkq[i]*h, p);
y[i]=f(m+xgkq[i]*h,p)*gkq_single(finner,integ_bnd[0],integ_bnd[1],1e-3,p,st);
}
get_integ_bounds_inner(integ_bnd, b, p);
y[12]=f(b,p)*gkq_single(finner,integ_bnd[0],integ_bnd[1],1e-3, p, st);
free(st->elements);
free(st);
double fa=y[0];
double fb=y[12];
double I_4= (h/6.0)*(y[0]+y[12]+5.0*(y[4]+y[8])); // 4-point gauss-lobatto
double I_7= (h/1470.0)*(77.0*(y[0]+y[12])+432.0*(y[2]+y[10])+ // 7-point kronrod
625.0*(y[4]+y[8])+672.0*y[6]);
double I_13= h*(0.0158271919734802*(y[0]+y[12])+0.0942738402188500*(y[1]+y[11])+0.155071987336585*(y[2]+y[10])+
0.188821573960182*(y[3]+y[9])+0.199773405226859*(y[4]+y[8])+0.224926465333340*(y[5]+y[7])+
0.242611071901408*y[6]); //13-point Kronrod
double Err1=fabs(I_7-I_13);
double Err2=fabs(I_4-I_13);
double r=(Err2 != 0.0) ? Err1/Err2 : 1.0;
double toler=(r > 0.0 && r < 1.0) ? TOL/r : TOL;
if(I_13 == 0)
I_13=b-a;
I_13=fabs(I_13);
printf("First approx I_13:%.4e\n",I_13);
//Prepare work and push onto stack
work_gkq work;
work.a = a;
work.b = b;
work.toler = toler;
work.I_13=I_13;
work.fa=fa;
work.fb=fb;
work.p=p;
work.f=f;
work.is_parent=1;
work.task_nr=0; //fange bei 0 das zählen an
for(i=0;i<5;i++)
work.inner_integrals[i]=0.0;
// ALLOC INNER WS FOR INTEGR.
// gsl_integration_workspace *gsinner2= gsl_integration_workspace_alloc (1000);
stack_t stack_inner;
create_stack(&stack_inner, sizeof(work_gkq));
work.stack_inner=stack_inner;
//push work on inner stack
//for comp of 5 inner intgrals
work_gkq winner;
double m_inner, h_inner, mll_inner,ml_inner, mr_inner, mrr_inner;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0], integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0], integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0], integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0], integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0], integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//push outer integtand work on outer stack
work.subtasks_left=5;
push_stack(stack, &work);
/*
int elemsinner=work.stack_inner->el_count;
int elout=stack->el_count;
work_gkq* top=(work_gkq*) stack->elements +(stack->el_count-1); //get top element
int inner2=0;
inner2=top->stack_inner->el_count;
*/
// printf("elemsinner:%d,ou:%d,in2:%d\n",elemsinner,elout, inner2);
// int in2= top->stack_inner->el_count;
result=gkq_adapt_double(stack);//,stack);
// result=gkq_adapt_serial(f,a,b,fa,fb,toler,I_13, p);
// gsl_integration_workspace_free(gsinner2);
// free(stack->elements);
return result;
}
double gkq_single(double (*f)(double, struct my_f_params*), double a, double b,
double TOL, struct my_f_params* p,stack_t stack)
{
double result=0.0;
// *********************************************
double m=0.5*(a+b);
double h=0.5*(b-a);
int i;
double y[13];
double fa=y[0]=f(a,p);
double fb=y[12]=f(b,p);
for(i=1;i<12;i++) //hier müssen direkt 13 integrale berechnet werden!!!
y[i]=f(m+xgkq[i]*h,p); //auch das sollte parallel erfolgen (pragma parfor?)
double I_4= (h/6.0)*(y[0]+y[12]+5.0*(y[4]+y[8])); // 4-point gauss-lobatto
double I_7= (h/1470.0)*(77.0*(y[0]+y[12])+432.0*(y[2]+y[10])+ // 7-point kronrod
625.0*(y[4]+y[8])+672.0*y[6]);
double I_13= h*(0.0158271919734802*(y[0]+y[12])+0.0942738402188500*(y[1]+y[11])+0.155071987336585*(y[2]+y[10])+
0.188821573960182*(y[3]+y[9])+0.199773405226859*(y[4]+y[8])+0.224926465333340*(y[5]+y[7])+
0.242611071901408*y[6]); //13-point Kronrod
double Err1=fabs(I_7-I_13);
double Err2=fabs(I_4-I_13);
double r=(Err2 != 0.0) ? Err1/Err2 : 1.0;
double toler=(r > 0.0 && r < 1.0) ? TOL/r : TOL;
if(I_13 == 0)
I_13=b-a;
I_13=fabs(I_13);
//Prepare work and push onto stack
work_gkq work;
work.a = a;
work.b = b;
work.toler = toler;
work.I_13=I_13;
work.fa=fa;
work.fb=fb;
work.p=p;
work.a=a;
work.f=f;
push_stack(stack, &work);
result=gkq_adapt_single(stack);//,stack);
return result;
}
double gkq_adapt_serial(double (*f)(double, struct my_f_params*), double a, double b,
double fa,double fb, double toler,double I_13, struct my_f_params* p)
{
double m = (a+b)/2;
double h = (b-a)/2;
double mll=m-alpha*h;
double ml=m-beta*h;
double mr=m+beta*h;
double mrr=m+alpha*h;
double fmll=f(mll,p);
double fml=f(ml,p);
double fm=f(m,p);
double fmr=f(mr,p);
double fmrr=f(mrr,p);
double I_4=h/6.0*(fa+fb+5.0*(fml+fmr)); // 4-point Gauss-Lobatto formula.
double I_7=h/1470.0*(77.0*(fa+fb)+432.0*(fmll+fmrr)+625.0*(fml+fmr)+672.0*fm);
if (fabs(I_7-I_4) <= toler*I_13 || mll <= a || b <= mrr)
{
if ((mll <= a || b <= mrr) && !terminate_serial) //Error
{
// out_of_tolerance=true; // Interval contains no more machine numbers
printf("OUT OF TOLERANCE !!!, mll:%.4e, a:%.4e, b:%.4e, mrr:%.4e\n", mll,b,b,mrr);
terminate_serial=1;
}
// printf("me ok:%d, a:%f,b:%f, tler:%.5e,I_4:%f,I_7:%f\n", omp_get_thread_num(), a,b,toler,I_4,I_7);
return I_7;
}
else
{
// printf("me NOOOO:%d, a:%f,b:%f, tler:%.5e,I_4:%f,I_7:%f\n", omp_get_thread_num(), a,b,toler,I_4,I_7);
return gkq_adapt_serial(f, a,mll,fa,fmll,toler,I_13,p) +
gkq_adapt_serial(f, mll,ml,fmll,fml,toler,I_13,p) +
gkq_adapt_serial(f, ml,m,fml,fm,toler,I_13,p) +
gkq_adapt_serial(f, m,mr,fm,fmr,toler,I_13,p) +
gkq_adapt_serial(f, mr,mrr,fmr,fmrr,toler,I_13,p) +
gkq_adapt_serial(f, mrr,b,fmrr,fb,toler,I_13,p);
}
}
double gkq_adapt_single(stack_t stack)
{
work_gkq work;
// work.iter=0;
int ready, idle, busy;
double integral_result = 0.0;
busy = 0;
terminate_gkq=0;
#pragma omp parallel default(none) \
shared(stack, integral_result,busy) \
private(work, idle, ready)
{
// printf("me:%d, err:%d\n",omp_get_thread_num(),simpson_error);
ready = 0;
idle = 1;
while(!ready) // && !terminate_gkq)// && !simpson_error) //<-- so NICHT!
{
#pragma omp critical (stack)
{
if (!empty_stack(stack))
{
/* we have new work */
pop_stack(stack, &work);
if (idle)
{
/* say others i'm busy */
busy += 1;
idle = 0;
}
}
else
{
/* no new work on stack */
if (!idle){
busy -= 1;
idle = 1;
}
/* nobody has anything to do; let us leave the loop */
if (busy == 0)
{
ready = 1;
}
}
} /* end critical(stack) */
if (idle)
continue; //if ready==1 --> leave loop
// double I_prev=work.I_prev;
double (*f)(double, struct my_f_params*)=work.f;
double a = work.a;
double b = work.b;
double toler = work.toler;
double I_13=work.I_13;
double fa=work.fa;
double fb=work.fb;
// int iter=work.iter;
// double *y= work.y; // brauch ich nicht!
struct my_f_params * p = work.p;
double m = (a+b)/2;
double h = (b -a)/2;
double mll=m-alpha*h;
double ml=m-beta*h;
double mr=m+beta*h;
double mrr=m+alpha*h;
double fmll=f(mll,p);
double fml=f(ml,p);
double fm=f(m,p);
double fmr=f(mr,p);
double fmrr=f(mrr,p);
double I_4=h/6.0*(fa+fb+5.0*(fml+fmr)); // 4-point Gauss-Lobatto formula.
double I_7=h/1470.0*(77.0*(fa+fb)+432.0*(fmll+fmrr)+625.0*(fml+fmr)+672.0*fm);
// if(myid==1)
// printf("I_7:%.4e, I_13:%.4e,I_4:%.4e, minus:%.4e, to:%.4e\n",I_7,I_13,I_4,I_7-I_4, toler*I_13);
// int maxiter=50; //max. subdivisions
// double abstol=1e-30;
// work.I_prev=I_7; // für abstolcheck in nächster recursion
if (fabs(I_7-I_4) <= toler*I_13 || mll <= a || b <= mrr) // || iter > maxiter || fabs(I_7-I_prev) < abstol )
{
if ((mll <= a || b <= mrr)) //Error
{
// out_of_tolerance=true; // Interval contains no more machine numbers
// printf("OUT OF TOLERANCE !!!, mll:%.4e, a:%.4e, b:%.4e, mrr:%.4e,I_7-I_4:%.4e, tol:%.4e,I_13:%.4e\n",
// mll,b,b,mrr,I_7-I_4, toler*I_13,I_13);
}
#pragma omp critical (integral_result)
{
integral_result += I_7; //Terminate recursion.
}
}
else //subdivide interval and push new work on stack
{
#pragma omp critical (stack)
{
// work.iter=iter+1;
work.a=a;
work.b=mll;
work.fa=fa;
work.fb=fmll;
push_stack(stack, &work);
work.a=mll;
work.b=ml;
work.fa=fmll;
work.fb=fml;
push_stack(stack, &work);
work.a=ml;
work.b=m;
work.fa=fml;
work.fb=fm;
push_stack(stack, &work);
work.a=m;
work.b=mr;
work.fa=fm;
work.fb=fmr;
push_stack(stack, &work);
work.a=mr;
work.b=mrr;
work.fa=fmr;
work.fb=fmrr;
push_stack(stack, &work);
work.a=mrr;
work.b=b;
work.fa=fmrr;
work.fb=fb;
push_stack(stack, &work);
} // pragma critical stack
} // else ..non-acceptable error
} // while
} /* end omp parallel */
return integral_result;
}
double gkq_adapt_double(stack_t stack)
{
work_gkq work;
work_gkq* pwork_outer;
int ready, idle, busy;
double integral_result = 0.0;
busy = 0;
int myid;
int elcnt; //nr of outer task elements on stack
#pragma omp parallel default(none) \
shared(stack, integral_result,busy,iter,elcnt) \
private(work, idle, ready,myid,pwork_outer)
{
myid=omp_get_thread_num();
ready = 0;
idle = 1;
while(!ready) // && !terminate_gkq)// && !simpson_error) //<-- so NICHT!
{
// printf("\n NEXT, curapprox:%.4e\n",integral_result);
#pragma omp critical (stack)
{
//pointer to outer work element
//work_gkq* pwork=(work_gkq*) stack->elements + stack->el_count*stack->el_size;
elcnt=stack->el_count;
// if(elcnt>1)
// getchar();
stack_t stack_inner;
if(elcnt>0)
{
pwork_outer=(work_gkq*) stack->elements +(elcnt-1); //get top element
//IST WOHL SUBOPTIMAL
while(pwork_outer->task_nr > 0 && pwork_outer->subtasks_left==0) //this task is complete, goto next outer task
{
printf("myid:%d, outer task nr:%d, complete:subs:%d, decrement\n",myid,pwork_outer->task_nr,pwork_outer->subtasks_left);
pwork_outer--;
}
stack_inner=pwork_outer->stack_inner;
}
else
{
stack_inner=NULL;
pwork_outer=NULL;
}
printf("myid:%d,elcnt:%d,sinner:%p,pwout:%p\n",myid,elcnt,stack_inner,pwork_outer);
{ //stackinner gehört zu work, und work ist private!
if(!empty_stack(stack_inner))
{
printf("myid:%d,iter:%d, inner stack not empty,pop..\n",myid,iter);
pop_stack(stack_inner, &work);
//if letztes inner-work elem, blockiere outer work elem bis ich das integral hab
iter++;
if (idle)
{
busy += 1;
idle = 0;
}
}
else //inner stack is empty, pop from outer
{
printf("myid:%d,iter:%d, inner stack is empty?\n",myid,iter);
if (!empty_stack(stack)) //work elem ist blockiert solange inner integs nicht vollständig
{
// printf("myid:%d,iter:%d, inner stack empty. work on outer with cnt=%d\n",myid,iter,pwork_outer->subtasks_left);
if(pwork_outer->subtasks_left==0)
{
printf("myid:%d,iter:%d, outer stack not empty\n",myid,iter);
pop_stack(stack, &work);
iter++;
if (idle)
{
busy += 1;
idle = 0;
}
}
else
{
printf("myid:%d,iter:%d, inner integs not complete:%d\n",myid,iter,pwork_outer->subtasks_left);
if (!idle)
{
busy -= 1;
idle = 1;
}
}
}
else //auch outer stack ist leer
{
printf("myid:%d,iter:%d, outer stack empty too\n",myid,iter);
if (!idle)
{
busy -= 1;
idle = 1;
}
if (busy == 0)
{
ready = 1;
}
}
}
} // critical inner stack
} //critical outer stack
if (idle)
{
printf("myid:%d,iter:%d, noth8ing to do\n",myid,iter);
continue; //if ready==1 --> leave loop
}
//work on inner tasks first, if available
if(work.is_parent==0)
{
printf("myid:%d,iter:%d,tasknr:%d,innertask:%d, left subs:%d\n",
myid,iter,work.task_nr,work.subtask_nr,pwork_outer->subtasks_left); //subtasks nr nicht immer aktuell!
// getchar();
double (*f)(double, struct my_f_params*) = work.f;
double a = work.a;
double b = work.b;
double toler = work.toler;
double I_13=work.I_13;
double fa=work.fa;
double fb=work.fb;
// double *y= work.y; // brauch ich nicht!
struct my_f_params * p = work.p;
double m = (a+b)/2;
double h = (b -a)/2;
double mll=m-alpha*h;
double ml=m-beta*h;
double mr=m+beta*h;
double mrr=m+alpha*h;
double fmll=f(mll,p);
double fml=f(ml,p);
double fm=f(m,p);
double fmr=f(mr,p);
double fmrr=f(mrr,p);
double I_4=h/6.0*(fa+fb+5.0*(fml+fmr)); // 4-point Gauss-Lobatto formula.
double I_7=h/1470.0*(77.0*(fa+fb)+432.0*(fmll+fmrr)+625.0*(fml+fmr)+672.0*fm);
// printf("myid:%d,iter:%d non parent checkpoint 1 passed\n",myid,iter);
if (fabs(I_7-I_4) <= toler*I_13 || mll <= a || b <= mrr)
{
if ((mll <= a || b <= mrr))// && !terminate_gkq) //Error
{
printf("OUT OF TOLERANCE !!!, mll:%.4e, a:%.4e, b:%.4e, mrr:%.4e\n", mll,a,b,mrr);
}
// printf("myid:%d,iter:%d non parent checkpoint 2.1 passed\n",myid,iter);
int tasknr=work.subtask_nr;
printf("myid:%d,iter:%d, winner error acceptable: I_7:%.4e, I_4:%.4e, I_13:%.4e,toler*I_13:%.4e,toler:%.4e\n",
myid,iter,I_7,I_4, I_13,toler*I_13,toler);
#pragma omp critical (innerinteg) // workers greifen auf selbes array zu.nur an anderer Stelle..evtl. besser 6 versch.doubles statt array?
{
double *inner_integrals=pwork_outer->inner_integrals;
inner_integrals[tasknr]+=I_7;
pwork_outer->subtasks_left=pwork_outer->subtasks_left-1; //hier sollte subtasks_left aktuell sein
printf("myid:%d,iter:%d, inner_integral[%d]:%.4e added:%.4e,subtask left:%d\n",
myid,iter,tasknr,inner_integrals[tasknr],I_7,pwork_outer->subtasks_left);
}
}
else //subdivide interval and push new work on stack
{
stack_t stack_inner;
#pragma omp critical //(stack) //(stack_inner)
{
stack_inner = pwork_outer->stack_inner;
pwork_outer->subtasks_left=pwork_outer->subtasks_left-1;//weil dieser stack durch 6 andere ersetzt wird
work.task_nr=pwork_outer->task_nr; //inner task soll auch wissen wer sein outer task is (bisher nur für debug)
//subtasknr bleibt dieselbe (bezogen auf parent task)
work.a=a;
work.b=mll;
work.fa=fa;
work.fb=fmll;
push_stack(stack_inner, &work);
work.a=mll;
work.b=ml;
work.fa=fmll;
work.fb=fml;
push_stack(stack_inner, &work);
work.a=ml;
work.b=m;
work.fa=fml;
work.fb=fm;
push_stack(stack_inner, &work);
work.a=m;
work.b=mr;
work.fa=fm;
work.fb=fmr;
push_stack(stack_inner, &work);
work.a=mr;
work.b=mrr;
work.fa=fmr;
work.fb=fmrr;
push_stack(stack_inner, &work);
work.a=mrr;
work.b=b;
work.fa=fmrr;
work.fb=fb;
push_stack(stack_inner, &work);
pwork_outer->subtasks_left=pwork_outer->subtasks_left+6;
printf("myid:%d,iter:%d, inner_integral approx not accurate..subdivided: I_4:%f,I_7:%f,I_13:%f,left subs:%d\n",
myid,iter,I_4,I_7,I_13,pwork_outer->subtasks_left);
} // pragma critical stack_inner
} // else ..non-acceptable error
}// !isparent
else //parent task without any inner tasks
{
printf("myid:%d,iter:%d, work is parent,subs left:%d\n",myid,iter,work.subtasks_left);
// stack_t stack_inner = work.stack_inner; //brauch ich hier nicht
// free(stack_inner->elements);
// free(stack_inner);
double (*f)(double, struct my_f_params*) = work.f;
double a = work.a;
double b = work.b;
double toler = work.toler;
double I_13=work.I_13;
double fa=work.fa;
double fb=work.fb;
// double *y= work.y; // brauch ich nicht!
struct my_f_params * p = work.p;
double m = (a+b)/2;
double h = (b -a)/2;
double mll=m-alpha*h;
double ml=m-beta*h;
double mr=m+beta*h;
double mrr=m+alpha*h;
//the inner intergrals are pre-calculated and saved in an array
//the function only calculates a prefactor...
double fmll=f(mll,p)*work.inner_integrals[0];
double fml= f(ml,p)*work.inner_integrals[1];
double fm= f(m,p)*work.inner_integrals[2];
double fmr= f(mr,p)*work.inner_integrals[3];
double fmrr= f(mrr,p)*work.inner_integrals[4];
double I_4=h/6.0*(fa+fb+5.0*(fml+fmr)); // 4-point Gauss-Lobatto formula.
double I_7=h/1470.0*(77.0*(fa+fb)+432.0*(fmll+fmrr)+625.0*(fml+fmr)+672.0*fm);
if (fabs(I_7-I_4) <= toler*I_13 || mll <= a || b <= mrr)
{
if ((mll <= a || b <= mrr))// && !terminate_gkq) //Error
{
// out_of_tolerance=true; // Interval contains no more machine numbers
printf("outer task OUT OF TOLERANCE !!!, mll:%.4e, a:%.4e, b:%.4e, mrr:%.4e\n", mll,b,b,mrr);
}
//#pragma omp critical(integal_result)
{
#pragma omp atomic
integral_result += I_7;
printf("myid:%d,iter:%d, I7 added, integs: 0=%f,1=%f,2=%f,3=%f,4=%f\n", myid,iter,
work.inner_integrals[0],work.inner_integrals[1],work.inner_integrals[2],
work.inner_integrals[3],work.inner_integrals[4]);
}
}
else //subdivide interval and push new work on stack
{
printf("myid:%d,iter:%d, iouter_integral approx not accurate: I_7=%.4e,I_4=%.4e,I_13=%.4e, toler*I_13:%.4e\n"
"integs: 0=%f,1=%f,2=%f,3=%f,4=%f..subdivide\n",
myid,iter,I_7,I_4,I_13,I_13*toler,work.inner_integrals[0],work.inner_integrals[1],work.inner_integrals[2],
work.inner_integrals[3],work.inner_integrals[4] );
stack_t stack_inner = work.stack_inner;
#pragma omp critical (stack)
{
//create sub-stack for inner-integrals
work.subtasks_left=5; //for all new outer work elements
create_stack(&stack_inner, sizeof(work_gkq));
work.is_parent=1;
work_gkq winner;
//outer task elem
work.a=a;
work.b=mll;
work.fa=fa;
work.fb=fmll;
double m_inner= (work.a+work.b)/2;
double h_inner= (work.a+work.b)/2;
double mll_inner=(m_inner-alpha*h_inner);
double ml_inner= (m_inner-beta*h_inner);
double mr_inner= (m_inner+beta*h_inner);
double mrr_inner=(m_inner+alpha*h_inner);
//Jedes subinterval bekommt 5 inner tasks für die inneren integrale!
int curcnt=stack->el_count;
work.task_nr=curcnt; //weil ich bei 0 anfange zu zählen und push_stack gleich el_count incrementiert
double integ_bnd[2];
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.task_nr=work.task_nr;
winner.subtask_nr=0;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
push_stack(stack, &work);
// *************************
// * 2nd subinterval
// **************************
work.a=mll;
work.b=ml;
work.fa=fmll;
work.fb=fml;
work.task_nr++;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
// #pragma omp critical (stack)
// {
push_stack(stack, &work);
// }
// *************************
// * 3rd subinterval
// **************************
work.a=ml;
work.b=m;
work.fa=fml;
work.fb=fm;
work.task_nr++;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2, integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
// #pragma omp critical (stack)
// {
push_stack(stack, &work);
// }
// *************************
// * 4th subinterval
// **************************
work.a=m;
work.b=mr;
work.fa=fm;
work.fb=fmr;
work.task_nr++;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
// #pragma omp critical (stack)
{
push_stack(stack, &work);
}
// *************************
// * 5th subinterval
// **************************
work.a=mr;
work.b=mrr;
work.fa=fmr;
work.fb=fmrr;
work.task_nr++;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
// #pragma omp critical (stack)
{
push_stack(stack, &work);
}
// *************************
// * 6th subinterval
// **************************
work.a=mrr;
work.b=b;
work.fa=fmrr;
work.fb=fb;
work.task_nr++;
m_inner= (work.a+work.b)/2;
h_inner= (work.a+work.b)/2;
mll_inner=(m_inner-alpha*h_inner);
ml_inner= (m_inner-beta*h_inner);
mr_inner= (m_inner+beta*h_inner);
mrr_inner=(m_inner+alpha*h_inner);
//1st subtask
get_integ_bounds_inner(integ_bnd, mll_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=0;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//2nd subtask
get_integ_bounds_inner(integ_bnd, ml_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=1;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//3rd subtask
get_integ_bounds_inner(integ_bnd, m_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=2;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//4th subtask
get_integ_bounds_inner(integ_bnd, mr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=3;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
//5th subtask
get_integ_bounds_inner(integ_bnd, mrr_inner, work.p);
winner=gkq_create_inner_task(inner_integrand2,integ_bnd[0],integ_bnd[1],1e-3, work.p);
winner.subtask_nr=4;
winner.task_nr=work.task_nr;
push_stack(work.stack_inner,&winner);
// #pragma omp critical (stack)
{
push_stack(stack, &work);
}
printf("myid:%d,iter:%d, 6 subinvtervals with 5 subtasks each pushed to stack outer\n",myid,iter);
} // pragma critical
} // else ..non-acceptable error
}
} // while
} /* end omp parallel */
return integral_result;
}
//SIMPSON
// *****************************************************************************************************************************************
// double integral2(double (*f)(double, struct my_f_params*), stack_t stack)
// {
// work_t work;
// int ready, idle, busy;
// double integral_result = 0.0;
// busy = 0;
// #pragma omp parallel default(none) \
// shared(stack, integral_result,f,busy,simpson_error) \
// private(work, idle, ready)
// {
// // printf("me:%d, err:%d\n",omp_get_thread_num(),simpson_error);
// ready = 0;
// idle = 1;
// while(!ready && !simpson_error) //<-- so NICHT!
// {
// #pragma omp critical (stack)
// {
// if (!empty_stack(stack))
// {
// /* we have new work */
// pop_stack(stack, &work);
// if (idle)
// {
// /* say others i'm busy */
// busy += 1;
// idle = 0;
// }
// }
// else{
// /* no new work on stack */
// if (!idle){
// busy -= 1;
// idle = 1;
// }
// /* nobody has anything to do; let us leave the loop */
// if (busy == 0)
// {
// ready = 1;
// }
// }
// } /* end critical(stack) */
// if (idle)
// continue; //if ready==1 --> leave loop
// double b = work.b;
// double a = work.a;
// double tol = work.tol;
// double S=work.S; // previous TOTAL integral
// double fa=work.fa;
// double fb=work.fb;
// double fm=work.fm;
// int rec=work.rec;
// double h = (b - a)/2;
// double mid = (a+b)/2;
// double lm=(a+mid)/2;
// double rm=(mid+b)/2;
// double flm=f(lm,work.p);
// double frm=f(rm,work.p);
// double Sl=h/6*(fa+4*flm+fm);
// double Sr=h/6*(fm+4*frm+fb);
// double delta=Sl+Sr-S;
// // serious numerical trouble: it won't converge
// if ((tol/2 == tol) || fabs(a-lm) <= tol) // || tol < tolmax)
// {
// simpson_error = 1;
// #pragma omp critical (integral_result)
// integral_result = S;
// }
// //need something against spurious convergence
// //for now: iter > 5 (c.f. numerical recipes)
// if( work.iter > 5 && (rec <= 0 || fabs(delta) <= 15*tol)) //error acceptable
// {
// #pragma omp critical (integral_result)
// integral_result += Sl+Sr+delta/15;
// }
// else // error not acceptable
// {
// //push new subintervals to stack
// work.a = a;
// work.b = mid;
// work.tol = tol/2;
// work.S = Sl;
// work.fa=fa;
// work.fb=fm;
// work.fm=flm;
// work.rec=rec-1;
// work.iter=work.iter+1;
// #pragma omp critical (stack)
// {
// //LEFT
// push_stack(stack, &work);
// //prepare RIGHT side and push to stack
// work.a = mid;
// work.b = b;
// work.tol = tol/2;
// work.S=Sr;
// work.fa=fm;
// work.fb=fb;
// work.fm=frm;
// work.rec=rec-1;
// push_stack(stack, &work);
// }
// }
// } /* while */
// } /* end omp parallel */
// return integral_result;
// } |
GB_binop__eq_fc32.c |
//------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__eq_fc32)
// A.*B function (eWiseMult): GB (_AemultB_08__eq_fc32)
// A.*B function (eWiseMult): GB (_AemultB_02__eq_fc32)
// A.*B function (eWiseMult): GB (_AemultB_04__eq_fc32)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__eq_fc32)
// A*D function (colscale): GB ((none))
// D*A function (rowscale): GB ((none))
// C+=B function (dense accum): GB (_Cdense_accumB__eq_fc32)
// C+=b function (dense accum): GB (_Cdense_accumb__eq_fc32)
// C+=A+B function (dense ewise3): GB ((none))
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__eq_fc32)
// C=scalar+B GB (_bind1st__eq_fc32)
// C=scalar+B' GB (_bind1st_tran__eq_fc32)
// C=A+scalar GB (_bind2nd__eq_fc32)
// C=A'+scalar GB (_bind2nd_tran__eq_fc32)
// C type: bool
// A type: GxB_FC32_t
// A pattern? 0
// B type: GxB_FC32_t
// B pattern? 0
// BinaryOp: cij = GB_FC32_eq (aij, bij)
#define GB_ATYPE \
GxB_FC32_t
#define GB_BTYPE \
GxB_FC32_t
#define GB_CTYPE \
bool
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
0
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
0
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
GxB_FC32_t aij = GBX (Ax, pA, A_iso)
// true if values of A are not used
#define GB_A_IS_PATTERN \
0 \
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
GxB_FC32_t bij = GBX (Bx, pB, B_iso)
// true if values of B are not used
#define GB_B_IS_PATTERN \
0 \
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
bool t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = (crealf (GBX (Ax, pA, A_iso)) != 0) || (cimagf (GBX (Ax, pA, A_iso)) != 0)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = (crealf (GBX (Bx, pB, B_iso)) != 0) || (cimagf (GBX (Bx, pB, B_iso)) != 0)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = GB_FC32_eq (x, y) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_EQ || GxB_NO_FC32 || GxB_NO_EQ_FC32)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
void GB (_Cdense_ewise3_noaccum__eq_fc32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_noaccum_template.c"
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__eq_fc32)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if 0
{
#include "GB_dense_subassign_23_template.c"
}
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__eq_fc32)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if 0
{
// get the scalar b for C += b, of type GxB_FC32_t
GxB_FC32_t bwork = (*((GxB_FC32_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix D,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *restrict Cx = (bool *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info GB ((none))
(
GrB_Matrix C,
const GrB_Matrix D,
const GrB_Matrix B,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *restrict Cx = (bool *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__eq_fc32)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool is_eWiseUnion,
const GB_void *alpha_scalar_in,
const GB_void *beta_scalar_in,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
GxB_FC32_t alpha_scalar ;
GxB_FC32_t beta_scalar ;
if (is_eWiseUnion)
{
alpha_scalar = (*((GxB_FC32_t *) alpha_scalar_in)) ;
beta_scalar = (*((GxB_FC32_t *) beta_scalar_in )) ;
}
#include "GB_add_template.c"
GB_FREE_WORKSPACE ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__eq_fc32)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__eq_fc32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__eq_fc32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__eq_fc32)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__eq_fc32)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
bool *Cx = (bool *) Cx_output ;
GxB_FC32_t x = (*((GxB_FC32_t *) x_input)) ;
GxB_FC32_t *Bx = (GxB_FC32_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
GxB_FC32_t bij = GBX (Bx, p, false) ;
Cx [p] = GB_FC32_eq (x, bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__eq_fc32)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
bool *Cx = (bool *) Cx_output ;
GxB_FC32_t *Ax = (GxB_FC32_t *) Ax_input ;
GxB_FC32_t y = (*((GxB_FC32_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
GxB_FC32_t aij = GBX (Ax, p, false) ;
Cx [p] = GB_FC32_eq (aij, y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
GxB_FC32_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = GB_FC32_eq (x, aij) ; \
}
GrB_Info GB (_bind1st_tran__eq_fc32)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
GxB_FC32_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GxB_FC32_t x = (*((const GxB_FC32_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
GxB_FC32_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
GxB_FC32_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = GB_FC32_eq (aij, y) ; \
}
GrB_Info GB (_bind2nd_tran__eq_fc32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GxB_FC32_t y = (*((const GxB_FC32_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
9.c | #include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <glib.h>
#include <omp.h>
// libgsl0-dev
#include <gsl/gsl_math.h>
#include <gsl/gsl_rng.h>
#include <sys/time.h>
#include <gsl/gsl_rng.h>
#include <gsl/gsl_sort_double.h>
double pi(int n, int batch) {
const gsl_rng_type * T;
gsl_rng * r;
gsl_rng_env_setup();
T = gsl_rng_default;
r = gsl_rng_alloc (T);
int end = (n / batch);
double sum = 0;
int i, j;
// GRand * grand = g_rand_new ();
//g_rand_double(grand);
#pragma omp parallel for default(shared) firstprivate(r) private(j) reduction (+:sum)
for (i = 0; i < end; i++) {
for (j=0; j<batch; j++) {
double a = gsl_rng_uniform_pos (r);
double b = gsl_rng_uniform_pos (r);
double c = (a * a) + (b * b);
if (c <= 1) {
sum += c;
}
}
}
return 8 * sum / n;
}
int main () {
printf("%0.15f\n", pi(100000000, 1000));
}
|
dcraw.c | #ifndef IGNOREALL
/*
dcraw.c -- Dave Coffin's raw photo decoder
Copyright 1997-2015 by Dave Coffin, dcoffin a cybercom o net
This is a command-line ANSI C program to convert raw photos from
any digital camera on any computer running any operating system.
No license is required to download and use dcraw.c. However,
to lawfully redistribute dcraw, you must either (a) offer, at
no extra charge, full source code* for all executable files
containing RESTRICTED functions, (b) distribute this code under
the GPL Version 2 or later, (c) remove all RESTRICTED functions,
re-implement them, or copy them from an earlier, unrestricted
Revision of dcraw.c, or (d) purchase a license from the author.
The functions that process Foveon images have been RESTRICTED
since Revision 1.237. All other code remains free for all uses.
*If you have not modified dcraw.c in any way, a link to my
homepage qualifies as "full source code".
$Revision: 1.47 $
$Date: 2015/11/29 14:49:27 $
*/
/*@out DEFINES
#ifndef USE_JPEG
#define NO_JPEG
#endif
#ifndef USE_JASPER
#define NO_JASPER
#endif
@end DEFINES */
#define NO_LCMS
#define DCRAW_VERBOSE
//@out DEFINES
#define DCRAW_VERSION "9.26"
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#define _USE_MATH_DEFINES
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <float.h>
#include <limits.h>
#include <math.h>
#include <setjmp.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <sys/types.h>
//@end DEFINES
#if defined(DJGPP) || defined(__MINGW32__)
#define fseeko fseek
#define ftello ftell
#else
#define fgetc getc_unlocked
#endif
//@out DEFINES
#ifdef __CYGWIN__
#include <io.h>
#endif
#ifdef WIN32
#include <sys/utime.h>
#include <winsock2.h>
#pragma comment(lib, "ws2_32.lib")
/* Visual Studio 2015 / VC 14 / MSVC 19.00 finally has snprintf() */
#if defined(_MSC_VER) && (_MSC_VER < 1900)
#define snprintf _snprintf
#endif // _MSC_VER
#define strcasecmp stricmp
#define strncasecmp strnicmp
//@end DEFINES
typedef __int64 INT64;
typedef unsigned __int64 UINT64;
//@out DEFINES
#else
#include <unistd.h>
#include <utime.h>
#include <netinet/in.h>
typedef long long INT64;
typedef unsigned long long UINT64;
#endif
#ifdef NODEPS
#define NO_JASPER
#define NO_JPEG
#define NO_LCMS
#endif
#ifndef NO_JASPER
#include <jasper/jasper.h> /* Decode Red camera movies */
#endif
#ifndef NO_JPEG
#include <jpeglib.h> /* Decode compressed Kodak DC120 photos */
#endif /* and Adobe Lossy DNGs */
#ifndef NO_LCMS
#ifdef USE_LCMS
#include <lcms.h> /* Support color profiles */
#else
#include <lcms2.h> /* Support color profiles */
#endif
#endif
#ifdef LOCALEDIR
#include <libintl.h>
#define _(String) gettext(String)
#else
#define _(String) (String)
#endif
#ifdef LJPEG_DECODE
#error Please compile dcraw.c by itself.
#error Do not link it with ljpeg_decode.
#endif
#ifndef LONG_BIT
#define LONG_BIT (8 * sizeof (long))
#endif
//@end DEFINES
#if !defined(uchar)
#define uchar unsigned char
#endif
#if !defined(ushort)
#define ushort unsigned short
#endif
/*
All global variables are defined here, and all functions that
access them are prefixed with "CLASS". Note that a thread-safe
C++ class cannot have non-const static local variables.
*/
FILE *ifp, *ofp;
short order;
const char *ifname;
char *meta_data, xtrans[6][6], xtrans_abs[6][6];
char cdesc[5], desc[512], make[64], model[64], model2[64], artist[64],software[64];
float flash_used, canon_ev, iso_speed, shutter, aperture, focal_len;
time_t timestamp;
off_t strip_offset, data_offset;
off_t thumb_offset, meta_offset, profile_offset;
unsigned shot_order, kodak_cbpp, exif_cfa, unique_id;
unsigned thumb_length, meta_length, profile_length;
unsigned thumb_misc, *oprof, fuji_layout, shot_select=0, multi_out=0;
unsigned tiff_nifds, tiff_samples, tiff_bps, tiff_compress;
unsigned black, maximum, mix_green, raw_color, zero_is_bad;
unsigned zero_after_ff, is_raw, dng_version, is_foveon, data_error;
unsigned tile_width, tile_length, gpsdata[32], load_flags;
unsigned flip, tiff_flip, filters, colors;
ushort raw_height, raw_width, height, width, top_margin, left_margin;
ushort shrink, iheight, iwidth, fuji_width, thumb_width, thumb_height;
ushort *raw_image, (*image)[4], cblack[4102];
ushort white[8][8], curve[0x10000], cr2_slice[3], sraw_mul[4];
double pixel_aspect, aber[4]={1,1,1,1}, gamm[6]={ 0.45,4.5,0,0,0,0 };
float bright=1, user_mul[4]={0,0,0,0}, threshold=0;
int mask[8][4];
int half_size=0, four_color_rgb=0, document_mode=0, highlight=0;
int verbose=0, use_auto_wb=0, use_camera_wb=0, use_camera_matrix=1;
int output_color=1, output_bps=8, output_tiff=0, med_passes=0;
int no_auto_bright=0;
unsigned greybox[4] = { 0, 0, UINT_MAX, UINT_MAX };
float cam_mul[4], pre_mul[4], cmatrix[3][4], rgb_cam[3][4];
const double xyz_rgb[3][3] = { /* XYZ from RGB */
{ 0.412453, 0.357580, 0.180423 },
{ 0.212671, 0.715160, 0.072169 },
{ 0.019334, 0.119193, 0.950227 } };
const float d65_white[3] = { 0.950456, 1, 1.088754 };
int histogram[4][0x2000];
void (*write_thumb)(), (*write_fun)();
void (*load_raw)(), (*thumb_load_raw)();
jmp_buf failure;
struct decode {
struct decode *branch[2];
int leaf;
} first_decode[2048], *second_decode, *free_decode;
struct tiff_ifd {
int t_width, t_height, bps, comp, phint, offset, t_flip, samples, bytes;
int t_tile_width, t_tile_length,sample_format,predictor;
float t_shutter;
} tiff_ifd[10];
struct ph1 {
int format, key_off, tag_21a;
int t_black, split_col, black_col, split_row, black_row;
float tag_210;
} ph1;
#define CLASS
//@out DEFINES
#define FORC(cnt) for (c=0; c < cnt; c++)
#define FORC3 FORC(3)
#define FORC4 FORC(4)
#define FORCC FORC(colors)
#define SQR(x) ((x)*(x))
#define ABS(x) (((int)(x) ^ ((int)(x) >> 31)) - ((int)(x) >> 31))
#define MIN(a,b) ((a) < (b) ? (a) : (b))
#define MAX(a,b) ((a) > (b) ? (a) : (b))
#define LIM(x,min,max) MAX(min,MIN(x,max))
#define ULIM(x,y,z) ((y) < (z) ? LIM(x,y,z) : LIM(x,z,y))
#define CLIP(x) LIM((int)(x),0,65535)
#define SWAP(a,b) { a=a+b; b=a-b; a=a-b; }
#define my_swap(type, i, j) {type t = i; i = j; j = t;}
/*
In order to inline this calculation, I make the risky
assumption that all filter patterns can be described
by a repeating pattern of eight rows and two columns
Do not use the FC or BAYER macros with the Leaf CatchLight,
because its pattern is 16x16, not 2x8.
Return values are either 0/1/2/3 = G/M/C/Y or 0/1/2/3 = R/G1/B/G2
PowerShot 600 PowerShot A50 PowerShot Pro70 Pro90 & G1
0xe1e4e1e4: 0x1b4e4b1e: 0x1e4b4e1b: 0xb4b4b4b4:
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
0 G M G M G M 0 C Y C Y C Y 0 Y C Y C Y C 0 G M G M G M
1 C Y C Y C Y 1 M G M G M G 1 M G M G M G 1 Y C Y C Y C
2 M G M G M G 2 Y C Y C Y C 2 C Y C Y C Y
3 C Y C Y C Y 3 G M G M G M 3 G M G M G M
4 C Y C Y C Y 4 Y C Y C Y C
PowerShot A5 5 G M G M G M 5 G M G M G M
0x1e4e1e4e: 6 Y C Y C Y C 6 C Y C Y C Y
7 M G M G M G 7 M G M G M G
0 1 2 3 4 5
0 C Y C Y C Y
1 G M G M G M
2 C Y C Y C Y
3 M G M G M G
All RGB cameras use one of these Bayer grids:
0x16161616: 0x61616161: 0x49494949: 0x94949494:
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
0 B G B G B G 0 G R G R G R 0 G B G B G B 0 R G R G R G
1 G R G R G R 1 B G B G B G 1 R G R G R G 1 G B G B G B
2 B G B G B G 2 G R G R G R 2 G B G B G B 2 R G R G R G
3 G R G R G R 3 B G B G B G 3 R G R G R G 3 G B G B G B
*/
#define RAW(row,col) \
raw_image[(row)*raw_width+(col)]
//@end DEFINES
#define FC(row,col) \
(filters >> ((((row) << 1 & 14) + ((col) & 1)) << 1) & 3)
//@out DEFINES
#define BAYER(row,col) \
image[((row) >> shrink)*iwidth + ((col) >> shrink)][FC(row,col)]
#define BAYER2(row,col) \
image[((row) >> shrink)*iwidth + ((col) >> shrink)][fcol(row,col)]
//@end DEFINES
/* @out COMMON
#include <math.h>
#define CLASS LibRaw::
#include "libraw/libraw_types.h"
#define LIBRAW_LIBRARY_BUILD
#define LIBRAW_IO_REDEFINED
#include "libraw/libraw.h"
#include "internal/defines.h"
#include "internal/var_defines.h"
@end COMMON */
//@out COMMON
int CLASS fcol (int row, int col)
{
static const char filter[16][16] =
{ { 2,1,1,3,2,3,2,0,3,2,3,0,1,2,1,0 },
{ 0,3,0,2,0,1,3,1,0,1,1,2,0,3,3,2 },
{ 2,3,3,2,3,1,1,3,3,1,2,1,2,0,0,3 },
{ 0,1,0,1,0,2,0,2,2,0,3,0,1,3,2,1 },
{ 3,1,1,2,0,1,0,2,1,3,1,3,0,1,3,0 },
{ 2,0,0,3,3,2,3,1,2,0,2,0,3,2,2,1 },
{ 2,3,3,1,2,1,2,1,2,1,1,2,3,0,0,1 },
{ 1,0,0,2,3,0,0,3,0,3,0,3,2,1,2,3 },
{ 2,3,3,1,1,2,1,0,3,2,3,0,2,3,1,3 },
{ 1,0,2,0,3,0,3,2,0,1,1,2,0,1,0,2 },
{ 0,1,1,3,3,2,2,1,1,3,3,0,2,1,3,2 },
{ 2,3,2,0,0,1,3,0,2,0,1,2,3,0,1,0 },
{ 1,3,1,2,3,2,3,2,0,2,0,1,1,0,3,0 },
{ 0,2,0,3,1,0,0,1,1,3,3,2,3,2,2,1 },
{ 2,1,3,2,3,1,2,1,0,3,0,2,0,2,0,2 },
{ 0,3,1,0,0,2,0,3,2,1,3,1,1,3,1,3 } };
if (filters == 1) return filter[(row+top_margin)&15][(col+left_margin)&15];
if (filters == 9) return xtrans[(row+6) % 6][(col+6) % 6];
return FC(row,col);
}
size_t strnlen(const char *s, size_t n)
{
const char *p = (const char *)memchr(s, 0, n);
return(p ? p-s : n);
}
#ifndef __GLIBC__
char *my_memmem (char *haystack, size_t haystacklen,
char *needle, size_t needlelen)
{
char *c;
for (c = haystack; c <= haystack + haystacklen - needlelen; c++)
if (!memcmp (c, needle, needlelen))
return c;
return 0;
}
#define memmem my_memmem
char *my_strcasestr (char *haystack, const char *needle)
{
char *c;
for (c = haystack; *c; c++)
if (!strncasecmp(c, needle, strlen(needle)))
return c;
return 0;
}
#define strcasestr my_strcasestr
#endif
int my_strlen(const char *str)
{
return (int)strnlen(str,0x7fffffff);
}
#define strlen(a) my_strlen((a))
//@end COMMON
void CLASS merror (void *ptr, const char *where)
{
if (ptr) return;
fprintf (stderr,_("%s: Out of memory in %s\n"), ifname, where);
longjmp (failure, 1);
}
void CLASS derror()
{
if (!data_error) {
fprintf (stderr, "%s: ", ifname);
if (feof(ifp))
fprintf (stderr,_("Unexpected end of file\n"));
else
fprintf (stderr,_("Corrupt data near 0x%llx\n"), (INT64) ftello(ifp));
}
data_error++;
}
//@out COMMON
ushort CLASS sget2 (uchar *s)
{
if (order == 0x4949) /* "II" means little-endian */
return s[0] | s[1] << 8;
else /* "MM" means big-endian */
return s[0] << 8 | s[1];
}
// DNG was written by:
#define CameraDNG 1
#define AdobeDNG 2
#ifdef LIBRAW_LIBRARY_BUILD
static ushort saneSonyCameraInfo(uchar a, uchar b, uchar c, uchar d, uchar e, uchar f){
if ((a >> 4) > 9) return 0;
else if ((a & 0x0f) > 9) return 0;
else if ((b >> 4) > 9) return 0;
else if ((b & 0x0f) > 9) return 0;
else if ((c >> 4) > 9) return 0;
else if ((c & 0x0f) > 9) return 0;
else if ((d >> 4) > 9) return 0;
else if ((d & 0x0f) > 9) return 0;
else if ((e >> 4) > 9) return 0;
else if ((e & 0x0f) > 9) return 0;
else if ((f >> 4) > 9) return 0;
else if ((f & 0x0f) > 9) return 0;
return 1;
}
static ushort bcd2dec(uchar data){
if ((data >> 4) > 9) return 0;
else if ((data & 0x0f) > 9) return 0;
else return (data >> 4) * 10 + (data & 0x0f);
}
static uchar SonySubstitution[257] = "\x00\x01\x32\xb1\x0a\x0e\x87\x28\x02\xcc\xca\xad\x1b\xdc\x08\xed\x64\x86\xf0\x4f\x8c\x6c\xb8\xcb\x69\xc4\x2c\x03\x97\xb6\x93\x7c\x14\xf3\xe2\x3e\x30\x8e\xd7\x60\x1c\xa1\xab\x37\xec\x75\xbe\x23\x15\x6a\x59\x3f\xd0\xb9\x96\xb5\x50\x27\x88\xe3\x81\x94\xe0\xc0\x04\x5c\xc6\xe8\x5f\x4b\x70\x38\x9f\x82\x80\x51\x2b\xc5\x45\x49\x9b\x21\x52\x53\x54\x85\x0b\x5d\x61\xda\x7b\x55\x26\x24\x07\x6e\x36\x5b\x47\xb7\xd9\x4a\xa2\xdf\xbf\x12\x25\xbc\x1e\x7f\x56\xea\x10\xe6\xcf\x67\x4d\x3c\x91\x83\xe1\x31\xb3\x6f\xf4\x05\x8a\x46\xc8\x18\x76\x68\xbd\xac\x92\x2a\x13\xe9\x0f\xa3\x7a\xdb\x3d\xd4\xe7\x3a\x1a\x57\xaf\x20\x42\xb2\x9e\xc3\x8b\xf2\xd5\xd3\xa4\x7e\x1f\x98\x9c\xee\x74\xa5\xa6\xa7\xd8\x5e\xb0\xb4\x34\xce\xa8\x79\x77\x5a\xc1\x89\xae\x9a\x11\x33\x9d\xf5\x39\x19\x65\x78\x16\x71\xd2\xa9\x44\x63\x40\x29\xba\xa0\x8f\xe4\xd6\x3b\x84\x0d\xc2\x4e\x58\xdd\x99\x22\x6b\xc9\xbb\x17\x06\xe5\x7d\x66\x43\x62\xf6\xcd\x35\x90\x2e\x41\x8d\x6d\xaa\x09\x73\x95\x0c\xf1\x1d\xde\x4c\x2f\x2d\xf7\xd1\x72\xeb\xef\x48\xc7\xf8\xf9\xfa\xfb\xfc\xfd\xfe\xff";
ushort CLASS sget2Rev(uchar *s) // specific to some Canon Makernotes fields, where they have endian in reverse
{
if (order == 0x4d4d) /* "II" means little-endian, and we reverse to "MM" - big endian */
return s[0] | s[1] << 8;
else /* "MM" means big-endian... */
return s[0] << 8 | s[1];
}
#endif
ushort CLASS get2()
{
uchar str[2] = { 0xff,0xff };
fread (str, 1, 2, ifp);
return sget2(str);
}
unsigned CLASS sget4 (uchar *s)
{
if (order == 0x4949)
return s[0] | s[1] << 8 | s[2] << 16 | s[3] << 24;
else
return s[0] << 24 | s[1] << 16 | s[2] << 8 | s[3];
}
#define sget4(s) sget4((uchar *)s)
unsigned CLASS get4()
{
uchar str[4] = { 0xff,0xff,0xff,0xff };
fread (str, 1, 4, ifp);
return sget4(str);
}
unsigned CLASS getint (int type)
{
return type == 3 ? get2() : get4();
}
float CLASS int_to_float (int i)
{
union { int i; float f; } u;
u.i = i;
return u.f;
}
double CLASS getreal (int type)
{
union { char c[8]; double d; } u,v;
int i, rev;
switch (type) {
case 3: return (unsigned short) get2();
case 4: return (unsigned int) get4();
case 5:
u.d = (unsigned int) get4();
v.d = (unsigned int)get4();
return u.d / (v.d ? v.d : 1);
case 8: return (signed short) get2();
case 9: return (signed int) get4();
case 10:
u.d = (signed int) get4();
v.d = (signed int)get4();
return u.d / (v.d?v.d:1);
case 11: return int_to_float (get4());
case 12:
rev = 7 * ((order == 0x4949) == (ntohs(0x1234) == 0x1234));
for (i=0; i < 8; i++)
u.c[i ^ rev] = fgetc(ifp);
return u.d;
default: return fgetc(ifp);
}
}
void CLASS read_shorts (ushort *pixel, int count)
{
if (fread (pixel, 2, count, ifp) < count) derror();
if ((order == 0x4949) == (ntohs(0x1234) == 0x1234))
swab ((char*)pixel, (char*)pixel, count*2);
}
void CLASS cubic_spline (const int *x_, const int *y_, const int len)
{
float **A, *b, *c, *d, *x, *y;
int i, j;
A = (float **) calloc (((2*len + 4)*sizeof **A + sizeof *A), 2*len);
if (!A) return;
A[0] = (float *) (A + 2*len);
for (i = 1; i < 2*len; i++)
A[i] = A[0] + 2*len*i;
y = len + (x = i + (d = i + (c = i + (b = A[0] + i*i))));
for (i = 0; i < len; i++) {
x[i] = x_[i] / 65535.0;
y[i] = y_[i] / 65535.0;
}
for (i = len-1; i > 0; i--) {
b[i] = (y[i] - y[i-1]) / (x[i] - x[i-1]);
d[i-1] = x[i] - x[i-1];
}
for (i = 1; i < len-1; i++) {
A[i][i] = 2 * (d[i-1] + d[i]);
if (i > 1) {
A[i][i-1] = d[i-1];
A[i-1][i] = d[i-1];
}
A[i][len-1] = 6 * (b[i+1] - b[i]);
}
for(i = 1; i < len-2; i++) {
float v = A[i+1][i] / A[i][i];
for(j = 1; j <= len-1; j++)
A[i+1][j] -= v * A[i][j];
}
for(i = len-2; i > 0; i--) {
float acc = 0;
for(j = i; j <= len-2; j++)
acc += A[i][j]*c[j];
c[i] = (A[i][len-1] - acc) / A[i][i];
}
for (i = 0; i < 0x10000; i++) {
float x_out = (float)(i / 65535.0);
float y_out = 0;
for (j = 0; j < len-1; j++) {
if (x[j] <= x_out && x_out <= x[j+1]) {
float v = x_out - x[j];
y_out = y[j] +
((y[j+1] - y[j]) / d[j] - (2 * d[j] * c[j] + c[j+1] * d[j])/6) * v
+ (c[j] * 0.5) * v*v + ((c[j+1] - c[j]) / (6 * d[j])) * v*v*v;
}
}
curve[i] = y_out < 0.0 ? 0 : (y_out >= 1.0 ? 65535 :
(ushort)(y_out * 65535.0 + 0.5));
}
free (A);
}
void CLASS canon_600_fixed_wb (int temp)
{
static const short mul[4][5] = {
{ 667, 358,397,565,452 },
{ 731, 390,367,499,517 },
{ 1119, 396,348,448,537 },
{ 1399, 485,431,508,688 } };
int lo, hi, i;
float frac=0;
for (lo=4; --lo; )
if (*mul[lo] <= temp) break;
for (hi=0; hi < 3; hi++)
if (*mul[hi] >= temp) break;
if (lo != hi)
frac = (float) (temp - *mul[lo]) / (*mul[hi] - *mul[lo]);
for (i=1; i < 5; i++)
pre_mul[i-1] = 1 / (frac * mul[hi][i] + (1-frac) * mul[lo][i]);
}
/* Return values: 0 = white 1 = near white 2 = not white */
int CLASS canon_600_color (int ratio[2], int mar)
{
int clipped=0, target, miss;
if (flash_used) {
if (ratio[1] < -104)
{ ratio[1] = -104; clipped = 1; }
if (ratio[1] > 12)
{ ratio[1] = 12; clipped = 1; }
} else {
if (ratio[1] < -264 || ratio[1] > 461) return 2;
if (ratio[1] < -50)
{ ratio[1] = -50; clipped = 1; }
if (ratio[1] > 307)
{ ratio[1] = 307; clipped = 1; }
}
target = flash_used || ratio[1] < 197
? -38 - (398 * ratio[1] >> 10)
: -123 + (48 * ratio[1] >> 10);
if (target - mar <= ratio[0] &&
target + 20 >= ratio[0] && !clipped) return 0;
miss = target - ratio[0];
if (abs(miss) >= mar*4) return 2;
if (miss < -20) miss = -20;
if (miss > mar) miss = mar;
ratio[0] = target - miss;
return 1;
}
void CLASS canon_600_auto_wb()
{
int mar, row, col, i, j, st, count[] = { 0,0 };
int test[8], total[2][8], ratio[2][2], stat[2];
memset (&total, 0, sizeof total);
i = canon_ev + 0.5;
if (i < 10) mar = 150;
else if (i > 12) mar = 20;
else mar = 280 - 20 * i;
if (flash_used) mar = 80;
for (row=14; row < height-14; row+=4)
for (col=10; col < width; col+=2) {
for (i=0; i < 8; i++)
test[(i & 4) + FC(row+(i >> 1),col+(i & 1))] =
BAYER(row+(i >> 1),col+(i & 1));
for (i=0; i < 8; i++)
if (test[i] < 150 || test[i] > 1500) goto next;
for (i=0; i < 4; i++)
if (abs(test[i] - test[i+4]) > 50) goto next;
for (i=0; i < 2; i++) {
for (j=0; j < 4; j+=2)
ratio[i][j >> 1] = ((test[i*4+j+1]-test[i*4+j]) << 10) / test[i*4+j];
stat[i] = canon_600_color (ratio[i], mar);
}
if ((st = stat[0] | stat[1]) > 1) goto next;
for (i=0; i < 2; i++)
if (stat[i])
for (j=0; j < 2; j++)
test[i*4+j*2+1] = test[i*4+j*2] * (0x400 + ratio[i][j]) >> 10;
for (i=0; i < 8; i++)
total[st][i] += test[i];
count[st]++;
next: ;
}
if (count[0] | count[1]) {
st = count[0]*200 < count[1];
for (i=0; i < 4; i++)
pre_mul[i] = 1.0 / (total[st][i] + total[st][i+4]);
}
}
void CLASS canon_600_coeff()
{
static const short table[6][12] = {
{ -190,702,-1878,2390, 1861,-1349,905,-393, -432,944,2617,-2105 },
{ -1203,1715,-1136,1648, 1388,-876,267,245, -1641,2153,3921,-3409 },
{ -615,1127,-1563,2075, 1437,-925,509,3, -756,1268,2519,-2007 },
{ -190,702,-1886,2398, 2153,-1641,763,-251, -452,964,3040,-2528 },
{ -190,702,-1878,2390, 1861,-1349,905,-393, -432,944,2617,-2105 },
{ -807,1319,-1785,2297, 1388,-876,769,-257, -230,742,2067,-1555 } };
int t=0, i, c;
float mc, yc;
mc = pre_mul[1] / pre_mul[2];
yc = pre_mul[3] / pre_mul[2];
if (mc > 1 && mc <= 1.28 && yc < 0.8789) t=1;
if (mc > 1.28 && mc <= 2) {
if (yc < 0.8789) t=3;
else if (yc <= 2) t=4;
}
if (flash_used) t=5;
for (raw_color = i=0; i < 3; i++)
FORCC rgb_cam[i][c] = table[t][i*4 + c] / 1024.0;
}
void CLASS canon_600_load_raw()
{
uchar data[1120], *dp;
ushort *pix;
int irow, row;
for (irow=row=0; irow < height; irow++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (data, 1, 1120, ifp) < 1120) derror();
pix = raw_image + row*raw_width;
for (dp=data; dp < data+1120; dp+=10, pix+=8) {
pix[0] = (dp[0] << 2) + (dp[1] >> 6 );
pix[1] = (dp[2] << 2) + (dp[1] >> 4 & 3);
pix[2] = (dp[3] << 2) + (dp[1] >> 2 & 3);
pix[3] = (dp[4] << 2) + (dp[1] & 3);
pix[4] = (dp[5] << 2) + (dp[9] & 3);
pix[5] = (dp[6] << 2) + (dp[9] >> 2 & 3);
pix[6] = (dp[7] << 2) + (dp[9] >> 4 & 3);
pix[7] = (dp[8] << 2) + (dp[9] >> 6 );
}
if ((row+=2) > height) row = 1;
}
}
void CLASS canon_600_correct()
{
int row, col, val;
static const short mul[4][2] =
{ { 1141,1145 }, { 1128,1109 }, { 1178,1149 }, { 1128,1109 } };
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col++) {
if ((val = BAYER(row,col) - black) < 0) val = 0;
val = val * mul[row & 3][col & 1] >> 9;
BAYER(row,col) = val;
}
}
canon_600_fixed_wb(1311);
canon_600_auto_wb();
canon_600_coeff();
maximum = (0x3ff - black) * 1109 >> 9;
black = 0;
}
int CLASS canon_s2is()
{
unsigned row;
for (row=0; row < 100; row++) {
fseek (ifp, row*3340 + 3284, SEEK_SET);
if (getc(ifp) > 15) return 1;
}
return 0;
}
unsigned CLASS getbithuff (int nbits, ushort *huff)
{
#ifdef LIBRAW_NOTHREADS
static unsigned bitbuf=0;
static int vbits=0, reset=0;
#else
#define bitbuf tls->getbits.bitbuf
#define vbits tls->getbits.vbits
#define reset tls->getbits.reset
#endif
unsigned c;
if (nbits > 25) return 0;
if (nbits < 0)
return bitbuf = vbits = reset = 0;
if (nbits == 0 || vbits < 0) return 0;
while (!reset && vbits < nbits && (c = fgetc(ifp)) != EOF &&
!(reset = zero_after_ff && c == 0xff && fgetc(ifp))) {
bitbuf = (bitbuf << 8) + (uchar) c;
vbits += 8;
}
c = bitbuf << (32-vbits) >> (32-nbits);
if (huff) {
vbits -= huff[c] >> 8;
c = (uchar) huff[c];
} else
vbits -= nbits;
if (vbits < 0) derror();
return c;
#ifndef LIBRAW_NOTHREADS
#undef bitbuf
#undef vbits
#undef reset
#endif
}
#define getbits(n) getbithuff(n,0)
#define gethuff(h) getbithuff(*h,h+1)
/*
Construct a decode tree according the specification in *source.
The first 16 bytes specify how many codes should be 1-bit, 2-bit
3-bit, etc. Bytes after that are the leaf values.
For example, if the source is
{ 0,1,4,2,3,1,2,0,0,0,0,0,0,0,0,0,
0x04,0x03,0x05,0x06,0x02,0x07,0x01,0x08,0x09,0x00,0x0a,0x0b,0xff },
then the code is
00 0x04
010 0x03
011 0x05
100 0x06
101 0x02
1100 0x07
1101 0x01
11100 0x08
11101 0x09
11110 0x00
111110 0x0a
1111110 0x0b
1111111 0xff
*/
ushort * CLASS make_decoder_ref (const uchar **source)
{
int max, len, h, i, j;
const uchar *count;
ushort *huff;
count = (*source += 16) - 17;
for (max=16; max && !count[max]; max--);
huff = (ushort *) calloc (1 + (1 << max), sizeof *huff);
merror (huff, "make_decoder()");
huff[0] = max;
for (h=len=1; len <= max; len++)
for (i=0; i < count[len]; i++, ++*source)
for (j=0; j < 1 << (max-len); j++)
if (h <= 1 << max)
huff[h++] = len << 8 | **source;
return huff;
}
ushort * CLASS make_decoder (const uchar *source)
{
return make_decoder_ref (&source);
}
void CLASS crw_init_tables (unsigned table, ushort *huff[2])
{
static const uchar first_tree[3][29] = {
{ 0,1,4,2,3,1,2,0,0,0,0,0,0,0,0,0,
0x04,0x03,0x05,0x06,0x02,0x07,0x01,0x08,0x09,0x00,0x0a,0x0b,0xff },
{ 0,2,2,3,1,1,1,1,2,0,0,0,0,0,0,0,
0x03,0x02,0x04,0x01,0x05,0x00,0x06,0x07,0x09,0x08,0x0a,0x0b,0xff },
{ 0,0,6,3,1,1,2,0,0,0,0,0,0,0,0,0,
0x06,0x05,0x07,0x04,0x08,0x03,0x09,0x02,0x00,0x0a,0x01,0x0b,0xff },
};
static const uchar second_tree[3][180] = {
{ 0,2,2,2,1,4,2,1,2,5,1,1,0,0,0,139,
0x03,0x04,0x02,0x05,0x01,0x06,0x07,0x08,
0x12,0x13,0x11,0x14,0x09,0x15,0x22,0x00,0x21,0x16,0x0a,0xf0,
0x23,0x17,0x24,0x31,0x32,0x18,0x19,0x33,0x25,0x41,0x34,0x42,
0x35,0x51,0x36,0x37,0x38,0x29,0x79,0x26,0x1a,0x39,0x56,0x57,
0x28,0x27,0x52,0x55,0x58,0x43,0x76,0x59,0x77,0x54,0x61,0xf9,
0x71,0x78,0x75,0x96,0x97,0x49,0xb7,0x53,0xd7,0x74,0xb6,0x98,
0x47,0x48,0x95,0x69,0x99,0x91,0xfa,0xb8,0x68,0xb5,0xb9,0xd6,
0xf7,0xd8,0x67,0x46,0x45,0x94,0x89,0xf8,0x81,0xd5,0xf6,0xb4,
0x88,0xb1,0x2a,0x44,0x72,0xd9,0x87,0x66,0xd4,0xf5,0x3a,0xa7,
0x73,0xa9,0xa8,0x86,0x62,0xc7,0x65,0xc8,0xc9,0xa1,0xf4,0xd1,
0xe9,0x5a,0x92,0x85,0xa6,0xe7,0x93,0xe8,0xc1,0xc6,0x7a,0x64,
0xe1,0x4a,0x6a,0xe6,0xb3,0xf1,0xd3,0xa5,0x8a,0xb2,0x9a,0xba,
0x84,0xa4,0x63,0xe5,0xc5,0xf3,0xd2,0xc4,0x82,0xaa,0xda,0xe4,
0xf2,0xca,0x83,0xa3,0xa2,0xc3,0xea,0xc2,0xe2,0xe3,0xff,0xff },
{ 0,2,2,1,4,1,4,1,3,3,1,0,0,0,0,140,
0x02,0x03,0x01,0x04,0x05,0x12,0x11,0x06,
0x13,0x07,0x08,0x14,0x22,0x09,0x21,0x00,0x23,0x15,0x31,0x32,
0x0a,0x16,0xf0,0x24,0x33,0x41,0x42,0x19,0x17,0x25,0x18,0x51,
0x34,0x43,0x52,0x29,0x35,0x61,0x39,0x71,0x62,0x36,0x53,0x26,
0x38,0x1a,0x37,0x81,0x27,0x91,0x79,0x55,0x45,0x28,0x72,0x59,
0xa1,0xb1,0x44,0x69,0x54,0x58,0xd1,0xfa,0x57,0xe1,0xf1,0xb9,
0x49,0x47,0x63,0x6a,0xf9,0x56,0x46,0xa8,0x2a,0x4a,0x78,0x99,
0x3a,0x75,0x74,0x86,0x65,0xc1,0x76,0xb6,0x96,0xd6,0x89,0x85,
0xc9,0xf5,0x95,0xb4,0xc7,0xf7,0x8a,0x97,0xb8,0x73,0xb7,0xd8,
0xd9,0x87,0xa7,0x7a,0x48,0x82,0x84,0xea,0xf4,0xa6,0xc5,0x5a,
0x94,0xa4,0xc6,0x92,0xc3,0x68,0xb5,0xc8,0xe4,0xe5,0xe6,0xe9,
0xa2,0xa3,0xe3,0xc2,0x66,0x67,0x93,0xaa,0xd4,0xd5,0xe7,0xf8,
0x88,0x9a,0xd7,0x77,0xc4,0x64,0xe2,0x98,0xa5,0xca,0xda,0xe8,
0xf3,0xf6,0xa9,0xb2,0xb3,0xf2,0xd2,0x83,0xba,0xd3,0xff,0xff },
{ 0,0,6,2,1,3,3,2,5,1,2,2,8,10,0,117,
0x04,0x05,0x03,0x06,0x02,0x07,0x01,0x08,
0x09,0x12,0x13,0x14,0x11,0x15,0x0a,0x16,0x17,0xf0,0x00,0x22,
0x21,0x18,0x23,0x19,0x24,0x32,0x31,0x25,0x33,0x38,0x37,0x34,
0x35,0x36,0x39,0x79,0x57,0x58,0x59,0x28,0x56,0x78,0x27,0x41,
0x29,0x77,0x26,0x42,0x76,0x99,0x1a,0x55,0x98,0x97,0xf9,0x48,
0x54,0x96,0x89,0x47,0xb7,0x49,0xfa,0x75,0x68,0xb6,0x67,0x69,
0xb9,0xb8,0xd8,0x52,0xd7,0x88,0xb5,0x74,0x51,0x46,0xd9,0xf8,
0x3a,0xd6,0x87,0x45,0x7a,0x95,0xd5,0xf6,0x86,0xb4,0xa9,0x94,
0x53,0x2a,0xa8,0x43,0xf5,0xf7,0xd4,0x66,0xa7,0x5a,0x44,0x8a,
0xc9,0xe8,0xc8,0xe7,0x9a,0x6a,0x73,0x4a,0x61,0xc7,0xf4,0xc6,
0x65,0xe9,0x72,0xe6,0x71,0x91,0x93,0xa6,0xda,0x92,0x85,0x62,
0xf3,0xc5,0xb2,0xa4,0x84,0xba,0x64,0xa5,0xb3,0xd2,0x81,0xe5,
0xd3,0xaa,0xc4,0xca,0xf2,0xb1,0xe4,0xd1,0x83,0x63,0xea,0xc3,
0xe2,0x82,0xf1,0xa3,0xc2,0xa1,0xc1,0xe3,0xa2,0xe1,0xff,0xff }
};
if (table > 2) table = 2;
huff[0] = make_decoder ( first_tree[table]);
huff[1] = make_decoder (second_tree[table]);
}
/*
Return 0 if the image starts with compressed data,
1 if it starts with uncompressed low-order bits.
In Canon compressed data, 0xff is always followed by 0x00.
*/
int CLASS canon_has_lowbits()
{
uchar test[0x4000];
int ret=1, i;
fseek (ifp, 0, SEEK_SET);
fread (test, 1, sizeof test, ifp);
for (i=540; i < sizeof test - 1; i++)
if (test[i] == 0xff) {
if (test[i+1]) return 1;
ret=0;
}
return ret;
}
void CLASS canon_load_raw()
{
ushort *pixel, *prow, *huff[2];
int nblocks, lowbits, i, c, row, r, save, val;
int block, diffbuf[64], leaf, len, diff, carry=0, pnum=0, base[2];
crw_init_tables (tiff_compress, huff);
lowbits = canon_has_lowbits();
if (!lowbits) maximum = 0x3ff;
fseek (ifp, 540 + lowbits*raw_height*raw_width/4, SEEK_SET);
zero_after_ff = 1;
getbits(-1);
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row+=8) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
pixel = raw_image + row*raw_width;
nblocks = MIN (8, raw_height-row) * raw_width >> 6;
for (block=0; block < nblocks; block++) {
memset (diffbuf, 0, sizeof diffbuf);
for (i=0; i < 64; i++ ) {
leaf = gethuff(huff[i > 0]);
if (leaf == 0 && i) break;
if (leaf == 0xff) continue;
i += leaf >> 4;
len = leaf & 15;
if (len == 0) continue;
diff = getbits(len);
if ((diff & (1 << (len-1))) == 0)
diff -= (1 << len) - 1;
if (i < 64) diffbuf[i] = diff;
}
diffbuf[0] += carry;
carry = diffbuf[0];
for (i=0; i < 64; i++ ) {
if (pnum++ % raw_width == 0)
base[0] = base[1] = 512;
if ((pixel[(block << 6) + i] = base[i & 1] += diffbuf[i]) >> 10)
derror();
}
}
if (lowbits) {
save = ftell(ifp);
fseek (ifp, 26 + row*raw_width/4, SEEK_SET);
for (prow=pixel, i=0; i < raw_width*2; i++) {
c = fgetc(ifp);
for (r=0; r < 8; r+=2, prow++) {
val = (*prow << 2) + ((c >> r) & 3);
if (raw_width == 2672 && val < 512) val += 2;
*prow = val;
}
}
fseek (ifp, save, SEEK_SET);
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
FORC(2) free (huff[c]);
throw;
}
#endif
FORC(2) free (huff[c]);
}
//@end COMMON
struct jhead {
int algo, bits, high, wide, clrs, sraw, psv, restart, vpred[6];
ushort quant[64], idct[64], *huff[20], *free[20], *row;
};
//@out COMMON
int CLASS ljpeg_start (struct jhead *jh, int info_only)
{
ushort c, tag, len;
int cnt = 0;
uchar data[0x10000];
const uchar *dp;
memset (jh, 0, sizeof *jh);
jh->restart = INT_MAX;
if ((fgetc(ifp),fgetc(ifp)) != 0xd8) return 0;
do {
if(feof(ifp)) return 0;
if(cnt++ > 1024) return 0; // 1024 tags limit
if (!fread (data, 2, 2, ifp)) return 0;
tag = data[0] << 8 | data[1];
len = (data[2] << 8 | data[3]) - 2;
if (tag <= 0xff00) return 0;
fread (data, 1, len, ifp);
switch (tag) {
case 0xffc3: // start of frame; lossless, Huffman
jh->sraw = ((data[7] >> 4) * (data[7] & 15) - 1) & 3;
case 0xffc1:
case 0xffc0:
jh->algo = tag & 0xff;
jh->bits = data[0];
jh->high = data[1] << 8 | data[2];
jh->wide = data[3] << 8 | data[4];
jh->clrs = data[5] + jh->sraw;
if (len == 9 && !dng_version) getc(ifp);
break;
case 0xffc4: // define Huffman tables
if (info_only) break;
for (dp = data; dp < data+len && !((c = *dp++) & -20); )
jh->free[c] = jh->huff[c] = make_decoder_ref (&dp);
break;
case 0xffda: // start of scan
jh->psv = data[1+data[0]*2];
jh->bits -= data[3+data[0]*2] & 15;
break;
case 0xffdb:
FORC(64) jh->quant[c] = data[c*2+1] << 8 | data[c*2+2];
break;
case 0xffdd:
jh->restart = data[0] << 8 | data[1];
}
} while (tag != 0xffda);
if (info_only) return 1;
if (jh->clrs > 6 || !jh->huff[0]) return 0;
FORC(19) if (!jh->huff[c+1]) jh->huff[c+1] = jh->huff[c];
if (jh->sraw) {
FORC(4) jh->huff[2+c] = jh->huff[1];
FORC(jh->sraw) jh->huff[1+c] = jh->huff[0];
}
jh->row = (ushort *) calloc (jh->wide*jh->clrs, 4);
merror (jh->row, "ljpeg_start()");
return zero_after_ff = 1;
}
void CLASS ljpeg_end (struct jhead *jh)
{
int c;
FORC4 if (jh->free[c]) free (jh->free[c]);
free (jh->row);
}
int CLASS ljpeg_diff (ushort *huff)
{
int len, diff;
if(!huff)
#ifdef LIBRAW_LIBRARY_BUILD
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#else
longjmp (failure, 2);
#endif
len = gethuff(huff);
if (len == 16 && (!dng_version || dng_version >= 0x1010000))
return -32768;
diff = getbits(len);
if ((diff & (1 << (len-1))) == 0)
diff -= (1 << len) - 1;
return diff;
}
ushort * CLASS ljpeg_row (int jrow, struct jhead *jh)
{
int col, c, diff, pred, spred=0;
ushort mark=0, *row[3];
if (jrow * jh->wide % jh->restart == 0) {
FORC(6) jh->vpred[c] = 1 << (jh->bits-1);
if (jrow) {
fseek (ifp, -2, SEEK_CUR);
do mark = (mark << 8) + (c = fgetc(ifp));
while (c != EOF && mark >> 4 != 0xffd);
}
getbits(-1);
}
FORC3 row[c] = jh->row + jh->wide*jh->clrs*((jrow+c) & 1);
for (col=0; col < jh->wide; col++)
FORC(jh->clrs) {
diff = ljpeg_diff (jh->huff[c]);
if (jh->sraw && c <= jh->sraw && (col | c))
pred = spred;
else if (col) pred = row[0][-jh->clrs];
else pred = (jh->vpred[c] += diff) - diff;
if (jrow && col) switch (jh->psv) {
case 1: break;
case 2: pred = row[1][0]; break;
case 3: pred = row[1][-jh->clrs]; break;
case 4: pred = pred + row[1][0] - row[1][-jh->clrs]; break;
case 5: pred = pred + ((row[1][0] - row[1][-jh->clrs]) >> 1); break;
case 6: pred = row[1][0] + ((pred - row[1][-jh->clrs]) >> 1); break;
case 7: pred = (pred + row[1][0]) >> 1; break;
default: pred = 0;
}
if ((**row = pred + diff) >> jh->bits) derror();
if (c <= jh->sraw) spred = **row;
row[0]++; row[1]++;
}
return row[2];
}
void CLASS lossless_jpeg_load_raw()
{
int jwide, jhigh, jrow, jcol, val, jidx, i, j, row=0, col=0;
struct jhead jh;
ushort *rp;
if (!ljpeg_start (&jh, 0)) return;
if(jh.wide<1 || jh.high<1 || jh.clrs<1 || jh.bits <1)
#ifdef LIBRAW_LIBRARY_BUILD
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#else
longjmp (failure, 2);
#endif
jwide = jh.wide * jh.clrs;
jhigh = jh.high;
if(jh.clrs == 4 && jwide >= raw_width*2) jhigh *= 2;
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (jrow=0; jrow < jh.high; jrow++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
rp = ljpeg_row (jrow, &jh);
if (load_flags & 1)
row = jrow & 1 ? height-1-jrow/2 : jrow/2;
for (jcol=0; jcol < jwide; jcol++) {
val = curve[*rp++];
if (cr2_slice[0]) {
jidx = jrow*jwide + jcol;
i = jidx / (cr2_slice[1]*raw_height);
if ((j = i >= cr2_slice[0]))
i = cr2_slice[0];
jidx -= i * (cr2_slice[1]*raw_height);
row = jidx / cr2_slice[1+j];
col = jidx % cr2_slice[1+j] + i*cr2_slice[1];
}
if (raw_width == 3984 && (col -= 2) < 0)
col += (row--,raw_width);
if(row>raw_height)
#ifdef LIBRAW_LIBRARY_BUILD
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#else
longjmp (failure, 3);
#endif
if ((unsigned) row < raw_height) RAW(row,col) = val;
if (++col >= raw_width)
col = (row++,0);
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
ljpeg_end (&jh);
throw;
}
#endif
ljpeg_end (&jh);
}
void CLASS canon_sraw_load_raw()
{
struct jhead jh;
short *rp=0, (*ip)[4];
int jwide, slice, scol, ecol, row, col, jrow=0, jcol=0, pix[3], c;
int v[3]={0,0,0}, ver, hue;
char *cp;
if (!ljpeg_start (&jh, 0) || jh.clrs < 4) return;
jwide = (jh.wide >>= 1) * jh.clrs;
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (ecol=slice=0; slice <= cr2_slice[0]; slice++) {
scol = ecol;
ecol += cr2_slice[1] * 2 / jh.clrs;
if (!cr2_slice[0] || ecol > raw_width-1) ecol = raw_width & -2;
for (row=0; row < height; row += (jh.clrs >> 1) - 1) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
ip = (short (*)[4]) image + row*width;
for (col=scol; col < ecol; col+=2, jcol+=jh.clrs) {
if ((jcol %= jwide) == 0)
rp = (short *) ljpeg_row (jrow++, &jh);
if (col >= width) continue;
#ifdef LIBRAW_LIBRARY_BUILD
if(imgdata.params.sraw_ycc>=2)
{
FORC (jh.clrs-2)
{
ip[col + (c >> 1)*width + (c & 1)][0] = rp[jcol+c];
ip[col + (c >> 1)*width + (c & 1)][1] = ip[col + (c >> 1)*width + (c & 1)][2] = 8192;
}
ip[col][1] = rp[jcol+jh.clrs-2] - 8192;
ip[col][2] = rp[jcol+jh.clrs-1] - 8192;
}
else if(imgdata.params.sraw_ycc)
{
FORC (jh.clrs-2)
ip[col + (c >> 1)*width + (c & 1)][0] = rp[jcol+c];
ip[col][1] = rp[jcol+jh.clrs-2] - 8192;
ip[col][2] = rp[jcol+jh.clrs-1] - 8192;
}
else
#endif
{
FORC (jh.clrs-2)
ip[col + (c >> 1)*width + (c & 1)][0] = rp[jcol+c];
ip[col][1] = rp[jcol+jh.clrs-2] - 16384;
ip[col][2] = rp[jcol+jh.clrs-1] - 16384;
}
}
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
ljpeg_end (&jh);
throw ;
}
#endif
#ifdef LIBRAW_LIBRARY_BUILD
if(imgdata.params.sraw_ycc>=2)
{
ljpeg_end (&jh);
maximum = 0x3fff;
return;
}
#endif
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (cp=model2; *cp && !isdigit(*cp); cp++);
sscanf (cp, "%d.%d.%d", v, v+1, v+2);
ver = (v[0]*1000 + v[1])*1000 + v[2];
hue = (jh.sraw+1) << 2;
if (unique_id >= 0x80000281 || (unique_id == 0x80000218 && ver > 1000006))
hue = jh.sraw << 1;
ip = (short (*)[4]) image;
rp = ip[0];
for (row=0; row < height; row++, ip+=width) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (row & (jh.sraw >> 1))
{
for (col=0; col < width; col+=2)
for (c=1; c < 3; c++)
if (row == height-1)
{
ip[col][c] = ip[col-width][c];
}
else
{
ip[col][c] = (ip[col-width][c] + ip[col+width][c] + 1) >> 1;
}
}
for (col=1; col < width; col+=2)
for (c=1; c < 3; c++)
if (col == width-1)
ip[col][c] = ip[col-1][c];
else ip[col][c] = (ip[col-1][c] + ip[col+1][c] + 1) >> 1;
}
#ifdef LIBRAW_LIBRARY_BUILD
if(!imgdata.params.sraw_ycc)
#endif
for ( ; rp < ip[0]; rp+=4) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (unique_id == 0x80000218 ||
unique_id == 0x80000250 ||
unique_id == 0x80000261 ||
unique_id == 0x80000281 ||
unique_id == 0x80000287) {
rp[1] = (rp[1] << 2) + hue;
rp[2] = (rp[2] << 2) + hue;
pix[0] = rp[0] + (( 50*rp[1] + 22929*rp[2]) >> 14);
pix[1] = rp[0] + ((-5640*rp[1] - 11751*rp[2]) >> 14);
pix[2] = rp[0] + ((29040*rp[1] - 101*rp[2]) >> 14);
} else {
if (unique_id < 0x80000218) rp[0] -= 512;
pix[0] = rp[0] + rp[2];
pix[2] = rp[0] + rp[1];
pix[1] = rp[0] + ((-778*rp[1] - (rp[2] << 11)) >> 12);
}
FORC3 rp[c] = CLIP(pix[c] * sraw_mul[c] >> 10);
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
ljpeg_end (&jh);
throw ;
}
#endif
ljpeg_end (&jh);
maximum = 0x3fff;
}
void CLASS adobe_copy_pixel (unsigned row, unsigned col, ushort **rp)
{
int c;
if (tiff_samples == 2 && shot_select) (*rp)++;
if (raw_image) {
if (row < raw_height && col < raw_width)
RAW(row,col) = curve[**rp];
*rp += tiff_samples;
} else {
if (row < height && col < width)
FORC(tiff_samples)
image[row*width+col][c] = curve[(*rp)[c]];
*rp += tiff_samples;
}
if (tiff_samples == 2 && shot_select) (*rp)--;
}
void CLASS ljpeg_idct (struct jhead *jh)
{
int c, i, j, len, skip, coef;
float work[3][8][8];
static float cs[106] = { 0 };
static const uchar zigzag[80] =
{ 0, 1, 8,16, 9, 2, 3,10,17,24,32,25,18,11, 4, 5,12,19,26,33,
40,48,41,34,27,20,13, 6, 7,14,21,28,35,42,49,56,57,50,43,36,
29,22,15,23,30,37,44,51,58,59,52,45,38,31,39,46,53,60,61,54,
47,55,62,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63 };
if (!cs[0])
FORC(106) cs[c] = cos((c & 31)*M_PI/16)/2;
memset (work, 0, sizeof work);
work[0][0][0] = jh->vpred[0] += ljpeg_diff (jh->huff[0]) * jh->quant[0];
for (i=1; i < 64; i++ ) {
len = gethuff (jh->huff[16]);
i += skip = len >> 4;
if (!(len &= 15) && skip < 15) break;
coef = getbits(len);
if ((coef & (1 << (len-1))) == 0)
coef -= (1 << len) - 1;
((float *)work)[zigzag[i]] = coef * jh->quant[i];
}
FORC(8) work[0][0][c] *= M_SQRT1_2;
FORC(8) work[0][c][0] *= M_SQRT1_2;
for (i=0; i < 8; i++)
for (j=0; j < 8; j++)
FORC(8) work[1][i][j] += work[0][i][c] * cs[(j*2+1)*c];
for (i=0; i < 8; i++)
for (j=0; j < 8; j++)
FORC(8) work[2][i][j] += work[1][c][j] * cs[(i*2+1)*c];
FORC(64) jh->idct[c] = CLIP(((float *)work[2])[c]+0.5);
}
void CLASS lossless_dng_load_raw()
{
unsigned save, trow=0, tcol=0, jwide, jrow, jcol, row, col, i, j;
struct jhead jh;
ushort *rp;
while (trow < raw_height) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
save = ftell(ifp);
if (tile_length < INT_MAX)
fseek (ifp, get4(), SEEK_SET);
if (!ljpeg_start (&jh, 0)) break;
jwide = jh.wide;
if (filters) jwide *= jh.clrs;
jwide /= MIN (is_raw, tiff_samples);
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
switch (jh.algo) {
case 0xc1:
jh.vpred[0] = 16384;
getbits(-1);
for (jrow=0; jrow+7 < jh.high; jrow += 8) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (jcol=0; jcol+7 < jh.wide; jcol += 8) {
ljpeg_idct (&jh);
rp = jh.idct;
row = trow + jcol/tile_width + jrow*2;
col = tcol + jcol%tile_width;
for (i=0; i < 16; i+=2)
for (j=0; j < 8; j++)
adobe_copy_pixel (row+i, col+j, &rp);
}
}
break;
case 0xc3:
for (row=col=jrow=0; jrow < jh.high; jrow++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
rp = ljpeg_row (jrow, &jh);
for (jcol=0; jcol < jwide; jcol++) {
adobe_copy_pixel (trow+row, tcol+col, &rp);
if (++col >= tile_width || col >= raw_width)
row += 1 + (col = 0);
}
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
ljpeg_end (&jh);
throw ;
}
#endif
fseek (ifp, save+4, SEEK_SET);
if ((tcol += tile_width) >= raw_width)
trow += tile_length + (tcol = 0);
ljpeg_end (&jh);
}
}
void CLASS packed_dng_load_raw()
{
ushort *pixel, *rp;
int row, col;
pixel = (ushort *) calloc (raw_width, tiff_samples*sizeof *pixel);
merror (pixel, "packed_dng_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (tiff_bps == 16)
read_shorts (pixel, raw_width * tiff_samples);
else {
getbits(-1);
for (col=0; col < raw_width * tiff_samples; col++)
pixel[col] = getbits(tiff_bps);
}
for (rp=pixel, col=0; col < raw_width; col++)
adobe_copy_pixel (row, col, &rp);
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
free (pixel);
throw ;
}
#endif
free (pixel);
}
void CLASS pentax_load_raw()
{
ushort bit[2][15], huff[4097];
int dep, row, col, diff, c, i;
ushort vpred[2][2] = {{0,0},{0,0}}, hpred[2];
fseek (ifp, meta_offset, SEEK_SET);
dep = (get2() + 12) & 15;
fseek (ifp, 12, SEEK_CUR);
FORC(dep) bit[0][c] = get2();
FORC(dep) bit[1][c] = fgetc(ifp);
FORC(dep)
for (i=bit[0][c]; i <= ((bit[0][c]+(4096 >> bit[1][c])-1) & 4095); )
huff[++i] = bit[1][c] << 8 | c;
huff[0] = 12;
fseek (ifp, data_offset, SEEK_SET);
getbits(-1);
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
diff = ljpeg_diff (huff);
if (col < 2) hpred[col] = vpred[row & 1][col] += diff;
else hpred[col & 1] += diff;
RAW(row,col) = hpred[col & 1];
if (hpred[col & 1] >> tiff_bps) derror();
}
}
}
#ifdef LIBRAW_LIBRARY_BUILD
void CLASS nikon_coolscan_load_raw()
{
int bufsize = width*3*tiff_bps/8;
if(tiff_bps <= 8)
gamma_curve(1.0/imgdata.params.coolscan_nef_gamma,0.,1,255);
else
gamma_curve(1.0/imgdata.params.coolscan_nef_gamma,0.,1,65535);
fseek (ifp, data_offset, SEEK_SET);
unsigned char *buf = (unsigned char*)malloc(bufsize);
unsigned short *ubuf = (unsigned short *)buf;
for(int row = 0; row < raw_height; row++)
{
int red = fread (buf, 1, bufsize, ifp);
unsigned short (*ip)[4] = (unsigned short (*)[4]) image + row*width;
if(tiff_bps <= 8)
for(int col=0; col<width;col++)
{
ip[col][0] = curve[buf[col*3]];
ip[col][1] = curve[buf[col*3+1]];
ip[col][2] = curve[buf[col*3+2]];
ip[col][3]=0;
}
else
for(int col=0; col<width;col++)
{
ip[col][0] = curve[ubuf[col*3]];
ip[col][1] = curve[ubuf[col*3+1]];
ip[col][2] = curve[ubuf[col*3+2]];
ip[col][3]=0;
}
}
free(buf);
}
#endif
void CLASS nikon_load_raw()
{
static const uchar nikon_tree[][32] = {
{ 0,1,5,1,1,1,1,1,1,2,0,0,0,0,0,0, /* 12-bit lossy */
5,4,3,6,2,7,1,0,8,9,11,10,12 },
{ 0,1,5,1,1,1,1,1,1,2,0,0,0,0,0,0, /* 12-bit lossy after split */
0x39,0x5a,0x38,0x27,0x16,5,4,3,2,1,0,11,12,12 },
{ 0,1,4,2,3,1,2,0,0,0,0,0,0,0,0,0, /* 12-bit lossless */
5,4,6,3,7,2,8,1,9,0,10,11,12 },
{ 0,1,4,3,1,1,1,1,1,2,0,0,0,0,0,0, /* 14-bit lossy */
5,6,4,7,8,3,9,2,1,0,10,11,12,13,14 },
{ 0,1,5,1,1,1,1,1,1,1,2,0,0,0,0,0, /* 14-bit lossy after split */
8,0x5c,0x4b,0x3a,0x29,7,6,5,4,3,2,1,0,13,14 },
{ 0,1,4,2,2,3,1,2,0,0,0,0,0,0,0,0, /* 14-bit lossless */
7,6,8,5,9,4,10,3,11,12,2,0,1,13,14 } };
ushort *huff, ver0, ver1, vpred[2][2], hpred[2], csize;
int i, min, max, step=0, tree=0, split=0, row, col, len, shl, diff;
fseek (ifp, meta_offset, SEEK_SET);
ver0 = fgetc(ifp);
ver1 = fgetc(ifp);
if (ver0 == 0x49 || ver1 == 0x58)
fseek (ifp, 2110, SEEK_CUR);
if (ver0 == 0x46) tree = 2;
if (tiff_bps == 14) tree += 3;
read_shorts (vpred[0], 4);
max = 1 << tiff_bps & 0x7fff;
if ((csize = get2()) > 1)
step = max / (csize-1);
if (ver0 == 0x44 && ver1 == 0x20 && step > 0) {
for (i=0; i < csize; i++)
curve[i*step] = get2();
for (i=0; i < max; i++)
curve[i] = ( curve[i-i%step]*(step-i%step) +
curve[i-i%step+step]*(i%step) ) / step;
fseek (ifp, meta_offset+562, SEEK_SET);
split = get2();
} else if (ver0 != 0x46 && csize <= 0x4001)
read_shorts (curve, max=csize);
while (curve[max-2] == curve[max-1]) max--;
huff = make_decoder (nikon_tree[tree]);
fseek (ifp, data_offset, SEEK_SET);
getbits(-1);
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (min=row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (split && row == split) {
free (huff);
huff = make_decoder (nikon_tree[tree+1]);
max += (min = 16) << 1;
}
for (col=0; col < raw_width; col++) {
i = gethuff(huff);
len = i & 15;
shl = i >> 4;
diff = ((getbits(len-shl) << 1) + 1) << shl >> 1;
if ((diff & (1 << (len-1))) == 0)
diff -= (1 << len) - !shl;
if (col < 2) hpred[col] = vpred[row & 1][col] += diff;
else hpred[col & 1] += diff;
if ((ushort)(hpred[col & 1] + min) >= max) derror();
RAW(row,col) = curve[LIM((short)hpred[col & 1],0,0x3fff)];
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
free (huff);
throw;
}
#endif
free (huff);
}
void CLASS nikon_yuv_load_raw()
{
int row, col, yuv[4], rgb[3], b, c;
UINT64 bitbuf=0;
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
if (!(b = col & 1)) {
bitbuf = 0;
FORC(6) bitbuf |= (UINT64) fgetc(ifp) << c*8;
FORC(4) yuv[c] = (bitbuf >> c*12 & 0xfff) - (c >> 1 << 11);
}
rgb[0] = yuv[b] + 1.370705*yuv[3];
rgb[1] = yuv[b] - 0.337633*yuv[2] - 0.698001*yuv[3];
rgb[2] = yuv[b] + 1.732446*yuv[2];
FORC3 image[row*width+col][c] = curve[LIM(rgb[c],0,0xfff)] / cam_mul[c];
}
}
}
/*
Returns 1 for a Coolpix 995, 0 for anything else.
*/
int CLASS nikon_e995()
{
int i, histo[256];
const uchar often[] = { 0x00, 0x55, 0xaa, 0xff };
memset (histo, 0, sizeof histo);
fseek (ifp, -2000, SEEK_END);
for (i=0; i < 2000; i++)
histo[fgetc(ifp)]++;
for (i=0; i < 4; i++)
if (histo[often[i]] < 200)
return 0;
return 1;
}
/*
Returns 1 for a Coolpix 2100, 0 for anything else.
*/
int CLASS nikon_e2100()
{
uchar t[12];
int i;
fseek (ifp, 0, SEEK_SET);
for (i=0; i < 1024; i++) {
fread (t, 1, 12, ifp);
if (((t[2] & t[4] & t[7] & t[9]) >> 4
& t[1] & t[6] & t[8] & t[11] & 3) != 3)
return 0;
}
return 1;
}
void CLASS nikon_3700()
{
int bits, i;
uchar dp[24];
static const struct {
int bits;
char t_make[12], t_model[15];
} table[] = {
{ 0x00, "Pentax", "Optio 33WR" },
{ 0x03, "Nikon", "E3200" },
{ 0x32, "Nikon", "E3700" },
{ 0x33, "Olympus", "C740UZ" } };
fseek (ifp, 3072, SEEK_SET);
fread (dp, 1, 24, ifp);
bits = (dp[8] & 3) << 4 | (dp[20] & 3);
for (i=0; i < sizeof table / sizeof *table; i++)
if (bits == table[i].bits) {
strcpy (make, table[i].t_make );
strcpy (model, table[i].t_model);
}
}
/*
Separates a Minolta DiMAGE Z2 from a Nikon E4300.
*/
int CLASS minolta_z2()
{
int i, nz;
char tail[424];
fseek (ifp, -sizeof tail, SEEK_END);
fread (tail, 1, sizeof tail, ifp);
for (nz=i=0; i < sizeof tail; i++)
if (tail[i]) nz++;
return nz > 20;
}
//@end COMMON
void CLASS jpeg_thumb();
//@out COMMON
void CLASS ppm_thumb()
{
char *thumb;
thumb_length = thumb_width*thumb_height*3;
thumb = (char *) malloc (thumb_length);
merror (thumb, "ppm_thumb()");
fprintf (ofp, "P6\n%d %d\n255\n", thumb_width, thumb_height);
fread (thumb, 1, thumb_length, ifp);
fwrite (thumb, 1, thumb_length, ofp);
free (thumb);
}
void CLASS ppm16_thumb()
{
int i;
char *thumb;
thumb_length = thumb_width*thumb_height*3;
thumb = (char *) calloc (thumb_length, 2);
merror (thumb, "ppm16_thumb()");
read_shorts ((ushort *) thumb, thumb_length);
for (i=0; i < thumb_length; i++)
thumb[i] = ((ushort *) thumb)[i] >> 8;
fprintf (ofp, "P6\n%d %d\n255\n", thumb_width, thumb_height);
fwrite (thumb, 1, thumb_length, ofp);
free (thumb);
}
void CLASS layer_thumb()
{
int i, c;
char *thumb, map[][4] = { "012","102" };
colors = thumb_misc >> 5 & 7;
thumb_length = thumb_width*thumb_height;
thumb = (char *) calloc (colors, thumb_length);
merror (thumb, "layer_thumb()");
fprintf (ofp, "P%d\n%d %d\n255\n",
5 + (colors >> 1), thumb_width, thumb_height);
fread (thumb, thumb_length, colors, ifp);
for (i=0; i < thumb_length; i++)
FORCC putc (thumb[i+thumb_length*(map[thumb_misc >> 8][c]-'0')], ofp);
free (thumb);
}
void CLASS rollei_thumb()
{
unsigned i;
ushort *thumb;
thumb_length = thumb_width * thumb_height;
thumb = (ushort *) calloc (thumb_length, 2);
merror (thumb, "rollei_thumb()");
fprintf (ofp, "P6\n%d %d\n255\n", thumb_width, thumb_height);
read_shorts (thumb, thumb_length);
for (i=0; i < thumb_length; i++) {
putc (thumb[i] << 3, ofp);
putc (thumb[i] >> 5 << 2, ofp);
putc (thumb[i] >> 11 << 3, ofp);
}
free (thumb);
}
void CLASS rollei_load_raw()
{
uchar pixel[10];
unsigned iten=0, isix, i, buffer=0, todo[16];
isix = raw_width * raw_height * 5 / 8;
while (fread (pixel, 1, 10, ifp) == 10) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (i=0; i < 10; i+=2) {
todo[i] = iten++;
todo[i+1] = pixel[i] << 8 | pixel[i+1];
buffer = pixel[i] >> 2 | buffer << 6;
}
for ( ; i < 16; i+=2) {
todo[i] = isix++;
todo[i+1] = buffer >> (14-i)*5;
}
for (i=0; i < 16; i+=2)
raw_image[todo[i]] = (todo[i+1] & 0x3ff);
}
maximum = 0x3ff;
}
int CLASS raw (unsigned row, unsigned col)
{
return (row < raw_height && col < raw_width) ? RAW(row,col) : 0;
}
void CLASS phase_one_flat_field (int is_float, int nc)
{
ushort head[8];
unsigned wide, high, y, x, c, rend, cend, row, col;
float *mrow, num, mult[4];
read_shorts (head, 8);
if (head[2] * head[3] * head[4] * head[5] == 0) return;
wide = head[2] / head[4] + (head[2] % head[4] != 0);
high = head[3] / head[5] + (head[3] % head[5] != 0);
mrow = (float *) calloc (nc*wide, sizeof *mrow);
merror (mrow, "phase_one_flat_field()");
for (y=0; y < high; y++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (x=0; x < wide; x++)
for (c=0; c < nc; c+=2) {
num = is_float ? getreal(11) : get2()/32768.0;
if (y==0) mrow[c*wide+x] = num;
else mrow[(c+1)*wide+x] = (num - mrow[c*wide+x]) / head[5];
}
if (y==0) continue;
rend = head[1] + y*head[5];
for (row = rend-head[5];
row < raw_height && row < rend &&
row < head[1]+head[3]-head[5]; row++) {
for (x=1; x < wide; x++) {
for (c=0; c < nc; c+=2) {
mult[c] = mrow[c*wide+x-1];
mult[c+1] = (mrow[c*wide+x] - mult[c]) / head[4];
}
cend = head[0] + x*head[4];
for (col = cend-head[4];
col < raw_width &&
col < cend && col < head[0]+head[2]-head[4]; col++) {
c = nc > 2 ? FC(row-top_margin,col-left_margin) : 0;
if (!(c & 1)) {
c = RAW(row,col) * mult[c];
RAW(row,col) = LIM(c,0,65535);
}
for (c=0; c < nc; c+=2)
mult[c] += mult[c+1];
}
}
for (x=0; x < wide; x++)
for (c=0; c < nc; c+=2)
mrow[c*wide+x] += mrow[(c+1)*wide+x];
}
}
free (mrow);
}
int CLASS phase_one_correct()
{
unsigned entries, tag, data, save, col, row, type;
int len, i, j, k, cip, val[4], dev[4], sum, max;
int head[9], diff, mindiff=INT_MAX, off_412=0;
/* static */ const signed char dir[12][2] =
{ {-1,-1}, {-1,1}, {1,-1}, {1,1}, {-2,0}, {0,-2}, {0,2}, {2,0},
{-2,-2}, {-2,2}, {2,-2}, {2,2} };
float poly[8], num, cfrac, frac, mult[2], *yval[2]={NULL,NULL};
ushort *xval[2];
int qmult_applied = 0, qlin_applied = 0;
if (half_size || !meta_length) return 0;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Phase One correction...\n"));
#endif
fseek (ifp, meta_offset, SEEK_SET);
order = get2();
fseek (ifp, 6, SEEK_CUR);
fseek (ifp, meta_offset+get4(), SEEK_SET);
entries = get4(); get4();
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
while (entries--) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
tag = get4();
len = get4();
data = get4();
save = ftell(ifp);
fseek (ifp, meta_offset+data, SEEK_SET);
if (tag == 0x419) { /* Polynomial curve */
for (get4(), i=0; i < 8; i++)
poly[i] = getreal(11);
poly[3] += (ph1.tag_210 - poly[7]) * poly[6] + 1;
for (i=0; i < 0x10000; i++) {
num = (poly[5]*i + poly[3])*i + poly[1];
curve[i] = LIM(num,0,65535);
} goto apply; /* apply to right half */
} else if (tag == 0x41a) { /* Polynomial curve */
for (i=0; i < 4; i++)
poly[i] = getreal(11);
for (i=0; i < 0x10000; i++) {
for (num=0, j=4; j--; )
num = num * i + poly[j];
curve[i] = LIM(num+i,0,65535);
} apply: /* apply to whole image */
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col = (tag & 1)*ph1.split_col; col < raw_width; col++)
RAW(row,col) = curve[RAW(row,col)];
}
} else if (tag == 0x400) { /* Sensor defects */
while ((len -= 8) >= 0) {
col = get2();
row = get2();
type = get2(); get2();
if (col >= raw_width) continue;
if (type == 131 || type == 137) /* Bad column */
for (row=0; row < raw_height; row++)
if (FC(row-top_margin,col-left_margin) == 1) {
for (sum=i=0; i < 4; i++)
sum += val[i] = raw (row+dir[i][0], col+dir[i][1]);
for (max=i=0; i < 4; i++) {
dev[i] = abs((val[i] << 2) - sum);
if (dev[max] < dev[i]) max = i;
}
RAW(row,col) = (sum - val[max])/3.0 + 0.5;
} else {
for (sum=0, i=8; i < 12; i++)
sum += raw (row+dir[i][0], col+dir[i][1]);
RAW(row,col) = 0.5 + sum * 0.0732233 +
(raw(row,col-2) + raw(row,col+2)) * 0.3535534;
}
else if (type == 129) { /* Bad pixel */
if (row >= raw_height) continue;
j = (FC(row-top_margin,col-left_margin) != 1) * 4;
for (sum=0, i=j; i < j+8; i++)
sum += raw (row+dir[i][0], col+dir[i][1]);
RAW(row,col) = (sum + 4) >> 3;
}
}
} else if (tag == 0x401) { /* All-color flat fields */
phase_one_flat_field (1, 2);
} else if (tag == 0x416 || tag == 0x410) {
phase_one_flat_field (0, 2);
} else if (tag == 0x40b) { /* Red+blue flat field */
phase_one_flat_field (0, 4);
} else if (tag == 0x412) {
fseek (ifp, 36, SEEK_CUR);
diff = abs (get2() - ph1.tag_21a);
if (mindiff > diff) {
mindiff = diff;
off_412 = ftell(ifp) - 38;
}
} else if (tag == 0x41f && !qlin_applied) { /* Quadrant linearization */
ushort lc[2][2][16], ref[16];
int qr, qc;
for (qr = 0; qr < 2; qr++)
for (qc = 0; qc < 2; qc++)
for (i = 0; i < 16; i++)
lc[qr][qc][i] = get4();
for (i = 0; i < 16; i++) {
int v = 0;
for (qr = 0; qr < 2; qr++)
for (qc = 0; qc < 2; qc++)
v += lc[qr][qc][i];
ref[i] = (v + 2) >> 2;
}
for (qr = 0; qr < 2; qr++) {
for (qc = 0; qc < 2; qc++) {
int cx[19], cf[19];
for (i = 0; i < 16; i++) {
cx[1+i] = lc[qr][qc][i];
cf[1+i] = ref[i];
}
cx[0] = cf[0] = 0;
cx[17] = cf[17] = ((unsigned int)ref[15] * 65535) / lc[qr][qc][15];
cf[18] = cx[18] = 65535;
cubic_spline(cx, cf, 19);
for (row = (qr ? ph1.split_row : 0);
row < (qr ? raw_height : ph1.split_row); row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col = (qc ? ph1.split_col : 0);
col < (qc ? raw_width : ph1.split_col); col++)
RAW(row,col) = curve[RAW(row,col)];
}
}
}
qlin_applied = 1;
} else if (tag == 0x41e && !qmult_applied) { /* Quadrant multipliers */
float qmult[2][2] = { { 1, 1 }, { 1, 1 } };
get4(); get4(); get4(); get4();
qmult[0][0] = 1.0 + getreal(11);
get4(); get4(); get4(); get4(); get4();
qmult[0][1] = 1.0 + getreal(11);
get4(); get4(); get4();
qmult[1][0] = 1.0 + getreal(11);
get4(); get4(); get4();
qmult[1][1] = 1.0 + getreal(11);
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
i = qmult[row >= ph1.split_row][col >= ph1.split_col] * RAW(row,col);
RAW(row,col) = LIM(i,0,65535);
}
}
qmult_applied = 1;
} else if (tag == 0x431 && !qmult_applied) { /* Quadrant combined */
ushort lc[2][2][7], ref[7];
int qr, qc;
for (i = 0; i < 7; i++)
ref[i] = get4();
for (qr = 0; qr < 2; qr++)
for (qc = 0; qc < 2; qc++)
for (i = 0; i < 7; i++)
lc[qr][qc][i] = get4();
for (qr = 0; qr < 2; qr++) {
for (qc = 0; qc < 2; qc++) {
int cx[9], cf[9];
for (i = 0; i < 7; i++) {
cx[1+i] = ref[i];
cf[1+i] = ((unsigned) ref[i] * lc[qr][qc][i]) / 10000;
}
cx[0] = cf[0] = 0;
cx[8] = cf[8] = 65535;
cubic_spline(cx, cf, 9);
for (row = (qr ? ph1.split_row : 0);
row < (qr ? raw_height : ph1.split_row); row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col = (qc ? ph1.split_col : 0);
col < (qc ? raw_width : ph1.split_col); col++)
RAW(row,col) = curve[RAW(row,col)];
}
}
}
qmult_applied = 1;
qlin_applied = 1;
}
fseek (ifp, save, SEEK_SET);
}
if (off_412) {
fseek (ifp, off_412, SEEK_SET);
for (i=0; i < 9; i++) head[i] = get4() & 0x7fff;
yval[0] = (float *) calloc (head[1]*head[3] + head[2]*head[4], 6);
merror (yval[0], "phase_one_correct()");
yval[1] = (float *) (yval[0] + head[1]*head[3]);
xval[0] = (ushort *) (yval[1] + head[2]*head[4]);
xval[1] = (ushort *) (xval[0] + head[1]*head[3]);
get2();
for (i=0; i < 2; i++)
for (j=0; j < head[i+1]*head[i+3]; j++)
yval[i][j] = getreal(11);
for (i=0; i < 2; i++)
for (j=0; j < head[i+1]*head[i+3]; j++)
xval[i][j] = get2();
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
cfrac = (float) col * head[3] / raw_width;
cfrac -= cip = cfrac;
num = RAW(row,col) * 0.5;
for (i=cip; i < cip+2; i++) {
for (k=j=0; j < head[1]; j++)
if (num < xval[0][k = head[1]*i+j]) break;
frac = (j == 0 || j == head[1]) ? 0 :
(xval[0][k] - num) / (xval[0][k] - xval[0][k-1]);
mult[i-cip] = yval[0][k-1] * frac + yval[0][k] * (1-frac);
}
i = ((mult[0] * (1-cfrac) + mult[1] * cfrac) * row + num) * 2;
RAW(row,col) = LIM(i,0,65535);
}
}
free (yval[0]);
}
#ifdef LIBRAW_LIBRARY_BUILD
}
catch (...)
{
if(yval[0]) free(yval[0]);
return LIBRAW_CANCELLED_BY_CALLBACK;
}
#endif
return 0;
}
void CLASS phase_one_load_raw()
{
int a, b, i;
ushort akey, bkey, t_mask;
fseek (ifp, ph1.key_off, SEEK_SET);
akey = get2();
bkey = get2();
t_mask = ph1.format == 1 ? 0x5555:0x1354;
#ifdef LIBRAW_LIBRARY_BUILD
if (ph1.black_col || ph1.black_row )
{
imgdata.rawdata.ph1_cblack = (short(*)[2])calloc(raw_height*2,sizeof(ushort));
merror(imgdata.rawdata.ph1_cblack,"phase_one_load_raw()");
imgdata.rawdata.ph1_rblack = (short(*)[2])calloc(raw_width*2,sizeof(ushort));
merror(imgdata.rawdata.ph1_rblack,"phase_one_load_raw()");
if (ph1.black_col)
{
fseek (ifp, ph1.black_col, SEEK_SET);
read_shorts ((ushort *)imgdata.rawdata.ph1_cblack[0], raw_height*2);
}
if (ph1.black_row)
{
fseek (ifp, ph1.black_row, SEEK_SET);
read_shorts ((ushort *) imgdata.rawdata.ph1_rblack[0], raw_width*2);
}
}
#endif
fseek (ifp, data_offset, SEEK_SET);
read_shorts (raw_image, raw_width*raw_height);
if (ph1.format)
for (i=0; i < raw_width*raw_height; i+=2) {
a = raw_image[i+0] ^ akey;
b = raw_image[i+1] ^ bkey;
raw_image[i+0] = (a & t_mask) | (b & ~t_mask);
raw_image[i+1] = (b & t_mask) | (a & ~t_mask);
}
}
unsigned CLASS ph1_bithuff (int nbits, ushort *huff)
{
#ifndef LIBRAW_NOTHREADS
#define bitbuf tls->ph1_bits.bitbuf
#define vbits tls->ph1_bits.vbits
#else
static UINT64 bitbuf=0;
static int vbits=0;
#endif
unsigned c;
if (nbits == -1)
return bitbuf = vbits = 0;
if (nbits == 0) return 0;
if (vbits < nbits) {
bitbuf = bitbuf << 32 | get4();
vbits += 32;
}
c = bitbuf << (64-vbits) >> (64-nbits);
if (huff) {
vbits -= huff[c] >> 8;
return (uchar) huff[c];
}
vbits -= nbits;
return c;
#ifndef LIBRAW_NOTHREADS
#undef bitbuf
#undef vbits
#endif
}
#define ph1_bits(n) ph1_bithuff(n,0)
#define ph1_huff(h) ph1_bithuff(*h,h+1)
void CLASS phase_one_load_raw_c()
{
static const int length[] = { 8,7,6,9,11,10,5,12,14,13 };
int *offset, len[2], pred[2], row, col, i, j;
ushort *pixel;
short (*c_black)[2], (*r_black)[2];
#ifdef LIBRAW_LIBRARY_BUILD
if(ph1.format == 6)
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#endif
pixel = (ushort *) calloc (raw_width*3 + raw_height*4, 2);
merror (pixel, "phase_one_load_raw_c()");
offset = (int *) (pixel + raw_width);
fseek (ifp, strip_offset, SEEK_SET);
for (row=0; row < raw_height; row++)
offset[row] = get4();
c_black = (short (*)[2]) (offset + raw_height);
fseek (ifp, ph1.black_col, SEEK_SET);
if (ph1.black_col)
read_shorts ((ushort *) c_black[0], raw_height*2);
r_black = c_black + raw_height;
fseek (ifp, ph1.black_row, SEEK_SET);
if (ph1.black_row)
read_shorts ((ushort *) r_black[0], raw_width*2);
#ifdef LIBRAW_LIBRARY_BUILD
// Copy data to internal copy (ever if not read)
if (ph1.black_col || ph1.black_row )
{
imgdata.rawdata.ph1_cblack = (short(*)[2])calloc(raw_height*2,sizeof(ushort));
merror(imgdata.rawdata.ph1_cblack,"phase_one_load_raw_c()");
memmove(imgdata.rawdata.ph1_cblack,(ushort*)c_black[0],raw_height*2*sizeof(ushort));
imgdata.rawdata.ph1_rblack = (short(*)[2])calloc(raw_width*2,sizeof(ushort));
merror(imgdata.rawdata.ph1_rblack,"phase_one_load_raw_c()");
memmove(imgdata.rawdata.ph1_rblack,(ushort*)r_black[0],raw_width*2*sizeof(ushort));
}
#endif
for (i=0; i < 256; i++)
curve[i] = i*i / 3.969 + 0.5;
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
fseek (ifp, data_offset + offset[row], SEEK_SET);
ph1_bits(-1);
pred[0] = pred[1] = 0;
for (col=0; col < raw_width; col++) {
if (col >= (raw_width & -8))
len[0] = len[1] = 14;
else if ((col & 7) == 0)
for (i=0; i < 2; i++) {
for (j=0; j < 5 && !ph1_bits(1); j++);
if (j--) len[i] = length[j*2 + ph1_bits(1)];
}
if ((i = len[col & 1]) == 14)
pixel[col] = pred[col & 1] = ph1_bits(16);
else
pixel[col] = pred[col & 1] += ph1_bits(i) + 1 - (1 << (i - 1));
if (pred[col & 1] >> 16) derror();
if (ph1.format == 5 && pixel[col] < 256)
pixel[col] = curve[pixel[col]];
}
for (col=0; col < raw_width; col++) {
#ifndef LIBRAW_LIBRARY_BUILD
i = (pixel[col] << 2) - ph1.t_black
+ c_black[row][col >= ph1.split_col]
+ r_black[col][row >= ph1.split_row];
if (i > 0) RAW(row,col) = i;
#else
RAW(row,col) = pixel[col] << 2;
#endif
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (pixel);
throw;
}
#endif
free (pixel);
maximum = 0xfffc - ph1.t_black;
}
void CLASS hasselblad_load_raw()
{
struct jhead jh;
int shot, row, col, *back[5], len[2], diff[12], pred, sh, f, s, c;
unsigned upix, urow, ucol;
ushort *ip;
if (!ljpeg_start (&jh, 0)) return;
order = 0x4949;
ph1_bits(-1);
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
back[4] = (int *) calloc (raw_width, 3*sizeof **back);
merror (back[4], "hasselblad_load_raw()");
FORC3 back[c] = back[4] + c*raw_width;
cblack[6] >>= sh = tiff_samples > 1;
shot = LIM(shot_select, 1, tiff_samples) - 1;
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
FORC4 back[(c+3) & 3] = back[c];
for (col=0; col < raw_width; col+=2) {
for (s=0; s < tiff_samples*2; s+=2) {
FORC(2) len[c] = ph1_huff(jh.huff[0]);
FORC(2) {
diff[s+c] = ph1_bits(len[c]);
if ((diff[s+c] & (1 << (len[c]-1))) == 0)
diff[s+c] -= (1 << len[c]) - 1;
if (diff[s+c] == 65535) diff[s+c] = -32768;
}
}
for (s=col; s < col+2; s++) {
pred = 0x8000 + load_flags;
if (col) pred = back[2][s-2];
if (col && row > 1) switch (jh.psv) {
case 11: pred += back[0][s]/2 - back[0][s-2]/2; break;
}
f = (row & 1)*3 ^ ((col+s) & 1);
FORC (tiff_samples) {
pred += diff[(s & 1)*tiff_samples+c];
upix = pred >> sh & 0xffff;
if (raw_image && c == shot)
RAW(row,s) = upix;
if (image) {
urow = row-top_margin + (c & 1);
ucol = col-left_margin - ((c >> 1) & 1);
ip = &image[urow*width+ucol][f];
if (urow < height && ucol < width)
*ip = c < 4 ? upix : (*ip + upix) >> 1;
}
}
back[2][s] = pred;
}
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...){
free (back[4]);
ljpeg_end (&jh);
throw;
}
#endif
free (back[4]);
ljpeg_end (&jh);
if (image) mix_green = 1;
}
void CLASS leaf_hdr_load_raw()
{
ushort *pixel=0;
unsigned tile=0, r, c, row, col;
if (!filters) {
pixel = (ushort *) calloc (raw_width, sizeof *pixel);
merror (pixel, "leaf_hdr_load_raw()");
}
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
FORC(tiff_samples)
for (r=0; r < raw_height; r++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (r % tile_length == 0) {
fseek (ifp, data_offset + 4*tile++, SEEK_SET);
fseek (ifp, get4(), SEEK_SET);
}
if (filters && c != shot_select) continue;
if (filters) pixel = raw_image + r*raw_width;
read_shorts (pixel, raw_width);
if (!filters && (row = r - top_margin) < height)
for (col=0; col < width; col++)
image[row*width+col][c] = pixel[col+left_margin];
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
if(!filters) free(pixel);
throw;
}
#endif
if (!filters) {
maximum = 0xffff;
raw_color = 1;
free (pixel);
}
}
void CLASS unpacked_load_raw()
{
int row, col, bits=0;
while (1 << ++bits < maximum);
read_shorts (raw_image, raw_width*raw_height);
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++)
if ((RAW(row,col) >>= load_flags) >> bits
&& (unsigned) (row-top_margin) < height
&& (unsigned) (col-left_margin) < width) derror();
}
}
void CLASS unpacked_load_raw_reversed()
{
int row, col, bits=0;
while (1 << ++bits < maximum);
for (row=raw_height-1; row >= 0; row--)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
read_shorts (&raw_image[row*raw_width], raw_width);
for (col=0; col < raw_width; col++)
if ((RAW(row,col) >>= load_flags) >> bits
&& (unsigned) (row-top_margin) < height
&& (unsigned) (col-left_margin) < width) derror();
}
}
void CLASS sinar_4shot_load_raw()
{
ushort *pixel;
unsigned shot, row, col, r, c;
if (raw_image) {
shot = LIM (shot_select, 1, 4) - 1;
fseek (ifp, data_offset + shot*4, SEEK_SET);
fseek (ifp, get4(), SEEK_SET);
unpacked_load_raw();
return;
}
pixel = (ushort *) calloc (raw_width, sizeof *pixel);
merror (pixel, "sinar_4shot_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (shot=0; shot < 4; shot++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
fseek (ifp, data_offset + shot*4, SEEK_SET);
fseek (ifp, get4(), SEEK_SET);
for (row=0; row < raw_height; row++) {
read_shorts (pixel, raw_width);
if ((r = row-top_margin - (shot >> 1 & 1)) >= height) continue;
for (col=0; col < raw_width; col++) {
if ((c = col-left_margin - (shot & 1)) >= width) continue;
image[r*width+c][(row & 1)*3 ^ (~col & 1)] = pixel[col];
}
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
free(pixel);
throw;
}
#endif
free (pixel);
mix_green = 1;
}
void CLASS imacon_full_load_raw()
{
int row, col;
if (!image) return;
#ifdef LIBRAW_LIBRARY_BUILD
unsigned short *buf = (unsigned short *)malloc(width*3*sizeof(unsigned short));
merror(buf,"imacon_full_load_raw");
#endif
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
read_shorts(buf,width*3);
unsigned short (*rowp)[4] = &image[row*width];
for (col=0; col < width; col++)
{
rowp[col][0]=buf[col*3];
rowp[col][1]=buf[col*3+1];
rowp[col][2]=buf[col*3+2];
rowp[col][3]=0;
}
#else
for (col=0; col < width; col++)
read_shorts (image[row*width+col], 3);
#endif
}
#ifdef LIBRAW_LIBRARY_BUILD
free(buf);
#endif
}
void CLASS packed_load_raw()
{
int vbits=0, bwide, rbits, bite, half, irow, row, col, val, i;
UINT64 bitbuf=0;
bwide = raw_width * tiff_bps / 8;
bwide += bwide & load_flags >> 7;
rbits = bwide * 8 - raw_width * tiff_bps;
if (load_flags & 1) bwide = bwide * 16 / 15;
bite = 8 + (load_flags & 24);
half = (raw_height+1) >> 1;
for (irow=0; irow < raw_height; irow++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
row = irow;
if (load_flags & 2 &&
(row = irow % half * 2 + irow / half) == 1 &&
load_flags & 4) {
if (vbits=0, tiff_compress)
fseek (ifp, data_offset - (-half*bwide & -2048), SEEK_SET);
else {
fseek (ifp, 0, SEEK_END);
fseek (ifp, ftell(ifp) >> 3 << 2, SEEK_SET);
}
}
for (col=0; col < raw_width; col++) {
for (vbits -= tiff_bps; vbits < 0; vbits += bite) {
bitbuf <<= bite;
for (i=0; i < bite; i+=8)
bitbuf |= (unsigned) (fgetc(ifp) << i);
}
val = bitbuf << (64-tiff_bps-vbits) >> (64-tiff_bps);
RAW(row,col ^ (load_flags >> 6 & 1)) = val;
if (load_flags & 1 && (col % 10) == 9 && fgetc(ifp) &&
row < height+top_margin && col < width+left_margin) derror();
}
vbits -= rbits;
}
}
void CLASS nokia_load_raw()
{
uchar *data, *dp;
int rev, dwide, row, col, c;
double sum[]={0,0};
rev = 3 * (order == 0x4949);
dwide = (raw_width * 5 + 1) / 4;
data = (uchar *) malloc (dwide*2);
merror (data, "nokia_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (data+dwide, 1, dwide, ifp) < dwide) derror();
FORC(dwide) data[c] = data[dwide+(c ^ rev)];
for (dp=data, col=0; col < raw_width; dp+=5, col+=4)
FORC4 RAW(row,col+c) = (dp[c] << 2) | (dp[4] >> (c << 1) & 3);
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...){
free (data);
throw;
}
#endif
free (data);
maximum = 0x3ff;
if (strncmp(make,"OmniVision",10)) return;
row = raw_height/2;
FORC(width-1) {
sum[ c & 1] += SQR(RAW(row,c)-RAW(row+1,c+1));
sum[~c & 1] += SQR(RAW(row+1,c)-RAW(row,c+1));
}
if (sum[1] > sum[0]) filters = 0x4b4b4b4b;
}
void CLASS android_tight_load_raw()
{
uchar *data, *dp;
int bwide, row, col, c;
bwide = -(-5*raw_width >> 5) << 3;
data = (uchar *) malloc (bwide);
merror (data, "android_tight_load_raw()");
for (row=0; row < raw_height; row++) {
if (fread (data, 1, bwide, ifp) < bwide) derror();
for (dp=data, col=0; col < raw_width; dp+=5, col+=4)
FORC4 RAW(row,col+c) = (dp[c] << 2) | (dp[4] >> (c << 1) & 3);
}
free (data);
}
void CLASS android_loose_load_raw()
{
uchar *data, *dp;
int bwide, row, col, c;
UINT64 bitbuf=0;
bwide = (raw_width+5)/6 << 3;
data = (uchar *) malloc (bwide);
merror (data, "android_loose_load_raw()");
for (row=0; row < raw_height; row++) {
if (fread (data, 1, bwide, ifp) < bwide) derror();
for (dp=data, col=0; col < raw_width; dp+=8, col+=6) {
FORC(8) bitbuf = (bitbuf << 8) | dp[c^7];
FORC(6) RAW(row,col+c) = (bitbuf >> c*10) & 0x3ff;
}
}
free (data);
}
void CLASS canon_rmf_load_raw()
{
int row, col, bits, orow, ocol, c;
#ifdef LIBRAW_LIBRARY_BUILD
int *words = (int*)malloc(sizeof(int)*(raw_width/3+1));
merror(words,"canon_rmf_load_raw");
#endif
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
fread(words,sizeof(int),raw_width/3,ifp);
for (col=0; col < raw_width-2; col+=3)
{
bits = words[col/3];
FORC3 {
orow = row;
if ((ocol = col+c-4) < 0)
{
ocol += raw_width;
if ((orow -= 2) < 0)
orow += raw_height;
}
RAW(orow,ocol) = curve[bits >> (10*c+2) & 0x3ff];
}
}
#else
for (col=0; col < raw_width-2; col+=3) {
bits = get4();
FORC3 {
orow = row;
if ((ocol = col+c-4) < 0) {
ocol += raw_width;
if ((orow -= 2) < 0)
orow += raw_height;
}
RAW(orow,ocol) = curve[bits >> (10*c+2) & 0x3ff];
}
}
#endif
}
#ifdef LIBRAW_LIBRARY_BUILD
free(words);
#endif
maximum = curve[0x3ff];
}
unsigned CLASS pana_bits (int nbits)
{
#ifndef LIBRAW_NOTHREADS
#define buf tls->pana_bits.buf
#define vbits tls->pana_bits.vbits
#else
static uchar buf[0x4000];
static int vbits;
#endif
int byte;
if (!nbits) return vbits=0;
if (!vbits) {
fread (buf+load_flags, 1, 0x4000-load_flags, ifp);
fread (buf, 1, load_flags, ifp);
}
vbits = (vbits - nbits) & 0x1ffff;
byte = vbits >> 3 ^ 0x3ff0;
return (buf[byte] | buf[byte+1] << 8) >> (vbits & 7) & ~((~0u) << nbits);
#ifndef LIBRAW_NOTHREADS
#undef buf
#undef vbits
#endif
}
void CLASS panasonic_load_raw()
{
int row, col, i, j, sh=0, pred[2], nonz[2];
pana_bits(0);
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
if ((i = col % 14) == 0)
pred[0] = pred[1] = nonz[0] = nonz[1] = 0;
if (i % 3 == 2) sh = 4 >> (3 - pana_bits(2));
if (nonz[i & 1]) {
if ((j = pana_bits(8))) {
if ((pred[i & 1] -= 0x80 << sh) < 0 || sh == 4)
pred[i & 1] &= ~((~0u) << sh);
pred[i & 1] += j << sh;
}
} else if ((nonz[i & 1] = pana_bits(8)) || i > 11)
pred[i & 1] = nonz[i & 1] << 4 | pana_bits(4);
if ((RAW(row,col) = pred[col & 1]) > 4098 && col < width) derror();
}
}
}
void CLASS olympus_load_raw()
{
ushort huff[4096];
int row, col, nbits, sign, low, high, i, c, w, n, nw;
int acarry[2][3], *carry, pred, diff;
huff[n=0] = 0xc0c;
for (i=12; i--; )
FORC(2048 >> i) huff[++n] = (i+1) << 8 | i;
fseek (ifp, 7, SEEK_CUR);
getbits(-1);
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
memset (acarry, 0, sizeof acarry);
for (col=0; col < raw_width; col++) {
carry = acarry[col & 1];
i = 2 * (carry[2] < 3);
for (nbits=2+i; (ushort) carry[0] >> (nbits+i); nbits++);
low = (sign = getbits(3)) & 3;
sign = sign << 29 >> 31;
if ((high = getbithuff(12,huff)) == 12)
high = getbits(16-nbits) >> 1;
carry[0] = (high << nbits) | getbits(nbits);
diff = (carry[0] ^ sign) + carry[1];
carry[1] = (diff*3 + carry[1]) >> 5;
carry[2] = carry[0] > 16 ? 0 : carry[2]+1;
if (col >= width) continue;
if (row < 2 && col < 2) pred = 0;
else if (row < 2) pred = RAW(row,col-2);
else if (col < 2) pred = RAW(row-2,col);
else {
w = RAW(row,col-2);
n = RAW(row-2,col);
nw = RAW(row-2,col-2);
if ((w < nw && nw < n) || (n < nw && nw < w)) {
if (ABS(w-nw) > 32 || ABS(n-nw) > 32)
pred = w + n - nw;
else pred = (w + n) >> 1;
} else pred = ABS(w-nw) > ABS(n-nw) ? w : n;
}
if ((RAW(row,col) = pred + ((diff << 2) | low)) >> 12) derror();
}
}
}
void CLASS minolta_rd175_load_raw()
{
uchar pixel[768];
unsigned irow, box, row, col;
for (irow=0; irow < 1481; irow++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (pixel, 1, 768, ifp) < 768) derror();
box = irow / 82;
row = irow % 82 * 12 + ((box < 12) ? box | 1 : (box-12)*2);
switch (irow) {
case 1477: case 1479: continue;
case 1476: row = 984; break;
case 1480: row = 985; break;
case 1478: row = 985; box = 1;
}
if ((box < 12) && (box & 1)) {
for (col=0; col < 1533; col++, row ^= 1)
if (col != 1) RAW(row,col) = (col+1) & 2 ?
pixel[col/2-1] + pixel[col/2+1] : pixel[col/2] << 1;
RAW(row,1) = pixel[1] << 1;
RAW(row,1533) = pixel[765] << 1;
} else
for (col=row & 1; col < 1534; col+=2)
RAW(row,col) = pixel[col/2] << 1;
}
maximum = 0xff << 1;
}
void CLASS quicktake_100_load_raw()
{
uchar pixel[484][644];
static const short gstep[16] =
{ -89,-60,-44,-32,-22,-15,-8,-2,2,8,15,22,32,44,60,89 };
static const short rstep[6][4] =
{ { -3,-1,1,3 }, { -5,-1,1,5 }, { -8,-2,2,8 },
{ -13,-3,3,13 }, { -19,-4,4,19 }, { -28,-6,6,28 } };
static const short t_curve[256] =
{ 0,1,2,3,4,5,6,7,8,9,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,
28,29,30,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,53,
54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,74,75,76,77,78,
79,80,81,82,83,84,86,88,90,92,94,97,99,101,103,105,107,110,112,114,116,
118,120,123,125,127,129,131,134,136,138,140,142,144,147,149,151,153,155,
158,160,162,164,166,168,171,173,175,177,179,181,184,186,188,190,192,195,
197,199,201,203,205,208,210,212,214,216,218,221,223,226,230,235,239,244,
248,252,257,261,265,270,274,278,283,287,291,296,300,305,309,313,318,322,
326,331,335,339,344,348,352,357,361,365,370,374,379,383,387,392,396,400,
405,409,413,418,422,426,431,435,440,444,448,453,457,461,466,470,474,479,
483,487,492,496,500,508,519,531,542,553,564,575,587,598,609,620,631,643,
654,665,676,687,698,710,721,732,743,754,766,777,788,799,810,822,833,844,
855,866,878,889,900,911,922,933,945,956,967,978,989,1001,1012,1023 };
int rb, row, col, sharp, val=0;
getbits(-1);
memset (pixel, 0x80, sizeof pixel);
for (row=2; row < height+2; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=2+(row & 1); col < width+2; col+=2) {
val = ((pixel[row-1][col-1] + 2*pixel[row-1][col+1] +
pixel[row][col-2]) >> 2) + gstep[getbits(4)];
pixel[row][col] = val = LIM(val,0,255);
if (col < 4)
pixel[row][col-2] = pixel[row+1][~row & 1] = val;
if (row == 2)
pixel[row-1][col+1] = pixel[row-1][col+3] = val;
}
pixel[row][col] = val;
}
for (rb=0; rb < 2; rb++)
for (row=2+rb; row < height+2; row+=2)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=3-(row & 1); col < width+2; col+=2) {
if (row < 4 || col < 4) sharp = 2;
else {
val = ABS(pixel[row-2][col] - pixel[row][col-2])
+ ABS(pixel[row-2][col] - pixel[row-2][col-2])
+ ABS(pixel[row][col-2] - pixel[row-2][col-2]);
sharp = val < 4 ? 0 : val < 8 ? 1 : val < 16 ? 2 :
val < 32 ? 3 : val < 48 ? 4 : 5;
}
val = ((pixel[row-2][col] + pixel[row][col-2]) >> 1)
+ rstep[sharp][getbits(2)];
pixel[row][col] = val = LIM(val,0,255);
if (row < 4) pixel[row-2][col+2] = val;
if (col < 4) pixel[row+2][col-2] = val;
}
}
for (row=2; row < height+2; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=3-(row & 1); col < width+2; col+=2) {
val = ((pixel[row][col-1] + (pixel[row][col] << 2) +
pixel[row][col+1]) >> 1) - 0x100;
pixel[row][col] = LIM(val,0,255);
}
}
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col++)
RAW(row,col) = t_curve[pixel[row+2][col+2]];
}
maximum = 0x3ff;
}
#define radc_token(tree) ((signed char) getbithuff(8,huff[tree]))
#define FORYX for (y=1; y < 3; y++) for (x=col+1; x >= col; x--)
#define PREDICTOR (c ? (buf[c][y-1][x] + buf[c][y][x+1]) / 2 \
: (buf[c][y-1][x+1] + 2*buf[c][y-1][x] + buf[c][y][x+1]) / 4)
#ifdef __GNUC__
# if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)
# pragma GCC optimize("no-aggressive-loop-optimizations")
# endif
#endif
void CLASS kodak_radc_load_raw()
{
static const char src[] = {
1,1, 2,3, 3,4, 4,2, 5,7, 6,5, 7,6, 7,8,
1,0, 2,1, 3,3, 4,4, 5,2, 6,7, 7,6, 8,5, 8,8,
2,1, 2,3, 3,0, 3,2, 3,4, 4,6, 5,5, 6,7, 6,8,
2,0, 2,1, 2,3, 3,2, 4,4, 5,6, 6,7, 7,5, 7,8,
2,1, 2,4, 3,0, 3,2, 3,3, 4,7, 5,5, 6,6, 6,8,
2,3, 3,1, 3,2, 3,4, 3,5, 3,6, 4,7, 5,0, 5,8,
2,3, 2,6, 3,0, 3,1, 4,4, 4,5, 4,7, 5,2, 5,8,
2,4, 2,7, 3,3, 3,6, 4,1, 4,2, 4,5, 5,0, 5,8,
2,6, 3,1, 3,3, 3,5, 3,7, 3,8, 4,0, 5,2, 5,4,
2,0, 2,1, 3,2, 3,3, 4,4, 4,5, 5,6, 5,7, 4,8,
1,0, 2,2, 2,-2,
1,-3, 1,3,
2,-17, 2,-5, 2,5, 2,17,
2,-7, 2,2, 2,9, 2,18,
2,-18, 2,-9, 2,-2, 2,7,
2,-28, 2,28, 3,-49, 3,-9, 3,9, 4,49, 5,-79, 5,79,
2,-1, 2,13, 2,26, 3,39, 4,-16, 5,55, 6,-37, 6,76,
2,-26, 2,-13, 2,1, 3,-39, 4,16, 5,-55, 6,-76, 6,37
};
ushort huff[19][256];
int row, col, tree, nreps, rep, step, i, c, s, r, x, y, val;
short last[3] = { 16,16,16 }, mul[3], buf[3][3][386];
static const ushort pt[] =
{ 0,0, 1280,1344, 2320,3616, 3328,8000, 4095,16383, 65535,16383 };
for (i=2; i < 12; i+=2)
for (c=pt[i-2]; c <= pt[i]; c++)
curve[c] = (float)
(c-pt[i-2]) / (pt[i]-pt[i-2]) * (pt[i+1]-pt[i-1]) + pt[i-1] + 0.5;
for (s=i=0; i < sizeof src; i+=2)
FORC(256 >> src[i])
((ushort *)huff)[s++] = src[i] << 8 | (uchar) src[i+1];
s = kodak_cbpp == 243 ? 2 : 3;
FORC(256) huff[18][c] = (8-s) << 8 | c >> s << s | 1 << (s-1);
getbits(-1);
for (i=0; i < sizeof(buf)/sizeof(short); i++)
((short *)buf)[i] = 2048;
for (row=0; row < height; row+=4) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
FORC3 mul[c] = getbits(6);
FORC3 {
val = ((0x1000000/last[c] + 0x7ff) >> 12) * mul[c];
s = val > 65564 ? 10:12;
x = ~((~0u) << (s-1));
val <<= 12-s;
for (i=0; i < sizeof(buf[0])/sizeof(short); i++)
((short *)buf[c])[i] = (((short *)buf[c])[i] * val + x) >> s;
last[c] = mul[c];
for (r=0; r <= !c; r++) {
buf[c][1][width/2] = buf[c][2][width/2] = mul[c] << 7;
for (tree=1, col=width/2; col > 0; ) {
if ((tree = radc_token(tree))) {
col -= 2;
if (tree == 8)
FORYX buf[c][y][x] = (uchar) radc_token(18) * mul[c];
else
FORYX buf[c][y][x] = radc_token(tree+10) * 16 + PREDICTOR;
} else
do {
nreps = (col > 2) ? radc_token(9) + 1 : 1;
for (rep=0; rep < 8 && rep < nreps && col > 0; rep++) {
col -= 2;
FORYX buf[c][y][x] = PREDICTOR;
if (rep & 1) {
step = radc_token(10) << 4;
FORYX buf[c][y][x] += step;
}
}
} while (nreps == 9);
}
for (y=0; y < 2; y++)
for (x=0; x < width/2; x++) {
val = (buf[c][y+1][x] << 4) / mul[c];
if (val < 0) val = 0;
if (c) RAW(row+y*2+c-1,x*2+2-c) = val;
else RAW(row+r*2+y,x*2+y) = val;
}
memcpy (buf[c][0]+!c, buf[c][2], sizeof buf[c][0]-2*!c);
}
}
for (y=row; y < row+4; y++)
for (x=0; x < width; x++)
if ((x+y) & 1) {
r = x ? x-1 : x+1;
s = x+1 < width ? x+1 : x-1;
val = (RAW(y,x)-2048)*2 + (RAW(y,r)+RAW(y,s))/2;
if (val < 0) val = 0;
RAW(y,x) = val;
}
}
for (i=0; i < height*width; i++)
raw_image[i] = curve[raw_image[i]];
maximum = 0x3fff;
}
#undef FORYX
#undef PREDICTOR
#ifdef NO_JPEG
void CLASS kodak_jpeg_load_raw() {}
void CLASS lossy_dng_load_raw() {}
#else
#ifndef LIBRAW_LIBRARY_BUILD
METHODDEF(boolean)
fill_input_buffer (j_decompress_ptr cinfo)
{
static uchar jpeg_buffer[4096];
size_t nbytes;
nbytes = fread (jpeg_buffer, 1, 4096, ifp);
swab (jpeg_buffer, jpeg_buffer, nbytes);
cinfo->src->next_input_byte = jpeg_buffer;
cinfo->src->bytes_in_buffer = nbytes;
return TRUE;
}
void CLASS kodak_jpeg_load_raw()
{
struct jpeg_decompress_struct cinfo;
struct jpeg_error_mgr jerr;
JSAMPARRAY buf;
JSAMPLE (*pixel)[3];
int row, col;
cinfo.err = jpeg_std_error (&jerr);
jpeg_create_decompress (&cinfo);
jpeg_stdio_src (&cinfo, ifp);
cinfo.src->fill_input_buffer = fill_input_buffer;
jpeg_read_header (&cinfo, TRUE);
jpeg_start_decompress (&cinfo);
if ((cinfo.output_width != width ) ||
(cinfo.output_height*2 != height ) ||
(cinfo.output_components != 3 )) {
fprintf (stderr,_("%s: incorrect JPEG dimensions\n"), ifname);
jpeg_destroy_decompress (&cinfo);
longjmp (failure, 3);
}
buf = (*cinfo.mem->alloc_sarray)
((j_common_ptr) &cinfo, JPOOL_IMAGE, width*3, 1);
while (cinfo.output_scanline < cinfo.output_height) {
row = cinfo.output_scanline * 2;
jpeg_read_scanlines (&cinfo, buf, 1);
pixel = (JSAMPLE (*)[3]) buf[0];
for (col=0; col < width; col+=2) {
RAW(row+0,col+0) = pixel[col+0][1] << 1;
RAW(row+1,col+1) = pixel[col+1][1] << 1;
RAW(row+0,col+1) = pixel[col][0] + pixel[col+1][0];
RAW(row+1,col+0) = pixel[col][2] + pixel[col+1][2];
}
}
jpeg_finish_decompress (&cinfo);
jpeg_destroy_decompress (&cinfo);
maximum = 0xff << 1;
}
#else
struct jpegErrorManager {
struct jpeg_error_mgr pub;
};
static void jpegErrorExit (j_common_ptr cinfo)
{
jpegErrorManager* myerr = (jpegErrorManager*) cinfo->err;
throw LIBRAW_EXCEPTION_DECODE_JPEG;
}
// LibRaw's Kodak_jpeg_load_raw
void CLASS kodak_jpeg_load_raw()
{
if(data_size < 1)
throw LIBRAW_EXCEPTION_DECODE_JPEG;
int row, col;
jpegErrorManager jerr;
struct jpeg_decompress_struct cinfo;
cinfo.err = jpeg_std_error(&jerr.pub);
jerr.pub.error_exit = jpegErrorExit;
unsigned char *jpg_buf = (unsigned char *)malloc(data_size);
merror(jpg_buf,"kodak_jpeg_load_raw");
unsigned char *pixel_buf = (unsigned char*) malloc(width*3);
jpeg_create_decompress (&cinfo);
merror(pixel_buf,"kodak_jpeg_load_raw");
fread(jpg_buf,data_size,1,ifp);
swab ((char*)jpg_buf, (char*)jpg_buf, data_size);
try
{
jpeg_mem_src(&cinfo, jpg_buf, data_size);
int rc = jpeg_read_header(&cinfo, TRUE);
if(rc!=1)
throw LIBRAW_EXCEPTION_DECODE_JPEG;
jpeg_start_decompress (&cinfo);
if ((cinfo.output_width != width ) ||
(cinfo.output_height*2 != height ) ||
(cinfo.output_components != 3 ))
{
throw LIBRAW_EXCEPTION_DECODE_JPEG;
}
unsigned char *buf[1];
buf[0] = pixel_buf;
while (cinfo.output_scanline < cinfo.output_height)
{
checkCancel();
row = cinfo.output_scanline * 2;
jpeg_read_scanlines (&cinfo, buf, 1);
unsigned char (*pixel)[3] = (unsigned char (*)[3]) buf[0];
for (col=0; col < width; col+=2) {
RAW(row+0,col+0) = pixel[col+0][1] << 1;
RAW(row+1,col+1) = pixel[col+1][1] << 1;
RAW(row+0,col+1) = pixel[col][0] + pixel[col+1][0];
RAW(row+1,col+0) = pixel[col][2] + pixel[col+1][2];
}
}
}
catch (...)
{
jpeg_finish_decompress (&cinfo);
jpeg_destroy_decompress (&cinfo);
free(jpg_buf);
free(pixel_buf);
throw;
}
jpeg_finish_decompress (&cinfo);
jpeg_destroy_decompress (&cinfo);
free(jpg_buf);
free(pixel_buf);
maximum = 0xff << 1;
}
#endif
#ifndef LIBRAW_LIBRARY_BUILD
void CLASS gamma_curve (double pwr, double ts, int mode, int imax);
#endif
void CLASS lossy_dng_load_raw()
{
struct jpeg_decompress_struct cinfo;
struct jpeg_error_mgr jerr;
JSAMPARRAY buf;
JSAMPLE (*pixel)[3];
unsigned sorder=order, ntags, opcode, deg, i, j, c;
unsigned save=data_offset-4, trow=0, tcol=0, row, col;
ushort cur[3][256];
double coeff[9], tot;
if (meta_offset) {
fseek (ifp, meta_offset, SEEK_SET);
order = 0x4d4d;
ntags = get4();
while (ntags--) {
opcode = get4(); get4(); get4();
if (opcode != 8)
{ fseek (ifp, get4(), SEEK_CUR); continue; }
fseek (ifp, 20, SEEK_CUR);
if ((c = get4()) > 2) break;
fseek (ifp, 12, SEEK_CUR);
if ((deg = get4()) > 8) break;
for (i=0; i <= deg && i < 9; i++)
coeff[i] = getreal(12);
for (i=0; i < 256; i++) {
for (tot=j=0; j <= deg; j++)
tot += coeff[j] * pow(i/255.0, (int)j);
cur[c][i] = tot*0xffff;
}
}
order = sorder;
} else {
gamma_curve (1/2.4, 12.92, 1, 255);
FORC3 memcpy (cur[c], curve, sizeof cur[0]);
}
cinfo.err = jpeg_std_error (&jerr);
jpeg_create_decompress (&cinfo);
while (trow < raw_height) {
fseek (ifp, save+=4, SEEK_SET);
if (tile_length < INT_MAX)
fseek (ifp, get4(), SEEK_SET);
#ifdef LIBRAW_LIBRARY_BUILD
if(libraw_internal_data.internal_data.input->jpeg_src(&cinfo) == -1)
{
jpeg_destroy_decompress(&cinfo);
throw LIBRAW_EXCEPTION_DECODE_JPEG;
}
#else
jpeg_stdio_src (&cinfo, ifp);
#endif
jpeg_read_header (&cinfo, TRUE);
jpeg_start_decompress (&cinfo);
buf = (*cinfo.mem->alloc_sarray)
((j_common_ptr) &cinfo, JPOOL_IMAGE, cinfo.output_width*3, 1);
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
while (cinfo.output_scanline < cinfo.output_height &&
(row = trow + cinfo.output_scanline) < height) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
jpeg_read_scanlines (&cinfo, buf, 1);
pixel = (JSAMPLE (*)[3]) buf[0];
for (col=0; col < cinfo.output_width && tcol+col < width; col++) {
FORC3 image[row*width+tcol+col][c] = cur[c][pixel[col][c]];
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
jpeg_destroy_decompress (&cinfo);
throw;
}
#endif
jpeg_abort_decompress (&cinfo);
if ((tcol += tile_width) >= raw_width)
trow += tile_length + (tcol = 0);
}
jpeg_destroy_decompress (&cinfo);
maximum = 0xffff;
}
#endif
void CLASS kodak_dc120_load_raw()
{
static const int mul[4] = { 162, 192, 187, 92 };
static const int add[4] = { 0, 636, 424, 212 };
uchar pixel[848];
int row, shift, col;
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (pixel, 1, 848, ifp) < 848) derror();
shift = row * mul[row & 3] + add[row & 3];
for (col=0; col < width; col++)
RAW(row,col) = (ushort) pixel[(col + shift) % 848];
}
maximum = 0xff;
}
void CLASS eight_bit_load_raw()
{
uchar *pixel;
unsigned row, col;
pixel = (uchar *) calloc (raw_width, sizeof *pixel);
merror (pixel, "eight_bit_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (pixel, 1, raw_width, ifp) < raw_width) derror();
for (col=0; col < raw_width; col++)
RAW(row,col) = curve[pixel[col]];
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (pixel);
throw;
}
#endif
free (pixel);
maximum = curve[0xff];
}
void CLASS kodak_c330_load_raw()
{
uchar *pixel;
int row, col, y, cb, cr, rgb[3], c;
pixel = (uchar *) calloc (raw_width, 2*sizeof *pixel);
merror (pixel, "kodak_c330_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (fread (pixel, raw_width, 2, ifp) < 2) derror();
if (load_flags && (row & 31) == 31)
fseek (ifp, raw_width*32, SEEK_CUR);
for (col=0; col < width; col++) {
y = pixel[col*2];
cb = pixel[(col*2 & -4) | 1] - 128;
cr = pixel[(col*2 & -4) | 3] - 128;
rgb[1] = y - ((cb + cr + 2) >> 2);
rgb[2] = rgb[1] + cb;
rgb[0] = rgb[1] + cr;
FORC3 image[row*width+col][c] = curve[LIM(rgb[c],0,255)];
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (pixel);
throw;
}
#endif
free (pixel);
maximum = curve[0xff];
}
void CLASS kodak_c603_load_raw()
{
uchar *pixel;
int row, col, y, cb, cr, rgb[3], c;
pixel = (uchar *) calloc (raw_width, 3*sizeof *pixel);
merror (pixel, "kodak_c603_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if (~row & 1)
if (fread (pixel, raw_width, 3, ifp) < 3) derror();
for (col=0; col < width; col++) {
y = pixel[width*2*(row & 1) + col];
cb = pixel[width + (col & -2)] - 128;
cr = pixel[width + (col & -2)+1] - 128;
rgb[1] = y - ((cb + cr + 2) >> 2);
rgb[2] = rgb[1] + cb;
rgb[0] = rgb[1] + cr;
FORC3 image[row*width+col][c] = curve[LIM(rgb[c],0,255)];
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (pixel);
throw;
}
#endif
free (pixel);
maximum = curve[0xff];
}
void CLASS kodak_262_load_raw()
{
static const uchar kodak_tree[2][26] =
{ { 0,1,5,1,1,2,0,0,0,0,0,0,0,0,0,0, 0,1,2,3,4,5,6,7,8,9 },
{ 0,3,1,1,1,1,1,2,0,0,0,0,0,0,0,0, 0,1,2,3,4,5,6,7,8,9 } };
ushort *huff[2];
uchar *pixel;
int *strip, ns, c, row, col, chess, pi=0, pi1, pi2, pred, val;
FORC(2) huff[c] = make_decoder (kodak_tree[c]);
ns = (raw_height+63) >> 5;
pixel = (uchar *) malloc (raw_width*32 + ns*4);
merror (pixel, "kodak_262_load_raw()");
strip = (int *) (pixel + raw_width*32);
order = 0x4d4d;
FORC(ns) strip[c] = get4();
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
if ((row & 31) == 0) {
fseek (ifp, strip[row >> 5], SEEK_SET);
getbits(-1);
pi = 0;
}
for (col=0; col < raw_width; col++) {
chess = (row + col) & 1;
pi1 = chess ? pi-2 : pi-raw_width-1;
pi2 = chess ? pi-2*raw_width : pi-raw_width+1;
if (col <= chess) pi1 = -1;
if (pi1 < 0) pi1 = pi2;
if (pi2 < 0) pi2 = pi1;
if (pi1 < 0 && col > 1) pi1 = pi2 = pi-2;
pred = (pi1 < 0) ? 0 : (pixel[pi1] + pixel[pi2]) >> 1;
pixel[pi] = val = pred + ljpeg_diff (huff[chess]);
if (val >> 8) derror();
val = curve[pixel[pi++]];
RAW(row,col) = val;
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (pixel);
throw;
}
#endif
free (pixel);
FORC(2) free (huff[c]);
}
int CLASS kodak_65000_decode (short *out, int bsize)
{
uchar c, blen[768];
ushort raw[6];
INT64 bitbuf=0;
int save, bits=0, i, j, len, diff;
save = ftell(ifp);
bsize = (bsize + 3) & -4;
for (i=0; i < bsize; i+=2) {
c = fgetc(ifp);
if ((blen[i ] = c & 15) > 12 ||
(blen[i+1] = c >> 4) > 12 ) {
fseek (ifp, save, SEEK_SET);
for (i=0; i < bsize; i+=8) {
read_shorts (raw, 6);
out[i ] = raw[0] >> 12 << 8 | raw[2] >> 12 << 4 | raw[4] >> 12;
out[i+1] = raw[1] >> 12 << 8 | raw[3] >> 12 << 4 | raw[5] >> 12;
for (j=0; j < 6; j++)
out[i+2+j] = raw[j] & 0xfff;
}
return 1;
}
}
if ((bsize & 7) == 4) {
bitbuf = fgetc(ifp) << 8;
bitbuf += fgetc(ifp);
bits = 16;
}
for (i=0; i < bsize; i++) {
len = blen[i];
if (bits < len) {
for (j=0; j < 32; j+=8)
bitbuf += (INT64) fgetc(ifp) << (bits+(j^8));
bits += 32;
}
diff = bitbuf & (0xffff >> (16-len));
bitbuf >>= len;
bits -= len;
if ((diff & (1 << (len-1))) == 0)
diff -= (1 << len) - 1;
out[i] = diff;
}
return 0;
}
void CLASS kodak_65000_load_raw()
{
short buf[256];
int row, col, len, pred[2], ret, i;
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col+=256) {
pred[0] = pred[1] = 0;
len = MIN (256, width-col);
ret = kodak_65000_decode (buf, len);
for (i=0; i < len; i++)
if ((RAW(row,col+i) = curve[ret ? buf[i] :
(pred[i & 1] += buf[i])]) >> 12) derror();
}
}
}
void CLASS kodak_ycbcr_load_raw()
{
short buf[384], *bp;
int row, col, len, c, i, j, k, y[2][2], cb, cr, rgb[3];
ushort *ip;
if (!image) return;
unsigned int bits = (load_flags && load_flags > 9 && load_flags < 17)?load_flags:10;
for (row=0; row < height; row+=2)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col+=128) {
len = MIN (128, width-col);
kodak_65000_decode (buf, len*3);
y[0][1] = y[1][1] = cb = cr = 0;
for (bp=buf, i=0; i < len; i+=2, bp+=2) {
cb += bp[4];
cr += bp[5];
rgb[1] = -((cb + cr + 2) >> 2);
rgb[2] = rgb[1] + cb;
rgb[0] = rgb[1] + cr;
for (j=0; j < 2; j++)
for (k=0; k < 2; k++) {
if ((y[j][k] = y[j][k^1] + *bp++) >> bits) derror();
ip = image[(row+j)*width + col+i+k];
FORC3 ip[c] = curve[LIM(y[j][k]+rgb[c], 0, 0xfff)];
}
}
}
}
}
void CLASS kodak_rgb_load_raw()
{
short buf[768], *bp;
int row, col, len, c, i, rgb[3],ret;
ushort *ip=image[0];
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col+=256) {
len = MIN (256, width-col);
ret = kodak_65000_decode (buf, len*3);
memset (rgb, 0, sizeof rgb);
for (bp=buf, i=0; i < len; i++, ip+=4)
#ifdef LIBRAW_LIBRARY_BUILD
if(load_flags == 12)
{
FORC3 ip[c] = ret ? (*bp++) : (rgb[c] += *bp++);
}
else
#endif
FORC3 if ((ip[c] = ret ? (*bp++) : (rgb[c] += *bp++)) >> 12) derror();
}
}
}
void CLASS kodak_thumb_load_raw()
{
int row, col;
colors = thumb_misc >> 5;
for (row=0; row < height; row++)
for (col=0; col < width; col++)
read_shorts (image[row*width+col], colors);
maximum = (1 << (thumb_misc & 31)) - 1;
}
void CLASS sony_decrypt (unsigned *data, int len, int start, int key)
{
#ifndef LIBRAW_NOTHREADS
#define pad tls->sony_decrypt.pad
#define p tls->sony_decrypt.p
#else
static unsigned pad[128], p;
#endif
if (start) {
for (p=0; p < 4; p++)
pad[p] = key = key * 48828125 + 1;
pad[3] = pad[3] << 1 | (pad[0]^pad[2]) >> 31;
for (p=4; p < 127; p++)
pad[p] = (pad[p-4]^pad[p-2]) << 1 | (pad[p-3]^pad[p-1]) >> 31;
for (p=0; p < 127; p++)
pad[p] = htonl(pad[p]);
}
while (len--)
{
*data++ ^= pad[p & 127] = pad[(p+1) & 127] ^ pad[(p+65) & 127];
p++;
}
#ifndef LIBRAW_NOTHREADS
#undef pad
#undef p
#endif
}
void CLASS sony_load_raw()
{
uchar head[40];
ushort *pixel;
unsigned i, key, row, col;
fseek (ifp, 200896, SEEK_SET);
fseek (ifp, (unsigned) fgetc(ifp)*4 - 1, SEEK_CUR);
order = 0x4d4d;
key = get4();
fseek (ifp, 164600, SEEK_SET);
fread (head, 1, 40, ifp);
sony_decrypt ((unsigned *) head, 10, 1, key);
for (i=26; i-- > 22; )
key = key << 8 | head[i];
fseek (ifp, data_offset, SEEK_SET);
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
pixel = raw_image + row*raw_width;
if (fread (pixel, 2, raw_width, ifp) < raw_width) derror();
sony_decrypt ((unsigned *) pixel, raw_width/2, !row, key);
for (col=0; col < raw_width; col++)
if ((pixel[col] = ntohs(pixel[col])) >> 14) derror();
}
maximum = 0x3ff0;
}
void CLASS sony_arw_load_raw()
{
ushort huff[32770];
static const ushort tab[18] =
{ 0xf11,0xf10,0xe0f,0xd0e,0xc0d,0xb0c,0xa0b,0x90a,0x809,
0x708,0x607,0x506,0x405,0x304,0x303,0x300,0x202,0x201 };
int i, c, n, col, row, sum=0;
huff[0] = 15;
for (n=i=0; i < 18; i++)
FORC(32768 >> (tab[i] >> 8)) huff[++n] = tab[i];
getbits(-1);
for (col = raw_width; col--; )
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (row=0; row < raw_height+1; row+=2) {
if (row == raw_height) row = 1;
if ((sum += ljpeg_diff(huff)) >> 12) derror();
if (row < height) RAW(row,col) = sum;
}
}
}
void CLASS sony_arw2_load_raw()
{
uchar *data, *dp;
ushort pix[16];
int row, col, val, max, min, imax, imin, sh, bit, i;
data = (uchar *) malloc (raw_width+1);
merror (data, "sony_arw2_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
try {
#endif
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
fread (data, 1, raw_width, ifp);
for (dp=data, col=0; col < raw_width-30; dp+=16) {
max = 0x7ff & (val = sget4(dp));
min = 0x7ff & val >> 11;
imax = 0x0f & val >> 22;
imin = 0x0f & val >> 26;
for (sh=0; sh < 4 && 0x80 << sh <= max-min; sh++);
#ifdef LIBRAW_LIBRARY_BUILD
/* flag checks if outside of loop */
if(! (imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_ALLFLAGS) // no flag set
|| (imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_DELTATOVALUE)
)
{
for (bit=30, i=0; i < 16; i++)
if (i == imax) pix[i] = max;
else if (i == imin) pix[i] = min;
else {
pix[i] = ((sget2(dp+(bit >> 3)) >> (bit & 7) & 0x7f) << sh) + min;
if (pix[i] > 0x7ff) pix[i] = 0x7ff;
bit += 7;
}
}
else if(imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_BASEONLY)
{
for (bit=30, i=0; i < 16; i++)
if (i == imax) pix[i] = max;
else if (i == imin) pix[i] = min;
else pix[i]=0;
}
else if(imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_DELTAONLY)
{
for (bit=30, i=0; i < 16; i++)
if (i == imax) pix[i] = 0;
else if (i == imin) pix[i] = 0;
else {
pix[i] = ((sget2(dp+(bit >> 3)) >> (bit & 7) & 0x7f) << sh) + min;
if (pix[i] > 0x7ff) pix[i] = 0x7ff;
bit += 7;
}
}
else if(imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_DELTAZEROBASE)
{
for (bit=30, i=0; i < 16; i++)
if (i == imax) pix[i] = 0;
else if (i == imin) pix[i] = 0;
else {
pix[i] = ((sget2(dp+(bit >> 3)) >> (bit & 7) & 0x7f) << sh);
if (pix[i] > 0x7ff) pix[i] = 0x7ff;
bit += 7;
}
}
#else
/* unaltered dcraw processing */
for (bit=30, i=0; i < 16; i++)
if (i == imax) pix[i] = max;
else if (i == imin) pix[i] = min;
else {
pix[i] = ((sget2(dp+(bit >> 3)) >> (bit & 7) & 0x7f) << sh) + min;
if (pix[i] > 0x7ff) pix[i] = 0x7ff;
bit += 7;
}
#endif
#ifdef LIBRAW_LIBRARY_BUILD
if(imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_DELTATOVALUE)
{
for (i=0; i < 16; i++, col+=2)
{
unsigned slope = pix[i] < 1001? 2 : curve[pix[i]<<1]-curve[(pix[i]<<1)-2];
unsigned step = 1 << sh;
RAW(row,col)=curve[pix[i]<<1]>black+imgdata.params.sony_arw2_posterization_thr?
LIM(((slope*step*1000)/(curve[pix[i]<<1]-black)),0,10000):0;
}
}
else
{
for (i=0; i < 16; i++, col+=2)
RAW(row,col) = curve[pix[i] << 1];
}
#else
for (i=0; i < 16; i++, col+=2)
RAW(row,col) = curve[pix[i] << 1] >> 2;
#endif
col -= col & 1 ? 1:31;
}
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch(...) {
free (data);
throw;
}
if(imgdata.params.raw_processing_options & LIBRAW_PROCESSING_SONYARW2_DELTATOVALUE)
maximum=10000;
#endif
free (data);
}
void CLASS samsung_load_raw()
{
int row, col, c, i, dir, op[4], len[4];
order = 0x4949;
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
fseek (ifp, strip_offset+row*4, SEEK_SET);
fseek (ifp, data_offset+get4(), SEEK_SET);
ph1_bits(-1);
FORC4 len[c] = row < 2 ? 7:4;
for (col=0; col < raw_width; col+=16) {
dir = ph1_bits(1);
FORC4 op[c] = ph1_bits(2);
FORC4 switch (op[c]) {
case 3: len[c] = ph1_bits(4); break;
case 2: len[c]--; break;
case 1: len[c]++;
}
for (c=0; c < 16; c+=2) {
i = len[((c & 1) << 1) | (c >> 3)];
RAW(row,col+c) = ((signed) ph1_bits(i) << (32-i) >> (32-i)) +
(dir ? RAW(row+(~c | -2),col+c) : col ? RAW(row,col+(c | -2)) : 128);
if (c == 14) c = -1;
}
}
}
for (row=0; row < raw_height-1; row+=2)
for (col=0; col < raw_width-1; col+=2)
SWAP (RAW(row,col+1), RAW(row+1,col));
}
void CLASS samsung2_load_raw()
{
static const ushort tab[14] =
{ 0x304,0x307,0x206,0x205,0x403,0x600,0x709,
0x80a,0x90b,0xa0c,0xa0d,0x501,0x408,0x402 };
ushort huff[1026], vpred[2][2] = {{0,0},{0,0}}, hpred[2];
int i, c, n, row, col, diff;
huff[0] = 10;
for (n=i=0; i < 14; i++)
FORC(1024 >> (tab[i] >> 8)) huff[++n] = tab[i];
getbits(-1);
for (row=0; row < raw_height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < raw_width; col++) {
diff = ljpeg_diff (huff);
if (col < 2) hpred[col] = vpred[row & 1][col] += diff;
else hpred[col & 1] += diff;
RAW(row,col) = hpred[col & 1];
if (hpred[col & 1] >> tiff_bps) derror();
}
}
}
void CLASS samsung3_load_raw()
{
int opt, init, mag, pmode, row, tab, col, pred, diff, i, c;
ushort lent[3][2], len[4], *prow[2];
order = 0x4949;
fseek (ifp, 9, SEEK_CUR);
opt = fgetc(ifp);
init = (get2(),get2());
for (row=0; row < raw_height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
fseek (ifp, (data_offset-ftell(ifp)) & 15, SEEK_CUR);
ph1_bits(-1);
mag = 0; pmode = 7;
FORC(6) ((ushort *)lent)[c] = row < 2 ? 7:4;
prow[ row & 1] = &RAW(row-1,1-((row & 1) << 1)); // green
prow[~row & 1] = &RAW(row-2,0); // red and blue
for (tab=0; tab+15 < raw_width; tab+=16) {
if (~opt & 4 && !(tab & 63)) {
i = ph1_bits(2);
mag = i < 3 ? mag-'2'+"204"[i] : ph1_bits(12);
}
if (opt & 2)
pmode = 7 - 4*ph1_bits(1);
else if (!ph1_bits(1))
pmode = ph1_bits(3);
if (opt & 1 || !ph1_bits(1)) {
FORC4 len[c] = ph1_bits(2);
FORC4 {
i = ((row & 1) << 1 | (c & 1)) % 3;
len[c] = len[c] < 3 ? lent[i][0]-'1'+"120"[len[c]] : ph1_bits(4);
lent[i][0] = lent[i][1];
lent[i][1] = len[c];
}
}
FORC(16) {
col = tab + (((c & 7) << 1)^(c >> 3)^(row & 1));
pred = (pmode == 7 || row < 2)
? (tab ? RAW(row,tab-2+(col & 1)) : init)
: (prow[col & 1][col-'4'+"0224468"[pmode]] +
prow[col & 1][col-'4'+"0244668"[pmode]] + 1) >> 1;
diff = ph1_bits (i = len[c >> 2]);
if (diff >> (i-1)) diff -= 1 << i;
diff = diff * (mag*2+1) + mag;
RAW(row,col) = pred + diff;
}
}
}
}
#define HOLE(row) ((holes >> (((row) - raw_height) & 7)) & 1)
/* Kudos to Rich Taylor for figuring out SMaL's compression algorithm. */
void CLASS smal_decode_segment (unsigned seg[2][2], int holes)
{
uchar hist[3][13] = {
{ 7, 7, 0, 0, 63, 55, 47, 39, 31, 23, 15, 7, 0 },
{ 7, 7, 0, 0, 63, 55, 47, 39, 31, 23, 15, 7, 0 },
{ 3, 3, 0, 0, 63, 47, 31, 15, 0 } };
int low, high=0xff, carry=0, nbits=8;
int pix, s, count, bin, next, i, sym[3];
uchar diff, pred[]={0,0};
ushort data=0, range=0;
fseek (ifp, seg[0][1]+1, SEEK_SET);
getbits(-1);
for (pix=seg[0][0]; pix < seg[1][0]; pix++) {
for (s=0; s < 3; s++) {
data = data << nbits | getbits(nbits);
if (carry < 0)
carry = (nbits += carry+1) < 1 ? nbits-1 : 0;
while (--nbits >= 0)
if ((data >> nbits & 0xff) == 0xff) break;
if (nbits > 0)
data = ((data & ((1 << (nbits-1)) - 1)) << 1) |
((data + (((data & (1 << (nbits-1)))) << 1)) & ((~0u) << nbits));
if (nbits >= 0) {
data += getbits(1);
carry = nbits - 8;
}
count = ((((data-range+1) & 0xffff) << 2) - 1) / (high >> 4);
for (bin=0; hist[s][bin+5] > count; bin++);
low = hist[s][bin+5] * (high >> 4) >> 2;
if (bin) high = hist[s][bin+4] * (high >> 4) >> 2;
high -= low;
for (nbits=0; high << nbits < 128; nbits++);
range = (range+low) << nbits;
high <<= nbits;
next = hist[s][1];
if (++hist[s][2] > hist[s][3]) {
next = (next+1) & hist[s][0];
hist[s][3] = (hist[s][next+4] - hist[s][next+5]) >> 2;
hist[s][2] = 1;
}
if (hist[s][hist[s][1]+4] - hist[s][hist[s][1]+5] > 1) {
if (bin < hist[s][1])
for (i=bin; i < hist[s][1]; i++) hist[s][i+5]--;
else if (next <= bin)
for (i=hist[s][1]; i < bin; i++) hist[s][i+5]++;
}
hist[s][1] = next;
sym[s] = bin;
}
diff = sym[2] << 5 | sym[1] << 2 | (sym[0] & 3);
if (sym[0] & 4)
diff = diff ? -diff : 0x80;
if (ftell(ifp) + 12 >= seg[1][1])
diff = 0;
#ifdef LIBRAW_LIBRARY_BUILD
if(pix>=raw_width*raw_height)
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#endif
raw_image[pix] = pred[pix & 1] += diff;
if (!(pix & 1) && HOLE(pix / raw_width)) pix += 2;
}
maximum = 0xff;
}
void CLASS smal_v6_load_raw()
{
unsigned seg[2][2];
fseek (ifp, 16, SEEK_SET);
seg[0][0] = 0;
seg[0][1] = get2();
seg[1][0] = raw_width * raw_height;
seg[1][1] = INT_MAX;
smal_decode_segment (seg, 0);
}
int CLASS median4 (int *p)
{
int min, max, sum, i;
min = max = sum = p[0];
for (i=1; i < 4; i++) {
sum += p[i];
if (min > p[i]) min = p[i];
if (max < p[i]) max = p[i];
}
return (sum - min - max) >> 1;
}
void CLASS fill_holes (int holes)
{
int row, col, val[4];
for (row=2; row < height-2; row++) {
if (!HOLE(row)) continue;
for (col=1; col < width-1; col+=4) {
val[0] = RAW(row-1,col-1);
val[1] = RAW(row-1,col+1);
val[2] = RAW(row+1,col-1);
val[3] = RAW(row+1,col+1);
RAW(row,col) = median4(val);
}
for (col=2; col < width-2; col+=4)
if (HOLE(row-2) || HOLE(row+2))
RAW(row,col) = (RAW(row,col-2) + RAW(row,col+2)) >> 1;
else {
val[0] = RAW(row,col-2);
val[1] = RAW(row,col+2);
val[2] = RAW(row-2,col);
val[3] = RAW(row+2,col);
RAW(row,col) = median4(val);
}
}
}
void CLASS smal_v9_load_raw()
{
unsigned seg[256][2], offset, nseg, holes, i;
fseek (ifp, 67, SEEK_SET);
offset = get4();
nseg = (uchar) fgetc(ifp);
fseek (ifp, offset, SEEK_SET);
for (i=0; i < nseg*2; i++)
((unsigned *)seg)[i] = get4() + data_offset*(i & 1);
fseek (ifp, 78, SEEK_SET);
holes = fgetc(ifp);
fseek (ifp, 88, SEEK_SET);
seg[nseg][0] = raw_height * raw_width;
seg[nseg][1] = get4() + data_offset;
for (i=0; i < nseg; i++)
smal_decode_segment (seg+i, holes);
if (holes) fill_holes (holes);
}
void CLASS redcine_load_raw()
{
#ifndef NO_JASPER
int c, row, col;
jas_stream_t *in;
jas_image_t *jimg;
jas_matrix_t *jmat;
jas_seqent_t *data;
ushort *img, *pix;
jas_init();
#ifndef LIBRAW_LIBRARY_BUILD
in = jas_stream_fopen (ifname, "rb");
#else
in = (jas_stream_t*)ifp->make_jas_stream();
if(!in)
throw LIBRAW_EXCEPTION_DECODE_JPEG2000;
#endif
jas_stream_seek (in, data_offset+20, SEEK_SET);
jimg = jas_image_decode (in, -1, 0);
#ifndef LIBRAW_LIBRARY_BUILD
if (!jimg) longjmp (failure, 3);
#else
if(!jimg)
{
jas_stream_close (in);
throw LIBRAW_EXCEPTION_DECODE_JPEG2000;
}
#endif
jmat = jas_matrix_create (height/2, width/2);
merror (jmat, "redcine_load_raw()");
img = (ushort *) calloc ((height+2), (width+2)*2);
merror (img, "redcine_load_raw()");
#ifdef LIBRAW_LIBRARY_BUILD
bool fastexitflag = false;
try {
#endif
FORC4 {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
jas_image_readcmpt (jimg, c, 0, 0, width/2, height/2, jmat);
data = jas_matrix_getref (jmat, 0, 0);
for (row = c >> 1; row < height; row+=2)
for (col = c & 1; col < width; col+=2)
img[(row+1)*(width+2)+col+1] = data[(row/2)*(width/2)+col/2];
}
for (col=1; col <= width; col++) {
img[col] = img[2*(width+2)+col];
img[(height+1)*(width+2)+col] = img[(height-1)*(width+2)+col];
}
for (row=0; row < height+2; row++) {
img[row*(width+2)] = img[row*(width+2)+2];
img[(row+1)*(width+2)-1] = img[(row+1)*(width+2)-3];
}
for (row=1; row <= height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
pix = img + row*(width+2) + (col = 1 + (FC(row,1) & 1));
for ( ; col <= width; col+=2, pix+=2) {
c = (((pix[0] - 0x800) << 3) +
pix[-(width+2)] + pix[width+2] + pix[-1] + pix[1]) >> 2;
pix[0] = LIM(c,0,4095);
}
}
for (row=0; row < height; row++)
{
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col++)
RAW(row,col) = curve[img[(row+1)*(width+2)+col+1]];
}
#ifdef LIBRAW_LIBRARY_BUILD
} catch (...) {
fastexitflag=true;
}
#endif
free (img);
jas_matrix_destroy (jmat);
jas_image_destroy (jimg);
jas_stream_close (in);
#ifdef LIBRAW_LIBRARY_BUILD
if(fastexitflag)
throw LIBRAW_EXCEPTION_CANCELLED_BY_CALLBACK;
#endif
#endif
}
//@end COMMON
/* RESTRICTED code starts here */
void CLASS foveon_decoder (unsigned size, unsigned code)
{
static unsigned huff[1024];
struct decode *cur;
int i, len;
if (!code) {
for (i=0; i < size; i++)
huff[i] = get4();
memset (first_decode, 0, sizeof first_decode);
free_decode = first_decode;
}
cur = free_decode++;
if (free_decode > first_decode+2048) {
fprintf (stderr,_("%s: decoder table overflow\n"), ifname);
longjmp (failure, 2);
}
if (code)
for (i=0; i < size; i++)
if (huff[i] == code) {
cur->leaf = i;
return;
}
if ((len = code >> 27) > 26) return;
code = (len+1) << 27 | (code & 0x3ffffff) << 1;
cur->branch[0] = free_decode;
foveon_decoder (size, code);
cur->branch[1] = free_decode;
foveon_decoder (size, code+1);
}
void CLASS foveon_thumb()
{
unsigned bwide, row, col, bitbuf=0, bit=1, c, i;
char *buf;
struct decode *dindex;
short pred[3];
bwide = get4();
fprintf (ofp, "P6\n%d %d\n255\n", thumb_width, thumb_height);
if (bwide > 0) {
if (bwide < thumb_width*3) return;
buf = (char *) malloc (bwide);
merror (buf, "foveon_thumb()");
for (row=0; row < thumb_height; row++) {
fread (buf, 1, bwide, ifp);
fwrite (buf, 3, thumb_width, ofp);
}
free (buf);
return;
}
foveon_decoder (256, 0);
for (row=0; row < thumb_height; row++) {
memset (pred, 0, sizeof pred);
if (!bit) get4();
for (bit=col=0; col < thumb_width; col++)
FORC3 {
for (dindex=first_decode; dindex->branch[0]; ) {
if ((bit = (bit-1) & 31) == 31)
for (i=0; i < 4; i++)
bitbuf = (bitbuf << 8) + fgetc(ifp);
dindex = dindex->branch[bitbuf >> bit & 1];
}
pred[c] += dindex->leaf;
fputc (pred[c], ofp);
}
}
}
void CLASS foveon_sd_load_raw()
{
struct decode *dindex;
short diff[1024];
unsigned bitbuf=0;
int pred[3], row, col, bit=-1, c, i;
read_shorts ((ushort *) diff, 1024);
if (!load_flags) foveon_decoder (1024, 0);
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
memset (pred, 0, sizeof pred);
if (!bit && !load_flags && atoi(model+2) < 14) get4();
for (col=bit=0; col < width; col++) {
if (load_flags) {
bitbuf = get4();
FORC3 pred[2-c] += diff[bitbuf >> c*10 & 0x3ff];
}
else FORC3 {
for (dindex=first_decode; dindex->branch[0]; ) {
if ((bit = (bit-1) & 31) == 31)
for (i=0; i < 4; i++)
bitbuf = (bitbuf << 8) + fgetc(ifp);
dindex = dindex->branch[bitbuf >> bit & 1];
}
pred[c] += diff[dindex->leaf];
if (pred[c] >> 16 && ~pred[c] >> 16) derror();
}
FORC3 image[row*width+col][c] = pred[c];
}
}
}
void CLASS foveon_huff (ushort *huff)
{
int i, j, clen, code;
huff[0] = 8;
for (i=0; i < 13; i++) {
clen = getc(ifp);
code = getc(ifp);
for (j=0; j < 256 >> clen; )
huff[code+ ++j] = clen << 8 | i;
}
get2();
}
void CLASS foveon_dp_load_raw()
{
unsigned c, roff[4], row, col, diff;
ushort huff[512], vpred[2][2], hpred[2];
fseek (ifp, 8, SEEK_CUR);
foveon_huff (huff);
roff[0] = 48;
FORC3 roff[c+1] = -(-(roff[c] + get4()) & -16);
FORC3 {
fseek (ifp, data_offset+roff[c], SEEK_SET);
getbits(-1);
vpred[0][0] = vpred[0][1] = vpred[1][0] = vpred[1][1] = 512;
for (row=0; row < height; row++) {
#ifdef LIBRAW_LIBRARY_BUILD
checkCancel();
#endif
for (col=0; col < width; col++) {
diff = ljpeg_diff(huff);
if (col < 2) hpred[col] = vpred[row & 1][col] += diff;
else hpred[col & 1] += diff;
image[row*width+col][c] = hpred[col & 1];
}
}
}
}
void CLASS foveon_load_camf()
{
unsigned type, wide, high, i, j, row, col, diff;
ushort huff[258], vpred[2][2] = {{512,512},{512,512}}, hpred[2];
fseek (ifp, meta_offset, SEEK_SET);
type = get4(); get4(); get4();
wide = get4();
high = get4();
if (type == 2) {
fread (meta_data, 1, meta_length, ifp);
for (i=0; i < meta_length; i++) {
high = (high * 1597 + 51749) % 244944;
wide = high * (INT64) 301593171 >> 24;
meta_data[i] ^= ((((high << 8) - wide) >> 1) + wide) >> 17;
}
} else if (type == 4) {
free (meta_data);
meta_data = (char *) malloc (meta_length = wide*high*3/2);
merror (meta_data, "foveon_load_camf()");
foveon_huff (huff);
get4();
getbits(-1);
for (j=row=0; row < high; row++) {
for (col=0; col < wide; col++) {
diff = ljpeg_diff(huff);
if (col < 2) hpred[col] = vpred[row & 1][col] += diff;
else hpred[col & 1] += diff;
if (col & 1) {
meta_data[j++] = hpred[0] >> 4;
meta_data[j++] = hpred[0] << 4 | hpred[1] >> 8;
meta_data[j++] = hpred[1];
}
}
}
}
#ifdef DCRAW_VERBOSE
else
fprintf (stderr,_("%s has unknown CAMF type %d.\n"), ifname, type);
#endif
}
const char * CLASS foveon_camf_param (const char *block, const char *param)
{
unsigned idx, num;
char *pos, *cp, *dp;
for (idx=0; idx < meta_length; idx += sget4(pos+8)) {
pos = meta_data + idx;
if (strncmp (pos, "CMb", 3)) break;
if (pos[3] != 'P') continue;
if (strcmp (block, pos+sget4(pos+12))) continue;
cp = pos + sget4(pos+16);
num = sget4(cp);
dp = pos + sget4(cp+4);
while (num--) {
cp += 8;
if (!strcmp (param, dp+sget4(cp)))
return dp+sget4(cp+4);
}
}
return 0;
}
void * CLASS foveon_camf_matrix (unsigned dim[3], const char *name)
{
unsigned i, idx, type, ndim, size, *mat;
char *pos, *cp, *dp;
double dsize;
for (idx=0; idx < meta_length; idx += sget4(pos+8)) {
pos = meta_data + idx;
if (strncmp (pos, "CMb", 3)) break;
if (pos[3] != 'M') continue;
if (strcmp (name, pos+sget4(pos+12))) continue;
dim[0] = dim[1] = dim[2] = 1;
cp = pos + sget4(pos+16);
type = sget4(cp);
if ((ndim = sget4(cp+4)) > 3) break;
dp = pos + sget4(cp+8);
for (i=ndim; i--; ) {
cp += 12;
dim[i] = sget4(cp);
}
if ((dsize = (double) dim[0]*dim[1]*dim[2]) > meta_length/4) break;
mat = (unsigned *) malloc ((size = dsize) * 4);
merror (mat, "foveon_camf_matrix()");
for (i=0; i < size; i++)
if (type && type != 6)
mat[i] = sget4(dp + i*4);
else
mat[i] = sget4(dp + i*2) & 0xffff;
return mat;
}
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: \"%s\" matrix not found!\n"), ifname, name);
#endif
return 0;
}
int CLASS foveon_fixed (void *ptr, int size, const char *name)
{
void *dp;
unsigned dim[3];
if (!name) return 0;
dp = foveon_camf_matrix (dim, name);
if (!dp) return 0;
memcpy (ptr, dp, size*4);
free (dp);
return 1;
}
float CLASS foveon_avg (short *pix, int range[2], float cfilt)
{
int i;
float val, min=FLT_MAX, max=-FLT_MAX, sum=0;
for (i=range[0]; i <= range[1]; i++) {
sum += val = pix[i*4] + (pix[i*4]-pix[(i-1)*4]) * cfilt;
if (min > val) min = val;
if (max < val) max = val;
}
if (range[1] - range[0] == 1) return sum/2;
return (sum - min - max) / (range[1] - range[0] - 1);
}
short * CLASS foveon_make_curve (double max, double mul, double filt)
{
short *curve;
unsigned i, size;
double x;
if (!filt) filt = 0.8;
size = 4*M_PI*max / filt;
if (size == UINT_MAX) size--;
curve = (short *) calloc (size+1, sizeof *curve);
merror (curve, "foveon_make_curve()");
curve[0] = size;
for (i=0; i < size; i++) {
x = i*filt/max/4;
curve[i+1] = (cos(x)+1)/2 * tanh(i*filt/mul) * mul + 0.5;
}
return curve;
}
void CLASS foveon_make_curves
(short **curvep, float dq[3], float div[3], float filt)
{
double mul[3], max=0;
int c;
FORC3 mul[c] = dq[c]/div[c];
FORC3 if (max < mul[c]) max = mul[c];
FORC3 curvep[c] = foveon_make_curve (max, mul[c], filt);
}
int CLASS foveon_apply_curve (short *curve, int i)
{
if (abs(i) >= curve[0]) return 0;
return i < 0 ? -curve[1-i] : curve[1+i];
}
#define image ((short (*)[4]) image)
void CLASS foveon_interpolate()
{
static const short hood[] = { -1,-1, -1,0, -1,1, 0,-1, 0,1, 1,-1, 1,0, 1,1 };
short *pix, prev[3], *curve[8], (*shrink)[3];
float cfilt=0, ddft[3][3][2], ppm[3][3][3];
float cam_xyz[3][3], correct[3][3], last[3][3], trans[3][3];
float chroma_dq[3], color_dq[3], diag[3][3], div[3];
float (*black)[3], (*sgain)[3], (*sgrow)[3];
float fsum[3], val, frow, num;
int row, col, c, i, j, diff, sgx, irow, sum, min, max, limit;
int dscr[2][2], dstb[4], (*smrow[7])[3], total[4], ipix[3];
int work[3][3], smlast, smred, smred_p=0, dev[3];
int satlev[3], keep[4], active[4];
unsigned dim[3], *badpix;
double dsum=0, trsum[3];
char str[128];
const char* cp;
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Foveon interpolation...\n"));
#endif
foveon_load_camf();
foveon_fixed (dscr, 4, "DarkShieldColRange");
foveon_fixed (ppm[0][0], 27, "PostPolyMatrix");
foveon_fixed (satlev, 3, "SaturationLevel");
foveon_fixed (keep, 4, "KeepImageArea");
foveon_fixed (active, 4, "ActiveImageArea");
foveon_fixed (chroma_dq, 3, "ChromaDQ");
foveon_fixed (color_dq, 3,
foveon_camf_param ("IncludeBlocks", "ColorDQ") ?
"ColorDQ" : "ColorDQCamRGB");
if (foveon_camf_param ("IncludeBlocks", "ColumnFilter"))
foveon_fixed (&cfilt, 1, "ColumnFilter");
memset (ddft, 0, sizeof ddft);
if (!foveon_camf_param ("IncludeBlocks", "DarkDrift")
|| !foveon_fixed (ddft[1][0], 12, "DarkDrift"))
for (i=0; i < 2; i++) {
foveon_fixed (dstb, 4, i ? "DarkShieldBottom":"DarkShieldTop");
for (row = dstb[1]; row <= dstb[3]; row++)
for (col = dstb[0]; col <= dstb[2]; col++)
FORC3 ddft[i+1][c][1] += (short) image[row*width+col][c];
FORC3 ddft[i+1][c][1] /= (dstb[3]-dstb[1]+1) * (dstb[2]-dstb[0]+1);
}
if (!(cp = foveon_camf_param ("WhiteBalanceIlluminants", model2)))
{
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: Invalid white balance \"%s\"\n"), ifname, model2);
#endif
return; }
foveon_fixed (cam_xyz, 9, cp);
foveon_fixed (correct, 9,
foveon_camf_param ("WhiteBalanceCorrections", model2));
memset (last, 0, sizeof last);
for (i=0; i < 3; i++)
for (j=0; j < 3; j++)
FORC3 last[i][j] += correct[i][c] * cam_xyz[c][j];
#define LAST(x,y) last[(i+x)%3][(c+y)%3]
for (i=0; i < 3; i++)
FORC3 diag[c][i] = LAST(1,1)*LAST(2,2) - LAST(1,2)*LAST(2,1);
#undef LAST
FORC3 div[c] = diag[c][0]*0.3127 + diag[c][1]*0.329 + diag[c][2]*0.3583;
sprintf (str, "%sRGBNeutral", model2);
if (foveon_camf_param ("IncludeBlocks", str))
foveon_fixed (div, 3, str);
num = 0;
FORC3 if (num < div[c]) num = div[c];
FORC3 div[c] /= num;
memset (trans, 0, sizeof trans);
for (i=0; i < 3; i++)
for (j=0; j < 3; j++)
FORC3 trans[i][j] += rgb_cam[i][c] * last[c][j] * div[j];
FORC3 trsum[c] = trans[c][0] + trans[c][1] + trans[c][2];
dsum = (6*trsum[0] + 11*trsum[1] + 3*trsum[2]) / 20;
for (i=0; i < 3; i++)
FORC3 last[i][c] = trans[i][c] * dsum / trsum[i];
memset (trans, 0, sizeof trans);
for (i=0; i < 3; i++)
for (j=0; j < 3; j++)
FORC3 trans[i][j] += (i==c ? 32 : -1) * last[c][j] / 30;
foveon_make_curves (curve, color_dq, div, cfilt);
FORC3 chroma_dq[c] /= 3;
foveon_make_curves (curve+3, chroma_dq, div, cfilt);
FORC3 dsum += chroma_dq[c] / div[c];
curve[6] = foveon_make_curve (dsum, dsum, cfilt);
curve[7] = foveon_make_curve (dsum*2, dsum*2, cfilt);
sgain = (float (*)[3]) foveon_camf_matrix (dim, "SpatialGain");
if (!sgain) return;
sgrow = (float (*)[3]) calloc (dim[1], sizeof *sgrow);
sgx = (width + dim[1]-2) / (dim[1]-1);
black = (float (*)[3]) calloc (height, sizeof *black);
for (row=0; row < height; row++) {
for (i=0; i < 6; i++)
((float *)ddft[0])[i] = ((float *)ddft[1])[i] +
row / (height-1.0) * (((float *)ddft[2])[i] - ((float *)ddft[1])[i]);
FORC3 black[row][c] =
( foveon_avg (image[row*width]+c, dscr[0], cfilt) +
foveon_avg (image[row*width]+c, dscr[1], cfilt) * 3
- ddft[0][c][0] ) / 4 - ddft[0][c][1];
}
memcpy (black, black+8, sizeof *black*8);
memcpy (black+height-11, black+height-22, 11*sizeof *black);
memcpy (last, black, sizeof last);
for (row=1; row < height-1; row++) {
FORC3 if (last[1][c] > last[0][c]) {
if (last[1][c] > last[2][c])
black[row][c] = (last[0][c] > last[2][c]) ? last[0][c]:last[2][c];
} else
if (last[1][c] < last[2][c])
black[row][c] = (last[0][c] < last[2][c]) ? last[0][c]:last[2][c];
memmove (last, last+1, 2*sizeof last[0]);
memcpy (last[2], black[row+1], sizeof last[2]);
}
FORC3 black[row][c] = (last[0][c] + last[1][c])/2;
FORC3 black[0][c] = (black[1][c] + black[3][c])/2;
val = 1 - exp(-1/24.0);
memcpy (fsum, black, sizeof fsum);
for (row=1; row < height; row++)
FORC3 fsum[c] += black[row][c] =
(black[row][c] - black[row-1][c])*val + black[row-1][c];
memcpy (last[0], black[height-1], sizeof last[0]);
FORC3 fsum[c] /= height;
for (row = height; row--; )
FORC3 last[0][c] = black[row][c] =
(black[row][c] - fsum[c] - last[0][c])*val + last[0][c];
memset (total, 0, sizeof total);
for (row=2; row < height; row+=4)
for (col=2; col < width; col+=4) {
FORC3 total[c] += (short) image[row*width+col][c];
total[3]++;
}
for (row=0; row < height; row++)
FORC3 black[row][c] += fsum[c]/2 + total[c]/(total[3]*100.0);
for (row=0; row < height; row++) {
for (i=0; i < 6; i++)
((float *)ddft[0])[i] = ((float *)ddft[1])[i] +
row / (height-1.0) * (((float *)ddft[2])[i] - ((float *)ddft[1])[i]);
pix = image[row*width];
memcpy (prev, pix, sizeof prev);
frow = row / (height-1.0) * (dim[2]-1);
if ((irow = frow) == dim[2]-1) irow--;
frow -= irow;
for (i=0; i < dim[1]; i++)
FORC3 sgrow[i][c] = sgain[ irow *dim[1]+i][c] * (1-frow) +
sgain[(irow+1)*dim[1]+i][c] * frow;
for (col=0; col < width; col++) {
FORC3 {
diff = pix[c] - prev[c];
prev[c] = pix[c];
ipix[c] = pix[c] + floor ((diff + (diff*diff >> 14)) * cfilt
- ddft[0][c][1] - ddft[0][c][0] * ((float) col/width - 0.5)
- black[row][c] );
}
FORC3 {
work[0][c] = ipix[c] * ipix[c] >> 14;
work[2][c] = ipix[c] * work[0][c] >> 14;
work[1][2-c] = ipix[(c+1) % 3] * ipix[(c+2) % 3] >> 14;
}
FORC3 {
for (val=i=0; i < 3; i++)
for ( j=0; j < 3; j++)
val += ppm[c][i][j] * work[i][j];
ipix[c] = floor ((ipix[c] + floor(val)) *
( sgrow[col/sgx ][c] * (sgx - col%sgx) +
sgrow[col/sgx+1][c] * (col%sgx) ) / sgx / div[c]);
if (ipix[c] > 32000) ipix[c] = 32000;
pix[c] = ipix[c];
}
pix += 4;
}
}
free (black);
free (sgrow);
free (sgain);
if ((badpix = (unsigned *) foveon_camf_matrix (dim, "BadPixels"))) {
for (i=0; i < dim[0]; i++) {
col = (badpix[i] >> 8 & 0xfff) - keep[0];
row = (badpix[i] >> 20 ) - keep[1];
if ((unsigned)(row-1) > height-3 || (unsigned)(col-1) > width-3)
continue;
memset (fsum, 0, sizeof fsum);
for (sum=j=0; j < 8; j++)
if (badpix[i] & (1 << j)) {
FORC3 fsum[c] += (short)
image[(row+hood[j*2])*width+col+hood[j*2+1]][c];
sum++;
}
if (sum) FORC3 image[row*width+col][c] = fsum[c]/sum;
}
free (badpix);
}
/* Array for 5x5 Gaussian averaging of red values */
smrow[6] = (int (*)[3]) calloc (width*5, sizeof **smrow);
merror (smrow[6], "foveon_interpolate()");
for (i=0; i < 5; i++)
smrow[i] = smrow[6] + i*width;
/* Sharpen the reds against these Gaussian averages */
for (smlast=-1, row=2; row < height-2; row++) {
while (smlast < row+2) {
for (i=0; i < 6; i++)
smrow[(i+5) % 6] = smrow[i];
pix = image[++smlast*width+2];
for (col=2; col < width-2; col++) {
smrow[4][col][0] =
(pix[0]*6 + (pix[-4]+pix[4])*4 + pix[-8]+pix[8] + 8) >> 4;
pix += 4;
}
}
pix = image[row*width+2];
for (col=2; col < width-2; col++) {
smred = ( 6 * smrow[2][col][0]
+ 4 * (smrow[1][col][0] + smrow[3][col][0])
+ smrow[0][col][0] + smrow[4][col][0] + 8 ) >> 4;
if (col == 2)
smred_p = smred;
i = pix[0] + ((pix[0] - ((smred*7 + smred_p) >> 3)) >> 3);
if (i > 32000) i = 32000;
pix[0] = i;
smred_p = smred;
pix += 4;
}
}
/* Adjust the brighter pixels for better linearity */
min = 0xffff;
FORC3 {
i = satlev[c] / div[c];
if (min > i) min = i;
}
limit = min * 9 >> 4;
for (pix=image[0]; pix < image[height*width]; pix+=4) {
if (pix[0] <= limit || pix[1] <= limit || pix[2] <= limit)
continue;
min = max = pix[0];
for (c=1; c < 3; c++) {
if (min > pix[c]) min = pix[c];
if (max < pix[c]) max = pix[c];
}
if (min >= limit*2) {
pix[0] = pix[1] = pix[2] = max;
} else {
i = 0x4000 - ((min - limit) << 14) / limit;
i = 0x4000 - (i*i >> 14);
i = i*i >> 14;
FORC3 pix[c] += (max - pix[c]) * i >> 14;
}
}
/*
Because photons that miss one detector often hit another,
the sum R+G+B is much less noisy than the individual colors.
So smooth the hues without smoothing the total.
*/
for (smlast=-1, row=2; row < height-2; row++) {
while (smlast < row+2) {
for (i=0; i < 6; i++)
smrow[(i+5) % 6] = smrow[i];
pix = image[++smlast*width+2];
for (col=2; col < width-2; col++) {
FORC3 smrow[4][col][c] = (pix[c-4]+2*pix[c]+pix[c+4]+2) >> 2;
pix += 4;
}
}
pix = image[row*width+2];
for (col=2; col < width-2; col++) {
FORC3 dev[c] = -foveon_apply_curve (curve[7], pix[c] -
((smrow[1][col][c] + 2*smrow[2][col][c] + smrow[3][col][c]) >> 2));
sum = (dev[0] + dev[1] + dev[2]) >> 3;
FORC3 pix[c] += dev[c] - sum;
pix += 4;
}
}
for (smlast=-1, row=2; row < height-2; row++) {
while (smlast < row+2) {
for (i=0; i < 6; i++)
smrow[(i+5) % 6] = smrow[i];
pix = image[++smlast*width+2];
for (col=2; col < width-2; col++) {
FORC3 smrow[4][col][c] =
(pix[c-8]+pix[c-4]+pix[c]+pix[c+4]+pix[c+8]+2) >> 2;
pix += 4;
}
}
pix = image[row*width+2];
for (col=2; col < width-2; col++) {
for (total[3]=375, sum=60, c=0; c < 3; c++) {
for (total[c]=i=0; i < 5; i++)
total[c] += smrow[i][col][c];
total[3] += total[c];
sum += pix[c];
}
if (sum < 0) sum = 0;
j = total[3] > 375 ? (sum << 16) / total[3] : sum * 174;
FORC3 pix[c] += foveon_apply_curve (curve[6],
((j*total[c] + 0x8000) >> 16) - pix[c]);
pix += 4;
}
}
/* Transform the image to a different colorspace */
for (pix=image[0]; pix < image[height*width]; pix+=4) {
FORC3 pix[c] -= foveon_apply_curve (curve[c], pix[c]);
sum = (pix[0]+pix[1]+pix[1]+pix[2]) >> 2;
FORC3 pix[c] -= foveon_apply_curve (curve[c], pix[c]-sum);
FORC3 {
for (dsum=i=0; i < 3; i++)
dsum += trans[c][i] * pix[i];
if (dsum < 0) dsum = 0;
if (dsum > 24000) dsum = 24000;
ipix[c] = dsum + 0.5;
}
FORC3 pix[c] = ipix[c];
}
/* Smooth the image bottom-to-top and save at 1/4 scale */
shrink = (short (*)[3]) calloc ((height/4), (width/4)*sizeof *shrink);
merror (shrink, "foveon_interpolate()");
for (row = height/4; row--; )
for (col=0; col < width/4; col++) {
ipix[0] = ipix[1] = ipix[2] = 0;
for (i=0; i < 4; i++)
for (j=0; j < 4; j++)
FORC3 ipix[c] += image[(row*4+i)*width+col*4+j][c];
FORC3
if (row+2 > height/4)
shrink[row*(width/4)+col][c] = ipix[c] >> 4;
else
shrink[row*(width/4)+col][c] =
(shrink[(row+1)*(width/4)+col][c]*1840 + ipix[c]*141 + 2048) >> 12;
}
/* From the 1/4-scale image, smooth right-to-left */
for (row=0; row < (height & ~3); row++) {
ipix[0] = ipix[1] = ipix[2] = 0;
if ((row & 3) == 0)
for (col = width & ~3 ; col--; )
FORC3 smrow[0][col][c] = ipix[c] =
(shrink[(row/4)*(width/4)+col/4][c]*1485 + ipix[c]*6707 + 4096) >> 13;
/* Then smooth left-to-right */
ipix[0] = ipix[1] = ipix[2] = 0;
for (col=0; col < (width & ~3); col++)
FORC3 smrow[1][col][c] = ipix[c] =
(smrow[0][col][c]*1485 + ipix[c]*6707 + 4096) >> 13;
/* Smooth top-to-bottom */
if (row == 0)
memcpy (smrow[2], smrow[1], sizeof **smrow * width);
else
for (col=0; col < (width & ~3); col++)
FORC3 smrow[2][col][c] =
(smrow[2][col][c]*6707 + smrow[1][col][c]*1485 + 4096) >> 13;
/* Adjust the chroma toward the smooth values */
for (col=0; col < (width & ~3); col++) {
for (i=j=30, c=0; c < 3; c++) {
i += smrow[2][col][c];
j += image[row*width+col][c];
}
j = (j << 16) / i;
for (sum=c=0; c < 3; c++) {
ipix[c] = foveon_apply_curve (curve[c+3],
((smrow[2][col][c] * j + 0x8000) >> 16) - image[row*width+col][c]);
sum += ipix[c];
}
sum >>= 3;
FORC3 {
i = image[row*width+col][c] + ipix[c] - sum;
if (i < 0) i = 0;
image[row*width+col][c] = i;
}
}
}
free (shrink);
free (smrow[6]);
for (i=0; i < 8; i++)
free (curve[i]);
/* Trim off the black border */
active[1] -= keep[1];
active[3] -= 2;
i = active[2] - active[0];
for (row=0; row < active[3]-active[1]; row++)
memcpy (image[row*i], image[(row+active[1])*width+active[0]],
i * sizeof *image);
width = i;
height = row;
}
#undef image
/* RESTRICTED code ends here */
//@out COMMON
void CLASS crop_masked_pixels()
{
int row, col;
unsigned
#ifndef LIBRAW_LIBRARY_BUILD
r, raw_pitch = raw_width*2,
c, m, mblack[8], zero, val;
#else
c, m, zero, val;
#define mblack imgdata.color.black_stat
#endif
#ifndef LIBRAW_LIBRARY_BUILD
if (load_raw == &CLASS phase_one_load_raw ||
load_raw == &CLASS phase_one_load_raw_c)
phase_one_correct();
if (fuji_width) {
for (row=0; row < raw_height-top_margin*2; row++) {
for (col=0; col < fuji_width << !fuji_layout; col++) {
if (fuji_layout) {
r = fuji_width - 1 - col + (row >> 1);
c = col + ((row+1) >> 1);
} else {
r = fuji_width - 1 + row - (col >> 1);
c = row + ((col+1) >> 1);
}
if (r < height && c < width)
BAYER(r,c) = RAW(row+top_margin,col+left_margin);
}
}
} else {
for (row=0; row < height; row++)
for (col=0; col < width; col++)
BAYER2(row,col) = RAW(row+top_margin,col+left_margin);
}
#endif
if (mask[0][3] > 0) goto mask_set;
if (load_raw == &CLASS canon_load_raw ||
load_raw == &CLASS lossless_jpeg_load_raw) {
mask[0][1] = mask[1][1] += 2;
mask[0][3] -= 2;
goto sides;
}
if (load_raw == &CLASS canon_600_load_raw ||
load_raw == &CLASS sony_load_raw ||
(load_raw == &CLASS eight_bit_load_raw && strncmp(model,"DC2",3)) ||
load_raw == &CLASS kodak_262_load_raw ||
(load_raw == &CLASS packed_load_raw && (load_flags & 32))) {
sides:
mask[0][0] = mask[1][0] = top_margin;
mask[0][2] = mask[1][2] = top_margin+height;
mask[0][3] += left_margin;
mask[1][1] += left_margin+width;
mask[1][3] += raw_width;
}
if (load_raw == &CLASS nokia_load_raw) {
mask[0][2] = top_margin;
mask[0][3] = width;
}
mask_set:
memset (mblack, 0, sizeof mblack);
for (zero=m=0; m < 8; m++)
for (row=MAX(mask[m][0],0); row < MIN(mask[m][2],raw_height); row++)
for (col=MAX(mask[m][1],0); col < MIN(mask[m][3],raw_width); col++) {
c = FC(row-top_margin,col-left_margin);
mblack[c] += val = raw_image[(row)*raw_pitch/2+(col)];
mblack[4+c]++;
zero += !val;
}
if (load_raw == &CLASS canon_600_load_raw && width < raw_width) {
black = (mblack[0]+mblack[1]+mblack[2]+mblack[3]) /
(mblack[4]+mblack[5]+mblack[6]+mblack[7]) - 4;
#ifndef LIBRAW_LIBRARY_BUILD
canon_600_correct();
#endif
} else if (zero < mblack[4] && mblack[5] && mblack[6] && mblack[7]) {
FORC4 cblack[c] = mblack[c] / mblack[4+c];
cblack[4] = cblack[5] = cblack[6] = 0;
}
}
#ifdef LIBRAW_LIBRARY_BUILD
#undef mblack
#endif
void CLASS remove_zeroes()
{
unsigned row, col, tot, n, r, c;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_REMOVE_ZEROES,0,2);
#endif
for (row=0; row < height; row++)
for (col=0; col < width; col++)
if (BAYER(row,col) == 0) {
tot = n = 0;
for (r = row-2; r <= row+2; r++)
for (c = col-2; c <= col+2; c++)
if (r < height && c < width &&
FC(r,c) == FC(row,col) && BAYER(r,c))
tot += (n++,BAYER(r,c));
if (n) BAYER(row,col) = tot/n;
}
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_REMOVE_ZEROES,1,2);
#endif
}
//@end COMMON
/* @out FILEIO
#include <math.h>
#define CLASS LibRaw::
#include "libraw/libraw_types.h"
#define LIBRAW_LIBRARY_BUILD
#include "libraw/libraw.h"
#include "internal/defines.h"
#include "internal/var_defines.h"
@end FILEIO */
// @out FILEIO
/*
Seach from the current directory up to the root looking for
a ".badpixels" file, and fix those pixels now.
*/
void CLASS bad_pixels (const char *cfname)
{
FILE *fp=NULL;
#ifndef LIBRAW_LIBRARY_BUILD
char *fname, *cp, line[128];
int len, time, row, col, r, c, rad, tot, n, fixed=0;
#else
char *cp, line[128];
int time, row, col, r, c, rad, tot, n;
#ifdef DCRAW_VERBOSE
int fixed = 0;
#endif
#endif
if (!filters) return;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_BAD_PIXELS,0,2);
#endif
if (cfname)
fp = fopen (cfname, "r");
// @end FILEIO
else {
for (len=32 ; ; len *= 2) {
fname = (char *) malloc (len);
if (!fname) return;
if (getcwd (fname, len-16)) break;
free (fname);
if (errno != ERANGE) return;
}
#if defined(WIN32) || defined(DJGPP)
if (fname[1] == ':')
memmove (fname, fname+2, len-2);
for (cp=fname; *cp; cp++)
if (*cp == '\\') *cp = '/';
#endif
cp = fname + strlen(fname);
if (cp[-1] == '/') cp--;
while (*fname == '/') {
strcpy (cp, "/.badpixels");
if ((fp = fopen (fname, "r"))) break;
if (cp == fname) break;
while (*--cp != '/');
}
free (fname);
}
// @out FILEIO
if (!fp)
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_BADPIXELMAP;
#endif
return;
}
while (fgets (line, 128, fp)) {
cp = strchr (line, '#');
if (cp) *cp = 0;
if (sscanf (line, "%d %d %d", &col, &row, &time) != 3) continue;
if ((unsigned) col >= width || (unsigned) row >= height) continue;
if (time > timestamp) continue;
for (tot=n=0, rad=1; rad < 3 && n==0; rad++)
for (r = row-rad; r <= row+rad; r++)
for (c = col-rad; c <= col+rad; c++)
if ((unsigned) r < height && (unsigned) c < width &&
(r != row || c != col) && fcol(r,c) == fcol(row,col)) {
tot += BAYER2(r,c);
n++;
}
BAYER2(row,col) = tot/n;
#ifdef DCRAW_VERBOSE
if (verbose) {
if (!fixed++)
fprintf (stderr,_("Fixed dead pixels at:"));
fprintf (stderr, " %d,%d", col, row);
}
#endif
}
#ifdef DCRAW_VERBOSE
if (fixed) fputc ('\n', stderr);
#endif
fclose (fp);
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_BAD_PIXELS,1,2);
#endif
}
void CLASS subtract (const char *fname)
{
FILE *fp;
int dim[3]={0,0,0}, comment=0, number=0, error=0, nd=0, c, row, col;
ushort *pixel;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_DARK_FRAME,0,2);
#endif
if (!(fp = fopen (fname, "rb"))) {
#ifdef DCRAW_VERBOSE
perror (fname);
#endif
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_BAD_DARKFRAME_FILE;
#endif
return;
}
if (fgetc(fp) != 'P' || fgetc(fp) != '5') error = 1;
while (!error && nd < 3 && (c = fgetc(fp)) != EOF) {
if (c == '#') comment = 1;
if (c == '\n') comment = 0;
if (comment) continue;
if (isdigit(c)) number = 1;
if (number) {
if (isdigit(c)) dim[nd] = dim[nd]*10 + c -'0';
else if (isspace(c)) {
number = 0; nd++;
} else error = 1;
}
}
if (error || nd < 3) {
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s is not a valid PGM file!\n"), fname);
#endif
fclose (fp); return;
} else if (dim[0] != width || dim[1] != height || dim[2] != 65535) {
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s has the wrong dimensions!\n"), fname);
#endif
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_BAD_DARKFRAME_DIM;
#endif
fclose (fp); return;
}
pixel = (ushort *) calloc (width, sizeof *pixel);
merror (pixel, "subtract()");
for (row=0; row < height; row++) {
fread (pixel, 2, width, fp);
for (col=0; col < width; col++)
BAYER(row,col) = MAX (BAYER(row,col) - ntohs(pixel[col]), 0);
}
free (pixel);
fclose (fp);
memset (cblack, 0, sizeof cblack);
black = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_DARK_FRAME,1,2);
#endif
}
//@end FILEIO
//@out COMMON
static const uchar xlat[2][256] = {
{ 0xc1,0xbf,0x6d,0x0d,0x59,0xc5,0x13,0x9d,0x83,0x61,0x6b,0x4f,0xc7,0x7f,0x3d,0x3d,
0x53,0x59,0xe3,0xc7,0xe9,0x2f,0x95,0xa7,0x95,0x1f,0xdf,0x7f,0x2b,0x29,0xc7,0x0d,
0xdf,0x07,0xef,0x71,0x89,0x3d,0x13,0x3d,0x3b,0x13,0xfb,0x0d,0x89,0xc1,0x65,0x1f,
0xb3,0x0d,0x6b,0x29,0xe3,0xfb,0xef,0xa3,0x6b,0x47,0x7f,0x95,0x35,0xa7,0x47,0x4f,
0xc7,0xf1,0x59,0x95,0x35,0x11,0x29,0x61,0xf1,0x3d,0xb3,0x2b,0x0d,0x43,0x89,0xc1,
0x9d,0x9d,0x89,0x65,0xf1,0xe9,0xdf,0xbf,0x3d,0x7f,0x53,0x97,0xe5,0xe9,0x95,0x17,
0x1d,0x3d,0x8b,0xfb,0xc7,0xe3,0x67,0xa7,0x07,0xf1,0x71,0xa7,0x53,0xb5,0x29,0x89,
0xe5,0x2b,0xa7,0x17,0x29,0xe9,0x4f,0xc5,0x65,0x6d,0x6b,0xef,0x0d,0x89,0x49,0x2f,
0xb3,0x43,0x53,0x65,0x1d,0x49,0xa3,0x13,0x89,0x59,0xef,0x6b,0xef,0x65,0x1d,0x0b,
0x59,0x13,0xe3,0x4f,0x9d,0xb3,0x29,0x43,0x2b,0x07,0x1d,0x95,0x59,0x59,0x47,0xfb,
0xe5,0xe9,0x61,0x47,0x2f,0x35,0x7f,0x17,0x7f,0xef,0x7f,0x95,0x95,0x71,0xd3,0xa3,
0x0b,0x71,0xa3,0xad,0x0b,0x3b,0xb5,0xfb,0xa3,0xbf,0x4f,0x83,0x1d,0xad,0xe9,0x2f,
0x71,0x65,0xa3,0xe5,0x07,0x35,0x3d,0x0d,0xb5,0xe9,0xe5,0x47,0x3b,0x9d,0xef,0x35,
0xa3,0xbf,0xb3,0xdf,0x53,0xd3,0x97,0x53,0x49,0x71,0x07,0x35,0x61,0x71,0x2f,0x43,
0x2f,0x11,0xdf,0x17,0x97,0xfb,0x95,0x3b,0x7f,0x6b,0xd3,0x25,0xbf,0xad,0xc7,0xc5,
0xc5,0xb5,0x8b,0xef,0x2f,0xd3,0x07,0x6b,0x25,0x49,0x95,0x25,0x49,0x6d,0x71,0xc7 },
{ 0xa7,0xbc,0xc9,0xad,0x91,0xdf,0x85,0xe5,0xd4,0x78,0xd5,0x17,0x46,0x7c,0x29,0x4c,
0x4d,0x03,0xe9,0x25,0x68,0x11,0x86,0xb3,0xbd,0xf7,0x6f,0x61,0x22,0xa2,0x26,0x34,
0x2a,0xbe,0x1e,0x46,0x14,0x68,0x9d,0x44,0x18,0xc2,0x40,0xf4,0x7e,0x5f,0x1b,0xad,
0x0b,0x94,0xb6,0x67,0xb4,0x0b,0xe1,0xea,0x95,0x9c,0x66,0xdc,0xe7,0x5d,0x6c,0x05,
0xda,0xd5,0xdf,0x7a,0xef,0xf6,0xdb,0x1f,0x82,0x4c,0xc0,0x68,0x47,0xa1,0xbd,0xee,
0x39,0x50,0x56,0x4a,0xdd,0xdf,0xa5,0xf8,0xc6,0xda,0xca,0x90,0xca,0x01,0x42,0x9d,
0x8b,0x0c,0x73,0x43,0x75,0x05,0x94,0xde,0x24,0xb3,0x80,0x34,0xe5,0x2c,0xdc,0x9b,
0x3f,0xca,0x33,0x45,0xd0,0xdb,0x5f,0xf5,0x52,0xc3,0x21,0xda,0xe2,0x22,0x72,0x6b,
0x3e,0xd0,0x5b,0xa8,0x87,0x8c,0x06,0x5d,0x0f,0xdd,0x09,0x19,0x93,0xd0,0xb9,0xfc,
0x8b,0x0f,0x84,0x60,0x33,0x1c,0x9b,0x45,0xf1,0xf0,0xa3,0x94,0x3a,0x12,0x77,0x33,
0x4d,0x44,0x78,0x28,0x3c,0x9e,0xfd,0x65,0x57,0x16,0x94,0x6b,0xfb,0x59,0xd0,0xc8,
0x22,0x36,0xdb,0xd2,0x63,0x98,0x43,0xa1,0x04,0x87,0x86,0xf7,0xa6,0x26,0xbb,0xd6,
0x59,0x4d,0xbf,0x6a,0x2e,0xaa,0x2b,0xef,0xe6,0x78,0xb6,0x4e,0xe0,0x2f,0xdc,0x7c,
0xbe,0x57,0x19,0x32,0x7e,0x2a,0xd0,0xb8,0xba,0x29,0x00,0x3c,0x52,0x7d,0xa8,0x49,
0x3b,0x2d,0xeb,0x25,0x49,0xfa,0xa3,0xaa,0x39,0xa7,0xc5,0xa7,0x50,0x11,0x36,0xfb,
0xc6,0x67,0x4a,0xf5,0xa5,0x12,0x65,0x7e,0xb0,0xdf,0xaf,0x4e,0xb3,0x61,0x7f,0x2f } };
void CLASS gamma_curve (double pwr, double ts, int mode, int imax)
{
int i;
double g[6], bnd[2]={0,0}, r;
g[0] = pwr;
g[1] = ts;
g[2] = g[3] = g[4] = 0;
bnd[g[1] >= 1] = 1;
if (g[1] && (g[1]-1)*(g[0]-1) <= 0) {
for (i=0; i < 48; i++) {
g[2] = (bnd[0] + bnd[1])/2;
if (g[0]) bnd[(pow(g[2]/g[1],-g[0]) - 1)/g[0] - 1/g[2] > -1] = g[2];
else bnd[g[2]/exp(1-1/g[2]) < g[1]] = g[2];
}
g[3] = g[2] / g[1];
if (g[0]) g[4] = g[2] * (1/g[0] - 1);
}
if (g[0]) g[5] = 1 / (g[1]*SQR(g[3])/2 - g[4]*(1 - g[3]) +
(1 - pow(g[3],1+g[0]))*(1 + g[4])/(1 + g[0])) - 1;
else g[5] = 1 / (g[1]*SQR(g[3])/2 + 1
- g[2] - g[3] - g[2]*g[3]*(log(g[3]) - 1)) - 1;
if (!mode--) {
memcpy (gamm, g, sizeof gamm);
return;
}
for (i=0; i < 0x10000; i++) {
curve[i] = 0xffff;
if ((r = (double) i / imax) < 1)
curve[i] = 0x10000 * ( mode
? (r < g[3] ? r*g[1] : (g[0] ? pow( r,g[0])*(1+g[4])-g[4] : log(r)*g[2]+1))
: (r < g[2] ? r/g[1] : (g[0] ? pow((r+g[4])/(1+g[4]),1/g[0]) : exp((r-1)/g[2]))));
}
}
void CLASS pseudoinverse (double (*in)[3], double (*out)[3], int size)
{
double work[3][6], num;
int i, j, k;
for (i=0; i < 3; i++) {
for (j=0; j < 6; j++)
work[i][j] = j == i+3;
for (j=0; j < 3; j++)
for (k=0; k < size; k++)
work[i][j] += in[k][i] * in[k][j];
}
for (i=0; i < 3; i++) {
num = work[i][i];
for (j=0; j < 6; j++)
work[i][j] /= num;
for (k=0; k < 3; k++) {
if (k==i) continue;
num = work[k][i];
for (j=0; j < 6; j++)
work[k][j] -= work[i][j] * num;
}
}
for (i=0; i < size; i++)
for (j=0; j < 3; j++)
for (out[i][j]=k=0; k < 3; k++)
out[i][j] += work[j][k+3] * in[i][k];
}
void CLASS cam_xyz_coeff (float _rgb_cam[3][4], double cam_xyz[4][3])
{
double cam_rgb[4][3], inverse[4][3], num;
int i, j, k;
for (i=0; i < colors; i++) /* Multiply out XYZ colorspace */
for (j=0; j < 3; j++)
for (cam_rgb[i][j] = k=0; k < 3; k++)
cam_rgb[i][j] += cam_xyz[i][k] * xyz_rgb[k][j];
for (i=0; i < colors; i++) { /* Normalize cam_rgb so that */
for (num=j=0; j < 3; j++) /* cam_rgb * (1,1,1) is (1,1,1,1) */
num += cam_rgb[i][j];
if(num > 0.00001)
{
for (j=0; j < 3; j++)
cam_rgb[i][j] /= num;
pre_mul[i] = 1 / num;
}
else
{
for (j=0; j < 3; j++)
cam_rgb[i][j] = 0.0;
pre_mul[i] = 1.0;
}
}
pseudoinverse (cam_rgb, inverse, colors);
for (i=0; i < 3; i++)
for (j=0; j < colors; j++)
_rgb_cam[i][j] = inverse[j][i];
}
#ifdef COLORCHECK
void CLASS colorcheck()
{
#define NSQ 24
// Coordinates of the GretagMacbeth ColorChecker squares
// width, height, 1st_column, 1st_row
int cut[NSQ][4]; // you must set these
// ColorChecker Chart under 6500-kelvin illumination
static const double gmb_xyY[NSQ][3] = {
{ 0.400, 0.350, 10.1 }, // Dark Skin
{ 0.377, 0.345, 35.8 }, // Light Skin
{ 0.247, 0.251, 19.3 }, // Blue Sky
{ 0.337, 0.422, 13.3 }, // Foliage
{ 0.265, 0.240, 24.3 }, // Blue Flower
{ 0.261, 0.343, 43.1 }, // Bluish Green
{ 0.506, 0.407, 30.1 }, // Orange
{ 0.211, 0.175, 12.0 }, // Purplish Blue
{ 0.453, 0.306, 19.8 }, // Moderate Red
{ 0.285, 0.202, 6.6 }, // Purple
{ 0.380, 0.489, 44.3 }, // Yellow Green
{ 0.473, 0.438, 43.1 }, // Orange Yellow
{ 0.187, 0.129, 6.1 }, // Blue
{ 0.305, 0.478, 23.4 }, // Green
{ 0.539, 0.313, 12.0 }, // Red
{ 0.448, 0.470, 59.1 }, // Yellow
{ 0.364, 0.233, 19.8 }, // Magenta
{ 0.196, 0.252, 19.8 }, // Cyan
{ 0.310, 0.316, 90.0 }, // White
{ 0.310, 0.316, 59.1 }, // Neutral 8
{ 0.310, 0.316, 36.2 }, // Neutral 6.5
{ 0.310, 0.316, 19.8 }, // Neutral 5
{ 0.310, 0.316, 9.0 }, // Neutral 3.5
{ 0.310, 0.316, 3.1 } }; // Black
double gmb_cam[NSQ][4], gmb_xyz[NSQ][3];
double inverse[NSQ][3], cam_xyz[4][3], balance[4], num;
int c, i, j, k, sq, row, col, pass, count[4];
memset (gmb_cam, 0, sizeof gmb_cam);
for (sq=0; sq < NSQ; sq++) {
FORCC count[c] = 0;
for (row=cut[sq][3]; row < cut[sq][3]+cut[sq][1]; row++)
for (col=cut[sq][2]; col < cut[sq][2]+cut[sq][0]; col++) {
c = FC(row,col);
if (c >= colors) c -= 2;
gmb_cam[sq][c] += BAYER2(row,col);
BAYER2(row,col) = black + (BAYER2(row,col)-black)/2;
count[c]++;
}
FORCC gmb_cam[sq][c] = gmb_cam[sq][c]/count[c] - black;
gmb_xyz[sq][0] = gmb_xyY[sq][2] * gmb_xyY[sq][0] / gmb_xyY[sq][1];
gmb_xyz[sq][1] = gmb_xyY[sq][2];
gmb_xyz[sq][2] = gmb_xyY[sq][2] *
(1 - gmb_xyY[sq][0] - gmb_xyY[sq][1]) / gmb_xyY[sq][1];
}
pseudoinverse (gmb_xyz, inverse, NSQ);
for (pass=0; pass < 2; pass++) {
for (raw_color = i=0; i < colors; i++)
for (j=0; j < 3; j++)
for (cam_xyz[i][j] = k=0; k < NSQ; k++)
cam_xyz[i][j] += gmb_cam[k][i] * inverse[k][j];
cam_xyz_coeff (rgb_cam, cam_xyz);
FORCC balance[c] = pre_mul[c] * gmb_cam[20][c];
for (sq=0; sq < NSQ; sq++)
FORCC gmb_cam[sq][c] *= balance[c];
}
if (verbose) {
printf (" { \"%s %s\", %d,\n\t{", make, model, black);
num = 10000 / (cam_xyz[1][0] + cam_xyz[1][1] + cam_xyz[1][2]);
FORCC for (j=0; j < 3; j++)
printf ("%c%d", (c | j) ? ',':' ', (int) (cam_xyz[c][j] * num + 0.5));
puts (" } },");
}
#undef NSQ
}
#endif
void CLASS hat_transform (float *temp, float *base, int st, int size, int sc)
{
int i;
for (i=0; i < sc; i++)
temp[i] = 2*base[st*i] + base[st*(sc-i)] + base[st*(i+sc)];
for (; i+sc < size; i++)
temp[i] = 2*base[st*i] + base[st*(i-sc)] + base[st*(i+sc)];
for (; i < size; i++)
temp[i] = 2*base[st*i] + base[st*(i-sc)] + base[st*(2*size-2-(i+sc))];
}
#if !defined(LIBRAW_USE_OPENMP)
void CLASS wavelet_denoise()
{
float *fimg=0, *temp, thold, mul[2], avg, diff;
int scale=1, size, lev, hpass, lpass, row, col, nc, c, i, wlast, blk[2];
ushort *window[4];
static const float noise[] =
{ 0.8002,0.2735,0.1202,0.0585,0.0291,0.0152,0.0080,0.0044 };
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Wavelet denoising...\n"));
#endif
while (maximum << scale < 0x10000) scale++;
maximum <<= --scale;
black <<= scale;
FORC4 cblack[c] <<= scale;
if ((size = iheight*iwidth) < 0x15550000)
fimg = (float *) malloc ((size*3 + iheight + iwidth) * sizeof *fimg);
merror (fimg, "wavelet_denoise()");
temp = fimg + size*3;
if ((nc = colors) == 3 && filters) nc++;
FORC(nc) { /* denoise R,G1,B,G3 individually */
for (i=0; i < size; i++)
fimg[i] = 256 * sqrt((double)(image[i][c] << scale));
for (hpass=lev=0; lev < 5; lev++) {
lpass = size*((lev & 1)+1);
for (row=0; row < iheight; row++) {
hat_transform (temp, fimg+hpass+row*iwidth, 1, iwidth, 1 << lev);
for (col=0; col < iwidth; col++)
fimg[lpass + row*iwidth + col] = temp[col] * 0.25;
}
for (col=0; col < iwidth; col++) {
hat_transform (temp, fimg+lpass+col, iwidth, iheight, 1 << lev);
for (row=0; row < iheight; row++)
fimg[lpass + row*iwidth + col] = temp[row] * 0.25;
}
thold = threshold * noise[lev];
for (i=0; i < size; i++) {
fimg[hpass+i] -= fimg[lpass+i];
if (fimg[hpass+i] < -thold) fimg[hpass+i] += thold;
else if (fimg[hpass+i] > thold) fimg[hpass+i] -= thold;
else fimg[hpass+i] = 0;
if (hpass) fimg[i] += fimg[hpass+i];
}
hpass = lpass;
}
for (i=0; i < size; i++)
image[i][c] = CLIP(SQR(fimg[i]+fimg[lpass+i])/0x10000);
}
if (filters && colors == 3) { /* pull G1 and G3 closer together */
for (row=0; row < 2; row++) {
mul[row] = 0.125 * pre_mul[FC(row+1,0) | 1] / pre_mul[FC(row,0) | 1];
blk[row] = cblack[FC(row,0) | 1];
}
for (i=0; i < 4; i++)
window[i] = (ushort *) fimg + width*i;
for (wlast=-1, row=1; row < height-1; row++) {
while (wlast < row+1) {
for (wlast++, i=0; i < 4; i++)
window[(i+3) & 3] = window[i];
for (col = FC(wlast,1) & 1; col < width; col+=2)
window[2][col] = BAYER(wlast,col);
}
thold = threshold/512;
for (col = (FC(row,0) & 1)+1; col < width-1; col+=2) {
avg = ( window[0][col-1] + window[0][col+1] +
window[2][col-1] + window[2][col+1] - blk[~row & 1]*4 )
* mul[row & 1] + (window[1][col] + blk[row & 1]) * 0.5;
avg = avg < 0 ? 0 : sqrt(avg);
diff = sqrt((double)BAYER(row,col)) - avg;
if (diff < -thold) diff += thold;
else if (diff > thold) diff -= thold;
else diff = 0;
BAYER(row,col) = CLIP(SQR(avg+diff) + 0.5);
}
}
}
free (fimg);
}
#else /* LIBRAW_USE_OPENMP */
void CLASS wavelet_denoise()
{
float *fimg=0, *temp, thold, mul[2], avg, diff;
int scale=1, size, lev, hpass, lpass, row, col, nc, c, i, wlast, blk[2];
ushort *window[4];
static const float noise[] =
{ 0.8002,0.2735,0.1202,0.0585,0.0291,0.0152,0.0080,0.0044 };
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Wavelet denoising...\n"));
#endif
while (maximum << scale < 0x10000) scale++;
maximum <<= --scale;
black <<= scale;
FORC4 cblack[c] <<= scale;
if ((size = iheight*iwidth) < 0x15550000)
fimg = (float *) malloc ((size*3 + iheight + iwidth) * sizeof *fimg);
merror (fimg, "wavelet_denoise()");
temp = fimg + size*3;
if ((nc = colors) == 3 && filters) nc++;
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp parallel default(shared) private(i,col,row,thold,lev,lpass,hpass,temp,c) firstprivate(scale,size)
#endif
{
temp = (float*)malloc( (iheight + iwidth) * sizeof *fimg);
FORC(nc) { /* denoise R,G1,B,G3 individually */
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp for
#endif
for (i=0; i < size; i++)
fimg[i] = 256 * sqrt((double)(image[i][c] << scale));
for (hpass=lev=0; lev < 5; lev++) {
lpass = size*((lev & 1)+1);
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp for
#endif
for (row=0; row < iheight; row++) {
hat_transform (temp, fimg+hpass+row*iwidth, 1, iwidth, 1 << lev);
for (col=0; col < iwidth; col++)
fimg[lpass + row*iwidth + col] = temp[col] * 0.25;
}
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp for
#endif
for (col=0; col < iwidth; col++) {
hat_transform (temp, fimg+lpass+col, iwidth, iheight, 1 << lev);
for (row=0; row < iheight; row++)
fimg[lpass + row*iwidth + col] = temp[row] * 0.25;
}
thold = threshold * noise[lev];
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp for
#endif
for (i=0; i < size; i++) {
fimg[hpass+i] -= fimg[lpass+i];
if (fimg[hpass+i] < -thold) fimg[hpass+i] += thold;
else if (fimg[hpass+i] > thold) fimg[hpass+i] -= thold;
else fimg[hpass+i] = 0;
if (hpass) fimg[i] += fimg[hpass+i];
}
hpass = lpass;
}
#ifdef LIBRAW_LIBRARY_BUILD
#pragma omp for
#endif
for (i=0; i < size; i++)
image[i][c] = CLIP(SQR(fimg[i]+fimg[lpass+i])/0x10000);
}
free(temp);
} /* end omp parallel */
/* the following loops are hard to parallize, no idea yes,
* problem is wlast which is carrying dependency
* second part should be easyer, but did not yet get it right.
*/
if (filters && colors == 3) { /* pull G1 and G3 closer together */
for (row=0; row < 2; row++){
mul[row] = 0.125 * pre_mul[FC(row+1,0) | 1] / pre_mul[FC(row,0) | 1];
blk[row] = cblack[FC(row,0) | 1];
}
for (i=0; i < 4; i++)
window[i] = (ushort *) fimg + width*i;
for (wlast=-1, row=1; row < height-1; row++) {
while (wlast < row+1) {
for (wlast++, i=0; i < 4; i++)
window[(i+3) & 3] = window[i];
for (col = FC(wlast,1) & 1; col < width; col+=2)
window[2][col] = BAYER(wlast,col);
}
thold = threshold/512;
for (col = (FC(row,0) & 1)+1; col < width-1; col+=2) {
avg = ( window[0][col-1] + window[0][col+1] +
window[2][col-1] + window[2][col+1] - blk[~row & 1]*4 )
* mul[row & 1] + (window[1][col] + blk[row & 1]) * 0.5;
avg = avg < 0 ? 0 : sqrt(avg);
diff = sqrt((double)BAYER(row,col)) - avg;
if (diff < -thold) diff += thold;
else if (diff > thold) diff -= thold;
else diff = 0;
BAYER(row,col) = CLIP(SQR(avg+diff) + 0.5);
}
}
}
free (fimg);
}
#endif
// green equilibration
void CLASS green_matching()
{
int i,j;
double m1,m2,c1,c2;
int o1_1,o1_2,o1_3,o1_4;
int o2_1,o2_2,o2_3,o2_4;
ushort (*img)[4];
const int margin = 3;
int oj = 2, oi = 2;
float f;
const float thr = 0.01f;
if(half_size || shrink) return;
if(FC(oj, oi) != 3) oj++;
if(FC(oj, oi) != 3) oi++;
if(FC(oj, oi) != 3) oj--;
img = (ushort (*)[4]) calloc (height*width, sizeof *image);
merror (img, "green_matching()");
memcpy(img,image,height*width*sizeof *image);
for(j=oj;j<height-margin;j+=2)
for(i=oi;i<width-margin;i+=2){
o1_1=img[(j-1)*width+i-1][1];
o1_2=img[(j-1)*width+i+1][1];
o1_3=img[(j+1)*width+i-1][1];
o1_4=img[(j+1)*width+i+1][1];
o2_1=img[(j-2)*width+i][3];
o2_2=img[(j+2)*width+i][3];
o2_3=img[j*width+i-2][3];
o2_4=img[j*width+i+2][3];
m1=(o1_1+o1_2+o1_3+o1_4)/4.0;
m2=(o2_1+o2_2+o2_3+o2_4)/4.0;
c1=(abs(o1_1-o1_2)+abs(o1_1-o1_3)+abs(o1_1-o1_4)+abs(o1_2-o1_3)+abs(o1_3-o1_4)+abs(o1_2-o1_4))/6.0;
c2=(abs(o2_1-o2_2)+abs(o2_1-o2_3)+abs(o2_1-o2_4)+abs(o2_2-o2_3)+abs(o2_3-o2_4)+abs(o2_2-o2_4))/6.0;
if((img[j*width+i][3]<maximum*0.95)&&(c1<maximum*thr)&&(c2<maximum*thr))
{
f = image[j*width+i][3]*m1/m2;
image[j*width+i][3]=f>0xffff?0xffff:f;
}
}
free(img);
}
void CLASS scale_colors()
{
unsigned bottom, right, size, row, col, ur, uc, i, x, y, c, sum[8];
int val, dark, sat;
double dsum[8], dmin, dmax;
float scale_mul[4], fr, fc;
ushort *img=0, *pix;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_SCALE_COLORS,0,2);
#endif
if (user_mul[0])
memcpy (pre_mul, user_mul, sizeof pre_mul);
if (use_auto_wb || (use_camera_wb && cam_mul[0] == -1)) {
memset (dsum, 0, sizeof dsum);
bottom = MIN (greybox[1]+greybox[3], height);
right = MIN (greybox[0]+greybox[2], width);
for (row=greybox[1]; row < bottom; row += 8)
for (col=greybox[0]; col < right; col += 8) {
memset (sum, 0, sizeof sum);
for (y=row; y < row+8 && y < bottom; y++)
for (x=col; x < col+8 && x < right; x++)
FORC4 {
if (filters) {
c = fcol(y,x);
val = BAYER2(y,x);
} else
val = image[y*width+x][c];
if (val > maximum-25) goto skip_block;
if ((val -= cblack[c]) < 0) val = 0;
sum[c] += val;
sum[c+4]++;
if (filters) break;
}
FORC(8) dsum[c] += sum[c];
skip_block: ;
}
FORC4 if (dsum[c]) pre_mul[c] = dsum[c+4] / dsum[c];
}
if (use_camera_wb && cam_mul[0] != -1) {
memset (sum, 0, sizeof sum);
for (row=0; row < 8; row++)
for (col=0; col < 8; col++) {
c = FC(row,col);
if ((val = white[row][col] - cblack[c]) > 0)
sum[c] += val;
sum[c+4]++;
}
#ifdef LIBRAW_LIBRARY_BUILD
if(load_raw == &LibRaw::nikon_load_sraw)
{
// Nikon sRAW: camera WB already applied:
pre_mul[0]=pre_mul[1]=pre_mul[2]=pre_mul[3]=1.0;
}
else
#endif
if (sum[0] && sum[1] && sum[2] && sum[3])
FORC4 pre_mul[c] = (float) sum[c+4] / sum[c];
else if (cam_mul[0] && cam_mul[2])
memcpy (pre_mul, cam_mul, sizeof pre_mul);
else
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_BAD_CAMERA_WB;
#endif
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: Cannot use camera white balance.\n"), ifname);
#endif
}
}
#ifdef LIBRAW_LIBRARY_BUILD
// Nikon sRAW, daylight
if (load_raw == &LibRaw::nikon_load_sraw
&& !use_camera_wb && !use_auto_wb
&& cam_mul[0] > 0.001f && cam_mul[1] > 0.001f && cam_mul[2] > 0.001f )
{
for(c=0;c<3;c++)
pre_mul[c]/=cam_mul[c];
}
#endif
if (pre_mul[1] == 0) pre_mul[1] = 1;
if (pre_mul[3] == 0) pre_mul[3] = colors < 4 ? pre_mul[1] : 1;
dark = black;
sat = maximum;
if (threshold) wavelet_denoise();
maximum -= black;
for (dmin=DBL_MAX, dmax=c=0; c < 4; c++) {
if (dmin > pre_mul[c])
dmin = pre_mul[c];
if (dmax < pre_mul[c])
dmax = pre_mul[c];
}
if (!highlight) dmax = dmin;
FORC4 scale_mul[c] = (pre_mul[c] /= dmax) * 65535.0 / maximum;
#ifdef DCRAW_VERBOSE
if (verbose) {
fprintf (stderr,
_("Scaling with darkness %d, saturation %d, and\nmultipliers"), dark, sat);
FORC4 fprintf (stderr, " %f", pre_mul[c]);
fputc ('\n', stderr);
}
#endif
if (filters > 1000 && (cblack[4]+1)/2 == 1 && (cblack[5]+1)/2 == 1) {
FORC4 cblack[FC(c/2,c%2)] +=
cblack[6 + c/2 % cblack[4] * cblack[5] + c%2 % cblack[5]];
cblack[4] = cblack[5] = 0;
}
size = iheight*iwidth;
#ifdef LIBRAW_LIBRARY_BUILD
scale_colors_loop(scale_mul);
#else
for (i=0; i < size*4; i++) {
if (!(val = ((ushort *)image)[i])) continue;
if (cblack[4] && cblack[5])
val -= cblack[6 + i/4 / iwidth % cblack[4] * cblack[5] +
i/4 % iwidth % cblack[5]];
val -= cblack[i & 3];
val *= scale_mul[i & 3];
((ushort *)image)[i] = CLIP(val);
}
#endif
if ((aber[0] != 1 || aber[2] != 1) && colors == 3) {
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Correcting chromatic aberration...\n"));
#endif
for (c=0; c < 4; c+=2) {
if (aber[c] == 1) continue;
img = (ushort *) malloc (size * sizeof *img);
merror (img, "scale_colors()");
for (i=0; i < size; i++)
img[i] = image[i][c];
for (row=0; row < iheight; row++) {
ur = fr = (row - iheight*0.5) * aber[c] + iheight*0.5;
if (ur > iheight-2) continue;
fr -= ur;
for (col=0; col < iwidth; col++) {
uc = fc = (col - iwidth*0.5) * aber[c] + iwidth*0.5;
if (uc > iwidth-2) continue;
fc -= uc;
pix = img + ur*iwidth + uc;
image[row*iwidth+col][c] =
(pix[ 0]*(1-fc) + pix[ 1]*fc) * (1-fr) +
(pix[iwidth]*(1-fc) + pix[iwidth+1]*fc) * fr;
}
}
free(img);
}
}
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_SCALE_COLORS,1,2);
#endif
}
void CLASS pre_interpolate()
{
ushort (*img)[4];
int row, col, c;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_PRE_INTERPOLATE,0,2);
#endif
if (shrink) {
if (half_size) {
height = iheight;
width = iwidth;
if (filters == 9) {
for (row=0; row < 3; row++)
for (col=1; col < 4; col++)
if (!(image[row*width+col][0] | image[row*width+col][2]))
goto break2; break2:
for ( ; row < height; row+=3)
for (col=(col-1)%3+1; col < width-1; col+=3) {
img = image + row*width+col;
for (c=0; c < 3; c+=2)
img[0][c] = (img[-1][c] + img[1][c]) >> 1;
}
}
} else {
img = (ushort (*)[4]) calloc (height, width*sizeof *img);
merror (img, "pre_interpolate()");
for (row=0; row < height; row++)
for (col=0; col < width; col++) {
c = fcol(row,col);
img[row*width+col][c] = image[(row >> 1)*iwidth+(col >> 1)][c];
}
free (image);
image = img;
shrink = 0;
}
}
if (filters > 1000 && colors == 3) {
mix_green = four_color_rgb ^ half_size;
if (four_color_rgb | half_size) colors++;
else {
for (row = FC(1,0) >> 1; row < height; row+=2)
for (col = FC(row,1) & 1; col < width; col+=2)
image[row*width+col][1] = image[row*width+col][3];
filters &= ~((filters & 0x55555555) << 1);
}
}
if (half_size) filters = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_PRE_INTERPOLATE,1,2);
#endif
}
void CLASS border_interpolate (int border)
{
unsigned row, col, y, x, f, c, sum[8];
for (row=0; row < height; row++)
for (col=0; col < width; col++) {
if (col==border && row >= border && row < height-border)
col = width-border;
memset (sum, 0, sizeof sum);
for (y=row-1; y != row+2; y++)
for (x=col-1; x != col+2; x++)
if (y < height && x < width) {
f = fcol(y,x);
sum[f] += image[y*width+x][f];
sum[f+4]++;
}
f = fcol(row,col);
FORCC if (c != f && sum[c+4])
image[row*width+col][c] = sum[c] / sum[c+4];
}
}
void CLASS lin_interpolate_loop(int code[16][16][32],int size)
{
int row;
for (row=1; row < height-1; row++)
{
int col,*ip;
ushort *pix;
for (col=1; col < width-1; col++) {
int i;
int sum[4];
pix = image[row*width+col];
ip = code[row % size][col % size];
memset (sum, 0, sizeof sum);
for (i=*ip++; i--; ip+=3)
sum[ip[2]] += pix[ip[0]] << ip[1];
for (i=colors; --i; ip+=2)
pix[ip[0]] = sum[ip[0]] * ip[1] >> 8;
}
}
}
void CLASS lin_interpolate()
{
int code[16][16][32], size=16, *ip, sum[4];
int f, c, x, y, row, col, shift, color;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Bilinear interpolation...\n"));
#endif
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,0,3);
#endif
if (filters == 9) size = 6;
border_interpolate(1);
for (row=0; row < size; row++)
for (col=0; col < size; col++) {
ip = code[row][col]+1;
f = fcol(row,col);
memset (sum, 0, sizeof sum);
for (y=-1; y <= 1; y++)
for (x=-1; x <= 1; x++) {
shift = (y==0) + (x==0);
color = fcol(row+y,col+x);
if (color == f) continue;
*ip++ = (width*y + x)*4 + color;
*ip++ = shift;
*ip++ = color;
sum[color] += 1 << shift;
}
code[row][col][0] = (ip - code[row][col]) / 3;
FORCC
if (c != f) {
*ip++ = c;
*ip++ = sum[c]>0?256 / sum[c]:0;
}
}
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,1,3);
#endif
lin_interpolate_loop(code,size);
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,2,3);
#endif
}
/*
This algorithm is officially called:
"Interpolation using a Threshold-based variable number of gradients"
described in http://scien.stanford.edu/pages/labsite/1999/psych221/projects/99/tingchen/algodep/vargra.html
I've extended the basic idea to work with non-Bayer filter arrays.
Gradients are numbered clockwise from NW=0 to W=7.
*/
void CLASS vng_interpolate()
{
static const signed char *cp, terms[] = {
-2,-2,+0,-1,0,0x01, -2,-2,+0,+0,1,0x01, -2,-1,-1,+0,0,0x01,
-2,-1,+0,-1,0,0x02, -2,-1,+0,+0,0,0x03, -2,-1,+0,+1,1,0x01,
-2,+0,+0,-1,0,0x06, -2,+0,+0,+0,1,0x02, -2,+0,+0,+1,0,0x03,
-2,+1,-1,+0,0,0x04, -2,+1,+0,-1,1,0x04, -2,+1,+0,+0,0,0x06,
-2,+1,+0,+1,0,0x02, -2,+2,+0,+0,1,0x04, -2,+2,+0,+1,0,0x04,
-1,-2,-1,+0,0,0x80, -1,-2,+0,-1,0,0x01, -1,-2,+1,-1,0,0x01,
-1,-2,+1,+0,1,0x01, -1,-1,-1,+1,0,0x88, -1,-1,+1,-2,0,0x40,
-1,-1,+1,-1,0,0x22, -1,-1,+1,+0,0,0x33, -1,-1,+1,+1,1,0x11,
-1,+0,-1,+2,0,0x08, -1,+0,+0,-1,0,0x44, -1,+0,+0,+1,0,0x11,
-1,+0,+1,-2,1,0x40, -1,+0,+1,-1,0,0x66, -1,+0,+1,+0,1,0x22,
-1,+0,+1,+1,0,0x33, -1,+0,+1,+2,1,0x10, -1,+1,+1,-1,1,0x44,
-1,+1,+1,+0,0,0x66, -1,+1,+1,+1,0,0x22, -1,+1,+1,+2,0,0x10,
-1,+2,+0,+1,0,0x04, -1,+2,+1,+0,1,0x04, -1,+2,+1,+1,0,0x04,
+0,-2,+0,+0,1,0x80, +0,-1,+0,+1,1,0x88, +0,-1,+1,-2,0,0x40,
+0,-1,+1,+0,0,0x11, +0,-1,+2,-2,0,0x40, +0,-1,+2,-1,0,0x20,
+0,-1,+2,+0,0,0x30, +0,-1,+2,+1,1,0x10, +0,+0,+0,+2,1,0x08,
+0,+0,+2,-2,1,0x40, +0,+0,+2,-1,0,0x60, +0,+0,+2,+0,1,0x20,
+0,+0,+2,+1,0,0x30, +0,+0,+2,+2,1,0x10, +0,+1,+1,+0,0,0x44,
+0,+1,+1,+2,0,0x10, +0,+1,+2,-1,1,0x40, +0,+1,+2,+0,0,0x60,
+0,+1,+2,+1,0,0x20, +0,+1,+2,+2,0,0x10, +1,-2,+1,+0,0,0x80,
+1,-1,+1,+1,0,0x88, +1,+0,+1,+2,0,0x08, +1,+0,+2,-1,0,0x40,
+1,+0,+2,+1,0,0x10
}, chood[] = { -1,-1, -1,0, -1,+1, 0,+1, +1,+1, +1,0, +1,-1, 0,-1 };
ushort (*brow[5])[4], *pix;
int prow=8, pcol=2, *ip, *code[16][16], gval[8], gmin, gmax, sum[4];
int row, col, x, y, x1, x2, y1, y2, t, weight, grads, color, diag;
int g, diff, thold, num, c;
lin_interpolate();
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("VNG interpolation...\n"));
#endif
if (filters == 1) prow = pcol = 16;
if (filters == 9) prow = pcol = 6;
ip = (int *) calloc (prow*pcol, 1280);
merror (ip, "vng_interpolate()");
for (row=0; row < prow; row++) /* Precalculate for VNG */
for (col=0; col < pcol; col++) {
code[row][col] = ip;
for (cp=terms, t=0; t < 64; t++) {
y1 = *cp++; x1 = *cp++;
y2 = *cp++; x2 = *cp++;
weight = *cp++;
grads = *cp++;
color = fcol(row+y1,col+x1);
if (fcol(row+y2,col+x2) != color) continue;
diag = (fcol(row,col+1) == color && fcol(row+1,col) == color) ? 2:1;
if (abs(y1-y2) == diag && abs(x1-x2) == diag) continue;
*ip++ = (y1*width + x1)*4 + color;
*ip++ = (y2*width + x2)*4 + color;
*ip++ = weight;
for (g=0; g < 8; g++)
if (grads & 1<<g) *ip++ = g;
*ip++ = -1;
}
*ip++ = INT_MAX;
for (cp=chood, g=0; g < 8; g++) {
y = *cp++; x = *cp++;
*ip++ = (y*width + x) * 4;
color = fcol(row,col);
if (fcol(row+y,col+x) != color && fcol(row+y*2,col+x*2) == color)
*ip++ = (y*width + x) * 8 + color;
else
*ip++ = 0;
}
}
brow[4] = (ushort (*)[4]) calloc (width*3, sizeof **brow);
merror (brow[4], "vng_interpolate()");
for (row=0; row < 3; row++)
brow[row] = brow[4] + row*width;
for (row=2; row < height-2; row++) { /* Do VNG interpolation */
#ifdef LIBRAW_LIBRARY_BUILD
if(!((row-2)%256))RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,(row-2)/256+1,((height-3)/256)+1);
#endif
for (col=2; col < width-2; col++) {
pix = image[row*width+col];
ip = code[row % prow][col % pcol];
memset (gval, 0, sizeof gval);
while ((g = ip[0]) != INT_MAX) { /* Calculate gradients */
diff = ABS(pix[g] - pix[ip[1]]) << ip[2];
gval[ip[3]] += diff;
ip += 5;
if ((g = ip[-1]) == -1) continue;
gval[g] += diff;
while ((g = *ip++) != -1)
gval[g] += diff;
}
ip++;
gmin = gmax = gval[0]; /* Choose a threshold */
for (g=1; g < 8; g++) {
if (gmin > gval[g]) gmin = gval[g];
if (gmax < gval[g]) gmax = gval[g];
}
if (gmax == 0) {
memcpy (brow[2][col], pix, sizeof *image);
continue;
}
thold = gmin + (gmax >> 1);
memset (sum, 0, sizeof sum);
color = fcol(row,col);
for (num=g=0; g < 8; g++,ip+=2) { /* Average the neighbors */
if (gval[g] <= thold) {
FORCC
if (c == color && ip[1])
sum[c] += (pix[c] + pix[ip[1]]) >> 1;
else
sum[c] += pix[ip[0] + c];
num++;
}
}
FORCC { /* Save to buffer */
t = pix[color];
if (c != color)
t += (sum[c] - sum[color]) / num;
brow[2][col][c] = CLIP(t);
}
}
if (row > 3) /* Write buffer to image */
memcpy (image[(row-2)*width+2], brow[0]+2, (width-4)*sizeof *image);
for (g=0; g < 4; g++)
brow[(g-1) & 3] = brow[g];
}
memcpy (image[(row-2)*width+2], brow[0]+2, (width-4)*sizeof *image);
memcpy (image[(row-1)*width+2], brow[1]+2, (width-4)*sizeof *image);
free (brow[4]);
free (code[0][0]);
}
/*
Patterned Pixel Grouping Interpolation by Alain Desbiolles
*/
void CLASS ppg_interpolate()
{
int dir[5] = { 1, width, -1, -width, 1 };
int row, col, diff[2], guess[2], c, d, i;
ushort (*pix)[4];
border_interpolate(3);
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("PPG interpolation...\n"));
#endif
/* Fill in the green layer with gradients and pattern recognition: */
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,0,3);
#ifdef LIBRAW_USE_OPENMP
#pragma omp parallel for default(shared) private(guess, diff, row, col, d, c, i, pix) schedule(static)
#endif
#endif
for (row=3; row < height-3; row++)
for (col=3+(FC(row,3) & 1), c=FC(row,col); col < width-3; col+=2) {
pix = image + row*width+col;
for (i=0; (d=dir[i]) > 0; i++) {
guess[i] = (pix[-d][1] + pix[0][c] + pix[d][1]) * 2
- pix[-2*d][c] - pix[2*d][c];
diff[i] = ( ABS(pix[-2*d][c] - pix[ 0][c]) +
ABS(pix[ 2*d][c] - pix[ 0][c]) +
ABS(pix[ -d][1] - pix[ d][1]) ) * 3 +
( ABS(pix[ 3*d][1] - pix[ d][1]) +
ABS(pix[-3*d][1] - pix[-d][1]) ) * 2;
}
d = dir[i = diff[0] > diff[1]];
pix[0][1] = ULIM(guess[i] >> 2, pix[d][1], pix[-d][1]);
}
/* Calculate red and blue for each green pixel: */
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,1,3);
#ifdef LIBRAW_USE_OPENMP
#pragma omp parallel for default(shared) private(guess, diff, row, col, d, c, i, pix) schedule(static)
#endif
#endif
for (row=1; row < height-1; row++)
for (col=1+(FC(row,2) & 1), c=FC(row,col+1); col < width-1; col+=2) {
pix = image + row*width+col;
for (i=0; (d=dir[i]) > 0; c=2-c, i++)
pix[0][c] = CLIP((pix[-d][c] + pix[d][c] + 2*pix[0][1]
- pix[-d][1] - pix[d][1]) >> 1);
}
/* Calculate blue for red pixels and vice versa: */
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_INTERPOLATE,2,3);
#ifdef LIBRAW_USE_OPENMP
#pragma omp parallel for default(shared) private(guess, diff, row, col, d, c, i, pix) schedule(static)
#endif
#endif
for (row=1; row < height-1; row++)
for (col=1+(FC(row,1) & 1), c=2-FC(row,col); col < width-1; col+=2) {
pix = image + row*width+col;
for (i=0; (d=dir[i]+dir[i+1]) > 0; i++) {
diff[i] = ABS(pix[-d][c] - pix[d][c]) +
ABS(pix[-d][1] - pix[0][1]) +
ABS(pix[ d][1] - pix[0][1]);
guess[i] = pix[-d][c] + pix[d][c] + 2*pix[0][1]
- pix[-d][1] - pix[d][1];
}
if (diff[0] != diff[1])
pix[0][c] = CLIP(guess[diff[0] > diff[1]] >> 1);
else
pix[0][c] = CLIP((guess[0]+guess[1]) >> 2);
}
}
void CLASS cielab (ushort rgb[3], short lab[3])
{
int c, i, j, k;
float r, xyz[3];
#ifdef LIBRAW_NOTHREADS
static float cbrt[0x10000], xyz_cam[3][4];
#else
#define cbrt tls->ahd_data.cbrt
#define xyz_cam tls->ahd_data.xyz_cam
#endif
if (!rgb) {
#ifndef LIBRAW_NOTHREADS
if(cbrt[0] < -1.0f)
#endif
for (i=0; i < 0x10000; i++) {
r = i / 65535.0;
cbrt[i] = r > 0.008856 ? pow(r,1.f/3.0f) : 7.787f*r + 16.f/116.0f;
}
for (i=0; i < 3; i++)
for (j=0; j < colors; j++)
for (xyz_cam[i][j] = k=0; k < 3; k++)
xyz_cam[i][j] += xyz_rgb[i][k] * rgb_cam[k][j] / d65_white[i];
return;
}
xyz[0] = xyz[1] = xyz[2] = 0.5;
FORCC {
xyz[0] += xyz_cam[0][c] * rgb[c];
xyz[1] += xyz_cam[1][c] * rgb[c];
xyz[2] += xyz_cam[2][c] * rgb[c];
}
xyz[0] = cbrt[CLIP((int) xyz[0])];
xyz[1] = cbrt[CLIP((int) xyz[1])];
xyz[2] = cbrt[CLIP((int) xyz[2])];
lab[0] = 64 * (116 * xyz[1] - 16);
lab[1] = 64 * 500 * (xyz[0] - xyz[1]);
lab[2] = 64 * 200 * (xyz[1] - xyz[2]);
#ifndef LIBRAW_NOTHREADS
#undef cbrt
#undef xyz_cam
#endif
}
#define TS 512 /* Tile Size */
#define fcol(row,col) xtrans[(row+6) % 6][(col+6) % 6]
/*
Frank Markesteijn's algorithm for Fuji X-Trans sensors
*/
void CLASS xtrans_interpolate (int passes)
{
int c, d, f, g, h, i, v, ng, row, col, top, left, mrow, mcol;
int val, ndir, pass, hm[8], avg[4], color[3][8];
static const short orth[12] = { 1,0,0,1,-1,0,0,-1,1,0,0,1 },
patt[2][16] = { { 0,1,0,-1,2,0,-1,0,1,1,1,-1,0,0,0,0 },
{ 0,1,0,-2,1,0,-2,0,1,1,-2,-2,1,-1,-1,1 } },
dir[4] = { 1,TS,TS+1,TS-1 };
short allhex[3][3][2][8], *hex;
ushort min, max, sgrow, sgcol;
ushort (*rgb)[TS][TS][3], (*rix)[3], (*pix)[4];
short (*lab) [TS][3], (*lix)[3];
float (*drv)[TS][TS], diff[6], tr;
char (*homo)[TS][TS], *buffer;
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("%d-pass X-Trans interpolation...\n"), passes);
#endif
cielab (0,0);
ndir = 4 << (passes > 1);
buffer = (char *) malloc (TS*TS*(ndir*11+6));
merror (buffer, "xtrans_interpolate()");
rgb = (ushort(*)[TS][TS][3]) buffer;
lab = (short (*) [TS][3])(buffer + TS*TS*(ndir*6));
drv = (float (*)[TS][TS]) (buffer + TS*TS*(ndir*6+6));
homo = (char (*)[TS][TS]) (buffer + TS*TS*(ndir*10+6));
/* Map a green hexagon around each non-green pixel and vice versa: */
for (row=0; row < 3; row++)
for (col=0; col < 3; col++)
for (ng=d=0; d < 10; d+=2) {
g = fcol(row,col) == 1;
if (fcol(row+orth[d],col+orth[d+2]) == 1) ng=0; else ng++;
if (ng == 4) { sgrow = row; sgcol = col; }
if (ng == g+1) FORC(8) {
v = orth[d ]*patt[g][c*2] + orth[d+1]*patt[g][c*2+1];
h = orth[d+2]*patt[g][c*2] + orth[d+3]*patt[g][c*2+1];
allhex[row][col][0][c^(g*2 & d)] = h + v*width;
allhex[row][col][1][c^(g*2 & d)] = h + v*TS;
}
}
/* Set green1 and green3 to the minimum and maximum allowed values: */
for (row=2; row < height-2; row++)
for (min=~(max=0), col=2; col < width-2; col++) {
if (fcol(row,col) == 1 && (min=~(max=0))) continue;
pix = image + row*width + col;
hex = allhex[row % 3][col % 3][0];
if (!max) FORC(6) {
val = pix[hex[c]][1];
if (min > val) min = val;
if (max < val) max = val;
}
pix[0][1] = min;
pix[0][3] = max;
switch ((row-sgrow) % 3) {
case 1: if (row < height-3) { row++; col--; } break;
case 2: if ((min=~(max=0)) && (col+=2) < width-3 && row > 2) row--;
}
}
for (top=3; top < height-19; top += TS-16)
for (left=3; left < width-19; left += TS-16) {
mrow = MIN (top+TS, height-3);
mcol = MIN (left+TS, width-3);
for (row=top; row < mrow; row++)
for (col=left; col < mcol; col++)
memcpy (rgb[0][row-top][col-left], image[row*width+col], 6);
FORC3 memcpy (rgb[c+1], rgb[0], sizeof *rgb);
/* Interpolate green horizontally, vertically, and along both diagonals: */
for (row=top; row < mrow; row++)
for (col=left; col < mcol; col++) {
if ((f = fcol(row,col)) == 1) continue;
pix = image + row*width + col;
hex = allhex[row % 3][col % 3][0];
color[1][0] = 174 * (pix[ hex[1]][1] + pix[ hex[0]][1]) -
46 * (pix[2*hex[1]][1] + pix[2*hex[0]][1]);
color[1][1] = 223 * pix[ hex[3]][1] + pix[ hex[2]][1] * 33 +
92 * (pix[ 0 ][f] - pix[ -hex[2]][f]);
FORC(2) color[1][2+c] =
164 * pix[hex[4+c]][1] + 92 * pix[-2*hex[4+c]][1] + 33 *
(2*pix[0][f] - pix[3*hex[4+c]][f] - pix[-3*hex[4+c]][f]);
FORC4 rgb[c^!((row-sgrow) % 3)][row-top][col-left][1] =
LIM(color[1][c] >> 8,pix[0][1],pix[0][3]);
}
for (pass=0; pass < passes; pass++) {
if (pass == 1)
memcpy (rgb+=4, buffer, 4*sizeof *rgb);
/* Recalculate green from interpolated values of closer pixels: */
if (pass) {
for (row=top+2; row < mrow-2; row++)
for (col=left+2; col < mcol-2; col++) {
if ((f = fcol(row,col)) == 1) continue;
pix = image + row*width + col;
hex = allhex[row % 3][col % 3][1];
for (d=3; d < 6; d++) {
rix = &rgb[(d-2)^!((row-sgrow) % 3)][row-top][col-left];
val = rix[-2*hex[d]][1] + 2*rix[hex[d]][1]
- rix[-2*hex[d]][f] - 2*rix[hex[d]][f] + 3*rix[0][f];
rix[0][1] = LIM(val/3,pix[0][1],pix[0][3]);
}
}
}
/* Interpolate red and blue values for solitary green pixels: */
for (row=(top-sgrow+4)/3*3+sgrow; row < mrow-2; row+=3)
for (col=(left-sgcol+4)/3*3+sgcol; col < mcol-2; col+=3) {
rix = &rgb[0][row-top][col-left];
h = fcol(row,col+1);
memset (diff, 0, sizeof diff);
for (i=1, d=0; d < 6; d++, i^=TS^1, h^=2) {
for (c=0; c < 2; c++, h^=2) {
g = 2*rix[0][1] - rix[i<<c][1] - rix[-i<<c][1];
color[h][d] = g + rix[i<<c][h] + rix[-i<<c][h];
if (d > 1)
diff[d] += SQR (rix[i<<c][1] - rix[-i<<c][1]
- rix[i<<c][h] + rix[-i<<c][h]) + SQR(g);
}
if (d > 1 && (d & 1))
if (diff[d-1] < diff[d])
FORC(2) color[c*2][d] = color[c*2][d-1];
if (d < 2 || (d & 1)) {
FORC(2) rix[0][c*2] = CLIP(color[c*2][d]/2);
rix += TS*TS;
}
}
}
/* Interpolate red for blue pixels and vice versa: */
for (row=top+3; row < mrow-3; row++)
for (col=left+3; col < mcol-3; col++) {
if ((f = 2-fcol(row,col)) == 1) continue;
rix = &rgb[0][row-top][col-left];
c = (row-sgrow) % 3 ? TS:1;
h = 3 * (c ^ TS ^ 1);
for (d=0; d < 4; d++, rix += TS*TS) {
i = d > 1 || ((d ^ c) & 1) ||
((ABS(rix[0][1]-rix[c][1])+ABS(rix[0][1]-rix[-c][1])) <
2*(ABS(rix[0][1]-rix[h][1])+ABS(rix[0][1]-rix[-h][1]))) ? c:h;
rix[0][f] = CLIP((rix[i][f] + rix[-i][f] +
2*rix[0][1] - rix[i][1] - rix[-i][1])/2);
}
}
/* Fill in red and blue for 2x2 blocks of green: */
for (row=top+2; row < mrow-2; row++) if ((row-sgrow) % 3)
for (col=left+2; col < mcol-2; col++) if ((col-sgcol) % 3) {
rix = &rgb[0][row-top][col-left];
hex = allhex[row % 3][col % 3][1];
for (d=0; d < ndir; d+=2, rix += TS*TS)
if (hex[d] + hex[d+1]) {
g = 3*rix[0][1] - 2*rix[hex[d]][1] - rix[hex[d+1]][1];
for (c=0; c < 4; c+=2) rix[0][c] =
CLIP((g + 2*rix[hex[d]][c] + rix[hex[d+1]][c])/3);
} else {
g = 2*rix[0][1] - rix[hex[d]][1] - rix[hex[d+1]][1];
for (c=0; c < 4; c+=2) rix[0][c] =
CLIP((g + rix[hex[d]][c] + rix[hex[d+1]][c])/2);
}
}
}
rgb = (ushort(*)[TS][TS][3]) buffer;
mrow -= top;
mcol -= left;
/* Convert to CIELab and differentiate in all directions: */
for (d=0; d < ndir; d++) {
for (row=2; row < mrow-2; row++)
for (col=2; col < mcol-2; col++)
cielab (rgb[d][row][col], lab[row][col]);
for (f=dir[d & 3],row=3; row < mrow-3; row++)
for (col=3; col < mcol-3; col++) {
lix = &lab[row][col];
g = 2*lix[0][0] - lix[f][0] - lix[-f][0];
drv[d][row][col] = SQR(g)
+ SQR((2*lix[0][1] - lix[f][1] - lix[-f][1] + g*500/232))
+ SQR((2*lix[0][2] - lix[f][2] - lix[-f][2] - g*500/580));
}
}
/* Build homogeneity maps from the derivatives: */
memset(homo, 0, ndir*TS*TS);
for (row=4; row < mrow-4; row++)
for (col=4; col < mcol-4; col++) {
for (tr=FLT_MAX, d=0; d < ndir; d++)
if (tr > drv[d][row][col])
tr = drv[d][row][col];
tr *= 8;
for (d=0; d < ndir; d++)
for (v=-1; v <= 1; v++)
for (h=-1; h <= 1; h++)
if (drv[d][row+v][col+h] <= tr)
homo[d][row][col]++;
}
/* Average the most homogenous pixels for the final result: */
if (height-top < TS+4) mrow = height-top+2;
if (width-left < TS+4) mcol = width-left+2;
for (row = MIN(top,8); row < mrow-8; row++)
for (col = MIN(left,8); col < mcol-8; col++) {
for (d=0; d < ndir; d++)
for (hm[d]=0, v=-2; v <= 2; v++)
for (h=-2; h <= 2; h++)
hm[d] += homo[d][row+v][col+h];
for (d=0; d < ndir-4; d++)
if (hm[d] < hm[d+4]) hm[d ] = 0; else
if (hm[d] > hm[d+4]) hm[d+4] = 0;
for (max=hm[0],d=1; d < ndir; d++)
if (max < hm[d]) max = hm[d];
max -= max >> 3;
memset (avg, 0, sizeof avg);
for (d=0; d < ndir; d++)
if (hm[d] >= max) {
FORC3 avg[c] += rgb[d][row][col][c];
avg[3]++;
}
FORC3 image[(row+top)*width+col+left][c] = avg[c]/avg[3];
}
}
free(buffer);
border_interpolate(8);
}
#undef fcol
/*
Adaptive Homogeneity-Directed interpolation is based on
the work of Keigo Hirakawa, Thomas Parks, and Paul Lee.
*/
#ifdef LIBRAW_LIBRARY_BUILD
void CLASS ahd_interpolate_green_h_and_v(int top, int left, ushort (*out_rgb)[TS][TS][3])
{
int row, col;
int c, val;
ushort (*pix)[4];
const int rowlimit = MIN(top+TS, height-2);
const int collimit = MIN(left+TS, width-2);
for (row = top; row < rowlimit; row++) {
col = left + (FC(row,left) & 1);
for (c = FC(row,col); col < collimit; col+=2) {
pix = image + row*width+col;
val = ((pix[-1][1] + pix[0][c] + pix[1][1]) * 2
- pix[-2][c] - pix[2][c]) >> 2;
out_rgb[0][row-top][col-left][1] = ULIM(val,pix[-1][1],pix[1][1]);
val = ((pix[-width][1] + pix[0][c] + pix[width][1]) * 2
- pix[-2*width][c] - pix[2*width][c]) >> 2;
out_rgb[1][row-top][col-left][1] = ULIM(val,pix[-width][1],pix[width][1]);
}
}
}
void CLASS ahd_interpolate_r_and_b_in_rgb_and_convert_to_cielab(int top, int left, ushort (*inout_rgb)[TS][3], short (*out_lab)[TS][3])
{
unsigned row, col;
int c, val;
ushort (*pix)[4];
ushort (*rix)[3];
short (*lix)[3];
float xyz[3];
const unsigned num_pix_per_row = 4*width;
const unsigned rowlimit = MIN(top+TS-1, height-3);
const unsigned collimit = MIN(left+TS-1, width-3);
ushort *pix_above;
ushort *pix_below;
int t1, t2;
for (row = top+1; row < rowlimit; row++) {
pix = image + row*width + left;
rix = &inout_rgb[row-top][0];
lix = &out_lab[row-top][0];
for (col = left+1; col < collimit; col++) {
pix++;
pix_above = &pix[0][0] - num_pix_per_row;
pix_below = &pix[0][0] + num_pix_per_row;
rix++;
lix++;
c = 2 - FC(row, col);
if (c == 1) {
c = FC(row+1,col);
t1 = 2-c;
val = pix[0][1] + (( pix[-1][t1] + pix[1][t1]
- rix[-1][1] - rix[1][1] ) >> 1);
rix[0][t1] = CLIP(val);
val = pix[0][1] + (( pix_above[c] + pix_below[c]
- rix[-TS][1] - rix[TS][1] ) >> 1);
} else {
t1 = -4+c; /* -4+c: pixel of color c to the left */
t2 = 4+c; /* 4+c: pixel of color c to the right */
val = rix[0][1] + (( pix_above[t1] + pix_above[t2]
+ pix_below[t1] + pix_below[t2]
- rix[-TS-1][1] - rix[-TS+1][1]
- rix[+TS-1][1] - rix[+TS+1][1] + 1) >> 2);
}
rix[0][c] = CLIP(val);
c = FC(row,col);
rix[0][c] = pix[0][c];
cielab(rix[0],lix[0]);
}
}
}
void CLASS ahd_interpolate_r_and_b_and_convert_to_cielab(int top, int left, ushort (*inout_rgb)[TS][TS][3], short (*out_lab)[TS][TS][3])
{
int direction;
for (direction = 0; direction < 2; direction++) {
ahd_interpolate_r_and_b_in_rgb_and_convert_to_cielab(top, left, inout_rgb[direction], out_lab[direction]);
}
}
void CLASS ahd_interpolate_build_homogeneity_map(int top, int left, short (*lab)[TS][TS][3], char (*out_homogeneity_map)[TS][2])
{
int row, col;
int tr, tc;
int direction;
int i;
short (*lix)[3];
short (*lixs[2])[3];
short *adjacent_lix;
unsigned ldiff[2][4], abdiff[2][4], leps, abeps;
static const int dir[4] = { -1, 1, -TS, TS };
const int rowlimit = MIN(top+TS-2, height-4);
const int collimit = MIN(left+TS-2, width-4);
int homogeneity;
char (*homogeneity_map_p)[2];
memset (out_homogeneity_map, 0, 2*TS*TS);
for (row=top+2; row < rowlimit; row++) {
tr = row-top;
homogeneity_map_p = &out_homogeneity_map[tr][1];
for (direction=0; direction < 2; direction++) {
lixs[direction] = &lab[direction][tr][1];
}
for (col=left+2; col < collimit; col++) {
tc = col-left;
homogeneity_map_p++;
for (direction=0; direction < 2; direction++) {
lix = ++lixs[direction];
for (i=0; i < 4; i++) {
adjacent_lix = lix[dir[i]];
ldiff[direction][i] = ABS(lix[0][0]-adjacent_lix[0]);
abdiff[direction][i] = SQR(lix[0][1]-adjacent_lix[1])
+ SQR(lix[0][2]-adjacent_lix[2]);
}
}
leps = MIN(MAX(ldiff[0][0],ldiff[0][1]),
MAX(ldiff[1][2],ldiff[1][3]));
abeps = MIN(MAX(abdiff[0][0],abdiff[0][1]),
MAX(abdiff[1][2],abdiff[1][3]));
for (direction=0; direction < 2; direction++) {
homogeneity = 0;
for (i=0; i < 4; i++) {
if (ldiff[direction][i] <= leps && abdiff[direction][i] <= abeps) {
homogeneity++;
}
}
homogeneity_map_p[0][direction] = homogeneity;
}
}
}
}
void CLASS ahd_interpolate_combine_homogeneous_pixels(int top, int left, ushort (*rgb)[TS][TS][3], char (*homogeneity_map)[TS][2])
{
int row, col;
int tr, tc;
int i, j;
int direction;
int hm[2];
int c;
const int rowlimit = MIN(top+TS-3, height-5);
const int collimit = MIN(left+TS-3, width-5);
ushort (*pix)[4];
ushort (*rix[2])[3];
for (row=top+3; row < rowlimit; row++) {
tr = row-top;
pix = &image[row*width+left+2];
for (direction = 0; direction < 2; direction++) {
rix[direction] = &rgb[direction][tr][2];
}
for (col=left+3; col < collimit; col++) {
tc = col-left;
pix++;
for (direction = 0; direction < 2; direction++) {
rix[direction]++;
}
for (direction=0; direction < 2; direction++) {
hm[direction] = 0;
for (i=tr-1; i <= tr+1; i++) {
for (j=tc-1; j <= tc+1; j++) {
hm[direction] += homogeneity_map[i][j][direction];
}
}
}
if (hm[0] != hm[1]) {
memcpy(pix[0], rix[hm[1] > hm[0]][0], 3 * sizeof(ushort));
} else {
FORC3 {
pix[0][c] = (rix[0][0][c] + rix[1][0][c]) >> 1;
}
}
}
}
}
void CLASS ahd_interpolate()
{
int i, j, k, top, left;
float xyz_cam[3][4],r;
char *buffer;
ushort (*rgb)[TS][TS][3];
short (*lab)[TS][TS][3];
char (*homo)[TS][2];
int terminate_flag = 0;
cielab(0,0);
border_interpolate(5);
#ifdef LIBRAW_LIBRARY_BUILD
#ifdef LIBRAW_USE_OPENMP
#pragma omp parallel private(buffer,rgb,lab,homo,top,left,i,j,k) shared(xyz_cam,terminate_flag)
#endif
#endif
{
buffer = (char *) malloc (26*TS*TS); /* 1664 kB */
merror (buffer, "ahd_interpolate()");
rgb = (ushort(*)[TS][TS][3]) buffer;
lab = (short (*)[TS][TS][3])(buffer + 12*TS*TS);
homo = (char (*)[TS][2]) (buffer + 24*TS*TS);
#ifdef LIBRAW_LIBRARY_BUILD
#ifdef LIBRAW_USE_OPENMP
#pragma omp for schedule(dynamic)
#endif
#endif
for (top=2; top < height-5; top += TS-6){
#ifdef LIBRAW_LIBRARY_BUILD
#ifdef LIBRAW_USE_OPENMP
if(0== omp_get_thread_num())
#endif
if(callbacks.progress_cb) {
int rr = (*callbacks.progress_cb)(callbacks.progresscb_data,LIBRAW_PROGRESS_INTERPOLATE,top-2,height-7);
if(rr)
terminate_flag = 1;
}
#endif
for (left=2; !terminate_flag && (left < width-5); left += TS-6) {
ahd_interpolate_green_h_and_v(top, left, rgb);
ahd_interpolate_r_and_b_and_convert_to_cielab(top, left, rgb, lab);
ahd_interpolate_build_homogeneity_map(top, left, lab, homo);
ahd_interpolate_combine_homogeneous_pixels(top, left, rgb, homo);
}
}
free (buffer);
}
#ifdef LIBRAW_LIBRARY_BUILD
if(terminate_flag)
throw LIBRAW_EXCEPTION_CANCELLED_BY_CALLBACK;
#endif
}
#else
void CLASS ahd_interpolate()
{
int i, j, top, left, row, col, tr, tc, c, d, val, hm[2];
static const int dir[4] = { -1, 1, -TS, TS };
unsigned ldiff[2][4], abdiff[2][4], leps, abeps;
ushort (*rgb)[TS][TS][3], (*rix)[3], (*pix)[4];
short (*lab)[TS][TS][3], (*lix)[3];
char (*homo)[TS][TS], *buffer;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("AHD interpolation...\n"));
#endif
cielab (0,0);
border_interpolate(5);
buffer = (char *) malloc (26*TS*TS);
merror (buffer, "ahd_interpolate()");
rgb = (ushort(*)[TS][TS][3]) buffer;
lab = (short (*)[TS][TS][3])(buffer + 12*TS*TS);
homo = (char (*)[TS][TS]) (buffer + 24*TS*TS);
for (top=2; top < height-5; top += TS-6)
for (left=2; left < width-5; left += TS-6) {
/* Interpolate green horizontally and vertically: */
for (row=top; row < top+TS && row < height-2; row++) {
col = left + (FC(row,left) & 1);
for (c = FC(row,col); col < left+TS && col < width-2; col+=2) {
pix = image + row*width+col;
val = ((pix[-1][1] + pix[0][c] + pix[1][1]) * 2
- pix[-2][c] - pix[2][c]) >> 2;
rgb[0][row-top][col-left][1] = ULIM(val,pix[-1][1],pix[1][1]);
val = ((pix[-width][1] + pix[0][c] + pix[width][1]) * 2
- pix[-2*width][c] - pix[2*width][c]) >> 2;
rgb[1][row-top][col-left][1] = ULIM(val,pix[-width][1],pix[width][1]);
}
}
/* Interpolate red and blue, and convert to CIELab: */
for (d=0; d < 2; d++)
for (row=top+1; row < top+TS-1 && row < height-3; row++)
for (col=left+1; col < left+TS-1 && col < width-3; col++) {
pix = image + row*width+col;
rix = &rgb[d][row-top][col-left];
lix = &lab[d][row-top][col-left];
if ((c = 2 - FC(row,col)) == 1) {
c = FC(row+1,col);
val = pix[0][1] + (( pix[-1][2-c] + pix[1][2-c]
- rix[-1][1] - rix[1][1] ) >> 1);
rix[0][2-c] = CLIP(val);
val = pix[0][1] + (( pix[-width][c] + pix[width][c]
- rix[-TS][1] - rix[TS][1] ) >> 1);
} else
val = rix[0][1] + (( pix[-width-1][c] + pix[-width+1][c]
+ pix[+width-1][c] + pix[+width+1][c]
- rix[-TS-1][1] - rix[-TS+1][1]
- rix[+TS-1][1] - rix[+TS+1][1] + 1) >> 2);
rix[0][c] = CLIP(val);
c = FC(row,col);
rix[0][c] = pix[0][c];
cielab (rix[0],lix[0]);
}
/* Build homogeneity maps from the CIELab images: */
memset (homo, 0, 2*TS*TS);
for (row=top+2; row < top+TS-2 && row < height-4; row++) {
tr = row-top;
for (col=left+2; col < left+TS-2 && col < width-4; col++) {
tc = col-left;
for (d=0; d < 2; d++) {
lix = &lab[d][tr][tc];
for (i=0; i < 4; i++) {
ldiff[d][i] = ABS(lix[0][0]-lix[dir[i]][0]);
abdiff[d][i] = SQR(lix[0][1]-lix[dir[i]][1])
+ SQR(lix[0][2]-lix[dir[i]][2]);
}
}
leps = MIN(MAX(ldiff[0][0],ldiff[0][1]),
MAX(ldiff[1][2],ldiff[1][3]));
abeps = MIN(MAX(abdiff[0][0],abdiff[0][1]),
MAX(abdiff[1][2],abdiff[1][3]));
for (d=0; d < 2; d++)
for (i=0; i < 4; i++)
if (ldiff[d][i] <= leps && abdiff[d][i] <= abeps)
homo[d][tr][tc]++;
}
}
/* Combine the most homogenous pixels for the final result: */
for (row=top+3; row < top+TS-3 && row < height-5; row++) {
tr = row-top;
for (col=left+3; col < left+TS-3 && col < width-5; col++) {
tc = col-left;
for (d=0; d < 2; d++)
for (hm[d]=0, i=tr-1; i <= tr+1; i++)
for (j=tc-1; j <= tc+1; j++)
hm[d] += homo[d][i][j];
if (hm[0] != hm[1])
FORC3 image[row*width+col][c] = rgb[hm[1] > hm[0]][tr][tc][c];
else
FORC3 image[row*width+col][c] =
(rgb[0][tr][tc][c] + rgb[1][tr][tc][c]) >> 1;
}
}
}
free (buffer);
}
#endif
#undef TS
void CLASS median_filter()
{
ushort (*pix)[4];
int pass, c, i, j, k, med[9];
static const uchar opt[] = /* Optimal 9-element median search */
{ 1,2, 4,5, 7,8, 0,1, 3,4, 6,7, 1,2, 4,5, 7,8,
0,3, 5,8, 4,7, 3,6, 1,4, 2,5, 4,7, 4,2, 6,4, 4,2 };
for (pass=1; pass <= med_passes; pass++) {
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_MEDIAN_FILTER,pass-1,med_passes);
#endif
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Median filter pass %d...\n"), pass);
#endif
for (c=0; c < 3; c+=2) {
for (pix = image; pix < image+width*height; pix++)
pix[0][3] = pix[0][c];
for (pix = image+width; pix < image+width*(height-1); pix++) {
if ((pix-image+1) % width < 2) continue;
for (k=0, i = -width; i <= width; i += width)
for (j = i-1; j <= i+1; j++)
med[k++] = pix[j][3] - pix[j][1];
for (i=0; i < sizeof opt; i+=2)
if (med[opt[i]] > med[opt[i+1]])
SWAP (med[opt[i]] , med[opt[i+1]]);
pix[0][c] = CLIP(med[4] + pix[0][1]);
}
}
}
}
void CLASS blend_highlights()
{
int clip=INT_MAX, row, col, c, i, j;
static const float trans[2][4][4] =
{ { { 1,1,1 }, { 1.7320508,-1.7320508,0 }, { -1,-1,2 } },
{ { 1,1,1,1 }, { 1,-1,1,-1 }, { 1,1,-1,-1 }, { 1,-1,-1,1 } } };
static const float itrans[2][4][4] =
{ { { 1,0.8660254,-0.5 }, { 1,-0.8660254,-0.5 }, { 1,0,1 } },
{ { 1,1,1,1 }, { 1,-1,1,-1 }, { 1,1,-1,-1 }, { 1,-1,-1,1 } } };
float cam[2][4], lab[2][4], sum[2], chratio;
if ((unsigned) (colors-3) > 1) return;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Blending highlights...\n"));
#endif
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_HIGHLIGHTS,0,2);
#endif
FORCC if (clip > (i = 65535*pre_mul[c])) clip = i;
for (row=0; row < height; row++)
for (col=0; col < width; col++) {
FORCC if (image[row*width+col][c] > clip) break;
if (c == colors) continue;
FORCC {
cam[0][c] = image[row*width+col][c];
cam[1][c] = MIN(cam[0][c],clip);
}
for (i=0; i < 2; i++) {
FORCC for (lab[i][c]=j=0; j < colors; j++)
lab[i][c] += trans[colors-3][c][j] * cam[i][j];
for (sum[i]=0,c=1; c < colors; c++)
sum[i] += SQR(lab[i][c]);
}
chratio = sqrt(sum[1]/sum[0]);
for (c=1; c < colors; c++)
lab[0][c] *= chratio;
FORCC for (cam[0][c]=j=0; j < colors; j++)
cam[0][c] += itrans[colors-3][c][j] * lab[0][j];
FORCC image[row*width+col][c] = cam[0][c] / colors;
}
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_HIGHLIGHTS,1,2);
#endif
}
#define SCALE (4 >> shrink)
void CLASS recover_highlights()
{
float *map, sum, wgt, grow;
int hsat[4], count, spread, change, val, i;
unsigned high, wide, mrow, mcol, row, col, kc, c, d, y, x;
ushort *pixel;
static const signed char dir[8][2] =
{ {-1,-1}, {-1,0}, {-1,1}, {0,1}, {1,1}, {1,0}, {1,-1}, {0,-1} };
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Rebuilding highlights...\n"));
#endif
grow = pow (2.0, 4-highlight);
FORCC hsat[c] = 32000 * pre_mul[c];
for (kc=0, c=1; c < colors; c++)
if (pre_mul[kc] < pre_mul[c]) kc = c;
high = height / SCALE;
wide = width / SCALE;
map = (float *) calloc (high, wide*sizeof *map);
merror (map, "recover_highlights()");
FORCC if (c != kc) {
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_HIGHLIGHTS,c-1,colors-1);
#endif
memset (map, 0, high*wide*sizeof *map);
for (mrow=0; mrow < high; mrow++)
for (mcol=0; mcol < wide; mcol++) {
sum = wgt = count = 0;
for (row = mrow*SCALE; row < (mrow+1)*SCALE; row++)
for (col = mcol*SCALE; col < (mcol+1)*SCALE; col++) {
pixel = image[row*width+col];
if (pixel[c] / hsat[c] == 1 && pixel[kc] > 24000) {
sum += pixel[c];
wgt += pixel[kc];
count++;
}
}
if (count == SCALE*SCALE)
map[mrow*wide+mcol] = sum / wgt;
}
for (spread = 32/grow; spread--; ) {
for (mrow=0; mrow < high; mrow++)
for (mcol=0; mcol < wide; mcol++) {
if (map[mrow*wide+mcol]) continue;
sum = count = 0;
for (d=0; d < 8; d++) {
y = mrow + dir[d][0];
x = mcol + dir[d][1];
if (y < high && x < wide && map[y*wide+x] > 0) {
sum += (1 + (d & 1)) * map[y*wide+x];
count += 1 + (d & 1);
}
}
if (count > 3)
map[mrow*wide+mcol] = - (sum+grow) / (count+grow);
}
for (change=i=0; i < high*wide; i++)
if (map[i] < 0) {
map[i] = -map[i];
change = 1;
}
if (!change) break;
}
for (i=0; i < high*wide; i++)
if (map[i] == 0) map[i] = 1;
for (mrow=0; mrow < high; mrow++)
for (mcol=0; mcol < wide; mcol++) {
for (row = mrow*SCALE; row < (mrow+1)*SCALE; row++)
for (col = mcol*SCALE; col < (mcol+1)*SCALE; col++) {
pixel = image[row*width+col];
if (pixel[c] / hsat[c] > 1) {
val = pixel[kc] * map[mrow*wide+mcol];
if (pixel[c] < val) pixel[c] = CLIP(val);
}
}
}
}
free (map);
}
#undef SCALE
void CLASS tiff_get (unsigned base,
unsigned *tag, unsigned *type, unsigned *len, unsigned *save)
{
*tag = get2();
*type = get2();
*len = get4();
*save = ftell(ifp) + 4;
if (*len * ("11124811248484"[*type < 14 ? *type:0]-'0') > 4)
fseek (ifp, get4()+base, SEEK_SET);
}
void CLASS parse_thumb_note (int base, unsigned toff, unsigned tlen)
{
unsigned entries, tag, type, len, save;
entries = get2();
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
if (tag == toff) thumb_offset = get4()+base;
if (tag == tlen) thumb_length = get4();
fseek (ifp, save, SEEK_SET);
}
}
//@end COMMON
int CLASS parse_tiff_ifd (int base);
//@out COMMON
static float powf_lim(float a, float b, float limup)
{
return (b>limup || b < -limup)?0.f:powf(a,b);
}
static float powf64(float a, float b)
{
return powf_lim(a,b,64.f);
}
#ifdef LIBRAW_LIBRARY_BUILD
static float my_roundf(float x) {
float t;
if (x >= 0.0) {
t = ceilf(x);
if (t - x > 0.5) t -= 1.0;
return t;
} else {
t = ceilf(-x);
if (t + x > 0.5) t -= 1.0;
return -t;
}
}
static float _CanonConvertAperture(ushort in)
{
if ((in == (ushort)0xffe0) || (in == (ushort)0x7fff)) return 0.0f;
return powf64(2.0, in/64.0);
}
void CLASS setCanonBodyFeatures (unsigned id)
{
imgdata.lens.makernotes.CamID = id;
if (
(id == 0x80000001) || // 1D
(id == 0x80000174) || // 1D2
(id == 0x80000232) || // 1D2N
(id == 0x80000169) || // 1D3
(id == 0x80000281) // 1D4
)
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSH;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Canon_EF;
}
else
if (
(id == 0x80000167) || // 1Ds
(id == 0x80000188) || // 1Ds2
(id == 0x80000215) || // 1Ds3
(id == 0x80000213) || // 5D
(id == 0x80000218) || // 5D2
(id == 0x80000285) || // 5D3
(id == 0x80000302) || // 6D
(id == 0x80000269) || // 1DX
(id == 0x80000324) || // 1DC
(id == 0x80000382) || // 5DS
(id == 0x80000401) // 5DS R
)
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_FF;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Canon_EF;
}
else
if (
(id == 0x80000331) || // M
(id == 0x80000355) || // M2
(id == 0x80000374) || // M3
(id == 0x80000384) // M10
)
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Canon_EF_M;
}
else
if (
(id == 0x01140000) || // D30
(id == 0x01668000) || // D60
(id > 0x80000000)
)
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Canon_EF;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Unknown;
}
else
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
}
return;
}
void CLASS processCanonCameraInfo (unsigned id, uchar *CameraInfo, unsigned maxlen)
{
ushort iCanonLensID = 0, iCanonMaxFocal = 0, iCanonMinFocal = 0, iCanonLens = 0, iCanonCurFocal = 0, iCanonFocalType = 0;
CameraInfo[0] = 0;
CameraInfo[1] = 0;
switch (id) {
case 0x80000001: // 1D
case 0x80000167: // 1DS
iCanonCurFocal = 10;
iCanonLensID = 13;
iCanonMinFocal = 14;
iCanonMaxFocal = 16;
if (!imgdata.lens.makernotes.CurFocal)
imgdata.lens.makernotes.CurFocal = sget2(CameraInfo + iCanonCurFocal);
if (!imgdata.lens.makernotes.MinFocal)
imgdata.lens.makernotes.MinFocal = sget2(CameraInfo + iCanonMinFocal);
if (!imgdata.lens.makernotes.MaxFocal)
imgdata.lens.makernotes.MaxFocal = sget2(CameraInfo + iCanonMaxFocal);
break;
case 0x80000174: // 1DMkII
case 0x80000188: // 1DsMkII
iCanonCurFocal = 9;
iCanonLensID = 12;
iCanonMinFocal = 17;
iCanonMaxFocal = 19;
iCanonFocalType = 45;
break;
case 0x80000232: // 1DMkII N
iCanonCurFocal = 9;
iCanonLensID = 12;
iCanonMinFocal = 17;
iCanonMaxFocal = 19;
break;
case 0x80000169: // 1DMkIII
case 0x80000215: // 1DsMkIII
iCanonCurFocal = 29;
iCanonLensID = 273;
iCanonMinFocal = 275;
iCanonMaxFocal = 277;
break;
case 0x80000281: // 1DMkIV
iCanonCurFocal = 30;
iCanonLensID = 335;
iCanonMinFocal = 337;
iCanonMaxFocal = 339;
break;
case 0x80000269: // 1D X
iCanonCurFocal = 35;
iCanonLensID = 423;
iCanonMinFocal = 425;
iCanonMaxFocal = 427;
break;
case 0x80000213: // 5D
iCanonCurFocal = 40;
if (!sget2Rev(CameraInfo + 12)) iCanonLensID = 151;
else iCanonLensID = 12;
iCanonMinFocal = 147;
iCanonMaxFocal = 149;
break;
case 0x80000218: // 5DMkII
iCanonCurFocal = 30;
iCanonLensID = 230;
iCanonMinFocal = 232;
iCanonMaxFocal = 234;
break;
case 0x80000285: // 5DMkIII
iCanonCurFocal = 35;
iCanonLensID = 339;
iCanonMinFocal = 341;
iCanonMaxFocal = 343;
break;
case 0x80000302: // 6D
iCanonCurFocal = 35;
iCanonLensID = 353;
iCanonMinFocal = 355;
iCanonMaxFocal = 357;
break;
case 0x80000250: // 7D
iCanonCurFocal = 30;
iCanonLensID = 274;
iCanonMinFocal = 276;
iCanonMaxFocal = 278;
break;
case 0x80000190: // 40D
iCanonCurFocal = 29;
iCanonLensID = 214;
iCanonMinFocal = 216;
iCanonMaxFocal = 218;
iCanonLens = 2347;
break;
case 0x80000261: // 50D
iCanonCurFocal = 30;
iCanonLensID = 234;
iCanonMinFocal = 236;
iCanonMaxFocal = 238;
break;
case 0x80000287: // 60D
iCanonCurFocal = 30;
iCanonLensID = 232;
iCanonMinFocal = 234;
iCanonMaxFocal = 236;
break;
case 0x80000325: // 70D
iCanonCurFocal = 35;
iCanonLensID = 358;
iCanonMinFocal = 360;
iCanonMaxFocal = 362;
break;
case 0x80000176: // 450D
iCanonCurFocal = 29;
iCanonLensID = 222;
iCanonLens = 2355;
break;
case 0x80000252: // 500D
iCanonCurFocal = 30;
iCanonLensID = 246;
iCanonMinFocal = 248;
iCanonMaxFocal = 250;
break;
case 0x80000270: // 550D
iCanonCurFocal = 30;
iCanonLensID = 255;
iCanonMinFocal = 257;
iCanonMaxFocal = 259;
break;
case 0x80000286: // 600D
case 0x80000288: // 1100D
iCanonCurFocal = 30;
iCanonLensID = 234;
iCanonMinFocal = 236;
iCanonMaxFocal = 238;
break;
case 0x80000301: // 650D
case 0x80000326: // 700D
iCanonCurFocal = 35;
iCanonLensID = 295;
iCanonMinFocal = 297;
iCanonMaxFocal = 299;
break;
case 0x80000254: // 1000D
iCanonCurFocal = 29;
iCanonLensID = 226;
iCanonMinFocal = 228;
iCanonMaxFocal = 230;
iCanonLens = 2359;
break;
}
if (iCanonFocalType)
{
if(iCanonFocalType>=maxlen) return; // broken;
imgdata.lens.makernotes.FocalType = CameraInfo[iCanonFocalType];
if (!imgdata.lens.makernotes.FocalType) // zero means 'fixed' here, replacing with standard '1'
imgdata.lens.makernotes.FocalType = 1;
}
if (!imgdata.lens.makernotes.CurFocal)
{
if(iCanonCurFocal>=maxlen) return; // broken;
imgdata.lens.makernotes.CurFocal = sget2Rev(CameraInfo + iCanonCurFocal);
}
if (!imgdata.lens.makernotes.LensID)
{
if(iCanonLensID>=maxlen) return; // broken;
imgdata.lens.makernotes.LensID = sget2Rev(CameraInfo + iCanonLensID);
}
if (!imgdata.lens.makernotes.MinFocal)
{
if(iCanonMinFocal>=maxlen) return; // broken;
imgdata.lens.makernotes.MinFocal = sget2Rev(CameraInfo + iCanonMinFocal);
}
if (!imgdata.lens.makernotes.MaxFocal)
{
if(iCanonMaxFocal>=maxlen) return; // broken;
imgdata.lens.makernotes.MaxFocal = sget2Rev(CameraInfo + iCanonMaxFocal);
}
if (!imgdata.lens.makernotes.Lens[0] && iCanonLens) {
if(iCanonLens+64>=maxlen) return; // broken;
if (CameraInfo[iCanonLens] < 65) // non-Canon lens
{
memcpy(imgdata.lens.makernotes.Lens, CameraInfo + iCanonLens, 64);
}
else if (!strncmp((char *)CameraInfo + iCanonLens, "EF-S", 4))
{
memcpy(imgdata.lens.makernotes.Lens, "EF-S ", 5);
memcpy(imgdata.lens.makernotes.LensFeatures_pre, "EF-E", 4);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_S;
memcpy(imgdata.lens.makernotes.Lens + 5, CameraInfo + iCanonLens + 4, 60);
}
else if (!strncmp((char *)CameraInfo + iCanonLens, "TS-E", 4)) {
memcpy(imgdata.lens.makernotes.Lens, "TS-E ", 5);
memcpy(imgdata.lens.makernotes.LensFeatures_pre, "TS-E", 4);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
memcpy(imgdata.lens.makernotes.Lens + 5, CameraInfo + iCanonLens + 4, 60);
}
else if (!strncmp((char *)CameraInfo + iCanonLens, "MP-E", 4)) {
memcpy(imgdata.lens.makernotes.Lens, "MP-E ", 5);
memcpy(imgdata.lens.makernotes.LensFeatures_pre, "MP-E", 4);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
memcpy(imgdata.lens.makernotes.Lens + 5, CameraInfo + iCanonLens + 4, 60);
}
else if (!strncmp((char *)CameraInfo + iCanonLens, "EF-M", 4)) {
memcpy(imgdata.lens.makernotes.Lens, "EF-M ", 5);
memcpy(imgdata.lens.makernotes.LensFeatures_pre, "EF-M", 4);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_M;
memcpy(imgdata.lens.makernotes.Lens + 5, CameraInfo + iCanonLens + 4, 60);
}
else {
memcpy(imgdata.lens.makernotes.Lens, CameraInfo + iCanonLens, 2);
memcpy(imgdata.lens.makernotes.LensFeatures_pre, "EF", 2);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
imgdata.lens.makernotes.Lens[2] = 32;
memcpy(imgdata.lens.makernotes.Lens + 3, CameraInfo + iCanonLens + 2, 62);
}
}
return;
}
void CLASS Canon_WBpresets (int skip1, int skip2)
{
int c;
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ (c >> 1)] = get2();
if (skip1) fseek(ifp, skip1, SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c ^ (c >> 1)] = get2();
if (skip1) fseek(ifp, skip1, SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ (c >> 1)] = get2();
if (skip1) fseek(ifp, skip1, SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ (c >> 1)] = get2();
if (skip1) fseek(ifp, skip1, SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][c ^ (c >> 1)] = get2();
if (skip2) fseek(ifp, skip2, SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c ^ (c >> 1)] = get2();
return;
}
void CLASS Canon_WBCTpresets (short WBCTversion)
{
if (WBCTversion == 0)
for (int i=0; i<15; i++)// tint, as shot R, as shot B, CСT
{
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 1.0f;
fseek (ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][1] = 1024.0f / (float)get2();
imgdata.color.WBCT_Coeffs[i][3] = 1024.0f / (float)get2();
imgdata.color.WBCT_Coeffs[i][0] = get2();
}
else if (WBCTversion == 1)
for (int i=0; i<15; i++) // as shot R, as shot B, tint, CСT
{
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 1.0f;
imgdata.color.WBCT_Coeffs[i][1] = 1024.0f / (float)get2();
imgdata.color.WBCT_Coeffs[i][3] = 1024.0f / (float)get2();
fseek (ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][0] = get2();
}
else if ((WBCTversion == 2) &&
((unique_id == 0x80000374) || // M3
(unique_id == 0x80000384))) // M10
for (int i=0; i<15; i++) // tint, offset, as shot R, as shot B, CСT
{
fseek (ifp, 2, SEEK_CUR);
fseek (ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 1.0f;
imgdata.color.WBCT_Coeffs[i][1] = 1024.0f / (float)get2();
imgdata.color.WBCT_Coeffs[i][3] = 1024.0f / (float)get2();
imgdata.color.WBCT_Coeffs[i][0] = get2();
}
else if ((WBCTversion == 2) &&
((unique_id == 0x03950000) || (unique_id == 0x03930000))) // G5 X
for (int i=0; i<15; i++) // tint, offset, as shot R, as shot B, CСT
{
fseek (ifp, 2, SEEK_CUR);
fseek (ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 1.0f;
imgdata.color.WBCT_Coeffs[i][1] = (float)get2() / 512.0f;
imgdata.color.WBCT_Coeffs[i][3] = (float)get2() / 512.0f;
imgdata.color.WBCT_Coeffs[i][0] = get2();
}
return;
}
void CLASS processNikonLensData (uchar *LensData, unsigned len)
{
ushort i;
if (!(imgdata.lens.nikon.NikonLensType & 0x01))
{
imgdata.lens.makernotes.LensFeatures_pre[0] = 'A';
imgdata.lens.makernotes.LensFeatures_pre[1] = 'F';
}
else
{
imgdata.lens.makernotes.LensFeatures_pre[0] = 'M';
imgdata.lens.makernotes.LensFeatures_pre[1] = 'F';
}
if (imgdata.lens.nikon.NikonLensType & 0x02)
{
if (imgdata.lens.nikon.NikonLensType & 0x04)
imgdata.lens.makernotes.LensFeatures_suf[0] = 'G';
else
imgdata.lens.makernotes.LensFeatures_suf[0] = 'D';
imgdata.lens.makernotes.LensFeatures_suf[1] = ' ';
}
if (imgdata.lens.nikon.NikonLensType & 0x08)
{
imgdata.lens.makernotes.LensFeatures_suf[2] = 'V';
imgdata.lens.makernotes.LensFeatures_suf[3] = 'R';
}
if (imgdata.lens.nikon.NikonLensType & 0x10)
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Nikon_CX;
else
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Nikon_F;
if (imgdata.lens.nikon.NikonLensType & 0x20)
{
strcpy(imgdata.lens.makernotes.Adapter, "FT-1");
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Nikon_F;
}
imgdata.lens.nikon.NikonLensType = imgdata.lens.nikon.NikonLensType & 0xdf;
if (len < 20) {
switch (len) {
case 9:
i = 2;
break;
case 15:
i = 7;
break;
case 16:
i = 8;
break;
}
imgdata.lens.nikon.NikonLensIDNumber = LensData[i];
imgdata.lens.nikon.NikonLensFStops = LensData[i + 1];
imgdata.lens.makernotes.LensFStops = (float)imgdata.lens.nikon.NikonLensFStops /12.0f;
if (fabsf(imgdata.lens.makernotes.MinFocal) < 1.1f)
{
if ((imgdata.lens.nikon.NikonLensType ^ (uchar)0x01) || LensData[i + 2])
imgdata.lens.makernotes.MinFocal = 5.0f * powf64(2.0f, (float)LensData[i + 2] / 24.0f);
if ((imgdata.lens.nikon.NikonLensType ^ (uchar)0x01) || LensData[i + 3])
imgdata.lens.makernotes.MaxFocal = 5.0f * powf64(2.0f, (float)LensData[i + 3] / 24.0f);
if ((imgdata.lens.nikon.NikonLensType ^ (uchar)0x01) || LensData[i + 4])
imgdata.lens.makernotes.MaxAp4MinFocal = powf64(2.0f, (float)LensData[i + 4] / 24.0f);
if ((imgdata.lens.nikon.NikonLensType ^ (uchar)0x01) || LensData[i + 5])
imgdata.lens.makernotes.MaxAp4MaxFocal = powf64(2.0f, (float)LensData[i + 5] / 24.0f);
}
imgdata.lens.nikon.NikonMCUVersion = LensData[i + 6];
if (i != 2)
{
if ((LensData[i - 1]) &&
(fabsf(imgdata.lens.makernotes.CurFocal) < 1.1f))
imgdata.lens.makernotes.CurFocal = 5.0f * powf64(2.0f, (float)LensData[i - 1] / 24.0f);
if (LensData[i + 7]) imgdata.lens.nikon.NikonEffectiveMaxAp = powf64(2.0f, (float)LensData[i + 7] / 24.0f);
}
imgdata.lens.makernotes.LensID =
(unsigned long long) LensData[i] << 56 |
(unsigned long long) LensData[i + 1] << 48 |
(unsigned long long) LensData[i + 2] << 40 |
(unsigned long long) LensData[i + 3] << 32 |
(unsigned long long) LensData[i + 4] << 24 |
(unsigned long long) LensData[i + 5] << 16 |
(unsigned long long) LensData[i + 6] << 8 |
(unsigned long long) imgdata.lens.nikon.NikonLensType;
}
else if ((len == 459) || (len == 590))
{
memcpy(imgdata.lens.makernotes.Lens, LensData + 390, 64);
}
else if (len == 509)
{
memcpy(imgdata.lens.makernotes.Lens, LensData + 391, 64);
}
else if (len == 879)
{
memcpy(imgdata.lens.makernotes.Lens, LensData + 680, 64);
}
return;
}
void CLASS setOlympusBodyFeatures (unsigned long long id)
{
imgdata.lens.makernotes.CamID = id;
if ((id == 0x4434303430ULL) || // E-1
(id == 0x4434303431ULL) || // E-300
((id & 0x00ffff0000ULL) == 0x0030300000ULL))
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_FT;
if ((id == 0x4434303430ULL) || // E-1
(id == 0x4434303431ULL) || // E-330
((id >= 0x5330303033ULL) && (id <= 0x5330303138ULL)) || // E-330 to E-520
(id == 0x5330303233ULL) || // E-620
(id == 0x5330303239ULL) || // E-450
(id == 0x5330303330ULL) || // E-600
(id == 0x5330303333ULL)) // E-5
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FT;
}
else
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_mFT;
}
}
else
{
imgdata.lens.makernotes.LensMount =
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
return;
}
void CLASS setPentaxBodyFeatures (unsigned id)
{
imgdata.lens.makernotes.CamID = id;
switch (id) {
case 0x12994:
case 0x12aa2:
case 0x12b1a:
case 0x12b60:
case 0x12b62:
case 0x12b7e:
case 0x12b80:
case 0x12b9c:
case 0x12b9d:
case 0x12ba2:
case 0x12c1e:
case 0x12c20:
case 0x12cd2:
case 0x12cd4:
case 0x12cfa:
case 0x12d72:
case 0x12d73:
case 0x12db8:
case 0x12dfe:
case 0x12e6c:
case 0x12e76:
case 0x12ef8:
case 0x12f52:
case 0x12f70:
case 0x12f71:
case 0x12fb6:
case 0x12fc0:
case 0x12fca:
case 0x1301a:
case 0x13024:
case 0x1309c:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Pentax_K;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Pentax_K;
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
break;
case 0x12e08:
case 0x13010:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Pentax_645;
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_MF;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Pentax_645;
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_MF;
break;
case 0x12ee4:
case 0x12f66:
case 0x12f7a:
case 0x1302e:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Pentax_Q;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Pentax_Q;
break;
default:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
return;
}
void CLASS PentaxLensInfo (unsigned id, unsigned len) // tag 0x0207
{
ushort iLensData = 0;
uchar *table_buf;
table_buf = (uchar*)malloc(MAX(len,128));
fread(table_buf, len, 1, ifp);
if ((id < 0x12b9c) ||
(((id == 0x12b9c) || // K100D
(id == 0x12b9d) || // K110D
(id == 0x12ba2)) && // K100D Super
((!table_buf[20] ||
(table_buf[20] == 0xff)))))
{
iLensData = 3;
if (imgdata.lens.makernotes.LensID == -1)
imgdata.lens.makernotes.LensID =
(((unsigned)table_buf[0]) << 8) + table_buf[1];
}
else switch (len)
{
case 90: // LensInfo3
iLensData = 13;
if (imgdata.lens.makernotes.LensID == -1)
imgdata.lens.makernotes.LensID =
((unsigned)((table_buf[1] & 0x0f) + table_buf[3]) <<8) + table_buf[4];
break;
case 91: // LensInfo4
iLensData = 12;
if (imgdata.lens.makernotes.LensID == -1)
imgdata.lens.makernotes.LensID =
((unsigned)((table_buf[1] & 0x0f) + table_buf[3]) <<8) + table_buf[4];
break;
case 80: // LensInfo5
case 128:
iLensData = 15;
if (imgdata.lens.makernotes.LensID == -1)
imgdata.lens.makernotes.LensID =
((unsigned)((table_buf[1] & 0x0f) + table_buf[4]) <<8) + table_buf[5];
break;
default:
if (id >= 0x12b9c) // LensInfo2
{
iLensData = 4;
if (imgdata.lens.makernotes.LensID == -1)
imgdata.lens.makernotes.LensID =
((unsigned)((table_buf[0] & 0x0f) + table_buf[2]) <<8) + table_buf[3];
}
}
if (iLensData)
{
if (table_buf[iLensData+9] &&
(fabs(imgdata.lens.makernotes.CurFocal) < 0.1f))
imgdata.lens.makernotes.CurFocal =
10*(table_buf[iLensData+9]>>2) * powf64(4, (table_buf[iLensData+9] & 0x03)-2);
if (table_buf[iLensData+10] & 0xf0)
imgdata.lens.makernotes.MaxAp4CurFocal =
powf64(2.0f, (float)((table_buf[iLensData+10] & 0xf0) >>4)/4.0f);
if (table_buf[iLensData+10] & 0x0f)
imgdata.lens.makernotes.MinAp4CurFocal =
powf64(2.0f, (float)((table_buf[iLensData+10] & 0x0f) + 10)/4.0f);
if (iLensData != 12)
{
switch (table_buf[iLensData] & 0x06)
{
case 0: imgdata.lens.makernotes.MinAp4MinFocal = 22.0f; break;
case 2: imgdata.lens.makernotes.MinAp4MinFocal = 32.0f; break;
case 4: imgdata.lens.makernotes.MinAp4MinFocal = 45.0f; break;
case 6: imgdata.lens.makernotes.MinAp4MinFocal = 16.0f; break;
}
if (table_buf[iLensData] & 0x70)
imgdata.lens.makernotes.LensFStops =
((float)(((table_buf[iLensData] & 0x70) >> 4) ^ 0x07)) / 2.0f + 5.0f;
imgdata.lens.makernotes.MinFocusDistance = (float)(table_buf[iLensData+3] & 0xf8);
imgdata.lens.makernotes.FocusRangeIndex = (float)(table_buf[iLensData+3] & 0x07);
if ((table_buf[iLensData+14] > 1) &&
(fabs(imgdata.lens.makernotes.MaxAp4CurFocal) < 0.7f))
imgdata.lens.makernotes.MaxAp4CurFocal =
powf64(2.0f, (float)((table_buf[iLensData+14] & 0x7f) -1)/32.0f);
}
else if ((id != 0x12e76) && // K-5
(table_buf[iLensData+15] > 1) &&
(fabs(imgdata.lens.makernotes.MaxAp4CurFocal) < 0.7f))
{
imgdata.lens.makernotes.MaxAp4CurFocal =
powf64(2.0f, (float)((table_buf[iLensData+15] & 0x7f) -1)/32.0f);
}
}
free(table_buf);
return;
}
void CLASS setPhaseOneFeatures (unsigned id) {
ushort i;
static const struct {
ushort id;
char t_model[32];
} p1_unique[] = {
// Phase One section:
{1, "Hasselblad V"},
{10, "PhaseOne/Mamiya"},
{12, "Contax 645"},
{16, "Hasselblad V"},
{17, "Hasselblad V"},
{18, "Contax 645"},
{19, "PhaseOne/Mamiya"},
{20, "Hasselblad V"},
{21, "Contax 645"},
{22, "PhaseOne/Mamiya"},
{23, "Hasselblad V"},
{24, "Hasselblad H"},
{25, "PhaseOne/Mamiya"},
{32, "Contax 645"},
{34, "Hasselblad V"},
{35, "Hasselblad V"},
{36, "Hasselblad H"},
{37, "Contax 645"},
{38, "PhaseOne/Mamiya"},
{39, "Hasselblad V"},
{40, "Hasselblad H"},
{41, "Contax 645"},
{42, "PhaseOne/Mamiya"},
{44, "Hasselblad V"},
{45, "Hasselblad H"},
{46, "Contax 645"},
{47, "PhaseOne/Mamiya"},
{48, "Hasselblad V"},
{49, "Hasselblad H"},
{50, "Contax 645"},
{51, "PhaseOne/Mamiya"},
{52, "Hasselblad V"},
{53, "Hasselblad H"},
{54, "Contax 645"},
{55, "PhaseOne/Mamiya"},
{67, "Hasselblad V"},
{68, "Hasselblad H"},
{69, "Contax 645"},
{70, "PhaseOne/Mamiya"},
{71, "Hasselblad V"},
{72, "Hasselblad H"},
{73, "Contax 645"},
{74, "PhaseOne/Mamiya"},
{76, "Hasselblad V"},
{77, "Hasselblad H"},
{78, "Contax 645"},
{79, "PhaseOne/Mamiya"},
{80, "Hasselblad V"},
{81, "Hasselblad H"},
{82, "Contax 645"},
{83, "PhaseOne/Mamiya"},
{84, "Hasselblad V"},
{85, "Hasselblad H"},
{86, "Contax 645"},
{87, "PhaseOne/Mamiya"},
{99, "Hasselblad V"},
{100, "Hasselblad H"},
{101, "Contax 645"},
{102, "PhaseOne/Mamiya"},
{103, "Hasselblad V"},
{104, "Hasselblad H"},
{105, "PhaseOne/Mamiya"},
{106, "Contax 645"},
{112, "Hasselblad V"},
{113, "Hasselblad H"},
{114, "Contax 645"},
{115, "PhaseOne/Mamiya"},
{131, "Hasselblad V"},
{132, "Hasselblad H"},
{133, "Contax 645"},
{134, "PhaseOne/Mamiya"},
{135, "Hasselblad V"},
{136, "Hasselblad H"},
{137, "Contax 645"},
{138, "PhaseOne/Mamiya"},
{140, "Hasselblad V"},
{141, "Hasselblad H"},
{142, "Contax 645"},
{143, "PhaseOne/Mamiya"},
{148, "Hasselblad V"},
{149, "Hasselblad H"},
{150, "Contax 645"},
{151, "PhaseOne/Mamiya"},
{160, "A-250"},
{161, "A-260"},
{162, "A-280"},
{167, "Hasselblad V"},
{168, "Hasselblad H"},
{169, "Contax 645"},
{170, "PhaseOne/Mamiya"},
{172, "Hasselblad V"},
{173, "Hasselblad H"},
{174, "Contax 645"},
{175, "PhaseOne/Mamiya"},
{176, "Hasselblad V"},
{177, "Hasselblad H"},
{178, "Contax 645"},
{179, "PhaseOne/Mamiya"},
{180, "Hasselblad V"},
{181, "Hasselblad H"},
{182, "Contax 645"},
{183, "PhaseOne/Mamiya"},
{208, "Hasselblad V"},
{211, "PhaseOne/Mamiya"},
{448, "Phase One 645AF"},
{457, "Phase One 645DF"},
{471, "Phase One 645DF+"},
{704, "Phase One iXA"},
{705, "Phase One iXA - R"},
{706, "Phase One iXU 150"},
{707, "Phase One iXU 150 - NIR"},
{708, "Phase One iXU 180"},
{721, "Phase One iXR"},
// Leaf section:
{333,"Mamiya"},
{329,"Universal"},
{330,"Hasselblad H1/H2"},
{332,"Contax"},
{336,"AFi"},
{327,"Mamiya"},
{324,"Universal"},
{325,"Hasselblad H1/H2"},
{326,"Contax"},
{335,"AFi"},
{340,"Mamiya"},
{337,"Universal"},
{338,"Hasselblad H1/H2"},
{339,"Contax"},
{323,"Mamiya"},
{320,"Universal"},
{322,"Hasselblad H1/H2"},
{321,"Contax"},
{334,"AFi"},
{369,"Universal"},
{370,"Mamiya"},
{371,"Hasselblad H1/H2"},
{372,"Contax"},
{373,"Afi"},
};
imgdata.lens.makernotes.CamID = id;
if (id && !imgdata.lens.makernotes.body[0]) {
for (i=0; i < sizeof p1_unique / sizeof *p1_unique; i++)
if (id == p1_unique[i].id) {
strcpy(imgdata.lens.makernotes.body,p1_unique[i].t_model);
}
}
return;
}
void CLASS setSonyBodyFeatures (unsigned id) {
imgdata.lens.makernotes.CamID = id;
if ( // FF cameras
(id == 257) || // a900
(id == 269) || // a850
(id == 340) || // ILCE-7M2
(id == 318) || // ILCE-7S
(id == 350) || // ILCE-7SM2
(id == 311) || // ILCE-7R
(id == 347) || // ILCE-7RM2
(id == 306) || // ILCE-7
(id == 298) || // DSC-RX1
(id == 299) || // NEX-VG900
(id == 310) || // DSC-RX1R
(id == 344) || // DSC-RX1RM2
(id == 294) // SLT-99, Hasselblad HV
)
{
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_FF;
}
else
{
if ((id != 002) && // DSC-R1
(id != 297) && // DSC-RX100
(id != 308) && // DSC-RX100M2
(id != 309) && // DSC-RX10
(id != 317) && // DSC-RX100M3
(id != 341) && // DSC-RX100M4
(id != 342) // DSC-RX10M2
)
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
}
if ( // E-mount cameras
// ILCE:
(id == 302) ||
(id == 306) ||
(id == 311) ||
(id == 312) ||
(id == 313) ||
(id == 318) ||
(id == 339) ||
(id == 340) ||
(id == 346) ||
(id == 347) ||
(id == 350) ||
// NEX:
(id == 278) ||
(id == 279) ||
(id == 284) ||
(id == 288) ||
(id == 289) ||
(id == 290) ||
(id == 293) ||
(id == 295) ||
(id == 296) ||
(id == 299) ||
(id == 300) ||
(id == 305) ||
(id == 307)
)
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Sony_E;
}
else if ( // A-mount cameras
// DSLR:
(id == 256) ||
(id == 257) ||
(id == 258) ||
(id == 259) ||
(id == 260) ||
(id == 261) ||
(id == 262) ||
(id == 263) ||
(id == 264) ||
(id == 265) ||
(id == 266) ||
(id == 269) ||
(id == 270) ||
(id == 273) ||
(id == 274) ||
(id == 275) ||
(id == 282) ||
(id == 283) ||
// SLT:
(id == 280) ||
(id == 281) ||
(id == 285) ||
(id == 286) ||
(id == 287) ||
(id == 291) ||
(id == 292) ||
(id == 294) ||
(id == 303) ||
// ILCA:
(id == 319)
)
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Minolta_A;
}
else if ( // DSC
(id == 002) || // DSC-R1
(id == 297) || // DSC-RX100
(id == 298) || // DSC-RX1
(id == 308) || // DSC-RX100M2
(id == 309) || // DSC-RX10
(id == 310) || // DSC-RX1R
(id == 344) || // DSC-RX1RM2
(id == 317) || // DSC-RX100M3
(id == 341) || // DSC-RX100M4
(id == 342) // DSC-RX10M2
)
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
}
return;
}
void CLASS parseSonyLensType2 (uchar a, uchar b) {
ushort lid2;
lid2 = (((ushort)a)<<8) | ((ushort)b);
if (!lid2) return;
if (lid2 < 0x100)
{
imgdata.lens.makernotes.AdapterID = lid2;
switch (lid2) {
case 1:
case 2:
case 3:
case 6:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Minolta_A;
break;
case 44:
case 78:
case 239:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
break;
}
}
else
imgdata.lens.makernotes.LensID = lid2;
return;
}
void CLASS parseSonyLensFeatures (uchar a, uchar b) {
ushort features;
features = (((ushort)a)<<8) | ((ushort)b);
if ((imgdata.lens.makernotes.LensMount == LIBRAW_MOUNT_Canon_EF) || !features)
return;
imgdata.lens.makernotes.LensFeatures_pre[0] = 0;
imgdata.lens.makernotes.LensFeatures_suf[0] = 0;
if ((features & 0x0200) && (features & 0x0100)) strcpy(imgdata.lens.makernotes.LensFeatures_pre, "E");
else if (features & 0x0200) strcpy(imgdata.lens.makernotes.LensFeatures_pre, "FE");
else if (features & 0x0100) strcpy(imgdata.lens.makernotes.LensFeatures_pre, "DT");
if (!imgdata.lens.makernotes.LensFormat && !imgdata.lens.makernotes.LensMount)
{
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_FF;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Minolta_A;
if ((features & 0x0200) && (features & 0x0100)) {
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_APSC;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Sony_E;
} else if (features & 0x0200) {
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Sony_E;
} else if (features & 0x0100) {
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_APSC;
}
}
if (features & 0x4000)
strncat(imgdata.lens.makernotes.LensFeatures_pre, " PZ", sizeof(imgdata.lens.makernotes.LensFeatures_pre));
if (features & 0x0008)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " G", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
else if (features & 0x0004)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " ZA", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if ((features & 0x0020) && (features & 0x0040))
strncat(imgdata.lens.makernotes.LensFeatures_suf, " Macro", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
else if (features & 0x0020)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " STF", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
else if (features & 0x0040)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " Reflex", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
else if (features & 0x0080)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " Fisheye", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if (features & 0x0001)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " SSM", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
else if (features & 0x0002)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " SAM", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if (features & 0x8000)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " OSS", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if (features & 0x2000)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " LE", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if (features & 0x0800)
strncat(imgdata.lens.makernotes.LensFeatures_suf, " II", sizeof(imgdata.lens.makernotes.LensFeatures_suf));
if (imgdata.lens.makernotes.LensFeatures_suf[0] == ' ')
memmove(imgdata.lens.makernotes.LensFeatures_suf, imgdata.lens.makernotes.LensFeatures_suf+1, strlen(imgdata.lens.makernotes.LensFeatures_suf));
return;
}
void CLASS process_Sony_0x940c (uchar * buf)
{
ushort lid2;
if (imgdata.lens.makernotes.LensMount != LIBRAW_MOUNT_Canon_EF)
{
switch (SonySubstitution[buf[0x0008]]) {
case 1:
case 5:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Minolta_A;
break;
case 4:
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Sony_E;
break;
}
}
lid2 = (((ushort)SonySubstitution[buf[0x000a]])<<8) |
((ushort)SonySubstitution[buf[0x0009]]);
if ((lid2 > 0) && (lid2 < 32784))
parseSonyLensType2 (SonySubstitution[buf[0x000a]], // LensType2 - Sony lens ids
SonySubstitution[buf[0x0009]]);
return;
}
void CLASS process_Sony_0x9050 (uchar * buf, unsigned id)
{
ushort lid;
if ((imgdata.lens.makernotes.CameraMount != LIBRAW_MOUNT_Sony_E) &&
(imgdata.lens.makernotes.CameraMount != LIBRAW_MOUNT_FixedLens))
{
if (buf[0])
imgdata.lens.makernotes.MaxAp4CurFocal =
my_roundf(powf64(2.0f, ((float)SonySubstitution[buf[0]] / 8.0 - 1.06f) / 2.0f)*10.0f) / 10.0f;
if (buf[1])
imgdata.lens.makernotes.MinAp4CurFocal =
my_roundf(powf64(2.0f, ((float)SonySubstitution[buf[1]] / 8.0 - 1.06f) / 2.0f)*10.0f) / 10.0f;
}
if (imgdata.lens.makernotes.CameraMount != LIBRAW_MOUNT_FixedLens)
{
if (buf[0x3d] | buf[0x3c])
{
lid = SonySubstitution[buf[0x3d]] << 8 |
SonySubstitution[buf[0x3c]];
imgdata.lens.makernotes.CurAp =
powf64(2.0f, ((float)lid/256.0f - 16.0f) / 2.0f);
}
if (buf[0x105] && (imgdata.lens.makernotes.LensMount != LIBRAW_MOUNT_Canon_EF))
imgdata.lens.makernotes.LensMount =
SonySubstitution[buf[0x105]];
if (buf[0x106])
imgdata.lens.makernotes.LensFormat =
SonySubstitution[buf[0x106]];
}
if (imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Sony_E)
{
parseSonyLensType2 (SonySubstitution[buf[0x0108]], // LensType2 - Sony lens ids
SonySubstitution[buf[0x0107]]);
}
if ((imgdata.lens.makernotes.LensID == -1) &&
(imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Minolta_A) &&
(buf[0x010a] | buf[0x0109]))
{
imgdata.lens.makernotes.LensID = // LensType - Minolta/Sony lens ids
SonySubstitution[buf[0x010a]] << 8 |
SonySubstitution[buf[0x0109]];
if ((imgdata.lens.makernotes.LensID > 61184) &&
(imgdata.lens.makernotes.LensID < 65535))
{
imgdata.lens.makernotes.LensID -= 61184;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
}
}
if ((id >= 286) && (id <= 293))
// "SLT-A65", "SLT-A77", "NEX-7", "NEX-VG20E",
// "SLT-A37", "SLT-A57", "NEX-F3", "Lunar"
parseSonyLensFeatures (SonySubstitution[buf[0x115]],
SonySubstitution[buf[0x116]]);
else if (imgdata.lens.makernotes.CameraMount != LIBRAW_MOUNT_FixedLens)
parseSonyLensFeatures (SonySubstitution[buf[0x116]],
SonySubstitution[buf[0x117]]);
return;
}
void CLASS parse_makernote_0xc634(int base, int uptag, unsigned dng_writer)
{
unsigned ver97 = 0, offset = 0, entries, tag, type, len, save, c;
unsigned i;
uchar NikonKey, ci, cj, ck;
unsigned serial = 0;
unsigned NikonLensDataVersion = 0;
unsigned lenNikonLensData = 0;
uchar *CanonCameraInfo;
unsigned lenCanonCameraInfo = 0;
uchar *table_buf;
uchar *table_buf_0x9050;
ushort table_buf_0x9050_present = 0;
uchar *table_buf_0x940c;
ushort table_buf_0x940c_present = 0;
short morder, sorder = order;
char buf[10];
fread(buf, 1, 10, ifp);
if (!strcmp(buf, "Nikon")) {
base = ftell(ifp);
order = get2();
if (get2() != 42) goto quit;
offset = get4();
fseek(ifp, offset - 8, SEEK_CUR);
}
else if (!strcmp(buf, "OLYMPUS") ||
!strcmp(buf, "PENTAX ") ||
(!strncmp(make, "SAMSUNG", 7) && (dng_writer == CameraDNG))) {
base = ftell(ifp) - 10;
fseek(ifp, -2, SEEK_CUR);
order = get2();
if (buf[0] == 'O') get2();
}
else if (!strncmp(buf, "SONY", 4) ||
!strcmp(buf, "Panasonic")) {
goto nf;
}
else if (!strncmp(buf, "FUJIFILM", 8)) {
base = ftell(ifp) - 10;
nf: order = 0x4949;
fseek(ifp, 2, SEEK_CUR);
}
else if (!strcmp(buf, "OLYMP") ||
!strcmp(buf, "LEICA") ||
!strcmp(buf, "Ricoh") ||
!strcmp(buf, "EPSON"))
fseek(ifp, -2, SEEK_CUR);
else if (!strcmp(buf, "AOC") ||
!strcmp(buf, "QVC"))
fseek(ifp, -4, SEEK_CUR);
else {
fseek(ifp, -10, SEEK_CUR);
if ((!strncmp(make, "SAMSUNG", 7) &&
(dng_writer == AdobeDNG)))
base = ftell(ifp);
}
entries = get2();
if (entries > 1000) return;
morder = order;
while (entries--) {
order = morder;
tiff_get(base, &tag, &type, &len, &save);
tag |= uptag << 16;
if(len > 100*1024*1024) goto next; // 100Mb tag? No!
if (!strncmp(make, "Canon",5))
{
if (tag == 0x0001) // camera settings
{
fseek(ifp, 44, SEEK_CUR);
imgdata.lens.makernotes.LensID = get2();
imgdata.lens.makernotes.MaxFocal = get2();
imgdata.lens.makernotes.MinFocal = get2();
imgdata.lens.makernotes.CanonFocalUnits = get2();
if (imgdata.lens.makernotes.CanonFocalUnits != 1)
{
imgdata.lens.makernotes.MaxFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
imgdata.lens.makernotes.MinFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
imgdata.lens.makernotes.MaxAp = _CanonConvertAperture(get2());
imgdata.lens.makernotes.MinAp = _CanonConvertAperture(get2());
}
else if (tag == 0x0002) // focal length
{
imgdata.lens.makernotes.FocalType = get2();
imgdata.lens.makernotes.CurFocal = get2();
if ((imgdata.lens.makernotes.CanonFocalUnits != 1) &&
imgdata.lens.makernotes.CanonFocalUnits)
{
imgdata.lens.makernotes.CurFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
}
else if (tag == 0x0004) // shot info
{
short tempAp;
fseek(ifp, 8, SEEK_CUR);
if ((tempAp = get2()) != 0x7fff)
imgdata.lens.makernotes.CurAp = _CanonConvertAperture(tempAp);
if (imgdata.lens.makernotes.CurAp < 0.7f)
{
fseek(ifp, 32, SEEK_CUR);
imgdata.lens.makernotes.CurAp = _CanonConvertAperture(get2());
}
if (!aperture) aperture = imgdata.lens.makernotes.CurAp;
}
else if (tag == 0x000d) // camera info
{
CanonCameraInfo = (uchar*)malloc(len);
fread(CanonCameraInfo, len, 1, ifp);
lenCanonCameraInfo = len;
}
else if (tag == 0x10) // Canon ModelID
{
unique_id = get4();
if (unique_id == 0x03740000) unique_id = 0x80000374; // M3
if (unique_id == 0x03840000) unique_id = 0x80000384; // M10
setCanonBodyFeatures(unique_id);
if (lenCanonCameraInfo)
{
processCanonCameraInfo(unique_id, CanonCameraInfo,lenCanonCameraInfo);
free(CanonCameraInfo);
CanonCameraInfo = 0;
lenCanonCameraInfo = 0;
}
}
else if (tag == 0x0095 && // lens model tag
!imgdata.lens.makernotes.Lens[0])
{
fread(imgdata.lens.makernotes.Lens, 2, 1, ifp);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
if (imgdata.lens.makernotes.Lens[0] < 65) // non-Canon lens
fread(imgdata.lens.makernotes.Lens + 2, 62, 1, ifp);
else
{
char efs[2];
imgdata.lens.makernotes.LensFeatures_pre[0] = imgdata.lens.makernotes.Lens[0];
imgdata.lens.makernotes.LensFeatures_pre[1] = imgdata.lens.makernotes.Lens[1];
fread(efs, 2, 1, ifp);
if (efs[0] == 45 && (efs[1] == 83 || efs[1] == 69 || efs[1] == 77))
{ // "EF-S, TS-E, MP-E, EF-M" lenses
imgdata.lens.makernotes.Lens[2] = imgdata.lens.makernotes.LensFeatures_pre[2] = efs[0];
imgdata.lens.makernotes.Lens[3] = imgdata.lens.makernotes.LensFeatures_pre[3] = efs[1];
imgdata.lens.makernotes.Lens[4] = 32;
if (efs[1] == 83)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_S;
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_APSC;
}
else if (efs[1] == 77)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_M;
}
}
else
{ // "EF" lenses
imgdata.lens.makernotes.Lens[2] = 32;
imgdata.lens.makernotes.Lens[3] = efs[0];
imgdata.lens.makernotes.Lens[4] = efs[1];
}
fread(imgdata.lens.makernotes.Lens + 5, 58, 1, ifp);
}
}
}
else if (!strncmp(make, "FUJI", 4))
switch (tag) {
case 0x1404: imgdata.lens.makernotes.MinFocal = getreal(type); break;
case 0x1405: imgdata.lens.makernotes.MaxFocal = getreal(type); break;
case 0x1406: imgdata.lens.makernotes.MaxAp4MinFocal = getreal(type); break;
case 0x1407: imgdata.lens.makernotes.MaxAp4MaxFocal = getreal(type); break;
}
else if (!strncasecmp(make, "LEICA", 5))
{
if (((tag == 0x035e) || (tag == 0x035f)) && (type == 10) && (len == 9))
{
int ind = tag == 0x035e?0:1;
for (int j=0; j < 3; j++)
FORCC imgdata.color.dng_color[ind].forwardmatrix[c][j]= getreal(type);
}
if ((tag == 0x0303) && (type != 4))
{
fread(imgdata.lens.makernotes.Lens, MIN(len,127), 1, ifp);
}
if ((tag == 0x3405) ||
(tag == 0x0310) ||
(tag == 0x34003405))
{
imgdata.lens.makernotes.LensID = get4();
imgdata.lens.makernotes.LensID =
((imgdata.lens.makernotes.LensID>>2)<<8) |
(imgdata.lens.makernotes.LensID & 0x3);
if (imgdata.lens.makernotes.LensID != -1)
{
if ((model[0] == 'M') ||
!strncasecmp (model, "LEICA M", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_M;
if (imgdata.lens.makernotes.LensID)
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Leica_M;
}
else if ((model[0] == 'S') ||
!strncasecmp (model, "LEICA S", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_S;
if (imgdata.lens.makernotes.Lens[0])
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Leica_S;
}
}
}
else if (
((tag == 0x0313) || (tag == 0x34003406)) &&
(fabs(imgdata.lens.makernotes.CurAp) < 0.17f) &&
((type == 10) || (type == 5))
)
{
imgdata.lens.makernotes.CurAp = getreal(type);
if (imgdata.lens.makernotes.CurAp > 126.3)
imgdata.lens.makernotes.CurAp = 0.0f;
}
else if (tag == 0x3400)
{
parse_makernote (base, 0x3400);
}
}
else if (!strncmp(make, "NIKON", 5))
{
if (tag == 0x1d) // serial number
while ((c = fgetc(ifp)) && c != EOF)
serial = serial * 10 + (isdigit(c) ? c - '0' : c % 10);
else if (tag == 0x000a)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
else if (tag == 0x0082) // lens attachment
{
fread(imgdata.lens.makernotes.Attachment, MIN(len,127), 1, ifp);
}
else if (tag == 0x0083) // lens type
{
imgdata.lens.nikon.NikonLensType = fgetc(ifp);
}
else if (tag == 0x0084) // lens
{
imgdata.lens.makernotes.MinFocal = getreal(type);
imgdata.lens.makernotes.MaxFocal = getreal(type);
imgdata.lens.makernotes.MaxAp4MinFocal = getreal(type);
imgdata.lens.makernotes.MaxAp4MaxFocal = getreal(type);
}
else if (tag == 0x008b) // lens f-stops
{
uchar a, b, c;
a = fgetc(ifp);
b = fgetc(ifp);
c = fgetc(ifp);
if (c)
{
imgdata.lens.nikon.NikonLensFStops = a*b*(12/c);
imgdata.lens.makernotes.LensFStops =
(float)imgdata.lens.nikon.NikonLensFStops /12.0f;
}
}
else if (tag == 0x0093)
{
i = get2();
if ((i == 7) || (i == 9))
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
}
else if (tag == 0x0097)
{
for (i=0; i < 4; i++)
ver97 = ver97 * 10 + fgetc(ifp)-'0';
if (ver97 == 601) // Coolpix A
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
}
else if (tag == 0x0098) // contains lens data
{
for (i = 0; i < 4; i++)
{
NikonLensDataVersion = NikonLensDataVersion * 10 + fgetc(ifp) - '0';
}
switch (NikonLensDataVersion)
{
case 100: lenNikonLensData = 9; break;
case 101:
case 201: // encrypted, starting from v.201
case 202:
case 203: lenNikonLensData = 15; break;
case 204: lenNikonLensData = 16; break;
case 400: lenNikonLensData = 459; break;
case 401: lenNikonLensData = 590; break;
case 402: lenNikonLensData = 509; break;
case 403: lenNikonLensData = 879; break;
}
if(lenNikonLensData)
{
table_buf = (uchar*)malloc(lenNikonLensData);
fread(table_buf, lenNikonLensData, 1, ifp);
if ((NikonLensDataVersion < 201) && lenNikonLensData)
{
processNikonLensData(table_buf, lenNikonLensData);
free(table_buf);
lenNikonLensData = 0;
}
}
}
else if (tag == 0xa7) // shutter count
{
NikonKey = fgetc(ifp) ^ fgetc(ifp) ^ fgetc(ifp) ^ fgetc(ifp);
if ((NikonLensDataVersion > 200) && lenNikonLensData)
{
ci = xlat[0][serial & 0xff];
cj = xlat[1][NikonKey];
ck = 0x60;
for (i = 0; i < lenNikonLensData; i++)
table_buf[i] ^= (cj += ci * ck++);
processNikonLensData(table_buf, lenNikonLensData);
free(table_buf);
lenNikonLensData = 0;
}
}
else if (tag == 37 && (!iso_speed || iso_speed == 65535))
{
unsigned char cc;
fread(&cc, 1, 1, ifp);
iso_speed = (int)(100.0 * powf64(2.0, (double)(cc) / 12.0 - 5.0));
break;
}
}
else if (!strncmp(make, "OLYMPUS", 7))
{
if (tag == 0x2010)
{
fseek(ifp, save - 4, SEEK_SET);
fseek(ifp, base + get4(), SEEK_SET);
parse_makernote_0xc634(base, 0x2010, dng_writer);
}
switch (tag) {
case 0x0207:
case 0x20100100:
{
uchar sOlyID[8];
unsigned long long OlyID;
fread (sOlyID, MIN(len,7), 1, ifp);
sOlyID[7] = 0;
OlyID = sOlyID[0];
i = 1;
while (i < 7 && sOlyID[i])
{
OlyID = OlyID << 8 | sOlyID[i];
i++;
}
setOlympusBodyFeatures(OlyID);
}
break;
case 0x1002:
imgdata.lens.makernotes.CurAp = powf64(2.0f, getreal(type)/2);
break;
case 0x20100201:
imgdata.lens.makernotes.LensID =
(unsigned long long)fgetc(ifp)<<16 |
(unsigned long long)(fgetc(ifp), fgetc(ifp))<<8 |
(unsigned long long)fgetc(ifp);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FT;
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_FT;
if (((imgdata.lens.makernotes.LensID < 0x20000) ||
(imgdata.lens.makernotes.LensID > 0x4ffff)) &&
(imgdata.lens.makernotes.LensID & 0x10))
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_mFT;
}
break;
case 0x20100203:
fread(imgdata.lens.makernotes.Lens, MIN(len,127), 1, ifp);
break;
case 0x20100205:
imgdata.lens.makernotes.MaxAp4MinFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100206:
imgdata.lens.makernotes.MaxAp4MaxFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100207:
imgdata.lens.makernotes.MinFocal = (float)get2();
break;
case 0x20100208:
imgdata.lens.makernotes.MaxFocal = (float)get2();
if (imgdata.lens.makernotes.MaxFocal > 1000.0f)
imgdata.lens.makernotes.MaxFocal = imgdata.lens.makernotes.MinFocal;
break;
case 0x2010020a:
imgdata.lens.makernotes.MaxAp4CurFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100301:
imgdata.lens.makernotes.TeleconverterID = fgetc(ifp) << 8;
fgetc(ifp);
imgdata.lens.makernotes.TeleconverterID =
imgdata.lens.makernotes.TeleconverterID | fgetc(ifp);
break;
case 0x20100303:
fread(imgdata.lens.makernotes.Teleconverter, MIN(len,127), 1, ifp);
break;
case 0x20100403:
fread(imgdata.lens.makernotes.Attachment, MIN(len,127), 1, ifp);
break;
}
}
else if (!strncmp(make, "PENTAX", 6) ||
!strncmp(model, "PENTAX", 6) ||
(!strncmp(make, "SAMSUNG", 7) && (dng_writer == CameraDNG)))
{
if (tag == 0x0005)
{
unique_id = get4();
setPentaxBodyFeatures(unique_id);
}
else if (tag == 0x0013)
{
imgdata.lens.makernotes.CurAp = (float)get2()/10.0f;
}
else if (tag == 0x001d)
{
imgdata.lens.makernotes.CurFocal = (float)get4()/100.0f;
}
else if (tag == 0x003f)
{
imgdata.lens.makernotes.LensID = fgetc(ifp) << 8 | fgetc(ifp);
}
else if (tag == 0x0207)
{
PentaxLensInfo(imgdata.lens.makernotes.CamID, len);
}
else if (tag == 0x020d)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ (c >> 1)] = get2();
}
else if (tag == 0x020e)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c ^ (c >> 1)] = get2();
}
else if (tag == 0x020f)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0210)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0211)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0212)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0213)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0214)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0221)
{
int nWB = get2();
if(nWB<=sizeof(imgdata.color.WBCT_Coeffs)/sizeof(imgdata.color.WBCT_Coeffs[0]))
for (int i = 0; i < nWB; i++)
{
imgdata.color.WBCT_Coeffs[i][0] = (unsigned)0xcfc6 - get2();
fseek(ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][1] = get2();
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 0x2000;
imgdata.color.WBCT_Coeffs[i][3] = get2();
}
}
else if (tag == 0x022d)
{
fseek (ifp,2,SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0239) // Q-series lens info (LensInfoQ)
{
char LensInfo [20];
fseek (ifp, 12, SEEK_CUR);
fread(imgdata.lens.makernotes.Lens, 30, 1, ifp);
strcat(imgdata.lens.makernotes.Lens, " ");
fread(LensInfo, 20, 1, ifp);
strcat(imgdata.lens.makernotes.Lens, LensInfo);
}
}
else if (!strncmp(make, "SAMSUNG", 7) &&
(dng_writer == AdobeDNG))
{
if (tag == 0x0002)
{
if(get4() == 0x2000)
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Samsung_NX;
}
else if (!strncmp(model, "NX mini", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Samsung_NX_M;
}
else
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
}
}
else if (tag == 0x0003)
{
imgdata.lens.makernotes.CamID = unique_id = get4();
}
else if (tag == 0xa003)
{
imgdata.lens.makernotes.LensID = get2();
if (imgdata.lens.makernotes.LensID)
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Samsung_NX;
}
else if (tag == 0xa019)
{
imgdata.lens.makernotes.CurAp = getreal(type);
}
else if (tag == 0xa01a)
{
imgdata.lens.makernotes.FocalLengthIn35mmFormat = get4() / 10.0f;
if (imgdata.lens.makernotes.FocalLengthIn35mmFormat < 10.0f)
imgdata.lens.makernotes.FocalLengthIn35mmFormat *= 10.0f;
}
}
else if (!strncasecmp(make, "SONY", 4) ||
!strncasecmp(make, "Konica", 6) ||
!strncasecmp(make, "Minolta", 7) ||
(!strncasecmp(make, "Hasselblad", 10) &&
(!strncasecmp(model, "Stellar", 7) ||
!strncasecmp(model, "Lunar", 5) ||
!strncasecmp(model, "Lusso", 5) ||
!strncasecmp(model, "HV",2))))
{
ushort lid;
if (tag == 0xb001) // Sony ModelID
{
unique_id = get2();
setSonyBodyFeatures(unique_id);
if (table_buf_0x9050_present)
{
process_Sony_0x9050(table_buf_0x9050, unique_id);
free (table_buf_0x9050);
table_buf_0x9050_present = 0;
}
if (table_buf_0x940c_present)
{
if (imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Sony_E)
{
process_Sony_0x940c(table_buf_0x940c);
}
free (table_buf_0x940c);
table_buf_0x940c_present = 0;
}
}
else if ((tag == 0x0010) && // CameraInfo
strncasecmp(model, "DSLR-A100", 9) &&
strncasecmp(model, "NEX-5C", 6) &&
!strncasecmp(make, "SONY", 4) &&
((len == 368) || // a700
(len == 5478) || // a850, a900
(len == 5506) || // a200, a300, a350
(len == 6118) || // a230, a290, a330, a380, a390
// a450, a500, a550, a560, a580
// a33, a35, a55
// NEX3, NEX5, NEX5C, NEXC3, VG10E
(len == 15360))
)
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
if (memcmp(table_buf, "\xff\xff\xff\xff\xff\xff\xff\xff", 8) &&
memcmp(table_buf, "\x00\x00\x00\x00\x00\x00\x00\x00", 8))
{
switch (len) {
case 368:
case 5478:
// a700, a850, a900: CameraInfo
if (saneSonyCameraInfo(table_buf[0], table_buf[3], table_buf[2], table_buf[5], table_buf[4], table_buf[7]))
{
if (table_buf[0] | table_buf[3])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[0]) * 100 + bcd2dec(table_buf[3]);
if (table_buf[2] | table_buf[5])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[2]) * 100 + bcd2dec(table_buf[5]);
if (table_buf[4])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[4]) / 10.0f;
if (table_buf[4])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[7]) / 10.0f;
parseSonyLensFeatures(table_buf[1], table_buf[6]);
}
break;
default:
// CameraInfo2 & 3
if (saneSonyCameraInfo(table_buf[1], table_buf[2], table_buf[3], table_buf[4], table_buf[5], table_buf[6]))
{
if (table_buf[1] | table_buf[2])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[1]) * 100 + bcd2dec(table_buf[2]);
if (table_buf[3] | table_buf[4])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[3]) * 100 + bcd2dec(table_buf[4]);
if (table_buf[5])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[5]) / 10.0f;
if (table_buf[6])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[6]) / 10.0f;
parseSonyLensFeatures(table_buf[0], table_buf[7]);
}
}
}
free(table_buf);
}
else if (tag == 0x0105) // Teleconverter
{
imgdata.lens.makernotes.TeleconverterID = get2();
}
else if (tag == 0x0114) // CameraSettings
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
switch (len) {
case 280:
case 364:
case 332:
// CameraSettings and CameraSettings2 are big endian
if (table_buf[2] | table_buf[3])
{
lid = (((ushort)table_buf[2])<<8) |
((ushort)table_buf[3]);
imgdata.lens.makernotes.CurAp =
powf64(2.0f, ((float)lid/8.0f-1.0f)/2.0f);
}
break;
case 1536:
case 2048:
// CameraSettings3 are little endian
parseSonyLensType2(table_buf[1016], table_buf[1015]);
if (imgdata.lens.makernotes.LensMount != LIBRAW_MOUNT_Canon_EF)
{
switch (table_buf[153]) {
case 16: imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Minolta_A; break;
case 17: imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Sony_E; break;
}
}
break;
}
free(table_buf);
}
else if (tag == 0x9050) // little endian
{
table_buf_0x9050 = (uchar*)malloc(len);
table_buf_0x9050_present = 1;
fread(table_buf_0x9050, len, 1, ifp);
if (imgdata.lens.makernotes.CamID)
{
process_Sony_0x9050(table_buf_0x9050, imgdata.lens.makernotes.CamID);
free (table_buf_0x9050);
table_buf_0x9050_present = 0;
}
}
else if (tag == 0x940c)
{
table_buf_0x940c = (uchar*)malloc(len);
table_buf_0x940c_present = 1;
fread(table_buf_0x940c, len, 1, ifp);
if ((imgdata.lens.makernotes.CamID) &&
(imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Sony_E))
{
process_Sony_0x940c(table_buf_0x940c);
free(table_buf_0x940c);
table_buf_0x940c_present = 0;
}
}
else if (((tag == 0xb027) || (tag == 0x010c)) && (imgdata.lens.makernotes.LensID == -1))
{
imgdata.lens.makernotes.LensID = get4();
if ((imgdata.lens.makernotes.LensID > 61184) &&
(imgdata.lens.makernotes.LensID < 65535))
{
imgdata.lens.makernotes.LensID -= 61184;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
}
if (tag == 0x010c) imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Minolta_A;
}
else if (tag == 0xb02a) // Sony LensSpec
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
if (saneSonyCameraInfo(table_buf[1], table_buf[2], table_buf[3], table_buf[4], table_buf[5], table_buf[6]))
{
if (table_buf[1] | table_buf[2])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[1]) * 100 + bcd2dec(table_buf[2]);
if (table_buf[3] | table_buf[4])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[3]) * 100 + bcd2dec(table_buf[4]);
if (table_buf[5])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[5]) / 10.0f;
if (table_buf[6])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[6]) / 10.0f;
parseSonyLensFeatures(table_buf[0], table_buf[7]);
}
free(table_buf);
}
}
next:
fseek (ifp, save, SEEK_SET);
}
quit:
order = sorder;
}
#else
void CLASS parse_makernote_0xc634(int base, int uptag, unsigned dng_writer)
{
/*placeholder */
}
#endif
void CLASS parse_makernote (int base, int uptag)
{
unsigned offset=0, entries, tag, type, len, save, c;
unsigned ver97=0, serial=0, i, wbi=0, wb[4]={0,0,0,0};
uchar buf97[324], ci, cj, ck;
short morder, sorder=order;
char buf[10];
unsigned SamsungKey[11];
static const double rgb_adobe[3][3] = // inv(sRGB2XYZ_D65) * AdobeRGB2XYZ_D65
{{ 1.398283396477404, -0.398283116703571, 4.427165001263944E-08},
{-1.233904514232401E-07, 0.999999995196570, 3.126724276714121e-08},
{ 4.561487232726535E-08, -0.042938290466635, 1.042938250416105 }};
float adobe_cam [3][3];
uchar NikonKey;
#ifdef LIBRAW_LIBRARY_BUILD
unsigned NikonLensDataVersion = 0;
unsigned lenNikonLensData = 0;
uchar *CanonCameraInfo;
unsigned lenCanonCameraInfo = 0;
uchar *table_buf;
uchar *table_buf_0x9050;
ushort table_buf_0x9050_present = 0;
uchar *table_buf_0x940c;
ushort table_buf_0x940c_present = 0;
#endif
/*
The MakerNote might have its own TIFF header (possibly with
its own byte-order!), or it might just be a table.
*/
if (!strncmp(make,"Nokia",5)) return;
fread (buf, 1, 10, ifp);
if (!strncmp (buf,"KDK" ,3) || /* these aren't TIFF tables */
!strncmp (buf,"VER" ,3) ||
!strncmp (buf,"IIII",4) ||
!strncmp (buf,"MMMM",4)) return;
if (!strncmp (buf,"KC" ,2) || /* Konica KD-400Z, KD-510Z */
!strncmp (buf,"MLY" ,3)) { /* Minolta DiMAGE G series */
order = 0x4d4d;
while ((i=ftell(ifp)) < data_offset && i < 16384) {
wb[0] = wb[2]; wb[2] = wb[1]; wb[1] = wb[3];
wb[3] = get2();
if (wb[1] == 256 && wb[3] == 256 &&
wb[0] > 256 && wb[0] < 640 && wb[2] > 256 && wb[2] < 640)
FORC4 cam_mul[c] = wb[c];
}
goto quit;
}
if (!strcmp (buf,"Nikon")) {
base = ftell(ifp);
order = get2();
if (get2() != 42) goto quit;
offset = get4();
fseek (ifp, offset-8, SEEK_CUR);
} else if (!strcmp (buf,"OLYMPUS") ||
!strcmp (buf,"PENTAX ")) {
base = ftell(ifp)-10;
fseek (ifp, -2, SEEK_CUR);
order = get2();
if (buf[0] == 'O') get2();
} else if (!strncmp (buf,"SONY",4) ||
!strcmp (buf,"Panasonic")) {
goto nf;
} else if (!strncmp (buf,"FUJIFILM",8)) {
base = ftell(ifp)-10;
nf: order = 0x4949;
fseek (ifp, 2, SEEK_CUR);
} else if (!strcmp (buf,"OLYMP") ||
!strcmp (buf,"LEICA") ||
!strcmp (buf,"Ricoh") ||
!strcmp (buf,"EPSON"))
fseek (ifp, -2, SEEK_CUR);
else if (!strcmp (buf,"AOC") ||
!strcmp (buf,"QVC"))
fseek (ifp, -4, SEEK_CUR);
else {
fseek (ifp, -10, SEEK_CUR);
if (!strncmp(make,"SAMSUNG",7))
base = ftell(ifp);
}
// adjust pos & base for Leica M8/M9/M Mono tags and dir in tag 0x3400
if (!strncasecmp(make, "LEICA", 5))
{
if (!strncmp(model, "M8", 2) ||
!strncasecmp(model, "Leica M8", 8) ||
!strncasecmp(model, "LEICA X", 7))
{
base = ftell(ifp)-8;
}
else if (!strncasecmp(model, "LEICA M (Typ 240)", 17))
{
base = 0;
}
else if (!strncmp(model, "M9", 2) ||
!strncasecmp(model, "Leica M9", 8) ||
!strncasecmp(model, "M Monochrom", 11) ||
!strncasecmp(model, "Leica M Monochrom", 11))
{
if (!uptag)
{
base = ftell(ifp) - 10;
fseek (ifp, 8, SEEK_CUR);
}
else if (uptag == 0x3400)
{
fseek (ifp, 10, SEEK_CUR);
base += 10;
}
}
else if (!strncasecmp(model, "LEICA T", 7))
{
base = ftell(ifp)-8;
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_T;
#endif
}
#ifdef LIBRAW_LIBRARY_BUILD
else if (!strncasecmp(model, "LEICA SL", 8))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_SL;
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_FF;
}
#endif
}
entries = get2();
if (entries > 1000) return;
morder = order;
while (entries--) {
order = morder;
tiff_get (base, &tag, &type, &len, &save);
tag |= uptag << 16;
if(len > 100*1024*1024) continue; // 100Mb tag? No!
#ifdef LIBRAW_LIBRARY_BUILD
INT64 _pos = ftell(ifp);
if (!strncmp(make, "Canon",5))
{
if (tag == 0x0001) // camera settings
{
fseek(ifp, 44, SEEK_CUR);
imgdata.lens.makernotes.LensID = get2();
imgdata.lens.makernotes.MaxFocal = get2();
imgdata.lens.makernotes.MinFocal = get2();
imgdata.lens.makernotes.CanonFocalUnits = get2();
if (imgdata.lens.makernotes.CanonFocalUnits != 1)
{
imgdata.lens.makernotes.MaxFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
imgdata.lens.makernotes.MinFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
imgdata.lens.makernotes.MaxAp = _CanonConvertAperture(get2());
imgdata.lens.makernotes.MinAp = _CanonConvertAperture(get2());
}
else if (tag == 0x0002) // focal length
{
imgdata.lens.makernotes.FocalType = get2();
imgdata.lens.makernotes.CurFocal = get2();
if ((imgdata.lens.makernotes.CanonFocalUnits != 1) &&
imgdata.lens.makernotes.CanonFocalUnits)
{
imgdata.lens.makernotes.CurFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
}
else if (tag == 0x0004) // shot info
{
short tempAp;
fseek(ifp, 8, SEEK_CUR);
if ((tempAp = get2()) != 0x7fff)
imgdata.lens.makernotes.CurAp = _CanonConvertAperture(tempAp);
if (imgdata.lens.makernotes.CurAp < 0.7f)
{
fseek(ifp, 32, SEEK_CUR);
imgdata.lens.makernotes.CurAp = _CanonConvertAperture(get2());
}
if (!aperture) aperture = imgdata.lens.makernotes.CurAp;
}
else if (tag == 0x000d) // camera info
{
CanonCameraInfo = (uchar*)malloc(len);
fread(CanonCameraInfo, len, 1, ifp);
lenCanonCameraInfo = len;
}
else if (tag == 0x0095 && // lens model tag
!imgdata.lens.makernotes.Lens[0])
{
fread(imgdata.lens.makernotes.Lens, 2, 1, ifp);
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
if (imgdata.lens.makernotes.Lens[0] < 65) // non-Canon lens
fread(imgdata.lens.makernotes.Lens + 2, 62, 1, ifp);
else
{
char efs[2];
imgdata.lens.makernotes.LensFeatures_pre[0] = imgdata.lens.makernotes.Lens[0];
imgdata.lens.makernotes.LensFeatures_pre[1] = imgdata.lens.makernotes.Lens[1];
fread(efs, 2, 1, ifp);
if (efs[0] == 45 && (efs[1] == 83 || efs[1] == 69 || efs[1] == 77))
{ // "EF-S, TS-E, MP-E, EF-M" lenses
imgdata.lens.makernotes.Lens[2] = imgdata.lens.makernotes.LensFeatures_pre[2] = efs[0];
imgdata.lens.makernotes.Lens[3] = imgdata.lens.makernotes.LensFeatures_pre[3] = efs[1];
imgdata.lens.makernotes.Lens[4] = 32;
if (efs[1] == 83)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_S;
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_APSC;
}
else if (efs[1] == 77)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF_M;
}
}
else
{ // "EF" lenses
imgdata.lens.makernotes.Lens[2] = 32;
imgdata.lens.makernotes.Lens[3] = efs[0];
imgdata.lens.makernotes.Lens[4] = efs[1];
}
fread(imgdata.lens.makernotes.Lens + 5, 58, 1, ifp);
}
}
}
else if (!strncmp(make, "FUJI", 4))
switch (tag) {
case 0x1404: imgdata.lens.makernotes.MinFocal = getreal(type); break;
case 0x1405: imgdata.lens.makernotes.MaxFocal = getreal(type); break;
case 0x1406: imgdata.lens.makernotes.MaxAp4MinFocal = getreal(type); break;
case 0x1407: imgdata.lens.makernotes.MaxAp4MaxFocal = getreal(type); break;
}
else if (!strncasecmp(make, "LEICA", 5))
{
if (((tag == 0x035e) || (tag == 0x035f)) && (type == 10) && (len == 9))
{
int ind = tag == 0x035e?0:1;
for (int j=0; j < 3; j++)
FORCC imgdata.color.dng_color[ind].forwardmatrix[c][j]= getreal(type);
}
if ((tag == 0x0303) && (type != 4))
{
fread(imgdata.lens.makernotes.Lens, MIN(len,127), 1, ifp);
}
if ((tag == 0x3405) ||
(tag == 0x0310) ||
(tag == 0x34003405))
{
imgdata.lens.makernotes.LensID = get4();
imgdata.lens.makernotes.LensID =
((imgdata.lens.makernotes.LensID>>2)<<8) |
(imgdata.lens.makernotes.LensID & 0x3);
if (imgdata.lens.makernotes.LensID != -1)
{
if ((model[0] == 'M') ||
!strncasecmp (model, "LEICA M", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_M;
if (imgdata.lens.makernotes.LensID)
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Leica_M;
}
else if ((model[0] == 'S') ||
!strncasecmp (model, "LEICA S", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_S;
if (imgdata.lens.makernotes.Lens[0])
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Leica_S;
}
}
}
else if (
((tag == 0x0313) || (tag == 0x34003406)) &&
(fabs(imgdata.lens.makernotes.CurAp) < 0.17f) &&
((type == 10) || (type == 5))
)
{
imgdata.lens.makernotes.CurAp = getreal(type);
if (imgdata.lens.makernotes.CurAp > 126.3)
imgdata.lens.makernotes.CurAp = 0.0f;
}
else if (tag == 0x3400)
{
parse_makernote (base, 0x3400);
}
}
else if (!strncmp(make, "NIKON",5))
{
if (tag == 0x000a)
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
else if (tag == 0x0082) // lens attachment
{
fread(imgdata.lens.makernotes.Attachment, MIN(len,127), 1, ifp);
}
else if (tag == 0x0083) // lens type
{
imgdata.lens.nikon.NikonLensType = fgetc(ifp);
}
else if (tag == 0x0084) // lens
{
imgdata.lens.makernotes.MinFocal = getreal(type);
imgdata.lens.makernotes.MaxFocal = getreal(type);
imgdata.lens.makernotes.MaxAp4MinFocal = getreal(type);
imgdata.lens.makernotes.MaxAp4MaxFocal = getreal(type);
}
else if (tag == 0x008b) // lens f-stops
{
uchar a, b, c;
a = fgetc(ifp);
b = fgetc(ifp);
c = fgetc(ifp);
if (c)
{
imgdata.lens.nikon.NikonLensFStops = a*b*(12/c);
imgdata.lens.makernotes.LensFStops =
(float)imgdata.lens.nikon.NikonLensFStops /12.0f;
}
}
else if (tag == 0x0093)
{
i = get2();
if ((i == 7) || (i == 9))
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
}
else if (tag == 0x0098) // contains lens data
{
for (i = 0; i < 4; i++)
{
NikonLensDataVersion = NikonLensDataVersion * 10 + fgetc(ifp) - '0';
}
switch (NikonLensDataVersion)
{
case 100: lenNikonLensData = 9; break;
case 101:
case 201: // encrypted, starting from v.201
case 202:
case 203: lenNikonLensData = 15; break;
case 204: lenNikonLensData = 16; break;
case 400: lenNikonLensData = 459; break;
case 401: lenNikonLensData = 590; break;
case 402: lenNikonLensData = 509; break;
case 403: lenNikonLensData = 879; break;
}
if(lenNikonLensData>0)
{
table_buf = (uchar*)malloc(lenNikonLensData);
fread(table_buf, lenNikonLensData, 1, ifp);
if ((NikonLensDataVersion < 201) && lenNikonLensData)
{
processNikonLensData(table_buf, lenNikonLensData);
free(table_buf);
lenNikonLensData = 0;
}
}
}
}
else if (!strncmp(make, "OLYMPUS", 7))
{
switch (tag) {
case 0x0207:
case 0x20100100:
{
uchar sOlyID[8];
unsigned long long OlyID;
fread (sOlyID, MIN(len,7), 1, ifp);
sOlyID[7] = 0;
OlyID = sOlyID[0];
i = 1;
while (i < 7 && sOlyID[i])
{
OlyID = OlyID << 8 | sOlyID[i];
i++;
}
setOlympusBodyFeatures(OlyID);
}
break;
case 0x1002:
imgdata.lens.makernotes.CurAp = powf64(2.0f, getreal(type)/2);
break;
case 0x20401112:
imgdata.sizes.OlympusCropID = get2();
break;
case 0x20401113:
FORC4 imgdata.sizes.OlympusFrame[c] = get2();
break;
case 0x20100201:
{
unsigned long long oly_lensid [3];
oly_lensid[0] = fgetc(ifp);
fgetc(ifp);
oly_lensid[1] = fgetc(ifp);
oly_lensid[2] = fgetc(ifp);
imgdata.lens.makernotes.LensID =
(oly_lensid[0] << 16) | (oly_lensid[1] << 8) | oly_lensid[2];
}
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FT;
imgdata.lens.makernotes.LensFormat = LIBRAW_FORMAT_FT;
if (((imgdata.lens.makernotes.LensID < 0x20000) ||
(imgdata.lens.makernotes.LensID > 0x4ffff)) &&
(imgdata.lens.makernotes.LensID & 0x10))
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_mFT;
}
break;
case 0x20100203:
fread(imgdata.lens.makernotes.Lens, MIN(len,127), 1, ifp);
break;
case 0x20100205:
imgdata.lens.makernotes.MaxAp4MinFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100206:
imgdata.lens.makernotes.MaxAp4MaxFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100207:
imgdata.lens.makernotes.MinFocal = (float)get2();
break;
case 0x20100208:
imgdata.lens.makernotes.MaxFocal = (float)get2();
if (imgdata.lens.makernotes.MaxFocal > 1000.0f)
imgdata.lens.makernotes.MaxFocal = imgdata.lens.makernotes.MinFocal;
break;
case 0x2010020a:
imgdata.lens.makernotes.MaxAp4CurFocal = powf64(sqrt(2.0f), get2() / 256.0f);
break;
case 0x20100301:
imgdata.lens.makernotes.TeleconverterID = fgetc(ifp) << 8;
fgetc(ifp);
imgdata.lens.makernotes.TeleconverterID =
imgdata.lens.makernotes.TeleconverterID | fgetc(ifp);
break;
case 0x20100303:
fread(imgdata.lens.makernotes.Teleconverter, MIN(len,127), 1, ifp);
break;
case 0x20100403:
fread(imgdata.lens.makernotes.Attachment, MIN(len,127), 1, ifp);
break;
}
}
else if ((!strncmp(make, "PENTAX", 6) || !strncmp(make, "RICOH", 5)) &&
!strncmp(model, "GR", 2))
{
if ((tag == 0x1001) && (type == 3))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
imgdata.lens.makernotes.LensID = -1;
imgdata.lens.makernotes.FocalType = 1;
}
else if ((tag == 0x1017) && (get2() == 2))
{
strcpy(imgdata.lens.makernotes.Attachment, "Wide-Angle Adapter");
}
else if (tag == 0x1500)
{
imgdata.lens.makernotes.CurFocal = getreal(type);
}
}
else if (!strncmp(make, "RICOH", 5) &&
strncmp(model, "PENTAX", 6))
{
if ((tag == 0x1017) && (get2() == 2))
{
strcpy(imgdata.lens.makernotes.Attachment, "Wide-Angle Adapter");
}
else if (tag == 0x1500)
{
imgdata.lens.makernotes.CurFocal = getreal(type);
}
else if ((tag == 0x2001) && !strncmp(model, "GXR", 3))
{
short ntags, cur_tag;
fseek(ifp, 20, SEEK_CUR);
ntags = get2();
cur_tag = get2();
while (cur_tag != 0x002c)
{
fseek(ifp, 10, SEEK_CUR);
cur_tag = get2();
}
fseek(ifp, 6, SEEK_CUR);
fseek(ifp, get4()+34, SEEK_SET);
imgdata.lens.makernotes.LensID = getc(ifp) - '0';
switch(imgdata.lens.makernotes.LensID)
{
case 1:
case 2:
case 3:
case 5:
case 6:
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_RicohModule;
break;
case 8:
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Leica_M;
imgdata.lens.makernotes.CameraFormat = LIBRAW_FORMAT_APSC;
imgdata.lens.makernotes.LensID = -1;
break;
default:
imgdata.lens.makernotes.LensID = -1;
}
}
}
else if ((!strncmp(make, "PENTAX", 6) ||
!strncmp(model, "PENTAX", 6) ||
(!strncmp(make, "SAMSUNG", 7) && dng_version)) &&
strncmp(model, "GR", 2))
{
if (tag == 0x0005)
{
unique_id = get4();
setPentaxBodyFeatures(unique_id);
}
else if (tag == 0x0013)
{
imgdata.lens.makernotes.CurAp = (float)get2()/10.0f;
}
else if (tag == 0x001d)
{
imgdata.lens.makernotes.CurFocal = (float)get4()/100.0f;
}
else if (tag == 0x003f)
{
imgdata.lens.makernotes.LensID = fgetc(ifp) << 8 | fgetc(ifp);
}
else if (tag == 0x0207)
{
PentaxLensInfo(imgdata.lens.makernotes.CamID, len);
}
else if (tag == 0x020d)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ (c >> 1)] = get2();
}
else if (tag == 0x020e)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c ^ (c >> 1)] = get2();
}
else if (tag == 0x020f)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0210)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0211)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0212)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0213)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0214)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0221)
{
int nWB = get2();
if(nWB<=sizeof(imgdata.color.WBCT_Coeffs)/sizeof(imgdata.color.WBCT_Coeffs[0]))
for (int i = 0; i < nWB; i++)
{
imgdata.color.WBCT_Coeffs[i][0] = (unsigned)0xcfc6 - get2();
fseek(ifp, 2, SEEK_CUR);
imgdata.color.WBCT_Coeffs[i][1] = get2();
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] = 0x2000;
imgdata.color.WBCT_Coeffs[i][3] = get2();
}
}
else if (tag == 0x022d)
{
fseek (ifp,2,SEEK_CUR);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c ^ (c >> 1)] = get2();
getc(ifp);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][c ^ (c >> 1)] = get2();
}
else if (tag == 0x0239) // Q-series lens info (LensInfoQ)
{
char LensInfo [20];
fseek (ifp, 2, SEEK_CUR);
fread(imgdata.lens.makernotes.Lens, 30, 1, ifp);
strcat(imgdata.lens.makernotes.Lens, " ");
fread(LensInfo, 20, 1, ifp);
strcat(imgdata.lens.makernotes.Lens, LensInfo);
}
}
else if (!strncmp(make, "SAMSUNG", 7))
{
if (tag == 0x0002)
{
if(get4() == 0x2000)
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Samsung_NX;
}
else if (!strncmp(model, "NX mini", 7))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Samsung_NX_M;
}
else
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
}
}
else if (tag == 0x0003)
{
unique_id = imgdata.lens.makernotes.CamID = get4();
}
else if (tag == 0xa003)
{
imgdata.lens.makernotes.LensID = get2();
if (imgdata.lens.makernotes.LensID)
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Samsung_NX;
}
else if (tag == 0xa019)
{
imgdata.lens.makernotes.CurAp = getreal(type);
}
else if (tag == 0xa01a)
{
imgdata.lens.makernotes.FocalLengthIn35mmFormat = get4() / 10.0f;
if (imgdata.lens.makernotes.FocalLengthIn35mmFormat < 10.0f)
imgdata.lens.makernotes.FocalLengthIn35mmFormat *= 10.0f;
}
}
else if (!strncasecmp(make, "SONY", 4) ||
!strncasecmp(make, "Konica", 6) ||
!strncasecmp(make, "Minolta", 7) ||
(!strncasecmp(make, "Hasselblad", 10) &&
(!strncasecmp(model, "Stellar", 7) ||
!strncasecmp(model, "Lunar", 5) ||
!strncasecmp(model, "Lusso", 5) ||
!strncasecmp(model, "HV",2))))
{
ushort lid;
if (tag == 0xb001) // Sony ModelID
{
unique_id = get2();
setSonyBodyFeatures(unique_id);
if (table_buf_0x9050_present)
{
process_Sony_0x9050(table_buf_0x9050, unique_id);
free (table_buf_0x9050);
table_buf_0x9050_present = 0;
}
if (table_buf_0x940c_present)
{
if (imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Sony_E)
{
process_Sony_0x940c(table_buf_0x940c);
}
free (table_buf_0x940c);
table_buf_0x940c_present = 0;
}
}
else if ((tag == 0x0010) && // CameraInfo
strncasecmp(model, "DSLR-A100", 9) &&
strncasecmp(model, "NEX-5C", 6) &&
!strncasecmp(make, "SONY", 4) &&
((len == 368) || // a700
(len == 5478) || // a850, a900
(len == 5506) || // a200, a300, a350
(len == 6118) || // a230, a290, a330, a380, a390
// a450, a500, a550, a560, a580
// a33, a35, a55
// NEX3, NEX5, NEX5C, NEXC3, VG10E
(len == 15360))
)
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
if (memcmp(table_buf, "\xff\xff\xff\xff\xff\xff\xff\xff", 8) &&
memcmp(table_buf, "\x00\x00\x00\x00\x00\x00\x00\x00", 8))
{
switch (len)
{
case 368:
case 5478:
// a700, a850, a900: CameraInfo
if (table_buf[0] | table_buf[3])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[0]) * 100 + bcd2dec(table_buf[3]);
if (table_buf[2] | table_buf[5])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[2]) * 100 + bcd2dec(table_buf[5]);
if (table_buf[4])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[4]) / 10.0f;
if (table_buf[4])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[7]) / 10.0f;
parseSonyLensFeatures(table_buf[1], table_buf[6]);
break;
default:
// CameraInfo2 & 3
if (table_buf[1] | table_buf[2])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[1]) * 100 + bcd2dec(table_buf[2]);
if (table_buf[3] | table_buf[4])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[3]) * 100 + bcd2dec(table_buf[4]);
if (table_buf[5])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[5]) / 10.0f;
if (table_buf[6])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[6]) / 10.0f;
parseSonyLensFeatures(table_buf[0], table_buf[7]);
}
}
free(table_buf);
}
else if (tag == 0x0105) // Teleconverter
{
imgdata.lens.makernotes.TeleconverterID = get2();
}
else if (tag == 0x0114) // CameraSettings
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
switch (len) {
case 280:
case 364:
case 332:
// CameraSettings and CameraSettings2 are big endian
if (table_buf[2] | table_buf[3])
{
lid = (((ushort)table_buf[2])<<8) |
((ushort)table_buf[3]);
imgdata.lens.makernotes.CurAp =
powf64(2.0f, ((float)lid/8.0f-1.0f)/2.0f);
}
break;
case 1536:
case 2048:
// CameraSettings3 are little endian
parseSonyLensType2(table_buf[1016], table_buf[1015]);
if (imgdata.lens.makernotes.LensMount != LIBRAW_MOUNT_Canon_EF)
{
switch (table_buf[153]) {
case 16: imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Minolta_A; break;
case 17: imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Sony_E; break;
}
}
break;
}
free(table_buf);
}
else if (tag == 0x9050) // little endian
{
table_buf_0x9050 = (uchar*)malloc(len);
table_buf_0x9050_present = 1;
fread(table_buf_0x9050, len, 1, ifp);
if (imgdata.lens.makernotes.CamID)
{
process_Sony_0x9050(table_buf_0x9050, imgdata.lens.makernotes.CamID);
free (table_buf_0x9050);
table_buf_0x9050_present = 0;
}
}
else if (tag == 0x940c)
{
table_buf_0x940c = (uchar*)malloc(len);
table_buf_0x940c_present = 1;
fread(table_buf_0x940c, len, 1, ifp);
if ((imgdata.lens.makernotes.CamID) &&
(imgdata.lens.makernotes.CameraMount == LIBRAW_MOUNT_Sony_E))
{
process_Sony_0x940c(table_buf_0x940c);
free(table_buf_0x940c);
table_buf_0x940c_present = 0;
}
}
else if (((tag == 0xb027) || (tag == 0x010c)) && (imgdata.lens.makernotes.LensID == -1))
{
imgdata.lens.makernotes.LensID = get4();
if ((imgdata.lens.makernotes.LensID > 61184) &&
(imgdata.lens.makernotes.LensID < 65535))
{
imgdata.lens.makernotes.LensID -= 61184;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Canon_EF;
}
if (tag == 0x010c) imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Minolta_A;
}
else if (tag == 0xb02a) // Sony LensSpec
{
table_buf = (uchar*)malloc(len);
fread(table_buf, len, 1, ifp);
if (table_buf[1] | table_buf[2])
imgdata.lens.makernotes.MinFocal =
bcd2dec(table_buf[1]) * 100 + bcd2dec(table_buf[2]);
if (table_buf[3] | table_buf[4])
imgdata.lens.makernotes.MaxFocal =
bcd2dec(table_buf[3]) * 100 + bcd2dec(table_buf[4]);
if (table_buf[5])
imgdata.lens.makernotes.MaxAp4MinFocal = bcd2dec(table_buf[5]) / 10.0f;
if (table_buf[6])
imgdata.lens.makernotes.MaxAp4MaxFocal = bcd2dec(table_buf[6]) / 10.0f;
parseSonyLensFeatures(table_buf[0], table_buf[7]);
free(table_buf);
}
}
fseek(ifp,_pos,SEEK_SET);
#endif
if (tag == 2 && strstr(make,"NIKON") && !iso_speed)
iso_speed = (get2(),get2());
if (tag == 37 && strstr(make,"NIKON") && (!iso_speed || iso_speed == 65535))
{
unsigned char cc;
fread(&cc,1,1,ifp);
iso_speed = int(100.0 * powf64(2.0f,float(cc)/12.0-5.0));
}
if (tag == 4 && len > 26 && len < 35) {
if ((i=(get4(),get2())) != 0x7fff && (!iso_speed || iso_speed == 65535))
iso_speed = 50 * powf64(2.0, i/32.0 - 4);
#ifdef LIBRAW_LIBRARY_BUILD
get4();
#else
if ((i=(get2(),get2())) != 0x7fff && !aperture)
aperture = powf64(2.0, i/64.0);
#endif
if ((i=get2()) != 0xffff && !shutter)
shutter = powf64(2.0, (short) i/-32.0);
wbi = (get2(),get2());
shot_order = (get2(),get2());
}
if ((tag == 4 || tag == 0x114) && !strncmp(make,"KONICA",6)) {
fseek (ifp, tag == 4 ? 140:160, SEEK_CUR);
switch (get2()) {
case 72: flip = 0; break;
case 76: flip = 6; break;
case 82: flip = 5; break;
}
}
if (tag == 7 && type == 2 && len > 20)
fgets (model2, 64, ifp);
if (tag == 8 && type == 4)
shot_order = get4();
if (tag == 9 && !strncmp(make,"Canon",5))
fread (artist, 64, 1, ifp);
if (tag == 0xc && len == 4)
FORC3 cam_mul[(c << 1 | c >> 1) & 3] = getreal(type);
if (tag == 0xd && type == 7 && get2() == 0xaaaa) {
for (c=i=2; (ushort) c != 0xbbbb && i < len; i++)
c = c << 8 | fgetc(ifp);
while ((i+=4) < len-5)
if (get4() == 257 && (i=len) && (c = (get4(),fgetc(ifp))) < 3)
flip = "065"[c]-'0';
}
if (tag == 0x10 && type == 4)
{
unique_id = get4();
#ifdef LIBRAW_LIBRARY_BUILD
if (unique_id == 0x03740000) unique_id = 0x80000374;
if (unique_id == 0x03840000) unique_id = 0x80000384;
setCanonBodyFeatures(unique_id);
if (lenCanonCameraInfo)
{
processCanonCameraInfo(unique_id, CanonCameraInfo,lenCanonCameraInfo);
free(CanonCameraInfo);
CanonCameraInfo = 0;
lenCanonCameraInfo = 0;
}
#endif
}
#ifdef LIBRAW_LIBRARY_BUILD
INT64 _pos2 = ftell(ifp);
if (!strncasecmp(make,"Olympus",7))
{
short nWB, tWB;
if ((tag == 0x20400102) && (len == 2) &&
(!strncasecmp(model, "E-410", 5) || !strncasecmp(model, "E-510", 5)))
{
int i;
for (i=0; i<64; i++)
imgdata.color.WBCT_Coeffs[i][2] = imgdata.color.WBCT_Coeffs[i][4] =
imgdata.color.WB_Coeffs[i][1] = imgdata.color.WB_Coeffs[i][3] = 0x100;
for (i=64; i<256; i++)
imgdata.color.WB_Coeffs[i][1] = imgdata.color.WB_Coeffs[i][3] = 0x100;
}
if ((tag >= 0x20400102) && (tag <= 0x2040010d))
{
ushort CT;
nWB = tag-0x20400102;
switch (nWB)
{
case 0 : CT = 3000; tWB = LIBRAW_WBI_Tungsten; break;
case 1 : CT = 3300; tWB = 0x100; break;
case 2 : CT = 3600; tWB = 0x100; break;
case 3 : CT = 3900; tWB = 0x100; break;
case 4 : CT = 4000; tWB = LIBRAW_WBI_FL_W; break;
case 5 : CT = 4300; tWB = 0x100; break;
case 6 : CT = 4500; tWB = LIBRAW_WBI_FL_D; break;
case 7 : CT = 4800; tWB = 0x100; break;
case 8 : CT = 5300; tWB = LIBRAW_WBI_FineWeather; break;
case 9 : CT = 6000; tWB = LIBRAW_WBI_Cloudy; break;
case 10: CT = 6600; tWB = LIBRAW_WBI_FL_N; break;
case 11: CT = 7500; tWB = LIBRAW_WBI_Shade; break;
default: CT = 0; tWB = 0x100;
}
if (CT)
{
imgdata.color.WBCT_Coeffs[nWB][0] = CT;
imgdata.color.WBCT_Coeffs[nWB][1] = get2();
imgdata.color.WBCT_Coeffs[nWB][3] = get2();
if (len == 4)
{
imgdata.color.WBCT_Coeffs[nWB][2] = get2();
imgdata.color.WBCT_Coeffs[nWB][4] = get2();
}
}
if (tWB != 0x100)
FORC4 imgdata.color.WB_Coeffs[tWB][c] = imgdata.color.WBCT_Coeffs[nWB][c+1];
}
if ((tag >= 0x20400113) && (tag <= 0x2040011e))
{
nWB = tag-0x20400113;
imgdata.color.WBCT_Coeffs[nWB][2] = imgdata.color.WBCT_Coeffs[nWB][4] = get2();
switch (nWB)
{
case 0: tWB = LIBRAW_WBI_Tungsten; break;
case 4: tWB = LIBRAW_WBI_FL_W; break;
case 6: tWB = LIBRAW_WBI_FL_D; break;
case 8: tWB = LIBRAW_WBI_FineWeather; break;
case 9: tWB = LIBRAW_WBI_Cloudy; break;
case 10: tWB = LIBRAW_WBI_FL_N; break;
case 11: tWB = LIBRAW_WBI_Shade; break;
default: tWB = 0x100;
}
if (tWB != 0x100)
imgdata.color.WB_Coeffs[tWB][1] = imgdata.color.WB_Coeffs[tWB][3] =
imgdata.color.WBCT_Coeffs[nWB][2];
}
if (tag == 0x20400121)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][2] = get2();
if (len == 4)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][1] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][3] = get2();
}
}
if (tag == 0x2040011f)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][3] = get2();
}
if (tag == 0x30000120)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][2] = get2();
if (len == 2)
{
for (int i=0; i<256; i++)
imgdata.color.WB_Coeffs[i][1] = imgdata.color.WB_Coeffs[i][3] = 0x100;
}
}
if (tag == 0x30000121)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][2] = get2();
}
if (tag == 0x30000122)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][2] = get2();
}
if (tag == 0x30000123)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][2] = get2();
}
if (tag == 0x30000124)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Sunset][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Sunset][2] = get2();
}
if (tag == 0x30000130)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][2] = get2();
}
if (tag == 0x30000131)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][2] = get2();
}
if (tag == 0x30000132)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][2] = get2();
}
if (tag == 0x30000133)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][2] = get2();
}
if(tag == 0x20400805 && len == 2)
{
imgdata.color.OlympusSensorCalibration[0]=getreal(type);
imgdata.color.OlympusSensorCalibration[1]=getreal(type);
}
}
if(tag == 0x20400805 && len == 2 && !strncasecmp(make,"Olympus",7))
{
imgdata.color.OlympusSensorCalibration[0]=getreal(type);
imgdata.color.OlympusSensorCalibration[1]=getreal(type);
}
if ((tag == 0x00a9) && !strncasecmp(make,"Canon",5))
{
long int save1 = ftell(ifp);
fseek (ifp, save1+(0x5<<1), SEEK_SET);
Canon_WBpresets(0,0);
fseek (ifp, save1, SEEK_SET);
}
if (tag == 0x4001 && len > 500 && !strncasecmp(make,"Canon",5))
{
long int save1 = ftell(ifp);
switch (len)
{
case 582:
imgdata.color.canon_makernotes.CanonColorDataVer = 1; // 20D / 350D
{
fseek (ifp, save1+(0x23<<1), SEEK_SET);
Canon_WBpresets(2,2);
fseek (ifp, save1+(0x4b<<1), SEEK_SET);
Canon_WBCTpresets (1); // ABCT
}
break;
case 653:
imgdata.color.canon_makernotes.CanonColorDataVer = 2; // 1Dmk2 / 1DsMK2
{
fseek (ifp, save1+(0x27<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0xa4<<1), SEEK_SET);
Canon_WBCTpresets (1); // ABCT
}
break;
case 796:
imgdata.color.canon_makernotes.CanonColorDataVer = 3; // 1DmkIIN / 5D / 30D / 400D
imgdata.color.canon_makernotes.CanonColorDataSubVer = get2();
{
fseek (ifp, save1+(0x4e<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0x85<<1), SEEK_SET);
Canon_WBCTpresets (0); // BCAT
fseek (ifp, save1+(0x0c4<<1), SEEK_SET); // offset 196 short
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
break;
// 1DmkIII / 1DSmkIII / 1DmkIV / 5DmkII
// 7D / 40D / 50D / 60D / 450D / 500D
// 550D / 1000D / 1100D
case 674: case 692: case 702: case 1227: case 1250:
case 1251: case 1337: case 1338: case 1346:
imgdata.color.canon_makernotes.CanonColorDataVer = 4;
imgdata.color.canon_makernotes.CanonColorDataSubVer = get2();
{
fseek (ifp, save1+(0x53<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0xa8<<1), SEEK_SET);
Canon_WBCTpresets (0); // BCAT
fseek (ifp, save1+(0x0e7<<1), SEEK_SET); // offset 231 short
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
if ((imgdata.color.canon_makernotes.CanonColorDataSubVer == 4)
|| (imgdata.color.canon_makernotes.CanonColorDataSubVer == 5))
{
fseek (ifp, save1+(0x2b9<<1), SEEK_SET); // offset 697 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
}
else if ((imgdata.color.canon_makernotes.CanonColorDataSubVer == 6) ||
(imgdata.color.canon_makernotes.CanonColorDataSubVer == 7))
{
fseek (ifp, save1+(0x2d0<<1), SEEK_SET); // offset 720 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
}
else if (imgdata.color.canon_makernotes.CanonColorDataSubVer == 9)
{
fseek (ifp, save1+(0x2d4<<1), SEEK_SET); // offset 724 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
}
break;
case 5120:
imgdata.color.canon_makernotes.CanonColorDataVer = 5; // PowerSot G10, G12, G5 X, EOS M3
{
fseek (ifp, save1+(0x56<<1), SEEK_SET);
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Other][c ^ (c >> 1)] = get2();
get2();
Canon_WBpresets(2,12);
fseek (ifp, save1+(0xba<<1), SEEK_SET);
Canon_WBCTpresets (2); // BCADT
fseek (ifp, save1+(0x108<<1), SEEK_SET); // offset 264 short
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
break;
case 1273: case 1275:
imgdata.color.canon_makernotes.CanonColorDataVer = 6; // 600D / 1200D
imgdata.color.canon_makernotes.CanonColorDataSubVer = get2();
{
fseek (ifp, save1+(0x67<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0xbc<<1), SEEK_SET);
Canon_WBCTpresets (0); // BCAT
fseek (ifp, save1+(0x0fb<<1), SEEK_SET); // offset 251 short
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
fseek (ifp, save1+(0x1e4<<1), SEEK_SET); // offset 484 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
break;
// 1DX / 5DmkIII / 6D / 100D / 650D / 700D / EOS M / 7DmkII / 750D / 760D
case 1312: case 1313: case 1316: case 1506:
imgdata.color.canon_makernotes.CanonColorDataVer = 7;
imgdata.color.canon_makernotes.CanonColorDataSubVer = get2();
{
fseek (ifp, save1+(0x80<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0xd5<<1), SEEK_SET);
Canon_WBCTpresets (0); // BCAT
fseek (ifp, save1+(0x114<<1), SEEK_SET); // offset 276 shorts
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
if (imgdata.color.canon_makernotes.CanonColorDataSubVer == 10)
{
fseek (ifp, save1+(0x1fd<<1), SEEK_SET); // offset 509 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
} else if (imgdata.color.canon_makernotes.CanonColorDataSubVer == 11)
{
fseek (ifp, save1+(0x2dd<<1), SEEK_SET); // offset 733 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
}
break;
// 5DS / 5DS R
case 1560:
imgdata.color.canon_makernotes.CanonColorDataVer = 8;
imgdata.color.canon_makernotes.CanonColorDataSubVer = get2();
{
fseek (ifp, save1+(0x85<<1), SEEK_SET);
Canon_WBpresets(2,12);
fseek (ifp, save1+(0x107<<1), SEEK_SET);
Canon_WBCTpresets (0); // BCAT
fseek (ifp, save1+(0x146<<1), SEEK_SET); // offset 326 shorts
int bls=0;
FORC4 bls+=get2();
imgdata.color.canon_makernotes.AverageBlackLevel = bls/4;
}
fseek (ifp, save1+(0x30f<<1), SEEK_SET); // offset 783 shorts
imgdata.color.canon_makernotes.SpecularWhiteLevel = get2();
break;
}
fseek (ifp, save1, SEEK_SET);
}
fseek(ifp,_pos2,SEEK_SET);
#endif
if (tag == 0x11 && is_raw && !strncmp(make,"NIKON",5)) {
fseek (ifp, get4()+base, SEEK_SET);
parse_tiff_ifd (base);
}
if (tag == 0x14 && type == 7) {
if (len == 2560) {
fseek (ifp, 1248, SEEK_CUR);
goto get2_256;
}
fread (buf, 1, 10, ifp);
if (!strncmp(buf,"NRW ",4)) {
fseek (ifp, strcmp(buf+4,"0100") ? 46:1546, SEEK_CUR);
cam_mul[0] = get4() << 2;
cam_mul[1] = get4() + get4();
cam_mul[2] = get4() << 2;
}
}
if (tag == 0x15 && type == 2 && is_raw)
fread (model, 64, 1, ifp);
if (strstr(make,"PENTAX")) {
if (tag == 0x1b) tag = 0x1018;
if (tag == 0x1c) tag = 0x1017;
}
if (tag == 0x1d)
while ((c = fgetc(ifp)) && c != EOF)
serial = serial*10 + (isdigit(c) ? c - '0' : c % 10);
if (tag == 0x29 && type == 1) { // Canon PowerShot G9
c = wbi < 18 ? "012347800000005896"[wbi]-'0' : 0;
fseek (ifp, 8 + c*32, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1) ^ 1] = get4();
}
#ifndef LIBRAW_LIBRARY_BUILD
// works for some files, but not all
if (tag == 0x3d && type == 3 && len == 4)
FORC4 cblack[c ^ c >> 1] = get2() >> (14-tiff_ifd[2].bps);
#endif
if (tag == 0x81 && type == 4) {
data_offset = get4();
fseek (ifp, data_offset + 41, SEEK_SET);
raw_height = get2() * 2;
raw_width = get2();
filters = 0x61616161;
}
if ((tag == 0x81 && type == 7) ||
(tag == 0x100 && type == 7) ||
(tag == 0x280 && type == 1)) {
thumb_offset = ftell(ifp);
thumb_length = len;
}
if (tag == 0x88 && type == 4 && (thumb_offset = get4()))
thumb_offset += base;
if (tag == 0x89 && type == 4)
thumb_length = get4();
if (tag == 0x8c || tag == 0x96)
meta_offset = ftell(ifp);
if (tag == 0x97) {
for (i=0; i < 4; i++)
ver97 = ver97 * 10 + fgetc(ifp)-'0';
switch (ver97) {
case 100:
fseek (ifp, 68, SEEK_CUR);
FORC4 cam_mul[(c >> 1) | ((c & 1) << 1)] = get2();
break;
case 102:
fseek (ifp, 6, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1)] = get2();
break;
case 103:
fseek (ifp, 16, SEEK_CUR);
FORC4 cam_mul[c] = get2();
}
if (ver97 >= 200) {
if (ver97 != 205) fseek (ifp, 280, SEEK_CUR);
fread (buf97, 324, 1, ifp);
}
}
if (tag == 0xa1 && type == 7) {
order = 0x4949;
fseek (ifp, 140, SEEK_CUR);
FORC3 cam_mul[c] = get4();
}
if (tag == 0xa4 && type == 3) {
fseek (ifp, wbi*48, SEEK_CUR);
FORC3 cam_mul[c] = get2();
}
if (tag == 0xa7) { // shutter count
NikonKey = fgetc(ifp)^fgetc(ifp)^fgetc(ifp)^fgetc(ifp);
if ( (unsigned) (ver97-200) < 17) {
ci = xlat[0][serial & 0xff];
cj = xlat[1][NikonKey];
ck = 0x60;
for (i=0; i < 324; i++)
buf97[i] ^= (cj += ci * ck++);
i = "66666>666;6A;:;55"[ver97-200] - '0';
FORC4 cam_mul[c ^ (c >> 1) ^ (i & 1)] =
sget2 (buf97 + (i & -2) + c*2);
}
#ifdef LIBRAW_LIBRARY_BUILD
if ((NikonLensDataVersion > 200) && lenNikonLensData)
{
ci = xlat[0][serial & 0xff];
cj = xlat[1][NikonKey];
ck = 0x60;
for (i = 0; i < lenNikonLensData; i++)
table_buf[i] ^= (cj += ci * ck++);
processNikonLensData(table_buf, lenNikonLensData);
lenNikonLensData = 0;
free(table_buf);
}
if (ver97 == 601) // Coolpix A
{
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
}
#endif
}
if(tag == 0xb001 && type == 3) // Sony ModelID
{
unique_id = get2();
}
if (tag == 0x200 && len == 3)
shot_order = (get4(),get4());
if (tag == 0x200 && len == 4)
FORC4 cblack[c ^ c >> 1] = get2();
if (tag == 0x201 && len == 4)
FORC4 cam_mul[c ^ (c >> 1)] = get2();
if (tag == 0x220 && type == 7)
meta_offset = ftell(ifp);
if (tag == 0x401 && type == 4 && len == 4)
FORC4 cblack[c ^ c >> 1] = get4();
#ifdef LIBRAW_LIBRARY_BUILD
// not corrected for file bitcount, to be patched in open_datastream
if (tag == 0x03d && strstr(make,"NIKON") && len == 4)
{
FORC4 cblack[c ^ c >> 1] = get2();
i = cblack[3];
FORC3 if(i>cblack[c]) i = cblack[c];
FORC4 cblack[c]-=i;
black += i;
}
#endif
if (tag == 0xe01) { /* Nikon Capture Note */
#ifdef LIBRAW_LIBRARY_BUILD
int loopc = 0;
#endif
order = 0x4949;
fseek (ifp, 22, SEEK_CUR);
for (offset=22; offset+22 < len; offset += 22+i) {
#ifdef LIBRAW_LIBRARY_BUILD
if(loopc++>1024)
throw LIBRAW_EXCEPTION_IO_CORRUPT;
#endif
tag = get4();
fseek (ifp, 14, SEEK_CUR);
i = get4()-4;
if (tag == 0x76a43207) flip = get2();
else fseek (ifp, i, SEEK_CUR);
}
}
if (tag == 0xe80 && len == 256 && type == 7) {
fseek (ifp, 48, SEEK_CUR);
cam_mul[0] = get2() * 508 * 1.078 / 0x10000;
cam_mul[2] = get2() * 382 * 1.173 / 0x10000;
}
if (tag == 0xf00 && type == 7) {
if (len == 614)
fseek (ifp, 176, SEEK_CUR);
else if (len == 734 || len == 1502)
fseek (ifp, 148, SEEK_CUR);
else goto next;
goto get2_256;
}
if ((tag == 0x1011 && len == 9) || tag == 0x20400200)
{
if(!strncasecmp(make,"Olympus", 7))
{
int j,k;
for (i=0; i < 3; i++)
FORC3 adobe_cam[i][c] = ((short) get2()) / 256.0;
for (i=0; i < 3; i++)
for (j=0; j < 3; j++)
for (cmatrix[i][j] = k=0; k < 3; k++)
cmatrix[i][j] += rgb_adobe[i][k] * adobe_cam[k][j];
}
else
for (i=0; i < 3; i++)
FORC3 cmatrix[i][c] = ((short) get2()) / 256.0;
}
if ((tag == 0x1012 || tag == 0x20400600) && len == 4)
FORC4 cblack[c ^ c >> 1] = get2();
if (tag == 0x1017 || tag == 0x20400100)
cam_mul[0] = get2() / 256.0;
if (tag == 0x1018 || tag == 0x20400100)
cam_mul[2] = get2() / 256.0;
if (tag == 0x2011 && len == 2) {
get2_256:
order = 0x4d4d;
cam_mul[0] = get2() / 256.0;
cam_mul[2] = get2() / 256.0;
}
if ((tag | 0x70) == 0x2070 && (type == 4 || type == 13))
fseek (ifp, get4()+base, SEEK_SET);
if ((tag == 0x2020) && ((type == 7) || (type == 13)))
parse_thumb_note (base, 257, 258);
if (tag == 0x2040)
parse_makernote (base, 0x2040);
#ifdef LIBRAW_LIBRARY_BUILD
// IB start
if (tag == 0x2010)
{
INT64 _pos3 = ftell(ifp);
parse_makernote(base, 0x2010);
fseek(ifp,_pos3,SEEK_SET);
}
if ((tag == 0x3000) && !strncasecmp(make,"Olympus",7))
{
INT64 _pos3 = ftell(ifp);
parse_makernote(base, 0x3000);
fseek(ifp,_pos3,SEEK_SET);
}
// IB end
#endif
if (tag == 0xb028) {
fseek (ifp, get4()+base, SEEK_SET);
parse_thumb_note (base, 136, 137);
}
if (tag == 0x4001 && len > 500 && len < 100000) {
i = len == 582 ? 50 : len == 653 ? 68 : len == 5120 ? 142 : 126;
fseek (ifp, i, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1)] = get2();
for (i+=18; i <= len; i+=10) {
get2();
FORC4 sraw_mul[c ^ (c >> 1)] = get2();
if (sraw_mul[1] == 1170) break;
}
}
if(!strncasecmp(make,"Samsung",7))
{
if (tag == 0xa020) // get the full Samsung encryption key
for (i=0; i<11; i++) SamsungKey[i] = get4();
if (tag == 0xa021) // get and decode Samsung cam_mul array
FORC4 cam_mul[c ^ (c >> 1)] = get4() - SamsungKey[c];
#ifdef LIBRAW_LIBRARY_BUILD
if (tag == 0xa023)
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][0] = get4() - SamsungKey[8];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][1] = get4() - SamsungKey[9];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][3] = get4() - SamsungKey[10];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][2] = get4() - SamsungKey[0];
if (imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][0] < (imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][1]>>1))
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][1] >> 4;
imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Ill_A][3] >> 4;
}
}
if (tag == 0xa024)
{
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][c ^ (c >> 1)] = get4() - SamsungKey[c+1];
if (imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][0] < (imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][1]>>1))
{
imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][1] >> 4;
imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_D65][3] >> 4;
}
}
#endif
if (tag == 0xa030 && len == 9) // get and decode Samsung color matrix
for (i=0; i < 3; i++)
FORC3 cmatrix[i][c] = (short)((get4() + SamsungKey[i*3+c]))/256.0;
if (tag == 0xa028)
FORC4 cblack[c ^ (c >> 1)] = get4() - SamsungKey[c];
}
else
{
// Somebody else use 0xa021 and 0xa028?
if (tag == 0xa021)
FORC4 cam_mul[c ^ (c >> 1)] = get4();
if (tag == 0xa028)
FORC4 cam_mul[c ^ (c >> 1)] -= get4();
}
if (tag == 0x4021 && get4() && get4())
FORC4 cam_mul[c] = 1024;
next:
fseek (ifp, save, SEEK_SET);
}
quit:
order = sorder;
}
/*
Since the TIFF DateTime string has no timezone information,
assume that the camera's clock was set to Universal Time.
*/
void CLASS get_timestamp (int reversed)
{
struct tm t;
char str[20];
int i;
str[19] = 0;
if (reversed)
for (i=19; i--; ) str[i] = fgetc(ifp);
else
fread (str, 19, 1, ifp);
memset (&t, 0, sizeof t);
if (sscanf (str, "%d:%d:%d %d:%d:%d", &t.tm_year, &t.tm_mon,
&t.tm_mday, &t.tm_hour, &t.tm_min, &t.tm_sec) != 6)
return;
t.tm_year -= 1900;
t.tm_mon -= 1;
t.tm_isdst = -1;
if (mktime(&t) > 0)
timestamp = mktime(&t);
}
void CLASS parse_exif (int base)
{
unsigned kodak, entries, tag, type, len, save, c;
double expo,ape;
kodak = !strncmp(make,"EASTMAN",7) && tiff_nifds < 3;
entries = get2();
if(!strncmp(make,"Hasselblad",10) && (tiff_nifds > 3) && (entries > 512)) return;
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
#ifdef LIBRAW_LIBRARY_BUILD
if(callbacks.exif_cb)
{
int savepos = ftell(ifp);
callbacks.exif_cb(callbacks.exifparser_data,tag,type,len,order,ifp);
fseek(ifp,savepos,SEEK_SET);
}
#endif
switch (tag) {
#ifdef LIBRAW_LIBRARY_BUILD
case 0xa405: // FocalLengthIn35mmFormat
imgdata.lens.FocalLengthIn35mmFormat = get2();
break;
case 0xa432: // LensInfo, 42034dec, Lens Specification per EXIF standard
imgdata.lens.MinFocal = getreal(type);
imgdata.lens.MaxFocal = getreal(type);
imgdata.lens.MaxAp4MinFocal = getreal(type);
imgdata.lens.MaxAp4MaxFocal = getreal(type);
break;
case 0xc630: // DNG LensInfo, Lens Specification per EXIF standard
imgdata.lens.dng.MinFocal = getreal(type);
imgdata.lens.dng.MaxFocal = getreal(type);
imgdata.lens.dng.MaxAp4MinFocal = getreal(type);
imgdata.lens.dng.MaxAp4MaxFocal = getreal(type);
break;
case 0xa433: // LensMake
fread(imgdata.lens.LensMake, MIN(len,sizeof(imgdata.lens.LensMake)), 1, ifp);
break;
case 0xa434: // LensModel
fread(imgdata.lens.Lens, MIN(len, sizeof(imgdata.lens.LensMake)), 1, ifp);
if (!strncmp(imgdata.lens.Lens, "----", 4))
imgdata.lens.Lens[0] = 0;
break;
case 0x9205:
imgdata.lens.EXIF_MaxAp = powf64(2.0f, (getreal(type) / 2.0f));
break;
#endif
case 33434: tiff_ifd[tiff_nifds-1].t_shutter =
shutter = getreal(type); break;
case 33437: aperture = getreal(type); break; // 0x829d FNumber
case 34855: iso_speed = get2(); break;
case 34866:
if (iso_speed == 0xffff && (!strncasecmp(make, "SONY",4) || !strncasecmp(make, "CANON",5)))
iso_speed = getreal(type);
break;
case 36867:
case 36868: get_timestamp(0); break;
case 37377: if ((expo = -getreal(type)) < 128 && shutter == 0.)
tiff_ifd[tiff_nifds-1].t_shutter =
shutter = powf64(2.0, expo);
break;
case 37378: // 0x9202 ApertureValue
if ((fabs(ape = getreal(type))<256.0) && (!aperture))
aperture = powf64(2.0, ape/2);
break;
case 37385: flash_used = getreal(type); break;
case 37386: focal_len = getreal(type); break;
case 37500: parse_makernote (base, 0); break; // tag 0x927c
case 40962: if (kodak) raw_width = get4(); break;
case 40963: if (kodak) raw_height = get4(); break;
case 41730:
if (get4() == 0x20002)
for (exif_cfa=c=0; c < 8; c+=2)
exif_cfa |= fgetc(ifp) * 0x01010101 << c;
}
fseek (ifp, save, SEEK_SET);
}
}
#ifdef LIBRAW_LIBRARY_BUILD
void CLASS parse_gps_libraw(int base)
{
unsigned entries, tag, type, len, save, c;
entries = get2();
if (entries > 200)
return;
if (entries > 0)
imgdata.other.parsed_gps.gpsparsed = 1;
while (entries--) {
tiff_get(base, &tag, &type, &len, &save);
switch (tag) {
case 1: imgdata.other.parsed_gps.latref = getc(ifp); break;
case 3: imgdata.other.parsed_gps.longref = getc(ifp); break;
case 5: imgdata.other.parsed_gps.altref = getc(ifp); break;
case 2:
if (len == 3)
FORC(3) imgdata.other.parsed_gps.latitude[c] = getreal(type);
break;
case 4:
if (len == 3)
FORC(3) imgdata.other.parsed_gps.longtitude[c] = getreal(type);
break;
case 7:
if (len == 3)
FORC(3) imgdata.other.parsed_gps.gpstimestamp[c] = getreal(type);
break;
case 6:
imgdata.other.parsed_gps.altitude = getreal(type);
break;
case 9: imgdata.other.parsed_gps.gpsstatus = getc(ifp); break;
}
fseek(ifp, save, SEEK_SET);
}
}
#endif
void CLASS parse_gps (int base)
{
unsigned entries, tag, type, len, save, c;
entries = get2();
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
switch (tag) {
case 1: case 3: case 5:
gpsdata[29+tag/2] = getc(ifp); break;
case 2: case 4: case 7:
FORC(6) gpsdata[tag/3*6+c] = get4(); break;
case 6:
FORC(2) gpsdata[18+c] = get4(); break;
case 18: case 29:
fgets ((char *) (gpsdata+14+tag/3), MIN(len,12), ifp);
}
fseek (ifp, save, SEEK_SET);
}
}
void CLASS romm_coeff (float romm_cam[3][3])
{
static const float rgb_romm[3][3] = /* ROMM == Kodak ProPhoto */
{ { 2.034193, -0.727420, -0.306766 },
{ -0.228811, 1.231729, -0.002922 },
{ -0.008565, -0.153273, 1.161839 } };
int i, j, k;
for (i=0; i < 3; i++)
for (j=0; j < 3; j++)
for (cmatrix[i][j] = k=0; k < 3; k++)
cmatrix[i][j] += rgb_romm[i][k] * romm_cam[k][j];
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.color.digitalBack_color=1;
#endif
}
void CLASS parse_mos (int offset)
{
char data[40];
int skip, from, i, c, neut[4], planes=0, frot=0;
static const char *mod[] =
{ "","DCB2","Volare","Cantare","CMost","Valeo 6","Valeo 11","Valeo 22",
"Valeo 11p","Valeo 17","","Aptus 17","Aptus 22","Aptus 75","Aptus 65",
"Aptus 54S","Aptus 65S","Aptus 75S","AFi 5","AFi 6","AFi 7",
"AFi-II 7","Aptus-II 7","","Aptus-II 6","","","Aptus-II 10","Aptus-II 5",
"","","","","Aptus-II 10R","Aptus-II 8","","Aptus-II 12","","AFi-II 12" };
float romm_cam[3][3];
fseek (ifp, offset, SEEK_SET);
while (1) {
if (get4() != 0x504b5453) break;
get4();
fread (data, 1, 40, ifp);
skip = get4();
from = ftell(ifp);
// IB start
#ifdef LIBRAW_LIBRARY_BUILD
if (!strcmp(data,"CameraObj_camera_type")) {
fread(imgdata.lens.makernotes.body, MIN(skip,63), 1, ifp);
}
#endif
// IB end
if (!strcmp(data,"JPEG_preview_data")) {
thumb_offset = from;
thumb_length = skip;
}
if (!strcmp(data,"icc_camera_profile")) {
profile_offset = from;
profile_length = skip;
}
if (!strcmp(data,"ShootObj_back_type")) {
fscanf (ifp, "%d", &i);
if ((unsigned) i < sizeof mod / sizeof (*mod))
strcpy (model, mod[i]);
}
if (!strcmp(data,"icc_camera_to_tone_matrix")) {
for (i=0; i < 9; i++)
((float *)romm_cam)[i] = int_to_float(get4());
romm_coeff (romm_cam);
}
if (!strcmp(data,"CaptProf_color_matrix")) {
for (i=0; i < 9; i++)
fscanf (ifp, "%f", (float *)romm_cam + i);
romm_coeff (romm_cam);
}
if (!strcmp(data,"CaptProf_number_of_planes"))
fscanf (ifp, "%d", &planes);
if (!strcmp(data,"CaptProf_raw_data_rotation"))
fscanf (ifp, "%d", &flip);
if (!strcmp(data,"CaptProf_mosaic_pattern"))
FORC4 {
fscanf (ifp, "%d", &i);
if (i == 1) frot = c ^ (c >> 1);
}
if (!strcmp(data,"ImgProf_rotation_angle")) {
fscanf (ifp, "%d", &i);
flip = i - flip;
}
if (!strcmp(data,"NeutObj_neutrals") && !cam_mul[0]) {
FORC4 fscanf (ifp, "%d", neut+c);
FORC3 cam_mul[c] = (float) neut[0] / neut[c+1];
}
if (!strcmp(data,"Rows_data"))
load_flags = get4();
parse_mos (from);
fseek (ifp, skip+from, SEEK_SET);
}
if (planes)
filters = (planes == 1) * 0x01010101 *
(uchar) "\x94\x61\x16\x49"[(flip/90 + frot) & 3];
}
void CLASS linear_table (unsigned len)
{
int i;
if (len > 0x10000) len = 0x10000;
read_shorts (curve, len);
for (i=len; i < 0x10000; i++)
curve[i] = curve[i-1];
maximum = curve[len<0x1000?0xfff:len-1];
}
#ifdef LIBRAW_LIBRARY_BUILD
void CLASS Kodak_WB_0x08tags (int wb, unsigned type)
{
float mul[3]={1,1,1}, num, mul2;
int c;
FORC3 mul[c] = (num=getreal(type))==0 ? 1 : num;
imgdata.color.WB_Coeffs[wb][1] = imgdata.color.WB_Coeffs[wb][3] = mul[1];
mul2 = mul[1] * mul[1];
imgdata.color.WB_Coeffs[wb][0] = mul2 / mul[0];
imgdata.color.WB_Coeffs[wb][2] = mul2 / mul[2];
return;
}
/* Thanks to Alexey Danilchenko for wb as-shot parsing code */
void CLASS parse_kodak_ifd (int base)
{
unsigned entries, tag, type, len, save;
int i, c, wbi=-2;
float mul[3]={1,1,1}, num;
static const int wbtag[] = { 64037,64040,64039,64041,-1,-1,64042 };
entries = get2();
if (entries > 1024) return;
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
#ifdef LIBRAW_LIBRARY_BUILD
if(callbacks.exif_cb)
{
int savepos = ftell(ifp);
callbacks.exif_cb(callbacks.exifparser_data,tag | 0x20000,type,len,order,ifp);
fseek(ifp,savepos,SEEK_SET);
}
#endif
if (tag == 1020) wbi = getint(type);
if (tag == 1021 && len == 72) { /* WB set in software */
fseek (ifp, 40, SEEK_CUR);
FORC3 cam_mul[c] = 2048.0 / get2();
wbi = -2;
}
if (tag == 0x0848) Kodak_WB_0x08tags(LIBRAW_WBI_Daylight, type);
if (tag == 0x0849) Kodak_WB_0x08tags(LIBRAW_WBI_Tungsten, type);
if (tag == 0x084a) Kodak_WB_0x08tags(LIBRAW_WBI_Fluorescent, type);
if (tag == 0x084b) Kodak_WB_0x08tags(LIBRAW_WBI_Flash, type);
if (tag == 0xfa27)
{
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c] = get4();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][1];
}
if (tag == 0xfa28)
{
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][c] = get4();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][1];
}
if (tag == 0xfa29)
{
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c] = get4();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][1];
}
if (tag == 0xfa2a)
{
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c] = get4();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][1];
}
if (tag == 2120 + wbi ||
(wbi<0 && tag == 2125)) /* use Auto WB if illuminant index is not set */
{
FORC3 mul[c] = (num=getreal(type))==0 ? 1 : num;
FORC3 cam_mul[c] = mul[1] / mul[c]; /* normalise against green */
}
if (tag == 2317) linear_table (len);
if (tag == 0x903) iso_speed = getreal(type);
//if (tag == 6020) iso_speed = getint(type);
if (tag == 64013) wbi = fgetc(ifp);
if ((unsigned) wbi < 7 && tag == wbtag[wbi])
FORC3 cam_mul[c] = get4();
if (tag == 64019) width = getint(type);
if (tag == 64020) height = (getint(type)+1) & -2;
fseek (ifp, save, SEEK_SET);
}
}
#else
void CLASS parse_kodak_ifd (int base)
{
unsigned entries, tag, type, len, save;
int i, c, wbi=-2, wbtemp=6500;
float mul[3]={1,1,1}, num;
static const int wbtag[] = { 64037,64040,64039,64041,-1,-1,64042 };
entries = get2();
if (entries > 1024) return;
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
if (tag == 1020) wbi = getint(type);
if (tag == 1021 && len == 72) { /* WB set in software */
fseek (ifp, 40, SEEK_CUR);
FORC3 cam_mul[c] = 2048.0 / get2();
wbi = -2;
}
if (tag == 2118) wbtemp = getint(type);
if (tag == 2120 + wbi && wbi >= 0)
FORC3 cam_mul[c] = 2048.0 / getreal(type);
if (tag == 2130 + wbi)
FORC3 mul[c] = getreal(type);
if (tag == 2140 + wbi && wbi >= 0)
FORC3 {
for (num=i=0; i < 4; i++)
num += getreal(type) * pow (wbtemp/100.0, i);
cam_mul[c] = 2048 / (num * mul[c]);
}
if (tag == 2317) linear_table (len);
if (tag == 6020) iso_speed = getint(type);
if (tag == 64013) wbi = fgetc(ifp);
if ((unsigned) wbi < 7 && tag == wbtag[wbi])
FORC3 cam_mul[c] = get4();
if (tag == 64019) width = getint(type);
if (tag == 64020) height = (getint(type)+1) & -2;
fseek (ifp, save, SEEK_SET);
}
}
#endif
//@end COMMON
void CLASS parse_minolta (int base);
int CLASS parse_tiff (int base);
//@out COMMON
int CLASS parse_tiff_ifd (int base)
{
unsigned entries, tag, type, len, plen=16, save;
int ifd, use_cm=0, cfa, i, j, c, ima_len=0;
char *cbuf, *cp;
uchar cfa_pat[16], cfa_pc[] = { 0,1,2,3 }, tab[256];
double fm[3][4], cc[4][4], cm[4][3], cam_xyz[4][3], num;
double ab[]={ 1,1,1,1 }, asn[] = { 0,0,0,0 }, xyz[] = { 1,1,1 };
unsigned sony_curve[] = { 0,0,0,0,0,4095 };
unsigned *buf, sony_offset=0, sony_length=0, sony_key=0;
struct jhead jh;
int pana_raw = 0;
#ifndef LIBRAW_LIBRARY_BUILD
FILE *sfp;
#endif
if (tiff_nifds >= sizeof tiff_ifd / sizeof tiff_ifd[0])
return 1;
ifd = tiff_nifds++;
for (j=0; j < 4; j++)
for (i=0; i < 4; i++)
cc[j][i] = i == j;
entries = get2();
if (entries > 512) return 1;
while (entries--) {
tiff_get (base, &tag, &type, &len, &save);
#ifdef LIBRAW_LIBRARY_BUILD
if(callbacks.exif_cb)
{
int savepos = ftell(ifp);
callbacks.exif_cb(callbacks.exifparser_data,tag|(pana_raw?0x30000:0),type,len,order,ifp);
fseek(ifp,savepos,SEEK_SET);
}
#endif
#ifdef LIBRAW_LIBRARY_BUILD
if (!strncasecmp(make, "SONY", 4) ||
(!strncasecmp(make, "Hasselblad", 10) &&
(!strncasecmp(model, "Stellar", 7) ||
!strncasecmp(model, "Lunar", 5) ||
!strncasecmp(model, "HV",2))))
{
switch (tag) {
case 0x7480:
case 0x7820:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][1];
break;
case 0x7481:
case 0x7821:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][1];
break;
case 0x7482:
case 0x7822:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][1];
break;
case 0x7483:
case 0x7823:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][1];
break;
case 0x7484:
case 0x7824:
imgdata.color.WBCT_Coeffs[0][0] = 4500;
FORC3 imgdata.color.WBCT_Coeffs[0][c+1] = get2();
imgdata.color.WBCT_Coeffs[0][4] = imgdata.color.WBCT_Coeffs[0][2];
break;
case 0x7486:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Fluorescent][1];
break;
case 0x7825:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][1];
break;
case 0x7826:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][1];
break;
case 0x7827:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][1];
break;
case 0x7828:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][1];
break;
case 0x7829:
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][c] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][3] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][1];
break;
case 0x782a:
imgdata.color.WBCT_Coeffs[1][0] = 8500;
FORC3 imgdata.color.WBCT_Coeffs[1][c+1] = get2();
imgdata.color.WBCT_Coeffs[1][4] = imgdata.color.WBCT_Coeffs[1][2];
break;
case 0x782b:
imgdata.color.WBCT_Coeffs[2][0] = 6000;
FORC3 imgdata.color.WBCT_Coeffs[2][c+1] = get2();
imgdata.color.WBCT_Coeffs[2][4] = imgdata.color.WBCT_Coeffs[2][2];
break;
case 0x782c:
imgdata.color.WBCT_Coeffs[3][0] = 3200;
FORC3 imgdata.color.WB_Coeffs[LIBRAW_WBI_StudioTungsten][c] = imgdata.color.WBCT_Coeffs[3][c+1] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_StudioTungsten][3] = imgdata.color.WBCT_Coeffs[3][4] = imgdata.color.WB_Coeffs[LIBRAW_WBI_StudioTungsten][1];
break;
case 0x782d:
imgdata.color.WBCT_Coeffs[4][0] = 2500;
FORC3 imgdata.color.WBCT_Coeffs[4][c+1] = get2();
imgdata.color.WBCT_Coeffs[4][4] = imgdata.color.WBCT_Coeffs[4][2];
break;
}
}
#endif
switch (tag) {
case 1: if(len==4) pana_raw = get4(); break;
case 5: width = get2(); break;
case 6: height = get2(); break;
case 7: width += get2(); break;
case 9: if ((i = get2())) filters = i;
#ifdef LIBRAW_LIBRARY_BUILD
if(pana_raw && len == 1 && type ==3)
pana_black[3]+=i;
#endif
break;
case 8:
case 10:
#ifdef LIBRAW_LIBRARY_BUILD
if(pana_raw && len == 1 && type ==3)
pana_black[3]+=get2();
#endif
break;
case 17: case 18:
if (type == 3 && len == 1)
cam_mul[(tag-17)*2] = get2() / 256.0;
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 19:
if(pana_raw)
{
ushort nWB, cnt, tWB;
nWB = get2();
if (nWB > 0x100) break;
for (cnt=0; cnt<nWB; cnt++)
{
tWB = get2();
if (tWB < 0x100)
{
imgdata.color.WB_Coeffs[tWB][0] = get2();
imgdata.color.WB_Coeffs[tWB][2] = get2();
imgdata.color.WB_Coeffs[tWB][1] = imgdata.color.WB_Coeffs[tWB][3] = 0x100;
} else get4();
}
}
break;
#endif
case 23:
if (type == 3) iso_speed = get2();
break;
case 28: case 29: case 30:
#ifdef LIBRAW_LIBRARY_BUILD
if(pana_raw && len == 1 && type ==3)
{
pana_black[tag-28] = get2();
}
else
#endif
{
cblack[tag-28] = get2();
cblack[3] = cblack[1];
}
break;
case 36: case 37: case 38:
cam_mul[tag-36] = get2();
break;
case 39:
#ifdef LIBRAW_LIBRARY_BUILD
if(pana_raw)
{
ushort nWB, cnt, tWB;
nWB = get2();
if (nWB > 0x100) break;
for (cnt=0; cnt<nWB; cnt++)
{
tWB = get2();
if (tWB < 0x100)
{
imgdata.color.WB_Coeffs[tWB][0] = get2();
imgdata.color.WB_Coeffs[tWB][1] = imgdata.color.WB_Coeffs[tWB][3] = get2();
imgdata.color.WB_Coeffs[tWB][2] = get2();
} else fseek(ifp, 6, SEEK_CUR);
}
}
break;
#endif
if (len < 50 || cam_mul[0]) break;
fseek (ifp, 12, SEEK_CUR);
FORC3 cam_mul[c] = get2();
break;
case 46:
if (type != 7 || fgetc(ifp) != 0xff || fgetc(ifp) != 0xd8) break;
thumb_offset = ftell(ifp) - 2;
thumb_length = len;
break;
case 61440: /* Fuji HS10 table */
fseek (ifp, get4()+base, SEEK_SET);
parse_tiff_ifd (base);
break;
case 2: case 256: case 61441: /* ImageWidth */
tiff_ifd[ifd].t_width = getint(type);
break;
case 3: case 257: case 61442: /* ImageHeight */
tiff_ifd[ifd].t_height = getint(type);
break;
case 258: /* BitsPerSample */
case 61443:
tiff_ifd[ifd].samples = len & 7;
tiff_ifd[ifd].bps = getint(type);
break;
case 61446:
raw_height = 0;
if (tiff_ifd[ifd].bps > 12) break;
load_raw = &CLASS packed_load_raw;
load_flags = get4() ? 24:80;
break;
case 259: /* Compression */
tiff_ifd[ifd].comp = getint(type);
break;
case 262: /* PhotometricInterpretation */
tiff_ifd[ifd].phint = get2();
break;
case 270: /* ImageDescription */
fread (desc, 512, 1, ifp);
break;
case 271: /* Make */
fgets (make, 64, ifp);
break;
case 272: /* Model */
fgets (model, 64, ifp);
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 278:
tiff_ifd[ifd].rows_per_strip = getint(type);
break;
#endif
case 280: /* Panasonic RW2 offset */
if (type != 4) break;
load_raw = &CLASS panasonic_load_raw;
load_flags = 0x2008;
case 273: /* StripOffset */
#ifdef LIBRAW_LIBRARY_BUILD
if(len > 1 && len < 16384)
{
off_t sav = ftell(ifp);
tiff_ifd[ifd].strip_offsets = (int*)calloc(len,sizeof(int));
tiff_ifd[ifd].strip_offsets_count = len;
for(int i=0; i< len; i++)
tiff_ifd[ifd].strip_offsets[i]=get4()+base;
fseek(ifp,sav,SEEK_SET); // restore position
}
/* fallback */
#endif
case 513: /* JpegIFOffset */
case 61447:
tiff_ifd[ifd].offset = get4()+base;
if (!tiff_ifd[ifd].bps && tiff_ifd[ifd].offset > 0) {
fseek (ifp, tiff_ifd[ifd].offset, SEEK_SET);
if (ljpeg_start (&jh, 1)) {
tiff_ifd[ifd].comp = 6;
tiff_ifd[ifd].t_width = jh.wide;
tiff_ifd[ifd].t_height = jh.high;
tiff_ifd[ifd].bps = jh.bits;
tiff_ifd[ifd].samples = jh.clrs;
if (!(jh.sraw || (jh.clrs & 1)))
tiff_ifd[ifd].t_width *= jh.clrs;
if ((tiff_ifd[ifd].t_width > 4*tiff_ifd[ifd].t_height) & ~jh.clrs) {
tiff_ifd[ifd].t_width /= 2;
tiff_ifd[ifd].t_height *= 2;
}
i = order;
parse_tiff (tiff_ifd[ifd].offset + 12);
order = i;
}
}
break;
case 274: /* Orientation */
tiff_ifd[ifd].t_flip = "50132467"[get2() & 7]-'0';
break;
case 277: /* SamplesPerPixel */
tiff_ifd[ifd].samples = getint(type) & 7;
break;
case 279: /* StripByteCounts */
#ifdef LIBRAW_LIBRARY_BUILD
if(len > 1 && len < 16384)
{
off_t sav = ftell(ifp);
tiff_ifd[ifd].strip_byte_counts = (int*)calloc(len,sizeof(int));
tiff_ifd[ifd].strip_byte_counts_count = len;
for(int i=0; i< len; i++)
tiff_ifd[ifd].strip_byte_counts[i]=get4();
fseek(ifp,sav,SEEK_SET); // restore position
}
/* fallback */
#endif
case 514:
case 61448:
tiff_ifd[ifd].bytes = get4();
break;
case 61454:
FORC3 cam_mul[(4-c) % 3] = getint(type);
break;
case 305: case 11: /* Software */
fgets (software, 64, ifp);
if (!strncmp(software,"Adobe",5) ||
!strncmp(software,"dcraw",5) ||
!strncmp(software,"UFRaw",5) ||
!strncmp(software,"Bibble",6) ||
!strcmp (software,"Digital Photo Professional"))
is_raw = 0;
break;
case 306: /* DateTime */
get_timestamp(0);
break;
case 315: /* Artist */
fread (artist, 64, 1, ifp);
break;
case 317:
tiff_ifd[ifd].predictor = getint(type);
break;
case 322: /* TileWidth */
tiff_ifd[ifd].t_tile_width = getint(type);
break;
case 323: /* TileLength */
tiff_ifd[ifd].t_tile_length = getint(type);
break;
case 324: /* TileOffsets */
tiff_ifd[ifd].offset = len > 1 ? ftell(ifp) : get4();
if (len == 1)
tiff_ifd[ifd].t_tile_width = tiff_ifd[ifd].t_tile_length = 0;
if (len == 4) {
load_raw = &CLASS sinar_4shot_load_raw;
is_raw = 5;
}
break;
case 325:
tiff_ifd[ifd].bytes = len > 1 ? ftell(ifp): get4();
break;
case 330: /* SubIFDs */
if (!strcmp(model,"DSLR-A100") && tiff_ifd[ifd].t_width == 3872) {
load_raw = &CLASS sony_arw_load_raw;
data_offset = get4()+base;
ifd++; break;
}
#ifdef LIBRAW_LIBRARY_BUILD
if (!strncmp(make,"Hasselblad",10) && libraw_internal_data.unpacker_data.hasselblad_parser_flag) {
fseek (ifp, ftell(ifp)+4, SEEK_SET);
fseek (ifp, get4()+base, SEEK_SET);
parse_tiff_ifd (base);
break;
}
#endif
if(len > 1000) len=1000; /* 1000 SubIFDs is enough */
while (len--) {
i = ftell(ifp);
fseek (ifp, get4()+base, SEEK_SET);
if (parse_tiff_ifd (base)) break;
fseek (ifp, i+4, SEEK_SET);
}
break;
case 339:
tiff_ifd[ifd].sample_format = getint(type);
break;
case 400:
strcpy (make, "Sarnoff");
maximum = 0xfff;
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 700:
if((type == 1 || type == 2 || type == 6 || type == 7) && len > 1 && len < 5100000)
{
xmpdata = (char*)malloc(xmplen = len+1);
fread(xmpdata,len,1,ifp);
xmpdata[len]=0;
}
break;
#endif
case 28688:
FORC4 sony_curve[c+1] = get2() >> 2 & 0xfff;
for (i=0; i < 5; i++)
for (j = sony_curve[i]+1; j <= sony_curve[i+1]; j++)
curve[j] = curve[j-1] + (1 << i);
break;
case 29184: sony_offset = get4(); break;
case 29185: sony_length = get4(); break;
case 29217: sony_key = get4(); break;
case 29264:
parse_minolta (ftell(ifp));
raw_width = 0;
break;
case 29443:
FORC4 cam_mul[c ^ (c < 2)] = get2();
break;
case 29459:
FORC4 cam_mul[c] = get2();
i = (cam_mul[1] == 1024 && cam_mul[2] == 1024) << 1;
SWAP (cam_mul[i],cam_mul[i+1])
break;
case 30720: // Sony matrix, Sony_SR2SubIFD_0x7800
for (i=0; i < 3; i++)
FORC3 cmatrix[i][c] = ((short) get2()) / 1024.0;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr, _(" Sony matrix:\n%f %f %f\n%f %f %f\n%f %f %f\n"), cmatrix[0][0], cmatrix[0][1], cmatrix[0][2], cmatrix[1][0], cmatrix[1][1], cmatrix[1][2], cmatrix[2][0], cmatrix[2][1], cmatrix[2][2]);
#endif
break;
case 29456: // Sony black level, Sony_SR2SubIFD_0x7310, no more needs to be divided by 4
FORC4 cblack[c ^ c >> 1] = get2();
i = cblack[3];
FORC3 if(i>cblack[c]) i = cblack[c];
FORC4 cblack[c]-=i;
black = i;
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr, _("...Sony black: %u cblack: %u %u %u %u\n"),black, cblack[0],cblack[1],cblack[2], cblack[3]);
#endif
break;
case 33405: /* Model2 */
fgets (model2, 64, ifp);
break;
case 33421: /* CFARepeatPatternDim */
if (get2() == 6 && get2() == 6)
filters = 9;
break;
case 33422: /* CFAPattern */
if (filters == 9) {
FORC(36) ((char *)xtrans)[c] = fgetc(ifp) & 3;
break;
}
case 64777: /* Kodak P-series */
if(len == 36)
{
filters = 9;
colors = 3;
FORC(36) xtrans[0][c] = fgetc(ifp) & 3;
}
else if(len > 0)
{
if ((plen=len) > 16) plen = 16;
fread (cfa_pat, 1, plen, ifp);
for (colors=cfa=i=0; i < plen && colors < 4; i++) {
colors += !(cfa & (1 << cfa_pat[i]));
cfa |= 1 << cfa_pat[i];
}
if (cfa == 070) memcpy (cfa_pc,"\003\004\005",3); /* CMY */
if (cfa == 072) memcpy (cfa_pc,"\005\003\004\001",4); /* GMCY */
goto guess_cfa_pc;
}
break;
case 33424:
case 65024:
fseek (ifp, get4()+base, SEEK_SET);
parse_kodak_ifd (base);
break;
case 33434: /* ExposureTime */
tiff_ifd[ifd].t_shutter = shutter = getreal(type);
break;
case 33437: /* FNumber */
aperture = getreal(type);
break;
#ifdef LIBRAW_LIBRARY_BUILD
// IB start
case 0xa405: // FocalLengthIn35mmFormat
imgdata.lens.FocalLengthIn35mmFormat = get2();
break;
case 0xa432: // LensInfo, 42034dec, Lens Specification per EXIF standard
imgdata.lens.MinFocal = getreal(type);
imgdata.lens.MaxFocal = getreal(type);
imgdata.lens.MaxAp4MinFocal = getreal(type);
imgdata.lens.MaxAp4MaxFocal = getreal(type);
break;
case 0xc630: // DNG LensInfo, Lens Specification per EXIF standard
imgdata.lens.MinFocal = getreal(type);
imgdata.lens.MaxFocal = getreal(type);
imgdata.lens.MaxAp4MinFocal = getreal(type);
imgdata.lens.MaxAp4MaxFocal = getreal(type);
break;
case 0xa433: // LensMake
fread(imgdata.lens.LensMake, MIN(len, sizeof(imgdata.lens.LensMake)), 1, ifp);
break;
case 0xa434: // LensModel
fread(imgdata.lens.Lens, MIN(len, sizeof(imgdata.lens.Lens)), 1, ifp);
if (!strncmp(imgdata.lens.Lens, "----", 4))
imgdata.lens.Lens[0] = 0;
break;
case 0x9205:
imgdata.lens.EXIF_MaxAp = powf64(2.0f, (getreal(type) / 2.0f));
break;
// IB end
#endif
case 34306: /* Leaf white balance */
FORC4 cam_mul[c ^ 1] = 4096.0 / get2();
break;
case 34307: /* Leaf CatchLight color matrix */
fread (software, 1, 7, ifp);
if (strncmp(software,"MATRIX",6)) break;
colors = 4;
for (raw_color = i=0; i < 3; i++) {
FORC4 fscanf (ifp, "%f", &rgb_cam[i][c^1]);
if (!use_camera_wb) continue;
num = 0;
FORC4 num += rgb_cam[i][c];
FORC4 rgb_cam[i][c] /= num;
}
break;
case 34310: /* Leaf metadata */
parse_mos (ftell(ifp));
case 34303:
strcpy (make, "Leaf");
break;
case 34665: /* EXIF tag */
fseek (ifp, get4()+base, SEEK_SET);
parse_exif (base);
break;
case 34853: /* GPSInfo tag */
{
unsigned pos;
fseek(ifp, pos = (get4() + base), SEEK_SET);
parse_gps(base);
#ifdef LIBRAW_LIBRARY_BUILD
fseek(ifp, pos, SEEK_SET);
parse_gps_libraw(base);
#endif
}
break;
case 34675: /* InterColorProfile */
case 50831: /* AsShotICCProfile */
profile_offset = ftell(ifp);
profile_length = len;
break;
case 37122: /* CompressedBitsPerPixel */
kodak_cbpp = get4();
break;
case 37386: /* FocalLength */
focal_len = getreal(type);
break;
case 37393: /* ImageNumber */
shot_order = getint(type);
break;
case 37400: /* old Kodak KDC tag */
for (raw_color = i=0; i < 3; i++) {
getreal(type);
FORC3 rgb_cam[i][c] = getreal(type);
}
break;
case 40976:
strip_offset = get4();
switch (tiff_ifd[ifd].comp) {
case 32770: load_raw = &CLASS samsung_load_raw; break;
case 32772: load_raw = &CLASS samsung2_load_raw; break;
case 32773: load_raw = &CLASS samsung3_load_raw; break;
}
break;
case 46275: /* Imacon tags */
strcpy (make, "Imacon");
data_offset = ftell(ifp);
ima_len = len;
break;
case 46279:
if (!ima_len) break;
fseek (ifp, 38, SEEK_CUR);
case 46274:
fseek (ifp, 40, SEEK_CUR);
raw_width = get4();
raw_height = get4();
left_margin = get4() & 7;
width = raw_width - left_margin - (get4() & 7);
top_margin = get4() & 7;
height = raw_height - top_margin - (get4() & 7);
if (raw_width == 7262 && ima_len == 234317952 ) {
height = 5412;
width = 7216;
left_margin = 7;
filters=0;
} else if (raw_width == 7262) {
height = 5444;
width = 7244;
left_margin = 7;
}
fseek (ifp, 52, SEEK_CUR);
FORC3 cam_mul[c] = getreal(11);
fseek (ifp, 114, SEEK_CUR);
flip = (get2() >> 7) * 90;
if (width * height * 6 == ima_len) {
if (flip % 180 == 90) SWAP(width,height);
raw_width = width;
raw_height = height;
left_margin = top_margin = filters = flip = 0;
}
sprintf (model, "Ixpress %d-Mp", height*width/1000000);
load_raw = &CLASS imacon_full_load_raw;
if (filters) {
if (left_margin & 1) filters = 0x61616161;
load_raw = &CLASS unpacked_load_raw;
}
maximum = 0xffff;
break;
case 50454: /* Sinar tag */
case 50455:
if (!(cbuf = (char *) malloc(len))) break;
#ifndef LIBRAW_LIBRARY_BUILD
fread (cbuf, 1, len, ifp);
#else
if(fread (cbuf, 1, len, ifp) != len)
throw LIBRAW_EXCEPTION_IO_CORRUPT; // cbuf to be free'ed in recycle
#endif
for (cp = cbuf-1; cp && cp < cbuf+len; cp = strchr(cp,'\n'))
if (!strncmp (++cp,"Neutral ",8))
sscanf (cp+8, "%f %f %f", cam_mul, cam_mul+1, cam_mul+2);
free (cbuf);
break;
case 50458:
if (!make[0]) strcpy (make, "Hasselblad");
break;
case 50459: /* Hasselblad tag */
#ifdef LIBRAW_LIBRARY_BUILD
libraw_internal_data.unpacker_data.hasselblad_parser_flag=1;
#endif
i = order;
j = ftell(ifp);
c = tiff_nifds;
order = get2();
fseek (ifp, j+(get2(),get4()), SEEK_SET);
parse_tiff_ifd (j);
maximum = 0xffff;
tiff_nifds = c;
order = i;
break;
case 50706: /* DNGVersion */
FORC4 dng_version = (dng_version << 8) + fgetc(ifp);
if (!make[0]) strcpy (make, "DNG");
is_raw = 1;
break;
case 50708: /* UniqueCameraModel */
if (model[0]) break;
fgets (make, 64, ifp);
if ((cp = strchr(make,' '))) {
strcpy(model,cp+1);
*cp = 0;
}
break;
case 50710: /* CFAPlaneColor */
if (filters == 9) break;
if (len > 4) len = 4;
colors = len;
fread (cfa_pc, 1, colors, ifp);
guess_cfa_pc:
FORCC tab[cfa_pc[c]] = c;
cdesc[c] = 0;
for (i=16; i--; )
filters = filters << 2 | tab[cfa_pat[i % plen]];
filters -= !filters;
break;
case 50711: /* CFALayout */
if (get2() == 2) fuji_width = 1;
break;
case 291:
case 50712: /* LinearizationTable */
linear_table (len);
break;
case 50713: /* BlackLevelRepeatDim */
cblack[4] = get2();
cblack[5] = get2();
if (cblack[4] * cblack[5] > (sizeof(cblack) / sizeof (cblack[0]) - 6))
cblack[4] = cblack[5] = 1;
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 0xf00c: {
unsigned fwb[4];
FORC4 fwb[c] = get4();
if (fwb[3] < 0x100)
{
imgdata.color.WB_Coeffs[fwb[3]][0] = fwb[1];
imgdata.color.WB_Coeffs[fwb[3]][1] = imgdata.color.WB_Coeffs[fwb[3]][3] = fwb[0];
imgdata.color.WB_Coeffs[fwb[3]][2] = fwb[2];
if ((fwb[3] == 17) && libraw_internal_data.unpacker_data.lenRAFData>3)
{
long long f_save = ftell(ifp);
int fj, found = 0;
ushort *rafdata = (ushort*) malloc (sizeof(ushort)*libraw_internal_data.unpacker_data.lenRAFData);
fseek (ifp, libraw_internal_data.unpacker_data.posRAFData, SEEK_SET);
fread (rafdata, sizeof(ushort), libraw_internal_data.unpacker_data.lenRAFData, ifp);
fseek(ifp, f_save, SEEK_SET);
for (int fi=0; fi<(libraw_internal_data.unpacker_data.lenRAFData-3); fi++)
{
if ((fwb[0]==rafdata[fi]) && (fwb[1]==rafdata[fi+1]) && (fwb[2]==rafdata[fi+2]))
{
if (rafdata[fi-15] != fwb[0]) continue;
fi = fi - 15;
imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][3] = rafdata[fi];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][0] = rafdata[fi+1];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FineWeather][2] = rafdata[fi+2];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][3] = rafdata[fi+3];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][0] = rafdata[fi+4];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][2] = rafdata[fi+5];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][3] = rafdata[fi+6];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][0] = rafdata[fi+7];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][2] = rafdata[fi+8];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][3] = rafdata[fi+9];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][0] = rafdata[fi+10];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][2] = rafdata[fi+11];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][3] = rafdata[fi+12];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][0] = rafdata[fi+13];
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][2] = rafdata[fi+14];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][3] = rafdata[fi+15];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][0] = rafdata[fi+16];
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][2] = rafdata[fi+17];
fi += 111;
for (fj = fi; fj<(fi+15); fj+=3)
if (rafdata[fj] != rafdata[fi])
{
found = 1;
break;
}
if (found)
{
int FujiCCT_K [31] = {2500,2550,2650,2700,2800,2850,2950,3000,3100,3200,3300,3400,3600,3700,3800,4000,4200,4300,4500,4800,5000,5300,5600,5900,6300,6700,7100,7700,8300,9100,10000};
fj = fj - 93;
for (int iCCT=0; iCCT < 31; iCCT++)
{
imgdata.color.WBCT_Coeffs[iCCT][0] = FujiCCT_K[iCCT];
imgdata.color.WBCT_Coeffs[iCCT][1] = rafdata[iCCT*3+1+fj];
imgdata.color.WBCT_Coeffs[iCCT][2] = imgdata.color.WBCT_Coeffs[iCCT][4] = rafdata[iCCT*3+fj];
imgdata.color.WBCT_Coeffs[iCCT][3] = rafdata[iCCT*3+2+fj];
}
}
free (rafdata);
break;
}
}
}
}
FORC4 fwb[c] = get4();
if (fwb[3] < 0x100)
{
imgdata.color.WB_Coeffs[fwb[3]][0] = fwb[1];
imgdata.color.WB_Coeffs[fwb[3]][1] = imgdata.color.WB_Coeffs[fwb[3]][3] = fwb[0];
imgdata.color.WB_Coeffs[fwb[3]][2] = fwb[2];
}
}
break;
#endif
case 61450:
cblack[4] = cblack[5] = MIN(sqrt((double)len),64);
case 50714: /* BlackLevel */
if((cblack[4] * cblack[5] < 2) && len == 1)
{
black = getreal(type);
}
else if(cblack[4] * cblack[5] <= len)
{
FORC (cblack[4] * cblack[5])
cblack[6+c] = getreal(type);
black = 0;
}
break;
case 50715: /* BlackLevelDeltaH */
case 50716: /* BlackLevelDeltaV */
for (num=i=0; i < len && i < 65536; i++)
num += getreal(type);
black += num/len + 0.5;
break;
case 50717: /* WhiteLevel */
maximum = getint(type);
break;
case 50718: /* DefaultScale */
pixel_aspect = getreal(type);
pixel_aspect /= getreal(type);
if(pixel_aspect > 0.995 && pixel_aspect < 1.005)
pixel_aspect = 1.0;
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 50778:
imgdata.color.dng_color[0].illuminant = get2();
break;
case 50779:
imgdata.color.dng_color[1].illuminant = get2();
break;
#endif
case 50721: /* ColorMatrix1 */
case 50722: /* ColorMatrix2 */
#ifdef LIBRAW_LIBRARY_BUILD
i = tag == 50721?0:1;
#endif
FORCC for (j=0; j < 3; j++)
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.color.dng_color[i].colormatrix[c][j]=
#endif
cm[c][j] = getreal(type);
}
use_cm = 1;
break;
case 0xc714: /* ForwardMatrix1 */
case 0xc715: /* ForwardMatrix2 */
#ifdef LIBRAW_LIBRARY_BUILD
i = tag == 0xc714?0:1;
#endif
for (j=0; j < 3; j++)
FORCC
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.color.dng_color[i].forwardmatrix[c][j]=
#endif
fm[c][j] = getreal(type);
}
break;
case 50723: /* CameraCalibration1 */
case 50724: /* CameraCalibration2 */
#ifdef LIBRAW_LIBRARY_BUILD
j = tag == 50723?0:1;
#endif
for (i=0; i < colors; i++)
FORCC
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.color.dng_color[j].calibration[i][c]=
#endif
cc[i][c] = getreal(type);
}
break;
case 50727: /* AnalogBalance */
FORCC ab[c] = getreal(type);
break;
case 50728: /* AsShotNeutral */
FORCC asn[c] = getreal(type);
break;
case 50729: /* AsShotWhiteXY */
xyz[0] = getreal(type);
xyz[1] = getreal(type);
xyz[2] = 1 - xyz[0] - xyz[1];
FORC3 xyz[c] /= d65_white[c];
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 50730: /* DNG: Baseline Exposure */
baseline_exposure = getreal(type);
break;
#endif
// IB start
case 50740: /* tag 0xc634 : DNG Adobe, DNG Pentax, Sony SR2, DNG Private */
#ifdef LIBRAW_LIBRARY_BUILD
{
char mbuf[64];
unsigned short makernote_found = 0;
INT64 curr_pos, start_pos = ftell(ifp);
unsigned MakN_order, m_sorder = order;
unsigned MakN_length;
unsigned pos_in_original_raw;
fread(mbuf, 1, 6, ifp);
if (!strcmp(mbuf, "Adobe"))
{
order = 0x4d4d; // Adobe header is always in "MM" / big endian
curr_pos = start_pos + 6;
while (curr_pos + 8 - start_pos <= len)
{
fread(mbuf, 1, 4, ifp);
curr_pos += 8;
if (!strncmp(mbuf, "MakN", 4)) {
makernote_found = 1;
MakN_length = get4();
MakN_order = get2();
pos_in_original_raw = get4();
order = MakN_order;
parse_makernote_0xc634(curr_pos + 6 - pos_in_original_raw, 0, AdobeDNG);
break;
}
}
}
else
{
fread(mbuf + 6, 1, 2, ifp);
if (!strcmp(mbuf, "PENTAX ") ||
!strcmp(mbuf, "SAMSUNG"))
{
makernote_found = 1;
fseek(ifp, start_pos, SEEK_SET);
parse_makernote_0xc634(base, 0, CameraDNG);
}
}
fseek(ifp, start_pos, SEEK_SET);
order = m_sorder;
}
// IB end
#endif
if (dng_version) break;
parse_minolta (j = get4()+base);
fseek (ifp, j, SEEK_SET);
parse_tiff_ifd (base);
break;
case 50752:
read_shorts (cr2_slice, 3);
break;
case 50829: /* ActiveArea */
top_margin = getint(type);
left_margin = getint(type);
height = getint(type) - top_margin;
width = getint(type) - left_margin;
break;
case 50830: /* MaskedAreas */
for (i=0; i < len && i < 32; i++)
((int*)mask)[i] = getint(type);
black = 0;
break;
case 51009: /* OpcodeList2 */
meta_offset = ftell(ifp);
break;
case 64772: /* Kodak P-series */
if (len < 13) break;
fseek (ifp, 16, SEEK_CUR);
data_offset = get4();
fseek (ifp, 28, SEEK_CUR);
data_offset += get4();
load_raw = &CLASS packed_load_raw;
break;
case 65026:
if (type == 2) fgets (model2, 64, ifp);
}
fseek (ifp, save, SEEK_SET);
}
if (sony_length && (buf = (unsigned *) malloc(sony_length))) {
fseek (ifp, sony_offset, SEEK_SET);
fread (buf, sony_length, 1, ifp);
sony_decrypt (buf, sony_length/4, 1, sony_key);
#ifndef LIBRAW_LIBRARY_BUILD
sfp = ifp;
if ((ifp = tmpfile())) {
fwrite (buf, sony_length, 1, ifp);
fseek (ifp, 0, SEEK_SET);
parse_tiff_ifd (-sony_offset);
fclose (ifp);
}
ifp = sfp;
#else
if( !ifp->tempbuffer_open(buf,sony_length))
{
parse_tiff_ifd(-sony_offset);
ifp->tempbuffer_close();
}
#endif
free (buf);
}
for (i=0; i < colors; i++)
FORCC cc[i][c] *= ab[i];
if (use_cm) {
FORCC for (i=0; i < 3; i++)
for (cam_xyz[c][i]=j=0; j < colors; j++)
cam_xyz[c][i] += cc[c][j] * cm[j][i] * xyz[i];
cam_xyz_coeff (cmatrix, cam_xyz);
}
if (asn[0]) {
cam_mul[3] = 0;
FORCC cam_mul[c] = 1 / asn[c];
}
if (!use_cm)
FORCC pre_mul[c] /= cc[c][c];
return 0;
}
int CLASS parse_tiff (int base)
{
int doff;
fseek (ifp, base, SEEK_SET);
order = get2();
if (order != 0x4949 && order != 0x4d4d) return 0;
get2();
while ((doff = get4())) {
fseek (ifp, doff+base, SEEK_SET);
if (parse_tiff_ifd (base)) break;
}
return 1;
}
void CLASS apply_tiff()
{
int max_samp=0, ties=0, os, ns, raw=-1, thm=-1, i;
struct jhead jh;
thumb_misc = 16;
if (thumb_offset) {
fseek (ifp, thumb_offset, SEEK_SET);
if (ljpeg_start (&jh, 1)) {
if((unsigned)jh.bits<17 && (unsigned)jh.wide < 0x10000 && (unsigned)jh.high < 0x10000)
{
thumb_misc = jh.bits;
thumb_width = jh.wide;
thumb_height = jh.high;
}
}
}
for (i=tiff_nifds; i--; ) {
if (tiff_ifd[i].t_shutter)
shutter = tiff_ifd[i].t_shutter;
tiff_ifd[i].t_shutter = shutter;
}
for (i=0; i < tiff_nifds; i++) {
if (max_samp < tiff_ifd[i].samples)
max_samp = tiff_ifd[i].samples;
if (max_samp > 3) max_samp = 3;
os = raw_width*raw_height;
ns = tiff_ifd[i].t_width*tiff_ifd[i].t_height;
if ((tiff_ifd[i].comp != 6 || tiff_ifd[i].samples != 3) &&
unsigned(tiff_ifd[i].t_width | tiff_ifd[i].t_height) < 0x10000 &&
(unsigned)tiff_ifd[i].bps < 33 && (unsigned)tiff_ifd[i].samples < 13 &&
ns && ((ns > os && (ties = 1)) ||
(ns == os && shot_select == ties++))) {
raw_width = tiff_ifd[i].t_width;
raw_height = tiff_ifd[i].t_height;
tiff_bps = tiff_ifd[i].bps;
tiff_compress = tiff_ifd[i].comp;
data_offset = tiff_ifd[i].offset;
#ifdef LIBRAW_LIBRARY_BUILD
data_size = tiff_ifd[i].bytes;
#endif
tiff_flip = tiff_ifd[i].t_flip;
tiff_samples = tiff_ifd[i].samples;
tile_width = tiff_ifd[i].t_tile_width;
tile_length = tiff_ifd[i].t_tile_length;
shutter = tiff_ifd[i].t_shutter;
raw = i;
}
}
if (is_raw == 1 && ties) is_raw = ties;
if (!tile_width ) tile_width = INT_MAX;
if (!tile_length) tile_length = INT_MAX;
for (i=tiff_nifds; i--; )
if (tiff_ifd[i].t_flip) tiff_flip = tiff_ifd[i].t_flip;
if (raw >= 0 && !load_raw)
switch (tiff_compress) {
case 32767:
if (tiff_ifd[raw].bytes == raw_width*raw_height) {
tiff_bps = 12;
load_raw = &CLASS sony_arw2_load_raw; break;
}
if (!strncasecmp(make,"Sony",4) &&
tiff_ifd[raw].bytes == raw_width*raw_height*2) {
tiff_bps = 14;
load_raw = &CLASS unpacked_load_raw; break;
}
if (tiff_ifd[raw].bytes*8 != raw_width*raw_height*tiff_bps) {
raw_height += 8;
load_raw = &CLASS sony_arw_load_raw; break;
}
load_flags = 79;
case 32769:
load_flags++;
case 32770:
case 32773: goto slr;
case 0: case 1:
#ifdef LIBRAW_LIBRARY_BUILD
// Sony 14-bit uncompressed
if(!strncasecmp(make,"Sony",4) &&
tiff_ifd[raw].bytes == raw_width*raw_height*2)
{
tiff_bps = 14;
load_raw = &CLASS unpacked_load_raw;
break;
}
if(!strncasecmp(make,"Nikon",5) && !strncmp(software,"Nikon Scan",10))
{
load_raw = &CLASS nikon_coolscan_load_raw;
raw_color = 1;
filters = 0;
break;
}
#endif
if (!strncmp(make,"OLYMPUS",7) &&
tiff_ifd[raw].bytes*2 == raw_width*raw_height*3)
load_flags = 24;
if (tiff_ifd[raw].bytes*5 == raw_width*raw_height*8) {
load_flags = 81;
tiff_bps = 12;
} slr:
switch (tiff_bps) {
case 8: load_raw = &CLASS eight_bit_load_raw; break;
case 12: if (tiff_ifd[raw].phint == 2)
load_flags = 6;
load_raw = &CLASS packed_load_raw; break;
case 14: load_flags = 0;
case 16: load_raw = &CLASS unpacked_load_raw;
if (!strncmp(make,"OLYMPUS",7) &&
tiff_ifd[raw].bytes*7 > raw_width*raw_height)
load_raw = &CLASS olympus_load_raw;
}
break;
case 6: case 7: case 99:
load_raw = &CLASS lossless_jpeg_load_raw; break;
case 262:
load_raw = &CLASS kodak_262_load_raw; break;
case 34713:
if ((raw_width+9)/10*16*raw_height == tiff_ifd[raw].bytes) {
load_raw = &CLASS packed_load_raw;
load_flags = 1;
} else if (raw_width*raw_height*3 == tiff_ifd[raw].bytes*2) {
load_raw = &CLASS packed_load_raw;
if (model[0] == 'N') load_flags = 80;
} else if (raw_width*raw_height*3 == tiff_ifd[raw].bytes) {
load_raw = &CLASS nikon_yuv_load_raw;
gamma_curve (1/2.4, 12.92, 1, 4095);
memset (cblack, 0, sizeof cblack);
filters = 0;
} else if (raw_width*raw_height*2 == tiff_ifd[raw].bytes) {
load_raw = &CLASS unpacked_load_raw;
load_flags = 4;
order = 0x4d4d;
} else
#ifdef LIBRAW_LIBRARY_BUILD
if(raw_width*raw_height*3 == tiff_ifd[raw].bytes*2)
{
load_raw = &CLASS packed_load_raw;
load_flags=80;
}
else if(tiff_ifd[raw].rows_per_strip && tiff_ifd[raw].strip_offsets_count &&
tiff_ifd[raw].strip_offsets_count == tiff_ifd[raw].strip_byte_counts_count)
{
int fit = 1;
for(int i = 0; i < tiff_ifd[raw].strip_byte_counts_count-1; i++) // all but last
if(tiff_ifd[raw].strip_byte_counts[i]*2 != tiff_ifd[raw].rows_per_strip*raw_width*3)
{
fit = 0;
break;
}
if(fit)
load_raw = &CLASS nikon_load_striped_packed_raw;
else
load_raw = &CLASS nikon_load_raw; // fallback
}
else
#endif
load_raw = &CLASS nikon_load_raw; break;
case 65535:
load_raw = &CLASS pentax_load_raw; break;
case 65000:
switch (tiff_ifd[raw].phint) {
case 2: load_raw = &CLASS kodak_rgb_load_raw; filters = 0; break;
case 6: load_raw = &CLASS kodak_ycbcr_load_raw; filters = 0; break;
case 32803: load_raw = &CLASS kodak_65000_load_raw;
}
case 32867: case 34892: break;
#ifdef LIBRAW_LIBRARY_BUILD
case 8: break;
#endif
default: is_raw = 0;
}
if (!dng_version)
if ( ((tiff_samples == 3 && tiff_ifd[raw].bytes && tiff_bps != 14 &&
(tiff_compress & -16) != 32768)
|| (tiff_bps == 8 && !strcasestr(make,"Kodak") &&
!strstr(model2,"DEBUG RAW")))
&& strncmp(software,"Nikon Scan",10))
is_raw = 0;
for (i=0; i < tiff_nifds; i++)
if (i != raw && tiff_ifd[i].samples == max_samp &&
tiff_ifd[i].bps>0 && tiff_ifd[i].bps < 33
&& tiff_ifd[i].phint != 32803
&& tiff_ifd[i].phint != 34892
&& unsigned(tiff_ifd[i].t_width | tiff_ifd[i].t_height) < 0x10000 &&
tiff_ifd[i].t_width * tiff_ifd[i].t_height / (SQR(tiff_ifd[i].bps)+1) >
thumb_width * thumb_height / (SQR(thumb_misc)+1)
&& tiff_ifd[i].comp != 34892) {
thumb_width = tiff_ifd[i].t_width;
thumb_height = tiff_ifd[i].t_height;
thumb_offset = tiff_ifd[i].offset;
thumb_length = tiff_ifd[i].bytes;
thumb_misc = tiff_ifd[i].bps;
thm = i;
}
if (thm >= 0) {
thumb_misc |= tiff_ifd[thm].samples << 5;
switch (tiff_ifd[thm].comp) {
case 0:
write_thumb = &CLASS layer_thumb;
break;
case 1:
if (tiff_ifd[thm].bps <= 8)
write_thumb = &CLASS ppm_thumb;
else if (!strncmp(make,"Imacon",6))
write_thumb = &CLASS ppm16_thumb;
else
thumb_load_raw = &CLASS kodak_thumb_load_raw;
break;
case 65000:
thumb_load_raw = tiff_ifd[thm].phint == 6 ?
&CLASS kodak_ycbcr_load_raw : &CLASS kodak_rgb_load_raw;
}
}
}
void CLASS parse_minolta (int base)
{
int save, tag, len, offset, high=0, wide=0, i, c;
short sorder=order;
fseek (ifp, base, SEEK_SET);
if (fgetc(ifp) || fgetc(ifp)-'M' || fgetc(ifp)-'R') return;
order = fgetc(ifp) * 0x101;
offset = base + get4() + 8;
while ((save=ftell(ifp)) < offset) {
for (tag=i=0; i < 4; i++)
tag = tag << 8 | fgetc(ifp);
len = get4();
switch (tag) {
case 0x505244: /* PRD */
fseek (ifp, 8, SEEK_CUR);
high = get2();
wide = get2();
break;
#ifdef LIBRAW_LIBRARY_BUILD
case 0x524946: /* RIF */
if (!strncasecmp(model,"DSLR-A100", 9))
{
fseek(ifp, 8, SEEK_CUR);
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][2] = get2();
get4();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][0] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][2] = get2();
imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Flash][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_Shade][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_W][3] =
imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][1] = imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][3] = 0x100;
}
break;
#endif
case 0x574247: /* WBG */
get4();
i = strcmp(model,"DiMAGE A200") ? 0:3;
FORC4 cam_mul[c ^ (c >> 1) ^ i] = get2();
break;
case 0x545457: /* TTW */
parse_tiff (ftell(ifp));
data_offset = offset;
}
fseek (ifp, save+len+8, SEEK_SET);
}
raw_height = high;
raw_width = wide;
order = sorder;
}
/*
Many cameras have a "debug mode" that writes JPEG and raw
at the same time. The raw file has no header, so try to
to open the matching JPEG file and read its metadata.
*/
void CLASS parse_external_jpeg()
{
const char *file, *ext;
char *jname, *jfile, *jext;
#ifndef LIBRAW_LIBRARY_BUILD
FILE *save=ifp;
#else
#if defined(_WIN32) && !defined(__MINGW32__) && defined(_MSC_VER) && (_MSC_VER > 1310)
if(ifp->wfname())
{
std::wstring rawfile(ifp->wfname());
rawfile.replace(rawfile.length()-3,3,L"JPG");
if(!ifp->subfile_open(rawfile.c_str()))
{
parse_tiff (12);
thumb_offset = 0;
is_raw = 1;
ifp->subfile_close();
}
else
imgdata.process_warnings |= LIBRAW_WARN_NO_METADATA ;
return;
}
#endif
if(!ifp->fname())
{
imgdata.process_warnings |= LIBRAW_WARN_NO_METADATA ;
return;
}
#endif
ext = strrchr (ifname, '.');
file = strrchr (ifname, '/');
if (!file) file = strrchr (ifname, '\\');
#ifndef LIBRAW_LIBRARY_BUILD
if (!file) file = ifname-1;
#else
if (!file) file = (char*)ifname-1;
#endif
file++;
if (!ext || strlen(ext) != 4 || ext-file != 8) return;
jname = (char *) malloc (strlen(ifname) + 1);
merror (jname, "parse_external_jpeg()");
strcpy (jname, ifname);
jfile = file - ifname + jname;
jext = ext - ifname + jname;
if (strcasecmp (ext, ".jpg")) {
strcpy (jext, isupper(ext[1]) ? ".JPG":".jpg");
if (isdigit(*file)) {
memcpy (jfile, file+4, 4);
memcpy (jfile+4, file, 4);
}
} else
while (isdigit(*--jext)) {
if (*jext != '9') {
(*jext)++;
break;
}
*jext = '0';
}
#ifndef LIBRAW_LIBRARY_BUILD
if (strcmp (jname, ifname)) {
if ((ifp = fopen (jname, "rb"))) {
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Reading metadata from %s ...\n"), jname);
#endif
parse_tiff (12);
thumb_offset = 0;
is_raw = 1;
fclose (ifp);
}
}
#else
if (strcmp (jname, ifname))
{
if(!ifp->subfile_open(jname))
{
parse_tiff (12);
thumb_offset = 0;
is_raw = 1;
ifp->subfile_close();
}
else
imgdata.process_warnings |= LIBRAW_WARN_NO_METADATA ;
}
#endif
if (!timestamp)
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_METADATA ;
#endif
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("Failed to read metadata from %s\n"), jname);
#endif
}
free (jname);
#ifndef LIBRAW_LIBRARY_BUILD
ifp = save;
#endif
}
/*
CIFF block 0x1030 contains an 8x8 white sample.
Load this into white[][] for use in scale_colors().
*/
void CLASS ciff_block_1030()
{
static const ushort key[] = { 0x410, 0x45f3 };
int i, bpp, row, col, vbits=0;
unsigned long bitbuf=0;
if ((get2(),get4()) != 0x80008 || !get4()) return;
bpp = get2();
if (bpp != 10 && bpp != 12) return;
for (i=row=0; row < 8; row++)
for (col=0; col < 8; col++) {
if (vbits < bpp) {
bitbuf = bitbuf << 16 | (get2() ^ key[i++ & 1]);
vbits += 16;
}
white[row][col] = bitbuf >> (vbits -= bpp) & ~(-1 << bpp);
}
}
/*
Parse a CIFF file, better known as Canon CRW format.
*/
void CLASS parse_ciff (int offset, int length, int depth)
{
int tboff, nrecs, c, type, len, save, wbi=-1;
ushort key[] = { 0x410, 0x45f3 };
fseek (ifp, offset+length-4, SEEK_SET);
tboff = get4() + offset;
fseek (ifp, tboff, SEEK_SET);
nrecs = get2();
if ((nrecs | depth) > 127) return;
while (nrecs--) {
type = get2();
len = get4();
save = ftell(ifp) + 4;
fseek (ifp, offset+get4(), SEEK_SET);
if ((((type >> 8) + 8) | 8) == 0x38) {
parse_ciff (ftell(ifp), len, depth+1); /* Parse a sub-table */
}
if (type == 0x0810)
fread (artist, 64, 1, ifp);
if (type == 0x080a) {
fread (make, 64, 1, ifp);
fseek (ifp, strlen(make) - 63, SEEK_CUR);
fread (model, 64, 1, ifp);
}
if (type == 0x1810) {
width = get4();
height = get4();
pixel_aspect = int_to_float(get4());
flip = get4();
}
if (type == 0x1835) /* Get the decoder table */
tiff_compress = get4();
if (type == 0x2007) {
thumb_offset = ftell(ifp);
thumb_length = len;
}
if (type == 0x1818) {
shutter = powf64(2.0f, -int_to_float((get4(),get4())));
aperture = powf64(2.0f, int_to_float(get4())/2);
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.lens.makernotes.CurAp = aperture;
#endif
}
if (type == 0x102a) {
// iso_speed = pow (2.0, (get4(),get2())/32.0 - 4) * 50;
iso_speed = powf64(2.0f, ((get2(),get2()) + get2())/32.0f - 5.0f) * 100.0f;
#ifdef LIBRAW_LIBRARY_BUILD
aperture = _CanonConvertAperture((get2(),get2()));
imgdata.lens.makernotes.CurAp = aperture;
#else
aperture = powf64(2.0, (get2(),(short)get2())/64.0);
#endif
shutter = powf64(2.0,-((short)get2())/32.0);
wbi = (get2(),get2());
if (wbi > 17) wbi = 0;
fseek (ifp, 32, SEEK_CUR);
if (shutter > 1e6) shutter = get2()/10.0;
}
if (type == 0x102c) {
if (get2() > 512) { /* Pro90, G1 */
fseek (ifp, 118, SEEK_CUR);
FORC4 cam_mul[c ^ 2] = get2();
} else { /* G2, S30, S40 */
fseek (ifp, 98, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1) ^ 1] = get2();
}
}
#ifdef LIBRAW_LIBRARY_BUILD
if (type == 0x10a9)
{
INT64 o = ftell(ifp);
fseek (ifp, (0x5<<1), SEEK_CUR);
Canon_WBpresets(0,0);
fseek(ifp,o,SEEK_SET);
}
if (type == 0x102d)
{
INT64 o = ftell(ifp);
fseek(ifp, 44, SEEK_CUR);
imgdata.lens.makernotes.LensID = get2();
imgdata.lens.makernotes.MaxFocal = get2();
imgdata.lens.makernotes.MinFocal = get2();
imgdata.lens.makernotes.CanonFocalUnits = get2();
if (imgdata.lens.makernotes.CanonFocalUnits != 1)
{
imgdata.lens.makernotes.MaxFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
imgdata.lens.makernotes.MinFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
imgdata.lens.makernotes.MaxAp = _CanonConvertAperture(get2());
imgdata.lens.makernotes.MinAp = _CanonConvertAperture(get2());
fseek(ifp,o,SEEK_SET);
}
#endif
if (type == 0x0032) {
if (len == 768) { /* EOS D30 */
fseek (ifp, 72, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1)] = 1024.0 / get2();
if (!wbi) cam_mul[0] = -1; /* use my auto white balance */
} else if (!cam_mul[0]) {
if (get2() == key[0]) /* Pro1, G6, S60, S70 */
c = (strstr(model,"Pro1") ?
"012346000000000000":"01345:000000006008")[wbi]-'0'+ 2;
else { /* G3, G5, S45, S50 */
c = "023457000000006000"[wbi]-'0';
key[0] = key[1] = 0;
}
fseek (ifp, 78 + c*8, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1) ^ 1] = get2() ^ key[c & 1];
if (!wbi) cam_mul[0] = -1;
}
}
if (type == 0x10a9) { /* D60, 10D, 300D, and clones */
if (len > 66) wbi = "0134567028"[wbi]-'0';
fseek (ifp, 2 + wbi*8, SEEK_CUR);
FORC4 cam_mul[c ^ (c >> 1)] = get2();
}
if (type == 0x1030 && (0x18040 >> wbi & 1))
ciff_block_1030(); /* all that don't have 0x10a9 */
if (type == 0x1031) {
raw_width = (get2(),get2());
raw_height = get2();
}
if (type == 0x501c) {
iso_speed = len & 0xffff;
}
if (type == 0x5029) {
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.lens.makernotes.CurFocal = len >> 16;
imgdata.lens.makernotes.FocalType = len & 0xffff;
if (imgdata.lens.makernotes.FocalType == 2) {
imgdata.lens.makernotes.CanonFocalUnits = 32;
imgdata.lens.makernotes.CurFocal /= (float)imgdata.lens.makernotes.CanonFocalUnits;
}
focal_len = imgdata.lens.makernotes.CurFocal;
#else
focal_len = len >> 16;
if ((len & 0xffff) == 2) focal_len /= 32;
#endif
}
if (type == 0x5813) flash_used = int_to_float(len);
if (type == 0x5814) canon_ev = int_to_float(len);
if (type == 0x5817) shot_order = len;
if (type == 0x5834)
{
unique_id = len;
#ifdef LIBRAW_LIBRARY_BUILD
setCanonBodyFeatures(unique_id);
#endif
}
if (type == 0x580e) timestamp = len;
if (type == 0x180e) timestamp = get4();
#ifdef LOCALTIME
if ((type | 0x4000) == 0x580e)
timestamp = mktime (gmtime (×tamp));
#endif
fseek (ifp, save, SEEK_SET);
}
}
void CLASS parse_rollei()
{
char line[128], *val;
struct tm t;
fseek (ifp, 0, SEEK_SET);
memset (&t, 0, sizeof t);
do {
fgets (line, 128, ifp);
if ((val = strchr(line,'=')))
*val++ = 0;
else
val = line + strlen(line);
if (!strcmp(line,"DAT"))
sscanf (val, "%d.%d.%d", &t.tm_mday, &t.tm_mon, &t.tm_year);
if (!strcmp(line,"TIM"))
sscanf (val, "%d:%d:%d", &t.tm_hour, &t.tm_min, &t.tm_sec);
if (!strcmp(line,"HDR"))
thumb_offset = atoi(val);
if (!strcmp(line,"X "))
raw_width = atoi(val);
if (!strcmp(line,"Y "))
raw_height = atoi(val);
if (!strcmp(line,"TX "))
thumb_width = atoi(val);
if (!strcmp(line,"TY "))
thumb_height = atoi(val);
} while (strncmp(line,"EOHD",4));
data_offset = thumb_offset + thumb_width * thumb_height * 2;
t.tm_year -= 1900;
t.tm_mon -= 1;
if (mktime(&t) > 0)
timestamp = mktime(&t);
strcpy (make, "Rollei");
strcpy (model,"d530flex");
write_thumb = &CLASS rollei_thumb;
}
void CLASS parse_sinar_ia()
{
int entries, off;
char str[8], *cp;
order = 0x4949;
fseek (ifp, 4, SEEK_SET);
entries = get4();
fseek (ifp, get4(), SEEK_SET);
while (entries--) {
off = get4(); get4();
fread (str, 8, 1, ifp);
if (!strcmp(str,"META")) meta_offset = off;
if (!strcmp(str,"THUMB")) thumb_offset = off;
if (!strcmp(str,"RAW0")) data_offset = off;
}
fseek (ifp, meta_offset+20, SEEK_SET);
fread (make, 64, 1, ifp);
make[63] = 0;
if ((cp = strchr(make,' '))) {
strcpy (model, cp+1);
*cp = 0;
}
raw_width = get2();
raw_height = get2();
load_raw = &CLASS unpacked_load_raw;
thumb_width = (get4(),get2());
thumb_height = get2();
write_thumb = &CLASS ppm_thumb;
maximum = 0x3fff;
}
void CLASS parse_phase_one (int base)
{
unsigned entries, tag, type, len, data, save, i, c;
float romm_cam[3][3];
char *cp;
#ifdef LIBRAW_LIBRARY_BUILD
char body_id[3];
body_id[0] = 0;
#endif
memset (&ph1, 0, sizeof ph1);
fseek (ifp, base, SEEK_SET);
order = get4() & 0xffff;
if (get4() >> 8 != 0x526177) return; /* "Raw" */
fseek (ifp, get4()+base, SEEK_SET);
entries = get4();
get4();
while (entries--) {
tag = get4();
type = get4();
len = get4();
data = get4();
save = ftell(ifp);
fseek (ifp, base+data, SEEK_SET);
switch (tag) {
#ifdef LIBRAW_LIBRARY_BUILD
case 0x0102:
fread(body_id, 1, 3, ifp);
if ((body_id[0] == 0x4c) && (body_id[1] == 0x49)) {
body_id[1] = body_id[2];
}
unique_id = (((body_id[0] & 0x3f) << 5) | (body_id[1] & 0x3f)) - 0x41;
setPhaseOneFeatures(unique_id);
break;
case 0x0401:
if (type == 4) imgdata.lens.makernotes.CurAp = powf64(2.0f, (int_to_float(data)/2.0f));
else imgdata.lens.makernotes.CurAp = powf64(2.0f, (getreal(type)/2.0f));
break;
case 0x0403:
if (type == 4) imgdata.lens.makernotes.CurFocal = int_to_float(data);
else imgdata.lens.makernotes.CurFocal = getreal(type);
break;
case 0x0410:
fread(imgdata.lens.makernotes.body, 1, len, ifp);
break;
case 0x0412:
fread(imgdata.lens.makernotes.Lens, 1, len, ifp);
break;
case 0x0414:
if (type == 4) {
imgdata.lens.makernotes.MaxAp4CurFocal = powf64(2.0f, (int_to_float(data)/2.0f));
} else {
imgdata.lens.makernotes.MaxAp4CurFocal = powf64(2.0f, (getreal(type) / 2.0f));
}
break;
case 0x0415:
if (type == 4) {
imgdata.lens.makernotes.MinAp4CurFocal = powf64(2.0f, (int_to_float(data)/2.0f));
} else {
imgdata.lens.makernotes.MinAp4CurFocal = powf64(2.0f, (getreal(type) / 2.0f));
}
break;
case 0x0416:
if (type == 4) {
imgdata.lens.makernotes.MinFocal = int_to_float(data);
} else {
imgdata.lens.makernotes.MinFocal = getreal(type);
}
if (imgdata.lens.makernotes.MinFocal > 1000.0f)
{
imgdata.lens.makernotes.MinFocal = 0.0f;
}
break;
case 0x0417:
if (type == 4) {
imgdata.lens.makernotes.MaxFocal = int_to_float(data);
} else {
imgdata.lens.makernotes.MaxFocal = getreal(type);
}
break;
#endif
case 0x100: flip = "0653"[data & 3]-'0'; break;
case 0x106:
for (i=0; i < 9; i++)
((float *)romm_cam)[i] = getreal(11);
romm_coeff (romm_cam);
break;
case 0x107:
FORC3 cam_mul[c] = getreal(11);
break;
case 0x108: raw_width = data; break;
case 0x109: raw_height = data; break;
case 0x10a: left_margin = data; break;
case 0x10b: top_margin = data; break;
case 0x10c: width = data; break;
case 0x10d: height = data; break;
case 0x10e: ph1.format = data; break;
case 0x10f: data_offset = data+base; break;
case 0x110: meta_offset = data+base;
meta_length = len; break;
case 0x112: ph1.key_off = save - 4; break;
case 0x210: ph1.tag_210 = int_to_float(data); break;
case 0x21a: ph1.tag_21a = data; break;
case 0x21c: strip_offset = data+base; break;
case 0x21d: ph1.t_black = data; break;
case 0x222: ph1.split_col = data; break;
case 0x223: ph1.black_col = data+base; break;
case 0x224: ph1.split_row = data; break;
case 0x225: ph1.black_row = data+base; break;
case 0x301:
model[63] = 0;
fread (model, 1, 63, ifp);
if ((cp = strstr(model," camera"))) *cp = 0;
}
fseek (ifp, save, SEEK_SET);
}
#ifdef LIBRAW_LIBRARY_BUILD
if (!imgdata.lens.makernotes.body[0] && !body_id[0]) {
fseek (ifp, meta_offset, SEEK_SET);
order = get2();
fseek (ifp, 6, SEEK_CUR);
fseek (ifp, meta_offset+get4(), SEEK_SET);
entries = get4(); get4();
while (entries--) {
tag = get4();
len = get4();
data = get4();
save = ftell(ifp);
fseek (ifp, meta_offset+data, SEEK_SET);
if (tag == 0x0407) {
fread(body_id, 1, 3, ifp);
if ((body_id[0] == 0x4c) && (body_id[1] == 0x49)) {
body_id[1] = body_id[2];
}
unique_id = (((body_id[0] & 0x3f) << 5) | (body_id[1] & 0x3f)) - 0x41;
setPhaseOneFeatures(unique_id);
}
fseek (ifp, save, SEEK_SET);
}
}
#endif
load_raw = ph1.format < 3 ?
&CLASS phase_one_load_raw : &CLASS phase_one_load_raw_c;
maximum = 0xffff;
strcpy (make, "Phase One");
if (model[0]) return;
switch (raw_height) {
case 2060: strcpy (model,"LightPhase"); break;
case 2682: strcpy (model,"H 10"); break;
case 4128: strcpy (model,"H 20"); break;
case 5488: strcpy (model,"H 25"); break;
}
}
void CLASS parse_fuji (int offset)
{
unsigned entries, tag, len, save, c;
fseek (ifp, offset, SEEK_SET);
entries = get4();
if (entries > 255) return;
while (entries--) {
tag = get2();
len = get2();
save = ftell(ifp);
if (tag == 0x100) {
raw_height = get2();
raw_width = get2();
} else if (tag == 0x121) {
height = get2();
if ((width = get2()) == 4284) width += 3;
} else if (tag == 0x130) {
fuji_layout = fgetc(ifp) >> 7;
fuji_width = !(fgetc(ifp) & 8);
} else if (tag == 0x131) {
filters = 9;
FORC(36) xtrans_abs[0][35-c] = fgetc(ifp) & 3;
} else if (tag == 0x2ff0) {
FORC4 cam_mul[c ^ 1] = get2();
// IB start
#ifdef LIBRAW_LIBRARY_BUILD
} else if (tag == 0x9650) {
imgdata.color.FujiExpoMidPointShift = ((short)get2()) / ((float)get2());
} else if (tag == 0x2100) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Daylight][c ^ 1] = get2();
} else if (tag == 0x2200) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Cloudy][c ^ 1] = get2();
} else if (tag == 0x2300) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_D][c ^ 1] = get2();
} else if (tag == 0x2301) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_N][c ^ 1] = get2();
} else if (tag == 0x2302) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_WW][c ^ 1] = get2();
} else if (tag == 0x2310) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_FL_L][c ^ 1] = get2();
} else if (tag == 0x2400) {
FORC4 imgdata.color.WB_Coeffs[LIBRAW_WBI_Tungsten][c ^ 1] = get2();
#endif
// IB end
} else if (tag == 0xc000) {
c = order;
order = 0x4949;
if ((tag = get4()) > 10000) tag = get4();
width = tag;
height = get4();
#ifdef LIBRAW_LIBRARY_BUILD
libraw_internal_data.unpacker_data.posRAFData = save;
libraw_internal_data.unpacker_data.lenRAFData = (len>>1);
#endif
order = c;
}
fseek (ifp, save+len, SEEK_SET);
}
height <<= fuji_layout;
width >>= fuji_layout;
}
int CLASS parse_jpeg (int offset)
{
int len, save, hlen, mark;
fseek (ifp, offset, SEEK_SET);
if (fgetc(ifp) != 0xff || fgetc(ifp) != 0xd8) return 0;
while (fgetc(ifp) == 0xff && (mark = fgetc(ifp)) != 0xda) {
order = 0x4d4d;
len = get2() - 2;
save = ftell(ifp);
if (mark == 0xc0 || mark == 0xc3) {
fgetc(ifp);
raw_height = get2();
raw_width = get2();
}
order = get2();
hlen = get4();
if (get4() == 0x48454150) /* "HEAP" */
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
#endif
parse_ciff (save+hlen, len-hlen, 0);
}
if (parse_tiff (save+6)) apply_tiff();
fseek (ifp, save+len, SEEK_SET);
}
return 1;
}
void CLASS parse_riff()
{
unsigned i, size, end;
char tag[4], date[64], month[64];
static const char mon[12][4] =
{ "Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec" };
struct tm t;
order = 0x4949;
fread (tag, 4, 1, ifp);
size = get4();
end = ftell(ifp) + size;
if (!memcmp(tag,"RIFF",4) || !memcmp(tag,"LIST",4)) {
int maxloop = 1000;
get4();
while (ftell(ifp)+7 < end && !feof(ifp) && maxloop--)
parse_riff();
} else if (!memcmp(tag,"nctg",4)) {
while (ftell(ifp)+7 < end) {
i = get2();
size = get2();
if ((i+1) >> 1 == 10 && size == 20)
get_timestamp(0);
else fseek (ifp, size, SEEK_CUR);
}
} else if (!memcmp(tag,"IDIT",4) && size < 64) {
fread (date, 64, 1, ifp);
date[size] = 0;
memset (&t, 0, sizeof t);
if (sscanf (date, "%*s %s %d %d:%d:%d %d", month, &t.tm_mday,
&t.tm_hour, &t.tm_min, &t.tm_sec, &t.tm_year) == 6) {
for (i=0; i < 12 && strcasecmp(mon[i],month); i++);
t.tm_mon = i;
t.tm_year -= 1900;
if (mktime(&t) > 0)
timestamp = mktime(&t);
}
} else
fseek (ifp, size, SEEK_CUR);
}
void CLASS parse_qt (int end)
{
unsigned save, size;
char tag[4];
order = 0x4d4d;
while (ftell(ifp)+7 < end) {
save = ftell(ifp);
if ((size = get4()) < 8) return;
fread (tag, 4, 1, ifp);
if (!memcmp(tag,"moov",4) ||
!memcmp(tag,"udta",4) ||
!memcmp(tag,"CNTH",4))
parse_qt (save+size);
if (!memcmp(tag,"CNDA",4))
parse_jpeg (ftell(ifp));
fseek (ifp, save+size, SEEK_SET);
}
}
void CLASS parse_smal (int offset, int fsize)
{
int ver;
fseek (ifp, offset+2, SEEK_SET);
order = 0x4949;
ver = fgetc(ifp);
if (ver == 6)
fseek (ifp, 5, SEEK_CUR);
if (get4() != fsize) return;
if (ver > 6) data_offset = get4();
raw_height = height = get2();
raw_width = width = get2();
strcpy (make, "SMaL");
sprintf (model, "v%d %dx%d", ver, width, height);
if (ver == 6) load_raw = &CLASS smal_v6_load_raw;
if (ver == 9) load_raw = &CLASS smal_v9_load_raw;
}
void CLASS parse_cine()
{
unsigned off_head, off_setup, off_image, i;
order = 0x4949;
fseek (ifp, 4, SEEK_SET);
is_raw = get2() == 2;
fseek (ifp, 14, SEEK_CUR);
is_raw *= get4();
off_head = get4();
off_setup = get4();
off_image = get4();
timestamp = get4();
if ((i = get4())) timestamp = i;
fseek (ifp, off_head+4, SEEK_SET);
raw_width = get4();
raw_height = get4();
switch (get2(),get2()) {
case 8: load_raw = &CLASS eight_bit_load_raw; break;
case 16: load_raw = &CLASS unpacked_load_raw;
}
fseek (ifp, off_setup+792, SEEK_SET);
strcpy (make, "CINE");
sprintf (model, "%d", get4());
fseek (ifp, 12, SEEK_CUR);
switch ((i=get4()) & 0xffffff) {
case 3: filters = 0x94949494; break;
case 4: filters = 0x49494949; break;
default: is_raw = 0;
}
fseek (ifp, 72, SEEK_CUR);
switch ((get4()+3600) % 360) {
case 270: flip = 4; break;
case 180: flip = 1; break;
case 90: flip = 7; break;
case 0: flip = 2;
}
cam_mul[0] = getreal(11);
cam_mul[2] = getreal(11);
maximum = ~((~0u) << get4());
fseek (ifp, 668, SEEK_CUR);
shutter = get4()/1000000000.0;
fseek (ifp, off_image, SEEK_SET);
if (shot_select < is_raw)
fseek (ifp, shot_select*8, SEEK_CUR);
data_offset = (INT64) get4() + 8;
data_offset += (INT64) get4() << 32;
}
void CLASS parse_redcine()
{
unsigned i, len, rdvo;
order = 0x4d4d;
is_raw = 0;
fseek (ifp, 52, SEEK_SET);
width = get4();
height = get4();
fseek (ifp, 0, SEEK_END);
fseek (ifp, -(i = ftello(ifp) & 511), SEEK_CUR);
if (get4() != i || get4() != 0x52454f42) {
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: Tail is missing, parsing from head...\n"), ifname);
#endif
fseek (ifp, 0, SEEK_SET);
while ((len = get4()) != EOF) {
if (get4() == 0x52454456)
if (is_raw++ == shot_select)
data_offset = ftello(ifp) - 8;
fseek (ifp, len-8, SEEK_CUR);
}
} else {
rdvo = get4();
fseek (ifp, 12, SEEK_CUR);
is_raw = get4();
fseeko (ifp, rdvo+8 + shot_select*4, SEEK_SET);
data_offset = get4();
}
}
//@end COMMON
char * CLASS foveon_gets (int offset, char *str, int len)
{
int i;
fseek (ifp, offset, SEEK_SET);
for (i=0; i < len-1; i++)
if ((str[i] = get2()) == 0) break;
str[i] = 0;
return str;
}
void CLASS parse_foveon()
{
int entries, img=0, off, len, tag, save, i, wide, high, pent, poff[256][2];
char name[64], value[64];
order = 0x4949; /* Little-endian */
fseek (ifp, 36, SEEK_SET);
flip = get4();
fseek (ifp, -4, SEEK_END);
fseek (ifp, get4(), SEEK_SET);
if (get4() != 0x64434553) return; /* SECd */
entries = (get4(),get4());
while (entries--) {
off = get4();
len = get4();
tag = get4();
save = ftell(ifp);
fseek (ifp, off, SEEK_SET);
if (get4() != (0x20434553 | (tag << 24))) return;
switch (tag) {
case 0x47414d49: /* IMAG */
case 0x32414d49: /* IMA2 */
fseek (ifp, 8, SEEK_CUR);
pent = get4();
wide = get4();
high = get4();
if (wide > raw_width && high > raw_height) {
switch (pent) {
case 5: load_flags = 1;
case 6: load_raw = &CLASS foveon_sd_load_raw; break;
case 30: load_raw = &CLASS foveon_dp_load_raw; break;
default: load_raw = 0;
}
raw_width = wide;
raw_height = high;
data_offset = off+28;
is_foveon = 1;
}
fseek (ifp, off+28, SEEK_SET);
if (fgetc(ifp) == 0xff && fgetc(ifp) == 0xd8
&& thumb_length < len-28) {
thumb_offset = off+28;
thumb_length = len-28;
write_thumb = &CLASS jpeg_thumb;
}
if (++img == 2 && !thumb_length) {
thumb_offset = off+24;
thumb_width = wide;
thumb_height = high;
write_thumb = &CLASS foveon_thumb;
}
break;
case 0x464d4143: /* CAMF */
meta_offset = off+8;
meta_length = len-28;
break;
case 0x504f5250: /* PROP */
pent = (get4(),get4());
fseek (ifp, 12, SEEK_CUR);
off += pent*8 + 24;
if ((unsigned) pent > 256) pent=256;
for (i=0; i < pent*2; i++)
((int *)poff)[i] = off + get4()*2;
for (i=0; i < pent; i++) {
foveon_gets (poff[i][0], name, 64);
foveon_gets (poff[i][1], value, 64);
if (!strcmp (name, "ISO"))
iso_speed = atoi(value);
if (!strcmp (name, "CAMMANUF"))
strcpy (make, value);
if (!strcmp (name, "CAMMODEL"))
strcpy (model, value);
if (!strcmp (name, "WB_DESC"))
strcpy (model2, value);
if (!strcmp (name, "TIME"))
timestamp = atoi(value);
if (!strcmp (name, "EXPTIME"))
shutter = atoi(value) / 1000000.0;
if (!strcmp (name, "APERTURE"))
aperture = atof(value);
if (!strcmp (name, "FLENGTH"))
focal_len = atof(value);
#ifdef LIBRAW_LIBRARY_BUILD
if (!strcmp (name, "FLEQ35MM"))
imgdata.lens.makernotes.FocalLengthIn35mmFormat = atof(value);
if (!strcmp (name, "LENSARANGE"))
{
char *sp;
imgdata.lens.makernotes.MaxAp4CurFocal = imgdata.lens.makernotes.MinAp4CurFocal = atof(value);
sp = strrchr (value, ' ');
if (sp)
{
imgdata.lens.makernotes.MinAp4CurFocal = atof(sp);
if (imgdata.lens.makernotes.MaxAp4CurFocal > imgdata.lens.makernotes.MinAp4CurFocal)
my_swap (float, imgdata.lens.makernotes.MaxAp4CurFocal, imgdata.lens.makernotes.MinAp4CurFocal);
}
}
if (!strcmp (name, "LENSFRANGE"))
{
char *sp;
imgdata.lens.makernotes.MinFocal = imgdata.lens.makernotes.MaxFocal = atof(value);
sp = strrchr (value, ' ');
if (sp)
{
imgdata.lens.makernotes.MaxFocal = atof(sp);
if ((imgdata.lens.makernotes.MaxFocal + 0.17f) < imgdata.lens.makernotes.MinFocal)
my_swap (float, imgdata.lens.makernotes.MaxFocal, imgdata.lens.makernotes.MinFocal);
}
}
if (!strcmp (name, "LENSMODEL"))
{
imgdata.lens.makernotes.LensID = atoi(value);
if (imgdata.lens.makernotes.LensID)
imgdata.lens.makernotes.LensMount = Sigma_X3F;
}
}
#endif
}
#ifdef LOCALTIME
timestamp = mktime (gmtime (×tamp));
#endif
}
fseek (ifp, save, SEEK_SET);
}
}
//@out COMMON
/*
All matrices are from Adobe DNG Converter unless otherwise noted.
*/
void CLASS adobe_coeff (const char *t_make, const char *t_model
#ifdef LIBRAW_LIBRARY_BUILD
,int internal_only
#endif
)
{
static const struct {
const char *prefix;
int t_black, t_maximum, trans[12];
} table[] = {
{ "AgfaPhoto DC-833m", 0, 0, /* DJC */
{ 11438,-3762,-1115,-2409,9914,2497,-1227,2295,5300 } },
{ "Apple QuickTake", 0, 0, /* DJC */
{ 21392,-5653,-3353,2406,8010,-415,7166,1427,2078 } },
{ "Canon EOS D2000", 0, 0,
{ 24542,-10860,-3401,-1490,11370,-297,2858,-605,3225 } },
{ "Canon EOS D6000", 0, 0,
{ 20482,-7172,-3125,-1033,10410,-285,2542,226,3136 } },
{ "Canon EOS D30", 0, 0,
{ 9805,-2689,-1312,-5803,13064,3068,-2438,3075,8775 } },
{ "Canon EOS D60", 0, 0xfa0,
{ 6188,-1341,-890,-7168,14489,2937,-2640,3228,8483 } },
{ "Canon EOS 5DS", 0, 0x3c96,
{ 6250,-711,-808,-5153,12794,2636,-1249,2198,5610 } },
{ "Canon EOS 5D Mark III", 0, 0x3c80,
{ 6722,-635,-963,-4287,12460,2028,-908,2162,5668 } },
{ "Canon EOS 5D Mark II", 0, 0x3cf0,
{ 4716,603,-830,-7798,15474,2480,-1496,1937,6651 } },
{ "Canon EOS 5D", 0, 0xe6c,
{ 6347,-479,-972,-8297,15954,2480,-1968,2131,7649 } },
{ "Canon EOS 6D", 0, 0x3c82,
{ 8621,-2197,-787,-3150,11358,912,-1161,2400,4836 } },
{ "Canon EOS 7D Mark II", 0, 0x3510,
{ 7268,-1082,-969,-4186,11839,2663,-825,2029,5839 } },
{ "Canon EOS 7D", 0, 0x3510,
{ 6844,-996,-856,-3876,11761,2396,-593,1772,6198 } },
{ "Canon EOS 10D", 0, 0xfa0,
{ 8197,-2000,-1118,-6714,14335,2592,-2536,3178,8266 } },
{ "Canon EOS 20Da", 0, 0,
{ 14155,-5065,-1382,-6550,14633,2039,-1623,1824,6561 } },
{ "Canon EOS 20D", 0, 0xfff,
{ 6599,-537,-891,-8071,15783,2424,-1983,2234,7462 } },
{ "Canon EOS 30D", 0, 0,
{ 6257,-303,-1000,-7880,15621,2396,-1714,1904,7046 } },
{ "Canon EOS 40D", 0, 0x3f60,
{ 6071,-747,-856,-7653,15365,2441,-2025,2553,7315 } },
{ "Canon EOS 50D", 0, 0x3d93,
{ 4920,616,-593,-6493,13964,2784,-1774,3178,7005 } },
{ "Canon EOS 60D", 0, 0x2ff7,
{ 6719,-994,-925,-4408,12426,2211,-887,2129,6051 } },
{ "Canon EOS 70D", 0, 0x3bc7,
{ 7034,-804,-1014,-4420,12564,2058,-851,1994,5758 } },
{ "Canon EOS 100D", 0, 0x350f,
{ 6602,-841,-939,-4472,12458,2247,-975,2039,6148 } },
{ "Canon EOS 300D", 0, 0xfa0,
{ 8197,-2000,-1118,-6714,14335,2592,-2536,3178,8266 } },
{ "Canon EOS 350D", 0, 0xfff,
{ 6018,-617,-965,-8645,15881,2975,-1530,1719,7642 } },
{ "Canon EOS 400D", 0, 0xe8e,
{ 7054,-1501,-990,-8156,15544,2812,-1278,1414,7796 } },
{ "Canon EOS 450D", 0, 0x390d,
{ 5784,-262,-821,-7539,15064,2672,-1982,2681,7427 } },
{ "Canon EOS 500D", 0, 0x3479,
{ 4763,712,-646,-6821,14399,2640,-1921,3276,6561 } },
{ "Canon EOS 550D", 0, 0x3dd7,
{ 6941,-1164,-857,-3825,11597,2534,-416,1540,6039 } },
{ "Canon EOS 600D", 0, 0x3510,
{ 6461,-907,-882,-4300,12184,2378,-819,1944,5931 } },
{ "Canon EOS 650D", 0, 0x354d,
{ 6602,-841,-939,-4472,12458,2247,-975,2039,6148 } },
{ "Canon EOS 750D", 0, 0x3c00,
{ 6362,-823,-847,-4426,12109,2616,-743,1857,5635 } },
{ "Canon EOS 760D", 0, 0x3c00,
{ 6362,-823,-847,-4426,12109,2616,-743,1857,5635 } },
{ "Canon EOS 700D", 0, 0x3c00,
{ 6602,-841,-939,-4472,12458,2247,-975,2039,6148 } },
{ "Canon EOS 1000D", 0, 0xe43,
{ 6771,-1139,-977,-7818,15123,2928,-1244,1437,7533 } },
{ "Canon EOS 1100D", 0, 0x3510,
{ 6444,-904,-893,-4563,12308,2535,-903,2016,6728 } },
{ "Canon EOS 1200D", 0, 0x37c2,
{ 6461,-907,-882,-4300,12184,2378,-819,1944,5931 } },
{ "Canon EOS M3", 0, 0,
{ 6362,-823,-847,-4426,12109,2616,-743,1857,5635 } },
{ "Canon EOS M", 0, 0,
{ 6602,-841,-939,-4472,12458,2247,-975,2039,6148 } },
{ "Canon EOS-1Ds Mark III", 0, 0x3bb0,
{ 5859,-211,-930,-8255,16017,2353,-1732,1887,7448 } },
{ "Canon EOS-1Ds Mark II", 0, 0xe80,
{ 6517,-602,-867,-8180,15926,2378,-1618,1771,7633 } },
{ "Canon EOS-1D Mark IV", 0, 0x3bb0,
{ 6014,-220,-795,-4109,12014,2361,-561,1824,5787 } },
{ "Canon EOS-1D Mark III", 0, 0x3bb0,
{ 6291,-540,-976,-8350,16145,2311,-1714,1858,7326 } },
{ "Canon EOS-1D Mark II N", 0, 0xe80,
{ 6240,-466,-822,-8180,15825,2500,-1801,1938,8042 } },
{ "Canon EOS-1D Mark II", 0, 0xe80,
{ 6264,-582,-724,-8312,15948,2504,-1744,1919,8664 } },
{ "Canon EOS-1DS", 0, 0xe20,
{ 4374,3631,-1743,-7520,15212,2472,-2892,3632,8161 } },
{ "Canon EOS-1D C", 0, 0x3c4e,
{ 6847,-614,-1014,-4669,12737,2139,-1197,2488,6846 } },
{ "Canon EOS-1D X", 0, 0x3c4e,
{ 6847,-614,-1014,-4669,12737,2139,-1197,2488,6846 } },
{ "Canon EOS-1D", 0, 0xe20,
{ 6806,-179,-1020,-8097,16415,1687,-3267,4236,7690 } },
{ "Canon EOS C500", 853, 0, /* DJC */
{ 17851,-10604,922,-7425,16662,763,-3660,3636,22278 } },
{ "Canon PowerShot A530", 0, 0,
{ 0 } }, /* don't want the A5 matrix */
{ "Canon PowerShot A50", 0, 0,
{ -5300,9846,1776,3436,684,3939,-5540,9879,6200,-1404,11175,217 } },
{ "Canon PowerShot A5", 0, 0,
{ -4801,9475,1952,2926,1611,4094,-5259,10164,5947,-1554,10883,547 } },
{ "Canon PowerShot G10", 0, 0,
{ 11093,-3906,-1028,-5047,12492,2879,-1003,1750,5561 } },
{ "Canon PowerShot G11", 0, 0,
{ 12177,-4817,-1069,-1612,9864,2049,-98,850,4471 } },
{ "Canon PowerShot G12", 0, 0,
{ 13244,-5501,-1248,-1508,9858,1935,-270,1083,4366 } },
{ "Canon PowerShot G15", 0, 0,
{ 7474,-2301,-567,-4056,11456,2975,-222,716,4181 } },
{ "Canon PowerShot G16", 0, 0,
{ 14130,-8071,127,2199,6528,1551,3402,-1721,4960 } },
{ "Canon PowerShot G1 X Mark II", 0, 0,
{ 7378,-1255,-1043,-4088,12251,2048,-876,1946,5805 } },
{ "Canon PowerShot G1 X", 0, 0,
{ 7378,-1255,-1043,-4088,12251,2048,-876,1946,5805 } },
{ "Canon PowerShot G1", 0, 0,
{ -4778,9467,2172,4743,-1141,4344,-5146,9908,6077,-1566,11051,557 } },
{ "Canon PowerShot G2", 0, 0,
{ 9087,-2693,-1049,-6715,14382,2537,-2291,2819,7790 } },
{ "Canon PowerShot G3 X", 0, 0,
{ 9701,-3857,-921,-3149,11537,1817,-786,1817,5147 } },
{ "Canon PowerShot G3", 0, 0,
{ 9212,-2781,-1073,-6573,14189,2605,-2300,2844,7664 } },
{"Canon PowerShot G5 X",0, 0,
{ 9920,-3574,-806,-1943,9758,1942,-774,1848,4523 } }, /* LibRaw */
{ "Canon PowerShot G5", 0, 0,
{ 9757,-2872,-933,-5972,13861,2301,-1622,2328,7212 } },
{ "Canon PowerShot G6", 0, 0,
{ 9877,-3775,-871,-7613,14807,3072,-1448,1305,7485 } },
{ "Canon PowerShot G7 X", 0, 0,
{ 9602,-3823,-937,-2984,11495,1675,-407,1415,5049 } },
{"Canon PowerShot G9 X",0, 0,
{ 9659,-3410,-773,-1905,9734,1912,-626,1725,4711 } }, /* LibRaw */
{ "Canon PowerShot G9", 0, 0,
{ 7368,-2141,-598,-5621,13254,2625,-1418,1696,5743 } },
{ "Canon PowerShot Pro1", 0, 0,
{ 10062,-3522,-999,-7643,15117,2730,-765,817,7323 } },
{ "Canon PowerShot Pro70", 34, 0,
{ -4155,9818,1529,3939,-25,4522,-5521,9870,6610,-2238,10873,1342 } },
{ "Canon PowerShot Pro90", 0, 0,
{ -4963,9896,2235,4642,-987,4294,-5162,10011,5859,-1770,11230,577 } },
{ "Canon PowerShot S30", 0, 0,
{ 10566,-3652,-1129,-6552,14662,2006,-2197,2581,7670 } },
{ "Canon PowerShot S40", 0, 0,
{ 8510,-2487,-940,-6869,14231,2900,-2318,2829,9013 } },
{ "Canon PowerShot S45", 0, 0,
{ 8163,-2333,-955,-6682,14174,2751,-2077,2597,8041 } },
{ "Canon PowerShot S50", 0, 0,
{ 8882,-2571,-863,-6348,14234,2288,-1516,2172,6569 } },
{ "Canon PowerShot S60", 0, 0,
{ 8795,-2482,-797,-7804,15403,2573,-1422,1996,7082 } },
{ "Canon PowerShot S70", 0, 0,
{ 9976,-3810,-832,-7115,14463,2906,-901,989,7889 } },
{ "Canon PowerShot S90", 0, 0,
{ 12374,-5016,-1049,-1677,9902,2078,-83,852,4683 } },
{ "Canon PowerShot S95", 0, 0,
{ 13440,-5896,-1279,-1236,9598,1931,-180,1001,4651 } },
{ "Canon PowerShot S120", 0, 0,
{ 6961,-1685,-695,-4625,12945,1836,-1114,2152,5518 } },
{ "Canon PowerShot S110", 0, 0,
{ 8039,-2643,-654,-3783,11230,2930,-206,690,4194 } },
{ "Canon PowerShot S100", 0, 0,
{ 7968,-2565,-636,-2873,10697,2513,180,667,4211 } },
{ "Canon PowerShot SX1 IS", 0, 0,
{ 6578,-259,-502,-5974,13030,3309,-308,1058,4970 } },
{ "Canon PowerShot SX50 HS", 0, 0,
{ 12432,-4753,-1247,-2110,10691,1629,-412,1623,4926 } },
{ "Canon PowerShot SX60 HS", 0, 0,
{ 13161,-5451,-1344,-1989,10654,1531,-47,1271,4955 } },
{ "Canon PowerShot A3300", 0, 0, /* DJC */
{ 10826,-3654,-1023,-3215,11310,1906,0,999,4960 } },
{ "Canon PowerShot A470", 0, 0, /* DJC */
{ 12513,-4407,-1242,-2680,10276,2405,-878,2215,4734 } },
{ "Canon PowerShot A610", 0, 0, /* DJC */
{ 15591,-6402,-1592,-5365,13198,2168,-1300,1824,5075 } },
{ "Canon PowerShot A620", 0, 0, /* DJC */
{ 15265,-6193,-1558,-4125,12116,2010,-888,1639,5220 } },
{ "Canon PowerShot A630", 0, 0, /* DJC */
{ 14201,-5308,-1757,-6087,14472,1617,-2191,3105,5348 } },
{ "Canon PowerShot A640", 0, 0, /* DJC */
{ 13124,-5329,-1390,-3602,11658,1944,-1612,2863,4885 } },
{ "Canon PowerShot A650", 0, 0, /* DJC */
{ 9427,-3036,-959,-2581,10671,1911,-1039,1982,4430 } },
{ "Canon PowerShot A720", 0, 0, /* DJC */
{ 14573,-5482,-1546,-1266,9799,1468,-1040,1912,3810 } },
{ "Canon PowerShot S3 IS", 0, 0, /* DJC */
{ 14062,-5199,-1446,-4712,12470,2243,-1286,2028,4836 } },
{ "Canon PowerShot SX110 IS", 0, 0, /* DJC */
{ 14134,-5576,-1527,-1991,10719,1273,-1158,1929,3581 } },
{ "Canon PowerShot SX220", 0, 0, /* DJC */
{ 13898,-5076,-1447,-1405,10109,1297,-244,1860,3687 } },
{ "Casio EX-S20", 0, 0, /* DJC */
{ 11634,-3924,-1128,-4968,12954,2015,-1588,2648,7206 } },
{ "Casio EX-Z750", 0, 0, /* DJC */
{ 10819,-3873,-1099,-4903,13730,1175,-1755,3751,4632 } },
{ "Casio EX-Z10", 128, 0xfff, /* DJC */
{ 9790,-3338,-603,-2321,10222,2099,-344,1273,4799 } },
{ "CINE 650", 0, 0,
{ 3390,480,-500,-800,3610,340,-550,2336,1192 } },
{ "CINE 660", 0, 0,
{ 3390,480,-500,-800,3610,340,-550,2336,1192 } },
{ "CINE", 0, 0,
{ 20183,-4295,-423,-3940,15330,3985,-280,4870,9800 } },
{ "Contax N Digital", 0, 0xf1e,
{ 7777,1285,-1053,-9280,16543,2916,-3677,5679,7060 } },
{ "Epson R-D1", 0, 0,
{ 6827,-1878,-732,-8429,16012,2564,-704,592,7145 } },
{ "Fujifilm E550", 0, 0,
{ 11044,-3888,-1120,-7248,15168,2208,-1531,2277,8069 } },
{ "Fujifilm E900", 0, 0,
{ 9183,-2526,-1078,-7461,15071,2574,-2022,2440,8639 } },
{ "Fujifilm F5", 0, 0,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm F6", 0, 0,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm F77", 0, 0xfe9,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm F7", 0, 0,
{ 10004,-3219,-1201,-7036,15047,2107,-1863,2565,7736 } },
{ "Fujifilm F8", 0, 0,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm S100FS", 514, 0,
{ 11521,-4355,-1065,-6524,13767,3058,-1466,1984,6045 } },
{ "Fujifilm S1", 0, 0,
{ 12297,-4882,-1202,-2106,10691,1623,-88,1312,4790 } },
{ "Fujifilm S20Pro", 0, 0,
{ 10004,-3219,-1201,-7036,15047,2107,-1863,2565,7736 } },
{ "Fujifilm S20", 512, 0x3fff,
{ 11401,-4498,-1312,-5088,12751,2613,-838,1568,5941 } },
{ "Fujifilm S2Pro", 128, 0,
{ 12492,-4690,-1402,-7033,15423,1647,-1507,2111,7697 } },
{ "Fujifilm S3Pro", 0, 0,
{ 11807,-4612,-1294,-8927,16968,1988,-2120,2741,8006 } },
{ "Fujifilm S5Pro", 0, 0,
{ 12300,-5110,-1304,-9117,17143,1998,-1947,2448,8100 } },
{ "Fujifilm S5000", 0, 0,
{ 8754,-2732,-1019,-7204,15069,2276,-1702,2334,6982 } },
{ "Fujifilm S5100", 0, 0,
{ 11940,-4431,-1255,-6766,14428,2542,-993,1165,7421 } },
{ "Fujifilm S5500", 0, 0,
{ 11940,-4431,-1255,-6766,14428,2542,-993,1165,7421 } },
{ "Fujifilm S5200", 0, 0,
{ 9636,-2804,-988,-7442,15040,2589,-1803,2311,8621 } },
{ "Fujifilm S5600", 0, 0,
{ 9636,-2804,-988,-7442,15040,2589,-1803,2311,8621 } },
{ "Fujifilm S6", 0, 0,
{ 12628,-4887,-1401,-6861,14996,1962,-2198,2782,7091 } },
{ "Fujifilm S7000", 0, 0,
{ 10190,-3506,-1312,-7153,15051,2238,-2003,2399,7505 } },
{ "Fujifilm S9000", 0, 0,
{ 10491,-3423,-1145,-7385,15027,2538,-1809,2275,8692 } },
{ "Fujifilm S9500", 0, 0,
{ 10491,-3423,-1145,-7385,15027,2538,-1809,2275,8692 } },
{ "Fujifilm S9100", 0, 0,
{ 12343,-4515,-1285,-7165,14899,2435,-1895,2496,8800 } },
{ "Fujifilm S9600", 0, 0,
{ 12343,-4515,-1285,-7165,14899,2435,-1895,2496,8800 } },
{ "Fujifilm SL1000", 0, 0,
{ 11705,-4262,-1107,-2282,10791,1709,-555,1713,4945 } },
{ "Fujifilm IS-1", 0, 0,
{ 21461,-10807,-1441,-2332,10599,1999,289,875,7703 } },
{ "Fujifilm IS Pro", 0, 0,
{ 12300,-5110,-1304,-9117,17143,1998,-1947,2448,8100 } },
{ "Fujifilm HS10 HS11", 0, 0xf68,
{ 12440,-3954,-1183,-1123,9674,1708,-83,1614,4086 } },
{ "Fujifilm HS2", 0, 0,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm HS3", 0, 0,
{ 13690,-5358,-1474,-3369,11600,1998,-132,1554,4395 } },
{ "Fujifilm HS50EXR", 0, 0,
{ 12085,-4727,-953,-3257,11489,2002,-511,2046,4592 } },
{ "Fujifilm F900EXR", 0, 0,
{ 12085,-4727,-953,-3257,11489,2002,-511,2046,4592 } },
{ "Fujifilm X100S", 0, 0,
{ 10592,-4262,-1008,-3514,11355,2465,-870,2025,6386 } },
{ "Fujifilm X100T", 0, 0,
{ 10592,-4262,-1008,-3514,11355,2465,-870,2025,6386 } },
{ "Fujifilm X100", 0, 0,
{ 12161,-4457,-1069,-5034,12874,2400,-795,1724,6904 } },
{ "Fujifilm X10", 0, 0,
{ 13509,-6199,-1254,-4430,12733,1865,-331,1441,5022 } },
{ "Fujifilm X20", 0, 0,
{ 11768,-4971,-1133,-4904,12927,2183,-480,1723,4605 } },
{ "Fujifilm X30", 0, 0,
{ 12328,-5256,-1144,-4469,12927,1675,-87,1291,4351 } },
{ "Fujifilm X-Pro1", 0, 0,
{ 10413,-3996,-993,-3721,11640,2361,-733,1540,6011 } },
{ "Fujifilm X-A1", 0, 0,
{ 11086,-4555,-839,-3512,11310,2517,-815,1341,5940 } },
{ "Fujifilm X-A2", 0, 0,
{ 10763,-4560,-917,-3346,11311,2322,-475,1135,5843 } },
{ "Fujifilm X-E1", 0, 0,
{ 10413,-3996,-993,-3721,11640,2361,-733,1540,6011 } },
{ "Fujifilm X-E2", 0, 0,
{ 12066,-5927,-367,-1969,9878,1503,-721,2034,5453 } },
{ "Fujifilm XF1", 0, 0,
{ 13509,-6199,-1254,-4430,12733,1865,-331,1441,5022 } },
{ "Fujifilm X-M1", 0, 0,
{ 13193,-6685,-425,-2229,10458,1534,-878,1763,5217 } },
{ "Fujifilm X-S1", 0, 0,
{ 13509,-6199,-1254,-4430,12733,1865,-331,1441,5022 } },
{ "Fujifilm X-T10", 0, 0,
{ 10763,-4560,-917,-3346,11311,2322,-475,1135,5843 }},
{ "Fujifilm X-T1", 0, 0,
{ 8458,-2451,-855,-4597,12447,2407,-1475,2482,6526 } },
{ "Fujifilm XQ1", 0, 0,
{ 9252,-2704,-1064,-5893,14265,1717,-1101,2341,4349 } },
{ "Fujifilm XQ2", 0, 0,
{ 9252,-2704,-1064,-5893,14265,1717,-1101,2341,4349 } },
{ "Hasselblad Lunar", -512, 0,
{ 5491,-1192,-363,-4951,12342,2948,-911,1722,7192 } },
{ "Hasselblad Stellar", -800, 0,
{ 8651,-2754,-1057,-3464,12207,1373,-568,1398,4434 } },
{ "Hasselblad CFV", 0, 0, /* Adobe */
{ 8519, -3260, -280, -5081, 13459, 1738, -1449, 2960, 7809, } },
{ "Hasselblad H-16MP", 0, 0, /* LibRaw */
{ 17765,-5322,-1734,-6168,13354,2135,-264,2524,7440 } },
{ "Hasselblad H-22MP", 0, 0, /* LibRaw */
{ 17765,-5322,-1734,-6168,13354,2135,-264,2524,7440 } },
{ "Hasselblad H-31MP",0, 0, /* LibRaw */
{ 14480,-5448,-1686,-3534,13123,2260,384,2952,7232 } },
{ "Hasselblad H-39MP",0, 0, /* Adobe */
{ 3857,452, -46, -6008, 14477, 1596, -2627, 4481, 5718 } },
{ "Hasselblad H3D-50", 0, 0, /* Adobe */
{ 3857,452, -46, -6008, 14477, 1596, -2627, 4481, 5718 } },
{ "Hasselblad H4D-40",0, 0, /* LibRaw */
{ 6325,-860,-957,-6559,15945,266,167,770,5936 } },
{ "Hasselblad H4D-50",0, 0, /* LibRaw */
{ 15283,-6272,-465,-2030,16031,478,-2379,390,7965 } },
{ "Hasselblad H4D-60",0, 0, /* Adobe */
{ 9662, -684, -279, -4903, 12293, 2950, -344, 1669, 6024 } },
{ "Hasselblad H5D-50c",0, 0, /* Adobe */
{ 4932, -835, 141, -4878, 11868, 3437, -1138, 1961, 7067 } },
{ "Hasselblad H5D-50",0, 0, /* Adobe */
{ 5656, -659, -346, -3923, 12306, 1791, -1602, 3509, 5442 } },
{ "Imacon Ixpress", 0, 0, /* DJC */
{ 7025,-1415,-704,-5188,13765,1424,-1248,2742,6038 } },
{ "Kodak NC2000", 0, 0,
{ 13891,-6055,-803,-465,9919,642,2121,82,1291 } },
{ "Kodak DCS315C", -8, 0,
{ 17523,-4827,-2510,756,8546,-137,6113,1649,2250 } },
{ "Kodak DCS330C", -8, 0,
{ 20620,-7572,-2801,-103,10073,-396,3551,-233,2220 } },
{ "Kodak DCS420", 0, 0,
{ 10868,-1852,-644,-1537,11083,484,2343,628,2216 } },
{ "Kodak DCS460", 0, 0,
{ 10592,-2206,-967,-1944,11685,230,2206,670,1273 } },
{ "Kodak EOSDCS1", 0, 0,
{ 10592,-2206,-967,-1944,11685,230,2206,670,1273 } },
{ "Kodak EOSDCS3B", 0, 0,
{ 9898,-2700,-940,-2478,12219,206,1985,634,1031 } },
{ "Kodak DCS520C", -178, 0,
{ 24542,-10860,-3401,-1490,11370,-297,2858,-605,3225 } },
{ "Kodak DCS560C", -177, 0,
{ 20482,-7172,-3125,-1033,10410,-285,2542,226,3136 } },
{ "Kodak DCS620C", -177, 0,
{ 23617,-10175,-3149,-2054,11749,-272,2586,-489,3453 } },
{ "Kodak DCS620X", -176, 0,
{ 13095,-6231,154,12221,-21,-2137,895,4602,2258 } },
{ "Kodak DCS660C", -173, 0,
{ 18244,-6351,-2739,-791,11193,-521,3711,-129,2802 } },
{ "Kodak DCS720X", 0, 0,
{ 11775,-5884,950,9556,1846,-1286,-1019,6221,2728 } },
{ "Kodak DCS760C", 0, 0,
{ 16623,-6309,-1411,-4344,13923,323,2285,274,2926 } },
{ "Kodak DCS Pro SLR", 0, 0,
{ 5494,2393,-232,-6427,13850,2846,-1876,3997,5445 } },
{ "Kodak DCS Pro 14nx", 0, 0,
{ 5494,2393,-232,-6427,13850,2846,-1876,3997,5445 } },
{ "Kodak DCS Pro 14", 0, 0,
{ 7791,3128,-776,-8588,16458,2039,-2455,4006,6198 } },
{ "Kodak ProBack645", 0, 0,
{ 16414,-6060,-1470,-3555,13037,473,2545,122,4948 } },
{ "Kodak ProBack", 0, 0,
{ 21179,-8316,-2918,-915,11019,-165,3477,-180,4210 } },
{ "Kodak P712", 0, 0,
{ 9658,-3314,-823,-5163,12695,2768,-1342,1843,6044 } },
{ "Kodak P850", 0, 0xf7c,
{ 10511,-3836,-1102,-6946,14587,2558,-1481,1792,6246 } },
{ "Kodak P880", 0, 0xfff,
{ 12805,-4662,-1376,-7480,15267,2360,-1626,2194,7904 } },
{ "Kodak EasyShare Z980", 0, 0,
{ 11313,-3559,-1101,-3893,11891,2257,-1214,2398,4908 } },
{ "Kodak EasyShare Z981", 0, 0,
{ 12729,-4717,-1188,-1367,9187,2582,274,860,4411 } },
{ "Kodak EasyShare Z990", 0, 0xfed,
{ 11749,-4048,-1309,-1867,10572,1489,-138,1449,4522 } },
{ "Kodak EASYSHARE Z1015", 0, 0xef1,
{ 11265,-4286,-992,-4694,12343,2647,-1090,1523,5447 } },
{ "Leaf CMost", 0, 0,
{ 3952,2189,449,-6701,14585,2275,-4536,7349,6536 } },
{ "Leaf Valeo 6", 0, 0,
{ 3952,2189,449,-6701,14585,2275,-4536,7349,6536 } },
{ "Leaf Aptus 54S", 0, 0,
{ 8236,1746,-1314,-8251,15953,2428,-3673,5786,5771 } },
{ "Leaf Aptus 65", 0, 0,
{ 7914,1414,-1190,-8777,16582,2280,-2811,4605,5562 } },
{ "Leaf Aptus 75", 0, 0,
{ 7914,1414,-1190,-8777,16582,2280,-2811,4605,5562 } },
{ "Leaf Credo 40", 0, 0,
{ 8035, 435, -962, -6001, 13872, 2320, -1159, 3065, 5434 } },
{ "Leaf Credo 50", 0, 0,
{ 3984, 0, 0, 0, 10000, 0, 0, 0, 7666 } },
{ "Leaf Credo 60", 0, 0,
{ 8035, 435, -962, -6001, 13872,2320,-1159,3065,5434 } },
{ "Leaf Credo 80", 0, 0,
{ 6294, 686, -712, -5435, 13417, 2211, -1006, 2435, 5042 } },
{ "Leaf", 0, 0,
{ 8236,1746,-1314,-8251,15953,2428,-3673,5786,5771 } },
{ "Mamiya ZD", 0, 0,
{ 7645,2579,-1363,-8689,16717,2015,-3712,5941,5961 } },
{ "Micron 2010", 110, 0, /* DJC */
{ 16695,-3761,-2151,155,9682,163,3433,951,4904 } },
{ "Minolta DiMAGE 5", 0, 0xf7d,
{ 8983,-2942,-963,-6556,14476,2237,-2426,2887,8014 } },
{ "Minolta DiMAGE 7Hi", 0, 0xf7d,
{ 11368,-3894,-1242,-6521,14358,2339,-2475,3056,7285 } },
{ "Minolta DiMAGE 7", 0, 0xf7d,
{ 9144,-2777,-998,-6676,14556,2281,-2470,3019,7744 } },
{ "Minolta DiMAGE A1", 0, 0xf8b,
{ 9274,-2547,-1167,-8220,16323,1943,-2273,2720,8340 } },
{ "Minolta DiMAGE A200", 0, 0,
{ 8560,-2487,-986,-8112,15535,2771,-1209,1324,7743 } },
{ "Minolta DiMAGE A2", 0, 0xf8f,
{ 9097,-2726,-1053,-8073,15506,2762,-966,981,7763 } },
{ "Minolta DiMAGE Z2", 0, 0, /* DJC */
{ 11280,-3564,-1370,-4655,12374,2282,-1423,2168,5396 } },
{ "Minolta DYNAX 5", 0, 0xffb,
{ 10284,-3283,-1086,-7957,15762,2316,-829,882,6644 } },
{ "Minolta DYNAX 7", 0, 0xffb,
{ 10239,-3104,-1099,-8037,15727,2451,-927,925,6871 } },
{ "Motorola PIXL", 0, 0, /* DJC */
{ 8898,-989,-1033,-3292,11619,1674,-661,3178,5216 } },
{ "Nikon D100", 0, 0,
{ 5902,-933,-782,-8983,16719,2354,-1402,1455,6464 } },
{ "Nikon D1H", 0, 0,
{ 7577,-2166,-926,-7454,15592,1934,-2377,2808,8606 } },
{ "Nikon D1X", 0, 0,
{ 7702,-2245,-975,-9114,17242,1875,-2679,3055,8521 } },
{ "Nikon D1", 0, 0, /* multiplied by 2.218750, 1.0, 1.148438 */
{ 16772,-4726,-2141,-7611,15713,1972,-2846,3494,9521 } },
{ "Nikon D200", 0, 0xfbc,
{ 8367,-2248,-763,-8758,16447,2422,-1527,1550,8053 } },
{ "Nikon D2H", 0, 0,
{ 5710,-901,-615,-8594,16617,2024,-2975,4120,6830 } },
{ "Nikon D2X", 0, 0,
{ 10231,-2769,-1255,-8301,15900,2552,-797,680,7148 } },
{ "Nikon D3000", 0, 0,
{ 8736,-2458,-935,-9075,16894,2251,-1354,1242,8263 } },
{ "Nikon D3100", 0, 0,
{ 7911,-2167,-813,-5327,13150,2408,-1288,2483,7968 } },
{ "Nikon D3200", 0, 0xfb9,
{ 7013,-1408,-635,-5268,12902,2640,-1470,2801,7379 } },
{ "Nikon D3300", 0, 0,
{ 6988,-1384,-714,-5631,13410,2447,-1485,2204,7318 } },
{ "Nikon D300", 0, 0,
{ 9030,-1992,-715,-8465,16302,2255,-2689,3217,8069 } },
{ "Nikon D3X", 0, 0,
{ 7171,-1986,-648,-8085,15555,2718,-2170,2512,7457 } },
{ "Nikon D3S", 0, 0,
{ 8828,-2406,-694,-4874,12603,2541,-660,1509,7587 } },
{ "Nikon D3", 0, 0,
{ 8139,-2171,-663,-8747,16541,2295,-1925,2008,8093 } },
{ "Nikon D40X", 0, 0,
{ 8819,-2543,-911,-9025,16928,2151,-1329,1213,8449 } },
{ "Nikon D40", 0, 0,
{ 6992,-1668,-806,-8138,15748,2543,-874,850,7897 } },
{ "Nikon D4S", 0, 0,
{ 8598,-2848,-857,-5618,13606,2195,-1002,1773,7137 } },
{ "Nikon D4", 0, 0,
{ 8598,-2848,-857,-5618,13606,2195,-1002,1773,7137 } },
{ "Nikon Df", 0, 0,
{ 8598,-2848,-857,-5618,13606,2195,-1002,1773,7137 } },
{ "Nikon D5000", 0, 0xf00,
{ 7309,-1403,-519,-8474,16008,2622,-2433,2826,8064 } },
{ "Nikon D5100", 0, 0x3de6,
{ 8198,-2239,-724,-4871,12389,2798,-1043,2050,7181 } },
{ "Nikon D5200", 0, 0,
{ 8322,-3112,-1047,-6367,14342,2179,-988,1638,6394 } },
{ "Nikon D5300", 0, 0,
{ 6988,-1384,-714,-5631,13410,2447,-1485,2204,7318 } },
{ "Nikon D5500", 0, 0,
{ 8821,-2938,-785,-4178,12142,2287,-824,1651,6860 } },
{ "Nikon D50", 0, 0,
{ 7732,-2422,-789,-8238,15884,2498,-859,783,7330 } },
{ "Nikon D600", 0, 0x3e07,
{ 8178,-2245,-609,-4857,12394,2776,-1207,2086,7298 } },
{ "Nikon D610",0, 0,
{ 10426,-4005,-444,-3565,11764,1403,-1206,2266,6549 } },
{ "Nikon D60", 0, 0,
{ 8736,-2458,-935,-9075,16894,2251,-1354,1242,8263 } },
{ "Nikon D7000", 0, 0,
{ 8198,-2239,-724,-4871,12389,2798,-1043,2050,7181 } },
{ "Nikon D7100", 0, 0,
{ 8322,-3112,-1047,-6367,14342,2179,-988,1638,6394 } },
{ "Nikon D7200", 0, 0,
{ 8322,-3112,-1047,-6367,14342,2179,-988,1638,6394 } },
{ "Nikon D750", 0, 0,
{ 9020,-2890,-715,-4535,12436,2348,-934,1919,7086 } },
{ "Nikon D700", 0, 0,
{ 8139,-2171,-663,-8747,16541,2295,-1925,2008,8093 } },
{ "Nikon D70", 0, 0,
{ 7732,-2422,-789,-8238,15884,2498,-859,783,7330 } },
{ "Nikon D810A", 0, 0,
{ 11973, -5685, -888, -1965, 10326, 1901, -115, 1123, 7169 }},
{ "Nikon D810", 0, 0,
{ 9369,-3195,-791,-4488,12430,2301,-893,1796,6872 } },
{ "Nikon D800", 0, 0,
{ 7866,-2108,-555,-4869,12483,2681,-1176,2069,7501 } },
{ "Nikon D80", 0, 0,
{ 8629,-2410,-883,-9055,16940,2171,-1490,1363,8520 } },
{ "Nikon D90", 0, 0xf00,
{ 7309,-1403,-519,-8474,16008,2622,-2434,2826,8064 } },
{ "Nikon E700", 0, 0x3dd, /* DJC */
{ -3746,10611,1665,9621,-1734,2114,-2389,7082,3064,3406,6116,-244 } },
{ "Nikon E800", 0, 0x3dd, /* DJC */
{ -3746,10611,1665,9621,-1734,2114,-2389,7082,3064,3406,6116,-244 } },
{ "Nikon E950", 0, 0x3dd, /* DJC */
{ -3746,10611,1665,9621,-1734,2114,-2389,7082,3064,3406,6116,-244 } },
{ "Nikon E995", 0, 0, /* copied from E5000 */
{ -5547,11762,2189,5814,-558,3342,-4924,9840,5949,688,9083,96 } },
{ "Nikon E2100", 0, 0, /* copied from Z2, new white balance */
{ 13142,-4152,-1596,-4655,12374,2282,-1769,2696,6711 } },
{ "Nikon E2500", 0, 0,
{ -5547,11762,2189,5814,-558,3342,-4924,9840,5949,688,9083,96 } },
{ "Nikon E3200", 0, 0, /* DJC */
{ 9846,-2085,-1019,-3278,11109,2170,-774,2134,5745 } },
{ "Nikon E4300", 0, 0, /* copied from Minolta DiMAGE Z2 */
{ 11280,-3564,-1370,-4655,12374,2282,-1423,2168,5396 } },
{ "Nikon E4500", 0, 0,
{ -5547,11762,2189,5814,-558,3342,-4924,9840,5949,688,9083,96 } },
{ "Nikon E5000", 0, 0,
{ -5547,11762,2189,5814,-558,3342,-4924,9840,5949,688,9083,96 } },
{ "Nikon E5400", 0, 0,
{ 9349,-2987,-1001,-7919,15766,2266,-2098,2680,6839 } },
{ "Nikon E5700", 0, 0,
{ -5368,11478,2368,5537,-113,3148,-4969,10021,5782,778,9028,211 } },
{ "Nikon E8400", 0, 0,
{ 7842,-2320,-992,-8154,15718,2599,-1098,1342,7560 } },
{ "Nikon E8700", 0, 0,
{ 8489,-2583,-1036,-8051,15583,2643,-1307,1407,7354 } },
{ "Nikon E8800", 0, 0,
{ 7971,-2314,-913,-8451,15762,2894,-1442,1520,7610 } },
{ "Nikon COOLPIX A", 0, 0,
{ 8198,-2239,-724,-4871,12389,2798,-1043,2050,7181 } },
{ "Nikon COOLPIX P330", -200, 0,
{ 10321,-3920,-931,-2750,11146,1824,-442,1545,5539 } },
{ "Nikon COOLPIX P340", -200, 0,
{ 10321,-3920,-931,-2750,11146,1824,-442,1545,5539 } },
{ "Nikon COOLPIX P6000", 0, 0,
{ 9698,-3367,-914,-4706,12584,2368,-837,968,5801 } },
{ "Nikon COOLPIX P7000", 0, 0,
{ 11432,-3679,-1111,-3169,11239,2202,-791,1380,4455 } },
{ "Nikon COOLPIX P7100", 0, 0,
{ 11053,-4269,-1024,-1976,10182,2088,-526,1263,4469 } },
{ "Nikon COOLPIX P7700", -3200, 0,
{ 10321,-3920,-931,-2750,11146,1824,-442,1545,5539 } },
{ "Nikon COOLPIX P7800", -3200, 0,
{ 10321,-3920,-931,-2750,11146,1824,-442,1545,5539 } },
{ "Nikon 1 V3", -200, 0,
{ 5958,-1559,-571,-4021,11453,2939,-634,1548,5087 } },
{ "Nikon 1 J4", 0, 0,
{ 5958,-1559,-571,-4021,11453,2939,-634,1548,5087 } },
{ "Nikon 1 J5", 0, 0,
{ 7520,-2518,-645,-3844,12102,1945,-913,2249,6835} },
{ "Nikon 1 S2", -200, 0,
{ 6612,-1342,-618,-3338,11055,2623,-174,1792,5075 } },
{ "Nikon 1 V2", 0, 0,
{ 6588,-1305,-693,-3277,10987,2634,-355,2016,5106 } },
{ "Nikon 1 J3", 0, 0,
{ 8144,-2671,-473,-1740,9834,1601,-58,1971,4296 } },
{ "Nikon 1 AW1", 0, 0,
{ 6588,-1305,-693,-3277,10987,2634,-355,2016,5106 } },
{ "Nikon 1 ", 0, 0, /* J1, J2, S1, V1 */
{ 8994,-2667,-865,-4594,12324,2552,-699,1786,6260 } },
{ "Olympus AIR-A01", 0, 0xfe1,
{ 8992,-3093,-639,-2563,10721,2122,-437,1270,5473 } },
{ "Olympus C5050", 0, 0,
{ 10508,-3124,-1273,-6079,14294,1901,-1653,2306,6237 } },
{ "Olympus C5060", 0, 0,
{ 10445,-3362,-1307,-7662,15690,2058,-1135,1176,7602 } },
{ "Olympus C7070", 0, 0,
{ 10252,-3531,-1095,-7114,14850,2436,-1451,1723,6365 } },
{ "Olympus C70", 0, 0,
{ 10793,-3791,-1146,-7498,15177,2488,-1390,1577,7321 } },
{ "Olympus C80", 0, 0,
{ 8606,-2509,-1014,-8238,15714,2703,-942,979,7760 } },
{ "Olympus E-10", 0, 0xffc,
{ 12745,-4500,-1416,-6062,14542,1580,-1934,2256,6603 } },
{ "Olympus E-1", 0, 0,
{ 11846,-4767,-945,-7027,15878,1089,-2699,4122,8311 } },
{ "Olympus E-20", 0, 0xffc,
{ 13173,-4732,-1499,-5807,14036,1895,-2045,2452,7142 } },
{ "Olympus E-300", 0, 0,
{ 7828,-1761,-348,-5788,14071,1830,-2853,4518,6557 } },
{ "Olympus E-330", 0, 0,
{ 8961,-2473,-1084,-7979,15990,2067,-2319,3035,8249 } },
{ "Olympus E-30", 0, 0xfbc,
{ 8144,-1861,-1111,-7763,15894,1929,-1865,2542,7607 } },
{ "Olympus E-3", 0, 0xf99,
{ 9487,-2875,-1115,-7533,15606,2010,-1618,2100,7389 } },
{ "Olympus E-400", 0, 0,
{ 6169,-1483,-21,-7107,14761,2536,-2904,3580,8568 } },
{ "Olympus E-410", 0, 0xf6a,
{ 8856,-2582,-1026,-7761,15766,2082,-2009,2575,7469 } },
{ "Olympus E-420", 0, 0xfd7,
{ 8746,-2425,-1095,-7594,15612,2073,-1780,2309,7416 } },
{ "Olympus E-450", 0, 0xfd2,
{ 8745,-2425,-1095,-7594,15613,2073,-1780,2309,7416 } },
{ "Olympus E-500", 0, 0,
{ 8136,-1968,-299,-5481,13742,1871,-2556,4205,6630 } },
{ "Olympus E-510", 0, 0xf6a,
{ 8785,-2529,-1033,-7639,15624,2112,-1783,2300,7817 } },
{ "Olympus E-520", 0, 0xfd2,
{ 8344,-2322,-1020,-7596,15635,2048,-1748,2269,7287 } },
{ "Olympus E-5", 0, 0xeec,
{ 11200,-3783,-1325,-4576,12593,2206,-695,1742,7504 } },
{ "Olympus E-600", 0, 0xfaf,
{ 8453,-2198,-1092,-7609,15681,2008,-1725,2337,7824 } },
{ "Olympus E-620", 0, 0xfaf,
{ 8453,-2198,-1092,-7609,15681,2008,-1725,2337,7824 } },
{ "Olympus E-P1", 0, 0xffd,
{ 8343,-2050,-1021,-7715,15705,2103,-1831,2380,8235 } },
{ "Olympus E-P2", 0, 0xffd,
{ 8343,-2050,-1021,-7715,15705,2103,-1831,2380,8235 } },
{ "Olympus E-P3", 0, 0,
{ 7575,-2159,-571,-3722,11341,2725,-1434,2819,6271 } },
{ "Olympus E-P5", 0, 0,
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus E-PL1s", 0, 0,
{ 11409,-3872,-1393,-4572,12757,2003,-709,1810,7415 } },
{ "Olympus E-PL1", 0, 0,
{ 11408,-4289,-1215,-4286,12385,2118,-387,1467,7787 } },
{ "Olympus E-PL2", 0, 0xcf3,
{ 15030,-5552,-1806,-3987,12387,1767,-592,1670,7023 } },
{ "Olympus E-PL3", 0, 0,
{ 7575,-2159,-571,-3722,11341,2725,-1434,2819,6271 } },
{ "Olympus E-PL5", 0, 0xfcb,
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus E-PL6", 0, 0,
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus E-PL7", 0, 0,
{ 9197,-3190,-659,-2606,10830,2039,-458,1250,5458 } },
{ "Olympus E-PM1", 0, 0,
{ 7575,-2159,-571,-3722,11341,2725,-1434,2819,6271 } },
{ "Olympus E-PM2", 0, 0,
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus E-M10", 0, 0, /* Same for E-M10MarkII */
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus E-M1", 0, 0,
{ 7687,-1984,-606,-4327,11928,2721,-1381,2339,6452 } },
{ "Olympus E-M5MarkII", 0, 0,
{ 9422,-3258,-711,-2655,10898,2015,-512,1354,5512 } },
{ "Olympus E-M5", 0, 0xfe1,
{ 8380,-2630,-639,-2887,10725,2496,-627,1427,5438 } },
{ "Olympus SP350", 0, 0,
{ 12078,-4836,-1069,-6671,14306,2578,-786,939,7418 } },
{ "Olympus SP3", 0, 0,
{ 11766,-4445,-1067,-6901,14421,2707,-1029,1217,7572 } },
{ "Olympus SP500UZ", 0, 0xfff,
{ 9493,-3415,-666,-5211,12334,3260,-1548,2262,6482 } },
{ "Olympus SP510UZ", 0, 0xffe,
{ 10593,-3607,-1010,-5881,13127,3084,-1200,1805,6721 } },
{ "Olympus SP550UZ", 0, 0xffe,
{ 11597,-4006,-1049,-5432,12799,2957,-1029,1750,6516 } },
{ "Olympus SP560UZ", 0, 0xff9,
{ 10915,-3677,-982,-5587,12986,2911,-1168,1968,6223 } },
{ "Olympus SP570UZ", 0, 0,
{ 11522,-4044,-1146,-4736,12172,2904,-988,1829,6039 } },
{"Olympus SH-2", 0, 0,
{10156,-3425,-1077,-2611,11177,1624,-385,1592,5080}},
{ "Olympus STYLUS1",0, 0,
{ 11976,-5518,-545,-1419,10472,846,-475,1766,4524 } },
{"Olympus TG-4", 0, 0,
{11426,-4159,-1126,-2066,10678,1593,-120,1327,4998}},
{ "Olympus XZ-10", 0, 0,
{ 9777,-3483,-925,-2886,11297,1800,-602,1663,5134 } },
{ "Olympus XZ-1", 0, 0,
{ 10901,-4095,-1074,-1141,9208,2293,-62,1417,5158 } },
{ "Olympus XZ-2", 0, 0,
{ 9777,-3483,-925,-2886,11297,1800,-602,1663,5134 } },
{ "OmniVision", 0, 0, /* DJC */
{ 12782,-4059,-379,-478,9066,1413,1340,1513,5176 } },
{ "Pentax *ist DL2", 0, 0,
{ 10504,-2438,-1189,-8603,16207,2531,-1022,863,12242 } },
{ "Pentax *ist DL", 0, 0,
{ 10829,-2838,-1115,-8339,15817,2696,-837,680,11939 } },
{ "Pentax *ist DS2", 0, 0,
{ 10504,-2438,-1189,-8603,16207,2531,-1022,863,12242 } },
{ "Pentax *ist DS", 0, 0,
{ 10371,-2333,-1206,-8688,16231,2602,-1230,1116,11282 } },
{ "Pentax *ist D", 0, 0,
{ 9651,-2059,-1189,-8881,16512,2487,-1460,1345,10687 } },
{ "Pentax K10D", 0, 0,
{ 9566,-2863,-803,-7170,15172,2112,-818,803,9705 } },
{ "Pentax K1", 0, 0,
{ 11095,-3157,-1324,-8377,15834,2720,-1108,947,11688 } },
{ "Pentax K20D", 0, 0,
{ 9427,-2714,-868,-7493,16092,1373,-2199,3264,7180 } },
{ "Pentax K200D", 0, 0,
{ 9186,-2678,-907,-8693,16517,2260,-1129,1094,8524 } },
{ "Pentax K2000", 0, 0,
{ 11057,-3604,-1155,-5152,13046,2329,-282,375,8104 } },
{ "Pentax K-m", 0, 0,
{ 11057,-3604,-1155,-5152,13046,2329,-282,375,8104 } },
{ "Pentax K-x", 0, 0,
{ 8843,-2837,-625,-5025,12644,2668,-411,1234,7410 } },
{ "Pentax K-r", 0, 0,
{ 9895,-3077,-850,-5304,13035,2521,-883,1768,6936 } },
{ "Pentax K-3 II", 0, 0,
{7415,-2052,-721,-5186,12788,2682,-1446,2157,6773 } },
{ "Pentax K-3", 0, 0,
{ 7415,-2052,-721,-5186,12788,2682,-1446,2157,6773 } },
{ "Pentax K-5 II", 0, 0,
{ 8170,-2725,-639,-4440,12017,2744,-771,1465,6599 } },
{ "Pentax K-5", 0, 0,
{ 8713,-2833,-743,-4342,11900,2772,-722,1543,6247 } },
{ "Pentax K-7", 0, 0,
{ 9142,-2947,-678,-8648,16967,1663,-2224,2898,8615 } },
{ "Pentax K-S1", 0, 0,
{ 8512,-3211,-787,-4167,11966,2487,-638,1288,6054 } },
{ "Pentax K-S2", 0, 0,
{ 8130,-2556,-1157,-3882,12350,1689,-843,1491,6305 } },
{ "Pentax Q-S1", 0, 0,
{ 12995,-5593,-1107,-1879,10139,2027,-64,1233,4919 } },
{ "Pentax MX-1", 0, 0,
{ 8804,-2523,-1238,-2423,11627,860,-682,1774,4753 } },
{ "Pentax Q10", 0, 0,
{ 12995,-5593,-1107,-1879,10139,2027,-64,1233,4919 } },
{ "Pentax 645D", 0, 0x3e00,
{ 10646,-3593,-1158,-3329,11699,1831,-667,2874,6287 } },
{ "Panasonic DMC-CM1", -15, 0,
{ 8770, -3194,-820,-2871,11281,1803,-513,1552,4434 } },
{ "Panasonic DMC-FZ8", 0, 0xf7f,
{ 8986,-2755,-802,-6341,13575,3077,-1476,2144,6379 } },
{ "Panasonic DMC-FZ18", 0, 0,
{ 9932,-3060,-935,-5809,13331,2753,-1267,2155,5575 } },
{ "Panasonic DMC-FZ28", -15, 0xf96,
{ 10109,-3488,-993,-5412,12812,2916,-1305,2140,5543 } },
{ "Panasonic DMC-FZ300", -15, 0xfff,
{ 8378,-2798,-769,-3068,11410,1877,-538,1792,4623 } },
{ "Panasonic DMC-FZ330", -15, 0xfff, // same as FZ300
{ 8378,-2798,-769,-3068,11410,1877,-538,1792,4623 } },
{ "Panasonic DMC-FZ30", 0, 0xf94,
{ 10976,-4029,-1141,-7918,15491,2600,-1670,2071,8246 } },
{ "Panasonic DMC-FZ3", -15, 0,
{ 9938,-2780,-890,-4604,12393,2480,-1117,2304,4620 } },
{ "Panasonic DMC-FZ4", -15, 0,
{ 13639,-5535,-1371,-1698,9633,2430,316,1152,4108 } },
{ "Panasonic DMC-FZ50", 0, 0,
{ 7906,-2709,-594,-6231,13351,3220,-1922,2631,6537 } },
{ "Panasonic DMC-FZ7", -15, 0,
{ 11532,-4324,-1066,-2375,10847,1749,-564,1699,4351 } },
{ "Leica V-LUX1", 0, 0,
{ 7906,-2709,-594,-6231,13351,3220,-1922,2631,6537 } },
{ "Panasonic DMC-L10", -15, 0xf96,
{ 8025,-1942,-1050,-7920,15904,2100,-2456,3005,7039 } },
{ "Panasonic DMC-L1", 0, 0xf7f,
{ 8054,-1885,-1025,-8349,16367,2040,-2805,3542,7629 } },
{ "Leica DIGILUX 3", 0, 0xf7f,
{ 8054,-1885,-1025,-8349,16367,2040,-2805,3542,7629 } },
{ "Panasonic DMC-LC1", 0, 0,
{ 11340,-4069,-1275,-7555,15266,2448,-2960,3426,7685 } },
{ "Leica DIGILUX 2", 0, 0,
{ 11340,-4069,-1275,-7555,15266,2448,-2960,3426,7685 } },
{ "Panasonic DMC-LX100", -15, 0,
{ 8844,-3538,-768,-3709,11762,2200,-698,1792,5220 } },
{ "Leica D-LUX (Typ 109)", -15, 0,
{ 8844,-3538,-768,-3709,11762,2200,-698,1792,5220 } },
{ "Panasonic DMC-LF1", -15, 0,
{ 9379,-3267,-816,-3227,11560,1881,-926,1928,5340 } },
{ "Leica C (Typ 112)", -15, 0,
{ 9379,-3267,-816,-3227,11560,1881,-926,1928,5340 } },
{ "Panasonic DMC-LX1", 0, 0xf7f,
{ 10704,-4187,-1230,-8314,15952,2501,-920,945,8927 } },
{ "Leica D-Lux (Typ 109)", 0, 0xf7f,
{ 8844,-3538,-768,-3709,11762,2200,-698,1792,5220 } },
{ "Leica D-LUX2", 0, 0xf7f,
{ 10704,-4187,-1230,-8314,15952,2501,-920,945,8927 } },
{ "Panasonic DMC-LX2", 0, 0,
{ 8048,-2810,-623,-6450,13519,3272,-1700,2146,7049 } },
{ "Leica D-LUX3", 0, 0,
{ 8048,-2810,-623,-6450,13519,3272,-1700,2146,7049 } },
{ "Panasonic DMC-LX3", -15, 0,
{ 8128,-2668,-655,-6134,13307,3161,-1782,2568,6083 } },
{ "Leica D-LUX 4", -15, 0,
{ 8128,-2668,-655,-6134,13307,3161,-1782,2568,6083 } },
{ "Panasonic DMC-LX5", -15, 0,
{ 10909,-4295,-948,-1333,9306,2399,22,1738,4582 } },
{ "Leica D-LUX 5", -15, 0,
{ 10909,-4295,-948,-1333,9306,2399,22,1738,4582 } },
{ "Panasonic DMC-LX7", -15, 0,
{ 10148,-3743,-991,-2837,11366,1659,-701,1893,4899 } },
{ "Leica D-LUX 6", -15, 0,
{ 10148,-3743,-991,-2837,11366,1659,-701,1893,4899 } },
{ "Panasonic DMC-FZ1000", -15, 0,
{ 7830,-2696,-763,-3325,11667,1866,-641,1712,4824 } },
{ "Leica V-LUX (Typ 114)", 15, 0,
{ 7830,-2696,-763,-3325,11667,1866,-641,1712,4824 } },
{ "Panasonic DMC-FZ100", -15, 0xfff,
{ 16197,-6146,-1761,-2393,10765,1869,366,2238,5248 } },
{ "Leica V-LUX 2", -15, 0xfff,
{ 16197,-6146,-1761,-2393,10765,1869,366,2238,5248 } },
{ "Panasonic DMC-FZ150", -15, 0xfff,
{ 11904,-4541,-1189,-2355,10899,1662,-296,1586,4289 } },
{ "Leica V-LUX 3", -15, 0xfff,
{ 11904,-4541,-1189,-2355,10899,1662,-296,1586,4289 } },
{ "Panasonic DMC-FZ200", -15, 0xfff,
{ 8112,-2563,-740,-3730,11784,2197,-941,2075,4933 } },
{ "Leica V-LUX 4", -15, 0xfff,
{ 8112,-2563,-740,-3730,11784,2197,-941,2075,4933 } },
{ "Panasonic DMC-FX150", -15, 0xfff,
{ 9082,-2907,-925,-6119,13377,3058,-1797,2641,5609 } },
{ "Panasonic DMC-G10", 0, 0,
{ 10113,-3400,-1114,-4765,12683,2317,-377,1437,6710 } },
{ "Panasonic DMC-G1", -15, 0xf94,
{ 8199,-2065,-1056,-8124,16156,2033,-2458,3022,7220 } },
{ "Panasonic DMC-G2", -15, 0xf3c,
{ 10113,-3400,-1114,-4765,12683,2317,-377,1437,6710 } },
{ "Panasonic DMC-G3", -15, 0xfff,
{ 6763,-1919,-863,-3868,11515,2684,-1216,2387,5879 } },
{ "Panasonic DMC-G5", -15, 0xfff,
{ 7798,-2562,-740,-3879,11584,2613,-1055,2248,5434 } },
{ "Panasonic DMC-G6", -15, 0xfff,
{ 8294,-2891,-651,-3869,11590,2595,-1183,2267,5352 } },
{ "Panasonic DMC-G7", -15, 0xfff,
{7610,-2780,-576,-4614,12195,2733,-1375,2393,6490 }},
{ "Panasonic DMC-GF1", -15, 0xf92,
{ 7888,-1902,-1011,-8106,16085,2099,-2353,2866,7330 } },
{ "Panasonic DMC-GF2", -15, 0xfff,
{ 7888,-1902,-1011,-8106,16085,2099,-2353,2866,7330 } },
{ "Panasonic DMC-GF3", -15, 0xfff,
{ 9051,-2468,-1204,-5212,13276,2121,-1197,2510,6890 } },
{ "Panasonic DMC-GF5", -15, 0xfff,
{ 8228,-2945,-660,-3938,11792,2430,-1094,2278,5793 } },
{ "Panasonic DMC-GF6", -15, 0,
{ 8130,-2801,-946,-3520,11289,2552,-1314,2511,5791 } },
{ "Panasonic DMC-GF7", -15, 0,
{ 7610,-2780,-576,-4614,12195,2733,-1375,2393,6490 } },
{ "Panasonic DMC-GH1", -15, 0xf92,
{ 6299,-1466,-532,-6535,13852,2969,-2331,3112,5984 } },
{ "Panasonic DMC-GH2", -15, 0xf95,
{ 7780,-2410,-806,-3913,11724,2484,-1018,2390,5298 } },
{ "Panasonic DMC-GH3", -15, 0,
{ 6559,-1752,-491,-3672,11407,2586,-962,1875,5130 } },
{ "Panasonic DMC-GH4", -15, 0,
{ 7122,-2108,-512,-3155,11201,2231,-541,1423,5045 } },
{ "Panasonic DMC-GM1", -15, 0,
{ 6770,-1895,-744,-5232,13145,2303,-1664,2691,5703 } },
{ "Panasonic DMC-GM5", -15, 0,
{ 8238,-3244,-679,-3921,11814,2384,-836,2022,5852 } },
{ "Panasonic DMC-GX1", -15, 0,
{ 6763,-1919,-863,-3868,11515,2684,-1216,2387,5879 } },
{ "Panasonic DMC-GX7", -15,0,
{ 7610,-2780,-576,-4614,12195,2733,-1375,2393,6490 } },
{ "Panasonic DMC-GX8", -15,0,
{ 7564,-2263,-606,-3148,11239,2177,-540,1435,4853 } },
{ "Panasonic DMC-TZ6", -15, 0,
{ 8607,-2822,-808,-3755,11930,2049,-820,2060,5224 } },
{ "Panasonic DMC-ZS4", -15, 0,
{ 8607,-2822,-808,-3755,11930,2049,-820,2060,5224 } },
{ "Panasonic DMC-TZ7", -15, 0,
{ 8802,-3135,-789,-3151,11468,1904,-550,1745,4810 } },
{ "Panasonic DMC-ZS5", -15, 0,
{ 8802,-3135,-789,-3151,11468,1904,-550,1745,4810 } },
{ "Phase One H 20", 0, 0, /* DJC */
{ 1313,1855,-109,-6715,15908,808,-327,1840,6020 } },
{ "Phase One H 25", 0, 0,
{ 2905,732,-237,-8134,16626,1476,-3038,4253,7517 } },
{ "Phase One IQ250",0, 0,
{ 4396,-153,-249,-5267,12249,2657,-1397,2323,6014 } },
{ "Phase One P 2", 0, 0,
{ 2905,732,-237,-8134,16626,1476,-3038,4253,7517 } },
{ "Phase One P 30", 0, 0,
{ 4516,-245,-37,-7020,14976,2173,-3206,4671,7087 } },
{ "Phase One P 45", 0, 0,
{ 5053,-24,-117,-5684,14076,1702,-2619,4492,5849 } },
{ "Phase One P40", 0, 0,
{ 8035,435,-962,-6001,13872,2320,-1159,3065,5434 } },
{ "Phase One P65", 0, 0,
{ 8035,435,-962,-6001,13872,2320,-1159,3065,5434 } },
{ "Photron BC2-HD", 0, 0, /* DJC */
{ 14603,-4122,-528,-1810,9794,2017,-297,2763,5936 } },
{ "Red One", 704, 0xffff, /* DJC */
{ 21014,-7891,-2613,-3056,12201,856,-2203,5125,8042 } },
{ "Samsung EK-GN120", 0, 0, /* Adobe; Galaxy NX */
{ 7557,-2522,-739,-4679,12949,1894,-840,1777,5311 } },
{ "Samsung EX1", 0, 0x3e00,
{ 8898,-2498,-994,-3144,11328,2066,-760,1381,4576 } },
{ "Samsung EX2F", 0, 0x7ff,
{ 10648,-3897,-1055,-2022,10573,1668,-492,1611,4742 } },
{ "Samsung NX mini", 0, 0,
{ 5222,-1196,-550,-6540,14649,2009,-1666,2819,5657 } },
{ "Samsung NX3000", 0, 0,
{ 8060,-2933,-761,-4504,12890,1762,-630,1489,5227 } },
{ "Samsung NX30", 0, 0, /* NX30, NX300, NX300M */
{ 7557,-2522,-739,-4679,12949,1894,-840,1777,5311 } },
{ "Samsung NX2000", 0, 0,
{ 7557,-2522,-739,-4679,12949,1894,-840,1777,5311 } },
{ "Samsung NX2", 0, 0xfff, /* NX20, NX200, NX210 */
{ 6933,-2268,-753,-4921,13387,1647,-803,1641,6096 } },
{ "Samsung NX1000", 0, 0,
{ 6933,-2268,-753,-4921,13387,1647,-803,1641,6096 } },
{ "Samsung NX1100", 0, 0,
{ 6933,-2268,-753,-4921,13387,1647,-803,1641,6096 } },
{ "Samsung NX11", 0, 0,
{ 10332,-3234,-1168,-6111,14639,1520,-1352,2647,8331 } },
{ "Samsung NX10", 0, 0, /* also NX100 */
{ 10332,-3234,-1168,-6111,14639,1520,-1352,2647,8331 } },
{ "Samsung NX500", 0, 0,
{ 10686,-4042,-1052,-3595,13238,276,-464,1259,5931 } },
{ "Samsung NX5", 0, 0,
{ 10332,-3234,-1168,-6111,14639,1520,-1352,2647,8331 } },
{ "Samsung NX1", 0, 0,
{ 10686,-4042,-1052,-3595,13238,276,-464,1259,5931 } },
{ "Samsung WB2000", 0, 0xfff,
{ 12093,-3557,-1155,-1000,9534,1733,-22,1787,4576 } },
{ "Samsung GX-1", 0, 0,
{ 10504,-2438,-1189,-8603,16207,2531,-1022,863,12242 } },
{ "Samsung GX20", 0, 0, /* copied from Pentax K20D */
{ 9427,-2714,-868,-7493,16092,1373,-2199,3264,7180 } },
{ "Samsung S85", 0, 0, /* DJC */
{ 11885,-3968,-1473,-4214,12299,1916,-835,1655,5549 } },
// Foveon: LibRaw color data
{ "Sigma dp0 Quattro", 2047, 0,
{ 13801,-3390,-1016,5535,3802,877,1848,4245,3730 } },
{ "Sigma dp1 Quattro", 2047, 0,
{ 13801,-3390,-1016,5535,3802,877,1848,4245,3730 } },
{ "Sigma dp2 Quattro", 2047, 0,
{ 13801,-3390,-1016,5535,3802,877,1848,4245,3730 } },
{ "Sigma dp3 Quattro", 2047, 0,
{ 13801,-3390,-1016,5535,3802,877,1848,4245,3730 } },
{ "Sigma SD9", 15, 4095, /* LibRaw */
{ 14082,-2201,-1056,-5243,14788,167,-121,196,8881 } },
{ "Sigma SD10", 15, 16383, /* LibRaw */
{ 14082,-2201,-1056,-5243,14788,167,-121,196,8881 } },
{ "Sigma SD14", 15, 16383, /* LibRaw */
{ 14082,-2201,-1056,-5243,14788,167,-121,196,8881 } },
{ "Sigma SD15", 15, 4095, /* LibRaw */
{ 14082,-2201,-1056,-5243,14788,167,-121,196,8881 } },
// Merills + SD1
{ "Sigma SD1", 31, 4095, /* LibRaw */
{ 5133,-1895,-353,4978,744,144,3837,3069,2777 } },
{ "Sigma DP1 Merrill", 31, 4095, /* LibRaw */
{ 5133,-1895,-353,4978,744,144,3837,3069,2777 } },
{ "Sigma DP2 Merrill", 31, 4095, /* LibRaw */
{ 5133,-1895,-353,4978,744,144,3837,3069,2777 } },
{ "Sigma DP3 Merrill", 31, 4095, /* LibRaw */
{ 5133,-1895,-353,4978,744,144,3837,3069,2777 } },
// Sigma DP (non-Merill Versions)
{ "Sigma DP", 0, 4095, /* LibRaw */
// { 7401,-1169,-567,2059,3769,1510,664,3367,5328 } },
{ 13100,-3638,-847,6855,2369,580,2723,3218,3251 } },
{ "Sinar", 0, 0, /* DJC */
{ 16442,-2956,-2422,-2877,12128,750,-1136,6066,4559 } },
{ "Sony DSC-F828", 0, 0,
{ 7924,-1910,-777,-8226,15459,2998,-1517,2199,6818,-7242,11401,3481 } },
{ "Sony DSC-R1", -512, 0,
{ 8512,-2641,-694,-8042,15670,2526,-1821,2117,7414 } },
{ "Sony DSC-V3", 0, 0,
{ 7511,-2571,-692,-7894,15088,3060,-948,1111,8128 } },
{ "Sony DSC-RX100M", -800, 0, /* M2 and M3 and M4 */
{ 6596,-2079,-562,-4782,13016,1933,-970,1581,5181 } },
{ "Sony DSC-RX100", -800, 0,
{ 8651,-2754,-1057,-3464,12207,1373,-568,1398,4434 } },
{ "Sony DSC-RX10",0, 0, /* And M2 too */
{ 6679,-1825,-745,-5047,13256,1953,-1580,2422,5183 } },
{ "Sony DSC-RX1R", -512, 0,
{ 8195,-2800,-422,-4261,12273,1709,-1505,2400,5624 } },
{ "Sony DSC-RX1", -512, 0,
{ 6344,-1612,-462,-4863,12477,2681,-865,1786,6899 } },
{ "Sony DSLR-A100", 0, 0xfeb,
{ 9437,-2811,-774,-8405,16215,2290,-710,596,7181 } },
{ "Sony DSLR-A290", 0, 0,
{ 6038,-1484,-579,-9145,16746,2512,-875,746,7218 } },
{ "Sony DSLR-A2", 0, 0,
{ 9847,-3091,-928,-8485,16345,2225,-715,595,7103 } },
{ "Sony DSLR-A300", 0, 0,
{ 9847,-3091,-928,-8485,16345,2225,-715,595,7103 } },
{ "Sony DSLR-A330", 0, 0,
{ 9847,-3091,-929,-8485,16346,2225,-714,595,7103 } },
{ "Sony DSLR-A350", 0, 0xffc,
{ 6038,-1484,-578,-9146,16746,2513,-875,746,7217 } },
{ "Sony DSLR-A380", 0, 0,
{ 6038,-1484,-579,-9145,16746,2512,-875,746,7218 } },
{ "Sony DSLR-A390", 0, 0,
{ 6038,-1484,-579,-9145,16746,2512,-875,746,7218 } },
{ "Sony DSLR-A450", -512, 0xfeb,
{ 4950,-580,-103,-5228,12542,3029,-709,1435,7371 } },
{ "Sony DSLR-A580", -512, 0xfeb,
{ 5932,-1492,-411,-4813,12285,2856,-741,1524,6739 } },
{ "Sony DSLR-A500", -512, 0xfeb,
{ 6046,-1127,-278,-5574,13076,2786,-691,1419,7625 } },
{ "Sony DSLR-A5", -512, 0xfeb,
{ 4950,-580,-103,-5228,12542,3029,-709,1435,7371 } },
{ "Sony DSLR-A700", -512, 0,
{ 5775,-805,-359,-8574,16295,2391,-1943,2341,7249 } },
{ "Sony DSLR-A850", -512, 0,
{ 5413,-1162,-365,-5665,13098,2866,-608,1179,8440 } },
{ "Sony DSLR-A900", -512, 0,
{ 5209,-1072,-397,-8845,16120,2919,-1618,1803,8654 } },
{ "Sony ILCA-77M2", -512, 0,
{ 5991,-1732,-443,-4100,11989,2381,-704,1467,5992 } },
{ "Sony ILCE-7M2", -512, 0,
{ 5271,-712,-347,-6153,13653,2763,-1601,2366,7242 } },
{ "Sony ILCE-7SM2", -512, 0,
{ 5838,-1430,-246,-3497,11477,2297,-748,1885,5778 } },
{ "Sony ILCE-7S", -512, 0,
{ 5838,-1430,-246,-3497,11477,2297,-748,1885,5778 } },
{ "Sony ILCE-7RM2", -512, 0,
{ 6629,-1900,-483,-4618,12349,2550,-622,1381,6514 } },
{ "Sony ILCE-7R", -512, 0,
{ 4913,-541,-202,-6130,13513,2906,-1564,2151,7183 } },
{ "Sony ILCE-7", -512, 0,
{ 5271,-712,-347,-6153,13653,2763,-1601,2366,7242 } },
{ "Sony ILCE", -512, 0, /* 3000, 5000, 5100, 6000, and QX1 */
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony NEX-5N", -512, 0,
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony NEX-5R", -512, 0,
{ 6129,-1545,-418,-4930,12490,2743,-977,1693,6615 } },
{ "Sony NEX-5T", -512, 0,
{ 6129,-1545,-418,-4930,12490,2743,-977,1693,6615 } },
{ "Sony NEX-3N", -512, 0,
{ 6129,-1545,-418,-4930,12490,2743,-977,1693,6615 } },
{ "Sony NEX-3", -512, 0, /* Adobe */
{ 6549,-1550,-436,-4880,12435,2753,-854,1868,6976 } },
{ "Sony NEX-5", -512, 0, /* Adobe */
{ 6549,-1550,-436,-4880,12435,2753,-854,1868,6976 } },
{ "Sony NEX-6", -512, 0,
{ 6129,-1545,-418,-4930,12490,2743,-977,1693,6615 } },
{ "Sony NEX-7", -512, 0,
{ 5491,-1192,-363,-4951,12342,2948,-911,1722,7192 } },
{ "Sony NEX", -512, 0, /* NEX-C3, NEX-F3 */
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony SLT-A33", -512, 0,
{ 6069,-1221,-366,-5221,12779,2734,-1024,2066,6834 } },
{ "Sony SLT-A35", -512, 0,
{ 5986,-1618,-415,-4557,11820,3120,-681,1404,6971 } },
{ "Sony SLT-A37", -512, 0,
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony SLT-A55", -512, 0,
{ 5932,-1492,-411,-4813,12285,2856,-741,1524,6739 } },
{ "Sony SLT-A57", -512, 0,
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony SLT-A58", -512, 0,
{ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 } },
{ "Sony SLT-A65", -512, 0,
{ 5491,-1192,-363,-4951,12342,2948,-911,1722,7192 } },
{ "Sony SLT-A77", -512, 0,
{ 5491,-1192,-363,-4951,12342,2948,-911,1722,7192 } },
{ "Sony SLT-A99", -512, 0,
{ 6344,-1612,-462,-4863,12477,2681,-865,1786,6899 } },
};
double cam_xyz[4][3];
char name[130];
int i, j;
if(colors>4 || colors < 1) return;
int bl4=(cblack[0]+cblack[1]+cblack[2]+cblack[3])/4,bl64=0;
if(cblack[4]*cblack[5]>0)
{
for (unsigned c = 0; c < 4096 && c < cblack[4]*cblack[5]; c++)
bl64+=cblack[c+6];
bl64 /= cblack[4]*cblack[5];
}
int rblack = black+bl4+bl64;
sprintf (name, "%s %s", t_make, t_model);
for (i=0; i < sizeof table / sizeof *table; i++)
if (!strncasecmp(name, table[i].prefix, strlen(table[i].prefix))) {
if(!dng_version)
{
if (table[i].t_black>0)
{
black = (ushort) table[i].t_black;
memset(cblack,0,sizeof(cblack));
}
else if(table[i].t_black <0 && rblack == 0 )
{
black = (ushort) (-table[i].t_black);
memset(cblack,0,sizeof(cblack));
}
if (table[i].t_maximum)
maximum = (ushort) table[i].t_maximum;
}
if (table[i].trans[0]) {
for (raw_color = j=0; j < 12; j++)
#ifdef LIBRAW_LIBRARY_BUILD
if(internal_only)
imgdata.color.cam_xyz[0][j] = table[i].trans[j] / 10000.0;
else
imgdata.color.cam_xyz[0][j] =
#endif
((double*)cam_xyz)[j] = table[i].trans[j] / 10000.0;
#ifdef LIBRAW_LIBRARY_BUILD
if(!internal_only)
#endif
cam_xyz_coeff (rgb_cam, cam_xyz);
}
break;
}
}
void CLASS simple_coeff (int index)
{
static const float table[][12] = {
/* index 0 -- all Foveon cameras */
{ 1.4032,-0.2231,-0.1016,-0.5263,1.4816,0.017,-0.0112,0.0183,0.9113 },
/* index 1 -- Kodak DC20 and DC25 */
{ 2.25,0.75,-1.75,-0.25,-0.25,0.75,0.75,-0.25,-0.25,-1.75,0.75,2.25 },
/* index 2 -- Logitech Fotoman Pixtura */
{ 1.893,-0.418,-0.476,-0.495,1.773,-0.278,-1.017,-0.655,2.672 },
/* index 3 -- Nikon E880, E900, and E990 */
{ -1.936280, 1.800443, -1.448486, 2.584324,
1.405365, -0.524955, -0.289090, 0.408680,
-1.204965, 1.082304, 2.941367, -1.818705 }
};
int i, c;
for (raw_color = i=0; i < 3; i++)
FORCC rgb_cam[i][c] = table[index][i*colors+c];
}
short CLASS guess_byte_order (int words)
{
uchar test[4][2];
int t=2, msb;
double diff, sum[2] = {0,0};
fread (test[0], 2, 2, ifp);
for (words-=2; words--; ) {
fread (test[t], 2, 1, ifp);
for (msb=0; msb < 2; msb++) {
diff = (test[t^2][msb] << 8 | test[t^2][!msb])
- (test[t ][msb] << 8 | test[t ][!msb]);
sum[msb] += diff*diff;
}
t = (t+1) & 3;
}
return sum[0] < sum[1] ? 0x4d4d : 0x4949;
}
float CLASS find_green (int bps, int bite, int off0, int off1)
{
UINT64 bitbuf=0;
int vbits, col, i, c;
ushort img[2][2064];
double sum[]={0,0};
FORC(2) {
fseek (ifp, c ? off1:off0, SEEK_SET);
for (vbits=col=0; col < width; col++) {
for (vbits -= bps; vbits < 0; vbits += bite) {
bitbuf <<= bite;
for (i=0; i < bite; i+=8)
bitbuf |= (unsigned) (fgetc(ifp) << i);
}
img[c][col] = bitbuf << (64-bps-vbits) >> (64-bps);
}
}
FORC(width-1) {
sum[ c & 1] += ABS(img[0][c]-img[1][c+1]);
sum[~c & 1] += ABS(img[1][c]-img[0][c+1]);
}
return 100 * log(sum[0]/sum[1]);
}
/*
Identify which camera created this file, and set global variables
accordingly.
*/
void CLASS identify()
{
static const short pana[][6] = {
{ 3130, 1743, 4, 0, -6, 0 },
{ 3130, 2055, 4, 0, -6, 0 },
{ 3130, 2319, 4, 0, -6, 0 },
{ 3170, 2103, 18, 0,-42, 20 },
{ 3170, 2367, 18, 13,-42,-21 },
{ 3177, 2367, 0, 0, -1, 0 },
{ 3304, 2458, 0, 0, -1, 0 },
{ 3330, 2463, 9, 0, -5, 0 },
{ 3330, 2479, 9, 0,-17, 4 },
{ 3370, 1899, 15, 0,-44, 20 },
{ 3370, 2235, 15, 0,-44, 20 },
{ 3370, 2511, 15, 10,-44,-21 },
{ 3690, 2751, 3, 0, -8, -3 },
{ 3710, 2751, 0, 0, -3, 0 },
{ 3724, 2450, 0, 0, 0, -2 },
{ 3770, 2487, 17, 0,-44, 19 },
{ 3770, 2799, 17, 15,-44,-19 },
{ 3880, 2170, 6, 0, -6, 0 },
{ 4060, 3018, 0, 0, 0, -2 },
{ 4290, 2391, 3, 0, -8, -1 },
{ 4330, 2439, 17, 15,-44,-19 },
{ 4508, 2962, 0, 0, -3, -4 },
{ 4508, 3330, 0, 0, -3, -6 },
};
static const ushort canon[][11] = {
{ 1944, 1416, 0, 0, 48, 0 },
{ 2144, 1560, 4, 8, 52, 2, 0, 0, 0, 25 },
{ 2224, 1456, 48, 6, 0, 2 },
{ 2376, 1728, 12, 6, 52, 2 },
{ 2672, 1968, 12, 6, 44, 2 },
{ 3152, 2068, 64, 12, 0, 0, 16 },
{ 3160, 2344, 44, 12, 4, 4 },
{ 3344, 2484, 4, 6, 52, 6 },
{ 3516, 2328, 42, 14, 0, 0 },
{ 3596, 2360, 74, 12, 0, 0 },
{ 3744, 2784, 52, 12, 8, 12 },
{ 3944, 2622, 30, 18, 6, 2 },
{ 3948, 2622, 42, 18, 0, 2 },
{ 3984, 2622, 76, 20, 0, 2, 14 },
{ 4104, 3048, 48, 12, 24, 12 },
{ 4116, 2178, 4, 2, 0, 0 },
{ 4152, 2772, 192, 12, 0, 0 },
{ 4160, 3124, 104, 11, 8, 65 },
{ 4176, 3062, 96, 17, 8, 0, 0, 16, 0, 7, 0x49 },
{ 4192, 3062, 96, 17, 24, 0, 0, 16, 0, 0, 0x49 },
{ 4312, 2876, 22, 18, 0, 2 },
{ 4352, 2874, 62, 18, 0, 0 },
{ 4476, 2954, 90, 34, 0, 0 },
{ 4480, 3348, 12, 10, 36, 12, 0, 0, 0, 18, 0x49 },
{ 4480, 3366, 80, 50, 0, 0 },
{ 4496, 3366, 80, 50, 12, 0 },
{ 4768, 3516, 96, 16, 0, 0, 0, 16 },
{ 4832, 3204, 62, 26, 0, 0 },
{ 4832, 3228, 62, 51, 0, 0 },
{ 5108, 3349, 98, 13, 0, 0 },
{ 5120, 3318, 142, 45, 62, 0 },
{ 5280, 3528, 72, 52, 0, 0 }, /* EOS M */
{ 5344, 3516, 142, 51, 0, 0 },
{ 5344, 3584, 126,100, 0, 2 },
{ 5360, 3516, 158, 51, 0, 0 },
{ 5568, 3708, 72, 38, 0, 0 },
{ 5632, 3710, 96, 17, 0, 0, 0, 16, 0, 0, 0x49 },
{ 5712, 3774, 62, 20, 10, 2 },
{ 5792, 3804, 158, 51, 0, 0 },
{ 5920, 3950, 122, 80, 2, 0 },
{ 6096, 4056, 72, 34, 0, 0 }, /* EOS M3 */
{ 8896, 5920, 160, 64, 0, 0 },
};
static const struct {
ushort id;
char t_model[20];
} unique[] = {
{ 0x001, "EOS-1D" },
{ 0x167, "EOS-1DS" },
{ 0x168, "EOS 10D" },
{ 0x169, "EOS-1D Mark III" },
{ 0x170, "EOS 300D" },
{ 0x174, "EOS-1D Mark II" },
{ 0x175, "EOS 20D" },
{ 0x176, "EOS 450D" },
{ 0x188, "EOS-1Ds Mark II" },
{ 0x189, "EOS 350D" },
{ 0x190, "EOS 40D" },
{ 0x213, "EOS 5D" },
{ 0x215, "EOS-1Ds Mark III" },
{ 0x218, "EOS 5D Mark II" },
{ 0x232, "EOS-1D Mark II N" },
{ 0x234, "EOS 30D" },
{ 0x236, "EOS 400D" },
{ 0x250, "EOS 7D" },
{ 0x252, "EOS 500D" },
{ 0x254, "EOS 1000D" },
{ 0x261, "EOS 50D" },
{ 0x269, "EOS-1D X" },
{ 0x270, "EOS 550D" },
{ 0x281, "EOS-1D Mark IV" },
{ 0x285, "EOS 5D Mark III" },
{ 0x286, "EOS 600D" },
{ 0x287, "EOS 60D" },
{ 0x288, "EOS 1100D" },
{ 0x289, "EOS 7D Mark II" },
{ 0x301, "EOS 650D" },
{ 0x302, "EOS 6D" },
{ 0x324, "EOS-1D C" },
{ 0x325, "EOS 70D" },
{ 0x326, "EOS 700D" },
{ 0x327, "EOS 1200D" },
{ 0x331, "EOS M" },
{ 0x335, "EOS M2" },
{ 0x374, "EOS M3"}, /* temp */
{ 0x384, "EOS M10"}, /* temp */
{ 0x346, "EOS 100D" },
{ 0x347, "EOS 760D" },
{ 0x382, "EOS 5DS" },
{ 0x393, "EOS 750D" },
{ 0x401, "EOS 5DS R" },
}, sonique[] = {
{ 0x002, "DSC-R1" },
{ 0x100, "DSLR-A100" },
{ 0x101, "DSLR-A900" },
{ 0x102, "DSLR-A700" },
{ 0x103, "DSLR-A200" },
{ 0x104, "DSLR-A350" },
{ 0x105, "DSLR-A300" },
{ 0x106, "DSLR-A900" },
{ 0x107, "DSLR-A380" },
{ 0x108, "DSLR-A330" },
{ 0x109, "DSLR-A230" },
{ 0x10a, "DSLR-A290" },
{ 0x10d, "DSLR-A850" },
{ 0x10e, "DSLR-A850" },
{ 0x111, "DSLR-A550" },
{ 0x112, "DSLR-A500" },
{ 0x113, "DSLR-A450" },
{ 0x116, "NEX-5" },
{ 0x117, "NEX-3" },
{ 0x118, "SLT-A33" },
{ 0x119, "SLT-A55V" },
{ 0x11a, "DSLR-A560" },
{ 0x11b, "DSLR-A580" },
{ 0x11c, "NEX-C3" },
{ 0x11d, "SLT-A35" },
{ 0x11e, "SLT-A65V" },
{ 0x11f, "SLT-A77V" },
{ 0x120, "NEX-5N" },
{ 0x121, "NEX-7" },
{ 0x122, "NEX-VG20E"},
{ 0x123, "SLT-A37" },
{ 0x124, "SLT-A57" },
{ 0x125, "NEX-F3" },
{ 0x126, "SLT-A99V" },
{ 0x127, "NEX-6" },
{ 0x128, "NEX-5R" },
{ 0x129, "DSC-RX100" },
{ 0x12a, "DSC-RX1" },
{ 0x12b, "NEX-VG900" },
{ 0x12c, "NEX-VG30E" },
{ 0x12e, "ILCE-3000" },
{ 0x12f, "SLT-A58" },
{ 0x131, "NEX-3N" },
{ 0x132, "ILCE-7" },
{ 0x133, "NEX-5T" },
{ 0x134, "DSC-RX100M2" },
{ 0x135, "DSC-RX10" },
{ 0x136, "DSC-RX1R" },
{ 0x137, "ILCE-7R" },
{ 0x138, "ILCE-6000" },
{ 0x139, "ILCE-5000" },
{ 0x13d, "DSC-RX100M3" },
{ 0x13e, "ILCE-7S" },
{ 0x13f, "ILCA-77M2" },
{ 0x153, "ILCE-5100" },
{ 0x154, "ILCE-7M2" },
{ 0x155, "DSC-RX100M4" },
{ 0x156, "DSC-RX10M2" },
{ 0x158, "DSC-RX1RM2" },
{ 0x15a, "ILCE-QX1" },
{ 0x15b, "ILCE-7RM2" },
{ 0x15e, "ILCE-7SM2" },
};
static const struct {
unsigned fsize;
ushort rw, rh;
uchar lm, tm, rm, bm, lf, cf, max, flags;
char t_make[10], t_model[20];
ushort offset;
} table[] = {
{ 786432,1024, 768, 0, 0, 0, 0, 0,0x94,0,0,"AVT","F-080C" },
{ 1447680,1392,1040, 0, 0, 0, 0, 0,0x94,0,0,"AVT","F-145C" },
{ 1920000,1600,1200, 0, 0, 0, 0, 0,0x94,0,0,"AVT","F-201C" },
{ 5067304,2588,1958, 0, 0, 0, 0, 0,0x94,0,0,"AVT","F-510C" },
{ 5067316,2588,1958, 0, 0, 0, 0, 0,0x94,0,0,"AVT","F-510C",12 },
{ 10134608,2588,1958, 0, 0, 0, 0, 9,0x94,0,0,"AVT","F-510C" },
{ 10134620,2588,1958, 0, 0, 0, 0, 9,0x94,0,0,"AVT","F-510C",12 },
{ 16157136,3272,2469, 0, 0, 0, 0, 9,0x94,0,0,"AVT","F-810C" },
{ 15980544,3264,2448, 0, 0, 0, 0, 8,0x61,0,1,"AgfaPhoto","DC-833m" },
{ 9631728,2532,1902, 0, 0, 0, 0,96,0x61,0,0,"Alcatel","5035D" },
// Android Raw dumps id start
// File Size in bytes Horizontal Res Vertical Flag then bayer order eg 0x16 bbgr 0x94 rggb
{ 16424960,4208,3120, 0, 0, 0, 0, 1,0x16,0,0,"Sony","IMX135-mipi 13mp" },
{ 17522688,4212,3120, 0, 0, 0, 0, 0,0x16,0,0,"Sony","IMX135-QCOM" },
{ 10223360,2608,1960, 0, 0, 0, 0, 1,0x94,0,0,"Sony","IMX072-mipi" },
{ 5107712,2688,1520, 0, 0, 0, 0, 1,0x61,0,0,"HTC","UltraPixel" },
{ 1540857,2688,1520, 0, 0, 0, 0, 1,0x61,0,0,"Samsung","S3" },
{ 10223363,2688,1520, 0, 0, 0, 0, 1,0x61,0,0,"Samsung","GalaxyNexus" },
// Android Raw dumps id end
{ 20137344,3664,2748,0, 0, 0, 0,0x40,0x49,0,0,"Aptina","MT9J003",0xffff },
{ 2868726,1384,1036, 0, 0, 0, 0,64,0x49,0,8,"Baumer","TXG14",1078 },
{ 5298000,2400,1766,12,12,44, 2,40,0x94,0,2,"Canon","PowerShot SD300" },
{ 6553440,2664,1968, 4, 4,44, 4,40,0x94,0,2,"Canon","PowerShot A460" },
{ 6573120,2672,1968,12, 8,44, 0,40,0x94,0,2,"Canon","PowerShot A610" },
{ 6653280,2672,1992,10, 6,42, 2,40,0x94,0,2,"Canon","PowerShot A530" },
{ 7710960,2888,2136,44, 8, 4, 0,40,0x94,0,2,"Canon","PowerShot S3 IS" },
{ 9219600,3152,2340,36,12, 4, 0,40,0x94,0,2,"Canon","PowerShot A620" },
{ 9243240,3152,2346,12, 7,44,13,40,0x49,0,2,"Canon","PowerShot A470" },
{ 10341600,3336,2480, 6, 5,32, 3,40,0x94,0,2,"Canon","PowerShot A720 IS" },
{ 10383120,3344,2484,12, 6,44, 6,40,0x94,0,2,"Canon","PowerShot A630" },
{ 12945240,3736,2772,12, 6,52, 6,40,0x94,0,2,"Canon","PowerShot A640" },
{ 15636240,4104,3048,48,12,24,12,40,0x94,0,2,"Canon","PowerShot A650" },
{ 15467760,3720,2772, 6,12,30, 0,40,0x94,0,2,"Canon","PowerShot SX110 IS" },
{ 15534576,3728,2778,12, 9,44, 9,40,0x94,0,2,"Canon","PowerShot SX120 IS" },
{ 18653760,4080,3048,24,12,24,12,40,0x94,0,2,"Canon","PowerShot SX20 IS" },
{ 19131120,4168,3060,92,16, 4, 1,40,0x94,0,2,"Canon","PowerShot SX220 HS" },
{ 21936096,4464,3276,25,10,73,12,40,0x16,0,2,"Canon","PowerShot SX30 IS" },
{ 24724224,4704,3504, 8,16,56, 8,40,0x49,0,2,"Canon","PowerShot A3300 IS" },
{ 1976352,1632,1211, 0, 2, 0, 1, 0,0x94,0,1,"Casio","QV-2000UX" },
{ 3217760,2080,1547, 0, 0,10, 1, 0,0x94,0,1,"Casio","QV-3*00EX" },
{ 6218368,2585,1924, 0, 0, 9, 0, 0,0x94,0,1,"Casio","QV-5700" },
{ 7816704,2867,2181, 0, 0,34,36, 0,0x16,0,1,"Casio","EX-Z60" },
{ 2937856,1621,1208, 0, 0, 1, 0, 0,0x94,7,13,"Casio","EX-S20" },
{ 4948608,2090,1578, 0, 0,32,34, 0,0x94,7,1,"Casio","EX-S100" },
{ 6054400,2346,1720, 2, 0,32, 0, 0,0x94,7,1,"Casio","QV-R41" },
{ 7426656,2568,1928, 0, 0, 0, 0, 0,0x94,0,1,"Casio","EX-P505" },
{ 7530816,2602,1929, 0, 0,22, 0, 0,0x94,7,1,"Casio","QV-R51" },
{ 7542528,2602,1932, 0, 0,32, 0, 0,0x94,7,1,"Casio","EX-Z50" },
{ 7562048,2602,1937, 0, 0,25, 0, 0,0x16,7,1,"Casio","EX-Z500" },
{ 7753344,2602,1986, 0, 0,32,26, 0,0x94,7,1,"Casio","EX-Z55" },
{ 9313536,2858,2172, 0, 0,14,30, 0,0x94,7,1,"Casio","EX-P600" },
{ 10834368,3114,2319, 0, 0,27, 0, 0,0x94,0,1,"Casio","EX-Z750" },
{ 10843712,3114,2321, 0, 0,25, 0, 0,0x94,0,1,"Casio","EX-Z75" },
{ 10979200,3114,2350, 0, 0,32,32, 0,0x94,7,1,"Casio","EX-P700" },
{ 12310144,3285,2498, 0, 0, 6,30, 0,0x94,0,1,"Casio","EX-Z850" },
{ 12489984,3328,2502, 0, 0,47,35, 0,0x94,0,1,"Casio","EX-Z8" },
{ 15499264,3754,2752, 0, 0,82, 0, 0,0x94,0,1,"Casio","EX-Z1050" },
{ 18702336,4096,3044, 0, 0,24, 0,80,0x94,7,1,"Casio","EX-ZR100" },
{ 7684000,2260,1700, 0, 0, 0, 0,13,0x94,0,1,"Casio","QV-4000" },
{ 787456,1024, 769, 0, 1, 0, 0, 0,0x49,0,0,"Creative","PC-CAM 600" },
{ 28829184,4384,3288, 0, 0, 0, 0,36,0x61,0,0,"DJI" },
{ 15151104,4608,3288, 0, 0, 0, 0, 0,0x94,0,0,"Matrix" },
{ 3840000,1600,1200, 0, 0, 0, 0,65,0x49,0,0,"Foculus","531C" },
{ 307200, 640, 480, 0, 0, 0, 0, 0,0x94,0,0,"Generic" },
{ 62464, 256, 244, 1, 1, 6, 1, 0,0x8d,0,0,"Kodak","DC20" },
{ 124928, 512, 244, 1, 1,10, 1, 0,0x8d,0,0,"Kodak","DC20" },
{ 1652736,1536,1076, 0,52, 0, 0, 0,0x61,0,0,"Kodak","DCS200" },
{ 4159302,2338,1779, 1,33, 1, 2, 0,0x94,0,0,"Kodak","C330" },
{ 4162462,2338,1779, 1,33, 1, 2, 0,0x94,0,0,"Kodak","C330",3160 },
{ 2247168,1232, 912, 0, 0,16, 0, 0,0x00,0,0,"Kodak","C330" },
{ 3370752,1232, 912, 0, 0,16, 0, 0,0x00,0,0,"Kodak","C330" },
{ 6163328,2864,2152, 0, 0, 0, 0, 0,0x94,0,0,"Kodak","C603" },
{ 6166488,2864,2152, 0, 0, 0, 0, 0,0x94,0,0,"Kodak","C603",3160 },
{ 460800, 640, 480, 0, 0, 0, 0, 0,0x00,0,0,"Kodak","C603" },
{ 9116448,2848,2134, 0, 0, 0, 0, 0,0x00,0,0,"Kodak","C603" },
{ 12241200,4040,3030, 2, 0, 0,13, 0,0x49,0,0,"Kodak","12MP" },
{ 12272756,4040,3030, 2, 0, 0,13, 0,0x49,0,0,"Kodak","12MP",31556 },
{ 18000000,4000,3000, 0, 0, 0, 0, 0,0x00,0,0,"Kodak","12MP" },
{ 614400, 640, 480, 0, 3, 0, 0,64,0x94,0,0,"Kodak","KAI-0340" },
{ 15360000,3200,2400, 0, 0, 0, 0,96,0x16,0,0,"Lenovo","A820" },
{ 3884928,1608,1207, 0, 0, 0, 0,96,0x16,0,0,"Micron","2010",3212 },
{ 1138688,1534, 986, 0, 0, 0, 0, 0,0x61,0,0,"Minolta","RD175",513 },
{ 1581060,1305, 969, 0, 0,18, 6, 6,0x1e,4,1,"Nikon","E900" },
{ 2465792,1638,1204, 0, 0,22, 1, 6,0x4b,5,1,"Nikon","E950" },
{ 2940928,1616,1213, 0, 0, 0, 7,30,0x94,0,1,"Nikon","E2100" },
{ 4771840,2064,1541, 0, 0, 0, 1, 6,0xe1,0,1,"Nikon","E990" },
{ 4775936,2064,1542, 0, 0, 0, 0,30,0x94,0,1,"Nikon","E3700" },
{ 5865472,2288,1709, 0, 0, 0, 1, 6,0xb4,0,1,"Nikon","E4500" },
{ 5869568,2288,1710, 0, 0, 0, 0, 6,0x16,0,1,"Nikon","E4300" },
{ 7438336,2576,1925, 0, 0, 0, 1, 6,0xb4,0,1,"Nikon","E5000" },
{ 8998912,2832,2118, 0, 0, 0, 0,30,0x94,7,1,"Nikon","COOLPIX S6" },
{ 5939200,2304,1718, 0, 0, 0, 0,30,0x16,0,0,"Olympus","C770UZ" },
{ 3178560,2064,1540, 0, 0, 0, 0, 0,0x94,0,1,"Pentax","Optio S" },
{ 4841984,2090,1544, 0, 0,22, 0, 0,0x94,7,1,"Pentax","Optio S" },
{ 6114240,2346,1737, 0, 0,22, 0, 0,0x94,7,1,"Pentax","Optio S4" },
{ 10702848,3072,2322, 0, 0, 0,21,30,0x94,0,1,"Pentax","Optio 750Z" },
{ 4147200,1920,1080, 0, 0, 0, 0, 0,0x49,0,0,"Photron","BC2-HD" },
{ 4151666,1920,1080, 0, 0, 0, 0, 0,0x49,0,0,"Photron","BC2-HD",8 },
{ 13248000,2208,3000, 0, 0, 0, 0,13,0x61,0,0,"Pixelink","A782" },
{ 6291456,2048,1536, 0, 0, 0, 0,96,0x61,0,0,"RoverShot","3320AF" },
{ 311696, 644, 484, 0, 0, 0, 0, 0,0x16,0,8,"ST Micro","STV680 VGA" },
{ 16098048,3288,2448, 0, 0,24, 0, 9,0x94,0,1,"Samsung","S85" },
{ 16215552,3312,2448, 0, 0,48, 0, 9,0x94,0,1,"Samsung","S85" },
{ 20487168,3648,2808, 0, 0, 0, 0,13,0x94,5,1,"Samsung","WB550" },
{ 24000000,4000,3000, 0, 0, 0, 0,13,0x94,5,1,"Samsung","WB550" },
{ 12582980,3072,2048, 0, 0, 0, 0,33,0x61,0,0,"Sinar","",68 },
{ 33292868,4080,4080, 0, 0, 0, 0,33,0x61,0,0,"Sinar","",68 },
{ 44390468,4080,5440, 0, 0, 0, 0,33,0x61,0,0,"Sinar","",68 },
{ 1409024,1376,1024, 0, 0, 1, 0, 0,0x49,0,0,"Sony","XCD-SX910CR" },
{ 2818048,1376,1024, 0, 0, 1, 0,97,0x49,0,0,"Sony","XCD-SX910CR" },
};
static const char *corp[] =
{ "AgfaPhoto", "Canon", "Casio", "Epson", "Fujifilm",
"Mamiya", "Minolta", "Motorola", "Kodak", "Konica", "Leica",
"Nikon", "Nokia", "Olympus", "Pentax", "Phase One", "Ricoh",
"Samsung", "Sigma", "Sinar", "Sony" };
#ifdef LIBRAW_LIBRARY_BUILD
char head[64], *cp;
#else
char head[32], *cp;
#endif
int hlen, flen, fsize, zero_fsize=1, i, c;
struct jhead jh;
tiff_flip = flip = filters = UINT_MAX; /* unknown */
raw_height = raw_width = fuji_width = fuji_layout = cr2_slice[0] = 0;
maximum = height = width = top_margin = left_margin = 0;
cdesc[0] = desc[0] = artist[0] = make[0] = model[0] = model2[0] = 0;
iso_speed = shutter = aperture = focal_len = unique_id = 0;
tiff_nifds = 0;
memset (tiff_ifd, 0, sizeof tiff_ifd);
memset (gpsdata, 0, sizeof gpsdata);
memset (cblack, 0, sizeof cblack);
memset (white, 0, sizeof white);
memset (mask, 0, sizeof mask);
thumb_offset = thumb_length = thumb_width = thumb_height = 0;
load_raw = thumb_load_raw = 0;
write_thumb = &CLASS jpeg_thumb;
data_offset = meta_offset = meta_length = tiff_bps = tiff_compress = 0;
kodak_cbpp = zero_after_ff = dng_version = load_flags = 0;
timestamp = shot_order = tiff_samples = black = is_foveon = 0;
mix_green = profile_length = data_error = zero_is_bad = 0;
pixel_aspect = is_raw = raw_color = 1;
tile_width = tile_length = 0;
for (i=0; i < 4; i++) {
cam_mul[i] = i == 1;
pre_mul[i] = i < 3;
FORC3 cmatrix[c][i] = 0;
FORC3 rgb_cam[c][i] = c == i;
}
colors = 3;
for (i=0; i < 0x10000; i++) curve[i] = i;
order = get2();
hlen = get4();
fseek (ifp, 0, SEEK_SET);
#ifdef LIBRAW_LIBRARY_BUILD
fread (head, 1, 64, ifp);
libraw_internal_data.unpacker_data.lenRAFData = libraw_internal_data.unpacker_data.posRAFData = 0;
#else
fread (head, 1, 32, ifp);
#endif
fseek (ifp, 0, SEEK_END);
flen = fsize = ftell(ifp);
if ((cp = (char *) memmem (head, 32, (char*)"MMMM", 4)) ||
(cp = (char *) memmem (head, 32, (char*)"IIII", 4))) {
parse_phase_one (cp-head);
if (cp-head && parse_tiff(0)) apply_tiff();
} else if (order == 0x4949 || order == 0x4d4d) {
if (!memcmp (head+6,"HEAPCCDR",8)) {
data_offset = hlen;
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
#endif
parse_ciff (hlen, flen-hlen, 0);
load_raw = &CLASS canon_load_raw;
} else if (parse_tiff(0)) apply_tiff();
} else if (!memcmp (head,"\xff\xd8\xff\xe1",4) &&
!memcmp (head+6,"Exif",4)) {
fseek (ifp, 4, SEEK_SET);
data_offset = 4 + get2();
fseek (ifp, data_offset, SEEK_SET);
if (fgetc(ifp) != 0xff)
parse_tiff(12);
thumb_offset = 0;
} else if (!memcmp (head+25,"ARECOYK",7)) {
strcpy (make, "Contax");
strcpy (model,"N Digital");
fseek (ifp, 33, SEEK_SET);
get_timestamp(1);
fseek (ifp, 52, SEEK_SET);
switch (get4()) {
case 7: iso_speed = 25; break;
case 8: iso_speed = 32; break;
case 9: iso_speed = 40; break;
case 10: iso_speed = 50; break;
case 11: iso_speed = 64; break;
case 12: iso_speed = 80; break;
case 13: iso_speed = 100; break;
case 14: iso_speed = 125; break;
case 15: iso_speed = 160; break;
case 16: iso_speed = 200; break;
case 17: iso_speed = 250; break;
case 18: iso_speed = 320; break;
case 19: iso_speed = 400; break;
}
shutter = powf64(2.0f, (((float)get4())/8.0f)) / 16000.0f;
FORC4 cam_mul[c ^ (c >> 1)] = get4();
fseek (ifp, 88, SEEK_SET);
aperture = powf64(2.0f, ((float)get4())/16.0f);
fseek (ifp, 112, SEEK_SET);
focal_len = get4();
#ifdef LIBRAW_LIBRARY_BUILD
fseek (ifp, 104, SEEK_SET);
imgdata.lens.makernotes.MaxAp4CurFocal = powf64(2.0f, ((float)get4())/16.0f);
fseek (ifp, 124, SEEK_SET);
fread(imgdata.lens.makernotes.Lens, 32, 1, ifp);
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_Contax_N;
if (imgdata.lens.makernotes.Lens[0])
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_Contax_N;
#endif
} else if (!strcmp (head, "PXN")) {
strcpy (make, "Logitech");
strcpy (model,"Fotoman Pixtura");
} else if (!strcmp (head, "qktk")) {
strcpy (make, "Apple");
strcpy (model,"QuickTake 100");
load_raw = &CLASS quicktake_100_load_raw;
} else if (!strcmp (head, "qktn")) {
strcpy (make, "Apple");
strcpy (model,"QuickTake 150");
load_raw = &CLASS kodak_radc_load_raw;
} else if (!memcmp (head,"FUJIFILM",8)) {
#ifdef LIBRAW_LIBRARY_BUILD
strcpy(model, head+0x1c);
memcpy(model2, head+0x3c, 4);
model2[4]=0;
#endif
fseek (ifp, 84, SEEK_SET);
thumb_offset = get4();
thumb_length = get4();
fseek (ifp, 92, SEEK_SET);
parse_fuji (get4());
if (thumb_offset > 120) {
fseek (ifp, 120, SEEK_SET);
is_raw += (i = get4())?1:0;
if (is_raw == 2 && shot_select)
parse_fuji (i);
}
load_raw = &CLASS unpacked_load_raw;
fseek (ifp, 100+28*(shot_select > 0), SEEK_SET);
parse_tiff (data_offset = get4());
parse_tiff (thumb_offset+12);
apply_tiff();
} else if (!memcmp (head,"RIFF",4)) {
fseek (ifp, 0, SEEK_SET);
parse_riff();
} else if (!memcmp (head+4,"ftypqt ",9)) {
fseek (ifp, 0, SEEK_SET);
parse_qt (fsize);
is_raw = 0;
} else if (!memcmp (head,"\0\001\0\001\0@",6)) {
fseek (ifp, 6, SEEK_SET);
fread (make, 1, 8, ifp);
fread (model, 1, 8, ifp);
fread (model2, 1, 16, ifp);
data_offset = get2();
get2();
raw_width = get2();
raw_height = get2();
load_raw = &CLASS nokia_load_raw;
filters = 0x61616161;
} else if (!memcmp (head,"NOKIARAW",8)) {
strcpy (make, "NOKIA");
order = 0x4949;
fseek (ifp, 300, SEEK_SET);
data_offset = get4();
i = get4();
width = get2();
height = get2();
switch (tiff_bps = i*8 / (width * height)) {
case 8: load_raw = &CLASS eight_bit_load_raw; break;
case 10: load_raw = &CLASS nokia_load_raw;
}
raw_height = height + (top_margin = i / (width * tiff_bps/8) - height);
mask[0][3] = 1;
filters = 0x61616161;
} else if (!memcmp (head,"ARRI",4)) {
order = 0x4949;
fseek (ifp, 20, SEEK_SET);
width = get4();
height = get4();
strcpy (make, "ARRI");
fseek (ifp, 668, SEEK_SET);
fread (model, 1, 64, ifp);
data_offset = 4096;
load_raw = &CLASS packed_load_raw;
load_flags = 88;
filters = 0x61616161;
} else if (!memcmp (head,"XPDS",4)) {
order = 0x4949;
fseek (ifp, 0x800, SEEK_SET);
fread (make, 1, 41, ifp);
raw_height = get2();
raw_width = get2();
fseek (ifp, 56, SEEK_CUR);
fread (model, 1, 30, ifp);
data_offset = 0x10000;
load_raw = &CLASS canon_rmf_load_raw;
gamma_curve (0, 12.25, 1, 1023);
} else if (!memcmp (head+4,"RED1",4)) {
strcpy (make, "Red");
strcpy (model,"One");
parse_redcine();
load_raw = &CLASS redcine_load_raw;
gamma_curve (1/2.4, 12.92, 1, 4095);
filters = 0x49494949;
} else if (!memcmp (head,"DSC-Image",9))
parse_rollei();
else if (!memcmp (head,"PWAD",4))
parse_sinar_ia();
else if (!memcmp (head,"\0MRM",4))
parse_minolta(0);
else if (!memcmp (head,"FOVb",4))
{
#ifdef LIBRAW_LIBRARY_BUILD
#ifdef LIBRAW_DEMOSAIC_PACK_GPL2
if(!imgdata.params.force_foveon_x3f)
parse_foveon();
else
#endif
parse_x3f();
#else
#ifdef LIBRAW_DEMOSAIC_PACK_GPL2
parse_foveon();
#endif
#endif
}
else if (!memcmp (head,"CI",2))
parse_cine();
if(make[0] == 0)
for (zero_fsize=i=0; i < sizeof table / sizeof *table; i++)
if (fsize == table[i].fsize) {
strcpy (make, table[i].t_make );
#ifdef LIBRAW_LIBRARY_BUILD
if (!strncmp(make, "Canon",5))
{
imgdata.lens.makernotes.CameraMount = LIBRAW_MOUNT_FixedLens;
imgdata.lens.makernotes.LensMount = LIBRAW_MOUNT_FixedLens;
}
#endif
strcpy (model, table[i].t_model);
flip = table[i].flags >> 2;
zero_is_bad = table[i].flags & 2;
if (table[i].flags & 1)
parse_external_jpeg();
data_offset = table[i].offset == 0xffff?0:table[i].offset;
raw_width = table[i].rw;
raw_height = table[i].rh;
left_margin = table[i].lm;
top_margin = table[i].tm;
width = raw_width - left_margin - table[i].rm;
height = raw_height - top_margin - table[i].bm;
filters = 0x1010101 * table[i].cf;
colors = 4 - !((filters & filters >> 1) & 0x5555);
load_flags = table[i].lf;
switch (tiff_bps = (fsize-data_offset)*8 / (raw_width*raw_height)) {
case 6:
load_raw = &CLASS minolta_rd175_load_raw; break;
case 8:
load_raw = &CLASS eight_bit_load_raw; break;
case 10:
if ((fsize-data_offset)/raw_height*3 >= raw_width*4) {
load_raw = &CLASS android_loose_load_raw; break;
} else if (load_flags & 1) {
load_raw = &CLASS android_tight_load_raw; break;
}
case 12:
load_flags |= 128;
load_raw = &CLASS packed_load_raw; break;
case 16:
order = 0x4949 | 0x404 * (load_flags & 1);
tiff_bps -= load_flags >> 4;
tiff_bps -= load_flags = load_flags >> 1 & 7;
load_raw = table[i].offset == 0xffff ? &CLASS unpacked_load_raw_reversed : &CLASS unpacked_load_raw;
}
maximum = (1 << tiff_bps) - (1 << table[i].max);
}
if (zero_fsize) fsize = 0;
if (make[0] == 0) parse_smal (0, flen);
if (make[0] == 0) {
parse_jpeg(0);
fseek(ifp,0,SEEK_END);
int sz = ftell(ifp);
if (!(strncmp(model,"ov",2) && strncmp(model,"RP_OV",5)) && sz>=6404096 &&
!fseek (ifp, -6404096, SEEK_END) &&
fread (head, 1, 32, ifp) && !strcmp(head,"BRCMn")) {
strcpy (make, "OmniVision");
data_offset = ftell(ifp) + 0x8000-32;
width = raw_width;
raw_width = 2611;
load_raw = &CLASS nokia_load_raw;
filters = 0x16161616;
} else is_raw = 0;
}
for (i=0; i < sizeof corp / sizeof *corp; i++)
if (strcasestr (make, corp[i])) /* Simplify company names */
strcpy (make, corp[i]);
if ((!strncmp(make,"Kodak",5) || !strncmp(make,"Leica",5)) &&
((cp = strcasestr(model," DIGITAL CAMERA")) ||
(cp = strstr(model,"FILE VERSION"))))
*cp = 0;
if (!strncasecmp(model,"PENTAX",6))
strcpy (make, "Pentax");
cp = make + strlen(make); /* Remove trailing spaces */
while (*--cp == ' ') *cp = 0;
cp = model + strlen(model);
while (*--cp == ' ') *cp = 0;
i = strlen(make); /* Remove make from model */
if (!strncasecmp (model, make, i) && model[i++] == ' ')
memmove (model, model+i, 64-i);
if (!strncmp (model,"FinePix ",8))
strcpy (model, model+8);
if (!strncmp (model,"Digital Camera ",15))
strcpy (model, model+15);
desc[511] = artist[63] = make[63] = model[63] = model2[63] = 0;
if (!is_raw) goto notraw;
if (!height) height = raw_height;
if (!width) width = raw_width;
if (height == 2624 && width == 3936) /* Pentax K10D and Samsung GX10 */
{ height = 2616; width = 3896; }
if (height == 3136 && width == 4864) /* Pentax K20D and Samsung GX20 */
{ height = 3124; width = 4688; filters = 0x16161616; }
if (width == 4352 && (!strcmp(model,"K-r") || !strcmp(model,"K-x")))
{ width = 4309; filters = 0x16161616; }
if (width >= 4960 && !strncmp(model,"K-5",3))
{ left_margin = 10; width = 4950; filters = 0x16161616; }
if (width == 4736 && !strcmp(model,"K-7"))
{ height = 3122; width = 4684; filters = 0x16161616; top_margin = 2; }
if (width == 6080 && !strcmp(model,"K-3 II"))
{ left_margin = 4; width = 6040; }
if (width == 6080 && !strcmp(model,"K-3"))
{ left_margin = 4; width = 6040; }
if (width == 7424 && !strcmp(model,"645D"))
{ height = 5502; width = 7328; filters = 0x61616161; top_margin = 29;
left_margin = 48; }
if (height == 3014 && width == 4096) /* Ricoh GX200 */
width = 4014;
if (dng_version) {
if (filters == UINT_MAX) filters = 0;
if (filters) is_raw *= tiff_samples;
else colors = tiff_samples;
switch (tiff_compress) {
case 0: /* Compression not set, assuming uncompressed */
case 1: load_raw = &CLASS packed_dng_load_raw; break;
case 7: load_raw = &CLASS lossless_dng_load_raw; break;
#ifdef LIBRAW_LIBRARY_BUILD
case 8: load_raw = &CLASS deflate_dng_load_raw; break;
#endif
case 34892: load_raw = &CLASS lossy_dng_load_raw; break;
default: load_raw = 0;
}
if (!strncmp(make, "Canon",5) && unique_id)
{
for (i = 0; i < sizeof unique / sizeof *unique; i++)
if (unique_id == 0x80000000 + unique[i].id)
{
strcpy(model, unique[i].t_model);
break;
}
}
if (!strncasecmp(make, "Sony",4) && unique_id)
{
for (i = 0; i < sizeof sonique / sizeof *sonique; i++)
if (unique_id == sonique[i].id)
{
strcpy(model, sonique[i].t_model);
break;
}
}
goto dng_skip;
}
if (!strncmp(make,"Canon",5) && !fsize && tiff_bps != 15) {
if (!load_raw)
load_raw = &CLASS lossless_jpeg_load_raw;
for (i=0; i < sizeof canon / sizeof *canon; i++)
if (raw_width == canon[i][0] && raw_height == canon[i][1]) {
width = raw_width - (left_margin = canon[i][2]);
height = raw_height - (top_margin = canon[i][3]);
width -= canon[i][4];
height -= canon[i][5];
mask[0][1] = canon[i][6];
mask[0][3] = -canon[i][7];
mask[1][1] = canon[i][8];
mask[1][3] = -canon[i][9];
if (canon[i][10]) filters = canon[i][10] * 0x01010101;
}
if ((unique_id | 0x20000) == 0x2720000) {
left_margin = 8;
top_margin = 16;
}
}
if (!strncmp(make,"Canon",5) && unique_id)
{
for (i=0; i < sizeof unique / sizeof *unique; i++)
if (unique_id == 0x80000000 + unique[i].id)
{
adobe_coeff ("Canon", unique[i].t_model);
strcpy(model,unique[i].t_model);
}
}
if (!strncasecmp(make,"Sony",4) && unique_id)
{
for (i=0; i < sizeof sonique / sizeof *sonique; i++)
if (unique_id == sonique[i].id)
{
adobe_coeff ("Sony", sonique[i].t_model);
strcpy(model,sonique[i].t_model);
}
}
if (!strncmp(make,"Nikon",5)) {
if (!load_raw)
load_raw = &CLASS packed_load_raw;
if (model[0] == 'E')
load_flags |= !data_offset << 2 | 2;
}
/* Set parameters based on camera name (for non-DNG files). */
if (!strcmp(model,"KAI-0340")
&& find_green (16, 16, 3840, 5120) < 25) {
height = 480;
top_margin = filters = 0;
strcpy (model,"C603");
}
if (is_foveon) {
if (height*2 < width) pixel_aspect = 0.5;
if (height > width) pixel_aspect = 2;
filters = 0;
#ifdef LIBRAW_DEMOSAIC_PACK_GPL2
if(!imgdata.params.force_foveon_x3f)
simple_coeff(0);
#endif
} else if (!strncmp(make,"Canon",5) && tiff_bps == 15) {
switch (width) {
case 3344: width -= 66;
case 3872: width -= 6;
}
if (height > width) {
SWAP(height,width);
SWAP(raw_height,raw_width);
}
if (width == 7200 && height == 3888)
{
raw_width = width = 6480;
raw_height = height = 4320;
}
filters = 0;
tiff_samples = colors = 3;
load_raw = &CLASS canon_sraw_load_raw;
} else if (!strcmp(model,"PowerShot 600")) {
height = 613;
width = 854;
raw_width = 896;
colors = 4;
filters = 0xe1e4e1e4;
load_raw = &CLASS canon_600_load_raw;
} else if (!strcmp(model,"PowerShot A5") ||
!strcmp(model,"PowerShot A5 Zoom")) {
height = 773;
width = 960;
raw_width = 992;
pixel_aspect = 256/235.0;
filters = 0x1e4e1e4e;
goto canon_a5;
} else if (!strcmp(model,"PowerShot A50")) {
height = 968;
width = 1290;
raw_width = 1320;
filters = 0x1b4e4b1e;
goto canon_a5;
} else if (!strcmp(model,"PowerShot Pro70")) {
height = 1024;
width = 1552;
filters = 0x1e4b4e1b;
canon_a5:
colors = 4;
tiff_bps = 10;
load_raw = &CLASS packed_load_raw;
load_flags = 40;
} else if (!strcmp(model,"PowerShot Pro90 IS") ||
!strcmp(model,"PowerShot G1")) {
colors = 4;
filters = 0xb4b4b4b4;
} else if (!strcmp(model,"PowerShot A610")) {
if (canon_s2is()) strcpy (model+10, "S2 IS");
} else if (!strcmp(model,"PowerShot SX220 HS")) {
mask[1][3] = -4;
top_margin=16;
left_margin = 92;
} else if (!strcmp(model,"PowerShot S120")) {
raw_width = 4192;
raw_height = 3062;
width = 4022;
height = 3016;
mask[0][0] = top_margin = 31;
mask[0][2] = top_margin + height;
left_margin = 120;
mask[0][1] = 23;
mask[0][3] = 72;
} else if (!strcmp(model,"PowerShot G16")) {
mask[0][0] = 0;
mask[0][2] = 80;
mask[0][1] = 0;
mask[0][3] = 16;
top_margin = 29;
left_margin = 120;
width = raw_width-left_margin-48;
height = raw_height-top_margin-14;
} else if (!strcmp(model,"PowerShot SX50 HS")) {
top_margin = 17;
} else if (!strcmp(model,"EOS D2000C")) {
filters = 0x61616161;
black = curve[200];
} else if (!strcmp(model,"D1")) {
cam_mul[0] *= 256/527.0;
cam_mul[2] *= 256/317.0;
} else if (!strcmp(model,"D1X")) {
width -= 4;
pixel_aspect = 0.5;
} else if (!strcmp(model,"D40X") ||
!strcmp(model,"D60") ||
!strcmp(model,"D80") ||
!strcmp(model,"D3000")) {
height -= 3;
width -= 4;
} else if (!strcmp(model,"D3") ||
!strcmp(model,"D3S") ||
!strcmp(model,"D700")) {
width -= 4;
left_margin = 2;
} else if (!strcmp(model,"D3100")) {
width -= 28;
left_margin = 6;
} else if (!strcmp(model,"D5000") ||
!strcmp(model,"D90")) {
width -= 42;
} else if (!strcmp(model,"D5100") ||
!strcmp(model,"D7000") ||
!strcmp(model,"COOLPIX A")) {
width -= 44;
} else if (!strcmp(model,"D3200") ||
!strncmp(model,"D6",2) ||
!strncmp(model,"D800",4)) {
width -= 46;
} else if (!strcmp(model,"D4") ||
!strcmp(model,"Df")) {
width -= 52;
left_margin = 2;
} else if (!strncmp(model,"D40",3) ||
!strncmp(model,"D50",3) ||
!strncmp(model,"D70",3)) {
width--;
} else if (!strcmp(model,"D100")) {
if (load_flags)
raw_width = (width += 3) + 3;
} else if (!strcmp(model,"D200")) {
left_margin = 1;
width -= 4;
filters = 0x94949494;
} else if (!strncmp(model,"D2H",3)) {
left_margin = 6;
width -= 14;
} else if (!strncmp(model,"D2X",3)) {
if (width == 3264) width -= 32;
else width -= 8;
} else if (!strncmp(model,"D300",4)) {
width -= 32;
} else if (!strncmp(make,"Nikon",5) && raw_width == 4032) {
if(!strcmp(model,"COOLPIX P7700"))
{
adobe_coeff ("Nikon","COOLPIX P7700");
maximum = 65504;
load_flags = 0;
}
else if(!strcmp(model,"COOLPIX P7800"))
{
adobe_coeff ("Nikon","COOLPIX P7800");
maximum = 65504;
load_flags = 0;
}
else if(!strcmp(model,"COOLPIX P340"))
load_flags=0;
} else if (!strncmp(model,"COOLPIX P",9) && raw_width != 4032) {
load_flags = 24;
filters = 0x94949494;
if (model[9] == '7' && (iso_speed >= 400 || iso_speed==0) && !strstr(software,"V1.2") )
black = 255;
} else if (!strncmp(model,"1 ",2)) {
height -= 2;
} else if (fsize == 1581060) {
simple_coeff(3);
pre_mul[0] = 1.2085;
pre_mul[1] = 1.0943;
pre_mul[3] = 1.1103;
} else if (fsize == 3178560) {
cam_mul[0] *= 4;
cam_mul[2] *= 4;
} else if (fsize == 4771840) {
if (!timestamp && nikon_e995())
strcpy (model, "E995");
if (strcmp(model,"E995")) {
filters = 0xb4b4b4b4;
simple_coeff(3);
pre_mul[0] = 1.196;
pre_mul[1] = 1.246;
pre_mul[2] = 1.018;
}
} else if (fsize == 2940928) {
if (!timestamp && !nikon_e2100())
strcpy (model,"E2500");
if (!strcmp(model,"E2500")) {
height -= 2;
load_flags = 6;
colors = 4;
filters = 0x4b4b4b4b;
}
} else if (fsize == 4775936) {
if (!timestamp) nikon_3700();
if (model[0] == 'E' && atoi(model+1) < 3700)
filters = 0x49494949;
if (!strcmp(model,"Optio 33WR")) {
flip = 1;
filters = 0x16161616;
}
if (make[0] == 'O') {
i = find_green (12, 32, 1188864, 3576832);
c = find_green (12, 32, 2383920, 2387016);
if (abs(i) < abs(c)) {
SWAP(i,c);
load_flags = 24;
}
if (i < 0) filters = 0x61616161;
}
} else if (fsize == 5869568) {
if (!timestamp && minolta_z2()) {
strcpy (make, "Minolta");
strcpy (model,"DiMAGE Z2");
}
load_flags = 6 + 24*(make[0] == 'M');
} else if (fsize == 6291456) {
fseek (ifp, 0x300000, SEEK_SET);
if ((order = guess_byte_order(0x10000)) == 0x4d4d) {
height -= (top_margin = 16);
width -= (left_margin = 28);
maximum = 0xf5c0;
strcpy (make, "ISG");
model[0] = 0;
}
} else if (!strncmp(make,"Fujifilm",8)) {
if (!strcmp(model+7,"S2Pro")) {
strcpy (model,"S2Pro");
height = 2144;
width = 2880;
flip = 6;
} else if (load_raw != &CLASS packed_load_raw)
maximum = (is_raw == 2 && shot_select) ? 0x2f00 : 0x3e00;
top_margin = (raw_height - height) >> 2 << 1;
left_margin = (raw_width - width ) >> 2 << 1;
if (width == 2848 || width == 3664) filters = 0x16161616;
if (width == 4032 || width == 4952) left_margin = 0;
if (width == 3328 && (width -= 66)) left_margin = 34;
if (width == 4936) left_margin = 4;
if (!strcmp(model,"HS50EXR") ||
!strcmp(model,"F900EXR")) {
width += 2;
left_margin = 0;
filters = 0x16161616;
}
if(!strcmp(model,"S5500"))
{
height -= (top_margin=6);
}
if (fuji_layout) raw_width *= is_raw;
if (filters == 9)
FORC(36) ((char *)xtrans)[c] =
xtrans_abs[(c/6+top_margin) % 6][(c+left_margin) % 6];
} else if (!strcmp(model,"KD-400Z")) {
height = 1712;
width = 2312;
raw_width = 2336;
goto konica_400z;
} else if (!strcmp(model,"KD-510Z")) {
goto konica_510z;
} else if (!strncasecmp(make,"Minolta",7)) {
if (!load_raw && (maximum = 0xfff))
load_raw = &CLASS unpacked_load_raw;
if (!strncmp(model,"DiMAGE A",8)) {
if (!strcmp(model,"DiMAGE A200"))
filters = 0x49494949;
tiff_bps = 12;
load_raw = &CLASS packed_load_raw;
} else if (!strncmp(model,"ALPHA",5) ||
!strncmp(model,"DYNAX",5) ||
!strncmp(model,"MAXXUM",6)) {
sprintf (model+20, "DYNAX %-10s", model+6+(model[0]=='M'));
adobe_coeff (make, model+20);
load_raw = &CLASS packed_load_raw;
} else if (!strncmp(model,"DiMAGE G",8)) {
if (model[8] == '4') {
height = 1716;
width = 2304;
} else if (model[8] == '5') {
konica_510z:
height = 1956;
width = 2607;
raw_width = 2624;
} else if (model[8] == '6') {
height = 2136;
width = 2848;
}
data_offset += 14;
filters = 0x61616161;
konica_400z:
load_raw = &CLASS unpacked_load_raw;
maximum = 0x3df;
order = 0x4d4d;
}
} else if (!strcmp(model,"*ist D")) {
load_raw = &CLASS unpacked_load_raw;
data_error = -1;
} else if (!strcmp(model,"*ist DS")) {
height -= 2;
} else if (!strncmp(make,"Samsung",7) && raw_width == 4704) {
height -= top_margin = 8;
width -= 2 * (left_margin = 8);
load_flags = 32;
} else if (!strncmp(make,"Samsung",7) && !strcmp(model,"NX3000")) {
top_margin = 24;
left_margin = 64;
width = 5472;
height = 3648;
filters = 0x61616161;
colors = 3;
} else if (!strncmp(make,"Samsung",7) && raw_height == 3714) {
height -= top_margin = 18;
left_margin = raw_width - (width = 5536);
if (raw_width != 5600)
left_margin = top_margin = 0;
filters = 0x61616161;
colors = 3;
} else if (!strncmp(make,"Samsung",7) && raw_width == 5632) {
order = 0x4949;
height = 3694;
top_margin = 2;
width = 5574 - (left_margin = 32 + tiff_bps);
if (tiff_bps == 12) load_flags = 80;
} else if (!strncmp(make,"Samsung",7) && raw_width == 5664) {
height -= top_margin = 17;
left_margin = 96;
width = 5544;
filters = 0x49494949;
} else if (!strncmp(make,"Samsung",7) && raw_width == 6496) {
filters = 0x61616161;
#ifdef LIBRAW_LIBRARY_BUILD
if(!black && !cblack[0] && !cblack[1] && !cblack[2] && !cblack[3])
#endif
black = 1 << (tiff_bps - 7);
} else if (!strcmp(model,"EX1")) {
order = 0x4949;
height -= 20;
top_margin = 2;
if ((width -= 6) > 3682) {
height -= 10;
width -= 46;
top_margin = 8;
}
} else if (!strcmp(model,"WB2000")) {
order = 0x4949;
height -= 3;
top_margin = 2;
if ((width -= 10) > 3718) {
height -= 28;
width -= 56;
top_margin = 8;
}
} else if (strstr(model,"WB550")) {
strcpy (model, "WB550");
} else if (!strcmp(model,"EX2F")) {
height = 3045;
width = 4070;
top_margin = 3;
order = 0x4949;
filters = 0x49494949;
load_raw = &CLASS unpacked_load_raw;
} else if (!strcmp(model,"STV680 VGA")) {
black = 16;
} else if (!strcmp(model,"N95")) {
height = raw_height - (top_margin = 2);
} else if (!strcmp(model,"640x480")) {
gamma_curve (0.45, 4.5, 1, 255);
} else if (!strncmp(make,"Hasselblad",10)) {
if (load_raw == &CLASS lossless_jpeg_load_raw)
load_raw = &CLASS hasselblad_load_raw;
if (raw_width == 7262) {
height = 5444;
width = 7248;
top_margin = 4;
left_margin = 7;
filters = 0x61616161;
if(!strncasecmp(model,"H3D",3))
{
adobe_coeff("Hasselblad","H3DII-39");
strcpy(model,"H3DII-39");
}
} else if (raw_width == 7410 || raw_width == 8282) {
height -= 84;
width -= 82;
top_margin = 4;
left_margin = 41;
filters = 0x61616161;
adobe_coeff("Hasselblad","H4D-40");
strcpy(model,"H4D-40");
} else if (raw_width == 9044) {
if(black > 500)
{
top_margin = 12;
left_margin = 44;
width = 8956;
height = 6708;
memset(cblack,0,sizeof(cblack));
adobe_coeff("Hasselblad","H4D-60");
strcpy(model,"H4D-60");
black = 512;
}
else
{
height = 6716;
width = 8964;
top_margin = 8;
left_margin = 40;
black += load_flags = 256;
maximum = 0x8101;
strcpy(model,"H3DII-60");
}
} else if (raw_width == 4090) {
strcpy (model, "V96C");
height -= (top_margin = 6);
width -= (left_margin = 3) + 7;
filters = 0x61616161;
} else if (raw_width == 8282 && raw_height == 6240) {
if(!strncasecmp(model,"H5D",3))
{
/* H5D 50*/
left_margin = 54;
top_margin = 16;
width = 8176;
height = 6132;
black = 256;
strcpy(model,"H5D-50");
}
else if(!strncasecmp(model,"H3D",3))
{
black=0;
left_margin = 54;
top_margin = 16;
width = 8176;
height = 6132;
memset(cblack,0,sizeof(cblack));
adobe_coeff("Hasselblad","H3D-50");
strcpy(model,"H3D-50");
}
} else if (raw_width == 8374 && raw_height == 6304) {
/* H5D 50c*/
left_margin = 52;
top_margin = 100;
width = 8272;
height = 6200;
black = 256;
strcpy(model,"H5D-50c");
}
if (tiff_samples > 1) {
is_raw = tiff_samples+1;
if (!shot_select && !half_size) filters = 0;
}
} else if (!strncmp(make,"Sinar",5)) {
if (!load_raw) load_raw = &CLASS unpacked_load_raw;
if (is_raw > 1 && !shot_select && !half_size) filters = 0;
maximum = 0x3fff;
} else if (!strncmp(make,"Leaf",4)) {
maximum = 0x3fff;
fseek (ifp, data_offset, SEEK_SET);
if (ljpeg_start (&jh, 1) && jh.bits == 15)
maximum = 0x1fff;
if (tiff_samples > 1) filters = 0;
if (tiff_samples > 1 || tile_length < raw_height) {
load_raw = &CLASS leaf_hdr_load_raw;
raw_width = tile_width;
}
if ((width | height) == 2048) {
if (tiff_samples == 1) {
filters = 1;
strcpy (cdesc, "RBTG");
strcpy (model, "CatchLight");
top_margin = 8; left_margin = 18; height = 2032; width = 2016;
} else {
strcpy (model, "DCB2");
top_margin = 10; left_margin = 16; height = 2028; width = 2022;
}
} else if (width+height == 3144+2060) {
if (!model[0]) strcpy (model, "Cantare");
if (width > height) {
top_margin = 6; left_margin = 32; height = 2048; width = 3072;
filters = 0x61616161;
} else {
left_margin = 6; top_margin = 32; width = 2048; height = 3072;
filters = 0x16161616;
}
if (!cam_mul[0] || model[0] == 'V') filters = 0;
else is_raw = tiff_samples;
} else if (width == 2116) {
strcpy (model, "Valeo 6");
height -= 2 * (top_margin = 30);
width -= 2 * (left_margin = 55);
filters = 0x49494949;
} else if (width == 3171) {
strcpy (model, "Valeo 6");
height -= 2 * (top_margin = 24);
width -= 2 * (left_margin = 24);
filters = 0x16161616;
}
} else if (!strncmp(make,"Leica",5) || !strncmp(make,"Panasonic",9)) {
if (raw_width > 0&& ((flen - data_offset) / (raw_width*8/7) == raw_height) )
load_raw = &CLASS panasonic_load_raw;
if (!load_raw) {
load_raw = &CLASS unpacked_load_raw;
load_flags = 4;
}
zero_is_bad = 1;
if ((height += 12) > raw_height) height = raw_height;
for (i=0; i < sizeof pana / sizeof *pana; i++)
if (raw_width == pana[i][0] && raw_height == pana[i][1]) {
left_margin = pana[i][2];
top_margin = pana[i][3];
width += pana[i][4];
height += pana[i][5];
}
filters = 0x01010101 * (uchar) "\x94\x61\x49\x16"
[((filters-1) ^ (left_margin & 1) ^ (top_margin << 1)) & 3];
} else if (!strcmp(model,"C770UZ")) {
height = 1718;
width = 2304;
filters = 0x16161616;
load_raw = &CLASS packed_load_raw;
load_flags = 30;
} else if (!strncmp(make,"Olympus",7)) {
height += height & 1;
if (exif_cfa) filters = exif_cfa;
if (width == 4100) width -= 4;
if (width == 4080) width -= 24;
if (width == 9280) { width -= 6; height -= 6; }
if (load_raw == &CLASS unpacked_load_raw)
load_flags = 4;
tiff_bps = 12;
if (!strcmp(model,"E-300") ||
!strcmp(model,"E-500")) {
width -= 20;
if (load_raw == &CLASS unpacked_load_raw) {
maximum = 0xfc3;
memset (cblack, 0, sizeof cblack);
}
} else if (!strcmp(model,"STYLUS1")) {
width -= 14;
maximum = 0xfff;
} else if (!strcmp(model,"E-330")) {
width -= 30;
if (load_raw == &CLASS unpacked_load_raw)
maximum = 0xf79;
} else if (!strcmp(model,"SP550UZ")) {
thumb_length = flen - (thumb_offset = 0xa39800);
thumb_height = 480;
thumb_width = 640;
}
} else if (!strcmp(model,"N Digital")) {
height = 2047;
width = 3072;
filters = 0x61616161;
data_offset = 0x1a00;
load_raw = &CLASS packed_load_raw;
} else if (!strcmp(model,"DSC-F828")) {
width = 3288;
left_margin = 5;
mask[1][3] = -17;
data_offset = 862144;
load_raw = &CLASS sony_load_raw;
filters = 0x9c9c9c9c;
colors = 4;
strcpy (cdesc, "RGBE");
} else if (!strcmp(model,"DSC-V3")) {
width = 3109;
left_margin = 59;
mask[0][1] = 9;
data_offset = 787392;
load_raw = &CLASS sony_load_raw;
} else if (!strncmp(make,"Sony",4) && raw_width == 3984) {
width = 3925;
order = 0x4d4d;
} else if (!strncmp(make,"Sony",4) && raw_width == 4288) {
width -= 32;
} else if (!strncmp(make,"Sony",4) && raw_width == 4928) {
if (height < 3280) width -= 8;
} else if (!strncmp(make,"Sony",4) && raw_width == 5504) { // ILCE-3000//5000
width -= height > 3664 ? 8 : 32;
} else if (!strncmp(make,"Sony",4) && raw_width == 6048) {
width -= 24;
if (strstr(model,"RX1") || strstr(model,"A99"))
width -= 6;
} else if (!strncmp(make,"Sony",4) && raw_width == 7392) {
width -= 30;
} else if (!strncmp(make,"Sony",4) && raw_width == 8000) {
width -= 32;
} else if (!strcmp(model,"DSLR-A100")) {
if (width == 3880) {
height--;
width = ++raw_width;
} else {
height -= 4;
width -= 4;
order = 0x4d4d;
load_flags = 2;
}
filters = 0x61616161;
} else if (!strcmp(model,"DSLR-A350")) {
height -= 4;
} else if (!strcmp(model,"PIXL")) {
height -= top_margin = 4;
width -= left_margin = 32;
gamma_curve (0, 7, 1, 255);
} else if (!strcmp(model,"C603") || !strcmp(model,"C330")
|| !strcmp(model,"12MP")) {
order = 0x4949;
if (filters && data_offset) {
fseek (ifp, data_offset < 4096 ? 168 : 5252, SEEK_SET);
read_shorts (curve, 256);
} else gamma_curve (0, 3.875, 1, 255);
load_raw = filters ? &CLASS eight_bit_load_raw :
strcmp(model,"C330") ? &CLASS kodak_c603_load_raw :
&CLASS kodak_c330_load_raw;
load_flags = tiff_bps > 16;
tiff_bps = 8;
} else if (!strncasecmp(model,"EasyShare",9)) {
data_offset = data_offset < 0x15000 ? 0x15000 : 0x17000;
load_raw = &CLASS packed_load_raw;
} else if (!strncasecmp(make,"Kodak",5)) {
if (filters == UINT_MAX) filters = 0x61616161;
if (!strncmp(model,"NC2000",6) ||
!strncmp(model,"EOSDCS",6) ||
!strncmp(model,"DCS4",4)) {
width -= 4;
left_margin = 2;
if (model[6] == ' ') model[6] = 0;
if (!strcmp(model,"DCS460A")) goto bw;
} else if (!strcmp(model,"DCS660M")) {
black = 214;
goto bw;
} else if (!strcmp(model,"DCS760M")) {
bw: colors = 1;
filters = 0;
}
if (!strcmp(model+4,"20X"))
strcpy (cdesc, "MYCY");
if (strstr(model,"DC25")) {
strcpy (model, "DC25");
data_offset = 15424;
}
if (!strncmp(model,"DC2",3)) {
raw_height = 2 + (height = 242);
if (!strncmp(model, "DC290", 5))
iso_speed = 100;
if (!strncmp(model, "DC280", 5))
iso_speed = 70;
if (flen < 100000) {
raw_width = 256; width = 249;
pixel_aspect = (4.0*height) / (3.0*width);
} else {
raw_width = 512; width = 501;
pixel_aspect = (493.0*height) / (373.0*width);
}
top_margin = left_margin = 1;
colors = 4;
filters = 0x8d8d8d8d;
simple_coeff(1);
pre_mul[1] = 1.179;
pre_mul[2] = 1.209;
pre_mul[3] = 1.036;
load_raw = &CLASS eight_bit_load_raw;
} else if (!strcmp(model,"40")) {
strcpy (model, "DC40");
height = 512;
width = 768;
data_offset = 1152;
load_raw = &CLASS kodak_radc_load_raw;
} else if (strstr(model,"DC50")) {
strcpy (model, "DC50");
height = 512;
width = 768;
iso_speed=84;
data_offset = 19712;
load_raw = &CLASS kodak_radc_load_raw;
} else if (strstr(model,"DC120")) {
strcpy (model, "DC120");
height = 976;
width = 848;
iso_speed=160;
pixel_aspect = height/0.75/width;
load_raw = tiff_compress == 7 ?
&CLASS kodak_jpeg_load_raw : &CLASS kodak_dc120_load_raw;
} else if (!strcmp(model,"DCS200")) {
thumb_height = 128;
thumb_width = 192;
thumb_offset = 6144;
thumb_misc = 360;
iso_speed=140;
write_thumb = &CLASS layer_thumb;
black = 17;
}
} else if (!strcmp(model,"Fotoman Pixtura")) {
height = 512;
width = 768;
data_offset = 3632;
load_raw = &CLASS kodak_radc_load_raw;
filters = 0x61616161;
simple_coeff(2);
} else if (!strncmp(model,"QuickTake",9)) {
if (head[5]) strcpy (model+10, "200");
fseek (ifp, 544, SEEK_SET);
height = get2();
width = get2();
data_offset = (get4(),get2()) == 30 ? 738:736;
if (height > width) {
SWAP(height,width);
fseek (ifp, data_offset-6, SEEK_SET);
flip = ~get2() & 3 ? 5:6;
}
filters = 0x61616161;
} else if (!strncmp(make,"Rollei",6) && !load_raw) {
switch (raw_width) {
case 1316:
height = 1030;
width = 1300;
top_margin = 1;
left_margin = 6;
break;
case 2568:
height = 1960;
width = 2560;
top_margin = 2;
left_margin = 8;
}
filters = 0x16161616;
load_raw = &CLASS rollei_load_raw;
}
else if (!strcmp(model,"GRAS-50S5C")) {
height = 2048;
width = 2440;
load_raw = &CLASS unpacked_load_raw;
data_offset = 0;
filters = 0x49494949;
order = 0x4949;
maximum = 0xfffC;
} else if (!strcmp(model,"BB-500CL")) {
height = 2058;
width = 2448;
load_raw = &CLASS unpacked_load_raw;
data_offset = 0;
filters = 0x94949494;
order = 0x4949;
maximum = 0x3fff;
} else if (!strcmp(model,"BB-500GE")) {
height = 2058;
width = 2456;
load_raw = &CLASS unpacked_load_raw;
data_offset = 0;
filters = 0x94949494;
order = 0x4949;
maximum = 0x3fff;
} else if (!strcmp(model,"SVS625CL")) {
height = 2050;
width = 2448;
load_raw = &CLASS unpacked_load_raw;
data_offset = 0;
filters = 0x94949494;
order = 0x4949;
maximum = 0x0fff;
}
/* Early reject for damaged images */
if (!load_raw || height < 22 || width < 22 ||
#ifdef LIBRAW_LIBRARY_BUILD
(tiff_bps > 16 && load_raw != &LibRaw::deflate_dng_load_raw)
#else
tiff_bps > 16
#endif
|| tiff_samples > 4 || colors > 4 || colors < 1)
{
is_raw = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_IDENTIFY,1,2);
#endif
return;
}
if (!model[0])
sprintf (model, "%dx%d", width, height);
if (filters == UINT_MAX) filters = 0x94949494;
if (thumb_offset && !thumb_height) {
fseek (ifp, thumb_offset, SEEK_SET);
if (ljpeg_start (&jh, 1)) {
thumb_width = jh.wide;
thumb_height = jh.high;
}
}
dng_skip:
/* Early reject for damaged images */
if (!load_raw || height < 22 || width < 22 ||
#ifdef LIBRAW_LIBRARY_BUILD
(tiff_bps > 16 && load_raw != &LibRaw::deflate_dng_load_raw)
#else
tiff_bps > 16
#endif
|| tiff_samples > 4 || colors > 4 || colors < 1)
{
is_raw = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_IDENTIFY,1,2);
#endif
return;
}
if ((use_camera_matrix & (use_camera_wb || dng_version))
&& cmatrix[0][0] > 0.125) {
memcpy (rgb_cam, cmatrix, sizeof cmatrix);
raw_color = 0;
}
if (raw_color) adobe_coeff (make, model);
#ifdef LIBRAW_LIBRARY_BUILD
else if(imgdata.color.cam_xyz[0][0]<0.01)
adobe_coeff (make, model,1);
#endif
if (load_raw == &CLASS kodak_radc_load_raw)
if (raw_color) adobe_coeff ("Apple","Quicktake");
if (fuji_width) {
fuji_width = width >> !fuji_layout;
filters = fuji_width & 1 ? 0x94949494 : 0x49494949;
width = (height >> fuji_layout) + fuji_width;
height = width - 1;
pixel_aspect = 1;
} else {
if (raw_height < height) raw_height = height;
if (raw_width < width ) raw_width = width;
}
if (!tiff_bps) tiff_bps = 12;
if (!maximum)
{
maximum = (1 << tiff_bps) - 1;
if(maximum < 0x10000 && curve[maximum]>0 && load_raw == &CLASS sony_arw2_load_raw)
maximum = curve[maximum];
}
if (!load_raw || height < 22 || width < 22 ||
#ifdef LIBRAW_LIBRARY_BUILD
(tiff_bps > 16 && load_raw != &LibRaw::deflate_dng_load_raw)
#else
tiff_bps > 16
#endif
|| tiff_samples > 6 || colors > 4)
is_raw = 0;
if(raw_width < 22 || raw_width > 64000 || raw_height < 22 || raw_width > 64000)
is_raw = 0;
#ifdef NO_JASPER
if (load_raw == &CLASS redcine_load_raw) {
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: You must link dcraw with %s!!\n"),
ifname, "libjasper");
#endif
is_raw = 0;
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_JASPER;
#endif
}
#endif
#ifdef NO_JPEG
if (load_raw == &CLASS kodak_jpeg_load_raw ||
load_raw == &CLASS lossy_dng_load_raw) {
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s: You must link dcraw with %s!!\n"),
ifname, "libjpeg");
#endif
is_raw = 0;
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_JPEGLIB;
#endif
}
#endif
if (!cdesc[0])
strcpy (cdesc, colors == 3 ? "RGBG":"GMCY");
if (!raw_height) raw_height = height;
if (!raw_width ) raw_width = width;
if (filters > 999 && colors == 3)
filters |= ((filters >> 2 & 0x22222222) |
(filters << 2 & 0x88888888)) & filters << 1;
notraw:
if (flip == UINT_MAX) flip = tiff_flip;
if (flip == UINT_MAX) flip = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_IDENTIFY,1,2);
#endif
}
//@end COMMON
//@out FILEIO
#ifndef NO_LCMS
void CLASS apply_profile (const char *input, const char *output)
{
char *prof;
cmsHPROFILE hInProfile=0, hOutProfile=0;
cmsHTRANSFORM hTransform;
FILE *fp;
unsigned size;
if (strcmp (input, "embed"))
hInProfile = cmsOpenProfileFromFile (input, "r");
else if (profile_length) {
#ifndef LIBRAW_LIBRARY_BUILD
prof = (char *) malloc (profile_length);
merror (prof, "apply_profile()");
fseek (ifp, profile_offset, SEEK_SET);
fread (prof, 1, profile_length, ifp);
hInProfile = cmsOpenProfileFromMem (prof, profile_length);
free (prof);
#else
hInProfile = cmsOpenProfileFromMem (imgdata.color.profile, profile_length);
#endif
} else
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_EMBEDDED_PROFILE;
#endif
#ifdef DCRAW_VERBOSE
fprintf (stderr,_("%s has no embedded profile.\n"), ifname);
#endif
}
if (!hInProfile)
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_NO_INPUT_PROFILE;
#endif
return;
}
if (!output)
hOutProfile = cmsCreate_sRGBProfile();
else if ((fp = fopen (output, "rb"))) {
fread (&size, 4, 1, fp);
fseek (fp, 0, SEEK_SET);
oprof = (unsigned *) malloc (size = ntohl(size));
merror (oprof, "apply_profile()");
fread (oprof, 1, size, fp);
fclose (fp);
if (!(hOutProfile = cmsOpenProfileFromMem (oprof, size))) {
free (oprof);
oprof = 0;
}
}
#ifdef DCRAW_VERBOSE
else
fprintf (stderr,_("Cannot open file %s!\n"), output);
#endif
if (!hOutProfile)
{
#ifdef LIBRAW_LIBRARY_BUILD
imgdata.process_warnings |= LIBRAW_WARN_BAD_OUTPUT_PROFILE;
#endif
goto quit;
}
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Applying color profile...\n"));
#endif
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_APPLY_PROFILE,0,2);
#endif
hTransform = cmsCreateTransform (hInProfile, TYPE_RGBA_16,
hOutProfile, TYPE_RGBA_16, INTENT_PERCEPTUAL, 0);
cmsDoTransform (hTransform, image, image, width*height);
raw_color = 1; /* Don't use rgb_cam with a profile */
cmsDeleteTransform (hTransform);
cmsCloseProfile (hOutProfile);
quit:
cmsCloseProfile (hInProfile);
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_APPLY_PROFILE,1,2);
#endif
}
#endif
//@end FILEIO
//@out COMMON
void CLASS convert_to_rgb()
{
#ifndef LIBRAW_LIBRARY_BUILD
int row, col, c;
#endif
int i, j, k;
#ifndef LIBRAW_LIBRARY_BUILD
ushort *img;
float out[3];
#endif
float out_cam[3][4];
double num, inverse[3][3];
static const double xyzd50_srgb[3][3] =
{ { 0.436083, 0.385083, 0.143055 },
{ 0.222507, 0.716888, 0.060608 },
{ 0.013930, 0.097097, 0.714022 } };
static const double rgb_rgb[3][3] =
{ { 1,0,0 }, { 0,1,0 }, { 0,0,1 } };
static const double adobe_rgb[3][3] =
{ { 0.715146, 0.284856, 0.000000 },
{ 0.000000, 1.000000, 0.000000 },
{ 0.000000, 0.041166, 0.958839 } };
static const double wide_rgb[3][3] =
{ { 0.593087, 0.404710, 0.002206 },
{ 0.095413, 0.843149, 0.061439 },
{ 0.011621, 0.069091, 0.919288 } };
static const double prophoto_rgb[3][3] =
{ { 0.529317, 0.330092, 0.140588 },
{ 0.098368, 0.873465, 0.028169 },
{ 0.016879, 0.117663, 0.865457 } };
static const double (*out_rgb[])[3] =
{ rgb_rgb, adobe_rgb, wide_rgb, prophoto_rgb, xyz_rgb };
static const char *name[] =
{ "sRGB", "Adobe RGB (1998)", "WideGamut D65", "ProPhoto D65", "XYZ" };
static const unsigned phead[] =
{ 1024, 0, 0x2100000, 0x6d6e7472, 0x52474220, 0x58595a20, 0, 0, 0,
0x61637370, 0, 0, 0x6e6f6e65, 0, 0, 0, 0, 0xf6d6, 0x10000, 0xd32d };
unsigned pbody[] =
{ 10, 0x63707274, 0, 36, /* cprt */
0x64657363, 0, 40, /* desc */
0x77747074, 0, 20, /* wtpt */
0x626b7074, 0, 20, /* bkpt */
0x72545243, 0, 14, /* rTRC */
0x67545243, 0, 14, /* gTRC */
0x62545243, 0, 14, /* bTRC */
0x7258595a, 0, 20, /* rXYZ */
0x6758595a, 0, 20, /* gXYZ */
0x6258595a, 0, 20 }; /* bXYZ */
static const unsigned pwhite[] = { 0xf351, 0x10000, 0x116cc };
unsigned pcurve[] = { 0x63757276, 0, 1, 0x1000000 };
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_CONVERT_RGB,0,2);
#endif
gamma_curve (gamm[0], gamm[1], 0, 0);
memcpy (out_cam, rgb_cam, sizeof out_cam);
#ifndef LIBRAW_LIBRARY_BUILD
raw_color |= colors == 1 || document_mode ||
output_color < 1 || output_color > 5;
#else
raw_color |= colors == 1 ||
output_color < 1 || output_color > 5;
#endif
if (!raw_color) {
oprof = (unsigned *) calloc (phead[0], 1);
merror (oprof, "convert_to_rgb()");
memcpy (oprof, phead, sizeof phead);
if (output_color == 5) oprof[4] = oprof[5];
oprof[0] = 132 + 12*pbody[0];
for (i=0; i < pbody[0]; i++) {
oprof[oprof[0]/4] = i ? (i > 1 ? 0x58595a20 : 0x64657363) : 0x74657874;
pbody[i*3+2] = oprof[0];
oprof[0] += (pbody[i*3+3] + 3) & -4;
}
memcpy (oprof+32, pbody, sizeof pbody);
oprof[pbody[5]/4+2] = strlen(name[output_color-1]) + 1;
memcpy ((char *)oprof+pbody[8]+8, pwhite, sizeof pwhite);
pcurve[3] = (short)(256/gamm[5]+0.5) << 16;
for (i=4; i < 7; i++)
memcpy ((char *)oprof+pbody[i*3+2], pcurve, sizeof pcurve);
pseudoinverse ((double (*)[3]) out_rgb[output_color-1], inverse, 3);
for (i=0; i < 3; i++)
for (j=0; j < 3; j++) {
for (num = k=0; k < 3; k++)
num += xyzd50_srgb[i][k] * inverse[j][k];
oprof[pbody[j*3+23]/4+i+2] = num * 0x10000 + 0.5;
}
for (i=0; i < phead[0]/4; i++)
oprof[i] = htonl(oprof[i]);
strcpy ((char *)oprof+pbody[2]+8, "auto-generated by dcraw");
strcpy ((char *)oprof+pbody[5]+12, name[output_color-1]);
for (i=0; i < 3; i++)
for (j=0; j < colors; j++)
for (out_cam[i][j] = k=0; k < 3; k++)
out_cam[i][j] += out_rgb[output_color-1][i][k] * rgb_cam[k][j];
}
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr, raw_color ? _("Building histograms...\n") :
_("Converting to %s colorspace...\n"), name[output_color-1]);
#endif
#ifdef LIBRAW_LIBRARY_BUILD
convert_to_rgb_loop(out_cam);
#else
memset (histogram, 0, sizeof histogram);
for (img=image[0], row=0; row < height; row++)
for (col=0; col < width; col++, img+=4) {
if (!raw_color) {
out[0] = out[1] = out[2] = 0;
FORCC {
out[0] += out_cam[0][c] * img[c];
out[1] += out_cam[1][c] * img[c];
out[2] += out_cam[2][c] * img[c];
}
FORC3 img[c] = CLIP((int) out[c]);
}
else if (document_mode)
img[0] = img[fcol(row,col)];
FORCC histogram[c][img[c] >> 3]++;
}
#endif
if (colors == 4 && output_color) colors = 3;
#ifndef LIBRAW_LIBRARY_BUILD
if (document_mode && filters) colors = 1;
#endif
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_CONVERT_RGB,1,2);
#endif
}
void CLASS fuji_rotate()
{
int i, row, col;
double step;
float r, c, fr, fc;
unsigned ur, uc;
ushort wide, high, (*img)[4], (*pix)[4];
if (!fuji_width) return;
#ifdef DCRAW_VERBOSE
if (verbose)
fprintf (stderr,_("Rotating image 45 degrees...\n"));
#endif
fuji_width = (fuji_width - 1 + shrink) >> shrink;
step = sqrt(0.5);
wide = fuji_width / step;
high = (height - fuji_width) / step;
img = (ushort (*)[4]) calloc (high, wide*sizeof *img);
merror (img, "fuji_rotate()");
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_FUJI_ROTATE,0,2);
#endif
for (row=0; row < high; row++)
for (col=0; col < wide; col++) {
ur = r = fuji_width + (row-col)*step;
uc = c = (row+col)*step;
if (ur > height-2 || uc > width-2) continue;
fr = r - ur;
fc = c - uc;
pix = image + ur*width + uc;
for (i=0; i < colors; i++)
img[row*wide+col][i] =
(pix[ 0][i]*(1-fc) + pix[ 1][i]*fc) * (1-fr) +
(pix[width][i]*(1-fc) + pix[width+1][i]*fc) * fr;
}
free (image);
width = wide;
height = high;
image = img;
fuji_width = 0;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_FUJI_ROTATE,1,2);
#endif
}
void CLASS stretch()
{
ushort newdim, (*img)[4], *pix0, *pix1;
int row, col, c;
double rc, frac;
if (pixel_aspect == 1) return;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_STRETCH,0,2);
#endif
#ifdef DCRAW_VERBOSE
if (verbose) fprintf (stderr,_("Stretching the image...\n"));
#endif
if (pixel_aspect < 1) {
newdim = height / pixel_aspect + 0.5;
img = (ushort (*)[4]) calloc (width, newdim*sizeof *img);
merror (img, "stretch()");
for (rc=row=0; row < newdim; row++, rc+=pixel_aspect) {
frac = rc - (c = rc);
pix0 = pix1 = image[c*width];
if (c+1 < height) pix1 += width*4;
for (col=0; col < width; col++, pix0+=4, pix1+=4)
FORCC img[row*width+col][c] = pix0[c]*(1-frac) + pix1[c]*frac + 0.5;
}
height = newdim;
} else {
newdim = width * pixel_aspect + 0.5;
img = (ushort (*)[4]) calloc (height, newdim*sizeof *img);
merror (img, "stretch()");
for (rc=col=0; col < newdim; col++, rc+=1/pixel_aspect) {
frac = rc - (c = rc);
pix0 = pix1 = image[c];
if (c+1 < width) pix1 += 4;
for (row=0; row < height; row++, pix0+=width*4, pix1+=width*4)
FORCC img[row*newdim+col][c] = pix0[c]*(1-frac) + pix1[c]*frac + 0.5;
}
width = newdim;
}
free (image);
image = img;
#ifdef LIBRAW_LIBRARY_BUILD
RUN_CALLBACK(LIBRAW_PROGRESS_STRETCH,1,2);
#endif
}
int CLASS flip_index (int row, int col)
{
if (flip & 4) SWAP(row,col);
if (flip & 2) row = iheight - 1 - row;
if (flip & 1) col = iwidth - 1 - col;
return row * iwidth + col;
}
//@end COMMON
struct tiff_tag {
ushort tag, type;
int count;
union { char c[4]; short s[2]; int i; } val;
};
struct tiff_hdr {
ushort t_order, magic;
int ifd;
ushort pad, ntag;
struct tiff_tag tag[23];
int nextifd;
ushort pad2, nexif;
struct tiff_tag exif[4];
ushort pad3, ngps;
struct tiff_tag gpst[10];
short bps[4];
int rat[10];
unsigned gps[26];
char t_desc[512], t_make[64], t_model[64], soft[32], date[20], t_artist[64];
};
//@out COMMON
void CLASS tiff_set (struct tiff_hdr *th, ushort *ntag,
ushort tag, ushort type, int count, int val)
{
struct tiff_tag *tt;
int c;
tt = (struct tiff_tag *)(ntag+1) + (*ntag)++;
tt->val.i = val;
if (type == 1 && count <= 4)
FORC(4) tt->val.c[c] = val >> (c << 3);
else if (type == 2) {
count = strnlen((char *)th + val, count-1) + 1;
if (count <= 4)
FORC(4) tt->val.c[c] = ((char *)th)[val+c];
} else if (type == 3 && count <= 2)
FORC(2) tt->val.s[c] = val >> (c << 4);
tt->count = count;
tt->type = type;
tt->tag = tag;
}
#define TOFF(ptr) ((char *)(&(ptr)) - (char *)th)
void CLASS tiff_head (struct tiff_hdr *th, int full)
{
int c, psize=0;
struct tm *t;
memset (th, 0, sizeof *th);
th->t_order = htonl(0x4d4d4949) >> 16;
th->magic = 42;
th->ifd = 10;
th->rat[0] = th->rat[2] = 300;
th->rat[1] = th->rat[3] = 1;
FORC(6) th->rat[4+c] = 1000000;
th->rat[4] *= shutter;
th->rat[6] *= aperture;
th->rat[8] *= focal_len;
strncpy (th->t_desc, desc, 512);
strncpy (th->t_make, make, 64);
strncpy (th->t_model, model, 64);
strcpy (th->soft, "dcraw v"DCRAW_VERSION);
t = localtime (×tamp);
sprintf (th->date, "%04d:%02d:%02d %02d:%02d:%02d",
t->tm_year+1900,t->tm_mon+1,t->tm_mday,t->tm_hour,t->tm_min,t->tm_sec);
strncpy (th->t_artist, artist, 64);
if (full) {
tiff_set (th, &th->ntag, 254, 4, 1, 0);
tiff_set (th, &th->ntag, 256, 4, 1, width);
tiff_set (th, &th->ntag, 257, 4, 1, height);
tiff_set (th, &th->ntag, 258, 3, colors, output_bps);
if (colors > 2)
th->tag[th->ntag-1].val.i = TOFF(th->bps);
FORC4 th->bps[c] = output_bps;
tiff_set (th, &th->ntag, 259, 3, 1, 1);
tiff_set (th, &th->ntag, 262, 3, 1, 1 + (colors > 1));
}
tiff_set (th, &th->ntag, 270, 2, 512, TOFF(th->t_desc));
tiff_set (th, &th->ntag, 271, 2, 64, TOFF(th->t_make));
tiff_set (th, &th->ntag, 272, 2, 64, TOFF(th->t_model));
if (full) {
if (oprof) psize = ntohl(oprof[0]);
tiff_set (th, &th->ntag, 273, 4, 1, sizeof *th + psize);
tiff_set (th, &th->ntag, 277, 3, 1, colors);
tiff_set (th, &th->ntag, 278, 4, 1, height);
tiff_set (th, &th->ntag, 279, 4, 1, height*width*colors*output_bps/8);
} else
tiff_set (th, &th->ntag, 274, 3, 1, "12435867"[flip]-'0');
tiff_set (th, &th->ntag, 282, 5, 1, TOFF(th->rat[0]));
tiff_set (th, &th->ntag, 283, 5, 1, TOFF(th->rat[2]));
tiff_set (th, &th->ntag, 284, 3, 1, 1);
tiff_set (th, &th->ntag, 296, 3, 1, 2);
tiff_set (th, &th->ntag, 305, 2, 32, TOFF(th->soft));
tiff_set (th, &th->ntag, 306, 2, 20, TOFF(th->date));
tiff_set (th, &th->ntag, 315, 2, 64, TOFF(th->t_artist));
tiff_set (th, &th->ntag, 34665, 4, 1, TOFF(th->nexif));
if (psize) tiff_set (th, &th->ntag, 34675, 7, psize, sizeof *th);
tiff_set (th, &th->nexif, 33434, 5, 1, TOFF(th->rat[4]));
tiff_set (th, &th->nexif, 33437, 5, 1, TOFF(th->rat[6]));
tiff_set (th, &th->nexif, 34855, 3, 1, iso_speed);
tiff_set (th, &th->nexif, 37386, 5, 1, TOFF(th->rat[8]));
if (gpsdata[1]) {
tiff_set (th, &th->ntag, 34853, 4, 1, TOFF(th->ngps));
tiff_set (th, &th->ngps, 0, 1, 4, 0x202);
tiff_set (th, &th->ngps, 1, 2, 2, gpsdata[29]);
tiff_set (th, &th->ngps, 2, 5, 3, TOFF(th->gps[0]));
tiff_set (th, &th->ngps, 3, 2, 2, gpsdata[30]);
tiff_set (th, &th->ngps, 4, 5, 3, TOFF(th->gps[6]));
tiff_set (th, &th->ngps, 5, 1, 1, gpsdata[31]);
tiff_set (th, &th->ngps, 6, 5, 1, TOFF(th->gps[18]));
tiff_set (th, &th->ngps, 7, 5, 3, TOFF(th->gps[12]));
tiff_set (th, &th->ngps, 18, 2, 12, TOFF(th->gps[20]));
tiff_set (th, &th->ngps, 29, 2, 12, TOFF(th->gps[23]));
memcpy (th->gps, gpsdata, sizeof th->gps);
}
}
#ifdef LIBRAW_LIBRARY_BUILD
void CLASS jpeg_thumb_writer (FILE *tfp,char *t_humb,int t_humb_length)
{
ushort exif[5];
struct tiff_hdr th;
fputc (0xff, tfp);
fputc (0xd8, tfp);
if (strcmp (t_humb+6, "Exif")) {
memcpy (exif, "\xff\xe1 Exif\0\0", 10);
exif[1] = htons (8 + sizeof th);
fwrite (exif, 1, sizeof exif, tfp);
tiff_head (&th, 0);
fwrite (&th, 1, sizeof th, tfp);
}
fwrite (t_humb+2, 1, t_humb_length-2, tfp);
}
void CLASS jpeg_thumb()
{
char *thumb;
thumb = (char *) malloc (thumb_length);
merror (thumb, "jpeg_thumb()");
fread (thumb, 1, thumb_length, ifp);
jpeg_thumb_writer(ofp,thumb,thumb_length);
free (thumb);
}
#else
void CLASS jpeg_thumb()
{
char *thumb;
ushort exif[5];
struct tiff_hdr th;
thumb = (char *) malloc (thumb_length);
merror (thumb, "jpeg_thumb()");
fread (thumb, 1, thumb_length, ifp);
fputc (0xff, ofp);
fputc (0xd8, ofp);
if (strcmp (thumb+6, "Exif")) {
memcpy (exif, "\xff\xe1 Exif\0\0", 10);
exif[1] = htons (8 + sizeof th);
fwrite (exif, 1, sizeof exif, ofp);
tiff_head (&th, 0);
fwrite (&th, 1, sizeof th, ofp);
}
fwrite (thumb+2, 1, thumb_length-2, ofp);
free (thumb);
}
#endif
void CLASS write_ppm_tiff()
{
struct tiff_hdr th;
uchar *ppm;
ushort *ppm2;
int c, row, col, soff, rstep, cstep;
int perc, val, total, t_white=0x2000;
#ifdef LIBRAW_LIBRARY_BUILD
perc = width * height * auto_bright_thr;
#else
perc = width * height * 0.01; /* 99th percentile white level */
#endif
if (fuji_width) perc /= 2;
if (!((highlight & ~2) || no_auto_bright))
for (t_white=c=0; c < colors; c++) {
for (val=0x2000, total=0; --val > 32; )
if ((total += histogram[c][val]) > perc) break;
if (t_white < val) t_white = val;
}
gamma_curve (gamm[0], gamm[1], 2, (t_white << 3)/bright);
iheight = height;
iwidth = width;
if (flip & 4) SWAP(height,width);
ppm = (uchar *) calloc (width, colors*output_bps/8);
ppm2 = (ushort *) ppm;
merror (ppm, "write_ppm_tiff()");
if (output_tiff) {
tiff_head (&th, 1);
fwrite (&th, sizeof th, 1, ofp);
if (oprof)
fwrite (oprof, ntohl(oprof[0]), 1, ofp);
} else if (colors > 3)
fprintf (ofp,
"P7\nWIDTH %d\nHEIGHT %d\nDEPTH %d\nMAXVAL %d\nTUPLTYPE %s\nENDHDR\n",
width, height, colors, (1 << output_bps)-1, cdesc);
else
fprintf (ofp, "P%d\n%d %d\n%d\n",
colors/2+5, width, height, (1 << output_bps)-1);
soff = flip_index (0, 0);
cstep = flip_index (0, 1) - soff;
rstep = flip_index (1, 0) - flip_index (0, width);
for (row=0; row < height; row++, soff += rstep) {
for (col=0; col < width; col++, soff += cstep)
if (output_bps == 8)
FORCC ppm [col*colors+c] = curve[image[soff][c]] >> 8;
else FORCC ppm2[col*colors+c] = curve[image[soff][c]];
if (output_bps == 16 && !output_tiff && htons(0x55aa) != 0x55aa)
swab ((char*)ppm2, (char*)ppm2, width*colors*2);
fwrite (ppm, colors*output_bps/8, width, ofp);
}
free (ppm);
}
//@end COMMON
int CLASS main (int argc, const char **argv)
{
int arg, status=0, quality, i, c;
int timestamp_only=0, thumbnail_only=0, identify_only=0;
int user_qual=-1, user_black=-1, user_sat=-1, user_flip=-1;
int use_fuji_rotate=1, write_to_stdout=0, read_from_stdin=0;
const char *sp, *bpfile=0, *dark_frame=0, *write_ext;
char opm, opt, *ofname, *cp;
struct utimbuf ut;
#ifndef NO_LCMS
const char *cam_profile=0, *out_profile=0;
#endif
#ifndef LOCALTIME
putenv ((char *) "TZ=UTC");
#endif
#ifdef LOCALEDIR
setlocale (LC_CTYPE, "");
setlocale (LC_MESSAGES, "");
bindtextdomain ("dcraw", LOCALEDIR);
textdomain ("dcraw");
#endif
if (argc == 1) {
printf(_("\nRaw photo decoder \"dcraw\" v%s"), DCRAW_VERSION);
printf(_("\nby Dave Coffin, dcoffin a cybercom o net\n"));
printf(_("\nUsage: %s [OPTION]... [FILE]...\n\n"), argv[0]);
puts(_("-v Print verbose messages"));
puts(_("-c Write image data to standard output"));
puts(_("-e Extract embedded thumbnail image"));
puts(_("-i Identify files without decoding them"));
puts(_("-i -v Identify files and show metadata"));
puts(_("-z Change file dates to camera timestamp"));
puts(_("-w Use camera white balance, if possible"));
puts(_("-a Average the whole image for white balance"));
puts(_("-A <x y w h> Average a grey box for white balance"));
puts(_("-r <r g b g> Set custom white balance"));
puts(_("+M/-M Use/don't use an embedded color matrix"));
puts(_("-C <r b> Correct chromatic aberration"));
puts(_("-P <file> Fix the dead pixels listed in this file"));
puts(_("-K <file> Subtract dark frame (16-bit raw PGM)"));
puts(_("-k <num> Set the darkness level"));
puts(_("-S <num> Set the saturation level"));
puts(_("-n <num> Set threshold for wavelet denoising"));
puts(_("-H [0-9] Highlight mode (0=clip, 1=unclip, 2=blend, 3+=rebuild)"));
puts(_("-t [0-7] Flip image (0=none, 3=180, 5=90CCW, 6=90CW)"));
puts(_("-o [0-5] Output colorspace (raw,sRGB,Adobe,Wide,ProPhoto,XYZ)"));
#ifndef NO_LCMS
puts(_("-o <file> Apply output ICC profile from file"));
puts(_("-p <file> Apply camera ICC profile from file or \"embed\""));
#endif
puts(_("-d Document mode (no color, no interpolation)"));
puts(_("-D Document mode without scaling (totally raw)"));
puts(_("-j Don't stretch or rotate raw pixels"));
puts(_("-W Don't automatically brighten the image"));
puts(_("-b <num> Adjust brightness (default = 1.0)"));
puts(_("-g <p ts> Set custom gamma curve (default = 2.222 4.5)"));
puts(_("-q [0-3] Set the interpolation quality"));
puts(_("-h Half-size color image (twice as fast as \"-q 0\")"));
puts(_("-f Interpolate RGGB as four colors"));
puts(_("-m <num> Apply a 3x3 median filter to R-G and B-G"));
puts(_("-s [0..N-1] Select one raw image or \"all\" from each file"));
puts(_("-6 Write 16-bit instead of 8-bit"));
puts(_("-4 Linear 16-bit, same as \"-6 -W -g 1 1\""));
puts(_("-T Write TIFF instead of PPM"));
puts("");
return 1;
}
argv[argc] = "";
for (arg=1; (((opm = argv[arg][0]) - 2) | 2) == '+'; ) {
opt = argv[arg++][1];
if ((cp = (char *) strchr (sp="nbrkStqmHACg", opt)))
for (i=0; i < "114111111422"[cp-sp]-'0'; i++)
if (!isdigit(argv[arg+i][0])) {
fprintf (stderr,_("Non-numeric argument to \"-%c\"\n"), opt);
return 1;
}
switch (opt) {
case 'n': threshold = atof(argv[arg++]); break;
case 'b': bright = atof(argv[arg++]); break;
case 'r':
FORC4 user_mul[c] = atof(argv[arg++]); break;
case 'C': aber[0] = 1 / atof(argv[arg++]);
aber[2] = 1 / atof(argv[arg++]); break;
case 'g': gamm[0] = atof(argv[arg++]);
gamm[1] = atof(argv[arg++]);
if (gamm[0]) gamm[0] = 1/gamm[0]; break;
case 'k': user_black = atoi(argv[arg++]); break;
case 'S': user_sat = atoi(argv[arg++]); break;
case 't': user_flip = atoi(argv[arg++]); break;
case 'q': user_qual = atoi(argv[arg++]); break;
case 'm': med_passes = atoi(argv[arg++]); break;
case 'H': highlight = atoi(argv[arg++]); break;
case 's':
shot_select = abs(atoi(argv[arg]));
multi_out = !strcmp(argv[arg++],"all");
break;
case 'o':
if (isdigit(argv[arg][0]) && !argv[arg][1])
output_color = atoi(argv[arg++]);
#ifndef NO_LCMS
else out_profile = argv[arg++];
break;
case 'p': cam_profile = argv[arg++];
#endif
break;
case 'P': bpfile = argv[arg++]; break;
case 'K': dark_frame = argv[arg++]; break;
case 'z': timestamp_only = 1; break;
case 'e': thumbnail_only = 1; break;
case 'i': identify_only = 1; break;
case 'c': write_to_stdout = 1; break;
case 'v': verbose = 1; break;
case 'h': half_size = 1; break;
case 'f': four_color_rgb = 1; break;
case 'A': FORC4 greybox[c] = atoi(argv[arg++]);
case 'a': use_auto_wb = 1; break;
case 'w': use_camera_wb = 1; break;
case 'M': use_camera_matrix = 3 * (opm == '+'); break;
case 'I': read_from_stdin = 1; break;
case 'E': document_mode++;
case 'D': document_mode++;
case 'd': document_mode++;
case 'j': use_fuji_rotate = 0; break;
case 'W': no_auto_bright = 1; break;
case 'T': output_tiff = 1; break;
case '4': gamm[0] = gamm[1] =
no_auto_bright = 1;
case '6': output_bps = 16; break;
default:
fprintf (stderr,_("Unknown option \"-%c\".\n"), opt);
return 1;
}
}
if (arg == argc) {
fprintf (stderr,_("No files to process.\n"));
return 1;
}
if (write_to_stdout) {
if (isatty(1)) {
fprintf (stderr,_("Will not write an image to the terminal!\n"));
return 1;
}
#if defined(WIN32) || defined(DJGPP) || defined(__CYGWIN__)
if (setmode(1,O_BINARY) < 0) {
perror ("setmode()");
return 1;
}
#endif
}
for ( ; arg < argc; arg++) {
status = 1;
raw_image = 0;
image = 0;
oprof = 0;
meta_data = ofname = 0;
ofp = stdout;
if (setjmp (failure)) {
if (fileno(ifp) > 2) fclose(ifp);
if (fileno(ofp) > 2) fclose(ofp);
status = 1;
goto cleanup;
}
ifname = argv[arg];
if (!(ifp = fopen (ifname, "rb"))) {
perror (ifname);
continue;
}
status = (identify(),!is_raw);
if (user_flip >= 0)
flip = user_flip;
switch ((flip+3600) % 360) {
case 270: flip = 5; break;
case 180: flip = 3; break;
case 90: flip = 6;
}
if (timestamp_only) {
if ((status = !timestamp))
fprintf (stderr,_("%s has no timestamp.\n"), ifname);
else if (identify_only)
printf ("%10ld%10d %s\n", (long) timestamp, shot_order, ifname);
else {
if (verbose)
fprintf (stderr,_("%s time set to %d.\n"), ifname, (int) timestamp);
ut.actime = ut.modtime = timestamp;
utime (ifname, &ut);
}
goto next;
}
write_fun = &CLASS write_ppm_tiff;
if (thumbnail_only) {
if ((status = !thumb_offset)) {
fprintf (stderr,_("%s has no thumbnail.\n"), ifname);
goto next;
} else if (thumb_load_raw) {
load_raw = thumb_load_raw;
data_offset = thumb_offset;
height = thumb_height;
width = thumb_width;
filters = 0;
colors = 3;
} else {
fseek (ifp, thumb_offset, SEEK_SET);
write_fun = write_thumb;
goto thumbnail;
}
}
if (load_raw == &CLASS kodak_ycbcr_load_raw) {
height += height & 1;
width += width & 1;
}
if (identify_only && verbose && make[0]) {
printf (_("\nFilename: %s\n"), ifname);
printf (_("Timestamp: %s"), ctime(×tamp));
printf (_("Camera: %s %s\n"), make, model);
if (artist[0])
printf (_("Owner: %s\n"), artist);
if (dng_version) {
printf (_("DNG Version: "));
for (i=24; i >= 0; i -= 8)
printf ("%d%c", dng_version >> i & 255, i ? '.':'\n');
}
printf (_("ISO speed: %d\n"), (int) iso_speed);
printf (_("Shutter: "));
if (shutter > 0 && shutter < 1)
shutter = (printf ("1/"), 1 / shutter);
printf (_("%0.1f sec\n"), shutter);
printf (_("Aperture: f/%0.1f\n"), aperture);
printf (_("Focal length: %0.1f mm\n"), focal_len);
printf (_("Embedded ICC profile: %s\n"), profile_length ? _("yes"):_("no"));
printf (_("Number of raw images: %d\n"), is_raw);
if (pixel_aspect != 1)
printf (_("Pixel Aspect Ratio: %0.6f\n"), pixel_aspect);
if (thumb_offset)
printf (_("Thumb size: %4d x %d\n"), thumb_width, thumb_height);
printf (_("Full size: %4d x %d\n"), raw_width, raw_height);
} else if (!is_raw)
fprintf (stderr,_("Cannot decode file %s\n"), ifname);
if (!is_raw) goto next;
shrink = filters && (half_size || (!identify_only &&
(threshold || aber[0] != 1 || aber[2] != 1)));
iheight = (height + shrink) >> shrink;
iwidth = (width + shrink) >> shrink;
if (identify_only) {
if (verbose) {
if (document_mode == 3) {
top_margin = left_margin = fuji_width = 0;
height = raw_height;
width = raw_width;
}
iheight = (height + shrink) >> shrink;
iwidth = (width + shrink) >> shrink;
if (use_fuji_rotate) {
if (fuji_width) {
fuji_width = (fuji_width - 1 + shrink) >> shrink;
iwidth = fuji_width / sqrt(0.5);
iheight = (iheight - fuji_width) / sqrt(0.5);
} else {
if (pixel_aspect < 1) iheight = iheight / pixel_aspect + 0.5;
if (pixel_aspect > 1) iwidth = iwidth * pixel_aspect + 0.5;
}
}
if (flip & 4)
SWAP(iheight,iwidth);
printf (_("Image size: %4d x %d\n"), width, height);
printf (_("Output size: %4d x %d\n"), iwidth, iheight);
printf (_("Raw colors: %d"), colors);
if (filters) {
int fhigh = 2, fwide = 2;
if ((filters ^ (filters >> 8)) & 0xff) fhigh = 4;
if ((filters ^ (filters >> 16)) & 0xffff) fhigh = 8;
if (filters == 1) fhigh = fwide = 16;
if (filters == 9) fhigh = fwide = 6;
printf (_("\nFilter pattern: "));
for (i=0; i < fhigh; i++)
for (c = i && putchar('/') && 0; c < fwide; c++)
putchar (cdesc[fcol(i,c)]);
}
printf (_("\nDaylight multipliers:"));
FORCC printf (" %f", pre_mul[c]);
if (cam_mul[0] > 0) {
printf (_("\nCamera multipliers:"));
FORC4 printf (" %f", cam_mul[c]);
}
putchar ('\n');
} else
printf (_("%s is a %s %s image.\n"), ifname, make, model);
next:
fclose(ifp);
continue;
}
if (meta_length) {
meta_data = (char *) malloc (meta_length);
merror (meta_data, "main()");
}
if (filters || colors == 1) {
raw_image = (ushort *) calloc ((raw_height+7), raw_width*2);
merror (raw_image, "main()");
} else {
image = (ushort (*)[4]) calloc (iheight, iwidth*sizeof *image);
merror (image, "main()");
}
if (verbose)
fprintf (stderr,_("Loading %s %s image from %s ...\n"),
make, model, ifname);
if (shot_select >= is_raw)
fprintf (stderr,_("%s: \"-s %d\" requests a nonexistent image!\n"),
ifname, shot_select);
fseeko (ifp, data_offset, SEEK_SET);
if (raw_image && read_from_stdin)
fread (raw_image, 2, raw_height*raw_width, stdin);
else (*load_raw)();
if (document_mode == 3) {
top_margin = left_margin = fuji_width = 0;
height = raw_height;
width = raw_width;
}
iheight = (height + shrink) >> shrink;
iwidth = (width + shrink) >> shrink;
if (raw_image) {
image = (ushort (*)[4]) calloc (iheight, iwidth*sizeof *image);
merror (image, "main()");
crop_masked_pixels();
free (raw_image);
}
if (zero_is_bad) remove_zeroes();
bad_pixels (bpfile);
if (dark_frame) subtract (dark_frame);
quality = 2 + !fuji_width;
if (user_qual >= 0) quality = user_qual;
i = cblack[3];
FORC3 if (i > cblack[c]) i = cblack[c];
FORC4 cblack[c] -= i;
black += i;
i = cblack[6];
FORC (cblack[4] * cblack[5])
if (i > cblack[6+c]) i = cblack[6+c];
FORC (cblack[4] * cblack[5])
cblack[6+c] -= i;
black += i;
if (user_black >= 0) black = user_black;
FORC4 cblack[c] += black;
if (user_sat > 0) maximum = user_sat;
#ifdef COLORCHECK
colorcheck();
#endif
if (is_foveon) {
if (document_mode || load_raw == &CLASS foveon_dp_load_raw) {
for (i=0; i < height*width*4; i++)
if ((short) image[0][i] < 0) image[0][i] = 0;
} else foveon_interpolate();
} else if (document_mode < 2)
scale_colors();
pre_interpolate();
if (filters && !document_mode) {
if (quality == 0)
lin_interpolate();
else if (quality == 1 || colors > 3)
vng_interpolate();
else if (quality == 2 && filters > 1000)
ppg_interpolate();
else if (filters == 9)
xtrans_interpolate (quality*2-3);
else
ahd_interpolate();
}
if (mix_green)
for (colors=3, i=0; i < height*width; i++)
image[i][1] = (image[i][1] + image[i][3]) >> 1;
if (!is_foveon && colors == 3) median_filter();
if (!is_foveon && highlight == 2) blend_highlights();
if (!is_foveon && highlight > 2) recover_highlights();
if (use_fuji_rotate) fuji_rotate();
#ifndef NO_LCMS
if (cam_profile) apply_profile (cam_profile, out_profile);
#endif
convert_to_rgb();
if (use_fuji_rotate) stretch();
thumbnail:
if (write_fun == &CLASS jpeg_thumb)
write_ext = ".jpg";
else if (output_tiff && write_fun == &CLASS write_ppm_tiff)
write_ext = ".tiff";
else
write_ext = ".pgm\0.ppm\0.ppm\0.pam" + colors*5-5;
ofname = (char *) malloc (strlen(ifname) + 64);
merror (ofname, "main()");
if (write_to_stdout)
strcpy (ofname,_("standard output"));
else {
strcpy (ofname, ifname);
if ((cp = strrchr (ofname, '.'))) *cp = 0;
if (multi_out)
sprintf (ofname+strlen(ofname), "_%0*d",
snprintf(0,0,"%d",is_raw-1), shot_select);
if (thumbnail_only)
strcat (ofname, ".thumb");
strcat (ofname, write_ext);
ofp = fopen (ofname, "wb");
if (!ofp) {
status = 1;
perror (ofname);
goto cleanup;
}
}
if (verbose)
fprintf (stderr,_("Writing data to %s ...\n"), ofname);
(*write_fun)();
fclose(ifp);
if (ofp != stdout) fclose(ofp);
cleanup:
if (meta_data) free (meta_data);
if (ofname) free (ofname);
if (oprof) free (oprof);
if (image) free (image);
if (multi_out) {
if (++shot_select < is_raw) arg--;
else shot_select = 0;
}
}
return status;
}
#endif
|
omp-single-2.c | #include <omp.h>
extern void abort (void);
struct X
{
int a;
char b;
int c;
};
int
main()
{
int i = 0;
struct X x;
int bad = 0;
#pragma omp parallel private (i, x) shared (bad)
{
i = 5;
#pragma omp single copyprivate (i, x)
{
i++;
x.a = 23;
x.b = 42;
x.c = 26;
}
if (i != 6 || x.a != 23 || x.b != 42 || x.c != 26)
bad = 1;
}
if (bad)
abort ();
return 0;
}
|
convolution_3x3_pack1to4_bf16s.h | // Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2020 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
static void conv3x3s1_pack1to4_bf16s_neon(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt)
{
int inch = bottom_blob.c;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
#if __ARM_NEON && __aarch64__
Mat top_blob_fp32(outw, outh, opt.num_threads, (size_t)4u * 4 * 2, 4 * 2, opt.workspace_allocator);
#else
Mat top_blob_fp32(outw, outh, opt.num_threads, (size_t)4u * 4, 4, opt.workspace_allocator);
#endif
const float* bias = _bias;
int nn_outch = 0;
int remain_outch_start = 0;
#if __ARM_NEON && __aarch64__
nn_outch = outch >> 1;
remain_outch_start = nn_outch << 1;
#pragma omp parallel for num_threads(opt.num_threads)
for (int pp = 0; pp < nn_outch; pp++)
{
int p = pp * 2;
Mat out0 = top_blob_fp32.channel(get_omp_thread_num());
float32x4_t _bias0 = bias ? vld1q_f32((const float*)bias + p * 4) : vdupq_n_f32(0.f);
float32x4_t _bias1 = bias ? vld1q_f32((const float*)bias + (p + 1) * 4) : vdupq_n_f32(0.f);
{
float* ptr = (float*)out0;
for (int i = 0; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias0);
vst1q_f32(ptr + 8, _bias0);
vst1q_f32(ptr + 12, _bias0);
vst1q_f32(ptr + 16, _bias1);
vst1q_f32(ptr + 20, _bias1);
vst1q_f32(ptr + 24, _bias1);
vst1q_f32(ptr + 28, _bias1);
ptr += 32;
}
for (; j + 1 < outw; j += 2)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias0);
vst1q_f32(ptr + 8, _bias1);
vst1q_f32(ptr + 12, _bias1);
ptr += 16;
}
for (; j < outw; j++)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias1);
ptr += 8;
}
}
}
const unsigned short* k0 = kernel.channel(p);
const unsigned short* k1 = kernel.channel(p + 1);
int q = 0;
for (; q < inch - 1; q++)
{
float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
float32x4_t _k00_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1), 16));
float32x4_t _k01_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 4), 16));
float32x4_t _k02_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 8), 16));
float32x4_t _k10_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 12), 16));
float32x4_t _k11_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 16), 16));
float32x4_t _k12_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 20), 16));
float32x4_t _k20_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 24), 16));
float32x4_t _k21_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 28), 16));
float32x4_t _k22_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
asm volatile(
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1], #8 \n"
"ld1 {v1.s}[0], [%1] \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%0] \n"
"fmla v24.4s, %8.4s, v0.s[0] \n"
"fmla v25.4s, %8.4s, v0.s[1] \n"
"fmla v26.4s, %8.4s, v0.s[2] \n"
"fmla v27.4s, %8.4s, v0.s[3] \n"
"fmla v28.4s, %17.4s, v0.s[0] \n"
"fmla v29.4s, %17.4s, v0.s[1] \n"
"fmla v30.4s, %17.4s, v0.s[2] \n"
"fmla v31.4s, %17.4s, v0.s[3] \n"
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v2.4h}, [%2], #8 \n"
"ld1 {v3.s}[0], [%2] \n"
"fmla v24.4s, %9.4s, v0.s[1] \n"
"fmla v25.4s, %9.4s, v0.s[2] \n"
"fmla v26.4s, %9.4s, v0.s[3] \n"
"fmla v27.4s, %9.4s, v1.s[0] \n"
"fmla v28.4s, %18.4s, v0.s[1] \n"
"fmla v29.4s, %18.4s, v0.s[2] \n"
"fmla v30.4s, %18.4s, v0.s[3] \n"
"fmla v31.4s, %18.4s, v1.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v24.4s, %10.4s, v0.s[2] \n"
"fmla v25.4s, %10.4s, v0.s[3] \n"
"fmla v26.4s, %10.4s, v1.s[0] \n"
"fmla v27.4s, %10.4s, v1.s[1] \n"
"fmla v28.4s, %19.4s, v0.s[2] \n"
"fmla v29.4s, %19.4s, v0.s[3] \n"
"fmla v30.4s, %19.4s, v1.s[0] \n"
"fmla v31.4s, %19.4s, v1.s[1] \n"
"fmla v24.4s, %11.4s, v2.s[0] \n"
"fmla v25.4s, %11.4s, v2.s[1] \n"
"fmla v26.4s, %11.4s, v2.s[2] \n"
"fmla v27.4s, %11.4s, v2.s[3] \n"
"fmla v28.4s, %20.4s, v2.s[0] \n"
"fmla v29.4s, %20.4s, v2.s[1] \n"
"fmla v30.4s, %20.4s, v2.s[2] \n"
"fmla v31.4s, %20.4s, v2.s[3] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"ld1 {v1.s}[0], [%3] \n"
"fmla v24.4s, %12.4s, v2.s[1] \n"
"fmla v25.4s, %12.4s, v2.s[2] \n"
"fmla v26.4s, %12.4s, v2.s[3] \n"
"fmla v27.4s, %12.4s, v3.s[0] \n"
"fmla v28.4s, %21.4s, v2.s[1] \n"
"fmla v29.4s, %21.4s, v2.s[2] \n"
"fmla v30.4s, %21.4s, v2.s[3] \n"
"fmla v31.4s, %21.4s, v3.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %13.4s, v2.s[2] \n"
"fmla v25.4s, %13.4s, v2.s[3] \n"
"fmla v26.4s, %13.4s, v3.s[0] \n"
"fmla v27.4s, %13.4s, v3.s[1] \n"
"fmla v28.4s, %22.4s, v2.s[2] \n"
"fmla v29.4s, %22.4s, v2.s[3] \n"
"fmla v30.4s, %22.4s, v3.s[0] \n"
"fmla v31.4s, %22.4s, v3.s[1] \n"
"fmla v24.4s, %14.4s, v0.s[0] \n"
"fmla v25.4s, %14.4s, v0.s[1] \n"
"fmla v26.4s, %14.4s, v0.s[2] \n"
"fmla v27.4s, %14.4s, v0.s[3] \n"
"fmla v28.4s, %23.4s, v0.s[0] \n"
"fmla v29.4s, %23.4s, v0.s[1] \n"
"fmla v30.4s, %23.4s, v0.s[2] \n"
"fmla v31.4s, %23.4s, v0.s[3] \n"
"fmla v24.4s, %15.4s, v0.s[1] \n"
"fmla v25.4s, %15.4s, v0.s[2] \n"
"fmla v26.4s, %15.4s, v0.s[3] \n"
"fmla v27.4s, %15.4s, v1.s[0] \n"
"fmla v28.4s, %24.4s, v0.s[1] \n"
"fmla v29.4s, %24.4s, v0.s[2] \n"
"fmla v30.4s, %24.4s, v0.s[3] \n"
"fmla v31.4s, %24.4s, v1.s[0] \n"
"sub %0, %0, #64 \n"
"fmla v24.4s, %16.4s, v0.s[2] \n"
"fmla v25.4s, %16.4s, v0.s[3] \n"
"fmla v26.4s, %16.4s, v1.s[0] \n"
"fmla v27.4s, %16.4s, v1.s[1] \n"
"fmla v28.4s, %25.4s, v0.s[2] \n"
"fmla v29.4s, %25.4s, v0.s[3] \n"
"fmla v30.4s, %25.4s, v1.s[0] \n"
"fmla v31.4s, %25.4s, v1.s[1] \n"
"st1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
"st1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00_0), // %8
"w"(_k01_0), // %9
"w"(_k02_0), // %10
"w"(_k10_0), // %11
"w"(_k11_0), // %12
"w"(_k12_0), // %13
"w"(_k20_0), // %14
"w"(_k21_0), // %15
"w"(_k22_0), // %16
"w"(_k00_1), // %17
"w"(_k01_1), // %18
"w"(_k02_1), // %19
"w"(_k10_1), // %20
"w"(_k11_1), // %21
"w"(_k12_1), // %22
"w"(_k20_1), // %23
"w"(_k21_1), // %24
"w"(_k22_1) // %25
: "memory", "v0", "v1", "v2", "v3", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
for (; j + 1 < outw; j += 2)
{
asm volatile(
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1] \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v24.4s, %8.4s, v0.s[0] \n"
"fmla v25.4s, %8.4s, v0.s[1] \n"
"fmla v26.4s, %17.4s, v0.s[0] \n"
"fmla v27.4s, %17.4s, v0.s[1] \n"
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v1.4h}, [%2] \n"
"fmla v24.4s, %9.4s, v0.s[1] \n"
"fmla v25.4s, %9.4s, v0.s[2] \n"
"fmla v26.4s, %18.4s, v0.s[1] \n"
"fmla v27.4s, %18.4s, v0.s[2] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %10.4s, v0.s[2] \n"
"fmla v25.4s, %10.4s, v0.s[3] \n"
"fmla v26.4s, %19.4s, v0.s[2] \n"
"fmla v27.4s, %19.4s, v0.s[3] \n"
"fmla v24.4s, %11.4s, v1.s[0] \n"
"fmla v25.4s, %11.4s, v1.s[1] \n"
"fmla v26.4s, %20.4s, v1.s[0] \n"
"fmla v27.4s, %20.4s, v1.s[1] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3] \n"
"fmla v24.4s, %12.4s, v1.s[1] \n"
"fmla v25.4s, %12.4s, v1.s[2] \n"
"fmla v26.4s, %21.4s, v1.s[1] \n"
"fmla v27.4s, %21.4s, v1.s[2] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v24.4s, %13.4s, v1.s[2] \n"
"fmla v25.4s, %13.4s, v1.s[3] \n"
"fmla v26.4s, %22.4s, v1.s[2] \n"
"fmla v27.4s, %22.4s, v1.s[3] \n"
"fmla v24.4s, %14.4s, v0.s[0] \n"
"fmla v25.4s, %14.4s, v0.s[1] \n"
"fmla v26.4s, %23.4s, v0.s[0] \n"
"fmla v27.4s, %23.4s, v0.s[1] \n"
"add %1, %1, #8 \n"
"fmla v24.4s, %15.4s, v0.s[1] \n"
"fmla v25.4s, %15.4s, v0.s[2] \n"
"fmla v26.4s, %24.4s, v0.s[1] \n"
"fmla v27.4s, %24.4s, v0.s[2] \n"
"add %2, %2, #8 \n"
"fmla v24.4s, %16.4s, v0.s[2] \n"
"fmla v25.4s, %16.4s, v0.s[3] \n"
"fmla v26.4s, %25.4s, v0.s[2] \n"
"fmla v27.4s, %25.4s, v0.s[3] \n"
"add %3, %3, #8 \n"
"st1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00_0), // %8
"w"(_k01_0), // %9
"w"(_k02_0), // %10
"w"(_k10_0), // %11
"w"(_k11_0), // %12
"w"(_k12_0), // %13
"w"(_k20_0), // %14
"w"(_k21_0), // %15
"w"(_k22_0), // %16
"w"(_k00_1), // %17
"w"(_k01_1), // %18
"w"(_k02_1), // %19
"w"(_k10_1), // %20
"w"(_k11_1), // %21
"w"(_k12_1), // %22
"w"(_k20_1), // %23
"w"(_k21_1), // %24
"w"(_k22_1) // %25
: "memory", "v0", "v1", "v24", "v25", "v26", "v27");
}
for (; j < outw; j++)
{
float32x4_t _sum00 = vld1q_f32(outptr0);
float32x4_t _sum10 = vld1q_f32(outptr0 + 4);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
_sum00 = vfmaq_laneq_f32(_sum00, _k00_0, _r0, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k01_0, _r0, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k02_0, _r0, 2);
_sum00 = vfmaq_laneq_f32(_sum00, _k10_0, _r1, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k11_0, _r1, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k12_0, _r1, 2);
_sum00 = vfmaq_laneq_f32(_sum00, _k20_0, _r2, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k21_0, _r2, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k22_0, _r2, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k00_1, _r0, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k01_1, _r0, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k02_1, _r0, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k10_1, _r1, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k11_1, _r1, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k12_1, _r1, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k20_1, _r2, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k21_1, _r2, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k22_1, _r2, 2);
vst1q_f32(outptr0, _sum00);
vst1q_f32(outptr0 + 4, _sum10);
r0 += 1;
r1 += 1;
r2 += 1;
outptr0 += 8;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
k0 += 9 * 4;
k1 += 9 * 4;
}
for (; q < inch; q++)
{
unsigned short* outptr0_bf16 = top_blob.channel(p);
unsigned short* outptr1_bf16 = top_blob.channel(p + 1);
const float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
float32x4_t _k00_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1), 16));
float32x4_t _k01_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 4), 16));
float32x4_t _k02_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 8), 16));
float32x4_t _k10_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 12), 16));
float32x4_t _k11_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 16), 16));
float32x4_t _k12_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 20), 16));
float32x4_t _k20_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 24), 16));
float32x4_t _k21_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 28), 16));
float32x4_t _k22_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
asm volatile(
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"ld1 {v1.s}[0], [%3] \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%2], #64 \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%2], #64 \n"
"fmla v24.4s, %12.4s, v0.s[0] \n"
"fmla v25.4s, %12.4s, v0.s[1] \n"
"fmla v26.4s, %12.4s, v0.s[2] \n"
"fmla v27.4s, %12.4s, v0.s[3] \n"
"fmla v28.4s, %21.4s, v0.s[0] \n"
"fmla v29.4s, %21.4s, v0.s[1] \n"
"fmla v30.4s, %21.4s, v0.s[2] \n"
"fmla v31.4s, %21.4s, v0.s[3] \n"
"fmla v24.4s, %13.4s, v0.s[1] \n"
"fmla v25.4s, %13.4s, v0.s[2] \n"
"fmla v26.4s, %13.4s, v0.s[3] \n"
"fmla v27.4s, %13.4s, v1.s[0] \n"
"fmla v28.4s, %22.4s, v0.s[1] \n"
"fmla v29.4s, %22.4s, v0.s[2] \n"
"fmla v30.4s, %22.4s, v0.s[3] \n"
"fmla v31.4s, %22.4s, v1.s[0] \n"
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v2.4h}, [%4], #8 \n"
"ld1 {v3.s}[0], [%4] \n"
"fmla v24.4s, %14.4s, v0.s[2] \n"
"fmla v25.4s, %14.4s, v0.s[3] \n"
"fmla v26.4s, %14.4s, v1.s[0] \n"
"fmla v27.4s, %14.4s, v1.s[1] \n"
"fmla v28.4s, %23.4s, v0.s[2] \n"
"fmla v29.4s, %23.4s, v0.s[3] \n"
"fmla v30.4s, %23.4s, v1.s[0] \n"
"fmla v31.4s, %23.4s, v1.s[1] \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v24.4s, %15.4s, v2.s[0] \n"
"fmla v25.4s, %15.4s, v2.s[1] \n"
"fmla v26.4s, %15.4s, v2.s[2] \n"
"fmla v27.4s, %15.4s, v2.s[3] \n"
"fmla v28.4s, %24.4s, v2.s[0] \n"
"fmla v29.4s, %24.4s, v2.s[1] \n"
"fmla v30.4s, %24.4s, v2.s[2] \n"
"fmla v31.4s, %24.4s, v2.s[3] \n"
"fmla v24.4s, %16.4s, v2.s[1] \n"
"fmla v25.4s, %16.4s, v2.s[2] \n"
"fmla v26.4s, %16.4s, v2.s[3] \n"
"fmla v27.4s, %16.4s, v3.s[0] \n"
"fmla v28.4s, %25.4s, v2.s[1] \n"
"fmla v29.4s, %25.4s, v2.s[2] \n"
"fmla v30.4s, %25.4s, v2.s[3] \n"
"fmla v31.4s, %25.4s, v3.s[0] \n"
"prfm pldl1keep, [%5, #64] \n"
"ld1 {v0.4h}, [%5], #8 \n"
"ld1 {v1.s}[0], [%5] \n"
"fmla v24.4s, %17.4s, v2.s[2] \n"
"fmla v25.4s, %17.4s, v2.s[3] \n"
"fmla v26.4s, %17.4s, v3.s[0] \n"
"fmla v27.4s, %17.4s, v3.s[1] \n"
"fmla v28.4s, %26.4s, v2.s[2] \n"
"fmla v29.4s, %26.4s, v2.s[3] \n"
"fmla v30.4s, %26.4s, v3.s[0] \n"
"fmla v31.4s, %26.4s, v3.s[1] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %18.4s, v0.s[0] \n"
"fmla v25.4s, %18.4s, v0.s[1] \n"
"fmla v26.4s, %18.4s, v0.s[2] \n"
"fmla v27.4s, %18.4s, v0.s[3] \n"
"fmla v28.4s, %27.4s, v0.s[0] \n"
"fmla v29.4s, %27.4s, v0.s[1] \n"
"fmla v30.4s, %27.4s, v0.s[2] \n"
"fmla v31.4s, %27.4s, v0.s[3] \n"
"fmla v24.4s, %19.4s, v0.s[1] \n"
"fmla v25.4s, %19.4s, v0.s[2] \n"
"fmla v26.4s, %19.4s, v0.s[3] \n"
"fmla v27.4s, %19.4s, v1.s[0] \n"
"fmla v28.4s, %28.4s, v0.s[1] \n"
"fmla v29.4s, %28.4s, v0.s[2] \n"
"fmla v30.4s, %28.4s, v0.s[3] \n"
"fmla v31.4s, %28.4s, v1.s[0] \n"
"fmla v24.4s, %20.4s, v0.s[2] \n"
"fmla v25.4s, %20.4s, v0.s[3] \n"
"fmla v26.4s, %20.4s, v1.s[0] \n"
"fmla v27.4s, %20.4s, v1.s[1] \n"
"fmla v28.4s, %29.4s, v0.s[2] \n"
"fmla v29.4s, %29.4s, v0.s[3] \n"
"fmla v30.4s, %29.4s, v1.s[0] \n"
"fmla v31.4s, %29.4s, v1.s[1] \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"shrn v28.4h, v28.4s, #16 \n"
"shrn v29.4h, v29.4s, #16 \n"
"shrn v30.4h, v30.4s, #16 \n"
"shrn v31.4h, v31.4s, #16 \n"
"st1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%0], #32 \n"
"st1 {v28.4h, v29.4h, v30.4h, v31.4h}, [%1], #32 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr1_bf16), // %1
"=r"(outptr0), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2) // %5
: "0"(outptr0_bf16),
"1"(outptr1_bf16),
"2"(outptr0),
"3"(r0),
"4"(r1),
"5"(r2),
"w"(_k00_0), // %12
"w"(_k01_0), // %13
"w"(_k02_0), // %14
"w"(_k10_0), // %15
"w"(_k11_0), // %16
"w"(_k12_0), // %17
"w"(_k20_0), // %18
"w"(_k21_0), // %19
"w"(_k22_0), // %20
"w"(_k00_1), // %21
"w"(_k01_1), // %22
"w"(_k02_1), // %23
"w"(_k10_1), // %24
"w"(_k11_1), // %25
"w"(_k12_1), // %26
"w"(_k20_1), // %27
"w"(_k21_1), // %28
"w"(_k22_1) // %29
: "memory", "v0", "v1", "v2", "v3", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
for (; j + 1 < outw; j += 2)
{
asm volatile(
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3] \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%2], #64 \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v24.4s, %12.4s, v0.s[0] \n"
"fmla v25.4s, %12.4s, v0.s[1] \n"
"fmla v26.4s, %21.4s, v0.s[0] \n"
"fmla v27.4s, %21.4s, v0.s[1] \n"
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v1.4h}, [%4] \n"
"fmla v24.4s, %13.4s, v0.s[1] \n"
"fmla v25.4s, %13.4s, v0.s[2] \n"
"fmla v26.4s, %22.4s, v0.s[1] \n"
"fmla v27.4s, %22.4s, v0.s[2] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %14.4s, v0.s[2] \n"
"fmla v25.4s, %14.4s, v0.s[3] \n"
"fmla v26.4s, %23.4s, v0.s[2] \n"
"fmla v27.4s, %23.4s, v0.s[3] \n"
"fmla v24.4s, %15.4s, v1.s[0] \n"
"fmla v25.4s, %15.4s, v1.s[1] \n"
"fmla v26.4s, %24.4s, v1.s[0] \n"
"fmla v27.4s, %24.4s, v1.s[1] \n"
"prfm pldl1keep, [%5, #64] \n"
"ld1 {v0.4h}, [%5] \n"
"fmla v24.4s, %16.4s, v1.s[1] \n"
"fmla v25.4s, %16.4s, v1.s[2] \n"
"fmla v26.4s, %25.4s, v1.s[1] \n"
"fmla v27.4s, %25.4s, v1.s[2] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v24.4s, %17.4s, v1.s[2] \n"
"fmla v25.4s, %17.4s, v1.s[3] \n"
"fmla v26.4s, %26.4s, v1.s[2] \n"
"fmla v27.4s, %26.4s, v1.s[3] \n"
"fmla v24.4s, %18.4s, v0.s[0] \n"
"fmla v25.4s, %18.4s, v0.s[1] \n"
"fmla v26.4s, %27.4s, v0.s[0] \n"
"fmla v27.4s, %27.4s, v0.s[1] \n"
"fmla v24.4s, %19.4s, v0.s[1] \n"
"fmla v25.4s, %19.4s, v0.s[2] \n"
"fmla v26.4s, %28.4s, v0.s[1] \n"
"fmla v27.4s, %28.4s, v0.s[2] \n"
"add %3, %3, #8 \n"
"fmla v24.4s, %20.4s, v0.s[2] \n"
"fmla v25.4s, %20.4s, v0.s[3] \n"
"fmla v26.4s, %29.4s, v0.s[2] \n"
"fmla v27.4s, %29.4s, v0.s[3] \n"
"add %4, %4, #8 \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"add %5, %5, #8 \n"
"st1 {v24.4h, v25.4h}, [%0], #16 \n"
"st1 {v26.4h, v27.4h}, [%1], #16 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr1_bf16), // %1
"=r"(outptr0), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2) // %5
: "0"(outptr0_bf16),
"1"(outptr1_bf16),
"2"(outptr0),
"3"(r0),
"4"(r1),
"5"(r2),
"w"(_k00_0), // %12
"w"(_k01_0), // %13
"w"(_k02_0), // %14
"w"(_k10_0), // %15
"w"(_k11_0), // %16
"w"(_k12_0), // %17
"w"(_k20_0), // %18
"w"(_k21_0), // %19
"w"(_k22_0), // %20
"w"(_k00_1), // %21
"w"(_k01_1), // %22
"w"(_k02_1), // %23
"w"(_k10_1), // %24
"w"(_k11_1), // %25
"w"(_k12_1), // %26
"w"(_k20_1), // %27
"w"(_k21_1), // %28
"w"(_k22_1) // %29
: "memory", "v0", "v1", "v24", "v25", "v26", "v27");
}
for (; j < outw; j++)
{
float32x4_t _sum00 = vld1q_f32(outptr0);
float32x4_t _sum10 = vld1q_f32(outptr0 + 4);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
_sum00 = vfmaq_laneq_f32(_sum00, _k00_0, _r0, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k01_0, _r0, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k02_0, _r0, 2);
_sum00 = vfmaq_laneq_f32(_sum00, _k10_0, _r1, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k11_0, _r1, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k12_0, _r1, 2);
_sum00 = vfmaq_laneq_f32(_sum00, _k20_0, _r2, 0);
_sum00 = vfmaq_laneq_f32(_sum00, _k21_0, _r2, 1);
_sum00 = vfmaq_laneq_f32(_sum00, _k22_0, _r2, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k00_1, _r0, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k01_1, _r0, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k02_1, _r0, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k10_1, _r1, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k11_1, _r1, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k12_1, _r1, 2);
_sum10 = vfmaq_laneq_f32(_sum10, _k20_1, _r2, 0);
_sum10 = vfmaq_laneq_f32(_sum10, _k21_1, _r2, 1);
_sum10 = vfmaq_laneq_f32(_sum10, _k22_1, _r2, 2);
vst1_u16(outptr0_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum00), 16));
vst1_u16(outptr1_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum10), 16));
r0 += 1;
r1 += 1;
r2 += 1;
outptr0 += 8;
outptr0_bf16 += 4;
outptr1_bf16 += 4;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
k0 += 9 * 4;
k1 += 9 * 4;
}
}
#endif // __ARM_NEON && __aarch64__
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = remain_outch_start; p < outch; p++)
{
Mat out0 = top_blob_fp32.channel(get_omp_thread_num());
float32x4_t _bias0 = bias ? vld1q_f32((const float*)bias + p * 4) : vdupq_n_f32(0.f);
out0.fill(_bias0);
const unsigned short* k0 = kernel.channel(p);
int q = 0;
for (; q < inch - 1; q++)
{
float* outptr0 = out0.row(0);
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<unsigned short>(0);
const unsigned short* r1 = img0.row<unsigned short>(1);
const unsigned short* r2 = img0.row<unsigned short>(2);
float32x4_t _k00 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
#if __aarch64__
for (; j + 7 < outw; j += 8)
{
asm volatile(
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
// "prfm pldl1keep, [%0, #512] \n"
"ld1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%0] \n"
"prfm pldl1keep, [%1, #128] \n"
"ld1 {v0.4h, v1.4h}, [%1], #16 \n"
"ld1 {v2.s}[0], [%1] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %8.4s, v0.s[0] \n"
"fmla v25.4s, %8.4s, v0.s[1] \n"
"fmla v26.4s, %8.4s, v0.s[2] \n"
"fmla v27.4s, %8.4s, v0.s[3] \n"
"fmla v28.4s, %8.4s, v1.s[0] \n"
"fmla v29.4s, %8.4s, v1.s[1] \n"
"fmla v30.4s, %8.4s, v1.s[2] \n"
"fmla v31.4s, %8.4s, v1.s[3] \n"
"fmla v24.4s, %9.4s, v0.s[1] \n"
"fmla v25.4s, %9.4s, v0.s[2] \n"
"fmla v26.4s, %9.4s, v0.s[3] \n"
"fmla v27.4s, %9.4s, v1.s[0] \n"
"fmla v28.4s, %9.4s, v1.s[1] \n"
"fmla v29.4s, %9.4s, v1.s[2] \n"
"fmla v30.4s, %9.4s, v1.s[3] \n"
"fmla v31.4s, %9.4s, v2.s[0] \n"
"fmla v24.4s, %10.4s, v0.s[2] \n"
"fmla v25.4s, %10.4s, v0.s[3] \n"
"fmla v26.4s, %10.4s, v1.s[0] \n"
"fmla v27.4s, %10.4s, v1.s[1] \n"
"fmla v28.4s, %10.4s, v1.s[2] \n"
"fmla v29.4s, %10.4s, v1.s[3] \n"
"fmla v30.4s, %10.4s, v2.s[0] \n"
"fmla v31.4s, %10.4s, v2.s[1] \n"
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v4.4h, v5.4h}, [%2], #16 \n"
"ld1 {v2.s}[0], [%2] \n"
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %11.4s, v4.s[0] \n"
"fmla v25.4s, %11.4s, v4.s[1] \n"
"fmla v26.4s, %11.4s, v4.s[2] \n"
"fmla v27.4s, %11.4s, v4.s[3] \n"
"fmla v28.4s, %11.4s, v5.s[0] \n"
"fmla v29.4s, %11.4s, v5.s[1] \n"
"fmla v30.4s, %11.4s, v5.s[2] \n"
"fmla v31.4s, %11.4s, v5.s[3] \n"
"fmla v24.4s, %12.4s, v4.s[1] \n"
"fmla v25.4s, %12.4s, v4.s[2] \n"
"fmla v26.4s, %12.4s, v4.s[3] \n"
"fmla v27.4s, %12.4s, v5.s[0] \n"
"fmla v28.4s, %12.4s, v5.s[1] \n"
"fmla v29.4s, %12.4s, v5.s[2] \n"
"fmla v30.4s, %12.4s, v5.s[3] \n"
"fmla v31.4s, %12.4s, v2.s[0] \n"
"fmla v24.4s, %13.4s, v4.s[2] \n"
"fmla v25.4s, %13.4s, v4.s[3] \n"
"fmla v26.4s, %13.4s, v5.s[0] \n"
"fmla v27.4s, %13.4s, v5.s[1] \n"
"fmla v28.4s, %13.4s, v5.s[2] \n"
"fmla v29.4s, %13.4s, v5.s[3] \n"
"fmla v30.4s, %13.4s, v2.s[0] \n"
"fmla v31.4s, %13.4s, v2.s[1] \n"
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v0.4h, v1.4h}, [%3], #16 \n"
"ld1 {v2.s}[0], [%3] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %14.4s, v0.s[0] \n"
"fmla v25.4s, %14.4s, v0.s[1] \n"
"fmla v26.4s, %14.4s, v0.s[2] \n"
"fmla v27.4s, %14.4s, v0.s[3] \n"
"fmla v28.4s, %14.4s, v1.s[0] \n"
"fmla v29.4s, %14.4s, v1.s[1] \n"
"fmla v30.4s, %14.4s, v1.s[2] \n"
"fmla v31.4s, %14.4s, v1.s[3] \n"
"fmla v24.4s, %15.4s, v0.s[1] \n"
"fmla v25.4s, %15.4s, v0.s[2] \n"
"fmla v26.4s, %15.4s, v0.s[3] \n"
"fmla v27.4s, %15.4s, v1.s[0] \n"
"fmla v28.4s, %15.4s, v1.s[1] \n"
"fmla v29.4s, %15.4s, v1.s[2] \n"
"fmla v30.4s, %15.4s, v1.s[3] \n"
"fmla v31.4s, %15.4s, v2.s[0] \n"
"sub %0, %0, #64 \n"
"fmla v24.4s, %16.4s, v0.s[2] \n"
"fmla v25.4s, %16.4s, v0.s[3] \n"
"fmla v26.4s, %16.4s, v1.s[0] \n"
"fmla v27.4s, %16.4s, v1.s[1] \n"
"fmla v28.4s, %16.4s, v1.s[2] \n"
"fmla v29.4s, %16.4s, v1.s[3] \n"
"fmla v30.4s, %16.4s, v2.s[0] \n"
"fmla v31.4s, %16.4s, v2.s[1] \n"
"st1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
"st1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "memory", "v0", "v1", "v2", "v4", "v5", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
#endif // __aarch64__
for (; j + 3 < outw; j += 4)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1], #8 \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0] \n"
"shll v0.4s, v0.4h, #16 \n"
"ld1 {v1.s}[0], [%1] \n"
"fmla v24.4s, %8.4s, v0.s[0] \n"
"fmla v25.4s, %8.4s, v0.s[1] \n"
"fmla v26.4s, %8.4s, v0.s[2] \n"
"fmla v27.4s, %8.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %9.4s, v0.s[1] \n"
"fmla v25.4s, %9.4s, v0.s[2] \n"
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v2.4h}, [%2], #8 \n"
"fmla v26.4s, %9.4s, v0.s[3] \n"
"fmla v27.4s, %9.4s, v1.s[0] \n"
"ld1 {v3.s}[0], [%2] \n"
"fmla v24.4s, %10.4s, v0.s[2] \n"
"fmla v25.4s, %10.4s, v0.s[3] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v26.4s, %10.4s, v1.s[0] \n"
"fmla v27.4s, %10.4s, v1.s[1] \n"
"fmla v24.4s, %11.4s, v2.s[0] \n"
"fmla v25.4s, %11.4s, v2.s[1] \n"
"fmla v26.4s, %11.4s, v2.s[2] \n"
"fmla v27.4s, %11.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v24.4s, %12.4s, v2.s[1] \n"
"fmla v25.4s, %12.4s, v2.s[2] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"fmla v26.4s, %12.4s, v2.s[3] \n"
"fmla v27.4s, %12.4s, v3.s[0] \n"
"ld1 {v1.s}[0], [%3] \n"
"fmla v24.4s, %13.4s, v2.s[2] \n"
"fmla v25.4s, %13.4s, v2.s[3] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v26.4s, %13.4s, v3.s[0] \n"
"fmla v27.4s, %13.4s, v3.s[1] \n"
"fmla v24.4s, %14.4s, v0.s[0] \n"
"fmla v25.4s, %14.4s, v0.s[1] \n"
"fmla v26.4s, %14.4s, v0.s[2] \n"
"fmla v27.4s, %14.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %15.4s, v0.s[1] \n"
"fmla v25.4s, %15.4s, v0.s[2] \n"
"fmla v26.4s, %15.4s, v0.s[3] \n"
"fmla v27.4s, %15.4s, v1.s[0] \n"
"fmla v24.4s, %16.4s, v0.s[2] \n"
"fmla v25.4s, %16.4s, v0.s[3] \n"
"fmla v26.4s, %16.4s, v1.s[0] \n"
"fmla v27.4s, %16.4s, v1.s[1] \n"
"st1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "memory", "v0", "v1", "v2", "v3", "v24", "v25", "v26", "v27");
#else // __aarch64__
asm volatile(
"pld [%0, #512] \n"
"vldm %0, {d24-d31} \n"
"pld [%1, #64] \n"
"vld1.u16 {d1}, [%1]! \n"
"vld1.u32 {d2[0]}, [%1] \n"
"vshll.u16 q0, d1, #16 \n"
"vshll.u16 q1, d2, #16 \n"
"vmla.f32 q12, %q8, d0[0] \n"
"vmla.f32 q13, %q8, d0[1] \n"
"vmla.f32 q14, %q8, d1[0] \n"
"vmla.f32 q15, %q8, d1[1] \n"
"vmla.f32 q12, %q9, d0[1] \n"
"vmla.f32 q13, %q9, d1[0] \n"
"vmla.f32 q14, %q9, d1[1] \n"
"vmla.f32 q15, %q9, d2[0] \n"
"vmla.f32 q12, %q10, d1[0] \n"
"vmla.f32 q13, %q10, d1[1] \n"
"vmla.f32 q14, %q10, d2[0] \n"
"vmla.f32 q15, %q10, d2[1] \n"
"pld [%2, #64] \n"
"vld1.u16 {d5}, [%2]! \n"
"vld1.u32 {d3[0]}, [%2] \n"
"vshll.u16 q2, d5, #16 \n"
"vshll.u16 q1, d3, #16 \n"
"vmla.f32 q12, %q11, d4[0] \n"
"vmla.f32 q13, %q11, d4[1] \n"
"vmla.f32 q14, %q11, d5[0] \n"
"vmla.f32 q15, %q11, d5[1] \n"
"vmla.f32 q12, %q12, d4[1] \n"
"vmla.f32 q13, %q12, d5[0] \n"
"vmla.f32 q14, %q12, d5[1] \n"
"vmla.f32 q15, %q12, d2[0] \n"
"vmla.f32 q12, %q13, d5[0] \n"
"vmla.f32 q13, %q13, d5[1] \n"
"vmla.f32 q14, %q13, d2[0] \n"
"vmla.f32 q15, %q13, d2[1] \n"
"pld [%3, #64] \n"
"vld1.u16 {d1}, [%3]! \n"
"vld1.u32 {d2[0]}, [%3] \n"
"vshll.u16 q0, d1, #16 \n"
"vshll.u16 q1, d2, #16 \n"
"vmla.f32 q12, %q14, d0[0] \n"
"vmla.f32 q13, %q14, d0[1] \n"
"vmla.f32 q14, %q14, d1[0] \n"
"vmla.f32 q15, %q14, d1[1] \n"
"vmla.f32 q12, %q15, d0[1] \n"
"vmla.f32 q13, %q15, d1[0] \n"
"vmla.f32 q14, %q15, d1[1] \n"
"vmla.f32 q15, %q15, d2[0] \n"
"vmla.f32 q12, %q16, d1[0] \n"
"vmla.f32 q13, %q16, d1[1] \n"
"vmla.f32 q14, %q16, d2[0] \n"
"vmla.f32 q15, %q16, d2[1] \n"
"vstm %0!, {d24-d31} \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "memory", "q0", "q1", "q2", "q12", "q13", "q14", "q15");
#endif // __aarch64__
}
for (; j + 1 < outw; j += 2)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1] \n"
"prfm pldl1keep, [%0, #256] \n"
"ld1 {v28.4s, v29.4s}, [%0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmul v24.4s, %8.4s, v0.s[0] \n"
"fmul v25.4s, %8.4s, v0.s[1] \n"
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v1.4h}, [%2] \n"
"fmul v26.4s, %9.4s, v0.s[1] \n"
"fmul v27.4s, %9.4s, v0.s[2] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v28.4s, %10.4s, v0.s[2] \n"
"fmla v29.4s, %10.4s, v0.s[3] \n"
"fmla v24.4s, %11.4s, v1.s[0] \n"
"fmla v25.4s, %11.4s, v1.s[1] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3] \n"
"fmla v26.4s, %12.4s, v1.s[1] \n"
"fmla v27.4s, %12.4s, v1.s[2] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v28.4s, %13.4s, v1.s[2] \n"
"fmla v29.4s, %13.4s, v1.s[3] \n"
"fmla v24.4s, %14.4s, v0.s[0] \n"
"fmla v25.4s, %14.4s, v0.s[1] \n"
"fmla v26.4s, %15.4s, v0.s[1] \n"
"fmla v27.4s, %15.4s, v0.s[2] \n"
"fmla v28.4s, %16.4s, v0.s[2] \n"
"fmla v29.4s, %16.4s, v0.s[3] \n"
"add %1, %1, #4 \n"
"fadd v24.4s, v24.4s, v26.4s \n"
"fadd v25.4s, v25.4s, v27.4s \n"
"add %2, %2, #4 \n"
"fadd v28.4s, v28.4s, v24.4s \n"
"fadd v29.4s, v29.4s, v25.4s \n"
"add %3, %3, #4 \n"
"st1 {v28.4s, v29.4s}, [%0], #32 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "memory", "v0", "v1", "v24", "v25", "v26", "v27", "v28", "v29");
#else // __aarch64__
asm volatile(
"pld [%1, #64] \n"
"vld1.u16 {d1}, [%1] \n"
"pld [%0, #256] \n"
"vld1.f32 {d24-d27}, [%0 :128] \n"
"vshll.u16 q0, d1, #16 \n"
"vmul.f32 q14, %q8, d0[0] \n"
"vmul.f32 q15, %q8, d0[1] \n"
"vmla.f32 q12, %q9, d0[1] \n"
"vmla.f32 q13, %q9, d1[0] \n"
"pld [%2, #64] \n"
"vld1.u16 {d3}, [%2] \n"
"vmla.f32 q14, %q10, d1[0] \n"
"vmla.f32 q15, %q10, d1[1] \n"
"vshll.u16 q1, d3, #16 \n"
"vmla.f32 q12, %q11, d2[0] \n"
"vmla.f32 q13, %q11, d2[1] \n"
"vmla.f32 q14, %q12, d2[1] \n"
"vmla.f32 q15, %q12, d3[0] \n"
"pld [%3, #64] \n"
"vld1.u16 {d1}, [%3] \n"
"vmla.f32 q12, %q13, d3[0] \n"
"vmla.f32 q13, %q13, d3[1] \n"
"vshll.u16 q0, d1, #16 \n"
"vmla.f32 q14, %q14, d0[0] \n"
"vmla.f32 q15, %q14, d0[1] \n"
"vmla.f32 q12, %q15, d0[1] \n"
"vmla.f32 q13, %q15, d1[0] \n"
"add %1, %1, #4 \n"
"vmla.f32 q14, %q16, d1[0] \n"
"vmla.f32 q15, %q16, d1[1] \n"
"add %2, %2, #4 \n"
"vadd.f32 q12, q12, q14 \n"
"vadd.f32 q13, q13, q15 \n"
"add %3, %3, #4 \n"
"vst1.f32 {d24-d27}, [%0 :128]! \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "memory", "q0", "q1", "q12", "q13", "q14", "q15");
#endif // __aarch64__
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
#if __aarch64__
_sum0 = vfmaq_laneq_f32(_sum0, _k00, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22, _r2, 2);
#else
_sum0 = vmlaq_lane_f32(_sum0, _k00, vget_low_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k01, vget_low_f32(_r0), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k02, vget_high_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k10, vget_low_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k11, vget_low_f32(_r1), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k12, vget_high_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k20, vget_low_f32(_r2), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k21, vget_low_f32(_r2), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k22, vget_high_f32(_r2), 0);
#endif
vst1q_f32(outptr0, _sum0);
r0 += 1;
r1 += 1;
r2 += 1;
outptr0 += 4;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
k0 += 9 * 4;
}
for (; q < inch; q++)
{
unsigned short* outptr0_bf16 = top_blob.channel(p);
const float* outptr0 = out0.row(0);
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<unsigned short>(0);
const unsigned short* r1 = img0.row<unsigned short>(1);
const unsigned short* r2 = img0.row<unsigned short>(2);
float32x4_t _k00 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
#if __aarch64__
for (; j + 7 < outw; j += 8)
{
asm volatile(
"prfm pldl1keep, [%1, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%1], #64 \n"
"prfm pldl1keep, [%1, #512] \n"
"ld1 {v28.4s, v29.4s, v30.4s, v31.4s}, [%1], #64 \n"
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v0.4h, v1.4h}, [%2], #16 \n"
"ld1 {v2.s}[0], [%2] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %10.4s, v0.s[0] \n"
"fmla v25.4s, %10.4s, v0.s[1] \n"
"fmla v26.4s, %10.4s, v0.s[2] \n"
"fmla v27.4s, %10.4s, v0.s[3] \n"
"fmla v28.4s, %10.4s, v1.s[0] \n"
"fmla v29.4s, %10.4s, v1.s[1] \n"
"fmla v30.4s, %10.4s, v1.s[2] \n"
"fmla v31.4s, %10.4s, v1.s[3] \n"
"fmla v24.4s, %11.4s, v0.s[1] \n"
"fmla v25.4s, %11.4s, v0.s[2] \n"
"fmla v26.4s, %11.4s, v0.s[3] \n"
"fmla v27.4s, %11.4s, v1.s[0] \n"
"fmla v28.4s, %11.4s, v1.s[1] \n"
"fmla v29.4s, %11.4s, v1.s[2] \n"
"fmla v30.4s, %11.4s, v1.s[3] \n"
"fmla v31.4s, %11.4s, v2.s[0] \n"
"fmla v24.4s, %12.4s, v0.s[2] \n"
"fmla v25.4s, %12.4s, v0.s[3] \n"
"fmla v26.4s, %12.4s, v1.s[0] \n"
"fmla v27.4s, %12.4s, v1.s[1] \n"
"fmla v28.4s, %12.4s, v1.s[2] \n"
"fmla v29.4s, %12.4s, v1.s[3] \n"
"fmla v30.4s, %12.4s, v2.s[0] \n"
"fmla v31.4s, %12.4s, v2.s[1] \n"
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v4.4h, v5.4h}, [%3], #16 \n"
"ld1 {v2.s}[0], [%3] \n"
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %13.4s, v4.s[0] \n"
"fmla v25.4s, %13.4s, v4.s[1] \n"
"fmla v26.4s, %13.4s, v4.s[2] \n"
"fmla v27.4s, %13.4s, v4.s[3] \n"
"fmla v28.4s, %13.4s, v5.s[0] \n"
"fmla v29.4s, %13.4s, v5.s[1] \n"
"fmla v30.4s, %13.4s, v5.s[2] \n"
"fmla v31.4s, %13.4s, v5.s[3] \n"
"fmla v24.4s, %14.4s, v4.s[1] \n"
"fmla v25.4s, %14.4s, v4.s[2] \n"
"fmla v26.4s, %14.4s, v4.s[3] \n"
"fmla v27.4s, %14.4s, v5.s[0] \n"
"fmla v28.4s, %14.4s, v5.s[1] \n"
"fmla v29.4s, %14.4s, v5.s[2] \n"
"fmla v30.4s, %14.4s, v5.s[3] \n"
"fmla v31.4s, %14.4s, v2.s[0] \n"
"fmla v24.4s, %15.4s, v4.s[2] \n"
"fmla v25.4s, %15.4s, v4.s[3] \n"
"fmla v26.4s, %15.4s, v5.s[0] \n"
"fmla v27.4s, %15.4s, v5.s[1] \n"
"fmla v28.4s, %15.4s, v5.s[2] \n"
"fmla v29.4s, %15.4s, v5.s[3] \n"
"fmla v30.4s, %15.4s, v2.s[0] \n"
"fmla v31.4s, %15.4s, v2.s[1] \n"
"prfm pldl1keep, [%4, #128] \n"
"ld1 {v0.4h, v1.4h}, [%4], #16 \n"
"ld1 {v2.s}[0], [%4] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v24.4s, %16.4s, v0.s[0] \n"
"fmla v25.4s, %16.4s, v0.s[1] \n"
"fmla v26.4s, %16.4s, v0.s[2] \n"
"fmla v27.4s, %16.4s, v0.s[3] \n"
"fmla v28.4s, %16.4s, v1.s[0] \n"
"fmla v29.4s, %16.4s, v1.s[1] \n"
"fmla v30.4s, %16.4s, v1.s[2] \n"
"fmla v31.4s, %16.4s, v1.s[3] \n"
"fmla v24.4s, %17.4s, v0.s[1] \n"
"fmla v25.4s, %17.4s, v0.s[2] \n"
"fmla v26.4s, %17.4s, v0.s[3] \n"
"fmla v27.4s, %17.4s, v1.s[0] \n"
"fmla v28.4s, %17.4s, v1.s[1] \n"
"fmla v29.4s, %17.4s, v1.s[2] \n"
"fmla v30.4s, %17.4s, v1.s[3] \n"
"fmla v31.4s, %17.4s, v2.s[0] \n"
"fmla v24.4s, %18.4s, v0.s[2] \n"
"fmla v25.4s, %18.4s, v0.s[3] \n"
"fmla v26.4s, %18.4s, v1.s[0] \n"
"fmla v27.4s, %18.4s, v1.s[1] \n"
"fmla v28.4s, %18.4s, v1.s[2] \n"
"fmla v29.4s, %18.4s, v1.s[3] \n"
"fmla v30.4s, %18.4s, v2.s[0] \n"
"fmla v31.4s, %18.4s, v2.s[1] \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"shrn v28.4h, v28.4s, #16 \n"
"shrn v29.4h, v29.4s, #16 \n"
"shrn v30.4h, v30.4s, #16 \n"
"shrn v31.4h, v31.4s, #16 \n"
"st1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%0], #32 \n"
"st1 {v28.4h, v29.4h, v30.4h, v31.4h}, [%0], #32 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "memory", "v0", "v1", "v2", "v4", "v5", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
#endif // __aarch64__
for (; j + 3 < outw; j += 4)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v0.4h}, [%2], #8 \n"
"prfm pldl1keep, [%1, #512] \n"
"ld1 {v24.4s, v25.4s, v26.4s, v27.4s}, [%1], #64 \n"
"shll v0.4s, v0.4h, #16 \n"
"ld1 {v1.s}[0], [%2] \n"
"fmla v24.4s, %10.4s, v0.s[0] \n"
"fmla v25.4s, %10.4s, v0.s[1] \n"
"fmla v26.4s, %10.4s, v0.s[2] \n"
"fmla v27.4s, %10.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %11.4s, v0.s[1] \n"
"fmla v25.4s, %11.4s, v0.s[2] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v2.4h}, [%3], #8 \n"
"fmla v26.4s, %11.4s, v0.s[3] \n"
"fmla v27.4s, %11.4s, v1.s[0] \n"
"ld1 {v3.s}[0], [%3] \n"
"fmla v24.4s, %12.4s, v0.s[2] \n"
"fmla v25.4s, %12.4s, v0.s[3] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v26.4s, %12.4s, v1.s[0] \n"
"fmla v27.4s, %12.4s, v1.s[1] \n"
"fmla v24.4s, %13.4s, v2.s[0] \n"
"fmla v25.4s, %13.4s, v2.s[1] \n"
"fmla v26.4s, %13.4s, v2.s[2] \n"
"fmla v27.4s, %13.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v24.4s, %14.4s, v2.s[1] \n"
"fmla v25.4s, %14.4s, v2.s[2] \n"
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v0.4h}, [%4], #8 \n"
"fmla v26.4s, %14.4s, v2.s[3] \n"
"fmla v27.4s, %14.4s, v3.s[0] \n"
"ld1 {v1.s}[0], [%4] \n"
"fmla v24.4s, %15.4s, v2.s[2] \n"
"fmla v25.4s, %15.4s, v2.s[3] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v26.4s, %15.4s, v3.s[0] \n"
"fmla v27.4s, %15.4s, v3.s[1] \n"
"fmla v24.4s, %16.4s, v0.s[0] \n"
"fmla v25.4s, %16.4s, v0.s[1] \n"
"fmla v26.4s, %16.4s, v0.s[2] \n"
"fmla v27.4s, %16.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v24.4s, %17.4s, v0.s[1] \n"
"fmla v25.4s, %17.4s, v0.s[2] \n"
"fmla v26.4s, %17.4s, v0.s[3] \n"
"fmla v27.4s, %17.4s, v1.s[0] \n"
"fmla v24.4s, %18.4s, v0.s[2] \n"
"fmla v25.4s, %18.4s, v0.s[3] \n"
"fmla v26.4s, %18.4s, v1.s[0] \n"
"fmla v27.4s, %18.4s, v1.s[1] \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"st1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%0], #32 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "memory", "v0", "v1", "v2", "v3", "v24", "v25", "v26", "v27");
#else // __aarch64__
asm volatile(
"pld [%1, #512] \n"
"vldm %1!, {d24-d31} \n"
"pld [%2, #64] \n"
"vld1.u16 {d1}, [%2]! \n"
"vld1.u32 {d2[0]}, [%2] \n"
"vshll.u16 q0, d1, #16 \n"
"vshll.u16 q1, d2, #16 \n"
"vmla.f32 q12, %q10, d0[0] \n"
"vmla.f32 q13, %q10, d0[1] \n"
"vmla.f32 q14, %q10, d1[0] \n"
"vmla.f32 q15, %q10, d1[1] \n"
"vmla.f32 q12, %q11, d0[1] \n"
"vmla.f32 q13, %q11, d1[0] \n"
"vmla.f32 q14, %q11, d1[1] \n"
"vmla.f32 q15, %q11, d2[0] \n"
"vmla.f32 q12, %q12, d1[0] \n"
"vmla.f32 q13, %q12, d1[1] \n"
"vmla.f32 q14, %q12, d2[0] \n"
"vmla.f32 q15, %q12, d2[1] \n"
"pld [%3, #64] \n"
"vld1.u16 {d5}, [%3]! \n"
"vld1.u32 {d3[0]}, [%3] \n"
"vshll.u16 q2, d5, #16 \n"
"vshll.u16 q1, d3, #16 \n"
"vmla.f32 q12, %q13, d4[0] \n"
"vmla.f32 q13, %q13, d4[1] \n"
"vmla.f32 q14, %q13, d5[0] \n"
"vmla.f32 q15, %q13, d5[1] \n"
"vmla.f32 q12, %q14, d4[1] \n"
"vmla.f32 q13, %q14, d5[0] \n"
"vmla.f32 q14, %q14, d5[1] \n"
"vmla.f32 q15, %q14, d2[0] \n"
"vmla.f32 q12, %q15, d5[0] \n"
"vmla.f32 q13, %q15, d5[1] \n"
"vmla.f32 q14, %q15, d2[0] \n"
"vmla.f32 q15, %q15, d2[1] \n"
"pld [%4, #64] \n"
"vld1.u16 {d1}, [%4]! \n"
"vld1.u32 {d2[0]}, [%4] \n"
"vshll.u16 q0, d1, #16 \n"
"vshll.u16 q1, d2, #16 \n"
"vmla.f32 q12, %q16, d0[0] \n"
"vmla.f32 q13, %q16, d0[1] \n"
"vmla.f32 q14, %q16, d1[0] \n"
"vmla.f32 q15, %q16, d1[1] \n"
"vmla.f32 q12, %q17, d0[1] \n"
"vmla.f32 q13, %q17, d1[0] \n"
"vmla.f32 q14, %q17, d1[1] \n"
"vmla.f32 q15, %q17, d2[0] \n"
"vmla.f32 q12, %q18, d1[0] \n"
"vmla.f32 q13, %q18, d1[1] \n"
"vmla.f32 q14, %q18, d2[0] \n"
"vmla.f32 q15, %q18, d2[1] \n"
"vshrn.s32 d24, q12, #16 \n"
"vshrn.s32 d25, q13, #16 \n"
"vshrn.s32 d26, q14, #16 \n"
"vshrn.s32 d27, q15, #16 \n"
"vst1.u16 {d24-d27}, [%0 :64]! \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "memory", "q0", "q1", "q2", "q12", "q13", "q14", "q15");
#endif // __aarch64__
}
for (; j + 1 < outw; j += 2)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v0.4h}, [%2] \n"
"prfm pldl1keep, [%1, #256] \n"
"ld1 {v28.4s, v29.4s}, [%1], #32 \n"
"shll v0.4s, v0.4h, #16 \n"
"fmul v24.4s, %10.4s, v0.s[0] \n"
"fmul v25.4s, %10.4s, v0.s[1] \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v1.4h}, [%3] \n"
"fmul v26.4s, %11.4s, v0.s[1] \n"
"fmul v27.4s, %11.4s, v0.s[2] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v28.4s, %12.4s, v0.s[2] \n"
"fmla v29.4s, %12.4s, v0.s[3] \n"
"fmla v24.4s, %13.4s, v1.s[0] \n"
"fmla v25.4s, %13.4s, v1.s[1] \n"
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v0.4h}, [%4] \n"
"fmla v26.4s, %14.4s, v1.s[1] \n"
"fmla v27.4s, %14.4s, v1.s[2] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v28.4s, %15.4s, v1.s[2] \n"
"fmla v29.4s, %15.4s, v1.s[3] \n"
"fmla v24.4s, %16.4s, v0.s[0] \n"
"fmla v25.4s, %16.4s, v0.s[1] \n"
"fmla v26.4s, %17.4s, v0.s[1] \n"
"fmla v27.4s, %17.4s, v0.s[2] \n"
"fmla v28.4s, %18.4s, v0.s[2] \n"
"fmla v29.4s, %18.4s, v0.s[3] \n"
"add %1, %1, #4 \n"
"fadd v24.4s, v24.4s, v26.4s \n"
"fadd v25.4s, v25.4s, v27.4s \n"
"add %2, %2, #4 \n"
"fadd v28.4s, v28.4s, v24.4s \n"
"fadd v29.4s, v29.4s, v25.4s \n"
"add %3, %3, #4 \n"
"shrn v28.4h, v28.4s, #16 \n"
"shrn v29.4h, v29.4s, #16 \n"
"st1 {v28.4h, v29.4h}, [%0], #16 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "memory", "v0", "v1", "v24", "v25", "v26", "v27", "v28", "v29");
#else // __aarch64__
asm volatile(
"pld [%2, #64] \n"
"vld1.u16 {d1}, [%2] \n"
"pld [%1, #256] \n"
"vld1.f32 {d24-d27}, [%1 :128]! \n"
"vshll.u16 q0, d1, #16 \n"
"vmul.f32 q14, %q10, d0[0] \n"
"vmul.f32 q15, %q10, d0[1] \n"
"vmla.f32 q12, %q11, d0[1] \n"
"vmla.f32 q13, %q11, d1[0] \n"
"pld [%3, #64] \n"
"vld1.u16 {d3}, [%3] \n"
"vmla.f32 q14, %q12, d1[0] \n"
"vmla.f32 q15, %q12, d1[1] \n"
"vshll.u16 q1, d3, #16 \n"
"vmla.f32 q12, %q13, d2[0] \n"
"vmla.f32 q13, %q13, d2[1] \n"
"vmla.f32 q14, %q14, d2[1] \n"
"vmla.f32 q15, %q14, d3[0] \n"
"pld [%4, #64] \n"
"vld1.u16 {d1}, [%4] \n"
"vmla.f32 q12, %q15, d3[0] \n"
"vmla.f32 q13, %q15, d3[1] \n"
"vshll.u16 q0, d1, #16 \n"
"vmla.f32 q14, %q16, d0[0] \n"
"vmla.f32 q15, %q16, d0[1] \n"
"vmla.f32 q12, %q17, d0[1] \n"
"vmla.f32 q13, %q17, d1[0] \n"
"add %2, %2, #4 \n"
"vmla.f32 q14, %q18, d1[0] \n"
"vmla.f32 q15, %q18, d1[1] \n"
"add %3, %3, #4 \n"
"vadd.f32 q12, q12, q14 \n"
"vadd.f32 q13, q13, q15 \n"
"add %4, %4, #4 \n"
"vshrn.s32 d24, q12, #16 \n"
"vshrn.s32 d25, q13, #16 \n"
"vst1.f32 {d24-d25}, [%0 :64]! \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "memory", "q0", "q1", "q12", "q13", "q14", "q15");
#endif // __aarch64__
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
#if __aarch64__
_sum0 = vfmaq_laneq_f32(_sum0, _k00, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22, _r2, 2);
#else
_sum0 = vmlaq_lane_f32(_sum0, _k00, vget_low_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k01, vget_low_f32(_r0), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k02, vget_high_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k10, vget_low_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k11, vget_low_f32(_r1), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k12, vget_high_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k20, vget_low_f32(_r2), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k21, vget_low_f32(_r2), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k22, vget_high_f32(_r2), 0);
#endif
vst1_u16(outptr0_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum0), 16));
r0 += 1;
r1 += 1;
r2 += 1;
outptr0 += 4;
outptr0_bf16 += 4;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
k0 += 9 * 4;
}
}
}
static void conv3x3s2_pack1to4_bf16s_neon(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt)
{
int w = bottom_blob.w;
int inch = bottom_blob.c;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
#if __ARM_NEON && __aarch64__
Mat top_blob_fp32(outw, outh, opt.num_threads, (size_t)4u * 4 * 2, 4 * 2, opt.workspace_allocator);
#else
Mat top_blob_fp32(outw, outh, opt.num_threads, (size_t)4u * 4, 4, opt.workspace_allocator);
#endif
const int tailstep = w - 2 * outw + w;
const float* bias = _bias;
int nn_outch = 0;
int remain_outch_start = 0;
#if __ARM_NEON && __aarch64__
nn_outch = outch >> 1;
remain_outch_start = nn_outch << 1;
#pragma omp parallel for num_threads(opt.num_threads)
for (int pp = 0; pp < nn_outch; pp++)
{
int p = pp * 2;
Mat out0 = top_blob_fp32.channel(get_omp_thread_num());
float32x4_t _bias0 = bias ? vld1q_f32((const float*)bias + p * 4) : vdupq_n_f32(0.f);
float32x4_t _bias1 = bias ? vld1q_f32((const float*)bias + (p + 1) * 4) : vdupq_n_f32(0.f);
{
float* ptr = (float*)out0;
for (int i = 0; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias0);
vst1q_f32(ptr + 8, _bias0);
vst1q_f32(ptr + 12, _bias0);
vst1q_f32(ptr + 16, _bias1);
vst1q_f32(ptr + 20, _bias1);
vst1q_f32(ptr + 24, _bias1);
vst1q_f32(ptr + 28, _bias1);
ptr += 32;
}
for (; j + 1 < outw; j += 2)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias0);
vst1q_f32(ptr + 8, _bias1);
vst1q_f32(ptr + 12, _bias1);
ptr += 16;
}
for (; j < outw; j++)
{
vst1q_f32(ptr, _bias0);
vst1q_f32(ptr + 4, _bias1);
ptr += 8;
}
}
}
const unsigned short* k0 = kernel.channel(p);
const unsigned short* k1 = kernel.channel(p + 1);
int q = 0;
for (; q < inch - 1; q++)
{
float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
float32x4_t _k00_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1), 16));
float32x4_t _k01_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 4), 16));
float32x4_t _k02_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 8), 16));
float32x4_t _k10_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 12), 16));
float32x4_t _k11_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 16), 16));
float32x4_t _k12_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 20), 16));
float32x4_t _k20_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 24), 16));
float32x4_t _k21_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 28), 16));
float32x4_t _k22_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
asm volatile(
// r0
"prfm pldl1keep, [%1, #128] \n"
"ld1 {v0.4h, v1.4h}, [%1], #16 \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%0], #64 \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
// "prfm pldl1keep, [%0, #512] \n"
"ld1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%0] \n" // sum1
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %8.4s, v0.s[0] \n"
"fmla v7.4s, %8.4s, v0.s[2] \n"
"fmla v8.4s, %8.4s, v1.s[0] \n"
"fmla v9.4s, %8.4s, v1.s[2] \n"
"fmla v10.4s, %17.4s, v0.s[0] \n"
"fmla v11.4s, %17.4s, v0.s[2] \n"
"fmla v12.4s, %17.4s, v1.s[0] \n"
"fmla v13.4s, %17.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%1] \n"
"fmla v6.4s, %9.4s, v0.s[1] \n"
"fmla v7.4s, %9.4s, v0.s[3] \n"
"fmla v8.4s, %9.4s, v1.s[1] \n"
"fmla v9.4s, %9.4s, v1.s[3] \n"
"fmla v10.4s, %18.4s, v0.s[1] \n"
"fmla v11.4s, %18.4s, v0.s[3] \n"
"fmla v12.4s, %18.4s, v1.s[1] \n"
"fmla v13.4s, %18.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
// r1
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v2.4h, v3.4h}, [%2], #16 \n"
"fmla v6.4s, %10.4s, v0.s[2] \n"
"fmla v7.4s, %10.4s, v1.s[0] \n"
"fmla v8.4s, %10.4s, v1.s[2] \n"
"fmla v9.4s, %10.4s, v4.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v10.4s, %19.4s, v0.s[2] \n"
"fmla v11.4s, %19.4s, v1.s[0] \n"
"fmla v12.4s, %19.4s, v1.s[2] \n"
"fmla v13.4s, %19.4s, v4.s[0] \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v6.4s, %11.4s, v2.s[0] \n"
"fmla v7.4s, %11.4s, v2.s[2] \n"
"fmla v8.4s, %11.4s, v3.s[0] \n"
"fmla v9.4s, %11.4s, v3.s[2] \n"
"fmla v10.4s, %20.4s, v2.s[0] \n"
"fmla v11.4s, %20.4s, v2.s[2] \n"
"fmla v12.4s, %20.4s, v3.s[0] \n"
"fmla v13.4s, %20.4s, v3.s[2] \n"
"ld1 {v5.h}[0], [%2] \n"
"fmla v6.4s, %12.4s, v2.s[1] \n"
"fmla v7.4s, %12.4s, v2.s[3] \n"
"fmla v8.4s, %12.4s, v3.s[1] \n"
"fmla v9.4s, %12.4s, v3.s[3] \n"
"shll v5.4s, v5.4h, #16 \n"
"fmla v10.4s, %21.4s, v2.s[1] \n"
"fmla v11.4s, %21.4s, v2.s[3] \n"
"fmla v12.4s, %21.4s, v3.s[1] \n"
"fmla v13.4s, %21.4s, v3.s[3] \n"
// r2
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v0.4h, v1.4h}, [%3], #16 \n"
"fmla v6.4s, %13.4s, v2.s[2] \n"
"fmla v7.4s, %13.4s, v3.s[0] \n"
"fmla v8.4s, %13.4s, v3.s[2] \n"
"fmla v9.4s, %13.4s, v5.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %22.4s, v2.s[2] \n"
"fmla v11.4s, %22.4s, v3.s[0] \n"
"fmla v12.4s, %22.4s, v3.s[2] \n"
"fmla v13.4s, %22.4s, v5.s[0] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %14.4s, v0.s[0] \n"
"fmla v7.4s, %14.4s, v0.s[2] \n"
"fmla v8.4s, %14.4s, v1.s[0] \n"
"fmla v9.4s, %14.4s, v1.s[2] \n"
"fmla v10.4s, %23.4s, v0.s[0] \n"
"fmla v11.4s, %23.4s, v0.s[2] \n"
"fmla v12.4s, %23.4s, v1.s[0] \n"
"fmla v13.4s, %23.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%3] \n"
"fmla v6.4s, %15.4s, v0.s[1] \n"
"fmla v7.4s, %15.4s, v0.s[3] \n"
"fmla v8.4s, %15.4s, v1.s[1] \n"
"fmla v9.4s, %15.4s, v1.s[3] \n"
"fmla v10.4s, %24.4s, v0.s[1] \n"
"fmla v11.4s, %24.4s, v0.s[3] \n"
"fmla v12.4s, %24.4s, v1.s[1] \n"
"fmla v13.4s, %24.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
"fmla v6.4s, %16.4s, v0.s[2] \n"
"fmla v7.4s, %16.4s, v1.s[0] \n"
"fmla v8.4s, %16.4s, v1.s[2] \n"
"fmla v9.4s, %16.4s, v4.s[0] \n"
"sub %0, %0, #64 \n"
"fmla v10.4s, %25.4s, v0.s[2] \n"
"fmla v11.4s, %25.4s, v1.s[0] \n"
"fmla v12.4s, %25.4s, v1.s[2] \n"
"fmla v13.4s, %25.4s, v4.s[0] \n"
"st1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%0], #64 \n"
"st1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00_0), // %8
"w"(_k01_0), // %9
"w"(_k02_0), // %10
"w"(_k10_0), // %11
"w"(_k11_0), // %12
"w"(_k12_0), // %13
"w"(_k20_0), // %14
"w"(_k21_0), // %15
"w"(_k22_0), // %16
"w"(_k00_1), // %17
"w"(_k01_1), // %18
"w"(_k02_1), // %19
"w"(_k10_1), // %20
"w"(_k11_1), // %21
"w"(_k12_1), // %22
"w"(_k20_1), // %23
"w"(_k21_1), // %24
"w"(_k22_1) // %25
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13");
}
for (; j + 1 < outw; j += 2)
{
asm volatile(
// r0
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1], #8 \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%0] \n" // sum0 sum1
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %8.4s, v0.s[0] \n"
"fmla v11.4s, %8.4s, v0.s[2] \n"
"fmla v12.4s, %17.4s, v0.s[0] \n"
"fmla v13.4s, %17.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%1] \n"
"fmla v10.4s, %9.4s, v0.s[1] \n"
"fmla v11.4s, %9.4s, v0.s[3] \n"
"fmla v12.4s, %18.4s, v0.s[1] \n"
"fmla v13.4s, %18.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
// r1
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v2.4h}, [%2], #8 \n"
"fmla v10.4s, %10.4s, v0.s[2] \n"
"fmla v11.4s, %10.4s, v1.s[0] \n"
"fmla v12.4s, %19.4s, v0.s[2] \n"
"fmla v13.4s, %19.4s, v1.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v10.4s, %11.4s, v2.s[0] \n"
"fmla v11.4s, %11.4s, v2.s[2] \n"
"fmla v12.4s, %20.4s, v2.s[0] \n"
"fmla v13.4s, %20.4s, v2.s[2] \n"
"ld1 {v3.h}[0], [%2] \n"
"fmla v10.4s, %12.4s, v2.s[1] \n"
"fmla v11.4s, %12.4s, v2.s[3] \n"
"fmla v12.4s, %21.4s, v2.s[1] \n"
"fmla v13.4s, %21.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
// r2
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"fmla v10.4s, %13.4s, v2.s[2] \n"
"fmla v11.4s, %13.4s, v3.s[0] \n"
"fmla v12.4s, %22.4s, v2.s[2] \n"
"fmla v13.4s, %22.4s, v3.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %14.4s, v0.s[0] \n"
"fmla v11.4s, %14.4s, v0.s[2] \n"
"fmla v12.4s, %23.4s, v0.s[0] \n"
"fmla v13.4s, %23.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%3] \n"
"fmla v10.4s, %15.4s, v0.s[1] \n"
"fmla v11.4s, %15.4s, v0.s[3] \n"
"fmla v12.4s, %24.4s, v0.s[1] \n"
"fmla v13.4s, %24.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v10.4s, %16.4s, v0.s[2] \n"
"fmla v11.4s, %16.4s, v1.s[0] \n"
"fmla v12.4s, %25.4s, v0.s[2] \n"
"fmla v13.4s, %25.4s, v1.s[0] \n"
"st1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00_0), // %8
"w"(_k01_0), // %9
"w"(_k02_0), // %10
"w"(_k10_0), // %11
"w"(_k11_0), // %12
"w"(_k12_0), // %13
"w"(_k20_0), // %14
"w"(_k21_0), // %15
"w"(_k22_0), // %16
"w"(_k00_1), // %17
"w"(_k01_1), // %18
"w"(_k02_1), // %19
"w"(_k10_1), // %20
"w"(_k11_1), // %21
"w"(_k12_1), // %22
"w"(_k20_1), // %23
"w"(_k21_1), // %24
"w"(_k22_1) // %25
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13");
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _sum1 = vld1q_f32(outptr0 + 4);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
_sum0 = vfmaq_laneq_f32(_sum0, _k00_0, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01_0, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02_0, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10_0, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11_0, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12_0, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20_0, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21_0, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22_0, _r2, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k00_1, _r0, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k01_1, _r0, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k02_1, _r0, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k10_1, _r1, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k11_1, _r1, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k12_1, _r1, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k20_1, _r2, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k21_1, _r2, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k22_1, _r2, 2);
vst1q_f32(outptr0, _sum0);
vst1q_f32(outptr0 + 4, _sum1);
r0 += 2;
r1 += 2;
r2 += 2;
outptr0 += 8;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
k0 += 9 * 4;
k1 += 9 * 4;
}
for (; q < inch; q++)
{
unsigned short* outptr0_bf16 = top_blob.channel(p);
unsigned short* outptr1_bf16 = top_blob.channel(p + 1);
const float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22_0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
float32x4_t _k00_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1), 16));
float32x4_t _k01_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 4), 16));
float32x4_t _k02_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 8), 16));
float32x4_t _k10_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 12), 16));
float32x4_t _k11_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 16), 16));
float32x4_t _k12_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 20), 16));
float32x4_t _k20_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 24), 16));
float32x4_t _k21_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 28), 16));
float32x4_t _k22_1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k1 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
asm volatile(
// r0
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v0.4h, v1.4h}, [%3], #16 \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%2], #64 \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%2], #64 \n" // sum1
"fmla v6.4s, %12.4s, v0.s[0] \n"
"fmla v7.4s, %12.4s, v0.s[2] \n"
"fmla v8.4s, %12.4s, v1.s[0] \n"
"fmla v9.4s, %12.4s, v1.s[2] \n"
"fmla v10.4s, %21.4s, v0.s[0] \n"
"fmla v11.4s, %21.4s, v0.s[2] \n"
"fmla v12.4s, %21.4s, v1.s[0] \n"
"fmla v13.4s, %21.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%3] \n"
"fmla v6.4s, %13.4s, v0.s[1] \n"
"fmla v7.4s, %13.4s, v0.s[3] \n"
"fmla v8.4s, %13.4s, v1.s[1] \n"
"fmla v9.4s, %13.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
"fmla v10.4s, %22.4s, v0.s[1] \n"
"fmla v11.4s, %22.4s, v0.s[3] \n"
"fmla v12.4s, %22.4s, v1.s[1] \n"
"fmla v13.4s, %22.4s, v1.s[3] \n"
// r1
"prfm pldl1keep, [%4, #128] \n"
"ld1 {v2.4h, v3.4h}, [%4], #16 \n"
"fmla v6.4s, %14.4s, v0.s[2] \n"
"fmla v7.4s, %14.4s, v1.s[0] \n"
"fmla v8.4s, %14.4s, v1.s[2] \n"
"fmla v9.4s, %14.4s, v4.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v10.4s, %23.4s, v0.s[2] \n"
"fmla v11.4s, %23.4s, v1.s[0] \n"
"fmla v12.4s, %23.4s, v1.s[2] \n"
"fmla v13.4s, %23.4s, v4.s[0] \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v6.4s, %15.4s, v2.s[0] \n"
"fmla v7.4s, %15.4s, v2.s[2] \n"
"fmla v8.4s, %15.4s, v3.s[0] \n"
"fmla v9.4s, %15.4s, v3.s[2] \n"
"fmla v10.4s, %24.4s, v2.s[0] \n"
"fmla v11.4s, %24.4s, v2.s[2] \n"
"fmla v12.4s, %24.4s, v3.s[0] \n"
"fmla v13.4s, %24.4s, v3.s[2] \n"
"ld1 {v5.h}[0], [%4] \n"
"fmla v6.4s, %16.4s, v2.s[1] \n"
"fmla v7.4s, %16.4s, v2.s[3] \n"
"fmla v8.4s, %16.4s, v3.s[1] \n"
"fmla v9.4s, %16.4s, v3.s[3] \n"
"shll v5.4s, v5.4h, #16 \n"
"fmla v10.4s, %25.4s, v2.s[1] \n"
"fmla v11.4s, %25.4s, v2.s[3] \n"
"fmla v12.4s, %25.4s, v3.s[1] \n"
"fmla v13.4s, %25.4s, v3.s[3] \n"
// r2
"prfm pldl1keep, [%5, #128] \n"
"ld1 {v0.4h, v1.4h}, [%5], #16 \n"
"fmla v6.4s, %17.4s, v2.s[2] \n"
"fmla v7.4s, %17.4s, v3.s[0] \n"
"fmla v8.4s, %17.4s, v3.s[2] \n"
"fmla v9.4s, %17.4s, v5.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %26.4s, v2.s[2] \n"
"fmla v11.4s, %26.4s, v3.s[0] \n"
"fmla v12.4s, %26.4s, v3.s[2] \n"
"fmla v13.4s, %26.4s, v5.s[0] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %18.4s, v0.s[0] \n"
"fmla v7.4s, %18.4s, v0.s[2] \n"
"fmla v8.4s, %18.4s, v1.s[0] \n"
"fmla v9.4s, %18.4s, v1.s[2] \n"
"fmla v10.4s, %27.4s, v0.s[0] \n"
"fmla v11.4s, %27.4s, v0.s[2] \n"
"fmla v12.4s, %27.4s, v1.s[0] \n"
"fmla v13.4s, %27.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%5] \n"
"fmla v6.4s, %19.4s, v0.s[1] \n"
"fmla v7.4s, %19.4s, v0.s[3] \n"
"fmla v8.4s, %19.4s, v1.s[1] \n"
"fmla v9.4s, %19.4s, v1.s[3] \n"
"fmla v10.4s, %28.4s, v0.s[1] \n"
"fmla v11.4s, %28.4s, v0.s[3] \n"
"fmla v12.4s, %28.4s, v1.s[1] \n"
"fmla v13.4s, %28.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
"fmla v6.4s, %20.4s, v0.s[2] \n"
"fmla v7.4s, %20.4s, v1.s[0] \n"
"fmla v8.4s, %20.4s, v1.s[2] \n"
"fmla v9.4s, %20.4s, v4.s[0] \n"
"fmla v10.4s, %29.4s, v0.s[2] \n"
"fmla v11.4s, %29.4s, v1.s[0] \n"
"fmla v12.4s, %29.4s, v1.s[2] \n"
"fmla v13.4s, %29.4s, v4.s[0] \n"
"shrn v6.4h, v6.4s, #16 \n"
"shrn v7.4h, v7.4s, #16 \n"
"shrn v8.4h, v8.4s, #16 \n"
"shrn v9.4h, v9.4s, #16 \n"
"shrn v10.4h, v10.4s, #16 \n"
"shrn v11.4h, v11.4s, #16 \n"
"st1 {v6.4h, v7.4h, v8.4h, v9.4h}, [%0], #32 \n"
"shrn v12.4h, v12.4s, #16 \n"
"shrn v13.4h, v13.4s, #16 \n"
"st1 {v10.4h, v11.4h, v12.4h, v13.4h}, [%1], #32 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr1_bf16), // %1
"=r"(outptr0), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2) // %5
: "0"(outptr0_bf16),
"1"(outptr1_bf16),
"2"(outptr0),
"3"(r0),
"4"(r1),
"5"(r2),
"w"(_k00_0), // %12
"w"(_k01_0), // %13
"w"(_k02_0), // %14
"w"(_k10_0), // %15
"w"(_k11_0), // %16
"w"(_k12_0), // %17
"w"(_k20_0), // %18
"w"(_k21_0), // %19
"w"(_k22_0), // %20
"w"(_k00_1), // %21
"w"(_k01_1), // %22
"w"(_k02_1), // %23
"w"(_k10_1), // %24
"w"(_k11_1), // %25
"w"(_k12_1), // %26
"w"(_k20_1), // %27
"w"(_k21_1), // %28
"w"(_k22_1) // %29
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13");
}
for (; j + 1 < outw; j += 2)
{
asm volatile(
// r0
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"prfm pldl1keep, [%2, #512] \n"
"ld1 {v10.4s, v11.4s, v12.4s, v13.4s}, [%2], #64 \n" // sum0 sum1
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %12.4s, v0.s[0] \n"
"fmla v11.4s, %12.4s, v0.s[2] \n"
"fmla v12.4s, %21.4s, v0.s[0] \n"
"fmla v13.4s, %21.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%3] \n"
"fmla v10.4s, %13.4s, v0.s[1] \n"
"fmla v11.4s, %13.4s, v0.s[3] \n"
"fmla v12.4s, %22.4s, v0.s[1] \n"
"fmla v13.4s, %22.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
// r1
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v2.4h}, [%4], #8 \n"
"fmla v10.4s, %14.4s, v0.s[2] \n"
"fmla v11.4s, %14.4s, v1.s[0] \n"
"fmla v12.4s, %23.4s, v0.s[2] \n"
"fmla v13.4s, %23.4s, v1.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v10.4s, %15.4s, v2.s[0] \n"
"fmla v11.4s, %15.4s, v2.s[2] \n"
"fmla v12.4s, %24.4s, v2.s[0] \n"
"fmla v13.4s, %24.4s, v2.s[2] \n"
"ld1 {v3.h}[0], [%4] \n"
"fmla v10.4s, %16.4s, v2.s[1] \n"
"fmla v11.4s, %16.4s, v2.s[3] \n"
"fmla v12.4s, %25.4s, v2.s[1] \n"
"fmla v13.4s, %25.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
// r2
"prfm pldl1keep, [%5, #64] \n"
"ld1 {v0.4h}, [%5], #8 \n"
"fmla v10.4s, %17.4s, v2.s[2] \n"
"fmla v11.4s, %17.4s, v3.s[0] \n"
"fmla v12.4s, %26.4s, v2.s[2] \n"
"fmla v13.4s, %26.4s, v3.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v10.4s, %18.4s, v0.s[0] \n"
"fmla v11.4s, %18.4s, v0.s[2] \n"
"fmla v12.4s, %27.4s, v0.s[0] \n"
"fmla v13.4s, %27.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%5] \n"
"fmla v10.4s, %19.4s, v0.s[1] \n"
"fmla v11.4s, %19.4s, v0.s[3] \n"
"fmla v12.4s, %28.4s, v0.s[1] \n"
"fmla v13.4s, %28.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v10.4s, %20.4s, v0.s[2] \n"
"fmla v11.4s, %20.4s, v1.s[0] \n"
"fmla v12.4s, %29.4s, v0.s[2] \n"
"fmla v13.4s, %29.4s, v1.s[0] \n"
"shrn v10.4h, v10.4s, #16 \n"
"shrn v11.4h, v11.4s, #16 \n"
"shrn v12.4h, v12.4s, #16 \n"
"shrn v13.4h, v13.4s, #16 \n"
"st1 {v10.4h, v11.4h}, [%0], #16 \n"
"st1 {v12.4h, v13.4h}, [%1], #16 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr1_bf16), // %1
"=r"(outptr0), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2) // %5
: "0"(outptr0_bf16),
"1"(outptr1_bf16),
"2"(outptr0),
"3"(r0),
"4"(r1),
"5"(r2),
"w"(_k00_0), // %12
"w"(_k01_0), // %13
"w"(_k02_0), // %14
"w"(_k10_0), // %15
"w"(_k11_0), // %16
"w"(_k12_0), // %17
"w"(_k20_0), // %18
"w"(_k21_0), // %19
"w"(_k22_0), // %20
"w"(_k00_1), // %21
"w"(_k01_1), // %22
"w"(_k02_1), // %23
"w"(_k10_1), // %24
"w"(_k11_1), // %25
"w"(_k12_1), // %26
"w"(_k20_1), // %27
"w"(_k21_1), // %28
"w"(_k22_1) // %29
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13");
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _sum1 = vld1q_f32(outptr0 + 4);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
_sum0 = vfmaq_laneq_f32(_sum0, _k00_0, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01_0, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02_0, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10_0, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11_0, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12_0, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20_0, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21_0, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22_0, _r2, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k00_1, _r0, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k01_1, _r0, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k02_1, _r0, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k10_1, _r1, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k11_1, _r1, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k12_1, _r1, 2);
_sum1 = vfmaq_laneq_f32(_sum1, _k20_1, _r2, 0);
_sum1 = vfmaq_laneq_f32(_sum1, _k21_1, _r2, 1);
_sum1 = vfmaq_laneq_f32(_sum1, _k22_1, _r2, 2);
vst1_u16(outptr0_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum0), 16));
vst1_u16(outptr1_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum1), 16));
r0 += 2;
r1 += 2;
r2 += 2;
outptr0 += 8;
outptr0_bf16 += 4;
outptr1_bf16 += 4;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
k0 += 9 * 4;
k1 += 9 * 4;
}
}
#endif // __ARM_NEON && __aarch64__
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = remain_outch_start; p < outch; p++)
{
Mat out0 = top_blob_fp32.channel(get_omp_thread_num());
float32x4_t _bias0 = bias ? vld1q_f32((const float*)bias + p * 4) : vdupq_n_f32(0.f);
out0.fill(_bias0);
const unsigned short* k0 = kernel.channel(p);
int q = 0;
for (; q < inch - 1; q++)
{
float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
#if __aarch64__
asm volatile(
// r0
"prfm pldl1keep, [%1, #128] \n"
"ld1 {v0.4h, v1.4h}, [%1], #16 \n"
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%0] \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %8.4s, v0.s[0] \n"
"fmla v7.4s, %8.4s, v0.s[2] \n"
"fmla v8.4s, %8.4s, v1.s[0] \n"
"fmla v9.4s, %8.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%1] \n"
"fmla v6.4s, %9.4s, v0.s[1] \n"
"fmla v7.4s, %9.4s, v0.s[3] \n"
"fmla v8.4s, %9.4s, v1.s[1] \n"
"fmla v9.4s, %9.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
// r1
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v2.4h, v3.4h}, [%2], #16 \n"
"fmla v6.4s, %10.4s, v0.s[2] \n"
"fmla v7.4s, %10.4s, v1.s[0] \n"
"fmla v8.4s, %10.4s, v1.s[2] \n"
"fmla v9.4s, %10.4s, v4.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v6.4s, %11.4s, v2.s[0] \n"
"fmla v7.4s, %11.4s, v2.s[2] \n"
"fmla v8.4s, %11.4s, v3.s[0] \n"
"fmla v9.4s, %11.4s, v3.s[2] \n"
"ld1 {v5.h}[0], [%2] \n"
"fmla v6.4s, %12.4s, v2.s[1] \n"
"fmla v7.4s, %12.4s, v2.s[3] \n"
"fmla v8.4s, %12.4s, v3.s[1] \n"
"fmla v9.4s, %12.4s, v3.s[3] \n"
"shll v5.4s, v5.4h, #16 \n"
// r2
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v0.4h, v1.4h}, [%3], #16 \n"
"fmla v6.4s, %13.4s, v2.s[2] \n"
"fmla v7.4s, %13.4s, v3.s[0] \n"
"fmla v8.4s, %13.4s, v3.s[2] \n"
"fmla v9.4s, %13.4s, v5.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %14.4s, v0.s[0] \n"
"fmla v7.4s, %14.4s, v0.s[2] \n"
"fmla v8.4s, %14.4s, v1.s[0] \n"
"fmla v9.4s, %14.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%3] \n"
"fmla v6.4s, %15.4s, v0.s[1] \n"
"fmla v7.4s, %15.4s, v0.s[3] \n"
"fmla v8.4s, %15.4s, v1.s[1] \n"
"fmla v9.4s, %15.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
"fmla v6.4s, %16.4s, v0.s[2] \n"
"fmla v7.4s, %16.4s, v1.s[0] \n"
"fmla v8.4s, %16.4s, v1.s[2] \n"
"fmla v9.4s, %16.4s, v4.s[0] \n"
"st1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%0], #64 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9");
#else // __aarch64__
asm volatile(
// r0
"pld [%1, #128] \n"
"vld1.u16 {d12-d13}, [%1]! \n"
"pld [%0, #512] \n"
"vldm %0, {d0-d7} \n" // sum0
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%1] \n"
"vmla.f32 q0, %q8, d8[0] \n"
"vmla.f32 q1, %q8, d9[0] \n"
"vmla.f32 q2, %q8, d10[0] \n"
"vmla.f32 q3, %q8, d11[0] \n"
"vmla.f32 q0, %q9, d8[1] \n"
"vmla.f32 q1, %q9, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q9, d10[1] \n"
"vmla.f32 q3, %q9, d11[1] \n"
// r1
"pld [%2, #128] \n"
"vld1.u16 {d12-d13}, [%2]! \n"
"vmla.f32 q0, %q10, d9[0] \n"
"vmla.f32 q1, %q10, d10[0] \n"
"vmla.f32 q2, %q10, d11[0] \n"
"vmla.f32 q3, %q10, d8[0] \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%2] \n"
"vmla.f32 q0, %q11, d8[0] \n"
"vmla.f32 q1, %q11, d9[0] \n"
"vmla.f32 q2, %q11, d10[0] \n"
"vmla.f32 q3, %q11, d11[0] \n"
"vmla.f32 q0, %q12, d8[1] \n"
"vmla.f32 q1, %q12, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q12, d10[1] \n"
"vmla.f32 q3, %q12, d11[1] \n"
// r2
"pld [%3, #128] \n"
"vld1.u16 {d12-d13}, [%3]! \n"
"vmla.f32 q0, %q13, d9[0] \n"
"vmla.f32 q1, %q13, d10[0] \n"
"vmla.f32 q2, %q13, d11[0] \n"
"vmla.f32 q3, %q13, d8[0] \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%3] \n"
"vmla.f32 q0, %q14, d8[0] \n"
"vmla.f32 q1, %q14, d9[0] \n"
"vmla.f32 q2, %q14, d10[0] \n"
"vmla.f32 q3, %q14, d11[0] \n"
"vmla.f32 q0, %q15, d8[1] \n"
"vmla.f32 q1, %q15, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q15, d10[1] \n"
"vmla.f32 q3, %q15, d11[1] \n"
"vmla.f32 q0, %q16, d9[0] \n"
"vmla.f32 q1, %q16, d10[0] \n"
"vmla.f32 q2, %q16, d11[0] \n"
"vmla.f32 q3, %q16, d8[0] \n"
"vstm %0!, {d0-d7} \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6");
#endif // __aarch64__
}
for (; j + 1 < outw; j += 2)
{
#if __aarch64__
asm volatile(
// r0
"prfm pldl1keep, [%1, #64] \n"
"ld1 {v0.4h}, [%1], #8 \n"
"prfm pldl1keep, [%0, #256] \n"
"ld1 {v8.4s, v9.4s}, [%0] \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
"fmul v6.4s, %8.4s, v0.s[0] \n"
"fmul v7.4s, %8.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%1] \n"
"fmla v8.4s, %9.4s, v0.s[1] \n"
"fmla v9.4s, %9.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
// r1
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v2.4h}, [%2], #8 \n"
"fmla v6.4s, %10.4s, v0.s[2] \n"
"fmla v7.4s, %10.4s, v1.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v8.4s, %11.4s, v2.s[0] \n"
"fmla v9.4s, %11.4s, v2.s[2] \n"
"ld1 {v3.h}[0], [%2] \n"
"fmla v6.4s, %12.4s, v2.s[1] \n"
"fmla v7.4s, %12.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
// r2
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n"
"fmla v8.4s, %13.4s, v2.s[2] \n"
"fmla v9.4s, %13.4s, v3.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v6.4s, %14.4s, v0.s[0] \n"
"fmla v7.4s, %14.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%3] \n"
"fmla v8.4s, %15.4s, v0.s[1] \n"
"fmla v9.4s, %15.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %16.4s, v0.s[2] \n"
"fmla v7.4s, %16.4s, v1.s[0] \n"
"fadd v8.4s, v8.4s, v6.4s \n"
"fadd v9.4s, v9.4s, v7.4s \n"
"st1 {v8.4s, v9.4s}, [%0], #32 \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9");
#else // __aarch64__
asm volatile(
// r0
"pld [%1, #64] \n"
"vld1.u16 {d9}, [%1]! \n"
"pld [%0, #256] \n"
"vld1.f32 {d4-d7}, [%0] \n" // sum0
"vshll.u16 q4, d9, #16 \n"
"vmul.f32 q0, %q8, d8[0] \n"
"vmul.f32 q1, %q8, d9[0] \n"
"vld1.u16 {d11[]}, [%1] \n"
"vmla.f32 q2, %q9, d8[1] \n"
"vmla.f32 q3, %q9, d9[1] \n"
"vshll.u16 q5, d11, #16 \n"
// r1
"pld [%2, #64] \n"
"vld1.u16 {d13}, [%2]! \n"
"vmla.f32 q0, %q10, d9[0] \n"
"vmla.f32 q1, %q10, d10[0] \n"
"vshll.u16 q6, d13, #16 \n"
"vmla.f32 q2, %q11, d12[0] \n"
"vmla.f32 q3, %q11, d13[0] \n"
"vld1.u16 {d9[]}, [%2] \n"
"vmla.f32 q0, %q12, d12[1] \n"
"vmla.f32 q1, %q12, d13[1] \n"
"vshll.u16 q4, d9, #16 \n"
// r2
"pld [%3, #64] \n"
"vld1.u16 {d11}, [%3]! \n"
"vmla.f32 q2, %q13, d13[0] \n"
"vmla.f32 q3, %q13, d8[0] \n"
"vshll.u16 q5, d11, #16 \n"
"vmla.f32 q0, %q14, d10[0] \n"
"vmla.f32 q1, %q14, d11[0] \n"
"vld1.u16 {d13[]}, [%3] \n"
"vmla.f32 q2, %q15, d10[1] \n"
"vmla.f32 q3, %q15, d11[1] \n"
"vshll.u16 q6, d13, #16 \n"
"vmla.f32 q0, %q16, d11[0] \n"
"vmla.f32 q1, %q16, d12[0] \n"
"vadd.f32 q2, q2, q0 \n"
"vadd.f32 q3, q3, q1 \n"
"vst1.f32 {d4-d7}, [%0]! \n"
: "=r"(outptr0), // %0
"=r"(r0), // %1
"=r"(r1), // %2
"=r"(r2) // %3
: "0"(outptr0),
"1"(r0),
"2"(r1),
"3"(r2),
"w"(_k00), // %8
"w"(_k01), // %9
"w"(_k02), // %10
"w"(_k10), // %11
"w"(_k11), // %12
"w"(_k12), // %13
"w"(_k20), // %14
"w"(_k21), // %15
"w"(_k22) // %16
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6");
#endif // __aarch64__
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
#if __aarch64__
_sum0 = vfmaq_laneq_f32(_sum0, _k00, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22, _r2, 2);
#else
_sum0 = vmlaq_lane_f32(_sum0, _k00, vget_low_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k01, vget_low_f32(_r0), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k02, vget_high_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k10, vget_low_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k11, vget_low_f32(_r1), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k12, vget_high_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k20, vget_low_f32(_r2), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k21, vget_low_f32(_r2), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k22, vget_high_f32(_r2), 0);
#endif
vst1q_f32(outptr0, _sum0);
r0 += 2;
r1 += 2;
r2 += 2;
outptr0 += 4;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
k0 += 9 * 4;
}
for (; q < inch; q++)
{
unsigned short* outptr0_bf16 = top_blob.channel(p);
const float* outptr0 = out0;
const Mat img0 = bottom_blob.channel(q);
const unsigned short* r0 = img0.row<const unsigned short>(0);
const unsigned short* r1 = img0.row<const unsigned short>(1);
const unsigned short* r2 = img0.row<const unsigned short>(2);
float32x4_t _k00 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0), 16));
float32x4_t _k01 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 4), 16));
float32x4_t _k02 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 8), 16));
float32x4_t _k10 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 12), 16));
float32x4_t _k11 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 16), 16));
float32x4_t _k12 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 20), 16));
float32x4_t _k20 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 24), 16));
float32x4_t _k21 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 28), 16));
float32x4_t _k22 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(k0 + 32), 16));
int i = 0;
for (; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
#if __aarch64__
asm volatile(
// r0
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v0.4h, v1.4h}, [%2], #16 \n"
"prfm pldl1keep, [%1, #512] \n"
"ld1 {v6.4s, v7.4s, v8.4s, v9.4s}, [%1], #64 \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %10.4s, v0.s[0] \n"
"fmla v7.4s, %10.4s, v0.s[2] \n"
"fmla v8.4s, %10.4s, v1.s[0] \n"
"fmla v9.4s, %10.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%2] \n"
"fmla v6.4s, %11.4s, v0.s[1] \n"
"fmla v7.4s, %11.4s, v0.s[3] \n"
"fmla v8.4s, %11.4s, v1.s[1] \n"
"fmla v9.4s, %11.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
// r1
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v2.4h, v3.4h}, [%3], #16 \n"
"fmla v6.4s, %12.4s, v0.s[2] \n"
"fmla v7.4s, %12.4s, v1.s[0] \n"
"fmla v8.4s, %12.4s, v1.s[2] \n"
"fmla v9.4s, %12.4s, v4.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v6.4s, %13.4s, v2.s[0] \n"
"fmla v7.4s, %13.4s, v2.s[2] \n"
"fmla v8.4s, %13.4s, v3.s[0] \n"
"fmla v9.4s, %13.4s, v3.s[2] \n"
"ld1 {v5.h}[0], [%3] \n"
"fmla v6.4s, %14.4s, v2.s[1] \n"
"fmla v7.4s, %14.4s, v2.s[3] \n"
"fmla v8.4s, %14.4s, v3.s[1] \n"
"fmla v9.4s, %14.4s, v3.s[3] \n"
"shll v5.4s, v5.4h, #16 \n"
// r2
"prfm pldl1keep, [%4, #128] \n"
"ld1 {v0.4h, v1.4h}, [%4], #16 \n"
"fmla v6.4s, %15.4s, v2.s[2] \n"
"fmla v7.4s, %15.4s, v3.s[0] \n"
"fmla v8.4s, %15.4s, v3.s[2] \n"
"fmla v9.4s, %15.4s, v5.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %16.4s, v0.s[0] \n"
"fmla v7.4s, %16.4s, v0.s[2] \n"
"fmla v8.4s, %16.4s, v1.s[0] \n"
"fmla v9.4s, %16.4s, v1.s[2] \n"
"ld1 {v4.h}[0], [%4] \n"
"fmla v6.4s, %17.4s, v0.s[1] \n"
"fmla v7.4s, %17.4s, v0.s[3] \n"
"fmla v8.4s, %17.4s, v1.s[1] \n"
"fmla v9.4s, %17.4s, v1.s[3] \n"
"shll v4.4s, v4.4h, #16 \n"
"fmla v6.4s, %18.4s, v0.s[2] \n"
"fmla v7.4s, %18.4s, v1.s[0] \n"
"fmla v8.4s, %18.4s, v1.s[2] \n"
"fmla v9.4s, %18.4s, v4.s[0] \n"
"shrn v6.4h, v6.4s, #16 \n"
"shrn v7.4h, v7.4s, #16 \n"
"shrn v8.4h, v8.4s, #16 \n"
"shrn v9.4h, v9.4s, #16 \n"
"st1 {v6.4h, v7.4h, v8.4h, v9.4h}, [%0], #32 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9");
#else // __aarch64__
asm volatile(
// r0
"pld [%2, #128] \n"
"vld1.u16 {d12-d13}, [%2]! \n"
"pld [%1, #512] \n"
"vldm %1!, {d0-d7} \n" // sum0
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%2] \n"
"vmla.f32 q0, %q10, d8[0] \n"
"vmla.f32 q1, %q10, d9[0] \n"
"vmla.f32 q2, %q10, d10[0] \n"
"vmla.f32 q3, %q10, d11[0] \n"
"vmla.f32 q0, %q11, d8[1] \n"
"vmla.f32 q1, %q11, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q11, d10[1] \n"
"vmla.f32 q3, %q11, d11[1] \n"
// r1
"pld [%3, #128] \n"
"vld1.u16 {d12-d13}, [%3]! \n"
"vmla.f32 q0, %q12, d9[0] \n"
"vmla.f32 q1, %q12, d10[0] \n"
"vmla.f32 q2, %q12, d11[0] \n"
"vmla.f32 q3, %q12, d8[0] \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%3] \n"
"vmla.f32 q0, %q13, d8[0] \n"
"vmla.f32 q1, %q13, d9[0] \n"
"vmla.f32 q2, %q13, d10[0] \n"
"vmla.f32 q3, %q13, d11[0] \n"
"vmla.f32 q0, %q14, d8[1] \n"
"vmla.f32 q1, %q14, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q14, d10[1] \n"
"vmla.f32 q3, %q14, d11[1] \n"
// r2
"pld [%4, #128] \n"
"vld1.u16 {d12-d13}, [%4]! \n"
"vmla.f32 q0, %q15, d9[0] \n"
"vmla.f32 q1, %q15, d10[0] \n"
"vmla.f32 q2, %q15, d11[0] \n"
"vmla.f32 q3, %q15, d8[0] \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vld1.u16 {d12[0]}, [%4] \n"
"vmla.f32 q0, %q16, d8[0] \n"
"vmla.f32 q1, %q16, d9[0] \n"
"vmla.f32 q2, %q16, d10[0] \n"
"vmla.f32 q3, %q16, d11[0] \n"
"vmla.f32 q0, %q17, d8[1] \n"
"vmla.f32 q1, %q17, d9[1] \n"
"vshl.u32 d8, d12, #16 \n"
"vmla.f32 q2, %q17, d10[1] \n"
"vmla.f32 q3, %q17, d11[1] \n"
"vmla.f32 q0, %q18, d9[0] \n"
"vmla.f32 q1, %q18, d10[0] \n"
"vmla.f32 q2, %q18, d11[0] \n"
"vmla.f32 q3, %q18, d8[0] \n"
"vshrn.u32 d0, q0, #16 \n"
"vshrn.u32 d1, q1, #16 \n"
"vshrn.u32 d2, q2, #16 \n"
"vshrn.u32 d3, q3, #16 \n"
"vst1.u16 {d0-d3}, [%0 :64]! \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6");
#endif // __aarch64__
}
for (; j + 1 < outw; j += 2)
{
#if __aarch64__
asm volatile(
// r0
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v0.4h}, [%2], #8 \n"
"prfm pldl1keep, [%1, #256] \n"
"ld1 {v8.4s, v9.4s}, [%1], #32 \n" // sum0
"shll v0.4s, v0.4h, #16 \n"
"fmul v6.4s, %10.4s, v0.s[0] \n"
"fmul v7.4s, %10.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%2] \n"
"fmla v8.4s, %11.4s, v0.s[1] \n"
"fmla v9.4s, %11.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
// r1
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v2.4h}, [%3], #8 \n"
"fmla v6.4s, %12.4s, v0.s[2] \n"
"fmla v7.4s, %12.4s, v1.s[0] \n"
"shll v2.4s, v2.4h, #16 \n"
"fmla v8.4s, %13.4s, v2.s[0] \n"
"fmla v9.4s, %13.4s, v2.s[2] \n"
"ld1 {v3.h}[0], [%3] \n"
"fmla v6.4s, %14.4s, v2.s[1] \n"
"fmla v7.4s, %14.4s, v2.s[3] \n"
"shll v3.4s, v3.4h, #16 \n"
// r2
"prfm pldl1keep, [%4, #64] \n"
"ld1 {v0.4h}, [%4], #8 \n"
"fmla v8.4s, %15.4s, v2.s[2] \n"
"fmla v9.4s, %15.4s, v3.s[0] \n"
"shll v0.4s, v0.4h, #16 \n"
"fmla v6.4s, %16.4s, v0.s[0] \n"
"fmla v7.4s, %16.4s, v0.s[2] \n"
"ld1 {v1.h}[0], [%4] \n"
"fmla v8.4s, %17.4s, v0.s[1] \n"
"fmla v9.4s, %17.4s, v0.s[3] \n"
"shll v1.4s, v1.4h, #16 \n"
"fmla v6.4s, %18.4s, v0.s[2] \n"
"fmla v7.4s, %18.4s, v1.s[0] \n"
"fadd v8.4s, v8.4s, v6.4s \n"
"fadd v9.4s, v9.4s, v7.4s \n"
"shrn v8.4h, v8.4s, #16 \n"
"shrn v9.4h, v9.4s, #16 \n"
"st1 {v8.4h, v9.4h}, [%0], #16 \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9");
#else // __aarch64__
asm volatile(
// r0
"pld [%2, #64] \n"
"vld1.u16 {d9}, [%2]! \n"
"pld [%1, #256] \n"
"vld1.f32 {d4-d7}, [%1]! \n" // sum0
"vshll.u16 q4, d9, #16 \n"
"vmul.f32 q0, %q10, d8[0] \n"
"vmul.f32 q1, %q10, d9[0] \n"
"vld1.u16 {d11[]}, [%2] \n"
"vmla.f32 q2, %q11, d8[1] \n"
"vmla.f32 q3, %q11, d9[1] \n"
"vshll.u16 q5, d11, #16 \n"
// r1
"pld [%3, #64] \n"
"vld1.u16 {d13}, [%3]! \n"
"vmla.f32 q0, %q12, d9[0] \n"
"vmla.f32 q1, %q12, d10[0] \n"
"vshll.u16 q6, d13, #16 \n"
"vmla.f32 q2, %q13, d12[0] \n"
"vmla.f32 q3, %q13, d13[0] \n"
"vld1.u16 {d9[]}, [%3] \n"
"vmla.f32 q0, %q14, d12[1] \n"
"vmla.f32 q1, %q14, d13[1] \n"
"vshll.u16 q4, d9, #16 \n"
// r2
"pld [%4, #64] \n"
"vld1.u16 {d11}, [%4]! \n"
"vmla.f32 q2, %q15, d13[0] \n"
"vmla.f32 q3, %q15, d8[0] \n"
"vshll.u16 q5, d11, #16 \n"
"vmla.f32 q0, %q16, d10[0] \n"
"vmla.f32 q1, %q16, d11[0] \n"
"vld1.u16 {d13[]}, [%4] \n"
"vmla.f32 q2, %q17, d10[1] \n"
"vmla.f32 q3, %q17, d11[1] \n"
"vshll.u16 q6, d13, #16 \n"
"vmla.f32 q0, %q18, d11[0] \n"
"vmla.f32 q1, %q18, d12[0] \n"
"vadd.f32 q2, q2, q0 \n"
"vadd.f32 q3, q3, q1 \n"
"vshrn.u32 d2, q2, #16 \n"
"vshrn.u32 d3, q3, #16 \n"
"vst1.u16 {d2-d3}, [%0 :64]! \n"
: "=r"(outptr0_bf16), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(outptr0_bf16),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k00), // %10
"w"(_k01), // %11
"w"(_k02), // %12
"w"(_k10), // %13
"w"(_k11), // %14
"w"(_k12), // %15
"w"(_k20), // %16
"w"(_k21), // %17
"w"(_k22) // %18
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6");
#endif // __aarch64__
}
for (; j < outw; j++)
{
float32x4_t _sum0 = vld1q_f32(outptr0);
float32x4_t _r0 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r0), 16));
float32x4_t _r1 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r1), 16));
float32x4_t _r2 = vreinterpretq_f32_u32(vshll_n_u16(vld1_u16(r2), 16));
#if __aarch64__
_sum0 = vfmaq_laneq_f32(_sum0, _k00, _r0, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k01, _r0, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k02, _r0, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k10, _r1, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k11, _r1, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k12, _r1, 2);
_sum0 = vfmaq_laneq_f32(_sum0, _k20, _r2, 0);
_sum0 = vfmaq_laneq_f32(_sum0, _k21, _r2, 1);
_sum0 = vfmaq_laneq_f32(_sum0, _k22, _r2, 2);
#else
_sum0 = vmlaq_lane_f32(_sum0, _k00, vget_low_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k01, vget_low_f32(_r0), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k02, vget_high_f32(_r0), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k10, vget_low_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k11, vget_low_f32(_r1), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k12, vget_high_f32(_r1), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k20, vget_low_f32(_r2), 0);
_sum0 = vmlaq_lane_f32(_sum0, _k21, vget_low_f32(_r2), 1);
_sum0 = vmlaq_lane_f32(_sum0, _k22, vget_high_f32(_r2), 0);
#endif
vst1_u16(outptr0_bf16, vshrn_n_u32(vreinterpretq_u32_f32(_sum0), 16));
r0 += 2;
r1 += 2;
r2 += 2;
outptr0 += 4;
outptr0_bf16 += 4;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
k0 += 9 * 4;
}
}
}
|
dqp3.c | /*! @copyright (c) 2017 King Abdullah University of Science and
* Technology (KAUST). All rights reserved.
*
* STARS-H is a software package, provided by King Abdullah
* University of Science and Technology (KAUST)
*
* @file src/backends/mpi/blrm/dqp3.c
* @version 1.3.0
* @author Aleksandr Mikhalev
* @date 2017-11-07
* */
#include "common.h"
#include "starsh.h"
#include "starsh-mpi.h"
int starsh_blrm__dqp3_mpi(STARSH_blrm **matrix, STARSH_blrf *format,
int maxrank, double tol, int onfly)
//! Approximate each tile of BLR matrix with RRQR (GEQP3 function).
/*!
* @param[out] matrix: Address of pointer to @ref STARSH_blrm object.
* @param[in] format: Block low-rank format.
* @param[in] maxrank: Maximum possible rank.
* @param[in] tol: Relative error tolerance.
* @param[in] onfly: Whether not to store dense blocks.
* @return Error code @ref STARSH_ERRNO.
* @ingroup blrm
* */
{
STARSH_blrf *F = format;
STARSH_problem *P = F->problem;
STARSH_kernel *kernel = P->kernel;
STARSH_int nblocks_far = F->nblocks_far;
STARSH_int nblocks_near = F->nblocks_near;
STARSH_int nblocks_far_local = F->nblocks_far_local;
STARSH_int nblocks_near_local = F->nblocks_near_local;
// Shortcuts to information about clusters
STARSH_cluster *RC = F->row_cluster;
STARSH_cluster *CC = F->col_cluster;
void *RD = RC->data, *CD = CC->data;
// Following values default to given block low-rank format F, but they are
// changed when there are false far-field blocks.
STARSH_int new_nblocks_far = F->nblocks_far;
STARSH_int new_nblocks_near = F->nblocks_near;
STARSH_int new_nblocks_far_local = F->nblocks_far_local;
STARSH_int new_nblocks_near_local = F->nblocks_near_local;
STARSH_int *block_far = F->block_far;
STARSH_int *block_near = F->block_near;
STARSH_int *block_far_local = F->block_far_local;
STARSH_int *block_near_local = F->block_near_local;
// Places to store low-rank factors, dense blocks and ranks
Array **far_U = NULL, **far_V = NULL, **near_D = NULL;
int *far_rank = NULL;
double *alloc_U = NULL, *alloc_V = NULL, *alloc_D = NULL;
size_t offset_U = 0, offset_V = 0, offset_D = 0;
STARSH_int lbi, lbj, bi, bj = 0;
double drsdd_time = 0, kernel_time = 0;
const int oversample = starsh_params.oversample;
// Init buffers to store low-rank factors of far-field blocks if needed
if(nblocks_far > 0)
{
STARSH_MALLOC(far_U, nblocks_far_local);
STARSH_MALLOC(far_V, nblocks_far_local);
STARSH_MALLOC(far_rank, nblocks_far_local);
size_t size_U = 0, size_V = 0;
// Simple cycle over all far-field blocks
for(lbi = 0; lbi < nblocks_far_local; lbi++)
{
STARSH_int bi = block_far_local[lbi];
// Get indexes of corresponding block row and block column
STARSH_int i = block_far[2*bi];
STARSH_int j = block_far[2*bi+1];
// Get corresponding sizes and minimum of them
size_U += RC->size[i];
size_V += CC->size[j];
}
size_U *= maxrank;
size_V *= maxrank;
STARSH_MALLOC(alloc_U, size_U);
STARSH_MALLOC(alloc_V, size_V);
for(lbi = 0; lbi < nblocks_far_local; lbi++)
{
STARSH_int bi = block_far_local[lbi];
// Get indexes of corresponding block row and block column
STARSH_int i = block_far[2*bi];
STARSH_int j = block_far[2*bi+1];
// Get corresponding sizes and minimum of them
size_t nrows = RC->size[i], ncols = CC->size[j];
int shape_U[] = {nrows, maxrank};
int shape_V[] = {ncols, maxrank};
double *U = alloc_U+offset_U, *V = alloc_V+offset_V;
offset_U += nrows*maxrank;
offset_V += ncols*maxrank;
array_from_buffer(far_U+lbi, 2, shape_U, 'd', 'F', U);
array_from_buffer(far_V+lbi, 2, shape_V, 'd', 'F', V);
}
offset_U = 0;
offset_V = 0;
}
// Work variables
int info;
// Simple cycle over all far-field admissible blocks
#pragma omp parallel for schedule(dynamic, 1)
for(lbi = 0; lbi < nblocks_far_local; lbi++)
{
STARSH_int bi = block_far_local[lbi];
// Get indexes of corresponding block row and block column
STARSH_int i = block_far[2*bi];
STARSH_int j = block_far[2*bi+1];
// Get corresponding sizes and minimum of them
int nrows = RC->size[i];
int ncols = CC->size[j];
int mn = nrows < ncols ? nrows : ncols;
int mn2 = maxrank+oversample;
if(mn2 > mn)
mn2 = mn;
// Get size of temporary arrays
int lwork = 3*ncols+1, lwork_sdd = (4*(size_t)mn2+7)*mn2;
if(lwork_sdd > lwork)
lwork = lwork_sdd;
lwork += (size_t)mn2*(2*ncols+mn2+1)+mn;
int liwork = ncols, liwork_sdd = 8*mn2;
if(liwork_sdd > liwork)
liwork = liwork_sdd;
double *D, *work;
int *iwork;
int info;
// Allocate temporary arrays
STARSH_PMALLOC(D, (size_t)nrows*(size_t)ncols, info);
STARSH_PMALLOC(iwork, liwork, info);
STARSH_PMALLOC(work, lwork, info);
// Compute elements of a block
#ifdef OPENMP
double time0 = omp_get_wtime();
#endif
kernel(nrows, ncols, RC->pivot+RC->start[i], CC->pivot+CC->start[j],
RD, CD, D, nrows);
#ifdef OPENMP
double time1 = omp_get_wtime();
#endif
starsh_dense_dlrqp3(nrows, ncols, D, nrows, far_U[lbi]->data, nrows,
far_V[lbi]->data, ncols, far_rank+lbi, maxrank, oversample,
tol, work, lwork, iwork);
#ifdef OPENMP
double time2 = omp_get_wtime();
#pragma omp critical
{
drsdd_time += time2-time1;
kernel_time += time1-time0;
}
#endif
// Free temporary arrays
free(D);
free(work);
free(iwork);
}
// Get number of false far-field blocks
STARSH_int nblocks_false_far_local = 0;
STARSH_int *false_far_local = NULL;
for(lbi = 0; lbi < nblocks_far_local; lbi++)
if(far_rank[lbi] == -1)
nblocks_false_far_local++;
if(nblocks_false_far_local > 0)
{
// IMPORTANT: `false_far` and `false_far_local` must be in
// ascending order for later code to work normally
STARSH_MALLOC(false_far_local, nblocks_false_far_local);
lbj = 0;
for(lbi = 0; lbi < nblocks_far_local; lbi++)
if(far_rank[lbi] == -1)
false_far_local[lbj++] = block_far_local[lbi];
}
// Sync list of all false far-field blocks
STARSH_int nblocks_false_far = 0;
int int_nblocks_false_far_local = nblocks_false_far_local;
int *mpi_recvcount, *mpi_offset;
int mpi_size, mpi_rank;
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
STARSH_MALLOC(mpi_recvcount, mpi_size);
STARSH_MALLOC(mpi_offset, mpi_size);
MPI_Allgather(&int_nblocks_false_far_local, 1, MPI_INT, mpi_recvcount,
1, MPI_INT, MPI_COMM_WORLD);
for(bi = 0; bi < mpi_size; bi++)
nblocks_false_far += mpi_recvcount[bi];
mpi_offset[0] = 0;
for(bi = 1; bi < mpi_size; bi++)
mpi_offset[bi] = mpi_offset[bi-1]+mpi_recvcount[bi-1];
STARSH_int *false_far = NULL;
if(nblocks_false_far > 0)
STARSH_MALLOC(false_far, nblocks_false_far);
MPI_Allgatherv(false_far_local, nblocks_false_far_local, my_MPI_SIZE_T,
false_far, mpi_recvcount, mpi_offset, my_MPI_SIZE_T,
MPI_COMM_WORLD);
free(mpi_recvcount);
free(mpi_offset);
// Make false_far be in ascending order
qsort(false_far, nblocks_false_far, sizeof(*false_far), cmp_size_t);
if(nblocks_false_far > 0)
{
// Update list of near-field blocks
new_nblocks_near = nblocks_near+nblocks_false_far;
new_nblocks_near_local = nblocks_near_local+nblocks_false_far_local;
STARSH_MALLOC(block_near, 2*new_nblocks_near);
if(new_nblocks_near_local > 0)
STARSH_MALLOC(block_near_local, new_nblocks_near_local);
// At first get all near-field blocks, assumed to be dense
#pragma omp parallel for schedule(static)
for(bi = 0; bi < 2*nblocks_near; bi++)
block_near[bi] = F->block_near[bi];
#pragma omp parallel for schedule(static)
for(lbi = 0; lbi < nblocks_near_local; lbi++)
block_near_local[lbi] = F->block_near_local[lbi];
// Add false far-field blocks
#pragma omp parallel for schedule(static)
for(bi = 0; bi < nblocks_false_far; bi++)
{
STARSH_int bj = false_far[bi];
block_near[2*(bi+nblocks_near)] = F->block_far[2*bj];
block_near[2*(bi+nblocks_near)+1] = F->block_far[2*bj+1];
}
bi = 0;
for(lbi = 0; lbi < nblocks_false_far_local; lbi++)
{
lbj = false_far_local[lbi];
while(bi < nblocks_false_far && false_far[bi] < lbj)
bi++;
block_near_local[nblocks_near_local+lbi] = nblocks_near+bi;
}
// Update list of far-field blocks
new_nblocks_far = nblocks_far-nblocks_false_far;
new_nblocks_far_local = nblocks_far_local-nblocks_false_far_local;
if(new_nblocks_far > 0)
{
STARSH_MALLOC(block_far, 2*new_nblocks_far);
if(new_nblocks_far_local > 0)
STARSH_MALLOC(block_far_local, new_nblocks_far_local);
bj = 0;
lbi = 0;
lbj = 0;
for(bi = 0; bi < nblocks_far; bi++)
{
// `false_far` must be in ascending order for this to work
if(bj < nblocks_false_far && false_far[bj] == bi)
{
if(nblocks_false_far_local > lbj &&
false_far_local[lbj] == bi)
{
lbi++;
lbj++;
}
bj++;
}
else
{
block_far[2*(bi-bj)] = F->block_far[2*bi];
block_far[2*(bi-bj)+1] = F->block_far[2*bi+1];
if(nblocks_far_local > lbi &&
F->block_far_local[lbi] == bi)
{
block_far_local[lbi-lbj] = bi-bj;
lbi++;
}
}
}
}
// Update format by creating new format
STARSH_blrf *F2;
info = starsh_blrf_new_from_coo_mpi(&F2, P, F->symm, RC, CC,
new_nblocks_far, block_far, new_nblocks_far_local,
block_far_local, new_nblocks_near, block_near,
new_nblocks_near_local, block_near_local, F->type);
// Swap internal data of formats and free unnecessary data
STARSH_blrf tmp_blrf = *F;
*F = *F2;
*F2 = tmp_blrf;
if(mpi_rank == 0)
STARSH_WARNING("`F` was modified due to false far-field blocks");
starsh_blrf_free(F2);
}
// Compute near-field blocks if needed
if(onfly == 0 && new_nblocks_near > 0)
{
STARSH_MALLOC(near_D, new_nblocks_near_local);
size_t size_D = 0;
// Simple cycle over all near-field blocks
for(lbi = 0; lbi < new_nblocks_near_local; lbi++)
{
STARSH_int bi = block_near_local[lbi];
// Get indexes of corresponding block row and block column
STARSH_int i = block_near[2*bi];
STARSH_int j = block_near[2*bi+1];
// Get corresponding sizes and minimum of them
size_t nrows = RC->size[i];
size_t ncols = CC->size[j];
// Update size_D
size_D += nrows*ncols;
}
STARSH_MALLOC(alloc_D, size_D);
// For each near-field block compute its elements
#pragma omp parallel for schedule(dynamic, 1)
for(lbi = 0; lbi < new_nblocks_near_local; lbi++)
{
STARSH_int bi = block_near_local[lbi];
// Get indexes of corresponding block row and block column
STARSH_int i = block_near[2*bi];
STARSH_int j = block_near[2*bi+1];
// Get corresponding sizes and minimum of them
int nrows = RC->size[i];
int ncols = CC->size[j];
int shape[2] = {nrows, ncols};
double *D;
#pragma omp critical
{
D = alloc_D+offset_D;
offset_D += nrows*ncols;
//array_from_buffer(near_D+lbi, 2, shape, 'd', 'F', D);
//offset_D += near_D[lbi]->size;
}
array_from_buffer(near_D+lbi, 2, shape, 'd', 'F', D);
#ifdef OPENMP
double time0 = omp_get_wtime();
#endif
kernel(nrows, ncols, RC->pivot+RC->start[i],
CC->pivot+CC->start[j], RD, CD, D, nrows);
#ifdef OPENMP
double time1 = omp_get_wtime();
#pragma omp critical
kernel_time += time1-time0;
#endif
}
}
// Change sizes of far_rank, far_U and far_V if there were false
// far-field blocks
lbj = 0;
for(lbi = 0; lbi < nblocks_far_local; lbi++)
{
if(far_rank[lbi] == -1)
lbj++;
else
{
int shape_U[2] = {far_U[lbi]->shape[0], far_rank[lbi]};
int shape_V[2] = {far_V[lbi]->shape[0], far_rank[lbi]};
array_from_buffer(far_U+lbi-lbj, 2, shape_U, 'd', 'F',
far_U[lbi]->data);
array_from_buffer(far_V+lbi-lbj, 2, shape_V, 'd', 'F',
far_V[lbi]->data);
far_rank[lbi-lbj] = far_rank[lbi];
}
}
if(nblocks_false_far_local > 0 && new_nblocks_far_local > 0)
{
STARSH_REALLOC(far_rank, new_nblocks_far_local);
STARSH_REALLOC(far_U, new_nblocks_far_local);
STARSH_REALLOC(far_V, new_nblocks_far_local);
}
// If all far-field blocks are false, then dealloc buffers
if(new_nblocks_far_local == 0 && nblocks_far_local > 0)
{
block_far = NULL;
free(far_rank);
far_rank = NULL;
free(far_U);
far_U = NULL;
free(far_V);
far_V = NULL;
free(alloc_U);
alloc_U = NULL;
free(alloc_V);
alloc_V = NULL;
}
// Dealloc list of false far-field blocks if it is not empty
if(nblocks_false_far > 0)
free(false_far);
if(nblocks_false_far_local > 0)
free(false_far_local);
// Finish with creating instance of Block Low-Rank Matrix with given
// buffers
#ifdef OPENMP
double mpi_drsdd_time = 0, mpi_kernel_time = 0;
MPI_Reduce(&drsdd_time, &mpi_drsdd_time, 1, MPI_DOUBLE, MPI_SUM, 0,
MPI_COMM_WORLD);
MPI_Reduce(&kernel_time, &mpi_kernel_time, 1, MPI_DOUBLE, MPI_SUM, 0,
MPI_COMM_WORLD);
if(mpi_rank == 0)
{
//STARSH_WARNING("DRSDD kernel total time: %e secs", mpi_drsdd_time);
//STARSH_WARNING("MATRIX kernel total time: %e secs", mpi_kernel_time);
}
#endif
return starsh_blrm_new_mpi(matrix, F, far_rank, far_U, far_V, onfly,
near_D, alloc_U, alloc_V, alloc_D, '1');
}
|
GB_binop__plus_int8.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__plus_int8)
// A.*B function (eWiseMult): GB (_AemultB)
// A.*B function (eWiseMult): GB (_AemultB_02__plus_int8)
// A.*B function (eWiseMult): GB (_AemultB_03__plus_int8)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__plus_int8)
// A*D function (colscale): GB (_AxD__plus_int8)
// D*A function (rowscale): GB (_DxB__plus_int8)
// C+=B function (dense accum): GB (_Cdense_accumB__plus_int8)
// C+=b function (dense accum): GB (_Cdense_accumb__plus_int8)
// C+=A+B function (dense ewise3): GB (_Cdense_ewise3_accum__plus_int8)
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__plus_int8)
// C=scalar+B GB (_bind1st__plus_int8)
// C=scalar+B' GB (_bind1st_tran__plus_int8)
// C=A+scalar GB (_bind2nd__plus_int8)
// C=A'+scalar GB (_bind2nd_tran__plus_int8)
// C type: int8_t
// A type: int8_t
// B,b type: int8_t
// BinaryOp: cij = (aij + bij)
#define GB_ATYPE \
int8_t
#define GB_BTYPE \
int8_t
#define GB_CTYPE \
int8_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int8_t aij = Ax [pA]
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB) \
int8_t bij = Bx [pB]
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int8_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA) \
cij = Ax [pA]
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB) \
cij = Bx [pB]
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z, x, y, i, j) \
z = (x + y) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_PLUS || GxB_NO_INT8 || GxB_NO_PLUS_INT8)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB (_Cdense_ewise3_accum__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_ewise3_noaccum__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__plus_int8)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int8_t
int8_t bwork = (*((int8_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *restrict Cx = (int8_t *) C->x ;
#include "GB_AxB_colscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *restrict Cx = (int8_t *) C->x ;
#include "GB_AxB_rowscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__plus_int8)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
#include "GB_add_template.c"
GB_FREE_WORK ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_01__plus_int8)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_01_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_03__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_03_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__plus_int8)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__plus_int8)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t *Cx = (int8_t *) Cx_output ;
int8_t x = (*((int8_t *) x_input)) ;
int8_t *Bx = (int8_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Bb, p)) continue ;
int8_t bij = Bx [p] ;
Cx [p] = (x + bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__plus_int8)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int8_t *Cx = (int8_t *) Cx_output ;
int8_t *Ax = (int8_t *) Ax_input ;
int8_t y = (*((int8_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int8_t aij = Ax [p] ;
Cx [p] = (aij + y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int8_t aij = Ax [pA] ; \
Cx [pC] = (x + aij) ; \
}
GrB_Info GB (_bind1st_tran__plus_int8)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int8_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t x = (*((const int8_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int8_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int8_t aij = Ax [pA] ; \
Cx [pC] = (aij + y) ; \
}
GrB_Info GB (_bind2nd_tran__plus_int8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int8_t y = (*((const int8_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
wand-view.c | /*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% W W AAA N N DDDD %
% W W A A NN N D D %
% W W W AAAAA N N N D D %
% WW WW A A N NN D D %
% W W A A N N DDDD %
% %
% V V IIIII EEEEE W W %
% V V I E W W %
% V V I EEE W W W %
% V V I E WW WW %
% V IIIII EEEEE W W %
% %
% %
% MagickWand Wand View Methods %
% %
% Software Design %
% Cristy %
% March 2003 %
% %
% %
% Copyright 1999-2018 ImageMagick Studio LLC, a non-profit organization %
% dedicated to making software imaging solutions freely available. %
% %
% You may not use this file except in compliance with the License. You may %
% obtain a copy of the License at %
% %
% https://www.imagemagick.org/script/license.php %
% %
% Unless required by applicable law or agreed to in writing, software %
% distributed under the License is distributed on an "AS IS" BASIS, %
% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. %
% See the License for the specific language governing permissions and %
% limitations under the License. %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%
%
*/
/*
Include declarations.
*/
#include "wand/studio.h"
#include "wand/MagickWand.h"
#include "wand/magick-wand-private.h"
#include "wand/wand.h"
#include "magick/monitor-private.h"
#include "magick/thread-private.h"
/*
Define declarations.
*/
#define WandViewId "WandView"
/*
Typedef declarations.
*/
struct _WandView
{
size_t
id;
char
name[MaxTextExtent],
*description;
RectangleInfo
extent;
MagickWand
*wand;
CacheView
*view;
size_t
number_threads;
PixelWand
***pixel_wands;
ExceptionInfo
*exception;
MagickBooleanType
debug;
size_t
signature;
};
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% C l o n e W a n d V i e w %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% CloneWandView() makes a copy of the specified wand view.
%
% The format of the CloneWandView method is:
%
% WandView *CloneWandView(const WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
WandExport WandView *CloneWandView(const WandView *wand_view)
{
WandView
*clone_view;
register ssize_t
i;
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
if (wand_view->debug != MagickFalse)
(void) LogMagickEvent(WandEvent,GetMagickModule(),"%s",wand_view->name);
clone_view=(WandView *) AcquireMagickMemory(sizeof(*clone_view));
if (clone_view == (WandView *) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
wand_view->name);
(void) memset(clone_view,0,sizeof(*clone_view));
clone_view->id=AcquireWandId();
(void) FormatLocaleString(clone_view->name,MaxTextExtent,"%s-%.20g",
WandViewId,(double) clone_view->id);
clone_view->description=ConstantString(wand_view->description);
clone_view->view=CloneCacheView(wand_view->view);
clone_view->extent=wand_view->extent;
clone_view->number_threads=wand_view->number_threads;
clone_view->exception=AcquireExceptionInfo();
InheritException(clone_view->exception,wand_view->exception);
for (i=0; i < (ssize_t) wand_view->number_threads; i++)
clone_view->pixel_wands[i]=ClonePixelWands((const PixelWand **)
wand_view->pixel_wands[i],wand_view->extent.width);
clone_view->debug=wand_view->debug;
if (clone_view->debug != MagickFalse)
(void) LogMagickEvent(WandEvent,GetMagickModule(),"%s",clone_view->name);
clone_view->signature=WandSignature;
return(clone_view);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% D e s t r o y W a n d V i e w %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% DestroyWandView() deallocates memory associated with a wand view.
%
% The format of the DestroyWandView method is:
%
% WandView *DestroyWandView(WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
static PixelWand ***DestroyPixelsThreadSet(PixelWand ***pixel_wands,
const size_t number_wands,const size_t number_threads)
{
register ssize_t
i;
assert(pixel_wands != (PixelWand ***) NULL);
for (i=0; i < (ssize_t) number_threads; i++)
if (pixel_wands[i] != (PixelWand **) NULL)
pixel_wands[i]=DestroyPixelWands(pixel_wands[i],number_wands);
pixel_wands=(PixelWand ***) RelinquishMagickMemory(pixel_wands);
return(pixel_wands);
}
WandExport WandView *DestroyWandView(WandView *wand_view)
{
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
wand_view->pixel_wands=DestroyPixelsThreadSet(wand_view->pixel_wands,
wand_view->extent.width,wand_view->number_threads);
wand_view->view=DestroyCacheView(wand_view->view);
wand_view->exception=DestroyExceptionInfo(wand_view->exception);
wand_view->signature=(~WandSignature);
RelinquishWandId(wand_view->id);
wand_view=(WandView *) RelinquishMagickMemory(wand_view);
return(wand_view);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% D u p l e x T r a n s f e r W a n d V i e w I t e r a t o r %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% DuplexTransferWandViewIterator() iterates over three wand views in
% parallel and calls your transfer method for each scanline of the view. The
% source and duplex pixel extent is not confined to the image canvas-- that is
% you can include negative offsets or widths or heights that exceed the image
% dimension. However, the destination wand view is confined to the image
% canvas-- that is no negative offsets or widths or heights that exceed the
% image dimension are permitted.
%
% The callback signature is:
%
% MagickBooleanType DuplexTransferImageViewMethod(const WandView *source,
% const WandView *duplex,WandView *destination,const ssize_t y,
% const int thread_id,void *context)
%
% Use this pragma if the view is not single threaded:
%
% #pragma omp critical
%
% to define a section of code in your callback transfer method that must be
% executed by a single thread at a time.
%
% The format of the DuplexTransferWandViewIterator method is:
%
% MagickBooleanType DuplexTransferWandViewIterator(WandView *source,
% WandView *duplex,WandView *destination,
% DuplexTransferWandViewMethod transfer,void *context)
%
% A description of each parameter follows:
%
% o source: the source wand view.
%
% o duplex: the duplex wand view.
%
% o destination: the destination wand view.
%
% o transfer: the transfer callback method.
%
% o context: the user defined context.
%
*/
WandExport MagickBooleanType DuplexTransferWandViewIterator(WandView *source,
WandView *duplex,WandView *destination,DuplexTransferWandViewMethod transfer,
void *context)
{
ExceptionInfo
*exception;
Image
*destination_image,
*duplex_image,
*source_image;
MagickBooleanType
status;
MagickOffsetType
progress;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
size_t
height;
#endif
ssize_t
y;
assert(source != (WandView *) NULL);
assert(source->signature == WandSignature);
if (transfer == (DuplexTransferWandViewMethod) NULL)
return(MagickFalse);
source_image=source->wand->images;
duplex_image=duplex->wand->images;
destination_image=destination->wand->images;
if (SetImageStorageClass(destination_image,DirectClass) == MagickFalse)
return(MagickFalse);
status=MagickTrue;
progress=0;
exception=destination->exception;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
height=(size_t) (source->extent.height-source->extent.y);
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,destination_image,height,1)
#endif
for (y=source->extent.y; y < (ssize_t) source->extent.height; y++)
{
const int
id = GetOpenMPThreadId();
MagickBooleanType
sync;
register const IndexPacket
*magick_restrict duplex_indexes,
*magick_restrict indexes;
register const PixelPacket
*magick_restrict duplex_pixels,
*magick_restrict pixels;
register IndexPacket
*magick_restrict destination_indexes;
register ssize_t
x;
register PixelPacket
*magick_restrict destination_pixels;
if (status == MagickFalse)
continue;
pixels=GetCacheViewVirtualPixels(source->view,source->extent.x,y,
source->extent.width,1,source->exception);
if (pixels == (const PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewVirtualIndexQueue(source->view);
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetQuantumColor(source->pixel_wands[id][x],pixels+x);
if (source_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetBlackQuantum(source->pixel_wands[id][x],
GetPixelBlack(indexes+x));
if (source_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetIndex(source->pixel_wands[id][x],
GetPixelIndex(indexes+x));
duplex_pixels=GetCacheViewVirtualPixels(duplex->view,duplex->extent.x,y,
duplex->extent.width,1,duplex->exception);
if (duplex_pixels == (const PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
duplex_indexes=GetCacheViewVirtualIndexQueue(duplex->view);
for (x=0; x < (ssize_t) duplex->extent.width; x++)
PixelSetQuantumColor(duplex->pixel_wands[id][x],duplex_pixels+x);
if (duplex_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) duplex->extent.width; x++)
PixelSetBlackQuantum(duplex->pixel_wands[id][x],
GetPixelBlack(duplex_indexes+x));
if (duplex_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) duplex->extent.width; x++)
PixelSetIndex(duplex->pixel_wands[id][x],
GetPixelIndex(duplex_indexes+x));
destination_pixels=GetCacheViewAuthenticPixels(destination->view,
destination->extent.x,y,destination->extent.width,1,exception);
if (destination_pixels == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
destination_indexes=GetCacheViewAuthenticIndexQueue(destination->view);
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetQuantumColor(destination->pixel_wands[id][x],
destination_pixels+x);
if (destination_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetBlackQuantum(destination->pixel_wands[id][x],
GetPixelBlack(destination_indexes+x));
if (destination_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetIndex(destination->pixel_wands[id][x],
GetPixelIndex(destination_indexes+x));
if (transfer(source,duplex,destination,y,id,context) == MagickFalse)
status=MagickFalse;
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelGetQuantumColor(destination->pixel_wands[id][x],
destination_pixels+x);
if (destination_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) destination->extent.width; x++)
SetPixelBlack(destination_indexes+x,PixelGetBlackQuantum(
destination->pixel_wands[id][x]));
sync=SyncCacheViewAuthenticPixels(destination->view,exception);
if (sync == MagickFalse)
{
InheritException(destination->exception,GetCacheViewException(
source->view));
status=MagickFalse;
}
if (source_image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickWand_DuplexTransferWandViewIterator)
#endif
proceed=SetImageProgress(source_image,source->description,progress++,
source->extent.height);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t W a n d V i e w E x c e p t i o n %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetWandViewException() returns the severity, reason, and description of any
% error that occurs when utilizing a wand view.
%
% The format of the GetWandViewException method is:
%
% char *GetWandViewException(const WandView *wand_view,
% ExceptionType *severity)
%
% A description of each parameter follows:
%
% o wand_view: the pixel wand_view.
%
% o severity: the severity of the error is returned here.
%
*/
WandExport char *GetWandViewException(const WandView *wand_view,
ExceptionType *severity)
{
char
*description;
assert(wand_view != (const WandView *) NULL);
assert(wand_view->signature == WandSignature);
if (wand_view->debug != MagickFalse)
(void) LogMagickEvent(WandEvent,GetMagickModule(),"%s",wand_view->name);
assert(severity != (ExceptionType *) NULL);
*severity=wand_view->exception->severity;
description=(char *) AcquireQuantumMemory(2UL*MaxTextExtent,
sizeof(*description));
if (description == (char *) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
wand_view->name);
*description='\0';
if (wand_view->exception->reason != (char *) NULL)
(void) CopyMagickString(description,GetLocaleExceptionMessage(
wand_view->exception->severity,wand_view->exception->reason),
MaxTextExtent);
if (wand_view->exception->description != (char *) NULL)
{
(void) ConcatenateMagickString(description," (",MaxTextExtent);
(void) ConcatenateMagickString(description,GetLocaleExceptionMessage(
wand_view->exception->severity,wand_view->exception->description),
MaxTextExtent);
(void) ConcatenateMagickString(description,")",MaxTextExtent);
}
return(description);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t W a n d V i e w E x t e n t %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetWandViewExtent() returns the wand view extent.
%
% The format of the GetWandViewExtent method is:
%
% RectangleInfo GetWandViewExtent(const WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
WandExport RectangleInfo GetWandViewExtent(const WandView *wand_view)
{
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
return(wand_view->extent);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t W a n d V i e w I t e r a t o r %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetWandViewIterator() iterates over the wand view in parallel and calls
% your get method for each scanline of the view. The pixel extent is
% not confined to the image canvas-- that is you can include negative offsets
% or widths or heights that exceed the image dimension. Any updates to
% the pixels in your callback are ignored.
%
% The callback signature is:
%
% MagickBooleanType GetImageViewMethod(const WandView *source,
% const ssize_t y,const int thread_id,void *context)
%
% Use this pragma if the view is not single threaded:
%
% #pragma omp critical
%
% to define a section of code in your callback get method that must be
% executed by a single thread at a time.
%
% The format of the GetWandViewIterator method is:
%
% MagickBooleanType GetWandViewIterator(WandView *source,
% GetWandViewMethod get,void *context)
%
% A description of each parameter follows:
%
% o source: the source wand view.
%
% o get: the get callback method.
%
% o context: the user defined context.
%
*/
WandExport MagickBooleanType GetWandViewIterator(WandView *source,
GetWandViewMethod get,void *context)
{
Image
*source_image;
MagickBooleanType
status;
MagickOffsetType
progress;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
size_t
height;
#endif
ssize_t
y;
assert(source != (WandView *) NULL);
assert(source->signature == WandSignature);
if (get == (GetWandViewMethod) NULL)
return(MagickFalse);
source_image=source->wand->images;
status=MagickTrue;
progress=0;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
height=(size_t) (source->extent.height-source->extent.y);
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,source_image,height,1)
#endif
for (y=source->extent.y; y < (ssize_t) source->extent.height; y++)
{
const int
id = GetOpenMPThreadId();
register const IndexPacket
*indexes;
register const PixelPacket
*pixels;
register ssize_t
x;
if (status == MagickFalse)
continue;
pixels=GetCacheViewVirtualPixels(source->view,source->extent.x,y,
source->extent.width,1,source->exception);
if (pixels == (const PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewVirtualIndexQueue(source->view);
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetQuantumColor(source->pixel_wands[id][x],pixels+x);
if (source_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetBlackQuantum(source->pixel_wands[id][x],
GetPixelBlack(indexes+x));
if (source_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetIndex(source->pixel_wands[id][x],
GetPixelIndex(indexes+x));
if (get(source,y,id,context) == MagickFalse)
status=MagickFalse;
if (source_image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickWand_GetWandViewIterator)
#endif
proceed=SetImageProgress(source_image,source->description,progress++,
source->extent.height);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t W a n d V i e w P i x e l s %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetWandViewPixels() returns the wand view pixel_wands.
%
% The format of the GetWandViewPixels method is:
%
% PixelWand *GetWandViewPixels(const WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
WandExport PixelWand **GetWandViewPixels(const WandView *wand_view)
{
const int
id = GetOpenMPThreadId();
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
return(wand_view->pixel_wands[id]);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t W a n d V i e w W a n d %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetWandViewWand() returns the magick wand associated with the wand view.
%
% The format of the GetWandViewWand method is:
%
% MagickWand *GetWandViewWand(const WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
WandExport MagickWand *GetWandViewWand(const WandView *wand_view)
{
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
return(wand_view->wand);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% I s W a n d V i e w %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% IsWandView() returns MagickTrue if the the parameter is verified as a wand
% view object.
%
% The format of the IsWandView method is:
%
% MagickBooleanType IsWandView(const WandView *wand_view)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
*/
WandExport MagickBooleanType IsWandView(const WandView *wand_view)
{
size_t
length;
if (wand_view == (const WandView *) NULL)
return(MagickFalse);
if (wand_view->signature != WandSignature)
return(MagickFalse);
length=strlen(WandViewId);
if (LocaleNCompare(wand_view->name,WandViewId,length) != 0)
return(MagickFalse);
return(MagickTrue);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% N e w W a n d V i e w %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% NewWandView() returns a wand view required for all other methods in the
% Wand View API.
%
% The format of the NewWandView method is:
%
% WandView *NewWandView(MagickWand *wand)
%
% A description of each parameter follows:
%
% o wand: the wand.
%
*/
static PixelWand ***AcquirePixelsThreadSet(const size_t number_wands,
const size_t number_threads)
{
PixelWand
***pixel_wands;
register ssize_t
i;
pixel_wands=(PixelWand ***) AcquireQuantumMemory(number_threads,
sizeof(*pixel_wands));
if (pixel_wands == (PixelWand ***) NULL)
return((PixelWand ***) NULL);
(void) memset(pixel_wands,0,number_threads*sizeof(*pixel_wands));
for (i=0; i < (ssize_t) number_threads; i++)
{
pixel_wands[i]=NewPixelWands(number_wands);
if (pixel_wands[i] == (PixelWand **) NULL)
return(DestroyPixelsThreadSet(pixel_wands,number_wands,number_threads));
}
return(pixel_wands);
}
WandExport WandView *NewWandView(MagickWand *wand)
{
WandView
*wand_view;
assert(wand != (MagickWand *) NULL);
assert(wand->signature == WandSignature);
wand_view=(WandView *) AcquireMagickMemory(sizeof(*wand_view));
if (wand_view == (WandView *) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
GetExceptionMessage(errno));
(void) memset(wand_view,0,sizeof(*wand_view));
wand_view->id=AcquireWandId();
(void) FormatLocaleString(wand_view->name,MaxTextExtent,"%s-%.20g",
WandViewId,(double) wand_view->id);
wand_view->description=ConstantString("WandView");
wand_view->wand=wand;
wand_view->exception=AcquireExceptionInfo();
wand_view->view=AcquireVirtualCacheView(wand_view->wand->images,
wand_view->exception);
wand_view->extent.width=wand->images->columns;
wand_view->extent.height=wand->images->rows;
wand_view->number_threads=GetOpenMPMaximumThreads();
wand_view->pixel_wands=AcquirePixelsThreadSet(wand_view->extent.width,
wand_view->number_threads);
if (wand_view->pixel_wands == (PixelWand ***) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
GetExceptionMessage(errno));
wand_view->debug=IsEventLogging();
wand_view->signature=WandSignature;
return(wand_view);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% N e w W a n d V i e w E x t e n t %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% NewWandViewExtent() returns a wand view required for all other methods
% in the Wand View API.
%
% The format of the NewWandViewExtent method is:
%
% WandView *NewWandViewExtent(MagickWand *wand,const ssize_t x,
% const ssize_t y,const size_t width,const size_t height)
%
% A description of each parameter follows:
%
% o wand: the magick wand.
%
% o x,y,columns,rows: These values define the perimeter of a extent of
% pixel_wands view.
%
*/
WandExport WandView *NewWandViewExtent(MagickWand *wand,const ssize_t x,
const ssize_t y,const size_t width,const size_t height)
{
WandView
*wand_view;
assert(wand != (MagickWand *) NULL);
assert(wand->signature == WandSignature);
wand_view=(WandView *) AcquireMagickMemory(sizeof(*wand_view));
if (wand_view == (WandView *) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
GetExceptionMessage(errno));
(void) memset(wand_view,0,sizeof(*wand_view));
wand_view->id=AcquireWandId();
(void) FormatLocaleString(wand_view->name,MaxTextExtent,"%s-%.20g",
WandViewId,(double) wand_view->id);
wand_view->description=ConstantString("WandView");
wand_view->exception=AcquireExceptionInfo();
wand_view->view=AcquireVirtualCacheView(wand_view->wand->images,
wand_view->exception);
wand_view->wand=wand;
wand_view->extent.width=width;
wand_view->extent.height=height;
wand_view->extent.x=x;
wand_view->extent.y=y;
wand_view->number_threads=GetOpenMPMaximumThreads();
wand_view->pixel_wands=AcquirePixelsThreadSet(wand_view->extent.width,
wand_view->number_threads);
if (wand_view->pixel_wands == (PixelWand ***) NULL)
ThrowWandFatalException(ResourceLimitFatalError,"MemoryAllocationFailed",
GetExceptionMessage(errno));
wand_view->debug=IsEventLogging();
wand_view->signature=WandSignature;
return(wand_view);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% S e t W a n d V i e w D e s c r i p t i o n %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% SetWandViewDescription() associates a description with an image view.
%
% The format of the SetWandViewDescription method is:
%
% void SetWandViewDescription(WandView *image_view,const char *description)
%
% A description of each parameter follows:
%
% o wand_view: the wand view.
%
% o description: the wand view description.
%
*/
MagickExport void SetWandViewDescription(WandView *wand_view,
const char *description)
{
assert(wand_view != (WandView *) NULL);
assert(wand_view->signature == WandSignature);
wand_view->description=ConstantString(description);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% S e t W a n d V i e w I t e r a t o r %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% SetWandViewIterator() iterates over the wand view in parallel and calls
% your set method for each scanline of the view. The pixel extent is
% confined to the image canvas-- that is no negative offsets or widths or
% heights that exceed the image dimension. The pixels are initiallly
% undefined and any settings you make in the callback method are automagically
% synced back to your image.
%
% The callback signature is:
%
% MagickBooleanType SetImageViewMethod(ImageView *destination,
% const ssize_t y,const int thread_id,void *context)
%
% Use this pragma if the view is not single threaded:
%
% #pragma omp critical
%
% to define a section of code in your callback set method that must be
% executed by a single thread at a time.
%
% The format of the SetWandViewIterator method is:
%
% MagickBooleanType SetWandViewIterator(WandView *destination,
% SetWandViewMethod set,void *context)
%
% A description of each parameter follows:
%
% o destination: the wand view.
%
% o set: the set callback method.
%
% o context: the user defined context.
%
*/
WandExport MagickBooleanType SetWandViewIterator(WandView *destination,
SetWandViewMethod set,void *context)
{
ExceptionInfo
*exception;
Image
*destination_image;
MagickBooleanType
status;
MagickOffsetType
progress;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
size_t
height;
#endif
ssize_t
y;
assert(destination != (WandView *) NULL);
assert(destination->signature == WandSignature);
if (set == (SetWandViewMethod) NULL)
return(MagickFalse);
destination_image=destination->wand->images;
if (SetImageStorageClass(destination_image,DirectClass) == MagickFalse)
return(MagickFalse);
status=MagickTrue;
progress=0;
exception=destination->exception;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
height=(size_t) (destination->extent.height-destination->extent.y);
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(destination_image,destination_image,height,1)
#endif
for (y=destination->extent.y; y < (ssize_t) destination->extent.height; y++)
{
const int
id = GetOpenMPThreadId();
MagickBooleanType
sync;
register IndexPacket
*magick_restrict indexes;
register ssize_t
x;
register PixelPacket
*magick_restrict pixels;
if (status == MagickFalse)
continue;
pixels=GetCacheViewAuthenticPixels(destination->view,destination->extent.x,
y,destination->extent.width,1,exception);
if (pixels == (PixelPacket *) NULL)
{
InheritException(destination->exception,GetCacheViewException(
destination->view));
status=MagickFalse;
continue;
}
indexes=GetCacheViewAuthenticIndexQueue(destination->view);
if (set(destination,y,id,context) == MagickFalse)
status=MagickFalse;
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelGetQuantumColor(destination->pixel_wands[id][x],pixels+x);
if (destination_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) destination->extent.width; x++)
SetPixelBlack(indexes+x,PixelGetBlackQuantum(
destination->pixel_wands[id][x]));
sync=SyncCacheViewAuthenticPixels(destination->view,exception);
if (sync == MagickFalse)
{
InheritException(destination->exception,GetCacheViewException(
destination->view));
status=MagickFalse;
}
if (destination_image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickWand_SetWandViewIterator)
#endif
proceed=SetImageProgress(destination_image,destination->description,
progress++,destination->extent.height);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% S e t W a n d V i e w T h r e a d s %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% SetWandViewThreads() sets the number of threads in a thread team.
%
% The format of the SetWandViewDescription method is:
%
% void SetWandViewThreads(WandView *image_view,
% const size_t number_threads)
%
% A description of each parameter follows:
%
% o image_view: the image view.
%
% o number_threads: the number of threads in a thread team.
%
*/
MagickExport void SetWandViewThreads(WandView *image_view,
const size_t number_threads)
{
assert(image_view != (WandView *) NULL);
assert(image_view->signature == MagickCoreSignature);
image_view->number_threads=number_threads;
if (number_threads > (size_t) GetMagickResourceLimit(ThreadResource))
image_view->number_threads=GetOpenMPMaximumThreads();
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% T r a n s f e r W a n d V i e w I t e r a t o r %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% TransferWandViewIterator() iterates over two wand views in parallel and
% calls your transfer method for each scanline of the view. The source pixel
% extent is not confined to the image canvas-- that is you can include
% negative offsets or widths or heights that exceed the image dimension.
% However, the destination wand view is confined to the image canvas-- that
% is no negative offsets or widths or heights that exceed the image dimension
% are permitted.
%
% The callback signature is:
%
% MagickBooleanType TransferImageViewMethod(const WandView *source,
% WandView *destination,const ssize_t y,const int thread_id,
% void *context)
%
% Use this pragma if the view is not single threaded:
%
% #pragma omp critical
%
% to define a section of code in your callback transfer method that must be
% executed by a single thread at a time.
%
% The format of the TransferWandViewIterator method is:
%
% MagickBooleanType TransferWandViewIterator(WandView *source,
% WandView *destination,TransferWandViewMethod transfer,void *context)
%
% A description of each parameter follows:
%
% o source: the source wand view.
%
% o destination: the destination wand view.
%
% o transfer: the transfer callback method.
%
% o context: the user defined context.
%
*/
WandExport MagickBooleanType TransferWandViewIterator(WandView *source,
WandView *destination,TransferWandViewMethod transfer,void *context)
{
ExceptionInfo
*exception;
Image
*destination_image,
*source_image;
MagickBooleanType
status;
MagickOffsetType
progress;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
size_t
height;
#endif
ssize_t
y;
assert(source != (WandView *) NULL);
assert(source->signature == WandSignature);
if (transfer == (TransferWandViewMethod) NULL)
return(MagickFalse);
source_image=source->wand->images;
destination_image=destination->wand->images;
if (SetImageStorageClass(destination_image,DirectClass) == MagickFalse)
return(MagickFalse);
status=MagickTrue;
progress=0;
exception=destination->exception;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
height=(size_t) (source->extent.height-source->extent.y);
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,destination_image,height,1)
#endif
for (y=source->extent.y; y < (ssize_t) source->extent.height; y++)
{
const int
id = GetOpenMPThreadId();
MagickBooleanType
sync;
register const IndexPacket
*magick_restrict indexes;
register const PixelPacket
*magick_restrict pixels;
register IndexPacket
*magick_restrict destination_indexes;
register ssize_t
x;
register PixelPacket
*magick_restrict destination_pixels;
if (status == MagickFalse)
continue;
pixels=GetCacheViewVirtualPixels(source->view,source->extent.x,y,
source->extent.width,1,source->exception);
if (pixels == (const PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewVirtualIndexQueue(source->view);
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetQuantumColor(source->pixel_wands[id][x],pixels+x);
if (source_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetBlackQuantum(source->pixel_wands[id][x],
GetPixelBlack(indexes+x));
if (source_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetIndex(source->pixel_wands[id][x],
GetPixelIndex(indexes+x));
destination_pixels=GetCacheViewAuthenticPixels(destination->view,
destination->extent.x,y,destination->extent.width,1,exception);
if (destination_pixels == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
destination_indexes=GetCacheViewAuthenticIndexQueue(destination->view);
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetQuantumColor(destination->pixel_wands[id][x],pixels+x);
if (destination_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetBlackQuantum(destination->pixel_wands[id][x],
GetPixelBlack(indexes+x));
if (destination_image->storage_class == PseudoClass)
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelSetIndex(destination->pixel_wands[id][x],
GetPixelIndex(indexes+x));
if (transfer(source,destination,y,id,context) == MagickFalse)
status=MagickFalse;
for (x=0; x < (ssize_t) destination->extent.width; x++)
PixelGetQuantumColor(destination->pixel_wands[id][x],
destination_pixels+x);
if (destination_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) destination->extent.width; x++)
SetPixelBlack(destination_indexes+x,PixelGetBlackQuantum(
destination->pixel_wands[id][x]));
sync=SyncCacheViewAuthenticPixels(destination->view,exception);
if (sync == MagickFalse)
{
InheritException(destination->exception,GetCacheViewException(
source->view));
status=MagickFalse;
}
if (source_image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickWand_TransferWandViewIterator)
#endif
proceed=SetImageProgress(source_image,source->description,progress++,
source->extent.height);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% U p d a t e W a n d V i e w I t e r a t o r %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% UpdateWandViewIterator() iterates over the wand view in parallel and calls
% your update method for each scanline of the view. The pixel extent is
% confined to the image canvas-- that is no negative offsets or widths or
% heights that exceed the image dimension are permitted. Updates to pixels
% in your callback are automagically synced back to the image.
%
% The callback signature is:
%
% MagickBooleanType UpdateImageViewMethod(WandView *source,const ssize_t y,
% const int thread_id,void *context)
%
% Use this pragma if the view is not single threaded:
%
% #pragma omp critical
%
% to define a section of code in your callback update method that must be
% executed by a single thread at a time.
%
% The format of the UpdateWandViewIterator method is:
%
% MagickBooleanType UpdateWandViewIterator(WandView *source,
% UpdateWandViewMethod update,void *context)
%
% A description of each parameter follows:
%
% o source: the source wand view.
%
% o update: the update callback method.
%
% o context: the user defined context.
%
*/
WandExport MagickBooleanType UpdateWandViewIterator(WandView *source,
UpdateWandViewMethod update,void *context)
{
ExceptionInfo
*exception;
Image
*source_image;
MagickBooleanType
status;
MagickOffsetType
progress;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
size_t
height;
#endif
ssize_t
y;
assert(source != (WandView *) NULL);
assert(source->signature == WandSignature);
if (update == (UpdateWandViewMethod) NULL)
return(MagickFalse);
source_image=source->wand->images;
if (SetImageStorageClass(source_image,DirectClass) == MagickFalse)
return(MagickFalse);
status=MagickTrue;
progress=0;
exception=source->exception;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
height=(size_t) (source->extent.height-source->extent.y);
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,source_image,height,1)
#endif
for (y=source->extent.y; y < (ssize_t) source->extent.height; y++)
{
const int
id = GetOpenMPThreadId();
register IndexPacket
*magick_restrict indexes;
register ssize_t
x;
register PixelPacket
*magick_restrict pixels;
if (status == MagickFalse)
continue;
pixels=GetCacheViewAuthenticPixels(source->view,source->extent.x,y,
source->extent.width,1,exception);
if (pixels == (PixelPacket *) NULL)
{
InheritException(source->exception,GetCacheViewException(
source->view));
status=MagickFalse;
continue;
}
indexes=GetCacheViewAuthenticIndexQueue(source->view);
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetQuantumColor(source->pixel_wands[id][x],pixels+x);
if (source_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelSetBlackQuantum(source->pixel_wands[id][x],
GetPixelBlack(indexes+x));
if (update(source,y,id,context) == MagickFalse)
status=MagickFalse;
for (x=0; x < (ssize_t) source->extent.width; x++)
PixelGetQuantumColor(source->pixel_wands[id][x],pixels+x);
if (source_image->colorspace == CMYKColorspace)
for (x=0; x < (ssize_t) source->extent.width; x++)
SetPixelBlack(indexes+x,PixelGetBlackQuantum(
source->pixel_wands[id][x]));
if (SyncCacheViewAuthenticPixels(source->view,exception) == MagickFalse)
{
InheritException(source->exception,GetCacheViewException(source->view));
status=MagickFalse;
}
if (source_image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickWand_UpdateWandViewIterator)
#endif
proceed=SetImageProgress(source_image,source->description,progress++,
source->extent.height);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
return(status);
}
|
main.c | #include "rsbench.h"
int main(int argc, char * argv[])
{
// =====================================================================
// Initialization & Command Line Read-In
// =====================================================================
int version = 2;
int max_procs = omp_get_num_procs();
double start, stop;
srand(time(NULL));
// Process CLI Fields
Input input = read_CLI( argc, argv );
// Set number of OpenMP Threads
omp_set_num_threads(input.nthreads);
// =====================================================================
// Print-out of Input Summary
// =====================================================================
logo(version);
center_print("INPUT SUMMARY", 79);
border_print();
print_input_summary(input);
// =====================================================================
// Prepare Pole Paremeter Grids
// =====================================================================
border_print();
center_print("INITIALIZATION", 79);
border_print();
start = omp_get_wtime();
// Allocate & fill energy grids
printf("Generating resonance distributions...\n");
int * n_poles = generate_n_poles( input );
// Allocate & fill Window grids
printf("Generating window distributions...\n");
int * n_windows = generate_n_windows( input );
// Get material data
printf("Loading Hoogenboom-Martin material data...\n");
Materials materials = get_materials( input );
// Prepare full resonance grid
printf("Generating resonance parameter grid...\n");
Pole ** poles = generate_poles( input, n_poles );
// Prepare full Window grid
printf("Generating window parameter grid...\n");
Window ** windows = generate_window_params( input, n_windows, n_poles);
// Prepare 0K Resonances
printf("Generating 0K l_value data...\n");
double ** pseudo_K0RS = generate_pseudo_K0RS( input );
CalcDataPtrs data;
data.n_poles = n_poles;
data.n_windows = n_windows;
data.materials = materials;
data.poles = poles;
data.windows = windows;
data.pseudo_K0RS = pseudo_K0RS;
stop = omp_get_wtime();
printf("Initialization Complete. (%.2lf seconds)\n", stop-start);
// =====================================================================
// Cross Section (XS) Parallel Lookup Simulation Begins
// =====================================================================
border_print();
center_print("SIMULATION", 79);
border_print();
printf("Beginning Simulation.\n");
#ifndef STATUS
printf("Calculating XS's...\n");
#endif
#ifdef PAPI
/* initialize papi with one thread here */
if ( PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT){
fprintf(stderr, "PAPI library init error!\n");
exit(1);
}
#endif
start = omp_get_wtime();
unsigned long seed = rand();
int mat;
double E;
int i;
#pragma omp parallel default(none) \
private(seed, mat, E, i) \
shared(input, data)
{
double macro_xs[4];
int thread = omp_get_thread_num();
seed += thread;
#ifdef PAPI
int eventset = PAPI_NULL;
int num_papi_events;
#pragma omp critical
{
counter_init(&eventset, &num_papi_events);
}
#endif
complex double * sigTfactors =
(complex double *) malloc( input.numL * sizeof(complex double) );
#pragma omp for schedule(dynamic)
for( i = 0; i < input.lookups; i++ )
{
#ifdef STATUS
if( thread == 0 && i % 1000 == 0 )
printf("\rCalculating XS's... (%.0lf%% completed)",
(i / ( (double)input.lookups /
(double) input.nthreads )) /
(double) input.nthreads * 100.0);
#endif
mat = pick_mat( &seed );
E = rn( &seed );
calculate_macro_xs( macro_xs, mat, E, input, data, sigTfactors );
}
free(sigTfactors);
#ifdef PAPI
if( thread == 0 )
{
printf("\n");
border_print();
center_print("PAPI COUNTER RESULTS", 79);
border_print();
printf("Count \tSmybol \tDescription\n");
}
{
#pragma omp barrier
}
counter_stop(&eventset, num_papi_events);
#endif
}
stop = omp_get_wtime();
#ifndef PAPI
printf("\nSimulation Complete.\n");
#endif
// =====================================================================
// Print / Save Results and Exit
// =====================================================================
border_print();
center_print("RESULTS", 79);
border_print();
printf("Threads: %d\n", input.nthreads);
printf("Runtime: %.3lf seconds\n", stop-start);
printf("Lookups: "); fancy_int(input.lookups);
printf("Lookups/s: "); fancy_int((double) input.lookups / (stop-start));
border_print();
return 0;
}
|
GB_binop__rdiv_int64.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2020, All Rights Reserved.
// http://suitesparse.com See GraphBLAS/Doc/License.txt for license.
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_mkl.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB_AaddB__rdiv_int64
// A.*B function (eWiseMult): GB_AemultB__rdiv_int64
// A*D function (colscale): GB_AxD__rdiv_int64
// D*A function (rowscale): GB_DxB__rdiv_int64
// C+=B function (dense accum): GB_Cdense_accumB__rdiv_int64
// C+=b function (dense accum): GB_Cdense_accumb__rdiv_int64
// C+=A+B function (dense ewise3): GB_Cdense_ewise3_accum__rdiv_int64
// C=A+B function (dense ewise3): GB_Cdense_ewise3_noaccum__rdiv_int64
// C=scalar+B GB_bind1st__rdiv_int64
// C=scalar+B' GB_bind1st_tran__rdiv_int64
// C=A+scalar GB_bind2nd__rdiv_int64
// C=A'+scalar GB_bind2nd_tran__rdiv_int64
// C type: int64_t
// A type: int64_t
// B,b type: int64_t
// BinaryOp: cij = GB_IDIV_SIGNED (bij, aij, 64)
#define GB_ATYPE \
int64_t
#define GB_BTYPE \
int64_t
#define GB_CTYPE \
int64_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int64_t aij = Ax [pA]
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB) \
int64_t bij = Bx [pB]
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int64_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA) \
cij = Ax [pA]
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB) \
cij = Bx [pB]
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z, x, y) \
z = GB_IDIV_SIGNED (y, x, 64) ;
// op is second
#define GB_OP_IS_SECOND \
0
// op is plus_fp32 or plus_fp64
#define GB_OP_IS_PLUS_REAL \
0
// op is minus_fp32 or minus_fp64
#define GB_OP_IS_MINUS_REAL \
0
// GB_cblas_*axpy gateway routine, if it exists for this operator and type:
#define GB_CBLAS_AXPY \
(none)
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_RDIV || GxB_NO_INT64 || GxB_NO_RDIV_INT64)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB_Cdense_ewise3_accum__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_ewise3_noaccum__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumB__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumb__rdiv_int64
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int64_t
int64_t bwork = (*((int64_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_AxD__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *GB_RESTRICT Cx = (int64_t *) C->x ;
#include "GB_AxB_colscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_DxB__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *GB_RESTRICT Cx = (int64_t *) C->x ;
#include "GB_AxB_rowscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
GrB_Info GB_AaddB__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_add_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB_AemultB__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB_bind1st__rdiv_int64
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *Cx = (int64_t *) Cx_output ;
int64_t x = (*((int64_t *) x_input)) ;
int64_t *Bx = (int64_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
int64_t bij = Bx [p] ;
Cx [p] = GB_IDIV_SIGNED (bij, x, 64) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB_bind2nd__rdiv_int64
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int64_t *Cx = (int64_t *) Cx_output ;
int64_t *Ax = (int64_t *) Ax_input ;
int64_t y = (*((int64_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
int64_t aij = Ax [p] ;
Cx [p] = GB_IDIV_SIGNED (y, aij, 64) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typcasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int64_t aij = Ax [pA] ; \
Cx [pC] = GB_IDIV_SIGNED (aij, x, 64) ; \
}
GrB_Info GB_bind1st_tran__rdiv_int64
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *GB_RESTRICT *Rowcounts,
GBI_single_iterator Iter,
const int64_t *GB_RESTRICT A_slice,
int naslice
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int64_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t x = (*((const int64_t *) x_input)) ;
#define GB_PHASE_2_OF_2
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int64_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typcasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int64_t aij = Ax [pA] ; \
Cx [pC] = GB_IDIV_SIGNED (y, aij, 64) ; \
}
GrB_Info GB_bind2nd_tran__rdiv_int64
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *GB_RESTRICT *Rowcounts,
GBI_single_iterator Iter,
const int64_t *GB_RESTRICT A_slice,
int naslice
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t y = (*((const int64_t *) y_input)) ;
#define GB_PHASE_2_OF_2
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
polybench.c | /**
* polybench.c: This file is part of the PolyBench/C 3.2 test suite.
*
*
* Contact: Louis-Noel Pouchet <pouchet@cse.ohio-state.edu>
* Web address: http://polybench.sourceforge.net
* License: /LICENSE.OSU.txt
*/
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <assert.h>
#include <time.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sched.h>
#include <math.h>
#ifdef _OPENMP
# include <omp.h>
#endif
/* By default, collect PAPI counters on thread 0. */
#ifndef POLYBENCH_THREAD_MONITOR
# define POLYBENCH_THREAD_MONITOR 0
#endif
/* Total LLC cache size. By default 32+MB.. */
#ifndef POLYBENCH_CACHE_SIZE_KB
# define POLYBENCH_CACHE_SIZE_KB 32770
#endif
int polybench_papi_counters_threadid = POLYBENCH_THREAD_MONITOR;
double polybench_program_total_flops = 0;
#ifdef POLYBENCH_PAPI
# include <papi.h>
# define POLYBENCH_MAX_NB_PAPI_COUNTERS 96
char* _polybench_papi_eventlist[] = {
#include "papi_counters.list"
NULL
};
int polybench_papi_eventset;
int polybench_papi_eventlist[POLYBENCH_MAX_NB_PAPI_COUNTERS];
long_long polybench_papi_values[POLYBENCH_MAX_NB_PAPI_COUNTERS];
#endif
/* Timer code (gettimeofday). */
double polybench_t_start, polybench_t_end;
/* Timer code (RDTSC). */
unsigned long long int polybench_c_start, polybench_c_end;
static
double rtclock()
{
#ifdef POLYBENCH_TIME
struct timeval Tp;
int stat;
stat = gettimeofday (&Tp, NULL);
if (stat != 0)
printf ("Error return from gettimeofday: %d", stat);
return (Tp.tv_sec + Tp.tv_usec * 1.0e-6);
#else
return 0;
#endif
}
#ifdef POLYBENCH_CYCLE_ACCURATE_TIMER
static
unsigned long long int rdtsc()
{
unsigned long long int ret = 0;
unsigned int cycles_lo;
unsigned int cycles_hi;
__asm__ volatile ("RDTSC" : "=a" (cycles_lo), "=d" (cycles_hi));
ret = (unsigned long long int)cycles_hi << 32 | cycles_lo;
return ret;
}
#endif
void polybench_flush_cache()
{
int cs = POLYBENCH_CACHE_SIZE_KB * 1024 / sizeof(double);
double* flush = (double*) calloc (cs, sizeof(double));
int i;
double tmp = 0.0;
#ifdef _OPENMP
#pragma omp parallel for reduction(+:tmp)
#endif
for (i = 0; i < cs; i++)
tmp += flush[i];
assert (tmp <= 10.0);
free (flush);
}
#ifdef POLYBENCH_LINUX_FIFO_SCHEDULER
void polybench_linux_fifo_scheduler()
{
/* Use FIFO scheduler to limit OS interference. Program must be run
as root, and this works only for Linux kernels. */
struct sched_param schedParam;
schedParam.sched_priority = sched_get_priority_max (SCHED_FIFO);
sched_setscheduler (0, SCHED_FIFO, &schedParam);
}
void polybench_linux_standard_scheduler()
{
/* Restore to standard scheduler policy. */
struct sched_param schedParam;
schedParam.sched_priority = sched_get_priority_max (SCHED_OTHER);
sched_setscheduler (0, SCHED_OTHER, &schedParam);
}
#endif
#ifdef POLYBENCH_PAPI
static
void test_fail(char *file, int line, char *call, int retval)
{
char buf[128];
memset(buf, '\0', sizeof(buf));
if (retval != 0)
fprintf (stdout,"%-40s FAILED\nLine # %d\n", file, line);
else
{
fprintf (stdout,"%-40s SKIPPED\n", file);
fprintf (stdout,"Line # %d\n", line);
}
if (retval == PAPI_ESYS)
{
sprintf (buf, "System error in %s", call);
perror (buf);
}
else if (retval > 0)
fprintf (stdout,"Error: %s\n", call);
else if (retval == 0)
fprintf (stdout,"Error: %s\n", call);
else
{
char errstring[PAPI_MAX_STR_LEN];
PAPI_perror (retval, errstring, PAPI_MAX_STR_LEN);
fprintf (stdout,"Error in %s: %s\n", call, errstring);
}
fprintf (stdout,"\n");
if (PAPI_is_initialized ())
PAPI_shutdown ();
exit (1);
}
void polybench_papi_init()
{
# ifdef _OPENMP
#pragma omp parallel
{
#pragma omp master
{
if (omp_get_max_threads () < polybench_papi_counters_threadid)
polybench_papi_counters_threadid = omp_get_max_threads () - 1;
}
#pragma omp barrier
if (omp_get_thread_num () == polybench_papi_counters_threadid)
{
# endif
int retval;
polybench_papi_eventset = PAPI_NULL;
if ((retval = PAPI_library_init (PAPI_VER_CURRENT)) != PAPI_VER_CURRENT)
test_fail (__FILE__, __LINE__, "PAPI_library_init", retval);
if ((retval = PAPI_create_eventset (&polybench_papi_eventset))
!= PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_create_eventset", retval);
int k;
for (k = 0; _polybench_papi_eventlist[k]; ++k)
{
if ((retval =
PAPI_event_name_to_code (_polybench_papi_eventlist[k],
&(polybench_papi_eventlist[k])))
!= PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_event_name_to_code", retval);
}
polybench_papi_eventlist[k] = 0;
# ifdef _OPENMP
}
}
#pragma omp barrier
# endif
}
void polybench_papi_close()
{
# ifdef _OPENMP
#pragma omp parallel
{
if (omp_get_thread_num () == polybench_papi_counters_threadid)
{
# endif
int retval;
if ((retval = PAPI_destroy_eventset (&polybench_papi_eventset))
!= PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_destroy_eventset", retval);
if (PAPI_is_initialized ())
PAPI_shutdown ();
# ifdef _OPENMP
}
}
#pragma omp barrier
# endif
}
int polybench_papi_start_counter(int evid)
{
# ifndef POLYBENCH_NO_FLUSH_CACHE
polybench_flush_cache();
# endif
# ifdef _OPENMP
# pragma omp parallel
{
if (omp_get_thread_num () == polybench_papi_counters_threadid)
{
# endif
int retval = 1;
char descr[PAPI_MAX_STR_LEN];
PAPI_event_info_t evinfo;
PAPI_event_code_to_name (polybench_papi_eventlist[evid], descr);
if (PAPI_add_event (polybench_papi_eventset,
polybench_papi_eventlist[evid]) != PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_add_event", 1);
if (PAPI_get_event_info (polybench_papi_eventlist[evid], &evinfo)
!= PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_get_event_info", retval);
if ((retval = PAPI_start (polybench_papi_eventset)) != PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_start", retval);
# ifdef _OPENMP
}
}
#pragma omp barrier
# endif
return 0;
}
void polybench_papi_stop_counter(int evid)
{
# ifdef _OPENMP
# pragma omp parallel
{
if (omp_get_thread_num () == polybench_papi_counters_threadid)
{
# endif
int retval;
long_long values[1];
values[0] = 0;
if ((retval = PAPI_read (polybench_papi_eventset, &values[0]))
!= PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_read", retval);
if ((retval = PAPI_stop (polybench_papi_eventset, NULL)) != PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_stop", retval);
polybench_papi_values[evid] = values[0];
if ((retval = PAPI_remove_event
(polybench_papi_eventset,
polybench_papi_eventlist[evid])) != PAPI_OK)
test_fail (__FILE__, __LINE__, "PAPI_remove_event", retval);
# ifdef _OPENMP
}
}
#pragma omp barrier
# endif
}
void polybench_papi_print()
{
int verbose = 0;
# ifdef _OPENMP
# pragma omp parallel
{
if (omp_get_thread_num() == polybench_papi_counters_threadid)
{
#ifdef POLYBENCH_PAPI_VERBOSE
verbose = 1;
#endif
if (verbose)
printf ("On thread %d:\n", polybench_papi_counters_threadid);
#endif
int evid;
for (evid = 0; polybench_papi_eventlist[evid] != 0; ++evid)
{
if (verbose)
printf ("%s=", _polybench_papi_eventlist[evid]);
printf ("%llu ", polybench_papi_values[evid]);
if (verbose)
printf ("\n");
}
printf ("\n");
# ifdef _OPENMP
}
}
#pragma omp barrier
# endif
}
#endif
/* ! POLYBENCH_PAPI */
void polybench_prepare_instruments()
{
#ifndef POLYBENCH_NO_FLUSH_CACHE
polybench_flush_cache ();
#endif
#ifdef POLYBENCH_LINUX_FIFO_SCHEDULER
polybench_linux_fifo_scheduler ();
#endif
}
void polybench_timer_start()
{
polybench_prepare_instruments ();
#ifndef POLYBENCH_CYCLE_ACCURATE_TIMER
polybench_t_start = rtclock ();
#else
polybench_c_start = rdtsc ();
#endif
}
void polybench_timer_stop()
{
#ifndef POLYBENCH_CYCLE_ACCURATE_TIMER
polybench_t_end = rtclock ();
#else
polybench_c_end = rdtsc ();
#endif
#ifdef POLYBENCH_LINUX_FIFO_SCHEDULER
polybench_linux_standard_scheduler ();
#endif
}
void polybench_timer_print()
{
#ifdef POLYBENCH_GFLOPS
if (__polybench_program_total_flops == 0)
{
printf ("[PolyBench][WARNING] Program flops not defined, use polybench_set_program_flops(value)\n");
printf ("%0.6lf\n", polybench_t_end - polybench_t_start);
}
else
printf ("%0.2lf\n",
(__polybench_program_total_flops /
(double)(polybench_t_end - polybench_t_start)) / 1000000000);
#else
# ifndef POLYBENCH_CYCLE_ACCURATE_TIMER
printf ("%0.6f\n", polybench_t_end - polybench_t_start);
# else
printf ("%Ld\n", polybench_c_end - polybench_c_start);
# endif
#endif
}
static
void *
xmalloc (size_t num)
{
void* nnew = NULL;
int ret = posix_memalign (&nnew, 32, num);
if (! nnew || ret)
{
fprintf (stderr, "[PolyBench] posix_memalign: cannot allocate memory");
exit (1);
}
return nnew;
}
void* polybench_alloc_data(unsigned long long int n, int elt_size)
{
/// FIXME: detect overflow!
size_t val = n;
val *= elt_size;
void* ret = xmalloc (val);
return ret;
}
|
target-26.c | extern void abort (void);
#pragma omp declare target
int a[4] = { 2, 3, 4, 5 }, *b;
#pragma omp end declare target
int
main ()
{
int err;
int c[3] = { 6, 7, 8 };
b = c;
#pragma omp target map(to: a[0:2], b[0:2]) map(from: err)
err = a[0] != 2 || a[1] != 3 || a[2] != 4 || a[3] != 5 || b[0] != 6 || b[1] != 7;
if (err)
abort ();
a[1] = 9;
a[2] = 10;
#pragma omp target map(always,to:a[1:2]) map(from: err)
err = a[0] != 2 || a[1] != 9 || a[2] != 10 || a[3] != 5;
if (err)
abort ();
#pragma omp parallel firstprivate(a, b, c, err) num_threads (2)
#pragma omp single
{
b = c + 1;
a[0] = 11;
a[2] = 13;
c[1] = 14;
int d = 0;
#pragma omp target map(to: a[0:3], b[d:2]) map (from: err)
err = a[0] != 11 || a[1] != 9 || a[2] != 13 || b[0] != 14 || b[1] != 8;
if (err)
abort ();
}
return 0;
}
|
omp_dotprod_hybrid.c | /*****************************************************************************
* FILE: omp_dotprod_hybrid.c
* DESCRIPTION:
* This simple program is the hybrid version of a dot product and the fourth
* of four codes used to show the progression from a serial program to a
* hybrid MPI/OpenMP program. The relevant codes are:
* - omp_dotprod_serial.c - Serial version
* - omp_dotprod_openmp.c - OpenMP only version
* - omp_dotprod_mpi.c - MPI only version
* - omp_dotprod_hybrid.c - Hybrid MPI and OpenMP version
* SOURCE: Blaise Barney
* LAST REVISED: 06/02/17 Blaise Barney
******************************************************************************/
#include <mpi.h>
#include <omp.h>
#include <stdio.h>
#include <stdlib.h>
/* Define length of dot product vectors and number of OpenMP threads */
#define VECLEN 100
#define NUMTHREADS 8
int main (int argc, char* argv[])
{
int i, myid, tid, numprocs, len=VECLEN, threads=NUMTHREADS;
double *a, *b;
double mysum, allsum, sum, psum;
/* MPI Initialization */
MPI_Init (&argc, &argv);
MPI_Comm_size (MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank (MPI_COMM_WORLD, &myid);
/*
Each MPI task uses OpenMP to perform the dot product, obtains its partial sum,
and then calls MPI_Reduce to obtain the global sum.
*/
if (myid == 0)
printf("Starting omp_dotprod_hybrid. Using %d tasks...\n",numprocs);
/* Assign storage for dot product vectors */
a = (double*) malloc (len*threads*sizeof(double));
b = (double*) malloc (len*threads*sizeof(double));
/* Initialize dot product vectors */
for (i=0; i<len*threads; i++) {
a[i]=1.0;
b[i]=a[i];
}
/*
Perform the dot product in an OpenMP parallel region for loop with a sum reduction
For illustration purposes:
- Explicitly sets number of threads
- Gets and prints number of threads used
- Each thread keeps track of its partial sum
*/
/* Initialize OpenMP reduction sum */
sum = 0.0;
#pragma omp parallel private(i,tid,psum) num_threads(threads)
{
psum = 0.0;
tid = omp_get_thread_num();
if (tid ==0) {
threads = omp_get_num_threads();
printf("Task %d using %d threads\n",myid, threads);
}
#pragma omp for reduction(+:sum)
for (i=0; i<len*threads; i++) {
sum += (a[i] * b[i]);
psum = sum;
}
printf("Task %d thread %d partial sum = %f\n",myid, tid, psum);
}
/* Print this task's partial sum */
mysum = sum;
printf("Task %d partial sum = %f\n",myid, mysum);
/* After the dot product, perform a summation of results on each node */
MPI_Reduce (&mysum, &allsum, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
if (myid == 0)
printf ("Done. Hybrid version: global sum = %f \n", allsum);
free (a);
free (b);
MPI_Finalize();
}
|
convolution_sgemm_pack4_bf16s.h | // Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2021 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
static void im2col_sgemm_pack4_bf16s_neon(const Mat& bottom_im2col, Mat& top_blob, const Mat& kernel, const Mat& _bias, const Option& opt)
{
// Mat bottom_im2col(size, maxk, inch, 8u, 4, opt.workspace_allocator);
const int size = bottom_im2col.w;
const int maxk = bottom_im2col.h;
const int inch = bottom_im2col.c;
const int outch = top_blob.c;
const float* bias = _bias;
// permute
Mat tmp;
#if __aarch64__
if (size >= 12)
tmp.create(12 * maxk, inch, size / 12 + (size % 12) / 8 + (size % 12 % 8) / 4 + (size % 12 % 4) / 2 + size % 12 % 2, 8u, 4, opt.workspace_allocator);
else if (size >= 8)
tmp.create(8 * maxk, inch, size / 8 + (size % 8) / 4 + (size % 4) / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else if (size >= 4)
tmp.create(4 * maxk, inch, size / 4 + (size % 4) / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else if (size >= 2)
tmp.create(2 * maxk, inch, size / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else
tmp.create(maxk, inch, size, 8u, 4, opt.workspace_allocator);
#else
if (size >= 8)
tmp.create(8 * maxk, inch, size / 8 + (size % 8) / 4 + (size % 4) / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else if (size >= 4)
tmp.create(4 * maxk, inch, size / 4 + (size % 4) / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else if (size >= 2)
tmp.create(2 * maxk, inch, size / 2 + size % 2, 8u, 4, opt.workspace_allocator);
else
tmp.create(maxk, inch, size, 8u, 4, opt.workspace_allocator);
#endif
{
#if __aarch64__
int nn_size = size / 12;
int remain_size_start = 0;
#pragma omp parallel for num_threads(opt.num_threads)
for (int ii = 0; ii < nn_size; ii++)
{
int i = remain_size_start + ii * 12;
unsigned short* tmpptr = tmp.channel(i / 12);
for (int q = 0; q < inch; q++)
{
const unsigned short* img0 = (const unsigned short*)bottom_im2col.channel(q) + i * 4;
for (int k = 0; k < maxk; k++)
{
asm volatile(
"prfm pldl1keep, [%0, #512] \n"
"ld4 {v0.8h, v1.8h, v2.8h, v3.8h}, [%0], #64 \n"
"ld4 {v4.4h, v5.4h, v6.4h, v7.4h}, [%0] \n"
"st1 {v0.8h}, [%1], #16 \n"
"st1 {v4.4h}, [%1], #8 \n"
"st1 {v1.8h}, [%1], #16 \n"
"st1 {v5.4h}, [%1], #8 \n"
"sub %0, %0, #64 \n"
"st1 {v2.8h}, [%1], #16 \n"
"st1 {v6.4h}, [%1], #8 \n"
"st1 {v3.8h}, [%1], #16 \n"
"st1 {v7.4h}, [%1], #8 \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7");
img0 += size * 4;
}
}
}
remain_size_start += nn_size * 12;
nn_size = (size - remain_size_start) >> 3;
#else
int nn_size = size >> 3;
int remain_size_start = 0;
#endif
#pragma omp parallel for num_threads(opt.num_threads)
for (int ii = 0; ii < nn_size; ii++)
{
int i = remain_size_start + ii * 8;
#if __aarch64__
unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8);
#else
unsigned short* tmpptr = tmp.channel(i / 8);
#endif
for (int q = 0; q < inch; q++)
{
const unsigned short* img0 = (const unsigned short*)bottom_im2col.channel(q) + i * 4;
for (int k = 0; k < maxk; k++)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%0, #512] \n"
"ld1 {v0.8h, v1.8h, v2.8h, v3.8h}, [%0] \n"
"st1 {v0.8h, v1.8h, v2.8h, v3.8h}, [%1], #64 \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "v0", "v1", "v2", "v3");
#else
asm volatile(
"pld [%0, #256] \n"
"vld4.u16 {d0-d3}, [%0]! \n"
"pld [%0, #256] \n"
"vld4.u16 {d4-d7}, [%0] \n"
"sub %0, %0, #32 \n"
"vst1.u16 {d0}, [%1 :64]! \n"
"vst1.u16 {d4}, [%1 :64]! \n"
"vst1.u16 {d1}, [%1 :64]! \n"
"vst1.u16 {d5}, [%1 :64]! \n"
"vst1.u16 {d2}, [%1 :64]! \n"
"vst1.u16 {d6}, [%1 :64]! \n"
"vst1.u16 {d3}, [%1 :64]! \n"
"vst1.u16 {d7}, [%1 :64]! \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "q0", "q1", "q2", "q3");
#endif // __aarch64__
img0 += size * 4;
}
}
}
remain_size_start += nn_size << 3;
nn_size = (size - remain_size_start) >> 2;
#pragma omp parallel for num_threads(opt.num_threads)
for (int ii = 0; ii < nn_size; ii++)
{
int i = remain_size_start + ii * 4;
#if __aarch64__
unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4);
#else
unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4);
#endif
for (int q = 0; q < inch; q++)
{
const unsigned short* img0 = (const unsigned short*)bottom_im2col.channel(q) + i * 4;
for (int k = 0; k < maxk; k++)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%0, #256] \n"
"ld1 {v0.8h, v1.8h}, [%0] \n"
"st1 {v0.8h, v1.8h}, [%1], #32 \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "v0", "v1");
#else
asm volatile(
"pld [%0, #256] \n"
"vld1.u16 {d0-d3}, [%0 :128] \n"
"vst1.u16 {d0-d3}, [%1 :128]! \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "q0", "q1");
#endif // __aarch64__
img0 += size * 4;
}
}
}
remain_size_start += nn_size << 2;
nn_size = (size - remain_size_start) >> 1;
#pragma omp parallel for num_threads(opt.num_threads)
for (int ii = 0; ii < nn_size; ii++)
{
int i = remain_size_start + ii * 2;
#if __aarch64__
unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2);
#else
unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4 + (i % 4) / 2);
#endif
for (int q = 0; q < inch; q++)
{
const unsigned short* img0 = (const unsigned short*)bottom_im2col.channel(q) + i * 4;
for (int k = 0; k < maxk; k++)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%0, #128] \n"
"ld1 {v0.8h}, [%0] \n"
"st1 {v0.8h}, [%1], #16 \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "v0");
#else
asm volatile(
"pld [%0, #128] \n"
"vld1.u16 {d0-d1}, [%0 :128] \n"
"vst1.u16 {d0-d1}, [%1 :128]! \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "q0");
#endif // __aarch64__
img0 += size * 4;
}
}
}
remain_size_start += nn_size << 1;
#pragma omp parallel for num_threads(opt.num_threads)
for (int i = remain_size_start; i < size; i++)
{
#if __aarch64__
unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2);
#else
unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4 + (i % 4) / 2 + i % 2);
#endif
for (int q = 0; q < inch; q++)
{
const unsigned short* img0 = (const unsigned short*)bottom_im2col.channel(q) + i * 4;
for (int k = 0; k < maxk; k++)
{
#if __aarch64__
asm volatile(
"prfm pldl1keep, [%0, #64] \n"
"ld1 {v0.4h}, [%0] \n"
"st1 {v0.4h}, [%1], #8 \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "v0");
#else
asm volatile(
"pld [%0, #64] \n"
"vld1.u16 {d0}, [%0 :64] \n"
"vst1.u16 {d0}, [%1 :64]! \n"
: "=r"(img0), // %0
"=r"(tmpptr) // %1
: "0"(img0),
"1"(tmpptr)
: "memory", "q0");
#endif // __aarch64__
img0 += size * 4;
}
}
}
}
int remain_outch_start = 0;
#if __aarch64__
int nn_outch = outch >> 1;
remain_outch_start = nn_outch << 1;
#pragma omp parallel for num_threads(opt.num_threads)
for (int pp = 0; pp < nn_outch; pp++)
{
int p = pp * 2;
unsigned short* outptr0 = top_blob.channel(p);
unsigned short* outptr1 = top_blob.channel(p + 1);
const float zeros[8] = {0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f, 0.f};
const float* biasptr = bias ? bias + p * 4 : zeros;
int i = 0;
for (; i + 11 < size; i += 12)
{
const unsigned short* tmpptr = tmp.channel(i / 12);
const unsigned short* kptr0 = kernel.channel(p / 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v0.4s, v1.4s}, [%10] \n"
"mov v8.16b, v0.16b \n"
"mov v9.16b, v0.16b \n"
"mov v10.16b, v0.16b \n"
"mov v11.16b, v0.16b \n"
"mov v12.16b, v0.16b \n"
"mov v13.16b, v0.16b \n"
"mov v14.16b, v0.16b \n"
"mov v15.16b, v0.16b \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"mov v20.16b, v1.16b \n"
"mov v21.16b, v1.16b \n"
"mov v22.16b, v1.16b \n"
"mov v23.16b, v1.16b \n"
"mov v24.16b, v1.16b \n"
"mov v25.16b, v1.16b \n"
"mov v26.16b, v1.16b \n"
"mov v27.16b, v1.16b \n"
"mov v28.16b, v1.16b \n"
"mov v29.16b, v1.16b \n"
"mov v30.16b, v1.16b \n"
"mov v31.16b, v1.16b \n"
"0: \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%3], #32 \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v4.4h, v5.4h, v6.4h, v7.4h}, [%4], #32 \n" // w0011_01
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v6.4s, v6.4h, #16 \n"
"shll v7.4s, v7.4h, #16 \n"
"fmla v8.4s, v4.4s, v0.s[0] \n"
"fmla v9.4s, v4.4s, v0.s[1] \n"
"fmla v10.4s, v4.4s, v0.s[2] \n"
"fmla v11.4s, v4.4s, v0.s[3] \n"
"fmla v12.4s, v4.4s, v1.s[0] \n"
"fmla v13.4s, v4.4s, v1.s[1] \n"
"fmla v14.4s, v4.4s, v1.s[2] \n"
"fmla v15.4s, v4.4s, v1.s[3] \n"
"fmla v16.4s, v4.4s, v2.s[0] \n"
"fmla v17.4s, v4.4s, v2.s[1] \n"
"fmla v18.4s, v4.4s, v2.s[2] \n"
"fmla v19.4s, v4.4s, v2.s[3] \n"
"fmla v20.4s, v5.4s, v0.s[0] \n"
"fmla v21.4s, v5.4s, v0.s[1] \n"
"fmla v22.4s, v5.4s, v0.s[2] \n"
"fmla v23.4s, v5.4s, v0.s[3] \n"
"fmla v24.4s, v5.4s, v1.s[0] \n"
"fmla v25.4s, v5.4s, v1.s[1] \n"
"fmla v26.4s, v5.4s, v1.s[2] \n"
"fmla v27.4s, v5.4s, v1.s[3] \n"
"fmla v28.4s, v5.4s, v2.s[0] \n"
"fmla v29.4s, v5.4s, v2.s[1] \n"
"fmla v30.4s, v5.4s, v2.s[2] \n"
"fmla v31.4s, v5.4s, v2.s[3] \n"
"fmla v8.4s, v6.4s, v3.s[0] \n"
"fmla v9.4s, v6.4s, v3.s[1] \n"
"fmla v10.4s, v6.4s, v3.s[2] \n"
"fmla v11.4s, v6.4s, v3.s[3] \n"
"fmla v20.4s, v7.4s, v3.s[0] \n"
"fmla v21.4s, v7.4s, v3.s[1] \n"
"fmla v22.4s, v7.4s, v3.s[2] \n"
"fmla v23.4s, v7.4s, v3.s[3] \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%3], #32 \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v12.4s, v6.4s, v0.s[0] \n"
"fmla v13.4s, v6.4s, v0.s[1] \n"
"fmla v14.4s, v6.4s, v0.s[2] \n"
"fmla v15.4s, v6.4s, v0.s[3] \n"
"fmla v16.4s, v6.4s, v1.s[0] \n"
"fmla v17.4s, v6.4s, v1.s[1] \n"
"fmla v18.4s, v6.4s, v1.s[2] \n"
"fmla v19.4s, v6.4s, v1.s[3] \n"
"fmla v24.4s, v7.4s, v0.s[0] \n"
"fmla v25.4s, v7.4s, v0.s[1] \n"
"fmla v26.4s, v7.4s, v0.s[2] \n"
"fmla v27.4s, v7.4s, v0.s[3] \n"
"fmla v28.4s, v7.4s, v1.s[0] \n"
"fmla v29.4s, v7.4s, v1.s[1] \n"
"fmla v30.4s, v7.4s, v1.s[2] \n"
"fmla v31.4s, v7.4s, v1.s[3] \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v4.4h, v5.4h, v6.4h, v7.4h}, [%4], #32 \n" // w2233_01
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v6.4s, v6.4h, #16 \n"
"shll v7.4s, v7.4h, #16 \n"
"fmla v8.4s, v4.4s, v2.s[0] \n"
"fmla v9.4s, v4.4s, v2.s[1] \n"
"fmla v10.4s, v4.4s, v2.s[2] \n"
"fmla v11.4s, v4.4s, v2.s[3] \n"
"fmla v12.4s, v4.4s, v3.s[0] \n"
"fmla v13.4s, v4.4s, v3.s[1] \n"
"fmla v14.4s, v4.4s, v3.s[2] \n"
"fmla v15.4s, v4.4s, v3.s[3] \n"
"fmla v20.4s, v5.4s, v2.s[0] \n"
"fmla v21.4s, v5.4s, v2.s[1] \n"
"fmla v22.4s, v5.4s, v2.s[2] \n"
"fmla v23.4s, v5.4s, v2.s[3] \n"
"fmla v24.4s, v5.4s, v3.s[0] \n"
"fmla v25.4s, v5.4s, v3.s[1] \n"
"fmla v26.4s, v5.4s, v3.s[2] \n"
"fmla v27.4s, v5.4s, v3.s[3] \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%3], #32 \n"
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"fmla v16.4s, v4.4s, v0.s[0] \n"
"fmla v17.4s, v4.4s, v0.s[1] \n"
"fmla v18.4s, v4.4s, v0.s[2] \n"
"fmla v19.4s, v4.4s, v0.s[3] \n"
"fmla v28.4s, v5.4s, v0.s[0] \n"
"fmla v29.4s, v5.4s, v0.s[1] \n"
"fmla v30.4s, v5.4s, v0.s[2] \n"
"fmla v31.4s, v5.4s, v0.s[3] \n"
"fmla v8.4s, v6.4s, v1.s[0] \n"
"fmla v9.4s, v6.4s, v1.s[1] \n"
"fmla v10.4s, v6.4s, v1.s[2] \n"
"fmla v11.4s, v6.4s, v1.s[3] \n"
"fmla v12.4s, v6.4s, v2.s[0] \n"
"fmla v13.4s, v6.4s, v2.s[1] \n"
"fmla v14.4s, v6.4s, v2.s[2] \n"
"fmla v15.4s, v6.4s, v2.s[3] \n"
"fmla v16.4s, v6.4s, v3.s[0] \n"
"fmla v17.4s, v6.4s, v3.s[1] \n"
"fmla v18.4s, v6.4s, v3.s[2] \n"
"fmla v19.4s, v6.4s, v3.s[3] \n"
"subs %w0, %w0, #1 \n"
"fmla v20.4s, v7.4s, v1.s[0] \n"
"fmla v21.4s, v7.4s, v1.s[1] \n"
"fmla v22.4s, v7.4s, v1.s[2] \n"
"fmla v23.4s, v7.4s, v1.s[3] \n"
"fmla v24.4s, v7.4s, v2.s[0] \n"
"fmla v25.4s, v7.4s, v2.s[1] \n"
"fmla v26.4s, v7.4s, v2.s[2] \n"
"fmla v27.4s, v7.4s, v2.s[3] \n"
"fmla v28.4s, v7.4s, v3.s[0] \n"
"fmla v29.4s, v7.4s, v3.s[1] \n"
"fmla v30.4s, v7.4s, v3.s[2] \n"
"fmla v31.4s, v7.4s, v3.s[3] \n"
"bne 0b \n"
"shrn v8.4h, v8.4s, #16 \n"
"shrn v9.4h, v9.4s, #16 \n"
"shrn v10.4h, v10.4s, #16 \n"
"shrn v11.4h, v11.4s, #16 \n"
"shrn v12.4h, v12.4s, #16 \n"
"shrn v13.4h, v13.4s, #16 \n"
"shrn v14.4h, v14.4s, #16 \n"
"shrn v15.4h, v15.4s, #16 \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"shrn v20.4h, v20.4s, #16 \n"
"shrn v21.4h, v21.4s, #16 \n"
"shrn v22.4h, v22.4s, #16 \n"
"shrn v23.4h, v23.4s, #16 \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"shrn v28.4h, v28.4s, #16 \n"
"shrn v29.4h, v29.4s, #16 \n"
"shrn v30.4h, v30.4s, #16 \n"
"shrn v31.4h, v31.4s, #16 \n"
"st1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%1], #32 \n"
"st1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%2], #32 \n"
"st1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%1], #32 \n"
"st1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%2], #32 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
"st1 {v28.4h, v29.4h, v30.4h, v31.4h}, [%2], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr1), // %2
"=r"(tmpptr), // %3
"=r"(kptr0) // %4
: "0"(nn),
"1"(outptr0),
"2"(outptr1),
"3"(tmpptr),
"4"(kptr0),
"r"(biasptr) // %10
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
for (; i + 7 < size; i += 8)
{
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8);
const unsigned short* kptr0 = kernel.channel(p / 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v0.4s, v1.4s}, [%10] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"mov v20.16b, v0.16b \n"
"mov v21.16b, v0.16b \n"
"mov v22.16b, v0.16b \n"
"mov v23.16b, v0.16b \n"
"mov v24.16b, v1.16b \n"
"mov v25.16b, v1.16b \n"
"mov v26.16b, v1.16b \n"
"mov v27.16b, v1.16b \n"
"mov v28.16b, v1.16b \n"
"mov v29.16b, v1.16b \n"
"mov v30.16b, v1.16b \n"
"mov v31.16b, v1.16b \n"
"0: \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%3], #32 \n" // r0 r1 r2 r3
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%4], #32 \n" // w0011_01
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v18.4s, v8.4s, v2.s[0] \n"
"fmla v19.4s, v8.4s, v3.s[0] \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v4.4h, v5.4h, v6.4h, v7.4h}, [%3], #32 \n" // r4 r5 r6 r7
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v6.4s, v6.4h, #16 \n"
"shll v7.4s, v7.4h, #16 \n"
"fmla v20.4s, v8.4s, v4.s[0] \n"
"fmla v21.4s, v8.4s, v5.s[0] \n"
"fmla v22.4s, v8.4s, v6.s[0] \n"
"fmla v23.4s, v8.4s, v7.s[0] \n"
"fmla v24.4s, v9.4s, v0.s[0] \n"
"fmla v25.4s, v9.4s, v1.s[0] \n"
"fmla v26.4s, v9.4s, v2.s[0] \n"
"fmla v27.4s, v9.4s, v3.s[0] \n"
"fmla v28.4s, v9.4s, v4.s[0] \n"
"fmla v29.4s, v9.4s, v5.s[0] \n"
"fmla v30.4s, v9.4s, v6.s[0] \n"
"fmla v31.4s, v9.4s, v7.s[0] \n"
"fmla v16.4s, v10.4s, v0.s[1] \n"
"fmla v17.4s, v10.4s, v1.s[1] \n"
"fmla v18.4s, v10.4s, v2.s[1] \n"
"fmla v19.4s, v10.4s, v3.s[1] \n"
"fmla v20.4s, v10.4s, v4.s[1] \n"
"fmla v21.4s, v10.4s, v5.s[1] \n"
"fmla v22.4s, v10.4s, v6.s[1] \n"
"fmla v23.4s, v10.4s, v7.s[1] \n"
"fmla v24.4s, v11.4s, v0.s[1] \n"
"fmla v25.4s, v11.4s, v1.s[1] \n"
"fmla v26.4s, v11.4s, v2.s[1] \n"
"fmla v27.4s, v11.4s, v3.s[1] \n"
"fmla v28.4s, v11.4s, v4.s[1] \n"
"fmla v29.4s, v11.4s, v5.s[1] \n"
"fmla v30.4s, v11.4s, v6.s[1] \n"
"fmla v31.4s, v11.4s, v7.s[1] \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%4], #32 \n" // w2233_01
"shll v12.4s, v12.4h, #16 \n"
"shll v13.4s, v13.4h, #16 \n"
"shll v14.4s, v14.4h, #16 \n"
"shll v15.4s, v15.4h, #16 \n"
"fmla v16.4s, v12.4s, v0.s[2] \n"
"fmla v17.4s, v12.4s, v1.s[2] \n"
"fmla v18.4s, v12.4s, v2.s[2] \n"
"fmla v19.4s, v12.4s, v3.s[2] \n"
"fmla v20.4s, v12.4s, v4.s[2] \n"
"fmla v21.4s, v12.4s, v5.s[2] \n"
"fmla v22.4s, v12.4s, v6.s[2] \n"
"fmla v23.4s, v12.4s, v7.s[2] \n"
"fmla v24.4s, v13.4s, v0.s[2] \n"
"fmla v25.4s, v13.4s, v1.s[2] \n"
"fmla v26.4s, v13.4s, v2.s[2] \n"
"fmla v27.4s, v13.4s, v3.s[2] \n"
"fmla v28.4s, v13.4s, v4.s[2] \n"
"fmla v29.4s, v13.4s, v5.s[2] \n"
"fmla v30.4s, v13.4s, v6.s[2] \n"
"fmla v31.4s, v13.4s, v7.s[2] \n"
"fmla v16.4s, v14.4s, v0.s[3] \n"
"fmla v17.4s, v14.4s, v1.s[3] \n"
"fmla v18.4s, v14.4s, v2.s[3] \n"
"fmla v19.4s, v14.4s, v3.s[3] \n"
"fmla v20.4s, v14.4s, v4.s[3] \n"
"fmla v21.4s, v14.4s, v5.s[3] \n"
"fmla v22.4s, v14.4s, v6.s[3] \n"
"fmla v23.4s, v14.4s, v7.s[3] \n"
"subs %w0, %w0, #1 \n"
"fmla v24.4s, v15.4s, v0.s[3] \n"
"fmla v25.4s, v15.4s, v1.s[3] \n"
"fmla v26.4s, v15.4s, v2.s[3] \n"
"fmla v27.4s, v15.4s, v3.s[3] \n"
"fmla v28.4s, v15.4s, v4.s[3] \n"
"fmla v29.4s, v15.4s, v5.s[3] \n"
"fmla v30.4s, v15.4s, v6.s[3] \n"
"fmla v31.4s, v15.4s, v7.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"shrn v20.4h, v20.4s, #16 \n"
"shrn v21.4h, v21.4s, #16 \n"
"shrn v22.4h, v22.4s, #16 \n"
"shrn v23.4h, v23.4s, #16 \n"
"shrn v24.4h, v24.4s, #16 \n"
"shrn v25.4h, v25.4s, #16 \n"
"shrn v26.4h, v26.4s, #16 \n"
"shrn v27.4h, v27.4s, #16 \n"
"shrn v28.4h, v28.4s, #16 \n"
"shrn v29.4h, v29.4s, #16 \n"
"shrn v30.4h, v30.4s, #16 \n"
"shrn v31.4h, v31.4s, #16 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
"st1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%2], #32 \n"
"st1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%1], #32 \n"
"st1 {v28.4h, v29.4h, v30.4h, v31.4h}, [%2], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr1), // %2
"=r"(tmpptr), // %3
"=r"(kptr0) // %4
: "0"(nn),
"1"(outptr0),
"2"(outptr1),
"3"(tmpptr),
"4"(kptr0),
"r"(biasptr) // %10
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31");
}
for (; i + 3 < size; i += 4)
{
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4);
const unsigned short* kptr0 = kernel.channel(p / 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v0.4s, v1.4s}, [%10] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"mov v20.16b, v1.16b \n"
"mov v21.16b, v1.16b \n"
"mov v22.16b, v1.16b \n"
"mov v23.16b, v1.16b \n"
"0: \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%3], #32 \n" // r0 r1 r2 r3
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%4], #32 \n" // w0011_01
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v18.4s, v8.4s, v2.s[0] \n"
"fmla v19.4s, v8.4s, v3.s[0] \n"
"fmla v20.4s, v9.4s, v0.s[0] \n"
"fmla v21.4s, v9.4s, v1.s[0] \n"
"fmla v22.4s, v9.4s, v2.s[0] \n"
"fmla v23.4s, v9.4s, v3.s[0] \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%4], #32 \n" // w2233_01
"shll v12.4s, v12.4h, #16 \n"
"shll v13.4s, v13.4h, #16 \n"
"shll v14.4s, v14.4h, #16 \n"
"shll v15.4s, v15.4h, #16 \n"
"fmla v16.4s, v10.4s, v0.s[1] \n"
"fmla v17.4s, v10.4s, v1.s[1] \n"
"fmla v18.4s, v10.4s, v2.s[1] \n"
"fmla v19.4s, v10.4s, v3.s[1] \n"
"fmla v20.4s, v11.4s, v0.s[1] \n"
"fmla v21.4s, v11.4s, v1.s[1] \n"
"fmla v22.4s, v11.4s, v2.s[1] \n"
"fmla v23.4s, v11.4s, v3.s[1] \n"
"fmla v16.4s, v12.4s, v0.s[2] \n"
"fmla v17.4s, v12.4s, v1.s[2] \n"
"fmla v18.4s, v12.4s, v2.s[2] \n"
"fmla v19.4s, v12.4s, v3.s[2] \n"
"fmla v20.4s, v13.4s, v0.s[2] \n"
"fmla v21.4s, v13.4s, v1.s[2] \n"
"fmla v22.4s, v13.4s, v2.s[2] \n"
"fmla v23.4s, v13.4s, v3.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v14.4s, v0.s[3] \n"
"fmla v17.4s, v14.4s, v1.s[3] \n"
"fmla v18.4s, v14.4s, v2.s[3] \n"
"fmla v19.4s, v14.4s, v3.s[3] \n"
"fmla v20.4s, v15.4s, v0.s[3] \n"
"fmla v21.4s, v15.4s, v1.s[3] \n"
"fmla v22.4s, v15.4s, v2.s[3] \n"
"fmla v23.4s, v15.4s, v3.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"shrn v20.4h, v20.4s, #16 \n"
"shrn v21.4h, v21.4s, #16 \n"
"shrn v22.4h, v22.4s, #16 \n"
"shrn v23.4h, v23.4s, #16 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
"st1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%2], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr1), // %2
"=r"(tmpptr), // %3
"=r"(kptr0) // %4
: "0"(nn),
"1"(outptr0),
"2"(outptr1),
"3"(tmpptr),
"4"(kptr0),
"r"(biasptr) // %10
: "cc", "memory", "v0", "v1", "v2", "v3", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
}
for (; i + 1 < size; i += 2)
{
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2);
const unsigned short* kptr0 = kernel.channel(p / 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v0.4s, v1.4s}, [%10] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v1.16b \n"
"mov v19.16b, v1.16b \n"
"0: \n"
"prfm pldl1keep, [%3, #128] \n"
"ld1 {v0.4h, v1.4h}, [%3], #16 \n" // r0 r1
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%4], #32 \n" // w0011_01
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v18.4s, v9.4s, v0.s[0] \n"
"fmla v19.4s, v9.4s, v1.s[0] \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%4], #32 \n" // w2233_01
"shll v12.4s, v12.4h, #16 \n"
"shll v13.4s, v13.4h, #16 \n"
"shll v14.4s, v14.4h, #16 \n"
"shll v15.4s, v15.4h, #16 \n"
"fmla v16.4s, v10.4s, v0.s[1] \n"
"fmla v17.4s, v10.4s, v1.s[1] \n"
"fmla v18.4s, v11.4s, v0.s[1] \n"
"fmla v19.4s, v11.4s, v1.s[1] \n"
"fmla v16.4s, v12.4s, v0.s[2] \n"
"fmla v17.4s, v12.4s, v1.s[2] \n"
"fmla v18.4s, v13.4s, v0.s[2] \n"
"fmla v19.4s, v13.4s, v1.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v14.4s, v0.s[3] \n"
"fmla v17.4s, v14.4s, v1.s[3] \n"
"fmla v18.4s, v15.4s, v0.s[3] \n"
"fmla v19.4s, v15.4s, v1.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"st1 {v16.4h, v17.4h}, [%1], #16 \n"
"st1 {v18.4h, v19.4h}, [%2], #16 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr1), // %2
"=r"(tmpptr), // %3
"=r"(kptr0) // %4
: "0"(nn),
"1"(outptr0),
"2"(outptr1),
"3"(tmpptr),
"4"(kptr0),
"r"(biasptr) // %10
: "cc", "memory", "v0", "v1", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19");
}
for (; i < size; i++)
{
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2);
const unsigned short* kptr0 = kernel.channel(p / 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v16.4s, v17.4s}, [%10] \n"
"0: \n"
"prfm pldl1keep, [%3, #64] \n"
"ld1 {v0.4h}, [%3], #8 \n" // r0
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%4], #32 \n" // w0011_01
"shll v0.4s, v0.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v9.4s, v0.s[0] \n"
"prfm pldl1keep, [%4, #256] \n"
"ld1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%4], #32 \n" // w2233_01
"shll v12.4s, v12.4h, #16 \n"
"shll v13.4s, v13.4h, #16 \n"
"shll v14.4s, v14.4h, #16 \n"
"shll v15.4s, v15.4h, #16 \n"
"fmla v16.4s, v10.4s, v0.s[1] \n"
"fmla v17.4s, v11.4s, v0.s[1] \n"
"fmla v16.4s, v12.4s, v0.s[2] \n"
"fmla v17.4s, v13.4s, v0.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v14.4s, v0.s[3] \n"
"fmla v17.4s, v15.4s, v0.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"st1 {v16.4h}, [%1], #8 \n"
"st1 {v17.4h}, [%2], #8 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr1), // %2
"=r"(tmpptr), // %3
"=r"(kptr0) // %4
: "0"(nn),
"1"(outptr0),
"2"(outptr1),
"3"(tmpptr),
"4"(kptr0),
"r"(biasptr) // %10
: "cc", "memory", "v0", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17");
}
}
#endif // __aarch64__
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = remain_outch_start; p < outch; p++)
{
unsigned short* outptr0 = top_blob.channel(p);
const float zeros[4] = {0.f, 0.f, 0.f, 0.f};
const float* biasptr = bias ? bias + p * 4 : zeros;
int i = 0;
#if __aarch64__
for (; i + 11 < size; i += 12)
{
const unsigned short* tmpptr = tmp.channel(i / 12);
const unsigned short* kptr0 = kernel.channel(p / 2 + p % 2);
int nn = inch * maxk; // inch always > 0
asm volatile(
"ld1 {v0.4s}, [%8] \n"
"mov v8.16b, v0.16b \n"
"mov v9.16b, v0.16b \n"
"mov v10.16b, v0.16b \n"
"mov v11.16b, v0.16b \n"
"mov v12.16b, v0.16b \n"
"mov v13.16b, v0.16b \n"
"mov v14.16b, v0.16b \n"
"mov v15.16b, v0.16b \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"0: \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%2], #32 \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v4.4h, v5.4h, v6.4h, v7.4h}, [%3], #32 \n" // w0123_0
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v6.4s, v6.4h, #16 \n"
"shll v7.4s, v7.4h, #16 \n"
"fmla v8.4s, v4.4s, v0.s[0] \n"
"fmla v9.4s, v4.4s, v0.s[1] \n"
"fmla v10.4s, v4.4s, v0.s[2] \n"
"fmla v11.4s, v4.4s, v0.s[3] \n"
"fmla v12.4s, v4.4s, v1.s[0] \n"
"fmla v13.4s, v4.4s, v1.s[1] \n"
"fmla v14.4s, v4.4s, v1.s[2] \n"
"fmla v15.4s, v4.4s, v1.s[3] \n"
"fmla v16.4s, v4.4s, v2.s[0] \n"
"fmla v17.4s, v4.4s, v2.s[1] \n"
"fmla v18.4s, v4.4s, v2.s[2] \n"
"fmla v19.4s, v4.4s, v2.s[3] \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%2], #32 \n"
"shll v20.4s, v20.4h, #16 \n"
"shll v21.4s, v21.4h, #16 \n"
"shll v22.4s, v22.4h, #16 \n"
"shll v23.4s, v23.4h, #16 \n"
"fmla v8.4s, v5.4s, v3.s[0] \n"
"fmla v9.4s, v5.4s, v3.s[1] \n"
"fmla v10.4s, v5.4s, v3.s[2] \n"
"fmla v11.4s, v5.4s, v3.s[3] \n"
"fmla v12.4s, v5.4s, v20.s[0] \n"
"fmla v13.4s, v5.4s, v20.s[1] \n"
"fmla v14.4s, v5.4s, v20.s[2] \n"
"fmla v15.4s, v5.4s, v20.s[3] \n"
"fmla v16.4s, v5.4s, v21.s[0] \n"
"fmla v17.4s, v5.4s, v21.s[1] \n"
"fmla v18.4s, v5.4s, v21.s[2] \n"
"fmla v19.4s, v5.4s, v21.s[3] \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v24.4h, v25.4h, v26.4h, v27.4h}, [%2], #32 \n"
"shll v24.4s, v24.4h, #16 \n"
"shll v25.4s, v25.4h, #16 \n"
"shll v26.4s, v26.4h, #16 \n"
"shll v27.4s, v27.4h, #16 \n"
"fmla v8.4s, v6.4s, v22.s[0] \n"
"fmla v9.4s, v6.4s, v22.s[1] \n"
"fmla v10.4s, v6.4s, v22.s[2] \n"
"fmla v11.4s, v6.4s, v22.s[3] \n"
"fmla v12.4s, v6.4s, v23.s[0] \n"
"fmla v13.4s, v6.4s, v23.s[1] \n"
"fmla v14.4s, v6.4s, v23.s[2] \n"
"fmla v15.4s, v6.4s, v23.s[3] \n"
"fmla v16.4s, v6.4s, v24.s[0] \n"
"fmla v17.4s, v6.4s, v24.s[1] \n"
"fmla v18.4s, v6.4s, v24.s[2] \n"
"fmla v19.4s, v6.4s, v24.s[3] \n"
"subs %w0, %w0, #1 \n"
"fmla v8.4s, v7.4s, v25.s[0] \n"
"fmla v9.4s, v7.4s, v25.s[1] \n"
"fmla v10.4s, v7.4s, v25.s[2] \n"
"fmla v11.4s, v7.4s, v25.s[3] \n"
"fmla v12.4s, v7.4s, v26.s[0] \n"
"fmla v13.4s, v7.4s, v26.s[1] \n"
"fmla v14.4s, v7.4s, v26.s[2] \n"
"fmla v15.4s, v7.4s, v26.s[3] \n"
"fmla v16.4s, v7.4s, v27.s[0] \n"
"fmla v17.4s, v7.4s, v27.s[1] \n"
"fmla v18.4s, v7.4s, v27.s[2] \n"
"fmla v19.4s, v7.4s, v27.s[3] \n"
"bne 0b \n"
"shrn v8.4h, v8.4s, #16 \n"
"shrn v9.4h, v9.4s, #16 \n"
"shrn v10.4h, v10.4s, #16 \n"
"shrn v11.4h, v11.4s, #16 \n"
"shrn v12.4h, v12.4s, #16 \n"
"shrn v13.4h, v13.4s, #16 \n"
"shrn v14.4h, v14.4s, #16 \n"
"shrn v15.4h, v15.4s, #16 \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"st1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%1], #32 \n"
"st1 {v12.4h, v13.4h, v14.4h, v15.4h}, [%1], #32 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27");
}
#endif // __aarch64__
for (; i + 7 < size; i += 8)
{
#if __aarch64__
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8);
const unsigned short* kptr0 = kernel.channel(p / 2 + p % 2);
#else
const unsigned short* tmpptr = tmp.channel(i / 8);
const unsigned short* kptr0 = kernel.channel(p);
#endif
int nn = inch * maxk; // inch always > 0
#if __aarch64__
asm volatile(
"ld1 {v0.4s}, [%8] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"mov v20.16b, v0.16b \n"
"mov v21.16b, v0.16b \n"
"mov v22.16b, v0.16b \n"
"mov v23.16b, v0.16b \n"
"0: \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%2], #32 \n" // r0 r1 r2 r3
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%3], #32 \n" // w0123
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v18.4s, v8.4s, v2.s[0] \n"
"fmla v19.4s, v8.4s, v3.s[0] \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v4.4h, v5.4h, v6.4h, v7.4h}, [%2], #32 \n" // r4 r5 r6 r7
"shll v4.4s, v4.4h, #16 \n"
"shll v5.4s, v5.4h, #16 \n"
"shll v6.4s, v6.4h, #16 \n"
"shll v7.4s, v7.4h, #16 \n"
"fmla v20.4s, v8.4s, v4.s[0] \n"
"fmla v21.4s, v8.4s, v5.s[0] \n"
"fmla v22.4s, v8.4s, v6.s[0] \n"
"fmla v23.4s, v8.4s, v7.s[0] \n"
"fmla v16.4s, v9.4s, v0.s[1] \n"
"fmla v17.4s, v9.4s, v1.s[1] \n"
"fmla v18.4s, v9.4s, v2.s[1] \n"
"fmla v19.4s, v9.4s, v3.s[1] \n"
"fmla v20.4s, v9.4s, v4.s[1] \n"
"fmla v21.4s, v9.4s, v5.s[1] \n"
"fmla v22.4s, v9.4s, v6.s[1] \n"
"fmla v23.4s, v9.4s, v7.s[1] \n"
"fmla v16.4s, v10.4s, v0.s[2] \n"
"fmla v17.4s, v10.4s, v1.s[2] \n"
"fmla v18.4s, v10.4s, v2.s[2] \n"
"fmla v19.4s, v10.4s, v3.s[2] \n"
"fmla v20.4s, v10.4s, v4.s[2] \n"
"fmla v21.4s, v10.4s, v5.s[2] \n"
"fmla v22.4s, v10.4s, v6.s[2] \n"
"fmla v23.4s, v10.4s, v7.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v11.4s, v0.s[3] \n"
"fmla v17.4s, v11.4s, v1.s[3] \n"
"fmla v18.4s, v11.4s, v2.s[3] \n"
"fmla v19.4s, v11.4s, v3.s[3] \n"
"fmla v20.4s, v11.4s, v4.s[3] \n"
"fmla v21.4s, v11.4s, v5.s[3] \n"
"fmla v22.4s, v11.4s, v6.s[3] \n"
"fmla v23.4s, v11.4s, v7.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"shrn v20.4h, v20.4s, #16 \n"
"shrn v21.4h, v21.4s, #16 \n"
"shrn v22.4h, v22.4s, #16 \n"
"shrn v23.4h, v23.4s, #16 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
"st1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%1], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
#else
asm volatile(
"vld1.f32 {d0-d1}, [%8] \n"
"vmov q8, q0 \n"
"vmov q9, q0 \n"
"vmov q10, q0 \n"
"vmov q11, q0 \n"
"vmov q12, q0 \n"
"vmov q13, q0 \n"
"vmov q14, q0 \n"
"vmov q15, q0 \n"
"0: \n"
"pld [%2, #256] \n"
"vld1.u16 {d4-d7}, [%2]! \n"
"pld [%3, #256] \n"
"vld1.u16 {d12-d15}, [%3]! \n"
"vshll.u16 q0, d4, #16 \n"
"vshll.u16 q1, d5, #16 \n"
"vshll.u16 q2, d6, #16 \n"
"vshll.u16 q3, d7, #16 \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vshll.u16 q6, d14, #16 \n"
"vshll.u16 q7, d15, #16 \n"
"vmla.f32 q8, q4, d0[0] \n"
"vmla.f32 q9, q4, d0[1] \n"
"vmla.f32 q10, q4, d1[0] \n"
"vmla.f32 q11, q4, d1[1] \n"
"vmla.f32 q12, q4, d2[0] \n"
"vmla.f32 q13, q4, d2[1] \n"
"vmla.f32 q14, q4, d3[0] \n"
"vmla.f32 q15, q4, d3[1] \n"
"vmla.f32 q8, q5, d4[0] \n"
"vmla.f32 q9, q5, d4[1] \n"
"vmla.f32 q10, q5, d5[0] \n"
"vmla.f32 q11, q5, d5[1] \n"
"vmla.f32 q12, q5, d6[0] \n"
"vmla.f32 q13, q5, d6[1] \n"
"vmla.f32 q14, q5, d7[0] \n"
"vmla.f32 q15, q5, d7[1] \n"
"pld [%2, #256] \n"
"vld1.u16 {d4-d7}, [%2]! \n"
"vshll.u16 q0, d4, #16 \n"
"vshll.u16 q1, d5, #16 \n"
"vshll.u16 q2, d6, #16 \n"
"vshll.u16 q3, d7, #16 \n"
"vmla.f32 q8, q6, d0[0] \n"
"vmla.f32 q9, q6, d0[1] \n"
"vmla.f32 q10, q6, d1[0] \n"
"vmla.f32 q11, q6, d1[1] \n"
"vmla.f32 q12, q6, d2[0] \n"
"vmla.f32 q13, q6, d2[1] \n"
"vmla.f32 q14, q6, d3[0] \n"
"vmla.f32 q15, q6, d3[1] \n"
"subs %0, %0, #1 \n"
"vmla.f32 q8, q7, d4[0] \n"
"vmla.f32 q9, q7, d4[1] \n"
"vmla.f32 q10, q7, d5[0] \n"
"vmla.f32 q11, q7, d5[1] \n"
"vmla.f32 q12, q7, d6[0] \n"
"vmla.f32 q13, q7, d6[1] \n"
"vmla.f32 q14, q7, d7[0] \n"
"vmla.f32 q15, q7, d7[1] \n"
"bne 0b \n"
"vshrn.u32 d16, q8, #16 \n"
"vshrn.u32 d17, q9, #16 \n"
"vshrn.u32 d18, q10, #16 \n"
"vshrn.u32 d19, q11, #16 \n"
"vshrn.u32 d20, q12, #16 \n"
"vshrn.u32 d21, q13, #16 \n"
"vshrn.u32 d22, q14, #16 \n"
"vshrn.u32 d23, q15, #16 \n"
"vstm %1!, {d16-d23} \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15");
#endif
}
for (; i + 3 < size; i += 4)
{
#if __aarch64__
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4);
const unsigned short* kptr0 = kernel.channel(p / 2 + p % 2);
#else
const unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4);
const unsigned short* kptr0 = kernel.channel(p);
#endif
int nn = inch * maxk; // inch always > 0
#if __aarch64__
asm volatile(
"ld1 {v0.4s}, [%8] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"mov v18.16b, v0.16b \n"
"mov v19.16b, v0.16b \n"
"0: \n"
"prfm pldl1keep, [%2, #256] \n"
"ld1 {v0.4h, v1.4h, v2.4h, v3.4h}, [%2], #32 \n" // r0 r1 r2 r3
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%3], #32 \n" // w0123
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"shll v2.4s, v2.4h, #16 \n"
"shll v3.4s, v3.4h, #16 \n"
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v18.4s, v8.4s, v2.s[0] \n"
"fmla v19.4s, v8.4s, v3.s[0] \n"
"fmla v16.4s, v9.4s, v0.s[1] \n"
"fmla v17.4s, v9.4s, v1.s[1] \n"
"fmla v18.4s, v9.4s, v2.s[1] \n"
"fmla v19.4s, v9.4s, v3.s[1] \n"
"fmla v16.4s, v10.4s, v0.s[2] \n"
"fmla v17.4s, v10.4s, v1.s[2] \n"
"fmla v18.4s, v10.4s, v2.s[2] \n"
"fmla v19.4s, v10.4s, v3.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v11.4s, v0.s[3] \n"
"fmla v17.4s, v11.4s, v1.s[3] \n"
"fmla v18.4s, v11.4s, v2.s[3] \n"
"fmla v19.4s, v11.4s, v3.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"shrn v18.4h, v18.4s, #16 \n"
"shrn v19.4h, v19.4s, #16 \n"
"st1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%1], #32 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "v0", "v1", "v2", "v3", "v8", "v9", "v10", "v11", "v16", "v17", "v18", "v19");
#else
asm volatile(
"vld1.f32 {d0-d1}, [%8] \n"
"vmov q8, q0 \n"
"vmov q9, q0 \n"
"vmov q10, q0 \n"
"vmov q11, q0 \n"
"0: \n"
"pld [%2, #256] \n"
"vld1.u16 {d4-d7}, [%2]! \n"
"pld [%3, #256] \n"
"vld1.u16 {d12-d15}, [%3]! \n"
"vshll.u16 q0, d4, #16 \n"
"vshll.u16 q1, d5, #16 \n"
"vshll.u16 q2, d6, #16 \n"
"vshll.u16 q3, d7, #16 \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vshll.u16 q6, d14, #16 \n"
"vshll.u16 q7, d15, #16 \n"
"vmla.f32 q8, q4, d0[0] \n"
"vmla.f32 q9, q4, d2[0] \n"
"vmla.f32 q10, q4, d4[0] \n"
"vmla.f32 q11, q4, d6[0] \n"
"vmla.f32 q8, q5, d0[1] \n"
"vmla.f32 q9, q5, d2[1] \n"
"vmla.f32 q10, q5, d4[1] \n"
"vmla.f32 q11, q5, d6[1] \n"
"vmla.f32 q8, q6, d1[0] \n"
"vmla.f32 q9, q6, d3[0] \n"
"vmla.f32 q10, q6, d5[0] \n"
"vmla.f32 q11, q6, d7[0] \n"
"subs %0, %0, #1 \n"
"vmla.f32 q8, q7, d1[1] \n"
"vmla.f32 q9, q7, d3[1] \n"
"vmla.f32 q10, q7, d5[1] \n"
"vmla.f32 q11, q7, d7[1] \n"
"bne 0b \n"
"vshrn.u32 d16, q8, #16 \n"
"vshrn.u32 d17, q9, #16 \n"
"vshrn.u32 d18, q10, #16 \n"
"vshrn.u32 d19, q11, #16 \n"
"vst1.u16 {d16-d19}, [%1]! \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", "q8", "q9", "q10", "q11");
#endif
}
for (; i + 1 < size; i += 2)
{
#if __aarch64__
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2);
const unsigned short* kptr0 = kernel.channel(p / 2 + p % 2);
#else
const unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4 + (i % 4) / 2);
const unsigned short* kptr0 = kernel.channel(p);
#endif
int nn = inch * maxk; // inch always > 0
#if __aarch64__
asm volatile(
"ld1 {v0.4s}, [%8] \n"
"mov v16.16b, v0.16b \n"
"mov v17.16b, v0.16b \n"
"0: \n"
"prfm pldl1keep, [%2, #128] \n"
"ld1 {v0.4h, v1.4h}, [%2], #16 \n" // r0 r1
"shll v0.4s, v0.4h, #16 \n"
"shll v1.4s, v1.4h, #16 \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%3], #32 \n" // w0123
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v17.4s, v8.4s, v1.s[0] \n"
"fmla v16.4s, v9.4s, v0.s[1] \n"
"fmla v17.4s, v9.4s, v1.s[1] \n"
"fmla v16.4s, v10.4s, v0.s[2] \n"
"fmla v17.4s, v10.4s, v1.s[2] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v11.4s, v0.s[3] \n"
"fmla v17.4s, v11.4s, v1.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"shrn v17.4h, v17.4s, #16 \n"
"st1 {v16.4h, v17.4h}, [%1], #16 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "v0", "v1", "v8", "v9", "v10", "v11", "v16", "v17");
#else
asm volatile(
"vld1.f32 {d0-d1}, [%8] \n"
"vmov q8, q0 \n"
"vmov q9, q0 \n"
"0: \n"
"pld [%2, #128] \n"
"vld1.u16 {d4-d5}, [%2 :128]! \n"
"vshll.u16 q0, d4, #16 \n"
"vshll.u16 q1, d5, #16 \n"
"pld [%3, #256] \n"
"vld1.u16 {d12-d15}, [%3]! \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vshll.u16 q6, d14, #16 \n"
"vshll.u16 q7, d15, #16 \n"
"vmla.f32 q8, q4, d0[0] \n"
"vmla.f32 q9, q4, d2[0] \n"
"vmla.f32 q8, q5, d0[1] \n"
"vmla.f32 q9, q5, d2[1] \n"
"vmla.f32 q8, q6, d1[0] \n"
"vmla.f32 q9, q6, d3[0] \n"
"subs %0, %0, #1 \n"
"vmla.f32 q8, q7, d1[1] \n"
"vmla.f32 q9, q7, d3[1] \n"
"bne 0b \n"
"vshrn.u32 d16, q8, #16 \n"
"vshrn.u32 d17, q9, #16 \n"
"vst1.u16 {d16-d17}, [%1 :128]! \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "q0", "q1", "q4", "q5", "q6", "q7", "q8", "q9");
#endif
}
for (; i < size; i++)
{
#if __aarch64__
const unsigned short* tmpptr = tmp.channel(i / 12 + (i % 12) / 8 + (i % 12 % 8) / 4 + (i % 12 % 4) / 2 + i % 12 % 2);
const unsigned short* kptr0 = kernel.channel(p / 2 + p % 2);
#else
const unsigned short* tmpptr = tmp.channel(i / 8 + (i % 8) / 4 + (i % 4) / 2 + i % 2);
const unsigned short* kptr0 = kernel.channel(p);
#endif
int nn = inch * maxk; // inch always > 0
#if __aarch64__
asm volatile(
"ld1 {v16.4s}, [%8] \n"
"0: \n"
"prfm pldl1keep, [%2, #64] \n"
"ld1 {v0.4h}, [%2], #8 \n" // r0
"shll v0.4s, v0.4h, #16 \n"
"prfm pldl1keep, [%3, #256] \n"
"ld1 {v8.4h, v9.4h, v10.4h, v11.4h}, [%3], #32 \n" // w0123
"shll v8.4s, v8.4h, #16 \n"
"shll v9.4s, v9.4h, #16 \n"
"shll v10.4s, v10.4h, #16 \n"
"shll v11.4s, v11.4h, #16 \n"
"fmla v16.4s, v8.4s, v0.s[0] \n"
"fmla v16.4s, v9.4s, v0.s[1] \n"
"subs %w0, %w0, #1 \n"
"fmla v16.4s, v10.4s, v0.s[2] \n"
"fmla v16.4s, v11.4s, v0.s[3] \n"
"bne 0b \n"
"shrn v16.4h, v16.4s, #16 \n"
"st1 {v16.4h}, [%1], #8 \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "v0", "v8", "v9", "v10", "v11", "v16");
#else
asm volatile(
"vld1.f32 {d16-d17}, [%8] \n"
"0: \n"
"pld [%2, #64] \n"
"vld1.u16 {d1}, [%2 :64]! \n"
"vshll.u16 q0, d1, #16 \n"
"pld [%3, #256] \n"
"vld1.u16 {d12-d15}, [%3]! \n"
"vshll.u16 q4, d12, #16 \n"
"vshll.u16 q5, d13, #16 \n"
"vshll.u16 q6, d14, #16 \n"
"vshll.u16 q7, d15, #16 \n"
"vmla.f32 q8, q4, d0[0] \n"
"vmla.f32 q8, q5, d0[1] \n"
"subs %0, %0, #1 \n"
"vmla.f32 q8, q6, d1[0] \n"
"vmla.f32 q8, q7, d1[1] \n"
"bne 0b \n"
"vshrn.u32 d16, q8, #16 \n"
"vst1.u16 {d16}, [%1 :64]! \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(tmpptr), // %2
"=r"(kptr0) // %3
: "0"(nn),
"1"(outptr0),
"2"(tmpptr),
"3"(kptr0),
"r"(biasptr) // %8
: "cc", "memory", "q0", "q4", "q5", "q6", "q7", "q8");
#endif
}
}
}
static void convolution_im2col_sgemm_transform_kernel_pack4_bf16s_neon(const Mat& _kernel, Mat& kernel_tm, int inch, int outch, int kernel_w, int kernel_h)
{
const int maxk = kernel_w * kernel_h;
// interleave
// src = maxk-inch-outch
// dst = 4b-4a-maxk-inch/4a-outch/4b
Mat kernel = _kernel.reshape(maxk, inch, outch);
#if __aarch64__
kernel_tm.create(32 * maxk, inch / 4, outch / 8 + (outch % 8) / 4, (size_t)2u);
#else
kernel_tm.create(16 * maxk, inch / 4, outch / 4, (size_t)2u);
#endif
int q = 0;
#if __aarch64__
for (; q + 7 < outch; q += 8)
{
const Mat k0 = kernel.channel(q);
const Mat k1 = kernel.channel(q + 1);
const Mat k2 = kernel.channel(q + 2);
const Mat k3 = kernel.channel(q + 3);
const Mat k4 = kernel.channel(q + 4);
const Mat k5 = kernel.channel(q + 5);
const Mat k6 = kernel.channel(q + 6);
const Mat k7 = kernel.channel(q + 7);
unsigned short* g00 = kernel_tm.channel(q / 8);
for (int p = 0; p + 3 < inch; p += 4)
{
const float* k00 = k0.row(p);
const float* k01 = k0.row(p + 1);
const float* k02 = k0.row(p + 2);
const float* k03 = k0.row(p + 3);
const float* k10 = k1.row(p);
const float* k11 = k1.row(p + 1);
const float* k12 = k1.row(p + 2);
const float* k13 = k1.row(p + 3);
const float* k20 = k2.row(p);
const float* k21 = k2.row(p + 1);
const float* k22 = k2.row(p + 2);
const float* k23 = k2.row(p + 3);
const float* k30 = k3.row(p);
const float* k31 = k3.row(p + 1);
const float* k32 = k3.row(p + 2);
const float* k33 = k3.row(p + 3);
const float* k40 = k4.row(p);
const float* k41 = k4.row(p + 1);
const float* k42 = k4.row(p + 2);
const float* k43 = k4.row(p + 3);
const float* k50 = k5.row(p);
const float* k51 = k5.row(p + 1);
const float* k52 = k5.row(p + 2);
const float* k53 = k5.row(p + 3);
const float* k60 = k6.row(p);
const float* k61 = k6.row(p + 1);
const float* k62 = k6.row(p + 2);
const float* k63 = k6.row(p + 3);
const float* k70 = k7.row(p);
const float* k71 = k7.row(p + 1);
const float* k72 = k7.row(p + 2);
const float* k73 = k7.row(p + 3);
for (int k = 0; k < maxk; k++)
{
g00[0] = float32_to_bfloat16(k00[k]);
g00[1] = float32_to_bfloat16(k10[k]);
g00[2] = float32_to_bfloat16(k20[k]);
g00[3] = float32_to_bfloat16(k30[k]);
g00[4] = float32_to_bfloat16(k40[k]);
g00[5] = float32_to_bfloat16(k50[k]);
g00[6] = float32_to_bfloat16(k60[k]);
g00[7] = float32_to_bfloat16(k70[k]);
g00[8] = float32_to_bfloat16(k01[k]);
g00[9] = float32_to_bfloat16(k11[k]);
g00[10] = float32_to_bfloat16(k21[k]);
g00[11] = float32_to_bfloat16(k31[k]);
g00[12] = float32_to_bfloat16(k41[k]);
g00[13] = float32_to_bfloat16(k51[k]);
g00[14] = float32_to_bfloat16(k61[k]);
g00[15] = float32_to_bfloat16(k71[k]);
g00[16] = float32_to_bfloat16(k02[k]);
g00[17] = float32_to_bfloat16(k12[k]);
g00[18] = float32_to_bfloat16(k22[k]);
g00[19] = float32_to_bfloat16(k32[k]);
g00[20] = float32_to_bfloat16(k42[k]);
g00[21] = float32_to_bfloat16(k52[k]);
g00[22] = float32_to_bfloat16(k62[k]);
g00[23] = float32_to_bfloat16(k72[k]);
g00[24] = float32_to_bfloat16(k03[k]);
g00[25] = float32_to_bfloat16(k13[k]);
g00[26] = float32_to_bfloat16(k23[k]);
g00[27] = float32_to_bfloat16(k33[k]);
g00[28] = float32_to_bfloat16(k43[k]);
g00[29] = float32_to_bfloat16(k53[k]);
g00[30] = float32_to_bfloat16(k63[k]);
g00[31] = float32_to_bfloat16(k73[k]);
g00 += 32;
}
}
}
#endif // __aarch64__
for (; q + 3 < outch; q += 4)
{
const Mat k0 = kernel.channel(q);
const Mat k1 = kernel.channel(q + 1);
const Mat k2 = kernel.channel(q + 2);
const Mat k3 = kernel.channel(q + 3);
#if __aarch64__
unsigned short* g00 = kernel_tm.channel(q / 8 + (q % 8) / 4);
#else
unsigned short* g00 = kernel_tm.channel(q / 4);
#endif
for (int p = 0; p + 3 < inch; p += 4)
{
const float* k00 = k0.row(p);
const float* k01 = k0.row(p + 1);
const float* k02 = k0.row(p + 2);
const float* k03 = k0.row(p + 3);
const float* k10 = k1.row(p);
const float* k11 = k1.row(p + 1);
const float* k12 = k1.row(p + 2);
const float* k13 = k1.row(p + 3);
const float* k20 = k2.row(p);
const float* k21 = k2.row(p + 1);
const float* k22 = k2.row(p + 2);
const float* k23 = k2.row(p + 3);
const float* k30 = k3.row(p);
const float* k31 = k3.row(p + 1);
const float* k32 = k3.row(p + 2);
const float* k33 = k3.row(p + 3);
for (int k = 0; k < maxk; k++)
{
g00[0] = float32_to_bfloat16(k00[k]);
g00[1] = float32_to_bfloat16(k10[k]);
g00[2] = float32_to_bfloat16(k20[k]);
g00[3] = float32_to_bfloat16(k30[k]);
g00[4] = float32_to_bfloat16(k01[k]);
g00[5] = float32_to_bfloat16(k11[k]);
g00[6] = float32_to_bfloat16(k21[k]);
g00[7] = float32_to_bfloat16(k31[k]);
g00[8] = float32_to_bfloat16(k02[k]);
g00[9] = float32_to_bfloat16(k12[k]);
g00[10] = float32_to_bfloat16(k22[k]);
g00[11] = float32_to_bfloat16(k32[k]);
g00[12] = float32_to_bfloat16(k03[k]);
g00[13] = float32_to_bfloat16(k13[k]);
g00[14] = float32_to_bfloat16(k23[k]);
g00[15] = float32_to_bfloat16(k33[k]);
g00 += 16;
}
}
}
}
static void convolution_im2col_sgemm_pack4_bf16s_neon(const Mat& bottom_blob, Mat& top_blob, const Mat& kernel, const Mat& _bias, int kernel_w, int kernel_h, int dilation_w, int dilation_h, int stride_w, int stride_h, const Option& opt)
{
int w = bottom_blob.w;
int inch = bottom_blob.c;
int outw = top_blob.w;
int outh = top_blob.h;
const int size = outw * outh;
const int maxk = kernel_w * kernel_h;
// im2col
Mat bottom_im2col(size, maxk, inch, 8u, 4, opt.workspace_allocator);
{
const int gap = (w * stride_h - outw * stride_w) * 4;
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = 0; p < inch; p++)
{
const Mat img = bottom_blob.channel(p);
unsigned short* ptr = bottom_im2col.channel(p);
for (int u = 0; u < kernel_h; u++)
{
for (int v = 0; v < kernel_w; v++)
{
const unsigned short* sptr = img.row<const unsigned short>(dilation_h * u) + dilation_w * v * 4;
for (int i = 0; i < outh; i++)
{
int j = 0;
for (; j + 3 < outw; j += 4)
{
uint16x4_t _val0 = vld1_u16(sptr);
uint16x4_t _val1 = vld1_u16(sptr + stride_w * 4);
uint16x4_t _val2 = vld1_u16(sptr + stride_w * 8);
uint16x4_t _val3 = vld1_u16(sptr + stride_w * 12);
vst1_u16(ptr, _val0);
vst1_u16(ptr + 4, _val1);
vst1_u16(ptr + 8, _val2);
vst1_u16(ptr + 12, _val3);
sptr += stride_w * 16;
ptr += 16;
}
for (; j + 1 < outw; j += 2)
{
uint16x4_t _val0 = vld1_u16(sptr);
uint16x4_t _val1 = vld1_u16(sptr + stride_w * 4);
vst1_u16(ptr, _val0);
vst1_u16(ptr + 4, _val1);
sptr += stride_w * 8;
ptr += 8;
}
for (; j < outw; j++)
{
uint16x4_t _val = vld1_u16(sptr);
vst1_u16(ptr, _val);
sptr += stride_w * 4;
ptr += 4;
}
sptr += gap;
}
}
}
}
}
im2col_sgemm_pack4_bf16s_neon(bottom_im2col, top_blob, kernel, _bias, opt);
}
|
Typing.h | #ifndef INCLUDE_BAYESTYPING_TYPING_H
#define INCLUDE_BAYESTYPING_TYPING_H
#include <vector>
#include <seqan/sequence.h>
#include <omp.h>
#include "options.h"
class Typing
{
private:
int readlen;
int reflen;
int** scoreArr;
vector<int> candAllele;
private:
void calMaxHit();
//void pickCandidateAllele(double filter_threshold);
public:
vector<PairAlleles> answers;
vector<int> maxHit; // max scores for each reads gittest
public:
void bayesTyping(int mnoise);
int readTyping(CharString inTypingFile);
int writeTyping(CharString outTypingFile);
Typing(int **inScoreArr,int inReadlen, int inReflen);
void pickCandidateAllele(double filter_threshold);
};
Typing::Typing(int **inScoreArr,int inReadlen, int inReflen){
scoreArr = inScoreArr;
readlen = inReadlen;
reflen = inReflen;
maxHit=vector<int>(readlen,0);
}
void Typing::calMaxHit(){ // max scores for each reads
for(unsigned k=0;k<readlen;k++){
int maxscore=0;
for(int ci=0;ci<candAllele.size();ci++){
int i=candAllele[ci];
if(scoreArr[k][i]>maxscore)
maxscore=scoreArr[k][i];
}
if(maxscore>0){
maxHit[k] = maxscore;
}
}
}
void Typing::bayesTyping(int mnoise){
calMaxHit();
vector<int> candRead;
vector<PairAlleles> tmpanswers;
for(unsigned k=0;k<readlen;k++){
if(maxHit[k]>0){
candRead.push_back(k);
}
}
int maxalign=0;
//SEQAN_OMP_PRAGMA(parallel for)
for(unsigned ci=0;ci<candAllele.size()-1;ci++){
vector<int> difv;
for(int cj=ci+1;cj<candAllele.size();cj++){
int sum=0;
int i = candAllele[ci];
int j = candAllele[cj];
difv.clear();
for(int ck=0;ck<candRead.size();ck++){
int k=candRead[ck];
int maxscore=0;
if(scoreArr[k][i]>scoreArr[k][j]){
maxscore=scoreArr[k][i];
}
else{
maxscore=scoreArr[k][j];
}
if(maxHit[k]>maxscore)
difv.push_back(maxHit[k]-maxscore);
sum+=maxscore;
}
sort(difv.begin(),difv.end(),greater<int>());
for(int k=0;k<mnoise && k<difv.size();k++)
sum+=difv[k];
// #pragma omp critical(dataupdate)
if(sum>=maxalign){ // prepare for the outputs
maxalign = sum;
if(mnoise==0)
tmpanswers.push_back(PairAlleles(i,j,INT_MAX,sum));
else
tmpanswers.push_back(PairAlleles(i,j,difv[mnoise-1],sum));
}
}
}
for (vector<PairAlleles>::iterator it = tmpanswers.begin() ; it != tmpanswers.end(); ++it){
if(it->score==maxalign)
answers.push_back(*it);
}
}
void Typing::pickCandidateAllele(double filter_threshold){
vector<int> matchReads(reflen,0); // #reads with max score to a allele
int maxscore;
for(unsigned k=0;k<readlen;k++){
maxscore=0;
for(unsigned i=0;i<reflen;i++){
if(scoreArr[k][i]>maxscore)
maxscore=scoreArr[k][i];
}
for(unsigned i=0;i<reflen;i++){
if(scoreArr[k][i]==maxscore)
matchReads[i]+=1;
}
}
for(unsigned i=0;i<reflen;i++){
if(matchReads[i]>filter_threshold*(double)readlen){
candAllele.push_back(i);
}
}
}
int Typing::writeTyping(CharString outTypingFile){
ofstream ofs (toCString(outTypingFile));
if (ofs.is_open())
{
for(unsigned i=0;i<answers.size();i++){
ofs << answers[i].allele1 << "\t" << answers[i].allele2 << "\t" << answers[i].maxDiff << "\t" << answers[i].score << endl;
}
ofs.close();
return 1;
}
else{
return 0;
}
}
int Typing::readTyping(CharString inTypingFile){
ifstream ifs;
ifs.open (toCString(inTypingFile), ifstream::in);
answers.clear();
if (ifs.is_open())
{
int allele1,allele2,maxDiff,score;
while ((ifs >> allele1 >> allele2 >> maxDiff >> score).good()) {
answers.push_back(PairAlleles(allele1,allele2,maxDiff,score));
}
ifs.close();
return 1;
}
else
return 0;
}
#endif
|
omp_taskloop_num_tasks.c | // RUN: %libomp-compile-and-run
// RUN: %libomp-compile && env KMP_TASKLOOP_MIN_TASKS=1 %libomp-run
// These compilers don't support the taskloop construct
// UNSUPPORTED: gcc-4, gcc-5, icc-16
/*
* Test for taskloop
* Method: caculate how many times the iteration space is dispatched
* and judge if each dispatch has the requested grainsize
* It is possible for two adjacent chunks are executed by the same thread
*/
#include <stdio.h>
#include <omp.h>
#include <stdlib.h>
#include "omp_testsuite.h"
#define CFDMAX_SIZE 1120
int test_omp_taskloop_num_tasks()
{
int i;
int *tids;
int *tidsArray;
int count;
int result = 0;
int num_tasks;
for (num_tasks = 1; num_tasks < 120; ++num_tasks) {
count = 0;
tidsArray = (int *)malloc(sizeof(int) * CFDMAX_SIZE);
tids = tidsArray;
#pragma omp parallel shared(tids)
{
int i;
#pragma omp master
#pragma omp taskloop num_tasks(num_tasks)
for (i = 0; i < CFDMAX_SIZE; i++) {
tids[i] = omp_get_thread_num();
}
}
for (i = 0; i < CFDMAX_SIZE - 1; ++i) {
if (tids[i] != tids[i + 1]) {
count++;
}
}
if (count > num_tasks) {
fprintf(stderr, "counted too many tasks: (wanted %d, got %d)\n",
num_tasks, count);
result++;
}
}
return (result==0);
}
int main()
{
int i;
int num_failed=0;
for (i = 0; i < REPETITIONS; i++) {
if (!test_omp_taskloop_num_tasks()) {
num_failed++;
}
}
return num_failed;
}
|
GB_binop__plus_uint8.c |
//------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__plus_uint8)
// A.*B function (eWiseMult): GB (_AemultB_08__plus_uint8)
// A.*B function (eWiseMult): GB (_AemultB_02__plus_uint8)
// A.*B function (eWiseMult): GB (_AemultB_04__plus_uint8)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__plus_uint8)
// A*D function (colscale): GB (_AxD__plus_uint8)
// D*A function (rowscale): GB (_DxB__plus_uint8)
// C+=B function (dense accum): GB (_Cdense_accumB__plus_uint8)
// C+=b function (dense accum): GB (_Cdense_accumb__plus_uint8)
// C+=A+B function (dense ewise3): GB (_Cdense_ewise3_accum__plus_uint8)
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__plus_uint8)
// C=scalar+B GB (_bind1st__plus_uint8)
// C=scalar+B' GB (_bind1st_tran__plus_uint8)
// C=A+scalar GB (_bind2nd__plus_uint8)
// C=A'+scalar GB (_bind2nd_tran__plus_uint8)
// C type: uint8_t
// A type: uint8_t
// A pattern? 0
// B type: uint8_t
// B pattern? 0
// BinaryOp: cij = (aij + bij)
#define GB_ATYPE \
uint8_t
#define GB_BTYPE \
uint8_t
#define GB_CTYPE \
uint8_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
uint8_t aij = GBX (Ax, pA, A_iso)
// true if values of A are not used
#define GB_A_IS_PATTERN \
0 \
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
uint8_t bij = GBX (Bx, pB, B_iso)
// true if values of B are not used
#define GB_B_IS_PATTERN \
0 \
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
uint8_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = (x + y) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_PLUS || GxB_NO_UINT8 || GxB_NO_PLUS_UINT8)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB (_Cdense_ewise3_accum__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
void GB (_Cdense_ewise3_noaccum__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_noaccum_template.c"
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__plus_uint8)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type uint8_t
uint8_t bwork = (*((uint8_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix D,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint8_t *restrict Cx = (uint8_t *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix D,
const GrB_Matrix B,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint8_t *restrict Cx = (uint8_t *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__plus_uint8)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool is_eWiseUnion,
const GB_void *alpha_scalar_in,
const GB_void *beta_scalar_in,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
uint8_t alpha_scalar ;
uint8_t beta_scalar ;
if (is_eWiseUnion)
{
alpha_scalar = (*((uint8_t *) alpha_scalar_in)) ;
beta_scalar = (*((uint8_t *) beta_scalar_in )) ;
}
#include "GB_add_template.c"
GB_FREE_WORKSPACE ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__plus_uint8)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__plus_uint8)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__plus_uint8)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint8_t *Cx = (uint8_t *) Cx_output ;
uint8_t x = (*((uint8_t *) x_input)) ;
uint8_t *Bx = (uint8_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
uint8_t bij = GBX (Bx, p, false) ;
Cx [p] = (x + bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__plus_uint8)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
uint8_t *Cx = (uint8_t *) Cx_output ;
uint8_t *Ax = (uint8_t *) Ax_input ;
uint8_t y = (*((uint8_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
uint8_t aij = GBX (Ax, p, false) ;
Cx [p] = (aij + y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
uint8_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (x + aij) ; \
}
GrB_Info GB (_bind1st_tran__plus_uint8)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
uint8_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint8_t x = (*((const uint8_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
uint8_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
uint8_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (aij + y) ; \
}
GrB_Info GB (_bind2nd_tran__plus_uint8)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
uint8_t y = (*((const uint8_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
GB_binop__rminus_int16.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__rminus_int16)
// A.*B function (eWiseMult): GB (_AemultB_01__rminus_int16)
// A.*B function (eWiseMult): GB (_AemultB_02__rminus_int16)
// A.*B function (eWiseMult): GB (_AemultB_03__rminus_int16)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__rminus_int16)
// A*D function (colscale): GB (_AxD__rminus_int16)
// D*A function (rowscale): GB (_DxB__rminus_int16)
// C+=B function (dense accum): GB (_Cdense_accumB__rminus_int16)
// C+=b function (dense accum): GB (_Cdense_accumb__rminus_int16)
// C+=A+B function (dense ewise3): GB (_Cdense_ewise3_accum__rminus_int16)
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__rminus_int16)
// C=scalar+B GB (_bind1st__rminus_int16)
// C=scalar+B' GB (_bind1st_tran__rminus_int16)
// C=A+scalar GB (_bind2nd__rminus_int16)
// C=A'+scalar GB (_bind2nd_tran__rminus_int16)
// C type: int16_t
// A type: int16_t
// B,b type: int16_t
// BinaryOp: cij = (bij - aij)
#define GB_ATYPE \
int16_t
#define GB_BTYPE \
int16_t
#define GB_CTYPE \
int16_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
int16_t aij = GBX (Ax, pA, A_iso)
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
int16_t bij = GBX (Bx, pB, B_iso)
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int16_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = (y - x) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_RMINUS || GxB_NO_INT16 || GxB_NO_RMINUS_INT16)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB (_Cdense_ewise3_accum__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_ewise3_noaccum__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__rminus_int16)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int16_t
int16_t bwork = (*((int16_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *restrict Cx = (int16_t *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *restrict Cx = (int16_t *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__rminus_int16)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
#include "GB_add_template.c"
GB_FREE_WORK ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_01__rminus_int16)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_01_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_03__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_03_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__rminus_int16)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__rminus_int16)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *Cx = (int16_t *) Cx_output ;
int16_t x = (*((int16_t *) x_input)) ;
int16_t *Bx = (int16_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
int16_t bij = GBX (Bx, p, false) ;
Cx [p] = (bij - x) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__rminus_int16)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int16_t *Cx = (int16_t *) Cx_output ;
int16_t *Ax = (int16_t *) Ax_input ;
int16_t y = (*((int16_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int16_t aij = GBX (Ax, p, false) ;
Cx [p] = (y - aij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (aij - x) ; \
}
GrB_Info GB (_bind1st_tran__rminus_int16)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t x = (*((const int16_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = (y - aij) ; \
}
GrB_Info GB (_bind2nd_tran__rminus_int16)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t y = (*((const int16_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
target_enter_data.c | #pragma omp target enter data [clauses]
|
flip_op.h | /* Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include <algorithm>
#include <bitset>
#include <vector>
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/framework/operator.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
constexpr size_t dim_bitset_size = 64;
template <typename DeviceContext, typename T>
class FlipKernel : public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& ctx) const override;
};
template <typename T>
class FlipKernel<platform::CPUDeviceContext, T>
: public framework::OpKernel<T> {
public:
void Compute(const framework::ExecutionContext& ctx) const override {
const Tensor* x = ctx.Input<Tensor>("X");
Tensor* out = ctx.Output<Tensor>("Out");
auto flip_dims = ctx.template Attr<std::vector<int>>("axis");
auto x_dims = x->dims();
const int total_dims = x_dims.size();
std::bitset<dim_bitset_size> dim_bitset;
for (size_t i = 0; i < flip_dims.size(); ++i) {
int dim = flip_dims[i];
if (flip_dims[i] < 0) {
dim += total_dims;
}
dim_bitset[dim] = true;
}
auto x_strides = framework::stride(x_dims);
auto numel = x->numel();
const T* x_data = x->data<T>();
T* out_data = out->mutable_data<T>(ctx.GetPlace());
#ifdef PADDLE_WITH_MKLML
#pragma omp parallel for
#endif
for (int64_t i = 0; i < numel; ++i) {
int64_t cur_indices = i;
int64_t rem = 0;
int64_t dst_offset = 0;
for (int d = 0; d < total_dims; ++d) {
int64_t temp = cur_indices;
cur_indices = cur_indices / x_strides[d];
rem = temp - cur_indices * x_strides[d];
dst_offset += dim_bitset[d]
? (x_dims[d] - 1 - cur_indices) * x_strides[d]
: cur_indices * x_strides[d];
cur_indices = rem;
}
out_data[i] = x_data[dst_offset];
}
}
};
} // namespace operators
} // namespace paddle
|
ompfor3.c | /*
* Decremental loop iteration,
* Default loop scheduling
*/
#include <stdio.h>
#ifdef _OPENMP
#include <omp.h>
#endif
int a[20];
int main(void)
{
int i;
int j = 100;
#pragma omp parallel
{
#pragma omp single
printf ("Using %d threads.\n",omp_get_num_threads());
#pragma omp for nowait firstprivate(j) lastprivate(j)
for (i=19;i>-1;i-=3)
{
a[i]=i*2+j;
printf("Iteration %2d is carried out by thread %2d\n",\
i, omp_get_thread_num());
}
}
}
|
Scalar3D.h | /*
###################################################################################
#
# BCMTools
#
# Copyright (c) 2011-2014 Institute of Industrial Science, The University of Tokyo.
# All rights reserved.
#
# Copyright (c) 2012-2016 Advanced Institute for Computational Science (AICS), RIKEN.
# All rights reserved.
#
# Copyright (c) 2017 Research Institute for Information Technology (RIIT), Kyushu University.
# All rights reserved.
#
###################################################################################
*/
///
/// @file Scalar3D.h
/// @brief スカラーデータクラス
///
#ifndef SCALAR_3D_H
#define SCALAR_3D_H
#include "VCUpdater.h"
#include "DataClass.h"
#include "Index3D.h"
#ifdef BCMT_NAMESPACE
namespace BCMT_NAMESPACE {
#endif
/// スカラーデータクラス.
template <typename T>
class Scalar3D : public UpdatableDataClass {
private:
int nx0, ny0, nz0; ///< 内部配列サイズ(仮想セルを含めたセル分割数)
int c_0, c_j, c_k;
T* data; ///< 1次元データ配列
Index3DS index; ///< インデックスファンクタ
int vc0;
public:
/// コンストラクタ.
///
/// @param[in] size 分割数
/// @param[in] vc 仮想セル幅
///
Scalar3D(const ::Vec3i& size, int vc) : UpdatableDataClass(size, vc), index(size, vc) {
nx0 = size[0] + 2*vc;
ny0 = size[1] + 2*vc;
nz0 = size[2] + 2*vc;
data = new T[nx0*ny0*nz0];
vc0 = vc;
c_j = nx0;
c_k = nx0 * ny0;
c_0 = (1 + nx0 + nx0 * ny0) * vc;
}
/// コンストラクタ(for contiguous memory access).
///
/// @param[in] size 分割数
/// @param[in] vc 仮想セル幅
///
Scalar3D(const ::Vec3i& size, int vc, T* data0) : UpdatableDataClass(size, vc), index(size, vc) {
nx0 = size[0] + 2*vc;
ny0 = size[1] + 2*vc;
nz0 = size[2] + 2*vc;
data = data0;
vc0 = vc;
c_j = nx0;
c_k = nx0 * ny0;
c_0 = (1 + nx0 + nx0 * ny0) * vc;
}
/// デストラクタ.
~Scalar3D() {
delete[] data;
}
/// データ領域の取得.
T* getData() const { return data; }
int getVCsize() { return vc0; }
/// インデックスファンクタの取得.
Index3DS getIndex() const { return index; }
/// 3次元添字によるデータアクセス.
T& operator() (int i, int j, int k) {
return data[i + c_j * j + c_k * k + c_0];
}
/// 3次元添字によるデータアクセス.
const T& operator() (int i, int j, int k) const {
return data[i + c_j * j + c_k * k + c_0];
}
/*
/// 直方体領域からバッファへのデータコピー(シリアライズ).
///
/// @param[in] i0,j0,k0 コピー元の直方体領域の起点
/// @param[in] nx,ny,nz コピー元の直方体領域のサイズ
/// @param[out] buffer コピー先バッファのアドレス
///
void copyToBuffer(int i0, int j0, int k0, int nx, int ny, int nz, T* buffer) const;
/// バッファから直方体領域へのデータコピー(デシリアライズ).
///
/// @param[in] i0,j0,k0 コピー先の直方体領域の起点
/// @param[in] nx,ny,nz コピー先の直方体領域のサイズ
/// @param[in] buffer コピー元バッファのアドレス
///
void copyFromBuffer(int i0, int j0, int k0, int nx, int ny, int nz, const T* buffer);
/// 他データクラスの直方体領域から直方体領域へのデータコピー.
///
/// @param[in] i0,j0,k0 コピー元の直方体領域の起点
/// @param[in] i1,j1,k1 コピー先の直方体領域の起点
/// @param[in] nx,ny,nz 直方体領域のサイズ(コピー元/コピー先で共通)
/// @param[in] dataClass コピー元データクラス
///
/// @todo (i0,j0,k0)と(i1,j1,k1)を逆した方が分かりやすい？
///
void copyFromDataClass(int i0, int j0, int k0, int i1, int j1, int k1,
int nx, int ny, int nz, const DataClass* dataClass);
*/
private:
/// コピーコンストラクタ(コピー禁止).
Scalar3D(const Scalar3D<T>& rhs);
/// 代入演算子(コピー禁止).
Scalar3D& operator=(const Scalar3D<T>& rhs);
/*
void copyToBuffer_0(int i0, int j0, int k0, int nx, int ny, int nz,
const T* data, Index3DS index, T* buffer) const;
void copyFromBuffer_0(int i0, int j0, int k0, int nx, int ny, int nz,
const T* buffer, T* data, Index3DS index);
void copyFromDataClass_0(int i0, int j0, int k0, int i1, int j1, int k1,
int nx, int ny, int nz,
const T* sData, Index3DS sIndex,
T* dData, Index3DS dIndex);
*/
#define USE_PRIVATE_METHODS
public:
/// 直方体領域からバッファへのデータコピー(シリアライズ).
void copyToBuffer(int i0, int j0, int k0, int nx, int ny, int nz, T* buffer) const
{
#ifdef USE_PRIVATE_METHODS
copyToBuffer_0(i0, j0, k0, nx, ny, nz, data, index, buffer);
#else
for (int k = k0; k < k0 + nz; ++k) {
for (int j = j0; j < j0 + ny; ++j) {
for (int i = i0; i < i0 + nx; ++i) {
buffer[i-i0 + nx*(j-j0) + (nx*ny)*(k-k0)] = data[index(i,j,k)];
}
}
}
#endif
}
/// バッファから直方体領域へのデータコピー(デシリアライズ).
void copyFromBuffer(int i0, int j0, int k0, int nx, int ny, int nz, const T* buffer)
{
#ifdef USE_PRIVATE_METHODS
copyFromBuffer_0(i0, j0, k0, nx, ny, nz, buffer, data, index);
#else
for (int k = k0; k < k0 + nz; ++k) {
for (int j = j0; j < j0 + ny; ++j) {
for (int i = i0; i < i0 + nx; ++i) {
data[index(i,j,k)] = buffer[i-i0 + nx*(j-j0) + (nx*ny)*(k-k0)];
}
}
}
#endif
}
/// 他データクラスの直方体領域から直方体領域へのデータコピー.
void copyFromDataClass(int i0, int j0, int k0, int i1, int j1, int k1,
int nx, int ny, int nz, const DataClass* dataClass)
{
const Scalar3D<T>* s = dynamic_cast<const Scalar3D<T>*>(dataClass);
T* sData = s->getData();
Index3DS sIndex = s->getIndex();
#ifdef USE_PRIVATE_METHODS
copyFromDataClass_0(i0, j0, k0, i1, j1, k1, nx, ny, nz, sData, sIndex, data, index);
#else
for (int k = 0; k < nz; ++k) {
for (int j = 0; j < ny; ++j) {
for (int i = 0; i < nx; ++i) {
data[index(i0+i,j0+j,k0+k)] = sData[sIndex(i1+i,j1+j,k1+k)];
}
}
}
#endif
}
private:
void copyToBuffer_0(int i0, int j0, int k0, int nx, int ny, int nz,
const T* data, Index3DS index, T* buffer) const
{
#ifdef _BLOCK_IS_LARGE_
#pragma omp parallel for
#else
#endif
for (int k = k0; k < k0 + nz; ++k) {
for (int j = j0; j < j0 + ny; ++j) {
for (int i = i0; i < i0 + nx; ++i) {
buffer[i-i0 + nx*(j-j0) + (nx*ny)*(k-k0)] = data[index(i,j,k)];
}
}
}
}
void copyFromBuffer_0(int i0, int j0, int k0, int nx, int ny, int nz,
const T* buffer, T* data, Index3DS index)
{
#ifdef _BLOCK_IS_LARGE_
#pragma omp parallel for
#else
#endif
for (int k = k0; k < k0 + nz; ++k) {
for (int j = j0; j < j0 + ny; ++j) {
for (int i = i0; i < i0 + nx; ++i) {
data[index(i,j,k)] = buffer[i-i0 + nx*(j-j0) + (nx*ny)*(k-k0)];
}
}
}
}
void copyFromDataClass_0(int i0, int j0, int k0, int i1, int j1, int k1,
int nx, int ny, int nz,
const T* sData, Index3DS sIndex,
T* dData, Index3DS dIndex)
{
#ifdef _BLOCK_IS_LARGE_
#pragma omp parallel for
#else
#endif
for (int k = 0; k < nz; ++k) {
for (int j = 0; j < ny; ++j) {
for (int i = 0; i < nx; ++i) {
// dData[dIndex(i0+i,j0+j,k0+k)] = sData[sIndex(i1+i,j1+j,k1+k)];
int ii = dIndex(i0+i,j0+j,k0+k);
int jj = sIndex(i1+i,j1+j,k1+k);
dData[ii] = sData[jj];
}
}
}
}
};
#ifdef BCMT_NAMESPACE
} // namespace BCMT_NAMESPACE
#endif
#endif // SCALAR_3D_H
|
GB_unop__erf_fp32_fp32.c | //------------------------------------------------------------------------------
// GB_unop: hard-coded functions for each built-in unary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_atomics.h"
#include "GB_unop__include.h"
// C=unop(A) is defined by the following types and operators:
// op(A) function: GB (_unop_apply__erf_fp32_fp32)
// op(A') function: GB (_unop_tran__erf_fp32_fp32)
// C type: float
// A type: float
// cast: float cij = aij
// unaryop: cij = erff (aij)
#define GB_ATYPE \
float
#define GB_CTYPE \
float
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
float aij = Ax [pA]
#define GB_CX(p) Cx [p]
// unary operator
#define GB_OP(z, x) \
z = erff (x) ;
// casting
#define GB_CAST(z, aij) \
float z = aij ;
// cij = op (aij)
#define GB_CAST_OP(pC,pA) \
{ \
/* aij = Ax [pA] */ \
float aij = Ax [pA] ; \
/* Cx [pC] = op (cast (aij)) */ \
float z = aij ; \
Cx [pC] = erff (z) ; \
}
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_ERF || GxB_NO_FP32)
//------------------------------------------------------------------------------
// Cx = op (cast (Ax)): apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_apply__erf_fp32_fp32)
(
float *Cx, // Cx and Ax may be aliased
const float *Ax,
const int8_t *restrict Ab, // A->b if A is bitmap
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
if (Ab == NULL)
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
float aij = Ax [p] ;
float z = aij ;
Cx [p] = erff (z) ;
}
}
else
{
// bitmap case, no transpose; A->b already memcpy'd into C->b
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!Ab [p]) continue ;
float aij = Ax [p] ;
float z = aij ;
Cx [p] = erff (z) ;
}
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (cast (A')): transpose, typecast, and apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_tran__erf_fp32_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
GB_binop__rdiv_fp32.c |
//------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__rdiv_fp32)
// A.*B function (eWiseMult): GB (_AemultB_08__rdiv_fp32)
// A.*B function (eWiseMult): GB (_AemultB_02__rdiv_fp32)
// A.*B function (eWiseMult): GB (_AemultB_04__rdiv_fp32)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__rdiv_fp32)
// A*D function (colscale): GB (_AxD__rdiv_fp32)
// D*A function (rowscale): GB (_DxB__rdiv_fp32)
// C+=B function (dense accum): GB (_Cdense_accumB__rdiv_fp32)
// C+=b function (dense accum): GB (_Cdense_accumb__rdiv_fp32)
// C+=A+B function (dense ewise3): GB (_Cdense_ewise3_accum__rdiv_fp32)
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__rdiv_fp32)
// C=scalar+B GB (_bind1st__rdiv_fp32)
// C=scalar+B' GB (_bind1st_tran__rdiv_fp32)
// C=A+scalar GB (_bind2nd__rdiv_fp32)
// C=A'+scalar GB (_bind2nd_tran__rdiv_fp32)
// C type: float
// A type: float
// A pattern? 0
// B type: float
// B pattern? 0
// BinaryOp: cij = (bij / aij)
#define GB_ATYPE \
float
#define GB_BTYPE \
float
#define GB_CTYPE \
float
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
float aij = GBX (Ax, pA, A_iso)
// true if values of A are not used
#define GB_A_IS_PATTERN \
0 \
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
float bij = GBX (Bx, pB, B_iso)
// true if values of B are not used
#define GB_B_IS_PATTERN \
0 \
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
float t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = (y / x) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_RDIV || GxB_NO_FP32 || GxB_NO_RDIV_FP32)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB (_Cdense_ewise3_accum__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
void GB (_Cdense_ewise3_noaccum__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_noaccum_template.c"
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__rdiv_fp32)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type float
float bwork = (*((float *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_AxD__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix D,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
float *restrict Cx = (float *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB (_DxB__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix D,
const GrB_Matrix B,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
float *restrict Cx = (float *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__rdiv_fp32)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool is_eWiseUnion,
const GB_void *alpha_scalar_in,
const GB_void *beta_scalar_in,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
float alpha_scalar ;
float beta_scalar ;
if (is_eWiseUnion)
{
alpha_scalar = (*((float *) alpha_scalar_in)) ;
beta_scalar = (*((float *) beta_scalar_in )) ;
}
#include "GB_add_template.c"
GB_FREE_WORKSPACE ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__rdiv_fp32)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__rdiv_fp32)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__rdiv_fp32)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
float *Cx = (float *) Cx_output ;
float x = (*((float *) x_input)) ;
float *Bx = (float *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
float bij = GBX (Bx, p, false) ;
Cx [p] = (bij / x) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__rdiv_fp32)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
float *Cx = (float *) Cx_output ;
float *Ax = (float *) Ax_input ;
float y = (*((float *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
float aij = GBX (Ax, p, false) ;
Cx [p] = (y / aij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
float aij = GBX (Ax, pA, false) ; \
Cx [pC] = (aij / x) ; \
}
GrB_Info GB (_bind1st_tran__rdiv_fp32)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
float
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
float x = (*((const float *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
float
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
float aij = GBX (Ax, pA, false) ; \
Cx [pC] = (y / aij) ; \
}
GrB_Info GB (_bind2nd_tran__rdiv_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
float y = (*((const float *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
GB_binop__isge_int16.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2020, All Rights Reserved.
// http://suitesparse.com See GraphBLAS/Doc/License.txt for license.
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_mkl.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB_AaddB__isge_int16
// A.*B function (eWiseMult): GB_AemultB__isge_int16
// A*D function (colscale): GB_AxD__isge_int16
// D*A function (rowscale): GB_DxB__isge_int16
// C+=B function (dense accum): GB_Cdense_accumB__isge_int16
// C+=b function (dense accum): GB_Cdense_accumb__isge_int16
// C+=A+B function (dense ewise3): (none)
// C=A+B function (dense ewise3): GB_Cdense_ewise3_noaccum__isge_int16
// C=scalar+B GB_bind1st__isge_int16
// C=scalar+B' GB_bind1st_tran__isge_int16
// C=A+scalar GB_bind2nd__isge_int16
// C=A'+scalar GB_bind2nd_tran__isge_int16
// C type: int16_t
// A type: int16_t
// B,b type: int16_t
// BinaryOp: cij = (aij >= bij)
#define GB_ATYPE \
int16_t
#define GB_BTYPE \
int16_t
#define GB_CTYPE \
int16_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int16_t aij = Ax [pA]
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB) \
int16_t bij = Bx [pB]
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int16_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA) \
cij = Ax [pA]
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB) \
cij = Bx [pB]
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z, x, y) \
z = (x >= y) ;
// op is second
#define GB_OP_IS_SECOND \
0
// op is plus_fp32 or plus_fp64
#define GB_OP_IS_PLUS_REAL \
0
// op is minus_fp32 or minus_fp64
#define GB_OP_IS_MINUS_REAL \
0
// GB_cblas_*axpy gateway routine, if it exists for this operator and type:
#define GB_CBLAS_AXPY \
(none)
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_ISGE || GxB_NO_INT16 || GxB_NO_ISGE_INT16)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void (none)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_ewise3_noaccum__isge_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumB__isge_int16
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumb__isge_int16
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int16_t
int16_t bwork = (*((int16_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_AxD__isge_int16
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_colscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
GrB_Info GB_DxB__isge_int16
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_rowscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
GrB_Info GB_AaddB__isge_int16
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_add_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB_AemultB__isge_int16
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB_bind1st__isge_int16
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *Cx = (int16_t *) Cx_output ;
int16_t x = (*((int16_t *) x_input)) ;
int16_t *Bx = (int16_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
int16_t bij = Bx [p] ;
Cx [p] = (x >= bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB_bind2nd__isge_int16
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int16_t *Cx = (int16_t *) Cx_output ;
int16_t *Ax = (int16_t *) Ax_input ;
int16_t y = (*((int16_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
int16_t aij = Ax [p] ;
Cx [p] = (aij >= y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typcasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (x >= aij) ; \
}
GrB_Info GB_bind1st_tran__isge_int16
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *GB_RESTRICT *Rowcounts,
GBI_single_iterator Iter,
const int64_t *GB_RESTRICT A_slice,
int naslice
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t x = (*((const int16_t *) x_input)) ;
#define GB_PHASE_2_OF_2
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typcasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (aij >= y) ; \
}
GrB_Info GB_bind2nd_tran__isge_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *GB_RESTRICT *Rowcounts,
GBI_single_iterator Iter,
const int64_t *GB_RESTRICT A_slice,
int naslice
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t y = (*((const int16_t *) y_input)) ;
#define GB_PHASE_2_OF_2
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
resource_strings.h | #pragma once
#include <torch/csrc/jit/code_template.h>
namespace torch {
namespace jit {
namespace fuser {
namespace cpu {
/*with type_as not checking type of its input, a fusion group can have non-fp32
tensor as input. Correct code for this case is generated, however, nvrtc does
not know how to handle int*_t integer types, so typedefs help it handle those
cases*/
static auto type_declarations_template = CodeTemplate(R"(
#define POS_INFINITY INFINITY
#define NEG_INFINITY -INFINITY
typedef ${IndexType} IndexType;
template<typename T, size_t N>
struct TensorInfo {
T* data;
IndexType sizes[N];
IndexType strides[N];
};
template<typename T>
struct TensorInfo<T, 0> {
T * data;
};
)");
static auto cpu_compilation_unit_template = CodeTemplate(R"(
#include <math.h>
#include <cstddef>
#include <cstdint>
double rsqrt(double x) {
return 1.0/sqrt(x);
}
float rsqrtf(float x) {
return 1.0f/sqrtf(x);
}
double frac(double x) {
return x - trunc(x);
}
float fracf(float x) {
return x - truncf(x);
}
${type_declarations}
#ifdef _MSC_VER
template<size_t n> struct int_of_size;
#define DEFINE_INT_OF_SIZE(int_t) \
template<> struct int_of_size<sizeof(int_t)> { using type = int_t; }
DEFINE_INT_OF_SIZE(int64_t);
DEFINE_INT_OF_SIZE(int32_t);
DEFINE_INT_OF_SIZE(int16_t);
DEFINE_INT_OF_SIZE(int8_t);
#undef DEFINE_INT_OF_SIZE
template <typename T>
using int_same_size_t = typename int_of_size<sizeof(T)>::type;
#define IndexTypeLoop int_same_size_t<IndexType>
#define ToIndexTypeLoop(x) static_cast<IndexTypeLoop>(x)
#else
#define IndexTypeLoop IndexType
#define ToIndexTypeLoop(x) x
#endif
#define OMP_THRESHOLD 100000
static void ${kernelName}_kernel(IndexType totalElements, ${formals}) {
#pragma omp parallel for if(totalElements > OMP_THRESHOLD)
for (IndexTypeLoop linearIndex = 0;
linearIndex < ToIndexTypeLoop(totalElements);
linearIndex += 1) {
// Convert `linearIndex` into an offset of tensor:
${tensorOffsets}
// calculate the results
${kernelBody}
}
}
#ifdef _WIN32
#define JIT_API __declspec(dllexport)
#else
#define JIT_API
#endif
extern "C"
JIT_API void ${kernelName}(IndexType totalElements, void ** args) {
${kernelName}_kernel(totalElements ${,argument_loads});
}
)");
} // namespace cpu
} // namespace fuser
} // namespace jit
} // namespace torch
|
softmax-inl.h | /*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
/*!
* Copyright (c) 2017 by Contributors
* \file softmax-inl.h
* \brief
*/
#ifndef MXNET_OPERATOR_NN_SOFTMAX_INL_H_
#define MXNET_OPERATOR_NN_SOFTMAX_INL_H_
#include <algorithm>
#include <string>
#include <utility>
#include <vector>
#include "../mxnet_op.h"
#include "../operator_common.h"
#include "../tensor/broadcast_reduce_op.h"
namespace mxnet {
namespace op {
namespace mxnet_op {
struct softmax_fwd {
template<typename AType>
MSHADOW_XINLINE static AType Map(float a, AType b) {
return AType(expf(a)/b);
}
template<typename AType>
MSHADOW_XINLINE static AType Map(double a, AType b) {
return AType(exp(a)/b);
}
};
struct log_softmax_fwd {
template<typename DType>
MSHADOW_XINLINE static float Map(DType a, float b) {
return a - logf(b);
}
template<typename DType>
MSHADOW_XINLINE static double Map(DType a, double b) {
return a - log(b);
}
};
template<typename OP, bool negate, typename AType, typename DType, typename OType, int ndim>
inline void Softmax(Stream<cpu> *s, DType *in, OType *out,
Shape<ndim> shape, int axis, const DType temperature) {
index_t M = shape[axis];
index_t N = shape.Size()/M;
Shape<ndim> stride = calc_stride(shape);
Shape<ndim> sshape = shape;
sshape[axis] = 1;
index_t sa = stride[axis];
#pragma omp parallel for
for (int i = 0; i < static_cast<int>(N); ++i) {
index_t base = unravel_dot(i, sshape, stride);
DType mmax = negate ? -in[base] : in[base];
DType val;
for (index_t j = 1; j < M; ++j) {
val = negate ? -in[base + j*sa] : in[base + j*sa];
if (mmax < val) mmax = val;
}
AType sum = AType(0);
DType in_val;
// By default temperature is 1.0.
// Adding a branch here to save the CPU 'divide-by-1' computation at runtime
if (temperature == 1.0) {
for (index_t j = 0; j < M; ++j) {
in_val = negate ? -in[base + j*sa] : in[base + j*sa];
sum += std::exp(in_val - mmax);
}
for (index_t j = 0; j < M; ++j) {
in_val = negate ? -in[base + j*sa] : in[base + j*sa];
out[base + j*sa] = OP::Map(in_val - mmax, sum);
}
} else {
for (index_t j = 0; j < M; ++j) {
in_val = negate ? -in[base + j*sa] : in[base + j*sa];
sum += std::exp((in_val - mmax)/temperature);
}
for (index_t j = 0; j < M; ++j) {
in_val = negate ? -in[base + j*sa] : in[base + j*sa];
out[base + j*sa] = OP::Map((in_val - mmax)/temperature, sum);
}
}
}
}
struct softmax_bwd {
template<typename DType, typename AType>
MSHADOW_XINLINE static AType Map(DType ograd, DType out, AType sum) {
return AType(out * (ograd - sum));
}
};
struct log_softmax_bwd {
template<typename AType>
MSHADOW_XINLINE static AType Map(float ograd, float out, AType sum) {
return AType(ograd - expf(out)*sum);
}
template<typename AType>
MSHADOW_XINLINE static AType Map(double ograd, double out, AType sum) {
return AType(ograd - exp(out)*sum);
}
};
template<typename OP1, typename OP2, int Req, bool negate,
typename AType, typename DType, typename OType, int ndim>
inline void SoftmaxGrad(Stream<cpu> *s, OType *out, OType *ograd,
DType *igrad, Shape<ndim> shape, int axis,
const DType temperature) {
index_t M = shape[axis];
index_t N = shape.Size()/M;
Shape<ndim> stride = calc_stride(shape);
Shape<ndim> sshape = shape;
sshape[axis] = 1;
index_t sa = stride[axis];
#pragma omp parallel for
for (int i = 0; i < static_cast<int>(N); ++i) {
index_t base = unravel_dot(i, sshape, stride);
AType sum = AType(0);
for (index_t j = 0; j < M; ++j) {
sum += OP1::Map(ograd[base + j*sa], out[base + j*sa]);
}
// By default temperature is 1.0.
// Adding a branch here to save the CPU 'divide-by-1' computation at runtime
DType final_result;
if (temperature == 1.0) {
for (index_t j = 0; j < M; ++j) {
final_result = negate ?
-OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) :
OP2::Map(ograd[base + j*sa], out[base + j*sa], sum);
KERNEL_ASSIGN(igrad[base + j*sa], Req, final_result);
}
} else {
for (index_t j = 0; j < M; ++j) {
final_result = negate ?
-OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) / temperature :
OP2::Map(ograd[base + j*sa], out[base + j*sa], sum) / temperature;
KERNEL_ASSIGN(igrad[base + j*sa], Req, final_result);
}
}
}
}
#ifdef __CUDACC__
template<int x_bits, typename OP, bool negate, typename AType, int ndim,
typename DType, typename OType>
__global__ void softmax_compute_kernel(DType *in, OType *out, index_t M, int axis,
Shape<ndim> sshape, Shape<ndim> stride,
const double temperature) {
const unsigned x_size = 1 << x_bits;
__shared__ AType smem[x_size];
index_t sa = stride[axis];
index_t base = unravel_dot(blockIdx.x, sshape, stride);
index_t x = threadIdx.x;
red::maximum::SetInitValue(smem[x]);
for (index_t i = x; i < M; i += x_size) {
smem[x] = ::max(smem[x], negate ? -in[base + i*sa] : in[base + i*sa]);
}
__syncthreads();
cuda::Reduce1D<red::maximum, x_bits>(smem);
__syncthreads();
DType smax = smem[0];
__syncthreads();
red::sum::SetInitValue(smem[x]);
DType val;
for (index_t i = x; i < M; i += x_size) {
val = negate ? -in[base + i*sa]:in[base + i*sa];
smem[x] += static_cast<AType>(expf((val - smax) / static_cast<AType>(temperature)));
}
__syncthreads();
cuda::Reduce1D<red::sum, x_bits>(smem);
__syncthreads();
AType ssum = smem[0];
__syncthreads();
for (index_t i = x; i < M; i += x_size) {
val = negate ? -in[base + i*sa] : in[base + i*sa];
out[base + i*sa] = OP::Map((val - smax)/static_cast<DType>(temperature), ssum);
}
}
template<typename OP, bool negate, typename AType, typename DType, typename OType, int ndim>
inline void Softmax(Stream<gpu> *s, DType *in, OType *out,
Shape<ndim> shape, int axis, const double temperature) {
const int x_bits = 7;
const int x_size = 1 << x_bits;
index_t M = shape[axis];
index_t N = shape.Size()/M;
Shape<ndim> stride = calc_stride(shape);
Shape<ndim> sshape = shape;
sshape[axis] = 1;
softmax_compute_kernel<x_bits, OP, negate, AType, ndim>
<<<N, x_size, 0, mshadow::Stream<gpu>::GetStream(s)>>>(
in, out, M, axis, sshape, stride, temperature);
MSHADOW_CUDA_POST_KERNEL_CHECK(softmax_compute_kernel);
}
template<int x_bits, typename OP1, typename OP2, int Req, bool negate, typename AType, int ndim,
typename DType, typename OType>
__global__ void softmax_gradient_kernel(OType *out, OType *ograd, DType *igrad,
index_t M, int axis, Shape<ndim> sshape,
Shape<ndim> stride, const double temperature) {
const unsigned x_size = 1 << x_bits;
__shared__ AType smem[x_size];
index_t sa = stride[axis];
index_t base = unravel_dot(blockIdx.x, sshape, stride);
index_t x = threadIdx.x;
red::sum::SetInitValue(smem[x]);
for (index_t i = x; i < M; i += x_size) {
smem[x] += OP1::Map(ograd[base + i*sa], out[base + i*sa]);
}
__syncthreads();
cuda::Reduce1D<red::sum, x_bits>(smem);
__syncthreads();
AType ssum = smem[0];
__syncthreads();
DType final_result;
for (index_t i = x; i < M; i += x_size) {
final_result =
negate ?
-OP2::Map(ograd[base + i*sa], out[base + i*sa], ssum) :
OP2::Map(ograd[base + i*sa], out[base + i*sa], ssum);
KERNEL_ASSIGN(igrad[base + i*sa], Req, final_result / static_cast<DType>(temperature));
}
}
template<typename OP1, typename OP2, int Req, bool negate, typename AType, int ndim,
typename DType, typename OType>
inline void SoftmaxGrad(Stream<gpu> *s, OType *out, OType *ograd,
DType *igrad, Shape<ndim> shape, int axis,
const double temperature) {
const int x_bits = 7;
const int x_size = 1 << x_bits;
index_t M = shape[axis];
index_t N = shape.Size()/M;
Shape<ndim> stride = calc_stride(shape);
Shape<ndim> sshape = shape;
sshape[axis] = 1;
softmax_gradient_kernel<x_bits, OP1, OP2, Req, negate, AType, ndim>
<<<N, x_size, 0, mshadow::Stream<gpu>::GetStream(s)>>>(
out, ograd, igrad, M, axis, sshape, stride, temperature);
MSHADOW_CUDA_POST_KERNEL_CHECK(softmax_gradient_kernel);
}
#endif
} // namespace mxnet_op
struct SoftmaxParam : public dmlc::Parameter<SoftmaxParam> {
int axis;
dmlc::optional<double> temperature;
dmlc::optional<int> dtype;
DMLC_DECLARE_PARAMETER(SoftmaxParam) {
DMLC_DECLARE_FIELD(axis).set_default(-1)
.describe("The axis along which to compute softmax.");
DMLC_DECLARE_FIELD(temperature).set_default(dmlc::optional<double>())
.describe("Temperature parameter in softmax");
DMLC_DECLARE_FIELD(dtype)
.add_enum("float16", mshadow::kFloat16)
.add_enum("float32", mshadow::kFloat32)
.add_enum("float64", mshadow::kFloat64)
.set_default(dmlc::optional<int>())
.describe("DType of the output in case this can't be inferred. "
"Defaults to the same as input's dtype if not defined (dtype=None).");
}
};
static inline bool softmax_has_dtype_override(const nnvm::NodeAttrs& attrs) {
const SoftmaxParam& param = nnvm::get<SoftmaxParam>(attrs.parsed);
return param.dtype.has_value() && param.dtype.value() != -1;
}
static inline bool SoftmaxOpType(const nnvm::NodeAttrs& attrs,
std::vector<int>* in_attrs,
std::vector<int>* out_attrs) {
CHECK_EQ(in_attrs->size(), 1);
CHECK_EQ(out_attrs->size(), 1);
const SoftmaxParam& param = nnvm::get<SoftmaxParam>(attrs.parsed);
if (softmax_has_dtype_override(attrs)) {
TYPE_ASSIGN_CHECK(*out_attrs, 0, param.dtype.value());
type_assign(&(*in_attrs)[0], (*out_attrs)[0]);
return true;
} else {
return ElemwiseType<1, 1>(attrs, in_attrs, out_attrs);
}
}
static inline bool SoftmaxGradOpShape(const nnvm::NodeAttrs& attrs,
mxnet::ShapeVector *in_attrs,
mxnet::ShapeVector *out_attrs) {
if (softmax_has_dtype_override(attrs)) {
return ElemwiseShape<3, 1>(attrs, in_attrs, out_attrs);
} else {
return ElemwiseShape<2, 1>(attrs, in_attrs, out_attrs);
}
}
static inline bool SoftmaxGradOpType(const nnvm::NodeAttrs& attrs,
std::vector<int>* in_attrs,
std::vector<int>* out_attrs) {
CHECK_EQ(out_attrs->size(), 1);
if (softmax_has_dtype_override(attrs)) {
CHECK_EQ(in_attrs->size(), 3);
int in_dtype = (*in_attrs)[1];
int out_dtype = (*in_attrs)[2];
TYPE_ASSIGN_CHECK(*in_attrs, 0, out_dtype);
TYPE_ASSIGN_CHECK(*out_attrs, 0, in_dtype);
return (*out_attrs)[0] != -1 && (*in_attrs)[0] != -1;
} else {
CHECK_EQ(in_attrs->size(), 2);
int out_dtype = (*in_attrs)[1];
TYPE_ASSIGN_CHECK(*out_attrs, 0, out_dtype);
TYPE_ASSIGN_CHECK(*in_attrs, 0, out_dtype);
return (*out_attrs)[0] != -1 && (*in_attrs)[0] != -1;
}
}
static inline std::vector<std::pair<int, int> >
SoftmaxGradOpInplaceOption(const nnvm::NodeAttrs& attrs) {
if (softmax_has_dtype_override(attrs)) {
return std::vector<std::pair<int, int> >{{0, 0}, {1, 0}, {2, 0}};
} else {
return std::vector<std::pair<int, int> >{{0, 0}, {1, 0}};
}
}
static inline uint32_t SoftmaxGradOpNumInputs(const nnvm::NodeAttrs& attrs) {
return softmax_has_dtype_override(attrs) ? 3 : 2;
}
static inline std::vector<std::string> SoftmaxGradOpInputNames(const nnvm::NodeAttrs& attrs) {
if (softmax_has_dtype_override(attrs)) {
return std::vector<std::string>{"ograd", "data", "output"};
} else {
return std::vector<std::string>{"ograd", "output"};
}
}
struct SoftmaxFGradient {
const char *op_name;
std::vector<nnvm::NodeEntry> operator()(const nnvm::NodePtr& n,
const std::vector<nnvm::NodeEntry>& ograds) const {
if (softmax_has_dtype_override(n->attrs)) {
return ElemwiseGradUseInOut {op_name}(n, ograds);
} else {
return ElemwiseGradUseOut {op_name}(n, ograds);
}
}
};
template<typename xpu, typename OP, bool negate = false>
void SoftmaxCompute(const nnvm::NodeAttrs& attrs,
const OpContext& ctx,
const std::vector<TBlob>& inputs,
const std::vector<OpReqType>& req,
const std::vector<TBlob>& outputs) {
using namespace mxnet_op;
if (req[0] == kNullOp) return;
CHECK_NE(req[0], kAddTo);
const SoftmaxParam& param = nnvm::get<SoftmaxParam>(attrs.parsed);
int axis = CheckAxis(param.axis, inputs[0].ndim());
const double temperature = param.temperature.has_value() ?
param.temperature.value() : 1.0;
mxnet::TShape shape = AxisShapeCompact(inputs[0].shape_, &axis, true);
MXNET_REAL_ACC_TYPE_SWITCH(inputs[0].type_flag_, DType, AType, {
MSHADOW_REAL_TYPE_SWITCH(outputs[0].type_flag_, OType, {
if (shape.ndim() == 2) {
Softmax<OP, negate, AType>(
ctx.get_stream<xpu>(), inputs[0].dptr<DType>(),
outputs[0].dptr<OType>(), shape.get<2>(), axis,
static_cast<DType>(temperature));
} else {
Softmax<OP, negate, AType>(
ctx.get_stream<xpu>(), inputs[0].dptr<DType>(),
outputs[0].dptr<OType>(), shape.get<3>(), axis,
static_cast<DType>(temperature));
}
});
});
}
template<typename xpu, typename OP1, typename OP2, bool negate = false>
void SoftmaxGradCompute(const nnvm::NodeAttrs& attrs,
const OpContext& ctx,
const std::vector<TBlob>& inputs,
const std::vector<OpReqType>& req,
const std::vector<TBlob>& outputs) {
using namespace mxnet_op;
if (req[0] == kNullOp) return;
const SoftmaxParam& param = nnvm::get<SoftmaxParam>(attrs.parsed);
int axis = CheckAxis(param.axis, inputs[0].ndim());
const double temperature = param.temperature.has_value() ?
param.temperature.value() : 1.0;
mxnet::TShape shape = AxisShapeCompact(inputs[0].shape_, &axis, true);
int out_idx = softmax_has_dtype_override(attrs) ? 2 : 1;
MXNET_REAL_ACC_TYPE_SWITCH(inputs[0].type_flag_, OType, AType, {
MSHADOW_REAL_TYPE_SWITCH(outputs[0].type_flag_, DType, {
MXNET_ASSIGN_REQ_SWITCH(req[0], Req, {
if (shape.ndim() == 2) {
SoftmaxGrad<OP1, OP2, Req, negate, AType>(
ctx.get_stream<xpu>(), inputs[out_idx].dptr<OType>(),
inputs[0].dptr<OType>(), outputs[0].dptr<DType>(),
shape.get<2>(), axis, static_cast<DType>(temperature));
} else {
SoftmaxGrad<OP1, OP2, Req, negate, AType>(
ctx.get_stream<xpu>(), inputs[out_idx].dptr<OType>(),
inputs[0].dptr<OType>(), outputs[0].dptr<DType>(),
shape.get<3>(), axis, static_cast<DType>(temperature));
}
});
});
});
}
} // namespace op
} // namespace mxnet
#endif // MXNET_OPERATOR_NN_SOFTMAX_INL_H_
|
LookupTable.h | #ifndef _LOOKUPTABLE_H_
#define _LOOKUPTABLE_H_
/*
* LookupTable.h:
* Lookup operation, for embeddings
*
* Created on: Apr 22, 2017
* Author: mszhang
*/
#include "SparseParam.h"
#include "MyLib.h"
#include "Alphabet.h"
#include "Node.h"
#include "Graph.h"
#include "ModelUpdate.h"
class LookupTable {
public:
PAlphabet elems;
SparseParam E;
bool bFineTune;
int nDim;
int nVSize;
int nUNKId;
public:
LookupTable() {
nVSize = 0;
nDim = 0;
elems = NULL;
nUNKId = -1;
bFineTune = false;
}
//random initialization
inline void initial(PAlphabet alpha, int dim, bool fineTune = true) {
elems = alpha;
nVSize = elems->size();
nUNKId = elems->from_string(unknownkey);
initialWeights(dim, fineTune);
}
//initialization by pre-trained embeddings
inline bool initial(PAlphabet alpha, const string& inFile, bool fineTune = true, bool bNormalize = true) {
elems = alpha;
nVSize = elems->size();
nUNKId = elems->from_string(unknownkey);
return initialWeights(inFile, fineTune, bNormalize);
}
inline void initialWeights(int dim, bool tune) {
if (nVSize == 0) {
std::cout << "please check the alphabet" << std::endl;
return;
}
nDim = dim;
E.initial(nDim, nVSize);
//E.val.norm2one();
bFineTune = tune;
}
// default should be fineTune, just for initialization
inline bool initialWeights(const string& inFile, bool tune, bool normalize = true) {
if (nVSize == 0 || !elems->is_fixed()) {
std::cout << "please check the alphabet" << std::endl;
return false;
}
ifstream inf;
if (inf.is_open()) {
inf.close();
inf.clear();
}
inf.open(inFile.c_str());
string strLine, curWord;
int wordId;
vector<string> sLines;
sLines.clear();
while (1) {
if (!my_getline(inf, strLine)) {
break;
}
if (!strLine.empty()) {
sLines.push_back(strLine);
}
}
inf.close();
if (sLines.size() == 0) {
return false;
}
//find the first line, decide the wordDim;
vector<string> vecInfo;
split_bychar(sLines[0], vecInfo, ' ');
nDim = vecInfo.size() - 1;
E.initial(nDim, nVSize);
//E.val = 0;
DEV->zero(E.val);
std::cout << "word embedding dim is " << nDim << std::endl;
bool bHasUnknown = false;
unordered_set<int> indexers;
NRVec<dtype> sum(nDim);
sum = 0.0;
int count = 0;
int max_size = nDim * nVSize;
dtype* E_v = new dtype[max_size];
for(int idx = 0; idx < max_size; idx++) {
E_v[idx] = 0;
}
for (int idx = 0; idx < sLines.size(); idx++) {
split_bychar(sLines[idx], vecInfo, ' ');
if (vecInfo.size() != nDim + 1) {
std::cout << "error embedding file" << std::endl;
}
curWord = vecInfo[0];
//we assume the keys are normalized
wordId = elems->from_string(curWord);
if (wordId >= 0) {
count++;
if (nUNKId == wordId) {
bHasUnknown = true;
}
indexers.insert(wordId);
vector<dtype> cur_data(nDim);
for (int idy = 0; idy < nDim; idy++) {
dtype curValue = atof(vecInfo[idy + 1].c_str());
sum[idy] += curValue;
cur_data[idy] = curValue;
E_v[nDim * wordId + idy] += curValue;
//E.val[wordId][idy] += curValue;
}
}
}
if (count == 0) {
//E.val.random(sqrt(3.0 / nDim));
dtype val = sqrt(3.0 / nDim);
DEV->random_uniform(E.val, E.val.shape(), -val, val);
std::cout << "find no overlapped lexicons in the embedding file" << std::endl;
return false;
}
if (nUNKId >= 0 && !bHasUnknown) {
for (int idx = 0; idx < nDim; idx++) {
E_v[nUNKId * nDim + idx] = sum[idx] / (count + 1);
}
indexers.insert(nUNKId);
count++;
std::cout << unknownkey << " not found, using averaged value to initialize." << std::endl;
}
int oovWords = 0;
for (int id = 0; id < nVSize; id++) {
if (indexers.find(id) == indexers.end()) {
oovWords++;
for (int idy = 0; idy < nDim; idy++) {
E_v[id * nDim + idy] = nUNKId >= 0 ? E_v[nUNKId * nDim + idy] : sum[idy] / count;
}
}
}
std::cout << "OOV num is " << oovWords << ", total num is " << nVSize << ", embedding oov ratio is " << oovWords * 1.0 / nVSize << std::endl;
std::cout << "unknown id" << nUNKId << std::endl;
bFineTune = tune;
if (normalize) {
//E.val.norm2one();
}
DEV->set(E.val, E_v, nDim * nVSize);
delete []E_v;
return true;
}
inline void exportAdaParams(ModelUpdate& ada) {
if (bFineTune) {
ada.addParam(&E);
}
}
inline int getElemId(const string& strFeat) {
return elems->from_string(strFeat);
}
inline void save(std::ofstream &os) const {
E.save(os);
os << bFineTune << std::endl;
os << nDim << std::endl;
os << nVSize << std::endl;
os << nUNKId << std::endl;
}
//set alpha directly
inline void load(std::ifstream &is, PAlphabet alpha) {
E.load(is);
is >> bFineTune;
is >> nDim;
is >> nVSize;
is >> nUNKId;
elems = alpha;
}
};
class LookupNode : public Node {
public:
LookupTable* param;
int xid;
public:
LookupNode() {
xid = -1;
param = NULL;
node_type = "lookup";
}
inline void setParam(LookupTable* paramInit) {
param = paramInit;
}
inline void clearValue() {
Node::clearValue();
xid = -1;
}
public:
//notice the output
//this should be leaf nodes
void forward(Graph *cg, const string& strNorm) {
assert(param != NULL);
xid = param->getElemId(strNorm);
if (xid < 0 && param->nUNKId >= 0) {
xid = param->nUNKId;
}
if (param->bFineTune && xid < 0) {
std::cout << "Caution: unknown words are not modeled !" << std::endl;
}
degree = 0;
cg->addNode(this);
}
public:
inline PExecute generate(bool bTrain);
// better to rewrite for deep understanding
inline bool typeEqual(PNode other) {
bool result = Node::typeEqual(other);
if (!result) return false;
LookupNode* conv_other = (LookupNode*)other;
if (param != conv_other->param) {
return false;
}
return true;
}
// for which do no require merge
public:
void compute() {
if (xid >= 0) {
param->E.value(xid, val);
//cout << "shape: " << param->E.val.shape().to_string() << endl;
//cout << "id: "<< xid << endl;
} else {
//val.zero();
DEV->zero(val);
}
}
void backward() {
assert(param != NULL);
if (xid == param->nUNKId || (xid >= 0 && param->bFineTune)) {
param->E.loss(xid, loss);
}
}
};
//#if USE_GPU
//class LookupExecute :public Execute {
//public:
// bool bTrain;
//public:
// inline void forward() {
// int count = batch.size();
//
// for (int idx = 0; idx < count; idx++) {
// LookupNode* ptr = (LookupNode*)batch[idx];
// ptr->compute();
// ptr->forward_drop(bTrain);
// }
// }
//
// inline void backward() {
// int count = batch.size();
// for (int idx = 0; idx < count; idx++) {
// LookupNode* ptr = (LookupNode*)batch[idx];
// ptr->backward_drop();
// ptr->backward();
// }
// }
//};
//
//
//inline PExecute LookupNode::generate(bool bTrain) {
// LookupExecute* exec = new LookupExecute();
// exec->batch.push_back(this);
// exec->bTrain = bTrain;
// return exec;
//}
//#else
class LookupExecute :public Execute {
public:
bool bTrain;
//LDG::Tensor y;
public:
inline void forward() {
int count = batch.size();
//#pragma omp parallel for schedule(static,1)
vector<int> vec_indexes(count);
vector<LDG::PTensor> vec_val(count);
for (int idx = 0; idx < count; idx++) {
LookupNode* ptr = (LookupNode*)batch[idx];
vec_indexes[idx] = (ptr->xid);
vec_val[idx] = (&ptr->val);
ptr->degree = -1;
}
LookupNode* ptr = (LookupNode*)batch[0];
DEV->FLookup(ptr->param->E.val, vec_indexes, vec_val);
drop_value = ptr->drop_value;
if(drop_value > 0) {
if(bTrain)
DEV->Fdropout(vec_val, drop_value, mask, vec_val);
else
DEV->Fdropout(vec_val, drop_value, vec_val);
}
/*
for (int idx = 0; idx < count; idx++) {
LookupNode* ptr = (LookupNode*)batch[idx];
ptr->forward_drop(bTrain);
}
*/
}
inline void backward() {
int count = batch.size();
//#pragma omp parallel for schedule(static,1)
for (int idx = 0; idx < count; idx++) {
LookupNode* ptr = (LookupNode*)batch[idx];
// ptr->backward_drop();
//ptr->backward();
}
vector<LDG::PTensor> vec_loss;
vector<int> vec_xid;
for (int idx = 0; idx < count; idx++) {
LookupNode* ptr = (LookupNode*)batch[idx];
if(ptr->xid == ptr->param->nUNKId || (ptr->xid >= 0 && ptr->param->bFineTune)) {
vec_loss.push_back(&ptr->loss);
vec_xid.push_back(ptr->xid);
ptr->param->E.indexers[ptr->xid] = true;
}
}
LookupNode* ptr = (LookupNode*)batch[0];
if(vec_xid.size() > 0) {
if (drop_value > 0) {
DEV->Ddropout(vec_loss, mask);
}
DEV->DLookup(ptr->param->E.grad, vec_xid, vec_loss);
}
}
};
inline PExecute LookupNode::generate(bool bTrain) {
LookupExecute* exec = new LookupExecute();
exec->batch.push_back(this);
exec->bTrain = bTrain;
return exec;
}
//#endif
#endif /*_LOOKUPTABLE_H*/
|
strass.c | /*
* Matrices multiplication algorithms: a simple, strassen, and Intel BLAS
*
* This file is part of solution of a Intel Winter Summer School problem
* Copyright (c) 2010 Roman Tsisyk <roman@tsisyk.com>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following disclaimer
* in the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
*/
#include "benchmark.h"
#include <stddef.h>
#include <stdlib.h>
#include <stdio.h>
#include <omp.h>
/*
* TODO: malloc checks
*/
/*
* We use strassen algorithm if product size less that kMinStrassen
*/
static const size_t kMinStrassen = 32 * 32; // best on tested 2 x Xeon E5440
/*
* TODO: use col-major arrays for B matrix, its may improve prefetching
* TODO: replace at_r with inline functions
* _r functions for row-major arrays
*/
#define at_r(M, i, j) (M + (i * rstride##M + j))
#define at_r_ref(M, i, j) (*at_r(M, i, j))
void matmul_init() {
// omp_set_dynamic(1);
}
void matmul_set_num_threads(size_t count) {
omp_set_num_threads(count);
}
void matmul_fini() {
#ifdef WITH_MKL
mkl_free_buffers();
#endif
}
/*
void matmul_debug_print(data_t *A, size_t rstrideA, size_t height, size_t width) {
const data_t *end0 = A + height * rstrideA;
for (; A < end0; A += rstrideA) {
const data_t *end1 = A + width;
data_t *a = A;
for (; a < end1; a++) {
printf("%d ", *a);
}
printf("\n");
}
}
*/
inline void matmul_matmul(size_t heightA, size_t widthA, size_t widthB,
data_t *A, size_t rstrideA,
data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
// I dont want to use MKL here, please call directly matmul_mkl
if(heightA * widthB <= kMinStrassen)
{
matmul_simple(heightA, widthA, widthB, A, rstrideA, B, rstrideB, C, rstrideC);
//matmul_recursive_tile_caller(A,B,C,heightA,widthA,widthB,rstrideA);
}
else
{
matmul_strassen(heightA, widthA, widthB, A, rstrideA, B, rstrideB, C, rstrideC);
}
}
void matmul_matmul_caller( data_t *A, data_t *B, data_t *C, int rstrideA, int rstrideB, int rstrideC, int BW)
{
#pragma omp parallel shared(rstrideA, rstrideB, rstrideC, A, B, C)
{
#pragma omp single
{
{
matmul_matmul(rstrideA, rstrideB, rstrideC, A, rstrideA, B, rstrideB, C, rstrideC);\
}
}
}
}
/*
* Sum of matrices
*/
static void matmul_add_r(data_t *A, size_t rstrideA, data_t *B, size_t rstrideB,
data_t *R, size_t rstrideR, size_t height, size_t width) {
const data_t *end0 = A + height * rstrideA;
// its making no sense to parallelize here
for (; A < end0; A += rstrideA, B += rstrideB, R += rstrideR) {
const data_t *end1 = A + width;
data_t *a = A;
data_t *b = B;
data_t *r = R;
for (; a < end1; a++, b++, r++) {
*r = *a + *b;
}
}
}
/*
* Substraction of matrices
*/
static void matmul_sub_r(data_t *A, size_t rstrideA, data_t *B, size_t rstrideB,
data_t *R, size_t rstrideR, size_t height, size_t width) {
const data_t *end0 = A + height * rstrideA;
// its making no sense to parallelize here
for (; A < end0; A += rstrideA, B += rstrideB, R += rstrideR) {
const data_t *end1 = A + width;
data_t *a = A;
data_t *b = B;
data_t *r = R;
for (; a < end1; a++, b++, r++) {
*r = *a - *b;
}
}
}
/*
* Strassen P1 helper
* P = (A11 + A22) * (B11 + B22)
*/
static void matmul_strassen_P1 (size_t heightA, size_t widthA, size_t widthB,
data_t *A1,
data_t *A2,
size_t rstrideA,
data_t *B1,
data_t *B2,
size_t rstrideB,
data_t *P, size_t rstrideP)
{
const size_t heightB = widthA;
data_t *F = (data_t *) malloc(sizeof(data_t) * heightA * widthA);
const size_t rstrideF = widthA;
data_t *S = (data_t *) malloc(sizeof(data_t) * heightB * widthB);
const size_t rstrideS = widthB;
//#pragma omp parallel sections
{
//#pragma omp section
matmul_add_r(A1, rstrideA, A2, rstrideA, F, rstrideF, heightA, widthA);
//#pragma omp section
matmul_add_r(B1, rstrideB, B2, rstrideB, S, rstrideS, heightB, widthB);
}
// start recursion here
matmul_matmul(heightA, widthA, widthB, F, rstrideF, S, rstrideS, P,
rstrideP);
free(S);
free(F);
}
/*
* Strassen P2 and P5 helper
* P = (A1 + A2) * B
*/
static void matmul_strassen_P2_P5 (size_t heightA, size_t widthA, size_t widthB,
data_t *A1,
data_t *A2,
size_t rstrideA,
data_t *B,
size_t rstrideB,
data_t *P, size_t rstrideP)
{
data_t *F = (data_t *) malloc(sizeof(data_t) * heightA * widthA);
const size_t rstrideF = widthA;
matmul_add_r(A1, rstrideA, A2, rstrideA, F, rstrideF, heightA, widthA);
// start recursion here
matmul_matmul(heightA, widthA, widthB, F, rstrideF, B, rstrideB, P,
rstrideP);
free(F);
}
/*
* Strassen P3 and P4 helper
* P = A * (B1 - B2)
*/
static void matmul_strassen_P3_P4 (size_t heightA, size_t widthA, size_t widthB,
data_t *A,
size_t rstrideA,
data_t *B1,
data_t *B2,
size_t rstrideB,
data_t *P, size_t rstrideP)
{
const size_t heightB = widthA;
data_t *S = (data_t *) malloc(sizeof(data_t) * heightB * widthB);
const size_t rstrideS = widthB;
matmul_sub_r(B1, rstrideB, B2, rstrideB, S, rstrideS, heightB, widthB);
// start recursion here
matmul_matmul(heightA, widthA, widthB, A, rstrideA, S, rstrideS, P,
rstrideP);
free(S);
}
/*
* Strassen P6 and P7 helper
* P = (A1 - A2) * (B1 + B2)
*/
static void matmul_strassen_P6_P7 (size_t heightA, size_t widthA, size_t widthB,
data_t *A1,
data_t *A2,
size_t rstrideA,
data_t *B1,
data_t *B2,
size_t rstrideB,
data_t *P, size_t rstrideP)
{
const size_t heightB = widthA;
// Linux malloc is enough fast => not parallelize it below
data_t *F = (data_t *) malloc(sizeof(data_t) * heightA * widthA);
const size_t rstrideF = widthA;
data_t *S = (data_t *) malloc(sizeof(data_t) * heightB * widthB);
const size_t rstrideS = widthB;
//#pragma omp task
matmul_sub_r(A1, rstrideA, A2, rstrideA, F, rstrideF, heightA, widthA);
//#pragma omp task
matmul_add_r(B1, rstrideB, B2, rstrideB, S, rstrideS, heightB, widthB);
//#pragma omp taskwait
// start recursion here
matmul_matmul(heightA, widthA, widthB, F, rstrideF, S, rstrideS, P,
rstrideP);
free(S);
free(F);
}
/*
* Strassen C11 and C22 helper
* C = ([P1=C] + P2) + (P3 - P4)
*/
static void matmul_strassen_C11_C22 (
data_t *P1,
size_t rstrideP1,
data_t *P2,
size_t rstrideP2,
data_t *P3,
size_t rstrideP3,
data_t *P4,
size_t rstrideP4,
size_t heightP, size_t widthP)
{
data_t *S = (data_t *) malloc(sizeof(data_t) * heightP * widthP);
const size_t rstrideS = widthP;
//#pragma omp task
matmul_add_r(P1, rstrideP1, P2, rstrideP2, P1, rstrideP1, heightP, widthP);
//#pragma omp task
matmul_sub_r(P3, rstrideP3, P4, rstrideP4, S, rstrideS, heightP, widthP);
//#pragma omp taskwait
// sync
matmul_add_r(P1, rstrideP1, S, rstrideS, P1, rstrideP1, heightP, widthP);
free(S);
}
/*
* Strassen C12 and C21 helper
* C = [P1=C] + P2
*/
static void matmul_strassen_C12_C21 (
data_t *P1,
size_t rstrideP1,
data_t *P2,
size_t rstrideP2,
size_t heightP, size_t widthP)
{
matmul_add_r(P1, rstrideP1, P2, rstrideP2, P1, rstrideP1, heightP, widthP);
}
static void matmul_strassen_fix_heightA_odd(size_t heightA, size_t widthA,
size_t widthB, data_t *A, size_t rstrideA, data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
A += (heightA - 1) * rstrideA;
C += (heightA - 1) * rstrideC;
matmul_simple(1, widthA, widthB, A, rstrideA, B, rstrideB, C, rstrideC);
}
static void matmul_strassen_fix_widthB_odd(size_t heightA, size_t widthA, size_t widthB,
data_t *A, size_t rstrideA,
data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
B += (widthB - 1);
C += (widthB - 1);
// edge was fixed by fix_heightA
heightA &= ~1;
matmul_simple(heightA, widthA, 1, A, rstrideA, B, rstrideB, C, rstrideC);
}
static void matmul_strassen_fix_widthA_odd(size_t heightA, size_t widthA, size_t widthB,
data_t *A, size_t rstrideA,
data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
size_t i;
size_t j;
size_t k = widthA - 1;
// edges were fixed by fix_heightA and fix_widthB
heightA &= ~1;
widthB &= ~1;
// FIXME: replace at_h with pointers
//#pragma omp parallel for
for (i = 0; i < heightA; i++) {
for (j = 0; j < widthB; j++) {
at_r_ref(C, i, j) += at_r_ref(A, i, k) * at_r_ref(B, j, k);//k, j);
}
}
}
/*
* Strassen algorithm for multiplication
* @see http://en.wikipedia.org/wiki/Strassen_algorithm
*/
void matmul_strassen(size_t heightA, size_t widthA, size_t widthB,
data_t *A, size_t rstrideA,
data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
/*
* divide matrices first
*/
const size_t heightAh = heightA >> 1;
const size_t widthAh = widthA >> 1;
const size_t heightBh = widthAh;
const size_t widthBh = widthB >> 1;
// A
data_t *A11 = at_r(A, 0, 0);
data_t *A12 = at_r(A, 0, widthAh);
data_t *A21 = at_r(A, heightAh, 0);
data_t *A22 = at_r(A, heightAh, widthAh);
// B
data_t *B11 = at_r(B, 0, 0);
data_t *B12 = at_r(B, widthBh,0);
//data_t *B12 = at_r(B, 0, widthBh);
//data_t *B21 = at_r(B, heightBh, 0);
data_t *B21 = at_r(B, 0, heightBh);
data_t *B22 = at_r(B, heightBh, widthBh);
// C
data_t *C11 = at_r(C, 0, 0);
data_t *C12 = at_r(C, 0, widthBh);
data_t *C21 = at_r(C, heightAh, 0);
data_t *C22 = at_r(C, heightAh, widthBh);
const size_t heightP = heightAh;
const size_t widthP = widthBh;
data_t *P1 = (data_t *) malloc(sizeof(data_t) * heightP * widthP);
const size_t rstrideP1 = widthP;
data_t *P2 = C21;
const size_t rstrideP2 = rstrideC;
data_t *P3 = (data_t *) malloc(sizeof(data_t) * heightP * widthP);
const size_t rstrideP3 = widthP;
data_t *P4 = (data_t *) malloc(sizeof(data_t) * heightP * widthP);
const size_t rstrideP4 = widthP;
data_t *P5 = C12;
const size_t rstrideP5 = rstrideC;
data_t *P6 = C22;
const size_t rstrideP6 = rstrideC;
data_t *P7 = C11;
const size_t rstrideP7 = rstrideC;
// P1
#pragma omp task
matmul_strassen_P1(heightAh, widthAh, widthBh, A11, A22, rstrideA, B11,
B22, rstrideB, P1, rstrideP1);
// P2, P5
#pragma omp task
matmul_strassen_P2_P5(heightAh, widthAh, widthBh, A21, A22, rstrideA, B11,
rstrideB, P2, rstrideP2);
#pragma omp task
matmul_strassen_P2_P5(heightAh, widthAh, widthBh, A11, A12, rstrideA, B22,
rstrideB, P5, rstrideP5);
// P3, P4
#pragma omp task
matmul_strassen_P3_P4(heightAh, widthAh, widthBh, A11, rstrideA, B12, B22,
rstrideB, P3, rstrideP3);
#pragma omp task
matmul_strassen_P3_P4(heightAh, widthAh, widthBh, A22, rstrideA, B21, B11,
rstrideB, P4, rstrideP4);
// P6, P7
#pragma omp task
matmul_strassen_P6_P7(heightAh, widthAh, widthBh, A21, A11, rstrideA, B11,
B12, rstrideB, P6, rstrideP6);
#pragma omp task
matmul_strassen_P6_P7(heightAh, widthAh, widthBh, A12, A22, rstrideA, B21,
B22, rstrideB, P7, rstrideP7);
#pragma omp taskwait
// omp paralell sections
//#pragma omp task
matmul_strassen_C11_C22( P7, rstrideP7, P1, rstrideP1, P4, rstrideP4, P5,
rstrideP5, heightP, widthP);
//#pragma omp task
matmul_strassen_C11_C22( P6, rstrideP6, P1, rstrideP1, P3, rstrideP3, P2,
rstrideP2, heightP, widthP);
//#pragma omp taskwait
// omp paralell sections
//#pragma omp task
matmul_strassen_C12_C21(P5, rstrideP5, P3, rstrideP3, heightP, widthP);
//#pragma omp task
matmul_strassen_C12_C21(P2, rstrideP2, P4, rstrideP4, heightP, widthP);
//#pragma omp taskwait
// omp paralell sections
/*
* Fix odd
*/
//#pragma omp task
if (heightA & 1) // heightA is odd
matmul_strassen_fix_heightA_odd(heightA, widthA, widthB, A, rstrideA,
B, rstrideB, C, rstrideC);
//#pragma omp task
if (widthB & 1) // widthB is odd
matmul_strassen_fix_widthB_odd(heightA, widthA, widthB, A, rstrideA,
B, rstrideB, C, rstrideC);
//#pragma omp task
if (widthA & 1) // widthA is odd
matmul_strassen_fix_widthA_odd(heightA, widthA, widthB, A, rstrideA,
B, rstrideB, C, rstrideC);
//#pragma omp taskwait
// omp paralell sections
// sync
free(P4);
free(P3);
free(P1);
}
/*
* Simple, three-loop multiplication
*/
void matmul_simple(size_t heightA, size_t widthA, size_t widthB,
data_t *A, size_t rstrideA,
data_t *B, size_t rstrideB,
data_t *C, size_t rstrideC) {
size_t i;
size_t j;
size_t k;
#pragma omp parallel for
for (i = 0; i < heightA; i++) {
for (j = 0; j < widthB; j++) {
data_t sum = 0;
/* unsafe, slowly, no sense
#pragma omp parallel for reduction(+:sum) */
for(k = 0; k < widthA; k++) {
sum += at_r_ref(A, i, k) * at_r_ref(B, j, k); // k, j);
}
at_r_ref(C, i, j) = sum;
}
}
}
#undef at_r_ref
#undef at_r
|
jacobi-block-task-dep.c | # include "poisson.h"
/* #pragma omp task/taskwait version of SWEEP. */
void sweep_block_task_dep (int nx, int ny, double dx, double dy, double *f_,
int itold, int itnew, double *u_, double *unew_, int block_size)
{
int it;
#ifdef _OPENMP
double (*f)[nx][ny] = (double (*)[nx][ny])f_;
double (*u)[nx][ny] = (double (*)[nx][ny])u_;
double (*unew)[nx][ny] = (double (*)[nx][ny])unew_;
#endif
int block_x, block_y;
if (block_size == 0)
block_size = nx;
int max_blocks_x = (nx / block_size);
int max_blocks_y = (ny / block_size);
#pragma omp parallel \
shared(u_, unew_, f, max_blocks_x, max_blocks_y, nx, ny, dx, dy, itold, itnew, block_size) \
private(it, block_x, block_y)
#pragma omp single
{
for (it = itold + 1; it <= itnew; it++) {
// Save the current estimate.
for (block_x = 0; block_x < max_blocks_x; block_x++) {
for (block_y = 0; block_y < max_blocks_y; block_y++) {
#pragma omp task shared(u_, unew_, block_size, nx, ny) firstprivate(block_x, block_y) \
depend(in: unew[block_x * block_size: block_size][block_y * block_size: block_size]) \
depend(out: u[block_x * block_size: block_size][block_y * block_size: block_size])
copy_block(nx, ny, block_x, block_y, u_, unew_, block_size);
}
}
// Compute a new estimate.
for (block_x = 0; block_x < max_blocks_x; block_x++) {
for (block_y = 0; block_y < max_blocks_y; block_y++) {
int xdm1 = block_x == 0 ? 0 : 1;
int xdp1 = block_x == max_blocks_x-1 ? 0 : +1;
int ydp1 = block_y == max_blocks_y-1 ? 0 : +1;
int ydm1 = block_y == 0 ? 0 : 1;
#pragma omp task shared(u_, unew_, f_, dx, dy, nx, ny, block_size) firstprivate(block_x, block_y, xdm1, xdp1, ydp1, ydm1) \
depend(out: unew[block_x * block_size: block_size][block_y * block_size: block_size]) \
depend(in: f[block_x * block_size: block_size][block_y * block_size: block_size], \
u[block_x * block_size: block_size][block_y * block_size: block_size], \
u[(block_x - xdm1) * block_size: block_size][block_y * block_size: block_size], \
u[block_x * block_size: block_size][(block_y + ydp1)* block_size: block_size], \
u[block_x * block_size: block_size][(block_y - ydm1)* block_size: block_size], \
u[(block_x + xdp1)* block_size: block_size][block_y * block_size: block_size])
compute_estimate(block_x, block_y, u_, unew_, f_, dx, dy,
nx, ny, block_size);
}
}
}
}
}
|
oned_csc.c | /* Copyright (C) 2010-2011 The Trustees of Indiana University. */
/* */
/* Use, modification and distribution is subject to the Boost Software */
/* License, Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at */
/* http://www.boost.org/LICENSE_1_0.txt) */
/* */
/* Authors: Jeremiah Willcock */
/* Andrew Lumsdaine */
#include "common.h"
#include "oned_csc.h"
#include "redistribute.h"
#include <mpi.h>
#include <stdint.h>
#include <inttypes.h>
#include <stdlib.h>
#include <stddef.h>
#include <string.h>
#include <stdio.h>
#include <assert.h>
typedef struct temp_csc_graph {
size_t* restrict rowstarts;
int32_t* restrict column;// int64_t* restrict column;
size_t nlocalverts;
int lg_nglobalverts;
int32_t nglobalverts; //int64_t nglobalverts;
size_t nlocaledges;
size_t nlocaledges_allocated; /* Actual size of column */
int lg_local_queue_size;
size_t nrows; /* One less than size of rowstarts */
} temp_csc_graph;
static void make_empty_csc(temp_csc_graph* restrict const outg /* All fields NULL or 0 */) {
outg->rowstarts = (size_t*)xcalloc(1, sizeof(size_t));
outg->column = NULL; /* Realloc can enlarge a NULL pointer */
outg->nlocalverts = outg->nglobalverts = outg->nlocaledges = outg->nlocaledges_allocated = 0;
outg->lg_nglobalverts = -1;
outg->lg_local_queue_size = -1;
outg->nrows = 0;
}
static void make_csc(const packed_edge* restrict const inbuf, temp_csc_graph* restrict const outg /* Must have memory and nlocalverts/nglobalverts/nlocaledges filled in */) {
size_t nrows = outg->nrows;
size_t inbuf_size = outg->nlocaledges;
size_t* temp = (size_t*)xmalloc(nrows * sizeof(size_t));
size_t* restrict rowstarts = outg->rowstarts;
int32_t* restrict column = outg->column; // int64_t* restrict column = outg->column;
int lg_local_queue_size = outg->lg_local_queue_size;
{
size_t* restrict counts = temp;
memset(counts, 0, nrows * sizeof(size_t));
ptrdiff_t i;
#pragma omp parallel for
for (i = 0; i < (ptrdiff_t)inbuf_size; ++i) {
assert ((size_t)(SWIZZLE_VERTEX(get_v1_from_edge(&inbuf[i])) / ULONG_BITS) < nrows);
#pragma omp atomic
++counts[SWIZZLE_VERTEX(get_v1_from_edge(&inbuf[i])) / ULONG_BITS];
}
rowstarts[0] = 0;
for (i = 0; i < nrows; ++i) {
rowstarts[i + 1] = rowstarts[i] + counts[i];
}
}
{
size_t* restrict inserts = temp;
memcpy(inserts, rowstarts, nrows * sizeof(size_t));
ptrdiff_t i;
#pragma omp parallel for
for (i = 0; i < (ptrdiff_t)inbuf_size; ++i) {
int32_t v0 = get_v0_from_edge(&inbuf[i]);// int64_t v0 = get_v0_from_edge(&inbuf[i]);
int32_t v1 = SWIZZLE_VERTEX(get_v1_from_edge(&inbuf[i])); // int64_t v1 = SWIZZLE_VERTEX(get_v1_from_edge(&inbuf[i]));
// fprintf(stderr, "%d: Raw edge is (%" PRId64 ", %" PRId64 ") -> (%zu, %" PRId64 " = %" PRId64 ")\n", rank, v0, get_v1_from_edge(&inbuf[i]), VERTEX_LOCAL(v0), v1, UNSWIZZLE_VERTEX(v1));
size_t pos = __sync_fetch_and_add(&inserts[(v1) / ULONG_BITS], 1);
column[pos] = (v1 % ULONG_BITS) + VERTEX_LOCAL(v0) * ULONG_BITS;
// fprintf(stderr, "%d: Stored as (row %" PRId64 ", col %" PRId64 "/%" PRId64 ")\n", rank, (v1) / ULONG_BITS, column[pos] % ULONG_BITS, column[pos] / ULONG_BITS);
}
}
free(temp); temp = NULL;
}
/* Do merge: b = b union a */
static void merge_csc(temp_csc_graph* restrict const b,
temp_csc_graph* restrict const a) {
if (b->lg_local_queue_size == -1) { // b is empty
if (b->rowstarts != NULL) {free(b->rowstarts); b->rowstarts = NULL;}
if (b->column != NULL) {free(b->column); b->column = NULL;}
*b = *a;
a->rowstarts = NULL;
a->column = NULL;
return;
} else if (a->nglobalverts != b->nglobalverts) {
/* Redistribution wrapper should restart in this case, not try to do a merge. */
fprintf(stderr, "%d: a->nglobalverts=%" PRId64 " != b->nglobalverts=%" PRId64 "\n", rank, a->nglobalverts, b->nglobalverts);
MPI_Abort(MPI_COMM_WORLD, 5);
} else {
assert (a->lg_local_queue_size == b->lg_local_queue_size);
assert (a->nrows == b->nrows);
assert (a->lg_nglobalverts == b->lg_nglobalverts);
size_t a_nlocaledges = a->nlocaledges;
size_t b_nlocaledges = b->nlocaledges;
size_t nrows = b->nrows;
if (b_nlocaledges + a_nlocaledges > b->nlocaledges_allocated) {
size_t new_alloc = b_nlocaledges + a_nlocaledges + (1 << 16);
b->nlocaledges_allocated = new_alloc;
b->column = (int32_t*)xrealloc(b->column, new_alloc * sizeof(int32_t)); // b->column = (int64_t*)xrealloc(b->column, new_alloc * sizeof(int64_t));
}
ptrdiff_t i_plus_1;
/* This loop needs to be sequential. */
for (i_plus_1 = nrows; i_plus_1 > 0; --i_plus_1) {
ptrdiff_t i = i_plus_1 - 1;
memmove(&b->column[b->rowstarts[i] + a->rowstarts[i]],
&b->column[b->rowstarts[i]],
(b->rowstarts[i + 1] - b->rowstarts[i]) * sizeof(int32_t)); // (b->rowstarts[i + 1] - b->rowstarts[i]) * sizeof(int64_t));
}
/* This loop can be parallel. */
#pragma omp parallel for
for (i_plus_1 = nrows; i_plus_1 > 0; --i_plus_1) {
ptrdiff_t i = i_plus_1 - 1;
memcpy(&b->column[b->rowstarts[i + 1] + a->rowstarts[i]],
&a->column[a->rowstarts[i]],
(a->rowstarts[i + 1] - a->rowstarts[i]) * sizeof(int32_t)); // (a->rowstarts[i + 1] - a->rowstarts[i]) * sizeof(int64_t));
}
b_nlocaledges = b->nlocaledges = b_nlocaledges + a_nlocaledges;
ptrdiff_t i;
#pragma omp parallel for
for (i = 0; i <= nrows; ++i) {
b->rowstarts[i] += a->rowstarts[i];
}
free(a->column); a->column = NULL;
free(a->rowstarts); a->rowstarts = NULL;
}
}
#define CONV1D_FUNCNAME \
convert_graph_to_oned_csc_helper
#define CONV1D_EXTRA_PARAMS \
oned_csc_graph* const g
#define CONV1D_DECLARE_AND_INIT_GRAPH_SO_FAR \
temp_csc_graph graph_so_far = {NULL, NULL, 0, 0, 0}; \
make_empty_csc(&graph_so_far);
#define CONV1D_CALL_ON_EDGES(V0, V1, LG_NGLOBALVERTS_SO_FAR, CONT) \
CONT(VERTEX_OWNER((V0)), CONV1D_WRITE_EDGE_NORMAL) \
CONT(VERTEX_OWNER((V1)), CONV1D_WRITE_EDGE_FLIPPED)
#define CONV1D_WRITE_EDGE_NORMAL(BUF, V0, V1) \
write_edge(BUF, V0, V1);
#define CONV1D_WRITE_EDGE_FLIPPED(BUF, V0, V1) \
write_edge(BUF, V1, V0);
#define CONV1D_EDGE_BUFFER_TYPE \
packed_edge
#define CONV1D_EDGE_BUFFER_MPI_TYPE \
packed_edge_mpi_type
#define CONV1D_PRECOMPRESS_INCOMING_DATA(LG_NGLOBALVERTS_SO_FAR, EDGES_TO_RECV, EDGES_RECEIVED_THIS_BLOCK) \
size_t nlocalverts_so_far = (size_t)DIV_SIZE((UINT64_C(1) << (LG_NGLOBALVERTS_SO_FAR)) + size - 1); \
size_t t_nrows = (size_t)(MUL_SIZE((nlocalverts_so_far + ULONG_BITS * ULONG_BITS - 1) / ULONG_BITS / ULONG_BITS * ULONG_BITS)); \
temp_csc_graph t = { \
/* rowstarts */ (size_t*)xmalloc((t_nrows + 1) * sizeof(size_t)), \
/*(int64_t*)xmalloc((size_t)(EDGES_RECEIVED_THIS_BLOCK) * sizeof(int64_t)), */ \
/* column */ (int32_t*)xmalloc((size_t)(EDGES_RECEIVED_THIS_BLOCK) * sizeof(int32_t)), \
/* nlocalverts */ (size_t)(nlocalverts_so_far), \
/* lg_nglobalverts */ (int)(LG_NGLOBALVERTS_SO_FAR), \
/*(int64_t)(INT64_C(1) << (LG_NGLOBALVERTS_SO_FAR)),*/ \
/* nglobalverts */ (int32_t)(INT32_C(1) << (LG_NGLOBALVERTS_SO_FAR)), \
/* nlocaledges */ (size_t)(EDGES_RECEIVED_THIS_BLOCK), \
/* nlocaledges_allocated */ (size_t)(EDGES_RECEIVED_THIS_BLOCK), \
/* lg_local_queue_size */ -1, /* Filled in later */ \
/* nrows */ t_nrows \
}; \
{ \
t.lg_local_queue_size = lg_int32_t(DIV_SIZE(t_nrows)); /*t.lg_local_queue_size = lg_int64_t(DIV_SIZE(t_nrows));*/\
make_csc((EDGES_TO_RECV), &t); \
}
#define CONV1D_MERGE_INTO_GRAPH_SO_FAR \
size_t new_alloc = graph_so_far.nlocaledges + edges_received_this_block * (block_count - ITERATE_TUPLE_GRAPH_BLOCK_NUMBER); \
if (graph_so_far.lg_local_queue_size != -1 && new_alloc > graph_so_far.nlocaledges_allocated) { \
size_t new_alloc_real = new_alloc + (1 << 16); \
graph_so_far.nlocaledges_allocated = new_alloc_real; \
/* graph_so_far.column = (int64_t*)xrealloc(graph_so_far.column, new_alloc_real * sizeof(int64_t));*/\
graph_so_far.column = (int32_t*)xrealloc(graph_so_far.column, new_alloc_real * sizeof(int32_t)); \
} \
merge_csc(&graph_so_far, &t);
#define CONV1D_FREE_PRECOMPRESSED_DATA \
if (t.rowstarts != NULL) {free(t.rowstarts); t.rowstarts = NULL;} \
if (t.column != NULL) {free(t.column); t.column = NULL;}
#define CONV1D_BUILD_FINAL_DATA_STRUCTURE_FROM_GRAPH_SO_FAR \
g->nlocaledges = graph_so_far.nlocaledges; \
g->rowstarts = graph_so_far.rowstarts; \
graph_so_far.rowstarts = NULL; \
/*g->column = (int64_t*)xrealloc(graph_so_far.column, (size_t)g->nlocaledges * sizeof(int64_t));*/\
g->column = (int32_t*)xrealloc(graph_so_far.column, (size_t)g->nlocaledges * sizeof(int32_t)); \
graph_so_far.column = NULL; \
g->lg_local_queue_size = graph_so_far.lg_local_queue_size; \
size_t nlocalverts = graph_so_far.nlocalverts; \
g->nlocalverts = nlocalverts; \
g->max_nlocalverts = nlocalverts; /* Now same on all ranks */ \
g->lg_nglobalverts = graph_so_far.lg_nglobalverts; \
/* g->nglobalverts = INT64_C(1) << graph_so_far.lg_nglobalverts;*/\
g->nglobalverts = INT32_C(1) << graph_so_far.lg_nglobalverts;
#define CONV1D_CLEAR_GRAPH_SO_FAR \
free(graph_so_far.rowstarts); graph_so_far.rowstarts = NULL; \
free(graph_so_far.column); graph_so_far.column = NULL; \
graph_so_far.nlocalverts = graph_so_far.nlocaledges = graph_so_far.nlocaledges_allocated = 0; \
graph_so_far.lg_local_queue_size = -1; \
graph_so_far.nrows = 0;
static MAKE_REDISTRIBUTE_FUNC(CONV1D_FUNCNAME, CONV1D_EXTRA_PARAMS, CONV1D_DECLARE_AND_INIT_GRAPH_SO_FAR, CONV1D_CALL_ON_EDGES, CONV1D_EDGE_BUFFER_TYPE, CONV1D_EDGE_BUFFER_MPI_TYPE, CONV1D_PRECOMPRESS_INCOMING_DATA, CONV1D_MERGE_INTO_GRAPH_SO_FAR, CONV1D_FREE_PRECOMPRESSED_DATA, CONV1D_BUILD_FINAL_DATA_STRUCTURE_FROM_GRAPH_SO_FAR, CONV1D_CLEAR_GRAPH_SO_FAR)
void convert_graph_to_oned_csc(const tuple_graph* const tg, oned_csc_graph* const g) { \
g->tg = tg;
g->nlocaledges = 0;
convert_graph_to_oned_csc_helper(tg, g);
g->max_nlocalverts = (int32_t)(g->nlocalverts); // g->max_nlocalverts = (int64_t)(g->nlocalverts);
// MPI_Allreduce(MPI_IN_PLACE, &g->max_nlocalverts, 1, MPI_INT64_T, MPI_MAX, MPI_COMM_WORLD);
MPI_Allreduce(MPI_IN_PLACE, &g->max_nlocalverts, 1, MPI_INT32_T, MPI_MAX, MPI_COMM_WORLD);
// int64_t local_queue_summary_size = (g->max_nlocalverts + ULONG_BITS * ULONG_BITS - 1) / ULONG_BITS / ULONG_BITS;
int32_t local_queue_summary_size = (g->max_nlocalverts + ULONG_BITS * ULONG_BITS - 1) / ULONG_BITS / ULONG_BITS;
//int64_t local_queue_size = local_queue_summary_size * ULONG_BITS;
int32_t local_queue_size = local_queue_summary_size * ULONG_BITS;
if (g->lg_local_queue_size != lg_int32_t(local_queue_size)) // if (g->lg_local_queue_size != lg_int64_t(local_queue_size))
{
/*
fprintf(stderr, "%d: lg_local_queue_size mismatch: graph redistribution computed %d, convert_graph_to_oned_csc outer computed %d from %" PRId64 "\n", rank, g->lg_local_queue_size, lg_int64_t(local_queue_size), local_queue_size);
*/
fprintf(stderr, "%d: lg_local_queue_size mismatch: graph redistribution computed %d, convert_graph_to_oned_csc outer computed %d from %" PRId32 "\n", rank, g->lg_local_queue_size, lg_int32_t(local_queue_size), local_queue_size);
MPI_Abort(MPI_COMM_WORLD, 6);
}
}
void free_oned_csc_graph(oned_csc_graph* const g) {
if (g->rowstarts != NULL) {free(g->rowstarts); g->rowstarts = NULL;}
if (g->column != NULL) {free(g->column); g->column = NULL;}
}
|
val_omp.c | /* This file performs the following test: each OMP thread measures flops
for its provided tasks, and compares this to expected flop counts, each
thread having been provided with a random amount of work, such that the
time and order that they complete their measurements varies.
Specifically tested is the case where the value returned for some threads
actually corresponds to that for another thread reading its counter values
at the same time.
- It is based on zero_omp.c but ignored much of its functionality.
- It attempts to use the following two counters. It may use less
depending on hardware counter resource limitations. These are counted
in the default counting domain and default granularity, depending on
the platform. Usually this is the user domain (PAPI_DOM_USER) and
thread context (PAPI_GRN_THR).
+ PAPI_FP_INS
+ PAPI_TOT_CYC
Each thread inside the Thread routine:
- Do prework (MAX_FLOPS - flops)
- Get cyc.
- Get us.
- Start counters
- Do flops
- Stop and read counters
- Get us.
- Get cyc.
- Return flops
*/
#include "papi_test.h"
#ifdef _OPENMP
#include <omp.h>
#else
#error "This compiler does not understand OPENMP"
#endif
const int MAX_FLOPS = NUM_FLOPS;
extern int TESTS_QUIET; /* Declared in test_utils.c */
const PAPI_hw_info_t *hw_info = NULL;
long long Thread(int n)
{
int retval, num_tests = 1;
int EventSet1=PAPI_NULL;
int PAPI_event, mask1;
int num_events1;
long long flops;
long long **values;
long long elapsed_us, elapsed_cyc;
char event_name[PAPI_MAX_STR_LEN];
/* printf("Thread(n=%d) 0x%x started\n", n, omp_get_thread_num()); */
num_events1 = 2;
/* add PAPI_TOT_CYC and one of the events in PAPI_FP_INS, PAPI_FP_OPS or
PAPI_TOT_INS, depending on the availability of the event on the
platform */
EventSet1 = add_two_events(&num_events1, &PAPI_event, hw_info, &mask1);
retval = PAPI_event_code_to_name(PAPI_event, event_name);
if (retval != PAPI_OK)
test_fail(__FILE__, __LINE__, "PAPI_event_code_to_name", retval);
values = allocate_test_space(num_tests, num_events1);
do_flops(MAX_FLOPS - n); /* prework for balance */
elapsed_us = PAPI_get_real_usec();
elapsed_cyc = PAPI_get_real_cyc();
retval = PAPI_start(EventSet1);
if (retval != PAPI_OK)
test_fail(__FILE__, __LINE__, "PAPI_start", retval);
do_flops(n);
retval = PAPI_stop(EventSet1, values[0]);
if (retval != PAPI_OK)
test_fail(__FILE__, __LINE__, "PAPI_stop", retval);
flops = (values[0])[0];
elapsed_us = PAPI_get_real_usec() - elapsed_us;
elapsed_cyc = PAPI_get_real_cyc() - elapsed_cyc;
remove_test_events(&EventSet1, mask1);
if (!TESTS_QUIET) {
/*printf("Thread 0x%x %-12s : \t%lld\t%d\n", omp_get_thread_num(), event_name,
(values[0])[0], n);*/
#if 0
printf("Thread 0x%x PAPI_TOT_CYC: \t%lld\n", omp_get_thread_num(), (values[0])[1]);
printf("Thread 0x%x Real usec : \t%lld\n", omp_get_thread_num(), elapsed_us);
printf("Thread 0x%x Real cycles : \t%lld\n", omp_get_thread_num(), elapsed_cyc);
#endif
}
/* It is illegal for the threads to exit in OpenMP */
/* test_pass(__FILE__,0,0); */
free_test_space(values, num_tests);
PAPI_unregister_thread();
/* printf("Thread 0x%x finished\n", omp_get_thread_num()); */
return flops;
}
int main(int argc, char **argv)
{
int tid, retval;
int maxthr = omp_get_max_threads();
int flopper = 0;
long long *flops = calloc(maxthr, sizeof(long long));
long long *flopi = calloc(maxthr, sizeof(long long));
tests_quiet(argc, argv); /* Set TESTS_QUIET variable */
if (maxthr < 2)
test_skip(__FILE__, __LINE__, "omp_get_num_threads < 2", PAPI_EINVAL);
if ((flops == NULL) || (flopi == NULL))
test_fail(__FILE__, __LINE__, "calloc", PAPI_ENOMEM);
retval = PAPI_library_init(PAPI_VER_CURRENT);
if (retval != PAPI_VER_CURRENT)
test_fail(__FILE__, __LINE__, "PAPI_library_init", retval);
hw_info = PAPI_get_hardware_info();
if (hw_info == NULL)
test_fail(__FILE__, __LINE__, "PAPI_get_hardware_info", 2);
retval = PAPI_thread_init((unsigned long (*)(void)) (omp_get_thread_num));
if (retval != PAPI_OK)
if (retval == PAPI_ESBSTR)
test_skip(__FILE__, __LINE__, "PAPI_thread_init", retval);
else
test_fail(__FILE__, __LINE__, "PAPI_thread_init", retval);
flopper = Thread(65536) / 65536;
printf("flopper=%d\n", flopper);
for (int i=0; i<100000; i++)
#pragma omp parallel private(tid)
{
tid = omp_get_thread_num();
flopi[tid] = rand()*3;
flops[tid] = Thread((flopi[tid]/flopper)%MAX_FLOPS);
#pragma omp barrier
#pragma omp master
if (flops[tid] < flopi[tid]) {
printf("test iteration=%d\n", i);
for (int j=0; j<omp_get_num_threads(); j++) {
printf("Thread 0x%x Value %6lld %c %6lld", j, flops[j],
(flops[j]<flopi[j])?'<':'=', flopi[j]);
for (int k=0; k<omp_get_num_threads(); k++)
if ((k != j) && (flops[k] == flops[j]))
printf(" == Thread 0x%x!", k);
printf("\n");
}
test_fail(__FILE__, __LINE__, "value returned for thread", PAPI_EBUG);
}
}
test_pass(__FILE__, NULL, 0);
exit(0);
}
|
deconvolution_3x3.h | // Tencent is pleased to support the open source community by making ncnn available.
//
// Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
#if __ARM_NEON
#include <arm_neon.h>
#endif // __ARM_NEON
static void deconv3x3s1_neon(const Mat& bottom_blob, Mat& top_blob, const Mat& _kernel, const Mat& _bias)
{
int w = bottom_blob.w;
int h = bottom_blob.h;
int inch = bottom_blob.c;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
const float* kernel = _kernel;
const float* bias = _bias;
#pragma omp parallel for
for (int p=0; p<outch; p++)
{
Mat out = top_blob.channel(p);
const float bias0 = bias ? bias[p] : 0.f;
out.fill(bias0);
for (int q=0; q<inch; q++)
{
const float* img0 = bottom_blob.channel(q);
const float* kernel0 = kernel + p*inch*9 + q*9;
const float* r0 = img0;
const float* k0 = kernel0;
const float* k1 = kernel0 + 3;
const float* k2 = kernel0 + 6;
#if __ARM_NEON
float32x4_t _k0 = vld1q_f32(k0);
float32x4_t _k1 = vld1q_f32(k1);
float32x4_t _k2 = vld1q_f32(k2);
#endif // __ARM_NEON
for (int i = 0; i < h; i++)
{
float* outptr = out.data + out.w * i;
float* outptr0 = outptr;
float* outptr1 = outptr + outw;
float* outptr2 = outptr + outw*2;
int j = 0;
#if __ARM_NEON
for (; j+3 < w; j+=4)
{
float32x4_t _v = vld1q_f32(r0);
#if 0 // bad compiler generate slow instructions :(
// 0
float32x4_t _out00 = vld1q_f32(outptr0 + 0);
_out00 = vmlaq_lane_f32(_out00, _v, vget_low_f32(_k0), 0);
float32x4_t _out01 = vmulq_lane_f32(_v, vget_low_f32(_k0), 1);
// ext
float32x4_t _zero_out01 = vdupq_n_f32(0.f);
_zero_out01 = vextq_f32(_zero_out01, _out01, 3);
_out00 = vaddq_f32(_out00, _zero_out01);
//
float32x2_t _out00low = vget_low_f32(_out00);
float32x2_t _out00high = vget_high_f32(_out00);
_out00high = vmla_lane_f32(_out00high, vget_low_f32(_v), vget_high_f32(_k0), 0);
_out00 = vcombine_f32(_out00low, _out00high);
vst1q_f32(outptr0 + 0, _out00);
//
float32x2_t _out02high = vld1_f32(outptr0 + 4);
float32x2_t _out01_zero = vext_f32(vget_high_f32(_out01), vget_low_f32(_zero_out01), 1);
_out02high = vadd_f32(_out02high, _out01_zero);
_out02high = vmla_lane_f32(_out02high, vget_high_f32(_v), vget_high_f32(_k0), 0);
vst1_f32(outptr0 + 4, _out02high);
// 1
float32x4_t _out10 = vld1q_f32(outptr1 + 0);
_out10 = vmlaq_lane_f32(_out10, _v, vget_low_f32(_k1), 0);
float32x4_t _out11 = vmulq_lane_f32(_v, vget_low_f32(_k1), 1);
// ext
float32x4_t _zero_out11 = vdupq_n_f32(0.f);
_zero_out11 = vextq_f32(_zero_out11, _out11, 3);
_out10 = vaddq_f32(_out10, _zero_out11);
//
float32x2_t _out10low = vget_low_f32(_out10);
float32x2_t _out10high = vget_high_f32(_out10);
_out10high = vmla_lane_f32(_out10high, vget_low_f32(_v), vget_high_f32(_k1), 0);
_out10 = vcombine_f32(_out10low, _out10high);
vst1q_f32(outptr1 + 0, _out10);
//
float32x2_t _out12high = vld1_f32(outptr1 + 4);
float32x2_t _out11_zero = vext_f32(vget_high_f32(_out11), vget_low_f32(_zero_out11), 1);
_out12high = vadd_f32(_out12high, _out11_zero);
_out12high = vmla_lane_f32(_out12high, vget_high_f32(_v), vget_high_f32(_k1), 0);
vst1_f32(outptr1 + 4, _out12high);
// 2
float32x4_t _out20 = vld1q_f32(outptr2 + 0);
_out20 = vmlaq_lane_f32(_out20, _v, vget_low_f32(_k2), 0);
float32x4_t _out21 = vmulq_lane_f32(_v, vget_low_f32(_k2), 1);
// ext
float32x4_t _zero_out21 = vdupq_n_f32(0.f);
_zero_out21 = vextq_f32(_zero_out21, _out21, 3);
_out20 = vaddq_f32(_out20, _zero_out21);
//
float32x2_t _out20low = vget_low_f32(_out20);
float32x2_t _out20high = vget_high_f32(_out20);
_out20high = vmla_lane_f32(_out20high, vget_low_f32(_v), vget_high_f32(_k2), 0);
_out20 = vcombine_f32(_out20low, _out20high);
vst1q_f32(outptr2 + 0, _out20);
//
float32x2_t _out22high = vld1_f32(outptr2 + 4);
float32x2_t _out21_zero = vext_f32(vget_high_f32(_out21), vget_low_f32(_zero_out21), 1);
_out22high = vadd_f32(_out22high, _out21_zero);
_out22high = vmla_lane_f32(_out22high, vget_high_f32(_v), vget_high_f32(_k2), 0);
vst1_f32(outptr2 + 4, _out22high);
#else
//
float32x4_t _out00 = vld1q_f32(outptr0 + 0);
_out00 = vmlaq_lane_f32(_out00, _v, vget_low_f32(_k0), 0);
vst1q_f32(outptr0 + 0, _out00);
float32x4_t _out01 = vld1q_f32(outptr0 + 1);
_out01 = vmlaq_lane_f32(_out01, _v, vget_low_f32(_k0), 1);
vst1q_f32(outptr0 + 1, _out01);
float32x4_t _out02 = vld1q_f32(outptr0 + 2);
_out02 = vmlaq_lane_f32(_out02, _v, vget_high_f32(_k0), 0);
vst1q_f32(outptr0 + 2, _out02);
//
float32x4_t _out10 = vld1q_f32(outptr1 + 0);
_out10 = vmlaq_lane_f32(_out10, _v, vget_low_f32(_k1), 0);
vst1q_f32(outptr1 + 0, _out10);
float32x4_t _out11 = vld1q_f32(outptr1 + 1);
_out11 = vmlaq_lane_f32(_out11, _v, vget_low_f32(_k1), 1);
vst1q_f32(outptr1 + 1, _out11);
float32x4_t _out12 = vld1q_f32(outptr1 + 2);
_out12 = vmlaq_lane_f32(_out12, _v, vget_high_f32(_k1), 0);
vst1q_f32(outptr1 + 2, _out12);
//
float32x4_t _out20 = vld1q_f32(outptr2 + 0);
_out20 = vmlaq_lane_f32(_out20, _v, vget_low_f32(_k2), 0);
vst1q_f32(outptr2 + 0, _out20);
float32x4_t _out21 = vld1q_f32(outptr2 + 1);
_out21 = vmlaq_lane_f32(_out21, _v, vget_low_f32(_k2), 1);
vst1q_f32(outptr2 + 1, _out21);
float32x4_t _out22 = vld1q_f32(outptr2 + 2);
_out22 = vmlaq_lane_f32(_out22, _v, vget_high_f32(_k2), 0);
vst1q_f32(outptr2 + 2, _out22);
#endif
r0 += 4;
outptr0 += 4;
outptr1 += 4;
outptr2 += 4;
}
#endif // __ARM_NEON
for (; j < w; j++)
{
float val = r0[0];
outptr0[0] += val * k0[0];
outptr0[1] += val * k0[1];
outptr0[2] += val * k0[2];
outptr1[0] += val * k1[0];
outptr1[1] += val * k1[1];
outptr1[2] += val * k1[2];
outptr2[0] += val * k2[0];
outptr2[1] += val * k2[1];
outptr2[2] += val * k2[2];
r0++;
outptr0++;
outptr1++;
outptr2++;
}
}
}
}
}
|
simde-diagnostic.h | /* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person
* obtaining a copy of this software and associated documentation
* files (the "Software"), to deal in the Software without
* restriction, including without limitation the rights to use, copy,
* modify, merge, publish, distribute, sublicense, and/or sell copies
* of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
* Copyright:
* 2017-2020 Evan Nemerson <evan@nemerson.com>
*/
/* SIMDe targets a very wide range of standards and compilers, and our
* goal is to compile cleanly even with extremely aggressive warnings
* (i.e., -Weverything in clang, -Wextra in GCC, /W4 for MSVC, etc.)
* treated as errors.
*
* While our preference is to resolve the underlying issue a given
* diagnostic is warning us about, sometimes that's not possible.
* Fixing a warning in one compiler may cause problems in another.
* Sometimes a warning doesn't really apply to us (false positives),
* and sometimes adhering to a warning would mean dropping a feature
* we *know* the compiler supports since we have tested specifically
* for the compiler or feature.
*
* When practical, warnings are only disabled for specific code. For
* a list of warnings which are enabled by default in all SIMDe code,
* see SIMDE_DISABLE_UNWANTED_DIAGNOSTICS. Note that we restore the
* warning stack when SIMDe is done parsing, so code which includes
* SIMDe is not deprived of these warnings.
*/
#if !defined(SIMDE_DIAGNOSTIC_H)
#define SIMDE_DIAGNOSTIC_H
#include "hedley.h"
#include "simde-detect-clang.h"
/* This is only to help us implement functions like _mm_undefined_ps. */
#if defined(SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_)
#undef SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_
#endif
#if HEDLEY_HAS_WARNING("-Wuninitialized")
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("clang diagnostic ignored \"-Wuninitialized\"")
#elif HEDLEY_GCC_VERSION_CHECK(4,2,0)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("GCC diagnostic ignored \"-Wuninitialized\"")
#elif HEDLEY_PGI_VERSION_CHECK(19,10,0)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("diag_suppress 549")
#elif HEDLEY_SUNPRO_VERSION_CHECK(5,14,0) && defined(__cplusplus)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("error_messages(off,SEC_UNINITIALIZED_MEM_READ,SEC_UNDEFINED_RETURN_VALUE,unassigned)")
#elif HEDLEY_SUNPRO_VERSION_CHECK(5,14,0)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("error_messages(off,SEC_UNINITIALIZED_MEM_READ,SEC_UNDEFINED_RETURN_VALUE)")
#elif HEDLEY_SUNPRO_VERSION_CHECK(5,12,0) && defined(__cplusplus)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("error_messages(off,unassigned)")
#elif \
HEDLEY_TI_VERSION_CHECK(16,9,9) || \
HEDLEY_TI_CL6X_VERSION_CHECK(8,0,0) || \
HEDLEY_TI_CL7X_VERSION_CHECK(1,2,0) || \
HEDLEY_TI_CLPRU_VERSION_CHECK(2,3,2)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("diag_suppress 551")
#elif HEDLEY_INTEL_VERSION_CHECK(13,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ _Pragma("warning(disable:592)")
#elif HEDLEY_MSVC_VERSION_CHECK(19,0,0) && !defined(__MSVC_RUNTIME_CHECKS)
#define SIMDE_DIAGNOSTIC_DISABLE_UNINITIALIZED_ __pragma(warning(disable:4700))
#endif
/* GCC emits a lot of "notes" about the ABI being different for things
* in newer versions of GCC. We don't really care because all our
* functions are inlined and don't generate ABI. */
#if HEDLEY_GCC_VERSION_CHECK(7,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_PSABI_ _Pragma("GCC diagnostic ignored \"-Wpsabi\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_PSABI_
#endif
/* Since MMX uses x87 FP registers, you're supposed to call _mm_empty()
* after each MMX function before any floating point instructions.
* Some compilers warn about functions which use MMX functions but
* don't call _mm_empty(). However, since SIMDe is implementyng the
* MMX API we shouldn't be calling _mm_empty(); we leave it to the
* caller to invoke simde_mm_empty(). */
#if HEDLEY_INTEL_VERSION_CHECK(19,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_NO_EMMS_INSTRUCTION_ _Pragma("warning(disable:13200 13203)")
#elif defined(HEDLEY_MSVC_VERSION)
#define SIMDE_DIAGNOSTIC_DISABLE_NO_EMMS_INSTRUCTION_ __pragma(warning(disable:4799))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_NO_EMMS_INSTRUCTION_
#endif
/* Intel is pushing people to use OpenMP SIMD instead of Cilk+, so they
* emit a diagnostic if you use #pragma simd instead of
* #pragma omp simd. SIMDe supports OpenMP SIMD, you just need to
* compile with -qopenmp or -qopenmp-simd and define
* SIMDE_ENABLE_OPENMP. Cilk+ is just a fallback. */
#if HEDLEY_INTEL_VERSION_CHECK(18,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_SIMD_PRAGMA_DEPRECATED_ _Pragma("warning(disable:3948)")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_SIMD_PRAGMA_DEPRECATED_
#endif
/* MSVC emits a diagnostic when we call a function (like
* simde_mm_set_epi32) while initializing a struct. We currently do
* this a *lot* in the tests. */
#if \
defined(HEDLEY_MSVC_VERSION)
#define SIMDE_DIAGNOSTIC_DISABLE_NON_CONSTANT_AGGREGATE_INITIALIZER_ __pragma(warning(disable:4204))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_NON_CONSTANT_AGGREGATE_INITIALIZER_
#endif
/* This warning needs a lot of work. It is triggered if all you do is
* pass the value to memcpy/__builtin_memcpy, or if you initialize a
* member of the union, even if that member takes up the entire union.
* Last tested with clang-10, hopefully things will improve in the
* future; if clang fixes this I'd love to enable it. */
#if \
HEDLEY_HAS_WARNING("-Wconditional-uninitialized")
#define SIMDE_DIAGNOSTIC_DISABLE_CONDITIONAL_UNINITIALIZED_ _Pragma("clang diagnostic ignored \"-Wconditional-uninitialized\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_CONDITIONAL_UNINITIALIZED_
#endif
/* This warning is meant to catch things like `0.3 + 0.4 == 0.7`, which
* will is false. However, SIMDe uses these operations exclusively
* for things like _mm_cmpeq_ps, for which we really do want to check
* for equality (or inequality).
*
* If someone wants to put together a SIMDE_FLOAT_EQUAL(a, op, b) macro
* which just wraps a check in some code do disable this diagnostic I'd
* be happy to accept it. */
#if \
HEDLEY_HAS_WARNING("-Wfloat-equal") || \
HEDLEY_GCC_VERSION_CHECK(3,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL_ _Pragma("GCC diagnostic ignored \"-Wfloat-equal\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL_
#endif
/* This is because we use HEDLEY_STATIC_ASSERT for static assertions.
* If Hedley can't find an implementation it will preprocess to
* nothing, which means there will be a trailing semi-colon. */
#if HEDLEY_HAS_WARNING("-Wextra-semi")
#define SIMDE_DIAGNOSTIC_DISABLE_EXTRA_SEMI_ _Pragma("clang diagnostic ignored \"-Wextra-semi\"")
#elif HEDLEY_GCC_VERSION_CHECK(8,1,0) && defined(__cplusplus)
#define SIMDE_DIAGNOSTIC_DISABLE_EXTRA_SEMI_ _Pragma("GCC diagnostic ignored \"-Wextra-semi\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_EXTRA_SEMI_
#endif
/* We do use a few variadic macros, which technically aren't available
* until C99 and C++11, but every compiler I'm aware of has supported
* them for much longer. That said, usage is isolated to the test
* suite and compilers known to support them. */
#if HEDLEY_HAS_WARNING("-Wvariadic-macros") || HEDLEY_GCC_VERSION_CHECK(4,0,0)
#if HEDLEY_HAS_WARNING("-Wc++98-compat-pedantic")
#define SIMDE_DIAGNOSTIC_DISABLE_VARIADIC_MACROS_ \
_Pragma("clang diagnostic ignored \"-Wvariadic-macros\"") \
_Pragma("clang diagnostic ignored \"-Wc++98-compat-pedantic\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_VARIADIC_MACROS_ _Pragma("GCC diagnostic ignored \"-Wvariadic-macros\"")
#endif
#else
#define SIMDE_DIAGNOSTIC_DISABLE_VARIADIC_MACROS_
#endif
/* emscripten requires us to use a __wasm_unimplemented_simd128__ macro
* before we can access certain SIMD intrinsics, but this diagnostic
* warns about it being a reserved name. It is a reserved name, but
* it's reserved for the compiler and we are using it to convey
* information to the compiler. */
#if HEDLEY_HAS_WARNING("-Wdouble-promotion")
#define SIMDE_DIAGNOSTIC_DISABLE_RESERVED_ID_MACRO_ _Pragma("clang diagnostic ignored \"-Wreserved-id-macro\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_RESERVED_ID_MACRO_
#endif
/* clang 3.8 warns about the packed attribute being unnecessary when
* used in the _mm_loadu_* functions. That *may* be true for version
* 3.8, but for later versions it is crucial in order to make unaligned
* access safe. */
#if HEDLEY_HAS_WARNING("-Wpacked")
#define SIMDE_DIAGNOSTIC_DISABLE_PACKED_ _Pragma("clang diagnostic ignored \"-Wpacked\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_PACKED_
#endif
/* Triggered when assigning a float to a double implicitly. We use
* explicit casts in SIMDe, this is only used in the test suite. */
#if HEDLEY_HAS_WARNING("-Wdouble-promotion")
#define SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_ _Pragma("clang diagnostic ignored \"-Wdouble-promotion\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_
#endif
/* Several compilers treat conformant array parameters as VLAs. We
* test to make sure we're in C mode (C++ doesn't support CAPs), and
* that the version of the standard supports CAPs. We also reject
* some buggy compilers like MSVC (the logic is in Hedley if you want
* to take a look), but with certain warnings enabled some compilers
* still like to emit a diagnostic. */
#if HEDLEY_HAS_WARNING("-Wvla")
#define SIMDE_DIAGNOSTIC_DISABLE_VLA_ _Pragma("clang diagnostic ignored \"-Wvla\"")
#elif HEDLEY_GCC_VERSION_CHECK(4,3,0)
#define SIMDE_DIAGNOSTIC_DISABLE_VLA_ _Pragma("GCC diagnostic ignored \"-Wvla\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_VLA_
#endif
#if HEDLEY_HAS_WARNING("-Wused-but-marked-unused")
#define SIMDE_DIAGNOSTIC_DISABLE_USED_BUT_MARKED_UNUSED_ _Pragma("clang diagnostic ignored \"-Wused-but-marked-unused\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_USED_BUT_MARKED_UNUSED_
#endif
#if HEDLEY_HAS_WARNING("-Wunused-function")
#define SIMDE_DIAGNOSTIC_DISABLE_UNUSED_FUNCTION_ _Pragma("clang diagnostic ignored \"-Wunused-function\"")
#elif HEDLEY_GCC_VERSION_CHECK(3,4,0)
#define SIMDE_DIAGNOSTIC_DISABLE_UNUSED_FUNCTION_ _Pragma("GCC diagnostic ignored \"-Wunused-function\"")
#elif HEDLEY_MSVC_VERSION_CHECK(19,0,0) /* Likely goes back further */
#define SIMDE_DIAGNOSTIC_DISABLE_UNUSED_FUNCTION_ __pragma(warning(disable:4505))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_UNUSED_FUNCTION_
#endif
#if HEDLEY_HAS_WARNING("-Wpass-failed")
#define SIMDE_DIAGNOSTIC_DISABLE_PASS_FAILED_ _Pragma("clang diagnostic ignored \"-Wpass-failed\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_PASS_FAILED_
#endif
#if HEDLEY_HAS_WARNING("-Wpadded")
#define SIMDE_DIAGNOSTIC_DISABLE_PADDED_ _Pragma("clang diagnostic ignored \"-Wpadded\"")
#elif HEDLEY_MSVC_VERSION_CHECK(19,0,0) /* Likely goes back further */
#define SIMDE_DIAGNOSTIC_DISABLE_PADDED_ __pragma(warning(disable:4324))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_PADDED_
#endif
#if HEDLEY_HAS_WARNING("-Wzero-as-null-pointer-constant")
#define SIMDE_DIAGNOSTIC_DISABLE_ZERO_AS_NULL_POINTER_CONSTANT_ _Pragma("clang diagnostic ignored \"-Wzero-as-null-pointer-constant\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_ZERO_AS_NULL_POINTER_CONSTANT_
#endif
#if HEDLEY_HAS_WARNING("-Wold-style-cast")
#define SIMDE_DIAGNOSTIC_DISABLE_OLD_STYLE_CAST_ _Pragma("clang diagnostic ignored \"-Wold-style-cast\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_OLD_STYLE_CAST_
#endif
#if HEDLEY_HAS_WARNING("-Wcast-function-type") || HEDLEY_GCC_VERSION_CHECK(8,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_CAST_FUNCTION_TYPE_ _Pragma("GCC diagnostic ignored \"-Wcast-function-type\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_CAST_FUNCTION_TYPE_
#endif
/* clang will emit this warning when we use C99 extensions whan not in
* C99 mode, even though it does support this. In such cases we check
* the compiler and version first, so we know it's not a problem. */
#if HEDLEY_HAS_WARNING("-Wc99-extensions")
#define SIMDE_DIAGNOSTIC_DISABLE_C99_EXTENSIONS_ _Pragma("clang diagnostic ignored \"-Wc99-extensions\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_C99_EXTENSIONS_
#endif
/* https://github.com/simd-everywhere/simde/issues/277 */
#if defined(HEDLEY_GCC_VERSION) && HEDLEY_GCC_VERSION_CHECK(4,6,0) && !HEDLEY_GCC_VERSION_CHECK(6,4,0) && defined(__cplusplus)
#define SIMDE_DIAGNOSTIC_DISABLE_BUGGY_UNUSED_BUT_SET_VARIBALE_ _Pragma("GCC diagnostic ignored \"-Wunused-but-set-variable\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_BUGGY_UNUSED_BUT_SET_VARIBALE_
#endif
/* This is the warning that you normally define _CRT_SECURE_NO_WARNINGS
* to silence, but you have to do that before including anything and
* that would require reordering includes. */
#if defined(_MSC_VER)
#define SIMDE_DIAGNOSTIC_DISABLE_ANNEX_K_ __pragma(warning(disable:4996))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_ANNEX_K_
#endif
/* Some compilers, such as clang, may use `long long` for 64-bit
* integers, but `long long` triggers a diagnostic with
* -Wc++98-compat-pedantic which says 'long long' is incompatible with
* C++98. */
#if HEDLEY_HAS_WARNING("-Wc++98-compat-pedantic")
#define SIMDE_DIAGNOSTIC_DISABLE_CPP98_COMPAT_PEDANTIC_ _Pragma("clang diagnostic ignored \"-Wc++98-compat-pedantic\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_CPP98_COMPAT_PEDANTIC_
#endif
/* Some problem as above */
#if HEDLEY_HAS_WARNING("-Wc++11-long-long")
#define SIMDE_DIAGNOSTIC_DISABLE_CPP11_LONG_LONG_ _Pragma("clang diagnostic ignored \"-Wc++11-long-long\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_CPP11_LONG_LONG_
#endif
/* emscripten emits this whenever stdin/stdout/stderr is used in a
* macro. */
#if HEDLEY_HAS_WARNING("-Wdisabled-macro-expansion")
#define SIMDE_DIAGNOSTIC_DISABLE_DISABLED_MACRO_EXPANSION_ _Pragma("clang diagnostic ignored \"-Wdisabled-macro-expansion\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_DISABLED_MACRO_EXPANSION_
#endif
/* Clang uses C11 generic selections to implement some AltiVec
* functions, which triggers this diagnostic when not compiling
* in C11 mode */
#if HEDLEY_HAS_WARNING("-Wc11-extensions")
#define SIMDE_DIAGNOSTIC_DISABLE_C11_EXTENSIONS_ _Pragma("clang diagnostic ignored \"-Wc11-extensions\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_C11_EXTENSIONS_
#endif
/* Clang sometimes triggers this warning in macros in the AltiVec and
* NEON headers, or due to missing functions. */
#if HEDLEY_HAS_WARNING("-Wvector-conversion")
#define SIMDE_DIAGNOSTIC_DISABLE_VECTOR_CONVERSION_ _Pragma("clang diagnostic ignored \"-Wvector-conversion\"")
/* For NEON, the situation with -Wvector-conversion in clang < 10 is
* bad enough that we just disable the warning altogether. */
#if defined(SIMDE_ARCH_ARM) && SIMDE_DETECT_CLANG_VERSION_NOT(10,0,0)
#define SIMDE_DIAGNOSTIC_DISABLE_BUGGY_VECTOR_CONVERSION_ SIMDE_DIAGNOSTIC_DISABLE_VECTOR_CONVERSION_
#endif
#else
#define SIMDE_DIAGNOSTIC_DISABLE_VECTOR_CONVERSION_
#endif
#if !defined(SIMDE_DIAGNOSTIC_DISABLE_BUGGY_VECTOR_CONVERSION_)
#define SIMDE_DIAGNOSTIC_DISABLE_BUGGY_VECTOR_CONVERSION_
#endif
/* SLEEF triggers this a *lot* in their headers */
#if HEDLEY_HAS_WARNING("-Wignored-qualifiers")
#define SIMDE_DIAGNOSTIC_DISABLE_IGNORED_QUALIFIERS_ _Pragma("clang diagnostic ignored \"-Wignored-qualifiers\"")
#elif HEDLEY_GCC_VERSION_CHECK(4,3,0)
#define SIMDE_DIAGNOSTIC_DISABLE_IGNORED_QUALIFIERS_ _Pragma("GCC diagnostic ignored \"-Wignored-qualifiers\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_IGNORED_QUALIFIERS_
#endif
/* GCC emits this under some circumstances when using __int128 */
#if HEDLEY_GCC_VERSION_CHECK(4,8,0)
#define SIMDE_DIAGNOSTIC_DISABLE_PEDANTIC_ _Pragma("GCC diagnostic ignored \"-Wpedantic\"")
#else
#define SIMDE_DIAGNOSTIC_DISABLE_PEDANTIC_
#endif
/* MSVC doesn't like (__assume(0), code) and will warn about code being
* unreachable, but we want it there because not all compilers
* understand the unreachable macro and will complain if it is missing.
* I'm planning on adding a new macro to Hedley to handle this a bit
* more elegantly, but until then... */
#if defined(HEDLEY_MSVC_VERSION)
#define SIMDE_DIAGNOSTIC_DISABLE_UNREACHABLE_ __pragma(warning(disable:4702))
#else
#define SIMDE_DIAGNOSTIC_DISABLE_UNREACHABLE_
#endif
#define SIMDE_DISABLE_UNWANTED_DIAGNOSTICS \
SIMDE_DIAGNOSTIC_DISABLE_PSABI_ \
SIMDE_DIAGNOSTIC_DISABLE_NO_EMMS_INSTRUCTION_ \
SIMDE_DIAGNOSTIC_DISABLE_SIMD_PRAGMA_DEPRECATED_ \
SIMDE_DIAGNOSTIC_DISABLE_CONDITIONAL_UNINITIALIZED_ \
SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL_ \
SIMDE_DIAGNOSTIC_DISABLE_NON_CONSTANT_AGGREGATE_INITIALIZER_ \
SIMDE_DIAGNOSTIC_DISABLE_EXTRA_SEMI_ \
SIMDE_DIAGNOSTIC_DISABLE_VLA_ \
SIMDE_DIAGNOSTIC_DISABLE_USED_BUT_MARKED_UNUSED_ \
SIMDE_DIAGNOSTIC_DISABLE_UNUSED_FUNCTION_ \
SIMDE_DIAGNOSTIC_DISABLE_PASS_FAILED_ \
SIMDE_DIAGNOSTIC_DISABLE_CPP98_COMPAT_PEDANTIC_ \
SIMDE_DIAGNOSTIC_DISABLE_CPP11_LONG_LONG_ \
SIMDE_DIAGNOSTIC_DISABLE_BUGGY_UNUSED_BUT_SET_VARIBALE_ \
SIMDE_DIAGNOSTIC_DISABLE_BUGGY_VECTOR_CONVERSION_
#endif /* !defined(SIMDE_DIAGNOSTIC_H) */
|
HelloWorld.c | #include <stdio.h>
#include <stdlib.h>
#include <omp.h>
int main(void) {
int ID; /* Thread identification number*/
int numThreads; /* Current number of threads running */
/**************************************************************
* You can internally set the number of threads the program is to
* use, with
*
* omp_set_num_threads(4);
*
* but then you must recompile for each number of processors.
*
* omp_get_num_thread();
* returns the number of threads currently being used, so it will
* be 1 if outside a pragma (i.e., in the master thread).
*/
numThreads = omp_get_num_threads();
printf("Outside of the pragma, the number of threads is %d\n\n", numThreads);
#pragma omp parallel private(ID)
{
numThreads = omp_get_num_threads();
printf("Inside the pragma, the number of threads is %d\n\n", numThreads);
ID = omp_get_thread_num();
printf("Hello(%d)", ID);
printf(" world from process %d!\n\n", ID);
}
return 0;
}
|
vector_template.c | #include <stdlib.h>
#include <stdio.h>
#include <math.h>
#include <pthread.h>
#ifndef CVEC_TYPE
#include "cvec.h"
#define CVEC_TYPE cvec_float
#define CVEC_(N) cvec_ ## N
#endif
extern int cvec_njobs;
CVEC_TYPE * CVEC_(linspace)(CVEC_TYPE from, CVEC_TYPE to, cvec_uint len)
{
CVEC_TYPE *rv = malloc(len*sizeof(CVEC_TYPE));
double tof = (double)to;
double fromf = (double)from;
double stepf = ( tof - fromf ) / ( (double)(len - 1) );
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++) {
double iif = (double)i;
rv[i] = (CVEC_TYPE)round( (iif * stepf) + fromf );
}
return rv;
}
CVEC_TYPE * CVEC_(logspace)(CVEC_TYPE from, CVEC_TYPE to, cvec_uint len)
{
CVEC_TYPE lfrom = log(from), lto = log(to);
CVEC_TYPE *rv = CVEC_(linspace)(lfrom, lto, len);
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++) {
rv[i] = pow(10.0, rv[i]);
}
return rv;
}
CVEC_TYPE * CVEC_(apply)(
CVEC_TYPE* in,
cvec_uint len,
CVEC_TYPE (*f)(CVEC_TYPE))
{
CVEC_TYPE *rv = malloc(len*sizeof(CVEC_TYPE));
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++){
rv[i] = f(in[i]);
}
return rv;
}
CVEC_TYPE CVEC_(add)(CVEC_TYPE v1, CVEC_TYPE v2) { return v1 + v2; }
CVEC_TYPE CVEC_(subtract)(CVEC_TYPE v1, CVEC_TYPE v2) { return v1 - v2; }
CVEC_TYPE CVEC_(multiply)(CVEC_TYPE v1, CVEC_TYPE v2) { return v1 * v2; }
CVEC_TYPE CVEC_(divide)(CVEC_TYPE v1, CVEC_TYPE v2) { return v1 / v2; }
CVEC_TYPE CVEC_(pow(CVEC_TYPE v1, CVEC_TYPE v2) { return pow)(v1, v2); }
CVEC_TYPE * CVEC_(apply2)(
CVEC_TYPE *in1,
CVEC_TYPE *in2,
cvec_uint len,
CVEC_TYPE (*f)(CVEC_TYPE, CVEC_TYPE))
{
CVEC_TYPE *rv = malloc(len*sizeof(CVEC_TYPE));
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++){
rv[i] = f(in1[i], in2[i]);
}
return rv;
}
CVEC_TYPE * CVEC_(zeros)(cvec_uint len)
{
CVEC_TYPE *rv = malloc(len*sizeof(CVEC_TYPE));
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++){
rv[i] = 0.0;
}
return rv;
}
CVEC_TYPE *CVEC_(ones)(cvec_uint len)
{
CVEC_TYPE *rv = malloc(len*sizeof(CVEC_TYPE));
#pragma omp parallel for
for (cvec_uint i = 0; i < len; i++){
rv[i] = 1.0;
}
return rv;
}
CVEC_TYPE *CVEC_(copy)(CVEC_TYPE *source, cvec_uint len)
{
CVEC_TYPE *rv = malloc(sizeof(CVEC_TYPE)*len);
#pragma omp parallel for
for(cvec_uint i = 0; i < len; i++) {
rv[i] = source[i];
}
return rv;
}
CVEC_TYPE *CVEC_(cat)(
CVEC_TYPE *source,
cvec_uint len,
CVEC_TYPE *add,
cvec_uint addlen)
{
CVEC_TYPE *rv = malloc(sizeof(CVEC_TYPE)*(len+addlen));
#pragma omp parallel for
for(cvec_uint i = 0; i < len; i++) {
rv[i] = source[i];
}
#pragma omp parallel for
for(cvec_uint i = len; i < len+addlen; i++) {
rv[i] = add[i-len];
}
return rv;
}
CVEC_TYPE *CVEC_(diff)(CVEC_TYPE* x, cvec_uint len)
{
CVEC_TYPE *rv = malloc(sizeof(CVEC_TYPE)*(len-1));
cvec_uint i, j;
#pragma omp parallel for
for (i=0; i < (len-1); i++) {
j = i+1;
rv[i] = x[j] - x[i];
}
return rv;
}
CVEC_TYPE *CVEC_(slice)(
CVEC_TYPE *x,
cvec_uint len,
cvec_uint start,
cvec_uint stop,
cvec_uint skip)
{
if (stop > len)
cvec_ferr("cvec_slice","stop cannot exceed len");
if (skip == 0)
cvec_ferr(
"cvec_slice",
"cvec should not be less than one"
"if you want every value, set skip to 1.");
cvec_uint olen = (stop - start) / skip;
CVEC_TYPE *rv = malloc(sizeof(CVEC_TYPE)*olen);
for (cvec_uint i = start, j = 0; i < stop; i += skip, j++) {
rv[j] = x[i];
}
return rv;
}
CVEC_TYPE CVEC_(get_fit_sumse)(
CVEC_TYPE *x,
CVEC_TYPE *y,
cvec_uint len,
CVEC_TYPE *coefs,
cvec_uint ncoefs)
{
CVEC_TYPE calc_se(CVEC_TYPE o, CVEC_TYPE g) {
return pow(o - g, 2.0);
}
CVEC_TYPE calcfit(CVEC_TYPE xi) {
CVEC_TYPE rv = 0.0;
for (cvec_uint i = 0; i < ncoefs; i++)
rv += coefs[i] * pow(xi, (double)i);
return rv;
}
CVEC_TYPE *t = CVEC_(apply)(x, len, &calcfit);
CVEC_TYPE *se = CVEC_(apply2)(y, t, len, &calc_se);
CVEC_TYPE sumse = CVEC_(sum)(se, len);
free(t);
free(se);
return sumse;
}
CVEC_TYPE *CVEC_(polyfit)(
CVEC_TYPE *X,
CVEC_TYPE *Y,
cvec_uint len,
cvec_uint degree)
{
// https://github.com/natedomin/polyfit
cvec_uint maxdeg = 5;
CVEC_TYPE *B = calloc(maxdeg+1, sizeof(CVEC_TYPE));
CVEC_TYPE *P = calloc(((maxdeg+1)*2)+1, sizeof(CVEC_TYPE));
CVEC_TYPE *A = calloc((maxdeg+1)*2*(maxdeg+1), sizeof(CVEC_TYPE));
double x, y, powx;
cvec_uint i, j, k;
if (len <= degree || degree > maxdeg)
return NULL;
// Identify the column vector
for (i = 0; i < len; i++) {
x = X[i];
y = Y[i];
powx = 1.0;
for (j = 0; j < (degree + 1); j++) {
B[j] = B[j] + (y * powx);
powx = powx * x;
}
}
// Initialize the PowX array
P[0] = len;
// Compute the sum of the Powers of X
for (i = 0; i < len; i++) {
x = X[i];
powx = X[i];
for (j = 1; j < ((2 * (degree + 1)) + 1); j++) {
P[j] = P[j] + powx;
powx = powx * x;
}
}
// Initialize the reduction matrix
//
for (i = 0; i < (degree + 1); i++) {
for (j = 0; j < (degree + 1); j++) {
A[(i * (2 * (degree + 1))) + j] = P[i+j];
}
A[(i*(2 * (degree + 1))) + (i + (degree + 1))] = 1;
}
// Move the Identity matrix portion of the redux matrix
// to the left side (find the inverse of the left side
// of the redux matrix
for (i = 0; i < (degree + 1); i++) {
x = A[(i * (2 * (degree + 1))) + i];
if (x != 0) {
for (k = 0; k < (2 * (degree + 1)); k++) {
A[(i * (2 * (degree + 1))) + k] = A[(i * (2 * (degree + 1))) + k] / x;
}
for (j = 0; j < (degree + 1); j++) {
if ((j - i) != 0) {
y = A[(j * (2 * (degree + 1))) + i];
for (k = 0; k < (2 * (degree + 1)); k++) {
A[(j * (2 * (degree + 1))) + k] = A[(j * (2 * (degree + 1))) + k] - y * A[(i * (2 * (degree + 1))) + k];
}
}
}
}
else
{
// Cannot work with singular matrices
return NULL;
}
}
CVEC_TYPE *coefficients = calloc(degree+1, sizeof(CVEC_TYPE));
// Calculate and Identify the coefficients
for (i = 0; i < (degree + 1); i++) {
for (j = 0; j < (degree + 1); j++) {
x = 0;
for (k = 0; k < (degree + 1); k++) {
x = x + (A[(i * (2 * (degree + 1))) + (k + (degree + 1))] * B[k]);
}
coefficients[i] = x;
}
}
free(A);
free(B);
free(P);
return coefficients;
}
CVEC_TYPE *CVEC_(linearfit)(CVEC_TYPE *x, CVEC_TYPE *y, cvec_uint len)
{
CVEC_TYPE midx = CVEC_(average)(x, len);
CVEC_TYPE midy = CVEC_(average)(y, len);
CVEC_TYPE get_cvec_uint(CVEC_TYPE m) { return midy - m*midx; }
CVEC_TYPE get_sumse(
CVEC_TYPE *x,
CVEC_TYPE *y,
cvec_uint len,
CVEC_TYPE gradient) {
CVEC_TYPE cvec_uinterrupt = get_cvec_uint(gradient);
CVEC_TYPE applyfit(CVEC_TYPE v) { return cvec_uinterrupt+(gradient*v); }
CVEC_TYPE *fity = CVEC_(apply)(x, len, &applyfit);
CVEC_TYPE getse(CVEC_TYPE y1, CVEC_TYPE y2) {
return pow(y1 - y2, 2.0);
}
CVEC_TYPE *se = CVEC_(apply2)(y, fity, len, &getse);
CVEC_TYPE sumse = CVEC_(sum)(se, len);
free(se);
free(fity);
return sumse;
}
CVEC_TYPE change = 0.1, m = 1.0;
for (cvec_uint i = 0; i < len+100; i++) {
CVEC_TYPE err1 = get_sumse(x, y, len, m);
m += change;
CVEC_TYPE err2 = get_sumse(x, y, len, m);
if (err2 > err1) {
m -= change;
change *= -0.8;
}
else {
change *= 1.2;
}
}
CVEC_TYPE *coefs = malloc(sizeof(CVEC_TYPE)*2);
coefs[0] = get_cvec_uint(m);
coefs[1] = m;
return coefs;
}
CVEC_TYPE CVEC_(interpolate)(
CVEC_TYPE *x,
CVEC_TYPE *y,
cvec_uint len,
CVEC_TYPE ix)
{
//if (!CVEC_(in_order)(x, len))
// cvec_ferr("cvec_interpolate", "Interpolation expects ordered x.");
cvec_uint p1 = len, p2 = len;
if (ix < x[0]) {
cvec_warn("cvec_interpolate", "ix below start");
p1 = 0;
p2 = len*0.1;
if (p1 == p2) p2 ++;
}
else if (ix > x[len-1]) {
cvec_warn("cvec_interpolate", "ix after end");
p1 = len-1;
p2 = len*0.9;
if (p1 == p2) p2 --;
}
else {
// find surrounding xs
for (cvec_uint i = 0, j = 1; j < len; i++, j++) {
if (x[i] == ix)
return y[i];
if (x[j] == ix)
return y[j];
if (x[i] < ix && x[j] > ix) {
p1 = i;
p2 = j;
}
}
if (p1 == p2)
cvec_ferr("cvec_interpolate", "could not find points surrounding ix");
}
CVEC_TYPE m_num = (y[p1] - y[p2]);
CVEC_TYPE m_den = (x[p1] - x[p2]);
if (m_den == 0.0)
cvec_ferr("cvec_interpolate", "inf result!");
#ifdef FLOATING_TYPE
if (isnan(m_num) || isnan(m_den))
cvec_ferr("cvec_interpolate", "NaN result!");
#endif
CVEC_TYPE m = m_num / m_den;
CVEC_TYPE yi = (m * (ix - x[p2])) + y[p2];
return yi;
}
CVEC_TYPE *CVEC_(rearrange)(
CVEC_TYPE *x,
cvec_uint len,
cvec_uint *arrangement,
cvec_uint alen)
{
CVEC_TYPE *rv = CVEC_(ones)(alen);
for (cvec_uint i = 0; i < alen; i++) {
cvec_uint j = arrangement[i];
if (j >= len)
cvec_ferr(
"cvec_rearrange",
"new arrangement index outside of vector length (%u > %u)",
j, len);
rv[i] = x[j];
}
return rv;
}
typedef struct sc_td_t {
cvec_uint from;
cvec_uint to;
CVEC_TYPE *x;
CVEC_TYPE val;
} sc_td_t;
void *CVEC_(setconstthread)(void *vtd)
{
sc_td_t *td = (sc_td_t*)vtd;
for (cvec_uint i = td->from; i < td->to; i++) td->x[i] = td->val;
return NULL;
}
void CVEC_(set_constant)(CVEC_TYPE *x, cvec_uint len, CVEC_TYPE v)
{
pthread_t threads[cvec_njobs];
sc_td_t data[cvec_njobs];
cvec_uint each = len/cvec_njobs;
for (cvec_uint i = 0; i < cvec_njobs; i++) {
data[i].from = i*each;
data[i].to = (i == cvec_njobs-1) ? (len) : ((i+1)*each);
data[i].x = x;
data[i].val = v;
pthread_create(&threads[i], NULL, CVEC_(setconstthread), &data[i]);
}
for (cvec_uint i = 0; i < cvec_njobs; i++) {
pthread_join(threads[i], NULL);
}
}
CVEC_TYPE CVEC_(average)(CVEC_TYPE * in, cvec_uint len)
{
return CVEC_(mean)(in, len);
}
CVEC_TYPE CVEC_(mean)(CVEC_TYPE * in, cvec_uint len)
{
CVEC_TYPE sum = CVEC_(sum)(in, len);
return sum / ((CVEC_TYPE)len);
}
CVEC_TYPE CVEC_(median)(CVEC_TYPE * in, cvec_uint len)
{
cvec_uint midp = (len%2==0)?(len/2):((len+1)/2);
CVEC_TYPE *sorted;
CVEC_(sort)(in, len, NULL, &sorted);
CVEC_TYPE rv = sorted[midp];
free(sorted);
return rv;
}
CVEC_TYPE CVEC_(sumrange)(CVEC_TYPE *x, cvec_uint from, cvec_uint to)
{
CVEC_TYPE sum = 0.0;
for (cvec_uint i = from; i < to; i++)
sum += x[i];
return sum;
}
typedef struct sum_td_t {
cvec_uint from;
cvec_uint to;
CVEC_TYPE *x;
CVEC_TYPE *res;
cvec_uint id;
} sum_td_t;
void *CVEC_(sumthread)(void *vtd)
{
sum_td_t *td = (sum_td_t *)vtd;
td->res[td->id] += CVEC_(sumrange)(td->x, td->from, td->to);
return NULL;
}
CVEC_TYPE CVEC_(sum)(CVEC_TYPE * in, cvec_uint len)
{
CVEC_TYPE *sub_summed;
cvec_uint sub_len;
if (len < CVEC_LONG_LEN) {
sub_summed = in;
sub_len = len;
}
else {
sum_td_t data[cvec_njobs];
pthread_t threads[cvec_njobs];
cvec_uint each = len/cvec_njobs;
sub_summed = calloc(cvec_njobs, sizeof(CVEC_TYPE));
sub_len = cvec_njobs;
for (cvec_uint i = 0; i < cvec_njobs; i++) {
data[i].from = i*each;
data[i].to = (i == cvec_njobs-1) ? (len) : ((i+1)*each);
data[i].x = in;
data[i].res = sub_summed;
data[i].id = i;
pthread_create(&threads[i], NULL, CVEC_(sumthread), &data[i]);
}
for (cvec_uint i = 0; i < cvec_njobs; i++) {
pthread_join(threads[i], NULL);
}
}
CVEC_TYPE sum = CVEC_(sumrange)(sub_summed, 0, sub_len);
if (len >= CVEC_LONG_LEN)
free(sub_summed);
return sum;
}
CVEC_TYPE CVEC_(prod)(CVEC_TYPE * in, cvec_uint len)
{
CVEC_TYPE prod = 1.0;
for (cvec_uint i = 0; i < len; i++) {
prod *= in[i];
}
return prod;
}
|
sort.h | /*
// Copyright (c) 2000-2009, Texas Engineering Experiment Station (TEES), a
// component of the Texas A&M University System.
// All rights reserved.
// The information and source code contained herein is the exclusive
// property of TEES and may not be disclosed, examined or reproduced
// in whole or in part without explicit written authorization from TEES.
*/
#ifndef STAPL_BENCHMARKS_FMM_SORT_H
#define STAPL_BENCHMARKS_FMM_SORT_H
#include "types.h"
#ifndef _OPENMP
int omp_get_num_threads() {return 1;}
int omp_get_thread_num() {return 0;}
#else
#include <omp.h>
#endif
/// Custom radix sort for body and structures
class Sort
{
private:
/// Output buffer
Bodies output;
private:
/// Radixsorts the values using the keys
void radixsort(int * key, int * value, int size)
{
// Number of bits in one stride
const int bitStride = 8;
// Size of stride in decimal
const int stride = 1 << bitStride;
// Mask the bits in one stride
const int mask = stride - 1;
// Number of OpenMP threads
int numThreads;
// Maximum value of key
int maxKey = 0;
// Bucket for each thread
int (*bucketPerThread)[stride];
// MaxKey for each thread
int * maxKeyPerThread;
// Buffer for both key and value
int * buffer = new int [size];
// Permutation index
int * permutation = new int [size];
#ifdef _OPENMP
#pragma omp parallel
#endif
// Start OpenMP parallel clause
{
// Get number of OpenMP threads
numThreads = omp_get_num_threads();
#ifdef _OPENMP
#pragma omp single
#endif
// Start serial clause
{
// Allocate bucket per thread
bucketPerThread = new int [numThreads][stride]();
// Allocate maxKey per thread
maxKeyPerThread = new int [numThreads];
// Loop over threads
for (int i=0; i<numThreads; i++)
// Initialize maxKey per thread
maxKeyPerThread[i] = 0;
// End serial clause
}
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over keys
for ( int i=0; i<size; i++ )
// If key is larger than maxKey
if ( key[i] > maxKeyPerThread[omp_get_thread_num()] )
// Update maxKey per thread
maxKeyPerThread[omp_get_thread_num()] = key[i];
#ifdef _OPENMP
#pragma omp single
#endif
// Loop over threads
for ( int i=0; i<numThreads; i++ )
// Update maxKey
if ( maxKeyPerThread[i] > maxKey ) maxKey = maxKeyPerThread[i];
// While there are bits in maxKey to process
while ( maxKey > 0 ) {
// Initialize bucket
int bucket[stride] = {0};
#ifdef _OPENMP
#pragma omp single
#endif
// Loop over threads
for ( int t=0; t<numThreads; t++ )
// Loop over strides
for ( int i=0; i<stride; i++ )
// Initialize bucket per thread
bucketPerThread[t][i] = 0;
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over keys
for ( int i=0; i<size; i++ )
// Increment bucket
bucketPerThread[omp_get_thread_num()][key[i] & mask]++;
#ifdef _OPENMP
#pragma omp single
#endif
// Start serial clause
{
// Loop over threads
for ( int t=0; t<numThreads; t++ )
// Loop over strides
for ( int i=0; i<stride; i++ )
// Update bucket from all threads
bucket[i] += bucketPerThread[t][i];
// Loop over strides
for ( int i=1; i<stride; i++ )
// Scan bucket over strides
bucket[i] += bucket[i-1];
// Loop over keys backwards
for ( int i=size-1; i>=0; i-- )
// Reverse scan bucket to get permutation
permutation[i] = --bucket[key[i] & mask];
// End serial clause
}
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over values
for ( int i=0; i<size; i++ )
// Sort into buffer
buffer[permutation[i]] = value[i];
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over values
for ( int i=0; i<size; i++ )
// Copy back from buffer
value[i] = buffer[i];
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over keys
for ( int i=0; i<size; i++ )
// Sort into buffer
buffer[permutation[i]] = key[i];
#ifdef _OPENMP
#pragma omp for
#endif
// Loop over keys
for ( int i=0; i<size; i++ )
// Copy back from buffer and bit shift keys
key[i] = buffer[i] >> bitStride;
#ifdef _OPENMP
#pragma omp single
#endif
// Bit shift maxKey
maxKey >>= bitStride;
// End while for bits to process
}
// End OpenMP parallel clause
}
// Deallocate bucket per thread
delete[] bucketPerThread;
// Deallocate maxKey per thread
delete[] maxKeyPerThread;
// Deallocate buffer
delete[] buffer;
// Deallocate permutation index
delete[] permutation;
}
public:
/// Sort input according to ibody
Bodies ibody(Bodies & input)
{
// Size of bodies vector
const int size = input.size();
// Allocate key array
int * key = new int [size];
// Allocate index array
int * index = new int [size];
// Loop over input bodies
for (B_iter B=input.begin(); B!=input.end(); B++) {
// Body index
int i = B-input.begin();
// Copy IBODY to key array
key[i] = B->IBODY;
// Initialize index array
index[i] = i;
// End loop over input bodies
}
// Radix sort index according to key
radixsort(key,index,size);
// Resize output buffer
output.resize(size);
// Loop over output bodies
for (B_iter B=output.begin(); B!=output.end(); B++) {
// Body index
int i = B-output.begin();
// Permute according to index
*B = input[index[i]];
// End loop over output bodies
}
// Deallocate key array
delete[] key;
// Deallocate index array
delete[] index;
// Return output
return output;
}
/// Sort input according to irank
Bodies irank(Bodies & input)
{
// Size of bodies vector
const int size = input.size();
// Allocate key array
int * key = new int [size];
// Allocate index array
int * index = new int [size];
// Loop over input bodies
for (B_iter B=input.begin(); B!=input.end(); B++) {
// Body index
int i = B-input.begin();
// Copy IRANK to key array
key[i] = B->IRANK;
// Initialize index array
index[i] = i;
// End loop over input bodies
}
// Radix sort index according to key
radixsort(key,index,size);
// Resize output buffer
output.resize(size);
// Loop over output bodies
for (B_iter B=output.begin(); B!=output.end(); B++) {
// Body index
int i = B-output.begin();
// Permute according to index
*B = input[index[i]];
// End loop over output bodies
}
// Deallocate key array
delete[] key;
// Deallocate index array
delete[] index;
// Return output
return output;
}
/// Sort bodies back to original order
Bodies unsort(Bodies & bodies)
{
// Sort bodies
bodies = ibody(bodies);
return bodies;
}
};
#endif // STAPL_BENCHMARKS_FMM_SORT_H
|
mlp_example_f32_numa.c | /******************************************************************************
* Copyright (c) Intel Corporation - All rights reserved. *
* This file is part of the LIBXSMM library. *
* *
* For information on the license, see the LICENSE file. *
* Further information: https://github.com/hfp/libxsmm/ *
* SPDX-License-Identifier: BSD-3-Clause *
******************************************************************************/
/* Alexander Heinecke (Intel Corp.)
******************************************************************************/
#include <libxsmm.h>
#include <libxsmm_sync.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <math.h>
#if defined(_OPENMP)
# include <omp.h>
#endif
#include <numa.h>
#define CHECK_L1
/* include c-based dnn library */
#include "../common/dnn_common.h"
LIBXSMM_INLINE void my_init_buf(float* buf, size_t size, int initPos, int initOne)
{
int i;
zero_buf(buf, size);
for (i = 0; i < (int)size; ++i) {
buf[i] = (float)((initOne != 0) ? 1.0 : ((initPos != 0) ? libxsmm_rng_f64() : (0.05 - libxsmm_rng_f64()/10.0)));
}
}
typedef enum my_eltwise_fuse {
MY_ELTWISE_FUSE_NONE = 0,
MY_ELTWISE_FUSE_BIAS = 1,
MY_ELTWISE_FUSE_RELU = 2,
MY_ELTWISE_FUSE_BIAS_RELU = MY_ELTWISE_FUSE_BIAS | MY_ELTWISE_FUSE_RELU
} my_eltwise_fuse;
typedef enum my_pass {
MY_PASS_FWD = 1,
MY_PASS_BWD_D = 2,
MY_PASS_BWD_W = 4,
MY_PASS_BWD = 6
} my_pass;
typedef struct my_opt_config {
libxsmm_blasint C;
libxsmm_blasint K;
libxsmm_blasint bc;
libxsmm_blasint bk;
libxsmm_blasint threads;
float lr;
size_t scratch_size;
libxsmm_barrier* barrier;
} my_opt_config;
typedef struct my_smax_fwd_config {
libxsmm_blasint N;
libxsmm_blasint C;
libxsmm_blasint bn;
libxsmm_blasint bc;
libxsmm_blasint threads;
size_t scratch_size;
libxsmm_barrier* barrier;
} my_smax_fwd_config;
typedef struct my_smax_bwd_config {
libxsmm_blasint N;
libxsmm_blasint C;
libxsmm_blasint bn;
libxsmm_blasint bc;
libxsmm_blasint threads;
size_t scratch_size;
float loss_weight;
libxsmm_barrier* barrier;
} my_smax_bwd_config;
typedef struct my_fc_fwd_config {
libxsmm_blasint N;
libxsmm_blasint C;
libxsmm_blasint K;
libxsmm_blasint bn;
libxsmm_blasint bc;
libxsmm_blasint bk;
libxsmm_blasint threads;
my_eltwise_fuse fuse_type;
libxsmm_blasint fwd_bf;
libxsmm_blasint fwd_2d_blocking;
libxsmm_blasint fwd_col_teams;
libxsmm_blasint fwd_row_teams;
size_t scratch_size;
libxsmm_barrier* barrier;
libxsmm_smmfunction_reducebatch_strd gemm_fwd;
libxsmm_smmfunction_reducebatch_strd gemm_fwd2;
} my_fc_fwd_config;
typedef struct my_fc_bwd_config {
libxsmm_blasint N;
libxsmm_blasint C;
libxsmm_blasint K;
libxsmm_blasint bn;
libxsmm_blasint bc;
libxsmm_blasint bk;
libxsmm_blasint threads;
my_eltwise_fuse fuse_type;
libxsmm_blasint bwd_bf;
libxsmm_blasint bwd_2d_blocking;
libxsmm_blasint bwd_col_teams;
libxsmm_blasint bwd_row_teams;
libxsmm_blasint upd_bf;
libxsmm_blasint upd_2d_blocking;
libxsmm_blasint upd_col_teams;
libxsmm_blasint upd_row_teams;
libxsmm_blasint ifm_subtasks;
libxsmm_blasint ofm_subtasks;
size_t scratch_size;
libxsmm_barrier* barrier;
libxsmm_smmfunction_reducebatch_strd gemm_bwd;
libxsmm_smmfunction_reducebatch_strd gemm_bwd2;
libxsmm_smmfunction_reducebatch_strd gemm_upd;
libxsmm_smmfunction_reducebatch_strd gemm_upd2;
libxsmm_meltwfunction_unary norm_to_normT_kernel;
} my_fc_bwd_config;
typedef struct my_numa_thr_cfg {
int thr_s;
int thr_e;
int *blocksOFm_s;
int *blocksOFm_e;
int *blocksIFm_s;
int *blocksIFm_e;
int *blocksOFm_tr_s;
int *blocksOFm_tr_e;
int *blocksIFm_tr_s;
int *blocksIFm_tr_e;
float **scratch;
size_t *layer_size;
int **fwd_ofm_to_numa;
float *bwd_d_scratch;
size_t bwd_d_scratch_size;
float *bwd_w_scratch;
size_t bwd_w_layer_size;
} my_numa_thr_cfg;
my_fc_fwd_config setup_my_fc_fwd(libxsmm_blasint N, libxsmm_blasint C, libxsmm_blasint K, libxsmm_blasint bn,
libxsmm_blasint bc, libxsmm_blasint bk, libxsmm_blasint threads, my_eltwise_fuse fuse_type) {
my_fc_fwd_config res;
libxsmm_blasint lda = bk;
libxsmm_blasint ldb = bc;
libxsmm_blasint ldc = bk;
float alpha = 1.0f;
float beta = 1.0f;
float zerobeta = 0.0f;
/* setting up some handle values */
res.N = N;
res.C = C;
res.K = K;
res.bn = bn;
res.bc = bc;
res.bk = bk;
res.threads = threads;
res.fuse_type = fuse_type;
/* setup parallelization strategy */
if (threads == 16) {
res.fwd_bf = 1;
res.fwd_2d_blocking = 1;
res.fwd_col_teams = 2;
res.fwd_row_teams = 8;
} else {
res.fwd_bf = 1;
res.fwd_2d_blocking = 0;
res.fwd_col_teams = 1;
res.fwd_row_teams = 1;
}
#if 0
res.fwd_bf = atoi(getenv("FWD_BF"));
res.fwd_2d_blocking = atoi(getenv("FWD_2D_BLOCKING"));
res.fwd_col_teams = atoi(getenv("FWD_COL_TEAMS"));
res.fwd_row_teams = atoi(getenv("FWD_ROW_TEAMS"));
#endif
/* setting up the barrier */
res.barrier = libxsmm_barrier_create(threads, 1);
/* TPP creation */
res.gemm_fwd = libxsmm_smmdispatch_reducebatch_strd(res.bk, res.bn, res.bc,
res.bk*res.bc*sizeof(float), res.bc*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &beta, NULL, NULL);
if ( res.gemm_fwd == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_fwd failed. Bailing...!\n");
exit(-1);
}
res.gemm_fwd2 = libxsmm_smmdispatch_reducebatch_strd(res.bk, res.bn, res.bc,
res.bk*res.bc*sizeof(float), res.bc*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &zerobeta, NULL, NULL);
if ( res.gemm_fwd2 == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_fwd2 failed. Bailing...!\n");
exit(-1);
}
/* init scratch */
res.scratch_size = 0;
return res;
}
my_fc_bwd_config setup_my_fc_bwd(libxsmm_blasint N, libxsmm_blasint C, libxsmm_blasint K, libxsmm_blasint bn,
libxsmm_blasint bc, libxsmm_blasint bk, libxsmm_blasint threads, my_eltwise_fuse fuse_type) {
my_fc_bwd_config res;
libxsmm_blasint lda = bc;
libxsmm_blasint ldb = bk;
libxsmm_blasint ldc = bc;
float alpha = 1.0f;
float beta = 1.0f;
float zerobeta = 0.0f;
int updflags = LIBXSMM_GEMM_FLAGS( 'N', 'T' );
libxsmm_blasint updM;
libxsmm_blasint updN;
/* setting up some handle values */
res.N = N;
res.C = C;
res.K = K;
res.bn = bn;
res.bc = bc;
res.bk = bk;
res.threads = threads;
res.fuse_type = fuse_type;
/* setup parallelization strategy */
if (threads == 16) {
res.bwd_bf = 1;
res.bwd_2d_blocking = 1;
res.bwd_col_teams = 2;
res.bwd_row_teams = 8;
res.upd_bf = 1;
res.upd_2d_blocking = 0;
res.upd_col_teams = 1;
res.upd_row_teams = 1;
res.ifm_subtasks = 1;
res.ofm_subtasks = 1;
} else {
res.bwd_bf = 1;
res.bwd_2d_blocking = 0;
res.bwd_col_teams = 1;
res.bwd_row_teams = 1;
res.upd_bf = 1;
res.upd_2d_blocking = 0;
res.upd_col_teams = 1;
res.upd_row_teams = 1;
res.ifm_subtasks = 1;
res.ofm_subtasks = 1;
}
#if 0
res.bwd_bf = atoi(getenv("BWD_BF"));
res.bwd_2d_blocking = atoi(getenv("BWD_2D_BLOCKING"));
res.bwd_col_teams = atoi(getenv("BWD_COL_TEAMS"));
res.bwd_row_teams = atoi(getenv("BWD_ROW_TEAMS"));
res.upd_bf = atoi(getenv("UPD_BF"));
res.upd_2d_blocking = atoi(getenv("UPD_2D_BLOCKING"));
res.upd_col_teams = atoi(getenv("UPD_COL_TEAMS"));
res.upd_row_teams = atoi(getenv("UPD_ROW_TEAMS"));
res.ifm_subtasks = atoi(getenv("IFM_SUBTASKS"));
res.ofm_subtasks = atoi(getenv("OFM_SUBTASKS"));
#endif
/* setting up the barrier */
res.barrier = libxsmm_barrier_create(threads, 1);
/* TPP creation */
/* BWD GEMM */
res.gemm_bwd = libxsmm_smmdispatch_reducebatch_strd(res.bc, res.bn, res.bk,
res.bk*res.bc*sizeof(float), res.bk*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &beta, NULL, NULL);
if ( res.gemm_bwd == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_bwd failed. Bailing...!\n");
exit(-1);
}
res.gemm_bwd2 = libxsmm_smmdispatch_reducebatch_strd(res.bc, res.bn, res.bk,
res.bk*res.bc*sizeof(float), res.bk*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &zerobeta, NULL, NULL);
if ( res.gemm_bwd2 == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_bwd2 failed. Bailing...!\n");
exit(-1);
}
res.norm_to_normT_kernel = libxsmm_dispatch_meltw_unary(bk, bc, &ldaT, &lda, LIBXSMM_DATATYPE_F32, LIBXSMM_DATATYPE_F32, LIBXSMM_DATATYPE_F32, LIBXSMM_MELTW_FLAG_UNARY_NONE, LIBXSMM_MELTW_TYPE_UNARY_TRANSFORM_NORM_TO_NORMT);
if ( res.norm_to_normT_kernel == NULL ) {
fprintf( stderr, "JIT for TPP norm_to_normT_kernel failed. Bailing...!\n");
exit(-1);
}
/* UPD GEMM */
lda = res.bk;
ldb = res.bc;
ldc = res.bk;
updM = res.bk/res.ofm_subtasks;
updN = res.bc/res.ifm_subtasks;
res.gemm_upd = libxsmm_smmdispatch_reducebatch_strd(updM, updN, res.bn,
res.K*res.bn*sizeof(float), res.C*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &beta, &updflags, NULL);
if ( res.gemm_upd == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_upd failed. Bailing...!\n");
exit(-1);
}
res.gemm_upd2 = libxsmm_smmdispatch_reducebatch_strd(updM, updN, res.bn,
res.K*res.bn*sizeof(float), res.C*res.bn*sizeof(float),
&lda, &ldb, &ldc, &alpha, &zerobeta, &updflags, NULL);
if ( res.gemm_upd2 == NULL ) {
fprintf( stderr, "JIT for BRGEMM TPP gemm_upd2 failed. Bailing...!\n");
exit(-1);
}
/* init scratch */
res.scratch_size = sizeof(float) * ( (((size_t)res.C + (size_t)res.K) * (size_t)res.N) + ((size_t)res.C * (size_t)res.K) );
return res;
}
my_opt_config setup_my_opt(libxsmm_blasint C, libxsmm_blasint K, libxsmm_blasint bc, libxsmm_blasint bk,
libxsmm_blasint threads, float lr) {
my_opt_config res;
/* setting up some handle values */
res.C = C;
res.K = K;
res.bc = bc;
res.bk = bk;
res.threads = threads;
res.lr = lr;
/* setting up the barrier */
res.barrier = libxsmm_barrier_create(threads, 1);
/* init scratch */
res.scratch_size = 0;
return res;
}
my_smax_fwd_config setup_my_smax_fwd(libxsmm_blasint N, libxsmm_blasint C, libxsmm_blasint bn, libxsmm_blasint bc,
libxsmm_blasint threads) {
my_smax_fwd_config res;
/* setting up some handle values */
res.C = C;
res.N = N;
res.bc = bc;
res.bn = bn;
res.threads = threads;
/* setting up the barrier */
res.barrier = libxsmm_barrier_create(threads, 1);
/* init scratch */
res.scratch_size = 0;
return res;
}
my_smax_bwd_config setup_my_smax_bwd(libxsmm_blasint N, libxsmm_blasint C, libxsmm_blasint bn, libxsmm_blasint bc,
libxsmm_blasint threads, float loss_weight) {
my_smax_bwd_config res;
/* setting up some handle values */
res.C = C;
res.N = N;
res.bc = bc;
res.bn = bn;
res.threads = threads;
res.loss_weight = loss_weight;
/* setting up the barrier */
res.barrier = libxsmm_barrier_create(threads, 1);
/* init scratch */
res.scratch_size = 0;
return res;
}
void my_fc_fwd_exec( my_fc_fwd_config cfg, const float* in_act_ptr, float* out_act_ptr,
const float* bias_ptr, unsigned char* relu_ptr, int start_tid, int my_tid, void* scratch, my_numa_thr_cfg *numa_thr_cfg, int layer) {
const libxsmm_blasint nBlocksIFm = cfg.C / cfg.bc;
const libxsmm_blasint nBlocksOFm = cfg.K / cfg.bk;
const libxsmm_blasint nBlocksMB = cfg.N / cfg.bn;
/* computing first logical thread */
const libxsmm_blasint ltid = my_tid - start_tid;
/* number of tasks that could be run in parallel */
const libxsmm_blasint work = nBlocksOFm * nBlocksMB;
/* compute chunk size */
const libxsmm_blasint chunksize = (work % cfg.threads == 0) ?
(work / cfg.threads) : ((work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (ltid * chunksize < work) ? (ltid * chunksize) : work;
const libxsmm_blasint thr_end = ((ltid + 1) * chunksize < work) ? ((ltid + 1) * chunksize) : work;
/* loop variables */
libxsmm_blasint mb1ofm1 = 0, mb1 = 0, ofm1 = 0, ifm1 = 0, mb2 = 0, ofm2 = 0;
libxsmm_blasint N_tasks_per_thread = 0, M_tasks_per_thread = 0;
libxsmm_blasint my_M_start = 0, my_M_end = 0, my_N_start = 0, my_N_end = 0;
libxsmm_blasint my_col_id = 0, my_row_id = 0, col_teams = 0, row_teams = 0;
LIBXSMM_VLA_DECL(4, float, output, out_act_ptr, nBlocksOFm, cfg.bn, cfg.bk);
LIBXSMM_VLA_DECL(4, const float, input, in_act_ptr, nBlocksIFm, cfg.bn, cfg.bc);
LIBXSMM_VLA_DECL(4, const float, filter, numa_thr_cfg->scratch[layer], nBlocksIFm, cfg.bc, cfg.bk);
LIBXSMM_VLA_DECL(2, const float, bias, bias_ptr, cfg.bk);
LIBXSMM_VLA_DECL(4, unsigned char, relumask, relu_ptr, nBlocksOFm, cfg.bn, cfg.bk);
unsigned long long blocks = nBlocksIFm;
libxsmm_blasint CB_BLOCKS = nBlocksIFm, BF = 1;
LIBXSMM_UNUSED( scratch );
BF = cfg.fwd_bf;
CB_BLOCKS = nBlocksIFm/BF;
blocks = CB_BLOCKS;
col_teams = cfg.fwd_col_teams;
row_teams = cfg.fwd_row_teams;
my_row_id = ltid % row_teams;
my_col_id = ltid / row_teams;
N_tasks_per_thread = LIBXSMM_UPDIV(nBlocksMB, col_teams);
M_tasks_per_thread = LIBXSMM_UPDIV(nBlocksOFm, row_teams);
my_N_start = LIBXSMM_MIN(my_col_id * N_tasks_per_thread, nBlocksMB);
my_N_end = LIBXSMM_MIN((my_col_id+1) * N_tasks_per_thread, nBlocksMB);
my_M_start = LIBXSMM_MIN(my_row_id * M_tasks_per_thread, nBlocksOFm);
my_M_end = LIBXSMM_MIN((my_row_id+1) * M_tasks_per_thread, nBlocksOFm);
const libxsmm_blasint ofm_start = numa_thr_cfg->blocksOFm_s[layer];
/* lazy barrier init */
libxsmm_barrier_init(cfg.barrier, ltid);
if (cfg.fwd_2d_blocking == 1) {
if (BF > 1) {
for (ifm1 = 0; ifm1 < BF; ++ifm1) {
for (ofm1 = my_M_start; ofm1 < my_M_end; ++ofm1) {
for (mb1 = my_N_start; mb1 < my_N_end; ++mb1) {
/* Initialize output slice */
if ( ifm1 == 0 ) {
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_BIAS) == MY_ELTWISE_FUSE_BIAS ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = LIBXSMM_VLA_ACCESS(2, bias, ofm1, ofm2, cfg.bk);
}
}
} else {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (float)0;
}
}
}
}
/* BRGEMM */
cfg.gemm_fwd( &LIBXSMM_VLA_ACCESS(4, filter, ofm1, ifm1*CB_BLOCKS, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, ifm1*CB_BLOCKS, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
/* apply post BRGEMM fusion */
if ( ifm1 == BF-1 ) {
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
float l_cur_out = LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
LIBXSMM_VLA_ACCESS(4, relumask, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (unsigned char)(( l_cur_out > (float)0 ) ? 1 : 0);
l_cur_out = (l_cur_out > (float)0) ? l_cur_out : (float)0;
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = l_cur_out;
}
}
}
}
}
}
}
} else {
for (ofm1 = my_M_start; ofm1 < my_M_end; ++ofm1) {
for (mb1 = my_N_start; mb1 < my_N_end; ++mb1) {
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_BIAS) == MY_ELTWISE_FUSE_BIAS ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = LIBXSMM_VLA_ACCESS(2, bias, ofm1, ofm2, cfg.bk);
}
}
cfg.gemm_fwd( &LIBXSMM_VLA_ACCESS(4, filter, ofm1-ofm_start, 0, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, 0, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
} else {
cfg.gemm_fwd2( &LIBXSMM_VLA_ACCESS(4, filter, ofm1-ofm_start, 0, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, 0, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
}
/* post GEMM fusion */
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
float l_cur_out = LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
LIBXSMM_VLA_ACCESS(4, relumask, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (unsigned char)(( l_cur_out > (float)0 ) ? 1 : 0);
l_cur_out = ( l_cur_out > (float)0 ) ? l_cur_out : (float)0;
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = l_cur_out;
}
}
}
}
}
}
} else {
if (BF > 1) {
for ( ifm1 = 0; ifm1 < BF; ++ifm1 ) {
for ( mb1ofm1 = thr_begin; mb1ofm1 < thr_end; ++mb1ofm1 ) {
mb1 = mb1ofm1%nBlocksMB;
ofm1 = mb1ofm1/nBlocksMB;
/* Initialize output slice */
if ( ifm1 == 0 ) {
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_BIAS) == MY_ELTWISE_FUSE_BIAS ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = LIBXSMM_VLA_ACCESS(2, bias, ofm1, ofm2, cfg.bk);
}
}
} else {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (float)0;
}
}
}
}
/* BRGEMM */
cfg.gemm_fwd( &LIBXSMM_VLA_ACCESS(4, filter, ofm1, ifm1*CB_BLOCKS, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, ifm1*CB_BLOCKS, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
/* post GEMM fusion */
if ( ifm1 == BF-1 ) {
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
float l_cur_out = LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
LIBXSMM_VLA_ACCESS(4, relumask, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (unsigned char)(( l_cur_out > (float)0 ) ? 1 : 0);
l_cur_out = (l_cur_out > (float)0) ? l_cur_out : (float)0;
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = l_cur_out;
}
}
}
}
}
}
} else {
for ( mb1ofm1 = thr_begin; mb1ofm1 < thr_end; ++mb1ofm1 ) {
mb1 = mb1ofm1%nBlocksMB;
ofm1 = mb1ofm1/nBlocksMB;
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_BIAS) == MY_ELTWISE_FUSE_BIAS ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = LIBXSMM_VLA_ACCESS(2, bias, ofm1, ofm2, cfg.bk);
}
}
cfg.gemm_fwd( &LIBXSMM_VLA_ACCESS(4, filter, ofm1-ofm_start, 0, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, 0, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
} else {
cfg.gemm_fwd2( &LIBXSMM_VLA_ACCESS(4, filter, ofm1-ofm_start, 0, 0, 0, nBlocksIFm, cfg.bc, cfg.bk),
&LIBXSMM_VLA_ACCESS(4, input, mb1, 0, 0, 0, nBlocksIFm, cfg.bn, cfg.bc),
&LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, 0, 0, nBlocksOFm, cfg.bn, cfg.bk), &blocks);
}
/* post GEMM fusion */
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
float l_cur_out = LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
LIBXSMM_VLA_ACCESS(4, relumask, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = (unsigned char)(( l_cur_out > (float)0 ) ? 1 : 0);
l_cur_out = ( l_cur_out > (float)0 ) ? l_cur_out : (float)0;
LIBXSMM_VLA_ACCESS(4, output, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = l_cur_out;
}
}
}
}
}
}
libxsmm_barrier_wait(cfg.barrier, ltid);
}
void my_fc_bwd_d_transpose( my_fc_bwd_config cfg, int my_tid, my_numa_thr_cfg **numa_thr_cfg_, int numa_node, int layer, int *ofm_to_node) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
/* here we assume that input and output blocking is similar */
const libxsmm_blasint bk = cfg.bk;
const libxsmm_blasint bc = cfg.bc;
const libxsmm_blasint nBlocksIFm = cfg.C / bc;
const libxsmm_blasint nBlocksOFm = cfg.K / bk;
/* computing first logical thread */
const libxsmm_blasint ltid = my_tid - numa_thr_cfg[numa_node].thr_s;
const libxsmm_blasint l_nBlocksIFm = (numa_thr_cfg[numa_node].blocksIFm_tr_e[layer] - numa_thr_cfg[numa_node].blocksIFm_tr_s[layer]) + 1;
/* number of tasks for transpose that could be run in parallel */
const libxsmm_blasint transpose_work = l_nBlocksIFm * nBlocksOFm;
/* compute chunk size */
int thr = numa_thr_cfg[numa_node].thr_e - numa_thr_cfg[numa_node].thr_s;
const libxsmm_blasint transpose_chunksize = (transpose_work % thr == 0) ? (transpose_work / thr) : ((transpose_work / thr) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint transpose_thr_begin = (ltid * transpose_chunksize < transpose_work) ? (ltid * transpose_chunksize) : transpose_work;
const libxsmm_blasint transpose_thr_end = ((ltid + 1) * transpose_chunksize < transpose_work) ? ((ltid + 1) * transpose_chunksize) : transpose_work;
float *filter_tr = numa_thr_cfg[numa_node].bwd_d_scratch;
libxsmm_meltw_unary_param trans_param;
/* lazy barrier init */
libxsmm_barrier_init(cfg.barrier, my_tid);
/* transpose weight */
int ifm1ofm1 = 0;
for (ifm1ofm1 = transpose_thr_begin; ifm1ofm1 < transpose_thr_end; ++ifm1ofm1) {
const unsigned int ubk = (unsigned int)bk;
const unsigned int ubc = (unsigned int)bc;
int ofm1 = ifm1ofm1 / l_nBlocksIFm;
int ifm1 = ifm1ofm1 % l_nBlocksIFm;
my_numa_thr_cfg *l_numa_thr_cfg = &numa_thr_cfg[ofm_to_node[ofm1]];
float *inp = l_numa_thr_cfg->scratch[layer];
inp = inp + (ofm1 - l_numa_thr_cfg->blocksOFm_s[layer]) * nBlocksIFm * bc * bk
+ (ifm1 + numa_thr_cfg[numa_node].blocksIFm_tr_s[layer]) * bc * bk;
float *out = filter_tr + ifm1 * nBlocksOFm * bk * bc + ofm1 * bk * bc;
trans_param.in.primary = (void*)inp;
trans_param.out.primary = out;
cfg.norm_to_normT_kernel(&trans_param);
}
libxsmm_barrier_wait(cfg.barrier, my_tid);
}
void my_fc_bwd_exec( my_fc_bwd_config cfg, float* din_act_ptr,
float* dout_act_ptr, float* dwt_ptr, const float* in_act_ptr,
float* dbias_ptr, const unsigned char* relu_ptr, my_pass pass, int start_tid, int my_tid, void* scratch, my_numa_thr_cfg *numa_thr_cfg, int layer ) {
/* here we assume that input and output blocking is similar */
const libxsmm_blasint bn = cfg.bn;
const libxsmm_blasint bk = cfg.bk;
const libxsmm_blasint bc = cfg.bc;
const libxsmm_blasint nBlocksIFm = cfg.C / bc;
const libxsmm_blasint nBlocksOFm = cfg.K / bk;
const libxsmm_blasint nBlocksMB = cfg.N / bn;
/* computing first logical thread */
const libxsmm_blasint ltid = my_tid - start_tid;
/* number of tasks for transpose that could be run in parallel */
const libxsmm_blasint eltwise_work = nBlocksOFm * nBlocksMB;
/* compute chunk size */
const libxsmm_blasint eltwise_chunksize = (eltwise_work % cfg.threads == 0) ? (eltwise_work / cfg.threads) : ((eltwise_work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint eltwise_thr_begin = (ltid * eltwise_chunksize < eltwise_work) ? (ltid * eltwise_chunksize) : eltwise_work;
const libxsmm_blasint eltwise_thr_end = ((ltid + 1) * eltwise_chunksize < eltwise_work) ? ((ltid + 1) * eltwise_chunksize) : eltwise_work;
libxsmm_blasint mb1ofm1;
/* number of tasks for transpose that could be run in parallel */
const libxsmm_blasint dbias_work = nBlocksOFm;
/* compute chunk size */
const libxsmm_blasint dbias_chunksize = (dbias_work % cfg.threads == 0) ? (dbias_work / cfg.threads) : ((dbias_work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint dbias_thr_begin = (ltid * dbias_chunksize < dbias_work) ? (ltid * dbias_chunksize) : dbias_work;
const libxsmm_blasint dbias_thr_end = ((ltid + 1) * dbias_chunksize < dbias_work) ? ((ltid + 1) * dbias_chunksize) : dbias_work;
/* loop variables */
libxsmm_blasint ofm1 = 0, mb1 = 0, ofm2 = 0, mb2 = 0;
float *grad_output_ptr = (((cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU) ? ((float*)scratch)+(cfg.C*cfg.K) : dout_act_ptr);
LIBXSMM_VLA_DECL(4, const float, doutput_orig, dout_act_ptr, nBlocksOFm, bn, bk);
LIBXSMM_VLA_DECL(4, float, doutput, grad_output_ptr, nBlocksOFm, bn, bk);
LIBXSMM_VLA_DECL(2, float, dbias, dbias_ptr, cfg.bk);
LIBXSMM_VLA_DECL(4, const unsigned char, relumask, relu_ptr, nBlocksOFm, cfg.bn, cfg.bk);
const libxsmm_blasint ifm_start = numa_thr_cfg->blocksIFm_tr_s[layer];
/* lazy barrier init */
libxsmm_barrier_init(cfg.barrier, ltid);
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_RELU) == MY_ELTWISE_FUSE_RELU ) {
for ( mb1ofm1 = eltwise_thr_begin; mb1ofm1 < eltwise_thr_end; ++mb1ofm1 ) {
mb1 = mb1ofm1%nBlocksMB;
ofm1 = mb1ofm1/nBlocksMB;
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
float l_cur_out = LIBXSMM_VLA_ACCESS(4, doutput_orig, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
l_cur_out = (LIBXSMM_VLA_ACCESS(4, relumask, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) != 0) ? l_cur_out : (float)0;
LIBXSMM_VLA_ACCESS(4, doutput, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk) = l_cur_out;
}
}
}
/* wait for eltwise to finish */
libxsmm_barrier_wait(cfg.barrier, ltid);
}
if ( (cfg.fuse_type & MY_ELTWISE_FUSE_BIAS) == MY_ELTWISE_FUSE_BIAS ) {
for ( ofm1 = dbias_thr_begin; ofm1 < dbias_thr_end; ++ofm1 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS( 2, dbias, ofm1, ofm2, cfg.bk ) = 0.0f;
}
for ( mb1 = 0; mb1 < nBlocksMB; ++mb1 ) {
for ( mb2 = 0; mb2 < cfg.bn; ++mb2 ) {
for ( ofm2 = 0; ofm2 < cfg.bk; ++ofm2 ) {
LIBXSMM_VLA_ACCESS( 2, dbias, ofm1, ofm2, cfg.bk ) += LIBXSMM_VLA_ACCESS(4, doutput, mb1, ofm1, mb2, ofm2, nBlocksOFm, cfg.bn, cfg.bk);
}
}
}
}
/* wait for eltwise to finish */
libxsmm_barrier_wait(cfg.barrier, ltid);
}
if ( (pass & MY_PASS_BWD_D) == MY_PASS_BWD_D ) {
const libxsmm_blasint use_2d_blocking = cfg.bwd_2d_blocking;
/* number of tasks that could be run in parallel */
const libxsmm_blasint work = nBlocksIFm * nBlocksMB;
/* compute chunk size */
const libxsmm_blasint chunksize = (work % cfg.threads == 0) ? (work / cfg.threads) : ((work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (ltid * chunksize < work) ? (ltid * chunksize) : work;
const libxsmm_blasint thr_end = ((ltid + 1) * chunksize < work) ? ((ltid + 1) * chunksize) : work;
/* loop variables */
libxsmm_blasint ifm1 = 0, ifm2 = 0, mb1ifm1 = 0;
libxsmm_blasint N_tasks_per_thread = 0, M_tasks_per_thread = 0, my_M_start = 0, my_M_end = 0, my_N_start = 0, my_N_end = 0, my_col_id = 0, my_row_id = 0, col_teams = 0, row_teams = 0;
LIBXSMM_VLA_DECL(4, float, dinput, din_act_ptr, nBlocksIFm, bn, bc);
LIBXSMM_VLA_DECL(4, float, filter_tr, numa_thr_cfg->bwd_d_scratch, nBlocksOFm, bk, bc);
unsigned long long blocks = nBlocksOFm;
libxsmm_blasint KB_BLOCKS = nBlocksOFm, BF = 1;
BF = cfg.bwd_bf;
KB_BLOCKS = nBlocksOFm/BF;
blocks = KB_BLOCKS;
if (use_2d_blocking == 1) {
col_teams = cfg.bwd_col_teams;
row_teams = cfg.bwd_row_teams;
my_row_id = ltid % row_teams;
my_col_id = ltid / row_teams;
N_tasks_per_thread = LIBXSMM_UPDIV(nBlocksMB, col_teams);
M_tasks_per_thread = LIBXSMM_UPDIV(nBlocksIFm, row_teams);
my_N_start = LIBXSMM_MIN(my_col_id * N_tasks_per_thread, nBlocksMB);
my_N_end = LIBXSMM_MIN((my_col_id+1) * N_tasks_per_thread, nBlocksMB);
my_M_start = LIBXSMM_MIN(my_row_id * M_tasks_per_thread, nBlocksIFm);
my_M_end = LIBXSMM_MIN((my_row_id+1) * M_tasks_per_thread, nBlocksIFm);
}
if (use_2d_blocking == 1) {
if (BF > 1) {
for ( ofm1 = 0; ofm1 < BF; ++ofm1 ) {
for (ifm1 = my_M_start; ifm1 < my_M_end; ++ifm1) {
for (mb1 = my_N_start; mb1 < my_N_end; ++mb1) {
/* Initialize intermediate f32 tensor */
if ( ofm1 == 0 ) {
for ( mb2 = 0; mb2 < bn; ++mb2 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, mb2, ifm2, nBlocksIFm, bn, bc) = (float)0;
}
}
}
cfg.gemm_bwd( &LIBXSMM_VLA_ACCESS(4, filter_tr, ifm1, ofm1*KB_BLOCKS, 0, 0, nBlocksOFm, bk, bc ),
&LIBXSMM_VLA_ACCESS(4, doutput, mb1, ofm1*KB_BLOCKS, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, 0, 0, nBlocksIFm, bn, bc), &blocks);
}
}
}
} else {
for (ifm1 = my_M_start; ifm1 < my_M_end; ++ifm1) {
for (mb1 = my_N_start; mb1 < my_N_end; ++mb1) {
cfg.gemm_bwd2( &LIBXSMM_VLA_ACCESS(4, filter_tr, ifm1, 0, 0, 0, nBlocksOFm, bk, bc),
&LIBXSMM_VLA_ACCESS(4, doutput, mb1, 0, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, 0, 0, nBlocksIFm, bn, bc), &blocks);
}
}
}
} else {
if (BF > 1) {
for ( ofm1 = 0; ofm1 < BF; ++ofm1 ) {
for ( mb1ifm1 = thr_begin; mb1ifm1 < thr_end; ++mb1ifm1 ) {
mb1 = mb1ifm1%nBlocksMB;
ifm1 = mb1ifm1/nBlocksMB;
/* Initialize intermediate f32 tensor */
if ( ofm1 == 0 ) {
for ( mb2 = 0; mb2 < bn; ++mb2 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, mb2, ifm2, nBlocksIFm, bn, bc) = (float)0;
}
}
}
cfg.gemm_bwd( &LIBXSMM_VLA_ACCESS(4, filter_tr, ifm1, ofm1*KB_BLOCKS, 0, 0, nBlocksOFm, bk, bc ),
&LIBXSMM_VLA_ACCESS(4, doutput, mb1, ofm1*KB_BLOCKS, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, 0, 0, nBlocksIFm, bn, bc), &blocks);
}
}
} else {
for ( mb1ifm1 = thr_begin; mb1ifm1 < thr_end; ++mb1ifm1 ) {
mb1 = mb1ifm1%nBlocksMB;
ifm1 = mb1ifm1/nBlocksMB;
cfg.gemm_bwd2( &LIBXSMM_VLA_ACCESS(4, filter_tr, ifm1 - ifm_start, 0, 0, 0, nBlocksOFm, bk, bc ),
&LIBXSMM_VLA_ACCESS(4, doutput, mb1, 0, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, dinput, mb1, ifm1, 0, 0, nBlocksIFm, bn, bc), &blocks);
}
}
}
libxsmm_barrier_wait(cfg.barrier, ltid);
}
if ( (pass & MY_PASS_BWD_W) == MY_PASS_BWD_W ) {
/* number of tasks that could be run in parallel */
const libxsmm_blasint ofm_subtasks = (cfg.upd_2d_blocking == 1) ? 1 : cfg.ofm_subtasks;
const libxsmm_blasint ifm_subtasks = (cfg.upd_2d_blocking == 1) ? 1 : cfg.ifm_subtasks;
const libxsmm_blasint bbk = (cfg.upd_2d_blocking == 1) ? bk : bk/ofm_subtasks;
const libxsmm_blasint bbc = (cfg.upd_2d_blocking == 1) ? bc : bc/ifm_subtasks;
const libxsmm_blasint work = nBlocksIFm * ifm_subtasks * nBlocksOFm * ofm_subtasks;
const libxsmm_blasint Cck_work = nBlocksIFm * ifm_subtasks * ofm_subtasks;
const libxsmm_blasint Cc_work = nBlocksIFm * ifm_subtasks;
/* 2D blocking parameters */
libxsmm_blasint use_2d_blocking = cfg.upd_2d_blocking;
libxsmm_blasint N_tasks_per_thread = 0, M_tasks_per_thread = 0, my_M_start = 0, my_M_end = 0, my_N_start = 0, my_N_end = 0, my_col_id = 0, my_row_id = 0, col_teams = 0, row_teams = 0;
/* compute chunk size */
const libxsmm_blasint chunksize = (work % cfg.threads == 0) ? (work / cfg.threads) : ((work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (ltid * chunksize < work) ? (ltid * chunksize) : work;
const libxsmm_blasint thr_end = ((ltid + 1) * chunksize < work) ? ((ltid + 1) * chunksize) : work;
libxsmm_blasint BF = cfg.upd_bf;
/* loop variables */
libxsmm_blasint ifm1ofm1 = 0, ifm1 = 0, ifm2 = 0, bfn = 0, ii = 0, jj = 0;
/* Batch reduce related variables */
unsigned long long blocks = nBlocksMB/BF;
LIBXSMM_VLA_DECL(4, const float, input, in_act_ptr, nBlocksIFm, bn, bc);
LIBXSMM_VLA_DECL(4, float, dfilter, dwt_ptr, nBlocksIFm, bc, bk);
if (use_2d_blocking == 1) {
col_teams = cfg.upd_col_teams;
row_teams = cfg.upd_row_teams;
my_row_id = ltid % row_teams;
my_col_id = ltid / row_teams;
N_tasks_per_thread = LIBXSMM_UPDIV(nBlocksIFm, col_teams);
M_tasks_per_thread = LIBXSMM_UPDIV(nBlocksOFm, row_teams);
my_N_start = LIBXSMM_MIN(my_col_id * N_tasks_per_thread, nBlocksIFm);
my_N_end = LIBXSMM_MIN((my_col_id+1) * N_tasks_per_thread, nBlocksIFm);
my_M_start = LIBXSMM_MIN(my_row_id * M_tasks_per_thread, nBlocksOFm);
my_M_end = LIBXSMM_MIN((my_row_id+1) * M_tasks_per_thread, nBlocksOFm);
}
if (use_2d_blocking == 1) {
if (BF == 1) {
for (ofm1 = my_M_start; ofm1 < my_M_end; ++ofm1) {
for (ifm1 = my_N_start; ifm1 < my_N_end; ++ifm1) {
cfg.gemm_upd2(&LIBXSMM_VLA_ACCESS(4, doutput, 0, ofm1, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, input, 0, ifm1, 0, 0, nBlocksIFm, bn, bc),
&LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, 0, 0, nBlocksIFm, bc, bk), &blocks);
}
}
} else {
for (bfn = 0; bfn < BF; bfn++) {
for (ofm1 = my_M_start; ofm1 < my_M_end; ++ofm1) {
for (ifm1 = my_N_start; ifm1 < my_N_end; ++ifm1) {
/* initialize current work task to zero */
if (bfn == 0) {
for (ii = 0; ii<bc; ii++) {
for (jj = 0; jj<bk; jj++) {
LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, ii, jj, nBlocksIFm, bc, bk) = (float)0;
}
}
}
cfg.gemm_upd( &LIBXSMM_VLA_ACCESS(4, doutput, bfn*blocks, ofm1, 0, 0, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, input, bfn*blocks, ifm1, 0, 0, nBlocksIFm, bn, bc),
&LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, 0, 0, nBlocksIFm, bc, bk), &blocks);
}
}
}
}
} else {
if (BF == 1) {
for ( ifm1ofm1 = thr_begin; ifm1ofm1 < thr_end; ++ifm1ofm1 ) {
ofm1 = ifm1ofm1 / Cck_work;
ofm2 = (ifm1ofm1 % Cck_work) / Cc_work;
ifm1 = ((ifm1ofm1 % Cck_work) % Cc_work) / ifm_subtasks;
ifm2 = ((ifm1ofm1 % Cck_work) % Cc_work) % ifm_subtasks;
cfg.gemm_upd2( &LIBXSMM_VLA_ACCESS(4, doutput, 0, ofm1, 0, ofm2*bbk, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, input, 0, ifm1, 0, ifm2*bbc, nBlocksIFm, bn, bc),
&LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, ifm2*bbc, ofm2*bbk, nBlocksIFm, bc, bk), &blocks);
}
} else {
for (bfn = 0; bfn < BF; bfn++) {
for ( ifm1ofm1 = thr_begin; ifm1ofm1 < thr_end; ++ifm1ofm1 ) {
ofm1 = ifm1ofm1 / Cck_work;
ofm2 = (ifm1ofm1 % Cck_work) / Cc_work;
ifm1 = ((ifm1ofm1 % Cck_work) % Cc_work) / ifm_subtasks;
ifm2 = ((ifm1ofm1 % Cck_work) % Cc_work) % ifm_subtasks;
/* initialize current work task to zero */
if (bfn == 0) {
for (ii = 0; ii<bbc; ii++) {
for (jj = 0; jj<bbk; jj++) {
LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, ifm2*bbc+ii, ofm2*bbk+jj, nBlocksIFm, bc, bk) = (float)0;
}
}
}
cfg.gemm_upd( &LIBXSMM_VLA_ACCESS(4, doutput, bfn*blocks, ofm1, 0, ofm2*bbk, nBlocksOFm, bn, bk),
&LIBXSMM_VLA_ACCESS(4, input, bfn*blocks, ifm1, 0, ifm2*bbc, nBlocksIFm, bn, bc),
&LIBXSMM_VLA_ACCESS(4, dfilter, ofm1, ifm1, ifm2*bbc, ofm2*bbk, nBlocksIFm, bc, bk), &blocks);
}
}
}
}
libxsmm_barrier_wait(cfg.barrier, ltid);
}
}
void my_opt_exec( my_opt_config cfg, const float* delwt_ptr, int start_tid, int my_tid,
my_numa_thr_cfg *numa_thr_cfg, int l, my_fc_fwd_config my_fc_fwd) {
const libxsmm_blasint ltid = my_tid - numa_thr_cfg->thr_s;
const libxsmm_blasint nBlocksIFm = my_fc_fwd.C / my_fc_fwd.bc;
const libxsmm_blasint IFM_shift = my_fc_fwd.bc * my_fc_fwd.bk;
const libxsmm_blasint OFM_shift = nBlocksIFm * my_fc_fwd.bc * my_fc_fwd.bk;
const libxsmm_blasint work = ((numa_thr_cfg->blocksOFm_e[l] - numa_thr_cfg->blocksOFm_s[l]) + 1) * nBlocksIFm;
/* compute chunk size */
int thr = numa_thr_cfg->thr_e - numa_thr_cfg->thr_s;
const libxsmm_blasint chunksize = (work % thr == 0) ? (work / thr) : ((work / thr) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (ltid * chunksize < work) ? (ltid * chunksize) : work;
const libxsmm_blasint thr_end = ((ltid + 1) * chunksize < work) ? ((ltid + 1) * chunksize) : work;
libxsmm_barrier_init( cfg.barrier, my_tid );
__m512 vlr = _mm512_set1_ps( cfg.lr );
float *dw_prt = (float*)delwt_ptr + numa_thr_cfg->blocksOFm_s[l] * OFM_shift;
int j = 0, i = 0;
for (j = thr_begin; j < thr_end; j++) {
int ofm = j / nBlocksIFm;
int ifm = j % nBlocksIFm;
float *out = numa_thr_cfg->scratch[l] + ofm * OFM_shift + ifm * IFM_shift;
float *inp = dw_prt + ofm * OFM_shift + ifm * IFM_shift;
for (i = 0; i < IFM_shift; i += 16)
_mm512_storeu_ps( out+i, _mm512_sub_ps( _mm512_loadu_ps( out+i ), _mm512_mul_ps( vlr, _mm512_loadu_ps( inp + i ) ) ) ) ;
}
libxsmm_barrier_wait( cfg.barrier, my_tid );
}
void my_smax_fwd_exec( my_smax_fwd_config cfg, const float* in_act_ptr, float* out_act_ptr, const int* label_ptr, float* loss, int start_tid, int my_tid, void* scratch ) {
libxsmm_blasint bn = cfg.bn;
libxsmm_blasint Bn = cfg.N/cfg.bn;
libxsmm_blasint bc = cfg.bc;
libxsmm_blasint Bc = cfg.C/cfg.bc;
/* loop counters */
libxsmm_blasint i = 0;
libxsmm_blasint img1, img2, ifm1, ifm2;
/* computing first logical thread */
const libxsmm_blasint ltid = my_tid - start_tid;
/* number of tasks that could run in parallel for the batch */
const libxsmm_blasint n_work = Bn * bn;
/* compute chunk size */
const libxsmm_blasint n_chunksize = (n_work % cfg.threads == 0) ? (n_work / cfg.threads) : ((n_work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint n_thr_begin = (ltid * n_chunksize < n_work) ? (ltid * n_chunksize) : n_work;
const libxsmm_blasint n_thr_end = ((ltid + 1) * n_chunksize < n_work) ? ((ltid + 1) * n_chunksize) : n_work;
LIBXSMM_VLA_DECL(4, float, output, out_act_ptr, Bc, bn, bc);
LIBXSMM_VLA_DECL(4, const float, input, in_act_ptr, Bc, bn, bc);
LIBXSMM_VLA_DECL(2, const int, label, label_ptr, bn);
/* lazy barrier init */
libxsmm_barrier_init( cfg.barrier, ltid );
for ( i = n_thr_begin; i < n_thr_end; ++i ) {
float max = FLT_MIN;
float sum_of_exp = 0.0f;
img1 = i/bn;
img2 = i%bn;
/* set output to input and set compute max per image */
for ( ifm1 = 0; ifm1 < Bc; ++ifm1 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) = LIBXSMM_VLA_ACCESS( 4, input, img1, ifm1, img2, ifm2, Bc, bn, bc );
if ( LIBXSMM_VLA_ACCESS( 4, input, img1, ifm1, img2, ifm2, Bc, bn, bc ) > max ) {
max = LIBXSMM_VLA_ACCESS( 4, input, img1, ifm1, img2, ifm2, Bc, bn, bc );
}
}
}
/* sum exp over outputs */
for ( ifm1 = 0; ifm1 < Bc; ++ifm1 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) = (float)exp( (double)(LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) - max) );
sum_of_exp += LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc );
}
}
/* scale output */
sum_of_exp = 1.0f/sum_of_exp;
for ( ifm1 = 0; ifm1 < Bc; ++ifm1 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) = LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) * sum_of_exp;
}
}
}
libxsmm_barrier_wait( cfg.barrier, ltid );
/* calculate loss single threaded */
if ( ltid == 0 ) {
(*loss) = 0.0f;
for ( img1 = 0; img1 < Bn; ++img1 ) {
for ( img2 = 0; img2 <bn; ++img2 ) {
libxsmm_blasint ifm = (libxsmm_blasint)LIBXSMM_VLA_ACCESS( 2, label, img1, img2, bn );
libxsmm_blasint ifm1b = ifm/bc;
libxsmm_blasint ifm2b = ifm%bc;
float val = ( LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1b, img2, ifm2b, Bc, bn, bc ) > FLT_MIN ) ? LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1b, img2, ifm2b, Bc, bn, bc ) : FLT_MIN;
*loss = LIBXSMM_LOGF( val );
}
}
*loss = ((-1.0f)*(*loss))/cfg.N;
}
libxsmm_barrier_wait( cfg.barrier, ltid );
}
void my_smax_bwd_exec( my_smax_bwd_config cfg, float* delin_act_ptr, const float* out_act_ptr, const int* label_ptr, int start_tid, int my_tid, void* scratch ) {
libxsmm_blasint bn = cfg.bn;
libxsmm_blasint Bn = cfg.N/cfg.bn;
libxsmm_blasint bc = cfg.bc;
libxsmm_blasint Bc = cfg.C/cfg.bc;
/* loop counters */
libxsmm_blasint i = 0;
libxsmm_blasint img1, img2, ifm1, ifm2;
float rcp_N = 1.0f/cfg.N;
/* computing first logical thread */
const libxsmm_blasint ltid = my_tid - start_tid;
/* number of tasks that could run in parallel for the batch */
const libxsmm_blasint n_work = Bn * bn;
/* compute chunk size */
const libxsmm_blasint n_chunksize = (n_work % cfg.threads == 0) ? (n_work / cfg.threads) : ((n_work / cfg.threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint n_thr_begin = (ltid * n_chunksize < n_work) ? (ltid * n_chunksize) : n_work;
const libxsmm_blasint n_thr_end = ((ltid + 1) * n_chunksize < n_work) ? ((ltid + 1) * n_chunksize) : n_work;
LIBXSMM_VLA_DECL(4, const float, output, out_act_ptr, Bc, bn, bc);
LIBXSMM_VLA_DECL(4, float, dinput, delin_act_ptr, Bc, bn, bc);
LIBXSMM_VLA_DECL(2, const int, label, label_ptr, bn);
/* lazy barrier init */
libxsmm_barrier_init( cfg.barrier, ltid );
for ( i = n_thr_begin; i < n_thr_end; ++i ) {
img1 = i/bn;
img2 = i%bn;
/* set output to input and set compute max per image */
for ( ifm1 = 0; ifm1 < Bc; ++ifm1 ) {
for ( ifm2 = 0; ifm2 < bc; ++ifm2 ) {
if ( (ifm1*Bc)+ifm2 == (libxsmm_blasint)LIBXSMM_VLA_ACCESS( 2, label, img1, img2, bn ) ) {
LIBXSMM_VLA_ACCESS( 4, dinput, img1, ifm1, img2, ifm2, Bc, bn, bc ) =
( LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) - 1.0f ) * rcp_N * cfg.loss_weight;
} else {
LIBXSMM_VLA_ACCESS( 4, dinput, img1, ifm1, img2, ifm2, Bc, bn, bc ) =
LIBXSMM_VLA_ACCESS( 4, output, img1, ifm1, img2, ifm2, Bc, bn, bc ) * rcp_N * cfg.loss_weight;
}
}
}
}
libxsmm_barrier_wait( cfg.barrier, ltid );
}
void *numa_alloc_onnode_aligned(size_t size, int numa_node, int alignment_) {
#if 0
int alignment = alignment_ - 1;
size_t adj_size = sizeof(size_t) + alignment;
void *r_ptr = NULL;
void *t_ptr = numa_alloc_onnode(size + adj_size, numa_node);
if (t_ptr == NULL) return NULL;
r_ptr = (void *)(((size_t)t_ptr + adj_size) & ~alignment);
*((size_t*)r_ptr - 1) = (size_t)r_ptr - (size_t)t_ptr;
return r_ptr;
#else
return numa_alloc_onnode(size, numa_node);
#endif
}
void numa_free_aligned(void *ptr, size_t size) {
#if 0
if (ptr == NULL) return;
void *t_ptr = (void*)((size_t*)ptr - *((size_t*)ptr - 1));
numa_free(t_ptr, size);
#else
numa_free(ptr, size);
#endif
}
int setup_my_numa(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, int n_threads) {
int max_nodes = numa_max_node() + 1;
int max_cfg_nodes = numa_num_configured_nodes();
int max_cfg_cpus = numa_num_configured_cpus();
int max_task_cpus = numa_num_task_cpus();
my_numa_thr_cfg *numa_thr_cfg = (my_numa_thr_cfg *) malloc(sizeof(my_numa_thr_cfg) * max_cfg_nodes);
printf("NUMA configuration:\n");
printf("There are %d numa nodes on the system\n", max_nodes);
printf("There are %d configured numa nodes on the system\n", max_cfg_nodes);
printf("There are %d configured CPUs on the system\n", max_cfg_cpus);
printf("There are %d CPUs asigned for the current task\n", max_task_cpus);
struct bitmask* bmask = numa_bitmask_alloc(max_cfg_cpus);
int thr_count = 0, i = 0;
for (i = 0; i < max_cfg_nodes; i++) {
numa_node_to_cpus(i, bmask);
numa_thr_cfg[i].scratch = (float**) malloc(sizeof(float*) * num_layers);
numa_thr_cfg[i].layer_size = (size_t*)malloc(sizeof(size_t)*num_layers);
numa_thr_cfg[i].blocksOFm_s = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksOFm_e = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksIFm_s = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksIFm_e = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksOFm_tr_s = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksOFm_tr_e = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksIFm_tr_s = (int*)malloc(sizeof(int)*num_layers);
numa_thr_cfg[i].blocksIFm_tr_e = (int*)malloc(sizeof(int)*num_layers);
/*
printf("@@@@@ node %d size %zd cpus ", i, bmask->size);
size_t j = 0;
for(j = 0; j < bmask->size; j++)
printf("%d", numa_bitmask_isbitset(bmask, j));
printf("\n");
*/
int num_threads_in_mask = 0;
int t = 0;
for (t = 0; t < bmask->size; t++)
if (numa_bitmask_isbitset(bmask, t)) num_threads_in_mask++;
int node_threads = 0;
while(thr_count < n_threads && node_threads < num_threads_in_mask) {
if (numa_bitmask_isbitset(bmask, thr_count)) {
numa_thr_cfg[i].thr_s = thr_count;
break;
}
thr_count++; node_threads++;
}
while(thr_count < n_threads && node_threads < num_threads_in_mask) {
if (numa_bitmask_isbitset(bmask, thr_count))
numa_thr_cfg[i].thr_e = thr_count;
thr_count++; node_threads++;
}
}
*numa_thr_cfg_ = numa_thr_cfg;
return 1;
}
int setup_my_numa_fwd(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_fwd_config* my_fc_fwd) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int max_cfg_nodes = numa_num_configured_nodes();
int i = 0;
for (i = 0; i < max_cfg_nodes; i++) {
int l = 0;
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksOFm = my_fc_fwd[l].K / my_fc_fwd[l].bk;
const libxsmm_blasint nBlocksMB = my_fc_fwd[l].N / my_fc_fwd[l].bn;
if (my_fc_fwd[l].fwd_bf > 1) {
printf("@@@ NUMA ERROR: doesn't support this configuration\n");
return -1;
}
int thr = 0;
if (my_fc_fwd[l].fwd_2d_blocking == 1) {
libxsmm_blasint row_teams = my_fc_fwd[l].fwd_row_teams;
libxsmm_blasint M_tasks_per_thread = LIBXSMM_UPDIV(nBlocksOFm, row_teams);
numa_thr_cfg[i].blocksOFm_s[l] = nBlocksOFm;
numa_thr_cfg[i].blocksOFm_e[l] = 0;
for (thr = numa_thr_cfg[i].thr_s; thr <= numa_thr_cfg[i].thr_e; thr++) {
libxsmm_blasint my_row_id = thr % row_teams; /* ltid */
libxsmm_blasint my_M_start = LIBXSMM_MIN(my_row_id * M_tasks_per_thread, nBlocksOFm);
libxsmm_blasint my_M_end = LIBXSMM_MIN((my_row_id+1) * M_tasks_per_thread, nBlocksOFm);
numa_thr_cfg[i].blocksOFm_s[l] = (my_M_start < numa_thr_cfg[i].blocksOFm_s[l])
? my_M_start
: numa_thr_cfg[i].blocksOFm_s[l];
numa_thr_cfg[i].blocksOFm_e[l] = (my_M_end > numa_thr_cfg[i].blocksOFm_e[l])
? my_M_end
: numa_thr_cfg[i].blocksOFm_e[l];
}
} else {
numa_thr_cfg[i].blocksOFm_s[l] = nBlocksOFm;
numa_thr_cfg[i].blocksOFm_e[l] = 0;
for (thr = numa_thr_cfg[i].thr_s; thr <= numa_thr_cfg[i].thr_e; thr++) {
const libxsmm_blasint work = nBlocksOFm * nBlocksMB;
const libxsmm_blasint chunksize = (work % my_fc_fwd[l].threads == 0) ?
(work / my_fc_fwd[l].threads) : ((work / my_fc_fwd[l].threads) + 1);
const libxsmm_blasint thr_begin = (thr * chunksize < work) ? (thr * chunksize) : work;
const libxsmm_blasint thr_end = ((thr + 1) * chunksize < work) ? ((thr + 1) * chunksize) : work;
int ofm_s = thr_begin / nBlocksMB;
int ofm_e = (thr_end-1) / nBlocksMB;
numa_thr_cfg[i].blocksOFm_s[l] = (ofm_s < numa_thr_cfg[i].blocksOFm_s[l])
? ofm_s
: numa_thr_cfg[i].blocksOFm_s[l];
numa_thr_cfg[i].blocksOFm_e[l] = (ofm_e > numa_thr_cfg[i].blocksOFm_e[l])
? ofm_e
: numa_thr_cfg[i].blocksOFm_e[l];
}
#if 0
printf("numa_thr_cfg[%d].blocksOFm_s[%d] %d numa_thr_cfg[%d].blocksOFm_e[%d] %d\n",
i, l, numa_thr_cfg[i].blocksOFm_s[l], i, l, numa_thr_cfg[i].blocksOFm_e[l]);
#endif
}
}
}
return 1;
}
void set_fwd_ofm_to_node(int **fwd_ofm_to_node, my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_fwd_config* my_fc_fwd) {
int max_cfg_nodes = numa_num_configured_nodes();
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int l, ofm, i;
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksOFm = my_fc_fwd[l].K / my_fc_fwd[l].bk;
fwd_ofm_to_node[l] = (int*) malloc(sizeof(int) * nBlocksOFm);
int *l_fwd_ofm_to_node = fwd_ofm_to_node[l];
for (i = 0; i < max_cfg_nodes; i++) {
for (ofm = 0; ofm < nBlocksOFm; ofm++) {
if (ofm >= numa_thr_cfg[i].blocksOFm_s[l] && ofm <= numa_thr_cfg[i].blocksOFm_e[l])
l_fwd_ofm_to_node[ofm] = i;
}
}
}
#if 0
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksOFm = my_fc_fwd[l].K / my_fc_fwd[l].bk;
int *l_fwd_ofm_to_node = fwd_ofm_to_node[l];
for (ofm = 0; ofm < nBlocksOFm; ofm++)
printf("%d l_fwd_ofm_to_node[%d] %d | %d\n", l, ofm, l_fwd_ofm_to_node[ofm], nBlocksOFm);
}
#endif
}
void free_fwd_ofm_to_node(int **fwd_ofm_to_node, int num_layers) {
int l;
for (l = 0; l < num_layers; l++) {
free(fwd_ofm_to_node[l]);
}
}
int setup_my_numa_bwd_d(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_bwd_config* my_fc_bwd) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int max_cfg_nodes = numa_num_configured_nodes();
int i = 0;
for (i = 0; i < max_cfg_nodes; i++) {
int l = 0;
for (l = 0; l < num_layers; l++) {
if (my_fc_bwd[l].bwd_bf > 1) {
printf("@@@ NUMA ERROR: doesn't support this configuration\n");
return -1;
}
int thr = 0;
const libxsmm_blasint nBlocksIFm = my_fc_bwd[l].C / my_fc_bwd[l].bc;
const libxsmm_blasint nBlocksMB = my_fc_bwd[l].N / my_fc_bwd[l].bn;
if (my_fc_bwd[l].bwd_2d_blocking == 1) {
printf("@@@ NUMA ERROR: doesn't support this configuration\n");
return -1;
} else {
numa_thr_cfg[i].blocksIFm_tr_s[l] = nBlocksIFm;
numa_thr_cfg[i].blocksIFm_tr_e[l] = 0;
for (thr = numa_thr_cfg[i].thr_s; thr <= numa_thr_cfg[i].thr_e; thr++) {
/* number of tasks that could be run in parallel */
const libxsmm_blasint work = nBlocksIFm * nBlocksMB;
/* compute chunk size */
const libxsmm_blasint chunksize = (work % my_fc_bwd[l].threads == 0) ?
(work / my_fc_bwd[l].threads) : ((work / my_fc_bwd[l].threads) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (thr * chunksize < work) ? (thr * chunksize) : work;
const libxsmm_blasint thr_end = ((thr + 1) * chunksize < work) ? ((thr + 1) * chunksize) : work;
int ifm_s = thr_begin / nBlocksMB;
int ifm_e = (thr_end-1) / nBlocksMB;
numa_thr_cfg[i].blocksIFm_tr_s[l] = (ifm_s < numa_thr_cfg[i].blocksIFm_tr_s[l])
? ifm_s
: numa_thr_cfg[i].blocksIFm_tr_s[l];
numa_thr_cfg[i].blocksIFm_tr_e[l] = (ifm_e > numa_thr_cfg[i].blocksIFm_tr_e[l])
? ifm_e
: numa_thr_cfg[i].blocksIFm_tr_e[l];
}
#if 0
printf("numa_thr_cfg[%d].blocksIFm_tr_s[%d] %d numa_thr_cfg[%d].blocksIFm_tr_e[%d] %d\n",
i, l, numa_thr_cfg[i].blocksIFm_tr_s[l], i, l, numa_thr_cfg[i].blocksIFm_tr_e[l]);
#endif
}
}
}
return 1;
}
int allocate_numa_buffers_fwd(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_fwd_config* my_fc_fwd) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int max_cfg_nodes = numa_num_configured_nodes();
int i = 0, l = 0;
for (i = 0; i < max_cfg_nodes; i++) {
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksIFm = my_fc_fwd[l].C / my_fc_fwd[l].bc;
const libxsmm_blasint OFM_shift = nBlocksIFm * my_fc_fwd[l].bc * my_fc_fwd[l].bk;
int l_nBlocksOFm = (numa_thr_cfg[i].blocksOFm_e[l] - numa_thr_cfg[i].blocksOFm_s[l]) + 1;
if (l_nBlocksOFm <= 0)
continue;
numa_thr_cfg[i].layer_size[l] = sizeof(float) * ((l_nBlocksOFm) * OFM_shift);
numa_thr_cfg[i].scratch[l] = (float*)numa_alloc_onnode_aligned(numa_thr_cfg[i].layer_size[l], i, 2097152);
if (numa_thr_cfg[i].scratch[l] == NULL) {
printf("@@@ NUMA ERROR: cannot allocate on node #%d\n", i);
return -1;
}
}
}
return 1;
}
int allocate_numa_buffers_bwd_d(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_bwd_config* my_fc_bwd) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int max_cfg_nodes = numa_num_configured_nodes();
int i = 0, l = 0;
for (i = 0; i < max_cfg_nodes; i++) {
int l_nBlocksIFm = 0;
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksOFm = my_fc_bwd[l].K / my_fc_bwd[l].bk;
const libxsmm_blasint IFM_shift = nBlocksOFm * my_fc_bwd[l].bc * my_fc_bwd[l].bk;
if (l_nBlocksIFm <= ((numa_thr_cfg[i].blocksIFm_tr_e[l] - numa_thr_cfg[i].blocksIFm_tr_s[l]) + 1) * IFM_shift)
l_nBlocksIFm = ((numa_thr_cfg[i].blocksIFm_tr_e[l] - numa_thr_cfg[i].blocksIFm_tr_s[l]) + 1) * IFM_shift;
}
numa_thr_cfg[i].bwd_d_scratch_size = sizeof(float) * (l_nBlocksIFm);
numa_thr_cfg[i].bwd_d_scratch = (float*)numa_alloc_onnode_aligned(numa_thr_cfg[i].bwd_d_scratch_size, i, 2097152);
if (numa_thr_cfg[i].bwd_d_scratch == NULL) {
printf("@@@ NUMA ERROR: cannot allocate on node #%d\n", i);
return -1;
}
}
return 1;
}
int copy_to_numa_buffers_fwd_inf(my_numa_thr_cfg **numa_thr_cfg_, int num_layers, my_fc_fwd_config* my_fc_fwd, float **fil_libxsmm) {
my_numa_thr_cfg *numa_thr_cfg = *numa_thr_cfg_;
int max_cfg_nodes = numa_num_configured_nodes();
int i,l;
#ifndef COPY_ON_LOCAL_NODES
#pragma omp parallel for collapse(2) private (i,l)
#else
#pragma omp parallel private (i,l)
{
int tid = omp_get_thread_num();
#endif
for (i = 0; i < max_cfg_nodes; i++) {
#ifdef COPY_ON_LOCAL_NODES
if (tid >= numa_thr_cfg[i].thr_s && tid <= numa_thr_cfg[i].thr_e) {
numa_run_on_node(i);
}
if (tid == numa_thr_cfg[i].thr_s) {
#endif
for (l = 0; l < num_layers; l++) {
const libxsmm_blasint nBlocksIFm = my_fc_fwd[l].C / my_fc_fwd[l].bc;
const libxsmm_blasint BOFM_shift = nBlocksIFm * my_fc_fwd[l].bc * my_fc_fwd[l].bk;
int l_nBlocksOFm = (numa_thr_cfg[i].blocksOFm_e[l] - numa_thr_cfg[i].blocksOFm_s[l]) + 1;
int j = 0;
for (j = 0; j < l_nBlocksOFm ; j++) {
size_t l_BOFM_shift = j * BOFM_shift;
float *out = numa_thr_cfg[i].scratch[l] + l_BOFM_shift;
float *inp = fil_libxsmm[l] + numa_thr_cfg[i].blocksOFm_s[l] * BOFM_shift + l_BOFM_shift;
memcpy(out, inp, sizeof(float) * nBlocksIFm * my_fc_fwd[l].bc * my_fc_fwd[l].bk);
}
}
#ifdef COPY_ON_LOCAL_NODES
}
#endif
}
#ifdef COPY_ON_LOCAL_NODES
}
#endif
return 1;
}
int copy_to_numa_buffers_fwd(my_numa_thr_cfg *numa_thr_cfg, my_fc_fwd_config my_fc_fwd, float *fil_libxsmm, int numa_node, int l, int my_tid, int dir) {
const libxsmm_blasint ltid = my_tid - numa_thr_cfg->thr_s;
const libxsmm_blasint nBlocksIFm = my_fc_fwd.C / my_fc_fwd.bc;
const libxsmm_blasint IFM_shift = my_fc_fwd.bc * my_fc_fwd.bk;
const libxsmm_blasint OFM_shift = nBlocksIFm * my_fc_fwd.bc * my_fc_fwd.bk;
const libxsmm_blasint work = ((numa_thr_cfg->blocksOFm_e[l] - numa_thr_cfg->blocksOFm_s[l]) + 1) * nBlocksIFm;
/* compute chunk size */
int thr = numa_thr_cfg->thr_e - numa_thr_cfg->thr_s;
const libxsmm_blasint chunksize = (work % thr == 0) ? (work / thr) : ((work / thr) + 1);
/* compute thr_begin and thr_end */
const libxsmm_blasint thr_begin = (ltid * chunksize < work) ? (ltid * chunksize) : work;
const libxsmm_blasint thr_end = ((ltid + 1) * chunksize < work) ? ((ltid + 1) * chunksize) : work;
/*libxsmm_barrier_init( my_fc_fwd.barrier, my_tid );*/
float *inp, *out;
if (dir) {
inp = numa_thr_cfg->scratch[l];
out = fil_libxsmm + numa_thr_cfg->blocksOFm_s[l] * OFM_shift;
} else {
out = numa_thr_cfg->scratch[l];
inp = fil_libxsmm + numa_thr_cfg->blocksOFm_s[l] * OFM_shift;
}
int j = 0;
for (j = thr_begin; j < thr_end; j++) {
int ofm = j / nBlocksIFm;
int ifm = j % nBlocksIFm;
float *l_out = out + ofm * OFM_shift + ifm * IFM_shift;
float *l_inp = inp + ofm * OFM_shift + ifm * IFM_shift;
memcpy(l_out, l_inp, sizeof(float) * IFM_shift);
}
/*libxsmm_barrier_wait( my_fc_fwd.barrier, my_tid );*/
return 1;
}
int main(int argc, char* argv[])
{
float **act_libxsmm, **fil_libxsmm, **delact_libxsmm, **delfil_libxsmm;
float **bias_libxsmm, **delbias_libxsmm;
unsigned char **relumask_libxsmm;
int *label_libxsmm;
my_eltwise_fuse my_fuse;
my_fc_fwd_config* my_fc_fwd;
my_fc_bwd_config* my_fc_bwd;
my_opt_config* my_opt;
my_smax_fwd_config my_smax_fwd;
my_smax_bwd_config my_smax_bwd;
void* scratch = NULL;
size_t scratch_size = 0;
/* some parameters we can overwrite via cli,
default is some inner layer of overfeat */
int iters = 10; /* repetitions of benchmark */
int MB = 256; /* mini-batch size, "N" */
int fuse_type = 0; /* 0: nothing fused, 1: relu fused, 2: elementwise fused, 3: relu and elementwise fused */
char type = 'A'; /* 'A': ALL, 'F': FP, 'B': BP, 'U', WU */
int bn = 32;
int bk = 32;
int bc = 32;
int *C; /* number of input feature maps, "C" */
int num_layers = 0;
#if defined(_OPENMP)
int nThreads = omp_get_max_threads(); /* number of threads */
#else
int nThreads = 1; /* number of threads */
#endif
unsigned long long l_start, l_end;
unsigned long long *fwd_time, *bwd_time, *solver_time;
double l_total = 0.0;
double gflop = 0.0;
int i, j;
double fil_size = 0.0;
double act_size = 0.0;
float lr = 0.2f;
float loss_weight = 0.1f;
libxsmm_matdiff_info norms_fwd, norms_bwd, norms_upd, diff;
libxsmm_matdiff_clear(&norms_fwd);
libxsmm_matdiff_clear(&norms_bwd);
libxsmm_matdiff_clear(&norms_upd);
libxsmm_matdiff_clear(&diff);
if (argc > 1 && !strncmp(argv[1], "-h", 3)) {
printf("Usage: %s iters MB fuse_type type bn bk bc C1 C2 ... CN\n", argv[0]);
return 0;
}
libxsmm_rng_set_seed(1);
/* reading new values from cli */
i = 1;
num_layers = argc - 9;
if (argc > i) iters = atoi(argv[i++]);
if (argc > i) MB = atoi(argv[i++]);
if (argc > i) fuse_type = atoi(argv[i++]);
if (argc > i) type = *(argv[i++]);
if (argc > i) bn = atoi(argv[i++]);
if (argc > i) bk = atoi(argv[i++]);
if (argc > i) bc = atoi(argv[i++]);
/* allocate the number of channles buffer */
if ( num_layers < 1 ) {
printf("Usage: %s iters MB fuse_type type bn bk bc C1 C2 ... CN\n", argv[0]);
return 0;
}
C = (int*)malloc((num_layers+2)*sizeof(int));
for (j = 0 ; i < argc; ++i, ++j ) {
C[j] = atoi(argv[i]);
}
/* handle softmax config */
C[num_layers+1] = C[num_layers];
if (type != 'A' && type != 'F' && type != 'B') {
printf("type needs to be 'A' (All), 'F' (FP only), 'B' (BP only)\n");
return -1;
}
if ( (fuse_type < 0) || (fuse_type > 5) ) {
printf("fuse type needs to be 0 (None), 1 (Bias), 2 (ReLU), 3 (Sigmoid), 4 (Bias+ReLU), 5 (Bias+Sigmoid)\n");
return -1;
}
#if defined(__SSE3__)
_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
_MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
#endif
/* print some summary */
printf("##########################################\n");
printf("# Setting Up (Common) #\n");
printf("##########################################\n");
printf("PARAMS: N:%d\n", MB);
printf("PARAMS: Layers: %d\n", num_layers);
printf("PARAMS: ITERS:%d", iters); printf(" Threads:%d\n", nThreads);
for (i = 0; i < num_layers; ++i ) {
if (i == 0) {
act_size += (double)(MB*C[i]*sizeof(float))/(1024.0*1024.0);
printf("SIZE Activations %i (%dx%d): %10.2f MiB\n", i, MB, C[i], (double)(MB*C[i]*sizeof(float))/(1024.0*1024.0) );
}
act_size += (double)(MB*C[i+1]*sizeof(float))/(1024.0*1024.0);
fil_size += (double)(C[i]*C[i+1]*sizeof(float))/(1024.0*1024.0);
printf("SIZE Filter %i (%dx%d): %10.2f MiB\n", i, C[i], C[i+1], (double)(C[i]*C[i+1]*sizeof(float))/(1024.0*1024.0) );
printf("SIZE Activations %i (%dx%d): %10.2f MiB\n", i+1, MB, C[i+1], (double)(MB*C[i+1]*sizeof(float))/(1024.0*1024.0) );
}
act_size += (double)(MB*C[num_layers+1]*sizeof(float))/(1024.0*1024.0);
printf("SIZE Activations softmax (%dx%d): %10.2f MiB\n", MB, C[num_layers+1], (double)(MB*C[num_layers+1]*sizeof(float))/(1024.0*1024.0) );
printf("\nTOTAL SIZE Activations: %10.2f MiB\n", act_size );
printf("TOTAL SIZE Filter: %10.2f MiB\n", fil_size );
printf("TOTAL SIZE delActivations: %10.2f MiB\n", act_size );
printf("TOTAL SIZE delFilter: %10.2f MiB\n", fil_size );
printf("TOTAL SIZE MLP: %10.2f MiB\n", (2.0*fil_size) + (2.0*act_size) );
/* allocate data */
/* +2 because of the softwax layer */
act_libxsmm = (float**)malloc( (num_layers+2)*sizeof(float*) );
delact_libxsmm = (float**)malloc( (num_layers+1)*sizeof(float*) );
for ( i = 0 ; i < num_layers+2; ++i ) {
#ifdef ACT_NUMA_INTERLEAVED
act_libxsmm[i] = (float*)numa_alloc_interleaved( MB*C[i]*sizeof(float));
#else
act_libxsmm[i] = (float*)libxsmm_aligned_malloc( MB*C[i]*sizeof(float), 2097152);
#endif
/* softmax has no incoming gradients */
if ( i < num_layers+1 ) {
delact_libxsmm[i] = (float*)libxsmm_aligned_malloc( MB*C[i]*sizeof(float), 2097152);
}
}
fil_libxsmm = (float**)malloc( num_layers*sizeof(float*) );
delfil_libxsmm = (float**)malloc( num_layers*sizeof(float*) );
for ( i = 0 ; i < num_layers; ++i ) {
fil_libxsmm[i] = (float*)libxsmm_aligned_malloc( C[i]*C[i+1]*sizeof(float), 2097152);
delfil_libxsmm[i] = (float*)libxsmm_aligned_malloc( C[i]*C[i+1]*sizeof(float), 2097152);
}
bias_libxsmm = (float**)malloc( num_layers*sizeof(float*) );
delbias_libxsmm = (float**)malloc( num_layers*sizeof(float*) );
for ( i = 0 ; i < num_layers; ++i ) {
bias_libxsmm[i] = (float*)libxsmm_aligned_malloc( C[i+1]*sizeof(float), 2097152);
delbias_libxsmm[i] = (float*)libxsmm_aligned_malloc( C[i+1]*sizeof(float), 2097152);
}
relumask_libxsmm = (unsigned char**)malloc( num_layers*sizeof(unsigned char*) );
for ( i = 0 ; i < num_layers; ++i ) {
relumask_libxsmm[i] = (unsigned char*)libxsmm_aligned_malloc( MB*C[i+1]*sizeof(unsigned char), 2097152);
}
label_libxsmm = (int*)libxsmm_aligned_malloc( MB*sizeof(int), 2097152);
/* init data */
for ( i = 0 ; i < num_layers+2; ++i ) {
my_init_buf( act_libxsmm[i], MB*C[i], 0, 0 );
}
for ( i = 0 ; i < num_layers+1; ++i ) {
my_init_buf( delact_libxsmm[i], MB*C[i], 0, 0 );
}
for ( i = 0 ; i < num_layers; ++i ) {
my_init_buf( fil_libxsmm[i], C[i]*C[i+1], 0, 0 );
}
for ( i = 0 ; i < num_layers; ++i ) {
my_init_buf( delfil_libxsmm[i], C[i]*C[i+1], 0, 0 );
}
for ( i = 0 ; i < num_layers; ++i ) {
my_init_buf( bias_libxsmm[i], C[i+1], 0, 0 );
}
for ( i = 0 ; i < num_layers; ++i ) {
my_init_buf( delbias_libxsmm[i], C[i+1], 0, 0 );
}
for ( i = 0 ; i < num_layers; ++i ) {
zero_buf_uint8( relumask_libxsmm[i], MB*C[i+1] );
}
zero_buf_int32( label_libxsmm, MB );
printf("\n");
printf("##########################################\n");
printf("# Setting Up (custom-Storage) #\n");
printf("##########################################\n");
if ( fuse_type == 0 ) {
my_fuse = MY_ELTWISE_FUSE_NONE;
} else if ( fuse_type == 1 ) {
my_fuse = MY_ELTWISE_FUSE_BIAS;
} else if ( fuse_type == 2 ) {
my_fuse = MY_ELTWISE_FUSE_RELU;
} else if ( fuse_type == 4 ) {
my_fuse = MY_ELTWISE_FUSE_BIAS_RELU;
} else {
my_fuse = MY_ELTWISE_FUSE_NONE;
}
/* allocating handles */
my_fc_fwd = (my_fc_fwd_config*) malloc( num_layers*sizeof(my_fc_fwd_config) );
my_fc_bwd = (my_fc_bwd_config*) malloc( num_layers*sizeof(my_fc_bwd_config) );
my_opt = (my_opt_config*) malloc( num_layers*sizeof(my_opt_config) );
/* setting up handles + scratch */
for ( i = 0; i < num_layers; ++i ) {
my_fc_fwd[i] = setup_my_fc_fwd(MB, C[i], C[i+1], (MB % bn == 0) ? bn : MB,
(C[i ] % bc == 0) ? bc : C[i ],
(C[i+1] % bk == 0) ? bk : C[i+1],
nThreads, my_fuse);
my_fc_bwd[i] = setup_my_fc_bwd(MB, C[i], C[i+1], (MB % bn == 0) ? bn : MB,
(C[i ] % bc == 0) ? bc : C[i ],
(C[i+1] % bk == 0) ? bk : C[i+1],
nThreads, my_fuse);
my_opt[i] = setup_my_opt( C[i], C[i+1], (C[i ] % bc == 0) ? bc : C[i ],
(C[i+1] % bk == 0) ? bk : C[i+1],
nThreads, lr );
/* let's allocate and bind scratch */
if ( my_fc_fwd[i].scratch_size > 0 || my_fc_bwd[i].scratch_size > 0 || my_opt[i].scratch_size > 0 ) {
size_t alloc_size = LIBXSMM_MAX( LIBXSMM_MAX( my_fc_fwd[i].scratch_size, my_fc_bwd[i].scratch_size), my_opt[i].scratch_size );
if ( alloc_size > scratch_size ) {
if ( scratch != NULL ) libxsmm_free( scratch );
scratch_size = alloc_size;
scratch = libxsmm_aligned_scratch( scratch_size, 2097152 );
my_init_buf( (float*)(scratch), (scratch_size)/4, 0, 0 );
}
}
}
/* softmax+loss is treated as N+! layer */
my_smax_fwd = setup_my_smax_fwd( MB, C[num_layers+1], (MB % bn == 0) ? bn : MB,
(C[num_layers+1] % bk == 0) ? bk : C[num_layers+1],
nThreads );
my_smax_bwd = setup_my_smax_bwd( MB, C[num_layers+1], (MB % bn == 0) ? bn : MB,
(C[num_layers+1] % bk == 0) ? bk : C[num_layers+1],
nThreads, loss_weight );
if ( my_smax_fwd.scratch_size > 0 || my_smax_bwd.scratch_size > 0 ) {
size_t alloc_size = LIBXSMM_MAX( my_smax_fwd.scratch_size, my_smax_bwd.scratch_size );
if ( alloc_size > scratch_size ) {
if ( scratch != NULL ) libxsmm_free( scratch );
scratch_size = alloc_size;
scratch = libxsmm_aligned_scratch( scratch_size, 2097152 );
my_init_buf( (float*)(scratch), (scratch_size)/4, 0, 0 );
}
}
my_numa_thr_cfg *numa_thr_cfg;
/* Define numa configuration: #numa nodes, #threads on each node */
setup_my_numa(&numa_thr_cfg, num_layers, nThreads);
if ( type == 'F') {
printf("##########################################\n");
printf("# Performance - FWD (custom-Storage) #\n");
printf("##########################################\n");
setup_my_numa_fwd(&numa_thr_cfg, num_layers, my_fc_fwd);
allocate_numa_buffers_fwd(&numa_thr_cfg, num_layers, my_fc_fwd);
l_start = libxsmm_timer_tick();
#if defined(_OPENMP)
# pragma omp parallel private(i,j)
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
const int numa_node = numa_node_of_cpu(tid);
for ( i = 0; i < num_layers; ++i) {
copy_to_numa_buffers_fwd(&numa_thr_cfg[numa_node], my_fc_fwd[i], fil_libxsmm[i], numa_node, i, tid, 0);
}
for (j = 0; j < iters; ++j) {
for ( i = 0; i < num_layers; ++i) {
my_fc_fwd_exec( my_fc_fwd[i], act_libxsmm[i], act_libxsmm[i+1],
bias_libxsmm[i], relumask_libxsmm[i], 0, tid, scratch, &numa_thr_cfg[numa_node], i);
}
#ifdef USE_SOFTMAX
my_smax_fwd_exec( my_smax_fwd, act_libxsmm[num_layers], act_libxsmm[num_layers+1], label_libxsmm, &loss,
0, tid, scratch );
#endif
}
}
l_end = libxsmm_timer_tick();
l_total = libxsmm_timer_duration(l_start, l_end);
gflop = 0.0;
for ( i = 0; i < num_layers; ++i) {
gflop += (2.0*(double)MB*(double)C[i]*(double)C[i+1]*(double)iters) / (1000.0*1000.0*1000.0);
}
printf("GFLOP = %.5g\n", gflop/(double)iters);
printf("fp time = %.5g\n", ((double)(l_total/iters)));
printf("GFLOPS = %.5g\n", gflop/l_total);
printf("PERFDUMP,FP,%s,%i,%i,", LIBXSMM_VERSION, nThreads, MB );
for ( i = 0; i < num_layers; ++i ) {
printf("%i,", C[i] );
}
printf("%f,%f\n", ((double)(l_total/iters)), gflop/l_total);
/* Print some norms on last act for fwd and weights of first layer after all iterations */
libxsmm_matdiff(&norms_fwd, LIBXSMM_DATATYPE_F32, MB*C[num_layers], 1, act_libxsmm[num_layers], act_libxsmm[num_layers], 0, 0);
printf("L1 of act[num_layers] : %.25g\n", norms_fwd.l1_ref);
}
if (type == 'B') {
printf("##########################################\n");
printf("# NOT Supported: Performance - BWD (custom-Storage) #\n");
printf("##########################################\n");
exit( -1 );
#if 0
l_start = libxsmm_timer_tick();
#if defined(_OPENMP)
# pragma omp parallel private(i,j)
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
for (j = 0; j < iters; ++j) {
#ifdef USE_SOFTMAX
my_smax_bwd_exec( my_smax_bwd, delact_libxsmm[num_layers], act_libxsmm[num_layers+1], label_libxsmm,
0, tid, scratch );
#endif
for ( i = num_layers-1; i > 0; --i) {
my_fc_bwd_exec( my_fc_bwd[i], fil_libxsmm[i], delact_libxsmm[i], delact_libxsmm[i+1], delfil_libxsmm[i],
act_libxsmm[i], delbias_libxsmm[i], relumask_libxsmm[i], MY_PASS_BWD, 0, tid, scratch );
my_opt_exec( my_opt[i], fil_libxsmm[i], delfil_libxsmm[i], 0, tid, scratch );
}
my_fc_bwd_exec( my_fc_bwd[0], fil_libxsmm[0], delact_libxsmm[0], delact_libxsmm[0+1], delfil_libxsmm[0],
act_libxsmm[0], delbias_libxsmm[0], relumask_libxsmm[0], MY_PASS_BWD_W, 0, tid, scratch );
my_opt_exec( my_opt[0], fil_libxsmm[0], delfil_libxsmm[0], 0, tid, scratch );
}
}
l_end = libxsmm_timer_tick();
l_total = libxsmm_timer_duration(l_start, l_end);
gflop = 0.0;
for ( i = num_layers-1; i > 0; --i) {
gflop += (4.0*(double)MB*(double)C[i]*(double)C[i+1]*(double)iters) / (1000.0*1000.0*1000.0);
}
gflop += (2.0*(double)MB*(double)C[0]*(double)C[1]*(double)iters) / (1000.0*1000.0*1000.0);
printf("GFLOP = %.5g\n", gflop/(double)iters);
printf("fp time = %.5g\n", ((double)(l_total/iters)));
printf("GFLOPS = %.5g\n", gflop/l_total);
printf("PERFDUMP,BP,%s,%i,%i,", LIBXSMM_VERSION, nThreads, MB );
for ( i = 0; i < num_layers; ++i ) {
printf("%i,", C[i] );
}
printf("%f,%f\n", ((double)(l_total/iters)), gflop/l_total);
#endif
}
if (type == 'A') {
printf("##########################################\n");
printf("# Performance - FWD-BWD (custom-Storage) #\n");
printf("##########################################\n");
/* Timers: */
fwd_time = (unsigned long long *) malloc(sizeof(unsigned long long) * nThreads);
bwd_time = (unsigned long long *) malloc(sizeof(unsigned long long) * nThreads);
solver_time = (unsigned long long *) malloc(sizeof(unsigned long long) * nThreads);
/* Calculate chunks of weights used on each nume node on FWD based on FWD thread decomposition */
setup_my_numa_fwd(&numa_thr_cfg, num_layers, my_fc_fwd);
/* Calculate chunks of weights used on each nume node on BWD/d based on BWD/d thread decomposition */
setup_my_numa_bwd_d(&numa_thr_cfg, num_layers, my_fc_bwd);
/* NUMA aware allocations of buffers needed for FWD */
allocate_numa_buffers_fwd(&numa_thr_cfg, num_layers, my_fc_fwd);
/* NUMA aware allocations of buffers needed for BWD */
allocate_numa_buffers_bwd_d(&numa_thr_cfg, num_layers, my_fc_bwd);
/* Utility needed for transpoisition of weigths on BWD/d: get numa node based on current ofm */
int **fwd_ofm_to_node = (int**)malloc(sizeof(int*) * num_layers);
set_fwd_ofm_to_node(fwd_ofm_to_node, &numa_thr_cfg, num_layers, my_fc_fwd);
l_start = libxsmm_timer_tick();
#if defined(_OPENMP)
# pragma omp parallel private(i,j)
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
fwd_time[tid] = 0;
bwd_time[tid] = 0;
solver_time[tid] = 0;
const int numa_node = numa_node_of_cpu(tid);
for ( i = 0; i < num_layers; ++i) {
/* Copy original weights to NUMA FWD buffers. Threading decomposition is the same with FWD. */
copy_to_numa_buffers_fwd(&numa_thr_cfg[numa_node], my_fc_fwd[i], fil_libxsmm[i], numa_node, i, tid, 0);
}
for (j = 0; j < iters; ++j) {
unsigned long long fwd_time_start = libxsmm_timer_tick();
for ( i = 0; i < num_layers; ++i) {
/* FWD: Use weights from NUMA FWD buffers */
my_fc_fwd_exec( my_fc_fwd[i], act_libxsmm[i], act_libxsmm[i+1],
bias_libxsmm[i], relumask_libxsmm[i], 0, tid, scratch, &numa_thr_cfg[numa_node], i );
}
fwd_time[tid] += (libxsmm_timer_tick() - fwd_time_start);
#ifdef USE_SOFTMAX
my_smax_fwd_exec( my_smax_fwd, act_libxsmm[num_layers], act_libxsmm[num_layers+1], label_libxsmm, &loss,
0, tid, scratch );
my_smax_bwd_exec( my_smax_bwd, delact_libxsmm[num_layers], act_libxsmm[num_layers+1], label_libxsmm,
0, tid, scratch );
#endif
for ( i = num_layers-1; i > 0; --i) {
unsigned long long bwd_time_start = libxsmm_timer_tick();
/* Transpose weights from NUMA FWD buffers to NUMA BWD buffer. Threading decomposition is the same with BWD/d. */
my_fc_bwd_d_transpose( my_fc_bwd[i], tid , &numa_thr_cfg, numa_node, i, fwd_ofm_to_node[i] );
/* BWD/d: Use weights from NUMA BWD buffers */
my_fc_bwd_exec( my_fc_bwd[i], delact_libxsmm[i], delact_libxsmm[i+1], delfil_libxsmm[i],
act_libxsmm[i], delbias_libxsmm[i], relumask_libxsmm[i], MY_PASS_BWD, 0, tid, scratch, &numa_thr_cfg[numa_node], i );
bwd_time[tid] += (libxsmm_timer_tick() - bwd_time_start);
/* Solver: Update NUMA FWD buffers. Threading decomposition is the same with FWD. */
unsigned long long solver_time_start = libxsmm_timer_tick();
my_opt_exec( my_opt[i], delfil_libxsmm[i], 0, tid, &numa_thr_cfg[numa_node], i, my_fc_fwd[i] );
solver_time[tid] += (libxsmm_timer_tick() - solver_time_start);
}
/* BWD/w: todo */
unsigned long long bwd_time_start = libxsmm_timer_tick();
my_fc_bwd_exec( my_fc_bwd[0], delact_libxsmm[0], delact_libxsmm[0+1], delfil_libxsmm[0],
act_libxsmm[0], delbias_libxsmm[0], relumask_libxsmm[0], MY_PASS_BWD_W, 0, tid, scratch, &numa_thr_cfg[numa_node], 0 );
bwd_time[tid] += (libxsmm_timer_tick() - bwd_time_start);
/* Solver: Update NUMA FWD buffers. Threading decomposition is the same with FWD. */
unsigned long long solver_time_start = libxsmm_timer_tick();
my_opt_exec( my_opt[0], delfil_libxsmm[0], 0, tid, &numa_thr_cfg[numa_node], 0, my_fc_fwd[0] );
solver_time[tid] += (libxsmm_timer_tick() - solver_time_start);
}
/* Copy result from NUMA FWD Buffers to original weights. Threading decomposition is the same with FWD. */
for ( i = 0; i < num_layers; ++i) {
copy_to_numa_buffers_fwd(&numa_thr_cfg[numa_node], my_fc_fwd[i], fil_libxsmm[i], numa_node, i, tid, 1);
}
}
l_end = libxsmm_timer_tick();
l_total = libxsmm_timer_duration(l_start, l_end);
free_fwd_ofm_to_node(fwd_ofm_to_node, num_layers);
free(fwd_ofm_to_node);
#ifdef CHECK_L1
#if 1
/* Print some norms on last act for fwd and weights of first layer after all iterations */
libxsmm_matdiff(&norms_fwd, LIBXSMM_DATATYPE_F32, MB*C[num_layers], 1, act_libxsmm[num_layers], act_libxsmm[num_layers], 0, 0);
printf("L1 of act[num_layers] : %.25g\n", norms_fwd.l1_ref);
libxsmm_matdiff_reduce(&diff, &norms_fwd);
libxsmm_matdiff(&norms_bwd, LIBXSMM_DATATYPE_F32, C[0]*C[1], 1, fil_libxsmm[0], fil_libxsmm[0], 0, 0);
printf("L1 of wt[0] : %.25g\n", norms_bwd.l1_ref);
libxsmm_matdiff_reduce(&diff, &norms_bwd);
#else
{
int e = 0;
FILE *fileAct, *fileWt;
fileAct = fopen("acts.txt","w+");
if (fileAct != NULL) {
for (e = 0; e < MB*C[num_layers]; e++) {
fprintf(fileAct, "%.10g\n", *((float*)act_libxsmm[num_layers] + e));
}
fclose(fileAct);
}
fileWt = fopen("weights.txt","w+");
if (fileWt != NULL) {
for (e = 0; e < C[0]*C[1]; e++) {
fprintf(fileWt, "%.10g\n", *((float*)fil_libxsmm[0] + e));
}
fclose(fileWt);
}
}
#endif
#endif
gflop = 0.0;
for ( i = num_layers-1; i > 0; --i) {
gflop += (6.0*(double)MB*(double)C[i]*(double)C[i+1]*(double)iters) / (1000.0*1000.0*1000.0);
}
gflop += (4.0*(double)MB*(double)C[0]*(double)C[1]*(double)iters) / (1000.0*1000.0*1000.0);
printf("GFLOP = %.5g\n", gflop/(double)iters);
printf("fp time = %.5g\n", ((double)(l_total/iters)));
printf("GFLOPS = %.5g\n", gflop/l_total);
printf("PERFDUMP,BP,%s,%i,%i,", LIBXSMM_VERSION, nThreads, MB );
for ( i = 0; i < num_layers; ++i ) {
printf("%i,", C[i] );
}
printf("%f,%f\n", ((double)(l_total/iters)), gflop/l_total);
unsigned long long max_fwd_time = 0, max_bwd_time = 0, max_solver_time = 0;
for (i = 0; i < nThreads; i++) {
if (max_fwd_time < fwd_time[i]) max_fwd_time = fwd_time[i];
if (max_bwd_time < bwd_time[i]) max_bwd_time = bwd_time[i];
if (max_solver_time < solver_time[i]) max_solver_time = solver_time[i];
}
printf("Profiling: fwd_time = %lld, bwd_time = %lld, solver_time = %lld\n",
max_fwd_time, max_bwd_time, max_solver_time);
}
/* deallocate data */
if ( scratch != NULL ) {
libxsmm_free(scratch);
}
for ( i = 0; i < num_layers; ++i ) {
if ( i == 0 ) {
#ifdef ACT_NUMA_INTERLEAVED
numa_free(act_libxsmm[i], MB*C[i]*sizeof(float));
#else
libxsmm_free(act_libxsmm[i]);
#endif
libxsmm_free(delact_libxsmm[i]);
}
#ifdef ACT_NUMA_INTERLEAVED
numa_free(act_libxsmm[i+1], MB*C[i+1]*sizeof(float));
#else
libxsmm_free(act_libxsmm[i+1]);
#endif
libxsmm_free(delact_libxsmm[i+1]);
libxsmm_free(fil_libxsmm[i]);
libxsmm_free(delfil_libxsmm[i]);
libxsmm_free(bias_libxsmm[i]);
libxsmm_free(delbias_libxsmm[i]);
libxsmm_free(relumask_libxsmm[i]);
}
#ifdef ACT_NUMA_INTERLEAVED
numa_free(act_libxsmm[num_layers+1], MB*C[num_layers+1]*sizeof(float));
#else
libxsmm_free(act_libxsmm[num_layers+1]);
#endif
libxsmm_free(label_libxsmm);
for (i = 0; i < numa_num_configured_nodes(); i++) {
free(numa_thr_cfg[i].blocksOFm_s);
free(numa_thr_cfg[i].blocksOFm_e);
free(numa_thr_cfg[i].blocksIFm_tr_s);
free(numa_thr_cfg[i].blocksIFm_tr_e);
for (j = 0; j < num_layers; j++) {
numa_free_aligned(numa_thr_cfg[i].scratch[j], numa_thr_cfg[i].layer_size[j]);
}
free(numa_thr_cfg[i].scratch);
free(numa_thr_cfg[i].layer_size);
numa_free_aligned(numa_thr_cfg[i].bwd_d_scratch, numa_thr_cfg[i].bwd_d_scratch_size);
}
free(numa_thr_cfg);
free( my_opt );
free( my_fc_fwd );
free( my_fc_bwd );
free( act_libxsmm );
free( delact_libxsmm );
free( fil_libxsmm );
free( delfil_libxsmm );
free( bias_libxsmm );
free( delbias_libxsmm );
free( relumask_libxsmm );
free( C );
/* some empty lines at the end */
printf("\n\n\n");
return 0;
}
|
GB_binop__bxnor_int16.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_emult.h"
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB (_AaddB__bxnor_int16)
// A.*B function (eWiseMult): GB (_AemultB_08__bxnor_int16)
// A.*B function (eWiseMult): GB (_AemultB_02__bxnor_int16)
// A.*B function (eWiseMult): GB (_AemultB_04__bxnor_int16)
// A.*B function (eWiseMult): GB (_AemultB_bitmap__bxnor_int16)
// A*D function (colscale): GB ((none))
// D*A function (rowscale): GB ((none))
// C+=B function (dense accum): GB (_Cdense_accumB__bxnor_int16)
// C+=b function (dense accum): GB (_Cdense_accumb__bxnor_int16)
// C+=A+B function (dense ewise3): GB ((none))
// C=A+B function (dense ewise3): GB (_Cdense_ewise3_noaccum__bxnor_int16)
// C=scalar+B GB (_bind1st__bxnor_int16)
// C=scalar+B' GB (_bind1st_tran__bxnor_int16)
// C=A+scalar GB (_bind2nd__bxnor_int16)
// C=A'+scalar GB (_bind2nd_tran__bxnor_int16)
// C type: int16_t
// A type: int16_t
// B,b type: int16_t
// BinaryOp: cij = ~((aij) ^ (bij))
#define GB_ATYPE \
int16_t
#define GB_BTYPE \
int16_t
#define GB_CTYPE \
int16_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA,A_iso) \
int16_t aij = GBX (Ax, pA, A_iso)
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB,B_iso) \
int16_t bij = GBX (Bx, pB, B_iso)
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int16_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA,A_iso) \
cij = GBX (Ax, pA, A_iso)
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB,B_iso) \
cij = GBX (Bx, pB, B_iso)
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z,x,y,i,j) \
z = ~((x) ^ (y)) ;
// true if the binop must be flipped
#define GB_BINOP_FLIP \
0
// op is second
#define GB_OP_IS_SECOND \
0
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_BXNOR || GxB_NO_INT16 || GxB_NO_BXNOR_INT16)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_ewise3_noaccum__bxnor_int16)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumB__bxnor_int16)
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *B_ek_slicing, const int B_ntasks, const int B_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB (_Cdense_accumb__bxnor_int16)
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int16_t
int16_t bwork = (*((int16_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info GB ((none))
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *restrict Cx = (int16_t *) C->x ;
#include "GB_AxB_colscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info GB ((none))
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *restrict Cx = (int16_t *) C->x ;
#include "GB_AxB_rowscale_template.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// eWiseAdd: C=A+B, C<M>=A+B, C<!M>=A+B
//------------------------------------------------------------------------------
GrB_Info GB (_AaddB__bxnor_int16)
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
GB_WERK_DECLARE (M_ek_slicing, int64_t) ;
GB_WERK_DECLARE (A_ek_slicing, int64_t) ;
GB_WERK_DECLARE (B_ek_slicing, int64_t) ;
#include "GB_add_template.c"
GB_FREE_WORK ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, or C<M!>=A.*B where C is sparse/hyper
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_08__bxnor_int16)
(
GrB_Matrix C,
const int C_sparsity,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict C_to_M,
const int64_t *restrict C_to_A,
const int64_t *restrict C_to_B,
const GB_task_struct *restrict TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_08_meta.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<#> = A.*B when A is sparse/hyper and B is bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_02__bxnor_int16)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool flipxy,
const int64_t *restrict Cp_kfirst,
const int64_t *A_ek_slicing, const int A_ntasks, const int A_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#if GB_BINOP_FLIP
// The operator is not commutative, and does not have a flipped
// variant. For example z=atan2(y,x).
if (flipxy)
{
// use fmult(y,x)
#undef GB_FLIPPED
#define GB_FLIPPED 1
#include "GB_emult_02_template.c"
}
else
{
// use fmult(x,y)
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
}
#else
// No need to handle the flip: the operator is either commutative, or
// has been handled by changing z=div(y,x) to z=rdiv(x,y) for example.
#undef GB_FLIPPED
#define GB_FLIPPED 0
#include "GB_emult_02_template.c"
#endif
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C<M> = A.*B, M sparse/hyper, A and B bitmap/full
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_04__bxnor_int16)
(
GrB_Matrix C,
const GrB_Matrix M,
const bool Mask_struct,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *restrict Cp_kfirst,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_emult_04_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C=A.*B, C<M>=A.*B, C<!M>=A.*B where C is bitmap
//------------------------------------------------------------------------------
GrB_Info GB (_AemultB_bitmap__bxnor_int16)
(
GrB_Matrix C,
const int ewise_method,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *M_ek_slicing, const int M_ntasks, const int M_nthreads,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_bitmap_emult_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB (_bind1st__bxnor_int16)
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *restrict Bb,
int64_t bnz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *Cx = (int16_t *) Cx_output ;
int16_t x = (*((int16_t *) x_input)) ;
int16_t *Bx = (int16_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < bnz ; p++)
{
if (!GBB (Bb, p)) continue ;
int16_t bij = GBX (Bx, p, false) ;
Cx [p] = ~((x) ^ (bij)) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB (_bind2nd__bxnor_int16)
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *restrict Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int16_t *Cx = (int16_t *) Cx_output ;
int16_t *Ax = (int16_t *) Ax_input ;
int16_t y = (*((int16_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int16_t aij = GBX (Ax, p, false) ;
Cx [p] = ~((aij) ^ (y)) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = ~((x) ^ (aij)) ; \
}
GrB_Info GB (_bind1st_tran__bxnor_int16)
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t x = (*((const int16_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = GBX (Ax, pA, false) ; \
Cx [pC] = ~((aij) ^ (y)) ; \
}
GrB_Info GB (_bind2nd_tran__bxnor_int16)
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t y = (*((const int16_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
GB_subassign_04.c | //------------------------------------------------------------------------------
// GB_subassign_04: C(I,J) += A ; using S
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Method 04: C(I,J) += A ; using S
// M: NULL
// Mask_comp: false
// C_replace: false
// accum: present
// A: matrix
// S: constructed
// C: not bitmap: use GB_bitmap_assign instead
// A: any sparsity structure.
#include "GB_subassign_methods.h"
GrB_Info GB_subassign_04
(
GrB_Matrix C,
// input:
const GrB_Index *I,
const int64_t ni,
const int64_t nI,
const int Ikind,
const int64_t Icolon [3],
const GrB_Index *J,
const int64_t nj,
const int64_t nJ,
const int Jkind,
const int64_t Jcolon [3],
const GrB_BinaryOp accum,
const GrB_Matrix A,
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (!GB_IS_BITMAP (C)) ;
ASSERT (!GB_aliased (C, A)) ; // NO ALIAS of C==A
//--------------------------------------------------------------------------
// S = C(I,J)
//--------------------------------------------------------------------------
GB_EMPTY_TASKLIST ;
GB_OK (GB_subassign_symbolic (S, C, I, ni, J, nj, true, Context)) ;
//--------------------------------------------------------------------------
// get inputs
//--------------------------------------------------------------------------
GB_MATRIX_WAIT_IF_JUMBLED (A) ;
GB_GET_C ; // C must not be bitmap
GB_GET_A ;
GB_GET_S ;
GB_GET_ACCUM ;
//--------------------------------------------------------------------------
// Method 04: C(I,J) += A ; using S
//--------------------------------------------------------------------------
// Time: Close to Optimal. Every entry in A must be visited, and the
// corresponding entry in S must then be found. Time for this phase is
// Omega(nnz(A)), but S has already been constructed, in Omega(nnz(S))
// time. This method simply traverses all of A+S (like GB_add for
// computing A+S), the same as Method 02. Time taken is O(nnz(A)+nnz(S)).
// The only difference is that the traversal of A+S can terminate if A is
// exhausted. Entries in S but not A do not actually require any work
// (unlike Method 02, which must visit all entries in A+S).
// Method 02 and Method 04 are somewhat similar. They differ on how C is
// modified when the entry is present in S but not A.
// TODO: phase2 of Method 02 and 04 are identical and could be
// done in a single function.
// Compare with Method 16, which computes C(I,J)<!M> += A, using S.
//--------------------------------------------------------------------------
// Parallel: A+S (Methods 02, 04, 09, 10, 11, 12, 14, 16, 18, 20)
//--------------------------------------------------------------------------
if (A_is_bitmap)
{
// all of IxJ must be examined
GB_SUBASSIGN_IXJ_SLICE ;
}
else
{
// traverse all A+S
GB_SUBASSIGN_TWO_SLICE (A, S) ;
}
//--------------------------------------------------------------------------
// phase 1: create zombies, update entries, and count pending tuples
//--------------------------------------------------------------------------
if (A_is_bitmap)
{
//----------------------------------------------------------------------
// phase1: A is bitmap
//----------------------------------------------------------------------
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(+:nzombies)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//------------------------------------------------------------------
// get the task descriptor
//------------------------------------------------------------------
GB_GET_IXJ_TASK_DESCRIPTOR_PHASE1 (iA_start, iA_end) ;
//------------------------------------------------------------------
// compute all vectors in this task
//------------------------------------------------------------------
for (int64_t j = kfirst ; j <= klast ; j++)
{
//--------------------------------------------------------------
// get S(iA_start:iA_end,j)
//--------------------------------------------------------------
GB_GET_VECTOR_FOR_IXJ (S, iA_start) ;
int64_t pA_start = j * Avlen ;
//--------------------------------------------------------------
// do a 2-way merge of S(iA_start:iA_end,j) and A(ditto,j)
//--------------------------------------------------------------
for (int64_t iA = iA_start ; iA < iA_end ; iA++)
{
int64_t pA = pA_start + iA ;
bool Sfound = (pS < pS_end) && (GBI (Si, pS, Svlen) == iA) ;
bool Afound = Ab [pA] ;
if (Sfound && !Afound)
{
// ----[C . 1] or [X . 1]-------------------------------
// S (i,j) is present but A (i,j) is not
// [C . 1]: action: ( C ): no change, with accum
// [X . 1]: action: ( X ): still a zombie
GB_NEXT (S) ;
}
else if (!Sfound && Afound)
{
// ----[. A 1]------------------------------------------
// S (i,j) is not present, A (i,j) is present
// [. A 1]: action: ( insert )
task_pending++ ;
}
else if (Sfound && Afound)
{
// ----[C A 1] or [X A 1]-------------------------------
// both S (i,j) and A (i,j) present
// [C A 1]: action: ( =C+A ): apply accum
// [X A 1]: action: ( undelete ): zombie lives
GB_C_S_LOOKUP ;
GB_withaccum_C_A_1_matrix ;
GB_NEXT (S) ;
}
}
}
GB_PHASE1_TASK_WRAPUP ;
}
}
else
{
//----------------------------------------------------------------------
// phase1: A is hypersparse, sparse, or full
//----------------------------------------------------------------------
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(+:nzombies)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//------------------------------------------------------------------
// get the task descriptor
//------------------------------------------------------------------
GB_GET_TASK_DESCRIPTOR_PHASE1 ;
//------------------------------------------------------------------
// compute all vectors in this task
//------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//--------------------------------------------------------------
// get A(:,j) and S(:,j)
//--------------------------------------------------------------
int64_t j = GBH (Zh, k) ;
GB_GET_MAPPED (pA, pA_end, pA, pA_end, Ap, j, k, Z_to_X, Avlen);
GB_GET_MAPPED (pS, pS_end, pB, pB_end, Sp, j, k, Z_to_S, Svlen);
//--------------------------------------------------------------
// do a 2-way merge of S(:,j) and A(:,j)
//--------------------------------------------------------------
// jC = J [j] ; or J is a colon expression
// int64_t jC = GB_ijlist (J, j, Jkind, Jcolon) ;
// while both list S (:,j) and A (:,j) have entries
while (pS < pS_end && pA < pA_end)
{
int64_t iS = GBI (Si, pS, Svlen) ;
int64_t iA = GBI (Ai, pA, Avlen) ;
if (iS < iA)
{
// ----[C . 1] or [X . 1]-------------------------------
// S (i,j) is present but A (i,j) is not
// [C . 1]: action: ( C ): no change, with accum
// [X . 1]: action: ( X ): still a zombie
GB_NEXT (S) ;
}
else if (iA < iS)
{
// ----[. A 1]------------------------------------------
// S (i,j) is not present, A (i,j) is present
// [. A 1]: action: ( insert )
task_pending++ ;
GB_NEXT (A) ;
}
else
{
// ----[C A 1] or [X A 1]-------------------------------
// both S (i,j) and A (i,j) present
// [C A 1]: action: ( =C+A ): apply accum
// [X A 1]: action: ( undelete ): zombie lives
GB_C_S_LOOKUP ;
GB_withaccum_C_A_1_matrix ;
GB_NEXT (S) ;
GB_NEXT (A) ;
}
}
// ignore the remainder of S (:,j)
// List A (:,j) has entries. List S (:,j) exhausted.
task_pending += (pA_end - pA) ;
}
GB_PHASE1_TASK_WRAPUP ;
}
}
//--------------------------------------------------------------------------
// phase 2: insert pending tuples
//--------------------------------------------------------------------------
GB_PENDING_CUMSUM ;
if (A_is_bitmap)
{
//----------------------------------------------------------------------
// phase2: A is bitmap
//----------------------------------------------------------------------
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(&&:pending_sorted)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//------------------------------------------------------------------
// get the task descriptor
//------------------------------------------------------------------
GB_GET_IXJ_TASK_DESCRIPTOR_PHASE2 (iA_start, iA_end) ;
//------------------------------------------------------------------
// compute all vectors in this task
//------------------------------------------------------------------
for (int64_t j = kfirst ; j <= klast ; j++)
{
//--------------------------------------------------------------
// get S(iA_start:iA_end,j)
//--------------------------------------------------------------
GB_GET_VECTOR_FOR_IXJ (S, iA_start) ;
int64_t pA_start = j * Avlen ;
//--------------------------------------------------------------
// do a 2-way merge of S(iA_start:iA_end,j) and A(ditto,j)
//--------------------------------------------------------------
// jC = J [j] ; or J is a colon expression
int64_t jC = GB_ijlist (J, j, Jkind, Jcolon) ;
for (int64_t iA = iA_start ; iA < iA_end ; iA++)
{
int64_t pA = pA_start + iA ;
bool Sfound = (pS < pS_end) && (GBI (Si, pS, Svlen) == iA) ;
bool Afound = Ab [pA] ;
if (!Sfound && Afound)
{
// ----[. A 1]------------------------------------------
// S (i,j) is not present, A (i,j) is present
// [. A 1]: action: ( insert )
int64_t iC = GB_ijlist (I, iA, Ikind, Icolon) ;
GB_PENDING_INSERT (Ax +(pA*asize)) ;
GB_NEXT (A) ;
}
else if (Sfound)
{
// S (i,j) present
GB_NEXT (S) ;
}
}
}
GB_PHASE2_TASK_WRAPUP ;
}
}
else
{
//----------------------------------------------------------------------
// phase2: A is hypersparse, sparse, or full
//----------------------------------------------------------------------
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1) \
reduction(&&:pending_sorted)
for (taskid = 0 ; taskid < ntasks ; taskid++)
{
//------------------------------------------------------------------
// get the task descriptor
//------------------------------------------------------------------
GB_GET_TASK_DESCRIPTOR_PHASE2 ;
//------------------------------------------------------------------
// compute all vectors in this task
//------------------------------------------------------------------
for (int64_t k = kfirst ; k <= klast ; k++)
{
//--------------------------------------------------------------
// get A(:,j) and S(:,j)
//--------------------------------------------------------------
int64_t j = GBH (Zh, k) ;
GB_GET_MAPPED (pA, pA_end, pA, pA_end, Ap, j, k, Z_to_X, Avlen);
GB_GET_MAPPED (pS, pS_end, pB, pB_end, Sp, j, k, Z_to_S, Svlen);
//--------------------------------------------------------------
// do a 2-way merge of S(:,j) and A(:,j)
//--------------------------------------------------------------
// jC = J [j] ; or J is a colon expression
int64_t jC = GB_ijlist (J, j, Jkind, Jcolon) ;
// while both list S (:,j) and A (:,j) have entries
while (pS < pS_end && pA < pA_end)
{
int64_t iS = GBI (Si, pS, Svlen) ;
int64_t iA = GBI (Ai, pA, Avlen) ;
if (iS < iA)
{
GB_NEXT (S) ;
}
else if (iA < iS)
{
// ----[. A 1]------------------------------------------
// S (i,j) is not present, A (i,j) is present
// [. A 1]: action: ( insert )
int64_t iC = GB_ijlist (I, iA, Ikind, Icolon) ;
GB_PENDING_INSERT (Ax +(pA*asize)) ;
GB_NEXT (A) ;
}
else
{
GB_NEXT (S) ;
GB_NEXT (A) ;
}
}
// ignore the remainder of S (:,j)
// while list A (:,j) has entries. List S (:,j) exhausted.
while (pA < pA_end)
{
// ----[. A 1]----------------------------------------------
// S (i,j) is not present, A (i,j) is present
// [. A 1]: action: ( insert )
int64_t iA = GBI (Ai, pA, Avlen) ;
int64_t iC = GB_ijlist (I, iA, Ikind, Icolon) ;
GB_PENDING_INSERT (Ax +(pA*asize)) ;
GB_NEXT (A) ;
}
}
GB_PHASE2_TASK_WRAPUP ;
}
}
//--------------------------------------------------------------------------
// finalize the matrix and return result
//--------------------------------------------------------------------------
GB_SUBASSIGN_WRAPUP ;
}
|
line_ingredients.c |
/** @file line_ingredients.c Documented line_ingredients module
*
* This module includes functions that are needed for computing the line clustering and shot contributions.
*
* Azadeh Moradinezhad Dizgah, November 4th 2021
*
* In summary, the following functions can be called from other modules:
* -# Line_alloc_init() allocate memory and initizlized the line structure which contains 4 interpolators for first and second mass moments and linear and quadratic line biases.
* -# Line_free() frees the memory allocated to line structure
* -# Line_evaluate() evaluates the interpolators initialized in Line_alloc_init()
* -# mult_func() computes the multiplicity function needed for computing the halo mass function
* -# mass_func() computes the halo mass finction. Three options are available, Press-Schecter, Sheth-Tormen, Tinker
* -# mass_func_sims() reads in the measured mass function on Hidden-Valley simulations by Farnik, and convert it to compare with the theoretical predictions
* -# halo_bias() computes the halo biases assuming the above theoretical predictions of the halo mass function
* -# logSFR_Behroozi_read() reeds in the data file of Behroozi 2013 for SFR(M,z)
* -# logSFR_alloc_init() allocates memory for 2d interpolator of logSFR(M,z)
* -# SFR_behroozi_free() frees the memory allocated to logSFR interpolator
* -# logSFR_Behroozi() evaluates the logSFR_Behroozi interpolator
* -# luminosity() computes the line luminosity
* -# mass_moment1() computes the first mass moment
* -# mass_moment2() computes the first mass moment
* -# bias_lum_weighted() computes the luminosity-weighetd line bias
* -# p_sig_shot() computes the coefficient accounting for the scatter in L(M) in shot noise
* -# p_sig_Tbar() computes the coefficient accounting for the scatter in L(M) in mean brightness temprature
* -# mean_intens() compues the mean intensity of the line
* -# Tbar_line() compues the mean brightness temprature of the line
*
*/
#include "header.h"
/**
* Allocate the memory and initialize the the line structure. This structure contains interpolators for computing the luminosity-weighted mass moments and line biases
* For a given line defined with "line_type" variable, this function first computes the above four quantities for a wide range of redshifts.
* Next it iniialized 4 interpolators for these quantities, and store them in line structure.
*
* @param Cx Input: Cosmology structure
* @param line_type Inpute: name of the line to compute.
* It can be set to CII, CO10, CO21, CO32, CO43, CO54, CO65
* @param npoints_interp Input: number of interpolation points
* @param M_min Input: minimum halo mass for mass integrals
* @param mode_mf Inpute: theoretical model of halo mass function to use.
* It can be set to sheth-Tormen (ST), Tinker (TR) or Press-Schecter (PSC)
* @return the total clustering line power spectrum, including the 1- and 2-halo term
*/
struct Line * Line_alloc_init(struct Cosmology *Cx, long line_type, size_t npoints_interp, double M_min, long mode_mf)
{
struct Line * ThisLine;
// Allocation of the Line object
ThisLine = (struct Line *)malloc(sizeof(struct Line));
// Attributes
ThisLine->LineType = line_type;
int J=0;
if(line_type == CII){
ThisLine->line_freq = 1902. * pow(10.,9.); ///CII
J = 0;
}
else
{
if(line_type == CO10)
J = 1;
else if(line_type == CO21)
J = 2;
else if(line_type == CO32)
J = 3;
else if(line_type == CO43)
J = 4;
else if(line_type == CO54)
J = 5;
else if(line_type == CO65)
J = 6;
ThisLine->line_freq = J * 115.27 * pow(10.,9.); ////in unit of Hz for CO
}
// Allocation of interpolation objects
ThisLine -> mom1_accel_ptr = gsl_interp_accel_alloc();
ThisLine -> mom1_spline_ptr = gsl_spline_alloc(gsl_interp_linear, npoints_interp);
ThisLine -> mom2_accel_ptr = gsl_interp_accel_alloc();
ThisLine -> mom2_spline_ptr = gsl_spline_alloc(gsl_interp_linear, npoints_interp);
ThisLine -> b1_LW_accel_ptr = gsl_interp_accel_alloc();
ThisLine -> b1_LW_spline_ptr = gsl_spline_alloc(gsl_interp_linear, npoints_interp);
ThisLine -> b2_LW_accel_ptr = gsl_interp_accel_alloc();
ThisLine -> b2_LW_spline_ptr = gsl_spline_alloc(gsl_interp_linear, npoints_interp);
// Initialization of the interpolating objects
double *zz, *mom1, *mom2, *b1_LW, *b2_LW;
double zmin =0.0, zmax = gb.line_zmax;
// Allocate temporary arrays
mom1 = make_1Darray(npoints_interp);
mom2 = make_1Darray(npoints_interp);
b1_LW = make_1Darray(npoints_interp);
b2_LW = make_1Darray(npoints_interp);
zz = init_1Darray(npoints_interp,zmin,zmax);
double bias_arr[npoints_interp][2];
// Calculating values to be interpolated
#pragma omp parallel
#pragma omp for
for(int i=0;i<npoints_interp;i++){
mom1[i] = mass_moment1(Cx,zz[i],M_min,mode_mf,line_type);
mom2[i] = mass_moment2(Cx,zz[i],M_min,mode_mf,line_type);
bias_lum_weighted(Cx,zz[i],M_min,mode_mf,line_type,bias_arr[i]);
b1_LW[i] = bias_arr[i][0];
b2_LW[i] = bias_arr[i][1];
printf("hm %ld %d %12.6e %12.6e %12.6e %12.6e %12.6e \n", line_type, i, zz[i], mom1[i], mom2[i], b1_LW[i], b2_LW[i]);
// printf("process %d done computing redshift %12.6e\n", omp_get_thread_num(),zz[i]);
}
// Initialize interpolating object
gsl_spline_init(ThisLine -> mom1_spline_ptr, zz, mom1, npoints_interp);
gsl_spline_init(ThisLine -> mom2_spline_ptr, zz, mom2, npoints_interp);
gsl_spline_init(ThisLine -> b1_LW_spline_ptr, zz, b1_LW,npoints_interp);
gsl_spline_init(ThisLine -> b2_LW_spline_ptr, zz, b2_LW,npoints_interp);
printf("for line %d mom1, mom2, b1 and b2 interpolators initialized\n", line_type);
// Freeing temporary arrays
free(zz);
free(mom1);
free(mom2);
free(b1_LW);
free(b2_LW);
ThisLine->initialized = _TRUE_;
return ThisLine;
}
/**
* Free the line structure
*
* @param Lx Input: Pointer to line structure
* @return the error status
*/
int Line_free(struct Line * Lx)
{
gsl_interp_accel_free(Lx->mom1_accel_ptr);
gsl_spline_free(Lx->mom1_spline_ptr);
gsl_interp_accel_free(Lx->mom2_accel_ptr);
gsl_spline_free(Lx->mom2_spline_ptr);
gsl_interp_accel_free(Lx->b1_LW_accel_ptr);
gsl_spline_free(Lx->b1_LW_spline_ptr);
gsl_interp_accel_free(Lx->b2_LW_accel_ptr);
gsl_spline_free(Lx->b2_LW_spline_ptr);
free(Lx);
return _SUCCESS_;
}
/**
* Allocate the memory and initialize the the line structure. This structure contains interpolators for computing the luminosity-weighted mass moments and line biases
* For a given line defined with "line_type" variable, this function first computes the above four quantities for a wide range of redshifts.
* Next it iniialized 4 interpolators for these quantities, and store them in line structure.
*
* @param Lx Input: Pointer to the line structure
* @param zz Input: this is an array with 4 elements to determine which of the 4 interpolators should be evaluated.
* - If any of the elements are set to DO_NOT_EVALUATE, the quantitiy corresponding to that index is not computed. O
* - If any of the elements is set to z, the corresponding quantity would be evaluated at that redshift
* @param res Output: an array containing the results. The number of elements of this array depends on how the zz array is set.
* @return the error status
*/
int Line_evaluate(struct Line * Lx, double *zz, double *res)
{
if(zz[0] != DO_NOT_EVALUATE){
res[0] = gsl_spline_eval(Lx->mom1_spline_ptr, zz[0], Lx->mom1_accel_ptr);
}
if(zz[1] != DO_NOT_EVALUATE){
res[1] = gsl_spline_eval(Lx->mom2_spline_ptr, zz[1], Lx->mom2_accel_ptr);
}
if(zz[2] != DO_NOT_EVALUATE){
res[2] = gsl_spline_eval(Lx->b1_LW_spline_ptr, zz[2], Lx->b1_LW_accel_ptr);
}
if(zz[3] != DO_NOT_EVALUATE){
res[3] = gsl_spline_eval(Lx->b2_LW_spline_ptr, zz[3], Lx->b2_LW_accel_ptr);
}
return _SUCCESS_;
}
/**
* Compute the multiplicity function needed to compute the halo mass function
* Three models are implemented: Press-Schechter, Sheth-Tormen and Tinker
* see Pillepich et al arxiv: 0811.4176 for the expressions.
*
* @param sigma Input: variance of matter fluctuations
* @param mode_mf Input: switch for setting the model of mass function, can be set to PSC, ST, TR
* @return the multiplicity function
*/
double mult_func(double sigma, long mode_mf)
{
double f = 0;
double delta_c = 1.686;
double nu = delta_c/sigma;
if(mode_mf == PSC)
f = sqrt(2./M_PI)* nu * exp(-pow(nu,2.)/2.);
else if(mode_mf == ST){
double A = 0.3222;
double a = 0.707; /// In Barkana & Loeb Rev a = 0.75
double p = 0.3;
f = A * sqrt(2.*a/M_PI) * nu* exp(-(a*pow(nu,2.)/2.)) * (1.+pow(pow(nu,-2.)/a,p));
}
else if(mode_mf == TK){
//// Mass function of Tinker et al 2008 arxiv: 0803.2706 for Delta =200
double A = 0.186;
double a = 1.47;
double b = 2.57;
double c = 1.19;
f = A*(pow(sigma/b,-a) + 1.) *exp(-c/pow(sigma,2.));
//// Mass function of Tinker et al 2010 arxiv:1001.3162 for Delta =200
// double alpha = 0.368;
// double beta = 0.589;
// double gamma = 0.864;
// double phi = -0.729;
// double eta = -0.243;
// f = alpha * (1.+pow(beta*nu,-2.* phi))*pow(nu,2.*eta)*exp(-gamma*pow(nu,2.)/2.);
}
return f;
}
/**
* Compute the halo mass function for Press-Schechter, Sheth-Tormen and Tinker models
* see Pillepich et al arxiv: 0811.4176 for the expressions.
*
* @param Cx Input: Cosmology structure
* @param M Input: Halo mass function
* @param z Input: redshift
* @param mode_mf Input: switch for setting the model of mass function, can be set to PSC, ST, TR
* @return the halo mass function
*/
double mass_func(struct Cosmology *Cx, double M, double z, long mode_mf ) ///in unit of halos per Mpc^3 per solar mass, compared at z=0 with Murray etal https://arxiv.org/abs/1306.5140
{
double f = 0;
double R = R_scale(Cx,M);
double omega_m = Cx->cosmo_pars[3L] + Cx->cosmo_pars[4L];
double rho_m = omega_m* rhoc(Cx, 0.);
double sigma = sqrt(sig_sq(Cx, z, R));
/////If calculating derivative of sigma numerically
double Mmin = (1-0.05)*M;
double Mmax = (1+0.05)*M;
double Rmin = R_scale(Cx, Mmin);
double Rmax = R_scale(Cx, Mmax);
double sigma_min = sqrtl(sig_sq(Cx, z, Rmin));
double sigma_max = sqrtl(sig_sq(Cx, z, Rmax));
double der_lnsig_dM = (log(sigma_max) - log(sigma_min))/(2.*0.05*M);
f = - mult_func(sigma,mode_mf) * rho_m/M * der_lnsig_dM;
return f;
}
/**
* Read in the measured mass function of Hidden-valey sims and build an interpolator for HMF(M) for a fixed redshift.
* @param Cx Input: Cosmology structure
* @param M Input: halo mass
* @param z Input: redshift
* @param mode_mf Input: switch for setting the model of mass function, can be set to PSC, ST, TR
* @return the interpolated measured halo mass function
*/
double mass_func_sims(struct Cosmology *Cx, double M, double z, long mode_mf) ///M in unit of M_sun and HMF in unit of #-of-halos/Mpc^3/M_sun
{
///not to be used at z other than 2. This function is for testing purposes only. We test the theoretical predictions at z=2 if ST MF or measured MF are used.
///
char HMF_filename[FILENAME_MAX];
sprintf(HMF_filename,"%s/MithraLIMSims_HMF/hmf-z%s.00.dat.txt", gb.output_dir, (int)z);
int nlines = 0;
nlines = count_lines_in_file(HMF_filename);
double *log10M_input = make_1Darray(nlines);
double *hmf_input = make_1Darray(nlines);
double *log10hmf_in = make_1Darray(nlines);
static gsl_interp_accel *hmf_accel_ptr;
static gsl_spline *hmf_spline_ptr;
static int first = 1;
if(first ==1 )
{
FILE *HMF_file;
HMF_file = fopen(HMF_filename,"r");
char line[MAXL];
int i =0;
while(fgets(line, sizeof line, HMF_file) != NULL )
{
if(*line == '#') continue;
sscanf(line,"%lg %lg\n",&log10M_input[i],&hmf_input[i]);
log10hmf_in[i] = log10(1./log(10.)*hmf_input[i]);
// printf("%12.6e %12.6e \n", log10M_input[i],log10hmf_in[i]);
i += 1;
}
fclose(HMF_file);
hmf_accel_ptr = gsl_interp_accel_alloc();
hmf_spline_ptr = gsl_spline_alloc(gsl_interp_cspline,nlines);
gsl_spline_init(hmf_spline_ptr,log10M_input,log10hmf_in,nlines);
first = 0;
}
double log10M = log10(M*gb.h); //convert to Msun/h
double log10HMF = gsl_spline_eval(hmf_spline_ptr,log10M,hmf_accel_ptr); //in unit of (h/Mpc)^3
double HMF = pow(gb.h,3.)*pow(10.,log10HMF)/M; //convert to unit of (1/Mpc)^3
free(log10M_input);
free(hmf_input);
free(log10hmf_in);
return HMF;
}
/**
* computes the halo biases for three mass functions, press-schecter, Sheth-Tormen, and Tinker mass functions
*
* @param Cx Input: Cosmology structure
* @param M Input: halo mass
* @param z Input: redshift
* @param mode_mf Input: switch for setting the model of mass function, can be set to PSC, ST, TR
* @param bias_arr Output: the output array containning linear and quadratic local-in-matter halo biases, and quadratic and cubic tidal biases
* @return void
*/
void halo_bias(struct Cosmology *Cx, double M, double z, long mode_mf, double *bias_arr)
{
double delta_c = 1.686;
double R = R_scale(Cx, M);
double sigma = sqrt(sig_sq(Cx, z, R));
double nu = delta_c/sigma ;
double epsilon1 = 0., epsilon2 = 0., E1 = 0., E2 =0.;
double b1 =0., b2 =0., bg2=0., btd = 0.;
double alpha = 0., p = 0.;
double a =0., b =0., c=0.;
///Note that for PSC and ST mass functions, same form of the biases can be assumed, with different coefficents.
///See astro-ph/0006319
if(mode_mf == PSC){
epsilon1 = (pow(nu,2.)-1.)/delta_c;
epsilon2 = pow(nu/delta_c,2.)*(pow(nu,2.)-3.);
b1 = 1. + epsilon1;
b2 = 2. * (1.-17./21.) * epsilon1 + epsilon2;
}
else if(mode_mf == ST){
///Assuming spherical collapse
alpha = 0.707;
p = 0.3 ;
epsilon1 = (alpha*pow(nu,2.)-1.)/delta_c;
epsilon2 = alpha*pow(nu/delta_c,2.)*(alpha*pow(nu,2.)-3.);
E1 = (2.*p/delta_c)/(1.+pow(alpha*pow(nu,2.),p));
E2 = E1 * ((1.+2.*p)/delta_c + 2.*epsilon1);
b1 = 1. + epsilon1 + E1;
b2 = 2. * (1.-17./21.) * (epsilon1 + E1) + epsilon2 + E2;
////Assuming ellipsoidal collapse, astro-ph/9907024
// double a = 0.707;
// double b = 0.5;
// double c = 0.6;
// b1 = 1.+1./(sqrt(a) * delta_c) * (sqrt(a)*(a*pow(nu,2.)) + sqrt(a) *b * pow(a*pow(nu,2.),1.-c)\
// - pow(a*pow(nu,2.),c)/(pow(a*pow(nu,2.),c)+b*(1.-c)*(1.-c/2.)));
}
else if(mode_mf == TK){
//// Bias of Tinker et al 2008 arxiv: 0803.2706 for Delta =200
a = 0.707;
b = 0.35;
c = 0.80;
b1 = 1.+1./(sqrt(a)*delta_c)*(sqrt(a)*(a*pow(nu,2.))+sqrt(a)*b*pow(a*pow(nu,2.),1.-c)\
- pow(a*pow(nu,2.),c)/(pow(a*pow(nu,2.),c)+b*(1.-c)*(1.-c/2.)));
//// Bias of Tinker et al 2010 arxiv:1001.3162 for Delta =200
// double Delta = 200.;
// double y = log10(Delta);
// double A = 1. + 0.24 *y * exp(-pow(4./y,4.));
// double B = 0.183 ;
// double C = 0.019 + 0.107 *y + 0.19 *exp(-pow(4./y,4.));
// double aa = 0.44 * y - 0.88;
// double bb = 1.5;
// double cc = 2.4;
// b1 = 1. - A*pow(nu,aa)/(pow(nu,aa) + pow(delta_c,aa)) + B *pow(nu,bb) + C* pow(nu,cc);
}
bg2 = -2./7. * (b1-1.);
btd = 23./42. * (b1-1.);
bias_arr[0] = b1;
bias_arr[1] = b2;
bias_arr[2] = bg2;
bias_arr[3] = btd;
// printf("%12.6e %12.6e %12.6e %12.6e %12.6e %12.6e \n",nu,M*gb.h,b1,b2,bg2,btd);
return;
}
/**
* Read in the file for the star formation rate byy Behroozi et al 2013
*
* @param z_arr Output: pointer to an array of redshifts read from the file
* @param logM_arr Output: pointer to an array of halo masses read from the file
* @param log10SFR Output: pointer to an array of SFR read from the file
* @return void
*/
void logSFR_Behroozi_read(double *z_arr, double *logM_arr, double *log10SFR)
{
double result = 0.0;
long i, j, l, m;
FILE *ifp;
int verbose =1;
double M_min, M_max;
const gsl_interp2d_type *T = gsl_interp2d_bilinear;
size_t numlines = count_lines_in_file(gb.SFR_filename);
size_t num_z = 137 ;
size_t num_M = numlines/num_z;
double *zp1, *log10Mh, *log10SM;
zp1 = make_1Darray(numlines*sizeof(double));
log10Mh = make_1Darray(numlines*sizeof(double));
log10SM = make_1Darray(numlines*sizeof(double));
// Open the file
ifp = fopen(gb.SFR_filename,"r");
if(ifp == NULL)
{
printf("Failed to open the file");
exit(1);
}
//// Write a function that counts the columns of a file, read ead each line and parse the line by a given delimiter count_columns(filename, token) (return integer)
for(i=0L;i<numlines;i++){
fscanf(ifp,"%lg %lg %lg %lg \n", &zp1[i],&log10Mh[i],&log10SFR[i],&log10SM[i]);
// printf("%12.6e %12.5e %12.6e \n",zp1[i],log10Mh[i],log10SFR[i]);
}
gb.M_min = pow(10.,log10Mh[0]);
gb.M_max = pow(10.,log10Mh[numlines-1]);
gb.z_max = zp1[num_z-1] - 1;
//construct an array of unique values of log10Mh
j =0;
for(i=0L;i<numlines;i++){
if(log10Mh[i+1] != log10Mh[i]){
logM_arr[j] = log10Mh[i];
j += 1;
}
}
for(l=0;l<num_z;l++){
z_arr[l] = zp1[l] - 1. ;
}
// for(i=0;i<numlines;i++){
// if(logSFR_arr[i] == -1000)
// logSFR_arr[i] = 0.;
// else
// logSFR_arr[i] = log10SFR[i];
// }
free(zp1);
free(log10Mh);
free(log10SM);
fclose(ifp);
return;
}
/**
* Allocate memory and initialize the 2d interpolator for the star formation rate of Behroozi et al 2013 as a function of halo mass and redshift
*
* @return the error status
*/
int logSFR_alloc_init()
{
size_t numlines = count_lines_in_file(gb.SFR_filename);
size_t num_z = 137 ;
size_t num_M = numlines/num_z;
const gsl_interp2d_type *T = gsl_interp2d_bilinear;
gb.logM_accel_ptr = gsl_interp_accel_alloc();
gb.z_accel_ptr = gsl_interp_accel_alloc();
gb.logSFR_spline2d_ptr = gsl_spline2d_alloc(T, num_M, num_z);
double *z_arr = make_1Darray(num_z*sizeof(double));
double *logM_arr = make_1Darray(num_M*sizeof(double));
double *log10SFR = make_1Darray(numlines*sizeof(double));
double *logSFR_arr = make_1Darray(numlines*sizeof(double));
logSFR_Behroozi_read(z_arr,logM_arr,log10SFR);
int i = 0, l, m;
for(l=0;l<num_M;l++)
for(m=0;m<num_z;m++){
gsl_spline2d_set(gb.logSFR_spline2d_ptr, logSFR_arr, l, m, log10SFR[i]);
i += 1;
}
gsl_spline2d_init(gb.logSFR_spline2d_ptr, logM_arr, z_arr, logSFR_arr, num_M, num_z);
free(logM_arr);
free(z_arr);
free(logSFR_arr);
free(log10SFR);
return _SUCCESS_;
}
/**
* Free the memory allocated to the interpolators of star formation rate by Behroozi et al 2013
*
* @return the error status
*/
int SFR_Behroozi_free()
{
gsl_interp_accel_free(gb.logM_accel_ptr);
gsl_interp_accel_free(gb.z_accel_ptr);
gsl_spline2d_free(gb.logSFR_spline2d_ptr);
return _SUCCESS_;
}
/**
* Evaluate the SFR interpolator object for a given value of mass and redshift
*
* @param logM Input: log10 of halo mass
* @param z Input: redshift
* @return log10SFR
*/
double logSFR_Behroozi(double logM, double z)
{
double m1, m2, result = 0, eval;
if(pow(10.,logM)> gb.M_max || pow(10.,logM)<gb.M_min){
result = -1000.;
// printf("ERROR, choice of M is outside interpolation limits \n");
}
else if(pow(10.,logM)< gb.M_max && pow(10.,logM)>gb.M_min){
if(z <= gb.z_max)
eval = gsl_spline2d_eval(gb.logSFR_spline2d_ptr, logM, z, gb.logM_accel_ptr, gb.z_accel_ptr );
else if(z>gb.z_max){
m1 = gsl_spline2d_eval(gb.logSFR_spline2d_ptr, logM, gb.z_max, gb.logM_accel_ptr, gb.z_accel_ptr ) + 0.2943*(z-8.);
m2 = 3.3847-0.2413*z;
eval = GSL_MIN(m1,m2);
}
if(eval < 0. && fabs(eval) > 500.)
result = -1000.;
else
result = eval;
}
return result;
}
/**
* Compute the line specific luminosity in unit of solar luminosity
* For CO ladder, I am using the fits in Table 4 of Kamenetzky et al. arXiv:1508.05102, while for CII we use Silva et al arXiv:1410.4808
*
* @param M Input: halo mass
* @param z Input: redshift
* @param mode_lum Inpute: which luminosity model, basically which line considered
* @return line luminosity
*/
double luminosity(double M, double z, long mode_lum)
{
double f=0., L_CO = 0.;
double delta_MF=0. , nu_CO10=0. , L_IR=0. ,Lprime_CO=0.;
double a_CO = 0., b_CO = 0.;
double a_CII = 0., b_CII = 0.;
double logM = log10(M);
int J = 0;
if(mode_lum == CII){
//Foure models of Silva et al. We use M1
a_CII = 0.8475; ////M1
b_CII = 7.2203;
// a_LCII = 1.0000; ////M2
// b_LCII = 6.9647;
// a_LCII = 0.8727; ///M3
// b_LCII = 6.7250;
// a_LCII = 0.9231; ///M4
// b_LCII = 6.5234;
if(logSFR_Behroozi(logM, z) == -1000.)
f = 0.;
else
f = pow(10.,a_CII* logSFR_Behroozi(logM, z) + b_CII);
}
else{
delta_MF = 1.;
nu_CO10 = 115.27;
if(logSFR_Behroozi(logM, z) == -1000.)
L_IR = 0;
else
L_IR = pow(10.,logSFR_Behroozi(logM, z))/delta_MF * pow(10.,10.); //Kennicutt relation 1998
if(mode_lum == CO10){
a_CO = 1.27; /// a = 1.37 Charilli
b_CO =-1.0; /// b = -1.74
J = 1;
}
else if(mode_lum == CO21){
a_CO = 1.11;
b_CO = 0.6;
J = 2;
}
else if(mode_lum == CO32){
a_CO = 1.18;
b_CO = 0.1;
J = 3;
}
else if(mode_lum == CO43){
a_CO = 1.09;
b_CO = 1.2;
J =4;
}
else if(mode_lum == CO54){
a_CO = 1.05;
b_CO = 1.8;
J = 5;
}
else if(mode_lum == CO65){
a_CO = 1.04;
b_CO = 2.2;
J = 6;
}
Lprime_CO = pow(10.,(log10(L_IR)-b_CO)/a_CO); ///in unit of K km/s pc^2
L_CO = 4.9 * pow(10.,-5.) * pow(J,3.)* Lprime_CO; ///in unit of L_sun
f = L_CO;
}
return f;
}
/**
* The integrand function passed to hcubature integrator to compute the first moment of line luminosity
* @param nd Input: Dimensionality of the domain of integration
* @param x Input: integration variable
* @param p Input: integration parmaeters
* @param fdim Input: Dimensionality of the integrand function
* @param fvalue Input: Array of values of the integrand of dimension fdim
* return the error status
*/
int mass_moment1_integ(unsigned nd, // Number of dimensions in the domain space -- number of dim we're integrating over
const double *x, // The point at which the integrand is evaluated
void *p, // Pointer to a structure that holds the parameters
unsigned fdim, // Number of dimensions that the integrand return
double *fvalue // Array of values of the integrand of dimension fdim
)
{
double f = 0;
double result = 0;
double M = exp(x[0]);
struct integrand_parameters2 pij;
pij = *((struct integrand_parameters2 *)p);
struct Cosmology *Cx = pij.p1;
double z = pij.p4;
long mode_mf = pij.p13;
long mode_lum = pij.p14;
result = M * mass_func(Cx, M, z, mode_mf) * luminosity(M, z, mode_lum);
*fvalue = result;
return _SUCCESS_;
}
/**
* Compute the first luminosity moment.
*
* @param Cx Input: pointer to cosmology structure
* @param z Input: redshift
* @param M_min Input: minimum halo mass
* @param mode_mf Input: model of halo mass function to consider, PSC, ST, TR
* @param mode_lum Inpute: which luminosity model, basically which line considered
* @return the first mass moment
*/
double mass_moment1(struct Cosmology *Cx, double z, double M_min, long mode_mf, long mode_lum) /// in unit of M_sun/Mpc^3
{
struct integrand_parameters2 par;
unsigned fdim = 1; // Dimensionality of the integrand function
unsigned dim = 1; // Dimensionality of the domain of integration
double xmin[1], xmax[1]; // Integration limits: these are arrays of dimension dim
unsigned maxEval = 0; // Maximum number of integrand evaluations (0 for none)
double AbsErr = 0.0; // Required absolute error (0.0 for none)
double RelErr = 1.0e-2; // Required relative error (1.0e-3)
double result = 0.; // Final result
double error = 0.; // Error estimate on the result
error_norm norm = ERROR_INDIVIDUAL;
xmin[0] = log(M_min); ///In units of solar mass;
xmax[0] = log(1.e16); ///In units of solar mass
// xmin[0] = log(pow(10.,9.280699860662846135)/gb.h);
// xmax[0] = log(pow(10.,1.443569618680730571e+01)/gb.h);
par.p1 = Cx;
par.p4 = z;
par.p13 = mode_mf;
par.p14 = mode_lum;
hcubature(fdim, mass_moment1_integ, &par, dim, xmin, xmax, maxEval, AbsErr, RelErr, norm, &result, &error);
return result;
}
/**
* The integrand function passed to hcubature integrator to compute the second moment of line luminosity
* @param nd Input: Dimensionality of the domain of integration
* @param x Input: integration variable
* @param p Input: integration parmaeters
* @param fdim Input: Dimensionality of the integrand function
* @param fvalue Input: Array of values of the integrand of dimension fdim
* return the error status
*/
int mass_moment2_integ(unsigned nd, // Number of dimensions in the domain space -- number of dim we're integrating over
const double *x, // The point at which the integrand is evaluated
void *p, // Pointer to a structure that holds the parameters
unsigned fdim, // Number of dimensions that the integrand return
double *fvalue // Array of values of the integrand of dimension fdim
)
{
double f = 0;
double result = 0;
double M = exp(x[0]);
struct integrand_parameters2 pij;
pij = *((struct integrand_parameters2 *)p);
struct Cosmology *Cx = pij.p1;
double z = pij.p4;
long mode_mf = pij.p13;
long mode_lum = pij.p14;
result = M * mass_func(Cx, M, z, mode_mf)* pow(luminosity(M, z, mode_lum),2.);
*fvalue = result;
return _SUCCESS_;
}
/**
* Compute the second luminosity moment.
*
* @param Cx Input: pointer to cosmology structure
* @param z Input: redshift
* @param M_min Input: minimum halo mass
* @param mode_mf Input: model of halo mass function to consider, PSC, ST, TR
* @param mode_lum Inpute: which luminosity model, basically which line considered
* @return the second lum moment
*/
double mass_moment2(struct Cosmology *Cx, double z, double M_min, long mode_mf, long mode_lum) /// in unit of M_sun/Mpc^3
{
struct integrand_parameters2 par;
unsigned fdim = 1; // Dimensionality of the integrand function
unsigned dim = 1; // Dimensionality of the domain of integration
double xmin[1], xmax[1]; // Integration limits: these are arrays of dimension dim
unsigned maxEval = 0; // Maximum number of integrand evaluations (0 for none)
double AbsErr =0.0; // Required absolute error (0.0 for none)
double RelErr =1.0e-2; // Required relative error (1.0e-3)
double result =0.; // Final result
double error =0.; // Error estimate on the result
error_norm norm = ERROR_INDIVIDUAL;
xmin[0] = log(M_min); ///In units of solar mass;
xmax[0] = log(1.e16); ///In units of solar mass
// xmin[0] = log(pow(10.,9.280699860662846135)/gb.h);
// xmax[0] = log(pow(10.,1.443569618680730571e+01)/gb.h);
par.p1 = Cx;
par.p4 = z;
par.p13 = mode_mf;
par.p14 = mode_lum;
hcubature(fdim, mass_moment2_integ, &par, dim, xmin, xmax, maxEval, AbsErr, RelErr, norm, &result, &error);
return result;
}
/**
* The integrand function passed to hcubature integrator to compute the un-normalized luminosity-weighted line bias
* @param nd Input: Dimensionality of the domain of integration
* @param x Input: integration variable
* @param p Input: integration parmaeters
* @param fdim Input: Dimensionality of the integrand function
* @param fvalue Input: Array of values of the integrand of dimension fdim
* return the error status
*/
int bias_lum_weighted_integ(unsigned nd, // Number of dimensions in the domain space -- number of dim we're integrating over
const double *x, // The point at which the integrand is evaluated
void *p, // Pointer to a structure that holds the parameters
unsigned fdim, // Number of dimensions that the integrand return
double *fvalue // Array of values of the integrand of dimension fdim
)
{
double f = 0;
double result = 0;
double M = exp(x[0]);
struct integrand_parameters2 pij;
pij = *((struct integrand_parameters2 *)p);
struct Cosmology *Cx = pij.p1;
double z = pij.p4;
long mode_mf = pij.p13;
long mode_lum = pij.p14;
double MF = mass_func(Cx, M, z, mode_mf);
double lum = luminosity(M, z, mode_lum);
double *bias_arr = make_1Darray(4);
halo_bias(Cx, M, z, mode_mf, bias_arr);
double b1 = bias_arr[0];
double b2 = bias_arr[1];
fvalue[0] = M * MF * b1 * lum;
fvalue[1] = M * MF * b2 * lum;
return _SUCCESS_;
}
/**
* Compute the luminosity-weighted linear and quadratic line biases. The normalization of first mass moment is not included yet.
* The function bias_lum_weighted_integ() is the integrand and bias_lum_weighted() computes the bias
*
* @param Cx Input: pointer to cosmology structure
* @param z Input: redshift
* @param M_min Input: minimum halo mass
* @param mode_mf Input: model of halo mass function to consider, PSC, ST, TR
* @param mode_lum Inpute: which luminosity model, basically which line considered
* @param result Input: an output array of linear and quadratic line biases
* @return un-normalized line bias
*/
void bias_lum_weighted(struct Cosmology *Cx, double z, double M_min, long mode_mf, long mode_lum, double *result)
{
struct integrand_parameters2 par;
unsigned fdim = 2; // Dimensionality of the integrand function
unsigned dim = 1; // Dimensionality of the domain of integration
double xmin[1], xmax[1]; // Integration limits: these are arrays of dimension dim
unsigned maxEval = 0; // Maximum number of integrand evaluations (0 for none)
double AbsErr = 0.0; // Required absolute error (0.0 for none)
double RelErr = 1.0e-2; // Required relative error (1.0e-3)
double error[2]; // Error estimate on the result
error_norm norm = ERROR_INDIVIDUAL;
xmin[0] = log(M_min); ///In units of solar mass;
xmax[0] = log(1.e16); ///In units of solar mass
// xmin[0] = log(pow(10.,9.280699860662846135)/gb.h);
// xmax[0] = log(pow(10.,1.443569618680730571e+01)/gb.h);
par.p1 = Cx;
par.p4 = z;
par.p13 = mode_mf;
par.p14 = mode_lum;
hcubature(fdim, bias_lum_weighted_integ, &par, dim, xmin, xmax, maxEval, AbsErr, RelErr, norm, result, error);
return;
}
/**
* The integrand function passed to qags integrator to compute the scatter in shot ala Keating 2016
*
* @param x Input: integration variable
* @param par Input: integration parmaeters
* @return value of the integrand
*/
double p_sig_shot_integrand(double x, void *par)
{
double f=0;
double result = 0;
double scatter;
struct integrand_parameters2 pij;
pij = *((struct integrand_parameters2 *)par);
scatter = pij.p4 ;
result = pow(10.,2.*x)/sqrtl(2.*M_PI*pow(scatter,2.))* exp(-pow(x/scatter,2.)/2.);
return result;
}
/**
* Compute the scatter in shot noise due to scatter in luminosity-halo mass relation, assume log-normal scatter ala Keating
* Note that we set f_duty =1 unlike other LIM paper (ex. Lidz et al 2011).
*
* @param scatter Input: variance of the log-scatter
* @return the scatter coeff of shot
*/
double p_sig_shot(double scatter)
{
double result=0., error=0.;
gsl_integration_workspace *w = gsl_integration_workspace_alloc(1000000);
struct integrand_parameters2 par;
double xmin = -10.;
double xmax = 10.;
gsl_function F;
F.function = &p_sig_shot_integrand;
F.params = ∥
par.p4 = scatter;
gsl_integration_qags(&F,xmin,xmax,0.0,1.0e-2,1000000,w,&result,&error);
gsl_integration_workspace_free (w);
return result;
}
/**
* The integrand function passed to qags integrator to compute the scatter in Tbar ala Keating 2016
*
* @param x Input: integration variable
* @param par Input: integration parmaeters
* @return value of the integrand
*/
double p_sig_Tbar_integrand(double x, void *par)
{
double f=0;
double result = 0;
double scatter;
struct integrand_parameters2 pij;
pij = *((struct integrand_parameters2 *)par);
scatter = pij.p4 ;
result = pow(10.,x)/sqrtl(2.*M_PI*pow(scatter,2.))* exp(-pow(x/scatter,2.)/2.);
return result;
}
/**
* Compute the scatter in Tbar due to scatter in luminosity-halo mass relation, assume log-normal scatter ala Keating
* Note that we set f_duty =1 unlike other LIM paper (ex. Lidz et al 2011).
*
* @param scatter Input: variance of the log-scatter
* @return the scatter coeff of Tbar
*/
double p_sig_Tbar(double scatter)
{
double result=0., error=0.;
gsl_integration_workspace *w = gsl_integration_workspace_alloc(1000000);
struct integrand_parameters2 par;
double xmin = -10.;
double xmax = 10.;
gsl_function F;
F.function = &p_sig_Tbar_integrand;
F.params = ∥
par.p4 = scatter;
gsl_integration_qags(&F,xmin,xmax,0.0,1.0e-2,1000000,w,&result,&error);
gsl_integration_workspace_free (w);
return result;
}
/**
* Compute the linear and quadratic line biases, accounting ffor the normalization w.r.t. the first mass moment
*
* @param Lx Input: Pointer to line structure
* @param z Input: Redshift
* @param result Input: a pointer to an array containing the results of b1_line and b2_line
* @return void
*/
void line_bias(struct Line *Lx, double z, double *result)
{
double zz[4] = {z,DO_NOT_EVALUATE,z,z};
double res[4];
Line_evaluate(Lx, zz, res);
double mass_mom1 = res[0];
double b1_lumW = res[2];
double b2_lumW = res[3];
result[0] = b1_lumW/mass_mom1;
result[1] = b2_lumW/mass_mom1;
return;
}
/**
* Compute the line mean intensity in unit of erg Mpc^-2 Sr^-1
*
* @param Cx Input: Pointer to cosmology structure
* @param line_id Inpute: id of line of interest, an integer value
* @param z Input: Redshift
* @return the line mean intensity
*/
double mean_intens(struct Cosmology *Cx, size_t line_id, double z)
{
///Note: nu_J is the rest-frame emission frequency related to the observed frequency as nu_obs = nu_J/(1+z_J)
/// For a CO transition from J-> J-1, the rest-frame frequency is nu_J = J nu_CO where nu_Co = 115 GHz.
double nu_line = Cx->Lines[line_id]->line_freq;
double a = 1./(1.+z);
double L_sun = 3.846 * 1.e33; ///in unit of erg/s
double zz[4] = {z,DO_NOT_EVALUATE,DO_NOT_EVALUATE,DO_NOT_EVALUATE};
double res[4];
Line_evaluate(Cx->Lines[line_id], zz, res);
double mass_mom1 = res[0];
double y_tilde = gb.c/nu_line * pow(1.+z,2.)/Hubble(Cx,z);
double fac = 1./(4.*M_PI) * y_tilde * pow(1.+z,-2.)*L_sun ;
double f = fac * mass_mom1;
return f;
}
/**
* Compute the mean brightness temprature of CO in unit of microK, compared with Pullen et al and Lidz et al 2011
*
* @param Cx Input: Pointer to cosmology structure
* @param line_id Inpute: id of line of interest, an integer value
* @param z Input: Redshift
* @return the line mean temprature assuming Rayleigh-Jeans limit
*/
double Tbar_line(struct Cosmology *Cx, size_t line_id, double z)
{
double f = 0.;
double k_B = 1.38064852*1.e-16; ///Boltzmann constant in unit of erg K^-1
double fac = 3.086*1.e19; ////conversion factor from Mpc to km
double nu_line = Cx->Lines[line_id]->line_freq;
double sig_CO = 0.37;
double Tbar_scatter = p_sig_Tbar(sig_CO);
f = Tbar_scatter * pow(10.,6.)* pow(gb.c/nu_line,2.) *mean_intens(Cx,line_id,z) * pow(1.+z,2.)/(2.*k_B) * pow(fac,-2.); ///factor of 10^6 is the conversion factor from K to microK
return f;
}
|
DRB090-static-local-orig-yes.c | /*
Copyright (C) 1991-2018 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it andor
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http:www.gnu.org/licenses/>.
*/
/*
This header is separate from features.h so that the compiler can
include it implicitly at the start of every compilation. It must
not itself include <features.h> or any other header that includes
<features.h> because the implicit include comes before any feature
test macros that may be defined in a source file before it first
explicitly includes a system header. GCC knows the name of this
header in order to preinclude it.
*/
/*
glibc's intent is to support the IEC 559 math functionality, real
and complex. If the GCC (4.9 and later) predefined macros
specifying compiler intent are available, use them to determine
whether the overall intent is to support these features; otherwise,
presume an older compiler has intent to support these features and
define these macros by default.
*/
/*
wchar_t uses Unicode 10.0.0. Version 10.0 of the Unicode Standard is
synchronized with ISOIEC 10646:2017, fifth edition, plus
the following additions from Amendment 1 to the fifth edition:
- 56 emoji characters
- 285 hentaigana
- 3 additional Zanabazar Square characters
*/
/*
Copyright (c) 2017, Lawrence Livermore National Security, LLC.
Produced at the Lawrence Livermore National Laboratory
Written by Chunhua Liao, Pei-Hung Lin, Joshua Asplund,
Markus Schordan, and Ian Karlin
(email: liao6@llnl.gov, lin32@llnl.gov, asplund1@llnl.gov,
schordan1@llnl.gov, karlin1@llnl.gov)
LLNL-CODE-732144
All rights reserved.
This file is part of DataRaceBench. For details, see
https:github.comLLNL/dataracebench. Please also see the LICENSE file
for our additional BSD notice.
Redistribution and use in source and binary forms, with
or without modification, are permitted provided that the following
conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the disclaimer below.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the disclaimer (as noted below)
in the documentation and/or other materials provided with the
distribution.
* Neither the name of the LLNS/LLNL nor the names of its contributors
may be used to endorse or promote products derived from this
software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL LAWRENCE LIVERMORE NATIONAL
SECURITY, LLC, THE U.S. DEPARTMENT OF ENERGY OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
For a variable declared in a scope inside an OpenMP construct:
private if the variable has an automatic storage duration
* shared if the variable has a static storage duration.
Dependence pairs:
tmp@73:5 vs. tmp@73:5
tmp@73:5 vs. tmp@74:12
*/
#include<stdio.h>
int main(int argc, char * argv[])
{
int i;
int len = 100;
int a[len], b[len];
int _ret_val_0;
#pragma cetus private(i)
#pragma loop name main#0
#pragma cetus parallel
#pragma omp parallel for private(i)
for (i=0; i<len; i ++ )
{
a[i]=i;
b[i]=i;
}
/* static storage for a local variable */
{
static int tmp;
#pragma cetus private(i, tmp)
#pragma loop name main#1
#pragma cetus parallel
#pragma omp parallel for private(i, tmp)
for (i=0; i<len; i ++ )
{
tmp=(a[i]+i);
a[i]=tmp;
}
}
/* automatic storage for a local variable */
{
int tmp;
#pragma cetus private(i, tmp)
#pragma loop name main#2
#pragma cetus parallel
#pragma omp parallel for private(i, tmp)
for (i=0; i<len; i ++ )
{
tmp=(b[i]+i);
b[i]=tmp;
}
}
printf("a[50]=%d b[50]=%d\n", a[50], b[50]);
_ret_val_0=0;
return _ret_val_0;
}
|
boxloop_cuda.h | /******************************************************************************
* Copyright 1998-2019 Lawrence Livermore National Security, LLC and other
* HYPRE Project Developers. See the top-level COPYRIGHT file for details.
*
* SPDX-License-Identifier: (Apache-2.0 OR MIT)
******************************************************************************/
/******************************************************************************
*
* Header info for the BoxLoop
*
*****************************************************************************/
/*--------------------------------------------------------------------------
* BoxLoop macros:
*--------------------------------------------------------------------------*/
#ifndef HYPRE_NEWBOXLOOP_HEADER
#define HYPRE_NEWBOXLOOP_HEADER
#define HYPRE_LAMBDA [=] __host__ __device__
#define BLOCKSIZE 512
typedef struct hypre_Boxloop_struct
{
HYPRE_Int lsize0,lsize1,lsize2;
HYPRE_Int strides0,strides1,strides2;
HYPRE_Int bstart0,bstart1,bstart2;
HYPRE_Int bsize0,bsize1,bsize2;
} hypre_Boxloop;
#if 1
#define hypre_fence()
/*printf("\n hypre_newBoxLoop in %s(%d) function %s\n",__FILE__,__LINE__,__FUNCTION__);*/
#else
#define hypre_fence() \
{ \
cudaError err = cudaGetLastError(); \
if ( cudaSuccess != err ) \
{ \
printf("\n ERROR hypre_newBoxLoop: %s in %s(%d) function %s\n",cudaGetErrorString(err),__FILE__,__LINE__,__FUNCTION__); \
/* HYPRE_Int *p = NULL; *p = 1; */ \
} \
HYPRE_CUDA_CALL( cudaDeviceSynchronize() ); \
}
#endif
#ifdef __cplusplus
extern "C++" {
#endif
template <typename LOOP_BODY>
__global__ void forall_kernel(LOOP_BODY loop_body, HYPRE_Int length)
{
HYPRE_Int idx = blockDim.x * blockIdx.x + threadIdx.x;
if (idx < length)
{
loop_body(idx);
}
}
template<typename LOOP_BODY>
void BoxLoopforall(HYPRE_ExecutionPolicy policy, HYPRE_Int length, LOOP_BODY loop_body)
{
if (policy == HYPRE_EXEC_HOST)
{
#ifdef HYPRE_USING_OPENMP
#pragma omp parallel for HYPRE_SMP_SCHEDULE
#endif
for (HYPRE_Int idx = 0; idx < length; idx++)
{
loop_body(idx);
}
}
else if (policy == HYPRE_EXEC_DEVICE)
{
HYPRE_Int gridSize = (length + BLOCKSIZE - 1) / BLOCKSIZE;
const dim3 gDim(gridSize), bDim(BLOCKSIZE);
HYPRE_CUDA_LAUNCH( forall_kernel, gDim, bDim, loop_body, length );
}
}
template <typename LOOP_BODY>
__global__ void reductionforall_kernel(LOOP_BODY ReductionLoop,
HYPRE_Int length)
{
ReductionLoop(blockDim.x*blockIdx.x+threadIdx.x, blockDim.x*gridDim.x, length);
}
template<typename LOOP_BODY>
void ReductionBoxLoopforall(HYPRE_ExecutionPolicy policy, HYPRE_Int length, LOOP_BODY ReductionLoop)
{
if (length <= 0)
{
return;
}
if (policy == HYPRE_EXEC_HOST)
{
hypre_assert(0);
}
else if (policy == HYPRE_EXEC_DEVICE)
{
HYPRE_Int gridSize = (length + BLOCKSIZE - 1) / BLOCKSIZE;
gridSize = hypre_min(gridSize, 1024);
/*
hypre_printf("length= %d, blocksize = %d, gridsize = %d\n",
length, BLOCKSIZE, gridSize);
*/
const dim3 gDim(gridSize), bDim(BLOCKSIZE);
HYPRE_CUDA_LAUNCH( reductionforall_kernel, gDim, bDim, ReductionLoop, length );
}
}
#ifdef __cplusplus
}
#endif
#define hypre_BoxLoopIncK(k,box,hypre__i) \
HYPRE_Int hypre_boxD##k = 1; \
HYPRE_Int hypre__i = 0; \
hypre__i += (hypre_IndexD(local_idx, 0)*box.strides0 + box.bstart0) * hypre_boxD##k; \
hypre_boxD##k *= hypre_max(0, box.bsize0 + 1); \
hypre__i += (hypre_IndexD(local_idx, 1)*box.strides1 + box.bstart1) * hypre_boxD##k; \
hypre_boxD##k *= hypre_max(0, box.bsize1 + 1); \
hypre__i += (hypre_IndexD(local_idx, 2)*box.strides2 + box.bstart2) * hypre_boxD##k; \
hypre_boxD##k *= hypre_max(0, box.bsize2 + 1);
#define hypre_newBoxLoopInit(ndim,loop_size) \
HYPRE_Int hypre__tot = 1; \
for (HYPRE_Int hypre_d = 0;hypre_d < ndim;hypre_d ++) \
hypre__tot *= loop_size[hypre_d];
#define hypre_BasicBoxLoopInit(ndim,loop_size) \
HYPRE_Int hypre__tot = 1; \
for (HYPRE_Int hypre_d = 0;hypre_d < ndim;hypre_d ++) \
hypre__tot *= loop_size[hypre_d]; \
#define hypre_newBoxLoopDeclare(box) \
hypre_Index local_idx; \
HYPRE_Int idx_local = idx; \
hypre_IndexD(local_idx, 0) = idx_local % box.lsize0; \
idx_local = idx_local / box.lsize0; \
hypre_IndexD(local_idx, 1) = idx_local % box.lsize1; \
idx_local = idx_local / box.lsize1; \
hypre_IndexD(local_idx, 2) = idx_local % box.lsize2; \
#define hypre_newBoxLoop0Begin(ndim, loop_size) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{
#define hypre_newBoxLoop0End() \
}); \
hypre_fence(); \
}
#define hypre_BoxLoopDataDeclareK(k,ndim,loop_size,dbox,start,stride) \
hypre_Boxloop databox##k; \
databox##k.lsize0 = loop_size[0]; \
databox##k.strides0 = stride[0]; \
databox##k.bstart0 = start[0] - dbox->imin[0]; \
databox##k.bsize0 = dbox->imax[0]-dbox->imin[0]; \
if (ndim > 1) \
{ \
databox##k.lsize1 = loop_size[1]; \
databox##k.strides1 = stride[1]; \
databox##k.bstart1 = start[1] - dbox->imin[1]; \
databox##k.bsize1 = dbox->imax[1]-dbox->imin[1]; \
} \
else \
{ \
databox##k.lsize1 = 1; \
databox##k.strides1 = 0; \
databox##k.bstart1 = 0; \
databox##k.bsize1 = 0; \
} \
if (ndim == 3) \
{ \
databox##k.lsize2 = loop_size[2]; \
databox##k.strides2 = stride[2]; \
databox##k.bstart2 = start[2] - dbox->imin[2]; \
databox##k.bsize2 = dbox->imax[2]-dbox->imin[2]; \
} \
else \
{ \
databox##k.lsize2 = 1; \
databox##k.strides2 = 0; \
databox##k.bstart2 = 0; \
databox##k.bsize2 = 0; \
}
#define hypre_newBoxLoop1Begin(ndim, loop_size, \
dbox1, start1, stride1, i1) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1);
#define hypre_newBoxLoop1End(i1) \
}); \
hypre_fence(); \
}
#define hypre_newBoxLoop2Begin(ndim, loop_size, \
dbox1, start1, stride1, i1, \
dbox2, start2, stride2, i2) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
hypre_BoxLoopDataDeclareK(2,ndim,loop_size,dbox2,start2,stride2); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
hypre_BoxLoopIncK(2,databox2,i2);
#define hypre_newBoxLoop2End(i1, i2) \
}); \
hypre_fence(); \
}
#define hypre_newBoxLoop3Begin(ndim, loop_size, \
dbox1, start1, stride1, i1, \
dbox2, start2, stride2, i2, \
dbox3, start3, stride3, i3) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
hypre_BoxLoopDataDeclareK(2,ndim,loop_size,dbox2,start2,stride2); \
hypre_BoxLoopDataDeclareK(3,ndim,loop_size,dbox3,start3,stride3); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
hypre_BoxLoopIncK(2,databox2,i2); \
hypre_BoxLoopIncK(3,databox3,i3);
#define hypre_newBoxLoop3End(i1, i2,i3) \
}); \
hypre_fence(); \
}
#define hypre_newBoxLoop4Begin(ndim, loop_size, \
dbox1, start1, stride1, i1, \
dbox2, start2, stride2, i2, \
dbox3, start3, stride3, i3, \
dbox4, start4, stride4, i4) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
hypre_BoxLoopDataDeclareK(2,ndim,loop_size,dbox2,start2,stride2); \
hypre_BoxLoopDataDeclareK(3,ndim,loop_size,dbox3,start3,stride3); \
hypre_BoxLoopDataDeclareK(4,ndim,loop_size,dbox4,start4,stride4); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
hypre_BoxLoopIncK(2,databox2,i2); \
hypre_BoxLoopIncK(3,databox3,i3); \
hypre_BoxLoopIncK(4,databox4,i4);
#define hypre_newBoxLoop4End(i1, i2, i3, i4) \
}); \
hypre_fence(); \
}
#define zypre_BasicBoxLoopDataDeclareK(k,ndim,loop_size,stride) \
hypre_Boxloop databox##k; \
databox##k.lsize0 = loop_size[0]; \
databox##k.strides0 = stride[0]; \
databox##k.bstart0 = 0; \
databox##k.bsize0 = 0; \
if (ndim > 1) \
{ \
databox##k.lsize1 = loop_size[1]; \
databox##k.strides1 = stride[1]; \
databox##k.bstart1 = 0; \
databox##k.bsize1 = 0; \
} \
else \
{ \
databox##k.lsize1 = 1; \
databox##k.strides1 = 0; \
databox##k.bstart1 = 0; \
databox##k.bsize1 = 0; \
} \
if (ndim == 3) \
{ \
databox##k.lsize2 = loop_size[2]; \
databox##k.strides2 = stride[2]; \
databox##k.bstart2 = 0; \
databox##k.bsize2 = 0; \
} \
else \
{ \
databox##k.lsize2 = 1; \
databox##k.strides2 = 0; \
databox##k.bstart2 = 0; \
databox##k.bsize2 = 0; \
}
#define zypre_newBasicBoxLoop1Begin(ndim, loop_size, \
stride1, i1) \
{ \
hypre_BasicBoxLoopInit(ndim,loop_size); \
zypre_BasicBoxLoopDataDeclareK(1,ndim,loop_size,stride1); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
#define zypre_newBasicBoxLoop2Begin(ndim, loop_size, \
stride1, i1, \
stride2, i2) \
{ \
hypre_BasicBoxLoopInit(ndim,loop_size); \
zypre_BasicBoxLoopDataDeclareK(1,ndim,loop_size,stride1); \
zypre_BasicBoxLoopDataDeclareK(2,ndim,loop_size,stride2); \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),hypre__tot,HYPRE_LAMBDA (HYPRE_Int idx) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
hypre_BoxLoopIncK(2,databox2,i2); \
#define hypre_LoopBegin(size,idx) \
{ \
BoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()),size,HYPRE_LAMBDA (HYPRE_Int idx) \
{
#define hypre_LoopEnd() \
}); \
hypre_fence(); \
}
#define hypre_BoxLoopGetIndex(index) \
index[0] = hypre_IndexD(local_idx, 0); \
index[1] = hypre_IndexD(local_idx, 1); \
index[2] = hypre_IndexD(local_idx, 2);
#define hypre_BoxLoopBlock() 0
#define hypre_BoxLoop0Begin hypre_newBoxLoop0Begin
#define hypre_BoxLoop0For hypre_newBoxLoop0For
#define hypre_BoxLoop0End hypre_newBoxLoop0End
#define hypre_BoxLoop1Begin hypre_newBoxLoop1Begin
#define hypre_BoxLoop1For hypre_newBoxLoop1For
#define hypre_BoxLoop1End hypre_newBoxLoop1End
#define hypre_BoxLoop2Begin hypre_newBoxLoop2Begin
#define hypre_BoxLoop2For hypre_newBoxLoop2For
#define hypre_BoxLoop2End hypre_newBoxLoop2End
#define hypre_BoxLoop3Begin hypre_newBoxLoop3Begin
#define hypre_BoxLoop3For hypre_newBoxLoop3For
#define hypre_BoxLoop3End hypre_newBoxLoop3End
#define hypre_BoxLoop4Begin hypre_newBoxLoop4Begin
#define hypre_BoxLoop4For hypre_newBoxLoop4For
#define hypre_BoxLoop4End hypre_newBoxLoop4End
#define hypre_BasicBoxLoop1Begin zypre_newBasicBoxLoop1Begin
#define hypre_BasicBoxLoop2Begin zypre_newBasicBoxLoop2Begin
/* Reduction BoxLoop1*/
#define hypre_BoxLoop1ReductionBegin(ndim, loop_size, \
dbox1, start1, stride1, i1, \
reducesum) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
reducesum.nblocks = hypre_min( (hypre__tot+BLOCKSIZE-1)/BLOCKSIZE, 1024 ); \
ReductionBoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()), hypre__tot, \
HYPRE_LAMBDA (HYPRE_Int tid, HYPRE_Int nthreads, \
HYPRE_Int len) \
{ \
for (HYPRE_Int idx = tid; \
idx < len; \
idx += nthreads) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1);
#define hypre_BoxLoop1ReductionEnd(i1, reducesum) \
} \
reducesum.BlockReduce(); \
}); \
hypre_fence(); \
}
/* Reduction BoxLoop2 */
#define hypre_BoxLoop2ReductionBegin(ndim, loop_size, \
dbox1, start1, stride1, i1, \
dbox2, start2, stride2, i2, \
reducesum) \
{ \
hypre_newBoxLoopInit(ndim,loop_size); \
hypre_BoxLoopDataDeclareK(1,ndim,loop_size,dbox1,start1,stride1); \
hypre_BoxLoopDataDeclareK(2,ndim,loop_size,dbox2,start2,stride2); \
reducesum.nblocks = hypre_min( (hypre__tot+BLOCKSIZE-1)/BLOCKSIZE, 1024 ); \
ReductionBoxLoopforall(hypre_HandleStructExecPolicy(hypre_handle()), hypre__tot, \
HYPRE_LAMBDA (HYPRE_Int tid, HYPRE_Int nthreads, \
HYPRE_Int len) \
{ \
for (HYPRE_Int idx = tid; \
idx < len; \
idx += nthreads) \
{ \
hypre_newBoxLoopDeclare(databox1); \
hypre_BoxLoopIncK(1,databox1,i1); \
hypre_BoxLoopIncK(2,databox2,i2);
#define hypre_BoxLoop2ReductionEnd(i1, i2, reducesum) \
} \
reducesum.BlockReduce(); \
}); \
hypre_fence(); \
}
#endif
|
GB_unop__identity_uint16_fp32.c | //------------------------------------------------------------------------------
// GB_unop: hard-coded functions for each built-in unary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_atomics.h"
#include "GB_unop__include.h"
// C=unop(A) is defined by the following types and operators:
// op(A) function: GB (_unop_apply__identity_uint16_fp32)
// op(A') function: GB (_unop_tran__identity_uint16_fp32)
// C type: uint16_t
// A type: float
// cast: uint16_t cij = GB_cast_to_uint16_t ((double) (aij))
// unaryop: cij = aij
#define GB_ATYPE \
float
#define GB_CTYPE \
uint16_t
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
float aij = Ax [pA]
#define GB_CX(p) Cx [p]
// unary operator
#define GB_OP(z, x) \
z = x ;
// casting
#define GB_CAST(z, aij) \
uint16_t z = GB_cast_to_uint16_t ((double) (aij)) ;
// cij = op (aij)
#define GB_CAST_OP(pC,pA) \
{ \
/* aij = Ax [pA] */ \
float aij = Ax [pA] ; \
/* Cx [pC] = op (cast (aij)) */ \
uint16_t z = GB_cast_to_uint16_t ((double) (aij)) ; \
Cx [pC] = z ; \
}
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_IDENTITY || GxB_NO_UINT16 || GxB_NO_FP32)
//------------------------------------------------------------------------------
// Cx = op (cast (Ax)): apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_apply__identity_uint16_fp32)
(
uint16_t *Cx, // Cx and Ax may be aliased
const float *Ax,
const int8_t *restrict Ab, // A->b if A is bitmap
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
if (Ab == NULL)
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
float aij = Ax [p] ;
uint16_t z = GB_cast_to_uint16_t ((double) (aij)) ;
Cx [p] = z ;
}
}
else
{
// bitmap case, no transpose; A->b already memcpy'd into C->b
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!Ab [p]) continue ;
float aij = Ax [p] ;
uint16_t z = GB_cast_to_uint16_t ((double) (aij)) ;
Cx [p] = z ;
}
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (cast (A')): transpose, typecast, and apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_tran__identity_uint16_fp32)
(
GrB_Matrix C,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
driver.c | /*
Copyright (C) 2014, The University of Texas at Austin
This file is part of libflame and is available under the 3-Clause
BSD license, which can be found in the LICENSE file at the top-level
directory, or at http://opensource.org/licenses/BSD-3-Clause
*/
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include <time.h>
#include "omp.h"
#define ENABLE_OMP_THREADS 1
int main(int argc, char *argv[])
{
int n, nfirst, nlast, ninc, i, j,
nrepeats, nb_algi, rval;
double *Aref, *Ain, **Aout;
int **piv_buff, *piv_buff_ref;
double max_diff, avg_max_diff;
FILE *fpin = fopen("dgetrf.in", "rb");
FILE *fpout = fopen("dgetrf.out", "rb");
nfirst = 80;
nlast = 400;
ninc = 80;
nrepeats = 4;
for(n = nfirst; n <= nlast; n += ninc)
{
Ain = (double *) malloc(n * n * sizeof(double));
Aref = (double *) malloc(n * n * sizeof(double));
piv_buff_ref = (int *) malloc(n * sizeof(int));
piv_buff = (int **) malloc(nrepeats * sizeof(int *));
Aout = (double **) malloc(nrepeats * sizeof(double *));
for(i = 0; i < nrepeats; i++)
{
piv_buff[i] = (int *) malloc(n * sizeof(int));
Aout[i] = (double *) malloc(n * n * sizeof(double));
}
Aout[0][0] = 0;
/* read input and ref output (generate random) */
fread(Ain, sizeof(double), n * n, fpin);
fread(Aref, sizeof(double), n * n, fpout);
fread(piv_buff_ref, sizeof(int), n, fpout);
/* Initialize the output with input and send it to lapack function */
for(i = 0; i < nrepeats; i++)
{
for(j = 0; j < n * n; j++)
{
Aout[i][j] = Ain[j];
}
}
#if ENABLE_OMP_THREADS
#pragma omp parallel
#pragma omp for
#endif
for(i = 0; i < nrepeats; i++)
{
dgetrf_(&n, &n, Aout[i], &n, piv_buff[i], &rval);
}
avg_max_diff = 0;
for(i = 0; i < nrepeats; i++)
{
/* Max. Average difference */
max_diff = 0;
for(j = 0; j < n * n; j++)
{
max_diff = fmax(abs(Aout[i][j] - Aref[j]), max_diff);
}
avg_max_diff += (max_diff / nrepeats);
}
printf("Matrix Size:%dx%d\nMax Difference:%lf\n\n", n, n, avg_max_diff);
free(Ain);
free(Aref);
free(piv_buff);
for(i = 0; i < nrepeats; i++)
{
free(Aout[i]);
}
free(Aout);
}
fclose(fpin);
fclose(fpout);
}
|
composite.c | /*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% CCCC OOO M M PPPP OOO SSSSS IIIII TTTTT EEEEE %
% C O O MM MM P P O O SS I T E %
% C O O M M M PPPP O O SSS I T EEE %
% C O O M M P O O SS I T E %
% CCCC OOO M M P OOO SSSSS IIIII T EEEEE %
% %
% %
% MagickCore Image Composite Methods %
% %
% Software Design %
% Cristy %
% July 1992 %
% %
% %
% Copyright 1999-2018 ImageMagick Studio LLC, a non-profit organization %
% dedicated to making software imaging solutions freely available. %
% %
% You may not use this file except in compliance with the License. You may %
% obtain a copy of the License at %
% %
% https://imagemagick.org/script/license.php %
% %
% Unless required by applicable law or agreed to in writing, software %
% distributed under the License is distributed on an "AS IS" BASIS, %
% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. %
% See the License for the specific language governing permissions and %
% limitations under the License. %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%
%
*/
/*
Include declarations.
*/
#include "MagickCore/studio.h"
#include "MagickCore/artifact.h"
#include "MagickCore/cache.h"
#include "MagickCore/cache-private.h"
#include "MagickCore/cache-view.h"
#include "MagickCore/channel.h"
#include "MagickCore/client.h"
#include "MagickCore/color.h"
#include "MagickCore/color-private.h"
#include "MagickCore/colorspace.h"
#include "MagickCore/colorspace-private.h"
#include "MagickCore/composite.h"
#include "MagickCore/composite-private.h"
#include "MagickCore/constitute.h"
#include "MagickCore/draw.h"
#include "MagickCore/fx.h"
#include "MagickCore/gem.h"
#include "MagickCore/geometry.h"
#include "MagickCore/image.h"
#include "MagickCore/image-private.h"
#include "MagickCore/list.h"
#include "MagickCore/log.h"
#include "MagickCore/monitor.h"
#include "MagickCore/monitor-private.h"
#include "MagickCore/memory_.h"
#include "MagickCore/option.h"
#include "MagickCore/pixel-accessor.h"
#include "MagickCore/property.h"
#include "MagickCore/quantum.h"
#include "MagickCore/resample.h"
#include "MagickCore/resource_.h"
#include "MagickCore/string_.h"
#include "MagickCore/thread-private.h"
#include "MagickCore/threshold.h"
#include "MagickCore/token.h"
#include "MagickCore/utility.h"
#include "MagickCore/utility-private.h"
#include "MagickCore/version.h"
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% C o m p o s i t e I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% CompositeImage() returns the second image composited onto the first
% at the specified offset, using the specified composite method.
%
% The format of the CompositeImage method is:
%
% MagickBooleanType CompositeImage(Image *image,
% const Image *source_image,const CompositeOperator compose,
% const MagickBooleanType clip_to_self,const ssize_t x_offset,
% const ssize_t y_offset,ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the canvas image, modified by he composition
%
% o source_image: the source image.
%
% o compose: This operator affects how the composite is applied to
% the image. The operators and how they are utilized are listed here
% http://www.w3.org/TR/SVG12/#compositing.
%
% o clip_to_self: set to MagickTrue to limit composition to area composed.
%
% o x_offset: the column offset of the composited image.
%
% o y_offset: the row offset of the composited image.
%
% Extra Controls from Image meta-data in 'image' (artifacts)
%
% o "compose:args"
% A string containing extra numerical arguments for specific compose
% methods, generally expressed as a 'geometry' or a comma separated list
% of numbers.
%
% Compose methods needing such arguments include "BlendCompositeOp" and
% "DisplaceCompositeOp".
%
% o exception: return any errors or warnings in this structure.
%
*/
/*
Composition based on the SVG specification:
A Composition is defined by...
Color Function : f(Sc,Dc) where Sc and Dc are the normizalized colors
Blending areas : X = 1 for area of overlap, ie: f(Sc,Dc)
Y = 1 for source preserved
Z = 1 for canvas preserved
Conversion to transparency (then optimized)
Dca' = f(Sc, Dc)*Sa*Da + Y*Sca*(1-Da) + Z*Dca*(1-Sa)
Da' = X*Sa*Da + Y*Sa*(1-Da) + Z*Da*(1-Sa)
Where...
Sca = Sc*Sa normalized Source color divided by Source alpha
Dca = Dc*Da normalized Dest color divided by Dest alpha
Dc' = Dca'/Da' the desired color value for this channel.
Da' in in the follow formula as 'gamma' The resulting alpla value.
Most functions use a blending mode of over (X=1,Y=1,Z=1) this results in
the following optimizations...
gamma = Sa+Da-Sa*Da;
gamma = 1 - QuantumScale*alpha * QuantumScale*beta;
opacity = QuantumScale*alpha*beta; // over blend, optimized 1-Gamma
The above SVG definitions also define that Mathematical Composition
methods should use a 'Over' blending mode for Alpha Channel.
It however was not applied for composition modes of 'Plus', 'Minus',
the modulus versions of 'Add' and 'Subtract'.
Mathematical operator changes to be applied from IM v6.7...
1) Modulus modes 'Add' and 'Subtract' are obsoleted and renamed
'ModulusAdd' and 'ModulusSubtract' for clarity.
2) All mathematical compositions work as per the SVG specification
with regard to blending. This now includes 'ModulusAdd' and
'ModulusSubtract'.
3) When the special channel flag 'sync' (syncronize channel updates)
is turned off (enabled by default) then mathematical compositions are
only performed on the channels specified, and are applied
independantally of each other. In other words the mathematics is
performed as 'pure' mathematical operations, rather than as image
operations.
*/
static void HCLComposite(const MagickRealType hue,const MagickRealType chroma,
const MagickRealType luma,MagickRealType *red,MagickRealType *green,
MagickRealType *blue)
{
MagickRealType
b,
c,
g,
h,
m,
r,
x;
/*
Convert HCL to RGB colorspace.
*/
assert(red != (MagickRealType *) NULL);
assert(green != (MagickRealType *) NULL);
assert(blue != (MagickRealType *) NULL);
h=6.0*hue;
c=chroma;
x=c*(1.0-fabs(fmod(h,2.0)-1.0));
r=0.0;
g=0.0;
b=0.0;
if ((0.0 <= h) && (h < 1.0))
{
r=c;
g=x;
}
else
if ((1.0 <= h) && (h < 2.0))
{
r=x;
g=c;
}
else
if ((2.0 <= h) && (h < 3.0))
{
g=c;
b=x;
}
else
if ((3.0 <= h) && (h < 4.0))
{
g=x;
b=c;
}
else
if ((4.0 <= h) && (h < 5.0))
{
r=x;
b=c;
}
else
if ((5.0 <= h) && (h < 6.0))
{
r=c;
b=x;
}
m=luma-(0.298839*r+0.586811*g+0.114350*b);
*red=QuantumRange*(r+m);
*green=QuantumRange*(g+m);
*blue=QuantumRange*(b+m);
}
static void CompositeHCL(const MagickRealType red,const MagickRealType green,
const MagickRealType blue,MagickRealType *hue,MagickRealType *chroma,
MagickRealType *luma)
{
MagickRealType
b,
c,
g,
h,
max,
r;
/*
Convert RGB to HCL colorspace.
*/
assert(hue != (MagickRealType *) NULL);
assert(chroma != (MagickRealType *) NULL);
assert(luma != (MagickRealType *) NULL);
r=red;
g=green;
b=blue;
max=MagickMax(r,MagickMax(g,b));
c=max-(MagickRealType) MagickMin(r,MagickMin(g,b));
h=0.0;
if (c == 0)
h=0.0;
else
if (red == max)
h=fmod((g-b)/c+6.0,6.0);
else
if (green == max)
h=((b-r)/c)+2.0;
else
if (blue == max)
h=((r-g)/c)+4.0;
*hue=(h/6.0);
*chroma=QuantumScale*c;
*luma=QuantumScale*(0.298839*r+0.586811*g+0.114350*b);
}
static MagickBooleanType CompositeOverImage(Image *image,
const Image *source_image,const MagickBooleanType clip_to_self,
const ssize_t x_offset,const ssize_t y_offset,ExceptionInfo *exception)
{
#define CompositeImageTag "Composite/Image"
CacheView
*image_view,
*source_view;
const char
*value;
MagickBooleanType
clamp,
status;
MagickOffsetType
progress;
ssize_t
y;
/*
Composite image.
*/
status=MagickTrue;
progress=0;
clamp=MagickTrue;
value=GetImageArtifact(image,"compose:clamp");
if (value != (const char *) NULL)
clamp=IsStringTrue(value);
status=MagickTrue;
progress=0;
source_view=AcquireVirtualCacheView(source_image,exception);
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,image,image->rows,1)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
const Quantum
*pixels;
PixelInfo
canvas_pixel,
source_pixel;
register const Quantum
*magick_restrict p;
register Quantum
*magick_restrict q;
register ssize_t
x;
if (status == MagickFalse)
continue;
if (clip_to_self != MagickFalse)
{
if (y < y_offset)
continue;
if ((y-y_offset) >= (ssize_t) source_image->rows)
continue;
}
/*
If pixels is NULL, y is outside overlay region.
*/
pixels=(Quantum *) NULL;
p=(Quantum *) NULL;
if ((y >= y_offset) && ((y-y_offset) < (ssize_t) source_image->rows))
{
p=GetCacheViewVirtualPixels(source_view,0,y-y_offset,
source_image->columns,1,exception);
if (p == (const Quantum *) NULL)
{
status=MagickFalse;
continue;
}
pixels=p;
if (x_offset < 0)
p-=x_offset*GetPixelChannels(source_image);
}
q=GetCacheViewAuthenticPixels(image_view,0,y,image->columns,1,exception);
if (q == (Quantum *) NULL)
{
status=MagickFalse;
continue;
}
GetPixelInfo(image,&canvas_pixel);
GetPixelInfo(source_image,&source_pixel);
for (x=0; x < (ssize_t) image->columns; x++)
{
double
gamma;
MagickRealType
alpha,
Da,
Dc,
Dca,
Sa,
Sc,
Sca;
register ssize_t
i;
size_t
channels;
if (clip_to_self != MagickFalse)
{
if (x < x_offset)
{
q+=GetPixelChannels(image);
continue;
}
if ((x-x_offset) >= (ssize_t) source_image->columns)
break;
}
if ((pixels == (Quantum *) NULL) || (x < x_offset) ||
((x-x_offset) >= (ssize_t) source_image->columns))
{
Quantum
source[MaxPixelChannels];
/*
Virtual composite:
Sc: source color.
Dc: canvas color.
*/
(void) GetOneVirtualPixel(source_image,x-x_offset,y-y_offset,source,
exception);
for (i=0; i < (ssize_t) GetPixelChannels(image); i++)
{
MagickRealType
pixel;
PixelChannel channel = GetPixelChannelChannel(image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait source_traits=GetPixelChannelTraits(source_image,
channel);
if ((traits == UndefinedPixelTrait) ||
(source_traits == UndefinedPixelTrait))
continue;
if (channel == AlphaPixelChannel)
pixel=(MagickRealType) TransparentAlpha;
else
pixel=(MagickRealType) q[i];
q[i]=clamp != MagickFalse ? ClampPixel(pixel) :
ClampToQuantum(pixel);
}
q+=GetPixelChannels(image);
continue;
}
/*
Authentic composite:
Sa: normalized source alpha.
Da: normalized canvas alpha.
*/
Sa=QuantumScale*GetPixelAlpha(source_image,p);
Da=QuantumScale*GetPixelAlpha(image,q);
alpha=Sa+Da-Sa*Da;
for (i=0; i < (ssize_t) GetPixelChannels(image); i++)
{
MagickRealType
pixel;
PixelChannel channel = GetPixelChannelChannel(image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait source_traits=GetPixelChannelTraits(source_image,channel);
if (traits == UndefinedPixelTrait)
continue;
if ((source_traits == UndefinedPixelTrait) &&
(channel != AlphaPixelChannel))
continue;
if (channel == AlphaPixelChannel)
{
/*
Set alpha channel.
*/
pixel=QuantumRange*alpha;
q[i]=clamp != MagickFalse ? ClampPixel(pixel) :
ClampToQuantum(pixel);
continue;
}
/*
Sc: source color.
Dc: canvas color.
*/
Sc=(MagickRealType) GetPixelChannel(source_image,channel,p);
Dc=(MagickRealType) q[i];
if ((traits & CopyPixelTrait) != 0)
{
/*
Copy channel.
*/
q[i]=ClampToQuantum(Sc);
continue;
}
/*
Porter-Duff compositions:
Sca: source normalized color multiplied by alpha.
Dca: normalized canvas color multiplied by alpha.
*/
Sca=QuantumScale*Sa*Sc;
Dca=QuantumScale*Da*Dc;
gamma=PerceptibleReciprocal(alpha);
pixel=QuantumRange*gamma*(Sca+Dca*(1.0-Sa));
q[i]=clamp != MagickFalse ? ClampPixel(pixel) : ClampToQuantum(pixel);
}
p+=GetPixelChannels(source_image);
channels=GetPixelChannels(source_image);
if (p >= (pixels+channels*source_image->columns))
p=pixels;
q+=GetPixelChannels(image);
}
if (SyncCacheViewAuthenticPixels(image_view,exception) == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_CompositeImage)
#endif
proceed=SetImageProgress(image,CompositeImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
source_view=DestroyCacheView(source_view);
image_view=DestroyCacheView(image_view);
return(status);
}
MagickExport MagickBooleanType CompositeImage(Image *image,
const Image *composite,const CompositeOperator compose,
const MagickBooleanType clip_to_self,const ssize_t x_offset,
const ssize_t y_offset,ExceptionInfo *exception)
{
#define CompositeImageTag "Composite/Image"
CacheView
*source_view,
*image_view;
const char
*value;
GeometryInfo
geometry_info;
Image
*canvas_image,
*source_image;
MagickBooleanType
clamp,
status;
MagickOffsetType
progress;
MagickRealType
amount,
canvas_dissolve,
midpoint,
percent_luma,
percent_chroma,
source_dissolve,
threshold;
MagickStatusType
flags;
ssize_t
y;
assert(image != (Image *) NULL);
assert(image->signature == MagickCoreSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
assert(composite != (Image *) NULL);
assert(composite->signature == MagickCoreSignature);
if (SetImageStorageClass(image,DirectClass,exception) == MagickFalse)
return(MagickFalse);
source_image=CloneImage(composite,0,0,MagickTrue,exception);
if (source_image == (const Image *) NULL)
return(MagickFalse);
if (IsGrayColorspace(image->colorspace) != MagickFalse)
(void) SetImageColorspace(image,sRGBColorspace,exception);
(void) SetImageColorspace(source_image,image->colorspace,exception);
if ((compose == OverCompositeOp) || (compose == SrcOverCompositeOp))
{
status=CompositeOverImage(image,source_image,clip_to_self,x_offset,
y_offset,exception);
source_image=DestroyImage(source_image);
return(status);
}
amount=0.5;
canvas_image=(Image *) NULL;
canvas_dissolve=1.0;
clamp=MagickTrue;
value=GetImageArtifact(image,"compose:clamp");
if (value != (const char *) NULL)
clamp=IsStringTrue(value);
SetGeometryInfo(&geometry_info);
percent_luma=100.0;
percent_chroma=100.0;
source_dissolve=1.0;
threshold=0.05f;
switch (compose)
{
case CopyCompositeOp:
{
if ((x_offset < 0) || (y_offset < 0))
break;
if ((x_offset+(ssize_t) source_image->columns) > (ssize_t) image->columns)
break;
if ((y_offset+(ssize_t) source_image->rows) > (ssize_t) image->rows)
break;
status=MagickTrue;
source_view=AcquireVirtualCacheView(source_image,exception);
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(status) \
magick_number_threads(source_image,image,source_image->rows,1)
#endif
for (y=0; y < (ssize_t) source_image->rows; y++)
{
MagickBooleanType
sync;
register const Quantum
*p;
register Quantum
*q;
register ssize_t
x;
if (status == MagickFalse)
continue;
p=GetCacheViewVirtualPixels(source_view,0,y,source_image->columns,1,
exception);
q=GetCacheViewAuthenticPixels(image_view,x_offset,y+y_offset,
source_image->columns,1,exception);
if ((p == (const Quantum *) NULL) || (q == (Quantum *) NULL))
{
status=MagickFalse;
continue;
}
for (x=0; x < (ssize_t) source_image->columns; x++)
{
register ssize_t
i;
if (GetPixelReadMask(source_image,p) <= (QuantumRange/2))
{
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(image);
continue;
}
for (i=0; i < (ssize_t) GetPixelChannels(image); i++)
{
PixelChannel channel = GetPixelChannelChannel(image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait source_traits=GetPixelChannelTraits(source_image,
channel);
if (traits == UndefinedPixelTrait)
continue;
if (source_traits != UndefinedPixelTrait)
SetPixelChannel(image,channel,p[i],q);
else if (channel == AlphaPixelChannel)
SetPixelChannel(image,channel,OpaqueAlpha,q);
}
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(image);
}
sync=SyncCacheViewAuthenticPixels(image_view,exception);
if (sync == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_CompositeImage)
#endif
proceed=SetImageProgress(image,CompositeImageTag,
(MagickOffsetType) y,image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
source_view=DestroyCacheView(source_view);
image_view=DestroyCacheView(image_view);
source_image=DestroyImage(source_image);
return(status);
}
case IntensityCompositeOp:
{
if ((x_offset < 0) || (y_offset < 0))
break;
if ((x_offset+(ssize_t) source_image->columns) > (ssize_t) image->columns)
break;
if ((y_offset+(ssize_t) source_image->rows) > (ssize_t) image->rows)
break;
status=MagickTrue;
source_view=AcquireVirtualCacheView(source_image,exception);
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(status) \
magick_number_threads(source_image,image,source_image->rows,1)
#endif
for (y=0; y < (ssize_t) source_image->rows; y++)
{
MagickBooleanType
sync;
register const Quantum
*p;
register Quantum
*q;
register ssize_t
x;
if (status == MagickFalse)
continue;
p=GetCacheViewVirtualPixels(source_view,0,y,source_image->columns,1,
exception);
q=GetCacheViewAuthenticPixels(image_view,x_offset,y+y_offset,
source_image->columns,1,exception);
if ((p == (const Quantum *) NULL) || (q == (Quantum *) NULL))
{
status=MagickFalse;
continue;
}
for (x=0; x < (ssize_t) source_image->columns; x++)
{
if (GetPixelReadMask(source_image,p) <= (QuantumRange/2))
{
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(image);
continue;
}
SetPixelAlpha(image,clamp != MagickFalse ?
ClampPixel(GetPixelIntensity(source_image,p)) :
ClampToQuantum(GetPixelIntensity(source_image,p)),q);
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(image);
}
sync=SyncCacheViewAuthenticPixels(image_view,exception);
if (sync == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_CompositeImage)
#endif
proceed=SetImageProgress(image,CompositeImageTag,
(MagickOffsetType) y,image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
source_view=DestroyCacheView(source_view);
image_view=DestroyCacheView(image_view);
source_image=DestroyImage(source_image);
return(status);
}
case CopyAlphaCompositeOp:
case ChangeMaskCompositeOp:
{
/*
Modify canvas outside the overlaid region and require an alpha
channel to exist, to add transparency.
*/
if (image->alpha_trait == UndefinedPixelTrait)
(void) SetImageAlphaChannel(image,OpaqueAlphaChannel,exception);
break;
}
case BlurCompositeOp:
{
CacheView
*canvas_view;
MagickRealType
angle_range,
angle_start,
height,
width;
PixelInfo
pixel;
ResampleFilter
*resample_filter;
SegmentInfo
blur;
/*
Blur Image by resampling.
Blur Image dictated by an overlay gradient map: X = red_channel;
Y = green_channel; compose:args = x_scale[,y_scale[,angle]].
*/
canvas_image=CloneImage(image,0,0,MagickTrue,
exception);
if (canvas_image == (Image *) NULL)
{
source_image=DestroyImage(source_image);
return(MagickFalse);
}
/*
Gather the maximum blur sigma values from user.
*/
flags=NoValue;
value=GetImageArtifact(image,"compose:args");
if (value != (const char *) NULL)
flags=ParseGeometry(value,&geometry_info);
if ((flags & WidthValue) == 0)
{
(void) ThrowMagickException(exception,GetMagickModule(),OptionWarning,
"InvalidSetting","'%s' '%s'","compose:args",value);
source_image=DestroyImage(source_image);
canvas_image=DestroyImage(canvas_image);
return(MagickFalse);
}
/*
Users input sigma now needs to be converted to the EWA ellipse size.
The filter defaults to a sigma of 0.5 so to make this match the
users input the ellipse size needs to be doubled.
*/
width=height=geometry_info.rho*2.0;
if ((flags & HeightValue) != 0 )
height=geometry_info.sigma*2.0;
/*
Default the unrotated ellipse width and height axis vectors.
*/
blur.x1=width;
blur.x2=0.0;
blur.y1=0.0;
blur.y2=height;
/* rotate vectors if a rotation angle is given */
if ((flags & XValue) != 0 )
{
MagickRealType
angle;
angle=DegreesToRadians(geometry_info.xi);
blur.x1=width*cos(angle);
blur.x2=width*sin(angle);
blur.y1=(-height*sin(angle));
blur.y2=height*cos(angle);
}
/* Otherwise lets set a angle range and calculate in the loop */
angle_start=0.0;
angle_range=0.0;
if ((flags & YValue) != 0 )
{
angle_start=DegreesToRadians(geometry_info.xi);
angle_range=DegreesToRadians(geometry_info.psi)-angle_start;
}
/*
Set up a gaussian cylindrical filter for EWA Bluring.
As the minimum ellipse radius of support*1.0 the EWA algorithm
can only produce a minimum blur of 0.5 for Gaussian (support=2.0)
This means that even 'No Blur' will be still a little blurry!
The solution (as well as the problem of preventing any user
expert filter settings, is to set our own user settings, then
restore them afterwards.
*/
resample_filter=AcquireResampleFilter(image,exception);
SetResampleFilter(resample_filter,GaussianFilter);
/* do the variable blurring of each pixel in image */
GetPixelInfo(image,&pixel);
source_view=AcquireVirtualCacheView(source_image,exception);
canvas_view=AcquireAuthenticCacheView(canvas_image,exception);
for (y=0; y < (ssize_t) source_image->rows; y++)
{
MagickBooleanType
sync;
register const Quantum
*magick_restrict p;
register Quantum
*magick_restrict q;
register ssize_t
x;
if (((y+y_offset) < 0) || ((y+y_offset) >= (ssize_t) image->rows))
continue;
p=GetCacheViewVirtualPixels(source_view,0,y,source_image->columns,1,
exception);
q=QueueCacheViewAuthenticPixels(canvas_view,0,y,canvas_image->columns,1,
exception);
if ((p == (const Quantum *) NULL) || (q == (Quantum *) NULL))
break;
for (x=0; x < (ssize_t) source_image->columns; x++)
{
if (((x_offset+x) < 0) || ((x_offset+x) >= (ssize_t) image->columns))
{
p+=GetPixelChannels(source_image);
continue;
}
if (fabs((double) angle_range) > MagickEpsilon)
{
MagickRealType
angle;
angle=angle_start+angle_range*QuantumScale*
GetPixelBlue(source_image,p);
blur.x1=width*cos(angle);
blur.x2=width*sin(angle);
blur.y1=(-height*sin(angle));
blur.y2=height*cos(angle);
}
#if 0
if ( x == 10 && y == 60 ) {
(void) fprintf(stderr, "blur.x=%lf,%lf, blur.y=%lf,%lf\n",blur.x1,
blur.x2,blur.y1, blur.y2);
(void) fprintf(stderr, "scaled by=%lf,%lf\n",QuantumScale*
GetPixelRed(p),QuantumScale*GetPixelGreen(p));
#endif
ScaleResampleFilter(resample_filter,
blur.x1*QuantumScale*GetPixelRed(source_image,p),
blur.y1*QuantumScale*GetPixelGreen(source_image,p),
blur.x2*QuantumScale*GetPixelRed(source_image,p),
blur.y2*QuantumScale*GetPixelGreen(source_image,p) );
(void) ResamplePixelColor(resample_filter,(double) x_offset+x,
(double) y_offset+y,&pixel,exception);
SetPixelViaPixelInfo(canvas_image,&pixel,q);
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(canvas_image);
}
sync=SyncCacheViewAuthenticPixels(canvas_view,exception);
if (sync == MagickFalse)
break;
}
resample_filter=DestroyResampleFilter(resample_filter);
source_view=DestroyCacheView(source_view);
canvas_view=DestroyCacheView(canvas_view);
source_image=DestroyImage(source_image);
source_image=canvas_image;
break;
}
case DisplaceCompositeOp:
case DistortCompositeOp:
{
CacheView
*canvas_view;
MagickRealType
horizontal_scale,
vertical_scale;
PixelInfo
pixel;
PointInfo
center,
offset;
/*
Displace/Distort based on overlay gradient map:
X = red_channel; Y = green_channel;
compose:args = x_scale[,y_scale[,center.x,center.y]]
*/
canvas_image=CloneImage(image,0,0,MagickTrue,
exception);
if (canvas_image == (Image *) NULL)
{
source_image=DestroyImage(source_image);
return(MagickFalse);
}
SetGeometryInfo(&geometry_info);
flags=NoValue;
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
flags=ParseGeometry(value,&geometry_info);
if ((flags & (WidthValue | HeightValue)) == 0 )
{
if ((flags & AspectValue) == 0)
{
horizontal_scale=(MagickRealType) (source_image->columns-1)/2.0;
vertical_scale=(MagickRealType) (source_image->rows-1)/2.0;
}
else
{
horizontal_scale=(MagickRealType) (image->columns-1)/2.0;
vertical_scale=(MagickRealType) (image->rows-1)/2.0;
}
}
else
{
horizontal_scale=geometry_info.rho;
vertical_scale=geometry_info.sigma;
if ((flags & PercentValue) != 0)
{
if ((flags & AspectValue) == 0)
{
horizontal_scale*=(source_image->columns-1)/200.0;
vertical_scale*=(source_image->rows-1)/200.0;
}
else
{
horizontal_scale*=(image->columns-1)/200.0;
vertical_scale*=(image->rows-1)/200.0;
}
}
if ((flags & HeightValue) == 0)
vertical_scale=horizontal_scale;
}
/*
Determine fixed center point for absolute distortion map
Absolute distort ==
Displace offset relative to a fixed absolute point
Select that point according to +X+Y user inputs.
default = center of overlay image
arg flag '!' = locations/percentage relative to background image
*/
center.x=(MagickRealType) x_offset;
center.y=(MagickRealType) y_offset;
if (compose == DistortCompositeOp)
{
if ((flags & XValue) == 0)
if ((flags & AspectValue) != 0)
center.x=(MagickRealType) ((image->columns-1)/2.0);
else
center.x=(MagickRealType) (x_offset+(source_image->columns-1)/
2.0);
else
if ((flags & AspectValue) != 0)
center.x=geometry_info.xi;
else
center.x=(MagickRealType) (x_offset+geometry_info.xi);
if ((flags & YValue) == 0)
if ((flags & AspectValue) != 0)
center.y=(MagickRealType) ((image->rows-1)/2.0);
else
center.y=(MagickRealType) (y_offset+(source_image->rows-1)/2.0);
else
if ((flags & AspectValue) != 0)
center.y=geometry_info.psi;
else
center.y=(MagickRealType) (y_offset+geometry_info.psi);
}
/*
Shift the pixel offset point as defined by the provided,
displacement/distortion map. -- Like a lens...
*/
GetPixelInfo(image,&pixel);
image_view=AcquireVirtualCacheView(image,exception);
source_view=AcquireVirtualCacheView(source_image,exception);
canvas_view=AcquireAuthenticCacheView(canvas_image,exception);
for (y=0; y < (ssize_t) source_image->rows; y++)
{
MagickBooleanType
sync;
register const Quantum
*magick_restrict p;
register Quantum
*magick_restrict q;
register ssize_t
x;
if (((y+y_offset) < 0) || ((y+y_offset) >= (ssize_t) image->rows))
continue;
p=GetCacheViewVirtualPixels(source_view,0,y,source_image->columns,1,
exception);
q=QueueCacheViewAuthenticPixels(canvas_view,0,y,canvas_image->columns,1,
exception);
if ((p == (const Quantum *) NULL) || (q == (Quantum *) NULL))
break;
for (x=0; x < (ssize_t) source_image->columns; x++)
{
if (((x_offset+x) < 0) || ((x_offset+x) >= (ssize_t) image->columns))
{
p+=GetPixelChannels(source_image);
continue;
}
/*
Displace the offset.
*/
offset.x=(double) (horizontal_scale*(GetPixelRed(source_image,p)-
(((MagickRealType) QuantumRange+1.0)/2.0)))/(((MagickRealType)
QuantumRange+1.0)/2.0)+center.x+((compose == DisplaceCompositeOp) ?
x : 0);
offset.y=(double) (vertical_scale*(GetPixelGreen(source_image,p)-
(((MagickRealType) QuantumRange+1.0)/2.0)))/(((MagickRealType)
QuantumRange+1.0)/2.0)+center.y+((compose == DisplaceCompositeOp) ?
y : 0);
status=InterpolatePixelInfo(image,image_view,
UndefinedInterpolatePixel,(double) offset.x,(double) offset.y,
&pixel,exception);
if (status == MagickFalse)
break;
/*
Mask with the 'invalid pixel mask' in alpha channel.
*/
pixel.alpha=(MagickRealType) QuantumRange*(QuantumScale*pixel.alpha)*
(QuantumScale*GetPixelAlpha(source_image,p));
SetPixelViaPixelInfo(canvas_image,&pixel,q);
p+=GetPixelChannels(source_image);
q+=GetPixelChannels(canvas_image);
}
if (x < (ssize_t) source_image->columns)
break;
sync=SyncCacheViewAuthenticPixels(canvas_view,exception);
if (sync == MagickFalse)
break;
}
canvas_view=DestroyCacheView(canvas_view);
source_view=DestroyCacheView(source_view);
image_view=DestroyCacheView(image_view);
source_image=DestroyImage(source_image);
source_image=canvas_image;
break;
}
case DissolveCompositeOp:
{
/*
Geometry arguments to dissolve factors.
*/
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
{
flags=ParseGeometry(value,&geometry_info);
source_dissolve=geometry_info.rho/100.0;
canvas_dissolve=1.0;
if ((source_dissolve-MagickEpsilon) < 0.0)
source_dissolve=0.0;
if ((source_dissolve+MagickEpsilon) > 1.0)
{
canvas_dissolve=2.0-source_dissolve;
source_dissolve=1.0;
}
if ((flags & SigmaValue) != 0)
canvas_dissolve=geometry_info.sigma/100.0;
if ((canvas_dissolve-MagickEpsilon) < 0.0)
canvas_dissolve=0.0;
}
break;
}
case BlendCompositeOp:
{
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
{
flags=ParseGeometry(value,&geometry_info);
source_dissolve=geometry_info.rho/100.0;
canvas_dissolve=1.0-source_dissolve;
if ((flags & SigmaValue) != 0)
canvas_dissolve=geometry_info.sigma/100.0;
}
break;
}
case MathematicsCompositeOp:
{
/*
Just collect the values from "compose:args", setting.
Unused values are set to zero automagically.
Arguments are normally a comma separated list, so this probably should
be changed to some 'general comma list' parser, (with a minimum
number of values)
*/
SetGeometryInfo(&geometry_info);
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
(void) ParseGeometry(value,&geometry_info);
break;
}
case ModulateCompositeOp:
{
/*
Determine the luma and chroma scale.
*/
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
{
flags=ParseGeometry(value,&geometry_info);
percent_luma=geometry_info.rho;
if ((flags & SigmaValue) != 0)
percent_chroma=geometry_info.sigma;
}
break;
}
case ThresholdCompositeOp:
{
/*
Determine the amount and threshold.
*/
value=GetImageArtifact(image,"compose:args");
if (value != (char *) NULL)
{
flags=ParseGeometry(value,&geometry_info);
amount=geometry_info.rho;
threshold=geometry_info.sigma;
if ((flags & SigmaValue) == 0)
threshold=0.05f;
}
threshold*=QuantumRange;
break;
}
default:
break;
}
/*
Composite image.
*/
status=MagickTrue;
progress=0;
midpoint=((MagickRealType) QuantumRange+1.0)/2;
source_view=AcquireVirtualCacheView(source_image,exception);
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(progress,status) \
magick_number_threads(source_image,image,image->rows,1)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
const Quantum
*pixels;
MagickRealType
blue,
chroma,
green,
hue,
luma,
red;
PixelInfo
canvas_pixel,
source_pixel;
register const Quantum
*magick_restrict p;
register Quantum
*magick_restrict q;
register ssize_t
x;
if (status == MagickFalse)
continue;
if (clip_to_self != MagickFalse)
{
if (y < y_offset)
continue;
if ((y-y_offset) >= (ssize_t) source_image->rows)
continue;
}
/*
If pixels is NULL, y is outside overlay region.
*/
pixels=(Quantum *) NULL;
p=(Quantum *) NULL;
if ((y >= y_offset) && ((y-y_offset) < (ssize_t) source_image->rows))
{
p=GetCacheViewVirtualPixels(source_view,0,y-y_offset,
source_image->columns,1,exception);
if (p == (const Quantum *) NULL)
{
status=MagickFalse;
continue;
}
pixels=p;
if (x_offset < 0)
p-=x_offset*GetPixelChannels(source_image);
}
q=GetCacheViewAuthenticPixels(image_view,0,y,image->columns,1,exception);
if (q == (Quantum *) NULL)
{
status=MagickFalse;
continue;
}
hue=0.0;
chroma=0.0;
luma=0.0;
GetPixelInfo(image,&canvas_pixel);
GetPixelInfo(source_image,&source_pixel);
for (x=0; x < (ssize_t) image->columns; x++)
{
double
gamma;
MagickRealType
alpha,
Da,
Dc,
Dca,
DcaDa,
Sa,
SaSca,
Sc,
Sca;
register ssize_t
i;
size_t
channels;
if (clip_to_self != MagickFalse)
{
if (x < x_offset)
{
q+=GetPixelChannels(image);
continue;
}
if ((x-x_offset) >= (ssize_t) source_image->columns)
break;
}
if ((pixels == (Quantum *) NULL) || (x < x_offset) ||
((x-x_offset) >= (ssize_t) source_image->columns))
{
Quantum
source[MaxPixelChannels];
/*
Virtual composite:
Sc: source color.
Dc: canvas color.
*/
(void) GetOneVirtualPixel(source_image,x-x_offset,y-y_offset,source,
exception);
for (i=0; i < (ssize_t) GetPixelChannels(image); i++)
{
MagickRealType
pixel;
PixelChannel channel = GetPixelChannelChannel(image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait source_traits=GetPixelChannelTraits(source_image,
channel);
if ((traits == UndefinedPixelTrait) ||
(source_traits == UndefinedPixelTrait))
continue;
switch (compose)
{
case AlphaCompositeOp:
case ChangeMaskCompositeOp:
case CopyAlphaCompositeOp:
case DstAtopCompositeOp:
case DstInCompositeOp:
case InCompositeOp:
case OutCompositeOp:
case SrcInCompositeOp:
case SrcOutCompositeOp:
{
if (channel == AlphaPixelChannel)
pixel=(MagickRealType) TransparentAlpha;
else
pixel=(MagickRealType) q[i];
break;
}
case ClearCompositeOp:
case CopyCompositeOp:
case ReplaceCompositeOp:
case SrcCompositeOp:
{
if (channel == AlphaPixelChannel)
pixel=(MagickRealType) TransparentAlpha;
else
pixel=0.0;
break;
}
case BlendCompositeOp:
case DissolveCompositeOp:
{
if (channel == AlphaPixelChannel)
pixel=canvas_dissolve*GetPixelAlpha(source_image,source);
else
pixel=(MagickRealType) source[channel];
break;
}
default:
{
pixel=(MagickRealType) source[channel];
break;
}
}
q[i]=clamp != MagickFalse ? ClampPixel(pixel) :
ClampToQuantum(pixel);
}
q+=GetPixelChannels(image);
continue;
}
/*
Authentic composite:
Sa: normalized source alpha.
Da: normalized canvas alpha.
*/
Sa=QuantumScale*GetPixelAlpha(source_image,p);
Da=QuantumScale*GetPixelAlpha(image,q);
switch (compose)
{
case BumpmapCompositeOp:
{
alpha=GetPixelIntensity(source_image,p)*Sa;
break;
}
case ColorBurnCompositeOp:
case ColorDodgeCompositeOp:
case DarkenCompositeOp:
case DifferenceCompositeOp:
case DivideDstCompositeOp:
case DivideSrcCompositeOp:
case ExclusionCompositeOp:
case HardLightCompositeOp:
case HardMixCompositeOp:
case LinearBurnCompositeOp:
case LinearDodgeCompositeOp:
case LinearLightCompositeOp:
case LightenCompositeOp:
case MathematicsCompositeOp:
case MinusDstCompositeOp:
case MinusSrcCompositeOp:
case ModulusAddCompositeOp:
case ModulusSubtractCompositeOp:
case MultiplyCompositeOp:
case OverlayCompositeOp:
case PegtopLightCompositeOp:
case PinLightCompositeOp:
case ScreenCompositeOp:
case SoftLightCompositeOp:
case VividLightCompositeOp:
{
alpha=RoundToUnity(Sa+Da-Sa*Da);
break;
}
case DstAtopCompositeOp:
case DstInCompositeOp:
case InCompositeOp:
case SrcInCompositeOp:
{
alpha=Sa*Da;
break;
}
case DissolveCompositeOp:
{
alpha=source_dissolve*Sa*(-canvas_dissolve*Da)+source_dissolve*Sa+
canvas_dissolve*Da;
break;
}
case DstOverCompositeOp:
case OverCompositeOp:
case SrcOverCompositeOp:
{
alpha=Sa+Da-Sa*Da;
break;
}
case DstOutCompositeOp:
{
alpha=Da*(1.0-Sa);
break;
}
case OutCompositeOp:
case SrcOutCompositeOp:
{
alpha=Sa*(1.0-Da);
break;
}
case BlendCompositeOp:
case PlusCompositeOp:
{
alpha=RoundToUnity(source_dissolve*Sa+canvas_dissolve*Da);
break;
}
case XorCompositeOp:
{
alpha=Sa+Da-2.0*Sa*Da;
break;
}
default:
{
alpha=1.0;
break;
}
}
switch (compose)
{
case ColorizeCompositeOp:
case HueCompositeOp:
case LuminizeCompositeOp:
case ModulateCompositeOp:
case SaturateCompositeOp:
{
GetPixelInfoPixel(source_image,p,&source_pixel);
GetPixelInfoPixel(image,q,&canvas_pixel);
break;
}
default:
break;
}
for (i=0; i < (ssize_t) GetPixelChannels(image); i++)
{
MagickRealType
pixel,
sans;
PixelChannel channel = GetPixelChannelChannel(image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait source_traits = GetPixelChannelTraits(source_image,channel);
if (traits == UndefinedPixelTrait)
continue;
if ((channel == AlphaPixelChannel) &&
((traits & UpdatePixelTrait) != 0))
{
/*
Set alpha channel.
*/
switch (compose)
{
case AlphaCompositeOp:
{
pixel=QuantumRange*Sa;
break;
}
case AtopCompositeOp:
case CopyBlackCompositeOp:
case CopyBlueCompositeOp:
case CopyCyanCompositeOp:
case CopyGreenCompositeOp:
case CopyMagentaCompositeOp:
case CopyRedCompositeOp:
case CopyYellowCompositeOp:
case SrcAtopCompositeOp:
case DstCompositeOp:
case NoCompositeOp:
{
pixel=QuantumRange*Da;
break;
}
case ChangeMaskCompositeOp:
{
MagickBooleanType
equivalent;
if (Da < 0.5)
{
pixel=(MagickRealType) TransparentAlpha;
break;
}
equivalent=IsFuzzyEquivalencePixel(source_image,p,image,q);
if (equivalent != MagickFalse)
pixel=(MagickRealType) TransparentAlpha;
else
pixel=(MagickRealType) OpaqueAlpha;
break;
}
case ClearCompositeOp:
{
pixel=(MagickRealType) TransparentAlpha;
break;
}
case ColorizeCompositeOp:
case HueCompositeOp:
case LuminizeCompositeOp:
case SaturateCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=QuantumRange*Da;
break;
}
if (fabs((double) (QuantumRange*Da-TransparentAlpha)) < MagickEpsilon)
{
pixel=QuantumRange*Sa;
break;
}
if (Sa < Da)
{
pixel=QuantumRange*Da;
break;
}
pixel=QuantumRange*Sa;
break;
}
case CopyAlphaCompositeOp:
{
if (source_image->alpha_trait == UndefinedPixelTrait)
pixel=GetPixelIntensity(source_image,p);
else
pixel=QuantumRange*Sa;
break;
}
case CopyCompositeOp:
case DisplaceCompositeOp:
case DistortCompositeOp:
case DstAtopCompositeOp:
case ReplaceCompositeOp:
case SrcCompositeOp:
{
pixel=QuantumRange*Sa;
break;
}
case DarkenIntensityCompositeOp:
{
pixel=Sa*GetPixelIntensity(source_image,p) <
Da*GetPixelIntensity(image,q) ? Sa : Da;
break;
}
case DifferenceCompositeOp:
{
pixel=QuantumRange*fabs(Sa-Da);
break;
}
case LightenIntensityCompositeOp:
{
pixel=Sa*GetPixelIntensity(source_image,p) >
Da*GetPixelIntensity(image,q) ? Sa : Da;
break;
}
case ModulateCompositeOp:
{
pixel=QuantumRange*Da;
break;
}
case MultiplyCompositeOp:
{
pixel=QuantumRange*Sa*Da;
break;
}
case StereoCompositeOp:
{
pixel=QuantumRange*(Sa+Da)/2;
break;
}
default:
{
pixel=QuantumRange*alpha;
break;
}
}
q[i]=clamp != MagickFalse ? ClampPixel(pixel) :
ClampToQuantum(pixel);
continue;
}
if (source_traits == UndefinedPixelTrait)
continue;
/*
Sc: source color.
Dc: canvas color.
*/
Sc=(MagickRealType) GetPixelChannel(source_image,channel,p);
Dc=(MagickRealType) q[i];
if ((traits & CopyPixelTrait) != 0)
{
/*
Copy channel.
*/
q[i]=ClampToQuantum(Dc);
continue;
}
/*
Porter-Duff compositions:
Sca: source normalized color multiplied by alpha.
Dca: normalized canvas color multiplied by alpha.
*/
Sca=QuantumScale*Sa*Sc;
Dca=QuantumScale*Da*Dc;
SaSca=Sa*PerceptibleReciprocal(Sca);
DcaDa=Dca*PerceptibleReciprocal(Da);
switch (compose)
{
case DarkenCompositeOp:
case LightenCompositeOp:
case ModulusSubtractCompositeOp:
{
gamma=PerceptibleReciprocal(1.0-alpha);
break;
}
default:
{
gamma=PerceptibleReciprocal(alpha);
break;
}
}
pixel=Dc;
switch (compose)
{
case AlphaCompositeOp:
{
pixel=QuantumRange*Sa;
break;
}
case AtopCompositeOp:
case SrcAtopCompositeOp:
{
pixel=QuantumRange*(Sca*Da+Dca*(1.0-Sa));
break;
}
case BlendCompositeOp:
{
pixel=gamma*(source_dissolve*Sa*Sc+canvas_dissolve*Da*Dc);
break;
}
case BlurCompositeOp:
case CopyCompositeOp:
case ReplaceCompositeOp:
case SrcCompositeOp:
{
pixel=QuantumRange*Sca;
break;
}
case DisplaceCompositeOp:
case DistortCompositeOp:
{
pixel=Sc;
break;
}
case BumpmapCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
pixel=QuantumScale*GetPixelIntensity(source_image,p)*Dc;
break;
}
case ChangeMaskCompositeOp:
{
pixel=Dc;
break;
}
case ClearCompositeOp:
{
pixel=0.0;
break;
}
case ColorBurnCompositeOp:
{
if ((Sca == 0.0) && (Dca == Da))
{
pixel=QuantumRange*gamma*(Sa*Da+Dca*(1.0-Sa));
break;
}
if (Sca == 0.0)
{
pixel=QuantumRange*gamma*(Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*gamma*(Sa*Da-Sa*Da*MagickMin(1.0,(1.0-DcaDa)*
SaSca)+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case ColorDodgeCompositeOp:
{
if ((Sca*Da+Dca*Sa) >= Sa*Da)
pixel=QuantumRange*gamma*(Sa*Da+Sca*(1.0-Da)+Dca*(1.0-Sa));
else
pixel=QuantumRange*gamma*(Dca*Sa*Sa*PerceptibleReciprocal(Sa-Sca)+
Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case ColorizeCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
if (fabs((double) (QuantumRange*Da-TransparentAlpha)) < MagickEpsilon)
{
pixel=Sc;
break;
}
CompositeHCL(canvas_pixel.red,canvas_pixel.green,canvas_pixel.blue,
&sans,&sans,&luma);
CompositeHCL(source_pixel.red,source_pixel.green,source_pixel.blue,
&hue,&chroma,&sans);
HCLComposite(hue,chroma,luma,&red,&green,&blue);
switch (channel)
{
case RedPixelChannel: pixel=red; break;
case GreenPixelChannel: pixel=green; break;
case BluePixelChannel: pixel=blue; break;
default: pixel=Dc; break;
}
break;
}
case CopyAlphaCompositeOp:
{
pixel=Dc;
break;
}
case CopyBlackCompositeOp:
{
if (channel == BlackPixelChannel)
pixel=(MagickRealType) (QuantumRange-
GetPixelBlack(source_image,p));
break;
}
case CopyBlueCompositeOp:
case CopyYellowCompositeOp:
{
if (channel == BluePixelChannel)
pixel=(MagickRealType) GetPixelBlue(source_image,p);
break;
}
case CopyGreenCompositeOp:
case CopyMagentaCompositeOp:
{
if (channel == GreenPixelChannel)
pixel=(MagickRealType) GetPixelGreen(source_image,p);
break;
}
case CopyRedCompositeOp:
case CopyCyanCompositeOp:
{
if (channel == RedPixelChannel)
pixel=(MagickRealType) GetPixelRed(source_image,p);
break;
}
case DarkenCompositeOp:
{
/*
Darken is equivalent to a 'Minimum' method
OR a greyscale version of a binary 'Or'
OR the 'Intersection' of pixel sets.
*/
if ((Sca*Da) < (Dca*Sa))
{
pixel=QuantumRange*(Sca+Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*(Dca+Sca*(1.0-Da));
break;
}
case DarkenIntensityCompositeOp:
{
pixel=Sa*GetPixelIntensity(source_image,p) <
Da*GetPixelIntensity(image,q) ? Sc : Dc;
break;
}
case DifferenceCompositeOp:
{
pixel=QuantumRange*gamma*(Sca+Dca-2.0*MagickMin(Sca*Da,Dca*Sa));
break;
}
case DissolveCompositeOp:
{
pixel=gamma*(source_dissolve*Sa*Sc-source_dissolve*Sa*
canvas_dissolve*Da*Dc+canvas_dissolve*Da*Dc);
break;
}
case DivideDstCompositeOp:
{
if ((fabs((double) Sca) < MagickEpsilon) &&
(fabs((double) Dca) < MagickEpsilon))
{
pixel=QuantumRange*gamma*(Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
if (fabs((double) Dca) < MagickEpsilon)
{
pixel=QuantumRange*gamma*(Sa*Da+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*gamma*(Sca*Da*Da/Dca+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case DivideSrcCompositeOp:
{
if ((fabs((double) Dca) < MagickEpsilon) &&
(fabs((double) Sca) < MagickEpsilon))
{
pixel=QuantumRange*gamma*(Dca*(1.0-Sa)+Sca*(1.0-Da));
break;
}
if (fabs((double) Sca) < MagickEpsilon)
{
pixel=QuantumRange*gamma*(Da*Sa+Dca*(1.0-Sa)+Sca*(1.0-Da));
break;
}
pixel=QuantumRange*gamma*(Dca*Sa*SaSca+Dca*(1.0-Sa)+Sca*(1.0-Da));
break;
}
case DstAtopCompositeOp:
{
pixel=QuantumRange*(Dca*Sa+Sca*(1.0-Da));
break;
}
case DstCompositeOp:
case NoCompositeOp:
{
pixel=QuantumRange*Dca;
break;
}
case DstInCompositeOp:
{
pixel=QuantumRange*(Dca*Sa);
break;
}
case DstOutCompositeOp:
{
pixel=QuantumRange*(Dca*(1.0-Sa));
break;
}
case DstOverCompositeOp:
{
pixel=QuantumRange*gamma*(Dca+Sca*(1.0-Da));
break;
}
case ExclusionCompositeOp:
{
pixel=QuantumRange*gamma*(Sca*Da+Dca*Sa-2.0*Sca*Dca+Sca*(1.0-Da)+
Dca*(1.0-Sa));
break;
}
case HardLightCompositeOp:
{
if ((2.0*Sca) < Sa)
{
pixel=QuantumRange*gamma*(2.0*Sca*Dca+Sca*(1.0-Da)+Dca*(1.0-
Sa));
break;
}
pixel=QuantumRange*gamma*(Sa*Da-2.0*(Da-Dca)*(Sa-Sca)+Sca*(1.0-Da)+
Dca*(1.0-Sa));
break;
}
case HardMixCompositeOp:
{
pixel=gamma*(((Sca+Dca) < 1.0) ? 0.0 : QuantumRange);
break;
}
case HueCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
if (fabs((double) (QuantumRange*Da-TransparentAlpha)) < MagickEpsilon)
{
pixel=Sc;
break;
}
CompositeHCL(canvas_pixel.red,canvas_pixel.green,canvas_pixel.blue,
&hue,&chroma,&luma);
CompositeHCL(source_pixel.red,source_pixel.green,source_pixel.blue,
&hue,&sans,&sans);
HCLComposite(hue,chroma,luma,&red,&green,&blue);
switch (channel)
{
case RedPixelChannel: pixel=red; break;
case GreenPixelChannel: pixel=green; break;
case BluePixelChannel: pixel=blue; break;
default: pixel=Dc; break;
}
break;
}
case InCompositeOp:
case SrcInCompositeOp:
{
pixel=QuantumRange*(Sca*Da);
break;
}
case LinearBurnCompositeOp:
{
/*
LinearBurn: as defined by Abode Photoshop, according to
http://www.simplefilter.de/en/basics/mixmods.html is:
f(Sc,Dc) = Sc + Dc - 1
*/
pixel=QuantumRange*gamma*(Sca+Dca-Sa*Da);
break;
}
case LinearDodgeCompositeOp:
{
pixel=gamma*(Sa*Sc+Da*Dc);
break;
}
case LinearLightCompositeOp:
{
/*
LinearLight: as defined by Abode Photoshop, according to
http://www.simplefilter.de/en/basics/mixmods.html is:
f(Sc,Dc) = Dc + 2*Sc - 1
*/
pixel=QuantumRange*gamma*((Sca-Sa)*Da+Sca+Dca);
break;
}
case LightenCompositeOp:
{
if ((Sca*Da) > (Dca*Sa))
{
pixel=QuantumRange*(Sca+Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*(Dca+Sca*(1.0-Da));
break;
}
case LightenIntensityCompositeOp:
{
/*
Lighten is equivalent to a 'Maximum' method
OR a greyscale version of a binary 'And'
OR the 'Union' of pixel sets.
*/
pixel=Sa*GetPixelIntensity(source_image,p) >
Da*GetPixelIntensity(image,q) ? Sc : Dc;
break;
}
case LuminizeCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
if (fabs((double) (QuantumRange*Da-TransparentAlpha)) < MagickEpsilon)
{
pixel=Sc;
break;
}
CompositeHCL(canvas_pixel.red,canvas_pixel.green,canvas_pixel.blue,
&hue,&chroma,&luma);
CompositeHCL(source_pixel.red,source_pixel.green,source_pixel.blue,
&sans,&sans,&luma);
HCLComposite(hue,chroma,luma,&red,&green,&blue);
switch (channel)
{
case RedPixelChannel: pixel=red; break;
case GreenPixelChannel: pixel=green; break;
case BluePixelChannel: pixel=blue; break;
default: pixel=Dc; break;
}
break;
}
case MathematicsCompositeOp:
{
/*
'Mathematics' a free form user control mathematical composition
is defined as...
f(Sc,Dc) = A*Sc*Dc + B*Sc + C*Dc + D
Where the arguments A,B,C,D are (currently) passed to composite
as a command separated 'geometry' string in "compose:args" image
artifact.
A = a->rho, B = a->sigma, C = a->xi, D = a->psi
Applying the SVG transparency formula (see above), we get...
Dca' = Sa*Da*f(Sc,Dc) + Sca*(1.0-Da) + Dca*(1.0-Sa)
Dca' = A*Sca*Dca + B*Sca*Da + C*Dca*Sa + D*Sa*Da + Sca*(1.0-Da) +
Dca*(1.0-Sa)
*/
pixel=QuantumRange*gamma*(geometry_info.rho*Sca*Dca+
geometry_info.sigma*Sca*Da+geometry_info.xi*Dca*Sa+
geometry_info.psi*Sa*Da+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case MinusDstCompositeOp:
{
pixel=gamma*(Sa*Sc+Da*Dc-2.0*Da*Dc*Sa);
break;
}
case MinusSrcCompositeOp:
{
/*
Minus source from canvas.
f(Sc,Dc) = Sc - Dc
*/
pixel=gamma*(Da*Dc+Sa*Sc-2.0*Sa*Sc*Da);
break;
}
case ModulateCompositeOp:
{
ssize_t
offset;
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
offset=(ssize_t) (GetPixelIntensity(source_image,p)-midpoint);
if (offset == 0)
{
pixel=Dc;
break;
}
CompositeHCL(canvas_pixel.red,canvas_pixel.green,canvas_pixel.blue,
&hue,&chroma,&luma);
luma+=(0.01*percent_luma*offset)/midpoint;
chroma*=0.01*percent_chroma;
HCLComposite(hue,chroma,luma,&red,&green,&blue);
switch (channel)
{
case RedPixelChannel: pixel=red; break;
case GreenPixelChannel: pixel=green; break;
case BluePixelChannel: pixel=blue; break;
default: pixel=Dc; break;
}
break;
}
case ModulusAddCompositeOp:
{
pixel=Sc+Dc;
while (pixel > QuantumRange)
pixel-=QuantumRange;
while (pixel < 0.0)
pixel+=QuantumRange;
pixel=(Sa*Da*pixel+Sa*Sc*(1.0-Da)+Da*Dc*(1.0-Sa));
break;
}
case ModulusSubtractCompositeOp:
{
pixel=Sc-Dc;
while (pixel > QuantumRange)
pixel-=QuantumRange;
while (pixel < 0.0)
pixel+=QuantumRange;
pixel=(Sa*Da*pixel+Sa*Sc*(1.0-Da)+Da*Dc*(1.0-Sa));
break;
}
case MultiplyCompositeOp:
{
pixel=QuantumRange*gamma*(Sca*Dca+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case OutCompositeOp:
case SrcOutCompositeOp:
{
pixel=QuantumRange*(Sca*(1.0-Da));
break;
}
case OverCompositeOp:
case SrcOverCompositeOp:
{
pixel=QuantumRange*gamma*(Sca+Dca*(1.0-Sa));
break;
}
case OverlayCompositeOp:
{
if ((2.0*Dca) < Da)
{
pixel=QuantumRange*gamma*(2.0*Dca*Sca+Dca*(1.0-Sa)+Sca*(1.0-
Da));
break;
}
pixel=QuantumRange*gamma*(Da*Sa-2.0*(Sa-Sca)*(Da-Dca)+Dca*(1.0-Sa)+
Sca*(1.0-Da));
break;
}
case PegtopLightCompositeOp:
{
/*
PegTop: A Soft-Light alternative: A continuous version of the
Softlight function, producing very similar results.
f(Sc,Dc) = Dc^2*(1-2*Sc) + 2*Sc*Dc
http://www.pegtop.net/delphi/articles/blendmodes/softlight.htm.
*/
if (fabs((double) Da) < MagickEpsilon)
{
pixel=QuantumRange*gamma*(Sca);
break;
}
pixel=QuantumRange*gamma*(Dca*Dca*(Sa-2.0*Sca)/Da+Sca*(2.0*Dca+1.0-
Da)+Dca*(1.0-Sa));
break;
}
case PinLightCompositeOp:
{
/*
PinLight: A Photoshop 7 composition method
http://www.simplefilter.de/en/basics/mixmods.html
f(Sc,Dc) = Dc<2*Sc-1 ? 2*Sc-1 : Dc>2*Sc ? 2*Sc : Dc
*/
if ((Dca*Sa) < (Da*(2.0*Sca-Sa)))
{
pixel=QuantumRange*gamma*(Sca*(Da+1.0)-Sa*Da+Dca*(1.0-Sa));
break;
}
if ((Dca*Sa) > (2.0*Sca*Da))
{
pixel=QuantumRange*gamma*(Sca*Da+Sca+Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*gamma*(Sca*(1.0-Da)+Dca);
break;
}
case PlusCompositeOp:
{
pixel=QuantumRange*(Sca+Dca);
break;
}
case SaturateCompositeOp:
{
if (fabs((double) (QuantumRange*Sa-TransparentAlpha)) < MagickEpsilon)
{
pixel=Dc;
break;
}
if (fabs((double) (QuantumRange*Da-TransparentAlpha)) < MagickEpsilon)
{
pixel=Sc;
break;
}
CompositeHCL(canvas_pixel.red,canvas_pixel.green,canvas_pixel.blue,
&hue,&chroma,&luma);
CompositeHCL(source_pixel.red,source_pixel.green,source_pixel.blue,
&sans,&chroma,&sans);
HCLComposite(hue,chroma,luma,&red,&green,&blue);
switch (channel)
{
case RedPixelChannel: pixel=red; break;
case GreenPixelChannel: pixel=green; break;
case BluePixelChannel: pixel=blue; break;
default: pixel=Dc; break;
}
break;
}
case ScreenCompositeOp:
{
/*
Screen: a negated multiply:
f(Sc,Dc) = 1.0-(1.0-Sc)*(1.0-Dc)
*/
pixel=QuantumRange*gamma*(Sca+Dca-Sca*Dca);
break;
}
case SoftLightCompositeOp:
{
if ((2.0*Sca) < Sa)
{
pixel=QuantumRange*gamma*(Dca*(Sa+(2.0*Sca-Sa)*(1.0-DcaDa))+
Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
if (((2.0*Sca) > Sa) && ((4.0*Dca) <= Da))
{
pixel=QuantumRange*gamma*(Dca*Sa+Da*(2.0*Sca-Sa)*(4.0*DcaDa*
(4.0*DcaDa+1.0)*(DcaDa-1.0)+7.0*DcaDa)+Sca*(1.0-Da)+
Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*gamma*(Dca*Sa+Da*(2.0*Sca-Sa)*(pow(DcaDa,0.5)-
DcaDa)+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case StereoCompositeOp:
{
if (channel == RedPixelChannel)
pixel=(MagickRealType) GetPixelRed(source_image,p);
break;
}
case ThresholdCompositeOp:
{
MagickRealType
delta;
delta=Sc-Dc;
if ((MagickRealType) fabs((double) (2.0*delta)) < threshold)
{
pixel=gamma*Dc;
break;
}
pixel=gamma*(Dc+delta*amount);
break;
}
case VividLightCompositeOp:
{
/*
VividLight: A Photoshop 7 composition method. See
http://www.simplefilter.de/en/basics/mixmods.html.
f(Sc,Dc) = (2*Sc < 1) ? 1-(1-Dc)/(2*Sc) : Dc/(2*(1-Sc))
*/
if ((fabs((double) Sa) < MagickEpsilon) ||
(fabs((double) (Sca-Sa)) < MagickEpsilon))
{
pixel=QuantumRange*gamma*(Sa*Da+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
if ((2.0*Sca) <= Sa)
{
pixel=QuantumRange*gamma*(Sa*(Da+Sa*(Dca-Da)*
PerceptibleReciprocal(2.0*Sca))+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
pixel=QuantumRange*gamma*(Dca*Sa*Sa*PerceptibleReciprocal(2.0*
(Sa-Sca))+Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
case XorCompositeOp:
{
pixel=QuantumRange*(Sca*(1.0-Da)+Dca*(1.0-Sa));
break;
}
default:
{
pixel=Sc;
break;
}
}
q[i]=clamp != MagickFalse ? ClampPixel(pixel) : ClampToQuantum(pixel);
}
p+=GetPixelChannels(source_image);
channels=GetPixelChannels(source_image);
if (p >= (pixels+channels*source_image->columns))
p=pixels;
q+=GetPixelChannels(image);
}
if (SyncCacheViewAuthenticPixels(image_view,exception) == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_CompositeImage)
#endif
proceed=SetImageProgress(image,CompositeImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
source_view=DestroyCacheView(source_view);
image_view=DestroyCacheView(image_view);
if (canvas_image != (Image * ) NULL)
canvas_image=DestroyImage(canvas_image);
else
source_image=DestroyImage(source_image);
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% T e x t u r e I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% TextureImage() repeatedly tiles the texture image across and down the image
% canvas.
%
% The format of the TextureImage method is:
%
% MagickBooleanType TextureImage(Image *image,const Image *texture,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o texture_image: This image is the texture to layer on the background.
%
*/
MagickExport MagickBooleanType TextureImage(Image *image,const Image *texture,
ExceptionInfo *exception)
{
#define TextureImageTag "Texture/Image"
CacheView
*image_view,
*texture_view;
Image
*texture_image;
MagickBooleanType
status;
ssize_t
y;
assert(image != (Image *) NULL);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"...");
assert(image->signature == MagickCoreSignature);
if (texture == (const Image *) NULL)
return(MagickFalse);
if (SetImageStorageClass(image,DirectClass,exception) == MagickFalse)
return(MagickFalse);
texture_image=CloneImage(texture,0,0,MagickTrue,exception);
if (texture_image == (const Image *) NULL)
return(MagickFalse);
(void) TransformImageColorspace(texture_image,image->colorspace,exception);
(void) SetImageVirtualPixelMethod(texture_image,TileVirtualPixelMethod,
exception);
status=MagickTrue;
if ((image->compose != CopyCompositeOp) &&
((image->compose != OverCompositeOp) ||
(image->alpha_trait != UndefinedPixelTrait) ||
(texture_image->alpha_trait != UndefinedPixelTrait)))
{
/*
Tile texture onto the image background.
*/
for (y=0; y < (ssize_t) image->rows; y+=(ssize_t) texture_image->rows)
{
register ssize_t
x;
if (status == MagickFalse)
continue;
for (x=0; x < (ssize_t) image->columns; x+=(ssize_t) texture_image->columns)
{
MagickBooleanType
thread_status;
thread_status=CompositeImage(image,texture_image,image->compose,
MagickTrue,x+texture_image->tile_offset.x,y+
texture_image->tile_offset.y,exception);
if (thread_status == MagickFalse)
{
status=thread_status;
break;
}
}
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
proceed=SetImageProgress(image,TextureImageTag,(MagickOffsetType) y,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
(void) SetImageProgress(image,TextureImageTag,(MagickOffsetType)
image->rows,image->rows);
texture_image=DestroyImage(texture_image);
return(status);
}
/*
Tile texture onto the image background (optimized).
*/
status=MagickTrue;
texture_view=AcquireVirtualCacheView(texture_image,exception);
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static) shared(status) \
magick_number_threads(texture_image,image,image->rows,1)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
MagickBooleanType
sync;
register const Quantum
*p,
*pixels;
register ssize_t
x;
register Quantum
*q;
size_t
width;
if (status == MagickFalse)
continue;
pixels=GetCacheViewVirtualPixels(texture_view,texture_image->tile_offset.x,
(y+texture_image->tile_offset.y) % texture_image->rows,
texture_image->columns,1,exception);
q=QueueCacheViewAuthenticPixels(image_view,0,y,image->columns,1,exception);
if ((pixels == (const Quantum *) NULL) || (q == (Quantum *) NULL))
{
status=MagickFalse;
continue;
}
for (x=0; x < (ssize_t) image->columns; x+=(ssize_t) texture_image->columns)
{
register ssize_t
j;
p=pixels;
width=texture_image->columns;
if ((x+(ssize_t) width) > (ssize_t) image->columns)
width=image->columns-x;
for (j=0; j < (ssize_t) width; j++)
{
register ssize_t
i;
for (i=0; i < (ssize_t) GetPixelChannels(texture_image); i++)
{
PixelChannel channel = GetPixelChannelChannel(texture_image,i);
PixelTrait traits = GetPixelChannelTraits(image,channel);
PixelTrait texture_traits=GetPixelChannelTraits(texture_image,
channel);
if ((traits == UndefinedPixelTrait) ||
(texture_traits == UndefinedPixelTrait))
continue;
SetPixelChannel(image,channel,p[i],q);
}
p+=GetPixelChannels(texture_image);
q+=GetPixelChannels(image);
}
}
sync=SyncCacheViewAuthenticPixels(image_view,exception);
if (sync == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
proceed=SetImageProgress(image,TextureImageTag,(MagickOffsetType) y,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
texture_view=DestroyCacheView(texture_view);
image_view=DestroyCacheView(image_view);
texture_image=DestroyImage(texture_image);
return(status);
}
|
DRB047-doallchar-orig-no.c | /*
Copyright (c) 2017, Lawrence Livermore National Security, LLC.
Produced at the Lawrence Livermore National Laboratory
Written by Chunhua Liao, Pei-Hung Lin, Joshua Asplund,
Markus Schordan, and Ian Karlin
(email: liao6@llnl.gov, lin32@llnl.gov, asplund1@llnl.gov,
schordan1@llnl.gov, karlin1@llnl.gov)
LLNL-CODE-732144
All rights reserved.
This file is part of DataRaceBench. For details, see
https://github.com/LLNL/dataracebench. Please also see the LICENSE file
for our additional BSD notice.
Redistribution and use in source and binary forms, with
or without modification, are permitted provided that the following
conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the disclaimer below.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the disclaimer (as noted below)
in the documentation and/or other materials provided with the
distribution.
* Neither the name of the LLNS/LLNL nor the names of its contributors
may be used to endorse or promote products derived from this
software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL LAWRENCE LIVERMORE NATIONAL
SECURITY, LLC, THE U.S. DEPARTMENT OF ENERGY OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <stdio.h>
#include <stdlib.h>
/*
One dimension array computation
with finer granularity than traditional 4 bytes.
Dynamic tools monitoring 4-bytes elements may wrongfuly report race condition.
*/
char a[100];
int main()
{
int i;
#pragma omp parallel for simd
for (i=0;i<100;i++)
a[i]=i;
#pragma omp parallel for simd
for (i=0;i<100;i++)
a[i]=a[i]+1;
#pragma omp parallel for simd ordered
for (i=0;i<100;i++)
#pragma omp ordered simd
printf("%c\n",a[i]);
return 0;
}
|
libperf.c | /**
* Copyright (C) Mellanox Technologies Ltd. 2001-2019. ALL RIGHTS RESERVED.
* Copyright (C) UT-Battelle, LLC. 2015. ALL RIGHTS RESERVED.
* Copyright (C) The University of Tennessee and The University
* of Tennessee Research Foundation. 2015-2016. ALL RIGHTS RESERVED.
* Copyright (C) ARM Ltd. 2017-2020. ALL RIGHTS RESERVED.
* See file LICENSE for terms.
*/
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
#include <ucs/debug/log.h>
#include <ucs/arch/bitops.h>
#include <ucs/sys/module.h>
#include <ucs/sys/string.h>
#include <string.h>
#include <tools/perf/lib/libperf_int.h>
#include <unistd.h>
#if _OPENMP
#include <omp.h>
#endif /* _OPENMP */
#define ATOMIC_OP_CONFIG(_size, _op32, _op64, _op, _msg, _params, _status) \
_status = __get_atomic_flag((_size), (_op32), (_op64), (_op)); \
if (_status != UCS_OK) { \
ucs_error(UCT_PERF_TEST_PARAMS_FMT" does not support atomic %s for " \
"message size %zu bytes", UCT_PERF_TEST_PARAMS_ARG(_params), \
(_msg)[_op], (_size)); \
return _status; \
}
#define ATOMIC_OP_CHECK(_size, _attr, _required, _params, _msg) \
if (!ucs_test_all_flags(_attr, _required)) { \
if ((_params)->flags & UCX_PERF_TEST_FLAG_VERBOSE) { \
ucs_error(UCT_PERF_TEST_PARAMS_FMT" does not support required " \
#_size"-bit atomic: %s", UCT_PERF_TEST_PARAMS_ARG(_params), \
(_msg)[ucs_ffs64(~(_attr) & (_required))]); \
} \
return UCS_ERR_UNSUPPORTED; \
}
typedef struct {
union {
struct {
size_t dev_addr_len;
size_t iface_addr_len;
size_t ep_addr_len;
} uct;
struct {
size_t worker_addr_len;
size_t total_wireup_len;
} ucp;
};
size_t rkey_size;
unsigned long recv_buffer;
} ucx_perf_ep_info_t;
const ucx_perf_allocator_t* ucx_perf_mem_type_allocators[UCS_MEMORY_TYPE_LAST];
static const char *perf_iface_ops[] = {
[ucs_ilog2(UCT_IFACE_FLAG_AM_SHORT)] = "am short",
[ucs_ilog2(UCT_IFACE_FLAG_AM_BCOPY)] = "am bcopy",
[ucs_ilog2(UCT_IFACE_FLAG_AM_ZCOPY)] = "am zcopy",
[ucs_ilog2(UCT_IFACE_FLAG_PUT_SHORT)] = "put short",
[ucs_ilog2(UCT_IFACE_FLAG_PUT_BCOPY)] = "put bcopy",
[ucs_ilog2(UCT_IFACE_FLAG_PUT_ZCOPY)] = "put zcopy",
[ucs_ilog2(UCT_IFACE_FLAG_GET_SHORT)] = "get short",
[ucs_ilog2(UCT_IFACE_FLAG_GET_BCOPY)] = "get bcopy",
[ucs_ilog2(UCT_IFACE_FLAG_GET_ZCOPY)] = "get zcopy",
[ucs_ilog2(UCT_IFACE_FLAG_ERRHANDLE_PEER_FAILURE)] = "peer failure handler",
[ucs_ilog2(UCT_IFACE_FLAG_CONNECT_TO_IFACE)] = "connect to iface",
[ucs_ilog2(UCT_IFACE_FLAG_CONNECT_TO_EP)] = "connect to ep",
[ucs_ilog2(UCT_IFACE_FLAG_AM_DUP)] = "full reliability",
[ucs_ilog2(UCT_IFACE_FLAG_CB_SYNC)] = "sync callback",
[ucs_ilog2(UCT_IFACE_FLAG_CB_ASYNC)] = "async callback",
[ucs_ilog2(UCT_IFACE_FLAG_PENDING)] = "pending",
[ucs_ilog2(UCT_IFACE_FLAG_TAG_EAGER_SHORT)] = "tag eager short",
[ucs_ilog2(UCT_IFACE_FLAG_TAG_EAGER_BCOPY)] = "tag eager bcopy",
[ucs_ilog2(UCT_IFACE_FLAG_TAG_EAGER_ZCOPY)] = "tag eager zcopy",
[ucs_ilog2(UCT_IFACE_FLAG_TAG_RNDV_ZCOPY)] = "tag rndv zcopy",
[ucs_ilog2(UCT_IFACE_FLAG_EP_CHECK)] = "ep check",
[ucs_ilog2(UCT_IFACE_FLAG_EP_KEEPALIVE)] = "ep keepalive"
};
static const char *perf_atomic_op[] = {
[UCT_ATOMIC_OP_ADD] = "add",
[UCT_ATOMIC_OP_AND] = "and",
[UCT_ATOMIC_OP_OR] = "or" ,
[UCT_ATOMIC_OP_XOR] = "xor"
};
static const char *perf_atomic_fop[] = {
[UCT_ATOMIC_OP_ADD] = "fetch-add",
[UCT_ATOMIC_OP_AND] = "fetch-and",
[UCT_ATOMIC_OP_OR] = "fetch-or",
[UCT_ATOMIC_OP_XOR] = "fetch-xor",
[UCT_ATOMIC_OP_SWAP] = "swap",
[UCT_ATOMIC_OP_CSWAP] = "cswap"
};
/*
* This Quickselect routine is based on the algorithm described in
* "Numerical recipes in C", Second Edition,
* Cambridge University Press, 1992, Section 8.5, ISBN 0-521-43108-5
* This code by Nicolas Devillard - 1998. Public domain.
*/
static ucs_time_t __find_median_quick_select(ucs_time_t arr[], int n)
{
int low, high ;
int median;
int middle, ll, hh;
#define ELEM_SWAP(a,b) { register ucs_time_t t=(a);(a)=(b);(b)=t; }
low = 0 ; high = n-1 ; median = (low + high) / 2;
for (;;) {
if (high <= low) /* One element only */
return arr[median] ;
if (high == low + 1) { /* Two elements only */
if (arr[low] > arr[high])
ELEM_SWAP(arr[low], arr[high]) ;
return arr[median] ;
}
/* Find median of low, middle and high items; swap into position low */
middle = (low + high) / 2;
if (arr[middle] > arr[high]) ELEM_SWAP(arr[middle], arr[high]) ;
if (arr[low] > arr[high]) ELEM_SWAP(arr[low], arr[high]) ;
if (arr[middle] > arr[low]) ELEM_SWAP(arr[middle], arr[low]) ;
/* Swap low item (now in position middle) into position (low+1) */
ELEM_SWAP(arr[middle], arr[low+1]) ;
/* Nibble from each end towards middle, swapping items when stuck */
ll = low + 1;
hh = high;
for (;;) {
do ll++; while (arr[low] > arr[ll]) ;
do hh--; while (arr[hh] > arr[low]) ;
if (hh < ll)
break;
ELEM_SWAP(arr[ll], arr[hh]) ;
}
/* Swap middle item (in position low) back into correct position */
ELEM_SWAP(arr[low], arr[hh]) ;
/* Re-set active partition */
if (hh <= median)
low = ll;
if (hh >= median)
high = hh - 1;
}
}
static ucs_status_t
uct_perf_test_alloc_host(const ucx_perf_context_t *perf, size_t length,
unsigned flags, uct_allocated_memory_t *alloc_mem)
{
ucs_status_t status;
status = uct_iface_mem_alloc(perf->uct.iface, length,
flags, "perftest", alloc_mem);
if (status != UCS_OK) {
ucs_error("failed to allocate memory: %s", ucs_status_string(status));
return status;
}
ucs_assert(alloc_mem->md == perf->uct.md);
return UCS_OK;
}
static void uct_perf_test_free_host(const ucx_perf_context_t *perf,
uct_allocated_memory_t *alloc_mem)
{
uct_iface_mem_free(alloc_mem);
}
static void ucx_perf_test_memcpy_host(void *dst, ucs_memory_type_t dst_mem_type,
const void *src, ucs_memory_type_t src_mem_type,
size_t count)
{
if ((dst_mem_type != UCS_MEMORY_TYPE_HOST) ||
(src_mem_type != UCS_MEMORY_TYPE_HOST)) {
ucs_error("wrong memory type passed src - %d, dst - %d",
src_mem_type, dst_mem_type);
} else {
memcpy(dst, src, count);
}
}
static ucs_status_t uct_perf_test_alloc_mem(ucx_perf_context_t *perf)
{
ucx_perf_params_t *params = &perf->params;
ucs_status_t status;
unsigned flags;
size_t buffer_size;
if ((UCT_PERF_DATA_LAYOUT_ZCOPY == params->uct.data_layout) && params->iov_stride) {
buffer_size = params->msg_size_cnt * params->iov_stride;
} else {
buffer_size = ucx_perf_get_message_size(params);
}
/* TODO use params->alignment */
flags = (params->flags & UCX_PERF_TEST_FLAG_MAP_NONBLOCK) ?
UCT_MD_MEM_FLAG_NONBLOCK : 0;
flags |= UCT_MD_MEM_ACCESS_ALL;
/* Allocate send buffer memory */
status = perf->allocator->uct_alloc(perf, buffer_size * params->thread_count,
flags, &perf->uct.send_mem);
if (status != UCS_OK) {
goto err;
}
perf->send_buffer = perf->uct.send_mem.address;
/* Allocate receive buffer memory */
status = perf->allocator->uct_alloc(perf, buffer_size * params->thread_count,
flags, &perf->uct.recv_mem);
if (status != UCS_OK) {
goto err_free_send;
}
perf->recv_buffer = perf->uct.recv_mem.address;
/* Allocate IOV datatype memory */
perf->params.msg_size_cnt = params->msg_size_cnt;
perf->uct.iov = malloc(sizeof(*perf->uct.iov) *
perf->params.msg_size_cnt *
params->thread_count);
if (NULL == perf->uct.iov) {
status = UCS_ERR_NO_MEMORY;
ucs_error("Failed allocate send IOV(%lu) buffer: %s",
perf->params.msg_size_cnt, ucs_status_string(status));
goto err_free_recv;
}
ucs_debug("allocated memory. Send buffer %p, Recv buffer %p",
perf->send_buffer, perf->recv_buffer);
return UCS_OK;
err_free_recv:
perf->allocator->uct_free(perf, &perf->uct.recv_mem);
err_free_send:
perf->allocator->uct_free(perf, &perf->uct.send_mem);
err:
return status;
}
static void uct_perf_test_free_mem(ucx_perf_context_t *perf)
{
perf->allocator->uct_free(perf, &perf->uct.send_mem);
perf->allocator->uct_free(perf, &perf->uct.recv_mem);
free(perf->uct.iov);
}
void ucx_perf_test_start_clock(ucx_perf_context_t *perf)
{
ucs_time_t start_time = ucs_get_time();
perf->start_time_acc = ucs_get_accurate_time();
perf->end_time = (perf->params.max_time == 0.0) ? UINT64_MAX :
ucs_time_from_sec(perf->params.max_time) + start_time;
perf->prev_time = start_time;
perf->prev.time = start_time;
perf->prev.time_acc = perf->start_time_acc;
perf->current.time_acc = perf->start_time_acc;
}
/* Initialize/reset all parameters that could be modified by the warm-up run */
static void ucx_perf_test_prepare_new_run(ucx_perf_context_t *perf,
const ucx_perf_params_t *params)
{
unsigned i;
perf->max_iter = (perf->params.max_iter == 0) ? UINT64_MAX :
perf->params.max_iter;
perf->report_interval = ucs_time_from_sec(perf->params.report_interval);
perf->current.time = 0;
perf->current.msgs = 0;
perf->current.bytes = 0;
perf->current.iters = 0;
perf->prev.msgs = 0;
perf->prev.bytes = 0;
perf->prev.iters = 0;
perf->timing_queue_head = 0;
for (i = 0; i < TIMING_QUEUE_SIZE; ++i) {
perf->timing_queue[i] = 0;
}
ucx_perf_test_start_clock(perf);
}
static void ucx_perf_test_init(ucx_perf_context_t *perf,
const ucx_perf_params_t *params)
{
unsigned group_index;
perf->params = *params;
group_index = rte_call(perf, group_index);
if (0 == group_index) {
perf->allocator = ucx_perf_mem_type_allocators[params->send_mem_type];
} else {
perf->allocator = ucx_perf_mem_type_allocators[params->recv_mem_type];
}
ucx_perf_test_prepare_new_run(perf, params);
}
void ucx_perf_calc_result(ucx_perf_context_t *perf, ucx_perf_result_t *result)
{
ucs_time_t median;
double factor;
if ((perf->params.test_type == UCX_PERF_TEST_TYPE_PINGPONG) ||
(perf->params.test_type == UCX_PERF_TEST_TYPE_PINGPONG_WAIT_MEM)) {
factor = 2.0;
} else {
factor = 1.0;
}
result->iters = perf->current.iters;
result->bytes = perf->current.bytes;
result->elapsed_time = perf->current.time_acc - perf->start_time_acc;
/* Latency */
median = __find_median_quick_select(perf->timing_queue, TIMING_QUEUE_SIZE);
result->latency.typical = ucs_time_to_sec(median) / factor;
result->latency.moment_average =
(perf->current.time_acc - perf->prev.time_acc)
/ (perf->current.iters - perf->prev.iters)
/ factor;
result->latency.total_average =
(perf->current.time_acc - perf->start_time_acc)
/ perf->current.iters
/ factor;
/* Bandwidth */
result->bandwidth.typical = 0.0; // Undefined
result->bandwidth.moment_average =
(perf->current.bytes - perf->prev.bytes) /
(perf->current.time_acc - perf->prev.time_acc) * factor;
result->bandwidth.total_average =
perf->current.bytes /
(perf->current.time_acc - perf->start_time_acc) * factor;
/* Packet rate */
result->msgrate.typical = 0.0; // Undefined
result->msgrate.moment_average =
(perf->current.msgs - perf->prev.msgs) /
(perf->current.time_acc - perf->prev.time_acc) * factor;
result->msgrate.total_average =
perf->current.msgs /
(perf->current.time_acc - perf->start_time_acc) * factor;
}
static ucs_status_t ucx_perf_test_check_params(ucx_perf_params_t *params)
{
size_t it;
/* check if zero-size messages are requested and supported */
if ((/* they are not supported by: */
/* - UCT tests, except UCT AM Short/Bcopy */
(params->api == UCX_PERF_API_UCT) ||
(/* - UCP RMA and AMO tests */
(params->api == UCX_PERF_API_UCP) &&
(params->command != UCX_PERF_CMD_AM) &&
(params->command != UCX_PERF_CMD_TAG) &&
(params->command != UCX_PERF_CMD_TAG_SYNC) &&
(params->command != UCX_PERF_CMD_STREAM))) &&
ucx_perf_get_message_size(params) < 1) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Message size too small, need to be at least 1");
}
return UCS_ERR_INVALID_PARAM;
}
if ((params->api == UCX_PERF_API_UCP) &&
((params->send_mem_type != UCS_MEMORY_TYPE_HOST) ||
(params->recv_mem_type != UCS_MEMORY_TYPE_HOST)) &&
((params->command == UCX_PERF_CMD_PUT) ||
(params->command == UCX_PERF_CMD_GET) ||
(params->command == UCX_PERF_CMD_ADD) ||
(params->command == UCX_PERF_CMD_FADD) ||
(params->command == UCX_PERF_CMD_SWAP) ||
(params->command == UCX_PERF_CMD_CSWAP))) {
/* TODO: remove when support for non-HOST memory types will be added */
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("UCP doesn't support RMA/AMO for \"%s\"<->\"%s\" memory types",
ucs_memory_type_names[params->send_mem_type],
ucs_memory_type_names[params->recv_mem_type]);
}
return UCS_ERR_INVALID_PARAM;
}
if (params->max_outstanding < 1) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("max_outstanding, need to be at least 1");
}
return UCS_ERR_INVALID_PARAM;
}
/* check if particular message size fit into stride size */
if (params->iov_stride) {
for (it = 0; it < params->msg_size_cnt; ++it) {
if (params->msg_size_list[it] > params->iov_stride) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Buffer size %lu bigger than stride %lu",
params->msg_size_list[it], params->iov_stride);
}
return UCS_ERR_INVALID_PARAM;
}
}
}
return UCS_OK;
}
void uct_perf_ep_flush_b(ucx_perf_context_t *perf, int peer_index)
{
uct_ep_h ep = perf->uct.peers[peer_index].ep;
uct_completion_t comp;
ucs_status_t status;
int started;
started = 0;
comp.func = NULL;
comp.count = 2;
do {
if (!started) {
status = uct_ep_flush(ep, 0, &comp);
if (status == UCS_OK) {
--comp.count;
} else if (status == UCS_INPROGRESS) {
started = 1;
} else if (status != UCS_ERR_NO_RESOURCE) {
ucs_error("uct_ep_flush() failed: %s", ucs_status_string(status));
return;
}
}
uct_worker_progress(perf->uct.worker);
} while (comp.count > 1);
}
void uct_perf_iface_flush_b(ucx_perf_context_t *perf)
{
ucs_status_t status;
do {
status = uct_iface_flush(perf->uct.iface, 0, NULL);
uct_worker_progress(perf->uct.worker);
} while (status == UCS_INPROGRESS);
if (status != UCS_OK) {
ucs_error("uct_iface_flush() failed: %s", ucs_status_string(status));
}
}
static inline uint64_t __get_flag(uct_perf_data_layout_t layout, uint64_t short_f,
uint64_t bcopy_f, uint64_t zcopy_f)
{
return ((layout == UCT_PERF_DATA_LAYOUT_SHORT) ||
(layout == UCT_PERF_DATA_LAYOUT_SHORT_IOV)) ? short_f :
(layout == UCT_PERF_DATA_LAYOUT_BCOPY) ? bcopy_f :
(layout == UCT_PERF_DATA_LAYOUT_ZCOPY) ? zcopy_f :
0;
}
static inline ucs_status_t __get_atomic_flag(size_t size, uint64_t *op32,
uint64_t *op64, uint64_t op)
{
if (size == sizeof(uint32_t)) {
*op32 = UCS_BIT(op);
return UCS_OK;
} else if (size == sizeof(uint64_t)) {
*op64 = UCS_BIT(op);
return UCS_OK;
}
return UCS_ERR_UNSUPPORTED;
}
static inline size_t __get_max_size(uct_perf_data_layout_t layout, size_t short_m,
size_t bcopy_m, uint64_t zcopy_m)
{
return ((layout == UCT_PERF_DATA_LAYOUT_SHORT) ||
(layout == UCT_PERF_DATA_LAYOUT_SHORT_IOV)) ? short_m :
(layout == UCT_PERF_DATA_LAYOUT_BCOPY) ? bcopy_m :
(layout == UCT_PERF_DATA_LAYOUT_ZCOPY) ? zcopy_m :
0;
}
static ucs_status_t uct_perf_test_check_md_support(ucx_perf_params_t *params,
ucs_memory_type_t mem_type,
uct_md_attr_t *md_attr)
{
if (!(md_attr->cap.access_mem_types & UCS_BIT(mem_type)) &&
!(md_attr->cap.reg_mem_types & UCS_BIT(mem_type))) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Unsupported memory type %s by "UCT_PERF_TEST_PARAMS_FMT,
ucs_memory_type_names[mem_type],
UCT_PERF_TEST_PARAMS_ARG(params));
return UCS_ERR_INVALID_PARAM;
}
}
return UCS_OK;
}
static ucs_status_t uct_perf_test_check_capabilities(ucx_perf_params_t *params,
uct_iface_h iface, uct_md_h md)
{
uint64_t required_flags = 0;
uint64_t atomic_op32 = 0;
uint64_t atomic_op64 = 0;
uint64_t atomic_fop32 = 0;
uint64_t atomic_fop64 = 0;
uct_md_attr_t md_attr;
uct_iface_attr_t attr;
ucs_status_t status;
size_t min_size, max_size, max_iov, message_size;
status = uct_md_query(md, &md_attr);
if (status != UCS_OK) {
ucs_error("uct_md_query(%s) failed: %s",
params->uct.md_name, ucs_status_string(status));
return status;
}
status = uct_iface_query(iface, &attr);
if (status != UCS_OK) {
ucs_error("uct_iface_query("UCT_PERF_TEST_PARAMS_FMT") failed: %s",
UCT_PERF_TEST_PARAMS_ARG(params),
ucs_status_string(status));
return status;
}
min_size = 0;
max_iov = 1;
message_size = ucx_perf_get_message_size(params);
switch (params->command) {
case UCX_PERF_CMD_AM:
required_flags = __get_flag(params->uct.data_layout, UCT_IFACE_FLAG_AM_SHORT,
UCT_IFACE_FLAG_AM_BCOPY, UCT_IFACE_FLAG_AM_ZCOPY);
required_flags |= UCT_IFACE_FLAG_CB_SYNC;
min_size = __get_max_size(params->uct.data_layout, 0, 0,
attr.cap.am.min_zcopy);
max_size = __get_max_size(params->uct.data_layout, attr.cap.am.max_short,
attr.cap.am.max_bcopy, attr.cap.am.max_zcopy);
max_iov = attr.cap.am.max_iov;
break;
case UCX_PERF_CMD_PUT:
required_flags = __get_flag(params->uct.data_layout, UCT_IFACE_FLAG_PUT_SHORT,
UCT_IFACE_FLAG_PUT_BCOPY, UCT_IFACE_FLAG_PUT_ZCOPY);
min_size = __get_max_size(params->uct.data_layout, 0, 0,
attr.cap.put.min_zcopy);
max_size = __get_max_size(params->uct.data_layout, attr.cap.put.max_short,
attr.cap.put.max_bcopy, attr.cap.put.max_zcopy);
max_iov = attr.cap.put.max_iov;
break;
case UCX_PERF_CMD_GET:
required_flags = __get_flag(params->uct.data_layout, UCT_IFACE_FLAG_GET_SHORT,
UCT_IFACE_FLAG_GET_BCOPY, UCT_IFACE_FLAG_GET_ZCOPY);
min_size = __get_max_size(params->uct.data_layout, 0, 0,
attr.cap.get.min_zcopy);
max_size = __get_max_size(params->uct.data_layout, attr.cap.get.max_short,
attr.cap.get.max_bcopy, attr.cap.get.max_zcopy);
max_iov = attr.cap.get.max_iov;
break;
case UCX_PERF_CMD_ADD:
ATOMIC_OP_CONFIG(message_size, &atomic_op32, &atomic_op64, UCT_ATOMIC_OP_ADD,
perf_atomic_op, params, status);
max_size = 8;
break;
case UCX_PERF_CMD_FADD:
ATOMIC_OP_CONFIG(message_size, &atomic_fop32, &atomic_fop64, UCT_ATOMIC_OP_ADD,
perf_atomic_fop, params, status);
max_size = 8;
break;
case UCX_PERF_CMD_SWAP:
ATOMIC_OP_CONFIG(message_size, &atomic_fop32, &atomic_fop64, UCT_ATOMIC_OP_SWAP,
perf_atomic_fop, params, status);
max_size = 8;
break;
case UCX_PERF_CMD_CSWAP:
ATOMIC_OP_CONFIG(message_size, &atomic_fop32, &atomic_fop64, UCT_ATOMIC_OP_CSWAP,
perf_atomic_fop, params, status);
max_size = 8;
break;
default:
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Invalid test command");
}
return UCS_ERR_INVALID_PARAM;
}
status = ucx_perf_test_check_params(params);
if (status != UCS_OK) {
return status;
}
/* check atomics first */
ATOMIC_OP_CHECK(32, attr.cap.atomic32.op_flags, atomic_op32, params, perf_atomic_op);
ATOMIC_OP_CHECK(64, attr.cap.atomic64.op_flags, atomic_op64, params, perf_atomic_op);
ATOMIC_OP_CHECK(32, attr.cap.atomic32.fop_flags, atomic_fop32, params, perf_atomic_fop);
ATOMIC_OP_CHECK(64, attr.cap.atomic64.fop_flags, atomic_fop64, params, perf_atomic_fop);
/* check iface flags */
if (!(atomic_op32 | atomic_op64 | atomic_fop32 | atomic_fop64) &&
(!ucs_test_all_flags(attr.cap.flags, required_flags) || !required_flags)) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error(UCT_PERF_TEST_PARAMS_FMT" does not support operation %s",
UCT_PERF_TEST_PARAMS_ARG(params),
perf_iface_ops[ucs_ffs64(~attr.cap.flags & required_flags)]);
}
return UCS_ERR_UNSUPPORTED;
}
if (message_size < min_size) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Message size (%zu) is smaller than min supported (%zu)",
message_size, min_size);
}
return UCS_ERR_UNSUPPORTED;
}
if (message_size > max_size) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Message size (%zu) is larger than max supported (%zu)",
message_size, max_size);
}
return UCS_ERR_UNSUPPORTED;
}
if (params->command == UCX_PERF_CMD_AM) {
if ((params->uct.data_layout == UCT_PERF_DATA_LAYOUT_SHORT) &&
(params->uct.am_hdr_size != sizeof(uint64_t))) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Short AM header size must be 8 bytes");
}
return UCS_ERR_INVALID_PARAM;
}
if ((params->uct.data_layout == UCT_PERF_DATA_LAYOUT_ZCOPY) &&
(params->uct.am_hdr_size > attr.cap.am.max_hdr)) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("AM header size (%zu) is larger than max supported "
"(%zu)",
params->uct.am_hdr_size, attr.cap.am.max_hdr);
}
return UCS_ERR_UNSUPPORTED;
}
if (params->uct.am_hdr_size > message_size) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("AM header size (%zu) is larger than message size "
"(%zu)",
params->uct.am_hdr_size, message_size);
}
return UCS_ERR_INVALID_PARAM;
}
if (params->uct.fc_window > UCT_PERF_TEST_MAX_FC_WINDOW) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("AM flow-control window (%d) too large (should be <= %d)",
params->uct.fc_window, UCT_PERF_TEST_MAX_FC_WINDOW);
}
return UCS_ERR_INVALID_PARAM;
}
if ((params->flags & UCX_PERF_TEST_FLAG_ONE_SIDED) &&
(params->flags & UCX_PERF_TEST_FLAG_VERBOSE))
{
ucs_warn("Running active-message test with on-sided progress");
}
}
if ((UCT_PERF_DATA_LAYOUT_ZCOPY == params->uct.data_layout) ||
(UCT_PERF_DATA_LAYOUT_SHORT_IOV == params->uct.data_layout)) {
if (params->msg_size_cnt > max_iov) {
if ((params->flags & UCX_PERF_TEST_FLAG_VERBOSE) ||
!params->msg_size_cnt) {
ucs_error("Wrong number of IOV entries. Requested is %lu, "
"should be in the range 1...%lu", params->msg_size_cnt,
max_iov);
}
return UCS_ERR_UNSUPPORTED;
}
/* if msg_size_cnt == 1 the message size checked above */
if ((UCT_PERF_DATA_LAYOUT_ZCOPY == params->uct.data_layout) &&
(UCX_PERF_CMD_AM == params->command) && (params->msg_size_cnt > 1)) {
if (params->uct.am_hdr_size > params->msg_size_list[0]) {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("AM header size (%lu) larger than the first IOV "
"message size (%lu)",
params->uct.am_hdr_size,
params->msg_size_list[0]);
}
return UCS_ERR_INVALID_PARAM;
}
}
}
status = uct_perf_test_check_md_support(params, params->send_mem_type, &md_attr);
if (status != UCS_OK) {
return status;
}
status = uct_perf_test_check_md_support(params, params->recv_mem_type, &md_attr);
if (status != UCS_OK) {
return status;
}
return UCS_OK;
}
static ucs_status_t uct_perf_test_setup_endpoints(ucx_perf_context_t *perf)
{
const size_t buffer_size = ADDR_BUF_SIZE;
ucx_perf_ep_info_t info, *remote_info;
unsigned group_size, i, group_index;
uct_device_addr_t *dev_addr;
uct_iface_addr_t *iface_addr;
uct_ep_addr_t *ep_addr;
uct_iface_attr_t iface_attr;
uct_md_attr_t md_attr;
uct_ep_params_t ep_params;
void *rkey_buffer;
ucs_status_t status;
struct iovec vec[5];
void *buffer;
void *req;
buffer = malloc(buffer_size);
if (buffer == NULL) {
ucs_error("Failed to allocate RTE buffer");
status = UCS_ERR_NO_MEMORY;
goto err;
}
status = uct_iface_query(perf->uct.iface, &iface_attr);
if (status != UCS_OK) {
ucs_error("Failed to uct_iface_query: %s", ucs_status_string(status));
goto err_free;
}
status = uct_md_query(perf->uct.md, &md_attr);
if (status != UCS_OK) {
ucs_error("Failed to uct_md_query: %s", ucs_status_string(status));
goto err_free;
}
if (md_attr.cap.flags & (UCT_MD_FLAG_ALLOC|UCT_MD_FLAG_REG)) {
info.rkey_size = md_attr.rkey_packed_size;
} else {
info.rkey_size = 0;
}
info.uct.dev_addr_len = iface_attr.device_addr_len;
info.uct.iface_addr_len = iface_attr.iface_addr_len;
info.uct.ep_addr_len = iface_attr.ep_addr_len;
info.recv_buffer = (uintptr_t)perf->recv_buffer;
rkey_buffer = buffer;
dev_addr = UCS_PTR_BYTE_OFFSET(rkey_buffer, info.rkey_size);
iface_addr = UCS_PTR_BYTE_OFFSET(dev_addr, info.uct.dev_addr_len);
ep_addr = UCS_PTR_BYTE_OFFSET(iface_addr, info.uct.iface_addr_len);
ucs_assert_always(UCS_PTR_BYTE_OFFSET(ep_addr, info.uct.ep_addr_len) <=
UCS_PTR_BYTE_OFFSET(buffer, buffer_size));
status = uct_iface_get_device_address(perf->uct.iface, dev_addr);
if (status != UCS_OK) {
ucs_error("Failed to uct_iface_get_device_address: %s",
ucs_status_string(status));
goto err_free;
}
status = uct_iface_get_address(perf->uct.iface, iface_addr);
if (status != UCS_OK) {
ucs_error("Failed to uct_iface_get_address: %s", ucs_status_string(status));
goto err_free;
}
if (info.rkey_size > 0) {
memset(rkey_buffer, 0, info.rkey_size);
status = uct_md_mkey_pack(perf->uct.md, perf->uct.recv_mem.memh, rkey_buffer);
if (status != UCS_OK) {
ucs_error("Failed to uct_rkey_pack: %s", ucs_status_string(status));
goto err_free;
}
}
group_size = rte_call(perf, group_size);
group_index = rte_call(perf, group_index);
perf->uct.peers = calloc(group_size, sizeof(*perf->uct.peers));
if (perf->uct.peers == NULL) {
goto err_free;
}
ep_params.field_mask = UCT_EP_PARAM_FIELD_IFACE;
ep_params.iface = perf->uct.iface;
if (iface_attr.cap.flags & UCT_IFACE_FLAG_CONNECT_TO_EP) {
for (i = 0; i < group_size; ++i) {
if (i == group_index) {
continue;
}
status = uct_ep_create(&ep_params, &perf->uct.peers[i].ep);
if (status != UCS_OK) {
ucs_error("Failed to uct_ep_create: %s", ucs_status_string(status));
goto err_destroy_eps;
}
status = uct_ep_get_address(perf->uct.peers[i].ep, ep_addr);
if (status != UCS_OK) {
ucs_error("Failed to uct_ep_get_address: %s", ucs_status_string(status));
goto err_destroy_eps;
}
}
} else if (iface_attr.cap.flags & UCT_IFACE_FLAG_CONNECT_TO_IFACE) {
ep_params.field_mask |= UCT_EP_PARAM_FIELD_DEV_ADDR |
UCT_EP_PARAM_FIELD_IFACE_ADDR;
}
vec[0].iov_base = &info;
vec[0].iov_len = sizeof(info);
vec[1].iov_base = buffer;
vec[1].iov_len = info.rkey_size + info.uct.dev_addr_len +
info.uct.iface_addr_len + info.uct.ep_addr_len;
rte_call(perf, post_vec, vec, 2, &req);
rte_call(perf, exchange_vec, req);
for (i = 0; i < group_size; ++i) {
if (i == group_index) {
continue;
}
rte_call(perf, recv, i, buffer, buffer_size, req);
remote_info = buffer;
rkey_buffer = remote_info + 1;
dev_addr = UCS_PTR_BYTE_OFFSET(rkey_buffer, remote_info->rkey_size);
iface_addr = UCS_PTR_BYTE_OFFSET(dev_addr, remote_info->uct.dev_addr_len);
ep_addr = UCS_PTR_BYTE_OFFSET(iface_addr, remote_info->uct.iface_addr_len);
perf->uct.peers[i].remote_addr = remote_info->recv_buffer;
if (!uct_iface_is_reachable(perf->uct.iface, dev_addr,
remote_info->uct.iface_addr_len ?
iface_addr : NULL)) {
ucs_error("Destination is unreachable");
status = UCS_ERR_UNREACHABLE;
goto err_destroy_eps;
}
if (remote_info->rkey_size > 0) {
status = uct_rkey_unpack(perf->uct.cmpt, rkey_buffer,
&perf->uct.peers[i].rkey);
if (status != UCS_OK) {
ucs_error("Failed to uct_rkey_unpack: %s", ucs_status_string(status));
goto err_destroy_eps;
}
} else {
perf->uct.peers[i].rkey.handle = NULL;
perf->uct.peers[i].rkey.rkey = UCT_INVALID_RKEY;
}
if (iface_attr.cap.flags & UCT_IFACE_FLAG_CONNECT_TO_EP) {
status = uct_ep_connect_to_ep(perf->uct.peers[i].ep, dev_addr, ep_addr);
} else if (iface_attr.cap.flags & UCT_IFACE_FLAG_CONNECT_TO_IFACE) {
ep_params.dev_addr = dev_addr;
ep_params.iface_addr = iface_addr;
status = uct_ep_create(&ep_params, &perf->uct.peers[i].ep);
} else {
status = UCS_ERR_UNSUPPORTED;
}
if (status != UCS_OK) {
ucs_error("Failed to connect endpoint: %s", ucs_status_string(status));
goto err_destroy_eps;
}
}
uct_perf_iface_flush_b(perf);
free(buffer);
uct_perf_barrier(perf);
return UCS_OK;
err_destroy_eps:
for (i = 0; i < group_size; ++i) {
if (perf->uct.peers[i].rkey.rkey != UCT_INVALID_RKEY) {
uct_rkey_release(perf->uct.cmpt, &perf->uct.peers[i].rkey);
}
if (perf->uct.peers[i].ep != NULL) {
uct_ep_destroy(perf->uct.peers[i].ep);
}
}
free(perf->uct.peers);
err_free:
free(buffer);
err:
return status;
}
static void uct_perf_test_cleanup_endpoints(ucx_perf_context_t *perf)
{
unsigned group_size, group_index, i;
uct_perf_barrier(perf);
uct_iface_set_am_handler(perf->uct.iface, UCT_PERF_TEST_AM_ID, NULL, NULL, 0);
group_size = rte_call(perf, group_size);
group_index = rte_call(perf, group_index);
for (i = 0; i < group_size; ++i) {
if (i != group_index) {
if (perf->uct.peers[i].rkey.rkey != UCT_INVALID_RKEY) {
uct_rkey_release(perf->uct.cmpt, &perf->uct.peers[i].rkey);
}
if (perf->uct.peers[i].ep) {
uct_ep_destroy(perf->uct.peers[i].ep);
}
}
}
free(perf->uct.peers);
}
static ucs_status_t ucp_perf_test_fill_params(ucx_perf_params_t *params,
ucp_params_t *ucp_params)
{
ucs_status_t status;
size_t message_size;
message_size = ucx_perf_get_message_size(params);
switch (params->command) {
case UCX_PERF_CMD_PUT:
case UCX_PERF_CMD_GET:
ucp_params->features |= UCP_FEATURE_RMA;
break;
case UCX_PERF_CMD_ADD:
case UCX_PERF_CMD_FADD:
case UCX_PERF_CMD_SWAP:
case UCX_PERF_CMD_CSWAP:
if (message_size == sizeof(uint32_t)) {
ucp_params->features |= UCP_FEATURE_AMO32;
} else if (message_size == sizeof(uint64_t)) {
ucp_params->features |= UCP_FEATURE_AMO64;
} else {
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Atomic size should be either 32 or 64 bit");
}
return UCS_ERR_INVALID_PARAM;
}
break;
case UCX_PERF_CMD_TAG:
case UCX_PERF_CMD_TAG_SYNC:
ucp_params->features |= UCP_FEATURE_TAG;
break;
case UCX_PERF_CMD_STREAM:
ucp_params->features |= UCP_FEATURE_STREAM;
break;
case UCX_PERF_CMD_AM:
ucp_params->features |= UCP_FEATURE_AM;
break;
default:
if (params->flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Invalid test command");
}
return UCS_ERR_INVALID_PARAM;
}
if ((params->flags & UCX_PERF_TEST_FLAG_WAKEUP) ||
(params->wait_mode == UCX_PERF_WAIT_MODE_SLEEP)) {
ucp_params->features |= UCP_FEATURE_WAKEUP;
}
status = ucx_perf_test_check_params(params);
if (status != UCS_OK) {
return status;
}
return UCS_OK;
}
static ucs_status_t ucp_perf_test_alloc_iov_mem(ucp_perf_datatype_t datatype,
size_t iovcnt, unsigned thread_count,
ucp_dt_iov_t **iov_p)
{
ucp_dt_iov_t *iov;
if (UCP_PERF_DATATYPE_IOV == datatype) {
iov = malloc(sizeof(*iov) * iovcnt * thread_count);
if (NULL == iov) {
ucs_error("Failed allocate IOV buffer with iovcnt=%lu", iovcnt);
return UCS_ERR_NO_MEMORY;
}
*iov_p = iov;
}
return UCS_OK;
}
static ucs_status_t
ucp_perf_test_alloc_host(const ucx_perf_context_t *perf, size_t length,
void **address_p, ucp_mem_h *memh, int non_blk_flag)
{
ucp_mem_map_params_t mem_map_params;
ucp_mem_attr_t mem_attr;
ucs_status_t status;
mem_map_params.field_mask = UCP_MEM_MAP_PARAM_FIELD_ADDRESS |
UCP_MEM_MAP_PARAM_FIELD_LENGTH |
UCP_MEM_MAP_PARAM_FIELD_FLAGS;
mem_map_params.address = *address_p;
mem_map_params.length = length;
mem_map_params.flags = UCP_MEM_MAP_ALLOCATE;
if (perf->params.flags & UCX_PERF_TEST_FLAG_MAP_NONBLOCK) {
mem_map_params.flags |= non_blk_flag;
}
status = ucp_mem_map(perf->ucp.context, &mem_map_params, memh);
if (status != UCS_OK) {
goto err;
}
mem_attr.field_mask = UCP_MEM_ATTR_FIELD_ADDRESS;
status = ucp_mem_query(*memh, &mem_attr);
if (status != UCS_OK) {
goto err;
}
*address_p = mem_attr.address;
return UCS_OK;
err:
return status;
}
static void ucp_perf_test_free_host(const ucx_perf_context_t *perf,
void *address, ucp_mem_h memh)
{
ucs_status_t status;
status = ucp_mem_unmap(perf->ucp.context, memh);
if (status != UCS_OK) {
ucs_warn("ucp_mem_unmap() failed: %s", ucs_status_string(status));
}
}
static ucs_status_t ucp_perf_test_alloc_mem(ucx_perf_context_t *perf)
{
ucx_perf_params_t *params = &perf->params;
ucs_status_t status;
size_t buffer_size;
if (params->iov_stride) {
buffer_size = params->msg_size_cnt * params->iov_stride;
} else {
buffer_size = ucx_perf_get_message_size(params);
}
/* Allocate send buffer memory */
perf->send_buffer = NULL;
status = perf->allocator->ucp_alloc(perf, buffer_size * params->thread_count,
&perf->send_buffer, &perf->ucp.send_memh,
UCP_MEM_MAP_NONBLOCK);
if (status != UCS_OK) {
goto err;
}
/* Allocate receive buffer memory */
perf->recv_buffer = NULL;
status = perf->allocator->ucp_alloc(perf, buffer_size * params->thread_count,
&perf->recv_buffer, &perf->ucp.recv_memh,
0);
if (status != UCS_OK) {
goto err_free_send_buffer;
}
/* Allocate AM header */
if (params->ucp.am_hdr_size != 0) {
perf->ucp.am_hdr = malloc(params->ucp.am_hdr_size);
if (perf->ucp.am_hdr == NULL) {
goto err_free_buffers;
}
} else {
perf->ucp.am_hdr = NULL;
}
/* Allocate IOV datatype memory */
perf->ucp.send_iov = NULL;
status = ucp_perf_test_alloc_iov_mem(params->ucp.send_datatype,
perf->params.msg_size_cnt,
params->thread_count,
&perf->ucp.send_iov);
if (UCS_OK != status) {
goto err_free_am_hdr;
}
perf->ucp.recv_iov = NULL;
status = ucp_perf_test_alloc_iov_mem(params->ucp.recv_datatype,
perf->params.msg_size_cnt,
params->thread_count,
&perf->ucp.recv_iov);
if (UCS_OK != status) {
goto err_free_send_iov_buffers;
}
return UCS_OK;
err_free_send_iov_buffers:
free(perf->ucp.send_iov);
err_free_am_hdr:
free(perf->ucp.am_hdr);
err_free_buffers:
perf->allocator->ucp_free(perf, perf->recv_buffer, perf->ucp.recv_memh);
err_free_send_buffer:
perf->allocator->ucp_free(perf, perf->send_buffer, perf->ucp.send_memh);
err:
return UCS_ERR_NO_MEMORY;
}
static void ucp_perf_test_free_mem(ucx_perf_context_t *perf)
{
free(perf->ucp.recv_iov);
free(perf->ucp.send_iov);
free(perf->ucp.am_hdr);
perf->allocator->ucp_free(perf, perf->recv_buffer, perf->ucp.recv_memh);
perf->allocator->ucp_free(perf, perf->send_buffer, perf->ucp.send_memh);
}
static void ucp_perf_test_destroy_eps(ucx_perf_context_t* perf)
{
unsigned i, thread_count = perf->params.thread_count;
ucs_status_ptr_t *req;
ucs_status_t status;
for (i = 0; i < thread_count; ++i) {
if (perf->ucp.tctx[i].perf.ucp.rkey != NULL) {
ucp_rkey_destroy(perf->ucp.tctx[i].perf.ucp.rkey);
}
if (perf->ucp.tctx[i].perf.ucp.ep != NULL) {
req = ucp_ep_close_nb(perf->ucp.tctx[i].perf.ucp.ep,
UCP_EP_CLOSE_MODE_FLUSH);
if (UCS_PTR_IS_PTR(req)) {
do {
ucp_worker_progress(perf->ucp.tctx[i].perf.ucp.worker);
status = ucp_request_check_status(req);
} while (status == UCS_INPROGRESS);
ucp_request_release(req);
} else if (UCS_PTR_STATUS(req) != UCS_OK) {
ucs_warn("failed to close ep %p on thread %d: %s\n",
perf->ucp.tctx[i].perf.ucp.ep, i,
ucs_status_string(UCS_PTR_STATUS(req)));
}
}
}
}
static ucs_status_t ucp_perf_test_exchange_status(ucx_perf_context_t *perf,
ucs_status_t status)
{
unsigned group_size = rte_call(perf, group_size);
ucs_status_t collective_status = status;
struct iovec vec;
void *req = NULL;
unsigned i;
vec.iov_base = &status;
vec.iov_len = sizeof(status);
rte_call(perf, post_vec, &vec, 1, &req);
rte_call(perf, exchange_vec, req);
for (i = 0; i < group_size; ++i) {
rte_call(perf, recv, i, &status, sizeof(status), req);
if (status != UCS_OK) {
collective_status = status;
}
}
return collective_status;
}
static ucs_status_t ucp_perf_test_receive_remote_data(ucx_perf_context_t *perf)
{
unsigned thread_count = perf->params.thread_count;
void *rkey_buffer = NULL;
void *req = NULL;
unsigned group_size, group_index, i;
ucx_perf_ep_info_t *remote_info;
ucp_ep_params_t ep_params;
ucp_address_t *address;
ucs_status_t status;
size_t buffer_size;
void *buffer;
group_size = rte_call(perf, group_size);
group_index = rte_call(perf, group_index);
if (group_size != 2) {
ucs_error("perftest requires group size to be exactly 2 "
"(actual group size: %u)", group_size);
return UCS_ERR_UNSUPPORTED;
}
buffer_size = ADDR_BUF_SIZE * thread_count;
buffer = malloc(buffer_size);
if (buffer == NULL) {
ucs_error("failed to allocate RTE receive buffer");
status = UCS_ERR_NO_MEMORY;
goto err;
}
/* Initialize all endpoints and rkeys to NULL to handle error flow */
for (i = 0; i < thread_count; i++) {
perf->ucp.tctx[i].perf.ucp.ep = NULL;
perf->ucp.tctx[i].perf.ucp.rkey = NULL;
}
/* receive the data from the remote peer, extract the address from it
* (along with additional wireup info) and create an endpoint to the peer */
rte_call(perf, recv, 1 - group_index, buffer, buffer_size, req);
remote_info = buffer;
for (i = 0; i < thread_count; i++) {
address = (ucp_address_t*)(remote_info + 1);
rkey_buffer = UCS_PTR_BYTE_OFFSET(address,
remote_info->ucp.worker_addr_len);
perf->ucp.tctx[i].perf.ucp.remote_addr = remote_info->recv_buffer;
ep_params.field_mask = UCP_EP_PARAM_FIELD_REMOTE_ADDRESS;
ep_params.address = address;
status = ucp_ep_create(perf->ucp.tctx[i].perf.ucp.worker, &ep_params,
&perf->ucp.tctx[i].perf.ucp.ep);
if (status != UCS_OK) {
if (perf->params.flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("ucp_ep_create() failed: %s", ucs_status_string(status));
}
goto err_free_eps_buffer;
}
if (remote_info->rkey_size > 0) {
status = ucp_ep_rkey_unpack(perf->ucp.tctx[i].perf.ucp.ep, rkey_buffer,
&perf->ucp.tctx[i].perf.ucp.rkey);
if (status != UCS_OK) {
if (perf->params.flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_fatal("ucp_rkey_unpack() failed: %s", ucs_status_string(status));
}
goto err_free_eps_buffer;
}
} else {
perf->ucp.tctx[i].perf.ucp.rkey = NULL;
}
remote_info = UCS_PTR_BYTE_OFFSET(remote_info,
remote_info->ucp.total_wireup_len);
}
free(buffer);
return UCS_OK;
err_free_eps_buffer:
ucp_perf_test_destroy_eps(perf);
free(buffer);
err:
return status;
}
static ucs_status_t ucp_perf_test_send_local_data(ucx_perf_context_t *perf,
uint64_t features)
{
unsigned i, j, thread_count = perf->params.thread_count;
size_t address_length = 0;
void *rkey_buffer = NULL;
void *req = NULL;
ucx_perf_ep_info_t *info;
ucp_address_t *address;
ucs_status_t status;
struct iovec *vec;
size_t rkey_size;
if (features & (UCP_FEATURE_RMA|UCP_FEATURE_AMO32|UCP_FEATURE_AMO64)) {
status = ucp_rkey_pack(perf->ucp.context, perf->ucp.recv_memh,
&rkey_buffer, &rkey_size);
if (status != UCS_OK) {
if (perf->params.flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("ucp_rkey_pack() failed: %s", ucs_status_string(status));
}
goto err;
}
} else {
rkey_size = 0;
}
/* each thread has an iovec with 3 entries to send to the remote peer:
* ep_info, worker_address and rkey buffer */
vec = calloc(3 * thread_count, sizeof(struct iovec));
if (vec == NULL) {
ucs_error("failed to allocate iovec");
status = UCS_ERR_NO_MEMORY;
goto err_rkey_release;
}
/* get the worker address created for every thread and send it to the remote
* peer */
for (i = 0; i < thread_count; i++) {
status = ucp_worker_get_address(perf->ucp.tctx[i].perf.ucp.worker,
&address, &address_length);
if (status != UCS_OK) {
if (perf->params.flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("ucp_worker_get_address() failed: %s",
ucs_status_string(status));
}
goto err_free_workers_vec;
}
vec[i * 3].iov_base = malloc(sizeof(*info));
if (vec[i * 3].iov_base == NULL) {
ucs_error("failed to allocate vec entry for info");
status = UCS_ERR_NO_MEMORY;
ucp_worker_destroy(perf->ucp.tctx[i].perf.ucp.worker);
goto err_free_workers_vec;
}
info = vec[i * 3].iov_base;
info->ucp.worker_addr_len = address_length;
info->ucp.total_wireup_len = sizeof(*info) + address_length + rkey_size;
info->rkey_size = rkey_size;
info->recv_buffer = (uintptr_t)perf->ucp.tctx[i].perf.recv_buffer;
vec[(i * 3) + 0].iov_len = sizeof(*info);
vec[(i * 3) + 1].iov_base = address;
vec[(i * 3) + 1].iov_len = address_length;
vec[(i * 3) + 2].iov_base = rkey_buffer;
vec[(i * 3) + 2].iov_len = info->rkey_size;
address_length = 0;
}
/* send to the remote peer */
rte_call(perf, post_vec, vec, 3 * thread_count, &req);
rte_call(perf, exchange_vec, req);
if (features & (UCP_FEATURE_RMA|UCP_FEATURE_AMO32|UCP_FEATURE_AMO64)) {
ucp_rkey_buffer_release(rkey_buffer);
}
for (i = 0; i < thread_count; i++) {
free(vec[i * 3].iov_base);
ucp_worker_release_address(perf->ucp.tctx[i].perf.ucp.worker,
vec[(i * 3) + 1].iov_base);
}
free(vec);
return UCS_OK;
err_free_workers_vec:
for (j = 0; j < i; j++) {
ucp_worker_destroy(perf->ucp.tctx[i].perf.ucp.worker);
}
free(vec);
err_rkey_release:
if (features & (UCP_FEATURE_RMA|UCP_FEATURE_AMO32|UCP_FEATURE_AMO64)) {
ucp_rkey_buffer_release(rkey_buffer);
}
err:
return status;
}
static ucs_status_t ucp_perf_test_setup_endpoints(ucx_perf_context_t *perf,
uint64_t features)
{
ucs_status_t status;
unsigned i;
/* pack the local endpoints data and send to the remote peer */
status = ucp_perf_test_send_local_data(perf, features);
if (status != UCS_OK) {
goto err;
}
/* receive remote peer's endpoints' data and connect to them */
status = ucp_perf_test_receive_remote_data(perf);
if (status != UCS_OK) {
goto err;
}
/* sync status across all processes */
status = ucp_perf_test_exchange_status(perf, UCS_OK);
if (status != UCS_OK) {
goto err_destroy_eps;
}
/* force wireup completion */
for (i = 0; i < perf->params.thread_count; i++) {
status = ucp_worker_flush(perf->ucp.tctx[i].perf.ucp.worker);
if (status != UCS_OK) {
ucs_warn("ucp_worker_flush() failed on thread %d: %s",
i, ucs_status_string(status));
}
}
return status;
err_destroy_eps:
ucp_perf_test_destroy_eps(perf);
err:
(void)ucp_perf_test_exchange_status(perf, status);
return status;
}
static void ucp_perf_test_cleanup_endpoints(ucx_perf_context_t *perf)
{
ucp_perf_barrier(perf);
ucp_perf_test_destroy_eps(perf);
}
static void ucp_perf_test_destroy_workers(ucx_perf_context_t *perf)
{
unsigned i;
for (i = 0; i < perf->params.thread_count; i++) {
if (perf->ucp.tctx[i].perf.ucp.worker != NULL) {
ucp_worker_destroy(perf->ucp.tctx[i].perf.ucp.worker);
}
}
}
static void ucx_perf_set_warmup(ucx_perf_context_t* perf,
const ucx_perf_params_t* params)
{
perf->max_iter = ucs_min(params->warmup_iter,
ucs_div_round_up(params->max_iter, 10));
perf->report_interval = ULONG_MAX;
}
static ucs_status_t uct_perf_create_md(ucx_perf_context_t *perf)
{
uct_component_h *uct_components;
uct_component_attr_t component_attr;
uct_tl_resource_desc_t *tl_resources;
unsigned md_index, num_components;
unsigned tl_index, num_tl_resources;
unsigned cmpt_index;
ucs_status_t status;
uct_md_h md;
uct_md_config_t *md_config;
status = uct_query_components(&uct_components, &num_components);
if (status != UCS_OK) {
goto out;
}
for (cmpt_index = 0; cmpt_index < num_components; ++cmpt_index) {
component_attr.field_mask = UCT_COMPONENT_ATTR_FIELD_MD_RESOURCE_COUNT;
status = uct_component_query(uct_components[cmpt_index], &component_attr);
if (status != UCS_OK) {
goto out_release_components_list;
}
component_attr.field_mask = UCT_COMPONENT_ATTR_FIELD_MD_RESOURCES;
component_attr.md_resources = alloca(sizeof(*component_attr.md_resources) *
component_attr.md_resource_count);
status = uct_component_query(uct_components[cmpt_index], &component_attr);
if (status != UCS_OK) {
goto out_release_components_list;
}
for (md_index = 0; md_index < component_attr.md_resource_count; ++md_index) {
status = uct_md_config_read(uct_components[cmpt_index], NULL, NULL,
&md_config);
if (status != UCS_OK) {
goto out_release_components_list;
}
ucs_strncpy_zero(perf->params.uct.md_name,
component_attr.md_resources[md_index].md_name,
UCT_MD_NAME_MAX);
status = uct_md_open(uct_components[cmpt_index],
component_attr.md_resources[md_index].md_name,
md_config, &md);
uct_config_release(md_config);
if (status != UCS_OK) {
goto out_release_components_list;
}
status = uct_md_query_tl_resources(md, &tl_resources, &num_tl_resources);
if (status != UCS_OK) {
uct_md_close(md);
goto out_release_components_list;
}
for (tl_index = 0; tl_index < num_tl_resources; ++tl_index) {
if (!strcmp(perf->params.uct.tl_name, tl_resources[tl_index].tl_name) &&
!strcmp(perf->params.uct.dev_name, tl_resources[tl_index].dev_name))
{
uct_release_tl_resource_list(tl_resources);
perf->uct.cmpt = uct_components[cmpt_index];
perf->uct.md = md;
status = UCS_OK;
goto out_release_components_list;
}
}
uct_md_close(md);
uct_release_tl_resource_list(tl_resources);
}
}
ucs_error("Cannot use "UCT_PERF_TEST_PARAMS_FMT,
UCT_PERF_TEST_PARAMS_ARG(&perf->params));
status = UCS_ERR_NO_DEVICE;
out_release_components_list:
uct_release_component_list(uct_components);
out:
return status;
}
void uct_perf_barrier(ucx_perf_context_t *perf)
{
rte_call(perf, barrier, (void(*)(void*))uct_worker_progress,
(void*)perf->uct.worker);
}
void ucp_perf_barrier(ucx_perf_context_t *perf)
{
rte_call(perf, barrier, (void(*)(void*))ucp_worker_progress,
#if _OPENMP
(void*)perf->ucp.tctx[omp_get_thread_num()].perf.ucp.worker);
#else
(void*)perf->ucp.tctx[0].perf.ucp.worker);
#endif
}
static ucs_status_t uct_perf_setup(ucx_perf_context_t *perf)
{
ucx_perf_params_t *params = &perf->params;
uct_iface_config_t *iface_config;
ucs_status_t status;
uct_iface_params_t iface_params = {
.field_mask = UCT_IFACE_PARAM_FIELD_OPEN_MODE |
UCT_IFACE_PARAM_FIELD_STATS_ROOT |
UCT_IFACE_PARAM_FIELD_RX_HEADROOM |
UCT_IFACE_PARAM_FIELD_CPU_MASK,
.open_mode = UCT_IFACE_OPEN_MODE_DEVICE,
.mode.device.tl_name = params->uct.tl_name,
.mode.device.dev_name = params->uct.dev_name,
.stats_root = ucs_stats_get_root(),
.rx_headroom = 0
};
UCS_CPU_ZERO(&iface_params.cpu_mask);
status = ucs_async_context_init(&perf->uct.async, params->async_mode);
if (status != UCS_OK) {
goto out;
}
status = uct_worker_create(&perf->uct.async, params->thread_mode,
&perf->uct.worker);
if (status != UCS_OK) {
goto out_cleanup_async;
}
status = uct_perf_create_md(perf);
if (status != UCS_OK) {
goto out_destroy_worker;
}
status = uct_md_iface_config_read(perf->uct.md, params->uct.tl_name, NULL,
NULL, &iface_config);
if (status != UCS_OK) {
goto out_destroy_md;
}
status = uct_iface_open(perf->uct.md, perf->uct.worker, &iface_params,
iface_config, &perf->uct.iface);
uct_config_release(iface_config);
if (status != UCS_OK) {
ucs_error("Failed to open iface: %s", ucs_status_string(status));
goto out_destroy_md;
}
status = uct_perf_test_check_capabilities(params, perf->uct.iface,
perf->uct.md);
/* sync status across all processes */
status = ucp_perf_test_exchange_status(perf, status);
if (status != UCS_OK) {
goto out_iface_close;
}
status = uct_perf_test_alloc_mem(perf);
if (status != UCS_OK) {
goto out_iface_close;
}
/* Enable progress before `uct_iface_flush` and `uct_worker_progress` called
* to give a chance to finish connection for some transports (ib/ud, tcp).
* They may return UCS_INPROGRESS from `uct_iface_flush` when connections are
* in progress */
uct_iface_progress_enable(perf->uct.iface,
UCT_PROGRESS_SEND | UCT_PROGRESS_RECV);
status = uct_perf_test_setup_endpoints(perf);
if (status != UCS_OK) {
ucs_error("Failed to setup endpoints: %s", ucs_status_string(status));
goto out_free_mem;
}
return UCS_OK;
out_free_mem:
uct_perf_test_free_mem(perf);
out_iface_close:
uct_iface_close(perf->uct.iface);
out_destroy_md:
uct_md_close(perf->uct.md);
out_destroy_worker:
uct_worker_destroy(perf->uct.worker);
out_cleanup_async:
ucs_async_context_cleanup(&perf->uct.async);
out:
return status;
}
static void uct_perf_cleanup(ucx_perf_context_t *perf)
{
uct_perf_test_cleanup_endpoints(perf);
uct_perf_test_free_mem(perf);
uct_iface_close(perf->uct.iface);
uct_md_close(perf->uct.md);
uct_worker_destroy(perf->uct.worker);
ucs_async_context_cleanup(&perf->uct.async);
}
static void ucp_perf_request_init(void *req)
{
ucp_perf_request_t *request = req;
request->context = NULL;
}
static ucs_status_t ucp_perf_setup(ucx_perf_context_t *perf)
{
ucp_params_t ucp_params;
ucp_worker_params_t worker_params;
ucp_worker_attr_t worker_attr;
ucp_config_t *config;
ucs_status_t status;
unsigned i, thread_count;
size_t message_size;
ucp_params.field_mask = UCP_PARAM_FIELD_FEATURES |
UCP_PARAM_FIELD_REQUEST_SIZE |
UCP_PARAM_FIELD_REQUEST_INIT;
ucp_params.features = 0;
ucp_params.request_size = sizeof(ucp_perf_request_t);
ucp_params.request_init = ucp_perf_request_init;
if (perf->params.thread_count > 1) {
/* when there is more than one thread, a ucp_worker would be created for
* each. all of them will share the same ucp_context */
ucp_params.features |= UCP_PARAM_FIELD_MT_WORKERS_SHARED;
ucp_params.mt_workers_shared = 1;
}
status = ucp_perf_test_fill_params(&perf->params, &ucp_params);
if (status != UCS_OK) {
goto err;
}
status = ucp_config_read(NULL, NULL, &config);
if (status != UCS_OK) {
goto err;
}
status = ucp_init(&ucp_params, config, &perf->ucp.context);
ucp_config_release(config);
if (status != UCS_OK) {
goto err;
}
thread_count = perf->params.thread_count;
message_size = ucx_perf_get_message_size(&perf->params);
status = ucp_perf_test_alloc_mem(perf);
if (status != UCS_OK) {
ucs_warn("ucp test failed to allocate memory");
goto err_cleanup;
}
perf->ucp.tctx = calloc(thread_count, sizeof(ucx_perf_thread_context_t));
if (perf->ucp.tctx == NULL) {
ucs_warn("ucp test failed to allocate memory for thread contexts");
goto err_free_mem;
}
worker_params.field_mask = UCP_WORKER_PARAM_FIELD_THREAD_MODE;
worker_params.thread_mode = perf->params.thread_mode;
for (i = 0; i < thread_count; i++) {
perf->ucp.tctx[i].tid = i;
perf->ucp.tctx[i].perf = *perf;
/* Doctor the src and dst buffers to make them thread specific */
perf->ucp.tctx[i].perf.send_buffer =
UCS_PTR_BYTE_OFFSET(perf->send_buffer, i * message_size);
perf->ucp.tctx[i].perf.recv_buffer =
UCS_PTR_BYTE_OFFSET(perf->recv_buffer, i * message_size);
status = ucp_worker_create(perf->ucp.context, &worker_params,
&perf->ucp.tctx[i].perf.ucp.worker);
if (status != UCS_OK) {
goto err_free_tctx_destroy_workers;
}
}
if (perf->params.command == UCX_PERF_CMD_AM) {
/* Check that requested AM header size is not larger than max supported. */
worker_attr.field_mask = UCP_WORKER_ATTR_FIELD_MAX_AM_HEADER;
status = ucp_worker_query(perf->ucp.tctx[0].perf.ucp.worker,
&worker_attr);
if (status != UCS_OK) {
goto err_free_tctx_destroy_workers;
}
if (worker_attr.max_am_header < perf->params.ucp.am_hdr_size) {
ucs_error("AM header size (%zu) is larger than max supported (%zu)",
perf->params.ucp.am_hdr_size, worker_attr.max_am_header);
status = UCS_ERR_INVALID_PARAM;
goto err_free_tctx_destroy_workers;
}
}
status = ucp_perf_test_setup_endpoints(perf, ucp_params.features);
if (status != UCS_OK) {
if (perf->params.flags & UCX_PERF_TEST_FLAG_VERBOSE) {
ucs_error("Failed to setup endpoints: %s", ucs_status_string(status));
}
goto err_free_tctx_destroy_workers;
}
return UCS_OK;
err_free_tctx_destroy_workers:
ucp_perf_test_destroy_workers(perf);
free(perf->ucp.tctx);
err_free_mem:
ucp_perf_test_free_mem(perf);
err_cleanup:
ucp_cleanup(perf->ucp.context);
err:
return status;
}
static void ucp_perf_cleanup(ucx_perf_context_t *perf)
{
ucp_perf_test_cleanup_endpoints(perf);
ucp_perf_barrier(perf);
ucp_perf_test_free_mem(perf);
ucp_perf_test_destroy_workers(perf);
free(perf->ucp.tctx);
ucp_cleanup(perf->ucp.context);
}
static struct {
ucs_status_t (*setup)(ucx_perf_context_t *perf);
void (*cleanup)(ucx_perf_context_t *perf);
ucs_status_t (*run)(ucx_perf_context_t *perf);
void (*barrier)(ucx_perf_context_t *perf);
} ucx_perf_funcs[] = {
[UCX_PERF_API_UCT] = {uct_perf_setup, uct_perf_cleanup,
uct_perf_test_dispatch, uct_perf_barrier},
[UCX_PERF_API_UCP] = {ucp_perf_setup, ucp_perf_cleanup,
ucp_perf_test_dispatch, ucp_perf_barrier}
};
static ucs_status_t ucx_perf_thread_spawn(ucx_perf_context_t *perf,
ucx_perf_result_t* result);
ucs_status_t ucx_perf_run(const ucx_perf_params_t *params,
ucx_perf_result_t *result)
{
ucx_perf_context_t *perf;
ucs_status_t status;
ucx_perf_global_init();
if (params->command == UCX_PERF_CMD_LAST) {
ucs_error("Test is not selected");
status = UCS_ERR_INVALID_PARAM;
goto out;
}
if ((params->api != UCX_PERF_API_UCT) && (params->api != UCX_PERF_API_UCP)) {
ucs_error("Invalid test API parameter (should be UCT or UCP)");
status = UCS_ERR_INVALID_PARAM;
goto out;
}
perf = malloc(sizeof(*perf));
if (perf == NULL) {
status = UCS_ERR_NO_MEMORY;
goto out;
}
ucx_perf_test_init(perf, params);
if (perf->allocator == NULL) {
ucs_error("Unsupported memory types %s<->%s",
ucs_memory_type_names[params->send_mem_type],
ucs_memory_type_names[params->recv_mem_type]);
status = UCS_ERR_UNSUPPORTED;
goto out_free;
}
if ((params->api == UCX_PERF_API_UCT) &&
(perf->allocator->mem_type != UCS_MEMORY_TYPE_HOST)) {
ucs_warn("UCT tests also copy 2-byte values from %s memory to "
"%s memory, which may impact performance results",
ucs_memory_type_names[perf->allocator->mem_type],
ucs_memory_type_names[UCS_MEMORY_TYPE_HOST]);
}
status = perf->allocator->init(perf);
if (status != UCS_OK) {
goto out_free;
}
status = ucx_perf_funcs[params->api].setup(perf);
if (status != UCS_OK) {
goto out_free;
}
if (params->thread_count == 1) {
if (params->api == UCX_PERF_API_UCP) {
perf->ucp.worker = perf->ucp.tctx[0].perf.ucp.worker;
perf->ucp.ep = perf->ucp.tctx[0].perf.ucp.ep;
perf->ucp.remote_addr = perf->ucp.tctx[0].perf.ucp.remote_addr;
perf->ucp.rkey = perf->ucp.tctx[0].perf.ucp.rkey;
}
if (params->warmup_iter > 0) {
ucx_perf_set_warmup(perf, params);
status = ucx_perf_funcs[params->api].run(perf);
if (status != UCS_OK) {
goto out_cleanup;
}
ucx_perf_funcs[params->api].barrier(perf);
ucx_perf_test_prepare_new_run(perf, params);
}
/* Run test */
status = ucx_perf_funcs[params->api].run(perf);
ucx_perf_funcs[params->api].barrier(perf);
if (status == UCS_OK) {
ucx_perf_calc_result(perf, result);
rte_call(perf, report, result, perf->params.report_arg, 1, 0);
}
} else {
status = ucx_perf_thread_spawn(perf, result);
}
out_cleanup:
ucx_perf_funcs[params->api].cleanup(perf);
out_free:
free(perf);
out:
return status;
}
#if _OPENMP
static ucs_status_t ucx_perf_thread_run_test(void* arg)
{
ucx_perf_thread_context_t* tctx = (ucx_perf_thread_context_t*) arg; /* a single thread context */
ucx_perf_result_t* result = &tctx->result;
ucx_perf_context_t* perf = &tctx->perf;
ucx_perf_params_t* params = &perf->params;
ucs_status_t status;
/* new threads need explicit device association */
status = perf->allocator->init(perf);
if (status != UCS_OK) {
goto out;
}
if (params->warmup_iter > 0) {
ucx_perf_set_warmup(perf, params);
status = ucx_perf_funcs[params->api].run(perf);
ucx_perf_funcs[params->api].barrier(perf);
if (UCS_OK != status) {
goto out;
}
ucx_perf_test_prepare_new_run(perf, params);
}
/* Run test */
#pragma omp barrier
status = ucx_perf_funcs[params->api].run(perf);
ucx_perf_funcs[params->api].barrier(perf);
if (UCS_OK != status) {
goto out;
}
ucx_perf_calc_result(perf, result);
out:
return status;
}
static void ucx_perf_thread_report_aggregated_results(ucx_perf_context_t *perf)
{
ucx_perf_thread_context_t* tctx = perf->ucp.tctx; /* all the thread contexts on perf */
unsigned i, thread_count = perf->params.thread_count;
double lat_sum_total_avegare = 0.0;
ucx_perf_result_t agg_result;
agg_result.iters = tctx[0].result.iters;
agg_result.bytes = tctx[0].result.bytes;
agg_result.elapsed_time = tctx[0].result.elapsed_time;
agg_result.bandwidth.total_average = 0.0;
agg_result.bandwidth.typical = 0.0; /* Undefined since used only for latency calculations */
agg_result.latency.total_average = 0.0;
agg_result.msgrate.total_average = 0.0;
agg_result.msgrate.typical = 0.0; /* Undefined since used only for latency calculations */
/* when running with multiple threads, the moment average value is
* undefined since we don't capture the values of the last iteration */
agg_result.msgrate.moment_average = 0.0;
agg_result.bandwidth.moment_average = 0.0;
agg_result.latency.moment_average = 0.0;
agg_result.latency.typical = 0.0;
/* in case of multiple threads, we have to aggregate the results so that the
* final output of the result would show the performance numbers that were
* collected from all the threads.
* BW and message rate values will be the sum of their values from all
* the threads, while the latency value is the average latency from the
* threads. */
for (i = 0; i < thread_count; i++) {
agg_result.bandwidth.total_average += tctx[i].result.bandwidth.total_average;
agg_result.msgrate.total_average += tctx[i].result.msgrate.total_average;
lat_sum_total_avegare += tctx[i].result.latency.total_average;
}
agg_result.latency.total_average = lat_sum_total_avegare / thread_count;
rte_call(perf, report, &agg_result, perf->params.report_arg, 1, 1);
}
static ucs_status_t ucx_perf_thread_spawn(ucx_perf_context_t *perf,
ucx_perf_result_t* result)
{
ucx_perf_thread_context_t* tctx = perf->ucp.tctx; /* all the thread contexts on perf */
int ti, thread_count = perf->params.thread_count;
ucs_status_t* statuses;
ucs_status_t status;
omp_set_num_threads(thread_count);
statuses = calloc(thread_count, sizeof(ucs_status_t));
if (statuses == NULL) {
status = UCS_ERR_NO_MEMORY;
goto out;
}
#pragma omp parallel private(ti)
{
ti = omp_get_thread_num();
tctx[ti].status = ucx_perf_thread_run_test((void*)&tctx[ti]);
}
status = UCS_OK;
for (ti = 0; ti < thread_count; ti++) {
if (UCS_OK != tctx[ti].status) {
ucs_error("Thread %d failed to run test: %s", tctx[ti].tid,
ucs_status_string(tctx[ti].status));
status = tctx[ti].status;
}
}
ucx_perf_thread_report_aggregated_results(perf);
free(statuses);
out:
return status;
}
#else
static ucs_status_t ucx_perf_thread_spawn(ucx_perf_context_t *perf,
ucx_perf_result_t* result) {
ucs_error("Invalid test parameter (thread mode requested without OpenMP capabilities)");
return UCS_ERR_INVALID_PARAM;
}
#endif /* _OPENMP */
void ucx_perf_global_init()
{
static ucx_perf_allocator_t host_allocator = {
.mem_type = UCS_MEMORY_TYPE_HOST,
.init = ucs_empty_function_return_success,
.ucp_alloc = ucp_perf_test_alloc_host,
.ucp_free = ucp_perf_test_free_host,
.uct_alloc = uct_perf_test_alloc_host,
.uct_free = uct_perf_test_free_host,
.memcpy = ucx_perf_test_memcpy_host,
.memset = memset
};
UCS_MODULE_FRAMEWORK_DECLARE(ucx_perftest);
ucx_perf_mem_type_allocators[UCS_MEMORY_TYPE_HOST] = &host_allocator;
/* FIXME Memtype allocator modules must be loaded to global scope, otherwise
* alloc hooks, which are using dlsym() to get pointer to original function,
* do not work. Need to use bistro for memtype hooks to fix it.
*/
UCS_MODULE_FRAMEWORK_LOAD(ucx_perftest, UCS_MODULE_LOAD_FLAG_GLOBAL);
}
|
GB_unaryop__ainv_int16_int8.c | //------------------------------------------------------------------------------
// GB_unaryop: hard-coded functions for each built-in unary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2019, All Rights Reserved.
// http://suitesparse.com See GraphBLAS/Doc/License.txt for license.
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_iterator.h"
#include "GB_unaryop__include.h"
// C=unop(A) is defined by the following types and operators:
// op(A) function: GB_unop__ainv_int16_int8
// op(A') function: GB_tran__ainv_int16_int8
// C type: int16_t
// A type: int8_t
// cast: int16_t cij = (int16_t) aij
// unaryop: cij = -aij
#define GB_ATYPE \
int8_t
#define GB_CTYPE \
int16_t
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int8_t aij = Ax [pA]
#define GB_CX(p) Cx [p]
// unary operator
#define GB_OP(z, x) \
z = -x ;
// casting
#define GB_CASTING(z, x) \
int16_t z = (int16_t) x ;
// cij = op (cast (aij))
#define GB_CAST_OP(pC,pA) \
{ \
/* aij = Ax [pA] */ \
GB_GETA (aij, Ax, pA) ; \
/* Cx [pC] = op (cast (aij)) */ \
GB_CASTING (x, aij) ; \
GB_OP (GB_CX (pC), x) ; \
}
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_AINV || GxB_NO_INT16 || GxB_NO_INT8)
//------------------------------------------------------------------------------
// Cx = op (cast (Ax)): apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB_unop__ainv_int16_int8
(
int16_t *restrict Cx,
const int8_t *restrict Ax,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (int64_t p = 0 ; p < anz ; p++)
{
GB_CAST_OP (p, p) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (cast (A')): transpose, typecast, and apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB_tran__ainv_int16_int8
(
GrB_Matrix C,
const GrB_Matrix A,
int64_t *restrict *Rowcounts,
GBI_single_iterator Iter,
const int64_t *restrict A_slice,
int naslice
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#define GB_PHASE_2_OF_2
#include "GB_unaryop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
thr_omp.h | /* -*- c++ -*- -------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
Copyright (2003) Sandia Corporation. Under the terms of Contract
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
certain rights in this software. This software is distributed under
the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Axel Kohlmeyer (Temple U)
------------------------------------------------------------------------- */
#ifndef LMP_THR_OMP_H
#define LMP_THR_OMP_H
#include "pointers.h"
#include "error.h"
#include "fix_omp.h"
#include "thr_data.h"
namespace LAMMPS_NS {
// forward declarations
class Pair;
class Bond;
class Angle;
class Dihedral;
class Improper;
class KSpace;
class Fix;
class ThrOMP {
protected:
LAMMPS *lmp; // reference to base lammps object.
FixOMP *fix; // pointer to fix_omp;
const int thr_style;
int thr_error;
public:
ThrOMP(LAMMPS *, int);
virtual ~ThrOMP();
double memory_usage_thr();
inline void sync_threads() {
#if defined(_OPENMP)
#pragma omp barrier
#endif
{ ; }
};
enum {THR_NONE=0,THR_PAIR=1,THR_BOND=1<<1,THR_ANGLE=1<<2,
THR_DIHEDRAL=1<<3,THR_IMPROPER=1<<4,THR_KSPACE=1<<5,
THR_CHARMM=1<<6, /*THR_PROXY=1<<7,THR_HYBRID=1<<8, */
THR_FIX=1<<9,THR_INTGR=1<<10};
protected:
// extra ev_tally setup work for threaded styles
void ev_setup_thr(int, int, int, double *, double **, ThrData *);
// compute global per thread virial contribution from per-thread force
void virial_fdotr_compute_thr(double * const, const double * const * const,
const double * const * const,
const int, const int, const int);
// reduce per thread data as needed
void reduce_thr(void * const style, const int eflag, const int vflag,
ThrData * const thr);
// thread safe variant error abort support.
// signals an error condition in any thread by making
// thr_error > 0, if condition "cond" is true.
// will abort from thread 0 if thr_error is > 0
// otherwise return true.
// returns false if no error found on any thread.
// use return value to jump/return to end of threaded region.
bool check_error_thr(const bool cond, const int tid, const char *fname,
const int line, const char *errmsg) {
if (cond) {
#if defined(_OPENMP)
#pragma omp atomic
++thr_error;
#endif
if (tid > 0) return true;
else lmp->error->one(fname,line,errmsg);
} else {
if (thr_error > 0) {
if (tid == 0) lmp->error->one(fname,line,errmsg);
else return true;
} else return false;
}
return false;
};
protected:
// threading adapted versions of the ev_tally infrastructure
// style specific versions (need access to style class flags)
// Pair
void e_tally_thr(Pair * const, const int, const int, const int,
const int, const double, const double, ThrData * const);
void v_tally_thr(Pair * const, const int, const int, const int,
const int, const double * const, ThrData * const);
void ev_tally_thr(Pair * const, const int, const int, const int, const int,
const double, const double, const double, const double,
const double, const double, ThrData * const);
void ev_tally_xyz_thr(Pair * const, const int, const int, const int,
const int, const double, const double, const double,
const double, const double, const double,
const double, const double, ThrData * const);
void ev_tally_xyz_full_thr(Pair * const, const int, const double, const double,
const double, const double, const double,
const double, const double, const double, ThrData * const);
void ev_tally3_thr(Pair * const, const int, const int, const int, const double,
const double, const double * const, const double * const,
const double * const, const double * const, ThrData * const);
void ev_tally4_thr(Pair * const, const int, const int, const int, const int,
const double, const double * const, const double * const,
const double * const, const double * const, const double * const,
const double * const, ThrData * const);
// Bond
void ev_tally_thr(Bond * const, const int, const int, const int, const int,
const double, const double, const double, const double,
const double, ThrData * const);
// Angle
void ev_tally_thr(Angle * const, const int, const int, const int, const int, const int,
const double, const double * const, const double * const,
const double, const double, const double, const double, const double,
const double, ThrData * const thr);
void ev_tally13_thr(Angle * const, const int, const int, const int, const int,
const double, const double, const double, const double,
const double, ThrData * const thr);
// Dihedral
void ev_tally_thr(Dihedral * const, const int, const int, const int, const int, const int,
const int, const double, const double * const, const double * const,
const double * const, const double, const double, const double,
const double, const double, const double, const double, const double,
const double, ThrData * const);
// Improper
void ev_tally_thr(Improper * const, const int, const int, const int, const int, const int,
const int, const double, const double * const, const double * const,
const double * const, const double, const double, const double,
const double, const double, const double, const double, const double,
const double, ThrData * const);
// style independent versions
void v_tally2_thr(const int, const int, const double, const double * const, ThrData * const);
void v_tally3_thr(const int, const int, const int, const double * const, const double * const,
const double * const, const double * const, ThrData * const);
void v_tally4_thr(const int, const int, const int, const int, const double * const,
const double * const, const double * const, const double * const,
const double * const, const double * const, ThrData * const);
void ev_tally_list_thr(Pair * const, const int, const int * const,
const double * const, const double, const double,
ThrData * const);
};
// set loop range thread id, and force array offset for threaded runs.
static inline void loop_setup_thr(int &ifrom, int &ito, int &tid,
int inum, int nthreads)
{
#if defined(_OPENMP)
tid = omp_get_thread_num();
// each thread works on a fixed chunk of atoms.
const int idelta = 1 + inum/nthreads;
ifrom = tid*idelta;
ito = ((ifrom + idelta) > inum) ? inum : ifrom + idelta;
#else
tid = 0;
ifrom = 0;
ito = inum;
nthreads = 1;
#endif
}
// helpful definitions to help compilers optimizing code better
typedef struct { double x,y,z; } dbl3_t;
typedef struct { double x,y,z,w; } dbl4_t;
typedef struct { int a,b,t; } int3_t;
typedef struct { int a,b,c,t; } int4_t;
typedef struct { int a,b,c,d,t; } int5_t;
}
#endif
|
convolutiondepthwise_3x3_int8.h | // BUG1989 is pleased to support the open source community by supporting ncnn available.
//
// Copyright (C) 2019 BUG1989. All rights reserved.
//
// Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
// in compliance with the License. You may obtain a copy of the License at
//
// https://opensource.org/licenses/BSD-3-Clause
//
// Unless required by applicable law or agreed to in writing, software distributed
// under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
// CONDITIONS OF ANY KIND, either express or implied. See the License for the
// specific language governing permissions and limitations under the License.
static inline signed char float2int8(float v)
{
int int32 = round(v);
if (int32 > 127) return 127;
if (int32 < -128) return -128;
return (signed char)int32;
}
static void convdw3x3s1_int8_neon(const Mat &bottom_blob, Mat &top_blob, const Mat &_kernel, const Option& opt)
{
int w = bottom_blob.w;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = 0; p < outch; p++)
{
Mat out = top_blob.channel(p);
const signed char* kernel = (const signed char *)_kernel + p*9;
int* outptr0 = out;
int* outptr0n = outptr0 + outw;
const signed char* img0 = bottom_blob.channel(p);
const signed char* r0 = img0;
const signed char* r1 = img0 + w;
const signed char* r2 = img0 + w*2;
const signed char* r3 = img0 + w*3;
int i = 0;
#if __ARM_NEON
int8x16_t _k0123456789x = vld1q_s8(kernel);
int16x8_t _k_s16 = vmovl_s8(vget_low_s8(_k0123456789x));
int16x8_t _kn_s16 = vmovl_s8(vget_high_s8(_k0123456789x));
int16x4_t _k0123 = vget_low_s16(_k_s16);
int16x4_t _k4567 = vget_high_s16(_k_s16);
int16x4_t _k8xxx = vget_low_s16(_kn_s16);
#endif // __ARM_NEON
for (; i+1 < outh; i+=2)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"0: \n"
"ld1 {v4.8b, v5.8b}, [%3] \n"
"ld1 {v6.8b, v7.8b}, [%4] \n"
"ld1 {v8.8b, v9.8b}, [%5] \n"
"ld1 {v10.8b, v11.8b}, [%6] \n"
"add %3, %3, #8 \n"
"add %4, %4, #8 \n"
"add %5, %5, #8 \n"
"add %6, %6, #8 \n"
"ext v12.8b, v4.8b, v5.8b, #1 \n"
"ext v13.8b, v4.8b, v5.8b, #2 \n"
"ext v14.8b, v6.8b, v7.8b, #1 \n"
"ext v15.8b, v6.8b, v7.8b, #2 \n"
"ext v16.8b, v8.8b, v9.8b, #1 \n"
"ext v17.8b, v8.8b, v9.8b, #2 \n"
"ext v18.8b, v10.8b, v11.8b, #1 \n"
"ext v19.8b, v10.8b, v11.8b, #2 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v12.8h, v12.8b, #0 \n"// r01
"sshll v13.8h, v13.8b, #0 \n"// r02
"sshll v6.8h, v6.8b, #0 \n"// r10
"sshll v14.8h, v14.8b, #0 \n"// r11
"sshll v15.8h, v15.8b, #0 \n"// r12
"sshll v8.8h, v8.8b, #0 \n"// r20
"sshll v16.8h, v16.8b, #0 \n"// r21
"sshll v17.8h, v17.8b, #0 \n"// r22
"sshll v10.8h, v10.8b, #0 \n"// r30
"sshll v18.8h, v18.8b, #0 \n"// r31
"sshll v19.8h, v19.8b, #0 \n"// r32
// r0
"smull v20.4s, v4.4h, %14.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %14.h[0] \n"
"smull v22.4s, v12.4h, %14.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v12.8h, %14.h[1] \n"
"smull v24.4s, v13.4h, %14.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v13.8h, %14.h[2] \n"
// r1
"smull v26.4s, v6.4h, %14.h[0] \n"// (r10 - r17) * k00
"smull2 v27.4s, v6.8h, %14.h[0] \n"
"smull v28.4s, v14.4h, %14.h[1] \n"// (r11 - r18) * k01
"smull2 v29.4s, v14.8h, %14.h[1] \n"
"smull v30.4s, v15.4h, %14.h[2] \n"// (r12 - r19) * k02
"smull2 v31.4s, v15.8h, %14.h[2] \n"
"smlal v20.4s, v6.4h, %14.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v6.8h, %14.h[3] \n"
"smlal v22.4s, v14.4h, %15.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v14.8h, %15.h[0] \n"
"smlal v24.4s, v15.4h, %15.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v15.8h, %15.h[1] \n"
// r2
"smlal v26.4s, v8.4h, %14.h[3] \n"// (r20 - r27) * k03
"smlal2 v27.4s, v8.8h, %14.h[3] \n"
"smlal v28.4s, v16.4h, %15.h[0] \n"// (r21 - r28) * k04
"smlal2 v29.4s, v16.8h, %15.h[0] \n"
"smlal v30.4s, v17.4h, %15.h[1] \n"// (r22 - r29) * k05
"smlal2 v31.4s, v17.8h, %15.h[1] \n"
"smlal v20.4s, v8.4h, %15.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v8.8h, %15.h[2] \n"
"smlal v22.4s, v16.4h, %15.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v16.8h, %15.h[3] \n"
"smlal v24.4s, v17.4h, %16.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v17.8h, %16.h[0] \n"
// r3
"smlal v26.4s, v10.4h, %15.h[2] \n"// (r30 - r37) * k06
"smlal2 v27.4s, v10.8h, %15.h[2] \n"
"smlal v28.4s, v18.4h, %15.h[3] \n"// (r31 - r38) * k07
"smlal2 v29.4s, v18.8h, %15.h[3] \n"
"smlal v30.4s, v19.4h, %16.h[0] \n"// (r32 - r39) * k08
"smlal2 v31.4s, v19.8h, %16.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v26.4s, v26.4s, v28.4s \n"
"add v27.4s, v27.4s, v29.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
"add v26.4s, v26.4s, v30.4s \n"
"add v27.4s, v27.4s, v31.4s \n"
"st1 {v20.4s, v21.4s}, [%1], #32 \n"
"st1 {v26.4s, v27.4s}, [%2], #32 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr0n), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2), // %5
"=r"(r3) // %6
: "0"(nn),
"1"(outptr0),
"2"(outptr0n),
"3"(r0),
"4"(r1),
"5"(r2),
"6"(r3),
"w"(_k0123), // %14
"w"(_k4567), // %15
"w"(_k8xxx) // %16
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld1.s8 {d30-d31}, [%3] \n"// r0
"add %3, %3, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q5, d10 \n"// r01
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P14[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P14[0] \n"
"vmull.s16 q9, d10, %P14[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P14[1] \n"
"vmlal.s16 q7, d12, %P14[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P14[2] \n"
// r1
"vld1.s8 {d30-d31}, [%4] \n"// r1
"add %4, %4, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q5, d10 \n"// r11
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P14[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P14[3] \n"
"vmlal.s16 q9, d10, %P15[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P15[0] \n"
"vmlal.s16 q7, d12, %P15[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P15[1] \n"
// sum1
"vmull.s16 q11, d30, %P14[0] \n"// (r10 - r17) * k00
"vmull.s16 q12, d31, %P14[0] \n"
"vmull.s16 q13, d10, %P14[1] \n"// (r11 - r18) * k01
"vmull.s16 q14, d11, %P14[1] \n"
"vmlal.s16 q11, d12, %P14[2] \n"// (r12 - r19) * k02
"vmlal.s16 q12, d13, %P14[2] \n"
// r2
"vld1.s8 {d30-d31}, [%5] \n"// r2
"add %5, %5, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q5, d10 \n"// r21
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P15[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P15[2] \n"
"vmlal.s16 q9, d10, %P15[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P15[3] \n"
"vmlal.s16 q7, d12, %P16[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P16[0] \n"
// sum1
"vmlal.s16 q11, d30, %P14[3] \n"// (r20 - r27) * k03
"vmlal.s16 q12, d31, %P14[3] \n"
"vmlal.s16 q13, d10, %P15[0] \n"// (r21 - r28) * k04
"vmlal.s16 q14, d11, %P15[0] \n"
"vmlal.s16 q11, d12, %P15[1] \n"// (r22 - r29) * k05
"vmlal.s16 q12, d13, %P15[1] \n"
// r3
"vld1.s8 {d30-d31}, [%6] \n"// r3
"add %6, %6, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r30
"vmovl.s8 q5, d10 \n"// r31
"vmovl.s8 q6, d12 \n"// r32
// sum1
"vmlal.s16 q11, d30, %P15[2] \n"// (r30 - r37) * k06
"vmlal.s16 q12, d31, %P15[2] \n"
"vmlal.s16 q13, d10, %P15[3] \n"// (r31 - r38) * k07
"vmlal.s16 q14, d11, %P15[3] \n"
"vmlal.s16 q11, d12, %P16[0] \n"// (r32 - r39) * k08
"vmlal.s16 q12, d13, %P16[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vadd.s32 q11, q11, q13 \n"
"vadd.s32 q12, q12, q14 \n"
"vst1.s32 {d14-d17}, [%1]! \n"
"vst1.s32 {d22-d25}, [%2]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr0n), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2), // %5
"=r"(r3) // %6
: "0"(nn),
"1"(outptr0),
"2"(outptr0n),
"3"(r0),
"4"(r1),
"5"(r2),
"6"(r3),
"w"(_k0123), // %14
"w"(_k4567), // %15
"w"(_k8xxx) // %16
: "cc", "memory", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
// TODO NEON
int sum0 = 0;
int sum0n = 0;
sum0 += (int)r0[0] * kernel[0];
sum0 += (int)r0[1] * kernel[1];
sum0 += (int)r0[2] * kernel[2];
sum0 += (int)r1[0] * kernel[3];
sum0 += (int)r1[1] * kernel[4];
sum0 += (int)r1[2] * kernel[5];
sum0 += (int)r2[0] * kernel[6];
sum0 += (int)r2[1] * kernel[7];
sum0 += (int)r2[2] * kernel[8];
sum0n += (int)r1[0] * kernel[0];
sum0n += (int)r1[1] * kernel[1];
sum0n += (int)r1[2] * kernel[2];
sum0n += (int)r2[0] * kernel[3];
sum0n += (int)r2[1] * kernel[4];
sum0n += (int)r2[2] * kernel[5];
sum0n += (int)r3[0] * kernel[6];
sum0n += (int)r3[1] * kernel[7];
sum0n += (int)r3[2] * kernel[8];
*outptr0 = sum0;
*outptr0n = sum0n;
r0++;
r1++;
r2++;
r3++;
outptr0++;
outptr0n++;
}
r0 += 2 + w;
r1 += 2 + w;
r2 += 2 + w;
r3 += 2 + w;
outptr0 += outw;
outptr0n += outw;
}
for (; i < outh; i++)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"0: \n"
"ld1 {v4.8b, v5.8b}, [%2] \n"
"ld1 {v6.8b, v7.8b}, [%3] \n"
"ld1 {v8.8b, v9.8b}, [%4] \n"
"add %2, %2, #8 \n"
"add %3, %3, #8 \n"
"add %4, %4, #8 \n"
"ext v12.8b, v4.8b, v5.8b, #1 \n"
"ext v13.8b, v4.8b, v5.8b, #2 \n"
"ext v14.8b, v6.8b, v7.8b, #1 \n"
"ext v15.8b, v6.8b, v7.8b, #2 \n"
"ext v16.8b, v8.8b, v9.8b, #1 \n"
"ext v17.8b, v8.8b, v9.8b, #2 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v12.8h, v12.8b, #0 \n"// r01
"sshll v13.8h, v13.8b, #0 \n"// r02
"sshll v6.8h, v6.8b, #0 \n"// r10
"sshll v14.8h, v14.8b, #0 \n"// r11
"sshll v15.8h, v15.8b, #0 \n"// r12
"sshll v8.8h, v8.8b, #0 \n"// r20
"sshll v16.8h, v16.8b, #0 \n"// r21
"sshll v17.8h, v17.8b, #0 \n"// r22
// r0
"smull v20.4s, v4.4h, %10.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %10.h[0] \n"
"smull v22.4s, v12.4h, %10.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v12.8h, %10.h[1] \n"
"smull v24.4s, v13.4h, %10.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v13.8h, %10.h[2] \n"
// r1
"smlal v20.4s, v6.4h, %10.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v6.8h, %10.h[3] \n"
"smlal v22.4s, v14.4h, %11.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v14.8h, %11.h[0] \n"
"smlal v24.4s, v15.4h, %11.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v15.8h, %11.h[1] \n"
// r2
"smlal v20.4s, v8.4h, %11.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v8.8h, %11.h[2] \n"
"smlal v22.4s, v16.4h, %11.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v16.8h, %11.h[3] \n"
"smlal v24.4s, v17.4h, %12.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v17.8h, %12.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
"st1 {v20.4s, v21.4s}, [%1], #32 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx) // %12
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld1.s8 {d30-d31}, [%2] \n"// r0
"add %2, %2, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q5, d10 \n"// r01
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P10[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P10[0] \n"
"vmull.s16 q9, d10, %P10[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P10[1] \n"
"vmlal.s16 q7, d12, %P10[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P10[2] \n"
// r1
"vld1.s8 {d30-d31}, [%3] \n"// r1
"add %3, %3, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q5, d10 \n"// r11
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P10[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P10[3] \n"
"vmlal.s16 q9, d10, %P11[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P11[0] \n"
"vmlal.s16 q7, d12, %P11[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P11[1] \n"
// r2
"vld1.s8 {d30-d31}, [%4] \n"// r2
"add %4, %4, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q5, d10 \n"// r21
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P11[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P11[2] \n"
"vmlal.s16 q9, d10, %P11[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P11[3] \n"
"vmlal.s16 q7, d12, %P12[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P12[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vst1.s32 {d14-d17}, [%1]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx) // %12
: "cc", "memory", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
int sum = 0;
sum += (int)r0[0] * kernel[0];
sum += (int)r0[1] * kernel[1];
sum += (int)r0[2] * kernel[2];
sum += (int)r1[0] * kernel[3];
sum += (int)r1[1] * kernel[4];
sum += (int)r1[2] * kernel[5];
sum += (int)r2[0] * kernel[6];
sum += (int)r2[1] * kernel[7];
sum += (int)r2[2] * kernel[8];
*outptr0 = sum;
r0++;
r1++;
r2++;
outptr0++;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
}
}
static void convdw3x3s2_int8_neon(const Mat &bottom_blob, Mat &top_blob, const Mat &_kernel, const Option& opt)
{
int w = bottom_blob.w;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
const int tailstep = w - 2*outw + w;
#pragma omp parallel for num_threads(opt.num_threads)
for (int p=0; p<outch; p++)
{
Mat out = top_blob.channel(p);
const signed char* kernel = (const signed char*)_kernel + p*9;
int* outptr = out;
const signed char* img = bottom_blob.channel(p);
const signed char* r0 = img;
const signed char* r1 = img + w;
const signed char* r2 = img + w*2;
int i = 0;
#if __ARM_NEON
int8x16_t _k0123456789x = vld1q_s8(kernel);
int16x8_t _k_s16 = vmovl_s8(vget_low_s8(_k0123456789x));
int16x8_t _kn_s16 = vmovl_s8(vget_high_s8(_k0123456789x));
int16x4_t _k0123 = vget_low_s16(_k_s16);
int16x4_t _k4567 = vget_high_s16(_k_s16);
int16x4_t _k8xxx = vget_low_s16(_kn_s16);
#endif // __ARM_NEON
for (; i < outh; i++)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"0: \n"
"ld2 {v4.8b, v5.8b}, [%2], #16 \n"
"ld2 {v6.8b, v7.8b}, [%2] \n"
"ld2 {v8.8b, v9.8b}, [%3], #16 \n"
"ld2 {v10.8b, v11.8b}, [%3] \n"
"ld2 {v12.8b, v13.8b}, [%4], #16 \n"
"ld2 {v14.8b, v15.8b}, [%4] \n"
"ext v6.8b, v4.8b, v6.8b, #1 \n"
"ext v10.8b, v8.8b, v10.8b, #1 \n"
"ext v14.8b, v12.8b, v14.8b, #1 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v5.8h, v5.8b, #0 \n"// r01
"sshll v6.8h, v6.8b, #0 \n"// r02
"sshll v8.8h, v8.8b, #0 \n"// r10
"sshll v9.8h, v9.8b, #0 \n"// r11
"sshll v10.8h, v10.8b, #0 \n"// r12
"sshll v12.8h, v12.8b, #0 \n"// r20
"sshll v13.8h, v13.8b, #0 \n"// r21
"sshll v14.8h, v14.8b, #0 \n"// r22
// r0
"smull v20.4s, v4.4h, %10.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %10.h[0] \n"
"smull v22.4s, v5.4h, %10.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v5.8h, %10.h[1] \n"
"smull v24.4s, v6.4h, %10.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v6.8h, %10.h[2] \n"
// r1
"smlal v20.4s, v8.4h, %10.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v8.8h, %10.h[3] \n"
"smlal v22.4s, v9.4h, %11.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v9.8h, %11.h[0] \n"
"smlal v24.4s, v10.4h, %11.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v10.8h, %11.h[1] \n"
// r2
"smlal v20.4s, v12.4h, %11.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v12.8h, %11.h[2] \n"
"smlal v22.4s, v13.4h, %11.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v13.8h, %11.h[3] \n"
"smlal v24.4s, v14.4h, %12.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v14.8h, %12.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
"st1 {v20.4s, v21.4s}, [%1], #32 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx) // %12
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld2.s8 {d30-d31}, [%2]! \n"// r0
"vld2.s8 {d10-d11}, [%2] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r01
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P10[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P10[0] \n"
"vmull.s16 q9, d10, %P10[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P10[1] \n"
"vmlal.s16 q7, d12, %P10[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P10[2] \n"
// r1
"vld2.s8 {d30-d31}, [%3]! \n"// r1
"vld2.s8 {d10-d11}, [%3] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r11
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P10[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P10[3] \n"
"vmlal.s16 q9, d10, %P11[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P11[0] \n"
"vmlal.s16 q7, d12, %P11[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P11[1] \n"
// r2
"vld2.s8 {d30-d31}, [%4]! \n"// r2
"vld2.s8 {d10-d11}, [%4] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r21
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P11[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P11[2] \n"
"vmlal.s16 q9, d10, %P11[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P11[3] \n"
"vmlal.s16 q7, d12, %P12[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P12[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vst1.s32 {d14-d17}, [%1]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx) // %12
: "cc", "memory", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
int sum = 0;
sum += (int)r0[0] * kernel[0];
sum += (int)r0[1] * kernel[1];
sum += (int)r0[2] * kernel[2];
sum += (int)r1[0] * kernel[3];
sum += (int)r1[1] * kernel[4];
sum += (int)r1[2] * kernel[5];
sum += (int)r2[0] * kernel[6];
sum += (int)r2[1] * kernel[7];
sum += (int)r2[2] * kernel[8];
*outptr = sum;
r0 += 2;
r1 += 2;
r2 += 2;
outptr++;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
}
}
static void convdw3x3s1_int8_requant_neon(const Mat &bottom_blob, Mat &top_blob, const Mat &_kernel, const Mat &_bias, std::vector<float> scales_requant, const Option& opt)
{
int w = bottom_blob.w;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
const float* bias = _bias;
#pragma omp parallel for num_threads(opt.num_threads)
for (int p = 0; p < outch; p++)
{
Mat out = top_blob.channel(p);
const float bias0 = bias ? bias[p] : 0.f;
const float scale_requant_in = scales_requant[2*p];
const float scale_requant_out = scales_requant[2*p+1];
const signed char* kernel = (const signed char *)_kernel + p*9;
signed char* outptr0 = out;
signed char* outptr0n = outptr0 + outw;
const signed char* img0 = bottom_blob.channel(p);
const signed char* r0 = img0;
const signed char* r1 = img0 + w;
const signed char* r2 = img0 + w*2;
const signed char* r3 = img0 + w*3;
int i = 0;
#if __ARM_NEON
int8x16_t _k0123456789x = vld1q_s8(kernel);
int16x8_t _k_s16 = vmovl_s8(vget_low_s8(_k0123456789x));
int16x8_t _kn_s16 = vmovl_s8(vget_high_s8(_k0123456789x));
int16x4_t _k0123 = vget_low_s16(_k_s16);
int16x4_t _k4567 = vget_high_s16(_k_s16);
int16x4_t _k8xxx = vget_low_s16(_kn_s16);
#endif // __ARM_NEON
for (; i+1 < outh; i+=2)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"0: \n"
"ld1 {v4.8b, v5.8b}, [%3] \n"
"ld1 {v6.8b, v7.8b}, [%4] \n"
"ld1 {v8.8b, v9.8b}, [%5] \n"
"ld1 {v10.8b, v11.8b}, [%6] \n"
"add %3, %3, #8 \n"
"add %4, %4, #8 \n"
"add %5, %5, #8 \n"
"add %6, %6, #8 \n"
"ext v12.8b, v4.8b, v5.8b, #1 \n"
"ext v13.8b, v4.8b, v5.8b, #2 \n"
"ext v14.8b, v6.8b, v7.8b, #1 \n"
"ext v15.8b, v6.8b, v7.8b, #2 \n"
"ext v16.8b, v8.8b, v9.8b, #1 \n"
"ext v17.8b, v8.8b, v9.8b, #2 \n"
"ext v18.8b, v10.8b, v11.8b, #1 \n"
"ext v19.8b, v10.8b, v11.8b, #2 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v12.8h, v12.8b, #0 \n"// r01
"sshll v13.8h, v13.8b, #0 \n"// r02
"sshll v6.8h, v6.8b, #0 \n"// r10
"sshll v14.8h, v14.8b, #0 \n"// r11
"sshll v15.8h, v15.8b, #0 \n"// r12
"sshll v8.8h, v8.8b, #0 \n"// r20
"sshll v16.8h, v16.8b, #0 \n"// r21
"sshll v17.8h, v17.8b, #0 \n"// r22
"sshll v10.8h, v10.8b, #0 \n"// r30
"sshll v18.8h, v18.8b, #0 \n"// r31
"sshll v19.8h, v19.8b, #0 \n"// r32
// r0
"smull v20.4s, v4.4h, %14.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %14.h[0] \n"
"smull v22.4s, v12.4h, %14.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v12.8h, %14.h[1] \n"
"smull v24.4s, v13.4h, %14.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v13.8h, %14.h[2] \n"
// r1
"smull v26.4s, v6.4h, %14.h[0] \n"// (r10 - r17) * k00
"smull2 v27.4s, v6.8h, %14.h[0] \n"
"smull v28.4s, v14.4h, %14.h[1] \n"// (r11 - r18) * k01
"smull2 v29.4s, v14.8h, %14.h[1] \n"
"smull v30.4s, v15.4h, %14.h[2] \n"// (r12 - r19) * k02
"smull2 v31.4s, v15.8h, %14.h[2] \n"
"smlal v20.4s, v6.4h, %14.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v6.8h, %14.h[3] \n"
"smlal v22.4s, v14.4h, %15.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v14.8h, %15.h[0] \n"
"smlal v24.4s, v15.4h, %15.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v15.8h, %15.h[1] \n"
// r2
"smlal v26.4s, v8.4h, %14.h[3] \n"// (r20 - r27) * k03
"smlal2 v27.4s, v8.8h, %14.h[3] \n"
"smlal v28.4s, v16.4h, %15.h[0] \n"// (r21 - r28) * k04
"smlal2 v29.4s, v16.8h, %15.h[0] \n"
"smlal v30.4s, v17.4h, %15.h[1] \n"// (r22 - r29) * k05
"smlal2 v31.4s, v17.8h, %15.h[1] \n"
"smlal v20.4s, v8.4h, %15.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v8.8h, %15.h[2] \n"
"smlal v22.4s, v16.4h, %15.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v16.8h, %15.h[3] \n"
"smlal v24.4s, v17.4h, %16.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v17.8h, %16.h[0] \n"
// r3
"smlal v26.4s, v10.4h, %15.h[2] \n"// (r30 - r37) * k06
"smlal2 v27.4s, v10.8h, %15.h[2] \n"
"smlal v28.4s, v18.4h, %15.h[3] \n"// (r31 - r38) * k07
"smlal2 v29.4s, v18.8h, %15.h[3] \n"
"smlal v30.4s, v19.4h, %16.h[0] \n"// (r32 - r39) * k08
"smlal2 v31.4s, v19.8h, %16.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v26.4s, v26.4s, v28.4s \n"
"add v27.4s, v27.4s, v29.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
"add v26.4s, v26.4s, v30.4s \n"
"add v27.4s, v27.4s, v31.4s \n"
"dup v4.4s, %w17 \n" // bias
"dup v5.4s, %w18 \n" // scale_in
"dup v6.4s, %w19 \n" // scale_out
// top_s32 -> top_f32
"scvtf v20.4s, v20.4s \n"
"scvtf v21.4s, v21.4s \n"
"scvtf v26.4s, v26.4s \n"
"scvtf v27.4s, v27.4s \n"
// top_f32 = top_f32 * scale_in
"fmul v20.4s, v20.4s, v5.4s \n"
"fmul v21.4s, v21.4s, v5.4s \n"
"fmul v26.4s, v26.4s, v5.4s \n"
"fmul v27.4s, v27.4s, v5.4s \n"
// top_f32 = top_f32 + bias
"fadd v20.4s, v20.4s, v4.4s \n"
"fadd v21.4s, v21.4s, v4.4s \n"
"fadd v26.4s, v26.4s, v4.4s \n"
"fadd v27.4s, v27.4s, v4.4s \n"
// top_f32 = top_f32 * scale_out
"fmul v20.4s, v20.4s, v6.4s \n"
"fmul v21.4s, v21.4s, v6.4s \n"
"fmul v26.4s, v26.4s, v6.4s \n"
"fmul v27.4s, v27.4s, v6.4s \n"
// top_f32 -> top_s32
"fcvtas v20.4s, v20.4s \n"
"fcvtas v21.4s, v21.4s \n"
"fcvtas v26.4s, v26.4s \n"
"fcvtas v27.4s, v27.4s \n"
// top_s32 -> top_s16
"sqxtn v7.4h, v20.4s \n"
"sqxtn v9.4h, v26.4s \n"
"sqxtn2 v7.8h, v21.4s \n"
"sqxtn2 v9.8h, v27.4s \n"
// top_s16 -> top_s8
"sqxtn v8.8b, v7.8h \n"
"sqxtn v10.8b, v9.8h \n"
// save top_s8
"st1 {v8.8b}, [%1], #8 \n"
"st1 {v10.8b}, [%2], #8 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr0n), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2), // %5
"=r"(r3) // %6
: "0"(nn),
"1"(outptr0),
"2"(outptr0n),
"3"(r0),
"4"(r1),
"5"(r2),
"6"(r3),
"w"(_k0123), // %14
"w"(_k4567), // %15
"w"(_k8xxx), // %16
"r"(bias0), // %17
"r"(scale_requant_in), // %18
"r"(scale_requant_out) // %19
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld1.s8 {d30-d31}, [%3] \n"// r0
"add %3, %3, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q5, d10 \n"// r01
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P14[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P14[0] \n"
"vmull.s16 q9, d10, %P14[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P14[1] \n"
"vmlal.s16 q7, d12, %P14[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P14[2] \n"
// r1
"vld1.s8 {d30-d31}, [%4] \n"// r1
"add %4, %4, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q5, d10 \n"// r11
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P14[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P14[3] \n"
"vmlal.s16 q9, d10, %P15[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P15[0] \n"
"vmlal.s16 q7, d12, %P15[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P15[1] \n"
// sum1
"vmull.s16 q11, d30, %P14[0] \n"// (r10 - r17) * k00
"vmull.s16 q12, d31, %P14[0] \n"
"vmull.s16 q13, d10, %P14[1] \n"// (r11 - r18) * k01
"vmull.s16 q14, d11, %P14[1] \n"
"vmlal.s16 q11, d12, %P14[2] \n"// (r12 - r19) * k02
"vmlal.s16 q12, d13, %P14[2] \n"
// r2
"vld1.s8 {d30-d31}, [%5] \n"// r2
"add %5, %5, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q5, d10 \n"// r21
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P15[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P15[2] \n"
"vmlal.s16 q9, d10, %P15[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P15[3] \n"
"vmlal.s16 q7, d12, %P16[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P16[0] \n"
// sum1
"vmlal.s16 q11, d30, %P14[3] \n"// (r20 - r27) * k03
"vmlal.s16 q12, d31, %P14[3] \n"
"vmlal.s16 q13, d10, %P15[0] \n"// (r21 - r28) * k04
"vmlal.s16 q14, d11, %P15[0] \n"
"vmlal.s16 q11, d12, %P15[1] \n"// (r22 - r29) * k05
"vmlal.s16 q12, d13, %P15[1] \n"
// r3
"vld1.s8 {d30-d31}, [%6] \n"// r3
"add %6, %6, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r30
"vmovl.s8 q5, d10 \n"// r31
"vmovl.s8 q6, d12 \n"// r32
// sum1
"vmlal.s16 q11, d30, %P15[2] \n"// (r30 - r37) * k06
"vmlal.s16 q12, d31, %P15[2] \n"
"vmlal.s16 q13, d10, %P15[3] \n"// (r31 - r38) * k07
"vmlal.s16 q14, d11, %P15[3] \n"
"vmlal.s16 q11, d12, %P16[0] \n"// (r32 - r39) * k08
"vmlal.s16 q12, d13, %P16[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vadd.s32 q11, q11, q13 \n"
"vadd.s32 q12, q12, q14 \n"
"vdup.f32 q13, %17 \n" // bias
"vdup.f32 q14, %18 \n" // scale_in
"vdup.f32 q15, %19 \n" // scale_out
// top_s32 -> top_f32
"vcvt.f32.s32 q7, q7 \n"
"vcvt.f32.s32 q8, q8 \n"
// top_f32 = top_f32 * scale_int
"vmul.f32 q0, q7, q14 \n"
"vmul.f32 q1, q8, q14 \n"
// top_f32 = top_f32 + bias
"vadd.f32 q0, q0, q13 \n"
"vadd.f32 q1, q1, q13 \n"
// top_f32 = top_f32 * scale_out
"vmul.f32 q0, q0, q15 \n"
"vmul.f32 q1, q1, q15 \n"
// top_f32 -> top_s32
"vcvtr.s32.f32 s0, s0 \n"
"vcvtr.s32.f32 s1, s1 \n"
"vcvtr.s32.f32 s2, s2 \n"
"vcvtr.s32.f32 s3, s3 \n"
"vcvtr.s32.f32 s4, s4 \n"
"vcvtr.s32.f32 s5, s5 \n"
"vcvtr.s32.f32 s6, s6 \n"
"vcvtr.s32.f32 s7, s7 \n"
// top_s32 -> top_s16
"vqmovn.s32 d14, q0 \n"
"vqmovn.s32 d15, q1 \n"
// top_s16 -> top_s8
"vqmovn.s16 d14, q7 \n"
// save top_s8
"vst1.8 {d14}, [%1]! \n"
// top_s32 -> top_f32
"vcvt.f32.s32 q11, q11 \n"
"vcvt.f32.s32 q12, q12 \n"
// top_f32 = top_f32 * scale_int
"vmul.f32 q0, q11, q14 \n"
"vmul.f32 q1, q12, q14 \n"
// top_f32 = top_f32 + bias
"vadd.f32 q0, q0, q13 \n"
"vadd.f32 q1, q1, q13 \n"
// top_f32 = top_f32 * scale_out
"vmul.f32 q0, q0, q15 \n"
"vmul.f32 q1, q1, q15 \n"
// top_f32 -> top_s32
"vcvtr.s32.f32 s0, s0 \n"
"vcvtr.s32.f32 s1, s1 \n"
"vcvtr.s32.f32 s2, s2 \n"
"vcvtr.s32.f32 s3, s3 \n"
"vcvtr.s32.f32 s4, s4 \n"
"vcvtr.s32.f32 s5, s5 \n"
"vcvtr.s32.f32 s6, s6 \n"
"vcvtr.s32.f32 s7, s7 \n"
// top_s32 -> top_s16
"vqmovn.s32 d14, q0 \n"
"vqmovn.s32 d15, q1 \n"
// top_s16 -> top_s8
"vqmovn.s16 d14, q7 \n"
// save top_s8
"vst1.8 {d14}, [%2]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(outptr0n), // %2
"=r"(r0), // %3
"=r"(r1), // %4
"=r"(r2), // %5
"=r"(r3) // %6
: "0"(nn),
"1"(outptr0),
"2"(outptr0n),
"3"(r0),
"4"(r1),
"5"(r2),
"6"(r3),
"w"(_k0123), // %14
"w"(_k4567), // %15
"w"(_k8xxx), // %16
"r"(bias0), // %17
"r"(scale_requant_in), // %18
"r"(scale_requant_out) // %19
: "cc", "memory", "q0", "q1", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
// TODO NEON
int sum0 = 0;
int sum0n = 0;
sum0 += (int)r0[0] * kernel[0];
sum0 += (int)r0[1] * kernel[1];
sum0 += (int)r0[2] * kernel[2];
sum0 += (int)r1[0] * kernel[3];
sum0 += (int)r1[1] * kernel[4];
sum0 += (int)r1[2] * kernel[5];
sum0 += (int)r2[0] * kernel[6];
sum0 += (int)r2[1] * kernel[7];
sum0 += (int)r2[2] * kernel[8];
sum0n += (int)r1[0] * kernel[0];
sum0n += (int)r1[1] * kernel[1];
sum0n += (int)r1[2] * kernel[2];
sum0n += (int)r2[0] * kernel[3];
sum0n += (int)r2[1] * kernel[4];
sum0n += (int)r2[2] * kernel[5];
sum0n += (int)r3[0] * kernel[6];
sum0n += (int)r3[1] * kernel[7];
sum0n += (int)r3[2] * kernel[8];
*outptr0 = float2int8(((float)sum0 * scale_requant_in + bias0) * scale_requant_out);
*outptr0n = float2int8(((float)sum0n * scale_requant_in + bias0) * scale_requant_out);
r0++;
r1++;
r2++;
r3++;
outptr0++;
outptr0n++;
}
r0 += 2 + w;
r1 += 2 + w;
r2 += 2 + w;
r3 += 2 + w;
outptr0 += outw;
outptr0n += outw;
}
for (; i < outh; i++)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"dup v26.4s, %w13 \n"
"dup v27.4s, %w14 \n"
"dup v28.4s, %w15 \n"
"0: \n"
"ld1 {v4.8b, v5.8b}, [%2] \n"
"ld1 {v6.8b, v7.8b}, [%3] \n"
"ld1 {v8.8b, v9.8b}, [%4] \n"
"add %2, %2, #8 \n"
"add %3, %3, #8 \n"
"add %4, %4, #8 \n"
"ext v12.8b, v4.8b, v5.8b, #1 \n"
"ext v13.8b, v4.8b, v5.8b, #2 \n"
"ext v14.8b, v6.8b, v7.8b, #1 \n"
"ext v15.8b, v6.8b, v7.8b, #2 \n"
"ext v16.8b, v8.8b, v9.8b, #1 \n"
"ext v17.8b, v8.8b, v9.8b, #2 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v12.8h, v12.8b, #0 \n"// r01
"sshll v13.8h, v13.8b, #0 \n"// r02
"sshll v6.8h, v6.8b, #0 \n"// r10
"sshll v14.8h, v14.8b, #0 \n"// r11
"sshll v15.8h, v15.8b, #0 \n"// r12
"sshll v8.8h, v8.8b, #0 \n"// r20
"sshll v16.8h, v16.8b, #0 \n"// r21
"sshll v17.8h, v17.8b, #0 \n"// r22
// r0
"smull v20.4s, v4.4h, %10.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %10.h[0] \n"
"smull v22.4s, v12.4h, %10.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v12.8h, %10.h[1] \n"
"smull v24.4s, v13.4h, %10.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v13.8h, %10.h[2] \n"
// r1
"smlal v20.4s, v6.4h, %10.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v6.8h, %10.h[3] \n"
"smlal v22.4s, v14.4h, %11.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v14.8h, %11.h[0] \n"
"smlal v24.4s, v15.4h, %11.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v15.8h, %11.h[1] \n"
// r2
"smlal v20.4s, v8.4h, %11.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v8.8h, %11.h[2] \n"
"smlal v22.4s, v16.4h, %11.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v16.8h, %11.h[3] \n"
"smlal v24.4s, v17.4h, %12.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v17.8h, %12.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
// top_s32 -> top_f32
"scvtf v20.4s, v20.4s \n"
"scvtf v21.4s, v21.4s \n"
// top_f32 = top_f32 * scale_in
"fmul v20.4s, v20.4s, v27.4s \n"
"fmul v21.4s, v21.4s, v27.4s \n"
// top_f32 = top_f32 + bias
"fadd v20.4s, v20.4s, v26.4s \n"
"fadd v21.4s, v21.4s, v26.4s \n"
// top_f32 = top_f32 * scale_out
"fmul v20.4s, v20.4s, v28.4s \n"
"fmul v21.4s, v21.4s, v28.4s \n"
// top_f32 -> top_s32
"fcvtas v20.4s, v20.4s \n"
"fcvtas v21.4s, v21.4s \n"
// top_s32 -> top_s16
"sqxtn v7.4h, v20.4s \n"
"sqxtn2 v7.8h, v21.4s \n"
// top_s16 -> top_s8
"sqxtn v8.8b, v7.8h \n"
// save top_s8
"st1 {v8.8b}, [%1], #8 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx), // %12
"r"(bias0), // %13
"r"(scale_requant_in), // %14
"r"(scale_requant_out) // %15
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld1.s8 {d30-d31}, [%2] \n"// r0
"add %2, %2, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q5, d10 \n"// r01
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P10[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P10[0] \n"
"vmull.s16 q9, d10, %P10[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P10[1] \n"
"vmlal.s16 q7, d12, %P10[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P10[2] \n"
// r1
"vld1.s8 {d30-d31}, [%3] \n"// r1
"add %3, %3, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q5, d10 \n"// r11
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P10[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P10[3] \n"
"vmlal.s16 q9, d10, %P11[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P11[0] \n"
"vmlal.s16 q7, d12, %P11[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P11[1] \n"
// r2
"vld1.s8 {d30-d31}, [%4] \n"// r2
"add %4, %4, #8 \n"
"vext.s8 d10, d30, d31, #1 \n"
"vext.s8 d12, d30, d31, #2 \n"
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q5, d10 \n"// r21
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P11[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P11[2] \n"
"vmlal.s16 q9, d10, %P11[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P11[3] \n"
"vmlal.s16 q7, d12, %P12[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P12[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vdup.f32 q13, %13 \n" // bias
"vdup.f32 q14, %14 \n" // scale_in
"vdup.f32 q15, %15 \n" // scale_out
// top_s32 -> top_f32
"vcvt.f32.s32 q7, q7 \n"
"vcvt.f32.s32 q8, q8 \n"
// top_f32 = top_f32 * scale_int
"vmul.f32 q0, q7, q14 \n"
"vmul.f32 q1, q8, q14 \n"
// top_f32 = top_f32 + bias
"vadd.f32 q0, q0, q13 \n"
"vadd.f32 q1, q1, q13 \n"
// top_f32 = top_f32 * scale_out
"vmul.f32 q0, q0, q15 \n"
"vmul.f32 q1, q1, q15 \n"
// top_f32 -> top_s32
"vcvtr.s32.f32 s0, s0 \n"
"vcvtr.s32.f32 s1, s1 \n"
"vcvtr.s32.f32 s2, s2 \n"
"vcvtr.s32.f32 s3, s3 \n"
"vcvtr.s32.f32 s4, s4 \n"
"vcvtr.s32.f32 s5, s5 \n"
"vcvtr.s32.f32 s6, s6 \n"
"vcvtr.s32.f32 s7, s7 \n"
// top_s32 -> top_s16
"vqmovn.s32 d14, q0 \n"
"vqmovn.s32 d15, q1 \n"
// top_s16 -> top_s8
"vqmovn.s16 d14, q7 \n"
// save top_s8
"vst1.8 {d14}, [%1]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr0), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr0),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx), // %12
"r"(bias0), // %13
"r"(scale_requant_in), // %14
"r"(scale_requant_out) // %15
: "cc", "memory", "q0", "q1", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
int sum = 0;
sum += (int)r0[0] * kernel[0];
sum += (int)r0[1] * kernel[1];
sum += (int)r0[2] * kernel[2];
sum += (int)r1[0] * kernel[3];
sum += (int)r1[1] * kernel[4];
sum += (int)r1[2] * kernel[5];
sum += (int)r2[0] * kernel[6];
sum += (int)r2[1] * kernel[7];
sum += (int)r2[2] * kernel[8];
*outptr0 = float2int8(((float)sum * scale_requant_in + bias0) * scale_requant_out);
r0++;
r1++;
r2++;
outptr0++;
}
r0 += 2;
r1 += 2;
r2 += 2;
}
}
}
static void convdw3x3s2_int8_requant_neon(const Mat &bottom_blob, Mat &top_blob, const Mat &_kernel, const Mat &_bias, std::vector<float> scales_requant, const Option& opt)
{
int w = bottom_blob.w;
int outw = top_blob.w;
int outh = top_blob.h;
int outch = top_blob.c;
const float* bias = _bias;
const int tailstep = w - 2*outw + w;
#pragma omp parallel for num_threads(opt.num_threads)
for (int p=0; p<outch; p++)
{
Mat out = top_blob.channel(p);
const float bias0 = bias ? bias[p] : 0.f;
const float scale_requant_in = scales_requant[2*p];
const float scale_requant_out = scales_requant[2*p+1];
const signed char* kernel = (const signed char*)_kernel + p*9;
signed char* outptr = out;
const signed char* img = bottom_blob.channel(p);
const signed char* r0 = img;
const signed char* r1 = img + w;
const signed char* r2 = img + w*2;
int i = 0;
#if __ARM_NEON
int8x16_t _k0123456789x = vld1q_s8(kernel);
int16x8_t _k_s16 = vmovl_s8(vget_low_s8(_k0123456789x));
int16x8_t _kn_s16 = vmovl_s8(vget_high_s8(_k0123456789x));
int16x4_t _k0123 = vget_low_s16(_k_s16);
int16x4_t _k4567 = vget_high_s16(_k_s16);
int16x4_t _k8xxx = vget_low_s16(_kn_s16);
#endif // __ARM_NEON
for (; i < outh; i++)
{
#if __ARM_NEON
int nn = outw >> 3;
int remain = outw & 7;
#else
int remain = outw;
#endif // __ARM_NEON
#if __ARM_NEON
#if __aarch64__
if (nn > 0)
{
asm volatile(
"dup v26.4s, %w13 \n"
"dup v27.4s, %w14 \n"
"dup v28.4s, %w15 \n"
"0: \n"
"ld2 {v4.8b, v5.8b}, [%2], #16 \n"
"ld2 {v6.8b, v7.8b}, [%2] \n"
"ld2 {v8.8b, v9.8b}, [%3], #16 \n"
"ld2 {v10.8b, v11.8b}, [%3] \n"
"ld2 {v12.8b, v13.8b}, [%4], #16 \n"
"ld2 {v14.8b, v15.8b}, [%4] \n"
"ext v6.8b, v4.8b, v6.8b, #1 \n"
"ext v10.8b, v8.8b, v10.8b, #1 \n"
"ext v14.8b, v12.8b, v14.8b, #1 \n"
"sshll v4.8h, v4.8b, #0 \n"// r00
"sshll v5.8h, v5.8b, #0 \n"// r01
"sshll v6.8h, v6.8b, #0 \n"// r02
"sshll v8.8h, v8.8b, #0 \n"// r10
"sshll v9.8h, v9.8b, #0 \n"// r11
"sshll v10.8h, v10.8b, #0 \n"// r12
"sshll v12.8h, v12.8b, #0 \n"// r20
"sshll v13.8h, v13.8b, #0 \n"// r21
"sshll v14.8h, v14.8b, #0 \n"// r22
// r0
"smull v20.4s, v4.4h, %10.h[0] \n"// (r00 - r07) * k00
"smull2 v21.4s, v4.8h, %10.h[0] \n"
"smull v22.4s, v5.4h, %10.h[1] \n"// (r01 - r08) * k01
"smull2 v23.4s, v5.8h, %10.h[1] \n"
"smull v24.4s, v6.4h, %10.h[2] \n"// (r02 - r09) * k02
"smull2 v25.4s, v6.8h, %10.h[2] \n"
// r1
"smlal v20.4s, v8.4h, %10.h[3] \n"// (r10 - r17) * k03
"smlal2 v21.4s, v8.8h, %10.h[3] \n"
"smlal v22.4s, v9.4h, %11.h[0] \n"// (r11 - r18) * k04
"smlal2 v23.4s, v9.8h, %11.h[0] \n"
"smlal v24.4s, v10.4h, %11.h[1] \n"// (r12 - r19) * k05
"smlal2 v25.4s, v10.8h, %11.h[1] \n"
// r2
"smlal v20.4s, v12.4h, %11.h[2] \n"// (r20 - r27) * k06
"smlal2 v21.4s, v12.8h, %11.h[2] \n"
"smlal v22.4s, v13.4h, %11.h[3] \n"// (r21 - r28) * k07
"smlal2 v23.4s, v13.8h, %11.h[3] \n"
"smlal v24.4s, v14.4h, %12.h[0] \n"// (r22 - r29) * k08
"smlal2 v25.4s, v14.8h, %12.h[0] \n"
// add and save
"add v20.4s, v20.4s, v22.4s \n"
"add v21.4s, v21.4s, v23.4s \n"
"add v20.4s, v20.4s, v24.4s \n"
"add v21.4s, v21.4s, v25.4s \n"
// top_s32 -> top_f32
"scvtf v20.4s, v20.4s \n"
"scvtf v21.4s, v21.4s \n"
// top_f32 = top_f32 * scale_in
"fmul v20.4s, v20.4s, v27.4s \n"
"fmul v21.4s, v21.4s, v27.4s \n"
// top_f32 = top_f32 + bias
"fadd v20.4s, v20.4s, v26.4s \n"
"fadd v21.4s, v21.4s, v26.4s \n"
// top_f32 = top_f32 * scale_out
"fmul v20.4s, v20.4s, v28.4s \n"
"fmul v21.4s, v21.4s, v28.4s \n"
// top_f32 -> top_s32
"fcvtas v20.4s, v20.4s \n"
"fcvtas v21.4s, v21.4s \n"
// top_s32 -> top_s16
"sqxtn v7.4h, v20.4s \n"
"sqxtn2 v7.8h, v21.4s \n"
// top_s16 -> top_s8
"sqxtn v8.8b, v7.8h \n"
// save top_s8
"st1 {v8.8b}, [%1], #8 \n"
"subs %w0, %w0, #1 \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx), // %12
"r"(bias0), // %13
"r"(scale_requant_in), // %14
"r"(scale_requant_out) // %15
: "cc", "memory", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31"
);
}
#else
if (nn > 0)
{
asm volatile(
"0: \n"
// r0
"vld2.s8 {d30-d31}, [%2]! \n"// r0
"vld2.s8 {d10-d11}, [%2] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r01
"vmovl.s8 q15, d30 \n"// r00
"vmovl.s8 q6, d12 \n"// r02
// sum0
"vmull.s16 q7, d30, %P10[0] \n"// (r00 - r07) * k00
"vmull.s16 q8, d31, %P10[0] \n"
"vmull.s16 q9, d10, %P10[1] \n"// (r01 - r08) * k01
"vmull.s16 q10, d11, %P10[1] \n"
"vmlal.s16 q7, d12, %P10[2] \n"// (r02 - r09) * k02
"vmlal.s16 q8, d13, %P10[2] \n"
// r1
"vld2.s8 {d30-d31}, [%3]! \n"// r1
"vld2.s8 {d10-d11}, [%3] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r11
"vmovl.s8 q15, d30 \n"// r10
"vmovl.s8 q6, d12 \n"// r12
// sum0
"vmlal.s16 q7, d30, %P10[3] \n"// (r10 - r17) * k03
"vmlal.s16 q8, d31, %P10[3] \n"
"vmlal.s16 q9, d10, %P11[0] \n"// (r11 - r18) * k04
"vmlal.s16 q10, d11, %P11[0] \n"
"vmlal.s16 q7, d12, %P11[1] \n"// (r12 - r19) * k05
"vmlal.s16 q8, d13, %P11[1] \n"
// r2
"vld2.s8 {d30-d31}, [%4]! \n"// r2
"vld2.s8 {d10-d11}, [%4] \n"
"vext.s8 d12, d30, d10, #1 \n"
"vmovl.s8 q5, d31 \n"// r21
"vmovl.s8 q15, d30 \n"// r20
"vmovl.s8 q6, d12 \n"// r22
// sum0
"vmlal.s16 q7, d30, %P11[2] \n"// (r20 - r27) * k06
"vmlal.s16 q8, d31, %P11[2] \n"
"vmlal.s16 q9, d10, %P11[3] \n"// (r21 - r28) * k07
"vmlal.s16 q10, d11, %P11[3] \n"
"vmlal.s16 q7, d12, %P12[0] \n"// (r22 - r29) * k08
"vmlal.s16 q8, d13, %P12[0] \n"
"subs %0, %0, #1 \n"
// add and save
"vadd.s32 q7, q7, q9 \n"
"vadd.s32 q8, q8, q10 \n"
"vdup.f32 q11, %13 \n" // bias
"vdup.f32 q12, %14 \n" // scale_in
"vdup.f32 q13, %15 \n" // scale_out
// top_s32 -> top_f32
"vcvt.f32.s32 q7, q7 \n"
"vcvt.f32.s32 q8, q8 \n"
// top_f32 = top_f32 * scale_int
"vmul.f32 q0, q7, q12 \n"
"vmul.f32 q1, q8, q12 \n"
// top_f32 = top_f32 + bias
"vadd.f32 q0, q0, q11 \n"
"vadd.f32 q1, q1, q11 \n"
// top_f32 = top_f32 * scale_out
"vmul.f32 q0, q0, q13 \n"
"vmul.f32 q1, q1, q13 \n"
// top_f32 -> top_s32
"vcvtr.s32.f32 s0, s0 \n"
"vcvtr.s32.f32 s1, s1 \n"
"vcvtr.s32.f32 s2, s2 \n"
"vcvtr.s32.f32 s3, s3 \n"
"vcvtr.s32.f32 s4, s4 \n"
"vcvtr.s32.f32 s5, s5 \n"
"vcvtr.s32.f32 s6, s6 \n"
"vcvtr.s32.f32 s7, s7 \n"
// top_s32 -> top_s16
"vqmovn.s32 d14, q0 \n"
"vqmovn.s32 d15, q1 \n"
// top_s16 -> top_s8
"vqmovn.s16 d14, q7 \n"
// save top_s8
"vst1.8 {d14}, [%1]! \n"
"bne 0b \n"
: "=r"(nn), // %0
"=r"(outptr), // %1
"=r"(r0), // %2
"=r"(r1), // %3
"=r"(r2) // %4
: "0"(nn),
"1"(outptr),
"2"(r0),
"3"(r1),
"4"(r2),
"w"(_k0123), // %10
"w"(_k4567), // %11
"w"(_k8xxx), // %12
"r"(bias0), // %13
"r"(scale_requant_in), // %14
"r"(scale_requant_out) // %15
: "cc", "memory", "q0", "q1", "q5", "q6", "q7", "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // __aarch64__
#endif // __ARM_NEON
for (; remain>0; remain--)
{
int sum = 0;
sum += (int)r0[0] * kernel[0];
sum += (int)r0[1] * kernel[1];
sum += (int)r0[2] * kernel[2];
sum += (int)r1[0] * kernel[3];
sum += (int)r1[1] * kernel[4];
sum += (int)r1[2] * kernel[5];
sum += (int)r2[0] * kernel[6];
sum += (int)r2[1] * kernel[7];
sum += (int)r2[2] * kernel[8];
*outptr = float2int8(((float)sum * scale_requant_in + bias0) * scale_requant_out);
r0 += 2;
r1 += 2;
r2 += 2;
outptr++;
}
r0 += tailstep;
r1 += tailstep;
r2 += tailstep;
}
}
}
|
a.17.1.c | /* { dg-do compile } */
void
a17_1_wrong ()
{
union
{
int n;
float x;
} u;
#pragma omp parallel
{
#pragma omp atomic
u.n++;
#pragma omp atomic
u.x += 1.0;
/* Incorrect because the atomic constructs reference the same location
through incompatible types */
}
}
|
dem_structures_coupling_utilities.h | /*
* Author: Miguel Angel Celigueta
*
* maceli@cimne.upc.edu
*/
#ifndef KRATOS_STRUCTURES_DEM_COUPLING_UTILITIES_H
#define KRATOS_STRUCTURES_DEM_COUPLING_UTILITIES_H
// /* External includes */
// System includes
// Project includes
#include "includes/variables.h"
/* System includes */
#include <limits>
#include <iostream>
#include <iomanip>
/* External includes */
#ifdef _OPENMP
#include <omp.h>
#endif
/* Project includes */
#include "includes/define.h"
#include "includes/model_part.h"
#include "custom_conditions/RigidFace.h"
#include "custom_conditions/RigidEdge.h"
#include "DEM_application_variables.h"
#include "dem_structures_coupling_application_variables.h"
#include "custom_elements/spheric_continuum_particle.h"
namespace Kratos
{
class DemStructuresCouplingUtilities
{
public:
typedef ModelPart::NodesContainerType::ContainerType::iterator NodesIteratorType;
KRATOS_CLASS_POINTER_DEFINITION(DemStructuresCouplingUtilities);
/// Default constructor
DemStructuresCouplingUtilities(){}
/// Destructor
virtual ~DemStructuresCouplingUtilities(){}
//***************************************************************************************************************
//***************************************************************************************************************
void TransferStructuresSkinToDem(ModelPart& r_source_model_part, ModelPart& r_destination_model_part, Properties::Pointer props) {
std::string error = CheckProvidedProperties(props);
const int dimension = r_source_model_part.GetProcessInfo()[DOMAIN_SIZE];
if (error != "all_ok") KRATOS_ERROR << "The Dem Walls ModelPart has no valid Properties. Missing " << error << " . Exiting." << std::endl;
r_destination_model_part.Conditions().Sort();
int id = 1;
if (r_destination_model_part.Conditions().size()) id = (r_destination_model_part.ConditionsEnd()-1)->Id() + 1;
ModelPart::ConditionsContainerType& source_conditions = r_source_model_part.Conditions();
// Adding conditions
for (unsigned int i = 0; i < source_conditions.size(); i++) {
ModelPart::ConditionsContainerType::iterator it = r_source_model_part.ConditionsBegin() + i;
Geometry< Node<3> >::Pointer p_geometry = it->pGetGeometry();
Condition::Pointer cond;
if (dimension == 2) {
cond = Condition::Pointer(new RigidEdge3D(id, p_geometry, props));
} else {
cond = Condition::Pointer(new RigidFace3D(id, p_geometry, props));
}
cond->Set(DEMFlags::STICKY, true);
r_destination_model_part.AddCondition(cond); //TODO: add all of them in a single sentence! AddConditions. Use a temporary PointerVector as a list (not std::vector!).
id++;
}
// Adding nodes
r_destination_model_part.AddNodes(r_source_model_part.NodesBegin(), r_source_model_part.NodesEnd());
}
std::string CheckProvidedProperties(Properties::Pointer props) {
std::vector<const Variable<double>* > list_of_variables_double_to_check = {&FRICTION, &WALL_COHESION, &SEVERITY_OF_WEAR, &IMPACT_WEAR_SEVERITY, &BRINELL_HARDNESS, &YOUNG_MODULUS, &POISSON_RATIO};
std::vector<const Variable<bool>* > list_of_variables_bool_to_check = {&COMPUTE_WEAR};
for (int i=0; i<(int)list_of_variables_double_to_check.size(); i++) {
if(!props->Has(*list_of_variables_double_to_check[i])) return list_of_variables_double_to_check[i]->Name();
}
for (int i=0; i<(int)list_of_variables_bool_to_check.size(); i++) {
if(!props->Has(*list_of_variables_bool_to_check[i])) return list_of_variables_bool_to_check[i]->Name();
}
return "all_ok";
}
void SmoothLoadTrasferredToFem(ModelPart& r_model_part, const double portion_of_the_force_which_is_new) {
#pragma omp parallel for
for (int i=0; i<(int)r_model_part.Nodes().size(); i++) {
auto node_it = r_model_part.NodesBegin() + i;
array_1d<double, 3> averaged_force;
array_1d<double, 3>& node_dem_load = node_it->FastGetSolutionStepValue(DEM_SURFACE_LOAD);
noalias(averaged_force) = portion_of_the_force_which_is_new * node_dem_load + (1.0 - portion_of_the_force_which_is_new) * node_it->FastGetSolutionStepValue(DEM_SURFACE_LOAD, 1);
noalias(node_dem_load) = averaged_force;
}
}
void ComputeSandProduction(ModelPart& dem_model_part, ModelPart& outer_walls_model_part, const double time) {
const std::string sand_prod_filename = "sand_production_graph.txt";
static std::ofstream ofs_sand_prod_file;
static bool first_time_entered = true;
if (first_time_entered) {
ofs_sand_prod_file.open(sand_prod_filename, std::ofstream::out | std::ofstream::trunc);
first_time_entered = false;
}
ModelPart::ElementsContainerType& pElements = dem_model_part.GetCommunicator().LocalMesh().Elements();
double current_total_mass_in_grams = 0.0;
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
Element* raw_p_element = &(*it);
SphericParticle* p_sphere = dynamic_cast<SphericParticle*>(raw_p_element);
if (p_sphere->Is(ISOLATED)) continue;
const double particle_density = p_sphere->GetDensity();
const double particle_volume = p_sphere->CalculateVolume();
current_total_mass_in_grams += particle_volume * particle_density * 1.0e3;
}
static const double initial_total_mass_in_grams = current_total_mass_in_grams;
const double cumulative_sand_mass_in_grams = initial_total_mass_in_grams - current_total_mass_in_grams;
//ModelPart::ConditionsContainerType::iterator condition_begin = outer_walls_model_part.ConditionsBegin();
//const double face_pressure_in_psi = condition_begin->GetValue(POSITIVE_FACE_PRESSURE) * 0.000145;
ProcessInfo& r_process_info = dem_model_part.GetProcessInfo();
const double Pascals_to_psi_factor = 0.000145;
const double face_pressure_in_psi = fabs(r_process_info[TARGET_STRESS_Z]) * Pascals_to_psi_factor;
static std::ofstream sand_prod_file("sand_production_graph.txt", std::ios_base::out | std::ios_base::app);
sand_prod_file << time << " " << face_pressure_in_psi << " " << cumulative_sand_mass_in_grams << '\n';
sand_prod_file.flush();
}
void MarkBrokenSpheres(ModelPart& dem_model_part) {
ModelPart::ElementsContainerType& pElements = dem_model_part.GetCommunicator().LocalMesh().Elements();
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
Element* raw_p_element = &(*it);
SphericContinuumParticle* p_sphere = dynamic_cast<SphericContinuumParticle*>(raw_p_element);
if (p_sphere->Is(ISOLATED)) continue;
bool go_to_next_particle = false;
for (unsigned int i = 0; i < p_sphere->mContinuumInitialNeighborsSize; i++) {
if (!p_sphere->mIniNeighbourFailureId[i]) {
go_to_next_particle = true;
break;
}
}
if (go_to_next_particle) continue;
else p_sphere->Set(ISOLATED, true);
}
}
void ComputeSandProductionWithDepthFirstSearchNonRecursiveImplementation(ModelPart& dem_model_part, ModelPart& outer_walls_model_part, const double time) {
const std::string sand_prod_filename = "sand_production_graph_with_chunks_non_recursive.txt";
static std::ofstream ofs_sand_prod_file;
const std::string granulometry_distr_filename = "granulometry_distribution.txt";
static std::ofstream ofs_granulometry_distr_file;
static bool first_time_entered = true;
if (first_time_entered) {
ofs_sand_prod_file.open(sand_prod_filename, std::ofstream::out | std::ofstream::trunc);
ofs_granulometry_distr_file.open(granulometry_distr_filename, std::ofstream::out | std::ofstream::trunc);
first_time_entered = false;
}
ModelPart::ElementsContainerType& pElements = dem_model_part.GetCommunicator().LocalMesh().Elements();
std::vector<double> chunks_masses;
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
it->Set(VISITED, false);
}
std::vector<SphericContinuumParticle*> stack_of_particles_to_check;
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
Element* raw_p_element = &(*it);
SphericContinuumParticle* p_sphere = dynamic_cast<SphericContinuumParticle*>(raw_p_element);
double this_chunk_mass = 0.0;
stack_of_particles_to_check.push_back(p_sphere);
while (stack_of_particles_to_check.size()) {
SphericContinuumParticle* current_particle = stack_of_particles_to_check.back();
stack_of_particles_to_check.pop_back();
if (current_particle->Is(VISITED)) continue;
const double particle_density = current_particle->GetDensity();
const double particle_volume = current_particle->CalculateVolume();
this_chunk_mass += particle_volume * particle_density * 1.0e3;
current_particle->Set(VISITED, true);
for (size_t i = 0; i < current_particle->mContinuumInitialNeighborsSize; i++) {
SphericParticle* p_neighbour_sphere = current_particle->mNeighbourElements[i];
if (p_neighbour_sphere == NULL) continue;
if (p_neighbour_sphere->Is(VISITED)) continue; //not necessary, but saves increasing and decreasing stack_of_particles_to_check's size
if (current_particle->mIniNeighbourFailureId[i]) continue;
auto existing_element_it = dem_model_part.GetMesh(0).Elements().find(p_neighbour_sphere->Id());
if (existing_element_it == dem_model_part.GetMesh(0).ElementsEnd()) continue;
SphericContinuumParticle* p_neigh_cont_sphere = dynamic_cast<SphericContinuumParticle*>(p_neighbour_sphere);
stack_of_particles_to_check.push_back(p_neigh_cont_sphere);
}
}
if (this_chunk_mass) chunks_masses.push_back(this_chunk_mass);
}
const double max_mass_of_a_single_chunck = *std::max_element(chunks_masses.begin(), chunks_masses.end());
const double current_total_mass_in_grams = max_mass_of_a_single_chunck;
static const double initial_total_mass_in_grams = current_total_mass_in_grams;
const double cumulative_sand_mass_in_grams = initial_total_mass_in_grams - current_total_mass_in_grams;
ProcessInfo& r_process_info = dem_model_part.GetProcessInfo();
const double Pascals_to_psi_factor = 0.000145;
const double face_pressure_in_psi = fabs(r_process_info[TARGET_STRESS_Z]) * Pascals_to_psi_factor;
ofs_sand_prod_file << time << " " << face_pressure_in_psi << " " << cumulative_sand_mass_in_grams << '\n';
ofs_sand_prod_file.flush();
unsigned int number_of_time_steps_between_granulometry_prints = 1e9;
static unsigned int printing_counter = 0;
if (printing_counter == number_of_time_steps_between_granulometry_prints) {
ofs_granulometry_distr_file << time;
for (unsigned int k = 0; k < chunks_masses.size(); k++) ofs_granulometry_distr_file << " " << chunks_masses[k];
ofs_granulometry_distr_file << '\n';
printing_counter = 0;
}
printing_counter++;
ofs_granulometry_distr_file.flush();
}
void ComputeSandProductionWithDepthFirstSearch(ModelPart& dem_model_part, ModelPart& outer_walls_model_part, const double time) {
const std::string filename = "sand_production_graph_with_chunks.txt";
std::ifstream ifile(filename.c_str());
static bool first_time_entered = true;
if ((bool) ifile && first_time_entered) {
std::remove(filename.c_str());
first_time_entered = false;
}
ModelPart::ElementsContainerType& pElements = dem_model_part.GetCommunicator().LocalMesh().Elements();
std::vector<double> chunks_masses;
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
it->Set(VISITED, false);
}
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
Element* raw_p_element = &(*it);
SphericContinuumParticle* p_sphere = dynamic_cast<SphericContinuumParticle*>(raw_p_element);
double this_chunk_mass = 0.0;
if( it->IsNot(VISITED) ) {
DepthFirstSearchVisit(p_sphere, this_chunk_mass);
chunks_masses.push_back(this_chunk_mass);
}
}
const double max_mass_of_a_single_chunck = *std::max_element(chunks_masses.begin(), chunks_masses.end());
const double current_total_mass_in_grams = max_mass_of_a_single_chunck;
static const double initial_total_mass_in_grams = current_total_mass_in_grams;
const double cumulative_sand_mass_in_grams = initial_total_mass_in_grams - current_total_mass_in_grams;
ModelPart::ConditionsContainerType::iterator condition_begin = outer_walls_model_part.ConditionsBegin();
const double Pascals_to_psi_factor = 0.000145;
const double face_pressure_in_psi = condition_begin->GetValue(POSITIVE_FACE_PRESSURE) * Pascals_to_psi_factor;
static std::ofstream sand_prod_file(filename, std::ios_base::out | std::ios_base::app);
sand_prod_file << time << " " << face_pressure_in_psi << " " << cumulative_sand_mass_in_grams << '\n';
sand_prod_file.flush();
}
void DepthFirstSearchVisit(SphericContinuumParticle* p_sphere, double& this_chunk_mass) {
p_sphere->Set(VISITED, true);
const double particle_radius = p_sphere->GetRadius();
const double particle_density = p_sphere->GetDensity();
this_chunk_mass += (4.0/3.0) * Globals::Pi * particle_density * particle_radius * particle_radius * particle_radius * 1000.0;
for (size_t i=0; i<p_sphere->mContinuumInitialNeighborsSize; i++) {
SphericParticle* p_neighbour_sphere = p_sphere->mNeighbourElements[i];
if (p_neighbour_sphere==NULL) continue;
if (p_sphere->mIniNeighbourFailureId[i]) continue;
if (p_neighbour_sphere->IsNot(VISITED)) {
SphericContinuumParticle* p_neigh_cont_sphere = dynamic_cast<SphericContinuumParticle*>(p_neighbour_sphere);
DepthFirstSearchVisit(p_neigh_cont_sphere, this_chunk_mass);
}
}
}
void ComputeTriaxialSandProduction(ModelPart& dem_model_part, ModelPart& outer_walls_model_part_1, ModelPart& outer_walls_model_part_2, const double time) {
const std::string filename = "sand_production_graph.txt";
std::ifstream ifile(filename.c_str());
static bool first_time_entered = true;
if ((bool) ifile && first_time_entered) {
std::remove(filename.c_str());
first_time_entered = false;
}
ModelPart::ElementsContainerType& pElements = dem_model_part.GetCommunicator().LocalMesh().Elements();
double current_total_mass_in_grams = 0.0;
for (unsigned int k = 0; k < pElements.size(); k++) {
ModelPart::ElementsContainerType::iterator it = pElements.ptr_begin() + k;
Element* raw_p_element = &(*it);
SphericParticle* p_sphere = dynamic_cast<SphericParticle*>(raw_p_element);
if (p_sphere->Is(ISOLATED)) continue;
const double particle_radius = p_sphere->GetRadius();
const double particle_density = p_sphere->GetDensity();
current_total_mass_in_grams += (4.0/3.0) * Globals::Pi * particle_density * particle_radius * particle_radius * particle_radius * 1000.0;
}
static const double initial_total_mass_in_grams = current_total_mass_in_grams;
const double cumulative_sand_mass_in_grams = initial_total_mass_in_grams - current_total_mass_in_grams;
ModelPart::ConditionsContainerType::iterator condition_begin_1 = outer_walls_model_part_1.ConditionsBegin();
ModelPart::ConditionsContainerType::iterator condition_begin_2 = outer_walls_model_part_2.ConditionsBegin();
const double Pascals_to_psi_factor = 0.000145;
const double face_pressure_in_psi = (condition_begin_1->GetValue(POSITIVE_FACE_PRESSURE) +
condition_begin_2->GetValue(POSITIVE_FACE_PRESSURE) +
3.45e6) * Pascals_to_psi_factor * 0.33333333333333; // 3.45e6 is the sigma_z constant pressure
static std::ofstream sand_prod_file(filename, std::ios_base::out | std::ios_base::app);
sand_prod_file << time << " " << face_pressure_in_psi << " " << cumulative_sand_mass_in_grams << '\n';
sand_prod_file.flush();
}
//***************************************************************************************************************
//***************************************************************************************************************
/// Turn back information as a stemplate<class T, std::size_t dim> tring.
virtual std::string Info() const
{
return "";
}
/// Print information about this object.
virtual void PrintInfo(std::ostream& rOStream) const
{
}
/// Print object's data.
virtual void PrintData(std::ostream& rOStream) const
{
}
protected:
private:
/// Assignment operator
DemStructuresCouplingUtilities & operator=(DemStructuresCouplingUtilities const& rOther);
///@}
}; // Class DemStructuresCouplingUtilities
} // namespace Python.
#endif // KRATOS_STRUCTURES_DEM_COUPLING_UTILITIES_H
|
strMbacktest.c | //
// strMbacktest.c
// pstock
//
// Created by takayoshi on 2016/01/24.
// Copyright © 2016年 pgostation. All rights reserved.
//
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <libiomp/omp.h>
#include "strMbacktest.h"
#include "xmapString.h"
#include "calc.h"
static void strMdbacktest_in(const char* path, strMdata* data, strMsignalSum* signals,
const char** sellWords, int sellWordCount, const char** lcWords, int lcWordCount, const char** sig2Words, int sig2WordCount, const char** orderWords, int orderWordCount, int isKarauri, int split, int startDateIndex, int endDateIndex);
static int signal2filter(strMdata* data, strMsignalSum* signals, int dateIndex, const char** sig2Words, int sig2WordCount);
void strMbacktest_start(const char* path, strMdata* data, strMsignalSum* signals, int startDateIndex, int endDateIndex)
{
const char* sellWords[64] = {0x00};
int sellWordCount = 0;
const char* lcWords[64] = {0x00};
int lcWordCount = 0;
const char* sig2Words[64] = {0x00};
int sig2WordCount = 0;
const char* orderWords[64] = {0x00};
int orderWordCount = 0;
int isKarauri = 0;
int split = 10; //分割数
xmap_t* xmap;
//xmap取得
{
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/conf.xmap", path);
xmap = xmap_load(filePath);
if(xmap==NULL)
{
printf("Error: strMdbacktest_start(): conf.xmap is NULL\n");
return;
}
//期間指定の場合、結果保存ディレクトリを変える
if (startDateIndex!=0 || endDateIndex < data->dailys[MAX_DATA_COUNT-1].count)
{
for (int i=(int)strlen(path)-2; i>0; i--)
{
if (path[i] == '/')
{
char strategyName[256] = {0};
char parentPath[512] = {0};
strcpy(strategyName, &path[i+1]);
strncpy(parentPath, path, i);
strcat(parentPath, "-date/");
strcat(parentPath, strategyName);
path = parentPath;
//ディレクトリ作成
{
printf("mkdir %s\n", parentPath);
mkdir(parentPath, 0777);
}
break;
}
}
}
snprintf(filePath, sizeof(filePath)-1, "%s/saveconf.xmap", path);
xmap_saveWithFilename(xmap, filePath);
}
//スペースで区切って単語配列取得
{
const char *sellRule = xmap_get(xmap, "sell");
printf("sellRule:%s\n", sellRule);
int quoteFlag = 0;
int start = 0;
const int sellRuleLen = (int)strlen(sellRule);
for(int i=0; i<=sellRuleLen; i++)
{
if(!quoteFlag && ( sellRule[i]==' ' || sellRule[i]=='\n' || i==sellRuleLen) )
{
if(i-start==0)
{
start++;
continue;
}
char *buf = malloc(i-start+1);
strncpy(buf, &sellRule[start], i-start);
buf[i-start] = '\0';
sellWords[sellWordCount] = buf;
sellWordCount++;
start = i+1;
continue;
}
if(sellRule[i]=='"')
{
quoteFlag = !quoteFlag;
}
}
}
{
const char *lcRule = xmap_get(xmap, "lc");
printf("lcRule:%s\n", lcRule);
int quoteFlag = 0;
int start = 0;
const int lcRuleLen = (int)strlen(lcRule);
for(int i=0; i<=lcRuleLen; i++)
{
if(!quoteFlag && ( lcRule[i]==' ' || lcRule[i]=='\n' || i==lcRuleLen) )
{
if(i-start==0)
{
start++;
continue;
}
char *buf = malloc(i-start+1);
strncpy(buf, &lcRule[start], i-start);
buf[i-start] = '\0';
lcWords[lcWordCount] = buf;
lcWordCount++;
start = i+1;
continue;
}
if(lcRule[i]=='"')
{
quoteFlag = !quoteFlag;
}
}
}
{
const char *sig2Rule = xmap_get(xmap, "filter2");
printf("filter2:%s\n", sig2Rule);
int quoteFlag = 0;
int start = 0;
const int sig2RuleLen = (int)strlen(sig2Rule);
for(int i=0; i<=sig2RuleLen; i++)
{
if(!quoteFlag && ( sig2Rule[i]==' ' || sig2Rule[i]=='\n' || i==sig2RuleLen) )
{
if(i-start==0)
{
start++;
continue;
}
char *buf = malloc(i-start+1);
strncpy(buf, &sig2Rule[start], i-start);
buf[i-start] = '\0';
sig2Words[sig2WordCount] = buf;
sig2WordCount++;
start = i+1;
continue;
}
if(sig2Rule[i]=='"')
{
quoteFlag = !quoteFlag;
}
}
}
{
const char *orderRule = xmap_get(xmap, "order");
printf("order:%s\n", orderRule);
int quoteFlag = 0;
int start = 0;
const int orderRuleLen = (int)strlen(orderRule);
for(int i=0; i<=orderRuleLen; i++)
{
if(!quoteFlag && ( orderRule[i]==' ' || orderRule[i]=='\n' || i==orderRuleLen) )
{
if(i-start==0)
{
start++;
continue;
}
char *buf = malloc(i-start+1);
strncpy(buf, &orderRule[start], i-start);
buf[i-start] = '\0';
orderWords[orderWordCount] = buf;
orderWordCount++;
start = i+1;
continue;
}
if(orderRule[i]=='"')
{
quoteFlag = !quoteFlag;
}
}
}
{
const char *xRule = xmap_get(xmap, "rule");
printf("rule:%s\n", xRule);
if(strcmp(xRule,"空売り")==0)
{
isKarauri = 1;
}
}
{
const char *splitStr = xmap_get(xmap, "split");
if(splitStr!=NULL)
{
sscanf(splitStr, "%d", &split);
if(split>SPLIT_MAX){
split = SPLIT_MAX;
}
}
}
strMdbacktest_in(path, data, signals, sellWords, sellWordCount, lcWords, lcWordCount, sig2Words, sig2WordCount, orderWords, orderWordCount, isKarauri, split, startDateIndex, endDateIndex);
xmap_release(xmap);
}
static void strMdbacktest_in(const char* path,strMdata* data, strMsignalSum* signals,
const char** sellWords, int sellWordCount, const char** lcWords, int lcWordCount, const char** sig2Words, int sig2WordCount, const char** orderWords, int orderWordCount, int isKarauri, int split, int startDateIndex, int endDateIndex)
{
const long initialCache = 10000000;
long cache = initialCache;
int positionCount = 0;
strMsignal positionSignal[split];
const int positionSignalListMax = 20000;
strMsignal positionSignalList[positionSignalListMax] = {0x00};
int positionSignalListCounter = 0;
long cacheHistory[signals->count];
int positionCountHistory[signals->count];
strMsignal positionSignalHistory[signals->count][split];
int filter2History[signals->count];
long yearProfit = 0;
memset(positionSignal, 0x00, sizeof(strMsignal) * split);
//1日ずつずらしながらバックテスト
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
cacheHistory[dateIndex] = -1;
//発注シグナルがあるので、発注する
while(signals->dailyCount[dateIndex] > 0 && positionCount < split && cache>0)
{
//シグナル数フィルター
int signal2filterResult = signal2filter(data, signals, dateIndex, sig2Words, sig2WordCount);
filter2History[dateIndex] = signal2filterResult;
if(!signal2filterResult)
{
break;
}
{
double orderList[signals->dailyCount[dateIndex]];
//最も優先度の高いものを発注する
#pragma omp parallel
#pragma omp sections
{
#pragma omp section
for(int i=0; i<signals->dailyCount[dateIndex]/2; i++)
{
orderList[i] = -9999999999999.9;
//すでに保有していれば発注しない
int flag = 0;
for(int j=0; j<positionCount; j++)
{
if(positionSignal[j].codeIndex == signals->dailySignal[dateIndex][i].codeIndex)
{
flag = 1;
break;
}
}
if(flag){
orderList[i] = -9999999999999.9;
continue;
}
//変な銘柄は発注しない
int code = data->codes[signals->dailySignal[dateIndex][i].codeIndex];
if(code==1773){ //株価データが壊れている
continue;
}
//貸借銘柄じゃないので空売りできない
if (isKarauri)
{
int karauriOk = data->companys[signals->dailySignal[dateIndex][i].codeIndex].isKarauriOk;
if(!karauriOk){
continue;
}
}
//優先度を計算
orderList[i] = calc(orderWords, 0, orderWordCount-1, data, NULL, NULL, dateIndex, signals->dailySignal[dateIndex][i].codeIndex);
}
#pragma omp section
for(int i=signals->dailyCount[dateIndex]/2; i<signals->dailyCount[dateIndex]; i++)
{
orderList[i] = -9999999999999.9;
//すでに保有していれば発注しない
int flag = 0;
for(int j=0; j<positionCount; j++)
{
if(positionSignal[j].codeIndex == signals->dailySignal[dateIndex][i].codeIndex)
{
flag = 1;
break;
}
}
if(flag) {
orderList[i] = -9999999999999.9;
continue;
}
//変な銘柄は発注しない
int code = data->codes[signals->dailySignal[dateIndex][i].codeIndex];
if(code==1773){ //株価データが壊れている
continue;
}
//貸借銘柄じゃないので空売りできない
if (isKarauri)
{
int karauriOk = data->companys[signals->dailySignal[dateIndex][i].codeIndex].isKarauriOk;
if(!karauriOk){
continue;
}
}
//優先度を計算
orderList[i] = calc(orderWords, 0, orderWordCount-1, data, NULL, NULL, dateIndex, signals->dailySignal[dateIndex][i].codeIndex);
}
}
double orderMax = -9999999999999.9;
strMsignal* signalMax = NULL;
int buyValue = 0;
int buyUnit = 0;
for(int i=0; i<signals->dailyCount[dateIndex]; i++)
{
if(orderList[i] > orderMax)
{
int tmpValue = signals->dailySignal[dateIndex][i].buyValue;
if(tmpValue==-1){
tmpValue = data->dailys[signals->dailySignal[dateIndex][i].codeIndex].end[dateIndex];
}
long tmpCache = cache/(split - positionCount);
if (tmpCache > initialCache / split) {
tmpCache = initialCache / split;
}
int tmpUnit = (int)(tmpCache / tmpValue);
int companyUnit = data->companys[signals->dailySignal[dateIndex][i].codeIndex].unit;
if(companyUnit > 0 && tmpUnit < companyUnit)
{
continue;
}
int tmpUnit2 = 0;
if(companyUnit > 0){
tmpUnit2 = (tmpUnit / companyUnit) * companyUnit;
} else {
tmpUnit2 = tmpUnit;
}
if(tmpUnit2==0) {
continue;
}
orderMax = orderList[i];
signalMax = &signals->dailySignal[dateIndex][i];
buyValue = tmpValue;
buyUnit = tmpUnit2;
}
}
if(signalMax==NULL){
break;
}
//発注
long money = buyValue * buyUnit;
cache -= money;
positionCount++;
memcpy(&positionSignal[positionCount-1], signalMax, sizeof(strMsignal));
positionSignal[positionCount-1].buyUnit = buyUnit;
positionSignal[positionCount-1].buyDateIndex = dateIndex;
positionSignal[positionCount-1].isKarauri = isKarauri;
positionSignal[positionCount-1].sellReason = -1;
positionSignal[positionCount-1].sellDateIndex = endDateIndex;
positionSignal[positionCount-1].sellValue = -1;
positionSignal[positionCount-1].slippage = -1;
positionSignal[positionCount-1].zei = -1;
//過去の履歴に移動
positionSignal[positionCount-1].historyIndex = positionSignalListCounter;
if(positionSignalListCounter<positionSignalListMax){
memcpy(&positionSignalList[positionSignalListCounter], &positionSignal[positionCount-1], sizeof(strMsignal));
positionSignalListCounter++;
} else {
printf("Error:strMdbacktest_in():positionSignalList size over\n");
}
//printf("buy:cache=%ld, code=%d, positionCount=%d\n", cache, data->codes[positionSignal[positionCount-1].codeIndex], positionCount);
}
}
//手仕舞い条件を計算する
for(int i=positionCount-1; i>=0; i--)
{
if(dateIndex < endDateIndex-1 && positionSignal[i].buyDateIndex == dateIndex)
{
//本日の発注なので、まだ持っていない
continue;
}
float sellCalc = calc(sellWords, 0, sellWordCount-1, data, &positionSignal[i], NULL, dateIndex, positionSignal[i].codeIndex);
float lcCalc = 0.0;
if( sellCalc <= 0.0 ){
lcCalc = calc(lcWords, 0, lcWordCount-1, data, &positionSignal[i], NULL, dateIndex, positionSignal[i].codeIndex) > 0.0;
}
if( sellCalc > 0.0 ||
lcCalc > 0.0 ||
( dateIndex < endDateIndex-2 && data->dailys[positionSignal[i].codeIndex].start[dateIndex+1]==0 && data->dailys[positionSignal[i].codeIndex].start[dateIndex+2]==0 ) ||
dateIndex >= endDateIndex-1 )
{
//printf("sell:holdDays=%d code=%d buyValue=%d sellValue=%d ", dateIndex - positionSignal[i].buyDateIndex, data->codes[positionSignal[i].codeIndex], positionSignal[i].buyValue, data->dailys[positionSignal[i].codeIndex].start[dateIndex+1<signals->count?dateIndex+1:dateIndex]);
//手仕舞い
int zei = 0;
long slippage = 0;
int nextStart = 0;
long nextVolume = 0;
if(dateIndex+1<signals->count){
nextStart = data->dailys[positionSignal[i].codeIndex].start[dateIndex+1];
nextVolume = data->dailys[positionSignal[i].codeIndex].volume[dateIndex+1];
} else {
nextStart= data->dailys[positionSignal[i].codeIndex].end[dateIndex];
}
for(int j=2; nextStart==0 && dateIndex+j < signals->count && j<5; j++){
nextStart = data->dailys[positionSignal[i].codeIndex].start[dateIndex+j];
nextVolume = data->dailys[positionSignal[i].codeIndex].volume[dateIndex+j];
}
if(nextStart==0){
//閑散としすぎ。上場廃止？
nextStart = data->dailys[positionSignal[i].codeIndex].end[dateIndex];
nextVolume = 0;
}
if ( positionSignal[i].buyValue == -1 ) {
positionSignal[i].buyValue = nextStart;
}
if(isKarauri){
//空売りで仮想的に減らしていたお金を元に戻す
long money = positionSignal[i].buyValue * (long)positionSignal[i].buyUnit;
cache += money;
//空売った時と買い戻した値段の差額が利益/損失となる
long profit = (positionSignal[i].buyValue - nextStart) * (long)positionSignal[i].buyUnit;
cache += profit;
if(profit>0 && yearProfit+profit >= 0){
zei = profit * 0.2;
}
if(profit<0 && yearProfit > 0){
zei = profit * 0.2;
}
yearProfit += profit;
} else {
//買ったお金を元に戻す
long money = positionSignal[i].buyValue * (long)positionSignal[i].buyUnit;
cache += money;
//利益/損失
long profit = (nextStart - positionSignal[i].buyValue) * (long)positionSignal[i].buyUnit;
if ( profit > money * 3 ) {
profit = money * 3; //極端な銘柄の排除
}
cache += profit;
if(profit>0 && yearProfit+profit >= 0){
zei = profit * 0.2;
}
if(profit<0 && yearProfit > 0){
zei = profit * 0.2;
}
yearProfit += profit;
}
cache -= zei;
//スリッページ計算（発注時には計算せず、手仕舞い時のみ計算している）
if(nextVolume > 1000 && positionSignal[i].buyUnit >= nextVolume){
slippage = (nextStart * positionSignal[i].buyUnit) / 40;
}
else if(nextVolume > 1000 && positionSignal[i].buyUnit >= nextVolume/2){
slippage = (nextStart * positionSignal[i].buyUnit) / 60;
}
else if(nextVolume > 1000 && positionSignal[i].buyUnit >= nextVolume/5){
slippage = (nextStart * positionSignal[i].buyUnit) / 80;
}
else if(positionSignal[i].buyUnit >= nextVolume/10){
slippage = (nextStart * positionSignal[i].buyUnit) / 100;
}
else if(positionSignal[i].buyUnit >= nextVolume/20){
slippage = (nextStart * positionSignal[i].buyUnit) / 150;
}
else if(positionSignal[i].buyUnit >= nextVolume/50){
slippage = (nextStart * positionSignal[i].buyUnit) / 200;
}
else if(positionSignal[i].buyUnit >= nextVolume/100){
slippage = (nextStart * positionSignal[i].buyUnit) / 400;
}
cache -= slippage;
yearProfit -= slippage;
positionSignal[i].sellDateIndex = dateIndex;
positionSignal[i].sellValue = nextStart;
positionSignal[i].slippage = (int)slippage;
positionSignal[i].zei = zei;
if(sellCalc > 0.0) {
positionSignal[i].sellReason = 1;
} else if(lcCalc > 0.0 ) {
positionSignal[i].sellReason = 2;
} else if(dateIndex >= signals->count-1){
positionSignal[i].sellReason = 3;
} else {
positionSignal[i].sellReason = 0;
}
//過去の履歴に決済情報をコピー
if(positionSignal[i].historyIndex < positionSignalListMax){
memcpy(&positionSignalList[positionSignal[i].historyIndex], &positionSignal[i], sizeof(strMsignal));
}
//現在のポジションから削除
memcpy(&positionSignal[i], &positionSignal[positionCount-1], sizeof(strMsignal));
positionCount--;
//printf(" cache=%ld\n", cache);
}
}
//年が変わると、年間利益額をリセットする
if(dateIndex+1 < signals->count && data->date[dateIndex+1]>>16 > data->date[dateIndex]>>16)
{
yearProfit = 0;
}
//資産額計算のための履歴データ
{
cacheHistory[dateIndex] = cache;
positionCountHistory[dateIndex] = positionCount;
memcpy(positionSignalHistory[dateIndex], positionSignal, sizeof(positionSignal));
}
}
//資産額の計算
long stockHistory[signals->count];
memset(stockHistory, 0x00, sizeof(stockHistory));
{
for(int dateIndex=0; dateIndex<endDateIndex; dateIndex++)
{
for(int i=0; i<positionCountHistory[dateIndex]; i++)
{
int unit = positionSignalHistory[dateIndex][i].buyUnit;
int codeIndex = positionSignalHistory[dateIndex][i].codeIndex;
int lastEnd = data->dailys[codeIndex].end[dateIndex];
for(int j=1; lastEnd==0 && dateIndex-j > 0 && j<30; j++){
lastEnd = data->dailys[codeIndex].end[dateIndex-j];
}
long stock;
if(isKarauri){
//空売りで仮想的に減らしていたお金
long base = positionSignalHistory[dateIndex][i].buyValue * (long)unit;
//空売った時と買い戻した値段の差額が利益/損失となる
long profit = (positionSignalHistory[dateIndex][i].buyValue - lastEnd) * (long)unit;
stock = base + profit;
} else {
//買ったお金を戻す
long base = positionSignalHistory[dateIndex][i].buyValue * (long)unit;
//利益/損失
long profit = (lastEnd - positionSignalHistory[dateIndex][i].buyValue) * (long)unit;
if (profit > base * 3) {
profit = base * 3;
}
stock = base + profit;
}
stockHistory[dateIndex] += stock;
}
}
}
//allshisan.xmapの保存
{
xmap_t* xmap = xmap_load(NULL);
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "date%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%d", data->date[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "cache%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%ld", cacheHistory[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "stock%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%ld", stockHistory[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "positionCountHistory%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%d", positionCountHistory[dateIndex]);
xmap_set(xmap, key, value);
}
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/allshisan.xmap", path);
xmap_saveWithFilename(xmap, filePath);
xmap_release(xmap);
}
//allSignals.xmapの保存
{
xmap_t* xmap = xmap_load(NULL);
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "date");
snprintf(value, 32-1, "%d", data->date[signals->count]);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "signalCount");
snprintf(value, 32-1, "%d", positionSignalListCounter);
xmap_set(xmap, key, value);
char strategyName[256] = "";
for(int c=(int)strlen(path)-1; c>=0; c--)
{
if(path[c]=='/')
{
strcpy(strategyName, &path[c+1]);
if(strategyName[strlen(strategyName)-1]=='/'){
strategyName[strlen(strategyName)-1]='\0';
}
}
}
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 128);
snprintf(key, 32-1, "strategyName");
snprintf(value, 128-1, "%s", strategyName);
xmap_set(xmap, key, value);
for(int positionSignalListIndex=0; positionSignalListIndex < positionSignalListCounter; positionSignalListIndex++)
{
strMsignal* signal = &positionSignalList[positionSignalListIndex];
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "buyDateIndex%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->buyDateIndex);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "buyDate__%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", data->date[signal->buyDateIndex]);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "buyValue%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->buyValue);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "code%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", data->codes[signal->codeIndex]);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "rule%d", positionSignalListIndex);
snprintf(value, 32-1, "%s", signal->isKarauri?"空売り":"買い");
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "sellDateIndex%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->sellDateIndex);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "sellDate__%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", data->date[signal->sellDateIndex]);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "sellReason%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->sellReason);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "sellValue%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->sellValue);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "slippage%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->slippage);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "unit%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->buyUnit);
xmap_set(xmap, key, value);
key = apr_palloc(xmap->pool, 32);
value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "zei%d", positionSignalListIndex);
snprintf(value, 32-1, "%d", signal->zei);
xmap_set(xmap, key, value);
}
xmap_set(xmap, "HoldDate", "0");
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/allSignals.xmap", path);
xmap_saveWithFilename(xmap, filePath);
xmap_release(xmap);
}
//extraResult.xmapの保存
{
xmap_t* xmap = xmap_load(NULL);
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "date%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%d", data->date[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "signalCount%d", dateIndex-startDateIndex);
snprintf(value, 32-1, "%d", signals->dailyCount[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "codes%d", dateIndex-startDateIndex);
char* value = apr_palloc(xmap->pool, 6*signals->dailyCount[dateIndex]);
memset(value, 0x00, 6*signals->dailyCount[dateIndex]);
for(int i=0; i<signals->dailyCount[dateIndex]; i++)
{
char codeStr[16];
snprintf(codeStr, sizeof(codeStr)-1, "%d,", data->codes[signals->dailySignal[dateIndex][i].codeIndex]);
strcat(value, codeStr);
}
xmap_set(xmap, key, value);
}
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/extraResult.xmap", path);
xmap_saveWithFilename(xmap, filePath);
xmap_release(xmap);
}
//extraSignals.xmapの保存
{
xmap_t* xmap = xmap_load(NULL);
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
if(signals->dailyCount[dateIndex]==0){
continue;
}
char* key = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "codes%d", data->date[dateIndex]);
char* value = apr_palloc(xmap->pool, 6*signals->dailyCount[dateIndex]);
memset(value, 0x00, 6*signals->dailyCount[dateIndex]);
for(int i=0; i<signals->dailyCount[dateIndex]; i++)
{
char codeStr[16];
snprintf(codeStr, sizeof(codeStr)-1, "%d ", data->codes[signals->dailySignal[dateIndex][i].codeIndex]);
strcat(value, codeStr);
}
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "filter2%d", data->date[dateIndex]);
snprintf(value, 32-1, "%s", filter2History[dateIndex]?"true":"false");
xmap_set(xmap, key, value);
}
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/extraSignals.xmap", path);
xmap_saveWithFilename(xmap, filePath);
xmap_release(xmap);
}
//filter2signal.xmapの保存
{
xmap_t* xmap = xmap_load(NULL);
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
if(signals->dailyCount[dateIndex]==0){
continue;
}
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "signalCount%d", data->date[dateIndex]);
snprintf(value, 32-1, "%d ", signals->dailyCount[dateIndex]);
xmap_set(xmap, key, value);
}
for(int dateIndex=startDateIndex; dateIndex<endDateIndex; dateIndex++)
{
if(filter2History[dateIndex]){
char* key = apr_palloc(xmap->pool, 32);
char* value = apr_palloc(xmap->pool, 32);
snprintf(key, 32-1, "filter2_%d", data->date[dateIndex]);
snprintf(value, 32-1, "-");
xmap_set(xmap, key, value);
}
}
char filePath[256];
snprintf(filePath, sizeof(filePath)-1, "%s/filter2signal.xmap", path);
xmap_saveWithFilename(xmap, filePath);
xmap_release(xmap);
}
}
static int signal2filter(strMdata* data, strMsignalSum* signalSum, int dateIndex, const char** sig2Words, int sig2WordCount)
{
if(sig2WordCount==0) return 1;
int dummyCodeIndex = -1;
return calc(sig2Words, 0, sig2WordCount-1, data, NULL, signalSum, dateIndex, dummyCodeIndex) > 0.0;
}
|
Example_target_data.2.c | /*
* @@name: target_data.2c
* @@type: C
* @@compilable: yes
* @@linkable: no
* @@expect: success
* @@version: omp_4.0
*/
extern void init(float*, float*, int);
extern void init_again(float*, float*, int);
extern void output(float*, int);
void vec_mult(float *p, float *v1, float *v2, int N)
{
int i;
init(v1, v2, N);
#pragma omp target data map(from: p[0:N])
{
#pragma omp target map(to: v1[:N], v2[:N])
#pragma omp parallel for
for (i=0; i<N; i++)
p[i] = v1[i] * v2[i];
init_again(v1, v2, N);
#pragma omp target map(to: v1[:N], v2[:N])
#pragma omp parallel for
for (i=0; i<N; i++)
p[i] = p[i] + (v1[i] * v2[i]);
}
output(p, N);
}
|
libsais.c | /*--
This file is a part of libsais, a library for linear time
suffix array and burrows wheeler transform construction.
Copyright (c) 2021 Ilya Grebnov <ilya.grebnov@gmail.com>
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
Please see the file LICENSE for full copyright information.
--*/
#include "libsais_internal.h"
#include "libsais.h"
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#if defined(_OPENMP)
#include <omp.h>
#else
#define UNUSED(_x) (void)(_x)
#endif
typedef int32_t sa_sint_t;
typedef uint32_t sa_uint_t;
typedef ptrdiff_t fast_sint_t;
typedef size_t fast_uint_t;
#define SAINT_BIT (32)
#define SAINT_MAX INT32_MAX
#define SAINT_MIN INT32_MIN
#define ALPHABET_SIZE (1 << CHAR_BIT)
#define SUFFIX_GROUP_BIT (SAINT_BIT - 1)
#define SUFFIX_GROUP_MARKER (((sa_sint_t)1) << (SUFFIX_GROUP_BIT - 1))
#define BUCKETS_INDEX2(_c, _s) (((_c) << 1) + (_s))
#define BUCKETS_INDEX4(_c, _s) (((_c) << 2) + (_s))
#define LIBSAIS_PER_THREAD_CACHE_SIZE (24576)
typedef struct LIBSAIS_THREAD_CACHE
{
sa_sint_t symbol;
sa_sint_t index;
} LIBSAIS_THREAD_CACHE;
typedef union LIBSAIS_THREAD_STATE
{
struct
{
fast_sint_t position;
fast_sint_t count;
fast_sint_t m;
fast_sint_t last_lms_suffix;
sa_sint_t * buckets;
LIBSAIS_THREAD_CACHE * cache;
} state;
uint8_t padding[64];
} LIBSAIS_THREAD_STATE;
typedef struct LIBSAIS_CONTEXT
{
sa_sint_t * buckets;
LIBSAIS_THREAD_STATE * thread_state;
fast_sint_t threads;
} LIBSAIS_CONTEXT;
#if defined(__GNUC__) || defined(__clang__)
#define RESTRICT __restrict__
#elif defined(_MSC_VER) || defined(__INTEL_COMPILER)
#define RESTRICT __restrict
#else
#error Your compiler, configuration or platform is not supported.
#endif
#if defined(__has_builtin)
#if __has_builtin(__builtin_prefetch)
#define HAS_BUILTIN_PREFECTCH
#endif
#elif defined(__GNUC__) && __GNUC__ > 3
#define HAS_BUILTIN_PREFECTCH
#endif
#if defined(HAS_BUILTIN_PREFECTCH)
#define libsais_prefetch(address) __builtin_prefetch((const void *)(address), 0, 0)
#define libsais_prefetchw(address) __builtin_prefetch((const void *)(address), 1, 0)
#elif defined (_M_IX86) || defined (_M_AMD64)
#include <intrin.h>
#define libsais_prefetch(address) _mm_prefetch((const void *)(address), _MM_HINT_NTA)
#define libsais_prefetchw(address) _m_prefetchw((const void *)(address))
#elif defined (_M_ARM)
#include <intrin.h>
#define libsais_prefetch(address) __prefetch((const void *)(address))
#define libsais_prefetchw(address) __prefetchw((const void *)(address))
#elif defined (_M_ARM64)
#include <intrin.h>
#define libsais_prefetch(address) __prefetch2((const void *)(address), 1)
#define libsais_prefetchw(address) __prefetch2((const void *)(address), 17)
#else
#error Your compiler, configuration or platform is not supported.
#endif
static void * libsais_align_up(const void * address, size_t alignment)
{
return (void *)((((ptrdiff_t)address) + ((ptrdiff_t)alignment) - 1) & (-((ptrdiff_t)alignment)));
}
static void * libsais_alloc_aligned(size_t size, size_t alignment)
{
void * address = malloc(size + sizeof(short) + alignment - 1);
if (address != NULL)
{
void * aligned_address = libsais_align_up((void *)((ptrdiff_t)address + (ptrdiff_t)(sizeof(short))), alignment);
((short *)aligned_address)[-1] = (short)((ptrdiff_t)aligned_address - (ptrdiff_t)address);
return aligned_address;
}
return NULL;
}
static void libsais_free_aligned(void * aligned_address)
{
if (aligned_address != NULL)
{
free((void *)((ptrdiff_t)aligned_address - ((short *)aligned_address)[-1]));
}
}
static LIBSAIS_THREAD_STATE * libsais_alloc_thread_state(sa_sint_t threads)
{
LIBSAIS_THREAD_STATE * RESTRICT thread_state = (LIBSAIS_THREAD_STATE *)libsais_alloc_aligned((size_t)threads * sizeof(LIBSAIS_THREAD_STATE), 4096);
sa_sint_t * RESTRICT thread_buckets = (sa_sint_t *)libsais_alloc_aligned((size_t)threads * 4 * ALPHABET_SIZE * sizeof(sa_sint_t), 4096);
LIBSAIS_THREAD_CACHE * RESTRICT thread_cache = (LIBSAIS_THREAD_CACHE *)libsais_alloc_aligned((size_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE * sizeof(LIBSAIS_THREAD_CACHE), 4096);
if (thread_state != NULL && thread_buckets != NULL && thread_cache != NULL)
{
fast_sint_t t;
for (t = 0; t < threads; ++t)
{
thread_state[t].state.buckets = thread_buckets; thread_buckets += 4 * ALPHABET_SIZE;
thread_state[t].state.cache = thread_cache; thread_cache += LIBSAIS_PER_THREAD_CACHE_SIZE;
}
return thread_state;
}
libsais_free_aligned(thread_cache);
libsais_free_aligned(thread_buckets);
libsais_free_aligned(thread_state);
return NULL;
}
static void libsais_free_thread_state(LIBSAIS_THREAD_STATE * thread_state)
{
if (thread_state != NULL)
{
libsais_free_aligned(thread_state[0].state.cache);
libsais_free_aligned(thread_state[0].state.buckets);
libsais_free_aligned(thread_state);
}
}
static LIBSAIS_CONTEXT * libsais_create_ctx_main(sa_sint_t threads)
{
LIBSAIS_CONTEXT * RESTRICT ctx = (LIBSAIS_CONTEXT *)libsais_alloc_aligned(sizeof(LIBSAIS_CONTEXT), 64);
sa_sint_t * RESTRICT buckets = (sa_sint_t *)libsais_alloc_aligned(8 * ALPHABET_SIZE * sizeof(sa_sint_t), 4096);
LIBSAIS_THREAD_STATE * RESTRICT thread_state = threads > 1 ? libsais_alloc_thread_state(threads) : NULL;
if (ctx != NULL && buckets != NULL && (thread_state != NULL || threads == 1))
{
ctx->buckets = buckets;
ctx->threads = threads;
ctx->thread_state = thread_state;
return ctx;
}
libsais_free_thread_state(thread_state);
libsais_free_aligned(buckets);
libsais_free_aligned(ctx);
return NULL;
}
static void libsais_free_ctx_main(LIBSAIS_CONTEXT * ctx)
{
if (ctx != NULL)
{
libsais_free_thread_state(ctx->thread_state);
libsais_free_aligned(ctx->buckets);
libsais_free_aligned(ctx);
}
}
#if defined(_OPENMP)
static sa_sint_t libsais_count_negative_marked_suffixes(sa_sint_t * RESTRICT SA, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
sa_sint_t count = 0;
fast_sint_t i; for (i = omp_block_start; i < omp_block_start + omp_block_size; ++i) { count += (SA[i] < 0); }
return count;
}
static sa_sint_t libsais_count_zero_marked_suffixes(sa_sint_t * RESTRICT SA, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
sa_sint_t count = 0;
fast_sint_t i; for (i = omp_block_start; i < omp_block_start + omp_block_size; ++i) { count += (SA[i] == 0); }
return count;
}
static void libsais_place_cached_suffixes(sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&cache[i + 2 * prefetch_distance]);
libsais_prefetchw(&SA[cache[i + prefetch_distance + 0].symbol]);
libsais_prefetchw(&SA[cache[i + prefetch_distance + 1].symbol]);
libsais_prefetchw(&SA[cache[i + prefetch_distance + 2].symbol]);
libsais_prefetchw(&SA[cache[i + prefetch_distance + 3].symbol]);
SA[cache[i + 0].symbol] = cache[i + 0].index;
SA[cache[i + 1].symbol] = cache[i + 1].index;
SA[cache[i + 2].symbol] = cache[i + 2].index;
SA[cache[i + 3].symbol] = cache[i + 3].index;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
SA[cache[i].symbol] = cache[i].index;
}
}
static void libsais_compact_and_place_cached_suffixes(sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, l;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 3, l = omp_block_start; i < j; i += 4)
{
libsais_prefetchw(&cache[i + prefetch_distance]);
cache[l] = cache[i + 0]; l += cache[l].symbol >= 0;
cache[l] = cache[i + 1]; l += cache[l].symbol >= 0;
cache[l] = cache[i + 2]; l += cache[l].symbol >= 0;
cache[l] = cache[i + 3]; l += cache[l].symbol >= 0;
}
for (j += 3; i < j; i += 1)
{
cache[l] = cache[i]; l += cache[l].symbol >= 0;
}
libsais_place_cached_suffixes(SA, cache, omp_block_start, l - omp_block_start);
}
static void libsais_accumulate_counts_s32_2(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s]; }
}
static void libsais_accumulate_counts_s32_3(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s]; }
}
static void libsais_accumulate_counts_s32_4(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s]; }
}
static void libsais_accumulate_counts_s32_5(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
sa_sint_t * RESTRICT bucket04 = bucket03 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s] + bucket04[s]; }
}
static void libsais_accumulate_counts_s32_6(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
sa_sint_t * RESTRICT bucket04 = bucket03 - bucket_stride;
sa_sint_t * RESTRICT bucket05 = bucket04 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s] + bucket04[s] + bucket05[s]; }
}
static void libsais_accumulate_counts_s32_7(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
sa_sint_t * RESTRICT bucket04 = bucket03 - bucket_stride;
sa_sint_t * RESTRICT bucket05 = bucket04 - bucket_stride;
sa_sint_t * RESTRICT bucket06 = bucket05 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s] + bucket04[s] + bucket05[s] + bucket06[s]; }
}
static void libsais_accumulate_counts_s32_8(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
sa_sint_t * RESTRICT bucket04 = bucket03 - bucket_stride;
sa_sint_t * RESTRICT bucket05 = bucket04 - bucket_stride;
sa_sint_t * RESTRICT bucket06 = bucket05 - bucket_stride;
sa_sint_t * RESTRICT bucket07 = bucket06 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s] + bucket04[s] + bucket05[s] + bucket06[s] + bucket07[s]; }
}
static void libsais_accumulate_counts_s32_9(sa_sint_t * RESTRICT bucket00, fast_sint_t bucket_size, fast_sint_t bucket_stride)
{
sa_sint_t * RESTRICT bucket01 = bucket00 - bucket_stride;
sa_sint_t * RESTRICT bucket02 = bucket01 - bucket_stride;
sa_sint_t * RESTRICT bucket03 = bucket02 - bucket_stride;
sa_sint_t * RESTRICT bucket04 = bucket03 - bucket_stride;
sa_sint_t * RESTRICT bucket05 = bucket04 - bucket_stride;
sa_sint_t * RESTRICT bucket06 = bucket05 - bucket_stride;
sa_sint_t * RESTRICT bucket07 = bucket06 - bucket_stride;
sa_sint_t * RESTRICT bucket08 = bucket07 - bucket_stride;
fast_sint_t s; for (s = 0; s < bucket_size; s += 1) { bucket00[s] = bucket00[s] + bucket01[s] + bucket02[s] + bucket03[s] + bucket04[s] + bucket05[s] + bucket06[s] + bucket07[s] + bucket08[s]; }
}
static void libsais_accumulate_counts_s32(sa_sint_t * RESTRICT buckets, fast_sint_t bucket_size, fast_sint_t bucket_stride, fast_sint_t num_buckets)
{
while (num_buckets >= 9)
{
libsais_accumulate_counts_s32_9(buckets - (num_buckets - 9) * bucket_stride, bucket_size, bucket_stride); num_buckets -= 8;
}
switch (num_buckets)
{
case 1: break;
case 2: libsais_accumulate_counts_s32_2(buckets, bucket_size, bucket_stride); break;
case 3: libsais_accumulate_counts_s32_3(buckets, bucket_size, bucket_stride); break;
case 4: libsais_accumulate_counts_s32_4(buckets, bucket_size, bucket_stride); break;
case 5: libsais_accumulate_counts_s32_5(buckets, bucket_size, bucket_stride); break;
case 6: libsais_accumulate_counts_s32_6(buckets, bucket_size, bucket_stride); break;
case 7: libsais_accumulate_counts_s32_7(buckets, bucket_size, bucket_stride); break;
case 8: libsais_accumulate_counts_s32_8(buckets, bucket_size, bucket_stride); break;
}
}
#endif
static void libsais_gather_lms_suffixes_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, fast_sint_t m, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
if (omp_block_size > 0)
{
const fast_sint_t prefetch_distance = 128;
fast_sint_t i, j = omp_block_start + omp_block_size, c0 = T[omp_block_start + omp_block_size - 1], c1 = -1;
while (j < n && (c1 = T[j]) == c0) { ++j; }
fast_uint_t s = c0 >= c1;
for (i = omp_block_start + omp_block_size - 2, j = omp_block_start + 3; i >= j; i -= 4)
{
libsais_prefetch(&T[i - prefetch_distance]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 0); m -= ((s & 3) == 1);
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 1); m -= ((s & 3) == 1);
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 2); m -= ((s & 3) == 1);
}
for (j -= 3; i >= j; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
}
SA[m] = (sa_sint_t)(i + 1);
}
}
static void libsais_gather_lms_suffixes_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
libsais_gather_lms_suffixes_8u(T, SA, n, (fast_sint_t)n - 1, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
fast_sint_t t, m = 0; for (t = omp_num_threads - 1; t > omp_thread_num; --t) { m += thread_state[t].state.m; }
libsais_gather_lms_suffixes_8u(T, SA, n, (fast_sint_t)n - 1 - m, omp_block_start, omp_block_size);
#pragma omp barrier
if (thread_state[omp_thread_num].state.m > 0)
{
SA[(fast_sint_t)n - 1 - m] = (sa_sint_t)thread_state[omp_thread_num].state.last_lms_suffix;
}
}
#endif
}
}
static sa_sint_t libsais_gather_lms_suffixes_32s(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t i = n - 2;
sa_sint_t m = n - 1;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
for (; i >= 3; i -= 4)
{
libsais_prefetch(&T[i - prefetch_distance]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = i + 1; m -= ((s & 3) == 1);
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i - 0; m -= ((s & 3) == 1);
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = i - 1; m -= ((s & 3) == 1);
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i - 2; m -= ((s & 3) == 1);
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i + 1; m -= ((s & 3) == 1);
}
return n - 1 - m;
}
static sa_sint_t libsais_gather_compacted_lms_suffixes_32s(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t i = n - 2;
sa_sint_t m = n - 1;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
for (; i >= 3; i -= 4)
{
libsais_prefetch(&T[i - prefetch_distance]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = i + 1; m -= ((fast_sint_t)(s & 3) == (c0 >= 0));
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i - 0; m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = i - 1; m -= ((fast_sint_t)(s & 3) == (c0 >= 0));
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i - 2; m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = i + 1; m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
}
return n - 1 - m;
}
#if defined(_OPENMP)
static void libsais_count_lms_suffixes_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, 4 * (size_t)k * sizeof(sa_sint_t));
sa_sint_t i = n - 2;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
for (; i >= prefetch_distance + 3; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 0], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 1], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 2], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 3], 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
buckets[BUCKETS_INDEX4((fast_uint_t)c0, (s << 1) & 3)]++;
}
#endif
static void libsais_count_lms_suffixes_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, 2 * (size_t)k * sizeof(sa_sint_t));
sa_sint_t i = n - 2;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
for (; i >= prefetch_distance + 3; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 0], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 1], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 2], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 3], 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
buckets[BUCKETS_INDEX2((fast_uint_t)c0, 0)]++;
}
#if defined(_OPENMP)
static void libsais_count_compacted_lms_suffixes_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, 2 * (size_t)k * sizeof(sa_sint_t));
sa_sint_t i = n - 2;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
for (; i >= prefetch_distance + 3; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 0] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 1] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 2] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 3] & SAINT_MAX, 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, 0)]++;
}
#endif
static sa_sint_t libsais_count_and_gather_lms_suffixes_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
memset(buckets, 0, 4 * ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t m = omp_block_start + omp_block_size - 1;
if (omp_block_size > 0)
{
const fast_sint_t prefetch_distance = 128;
fast_sint_t i, j = m + 1, c0 = T[m], c1 = -1;
while (j < n && (c1 = T[j]) == c0) { ++j; }
fast_uint_t s = c0 >= c1;
for (i = m - 1, j = omp_block_start + 3; i >= j; i -= 4)
{
libsais_prefetch(&T[i - prefetch_distance]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 0); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 2); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
for (j -= 3; i >= j; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
c1 = (i >= 0) ? T[i] : -1; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
}
return (sa_sint_t)(omp_block_start + omp_block_size - 1 - m);
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_lms_suffixes_8u(T, SA, n, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.position = omp_block_start + omp_block_size;
thread_state[omp_thread_num].state.m = libsais_count_and_gather_lms_suffixes_8u(T, SA, n, thread_state[omp_thread_num].state.buckets, omp_block_start, omp_block_size);
if (thread_state[omp_thread_num].state.m > 0)
{
thread_state[omp_thread_num].state.last_lms_suffix = SA[thread_state[omp_thread_num].state.position - 1];
}
}
#pragma omp barrier
#pragma omp master
{
memset(buckets, 0, 4 * ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
m += (sa_sint_t)thread_state[t].state.m;
if (t != omp_num_threads - 1 && thread_state[t].state.m > 0)
{
memcpy(&SA[n - m], &SA[thread_state[t].state.position - thread_state[t].state.m], (size_t)thread_state[t].state.m * sizeof(sa_sint_t));
}
{
sa_sint_t * RESTRICT temp_bucket = thread_state[t].state.buckets;
fast_sint_t s; for (s = 0; s < 4 * ALPHABET_SIZE; s += 1) { sa_sint_t A = buckets[s], B = temp_bucket[s]; buckets[s] = A + B; temp_bucket[s] = A; }
}
}
}
}
#endif
}
return m;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
memset(buckets, 0, 4 * (size_t)k * sizeof(sa_sint_t));
fast_sint_t m = omp_block_start + omp_block_size - 1;
if (omp_block_size > 0)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j = m + 1, c0 = T[m], c1 = -1;
while (j < n && (c1 = T[j]) == c0) { ++j; }
fast_uint_t s = c0 >= c1;
for (i = m - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 0], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 1], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 2], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX4(T[i - prefetch_distance - 3], 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 0); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 2); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]++;
}
c1 = (i >= 0) ? T[i] : -1; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX4((fast_uint_t)c0, s & 3)]++;
}
return (sa_sint_t)(omp_block_start + omp_block_size - 1 - m);
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
memset(buckets, 0, 2 * (size_t)k * sizeof(sa_sint_t));
fast_sint_t m = omp_block_start + omp_block_size - 1;
if (omp_block_size > 0)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j = m + 1, c0 = T[m], c1 = -1;
while (j < n && (c1 = T[j]) == c0) { ++j; }
fast_uint_t s = c0 >= c1;
for (i = m - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 0], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 1], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 2], 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 3], 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 0); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 2); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
c1 = (i >= 0) ? T[i] : -1; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((s & 3) == 1);
buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
}
return (sa_sint_t)(omp_block_start + omp_block_size - 1 - m);
}
static sa_sint_t libsais_count_and_gather_compacted_lms_suffixes_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
memset(buckets, 0, 2 * (size_t)k * sizeof(sa_sint_t));
fast_sint_t m = omp_block_start + omp_block_size - 1;
if (omp_block_size > 0)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j = m + 1, c0 = T[m], c1 = -1;
while (j < n && (c1 = T[j]) == c0) { ++j; }
fast_uint_t s = c0 >= c1;
for (i = m - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 0] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 1] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 2] & SAINT_MAX, 0)]);
libsais_prefetchw(&buckets[BUCKETS_INDEX2(T[i - prefetch_distance - 3] & SAINT_MAX, 0)]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((fast_sint_t)(s & 3) == (c0 >= 0));
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 0); m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 1); m -= ((fast_sint_t)(s & 3) == (c0 >= 0));
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i - 2); m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((fast_sint_t)(s & 3) == (c1 >= 0));
c1 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c1, (s & 3) == 1)]++;
}
c1 = (i >= 0) ? T[i] : -1; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1))); SA[m] = (sa_sint_t)(i + 1); m -= ((fast_sint_t)(s & 3) == (c0 >= 0));
c0 &= SAINT_MAX; buckets[BUCKETS_INDEX2((fast_uint_t)c0, (s & 3) == 1)]++;
}
return (sa_sint_t)(omp_block_start + omp_block_size - 1 - m);
}
#if defined(_OPENMP)
static fast_sint_t libsais_get_bucket_stride(fast_sint_t free_space, fast_sint_t bucket_size, fast_sint_t num_buckets)
{
fast_sint_t bucket_size_1024 = (bucket_size + 1023) & (-1024); if (free_space / (num_buckets - 1) >= bucket_size_1024) { return bucket_size_1024; }
fast_sint_t bucket_size_16 = (bucket_size + 15) & (-16); if (free_space / (num_buckets - 1) >= bucket_size_16) { return bucket_size_16; }
return bucket_size;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_4k_fs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_lms_suffixes_32s_4k(T, SA, n, k, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
fast_sint_t bucket_size = 4 * (fast_sint_t)k;
fast_sint_t bucket_stride = libsais_get_bucket_stride(buckets - &SA[n], bucket_size, omp_num_threads);
{
thread_state[omp_thread_num].state.position = omp_block_start + omp_block_size;
thread_state[omp_thread_num].state.count = libsais_count_and_gather_lms_suffixes_32s_4k(T, SA, n, k, buckets - (omp_thread_num * bucket_stride), omp_block_start, omp_block_size);
}
#pragma omp barrier
if (omp_thread_num == omp_num_threads - 1)
{
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
m += (sa_sint_t)thread_state[t].state.count;
if (t != omp_num_threads - 1 && thread_state[t].state.count > 0)
{
memcpy(&SA[n - m], &SA[thread_state[t].state.position - thread_state[t].state.count], (size_t)thread_state[t].state.count * sizeof(sa_sint_t));
}
}
}
else
{
omp_num_threads = omp_num_threads - 1;
omp_block_stride = (bucket_size / omp_num_threads) & (-16);
omp_block_start = omp_thread_num * omp_block_stride;
omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : bucket_size - omp_block_start;
libsais_accumulate_counts_s32(buckets + omp_block_start, omp_block_size, bucket_stride, omp_num_threads + 1);
}
}
#endif
}
return m;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_2k_fs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_lms_suffixes_32s_2k(T, SA, n, k, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
fast_sint_t bucket_size = 2 * (fast_sint_t)k;
fast_sint_t bucket_stride = libsais_get_bucket_stride(buckets - &SA[n], bucket_size, omp_num_threads);
{
thread_state[omp_thread_num].state.position = omp_block_start + omp_block_size;
thread_state[omp_thread_num].state.count = libsais_count_and_gather_lms_suffixes_32s_2k(T, SA, n, k, buckets - (omp_thread_num * bucket_stride), omp_block_start, omp_block_size);
}
#pragma omp barrier
if (omp_thread_num == omp_num_threads - 1)
{
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
m += (sa_sint_t)thread_state[t].state.count;
if (t != omp_num_threads - 1 && thread_state[t].state.count > 0)
{
memcpy(&SA[n - m], &SA[thread_state[t].state.position - thread_state[t].state.count], (size_t)thread_state[t].state.count * sizeof(sa_sint_t));
}
}
}
else
{
omp_num_threads = omp_num_threads - 1;
omp_block_stride = (bucket_size / omp_num_threads) & (-16);
omp_block_start = omp_thread_num * omp_block_stride;
omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : bucket_size - omp_block_start;
libsais_accumulate_counts_s32(buckets + omp_block_start, omp_block_size, bucket_stride, omp_num_threads + 1);
}
}
#endif
}
return m;
}
static void libsais_count_and_gather_compacted_lms_suffixes_32s_2k_fs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
libsais_count_and_gather_compacted_lms_suffixes_32s_2k(T, SA, n, k, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
fast_sint_t bucket_size = 2 * (fast_sint_t)k;
fast_sint_t bucket_stride = libsais_get_bucket_stride(buckets - &SA[n + n], bucket_size, omp_num_threads);
{
thread_state[omp_thread_num].state.position = omp_block_start + omp_block_size;
thread_state[omp_thread_num].state.count = libsais_count_and_gather_compacted_lms_suffixes_32s_2k(T, SA + n, n, k, buckets - (omp_thread_num * bucket_stride), omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, m = 0; for (t = omp_num_threads - 1; t >= omp_thread_num; --t) { m += (sa_sint_t)thread_state[t].state.count; }
if (thread_state[omp_thread_num].state.count > 0)
{
memcpy(&SA[n - m], &SA[n + thread_state[omp_thread_num].state.position - thread_state[omp_thread_num].state.count], (size_t)thread_state[omp_thread_num].state.count * sizeof(sa_sint_t));
}
}
{
omp_block_stride = (bucket_size / omp_num_threads) & (-16);
omp_block_start = omp_thread_num * omp_block_stride;
omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : bucket_size - omp_block_start;
libsais_accumulate_counts_s32(buckets + omp_block_start, omp_block_size, bucket_stride, omp_num_threads);
}
}
#endif
}
}
#endif
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_4k_nofs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(2) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads);
fast_sint_t omp_num_threads = 1;
#endif
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_lms_suffixes_32s_4k(T, SA, n, k, buckets, 0, n);
}
#if defined(_OPENMP)
else if (omp_thread_num == 0)
{
libsais_count_lms_suffixes_32s_4k(T, n, k, buckets);
}
else
{
m = libsais_gather_lms_suffixes_32s(T, SA, n);
}
#endif
}
return m;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_2k_nofs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(2) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads);
fast_sint_t omp_num_threads = 1;
#endif
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_lms_suffixes_32s_2k(T, SA, n, k, buckets, 0, n);
}
#if defined(_OPENMP)
else if (omp_thread_num == 0)
{
libsais_count_lms_suffixes_32s_2k(T, n, k, buckets);
}
else
{
m = libsais_gather_lms_suffixes_32s(T, SA, n);
}
#endif
}
return m;
}
static sa_sint_t libsais_count_and_gather_compacted_lms_suffixes_32s_2k_nofs_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads)
{
sa_sint_t m = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(2) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads);
fast_sint_t omp_num_threads = 1;
#endif
if (omp_num_threads == 1)
{
m = libsais_count_and_gather_compacted_lms_suffixes_32s_2k(T, SA, n, k, buckets, 0, n);
}
#if defined(_OPENMP)
else if (omp_thread_num == 0)
{
libsais_count_compacted_lms_suffixes_32s_2k(T, n, k, buckets);
}
else
{
m = libsais_gather_compacted_lms_suffixes_32s(T, SA, n);
}
#endif
}
return m;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_4k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m;
#if defined(_OPENMP)
sa_sint_t max_threads = (sa_sint_t)((buckets - &SA[n]) / ((4 * (fast_sint_t)k + 15) & (-16))); if (max_threads > threads) { max_threads = threads; }
if (max_threads > 1 && n >= 65536 && n / k >= 2)
{
if (max_threads > n / 16 / k) { max_threads = n / 16 / k; }
m = libsais_count_and_gather_lms_suffixes_32s_4k_fs_omp(T, SA, n, k, buckets, max_threads > 2 ? max_threads : 2, thread_state);
}
else
#else
UNUSED(thread_state);
#endif
{
m = libsais_count_and_gather_lms_suffixes_32s_4k_nofs_omp(T, SA, n, k, buckets, threads);
}
return m;
}
static sa_sint_t libsais_count_and_gather_lms_suffixes_32s_2k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m;
#if defined(_OPENMP)
sa_sint_t max_threads = (sa_sint_t)((buckets - &SA[n]) / ((2 * (fast_sint_t)k + 15) & (-16))); if (max_threads > threads) { max_threads = threads; }
if (max_threads > 1 && n >= 65536 && n / k >= 2)
{
if (max_threads > n / 8 / k) { max_threads = n / 8 / k; }
m = libsais_count_and_gather_lms_suffixes_32s_2k_fs_omp(T, SA, n, k, buckets, max_threads > 2 ? max_threads : 2, thread_state);
}
else
#else
UNUSED(thread_state);
#endif
{
m = libsais_count_and_gather_lms_suffixes_32s_2k_nofs_omp(T, SA, n, k, buckets, threads);
}
return m;
}
static void libsais_count_and_gather_compacted_lms_suffixes_32s_2k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
sa_sint_t max_threads = (sa_sint_t)((buckets - &SA[n + n]) / ((2 * (fast_sint_t)k + 15) & (-16))); if (max_threads > threads) { max_threads = threads; }
if (max_threads > 1 && n >= 65536 && n / k >= 2)
{
if (max_threads > n / 8 / k) { max_threads = n / 8 / k; }
libsais_count_and_gather_compacted_lms_suffixes_32s_2k_fs_omp(T, SA, n, k, buckets, max_threads > 2 ? max_threads : 2, thread_state);
}
else
#else
UNUSED(thread_state);
#endif
{
libsais_count_and_gather_compacted_lms_suffixes_32s_2k_nofs_omp(T, SA, n, k, buckets, threads);
}
}
static void libsais_count_suffixes_32s(const sa_sint_t * RESTRICT T, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, (size_t)k * sizeof(sa_sint_t));
fast_sint_t i, j;
for (i = 0, j = (fast_sint_t)n - 7; i < j; i += 8)
{
libsais_prefetch(&T[i + prefetch_distance]);
buckets[T[i + 0]]++;
buckets[T[i + 1]]++;
buckets[T[i + 2]]++;
buckets[T[i + 3]]++;
buckets[T[i + 4]]++;
buckets[T[i + 5]]++;
buckets[T[i + 6]]++;
buckets[T[i + 7]]++;
}
for (j += 7; i < j; i += 1)
{
buckets[T[i]]++;
}
}
static void libsais_initialize_buckets_start_and_end_8u(sa_sint_t * RESTRICT buckets)
{
sa_sint_t * RESTRICT bucket_start = &buckets[6 * ALPHABET_SIZE];
sa_sint_t * RESTRICT bucket_end = &buckets[7 * ALPHABET_SIZE];
fast_sint_t i, j; sa_sint_t sum = 0;
for (i = BUCKETS_INDEX4(0, 0), j = 0; i <= BUCKETS_INDEX4(UCHAR_MAX, 0); i += BUCKETS_INDEX4(1, 0), j += 1)
{
bucket_start[j] = sum;
sum += buckets[i + BUCKETS_INDEX4(0, 0)] + buckets[i + BUCKETS_INDEX4(0, 1)] + buckets[i + BUCKETS_INDEX4(0, 2)] + buckets[i + BUCKETS_INDEX4(0, 3)];
bucket_end[j] = sum;
}
}
static void libsais_initialize_buckets_start_and_end_32s_6k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
sa_sint_t * RESTRICT bucket_start = &buckets[4 * k];
sa_sint_t * RESTRICT bucket_end = &buckets[5 * k];
fast_sint_t i, j; sa_sint_t sum = 0;
for (i = BUCKETS_INDEX4(0, 0), j = 0; i <= BUCKETS_INDEX4((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX4(1, 0), j += 1)
{
bucket_start[j] = sum;
sum += buckets[i + BUCKETS_INDEX4(0, 0)] + buckets[i + BUCKETS_INDEX4(0, 1)] + buckets[i + BUCKETS_INDEX4(0, 2)] + buckets[i + BUCKETS_INDEX4(0, 3)];
bucket_end[j] = sum;
}
}
static void libsais_initialize_buckets_start_and_end_32s_4k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
sa_sint_t * RESTRICT bucket_start = &buckets[2 * k];
sa_sint_t * RESTRICT bucket_end = &buckets[3 * k];
fast_sint_t i, j; sa_sint_t sum = 0;
for (i = BUCKETS_INDEX2(0, 0), j = 0; i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0), j += 1)
{
bucket_start[j] = sum;
sum += buckets[i + BUCKETS_INDEX2(0, 0)] + buckets[i + BUCKETS_INDEX2(0, 1)];
bucket_end[j] = sum;
}
}
static void libsais_initialize_buckets_end_32s_2k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
fast_sint_t i; sa_sint_t sum0 = 0;
for (i = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0))
{
sum0 += buckets[i + BUCKETS_INDEX2(0, 0)] + buckets[i + BUCKETS_INDEX2(0, 1)]; buckets[i + BUCKETS_INDEX2(0, 0)] = sum0;
}
}
static void libsais_initialize_buckets_start_and_end_32s_2k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
fast_sint_t i, j;
for (i = BUCKETS_INDEX2(0, 0), j = 0; i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0), j += 1)
{
buckets[j] = buckets[i];
}
buckets[k] = 0; memcpy(&buckets[k + 1], buckets, ((size_t)k - 1) * sizeof(sa_sint_t));
}
static void libsais_initialize_buckets_start_32s_1k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
fast_sint_t i; sa_sint_t sum = 0;
for (i = 0; i <= (fast_sint_t)k - 1; i += 1) { sa_sint_t tmp = buckets[i]; buckets[i] = sum; sum += tmp; }
}
static void libsais_initialize_buckets_end_32s_1k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
fast_sint_t i; sa_sint_t sum = 0;
for (i = 0; i <= (fast_sint_t)k - 1; i += 1) { sum += buckets[i]; buckets[i] = sum; }
}
static sa_sint_t libsais_initialize_buckets_for_lms_suffixes_radix_sort_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix)
{
{
fast_uint_t s = 0;
fast_sint_t c0 = T[first_lms_suffix];
fast_sint_t c1 = 0;
for (; --first_lms_suffix >= 0; )
{
c1 = c0; c0 = T[first_lms_suffix]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]--;
}
buckets[BUCKETS_INDEX4((fast_uint_t)c0, (s << 1) & 3)]--;
}
{
sa_sint_t * RESTRICT temp_bucket = &buckets[4 * ALPHABET_SIZE];
fast_sint_t i, j; sa_sint_t sum = 0;
for (i = BUCKETS_INDEX4(0, 0), j = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX4(UCHAR_MAX, 0); i += BUCKETS_INDEX4(1, 0), j += BUCKETS_INDEX2(1, 0))
{
temp_bucket[j + BUCKETS_INDEX2(0, 1)] = sum; sum += buckets[i + BUCKETS_INDEX4(0, 1)] + buckets[i + BUCKETS_INDEX4(0, 3)]; temp_bucket[j] = sum;
}
return sum;
}
}
static void libsais_initialize_buckets_for_lms_suffixes_radix_sort_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix)
{
buckets[BUCKETS_INDEX2(T[first_lms_suffix], 0)]++;
buckets[BUCKETS_INDEX2(T[first_lms_suffix], 1)]--;
fast_sint_t i; sa_sint_t sum0 = 0, sum1 = 0;
for (i = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0))
{
sum0 += buckets[i + BUCKETS_INDEX2(0, 0)] + buckets[i + BUCKETS_INDEX2(0, 1)];
sum1 += buckets[i + BUCKETS_INDEX2(0, 1)];
buckets[i + BUCKETS_INDEX2(0, 0)] = sum0;
buckets[i + BUCKETS_INDEX2(0, 1)] = sum1;
}
}
static sa_sint_t libsais_initialize_buckets_for_lms_suffixes_radix_sort_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix)
{
{
fast_uint_t s = 0;
fast_sint_t c0 = T[first_lms_suffix];
fast_sint_t c1 = 0;
for (; --first_lms_suffix >= 0; )
{
c1 = c0; c0 = T[first_lms_suffix]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
buckets[BUCKETS_INDEX4((fast_uint_t)c1, s & 3)]--;
}
buckets[BUCKETS_INDEX4((fast_uint_t)c0, (s << 1) & 3)]--;
}
{
sa_sint_t * RESTRICT temp_bucket = &buckets[4 * k];
fast_sint_t i, j; sa_sint_t sum = 0;
for (i = BUCKETS_INDEX4(0, 0), j = 0; i <= BUCKETS_INDEX4((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX4(1, 0), j += 1)
{
sum += buckets[i + BUCKETS_INDEX4(0, 1)] + buckets[i + BUCKETS_INDEX4(0, 3)]; temp_bucket[j] = sum;
}
return sum;
}
}
static void libsais_initialize_buckets_for_radix_and_partial_sorting_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix)
{
sa_sint_t * RESTRICT bucket_start = &buckets[2 * k];
sa_sint_t * RESTRICT bucket_end = &buckets[3 * k];
buckets[BUCKETS_INDEX2(T[first_lms_suffix], 0)]++;
buckets[BUCKETS_INDEX2(T[first_lms_suffix], 1)]--;
fast_sint_t i, j; sa_sint_t sum0 = 0, sum1 = 0;
for (i = BUCKETS_INDEX2(0, 0), j = 0; i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0), j += 1)
{
bucket_start[j] = sum1;
sum0 += buckets[i + BUCKETS_INDEX2(0, 1)];
sum1 += buckets[i + BUCKETS_INDEX2(0, 0)] + buckets[i + BUCKETS_INDEX2(0, 1)];
buckets[i + BUCKETS_INDEX2(0, 1)] = sum0;
bucket_end[j] = sum1;
}
}
static void libsais_radix_sort_lms_suffixes_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&SA[i - 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i - prefetch_distance - 0]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 1]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 2]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 3]]);
sa_sint_t p0 = SA[i - 0]; SA[--induction_bucket[BUCKETS_INDEX2(T[p0], 0)]] = p0;
sa_sint_t p1 = SA[i - 1]; SA[--induction_bucket[BUCKETS_INDEX2(T[p1], 0)]] = p1;
sa_sint_t p2 = SA[i - 2]; SA[--induction_bucket[BUCKETS_INDEX2(T[p2], 0)]] = p2;
sa_sint_t p3 = SA[i - 3]; SA[--induction_bucket[BUCKETS_INDEX2(T[p3], 0)]] = p3;
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[--induction_bucket[BUCKETS_INDEX2(T[p], 0)]] = p;
}
}
static void libsais_radix_sort_lms_suffixes_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536 && m >= 65536 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_num_threads = 1;
#endif
if (omp_num_threads == 1)
{
libsais_radix_sort_lms_suffixes_8u(T, SA, &buckets[4 * ALPHABET_SIZE], (fast_sint_t)n - (fast_sint_t)m + 1, (fast_sint_t)m - 1);
}
#if defined(_OPENMP)
else
{
{
sa_sint_t * RESTRICT src_bucket = &buckets[4 * ALPHABET_SIZE];
sa_sint_t * RESTRICT dst_bucket = thread_state[omp_thread_num].state.buckets;
fast_sint_t i, j;
for (i = BUCKETS_INDEX2(0, 0), j = BUCKETS_INDEX4(0, 1); i <= BUCKETS_INDEX2(UCHAR_MAX, 0); i += BUCKETS_INDEX2(1, 0), j += BUCKETS_INDEX4(1, 0))
{
dst_bucket[i] = src_bucket[i] - dst_bucket[j];
}
}
{
fast_sint_t t, omp_block_start = 0, omp_block_size = thread_state[omp_thread_num].state.m;
for (t = omp_num_threads - 1; t >= omp_thread_num; --t) omp_block_start += thread_state[t].state.m;
if (omp_block_start == (fast_sint_t)m && omp_block_size > 0)
{
omp_block_start -= 1; omp_block_size -= 1;
}
libsais_radix_sort_lms_suffixes_8u(T, SA, thread_state[omp_thread_num].state.buckets, (fast_sint_t)n - omp_block_start, omp_block_size);
}
}
#endif
}
}
static void libsais_radix_sort_lms_suffixes_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&SA[i - 3 * prefetch_distance]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 0]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 1]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 2]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 3]]);
libsais_prefetchw(&induction_bucket[T[SA[i - prefetch_distance - 0]]]);
libsais_prefetchw(&induction_bucket[T[SA[i - prefetch_distance - 1]]]);
libsais_prefetchw(&induction_bucket[T[SA[i - prefetch_distance - 2]]]);
libsais_prefetchw(&induction_bucket[T[SA[i - prefetch_distance - 3]]]);
sa_sint_t p0 = SA[i - 0]; SA[--induction_bucket[T[p0]]] = p0;
sa_sint_t p1 = SA[i - 1]; SA[--induction_bucket[T[p1]]] = p1;
sa_sint_t p2 = SA[i - 2]; SA[--induction_bucket[T[p2]]] = p2;
sa_sint_t p3 = SA[i - 3]; SA[--induction_bucket[T[p3]]] = p3;
}
for (j -= 2 * prefetch_distance + 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[--induction_bucket[T[p]]] = p;
}
}
static void libsais_radix_sort_lms_suffixes_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetch(&SA[i - 3 * prefetch_distance]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 0]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 1]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 2]]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 3]]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(T[SA[i - prefetch_distance - 0]], 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(T[SA[i - prefetch_distance - 1]], 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(T[SA[i - prefetch_distance - 2]], 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(T[SA[i - prefetch_distance - 3]], 0)]);
sa_sint_t p0 = SA[i - 0]; SA[--induction_bucket[BUCKETS_INDEX2(T[p0], 0)]] = p0;
sa_sint_t p1 = SA[i - 1]; SA[--induction_bucket[BUCKETS_INDEX2(T[p1], 0)]] = p1;
sa_sint_t p2 = SA[i - 2]; SA[--induction_bucket[BUCKETS_INDEX2(T[p2], 0)]] = p2;
sa_sint_t p3 = SA[i - 3]; SA[--induction_bucket[BUCKETS_INDEX2(T[p3], 0)]] = p3;
}
for (j -= 2 * prefetch_distance + 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[--induction_bucket[BUCKETS_INDEX2(T[p], 0)]] = p;
}
}
#if defined(_OPENMP)
static void libsais_radix_sort_lms_suffixes_32s_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i + prefetch_distance + 0]]);
libsais_prefetch(&T[SA[i + prefetch_distance + 1]]);
libsais_prefetch(&T[SA[i + prefetch_distance + 2]]);
libsais_prefetch(&T[SA[i + prefetch_distance + 3]]);
libsais_prefetchw(&cache[i + prefetch_distance]);
cache[i + 0].symbol = T[cache[i + 0].index = SA[i + 0]];
cache[i + 1].symbol = T[cache[i + 1].index = SA[i + 1]];
cache[i + 2].symbol = T[cache[i + 2].index = SA[i + 2]];
cache[i + 3].symbol = T[cache[i + 3].index = SA[i + 3]];
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
cache[i].symbol = T[cache[i].index = SA[i]];
}
}
static void libsais_radix_sort_lms_suffixes_32s_6k_block_sort(sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
libsais_prefetchw(&induction_bucket[cache[i - prefetch_distance - 0].symbol]);
libsais_prefetchw(&induction_bucket[cache[i - prefetch_distance - 1].symbol]);
libsais_prefetchw(&induction_bucket[cache[i - prefetch_distance - 2].symbol]);
libsais_prefetchw(&induction_bucket[cache[i - prefetch_distance - 3].symbol]);
cache[i - 0].symbol = --induction_bucket[cache[i - 0].symbol];
cache[i - 1].symbol = --induction_bucket[cache[i - 1].symbol];
cache[i - 2].symbol = --induction_bucket[cache[i - 2].symbol];
cache[i - 3].symbol = --induction_bucket[cache[i - 3].symbol];
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
cache[i].symbol = --induction_bucket[cache[i].symbol];
}
}
static void libsais_radix_sort_lms_suffixes_32s_2k_block_sort(sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 3; i >= j; i -= 4)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(cache[i - prefetch_distance - 0].symbol, 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(cache[i - prefetch_distance - 1].symbol, 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(cache[i - prefetch_distance - 2].symbol, 0)]);
libsais_prefetchw(&induction_bucket[BUCKETS_INDEX2(cache[i - prefetch_distance - 3].symbol, 0)]);
cache[i - 0].symbol = --induction_bucket[BUCKETS_INDEX2(cache[i - 0].symbol, 0)];
cache[i - 1].symbol = --induction_bucket[BUCKETS_INDEX2(cache[i - 1].symbol, 0)];
cache[i - 2].symbol = --induction_bucket[BUCKETS_INDEX2(cache[i - 2].symbol, 0)];
cache[i - 3].symbol = --induction_bucket[BUCKETS_INDEX2(cache[i - 3].symbol, 0)];
}
for (j -= prefetch_distance + 3; i >= j; i -= 1)
{
cache[i].symbol = --induction_bucket[BUCKETS_INDEX2(cache[i].symbol, 0)];
}
}
static void libsais_radix_sort_lms_suffixes_32s_6k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_radix_sort_lms_suffixes_32s_6k(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_radix_sort_lms_suffixes_32s_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_radix_sort_lms_suffixes_32s_6k_block_sort(induction_bucket, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
static void libsais_radix_sort_lms_suffixes_32s_2k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_radix_sort_lms_suffixes_32s_2k(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_radix_sort_lms_suffixes_32s_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_radix_sort_lms_suffixes_32s_2k_block_sort(induction_bucket, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
#endif
static void libsais_radix_sort_lms_suffixes_32s_6k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || m < 65536)
{
libsais_radix_sort_lms_suffixes_32s_6k(T, SA, induction_bucket, (fast_sint_t)n - (fast_sint_t)m + 1, (fast_sint_t)m - 1);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < (fast_sint_t)m - 1; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end >= m) { block_end = (fast_sint_t)m - 1; }
libsais_radix_sort_lms_suffixes_32s_6k_block_omp(T, SA, induction_bucket, thread_state[0].state.cache, (fast_sint_t)n - block_end, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_radix_sort_lms_suffixes_32s_2k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || m < 65536)
{
libsais_radix_sort_lms_suffixes_32s_2k(T, SA, induction_bucket, (fast_sint_t)n - (fast_sint_t)m + 1, (fast_sint_t)m - 1);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < (fast_sint_t)m - 1; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end >= m) { block_end = (fast_sint_t)m - 1; }
libsais_radix_sort_lms_suffixes_32s_2k_block_omp(T, SA, induction_bucket, thread_state[0].state.cache, (fast_sint_t)n - block_end, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static sa_sint_t libsais_radix_sort_lms_suffixes_32s_1k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t i = n - 2;
sa_sint_t m = 0;
fast_uint_t s = 1;
fast_sint_t c0 = T[n - 1];
fast_sint_t c1 = 0;
fast_sint_t c2 = 0;
for (; i >= prefetch_distance + 3; i -= 4)
{
libsais_prefetch(&T[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[T[i - prefetch_distance - 0]]);
libsais_prefetchw(&buckets[T[i - prefetch_distance - 1]]);
libsais_prefetchw(&buckets[T[i - prefetch_distance - 2]]);
libsais_prefetchw(&buckets[T[i - prefetch_distance - 3]]);
c1 = T[i - 0]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
if ((s & 3) == 1) { SA[--buckets[c2 = c0]] = i + 1; m++; }
c0 = T[i - 1]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
if ((s & 3) == 1) { SA[--buckets[c2 = c1]] = i - 0; m++; }
c1 = T[i - 2]; s = (s << 1) + (fast_uint_t)(c1 > (c0 - (fast_sint_t)(s & 1)));
if ((s & 3) == 1) { SA[--buckets[c2 = c0]] = i - 1; m++; }
c0 = T[i - 3]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
if ((s & 3) == 1) { SA[--buckets[c2 = c1]] = i - 2; m++; }
}
for (; i >= 0; i -= 1)
{
c1 = c0; c0 = T[i]; s = (s << 1) + (fast_uint_t)(c0 > (c1 - (fast_sint_t)(s & 1)));
if ((s & 3) == 1) { SA[--buckets[c2 = c1]] = i + 1; m++; }
}
if (m > 1)
{
SA[buckets[c2]] = 0;
}
return m;
}
static void libsais_radix_sort_set_markers_32s_6k(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&induction_bucket[i + 2 * prefetch_distance]);
libsais_prefetchw(&SA[induction_bucket[i + prefetch_distance + 0]]);
libsais_prefetchw(&SA[induction_bucket[i + prefetch_distance + 1]]);
libsais_prefetchw(&SA[induction_bucket[i + prefetch_distance + 2]]);
libsais_prefetchw(&SA[induction_bucket[i + prefetch_distance + 3]]);
SA[induction_bucket[i + 0]] |= SAINT_MIN;
SA[induction_bucket[i + 1]] |= SAINT_MIN;
SA[induction_bucket[i + 2]] |= SAINT_MIN;
SA[induction_bucket[i + 3]] |= SAINT_MIN;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
SA[induction_bucket[i]] |= SAINT_MIN;
}
}
static void libsais_radix_sort_set_markers_32s_4k(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&induction_bucket[BUCKETS_INDEX2(i + 2 * prefetch_distance, 0)]);
libsais_prefetchw(&SA[induction_bucket[BUCKETS_INDEX2(i + prefetch_distance + 0, 0)]]);
libsais_prefetchw(&SA[induction_bucket[BUCKETS_INDEX2(i + prefetch_distance + 1, 0)]]);
libsais_prefetchw(&SA[induction_bucket[BUCKETS_INDEX2(i + prefetch_distance + 2, 0)]]);
libsais_prefetchw(&SA[induction_bucket[BUCKETS_INDEX2(i + prefetch_distance + 3, 0)]]);
SA[induction_bucket[BUCKETS_INDEX2(i + 0, 0)]] |= SUFFIX_GROUP_MARKER;
SA[induction_bucket[BUCKETS_INDEX2(i + 1, 0)]] |= SUFFIX_GROUP_MARKER;
SA[induction_bucket[BUCKETS_INDEX2(i + 2, 0)]] |= SUFFIX_GROUP_MARKER;
SA[induction_bucket[BUCKETS_INDEX2(i + 3, 0)]] |= SUFFIX_GROUP_MARKER;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
SA[induction_bucket[BUCKETS_INDEX2(i, 0)]] |= SUFFIX_GROUP_MARKER;
}
}
static void libsais_radix_sort_set_markers_32s_6k_omp(sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && k >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = (((fast_sint_t)k - 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : (fast_sint_t)k - 1 - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = (fast_sint_t)k - 1;
#endif
libsais_radix_sort_set_markers_32s_6k(SA, induction_bucket, omp_block_start, omp_block_size);
}
}
static void libsais_radix_sort_set_markers_32s_4k_omp(sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && k >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = (((fast_sint_t)k - 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : (fast_sint_t)k - 1 - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = (fast_sint_t)k - 1;
#endif
libsais_radix_sort_set_markers_32s_4k(SA, induction_bucket, omp_block_start, omp_block_size);
}
}
static void libsais_initialize_buckets_for_partial_sorting_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count)
{
sa_sint_t * RESTRICT temp_bucket = &buckets[4 * ALPHABET_SIZE];
buckets[BUCKETS_INDEX4((fast_uint_t)T[first_lms_suffix], 1)]++;
fast_sint_t i, j; sa_sint_t sum0 = left_suffixes_count + 1, sum1 = 0;
for (i = BUCKETS_INDEX4(0, 0), j = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX4(UCHAR_MAX, 0); i += BUCKETS_INDEX4(1, 0), j += BUCKETS_INDEX2(1, 0))
{
temp_bucket[j + BUCKETS_INDEX2(0, 0)] = sum0;
sum0 += buckets[i + BUCKETS_INDEX4(0, 0)] + buckets[i + BUCKETS_INDEX4(0, 2)];
sum1 += buckets[i + BUCKETS_INDEX4(0, 1)];
buckets[j + BUCKETS_INDEX2(0, 0)] = sum0;
buckets[j + BUCKETS_INDEX2(0, 1)] = sum1;
}
}
static void libsais_initialize_buckets_for_partial_sorting_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count)
{
sa_sint_t * RESTRICT temp_bucket = &buckets[4 * k];
fast_sint_t i, j; sa_sint_t sum0 = left_suffixes_count + 1, sum1 = 0, sum2 = 0;
for (first_lms_suffix = T[first_lms_suffix], i = BUCKETS_INDEX4(0, 0), j = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX4((fast_sint_t)first_lms_suffix - 1, 0); i += BUCKETS_INDEX4(1, 0), j += BUCKETS_INDEX2(1, 0))
{
sa_sint_t SS = buckets[i + BUCKETS_INDEX4(0, 0)];
sa_sint_t LS = buckets[i + BUCKETS_INDEX4(0, 1)];
sa_sint_t SL = buckets[i + BUCKETS_INDEX4(0, 2)];
sa_sint_t LL = buckets[i + BUCKETS_INDEX4(0, 3)];
buckets[i + BUCKETS_INDEX4(0, 0)] = sum0;
buckets[i + BUCKETS_INDEX4(0, 1)] = sum2;
buckets[i + BUCKETS_INDEX4(0, 2)] = 0;
buckets[i + BUCKETS_INDEX4(0, 3)] = 0;
sum0 += SS + SL; sum1 += LS; sum2 += LS + LL;
temp_bucket[j + BUCKETS_INDEX2(0, 0)] = sum0;
temp_bucket[j + BUCKETS_INDEX2(0, 1)] = sum1;
}
for (sum1 += 1; i <= BUCKETS_INDEX4((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX4(1, 0), j += BUCKETS_INDEX2(1, 0))
{
sa_sint_t SS = buckets[i + BUCKETS_INDEX4(0, 0)];
sa_sint_t LS = buckets[i + BUCKETS_INDEX4(0, 1)];
sa_sint_t SL = buckets[i + BUCKETS_INDEX4(0, 2)];
sa_sint_t LL = buckets[i + BUCKETS_INDEX4(0, 3)];
buckets[i + BUCKETS_INDEX4(0, 0)] = sum0;
buckets[i + BUCKETS_INDEX4(0, 1)] = sum2;
buckets[i + BUCKETS_INDEX4(0, 2)] = 0;
buckets[i + BUCKETS_INDEX4(0, 3)] = 0;
sum0 += SS + SL; sum1 += LS; sum2 += LS + LL;
temp_bucket[j + BUCKETS_INDEX2(0, 0)] = sum0;
temp_bucket[j + BUCKETS_INDEX2(0, 1)] = sum1;
}
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[4 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 2);
sa_sint_t p0 = SA[i + 0]; d += (p0 < 0); p0 &= SAINT_MAX; sa_sint_t v0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] >= T[p0 - 1]);
SA[induction_bucket[v0]++] = (p0 - 1) | ((sa_sint_t)(distinct_names[v0] != d) << (SAINT_BIT - 1)); distinct_names[v0] = d;
sa_sint_t p1 = SA[i + 1]; d += (p1 < 0); p1 &= SAINT_MAX; sa_sint_t v1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] >= T[p1 - 1]);
SA[induction_bucket[v1]++] = (p1 - 1) | ((sa_sint_t)(distinct_names[v1] != d) << (SAINT_BIT - 1)); distinct_names[v1] = d;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] >= T[p - 1]);
SA[induction_bucket[v]++] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
return d;
}
#if defined(_OPENMP)
static void libsais_partial_sorting_scan_left_to_right_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size, LIBSAIS_THREAD_STATE * RESTRICT state)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
memset(buckets, 0, 4 * ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0; sa_sint_t d = 1;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 2);
sa_sint_t p0 = cache[count].index = SA[i + 0]; d += (p0 < 0); p0 &= SAINT_MAX; sa_sint_t v0 = cache[count++].symbol = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] >= T[p0 - 1]); induction_bucket[v0]++; distinct_names[v0] = d;
sa_sint_t p1 = cache[count].index = SA[i + 1]; d += (p1 < 0); p1 &= SAINT_MAX; sa_sint_t v1 = cache[count++].symbol = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] >= T[p1 - 1]); induction_bucket[v1]++; distinct_names[v1] = d;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = cache[count].index = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = cache[count++].symbol = BUCKETS_INDEX2(T[p - 1], T[p - 2] >= T[p - 1]); induction_bucket[v]++; distinct_names[v] = d;
}
state[0].state.position = (fast_sint_t)d - 1;
state[0].state.count = count;
}
static void libsais_partial_sorting_scan_left_to_right_8u_block_place(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t count, sa_sint_t d)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t i, j;
for (i = 0, j = count - 1; i < j; i += 2)
{
libsais_prefetch(&cache[i + prefetch_distance]);
sa_sint_t p0 = cache[i + 0].index; d += (p0 < 0); sa_sint_t v0 = cache[i + 0].symbol;
SA[induction_bucket[v0]++] = (p0 - 1) | ((sa_sint_t)(distinct_names[v0] != d) << (SAINT_BIT - 1)); distinct_names[v0] = d;
sa_sint_t p1 = cache[i + 1].index; d += (p1 < 0); sa_sint_t v1 = cache[i + 1].symbol;
SA[induction_bucket[v1]++] = (p1 - 1) | ((sa_sint_t)(distinct_names[v1] != d) << (SAINT_BIT - 1)); distinct_names[v1] = d;
}
for (j += 1; i < j; i += 1)
{
sa_sint_t p = cache[i].index; d += (p < 0); sa_sint_t v = cache[i].symbol;
SA[induction_bucket[v]++] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_left_to_right_8u(T, SA, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_left_to_right_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size, &thread_state[omp_thread_num]);
}
#pragma omp barrier
#pragma omp master
{
sa_sint_t * RESTRICT induction_bucket = &buckets[4 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t t;
for (t = 0; t < omp_num_threads; ++t)
{
sa_sint_t * RESTRICT temp_induction_bucket = &thread_state[t].state.buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT temp_distinct_names = &thread_state[t].state.buckets[2 * ALPHABET_SIZE];
fast_sint_t c;
for (c = 0; c < 2 * ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_induction_bucket[c]; induction_bucket[c] = A + B; temp_induction_bucket[c] = A; }
for (d -= 1, c = 0; c < 2 * ALPHABET_SIZE; c += 1) { sa_sint_t A = distinct_names[c], B = temp_distinct_names[c], D = B + d; distinct_names[c] = B > 0 ? D : A; temp_distinct_names[c] = A; }
d += 1 + (sa_sint_t)thread_state[t].state.position; thread_state[t].state.position = (fast_sint_t)d - thread_state[t].state.position;
}
}
#pragma omp barrier
{
libsais_partial_sorting_scan_left_to_right_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count, (sa_sint_t)thread_state[omp_thread_num].state.position);
}
}
#endif
}
return d;
}
#endif
static sa_sint_t libsais_partial_sorting_scan_left_to_right_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t left_suffixes_count, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t * RESTRICT induction_bucket = &buckets[4 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
SA[induction_bucket[BUCKETS_INDEX2(T[n - 1], T[n - 2] >= T[n - 1])]++] = (n - 1) | SAINT_MIN;
distinct_names[BUCKETS_INDEX2(T[n - 1], T[n - 2] >= T[n - 1])] = ++d;
if (threads == 1 || left_suffixes_count < 65536)
{
d = libsais_partial_sorting_scan_left_to_right_8u(T, SA, buckets, d, 0, left_suffixes_count);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start;
for (block_start = 0; block_start < left_suffixes_count; )
{
if (SA[block_start] == 0)
{
block_start++;
}
else
{
fast_sint_t block_max_end = block_start + ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end > left_suffixes_count) { block_max_end = left_suffixes_count;}
fast_sint_t block_end = block_start + 1; while (block_end < block_max_end && SA[block_end] != 0) { block_end++; }
fast_sint_t block_size = block_end - block_start;
if (block_size < 32)
{
for (; block_start < block_end; block_start += 1)
{
sa_sint_t p = SA[block_start]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] >= T[p - 1]);
SA[induction_bucket[v]++] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
}
else
{
d = libsais_partial_sorting_scan_left_to_right_8u_block_omp(T, SA, buckets, d, block_start, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
return d;
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 2 * prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 3 * prefetch_distance]);
libsais_prefetch(&T[SA[i + 2 * prefetch_distance + 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + 2 * prefetch_distance + 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i + 2 * prefetch_distance + 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + 2 * prefetch_distance + 1] & SAINT_MAX] - 2);
sa_sint_t p0 = SA[i + prefetch_distance + 0] & SAINT_MAX; sa_sint_t v0 = BUCKETS_INDEX4(T[p0 - (p0 > 0)], 0); libsais_prefetchw(&buckets[v0]);
sa_sint_t p1 = SA[i + prefetch_distance + 1] & SAINT_MAX; sa_sint_t v1 = BUCKETS_INDEX4(T[p1 - (p1 > 0)], 0); libsais_prefetchw(&buckets[v1]);
sa_sint_t p2 = SA[i + 0]; d += (p2 < 0); p2 &= SAINT_MAX; sa_sint_t v2 = BUCKETS_INDEX4(T[p2 - 1], T[p2 - 2] >= T[p2 - 1]);
SA[buckets[v2]++] = (p2 - 1) | ((sa_sint_t)(buckets[2 + v2] != d) << (SAINT_BIT - 1)); buckets[2 + v2] = d;
sa_sint_t p3 = SA[i + 1]; d += (p3 < 0); p3 &= SAINT_MAX; sa_sint_t v3 = BUCKETS_INDEX4(T[p3 - 1], T[p3 - 2] >= T[p3 - 1]);
SA[buckets[v3]++] = (p3 - 1) | ((sa_sint_t)(buckets[2 + v3] != d) << (SAINT_BIT - 1)); buckets[2 + v3] = d;
}
for (j += 2 * prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX4(T[p - 1], T[p - 2] >= T[p - 1]);
SA[buckets[v]++] = (p - 1) | ((sa_sint_t)(buckets[2 + v] != d) << (SAINT_BIT - 1)); buckets[2 + v] = d;
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[2 * k];
sa_sint_t * RESTRICT distinct_names = &buckets[0 * k];
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 2 * prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 3 * prefetch_distance]);
sa_sint_t s0 = SA[i + 2 * prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + 2 * prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i + 1 * prefetch_distance + 0]; if (s2 > 0) { const fast_sint_t Ts2 = T[(s2 & ~SUFFIX_GROUP_MARKER) - 1]; libsais_prefetchw(&induction_bucket[Ts2]); libsais_prefetchw(&distinct_names[BUCKETS_INDEX2(Ts2, 0)]); }
sa_sint_t s3 = SA[i + 1 * prefetch_distance + 1]; if (s3 > 0) { const fast_sint_t Ts3 = T[(s3 & ~SUFFIX_GROUP_MARKER) - 1]; libsais_prefetchw(&induction_bucket[Ts3]); libsais_prefetchw(&distinct_names[BUCKETS_INDEX2(Ts3, 0)]); }
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 & SAINT_MAX;
if (p0 > 0)
{
SA[i + 0] = 0; d += (p0 >> (SUFFIX_GROUP_BIT - 1)); p0 &= ~SUFFIX_GROUP_MARKER; sa_sint_t v0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] < T[p0 - 1]);
SA[induction_bucket[T[p0 - 1]]++] = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] < T[p0 - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v0] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v0] = d;
}
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 & SAINT_MAX;
if (p1 > 0)
{
SA[i + 1] = 0; d += (p1 >> (SUFFIX_GROUP_BIT - 1)); p1 &= ~SUFFIX_GROUP_MARKER; sa_sint_t v1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] < T[p1 - 1]);
SA[induction_bucket[T[p1 - 1]]++] = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] < T[p1 - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v1] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v1] = d;
}
}
for (j += 2 * prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX;
if (p > 0)
{
SA[i] = 0; d += (p >> (SUFFIX_GROUP_BIT - 1)); p &= ~SUFFIX_GROUP_MARKER; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] < T[p - 1]);
SA[induction_bucket[T[p - 1]]++] = (p - 1) | ((sa_sint_t)(T[p - 2] < T[p - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v] = d;
}
}
return d;
}
static void libsais_partial_sorting_scan_left_to_right_32s_1k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 2 * prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 3 * prefetch_distance]);
sa_sint_t s0 = SA[i + 2 * prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + 2 * prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i + 1 * prefetch_distance + 0]; if (s2 > 0) { libsais_prefetchw(&induction_bucket[T[s2 - 1]]); libsais_prefetch(&T[s2] - 2); }
sa_sint_t s3 = SA[i + 1 * prefetch_distance + 1]; if (s3 > 0) { libsais_prefetchw(&induction_bucket[T[s3 - 1]]); libsais_prefetch(&T[s3] - 2); }
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 & SAINT_MAX; if (p0 > 0) { SA[i + 0] = 0; SA[induction_bucket[T[p0 - 1]]++] = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] < T[p0 - 1]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 & SAINT_MAX; if (p1 > 0) { SA[i + 1] = 0; SA[induction_bucket[T[p1 - 1]]++] = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] < T[p1 - 1]) << (SAINT_BIT - 1)); }
}
for (j += 2 * prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { SA[i] = 0; SA[induction_bucket[T[p - 1]]++] = (p - 1) | ((sa_sint_t)(T[p - 2] < T[p - 1]) << (SAINT_BIT - 1)); }
}
}
#if defined(_OPENMP)
static void libsais_partial_sorting_scan_left_to_right_32s_6k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 2);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t p0 = cache[i + 0].index = SA[i + 0]; sa_sint_t symbol0 = 0; p0 &= SAINT_MAX; if (p0 != 0) { symbol0 = BUCKETS_INDEX4(T[p0 - 1], T[p0 - 2] >= T[p0 - 1]); } cache[i + 0].symbol = symbol0;
sa_sint_t p1 = cache[i + 1].index = SA[i + 1]; sa_sint_t symbol1 = 0; p1 &= SAINT_MAX; if (p1 != 0) { symbol1 = BUCKETS_INDEX4(T[p1 - 1], T[p1 - 2] >= T[p1 - 1]); } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = cache[i].index = SA[i]; sa_sint_t symbol = 0; p &= SAINT_MAX; if (p != 0) { symbol = BUCKETS_INDEX4(T[p - 1], T[p - 2] >= T[p - 1]); } cache[i].symbol = symbol;
}
}
static void libsais_partial_sorting_scan_left_to_right_32s_4k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; if (p0 > 0) { cache[i + 0].index = p0; p0 &= ~SUFFIX_GROUP_MARKER; symbol0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] < T[p0 - 1]); p0 = 0; } cache[i + 0].symbol = symbol0; SA[i + 0] = p0 & SAINT_MAX;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; if (p1 > 0) { cache[i + 1].index = p1; p1 &= ~SUFFIX_GROUP_MARKER; symbol1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] < T[p1 - 1]); p1 = 0; } cache[i + 1].symbol = symbol1; SA[i + 1] = p1 & SAINT_MAX;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; if (p > 0) { cache[i].index = p; p &= ~SUFFIX_GROUP_MARKER; symbol = BUCKETS_INDEX2(T[p - 1], T[p - 2] < T[p - 1]); p = 0; } cache[i].symbol = symbol; SA[i] = p & SAINT_MAX;
}
}
static void libsais_partial_sorting_scan_left_to_right_32s_1k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; if (p0 > 0) { cache[i + 0].index = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] < T[p0 - 1]) << (SAINT_BIT - 1)); symbol0 = T[p0 - 1]; p0 = 0; } cache[i + 0].symbol = symbol0; SA[i + 0] = p0 & SAINT_MAX;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; if (p1 > 0) { cache[i + 1].index = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] < T[p1 - 1]) << (SAINT_BIT - 1)); symbol1 = T[p1 - 1]; p1 = 0; } cache[i + 1].symbol = symbol1; SA[i + 1] = p1 & SAINT_MAX;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; if (p > 0) { cache[i].index = (p - 1) | ((sa_sint_t)(T[p - 2] < T[p - 1]) << (SAINT_BIT - 1)); symbol = T[p - 1]; p = 0; } cache[i].symbol = symbol; SA[i] = p & SAINT_MAX;
}
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_6k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, omp_block_end = omp_block_start + omp_block_size;
for (i = omp_block_start, j = omp_block_end - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&cache[i + 2 * prefetch_distance]);
libsais_prefetchw(&buckets[cache[i + prefetch_distance + 0].symbol]);
libsais_prefetchw(&buckets[cache[i + prefetch_distance + 1].symbol]);
sa_sint_t v0 = cache[i + 0].symbol, p0 = cache[i + 0].index; d += (p0 < 0); cache[i + 0].symbol = buckets[v0]++; cache[i + 0].index = (p0 - 1) | ((sa_sint_t)(buckets[2 + v0] != d) << (SAINT_BIT - 1)); buckets[2 + v0] = d;
if (cache[i + 0].symbol < omp_block_end) { sa_sint_t s = cache[i + 0].symbol, q = (cache[s].index = cache[i + 0].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] >= T[q - 1]); }
sa_sint_t v1 = cache[i + 1].symbol, p1 = cache[i + 1].index; d += (p1 < 0); cache[i + 1].symbol = buckets[v1]++; cache[i + 1].index = (p1 - 1) | ((sa_sint_t)(buckets[2 + v1] != d) << (SAINT_BIT - 1)); buckets[2 + v1] = d;
if (cache[i + 1].symbol < omp_block_end) { sa_sint_t s = cache[i + 1].symbol, q = (cache[s].index = cache[i + 1].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] >= T[q - 1]); }
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t v = cache[i].symbol, p = cache[i].index; d += (p < 0); cache[i].symbol = buckets[v]++; cache[i].index = (p - 1) | ((sa_sint_t)(buckets[2 + v] != d) << (SAINT_BIT - 1)); buckets[2 + v] = d;
if (cache[i].symbol < omp_block_end) { sa_sint_t s = cache[i].symbol, q = (cache[s].index = cache[i].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] >= T[q - 1]); }
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_4k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[2 * k];
sa_sint_t * RESTRICT distinct_names = &buckets[0 * k];
fast_sint_t i, j, omp_block_end = omp_block_start + omp_block_size;
for (i = omp_block_start, j = omp_block_end - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&cache[i + 2 * prefetch_distance]);
sa_sint_t s0 = cache[i + prefetch_distance + 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0 >> 1]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL); const sa_sint_t * Ds0 = &distinct_names[s0]; libsais_prefetchw(s0 >= 0 ? Ds0 : NULL);
sa_sint_t s1 = cache[i + prefetch_distance + 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1 >> 1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL); const sa_sint_t * Ds1 = &distinct_names[s1]; libsais_prefetchw(s1 >= 0 ? Ds1 : NULL);
sa_sint_t v0 = cache[i + 0].symbol;
if (v0 >= 0)
{
sa_sint_t p0 = cache[i + 0].index; d += (p0 >> (SUFFIX_GROUP_BIT - 1)); cache[i + 0].symbol = induction_bucket[v0 >> 1]++; cache[i + 0].index = (p0 - 1) | (v0 << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v0] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v0] = d;
if (cache[i + 0].symbol < omp_block_end) { sa_sint_t ni = cache[i + 0].symbol, np = cache[i + 0].index; if (np > 0) { cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] < T[np - 1]); np = 0; } cache[i + 0].index = np & SAINT_MAX; }
}
sa_sint_t v1 = cache[i + 1].symbol;
if (v1 >= 0)
{
sa_sint_t p1 = cache[i + 1].index; d += (p1 >> (SUFFIX_GROUP_BIT - 1)); cache[i + 1].symbol = induction_bucket[v1 >> 1]++; cache[i + 1].index = (p1 - 1) | (v1 << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v1] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v1] = d;
if (cache[i + 1].symbol < omp_block_end) { sa_sint_t ni = cache[i + 1].symbol, np = cache[i + 1].index; if (np > 0) { cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] < T[np - 1]); np = 0; } cache[i + 1].index = np & SAINT_MAX; }
}
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
sa_sint_t p = cache[i].index; d += (p >> (SUFFIX_GROUP_BIT - 1)); cache[i].symbol = induction_bucket[v >> 1]++; cache[i].index = (p - 1) | (v << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v] = d;
if (cache[i].symbol < omp_block_end) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; if (np > 0) { cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] < T[np - 1]); np = 0; } cache[i].index = np & SAINT_MAX; }
}
}
return d;
}
static void libsais_partial_sorting_scan_left_to_right_32s_1k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, omp_block_end = omp_block_start + omp_block_size;
for (i = omp_block_start, j = omp_block_end - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&cache[i + 2 * prefetch_distance]);
sa_sint_t s0 = cache[i + prefetch_distance + 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL);
sa_sint_t s1 = cache[i + prefetch_distance + 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL);
sa_sint_t v0 = cache[i + 0].symbol;
if (v0 >= 0)
{
cache[i + 0].symbol = induction_bucket[v0]++;
if (cache[i + 0].symbol < omp_block_end) { sa_sint_t ni = cache[i + 0].symbol, np = cache[i + 0].index; if (np > 0) { cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] < T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; np = 0; } cache[i + 0].index = np & SAINT_MAX; }
}
sa_sint_t v1 = cache[i + 1].symbol;
if (v1 >= 0)
{
cache[i + 1].symbol = induction_bucket[v1]++;
if (cache[i + 1].symbol < omp_block_end) { sa_sint_t ni = cache[i + 1].symbol, np = cache[i + 1].index; if (np > 0) { cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] < T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; np = 0; } cache[i + 1].index = np & SAINT_MAX; }
}
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
cache[i].symbol = induction_bucket[v]++;
if (cache[i].symbol < omp_block_end) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; if (np > 0) { cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] < T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; np = 0; } cache[i].index = np & SAINT_MAX; }
}
}
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_6k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_left_to_right_32s_6k(T, SA, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_left_to_right_32s_6k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
d = libsais_partial_sorting_scan_left_to_right_32s_6k_block_sort(T, buckets, d, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_4k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_left_to_right_32s_4k(T, SA, k, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_left_to_right_32s_4k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
d = libsais_partial_sorting_scan_left_to_right_32s_4k_block_sort(T, k, buckets, d, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
return d;
}
static void libsais_partial_sorting_scan_left_to_right_32s_1k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_partial_sorting_scan_left_to_right_32s_1k(T, SA, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_left_to_right_32s_1k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_partial_sorting_scan_left_to_right_32s_1k_block_sort(T, buckets, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
#endif
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_6k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t left_suffixes_count, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
SA[buckets[BUCKETS_INDEX4(T[n - 1], T[n - 2] >= T[n - 1])]++] = (n - 1) | SAINT_MIN;
buckets[2 + BUCKETS_INDEX4(T[n - 1], T[n - 2] >= T[n - 1])] = ++d;
if (threads == 1 || left_suffixes_count < 65536)
{
d = libsais_partial_sorting_scan_left_to_right_32s_6k(T, SA, buckets, d, 0, left_suffixes_count);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < left_suffixes_count; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end > left_suffixes_count) { block_end = left_suffixes_count; }
d = libsais_partial_sorting_scan_left_to_right_32s_6k_block_omp(T, SA, buckets, d, thread_state[0].state.cache, block_start, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
return d;
}
static sa_sint_t libsais_partial_sorting_scan_left_to_right_32s_4k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t * RESTRICT induction_bucket = &buckets[2 * k];
sa_sint_t * RESTRICT distinct_names = &buckets[0 * k];
SA[induction_bucket[T[n - 1]]++] = (n - 1) | ((sa_sint_t)(T[n - 2] < T[n - 1]) << (SAINT_BIT - 1)) | SUFFIX_GROUP_MARKER;
distinct_names[BUCKETS_INDEX2(T[n - 1], T[n - 2] < T[n - 1])] = ++d;
if (threads == 1 || n < 65536)
{
d = libsais_partial_sorting_scan_left_to_right_32s_4k(T, SA, k, buckets, d, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < n; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end > n) { block_end = n; }
d = libsais_partial_sorting_scan_left_to_right_32s_4k_block_omp(T, SA, k, buckets, d, thread_state[0].state.cache, block_start, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
return d;
}
static void libsais_partial_sorting_scan_left_to_right_32s_1k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
SA[buckets[T[n - 1]]++] = (n - 1) | ((sa_sint_t)(T[n - 2] < T[n - 1]) << (SAINT_BIT - 1));
if (threads == 1 || n < 65536)
{
libsais_partial_sorting_scan_left_to_right_32s_1k(T, SA, buckets, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < n; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end > n) { block_end = n; }
libsais_partial_sorting_scan_left_to_right_32s_1k_block_omp(T, SA, buckets, thread_state[0].state.cache, block_start, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_partial_sorting_shift_markers_8u_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, const sa_sint_t * RESTRICT buckets, sa_sint_t threads)
{
const fast_sint_t prefetch_distance = 32;
const sa_sint_t * RESTRICT temp_bucket = &buckets[4 * ALPHABET_SIZE];
fast_sint_t c;
#if defined(_OPENMP)
#pragma omp parallel for schedule(static, 1) num_threads(threads) if(threads > 1 && n >= 65536)
#else
UNUSED(threads); UNUSED(n);
#endif
for (c = BUCKETS_INDEX2(UCHAR_MAX, 0); c >= BUCKETS_INDEX2(1, 0); c -= BUCKETS_INDEX2(1, 0))
{
fast_sint_t i, j; sa_sint_t s = SAINT_MIN;
for (i = (fast_sint_t)temp_bucket[c] - 1, j = (fast_sint_t)buckets[c - BUCKETS_INDEX2(1, 0)] + 3; i >= j; i -= 4)
{
libsais_prefetchw(&SA[i - prefetch_distance]);
sa_sint_t p0 = SA[i - 0], q0 = (p0 & SAINT_MIN) ^ s; s = s ^ q0; SA[i - 0] = p0 ^ q0;
sa_sint_t p1 = SA[i - 1], q1 = (p1 & SAINT_MIN) ^ s; s = s ^ q1; SA[i - 1] = p1 ^ q1;
sa_sint_t p2 = SA[i - 2], q2 = (p2 & SAINT_MIN) ^ s; s = s ^ q2; SA[i - 2] = p2 ^ q2;
sa_sint_t p3 = SA[i - 3], q3 = (p3 & SAINT_MIN) ^ s; s = s ^ q3; SA[i - 3] = p3 ^ q3;
}
for (j -= 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i], q = (p & SAINT_MIN) ^ s; s = s ^ q; SA[i] = p ^ q;
}
}
}
static void libsais_partial_sorting_shift_markers_32s_6k_omp(sa_sint_t * RESTRICT SA, sa_sint_t k, const sa_sint_t * RESTRICT buckets, sa_sint_t threads)
{
const fast_sint_t prefetch_distance = 32;
const sa_sint_t * RESTRICT temp_bucket = &buckets[4 * k];
fast_sint_t c;
#if defined(_OPENMP)
#pragma omp parallel for schedule(static, 1) num_threads(threads) if(threads > 1 && k >= 65536)
#else
UNUSED(threads);
#endif
for (c = (fast_sint_t)k - 1; c >= 1; c -= 1)
{
fast_sint_t i, j; sa_sint_t s = SAINT_MIN;
for (i = (fast_sint_t)buckets[BUCKETS_INDEX4(c, 0)] - 1, j = (fast_sint_t)temp_bucket[BUCKETS_INDEX2(c - 1, 0)] + 3; i >= j; i -= 4)
{
libsais_prefetchw(&SA[i - prefetch_distance]);
sa_sint_t p0 = SA[i - 0], q0 = (p0 & SAINT_MIN) ^ s; s = s ^ q0; SA[i - 0] = p0 ^ q0;
sa_sint_t p1 = SA[i - 1], q1 = (p1 & SAINT_MIN) ^ s; s = s ^ q1; SA[i - 1] = p1 ^ q1;
sa_sint_t p2 = SA[i - 2], q2 = (p2 & SAINT_MIN) ^ s; s = s ^ q2; SA[i - 2] = p2 ^ q2;
sa_sint_t p3 = SA[i - 3], q3 = (p3 & SAINT_MIN) ^ s; s = s ^ q3; SA[i - 3] = p3 ^ q3;
}
for (j -= 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i], q = (p & SAINT_MIN) ^ s; s = s ^ q; SA[i] = p ^ q;
}
}
}
static void libsais_partial_sorting_shift_markers_32s_4k(sa_sint_t * RESTRICT SA, sa_sint_t n)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i; sa_sint_t s = SUFFIX_GROUP_MARKER;
for (i = (fast_sint_t)n - 1; i >= 3; i -= 4)
{
libsais_prefetchw(&SA[i - prefetch_distance]);
sa_sint_t p0 = SA[i - 0], q0 = ((p0 & SUFFIX_GROUP_MARKER) ^ s) & ((sa_sint_t)(p0 > 0) << ((SUFFIX_GROUP_BIT - 1))); s = s ^ q0; SA[i - 0] = p0 ^ q0;
sa_sint_t p1 = SA[i - 1], q1 = ((p1 & SUFFIX_GROUP_MARKER) ^ s) & ((sa_sint_t)(p1 > 0) << ((SUFFIX_GROUP_BIT - 1))); s = s ^ q1; SA[i - 1] = p1 ^ q1;
sa_sint_t p2 = SA[i - 2], q2 = ((p2 & SUFFIX_GROUP_MARKER) ^ s) & ((sa_sint_t)(p2 > 0) << ((SUFFIX_GROUP_BIT - 1))); s = s ^ q2; SA[i - 2] = p2 ^ q2;
sa_sint_t p3 = SA[i - 3], q3 = ((p3 & SUFFIX_GROUP_MARKER) ^ s) & ((sa_sint_t)(p3 > 0) << ((SUFFIX_GROUP_BIT - 1))); s = s ^ q3; SA[i - 3] = p3 ^ q3;
}
for (; i >= 0; i -= 1)
{
sa_sint_t p = SA[i], q = ((p & SUFFIX_GROUP_MARKER) ^ s) & ((sa_sint_t)(p > 0) << ((SUFFIX_GROUP_BIT - 1))); s = s ^ q; SA[i] = p ^ q;
}
}
static void libsais_partial_sorting_shift_buckets_32s_6k(sa_sint_t k, sa_sint_t * RESTRICT buckets)
{
sa_sint_t * RESTRICT temp_bucket = &buckets[4 * k];
fast_sint_t i;
for (i = BUCKETS_INDEX2(0, 0); i <= BUCKETS_INDEX2((fast_sint_t)k - 1, 0); i += BUCKETS_INDEX2(1, 0))
{
buckets[2 * i + BUCKETS_INDEX4(0, 0)] = temp_bucket[i + BUCKETS_INDEX2(0, 0)];
buckets[2 * i + BUCKETS_INDEX4(0, 1)] = temp_bucket[i + BUCKETS_INDEX2(0, 1)];
}
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetch(&SA[i - 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i - prefetch_distance - 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - prefetch_distance - 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i - prefetch_distance - 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - prefetch_distance - 1] & SAINT_MAX] - 2);
sa_sint_t p0 = SA[i - 0]; d += (p0 < 0); p0 &= SAINT_MAX; sa_sint_t v0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] > T[p0 - 1]);
SA[--induction_bucket[v0]] = (p0 - 1) | ((sa_sint_t)(distinct_names[v0] != d) << (SAINT_BIT - 1)); distinct_names[v0] = d;
sa_sint_t p1 = SA[i - 1]; d += (p1 < 0); p1 &= SAINT_MAX; sa_sint_t v1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] > T[p1 - 1]);
SA[--induction_bucket[v1]] = (p1 - 1) | ((sa_sint_t)(distinct_names[v1] != d) << (SAINT_BIT - 1)); distinct_names[v1] = d;
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] > T[p - 1]);
SA[--induction_bucket[v]] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
return d;
}
#if defined(_OPENMP)
static void libsais_partial_sorting_scan_right_to_left_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size, LIBSAIS_THREAD_STATE * RESTRICT state)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
memset(buckets, 0, 4 * ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0; sa_sint_t d = 1;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetch(&SA[i - 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i - prefetch_distance - 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - prefetch_distance - 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i - prefetch_distance - 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - prefetch_distance - 1] & SAINT_MAX] - 2);
sa_sint_t p0 = cache[count].index = SA[i - 0]; d += (p0 < 0); p0 &= SAINT_MAX; sa_sint_t v0 = cache[count++].symbol = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] > T[p0 - 1]); induction_bucket[v0]++; distinct_names[v0] = d;
sa_sint_t p1 = cache[count].index = SA[i - 1]; d += (p1 < 0); p1 &= SAINT_MAX; sa_sint_t v1 = cache[count++].symbol = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] > T[p1 - 1]); induction_bucket[v1]++; distinct_names[v1] = d;
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = cache[count].index = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = cache[count++].symbol = BUCKETS_INDEX2(T[p - 1], T[p - 2] > T[p - 1]); induction_bucket[v]++; distinct_names[v] = d;
}
state[0].state.position = (fast_sint_t)d - 1;
state[0].state.count = count;
}
static void libsais_partial_sorting_scan_right_to_left_8u_block_place(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t count, sa_sint_t d)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t i, j;
for (i = 0, j = count - 1; i < j; i += 2)
{
libsais_prefetch(&cache[i + prefetch_distance]);
sa_sint_t p0 = cache[i + 0].index; d += (p0 < 0); sa_sint_t v0 = cache[i + 0].symbol;
SA[--induction_bucket[v0]] = (p0 - 1) | ((sa_sint_t)(distinct_names[v0] != d) << (SAINT_BIT - 1)); distinct_names[v0] = d;
sa_sint_t p1 = cache[i + 1].index; d += (p1 < 0); sa_sint_t v1 = cache[i + 1].symbol;
SA[--induction_bucket[v1]] = (p1 - 1) | ((sa_sint_t)(distinct_names[v1] != d) << (SAINT_BIT - 1)); distinct_names[v1] = d;
}
for (j += 1; i < j; i += 1)
{
sa_sint_t p = cache[i].index; d += (p < 0); sa_sint_t v = cache[i].symbol;
SA[--induction_bucket[v]] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_right_to_left_8u(T, SA, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_right_to_left_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size, &thread_state[omp_thread_num]);
}
#pragma omp barrier
#pragma omp master
{
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
sa_sint_t * RESTRICT temp_induction_bucket = &thread_state[t].state.buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT temp_distinct_names = &thread_state[t].state.buckets[2 * ALPHABET_SIZE];
fast_sint_t c;
for (c = 0; c < 2 * ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_induction_bucket[c]; induction_bucket[c] = A - B; temp_induction_bucket[c] = A; }
for (d -= 1, c = 0; c < 2 * ALPHABET_SIZE; c += 1) { sa_sint_t A = distinct_names[c], B = temp_distinct_names[c], D = B + d; distinct_names[c] = B > 0 ? D : A; temp_distinct_names[c] = A; }
d += 1 + (sa_sint_t)thread_state[t].state.position; thread_state[t].state.position = (fast_sint_t)d - thread_state[t].state.position;
}
}
#pragma omp barrier
{
libsais_partial_sorting_scan_right_to_left_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count, (sa_sint_t)thread_state[omp_thread_num].state.position);
}
}
#endif
}
return d;
}
#endif
static void libsais_partial_sorting_scan_right_to_left_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
fast_sint_t scan_start = (fast_sint_t)left_suffixes_count + 1;
fast_sint_t scan_end = (fast_sint_t)n - (fast_sint_t)first_lms_suffix;
if (threads == 1 || (scan_end - scan_start) < 65536)
{
libsais_partial_sorting_scan_right_to_left_8u(T, SA, buckets, d, scan_start, scan_end - scan_start);
}
#if defined(_OPENMP)
else
{
sa_sint_t * RESTRICT induction_bucket = &buckets[0 * ALPHABET_SIZE];
sa_sint_t * RESTRICT distinct_names = &buckets[2 * ALPHABET_SIZE];
fast_sint_t block_start;
for (block_start = scan_end - 1; block_start >= scan_start; )
{
if (SA[block_start] == 0)
{
block_start--;
}
else
{
fast_sint_t block_max_end = block_start - ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end < scan_start) { block_max_end = scan_start - 1; }
fast_sint_t block_end = block_start - 1; while (block_end > block_max_end && SA[block_end] != 0) { block_end--; }
fast_sint_t block_size = block_start - block_end;
if (block_size < 32)
{
for (; block_start > block_end; block_start -= 1)
{
sa_sint_t p = SA[block_start]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] > T[p - 1]);
SA[--induction_bucket[v]] = (p - 1) | ((sa_sint_t)(distinct_names[v] != d) << (SAINT_BIT - 1)); distinct_names[v] = d;
}
}
else
{
d = libsais_partial_sorting_scan_right_to_left_8u_block_omp(T, SA, buckets, d, block_end + 1, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetch(&SA[i - 3 * prefetch_distance]);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i - 2 * prefetch_distance - 1] & SAINT_MAX] - 2);
sa_sint_t p0 = SA[i - prefetch_distance - 0] & SAINT_MAX; sa_sint_t v0 = BUCKETS_INDEX4(T[p0 - (p0 > 0)], 0); libsais_prefetchw(&buckets[v0]);
sa_sint_t p1 = SA[i - prefetch_distance - 1] & SAINT_MAX; sa_sint_t v1 = BUCKETS_INDEX4(T[p1 - (p1 > 0)], 0); libsais_prefetchw(&buckets[v1]);
sa_sint_t p2 = SA[i - 0]; d += (p2 < 0); p2 &= SAINT_MAX; sa_sint_t v2 = BUCKETS_INDEX4(T[p2 - 1], T[p2 - 2] > T[p2 - 1]);
SA[--buckets[v2]] = (p2 - 1) | ((sa_sint_t)(buckets[2 + v2] != d) << (SAINT_BIT - 1)); buckets[2 + v2] = d;
sa_sint_t p3 = SA[i - 1]; d += (p3 < 0); p3 &= SAINT_MAX; sa_sint_t v3 = BUCKETS_INDEX4(T[p3 - 1], T[p3 - 2] > T[p3 - 1]);
SA[--buckets[v3]] = (p3 - 1) | ((sa_sint_t)(buckets[2 + v3] != d) << (SAINT_BIT - 1)); buckets[2 + v3] = d;
}
for (j -= 2 * prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; d += (p < 0); p &= SAINT_MAX; sa_sint_t v = BUCKETS_INDEX4(T[p - 1], T[p - 2] > T[p - 1]);
SA[--buckets[v]] = (p - 1) | ((sa_sint_t)(buckets[2 + v] != d) << (SAINT_BIT - 1)); buckets[2 + v] = d;
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[3 * k];
sa_sint_t * RESTRICT distinct_names = &buckets[0 * k];
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 3 * prefetch_distance]);
sa_sint_t s0 = SA[i - 2 * prefetch_distance - 0]; const sa_sint_t * Ts0 = &T[s0 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - 2 * prefetch_distance - 1]; const sa_sint_t * Ts1 = &T[s1 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i - 1 * prefetch_distance - 0]; if (s2 > 0) { const fast_sint_t Ts2 = T[(s2 & ~SUFFIX_GROUP_MARKER) - 1]; libsais_prefetchw(&induction_bucket[Ts2]); libsais_prefetchw(&distinct_names[BUCKETS_INDEX2(Ts2, 0)]); }
sa_sint_t s3 = SA[i - 1 * prefetch_distance - 1]; if (s3 > 0) { const fast_sint_t Ts3 = T[(s3 & ~SUFFIX_GROUP_MARKER) - 1]; libsais_prefetchw(&induction_bucket[Ts3]); libsais_prefetchw(&distinct_names[BUCKETS_INDEX2(Ts3, 0)]); }
sa_sint_t p0 = SA[i - 0];
if (p0 > 0)
{
SA[i - 0] = 0; d += (p0 >> (SUFFIX_GROUP_BIT - 1)); p0 &= ~SUFFIX_GROUP_MARKER; sa_sint_t v0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] > T[p0 - 1]);
SA[--induction_bucket[T[p0 - 1]]] = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] > T[p0 - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v0] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v0] = d;
}
sa_sint_t p1 = SA[i - 1];
if (p1 > 0)
{
SA[i - 1] = 0; d += (p1 >> (SUFFIX_GROUP_BIT - 1)); p1 &= ~SUFFIX_GROUP_MARKER; sa_sint_t v1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] > T[p1 - 1]);
SA[--induction_bucket[T[p1 - 1]]] = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] > T[p1 - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v1] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v1] = d;
}
}
for (j -= 2 * prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i];
if (p > 0)
{
SA[i] = 0; d += (p >> (SUFFIX_GROUP_BIT - 1)); p &= ~SUFFIX_GROUP_MARKER; sa_sint_t v = BUCKETS_INDEX2(T[p - 1], T[p - 2] > T[p - 1]);
SA[--induction_bucket[T[p - 1]]] = (p - 1) | ((sa_sint_t)(T[p - 2] > T[p - 1]) << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v] = d;
}
}
return d;
}
static void libsais_partial_sorting_scan_right_to_left_32s_1k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 3 * prefetch_distance]);
sa_sint_t s0 = SA[i - 2 * prefetch_distance - 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - 2 * prefetch_distance - 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i - 1 * prefetch_distance - 0]; if (s2 > 0) { libsais_prefetchw(&induction_bucket[T[s2 - 1]]); libsais_prefetch(&T[s2] - 2); }
sa_sint_t s3 = SA[i - 1 * prefetch_distance - 1]; if (s3 > 0) { libsais_prefetchw(&induction_bucket[T[s3 - 1]]); libsais_prefetch(&T[s3] - 2); }
sa_sint_t p0 = SA[i - 0]; if (p0 > 0) { SA[i - 0] = 0; SA[--induction_bucket[T[p0 - 1]]] = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] > T[p0 - 1]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i - 1]; if (p1 > 0) { SA[i - 1] = 0; SA[--induction_bucket[T[p1 - 1]]] = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] > T[p1 - 1]) << (SAINT_BIT - 1)); }
}
for (j -= 2 * prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; if (p > 0) { SA[i] = 0; SA[--induction_bucket[T[p - 1]]] = (p - 1) | ((sa_sint_t)(T[p - 2] > T[p - 1]) << (SAINT_BIT - 1)); }
}
}
#if defined(_OPENMP)
static void libsais_partial_sorting_scan_right_to_left_32s_6k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 0] & SAINT_MAX] - 2);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 1);
libsais_prefetch(&T[SA[i + prefetch_distance + 1] & SAINT_MAX] - 2);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t p0 = cache[i + 0].index = SA[i + 0]; sa_sint_t symbol0 = 0; p0 &= SAINT_MAX; if (p0 != 0) { symbol0 = BUCKETS_INDEX4(T[p0 - 1], T[p0 - 2] > T[p0 - 1]); } cache[i + 0].symbol = symbol0;
sa_sint_t p1 = cache[i + 1].index = SA[i + 1]; sa_sint_t symbol1 = 0; p1 &= SAINT_MAX; if (p1 != 0) { symbol1 = BUCKETS_INDEX4(T[p1 - 1], T[p1 - 2] > T[p1 - 1]); } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = cache[i].index = SA[i]; sa_sint_t symbol = 0; p &= SAINT_MAX; if (p != 0) { symbol = BUCKETS_INDEX4(T[p - 1], T[p - 2] > T[p - 1]); } cache[i].symbol = symbol;
}
}
static void libsais_partial_sorting_scan_right_to_left_32s_4k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1 & ~SUFFIX_GROUP_MARKER] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; if (p0 > 0) { SA[i + 0] = 0; cache[i + 0].index = p0; p0 &= ~SUFFIX_GROUP_MARKER; symbol0 = BUCKETS_INDEX2(T[p0 - 1], T[p0 - 2] > T[p0 - 1]); } cache[i + 0].symbol = symbol0;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; if (p1 > 0) { SA[i + 1] = 0; cache[i + 1].index = p1; p1 &= ~SUFFIX_GROUP_MARKER; symbol1 = BUCKETS_INDEX2(T[p1 - 1], T[p1 - 2] > T[p1 - 1]); } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; if (p > 0) { SA[i] = 0; cache[i].index = p; p &= ~SUFFIX_GROUP_MARKER; symbol = BUCKETS_INDEX2(T[p - 1], T[p - 2] > T[p - 1]); } cache[i].symbol = symbol;
}
}
static void libsais_partial_sorting_scan_right_to_left_32s_1k_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; if (p0 > 0) { SA[i + 0] = 0; cache[i + 0].index = (p0 - 1) | ((sa_sint_t)(T[p0 - 2] > T[p0 - 1]) << (SAINT_BIT - 1)); symbol0 = T[p0 - 1]; } cache[i + 0].symbol = symbol0;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; if (p1 > 0) { SA[i + 1] = 0; cache[i + 1].index = (p1 - 1) | ((sa_sint_t)(T[p1 - 2] > T[p1 - 1]) << (SAINT_BIT - 1)); symbol1 = T[p1 - 1]; } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; if (p > 0) { SA[i] = 0; cache[i].index = (p - 1) | ((sa_sint_t)(T[p - 2] > T[p - 1]) << (SAINT_BIT - 1)); symbol = T[p - 1]; } cache[i].symbol = symbol;
}
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_6k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
libsais_prefetchw(&buckets[cache[i - prefetch_distance - 0].symbol]);
libsais_prefetchw(&buckets[cache[i - prefetch_distance - 1].symbol]);
sa_sint_t v0 = cache[i - 0].symbol, p0 = cache[i - 0].index; d += (p0 < 0); cache[i - 0].symbol = --buckets[v0]; cache[i - 0].index = (p0 - 1) | ((sa_sint_t)(buckets[2 + v0] != d) << (SAINT_BIT - 1)); buckets[2 + v0] = d;
if (cache[i - 0].symbol >= omp_block_start) { sa_sint_t s = cache[i - 0].symbol, q = (cache[s].index = cache[i - 0].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] > T[q - 1]); }
sa_sint_t v1 = cache[i - 1].symbol, p1 = cache[i - 1].index; d += (p1 < 0); cache[i - 1].symbol = --buckets[v1]; cache[i - 1].index = (p1 - 1) | ((sa_sint_t)(buckets[2 + v1] != d) << (SAINT_BIT - 1)); buckets[2 + v1] = d;
if (cache[i - 1].symbol >= omp_block_start) { sa_sint_t s = cache[i - 1].symbol, q = (cache[s].index = cache[i - 1].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] > T[q - 1]); }
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t v = cache[i].symbol, p = cache[i].index; d += (p < 0); cache[i].symbol = --buckets[v]; cache[i].index = (p - 1) | ((sa_sint_t)(buckets[2 + v] != d) << (SAINT_BIT - 1)); buckets[2 + v] = d;
if (cache[i].symbol >= omp_block_start) { sa_sint_t s = cache[i].symbol, q = (cache[s].index = cache[i].index) & SAINT_MAX; cache[s].symbol = BUCKETS_INDEX4(T[q - 1], T[q - 2] > T[q - 1]); }
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_4k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT induction_bucket = &buckets[3 * k];
sa_sint_t * RESTRICT distinct_names = &buckets[0 * k];
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
sa_sint_t s0 = cache[i - prefetch_distance - 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0 >> 1]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL); const sa_sint_t * Ds0 = &distinct_names[s0]; libsais_prefetchw(s0 >= 0 ? Ds0 : NULL);
sa_sint_t s1 = cache[i - prefetch_distance - 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1 >> 1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL); const sa_sint_t * Ds1 = &distinct_names[s1]; libsais_prefetchw(s1 >= 0 ? Ds1 : NULL);
sa_sint_t v0 = cache[i - 0].symbol;
if (v0 >= 0)
{
sa_sint_t p0 = cache[i - 0].index; d += (p0 >> (SUFFIX_GROUP_BIT - 1)); cache[i - 0].symbol = --induction_bucket[v0 >> 1]; cache[i - 0].index = (p0 - 1) | (v0 << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v0] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v0] = d;
if (cache[i - 0].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 0].symbol, np = cache[i - 0].index; if (np > 0) { cache[i - 0].index = 0; cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] > T[np - 1]); } }
}
sa_sint_t v1 = cache[i - 1].symbol;
if (v1 >= 0)
{
sa_sint_t p1 = cache[i - 1].index; d += (p1 >> (SUFFIX_GROUP_BIT - 1)); cache[i - 1].symbol = --induction_bucket[v1 >> 1]; cache[i - 1].index = (p1 - 1) | (v1 << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v1] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v1] = d;
if (cache[i - 1].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 1].symbol, np = cache[i - 1].index; if (np > 0) { cache[i - 1].index = 0; cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] > T[np - 1]); } }
}
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
sa_sint_t p = cache[i].index; d += (p >> (SUFFIX_GROUP_BIT - 1)); cache[i].symbol = --induction_bucket[v >> 1]; cache[i].index = (p - 1) | (v << (SAINT_BIT - 1)) | ((sa_sint_t)(distinct_names[v] != d) << (SUFFIX_GROUP_BIT - 1)); distinct_names[v] = d;
if (cache[i].symbol >= omp_block_start) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; if (np > 0) { cache[i].index = 0; cache[ni].index = np; np &= ~SUFFIX_GROUP_MARKER; cache[ni].symbol = BUCKETS_INDEX2(T[np - 1], T[np - 2] > T[np - 1]); } }
}
}
return d;
}
static void libsais_partial_sorting_scan_right_to_left_32s_1k_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
sa_sint_t s0 = cache[i - prefetch_distance - 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL);
sa_sint_t s1 = cache[i - prefetch_distance - 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL);
sa_sint_t v0 = cache[i - 0].symbol;
if (v0 >= 0)
{
cache[i - 0].symbol = --induction_bucket[v0];
if (cache[i - 0].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 0].symbol, np = cache[i - 0].index; if (np > 0) { cache[i - 0].index = 0; cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] > T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; } }
}
sa_sint_t v1 = cache[i - 1].symbol;
if (v1 >= 0)
{
cache[i - 1].symbol = --induction_bucket[v1];
if (cache[i - 1].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 1].symbol, np = cache[i - 1].index; if (np > 0) { cache[i - 1].index = 0; cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] > T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; }}
}
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
cache[i].symbol = --induction_bucket[v];
if (cache[i].symbol >= omp_block_start) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; if (np > 0) { cache[i].index = 0; cache[ni].index = (np - 1) | ((sa_sint_t)(T[np - 2] > T[np - 1]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np - 1]; } }
}
}
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_6k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_right_to_left_32s_6k(T, SA, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_right_to_left_32s_6k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
d = libsais_partial_sorting_scan_right_to_left_32s_6k_block_sort(T, buckets, d, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
return d;
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_4k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
d = libsais_partial_sorting_scan_right_to_left_32s_4k(T, SA, k, buckets, d, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_right_to_left_32s_4k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
d = libsais_partial_sorting_scan_right_to_left_32s_4k_block_sort(T, k, buckets, d, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
return d;
}
static void libsais_partial_sorting_scan_right_to_left_32s_1k_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_partial_sorting_scan_right_to_left_32s_1k(T, SA, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_partial_sorting_scan_right_to_left_32s_1k_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_partial_sorting_scan_right_to_left_32s_1k_block_sort(T, buckets, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
#endif
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_6k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
fast_sint_t scan_start = (fast_sint_t)left_suffixes_count + 1;
fast_sint_t scan_end = (fast_sint_t)n - (fast_sint_t)first_lms_suffix;
if (threads == 1 || (scan_end - scan_start) < 65536)
{
d = libsais_partial_sorting_scan_right_to_left_32s_6k(T, SA, buckets, d, scan_start, scan_end - scan_start);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = scan_end - 1; block_start >= scan_start; block_start = block_end)
{
block_end = block_start - (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end < scan_start) { block_end = scan_start - 1; }
d = libsais_partial_sorting_scan_right_to_left_32s_6k_block_omp(T, SA, buckets, d, thread_state[0].state.cache, block_end + 1, block_start - block_end, threads);
}
}
#else
UNUSED(thread_state);
#endif
return d;
}
static sa_sint_t libsais_partial_sorting_scan_right_to_left_32s_4k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t d, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || n < 65536)
{
d = libsais_partial_sorting_scan_right_to_left_32s_4k(T, SA, k, buckets, d, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = (fast_sint_t)n - 1; block_start >= 0; block_start = block_end)
{
block_end = block_start - (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end < 0) { block_end = -1; }
d = libsais_partial_sorting_scan_right_to_left_32s_4k_block_omp(T, SA, k, buckets, d, thread_state[0].state.cache, block_end + 1, block_start - block_end, threads);
}
}
#else
UNUSED(thread_state);
#endif
return d;
}
static void libsais_partial_sorting_scan_right_to_left_32s_1k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || n < 65536)
{
libsais_partial_sorting_scan_right_to_left_32s_1k(T, SA, buckets, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = (fast_sint_t)n - 1; block_start >= 0; block_start = block_end)
{
block_end = block_start - (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end < 0) { block_end = -1; }
libsais_partial_sorting_scan_right_to_left_32s_1k_block_omp(T, SA, buckets, thread_state[0].state.cache, block_end + 1, block_start - block_end, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static fast_sint_t libsais_partial_sorting_gather_lms_suffixes_32s_4k(sa_sint_t * RESTRICT SA, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, l;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 3, l = omp_block_start; i < j; i += 4)
{
libsais_prefetch(&SA[i + prefetch_distance]);
sa_sint_t s0 = SA[i + 0]; SA[l] = (s0 - SUFFIX_GROUP_MARKER) & (~SUFFIX_GROUP_MARKER); l += (s0 < 0);
sa_sint_t s1 = SA[i + 1]; SA[l] = (s1 - SUFFIX_GROUP_MARKER) & (~SUFFIX_GROUP_MARKER); l += (s1 < 0);
sa_sint_t s2 = SA[i + 2]; SA[l] = (s2 - SUFFIX_GROUP_MARKER) & (~SUFFIX_GROUP_MARKER); l += (s2 < 0);
sa_sint_t s3 = SA[i + 3]; SA[l] = (s3 - SUFFIX_GROUP_MARKER) & (~SUFFIX_GROUP_MARKER); l += (s3 < 0);
}
for (j += 3; i < j; i += 1)
{
sa_sint_t s = SA[i]; SA[l] = (s - SUFFIX_GROUP_MARKER) & (~SUFFIX_GROUP_MARKER); l += (s < 0);
}
return l;
}
static fast_sint_t libsais_partial_sorting_gather_lms_suffixes_32s_1k(sa_sint_t * RESTRICT SA, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, l;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 3, l = omp_block_start; i < j; i += 4)
{
libsais_prefetch(&SA[i + prefetch_distance]);
sa_sint_t s0 = SA[i + 0]; SA[l] = s0 & SAINT_MAX; l += (s0 < 0);
sa_sint_t s1 = SA[i + 1]; SA[l] = s1 & SAINT_MAX; l += (s1 < 0);
sa_sint_t s2 = SA[i + 2]; SA[l] = s2 & SAINT_MAX; l += (s2 < 0);
sa_sint_t s3 = SA[i + 3]; SA[l] = s3 & SAINT_MAX; l += (s3 < 0);
}
for (j += 3; i < j; i += 1)
{
sa_sint_t s = SA[i]; SA[l] = s & SAINT_MAX; l += (s < 0);
}
return l;
}
static void libsais_partial_sorting_gather_lms_suffixes_32s_4k_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
libsais_partial_sorting_gather_lms_suffixes_32s_4k(SA, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.position = omp_block_start;
thread_state[omp_thread_num].state.count = libsais_partial_sorting_gather_lms_suffixes_32s_4k(SA, omp_block_start, omp_block_size) - omp_block_start;
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t, position = 0;
for (t = 0; t < omp_num_threads; ++t)
{
if (t > 0 && thread_state[t].state.count > 0)
{
memmove(&SA[position], &SA[thread_state[t].state.position], (size_t)thread_state[t].state.count * sizeof(sa_sint_t));
}
position += thread_state[t].state.count;
}
}
}
#endif
}
}
static void libsais_partial_sorting_gather_lms_suffixes_32s_1k_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
libsais_partial_sorting_gather_lms_suffixes_32s_1k(SA, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.position = omp_block_start;
thread_state[omp_thread_num].state.count = libsais_partial_sorting_gather_lms_suffixes_32s_1k(SA, omp_block_start, omp_block_size) - omp_block_start;
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t, position = 0;
for (t = 0; t < omp_num_threads; ++t)
{
if (t > 0 && thread_state[t].state.count > 0)
{
memmove(&SA[position], &SA[thread_state[t].state.position], (size_t)thread_state[t].state.count * sizeof(sa_sint_t));
}
position += thread_state[t].state.count;
}
}
}
#endif
}
}
static void libsais_induce_partial_order_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
memset(&buckets[2 * ALPHABET_SIZE], 0, 2 * ALPHABET_SIZE * sizeof(sa_sint_t));
sa_sint_t d = libsais_partial_sorting_scan_left_to_right_8u_omp(T, SA, n, buckets, left_suffixes_count, 0, threads, thread_state);
libsais_partial_sorting_shift_markers_8u_omp(SA, n, buckets, threads);
libsais_partial_sorting_scan_right_to_left_8u_omp(T, SA, n, buckets, first_lms_suffix, left_suffixes_count, d, threads, thread_state);
}
static void libsais_induce_partial_order_32s_6k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t first_lms_suffix, sa_sint_t left_suffixes_count, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t d = libsais_partial_sorting_scan_left_to_right_32s_6k_omp(T, SA, n, buckets, left_suffixes_count, 0, threads, thread_state);
libsais_partial_sorting_shift_markers_32s_6k_omp(SA, k, buckets, threads);
libsais_partial_sorting_shift_buckets_32s_6k(k, buckets);
libsais_partial_sorting_scan_right_to_left_32s_6k_omp(T, SA, n, buckets, first_lms_suffix, left_suffixes_count, d, threads, thread_state);
}
static void libsais_induce_partial_order_32s_4k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
memset(buckets, 0, 2 * (size_t)k * sizeof(sa_sint_t));
sa_sint_t d = libsais_partial_sorting_scan_left_to_right_32s_4k_omp(T, SA, n, k, buckets, 0, threads, thread_state);
libsais_partial_sorting_shift_markers_32s_4k(SA, n);
libsais_partial_sorting_scan_right_to_left_32s_4k_omp(T, SA, n, k, buckets, d, threads, thread_state);
libsais_partial_sorting_gather_lms_suffixes_32s_4k_omp(SA, n, threads, thread_state);
}
static void libsais_induce_partial_order_32s_2k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_partial_sorting_scan_left_to_right_32s_1k_omp(T, SA, n, &buckets[1 * k], threads, thread_state);
libsais_partial_sorting_scan_right_to_left_32s_1k_omp(T, SA, n, &buckets[0 * k], threads, thread_state);
libsais_partial_sorting_gather_lms_suffixes_32s_1k_omp(SA, n, threads, thread_state);
}
static void libsais_induce_partial_order_32s_1k_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_start_32s_1k(k, buckets);
libsais_partial_sorting_scan_left_to_right_32s_1k_omp(T, SA, n, buckets, threads, thread_state);
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_end_32s_1k(k, buckets);
libsais_partial_sorting_scan_right_to_left_32s_1k_omp(T, SA, n, buckets, threads, thread_state);
libsais_partial_sorting_gather_lms_suffixes_32s_1k_omp(SA, n, threads, thread_state);
}
static sa_sint_t libsais_renumber_lms_suffixes_8u(sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t name, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 0] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 1] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 2] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 3] & SAINT_MAX) >> 1]);
sa_sint_t p0 = SA[i + 0]; SAm[(p0 & SAINT_MAX) >> 1] = name | SAINT_MIN; name += p0 < 0;
sa_sint_t p1 = SA[i + 1]; SAm[(p1 & SAINT_MAX) >> 1] = name | SAINT_MIN; name += p1 < 0;
sa_sint_t p2 = SA[i + 2]; SAm[(p2 & SAINT_MAX) >> 1] = name | SAINT_MIN; name += p2 < 0;
sa_sint_t p3 = SA[i + 3]; SAm[(p3 & SAINT_MAX) >> 1] = name | SAINT_MIN; name += p3 < 0;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
sa_sint_t p = SA[i]; SAm[(p & SAINT_MAX) >> 1] = name | SAINT_MIN; name += p < 0;
}
return name;
}
static fast_sint_t libsais_gather_marked_suffixes_8u(sa_sint_t * RESTRICT SA, sa_sint_t m, fast_sint_t l, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
l -= 1;
fast_sint_t i, j;
for (i = (fast_sint_t)m + omp_block_start + omp_block_size - 1, j = (fast_sint_t)m + omp_block_start + 3; i >= j; i -= 4)
{
libsais_prefetch(&SA[i - prefetch_distance]);
sa_sint_t s0 = SA[i - 0]; SA[l] = s0 & SAINT_MAX; l -= s0 < 0;
sa_sint_t s1 = SA[i - 1]; SA[l] = s1 & SAINT_MAX; l -= s1 < 0;
sa_sint_t s2 = SA[i - 2]; SA[l] = s2 & SAINT_MAX; l -= s2 < 0;
sa_sint_t s3 = SA[i - 3]; SA[l] = s3 & SAINT_MAX; l -= s3 < 0;
}
for (j -= 3; i >= j; i -= 1)
{
sa_sint_t s = SA[i]; SA[l] = s & SAINT_MAX; l -= s < 0;
}
l += 1;
return l;
}
static sa_sint_t libsais_renumber_lms_suffixes_8u_omp(sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t name = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && m >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (m / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : m - omp_block_start;
if (omp_num_threads == 1)
{
name = libsais_renumber_lms_suffixes_8u(SA, m, 0, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_count_negative_marked_suffixes(SA, omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, count = 0; for (t = 0; t < omp_thread_num; ++t) { count += thread_state[t].state.count; }
if (omp_thread_num == omp_num_threads - 1)
{
name = (sa_sint_t)(count + thread_state[omp_thread_num].state.count);
}
libsais_renumber_lms_suffixes_8u(SA, m, (sa_sint_t)count, omp_block_start, omp_block_size);
}
}
#endif
}
return name;
}
static void libsais_gather_marked_lms_suffixes_8u_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t fs, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 131072)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (((fast_sint_t)n >> 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : ((fast_sint_t)n >> 1) - omp_block_start;
if (omp_num_threads == 1)
{
libsais_gather_marked_suffixes_8u(SA, m, (fast_sint_t)n + (fast_sint_t)fs, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
if (omp_thread_num < omp_num_threads - 1)
{
thread_state[omp_thread_num].state.position = libsais_gather_marked_suffixes_8u(SA, m, (fast_sint_t)m + omp_block_start + omp_block_size, omp_block_start, omp_block_size);
thread_state[omp_thread_num].state.count = (fast_sint_t)m + omp_block_start + omp_block_size - thread_state[omp_thread_num].state.position;
}
else
{
thread_state[omp_thread_num].state.position = libsais_gather_marked_suffixes_8u(SA, m, (fast_sint_t)n + (fast_sint_t)fs, omp_block_start, omp_block_size);
thread_state[omp_thread_num].state.count = (fast_sint_t)n + (fast_sint_t)fs - thread_state[omp_thread_num].state.position;
}
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t, position = (fast_sint_t)n + (fast_sint_t)fs;
for (t = omp_num_threads - 1; t >= 0; --t)
{
position -= thread_state[t].state.count;
if (t != omp_num_threads - 1 && thread_state[t].state.count > 0)
{
memmove(&SA[position], &SA[thread_state[t].state.position], (size_t)thread_state[t].state.count * sizeof(sa_sint_t));
}
}
}
}
#endif
}
}
static sa_sint_t libsais_renumber_and_gather_lms_suffixes_8u_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t fs, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
memset(&SA[m], 0, ((size_t)n >> 1) * sizeof(sa_sint_t));
sa_sint_t name = libsais_renumber_lms_suffixes_8u_omp(SA, m, threads, thread_state);
if (name < m)
{
libsais_gather_marked_lms_suffixes_8u_omp(SA, n, m, fs, threads, thread_state);
}
else
{
fast_sint_t i; for (i = 0; i < m; i += 1) { SA[i] &= SAINT_MAX; }
}
return name;
}
static sa_sint_t libsais_renumber_distinct_lms_suffixes_32s_4k(sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t name, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
fast_sint_t i, j; sa_sint_t p0, p1, p2, p3 = 0;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 0] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 1] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 2] & SAINT_MAX) >> 1]);
libsais_prefetchw(&SAm[(SA[i + prefetch_distance + 3] & SAINT_MAX) >> 1]);
p0 = SA[i + 0]; SAm[(SA[i + 0] = p0 & SAINT_MAX) >> 1] = name | (p0 & p3 & SAINT_MIN); name += p0 < 0;
p1 = SA[i + 1]; SAm[(SA[i + 1] = p1 & SAINT_MAX) >> 1] = name | (p1 & p0 & SAINT_MIN); name += p1 < 0;
p2 = SA[i + 2]; SAm[(SA[i + 2] = p2 & SAINT_MAX) >> 1] = name | (p2 & p1 & SAINT_MIN); name += p2 < 0;
p3 = SA[i + 3]; SAm[(SA[i + 3] = p3 & SAINT_MAX) >> 1] = name | (p3 & p2 & SAINT_MIN); name += p3 < 0;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
p2 = p3; p3 = SA[i]; SAm[(SA[i] = p3 & SAINT_MAX) >> 1] = name | (p3 & p2 & SAINT_MIN); name += p3 < 0;
}
return name;
}
static void libsais_mark_distinct_lms_suffixes_32s(sa_sint_t * RESTRICT SA, sa_sint_t m, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j; sa_sint_t p0, p1, p2, p3 = 0;
for (i = (fast_sint_t)m + omp_block_start, j = (fast_sint_t)m + omp_block_start + omp_block_size - 3; i < j; i += 4)
{
libsais_prefetchw(&SA[i + prefetch_distance]);
p0 = SA[i + 0]; SA[i + 0] = p0 & (p3 | SAINT_MAX); p0 = (p0 == 0) ? p3 : p0;
p1 = SA[i + 1]; SA[i + 1] = p1 & (p0 | SAINT_MAX); p1 = (p1 == 0) ? p0 : p1;
p2 = SA[i + 2]; SA[i + 2] = p2 & (p1 | SAINT_MAX); p2 = (p2 == 0) ? p1 : p2;
p3 = SA[i + 3]; SA[i + 3] = p3 & (p2 | SAINT_MAX); p3 = (p3 == 0) ? p2 : p3;
}
for (j += 3; i < j; i += 1)
{
p2 = p3; p3 = SA[i]; SA[i] = p3 & (p2 | SAINT_MAX); p3 = (p3 == 0) ? p2 : p3;
}
}
static void libsais_clamp_lms_suffixes_length_32s(sa_sint_t * RESTRICT SA, sa_sint_t m, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 3; i < j; i += 4)
{
libsais_prefetchw(&SAm[i + prefetch_distance]);
SAm[i + 0] = (SAm[i + 0] < 0 ? SAm[i + 0] : 0) & SAINT_MAX;
SAm[i + 1] = (SAm[i + 1] < 0 ? SAm[i + 1] : 0) & SAINT_MAX;
SAm[i + 2] = (SAm[i + 2] < 0 ? SAm[i + 2] : 0) & SAINT_MAX;
SAm[i + 3] = (SAm[i + 3] < 0 ? SAm[i + 3] : 0) & SAINT_MAX;
}
for (j += 3; i < j; i += 1)
{
SAm[i] = (SAm[i] < 0 ? SAm[i] : 0) & SAINT_MAX;
}
}
static sa_sint_t libsais_renumber_distinct_lms_suffixes_32s_4k_omp(sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t name = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && m >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (m / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : m - omp_block_start;
if (omp_num_threads == 1)
{
name = libsais_renumber_distinct_lms_suffixes_32s_4k(SA, m, 1, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_count_negative_marked_suffixes(SA, omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, count = 1; for (t = 0; t < omp_thread_num; ++t) { count += thread_state[t].state.count; }
if (omp_thread_num == omp_num_threads - 1)
{
name = (sa_sint_t)(count + thread_state[omp_thread_num].state.count);
}
libsais_renumber_distinct_lms_suffixes_32s_4k(SA, m, (sa_sint_t)count, omp_block_start, omp_block_size);
}
}
#endif
}
return name - 1;
}
static void libsais_mark_distinct_lms_suffixes_32s_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 131072)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = (((fast_sint_t)n >> 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : ((fast_sint_t)n >> 1) - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = (fast_sint_t)n >> 1;
#endif
libsais_mark_distinct_lms_suffixes_32s(SA, m, omp_block_start, omp_block_size);
}
}
static void libsais_clamp_lms_suffixes_length_32s_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 131072)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = (((fast_sint_t)n >> 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : ((fast_sint_t)n >> 1) - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = (fast_sint_t)n >> 1;
#endif
libsais_clamp_lms_suffixes_length_32s(SA, m, omp_block_start, omp_block_size);
}
}
static sa_sint_t libsais_renumber_and_mark_distinct_lms_suffixes_32s_4k_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
memset(&SA[m], 0, ((size_t)n >> 1) * sizeof(sa_sint_t));
sa_sint_t name = libsais_renumber_distinct_lms_suffixes_32s_4k_omp(SA, m, threads, thread_state);
if (name < m)
{
libsais_mark_distinct_lms_suffixes_32s_omp(SA, n, m, threads);
}
return name;
}
static sa_sint_t libsais_renumber_and_mark_distinct_lms_suffixes_32s_1k_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
{
libsais_gather_lms_suffixes_32s(T, SA, n);
memset(&SA[m], 0, ((size_t)n - (size_t)m - (size_t)m) * sizeof(sa_sint_t));
fast_sint_t i, j;
for (i = (fast_sint_t)n - (fast_sint_t)m, j = (fast_sint_t)n - 1 - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 0]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 1]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 2]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 3]) >> 1]);
SAm[((sa_uint_t)SA[i + 0]) >> 1] = SA[i + 1] - SA[i + 0] + 1 + SAINT_MIN;
SAm[((sa_uint_t)SA[i + 1]) >> 1] = SA[i + 2] - SA[i + 1] + 1 + SAINT_MIN;
SAm[((sa_uint_t)SA[i + 2]) >> 1] = SA[i + 3] - SA[i + 2] + 1 + SAINT_MIN;
SAm[((sa_uint_t)SA[i + 3]) >> 1] = SA[i + 4] - SA[i + 3] + 1 + SAINT_MIN;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
SAm[((sa_uint_t)SA[i]) >> 1] = SA[i + 1] - SA[i] + 1 + SAINT_MIN;
}
SAm[((sa_uint_t)SA[n - 1]) >> 1] = 1 + SAINT_MIN;
}
{
libsais_clamp_lms_suffixes_length_32s_omp(SA, n, m, threads);
}
sa_sint_t name = 1;
{
fast_sint_t i, j, p = SA[0], plen = SAm[p >> 1]; sa_sint_t pdiff = SAINT_MIN;
for (i = 1, j = m - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 0]) >> 1]); libsais_prefetch(&T[((sa_uint_t)SA[i + prefetch_distance + 0])]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + prefetch_distance + 1]) >> 1]); libsais_prefetch(&T[((sa_uint_t)SA[i + prefetch_distance + 1])]);
fast_sint_t q = SA[i + 0], qlen = SAm[q >> 1]; sa_sint_t qdiff = SAINT_MIN;
if (plen == qlen) { fast_sint_t l = 0; do { if (T[p + l] != T[q + l]) { break; } } while (++l < qlen); qdiff = (l - qlen) & SAINT_MIN; }
SAm[p >> 1] = name | (pdiff & qdiff); name += (qdiff < 0);
p = SA[i + 1]; plen = SAm[p >> 1]; pdiff = SAINT_MIN;
if (qlen == plen) { fast_sint_t l = 0; do { if (T[q + l] != T[p + l]) { break; } } while (++l < plen); pdiff = (l - plen) & SAINT_MIN; }
SAm[q >> 1] = name | (qdiff & pdiff); name += (pdiff < 0);
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
fast_sint_t q = SA[i], qlen = SAm[q >> 1]; sa_sint_t qdiff = SAINT_MIN;
if (plen == qlen) { fast_sint_t l = 0; do { if (T[p + l] != T[q + l]) { break; } } while (++l < plen); qdiff = (l - plen) & SAINT_MIN; }
SAm[p >> 1] = name | (pdiff & qdiff); name += (qdiff < 0);
p = q; plen = qlen; pdiff = qdiff;
}
SAm[p >> 1] = name | pdiff; name++;
}
if (name <= m)
{
libsais_mark_distinct_lms_suffixes_32s_omp(SA, n, m, threads);
}
return name - 1;
}
static void libsais_reconstruct_lms_suffixes(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
const sa_sint_t * RESTRICT SAnm = &SA[n - m];
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&SAnm[SA[i + prefetch_distance + 0]]);
libsais_prefetch(&SAnm[SA[i + prefetch_distance + 1]]);
libsais_prefetch(&SAnm[SA[i + prefetch_distance + 2]]);
libsais_prefetch(&SAnm[SA[i + prefetch_distance + 3]]);
SA[i + 0] = SAnm[SA[i + 0]];
SA[i + 1] = SAnm[SA[i + 1]];
SA[i + 2] = SAnm[SA[i + 2]];
SA[i + 3] = SAnm[SA[i + 3]];
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
SA[i] = SAnm[SA[i]];
}
}
static void libsais_reconstruct_lms_suffixes_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && m >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = (m / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : m - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = m;
#endif
libsais_reconstruct_lms_suffixes(SA, n, m, omp_block_start, omp_block_size);
}
}
static void libsais_place_lms_suffixes_interval_8u(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
const sa_sint_t * RESTRICT bucket_end = &buckets[7 * ALPHABET_SIZE];
fast_sint_t c, j = n;
for (c = UCHAR_MAX - 1; c >= 0; --c)
{
fast_sint_t l = (fast_sint_t)buckets[BUCKETS_INDEX2(c, 1) + BUCKETS_INDEX2(1, 0)] - (fast_sint_t)buckets[BUCKETS_INDEX2(c, 1)];
if (l > 0)
{
fast_sint_t i = bucket_end[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_interval_32s_4k(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
const sa_sint_t * RESTRICT bucket_end = &buckets[3 * k];
fast_sint_t c, j = n;
for (c = (fast_sint_t)k - 2; c >= 0; --c)
{
fast_sint_t l = (fast_sint_t)buckets[BUCKETS_INDEX2(c, 1) + BUCKETS_INDEX2(1, 0)] - (fast_sint_t)buckets[BUCKETS_INDEX2(c, 1)];
if (l > 0)
{
fast_sint_t i = bucket_end[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_interval_32s_2k(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
fast_sint_t c, j = n;
for (c = BUCKETS_INDEX2((fast_sint_t)k - 2, 0); c >= BUCKETS_INDEX2(0, 0); c -= BUCKETS_INDEX2(1, 0))
{
fast_sint_t l = (fast_sint_t)buckets[c + BUCKETS_INDEX2(1, 1)] - (fast_sint_t)buckets[c + BUCKETS_INDEX2(0, 1)];
if (l > 0)
{
fast_sint_t i = buckets[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_interval_32s_1k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t k, sa_sint_t m, sa_sint_t * RESTRICT buckets)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t c = k - 1; fast_sint_t i, l = buckets[c];
for (i = (fast_sint_t)m - 1; i >= prefetch_distance + 3; i -= 4)
{
libsais_prefetch(&SA[i - 2 * prefetch_distance]);
libsais_prefetch(&T[SA[i - prefetch_distance - 0]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 1]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 2]]);
libsais_prefetch(&T[SA[i - prefetch_distance - 3]]);
sa_sint_t p0 = SA[i - 0]; if (T[p0] != c) { c = T[p0]; memset(&SA[buckets[c]], 0, (size_t)(l - buckets[c]) * sizeof(sa_sint_t)); l = buckets[c]; } SA[--l] = p0;
sa_sint_t p1 = SA[i - 1]; if (T[p1] != c) { c = T[p1]; memset(&SA[buckets[c]], 0, (size_t)(l - buckets[c]) * sizeof(sa_sint_t)); l = buckets[c]; } SA[--l] = p1;
sa_sint_t p2 = SA[i - 2]; if (T[p2] != c) { c = T[p2]; memset(&SA[buckets[c]], 0, (size_t)(l - buckets[c]) * sizeof(sa_sint_t)); l = buckets[c]; } SA[--l] = p2;
sa_sint_t p3 = SA[i - 3]; if (T[p3] != c) { c = T[p3]; memset(&SA[buckets[c]], 0, (size_t)(l - buckets[c]) * sizeof(sa_sint_t)); l = buckets[c]; } SA[--l] = p3;
}
for (; i >= 0; i -= 1)
{
sa_sint_t p = SA[i]; if (T[p] != c) { c = T[p]; memset(&SA[buckets[c]], 0, (size_t)(l - buckets[c]) * sizeof(sa_sint_t)); l = buckets[c]; } SA[--l] = p;
}
memset(&SA[0], 0, (size_t)l * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_histogram_32s_6k(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
const sa_sint_t * RESTRICT bucket_end = &buckets[5 * k];
fast_sint_t c, j = n;
for (c = (fast_sint_t)k - 2; c >= 0; --c)
{
fast_sint_t l = (fast_sint_t)buckets[BUCKETS_INDEX4(c, 1)];
if (l > 0)
{
fast_sint_t i = bucket_end[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_histogram_32s_4k(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
const sa_sint_t * RESTRICT bucket_end = &buckets[3 * k];
fast_sint_t c, j = n;
for (c = (fast_sint_t)k - 2; c >= 0; --c)
{
fast_sint_t l = (fast_sint_t)buckets[BUCKETS_INDEX2(c, 1)];
if (l > 0)
{
fast_sint_t i = bucket_end[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_place_lms_suffixes_histogram_32s_2k(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, const sa_sint_t * RESTRICT buckets)
{
fast_sint_t c, j = n;
for (c = BUCKETS_INDEX2((fast_sint_t)k - 2, 0); c >= BUCKETS_INDEX2(0, 0); c -= BUCKETS_INDEX2(1, 0))
{
fast_sint_t l = (fast_sint_t)buckets[c + BUCKETS_INDEX2(0, 1)];
if (l > 0)
{
fast_sint_t i = buckets[c];
if (j - i > 0)
{
memset(&SA[i], 0, (size_t)(j - i) * sizeof(sa_sint_t));
}
memmove(&SA[j = (i - l)], &SA[m -= (sa_sint_t)l], (size_t)l * sizeof(sa_sint_t));
}
}
memset(&SA[0], 0, (size_t)j * sizeof(sa_sint_t));
}
static void libsais_final_bwt_scan_left_to_right_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; SA[i + 0] = T[p0] | SAINT_MIN; SA[induction_bucket[T[p0]]++] = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; SA[i + 1] = T[p1] | SAINT_MIN; SA[induction_bucket[T[p1]]++] = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); }
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; SA[i] = T[p] | SAINT_MIN; SA[induction_bucket[T[p]]++] = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
}
static void libsais_final_sorting_scan_left_to_right_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 ^ SAINT_MIN; if (p0 > 0) { p0--; SA[induction_bucket[T[p0]]++] = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 ^ SAINT_MIN; if (p1 > 0) { p1--; SA[induction_bucket[T[p1]]++] = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); }
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p ^ SAINT_MIN; if (p > 0) { p--; SA[induction_bucket[T[p]]++] = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
}
static void libsais_final_sorting_scan_left_to_right_32s(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 2 * prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 3 * prefetch_distance]);
sa_sint_t s0 = SA[i + 2 * prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + 2 * prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i + 1 * prefetch_distance + 0]; if (s2 > 0) { libsais_prefetchw(&induction_bucket[T[s2 - 1]]); libsais_prefetch(&T[s2] - 2); }
sa_sint_t s3 = SA[i + 1 * prefetch_distance + 1]; if (s3 > 0) { libsais_prefetchw(&induction_bucket[T[s3 - 1]]); libsais_prefetch(&T[s3] - 2); }
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 ^ SAINT_MIN; if (p0 > 0) { p0--; SA[induction_bucket[T[p0]]++] = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 ^ SAINT_MIN; if (p1 > 0) { p1--; SA[induction_bucket[T[p1]]++] = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); }
}
for (j += 2 * prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p ^ SAINT_MIN; if (p > 0) { p--; SA[induction_bucket[T[p]]++] = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
}
#if defined(_OPENMP)
static fast_sint_t libsais_final_bwt_scan_left_to_right_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; SA[i + 0] = T[p0] | SAINT_MIN; buckets[cache[count].symbol = T[p0]]++; cache[count++].index = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; SA[i + 1] = T[p1] | SAINT_MIN; buckets[cache[count].symbol = T[p1]]++; cache[count++].index = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); }
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; SA[i] = T[p] | SAINT_MIN; buckets[cache[count].symbol = T[p]]++; cache[count++].index = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
return count;
}
static fast_sint_t libsais_final_sorting_scan_left_to_right_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i + 0]; SA[i + 0] = p0 ^ SAINT_MIN; if (p0 > 0) { p0--; buckets[cache[count].symbol = T[p0]]++; cache[count++].index = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i + 1]; SA[i + 1] = p1 ^ SAINT_MIN; if (p1 > 0) { p1--; buckets[cache[count].symbol = T[p1]]++; cache[count++].index = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); }
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t p = SA[i]; SA[i] = p ^ SAINT_MIN; if (p > 0) { p--; buckets[cache[count].symbol = T[p]]++; cache[count++].index = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
return count;
}
static void libsais_final_order_scan_left_to_right_8u_block_place(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t count)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = 0, j = count - 3; i < j; i += 4)
{
libsais_prefetch(&cache[i + prefetch_distance]);
SA[buckets[cache[i + 0].symbol]++] = cache[i + 0].index;
SA[buckets[cache[i + 1].symbol]++] = cache[i + 1].index;
SA[buckets[cache[i + 2].symbol]++] = cache[i + 2].index;
SA[buckets[cache[i + 3].symbol]++] = cache[i + 3].index;
}
for (j += 3; i < j; i += 1)
{
SA[buckets[cache[i].symbol]++] = cache[i].index;
}
}
static void libsais_final_sorting_scan_left_to_right_32s_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; SA[i + 0] = p0 ^ SAINT_MIN; if (p0 > 0) { p0--; cache[i + 0].index = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] < T[p0]) << (SAINT_BIT - 1)); symbol0 = T[p0]; } cache[i + 0].symbol = symbol0;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; SA[i + 1] = p1 ^ SAINT_MIN; if (p1 > 0) { p1--; cache[i + 1].index = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] < T[p1]) << (SAINT_BIT - 1)); symbol1 = T[p1]; } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; SA[i] = p ^ SAINT_MIN; if (p > 0) { p--; cache[i].index = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); symbol = T[p]; } cache[i].symbol = symbol;
}
}
static void libsais_final_sorting_scan_left_to_right_32s_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j, omp_block_end = omp_block_start + omp_block_size;
for (i = omp_block_start, j = omp_block_end - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&cache[i + 2 * prefetch_distance]);
sa_sint_t s0 = cache[i + prefetch_distance + 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL);
sa_sint_t s1 = cache[i + prefetch_distance + 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL);
sa_sint_t v0 = cache[i + 0].symbol;
if (v0 >= 0)
{
cache[i + 0].symbol = induction_bucket[v0]++;
if (cache[i + 0].symbol < omp_block_end) { sa_sint_t ni = cache[i + 0].symbol, np = cache[i + 0].index; cache[i + 0].index = np ^ SAINT_MIN; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] < T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
sa_sint_t v1 = cache[i + 1].symbol;
if (v1 >= 0)
{
cache[i + 1].symbol = induction_bucket[v1]++;
if (cache[i + 1].symbol < omp_block_end) { sa_sint_t ni = cache[i + 1].symbol, np = cache[i + 1].index; cache[i + 1].index = np ^ SAINT_MIN; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] < T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
cache[i].symbol = induction_bucket[v]++;
if (cache[i].symbol < omp_block_end) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; cache[i].index = np ^ SAINT_MIN; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] < T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
}
}
static void libsais_final_bwt_scan_left_to_right_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_bwt_scan_left_to_right_8u(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_final_bwt_scan_left_to_right_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t;
for (t = 0; t < omp_num_threads; ++t)
{
sa_sint_t * RESTRICT temp_bucket = thread_state[t].state.buckets;
fast_sint_t c; for (c = 0; c < ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_bucket[c]; induction_bucket[c] = A + B; temp_bucket[c] = A; }
}
}
#pragma omp barrier
{
libsais_final_order_scan_left_to_right_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count);
}
}
#endif
}
}
static void libsais_final_sorting_scan_left_to_right_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_sorting_scan_left_to_right_8u(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_final_sorting_scan_left_to_right_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t;
for (t = 0; t < omp_num_threads; ++t)
{
sa_sint_t * RESTRICT temp_bucket = thread_state[t].state.buckets;
fast_sint_t c; for (c = 0; c < ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_bucket[c]; induction_bucket[c] = A + B; temp_bucket[c] = A; }
}
}
#pragma omp barrier
{
libsais_final_order_scan_left_to_right_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count);
}
}
#endif
}
}
static void libsais_final_sorting_scan_left_to_right_32s_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_sorting_scan_left_to_right_32s(T, SA, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_final_sorting_scan_left_to_right_32s_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_final_sorting_scan_left_to_right_32s_block_sort(T, buckets, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
#endif
static void libsais_final_bwt_scan_left_to_right_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, fast_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
SA[induction_bucket[T[(sa_sint_t)n - 1]]++] = ((sa_sint_t)n - 1) | ((sa_sint_t)(T[(sa_sint_t)n - 2] < T[(sa_sint_t)n - 1]) << (SAINT_BIT - 1));
if (threads == 1 || n < 65536)
{
libsais_final_bwt_scan_left_to_right_8u(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start;
for (block_start = 0; block_start < n; )
{
if (SA[block_start] == 0)
{
block_start++;
}
else
{
fast_sint_t block_max_end = block_start + ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end > n) { block_max_end = n;}
fast_sint_t block_end = block_start + 1; while (block_end < block_max_end && SA[block_end] != 0) { block_end++; }
fast_sint_t block_size = block_end - block_start;
if (block_size < 32)
{
for (; block_start < block_end; block_start += 1)
{
sa_sint_t p = SA[block_start]; SA[block_start] = p & SAINT_MAX; if (p > 0) { p--; SA[block_start] = T[p] | SAINT_MIN; SA[induction_bucket[T[p]]++] = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
}
else
{
libsais_final_bwt_scan_left_to_right_8u_block_omp(T, SA, induction_bucket, block_start, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_final_sorting_scan_left_to_right_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, fast_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
SA[induction_bucket[T[(sa_sint_t)n - 1]]++] = ((sa_sint_t)n - 1) | ((sa_sint_t)(T[(sa_sint_t)n - 2] < T[(sa_sint_t)n - 1]) << (SAINT_BIT - 1));
if (threads == 1 || n < 65536)
{
libsais_final_sorting_scan_left_to_right_8u(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start;
for (block_start = 0; block_start < n; )
{
if (SA[block_start] == 0)
{
block_start++;
}
else
{
fast_sint_t block_max_end = block_start + ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end > n) { block_max_end = n;}
fast_sint_t block_end = block_start + 1; while (block_end < block_max_end && SA[block_end] != 0) { block_end++; }
fast_sint_t block_size = block_end - block_start;
if (block_size < 32)
{
for (; block_start < block_end; block_start += 1)
{
sa_sint_t p = SA[block_start]; SA[block_start] = p ^ SAINT_MIN; if (p > 0) { p--; SA[induction_bucket[T[p]]++] = p | ((sa_sint_t)(T[p - (p > 0)] < T[p]) << (SAINT_BIT - 1)); }
}
}
else
{
libsais_final_sorting_scan_left_to_right_8u_block_omp(T, SA, induction_bucket, block_start, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_final_sorting_scan_left_to_right_32s_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
SA[induction_bucket[T[n - 1]]++] = (n - 1) | ((sa_sint_t)(T[n - 2] < T[n - 1]) << (SAINT_BIT - 1));
if (threads == 1 || n < 65536)
{
libsais_final_sorting_scan_left_to_right_32s(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = 0; block_start < n; block_start = block_end)
{
block_end = block_start + (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end > n) { block_end = n; }
libsais_final_sorting_scan_left_to_right_32s_block_omp(T, SA, induction_bucket, thread_state[0].state.cache, block_start, block_end - block_start, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static sa_sint_t libsais_final_bwt_scan_right_to_left_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j; sa_sint_t index = -1;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 2 * prefetch_distance]);
sa_sint_t s0 = SA[i - prefetch_distance - 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - prefetch_distance - 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i - 0]; index = (p0 == 0) ? (sa_sint_t)(i - 0) : index;
SA[i - 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; uint8_t c0 = T[p0 - (p0 > 0)], c1 = T[p0]; SA[i - 0] = c1; sa_sint_t t = c0 | SAINT_MIN; SA[--induction_bucket[c1]] = (c0 <= c1) ? p0 : t; }
sa_sint_t p1 = SA[i - 1]; index = (p1 == 0) ? (sa_sint_t)(i - 1) : index;
SA[i - 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; uint8_t c0 = T[p1 - (p1 > 0)], c1 = T[p1]; SA[i - 1] = c1; sa_sint_t t = c0 | SAINT_MIN; SA[--induction_bucket[c1]] = (c0 <= c1) ? p1 : t; }
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; index = (p == 0) ? (sa_sint_t)i : index;
SA[i] = p & SAINT_MAX; if (p > 0) { p--; uint8_t c0 = T[p - (p > 0)], c1 = T[p]; SA[i] = c1; sa_sint_t t = c0 | SAINT_MIN; SA[--induction_bucket[c1]] = (c0 <= c1) ? p : t; }
}
return index;
}
static void libsais_final_sorting_scan_right_to_left_8u(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 2 * prefetch_distance]);
sa_sint_t s0 = SA[i - prefetch_distance - 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - prefetch_distance - 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i - 0]; SA[i - 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; SA[--induction_bucket[T[p0]]] = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] > T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i - 1]; SA[i - 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; SA[--induction_bucket[T[p1]]] = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] > T[p1]) << (SAINT_BIT - 1)); }
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; SA[--induction_bucket[T[p]]] = p | ((sa_sint_t)(T[p - (p > 0)] > T[p]) << (SAINT_BIT - 1)); }
}
}
static void libsais_final_sorting_scan_right_to_left_32s(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + 2 * prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 3 * prefetch_distance]);
sa_sint_t s0 = SA[i - 2 * prefetch_distance - 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - 2 * prefetch_distance - 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t s2 = SA[i - 1 * prefetch_distance - 0]; if (s2 > 0) { libsais_prefetchw(&induction_bucket[T[s2 - 1]]); libsais_prefetch(&T[s2] - 2); }
sa_sint_t s3 = SA[i - 1 * prefetch_distance - 1]; if (s3 > 0) { libsais_prefetchw(&induction_bucket[T[s3 - 1]]); libsais_prefetch(&T[s3] - 2); }
sa_sint_t p0 = SA[i - 0]; SA[i - 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; SA[--induction_bucket[T[p0]]] = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] > T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i - 1]; SA[i - 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; SA[--induction_bucket[T[p1]]] = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] > T[p1]) << (SAINT_BIT - 1)); }
}
for (j -= 2 * prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; SA[--induction_bucket[T[p]]] = p | ((sa_sint_t)(T[p - (p > 0)] > T[p]) << (SAINT_BIT - 1)); }
}
}
#if defined(_OPENMP)
static fast_sint_t libsais_final_bwt_scan_right_to_left_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 2 * prefetch_distance]);
sa_sint_t s0 = SA[i - prefetch_distance - 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - prefetch_distance - 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i - 0]; SA[i - 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; uint8_t c0 = T[p0 - (p0 > 0)], c1 = T[p0]; SA[i - 0] = c1; sa_sint_t t = c0 | SAINT_MIN; buckets[cache[count].symbol = c1]++; cache[count++].index = (c0 <= c1) ? p0 : t; }
sa_sint_t p1 = SA[i - 1]; SA[i - 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; uint8_t c0 = T[p1 - (p1 > 0)], c1 = T[p1]; SA[i - 1] = c1; sa_sint_t t = c0 | SAINT_MIN; buckets[cache[count].symbol = c1]++; cache[count++].index = (c0 <= c1) ? p1 : t; }
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; uint8_t c0 = T[p - (p > 0)], c1 = T[p]; SA[i] = c1; sa_sint_t t = c0 | SAINT_MIN; buckets[cache[count].symbol = c1]++; cache[count++].index = (c0 <= c1) ? p : t; }
}
return count;
}
static fast_sint_t libsais_final_sorting_scan_right_to_left_8u_block_prepare(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
memset(buckets, 0, ALPHABET_SIZE * sizeof(sa_sint_t));
fast_sint_t i, j, count = 0;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&SA[i - 2 * prefetch_distance]);
sa_sint_t s0 = SA[i - prefetch_distance - 0]; const uint8_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i - prefetch_distance - 1]; const uint8_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
sa_sint_t p0 = SA[i - 0]; SA[i - 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; buckets[cache[count].symbol = T[p0]]++; cache[count++].index = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] > T[p0]) << (SAINT_BIT - 1)); }
sa_sint_t p1 = SA[i - 1]; SA[i - 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; buckets[cache[count].symbol = T[p1]]++; cache[count++].index = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] > T[p1]) << (SAINT_BIT - 1)); }
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; buckets[cache[count].symbol = T[p]]++; cache[count++].index = p | ((sa_sint_t)(T[p - (p > 0)] > T[p]) << (SAINT_BIT - 1)); }
}
return count;
}
static void libsais_final_order_scan_right_to_left_8u_block_place(sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t count)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = 0, j = count - 3; i < j; i += 4)
{
libsais_prefetch(&cache[i + prefetch_distance]);
SA[--buckets[cache[i + 0].symbol]] = cache[i + 0].index;
SA[--buckets[cache[i + 1].symbol]] = cache[i + 1].index;
SA[--buckets[cache[i + 2].symbol]] = cache[i + 2].index;
SA[--buckets[cache[i + 3].symbol]] = cache[i + 3].index;
}
for (j += 3; i < j; i += 1)
{
SA[--buckets[cache[i].symbol]] = cache[i].index;
}
}
static void libsais_final_sorting_scan_right_to_left_32s_block_gather(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 1; i < j; i += 2)
{
libsais_prefetchw(&SA[i + 2 * prefetch_distance]);
sa_sint_t s0 = SA[i + prefetch_distance + 0]; const sa_sint_t * Ts0 = &T[s0] - 1; libsais_prefetch(s0 > 0 ? Ts0 : NULL); Ts0--; libsais_prefetch(s0 > 0 ? Ts0 : NULL);
sa_sint_t s1 = SA[i + prefetch_distance + 1]; const sa_sint_t * Ts1 = &T[s1] - 1; libsais_prefetch(s1 > 0 ? Ts1 : NULL); Ts1--; libsais_prefetch(s1 > 0 ? Ts1 : NULL);
libsais_prefetchw(&cache[i + prefetch_distance]);
sa_sint_t symbol0 = SAINT_MIN, p0 = SA[i + 0]; SA[i + 0] = p0 & SAINT_MAX; if (p0 > 0) { p0--; cache[i + 0].index = p0 | ((sa_sint_t)(T[p0 - (p0 > 0)] > T[p0]) << (SAINT_BIT - 1)); symbol0 = T[p0]; } cache[i + 0].symbol = symbol0;
sa_sint_t symbol1 = SAINT_MIN, p1 = SA[i + 1]; SA[i + 1] = p1 & SAINT_MAX; if (p1 > 0) { p1--; cache[i + 1].index = p1 | ((sa_sint_t)(T[p1 - (p1 > 0)] > T[p1]) << (SAINT_BIT - 1)); symbol1 = T[p1]; } cache[i + 1].symbol = symbol1;
}
for (j += prefetch_distance + 1; i < j; i += 1)
{
sa_sint_t symbol = SAINT_MIN, p = SA[i]; SA[i] = p & SAINT_MAX; if (p > 0) { p--; cache[i].index = p | ((sa_sint_t)(T[p - (p > 0)] > T[p]) << (SAINT_BIT - 1)); symbol = T[p]; } cache[i].symbol = symbol;
}
}
static void libsais_final_sorting_scan_right_to_left_32s_block_sort(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT induction_bucket, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = omp_block_start + omp_block_size - 1, j = omp_block_start + prefetch_distance + 1; i >= j; i -= 2)
{
libsais_prefetchw(&cache[i - 2 * prefetch_distance]);
sa_sint_t s0 = cache[i - prefetch_distance - 0].symbol; const sa_sint_t * Is0 = &induction_bucket[s0]; libsais_prefetchw(s0 >= 0 ? Is0 : NULL);
sa_sint_t s1 = cache[i - prefetch_distance - 1].symbol; const sa_sint_t * Is1 = &induction_bucket[s1]; libsais_prefetchw(s1 >= 0 ? Is1 : NULL);
sa_sint_t v0 = cache[i - 0].symbol;
if (v0 >= 0)
{
cache[i - 0].symbol = --induction_bucket[v0];
if (cache[i - 0].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 0].symbol, np = cache[i - 0].index; cache[i - 0].index = np & SAINT_MAX; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] > T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
sa_sint_t v1 = cache[i - 1].symbol;
if (v1 >= 0)
{
cache[i - 1].symbol = --induction_bucket[v1];
if (cache[i - 1].symbol >= omp_block_start) { sa_sint_t ni = cache[i - 1].symbol, np = cache[i - 1].index; cache[i - 1].index = np & SAINT_MAX; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] > T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
}
for (j -= prefetch_distance + 1; i >= j; i -= 1)
{
sa_sint_t v = cache[i].symbol;
if (v >= 0)
{
cache[i].symbol = --induction_bucket[v];
if (cache[i].symbol >= omp_block_start) { sa_sint_t ni = cache[i].symbol, np = cache[i].index; cache[i].index = np & SAINT_MAX; if (np > 0) { np--; cache[ni].index = np | ((sa_sint_t)(T[np - (np > 0)] > T[np]) << (SAINT_BIT - 1)); cache[ni].symbol = T[np]; } }
}
}
}
static void libsais_final_bwt_scan_right_to_left_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_bwt_scan_right_to_left_8u(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_final_bwt_scan_right_to_left_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
sa_sint_t * RESTRICT temp_bucket = thread_state[t].state.buckets;
fast_sint_t c; for (c = 0; c < ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_bucket[c]; induction_bucket[c] = A - B; temp_bucket[c] = A; }
}
}
#pragma omp barrier
{
libsais_final_order_scan_right_to_left_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count);
}
}
#endif
}
}
static void libsais_final_sorting_scan_right_to_left_8u_block_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT induction_bucket, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384 && omp_get_dynamic() == 0)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_sorting_scan_right_to_left_8u(T, SA, induction_bucket, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_final_sorting_scan_right_to_left_8u_block_prepare(T, SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t;
for (t = omp_num_threads - 1; t >= 0; --t)
{
sa_sint_t * RESTRICT temp_bucket = thread_state[t].state.buckets;
fast_sint_t c; for (c = 0; c < ALPHABET_SIZE; c += 1) { sa_sint_t A = induction_bucket[c], B = temp_bucket[c]; induction_bucket[c] = A - B; temp_bucket[c] = A; }
}
}
#pragma omp barrier
{
libsais_final_order_scan_right_to_left_8u_block_place(SA, thread_state[omp_thread_num].state.buckets, thread_state[omp_thread_num].state.cache, thread_state[omp_thread_num].state.count);
}
}
#endif
}
}
static void libsais_final_sorting_scan_right_to_left_32s_block_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t * RESTRICT buckets, LIBSAIS_THREAD_CACHE * RESTRICT cache, fast_sint_t block_start, fast_sint_t block_size, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && block_size >= 16384)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(cache);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (block_size / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : block_size - omp_block_start;
omp_block_start += block_start;
if (omp_num_threads == 1)
{
libsais_final_sorting_scan_right_to_left_32s(T, SA, buckets, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
libsais_final_sorting_scan_right_to_left_32s_block_gather(T, SA, cache - block_start, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
libsais_final_sorting_scan_right_to_left_32s_block_sort(T, buckets, cache - block_start, block_start, block_size);
}
#pragma omp barrier
{
libsais_compact_and_place_cached_suffixes(SA, cache - block_start, omp_block_start, omp_block_size);
}
}
#endif
}
}
#endif
static sa_sint_t libsais_final_bwt_scan_right_to_left_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t index = -1;
if (threads == 1 || n < 65536)
{
index = libsais_final_bwt_scan_right_to_left_8u(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start;
for (block_start = (fast_sint_t)n - 1; block_start >= 0; )
{
if (SA[block_start] == 0)
{
index = (sa_sint_t)block_start--;
}
else
{
fast_sint_t block_max_end = block_start - ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end < 0) { block_max_end = -1; }
fast_sint_t block_end = block_start - 1; while (block_end > block_max_end && SA[block_end] != 0) { block_end--; }
fast_sint_t block_size = block_start - block_end;
if (block_size < 32)
{
for (; block_start > block_end; block_start -= 1)
{
sa_sint_t p = SA[block_start]; SA[block_start] = p & SAINT_MAX; if (p > 0) { p--; uint8_t c0 = T[p - (p > 0)], c1 = T[p]; SA[block_start] = c1; sa_sint_t t = c0 | SAINT_MIN; SA[--induction_bucket[c1]] = (c0 <= c1) ? p : t; }
}
}
else
{
libsais_final_bwt_scan_right_to_left_8u_block_omp(T, SA, induction_bucket, block_end + 1, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
return index;
}
static void libsais_final_sorting_scan_right_to_left_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || n < 65536)
{
libsais_final_sorting_scan_right_to_left_8u(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start;
for (block_start = (fast_sint_t)n - 1; block_start >= 0; )
{
if (SA[block_start] == 0)
{
block_start--;
}
else
{
fast_sint_t block_max_end = block_start - ((fast_sint_t)threads) * (LIBSAIS_PER_THREAD_CACHE_SIZE - 16 * (fast_sint_t)threads); if (block_max_end < -1) { block_max_end = -1; }
fast_sint_t block_end = block_start - 1; while (block_end > block_max_end && SA[block_end] != 0) { block_end--; }
fast_sint_t block_size = block_start - block_end;
if (block_size < 32)
{
for (; block_start > block_end; block_start -= 1)
{
sa_sint_t p = SA[block_start]; SA[block_start] = p & SAINT_MAX; if (p > 0) { p--; SA[--induction_bucket[T[p]]] = p | ((sa_sint_t)(T[p - (p > 0)] > T[p]) << (SAINT_BIT - 1)); }
}
}
else
{
libsais_final_sorting_scan_right_to_left_8u_block_omp(T, SA, induction_bucket, block_end + 1, block_size, threads, thread_state);
block_start = block_end;
}
}
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_final_sorting_scan_right_to_left_32s_omp(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t * RESTRICT induction_bucket, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (threads == 1 || n < 65536)
{
libsais_final_sorting_scan_right_to_left_32s(T, SA, induction_bucket, 0, n);
}
#if defined(_OPENMP)
else
{
fast_sint_t block_start, block_end;
for (block_start = (fast_sint_t)n - 1; block_start >= 0; block_start = block_end)
{
block_end = block_start - (fast_sint_t)threads * LIBSAIS_PER_THREAD_CACHE_SIZE; if (block_end < 0) { block_end = -1; }
libsais_final_sorting_scan_right_to_left_32s_block_omp(T, SA, induction_bucket, thread_state[0].state.cache, block_end + 1, block_start - block_end, threads);
}
}
#else
UNUSED(thread_state);
#endif
}
static void libsais_clear_lms_suffixes_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT bucket_start, sa_sint_t * RESTRICT bucket_end, sa_sint_t threads)
{
fast_sint_t c;
#if defined(_OPENMP)
#pragma omp parallel for schedule(static, 1) num_threads(threads) if(threads > 1 && n >= 65536)
#else
UNUSED(threads); UNUSED(n);
#endif
for (c = 0; c < k; ++c)
{
if (bucket_end[c] > bucket_start[c])
{
memset(&SA[bucket_start[c]], 0, ((size_t)bucket_end[c] - (size_t)bucket_start[c]) * sizeof(sa_sint_t));
}
}
}
static sa_sint_t libsais_induce_final_order_8u_omp(const uint8_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t bwt, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (bwt)
{
libsais_final_bwt_scan_left_to_right_8u_omp(T, SA, n, &buckets[6 * ALPHABET_SIZE], threads, thread_state);
if (threads > 1 && n >= 65536) { libsais_clear_lms_suffixes_omp(SA, n, ALPHABET_SIZE, &buckets[6 * ALPHABET_SIZE], &buckets[7 * ALPHABET_SIZE], threads); }
return libsais_final_bwt_scan_right_to_left_8u_omp(T, SA, n, &buckets[7 * ALPHABET_SIZE], threads, thread_state);
}
else
{
libsais_final_sorting_scan_left_to_right_8u_omp(T, SA, n, &buckets[6 * ALPHABET_SIZE], threads, thread_state);
if (threads > 1 && n >= 65536) { libsais_clear_lms_suffixes_omp(SA, n, ALPHABET_SIZE, &buckets[6 * ALPHABET_SIZE], &buckets[7 * ALPHABET_SIZE], threads); }
libsais_final_sorting_scan_right_to_left_8u_omp(T, SA, n, &buckets[7 * ALPHABET_SIZE], threads, thread_state);
return 0;
}
}
static void libsais_induce_final_order_32s_6k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_final_sorting_scan_left_to_right_32s_omp(T, SA, n, &buckets[4 * k], threads, thread_state);
libsais_final_sorting_scan_right_to_left_32s_omp(T, SA, n, &buckets[5 * k], threads, thread_state);
}
static void libsais_induce_final_order_32s_4k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_final_sorting_scan_left_to_right_32s_omp(T, SA, n, &buckets[2 * k], threads, thread_state);
libsais_final_sorting_scan_right_to_left_32s_omp(T, SA, n, &buckets[3 * k], threads, thread_state);
}
static void libsais_induce_final_order_32s_2k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_final_sorting_scan_left_to_right_32s_omp(T, SA, n, &buckets[1 * k], threads, thread_state);
libsais_final_sorting_scan_right_to_left_32s_omp(T, SA, n, &buckets[0 * k], threads, thread_state);
}
static void libsais_induce_final_order_32s_1k(const sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_start_32s_1k(k, buckets);
libsais_final_sorting_scan_left_to_right_32s_omp(T, SA, n, buckets, threads, thread_state);
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_end_32s_1k(k, buckets);
libsais_final_sorting_scan_right_to_left_32s_omp(T, SA, n, buckets, threads, thread_state);
}
static sa_sint_t libsais_renumber_unique_and_nonunique_lms_suffixes_32s(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t f, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
sa_sint_t i, j;
for (i = (sa_sint_t)omp_block_start, j = (sa_sint_t)omp_block_start + (sa_sint_t)omp_block_size - 2 * (sa_sint_t)prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + 3 * prefetch_distance]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + 2 * prefetch_distance + 0]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + 2 * prefetch_distance + 1]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + 2 * prefetch_distance + 2]) >> 1]);
libsais_prefetchw(&SAm[((sa_uint_t)SA[i + 2 * prefetch_distance + 3]) >> 1]);
sa_uint_t q0 = (sa_uint_t)SA[i + prefetch_distance + 0]; const sa_sint_t * Tq0 = &T[q0]; libsais_prefetchw(SAm[q0 >> 1] < 0 ? Tq0 : NULL);
sa_uint_t q1 = (sa_uint_t)SA[i + prefetch_distance + 1]; const sa_sint_t * Tq1 = &T[q1]; libsais_prefetchw(SAm[q1 >> 1] < 0 ? Tq1 : NULL);
sa_uint_t q2 = (sa_uint_t)SA[i + prefetch_distance + 2]; const sa_sint_t * Tq2 = &T[q2]; libsais_prefetchw(SAm[q2 >> 1] < 0 ? Tq2 : NULL);
sa_uint_t q3 = (sa_uint_t)SA[i + prefetch_distance + 3]; const sa_sint_t * Tq3 = &T[q3]; libsais_prefetchw(SAm[q3 >> 1] < 0 ? Tq3 : NULL);
sa_uint_t p0 = (sa_uint_t)SA[i + 0]; sa_sint_t s0 = SAm[p0 >> 1]; if (s0 < 0) { T[p0] |= SAINT_MIN; f++; s0 = i + 0 + SAINT_MIN + f; } SAm[p0 >> 1] = s0 - f;
sa_uint_t p1 = (sa_uint_t)SA[i + 1]; sa_sint_t s1 = SAm[p1 >> 1]; if (s1 < 0) { T[p1] |= SAINT_MIN; f++; s1 = i + 1 + SAINT_MIN + f; } SAm[p1 >> 1] = s1 - f;
sa_uint_t p2 = (sa_uint_t)SA[i + 2]; sa_sint_t s2 = SAm[p2 >> 1]; if (s2 < 0) { T[p2] |= SAINT_MIN; f++; s2 = i + 2 + SAINT_MIN + f; } SAm[p2 >> 1] = s2 - f;
sa_uint_t p3 = (sa_uint_t)SA[i + 3]; sa_sint_t s3 = SAm[p3 >> 1]; if (s3 < 0) { T[p3] |= SAINT_MIN; f++; s3 = i + 3 + SAINT_MIN + f; } SAm[p3 >> 1] = s3 - f;
}
for (j += 2 * (sa_sint_t)prefetch_distance + 3; i < j; i += 1)
{
sa_uint_t p = (sa_uint_t)SA[i]; sa_sint_t s = SAm[p >> 1]; if (s < 0) { T[p] |= SAINT_MIN; f++; s = i + SAINT_MIN + f; } SAm[p >> 1] = s - f;
}
return f;
}
static void libsais_compact_unique_and_nonunique_lms_suffixes_32s(sa_sint_t * RESTRICT SA, sa_sint_t m, fast_sint_t * pl, fast_sint_t * pr, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAl = &SA[0];
sa_sint_t * RESTRICT SAr = &SA[0];
fast_sint_t i, j, l = *pl - 1, r = *pr - 1;
for (i = (fast_sint_t)m + omp_block_start + omp_block_size - 1, j = (fast_sint_t)m + omp_block_start + 3; i >= j; i -= 4)
{
libsais_prefetch(&SA[i - prefetch_distance]);
sa_sint_t p0 = SA[i - 0]; SAl[l] = p0 & SAINT_MAX; l -= p0 < 0; SAr[r] = p0 - 1; r -= p0 > 0;
sa_sint_t p1 = SA[i - 1]; SAl[l] = p1 & SAINT_MAX; l -= p1 < 0; SAr[r] = p1 - 1; r -= p1 > 0;
sa_sint_t p2 = SA[i - 2]; SAl[l] = p2 & SAINT_MAX; l -= p2 < 0; SAr[r] = p2 - 1; r -= p2 > 0;
sa_sint_t p3 = SA[i - 3]; SAl[l] = p3 & SAINT_MAX; l -= p3 < 0; SAr[r] = p3 - 1; r -= p3 > 0;
}
for (j -= 3; i >= j; i -= 1)
{
sa_sint_t p = SA[i]; SAl[l] = p & SAINT_MAX; l -= p < 0; SAr[r] = p - 1; r -= p > 0;
}
*pl = l + 1; *pr = r + 1;
}
#if defined(_OPENMP)
static sa_sint_t libsais_count_unique_suffixes(sa_sint_t * RESTRICT SA, sa_sint_t m, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
sa_sint_t * RESTRICT SAm = &SA[m];
fast_sint_t i, j; sa_sint_t f0 = 0, f1 = 0, f2 = 0, f3 = 0;
for (i = omp_block_start, j = omp_block_start + omp_block_size - prefetch_distance - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + 2 * prefetch_distance]);
libsais_prefetch(&SAm[((sa_uint_t)SA[i + prefetch_distance + 0]) >> 1]);
libsais_prefetch(&SAm[((sa_uint_t)SA[i + prefetch_distance + 1]) >> 1]);
libsais_prefetch(&SAm[((sa_uint_t)SA[i + prefetch_distance + 2]) >> 1]);
libsais_prefetch(&SAm[((sa_uint_t)SA[i + prefetch_distance + 3]) >> 1]);
f0 += SAm[((sa_uint_t)SA[i + 0]) >> 1] < 0;
f1 += SAm[((sa_uint_t)SA[i + 1]) >> 1] < 0;
f2 += SAm[((sa_uint_t)SA[i + 2]) >> 1] < 0;
f3 += SAm[((sa_uint_t)SA[i + 3]) >> 1] < 0;
}
for (j += prefetch_distance + 3; i < j; i += 1)
{
f0 += SAm[((sa_uint_t)SA[i]) >> 1] < 0;
}
return f0 + f1 + f2 + f3;
}
#endif
static sa_sint_t libsais_renumber_unique_and_nonunique_lms_suffixes_32s_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t m, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t f = 0;
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && m >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (m / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : m - omp_block_start;
if (omp_num_threads == 1)
{
f = libsais_renumber_unique_and_nonunique_lms_suffixes_32s(T, SA, m, 0, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_count_unique_suffixes(SA, m, omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, count = 0; for (t = 0; t < omp_thread_num; ++t) { count += thread_state[t].state.count; }
if (omp_thread_num == omp_num_threads - 1)
{
f = (sa_sint_t)(count + thread_state[omp_thread_num].state.count);
}
libsais_renumber_unique_and_nonunique_lms_suffixes_32s(T, SA, m, (sa_sint_t)count, omp_block_start, omp_block_size);
}
}
#endif
}
return f;
}
static void libsais_compact_unique_and_nonunique_lms_suffixes_32s_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t fs, sa_sint_t f, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 131072 && m < fs)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (((fast_sint_t)n >> 1) / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : ((fast_sint_t)n >> 1) - omp_block_start;
if (omp_num_threads == 1)
{
fast_sint_t l = m, r = (fast_sint_t)n + (fast_sint_t)fs;
libsais_compact_unique_and_nonunique_lms_suffixes_32s(SA, m, &l, &r, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.position = (fast_sint_t)m + ((fast_sint_t)n >> 1) + omp_block_start + omp_block_size;
thread_state[omp_thread_num].state.count = (fast_sint_t)m + omp_block_start + omp_block_size;
libsais_compact_unique_and_nonunique_lms_suffixes_32s(SA, m, &thread_state[omp_thread_num].state.position, &thread_state[omp_thread_num].state.count, omp_block_start, omp_block_size);
}
#pragma omp barrier
#pragma omp master
{
fast_sint_t t, position;
for (position = m, t = omp_num_threads - 1; t >= 0; --t)
{
fast_sint_t omp_block_end = t < omp_num_threads - 1 ? omp_block_stride * (t + 1) : ((fast_sint_t)n >> 1);
fast_sint_t count = ((fast_sint_t)m + ((fast_sint_t)n >> 1) + omp_block_end - thread_state[t].state.position);
if (count > 0)
{
position -= count; memcpy(&SA[position], &SA[thread_state[t].state.position], (size_t)count * sizeof(sa_sint_t));
}
}
for (position = (fast_sint_t)n + (fast_sint_t)fs, t = omp_num_threads - 1; t >= 0; --t)
{
fast_sint_t omp_block_end = t < omp_num_threads - 1 ? omp_block_stride * (t + 1) : ((fast_sint_t)n >> 1);
fast_sint_t count = ((fast_sint_t)m + omp_block_end - thread_state[t].state.count);
if (count > 0)
{
position -= count; memcpy(&SA[position], &SA[thread_state[t].state.count], (size_t)count * sizeof(sa_sint_t));
}
}
}
}
#endif
}
memcpy(&SA[(fast_sint_t)n + (fast_sint_t)fs - (fast_sint_t)m], &SA[(fast_sint_t)m - (fast_sint_t)f], (size_t)f * sizeof(sa_sint_t));
}
static sa_sint_t libsais_compact_lms_suffixes_32s_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t fs, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t f = libsais_renumber_unique_and_nonunique_lms_suffixes_32s_omp(T, SA, m, threads, thread_state);
libsais_compact_unique_and_nonunique_lms_suffixes_32s_omp(SA, n, m, fs, f, threads, thread_state);
return f;
}
static void libsais_merge_unique_lms_suffixes_32s(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, fast_sint_t l, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
const sa_sint_t * RESTRICT SAnm = &SA[(fast_sint_t)n - (fast_sint_t)m - 1 + l];
sa_sint_t i, j; fast_sint_t tmp = *SAnm++;
for (i = (sa_sint_t)omp_block_start, j = (sa_sint_t)omp_block_start + (sa_sint_t)omp_block_size - 6; i < j; i += 4)
{
libsais_prefetch(&T[i + prefetch_distance]);
sa_sint_t c0 = T[i + 0]; if (c0 < 0) { T[i + 0] = c0 & SAINT_MAX; SA[tmp] = i + 0; i++; tmp = *SAnm++; }
sa_sint_t c1 = T[i + 1]; if (c1 < 0) { T[i + 1] = c1 & SAINT_MAX; SA[tmp] = i + 1; i++; tmp = *SAnm++; }
sa_sint_t c2 = T[i + 2]; if (c2 < 0) { T[i + 2] = c2 & SAINT_MAX; SA[tmp] = i + 2; i++; tmp = *SAnm++; }
sa_sint_t c3 = T[i + 3]; if (c3 < 0) { T[i + 3] = c3 & SAINT_MAX; SA[tmp] = i + 3; i++; tmp = *SAnm++; }
}
for (j += 6; i < j; i += 1)
{
sa_sint_t c = T[i]; if (c < 0) { T[i] = c & SAINT_MAX; SA[tmp] = i; i++; tmp = *SAnm++; }
}
}
static void libsais_merge_nonunique_lms_suffixes_32s(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, fast_sint_t l, fast_sint_t omp_block_start, fast_sint_t omp_block_size)
{
const fast_sint_t prefetch_distance = 32;
const sa_sint_t * RESTRICT SAnm = &SA[(fast_sint_t)n - (fast_sint_t)m - 1 + l];
fast_sint_t i, j; sa_sint_t tmp = *SAnm++;
for (i = omp_block_start, j = omp_block_start + omp_block_size - 3; i < j; i += 4)
{
libsais_prefetch(&SA[i + prefetch_distance]);
if (SA[i + 0] == 0) { SA[i + 0] = tmp; tmp = *SAnm++; }
if (SA[i + 1] == 0) { SA[i + 1] = tmp; tmp = *SAnm++; }
if (SA[i + 2] == 0) { SA[i + 2] = tmp; tmp = *SAnm++; }
if (SA[i + 3] == 0) { SA[i + 3] = tmp; tmp = *SAnm++; }
}
for (j += 3; i < j; i += 1)
{
if (SA[i] == 0) { SA[i] = tmp; tmp = *SAnm++; }
}
}
static void libsais_merge_unique_lms_suffixes_32s_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : n - omp_block_start;
if (omp_num_threads == 1)
{
libsais_merge_unique_lms_suffixes_32s(T, SA, n, m, 0, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_count_negative_marked_suffixes(T, omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, count = 0; for (t = 0; t < omp_thread_num; ++t) { count += thread_state[t].state.count; }
libsais_merge_unique_lms_suffixes_32s(T, SA, n, m, count, omp_block_start, omp_block_size);
}
}
#endif
}
}
static void libsais_merge_nonunique_lms_suffixes_32s_omp(sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t f, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && m >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
#else
UNUSED(threads); UNUSED(thread_state);
fast_sint_t omp_thread_num = 0;
fast_sint_t omp_num_threads = 1;
#endif
fast_sint_t omp_block_stride = (m / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : m - omp_block_start;
if (omp_num_threads == 1)
{
libsais_merge_nonunique_lms_suffixes_32s(SA, n, m, f, omp_block_start, omp_block_size);
}
#if defined(_OPENMP)
else
{
{
thread_state[omp_thread_num].state.count = libsais_count_zero_marked_suffixes(SA, omp_block_start, omp_block_size);
}
#pragma omp barrier
{
fast_sint_t t, count = f; for (t = 0; t < omp_thread_num; ++t) { count += thread_state[t].state.count; }
libsais_merge_nonunique_lms_suffixes_32s(SA, n, m, count, omp_block_start, omp_block_size);
}
}
#endif
}
}
static void libsais_merge_compacted_lms_suffixes_32s_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t f, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
libsais_merge_unique_lms_suffixes_32s_omp(T, SA, n, m, threads, thread_state);
libsais_merge_nonunique_lms_suffixes_32s_omp(SA, n, m, f, threads, thread_state);
}
static void libsais_reconstruct_compacted_lms_suffixes_32s_2k_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t m, sa_sint_t fs, sa_sint_t f, sa_sint_t * RESTRICT buckets, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (f > 0)
{
memmove(&SA[n - m - 1], &SA[n + fs - m], (size_t)f * sizeof(sa_sint_t));
libsais_count_and_gather_compacted_lms_suffixes_32s_2k_omp(T, SA, n, k, buckets, threads, thread_state);
libsais_reconstruct_lms_suffixes_omp(SA, n, m - f, threads);
memcpy(&SA[n - m - 1 + f], &SA[0], ((size_t)m - (size_t)f) * sizeof(sa_sint_t));
memset(&SA[0], 0, (size_t)m * sizeof(sa_sint_t));
libsais_merge_compacted_lms_suffixes_32s_omp(T, SA, n, m, f, threads, thread_state);
}
else
{
libsais_count_and_gather_lms_suffixes_32s_2k(T, SA, n, k, buckets, 0, n);
libsais_reconstruct_lms_suffixes_omp(SA, n, m, threads);
}
}
static void libsais_reconstruct_compacted_lms_suffixes_32s_1k_omp(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t m, sa_sint_t fs, sa_sint_t f, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (f > 0)
{
memmove(&SA[n - m - 1], &SA[n + fs - m], (size_t)f * sizeof(sa_sint_t));
libsais_gather_compacted_lms_suffixes_32s(T, SA, n);
libsais_reconstruct_lms_suffixes_omp(SA, n, m - f, threads);
memcpy(&SA[n - m - 1 + f], &SA[0], ((size_t)m - (size_t)f) * sizeof(sa_sint_t));
memset(&SA[0], 0, (size_t)m * sizeof(sa_sint_t));
libsais_merge_compacted_lms_suffixes_32s_omp(T, SA, n, m, f, threads, thread_state);
}
else
{
libsais_gather_lms_suffixes_32s(T, SA, n);
libsais_reconstruct_lms_suffixes_omp(SA, n, m, threads);
}
}
static sa_sint_t libsais_main_32s(sa_sint_t * RESTRICT T, sa_sint_t * RESTRICT SA, sa_sint_t n, sa_sint_t k, sa_sint_t fs, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
if (k > 0 && fs / k >= 6)
{
sa_sint_t alignment = (fs - 1024) / k >= 6 ? 1024 : 16;
sa_sint_t * RESTRICT buckets = (fs - alignment) / k >= 6 ? (sa_sint_t *)libsais_align_up(&SA[n + fs - 6 * k - alignment], (size_t)alignment * sizeof(sa_sint_t)) : &SA[n + fs - 6 * k];
sa_sint_t m = libsais_count_and_gather_lms_suffixes_32s_4k_omp(T, SA, n, k, buckets, threads, thread_state);
if (m > 1)
{
memset(SA, 0, ((size_t)n - (size_t)m) * sizeof(sa_sint_t));
sa_sint_t first_lms_suffix = SA[n - m];
sa_sint_t left_suffixes_count = libsais_initialize_buckets_for_lms_suffixes_radix_sort_32s_6k(T, k, buckets, first_lms_suffix);
libsais_radix_sort_lms_suffixes_32s_6k_omp(T, SA, n, m, &buckets[4 * k], threads, thread_state);
libsais_radix_sort_set_markers_32s_6k_omp(SA, k, &buckets[4 * k], threads);
if (threads > 1 && n >= 65536) { memset(&SA[(fast_sint_t)n - (fast_sint_t)m], 0, (size_t)m * sizeof(sa_sint_t)); }
libsais_initialize_buckets_for_partial_sorting_32s_6k(T, k, buckets, first_lms_suffix, left_suffixes_count);
libsais_induce_partial_order_32s_6k_omp(T, SA, n, k, buckets, first_lms_suffix, left_suffixes_count, threads, thread_state);
sa_sint_t names = libsais_renumber_and_mark_distinct_lms_suffixes_32s_4k_omp(SA, n, m, threads, thread_state);
if (names < m)
{
sa_sint_t f = libsais_compact_lms_suffixes_32s_omp(T, SA, n, m, fs, threads, thread_state);
if (libsais_main_32s(SA + n + fs - m + f, SA, m - f, names - f, fs + n - 2 * m + f, threads, thread_state) != 0)
{
return -2;
}
libsais_reconstruct_compacted_lms_suffixes_32s_2k_omp(T, SA, n, k, m, fs, f, buckets, threads, thread_state);
}
else
{
libsais_count_lms_suffixes_32s_2k(T, n, k, buckets);
}
libsais_initialize_buckets_start_and_end_32s_4k(k, buckets);
libsais_place_lms_suffixes_histogram_32s_4k(SA, n, k, m, buckets);
libsais_induce_final_order_32s_4k(T, SA, n, k, buckets, threads, thread_state);
}
else
{
SA[0] = SA[n - 1];
libsais_initialize_buckets_start_and_end_32s_6k(k, buckets);
libsais_place_lms_suffixes_histogram_32s_6k(SA, n, k, m, buckets);
libsais_induce_final_order_32s_6k(T, SA, n, k, buckets, threads, thread_state);
}
return 0;
}
else if (k > 0 && fs / k >= 4)
{
sa_sint_t alignment = (fs - 1024) / k >= 4 ? 1024 : 16;
sa_sint_t * RESTRICT buckets = (fs - alignment) / k >= 4 ? (sa_sint_t *)libsais_align_up(&SA[n + fs - 4 * k - alignment], (size_t)alignment * sizeof(sa_sint_t)) : &SA[n + fs - 4 * k];
sa_sint_t m = libsais_count_and_gather_lms_suffixes_32s_2k_omp(T, SA, n, k, buckets, threads, thread_state);
if (m > 1)
{
libsais_initialize_buckets_for_radix_and_partial_sorting_32s_4k(T, k, buckets, SA[n - m]);
libsais_radix_sort_lms_suffixes_32s_2k_omp(T, SA, n, m, &buckets[1], threads, thread_state);
libsais_radix_sort_set_markers_32s_4k_omp(SA, k, &buckets[1], threads);
libsais_place_lms_suffixes_interval_32s_4k(SA, n, k, m - 1, buckets);
libsais_induce_partial_order_32s_4k_omp(T, SA, n, k, buckets, threads, thread_state);
sa_sint_t names = libsais_renumber_and_mark_distinct_lms_suffixes_32s_4k_omp(SA, n, m, threads, thread_state);
if (names < m)
{
sa_sint_t f = libsais_compact_lms_suffixes_32s_omp(T, SA, n, m, fs, threads, thread_state);
if (libsais_main_32s(SA + n + fs - m + f, SA, m - f, names - f, fs + n - 2 * m + f, threads, thread_state) != 0)
{
return -2;
}
libsais_reconstruct_compacted_lms_suffixes_32s_2k_omp(T, SA, n, k, m, fs, f, buckets, threads, thread_state);
}
else
{
libsais_count_lms_suffixes_32s_2k(T, n, k, buckets);
}
}
else
{
SA[0] = SA[n - 1];
}
libsais_initialize_buckets_start_and_end_32s_4k(k, buckets);
libsais_place_lms_suffixes_histogram_32s_4k(SA, n, k, m, buckets);
libsais_induce_final_order_32s_4k(T, SA, n, k, buckets, threads, thread_state);
return 0;
}
else if (k > 0 && fs / k >= 2)
{
sa_sint_t alignment = (fs - 1024) / k >= 2 ? 1024 : 16;
sa_sint_t * RESTRICT buckets = (fs - alignment) / k >= 2 ? (sa_sint_t *)libsais_align_up(&SA[n + fs - 2 * k - alignment], (size_t)alignment * sizeof(sa_sint_t)) : &SA[n + fs - 2 * k];
sa_sint_t m = libsais_count_and_gather_lms_suffixes_32s_2k_omp(T, SA, n, k, buckets, threads, thread_state);
if (m > 1)
{
libsais_initialize_buckets_for_lms_suffixes_radix_sort_32s_2k(T, k, buckets, SA[n - m]);
libsais_radix_sort_lms_suffixes_32s_2k_omp(T, SA, n, m, &buckets[1], threads, thread_state);
libsais_place_lms_suffixes_interval_32s_2k(SA, n, k, m - 1, buckets);
libsais_initialize_buckets_start_and_end_32s_2k(k, buckets);
libsais_induce_partial_order_32s_2k_omp(T, SA, n, k, buckets, threads, thread_state);
sa_sint_t names = libsais_renumber_and_mark_distinct_lms_suffixes_32s_1k_omp(T, SA, n, m, threads);
if (names < m)
{
sa_sint_t f = libsais_compact_lms_suffixes_32s_omp(T, SA, n, m, fs, threads, thread_state);
if (libsais_main_32s(SA + n + fs - m + f, SA, m - f, names - f, fs + n - 2 * m + f, threads, thread_state) != 0)
{
return -2;
}
libsais_reconstruct_compacted_lms_suffixes_32s_2k_omp(T, SA, n, k, m, fs, f, buckets, threads, thread_state);
}
else
{
libsais_count_lms_suffixes_32s_2k(T, n, k, buckets);
}
}
else
{
SA[0] = SA[n - 1];
}
libsais_initialize_buckets_end_32s_2k(k, buckets);
libsais_place_lms_suffixes_histogram_32s_2k(SA, n, k, m, buckets);
libsais_initialize_buckets_start_and_end_32s_2k(k, buckets);
libsais_induce_final_order_32s_2k(T, SA, n, k, buckets, threads, thread_state);
return 0;
}
else
{
sa_sint_t * buffer = fs < k ? (sa_sint_t *)libsais_alloc_aligned((size_t)k * sizeof(sa_sint_t), 4096) : (sa_sint_t *)NULL;
sa_sint_t alignment = fs - 1024 >= k ? 1024 : 16;
sa_sint_t * RESTRICT buckets = fs - alignment >= k ? (sa_sint_t *)libsais_align_up(&SA[n + fs - k - alignment], (size_t)alignment * sizeof(sa_sint_t)) : fs >= k ? &SA[n + fs - k] : buffer;
if (buckets == NULL) { return -2; }
memset(SA, 0, (size_t)n * sizeof(sa_sint_t));
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_end_32s_1k(k, buckets);
sa_sint_t m = libsais_radix_sort_lms_suffixes_32s_1k(T, SA, n, buckets);
if (m > 1)
{
libsais_induce_partial_order_32s_1k_omp(T, SA, n, k, buckets, threads, thread_state);
sa_sint_t names = libsais_renumber_and_mark_distinct_lms_suffixes_32s_1k_omp(T, SA, n, m, threads);
if (names < m)
{
if (buffer != NULL) { libsais_free_aligned(buffer); buckets = NULL; }
sa_sint_t f = libsais_compact_lms_suffixes_32s_omp(T, SA, n, m, fs, threads, thread_state);
if (libsais_main_32s(SA + n + fs - m + f, SA, m - f, names - f, fs + n - 2 * m + f, threads, thread_state) != 0)
{
return -2;
}
libsais_reconstruct_compacted_lms_suffixes_32s_1k_omp(T, SA, n, m, fs, f, threads, thread_state);
if (buckets == NULL) { buckets = buffer = (sa_sint_t *)libsais_alloc_aligned((size_t)k * sizeof(sa_sint_t), 4096); }
if (buckets == NULL) { return -2; }
}
libsais_count_suffixes_32s(T, n, k, buckets);
libsais_initialize_buckets_end_32s_1k(k, buckets);
libsais_place_lms_suffixes_interval_32s_1k(T, SA, k, m, buckets);
}
libsais_induce_final_order_32s_1k(T, SA, n, k, buckets, threads, thread_state);
libsais_free_aligned(buffer);
return 0;
}
}
int32_t libsais_main_32s_internal(int32_t * T, int32_t * SA, int32_t n, int32_t k, int32_t fs, int32_t threads)
{
LIBSAIS_THREAD_STATE * RESTRICT thread_state = threads > 1 ? libsais_alloc_thread_state(threads) : NULL;
sa_sint_t index = thread_state != NULL || threads == 1
? libsais_main_32s(T, SA, n, k, fs, threads, thread_state)
: -2;
libsais_free_thread_state(thread_state);
return index;
}
static sa_sint_t libsais_main_8u(const uint8_t * T, sa_sint_t * SA, sa_sint_t n, sa_sint_t * RESTRICT buckets, sa_sint_t bwt, sa_sint_t fs, sa_sint_t threads, LIBSAIS_THREAD_STATE * RESTRICT thread_state)
{
sa_sint_t m = libsais_count_and_gather_lms_suffixes_8u_omp(T, SA, n, buckets, threads, thread_state);
libsais_initialize_buckets_start_and_end_8u(buckets);
if (m > 0)
{
sa_sint_t first_lms_suffix = SA[n - m];
sa_sint_t left_suffixes_count = libsais_initialize_buckets_for_lms_suffixes_radix_sort_8u(T, buckets, first_lms_suffix);
if (threads > 1 && n >= 65536) { memset(SA, 0, ((size_t)n - (size_t)m) * sizeof(sa_sint_t)); }
libsais_radix_sort_lms_suffixes_8u_omp(T, SA, n, m, buckets, threads, thread_state);
if (threads > 1 && n >= 65536) { memset(&SA[(fast_sint_t)n - (fast_sint_t)m], 0, (size_t)m * sizeof(sa_sint_t)); }
libsais_initialize_buckets_for_partial_sorting_8u(T, buckets, first_lms_suffix, left_suffixes_count);
libsais_induce_partial_order_8u_omp(T, SA, n, buckets, first_lms_suffix, left_suffixes_count, threads, thread_state);
sa_sint_t names = libsais_renumber_and_gather_lms_suffixes_8u_omp(SA, n, m, fs, threads, thread_state);
if (names < m)
{
if (libsais_main_32s(SA + n + fs - m, SA, m, names, fs + n - 2 * m, threads, thread_state) != 0)
{
return -2;
}
libsais_gather_lms_suffixes_8u_omp(T, SA, n, threads, thread_state);
libsais_reconstruct_lms_suffixes_omp(SA, n, m, threads);
}
libsais_place_lms_suffixes_interval_8u(SA, n, m, buckets);
}
else
{
memset(SA, 0, (size_t)n * sizeof(sa_sint_t));
}
return libsais_induce_final_order_8u_omp(T, SA, n, bwt, buckets, threads, thread_state);
}
static sa_sint_t libsais_main(const uint8_t * T, sa_sint_t * SA, sa_sint_t n, sa_sint_t bwt, sa_sint_t fs, sa_sint_t threads)
{
LIBSAIS_THREAD_STATE * RESTRICT thread_state = threads > 1 ? libsais_alloc_thread_state(threads) : NULL;
sa_sint_t * RESTRICT buckets = (sa_sint_t *)libsais_alloc_aligned(8 * ALPHABET_SIZE * sizeof(sa_sint_t), 4096);
sa_sint_t index = buckets != NULL && (thread_state != NULL || threads == 1)
? libsais_main_8u(T, SA, n, buckets, bwt, fs, threads, thread_state)
: -2;
libsais_free_aligned(buckets);
libsais_free_thread_state(thread_state);
return index;
}
static sa_sint_t libsais_main_ctx(const LIBSAIS_CONTEXT * ctx, const uint8_t * T, sa_sint_t * SA, sa_sint_t n, sa_sint_t bwt, sa_sint_t fs)
{
return ctx != NULL && (ctx->buckets != NULL && (ctx->thread_state != NULL || ctx->threads == 1))
? libsais_main_8u(T, SA, n, ctx->buckets, bwt, fs, (sa_sint_t)ctx->threads, ctx->thread_state)
: -2;
}
static void libsais_bwt_copy_8u(uint8_t * RESTRICT U, sa_sint_t * RESTRICT A, sa_sint_t n)
{
const fast_sint_t prefetch_distance = 32;
fast_sint_t i, j;
for (i = 0, j = (fast_sint_t)n - 7; i < j; i += 8)
{
libsais_prefetch(&A[i + prefetch_distance]);
U[i + 0] = (uint8_t)A[i + 0];
U[i + 1] = (uint8_t)A[i + 1];
U[i + 2] = (uint8_t)A[i + 2];
U[i + 3] = (uint8_t)A[i + 3];
U[i + 4] = (uint8_t)A[i + 4];
U[i + 5] = (uint8_t)A[i + 5];
U[i + 6] = (uint8_t)A[i + 6];
U[i + 7] = (uint8_t)A[i + 7];
}
for (j += 7; i < j; i += 1)
{
U[i] = (uint8_t)A[i];
}
}
#if defined(_OPENMP)
static void libsais_bwt_copy_8u_omp(uint8_t * RESTRICT U, sa_sint_t * RESTRICT A, sa_sint_t n, sa_sint_t threads)
{
#if defined(_OPENMP)
#pragma omp parallel num_threads(threads) if(threads > 1 && n >= 65536)
#endif
{
#if defined(_OPENMP)
fast_sint_t omp_thread_num = omp_get_thread_num();
fast_sint_t omp_num_threads = omp_get_num_threads();
fast_sint_t omp_block_stride = ((fast_sint_t)n / omp_num_threads) & (-16);
fast_sint_t omp_block_start = omp_thread_num * omp_block_stride;
fast_sint_t omp_block_size = omp_thread_num < omp_num_threads - 1 ? omp_block_stride : (fast_sint_t)n - omp_block_start;
#else
UNUSED(threads);
fast_sint_t omp_block_start = 0;
fast_sint_t omp_block_size = (fast_sint_t)n;
#endif
libsais_bwt_copy_8u(U + omp_block_start, A + omp_block_start, (sa_sint_t)omp_block_size);
}
}
#endif
void * libsais_create_ctx(void)
{
return (void *)libsais_create_ctx_main(1);
}
void libsais_free_ctx(void * ctx)
{
libsais_free_ctx_main((LIBSAIS_CONTEXT *)ctx);
}
int32_t libsais(const uint8_t * T, int32_t * SA, int32_t n, int32_t fs)
{
if ((T == NULL) || (SA == NULL) || (n < 0) || (fs < 0))
{
return -1;
}
else if (n < 2)
{
if (n == 1) { SA[0] = 0; }
return 0;
}
return libsais_main(T, SA, n, 0, fs, 1);
}
int32_t libsais_ctx(const void * ctx, const uint8_t * T, int32_t * SA, int32_t n, int32_t fs)
{
if ((ctx == NULL) || (T == NULL) || (SA == NULL) || (n < 0) || (fs < 0))
{
return -1;
}
else if (n < 2)
{
if (n == 1) { SA[0] = 0; }
return 0;
}
return libsais_main_ctx((const LIBSAIS_CONTEXT *)ctx, T, SA, n, 0, fs);
}
int32_t libsais_bwt(const uint8_t * T, uint8_t * U, int32_t * A, int32_t n, int32_t fs)
{
if ((T == NULL) || (U == NULL) || (A == NULL) || (n < 0) || (fs < 0))
{
return -1;
}
else if (n <= 1)
{
if (n == 1) { U[0] = T[0]; }
return n;
}
sa_sint_t index = libsais_main(T, A, n, 1, fs, 1);
if (index >= 0)
{
U[0] = T[n - 1];
libsais_bwt_copy_8u(U + 1, A, index);
libsais_bwt_copy_8u(U + 1 + index, A + 1 + index, n - index - 1);
index++;
}
return index;
}
int32_t libsais_bwt_ctx(const void * ctx, const uint8_t * T, uint8_t * U, int32_t * A, int32_t n, int32_t fs)
{
if ((ctx == NULL) || (T == NULL) || (U == NULL) || (A == NULL) || (n < 0) || (fs < 0))
{
return -1;
}
else if (n <= 1)
{
if (n == 1) { U[0] = T[0]; }
return n;
}
sa_sint_t index = libsais_main_ctx((const LIBSAIS_CONTEXT *)ctx, T, A, n, 1, fs);
if (index >= 0)
{
U[0] = T[n - 1];
#if defined(_OPENMP)
libsais_bwt_copy_8u_omp(U + 1, A, index, (sa_sint_t)((const LIBSAIS_CONTEXT *)ctx)->threads);
libsais_bwt_copy_8u_omp(U + 1 + index, A + 1 + index, n - index - 1, (sa_sint_t)((const LIBSAIS_CONTEXT *)ctx)->threads);
#else
libsais_bwt_copy_8u(U + 1, A, index);
libsais_bwt_copy_8u(U + 1 + index, A + 1 + index, n - index - 1);
#endif
index++;
}
return index;
}
#if defined(_OPENMP)
void * libsais_create_ctx_omp(int32_t threads)
{
if (threads < 0) { return NULL; }
threads = threads > 0 ? threads : omp_get_max_threads();
return (void *)libsais_create_ctx_main(threads);
}
int32_t libsais_omp(const uint8_t * T, int32_t * SA, int32_t n, int32_t fs, int32_t threads)
{
if ((T == NULL) || (SA == NULL) || (n < 0) || (threads < 0) || (fs < 0))
{
return -1;
}
else if (n < 2)
{
if (n == 1) { SA[0] = 0; }
return 0;
}
threads = threads > 0 ? threads : omp_get_max_threads();
return libsais_main(T, SA, n, 0, fs, threads);
}
int32_t libsais_bwt_omp(const uint8_t * T, uint8_t * U, int32_t * A, int32_t n, int32_t fs, int32_t threads)
{
if ((T == NULL) || (U == NULL) || (A == NULL) || (n < 0) || (threads < 0) || (fs < 0))
{
return -1;
}
else if (n <= 1)
{
if (n == 1) { U[0] = T[0]; }
return n;
}
threads = threads > 0 ? threads : omp_get_max_threads();
sa_sint_t index = libsais_main(T, A, n, 1, fs, threads);
if (index >= 0)
{
U[0] = T[n - 1];
libsais_bwt_copy_8u_omp(U + 1, A, index, threads);
libsais_bwt_copy_8u_omp(U + 1 + index, A + 1 + index, n - index - 1, threads);
index++;
}
return index;
}
#endif
|
GB_binop__band_int16.c | //------------------------------------------------------------------------------
// GB_binop: hard-coded functions for each built-in binary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated/ folder, do not edit it (auto-generated).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_ek_slice.h"
#include "GB_dense.h"
#include "GB_atomics.h"
#include "GB_bitmap_assign_methods.h"
#include "GB_binop__include.h"
// C=binop(A,B) is defined by the following types and operators:
// A+B function (eWiseAdd): GB_AaddB__band_int16
// A.*B function (eWiseMult): GB_AemultB__band_int16
// A*D function (colscale): (none)
// D*A function (rowscale): (node)
// C+=B function (dense accum): GB_Cdense_accumB__band_int16
// C+=b function (dense accum): GB_Cdense_accumb__band_int16
// C+=A+B function (dense ewise3): (none)
// C=A+B function (dense ewise3): GB_Cdense_ewise3_noaccum__band_int16
// C=scalar+B GB_bind1st__band_int16
// C=scalar+B' GB_bind1st_tran__band_int16
// C=A+scalar GB_bind2nd__band_int16
// C=A'+scalar GB_bind2nd_tran__band_int16
// C type: int16_t
// A type: int16_t
// B,b type: int16_t
// BinaryOp: cij = (aij) & (bij)
#define GB_ATYPE \
int16_t
#define GB_BTYPE \
int16_t
#define GB_CTYPE \
int16_t
// true if the types of A and B are identical
#define GB_ATYPE_IS_BTYPE \
1
// true if the types of C and A are identical
#define GB_CTYPE_IS_ATYPE \
1
// true if the types of C and B are identical
#define GB_CTYPE_IS_BTYPE \
1
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int16_t aij = Ax [pA]
// bij = Bx [pB]
#define GB_GETB(bij,Bx,pB) \
int16_t bij = Bx [pB]
// declare scalar of the same type as C
#define GB_CTYPE_SCALAR(t) \
int16_t t
// cij = Ax [pA]
#define GB_COPY_A_TO_C(cij,Ax,pA) \
cij = Ax [pA]
// cij = Bx [pB]
#define GB_COPY_B_TO_C(cij,Bx,pB) \
cij = Bx [pB]
#define GB_CX(p) Cx [p]
// binary operator
#define GB_BINOP(z, x, y, i, j) \
z = (x) & (y) ;
// op is second
#define GB_OP_IS_SECOND \
0
// op is plus_fp32 or plus_fp64
#define GB_OP_IS_PLUS_REAL \
0
// op is minus_fp32 or minus_fp64
#define GB_OP_IS_MINUS_REAL \
0
// GB_cblas_*axpy gateway routine, if it exists for this operator and type:
#define GB_CBLAS_AXPY \
(none)
// do the numerical phases of GB_add and GB_emult
#define GB_PHASE_2_OF_2
// hard-coded loops can be vectorized
#define GB_PRAGMA_SIMD_VECTORIZE GB_PRAGMA_SIMD
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_BAND || GxB_NO_INT16 || GxB_NO_BAND_INT16)
//------------------------------------------------------------------------------
// C += A+B, all 3 matrices dense
//------------------------------------------------------------------------------
#if 0
// The op must be MIN, MAX, PLUS, MINUS, RMINUS, TIMES, DIV, or RDIV.
void (none)
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#include "GB_dense_ewise3_accum_template.c"
}
#endif
//------------------------------------------------------------------------------
// C = A+B, all 3 matrices dense
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_ewise3_noaccum__band_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GrB_Matrix B,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_dense_ewise3_noaccum_template.c"
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += B, accumulate a sparse matrix into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumB__band_int16
(
GrB_Matrix C,
const GrB_Matrix B,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
#include "GB_dense_subassign_23_template.c"
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C += b, accumulate a scalar into a dense matrix
//------------------------------------------------------------------------------
GrB_Info GB_Cdense_accumb__band_int16
(
GrB_Matrix C,
const GB_void *p_bwork,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
{
// get the scalar b for C += b, of type int16_t
int16_t bwork = (*((int16_t *) p_bwork)) ;
#include "GB_dense_subassign_22_template.c"
return (GrB_SUCCESS) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = A*D, column scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info (none)
(
GrB_Matrix C,
const GrB_Matrix A, bool A_is_pattern,
const GrB_Matrix D, bool D_is_pattern,
const int64_t *GB_RESTRICT kfirst_slice,
const int64_t *GB_RESTRICT klast_slice,
const int64_t *GB_RESTRICT pstart_slice,
const int ntasks,
const int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_colscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// C = D*B, row scale with diagonal D matrix
//------------------------------------------------------------------------------
#if 0
GrB_Info (node)
(
GrB_Matrix C,
const GrB_Matrix D, bool D_is_pattern,
const GrB_Matrix B, bool B_is_pattern,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *GB_RESTRICT Cx = (int16_t *) C->x ;
#include "GB_AxB_rowscale_meta.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
//------------------------------------------------------------------------------
// eWiseAdd: C = A+B or C<M> = A+B
//------------------------------------------------------------------------------
#undef GB_FREE_ALL
#define GB_FREE_ALL \
{ \
GB_ek_slice_free (&pstart_Mslice, &kfirst_Mslice, &klast_Mslice) ; \
GB_ek_slice_free (&pstart_Aslice, &kfirst_Aslice, &klast_Aslice) ; \
GB_ek_slice_free (&pstart_Bslice, &kfirst_Bslice, &klast_Bslice) ; \
}
GrB_Info GB_AaddB__band_int16
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const bool Ch_is_Mh,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *pstart_Mslice = NULL, *kfirst_Mslice = NULL, *klast_Mslice = NULL ;
int64_t *pstart_Aslice = NULL, *kfirst_Aslice = NULL, *klast_Aslice = NULL ;
int64_t *pstart_Bslice = NULL, *kfirst_Bslice = NULL, *klast_Bslice = NULL ;
#include "GB_add_template.c"
GB_FREE_ALL ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// eWiseMult: C = A.*B or C<M> = A.*B
//------------------------------------------------------------------------------
GrB_Info GB_AemultB__band_int16
(
GrB_Matrix C,
const int C_sparsity,
const GrB_Matrix M,
const bool Mask_struct,
const bool Mask_comp,
const GrB_Matrix A,
const GrB_Matrix B,
const int64_t *GB_RESTRICT C_to_M,
const int64_t *GB_RESTRICT C_to_A,
const int64_t *GB_RESTRICT C_to_B,
const GB_task_struct *GB_RESTRICT TaskList,
const int C_ntasks,
const int C_nthreads,
GB_Context Context
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t *pstart_Mslice = NULL, *kfirst_Mslice = NULL, *klast_Mslice = NULL ;
int64_t *pstart_Aslice = NULL, *kfirst_Aslice = NULL, *klast_Aslice = NULL ;
int64_t *pstart_Bslice = NULL, *kfirst_Bslice = NULL, *klast_Bslice = NULL ;
#include "GB_emult_template.c"
GB_FREE_ALL ;
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (x,Bx): apply a binary operator to a matrix with scalar bind1st
//------------------------------------------------------------------------------
GrB_Info GB_bind1st__band_int16
(
GB_void *Cx_output, // Cx and Bx may be aliased
const GB_void *x_input,
const GB_void *Bx_input,
const int8_t *GB_RESTRICT Bb,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t *Cx = (int16_t *) Cx_output ;
int16_t x = (*((int16_t *) x_input)) ;
int16_t *Bx = (int16_t *) Bx_input ;
int64_t p ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Bb, p)) continue ;
int16_t bij = Bx [p] ;
Cx [p] = (x) & (bij) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// Cx = op (Ax,y): apply a binary operator to a matrix with scalar bind2nd
//------------------------------------------------------------------------------
GrB_Info GB_bind2nd__band_int16
(
GB_void *Cx_output, // Cx and Ax may be aliased
const GB_void *Ax_input,
const GB_void *y_input,
const int8_t *GB_RESTRICT Ab,
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
int16_t *Cx = (int16_t *) Cx_output ;
int16_t *Ax = (int16_t *) Ax_input ;
int16_t y = (*((int16_t *) y_input)) ;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!GBB (Ab, p)) continue ;
int16_t aij = Ax [p] ;
Cx [p] = (aij) & (y) ;
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (x, A'): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (x, aij), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (x) & (aij) ; \
}
GrB_Info GB_bind1st_tran__band_int16
(
GrB_Matrix C,
const GB_void *x_input,
const GrB_Matrix A,
int64_t *GB_RESTRICT *Workspaces,
const int64_t *GB_RESTRICT A_slice,
int nworkspaces,
int nthreads
)
{
// GB_unop_transpose.c uses GB_ATYPE, but A is
// the 2nd input to binary operator z=f(x,y).
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t x = (*((const int16_t *) x_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
#undef GB_ATYPE
#define GB_ATYPE \
int16_t
}
//------------------------------------------------------------------------------
// C = op (A', y): transpose and apply a binary operator
//------------------------------------------------------------------------------
// cij = op (aij, y), no typecasting (in spite of the macro name)
#undef GB_CAST_OP
#define GB_CAST_OP(pC,pA) \
{ \
int16_t aij = Ax [pA] ; \
Cx [pC] = (aij) & (y) ; \
}
GrB_Info GB_bind2nd_tran__band_int16
(
GrB_Matrix C,
const GrB_Matrix A,
const GB_void *y_input,
int64_t *GB_RESTRICT *Workspaces,
const int64_t *GB_RESTRICT A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int16_t y = (*((const int16_t *) y_input)) ;
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
H2P_build_H2_UJ_proxy_levelup.c |
// Build H2 projection matrices using proxy points, level by level
void H2P_build_H2_UJ_proxy(H2Pack_p h2pack)
{
int pt_dim = h2pack->pt_dim;
int xpt_dim = h2pack->xpt_dim;
int krnl_dim = h2pack->krnl_dim;
int n_node = h2pack->n_node;
int n_leaf_node = h2pack->n_leaf_node;
int n_point = h2pack->n_point;
int n_thread = h2pack->n_thread;
int max_child = h2pack->max_child;
int max_level = h2pack->max_level;
int min_adm_level = h2pack->min_adm_level;
int stop_type = h2pack->QR_stop_type;
int *children = h2pack->children;
int *n_child = h2pack->n_child;
int *node_level = h2pack->node_level;
int *node_height = h2pack->node_height;
int *level_n_node = h2pack->level_n_node;
int *level_nodes = h2pack->level_nodes;
int *leaf_nodes = h2pack->height_nodes;
int *pt_cluster = h2pack->pt_cluster;
DTYPE *coord = h2pack->coord;
DTYPE *enbox = h2pack->enbox;
size_t *mat_size = h2pack->mat_size;
void *krnl_param = h2pack->krnl_param;
H2P_dense_mat_p *pp = h2pack->pp;
H2P_thread_buf_p *thread_buf = h2pack->tb;
kernel_eval_fptr krnl_eval = h2pack->krnl_eval;
void *stop_param = NULL;
if (stop_type == QR_RANK)
stop_param = &h2pack->QR_stop_rank;
if ((stop_type == QR_REL_NRM) || (stop_type == QR_ABS_NRM))
stop_param = &h2pack->QR_stop_tol;
// 1. Allocate U and J
h2pack->n_UJ = n_node;
h2pack->U = (H2P_dense_mat_p*) malloc(sizeof(H2P_dense_mat_p) * n_node);
h2pack->J = (H2P_int_vec_p*) malloc(sizeof(H2P_int_vec_p) * n_node);
h2pack->J_coord = (H2P_dense_mat_p*) malloc(sizeof(H2P_dense_mat_p) * n_node);
ASSERT_PRINTF(h2pack->U != NULL, "Failed to allocate %d U matrices\n", n_node);
ASSERT_PRINTF(h2pack->J != NULL, "Failed to allocate %d J matrices\n", n_node);
ASSERT_PRINTF(h2pack->J_coord != NULL, "Failed to allocate %d J_coord auxiliary matrices\n", n_node);
for (int i = 0; i < h2pack->n_UJ; i++)
{
h2pack->U[i] = NULL;
h2pack->J[i] = NULL;
h2pack->J_coord[i] = NULL;
}
H2P_dense_mat_p *U = h2pack->U;
H2P_int_vec_p *J = h2pack->J;
H2P_dense_mat_p *J_coord = h2pack->J_coord;
double *U_timers = (double*) malloc(sizeof(double) * n_thread * 8);
// 2. Initialize the row indices for leaf nodes: all points in that box
for (int j = 0; j < n_leaf_node; j++)
{
int node = leaf_nodes[j];
int pt_s = pt_cluster[node * 2];
int pt_e = pt_cluster[node * 2 + 1];
int node_npt = pt_e - pt_s + 1;
H2P_int_vec_init(&J[node], node_npt);
for (int k = 0; k < node_npt; k++)
J[node]->data[k] = pt_s + k;
J[node]->length = node_npt;
H2P_dense_mat_init(&J_coord[node], xpt_dim, node_npt);
copy_matrix_block(sizeof(DTYPE), xpt_dim, node_npt, coord + pt_s, n_point, J_coord[node]->data, node_npt);
}
// 3. Hierarchical construction level by level. min_adm_level is the
// highest level that still has admissible blocks.
for (int i = max_level; i >= min_adm_level; i--)
{
int *level_i_nodes = level_nodes + i * n_leaf_node;
int level_i_n_node = level_n_node[i];
int n_thread_i = MIN(level_i_n_node, n_thread);
int level = i;
// (1) Update row indices associated with clusters at level i
#pragma omp parallel num_threads(n_thread_i)
{
int tid = omp_get_thread_num();
thread_buf[tid]->timer = -get_wtime_sec();
#pragma omp for schedule(dynamic) nowait
for (int j = 0; j < level_i_n_node; j++)
{
int node = level_i_nodes[j];
int n_child_node = n_child[node];
if (n_child_node == 0) continue; // J[node] has already been prepared for leaf node
int *child_nodes = children + node * max_child;
int J_child_size = 0;
for (int i_child = 0; i_child < n_child_node; i_child++)
{
int i_child_node = child_nodes[i_child];
J_child_size += J[i_child_node]->length;
}
H2P_int_vec_init(&J[node], J_child_size);
for (int i_child = 0; i_child < n_child_node; i_child++)
{
int i_child_node = child_nodes[i_child];
H2P_int_vec_concatenate(J[node], J[i_child_node]);
}
}
thread_buf[tid]->timer += get_wtime_sec();
}
#pragma omp parallel num_threads(n_thread_i)
{
int tid = omp_get_thread_num();
H2P_int_vec_p inadm_skel_idx = thread_buf[tid]->idx0;
H2P_int_vec_p sub_idx = thread_buf[tid]->idx0;
H2P_int_vec_p ID_buff = thread_buf[tid]->idx1;
H2P_dense_mat_p node_skel_coord = thread_buf[tid]->mat0;
H2P_dense_mat_p inadm_skel_coord = thread_buf[tid]->mat1;
H2P_dense_mat_p A_block = thread_buf[tid]->mat2;
H2P_dense_mat_p QR_buff = thread_buf[tid]->mat1;
double st, et, krnl_t = 0.0, QR_t = 0.0, other_t = 0.0;
thread_buf[tid]->timer -= get_wtime_sec();
#pragma omp for schedule(dynamic) nowait
for (int j = 0; j < level_i_n_node; j++)
{
int node = level_i_nodes[j];
int height = node_height[node];
// (2) Gather current node's skeleton points (== all children nodes' skeleton points)
st = get_wtime_sec();
H2P_dense_mat_resize(node_skel_coord, xpt_dim, J[node]->length);
if (height == 0)
{
node_skel_coord = J_coord[node];
} else {
int n_child_node = n_child[node];
int *child_nodes = children + node * max_child;
int J_child_size = 0;
for (int i_child = 0; i_child < n_child_node; i_child++)
{
int i_child_node = child_nodes[i_child];
int src_ld = J_coord[i_child_node]->ncol;
int dst_ld = node_skel_coord->ncol;
DTYPE *src_mat = J_coord[i_child_node]->data;
DTYPE *dst_mat = node_skel_coord->data + J_child_size;
copy_matrix_block(sizeof(DTYPE), xpt_dim, src_ld, src_mat, src_ld, dst_mat, dst_ld);
J_child_size += J[i_child_node]->length;
}
} // End of "if (level == 0)"
et = get_wtime_sec();
other_t += et - st;
// (3) Shift current node's skeleton points so their center is at the original point
st = get_wtime_sec();
DTYPE *node_box = enbox + node * 2 * pt_dim;
int node_skel_npt = J[node]->length;
int node_pp_npt = pp[level]->ncol;
for (int k = 0; k < pt_dim; k++)
{
DTYPE box_center_k = node_box[k] + 0.5 * node_box[pt_dim + k];
DTYPE *node_skel_coord_k = node_skel_coord->data + k * node_skel_npt;
#pragma omp simd
for (int l = 0; l < node_skel_npt; l++)
node_skel_coord_k[l] -= box_center_k;
}
et = get_wtime_sec();
other_t += et - st;
// (4) Build the kernel matrix block
st = get_wtime_sec();
int A_blk_nrow = node_skel_npt * krnl_dim;
int A_blk_ncol = node_pp_npt * krnl_dim;
H2P_dense_mat_resize(A_block, A_blk_nrow, A_blk_ncol);
krnl_eval(
node_skel_coord->data, node_skel_npt, node_skel_npt,
pp[level]->data, node_pp_npt, node_pp_npt,
krnl_param, A_block->data, A_block->ld
);
et = get_wtime_sec();
krnl_t += et - st;
// (5) ID compress
// Note: A is transposed in ID compress, be careful when calculating the buffer size
st = get_wtime_sec();
if (krnl_dim == 1)
{
H2P_dense_mat_resize(QR_buff, A_block->nrow, 1);
} else {
int QR_buff_size = (2 * krnl_dim + 2) * A_block->ncol + (krnl_dim + 1) * A_block->nrow;
H2P_dense_mat_resize(QR_buff, QR_buff_size, 1);
}
H2P_int_vec_set_capacity(ID_buff, 4 * A_block->nrow);
H2P_ID_compress(
A_block, stop_type, stop_param, &U[node], sub_idx,
1, QR_buff->data, ID_buff->data, krnl_dim
);
et = get_wtime_sec();
QR_t += et - st;
// (6) Choose the skeleton points of this node
st = get_wtime_sec();
for (int k = 0; k < sub_idx->length; k++)
J[node]->data[k] = J[node]->data[sub_idx->data[k]];
J[node]->length = sub_idx->length;
H2P_dense_mat_init(&J_coord[node], xpt_dim, sub_idx->length);
H2P_gather_matrix_columns(
coord, n_point, J_coord[node]->data, J[node]->length,
xpt_dim, J[node]->data, J[node]->length
);
et = get_wtime_sec();
other_t += et - st;
} // End of j loop
thread_buf[tid]->timer += get_wtime_sec();
double *timers = U_timers + tid * 8;
timers[U_BUILD_KRNL_TIMER_IDX] = krnl_t;
timers[U_BUILD_QR_TIMER_IDX] = QR_t;
timers[U_BUILD_OTHER_TIMER_IDX] = other_t;
} // End of "pragma omp parallel"
if (h2pack->print_timers == 1)
{
double max_t = 0.0, avg_t = 0.0, min_t = 19241112.0;
for (int i = 0; i < n_thread_i; i++)
{
double thread_i_timer = thread_buf[i]->timer;
avg_t += thread_i_timer;
max_t = MAX(max_t, thread_i_timer);
min_t = MIN(min_t, thread_i_timer);
}
avg_t /= (double) n_thread_i;
INFO_PRINTF("Build U: level %d, %d/%d threads, %d nodes\n", i, n_thread_i, n_thread, level_n_node[i]);
INFO_PRINTF(" min/avg/max thread wall-time = %.3lf, %.3lf, %.3lf (s)\n", min_t, avg_t, max_t);
INFO_PRINTF("Build U subroutine time consumption:\n");
INFO_PRINTF(" tid, kernel evaluation, ID compress, misc, total\n");
for (int tid = 0; tid < n_thread_i; tid++)
{
double *timers = U_timers + 8 * tid;
INFO_PRINTF(
" %3d, %6.3lf, %6.3lf, %6.3lf, %6.3lf\n",
tid, timers[U_BUILD_KRNL_TIMER_IDX], timers[U_BUILD_QR_TIMER_IDX],
timers[U_BUILD_OTHER_TIMER_IDX], thread_buf[tid]->timer
);
}
} // End of "if (h2pack->print_timers == 1)"
} // End of i loop
// 3. Initialize other not touched U J & add statistic info
for (int i = 0; i < h2pack->n_UJ; i++)
{
if (U[i] == NULL)
{
H2P_dense_mat_init(&U[i], 1, 1);
U[i]->nrow = 0;
U[i]->ncol = 0;
U[i]->ld = 0;
} else {
mat_size[_U_SIZE_IDX] += U[i]->nrow * U[i]->ncol;
mat_size[MV_FWD_SIZE_IDX] += U[i]->nrow * U[i]->ncol;
mat_size[MV_FWD_SIZE_IDX] += U[i]->nrow + U[i]->ncol;
mat_size[MV_BWD_SIZE_IDX] += U[i]->nrow * U[i]->ncol;
mat_size[MV_BWD_SIZE_IDX] += U[i]->nrow + U[i]->ncol;
}
if (J[i] == NULL) H2P_int_vec_init(&J[i], 1);
if (J_coord[i] == NULL)
{
H2P_dense_mat_init(&J_coord[i], 1, 1);
J_coord[i]->nrow = 0;
J_coord[i]->ncol = 0;
J_coord[i]->ld = 0;
}
//printf("Node %3d: %d skeleton points\n", i, J[i]->length);
}
free(U_timers);
for (int i = 0; i < n_thread; i++)
H2P_thread_buf_reset(thread_buf[i]);
BLAS_SET_NUM_THREADS(n_thread);
}
|
comm.h | /*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
/**
*/
#ifndef MXNET_KVSTORE_COMM_H_
#define MXNET_KVSTORE_COMM_H_
#include <dmlc/omp.h>
#include <string>
#include <algorithm>
#include <utility>
#include <limits>
#include <vector>
#include <tuple>
#include <thread>
#include "mxnet/ndarray.h"
#include "../ndarray/ndarray_function.h"
#include "../operator/tensor/sparse_retain-inl.h"
namespace mxnet {
namespace kvstore {
/**
* \brief multiple device commmunication
*/
class Comm {
public:
Comm() {
pinned_ctx_ = Context::CPUPinned(0);
}
virtual ~Comm() { }
/**
* \brief init key with the data shape and storage shape
*/
virtual void Init(int key, const NDArrayStorageType stype,
const TShape& shape, int dtype = mshadow::kFloat32) = 0;
/**
* \brief returns src[0] + .. + src[src.size()-1]
*/
virtual const NDArray& Reduce(
int key, const std::vector<NDArray>& src, int priority) = 0;
/**
* \brief copy from src to dst[i] for every i
*/
virtual void Broadcast(
int key, const NDArray& src,
const std::vector<NDArray*> dst, int priority) = 0;
/**
* \brief broadcast src to dst[i] with target row_ids for every i
* \param dst a list of destination row_sparse NDArray and its target row_ids to broadcast,
where the row_ids are expected to be unique and sorted
* \param use_copy if set to true, directly copy src to dst[i] without looking up the
provided row_ids
*/
virtual void BroadcastRowSparse(int key, const NDArray& src,
const std::vector<std::pair<NDArray*, NDArray>>& dst,
const bool use_copy,
const int priority) = 0;
/**
* \brief return a pinned contex
*/
Context pinned_ctx() const {
return pinned_ctx_;
}
protected:
Context pinned_ctx_;
};
/**
* \brief an implemention of Comm that first copy data to CPU memeory, and then
* reduce there
*/
class CommCPU : public Comm {
public:
CommCPU() {
nthread_reduction_ = dmlc::GetEnv("MXNET_KVSTORE_REDUCTION_NTHREADS", 4);
bigarray_bound_ = dmlc::GetEnv("MXNET_KVSTORE_BIGARRAY_BOUND", 1000 * 1000);
// TODO(junwu) delete the following data member, now for benchmark only
is_serial_push_ = dmlc::GetEnv("MXNET_KVSTORE_SERIAL_PUSH", 0);
}
virtual ~CommCPU() { }
void Init(int key, const NDArrayStorageType stype, const TShape& shape,
int type = mshadow::kFloat32) override {
if (stype == kDefaultStorage) {
merge_buf_[key].merged = NDArray(shape, pinned_ctx_, false, type);
} else {
merge_buf_[key].merged = NDArray(stype, shape, pinned_ctx_, true, type);
}
}
const NDArray& Reduce(int key, const std::vector<NDArray>& src,
int priority) override {
auto& buf = merge_buf_[key];
// avoid extra copy for single device, but it may bring problems for
// abnormal usage of kvstore
if (src.size() == 1) {
if (src[0].storage_type() == kDefaultStorage) {
return src[0];
} else { // if sparse and only one GPU, always update weight on CPU
CopyFromTo(src[0], &buf.merged, priority);
return buf.merged;
}
}
if (buf.merged.storage_type() == kDefaultStorage) {
std::vector<Engine::VarHandle> const_vars(src.size() - 1);
std::vector<NDArray> reduce(src.size());
CopyFromTo(src[0], &buf.merged, priority);
reduce[0] = buf.merged;
if (buf.copy_buf.empty()) {
buf.copy_buf.resize(src.size()-1);
for (size_t j = 0; j < src.size() - 1; ++j) {
// allocate NDArray basd on storage type
buf.copy_buf[j] = NDArray(
src[0].shape(), pinned_ctx_, false, src[0].dtype());
}
}
for (size_t i = 1; i < src.size(); ++i) {
CopyFromTo(src[i], &(buf.copy_buf[i-1]), priority);
reduce[i] = buf.copy_buf[i-1];
const_vars[i-1] = reduce[i].var();
}
Engine::Get()->PushSync([reduce, this](RunContext rctx) {
ReduceSumCPU(reduce);
}, Context::CPU(), const_vars, {reduce[0].var()},
FnProperty::kCPUPrioritized, priority, PROFILER_MESSAGE("KVStoreReduce"));
} else {
// buf.merged is a sparse ndarray.
std::vector<Engine::VarHandle> const_vars(src.size());
std::vector<NDArray> reduce(src.size());
if (buf.copy_buf.empty()) {
buf.copy_buf.resize(src.size());
for (size_t j = 0; j < src.size(); ++j) {
buf.copy_buf[j] = NDArray(
src[0].storage_type(), src[0].shape(), pinned_ctx_, true, src[0].dtype());
}
}
for (size_t i = 0; i < src.size(); ++i) {
CopyFromTo(src[i], &(buf.copy_buf[i]), priority);
reduce[i] = buf.copy_buf[i];
const_vars[i] = reduce[i].var();
}
auto result = buf.merged;
Engine::Get()->PushSync([reduce, result, this](RunContext rctx) {
NDArray out = result;
Resource rsc = ResourceManager::Get()->Request(rctx.ctx,
ResourceRequest(ResourceRequest::kTempSpace));
is_serial_push_?
ReduceSumCPUExSerial(reduce, &out)
: mxnet::ndarray::ElementwiseSum(rctx.get_stream<cpu>(), rsc, reduce, &out);
}, Context::CPU(), const_vars, {result.var()},
FnProperty::kCPUPrioritized, priority, PROFILER_MESSAGE("KVStoreReduce"));
}
return buf.merged;
}
void Broadcast(int key, const NDArray& src,
const std::vector<NDArray*> dst, int priority) override {
int mask = src.ctx().dev_mask();
if (mask == Context::kCPU) {
for (auto d : dst) CopyFromTo(src, d, priority);
} else {
// first copy data to cpu, then broadcast
auto& buf = merge_buf_[key];
CopyFromTo(src, &buf.merged, priority);
for (auto d : dst) CopyFromTo(buf.merged, d, priority);
}
}
void BroadcastRowSparse(int key, const NDArray& src,
const std::vector<std::pair<NDArray*, NDArray>>& dst,
const bool use_copy,
const int priority) override {
using namespace mshadow;
CHECK_EQ(src.storage_type(), kRowSparseStorage)
<< "BroadcastRowSparse expects row-sparse src NDArray";
CHECK_EQ(src.ctx().dev_mask(), Context::kCPU)
<< "BroadcastRowSparse with src on gpu context not supported";
for (size_t i = 0; i < dst.size(); ++i) {
NDArray* out = dst[i].first;
NDArray row_id = dst[i].second;
if (use_copy) {
CopyFromTo(src, out, priority);
} else {
CHECK_EQ(out->storage_type(), kRowSparseStorage)
<< "BroadcastRowSparse expects row_sparse dst NDArray";
CHECK_EQ(row_id.ctx().dev_mask(), Context::kCPU)
<< "BroadcastRowSparse with row_indices on gpu context not supported";
// retain according to unique indices
const bool use_sparse_retain = (src.shape()[0] != src.storage_shape()[0])
|| (row_id.dtype() != out->aux_type(rowsparse::kIdx))
|| (out->ctx().dev_mask() != Context::kGPU);
if (use_sparse_retain) { // use sparse_retain op
const bool is_to_gpu = out->ctx().dev_mask() == Context::kGPU;
NDArray out_cpu = is_to_gpu? NDArray(kRowSparseStorage, src.shape(),
src.ctx(), true, src.dtype(), src.aux_types()) : *out;
Engine::Get()->PushSync([=](RunContext rctx) {
const TBlob& indices = row_id.data();
NDArray temp = out_cpu; // get rid of const qualifier
op::SparseRetainOpForwardRspImpl<cpu>(rctx.get_stream<cpu>(),
src, indices, kWriteTo,
&temp);
}, Context::CPU(), {src.var(), row_id.var()}, {out_cpu.var()},
FnProperty::kNormal, priority, PROFILER_MESSAGE("KVStoreSparseRetain"));
if (is_to_gpu) {
CopyFromTo(out_cpu, out, priority);
}
} else { // direct copy rows
Engine::Get()->PushSync([=](RunContext rctx) {
CopyRetainedRowsToGPU(rctx.get_stream<cpu>(), rctx.get_stream<gpu>(),
src, row_id, out);
}, out->ctx(), {src.var(), row_id.var()}, {out->var()},
FnProperty::kCopyToGPU, priority, PROFILER_MESSAGE("KVStoreCopyRetainedRowsToGPU"));
}
}
}
}
private:
/*!
* \brief When src is a rsp with full rows,
* simply copy retained rows directly from cpu to gpu
* without invoking sparse_retain op.
*/
void CopyRetainedRowsToGPU(mshadow::Stream<cpu>* cpu_stream,
mshadow::Stream<gpu>* gpu_stream,
const NDArray& src,
const NDArray& indices,
NDArray* dst) {
#if MXNET_USE_CUDA == 1
CHECK_EQ(src.storage_type(), kRowSparseStorage)
<< "CopyRetainedRowsToGPU expects row-sparse src NDArray";
CHECK_EQ(src.ctx().dev_mask(), Context::kCPU)
<< "CopyRetainedRowsToGPU with src on gpu context not supported";
CHECK_EQ(src.storage_shape()[0], src.shape()[0])
<< "CopyRetainedRowsToGPU only supports src rsp with full rows";
CHECK_EQ(indices.storage_type(), kDefaultStorage);
CHECK_EQ(indices.ctx().dev_mask(), Context::kCPU);
CHECK_EQ(dst->storage_type(), kRowSparseStorage);
CHECK_EQ(dst->ctx().dev_mask(), Context::kGPU);
CHECK_EQ(indices.dtype(), dst->aux_type(rowsparse::kIdx))
<< "CopyRetainedRowsToGPU only supports same data type for idx array and dst aux_data(0)";
if (!src.storage_initialized() || indices.data().Size() == 0U) {
op::FillZerosRspImpl(gpu_stream, dst);
return;
}
using namespace mshadow;
const TBlob& src_data = src.data();
const TBlob& idx_data = indices.data();
const size_t row_length = src.shape().ProdShape(1, src.shape().ndim());
const size_t num_rows_retained = idx_data.Size();
dst->CheckAndAlloc({Shape1(num_rows_retained)});
TBlob dst_data = dst->data();
TBlob dst_idx_data = dst->aux_data(rowsparse::kIdx);
MSHADOW_TYPE_SWITCH(src.dtype(), DType, {
MSHADOW_IDX_TYPE_SWITCH(indices.dtype(), IType, {
// copy idx array
Tensor<gpu, 1, IType> dst_idx_tensor = dst_idx_data.FlatTo1D<gpu, IType>(gpu_stream);
const Tensor<cpu, 1, IType> idx_tensor = idx_data.FlatTo1D<cpu, IType>(cpu_stream);
Copy(dst_idx_tensor, idx_tensor, gpu_stream);
// copy src data
const Tensor<cpu, 2, DType> src_data_tensor = src_data.get_with_shape<cpu, 2, DType>(
Shape2(src_data.shape_[0], row_length), cpu_stream);
Tensor<gpu, 2, DType> dst_data_tensor = dst_data.get_with_shape<gpu, 2, DType>(
Shape2(dst_data.shape_[0], row_length), gpu_stream);
for (size_t i = 0; i < num_rows_retained; ++i) {
Copy(dst_data_tensor[i], src_data_tensor[idx_tensor[i]], gpu_stream);
}
})
})
#else
LOG(FATAL) << "GPU not enabled";
#endif
}
// reduce sum into val[0]
inline void ReduceSumCPU(const std::vector<NDArray> &in_data) {
MSHADOW_TYPE_SWITCH(in_data[0].dtype(), DType, {
std::vector<DType*> dptr(in_data.size());
for (size_t i = 0; i < in_data.size(); ++i) {
TBlob data = in_data[i].data();
CHECK(data.CheckContiguous());
dptr[i] = data.FlatTo2D<cpu, DType>().dptr_;
}
size_t total = in_data[0].shape().Size();
ReduceSumCPUImpl(dptr, total);
});
}
// serial implementation of reduce sum for row sparse NDArray.
inline void ReduceSumCPUExSerial(const std::vector<NDArray> &in, NDArray *out) {
using namespace rowsparse;
using namespace mshadow;
auto stype = out->storage_type();
CHECK_EQ(stype, kRowSparseStorage) << "Unexpected storage type " << stype;
size_t total_num_rows = 0;
size_t num_in = in.size();
// skip the ones with empty indices and values
std::vector<bool> skip(num_in, false);
// the values tensor of the inputs
MSHADOW_TYPE_SWITCH(out->dtype(), DType, {
MSHADOW_IDX_TYPE_SWITCH(out->aux_type(kIdx), IType, {
std::vector<Tensor<cpu, 2, DType>> in_vals(num_in);
std::vector<Tensor<cpu, 1, IType>> in_indices(num_in);
// offset to the values tensor of all inputs
std::vector<size_t> offsets(num_in, 0);
std::vector<size_t> num_rows(num_in, 0);
for (size_t i = 0; i < num_in; i++) {
if (!in[i].storage_initialized()) {
skip[i] = true;
continue;
}
auto size = in[i].aux_shape(kIdx).Size();
num_rows[i] = size;
total_num_rows += size;
in_vals[i] = in[i].data().FlatTo2D<cpu, DType>();
in_indices[i] = in[i].aux_data(kIdx).FlatTo1D<cpu, IType>();
}
std::vector<IType> indices;
indices.reserve(total_num_rows);
// gather indices from all inputs
for (size_t i = 0; i < num_in; i++) {
for (size_t j = 0; j < num_rows[i]; j++) {
indices.emplace_back(in_indices[i][j]);
}
}
CHECK_EQ(indices.size(), total_num_rows);
// dedup indices
std::sort(indices.begin(), indices.end());
indices.resize(std::unique(indices.begin(), indices.end()) - indices.begin());
// the one left are unique non-zero rows
size_t nnr = indices.size();
// allocate memory for output
out->CheckAndAlloc({Shape1(nnr)});
auto idx_data = out->aux_data(kIdx).FlatTo1D<cpu, IType>();
auto val_data = out->data().FlatTo2D<cpu, DType>();
for (size_t i = 0; i < nnr; i++) {
// copy indices back
idx_data[i] = indices[i];
bool zeros = true;
for (size_t j = 0; j < num_in; j++) {
if (skip[j]) continue;
size_t offset = offsets[j];
if (offset < num_rows[j]) {
if (indices[i] == in_indices[j][offset]) {
if (zeros) {
Copy(val_data[i], in_vals[j][offset], nullptr);
zeros = false;
} else {
val_data[i] += in_vals[j][offset];
}
offsets[j] += 1;
}
}
}
}
});
});
}
template<typename DType>
inline static void ReduceSumCPU(
const std::vector<DType*> &dptr, size_t offset, index_t size) {
using namespace mshadow; // NOLINT(*)
Tensor<cpu, 1, DType> in_0(dptr[0] + offset, Shape1(size));
for (size_t i = 1; i < dptr.size(); i+=4) {
switch (dptr.size() - i) {
case 1: {
Tensor<cpu, 1, DType> in_1(dptr[i] + offset, Shape1(size));
in_0 += in_1;
break;
}
case 2: {
Tensor<cpu, 1, DType> in_1(dptr[i] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_2(dptr[i+1] + offset, Shape1(size));
in_0 += in_1 + in_2;
break;
}
case 3: {
Tensor<cpu, 1, DType> in_1(dptr[i] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_2(dptr[i+1] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_3(dptr[i+2] + offset, Shape1(size));
in_0 += in_1 + in_2 + in_3;
break;
}
default: {
Tensor<cpu, 1, DType> in_1(dptr[i] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_2(dptr[i+1] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_3(dptr[i+2] + offset, Shape1(size));
Tensor<cpu, 1, DType> in_4(dptr[i+3] + offset, Shape1(size));
in_0 += in_1 + in_2 + in_3 + in_4;
break;
}
}
}
}
template<typename DType>
inline void ReduceSumCPUImpl(std::vector<DType*> dptr, size_t total) {
const size_t step = std::min(bigarray_bound_, static_cast<size_t>(4 << 10));
long ntask = (total + step - 1) / step; // NOLINT(*)
if (total < bigarray_bound_ || nthread_reduction_ <= 1) {
ReduceSumCPU(dptr, 0, total);
} else {
#pragma omp parallel for schedule(static) num_threads(nthread_reduction_)
for (long j = 0; j < ntask; ++j) { // NOLINT(*)
size_t k = static_cast<size_t>(j);
size_t begin = std::min(k * step, total);
size_t end = std::min((k + 1) * step, total);
if (j == ntask - 1) CHECK_EQ(end, total);
ReduceSumCPU(dptr, begin, static_cast<index_t>(end - begin));
}
}
}
/// \brief temporal space for pushing and pulling
struct BufferEntry {
/// \brief the merged value
NDArray merged;
/// \brief the cpu buffer for gpu data
std::vector<NDArray> copy_buf;
};
std::unordered_map<int, BufferEntry> merge_buf_;
size_t bigarray_bound_;
int nthread_reduction_;
bool is_serial_push_;
};
/**
* \brief an implementation of Comm that performs reduction on device
* directly.
*
* It is faster if the total device-to-device bandwidths is larger than
* device-to-cpu, which is often true for 4 or 8 GPUs. But it uses more device
* memory.
*/
class CommDevice : public Comm {
public:
CommDevice() {
inited_ = false;
}
virtual ~CommDevice() { }
void Init(int key, const NDArrayStorageType stype, const TShape& shape,
int dtype = mshadow::kFloat32) override {
if (stype == kDefaultStorage) {
sorted_key_attrs_.push_back(std::make_tuple(key, shape, dtype));
} else {
LOG(FATAL) << "storage type " << stype << " not implemented for device yet";
}
}
const NDArray& Reduce(int key, const std::vector<NDArray>& src,
int priority) override {
// avoid extra copy for single device, but it may bring problems for
// abnormal usage of kvstore
if (src.size() == 1) {
return src[0];
}
if (!inited_) {
std::vector<Context> devs;
for (const auto& a : src) {
devs.push_back(a.ctx());
}
InitMergeBuffer(devs);
if (dmlc::GetEnv("MXNET_ENABLE_GPU_P2P", 1)) {
EnableP2P(devs);
}
}
auto& buf = merge_buf_[key];
std::vector<NDArray> reduce(src.size());
CopyFromTo(src[0], &(buf.merged), priority);
reduce[0] = buf.merged;
if (buf.copy_buf.empty()) {
// TODO(mli) this results in large device memory usage for huge ndarray,
// such as the largest fullc in VGG. consider to do segment reduce with
// NDArray.Slice or gpu direct memory access. for the latter, we need to
// remove some ctx check, and also it reduces 20% perf
buf.copy_buf.resize(src.size()-1);
for (size_t i = 0; i < src.size()-1; ++i) {
buf.copy_buf[i] = NDArray(
buf.merged.shape(), buf.merged.ctx(), false, buf.merged.dtype());
}
}
for (size_t i = 0; i < src.size()-1; ++i) {
CopyFromTo(src[i+1], &(buf.copy_buf[i]), priority);
reduce[i+1] = buf.copy_buf[i];
}
ElementwiseSum(reduce, &buf.merged);
return buf.merged;
}
void Broadcast(int key, const NDArray& src,
const std::vector<NDArray*> dst, int priority) override {
if (!inited_) {
// copy to a random device first
int dev_id = key % dst.size();
CopyFromTo(src, dst[dev_id], priority);
for (size_t i = 0; i < dst.size(); ++i) {
if (i != static_cast<size_t>(dev_id)) {
CopyFromTo(*dst[dev_id], dst[i], priority);
}
}
} else {
auto& buf = merge_buf_[key];
CopyFromTo(src, &buf.merged, priority);
for (auto d : dst) {
CopyFromTo(buf.merged, d, priority);
}
}
}
void BroadcastRowSparse(int key, const NDArray& src,
const std::vector<std::pair<NDArray*, NDArray>>& dst,
const bool use_copy,
const int priority) override {
LOG(FATAL) << "Not implemented yet";
}
private:
void EnableP2P(const std::vector<Context>& devs) {
#if MXNET_USE_CUDA
std::vector<int> gpus;
for (const auto& d : devs) {
if (d.dev_mask() == gpu::kDevMask) {
gpus.push_back(d.dev_id);
}
}
int n = static_cast<int>(gpus.size());
int enabled = 0;
std::vector<int> p2p(n*n);
for (int i = 0; i < n; ++i) {
cudaSetDevice(gpus[i]);
for (int j = 0; j < n; j++) {
int access;
cudaDeviceCanAccessPeer(&access, gpus[i], gpus[j]);
if (access) {
cudaError_t e = cudaDeviceEnablePeerAccess(gpus[j], 0);
if (e == cudaSuccess || e == cudaErrorPeerAccessAlreadyEnabled) {
++enabled;
p2p[i*n+j] = 1;
}
}
}
}
if (enabled != n*(n-1)) {
// print warning info if not fully enabled
LOG(WARNING) << "only " << enabled << " out of "
<< n*(n-1) << " GPU pairs are enabled direct access. "
<< "It may affect the performance. "
<< "You can set MXNET_ENABLE_GPU_P2P=0 to turn it off";
std::string access(n, '.');
for (int i = 0; i < n; ++i) {
for (int j = 0; j < n; ++j) {
access[j] = p2p[i*n+j] ? 'v' : '.';
}
LOG(WARNING) << access;
}
}
#endif
}
using KeyAttrs = std::tuple<int, TShape, int>;
// try to allocate buff on device evenly
void InitMergeBuffer(const std::vector<Context>& devs) {
std::sort(sorted_key_attrs_.begin(), sorted_key_attrs_.end(), [](
const KeyAttrs& a, const KeyAttrs& b) {
return std::get<1>(a).Size() > std::get<1>(b).Size();
});
std::unordered_map<int, std::pair<Context, size_t>> ctx_info;
for (auto d : devs) {
ctx_info[d.dev_id] = std::make_pair(d, 0);
}
for (size_t i = 0; i < sorted_key_attrs_.size(); ++i) {
int key = std::get<0>(sorted_key_attrs_[i]);
TShape s = std::get<1>(sorted_key_attrs_[i]);
int type = std::get<2>(sorted_key_attrs_[i]);
auto& buf = merge_buf_[key];
Context ctx;
size_t min_size = std::numeric_limits<size_t>::max();
for (auto it = ctx_info.begin(); it != ctx_info.end(); ++it) {
size_t size = it->second.second;
if (size <= min_size) {
ctx = it->second.first;
min_size = size;
}
}
buf.merged = NDArray(s, ctx, false, type);
ctx_info[ctx.dev_id].second += s.Size();
}
inited_ = true;
}
std::vector<KeyAttrs> sorted_key_attrs_;
/// \brief temporal space for pushing and pulling
struct BufferEntry {
/// \brief the merged value
NDArray merged;
/// \brief the gpu buffer
std::vector<NDArray> copy_buf;
};
std::unordered_map<int, BufferEntry> merge_buf_;
bool inited_;
};
} // namespace kvstore
} // namespace mxnet
#endif // MXNET_KVSTORE_COMM_H_
|
statistic.c | /*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% SSSSS TTTTT AAA TTTTT IIIII SSSSS TTTTT IIIII CCCC %
% SS T A A T I SS T I C %
% SSS T AAAAA T I SSS T I C %
% SS T A A T I SS T I C %
% SSSSS T A A T IIIII SSSSS T IIIII CCCC %
% %
% %
% MagickCore Image Statistical Methods %
% %
% Software Design %
% John Cristy %
% July 1992 %
% %
% %
% Copyright 1999-2013 ImageMagick Studio LLC, a non-profit organization %
% dedicated to making software imaging solutions freely available. %
% %
% You may not use this file except in compliance with the License. You may %
% obtain a copy of the License at %
% %
% http://www.imagemagick.org/script/license.php %
% %
% Unless required by applicable law or agreed to in writing, software %
% distributed under the License is distributed on an "AS IS" BASIS, %
% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. %
% See the License for the specific language governing permissions and %
% limitations under the License. %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
%
%
*/
/*
Include declarations.
*/
#include "magick/studio.h"
#include "magick/property.h"
#include "magick/animate.h"
#include "magick/blob.h"
#include "magick/blob-private.h"
#include "magick/cache.h"
#include "magick/cache-private.h"
#include "magick/cache-view.h"
#include "magick/client.h"
#include "magick/color.h"
#include "magick/color-private.h"
#include "magick/colorspace.h"
#include "magick/colorspace-private.h"
#include "magick/composite.h"
#include "magick/composite-private.h"
#include "magick/compress.h"
#include "magick/constitute.h"
#include "magick/deprecate.h"
#include "magick/display.h"
#include "magick/draw.h"
#include "magick/enhance.h"
#include "magick/exception.h"
#include "magick/exception-private.h"
#include "magick/gem.h"
#include "magick/geometry.h"
#include "magick/list.h"
#include "magick/image-private.h"
#include "magick/magic.h"
#include "magick/magick.h"
#include "magick/memory_.h"
#include "magick/module.h"
#include "magick/monitor.h"
#include "magick/monitor-private.h"
#include "magick/option.h"
#include "magick/paint.h"
#include "magick/pixel-private.h"
#include "magick/profile.h"
#include "magick/quantize.h"
#include "magick/random_.h"
#include "magick/random-private.h"
#include "magick/resource_.h"
#include "magick/segment.h"
#include "magick/semaphore.h"
#include "magick/signature-private.h"
#include "magick/statistic.h"
#include "magick/string_.h"
#include "magick/thread-private.h"
#include "magick/timer.h"
#include "magick/utility.h"
#include "magick/version.h"
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% E v a l u a t e I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% EvaluateImage() applies a value to the image with an arithmetic, relational,
% or logical operator to an image. Use these operations to lighten or darken
% an image, to increase or decrease contrast in an image, or to produce the
% "negative" of an image.
%
% The format of the EvaluateImageChannel method is:
%
% MagickBooleanType EvaluateImage(Image *image,
% const MagickEvaluateOperator op,const double value,
% ExceptionInfo *exception)
% MagickBooleanType EvaluateImages(Image *images,
% const MagickEvaluateOperator op,const double value,
% ExceptionInfo *exception)
% MagickBooleanType EvaluateImageChannel(Image *image,
% const ChannelType channel,const MagickEvaluateOperator op,
% const double value,ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o op: A channel op.
%
% o value: A value value.
%
% o exception: return any errors or warnings in this structure.
%
*/
static MagickPixelPacket **DestroyPixelThreadSet(MagickPixelPacket **pixels)
{
register ssize_t
i;
assert(pixels != (MagickPixelPacket **) NULL);
for (i=0; i < (ssize_t) GetMagickResourceLimit(ThreadResource); i++)
if (pixels[i] != (MagickPixelPacket *) NULL)
pixels[i]=(MagickPixelPacket *) RelinquishMagickMemory(pixels[i]);
pixels=(MagickPixelPacket **) RelinquishMagickMemory(pixels);
return(pixels);
}
static MagickPixelPacket **AcquirePixelThreadSet(const Image *image,
const size_t number_images)
{
register ssize_t
i,
j;
MagickPixelPacket
**pixels;
size_t
length,
number_threads;
number_threads=(size_t) GetMagickResourceLimit(ThreadResource);
pixels=(MagickPixelPacket **) AcquireQuantumMemory(number_threads,
sizeof(*pixels));
if (pixels == (MagickPixelPacket **) NULL)
return((MagickPixelPacket **) NULL);
(void) ResetMagickMemory(pixels,0,number_threads*sizeof(*pixels));
for (i=0; i < (ssize_t) number_threads; i++)
{
length=image->columns;
if (length < number_images)
length=number_images;
pixels[i]=(MagickPixelPacket *) AcquireQuantumMemory(length,
sizeof(**pixels));
if (pixels[i] == (MagickPixelPacket *) NULL)
return(DestroyPixelThreadSet(pixels));
for (j=0; j < (ssize_t) length; j++)
GetMagickPixelPacket(image,&pixels[i][j]);
}
return(pixels);
}
static inline double EvaluateMax(const double x,const double y)
{
if (x > y)
return(x);
return(y);
}
#if defined(__cplusplus) || defined(c_plusplus)
extern "C" {
#endif
static int IntensityCompare(const void *x,const void *y)
{
const MagickPixelPacket
*color_1,
*color_2;
int
intensity;
color_1=(const MagickPixelPacket *) x;
color_2=(const MagickPixelPacket *) y;
intensity=(int) MagickPixelIntensity(color_2)-
(int) MagickPixelIntensity(color_1);
return(intensity);
}
#if defined(__cplusplus) || defined(c_plusplus)
}
#endif
static inline double MagickMin(const double x,const double y)
{
if (x < y)
return(x);
return(y);
}
static MagickRealType ApplyEvaluateOperator(RandomInfo *random_info,
const Quantum pixel,const MagickEvaluateOperator op,
const MagickRealType value)
{
MagickRealType
result;
result=0.0;
switch (op)
{
case UndefinedEvaluateOperator:
break;
case AbsEvaluateOperator:
{
result=(MagickRealType) fabs((double) (pixel+value));
break;
}
case AddEvaluateOperator:
{
result=(MagickRealType) (pixel+value);
break;
}
case AddModulusEvaluateOperator:
{
/*
This returns a 'floored modulus' of the addition which is a
positive result. It differs from % or fmod() which returns a
'truncated modulus' result, where floor() is replaced by trunc()
and could return a negative result (which is clipped).
*/
result=pixel+value;
result-=(QuantumRange+1.0)*floor((double) result/(QuantumRange+1.0));
break;
}
case AndEvaluateOperator:
{
result=(MagickRealType) ((size_t) pixel & (size_t) (value+0.5));
break;
}
case CosineEvaluateOperator:
{
result=(MagickRealType) (QuantumRange*(0.5*cos((double) (2.0*MagickPI*
QuantumScale*pixel*value))+0.5));
break;
}
case DivideEvaluateOperator:
{
result=pixel/(value == 0.0 ? 1.0 : value);
break;
}
case ExponentialEvaluateOperator:
{
result=(MagickRealType) (QuantumRange*exp((double) (value*QuantumScale*
pixel)));
break;
}
case GaussianNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
GaussianNoise,value);
break;
}
case ImpulseNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
ImpulseNoise,value);
break;
}
case LaplacianNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
LaplacianNoise,value);
break;
}
case LeftShiftEvaluateOperator:
{
result=(MagickRealType) ((size_t) pixel << (size_t) (value+0.5));
break;
}
case LogEvaluateOperator:
{
if ((QuantumScale*pixel) >= MagickEpsilon)
result=(MagickRealType) (QuantumRange*log((double) (QuantumScale*value*
pixel+1.0))/log((double) (value+1.0)));
break;
}
case MaxEvaluateOperator:
{
result=(MagickRealType) EvaluateMax((double) pixel,value);
break;
}
case MeanEvaluateOperator:
{
result=(MagickRealType) (pixel+value);
break;
}
case MedianEvaluateOperator:
{
result=(MagickRealType) (pixel+value);
break;
}
case MinEvaluateOperator:
{
result=(MagickRealType) MagickMin((double) pixel,value);
break;
}
case MultiplicativeNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
MultiplicativeGaussianNoise,value);
break;
}
case MultiplyEvaluateOperator:
{
result=(MagickRealType) (value*pixel);
break;
}
case OrEvaluateOperator:
{
result=(MagickRealType) ((size_t) pixel | (size_t) (value+0.5));
break;
}
case PoissonNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
PoissonNoise,value);
break;
}
case PowEvaluateOperator:
{
result=(MagickRealType) (QuantumRange*pow((double) (QuantumScale*pixel),
(double) value));
break;
}
case RightShiftEvaluateOperator:
{
result=(MagickRealType) ((size_t) pixel >> (size_t) (value+0.5));
break;
}
case SetEvaluateOperator:
{
result=value;
break;
}
case SineEvaluateOperator:
{
result=(MagickRealType) (QuantumRange*(0.5*sin((double) (2.0*MagickPI*
QuantumScale*pixel*value))+0.5));
break;
}
case SubtractEvaluateOperator:
{
result=(MagickRealType) (pixel-value);
break;
}
case SumEvaluateOperator:
{
result=(MagickRealType) (pixel+value);
break;
}
case ThresholdEvaluateOperator:
{
result=(MagickRealType) (((MagickRealType) pixel <= value) ? 0 :
QuantumRange);
break;
}
case ThresholdBlackEvaluateOperator:
{
result=(MagickRealType) (((MagickRealType) pixel <= value) ? 0 : pixel);
break;
}
case ThresholdWhiteEvaluateOperator:
{
result=(MagickRealType) (((MagickRealType) pixel > value) ? QuantumRange :
pixel);
break;
}
case UniformNoiseEvaluateOperator:
{
result=(MagickRealType) GenerateDifferentialNoise(random_info,pixel,
UniformNoise,value);
break;
}
case XorEvaluateOperator:
{
result=(MagickRealType) ((size_t) pixel ^ (size_t) (value+0.5));
break;
}
}
return(result);
}
MagickExport MagickBooleanType EvaluateImage(Image *image,
const MagickEvaluateOperator op,const double value,ExceptionInfo *exception)
{
MagickBooleanType
status;
status=EvaluateImageChannel(image,CompositeChannels,op,value,exception);
return(status);
}
MagickExport Image *EvaluateImages(const Image *images,
const MagickEvaluateOperator op,ExceptionInfo *exception)
{
#define EvaluateImageTag "Evaluate/Image"
CacheView
*evaluate_view;
Image
*image;
MagickBooleanType
status;
MagickOffsetType
progress;
MagickPixelPacket
**restrict evaluate_pixels,
zero;
RandomInfo
**restrict random_info;
size_t
number_images;
ssize_t
y;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
unsigned long
key;
#endif
assert(images != (Image *) NULL);
assert(images->signature == MagickSignature);
if (images->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",images->filename);
assert(exception != (ExceptionInfo *) NULL);
assert(exception->signature == MagickSignature);
image=CloneImage(images,images->columns,images->rows,MagickTrue,exception);
if (image == (Image *) NULL)
return((Image *) NULL);
if (SetImageStorageClass(image,DirectClass) == MagickFalse)
{
InheritException(exception,&image->exception);
image=DestroyImage(image);
return((Image *) NULL);
}
number_images=GetImageListLength(images);
evaluate_pixels=AcquirePixelThreadSet(images,number_images);
if (evaluate_pixels == (MagickPixelPacket **) NULL)
{
image=DestroyImage(image);
(void) ThrowMagickException(exception,GetMagickModule(),
ResourceLimitError,"MemoryAllocationFailed","`%s'",images->filename);
return((Image *) NULL);
}
/*
Evaluate image pixels.
*/
status=MagickTrue;
progress=0;
GetMagickPixelPacket(images,&zero);
random_info=AcquireRandomInfoThreadSet();
#if defined(MAGICKCORE_OPENMP_SUPPORT)
key=GetRandomSecretKey(random_info[0]);
#endif
evaluate_view=AcquireAuthenticCacheView(image,exception);
if (op == MedianEvaluateOperator)
{
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,images,image->rows,key == ~0UL)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
CacheView
*image_view;
const Image
*next;
const int
id = GetOpenMPThreadId();
register IndexPacket
*restrict evaluate_indexes;
register MagickPixelPacket
*evaluate_pixel;
register PixelPacket
*restrict q;
register ssize_t
x;
if (status == MagickFalse)
continue;
q=QueueCacheViewAuthenticPixels(evaluate_view,0,y,
image->columns,1,exception);
if (q == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
evaluate_indexes=GetCacheViewAuthenticIndexQueue(evaluate_view);
evaluate_pixel=evaluate_pixels[id];
for (x=0; x < (ssize_t) image->columns; x++)
{
register ssize_t
i;
for (i=0; i < (ssize_t) number_images; i++)
evaluate_pixel[i]=zero;
next=images;
for (i=0; i < (ssize_t) number_images; i++)
{
register const IndexPacket
*indexes;
register const PixelPacket
*p;
image_view=AcquireVirtualCacheView(next,exception);
p=GetCacheViewVirtualPixels(image_view,x,y,1,1,exception);
if (p == (const PixelPacket *) NULL)
{
image_view=DestroyCacheView(image_view);
break;
}
indexes=GetCacheViewVirtualIndexQueue(image_view);
evaluate_pixel[i].red=ApplyEvaluateOperator(random_info[id],
GetPixelRed(p),op,evaluate_pixel[i].red);
evaluate_pixel[i].green=ApplyEvaluateOperator(random_info[id],
GetPixelGreen(p),op,evaluate_pixel[i].green);
evaluate_pixel[i].blue=ApplyEvaluateOperator(random_info[id],
GetPixelBlue(p),op,evaluate_pixel[i].blue);
evaluate_pixel[i].opacity=ApplyEvaluateOperator(random_info[id],
GetPixelOpacity(p),op,evaluate_pixel[i].opacity);
if (image->colorspace == CMYKColorspace)
evaluate_pixel[i].index=ApplyEvaluateOperator(random_info[id],
*indexes,op,evaluate_pixel[i].index);
image_view=DestroyCacheView(image_view);
next=GetNextImageInList(next);
}
qsort((void *) evaluate_pixel,number_images,sizeof(*evaluate_pixel),
IntensityCompare);
SetPixelRed(q,ClampToQuantum(evaluate_pixel[i/2].red));
SetPixelGreen(q,ClampToQuantum(evaluate_pixel[i/2].green));
SetPixelBlue(q,ClampToQuantum(evaluate_pixel[i/2].blue));
if (image->matte == MagickFalse)
SetPixelOpacity(q,ClampToQuantum(
evaluate_pixel[i/2].opacity));
else
SetPixelAlpha(q,ClampToQuantum(evaluate_pixel[i/2].opacity));
if (image->colorspace == CMYKColorspace)
SetPixelIndex(evaluate_indexes+i,ClampToQuantum(
evaluate_pixel[i/2].index));
q++;
}
if (SyncCacheViewAuthenticPixels(evaluate_view,exception) == MagickFalse)
status=MagickFalse;
if (images->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_EvaluateImages)
#endif
proceed=SetImageProgress(images,EvaluateImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
}
else
{
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,images,image->rows,key == ~0UL)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
CacheView
*image_view;
const Image
*next;
const int
id = GetOpenMPThreadId();
register IndexPacket
*restrict evaluate_indexes;
register ssize_t
i,
x;
register MagickPixelPacket
*evaluate_pixel;
register PixelPacket
*restrict q;
if (status == MagickFalse)
continue;
q=QueueCacheViewAuthenticPixels(evaluate_view,0,y,
image->columns,1,exception);
if (q == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
evaluate_indexes=GetCacheViewAuthenticIndexQueue(evaluate_view);
evaluate_pixel=evaluate_pixels[id];
for (x=0; x < (ssize_t) image->columns; x++)
evaluate_pixel[x]=zero;
next=images;
for (i=0; i < (ssize_t) number_images; i++)
{
register const IndexPacket
*indexes;
register const PixelPacket
*p;
image_view=AcquireVirtualCacheView(next,exception);
p=GetCacheViewVirtualPixels(image_view,0,y,next->columns,1,exception);
if (p == (const PixelPacket *) NULL)
{
image_view=DestroyCacheView(image_view);
break;
}
indexes=GetCacheViewVirtualIndexQueue(image_view);
for (x=0; x < (ssize_t) next->columns; x++)
{
evaluate_pixel[x].red=ApplyEvaluateOperator(random_info[id],
GetPixelRed(p),i == 0 ? AddEvaluateOperator : op,
evaluate_pixel[x].red);
evaluate_pixel[x].green=ApplyEvaluateOperator(random_info[id],
GetPixelGreen(p),i == 0 ? AddEvaluateOperator : op,
evaluate_pixel[x].green);
evaluate_pixel[x].blue=ApplyEvaluateOperator(random_info[id],
GetPixelBlue(p),i == 0 ? AddEvaluateOperator : op,
evaluate_pixel[x].blue);
evaluate_pixel[x].opacity=ApplyEvaluateOperator(random_info[id],
GetPixelOpacity(p),i == 0 ? AddEvaluateOperator : op,
evaluate_pixel[x].opacity);
if (image->colorspace == CMYKColorspace)
evaluate_pixel[x].index=ApplyEvaluateOperator(random_info[id],
GetPixelIndex(indexes+x),i == 0 ? AddEvaluateOperator : op,
evaluate_pixel[x].index);
p++;
}
image_view=DestroyCacheView(image_view);
next=GetNextImageInList(next);
}
if (op == MeanEvaluateOperator)
for (x=0; x < (ssize_t) image->columns; x++)
{
evaluate_pixel[x].red/=number_images;
evaluate_pixel[x].green/=number_images;
evaluate_pixel[x].blue/=number_images;
evaluate_pixel[x].opacity/=number_images;
evaluate_pixel[x].index/=number_images;
}
if (op == MultiplyEvaluateOperator)
for (x=0; x < (ssize_t) image->columns; x++)
{
register ssize_t
j;
for (j=0; j < (ssize_t) (number_images-1); j++)
{
evaluate_pixel[x].red*=(MagickRealType) QuantumScale;
evaluate_pixel[x].green*=(MagickRealType) QuantumScale;
evaluate_pixel[x].blue*=(MagickRealType) QuantumScale;
evaluate_pixel[x].opacity*=(MagickRealType) QuantumScale;
evaluate_pixel[x].index*=(MagickRealType) QuantumScale;
}
}
for (x=0; x < (ssize_t) image->columns; x++)
{
SetPixelRed(q,ClampToQuantum(evaluate_pixel[x].red));
SetPixelGreen(q,ClampToQuantum(evaluate_pixel[x].green));
SetPixelBlue(q,ClampToQuantum(evaluate_pixel[x].blue));
if (image->matte == MagickFalse)
SetPixelOpacity(q,ClampToQuantum(evaluate_pixel[x].opacity));
else
SetPixelAlpha(q,ClampToQuantum(evaluate_pixel[x].opacity));
if (image->colorspace == CMYKColorspace)
SetPixelIndex(evaluate_indexes+x,ClampToQuantum(
evaluate_pixel[x].index));
q++;
}
if (SyncCacheViewAuthenticPixels(evaluate_view,exception) == MagickFalse)
status=MagickFalse;
if (images->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_EvaluateImages)
#endif
proceed=SetImageProgress(images,EvaluateImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
}
evaluate_view=DestroyCacheView(evaluate_view);
evaluate_pixels=DestroyPixelThreadSet(evaluate_pixels);
random_info=DestroyRandomInfoThreadSet(random_info);
if (status == MagickFalse)
image=DestroyImage(image);
return(image);
}
MagickExport MagickBooleanType EvaluateImageChannel(Image *image,
const ChannelType channel,const MagickEvaluateOperator op,const double value,
ExceptionInfo *exception)
{
CacheView
*image_view;
MagickBooleanType
status;
MagickOffsetType
progress;
RandomInfo
**restrict random_info;
ssize_t
y;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
unsigned long
key;
#endif
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
assert(exception != (ExceptionInfo *) NULL);
assert(exception->signature == MagickSignature);
if (SetImageStorageClass(image,DirectClass) == MagickFalse)
{
InheritException(exception,&image->exception);
return(MagickFalse);
}
status=MagickTrue;
progress=0;
random_info=AcquireRandomInfoThreadSet();
#if defined(MAGICKCORE_OPENMP_SUPPORT)
key=GetRandomSecretKey(random_info[0]);
#endif
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,image,image->rows,key == ~0UL)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
const int
id = GetOpenMPThreadId();
register IndexPacket
*restrict indexes;
register PixelPacket
*restrict q;
register ssize_t
x;
if (status == MagickFalse)
continue;
q=GetCacheViewAuthenticPixels(image_view,0,y,image->columns,1,exception);
if (q == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewAuthenticIndexQueue(image_view);
for (x=0; x < (ssize_t) image->columns; x++)
{
if ((channel & RedChannel) != 0)
SetPixelRed(q,ClampToQuantum(ApplyEvaluateOperator(random_info[id],
GetPixelRed(q),op,value)));
if ((channel & GreenChannel) != 0)
SetPixelGreen(q,ClampToQuantum(ApplyEvaluateOperator(random_info[id],
GetPixelGreen(q),op,value)));
if ((channel & BlueChannel) != 0)
SetPixelBlue(q,ClampToQuantum(ApplyEvaluateOperator(random_info[id],
GetPixelBlue(q),op,value)));
if ((channel & OpacityChannel) != 0)
{
if (image->matte == MagickFalse)
SetPixelOpacity(q,ClampToQuantum(ApplyEvaluateOperator(
random_info[id],GetPixelOpacity(q),op,value)));
else
SetPixelAlpha(q,ClampToQuantum(ApplyEvaluateOperator(
random_info[id],(Quantum) GetPixelAlpha(q),op,value)));
}
if (((channel & IndexChannel) != 0) && (indexes != (IndexPacket *) NULL))
SetPixelIndex(indexes+x,ClampToQuantum(ApplyEvaluateOperator(
random_info[id],GetPixelIndex(indexes+x),op,value)));
q++;
}
if (SyncCacheViewAuthenticPixels(image_view,exception) == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_EvaluateImageChannel)
#endif
proceed=SetImageProgress(image,EvaluateImageTag,progress++,image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
image_view=DestroyCacheView(image_view);
random_info=DestroyRandomInfoThreadSet(random_info);
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% F u n c t i o n I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% FunctionImage() applies a value to the image with an arithmetic, relational,
% or logical operator to an image. Use these operations to lighten or darken
% an image, to increase or decrease contrast in an image, or to produce the
% "negative" of an image.
%
% The format of the FunctionImageChannel method is:
%
% MagickBooleanType FunctionImage(Image *image,
% const MagickFunction function,const ssize_t number_parameters,
% const double *parameters,ExceptionInfo *exception)
% MagickBooleanType FunctionImageChannel(Image *image,
% const ChannelType channel,const MagickFunction function,
% const ssize_t number_parameters,const double *argument,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o function: A channel function.
%
% o parameters: one or more parameters.
%
% o exception: return any errors or warnings in this structure.
%
*/
static Quantum ApplyFunction(Quantum pixel,const MagickFunction function,
const size_t number_parameters,const double *parameters,
ExceptionInfo *exception)
{
MagickRealType
result;
register ssize_t
i;
(void) exception;
result=0.0;
switch (function)
{
case PolynomialFunction:
{
/*
* Polynomial
* Parameters: polynomial constants, highest to lowest order
* For example: c0*x^3 + c1*x^2 + c2*x + c3
*/
result=0.0;
for (i=0; i < (ssize_t) number_parameters; i++)
result=result*QuantumScale*pixel + parameters[i];
result*=QuantumRange;
break;
}
case SinusoidFunction:
{
/* Sinusoid Function
* Parameters: Freq, Phase, Ampl, bias
*/
double freq,phase,ampl,bias;
freq = ( number_parameters >= 1 ) ? parameters[0] : 1.0;
phase = ( number_parameters >= 2 ) ? parameters[1] : 0.0;
ampl = ( number_parameters >= 3 ) ? parameters[2] : 0.5;
bias = ( number_parameters >= 4 ) ? parameters[3] : 0.5;
result=(MagickRealType) (QuantumRange*(ampl*sin((double) (2.0*MagickPI*
(freq*QuantumScale*pixel + phase/360.0) )) + bias ) );
break;
}
case ArcsinFunction:
{
/* Arcsin Function (peged at range limits for invalid results)
* Parameters: Width, Center, Range, Bias
*/
double width,range,center,bias;
width = ( number_parameters >= 1 ) ? parameters[0] : 1.0;
center = ( number_parameters >= 2 ) ? parameters[1] : 0.5;
range = ( number_parameters >= 3 ) ? parameters[2] : 1.0;
bias = ( number_parameters >= 4 ) ? parameters[3] : 0.5;
result = 2.0/width*(QuantumScale*pixel - center);
if ( result <= -1.0 )
result = bias - range/2.0;
else if ( result >= 1.0 )
result = bias + range/2.0;
else
result=(MagickRealType) (range/MagickPI*asin((double) result)+bias);
result *= QuantumRange;
break;
}
case ArctanFunction:
{
/* Arctan Function
* Parameters: Slope, Center, Range, Bias
*/
double slope,range,center,bias;
slope = ( number_parameters >= 1 ) ? parameters[0] : 1.0;
center = ( number_parameters >= 2 ) ? parameters[1] : 0.5;
range = ( number_parameters >= 3 ) ? parameters[2] : 1.0;
bias = ( number_parameters >= 4 ) ? parameters[3] : 0.5;
result=(MagickRealType) (MagickPI*slope*(QuantumScale*pixel-center));
result=(MagickRealType) (QuantumRange*(range/MagickPI*atan((double)
result) + bias ) );
break;
}
case UndefinedFunction:
break;
}
return(ClampToQuantum(result));
}
MagickExport MagickBooleanType FunctionImage(Image *image,
const MagickFunction function,const size_t number_parameters,
const double *parameters,ExceptionInfo *exception)
{
MagickBooleanType
status;
status=FunctionImageChannel(image,CompositeChannels,function,
number_parameters,parameters,exception);
return(status);
}
MagickExport MagickBooleanType FunctionImageChannel(Image *image,
const ChannelType channel,const MagickFunction function,
const size_t number_parameters,const double *parameters,
ExceptionInfo *exception)
{
#define FunctionImageTag "Function/Image "
CacheView
*image_view;
MagickBooleanType
status;
MagickOffsetType
progress;
ssize_t
y;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
assert(exception != (ExceptionInfo *) NULL);
assert(exception->signature == MagickSignature);
if (SetImageStorageClass(image,DirectClass) == MagickFalse)
{
InheritException(exception,&image->exception);
return(MagickFalse);
}
status=MagickTrue;
progress=0;
image_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,image,image->rows,1)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
register IndexPacket
*restrict indexes;
register ssize_t
x;
register PixelPacket
*restrict q;
if (status == MagickFalse)
continue;
q=GetCacheViewAuthenticPixels(image_view,0,y,image->columns,1,exception);
if (q == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewAuthenticIndexQueue(image_view);
for (x=0; x < (ssize_t) image->columns; x++)
{
if ((channel & RedChannel) != 0)
SetPixelRed(q,ApplyFunction(GetPixelRed(q),function,
number_parameters,parameters,exception));
if ((channel & GreenChannel) != 0)
SetPixelGreen(q,ApplyFunction(GetPixelGreen(q),function,
number_parameters,parameters,exception));
if ((channel & BlueChannel) != 0)
SetPixelBlue(q,ApplyFunction(GetPixelBlue(q),function,
number_parameters,parameters,exception));
if ((channel & OpacityChannel) != 0)
{
if (image->matte == MagickFalse)
SetPixelOpacity(q,ApplyFunction(GetPixelOpacity(q),function,
number_parameters,parameters,exception));
else
SetPixelAlpha(q,ApplyFunction((Quantum) GetPixelAlpha(q),function,
number_parameters,parameters,exception));
}
if (((channel & IndexChannel) != 0) && (indexes != (IndexPacket *) NULL))
SetPixelIndex(indexes+x,ApplyFunction(GetPixelIndex(indexes+x),function,
number_parameters,parameters,exception));
q++;
}
if (SyncCacheViewAuthenticPixels(image_view,exception) == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_FunctionImageChannel)
#endif
proceed=SetImageProgress(image,FunctionImageTag,progress++,image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
image_view=DestroyCacheView(image_view);
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
+ G e t I m a g e C h a n n e l E x t r e m a %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetImageChannelExtrema() returns the extrema of one or more image channels.
%
% The format of the GetImageChannelExtrema method is:
%
% MagickBooleanType GetImageChannelExtrema(const Image *image,
% const ChannelType channel,size_t *minima,size_t *maxima,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o minima: the minimum value in the channel.
%
% o maxima: the maximum value in the channel.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport MagickBooleanType GetImageExtrema(const Image *image,
size_t *minima,size_t *maxima,ExceptionInfo *exception)
{
return(GetImageChannelExtrema(image,CompositeChannels,minima,maxima,exception));
}
MagickExport MagickBooleanType GetImageChannelExtrema(const Image *image,
const ChannelType channel,size_t *minima,size_t *maxima,
ExceptionInfo *exception)
{
double
max,
min;
MagickBooleanType
status;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
status=GetImageChannelRange(image,channel,&min,&max,exception);
*minima=(size_t) ceil(min-0.5);
*maxima=(size_t) floor(max+0.5);
return(status);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t I m a g e C h a n n e l M e a n %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetImageChannelMean() returns the mean and standard deviation of one or more
% image channels.
%
% The format of the GetImageChannelMean method is:
%
% MagickBooleanType GetImageChannelMean(const Image *image,
% const ChannelType channel,double *mean,double *standard_deviation,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o mean: the average value in the channel.
%
% o standard_deviation: the standard deviation of the channel.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport MagickBooleanType GetImageMean(const Image *image,double *mean,
double *standard_deviation,ExceptionInfo *exception)
{
MagickBooleanType
status;
status=GetImageChannelMean(image,CompositeChannels,mean,standard_deviation,
exception);
return(status);
}
MagickExport MagickBooleanType GetImageChannelMean(const Image *image,
const ChannelType channel,double *mean,double *standard_deviation,
ExceptionInfo *exception)
{
ChannelStatistics
*channel_statistics;
size_t
channels;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
channel_statistics=GetImageChannelStatistics(image,exception);
if (channel_statistics == (ChannelStatistics *) NULL)
return(MagickFalse);
channels=0;
channel_statistics[CompositeChannels].mean=0.0;
channel_statistics[CompositeChannels].standard_deviation=0.0;
if ((channel & RedChannel) != 0)
{
channel_statistics[CompositeChannels].mean+=
channel_statistics[RedChannel].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[RedChannel].variance-
channel_statistics[RedChannel].mean*
channel_statistics[RedChannel].mean;
channels++;
}
if ((channel & GreenChannel) != 0)
{
channel_statistics[CompositeChannels].mean+=
channel_statistics[GreenChannel].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[GreenChannel].variance-
channel_statistics[GreenChannel].mean*
channel_statistics[GreenChannel].mean;
channels++;
}
if ((channel & BlueChannel) != 0)
{
channel_statistics[CompositeChannels].mean+=
channel_statistics[BlueChannel].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[BlueChannel].variance-
channel_statistics[BlueChannel].mean*
channel_statistics[BlueChannel].mean;
channels++;
}
if (((channel & OpacityChannel) != 0) &&
(image->matte != MagickFalse))
{
channel_statistics[CompositeChannels].mean+=
channel_statistics[OpacityChannel].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[OpacityChannel].variance-
channel_statistics[OpacityChannel].mean*
channel_statistics[OpacityChannel].mean;
channels++;
}
if (((channel & IndexChannel) != 0) &&
(image->colorspace == CMYKColorspace))
{
channel_statistics[CompositeChannels].mean+=
channel_statistics[BlackChannel].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[BlackChannel].variance-
channel_statistics[BlackChannel].mean*
channel_statistics[BlackChannel].mean;
channels++;
}
channel_statistics[CompositeChannels].mean/=channels;
channel_statistics[CompositeChannels].standard_deviation=
sqrt(channel_statistics[CompositeChannels].standard_deviation/channels);
*mean=channel_statistics[CompositeChannels].mean;
*standard_deviation=channel_statistics[CompositeChannels].standard_deviation;
channel_statistics=(ChannelStatistics *) RelinquishMagickMemory(
channel_statistics);
return(MagickTrue);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t I m a g e C h a n n e l K u r t o s i s %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetImageChannelKurtosis() returns the kurtosis and skewness of one or more
% image channels.
%
% The format of the GetImageChannelKurtosis method is:
%
% MagickBooleanType GetImageChannelKurtosis(const Image *image,
% const ChannelType channel,double *kurtosis,double *skewness,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o kurtosis: the kurtosis of the channel.
%
% o skewness: the skewness of the channel.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport MagickBooleanType GetImageKurtosis(const Image *image,
double *kurtosis,double *skewness,ExceptionInfo *exception)
{
MagickBooleanType
status;
status=GetImageChannelKurtosis(image,CompositeChannels,kurtosis,skewness,
exception);
return(status);
}
MagickExport MagickBooleanType GetImageChannelKurtosis(const Image *image,
const ChannelType channel,double *kurtosis,double *skewness,
ExceptionInfo *exception)
{
double
area,
mean,
standard_deviation,
sum_squares,
sum_cubes,
sum_fourth_power;
ssize_t
y;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
*kurtosis=0.0;
*skewness=0.0;
area=0.0;
mean=0.0;
standard_deviation=0.0;
sum_squares=0.0;
sum_cubes=0.0;
sum_fourth_power=0.0;
for (y=0; y < (ssize_t) image->rows; y++)
{
register const IndexPacket
*restrict indexes;
register const PixelPacket
*restrict p;
register ssize_t
x;
p=GetVirtualPixels(image,0,y,image->columns,1,exception);
if (p == (const PixelPacket *) NULL)
break;
indexes=GetVirtualIndexQueue(image);
for (x=0; x < (ssize_t) image->columns; x++)
{
if ((channel & RedChannel) != 0)
{
mean+=GetPixelRed(p);
sum_squares+=(double) GetPixelRed(p)*GetPixelRed(p);
sum_cubes+=(double) GetPixelRed(p)*GetPixelRed(p)*GetPixelRed(p);
sum_fourth_power+=(double) GetPixelRed(p)*GetPixelRed(p)*
GetPixelRed(p)*GetPixelRed(p);
area++;
}
if ((channel & GreenChannel) != 0)
{
mean+=GetPixelGreen(p);
sum_squares+=(double) GetPixelGreen(p)*GetPixelGreen(p);
sum_cubes+=(double) GetPixelGreen(p)*GetPixelGreen(p)*
GetPixelGreen(p);
sum_fourth_power+=(double) GetPixelGreen(p)*GetPixelGreen(p)*
GetPixelGreen(p)*GetPixelGreen(p);
area++;
}
if ((channel & BlueChannel) != 0)
{
mean+=GetPixelBlue(p);
sum_squares+=(double) GetPixelBlue(p)*GetPixelBlue(p);
sum_cubes+=(double) GetPixelBlue(p)*GetPixelBlue(p)*GetPixelBlue(p);
sum_fourth_power+=(double) GetPixelBlue(p)*GetPixelBlue(p)*
GetPixelBlue(p)*GetPixelBlue(p);
area++;
}
if ((channel & OpacityChannel) != 0)
{
mean+=GetPixelOpacity(p);
sum_squares+=(double) GetPixelOpacity(p)*GetPixelOpacity(p);
sum_cubes+=(double) GetPixelOpacity(p)*GetPixelOpacity(p)*
GetPixelOpacity(p);
sum_fourth_power+=(double) GetPixelOpacity(p)*GetPixelOpacity(p)*
GetPixelOpacity(p)*GetPixelOpacity(p);
area++;
}
if (((channel & IndexChannel) != 0) &&
(image->colorspace == CMYKColorspace))
{
mean+=GetPixelIndex(indexes+x);
sum_squares+=(double) GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x);
sum_cubes+=(double) GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x);
sum_fourth_power+=(double) GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x);
area++;
}
p++;
}
}
if (y < (ssize_t) image->rows)
return(MagickFalse);
if (area != 0.0)
{
mean/=area;
sum_squares/=area;
sum_cubes/=area;
sum_fourth_power/=area;
}
standard_deviation=sqrt(sum_squares-(mean*mean));
if (standard_deviation != 0.0)
{
*kurtosis=sum_fourth_power-4.0*mean*sum_cubes+6.0*mean*mean*sum_squares-
3.0*mean*mean*mean*mean;
*kurtosis/=standard_deviation*standard_deviation*standard_deviation*
standard_deviation;
*kurtosis-=3.0;
*skewness=sum_cubes-3.0*mean*sum_squares+2.0*mean*mean*mean;
*skewness/=standard_deviation*standard_deviation*standard_deviation;
}
return(y == (ssize_t) image->rows ? MagickTrue : MagickFalse);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t I m a g e C h a n n e l R a n g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetImageChannelRange() returns the range of one or more image channels.
%
% The format of the GetImageChannelRange method is:
%
% MagickBooleanType GetImageChannelRange(const Image *image,
% const ChannelType channel,double *minima,double *maxima,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the channel.
%
% o minima: the minimum value in the channel.
%
% o maxima: the maximum value in the channel.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport MagickBooleanType GetImageRange(const Image *image,
double *minima,double *maxima,ExceptionInfo *exception)
{
return(GetImageChannelRange(image,CompositeChannels,minima,maxima,exception));
}
MagickExport MagickBooleanType GetImageChannelRange(const Image *image,
const ChannelType channel,double *minima,double *maxima,
ExceptionInfo *exception)
{
MagickPixelPacket
pixel;
ssize_t
y;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
*maxima=(-MagickHuge);
*minima=MagickHuge;
GetMagickPixelPacket(image,&pixel);
for (y=0; y < (ssize_t) image->rows; y++)
{
register const IndexPacket
*restrict indexes;
register const PixelPacket
*restrict p;
register ssize_t
x;
p=GetVirtualPixels(image,0,y,image->columns,1,exception);
if (p == (const PixelPacket *) NULL)
break;
indexes=GetVirtualIndexQueue(image);
for (x=0; x < (ssize_t) image->columns; x++)
{
SetMagickPixelPacket(image,p,indexes+x,&pixel);
if ((channel & RedChannel) != 0)
{
if (pixel.red < *minima)
*minima=(double) pixel.red;
if (pixel.red > *maxima)
*maxima=(double) pixel.red;
}
if ((channel & GreenChannel) != 0)
{
if (pixel.green < *minima)
*minima=(double) pixel.green;
if (pixel.green > *maxima)
*maxima=(double) pixel.green;
}
if ((channel & BlueChannel) != 0)
{
if (pixel.blue < *minima)
*minima=(double) pixel.blue;
if (pixel.blue > *maxima)
*maxima=(double) pixel.blue;
}
if (((channel & OpacityChannel) != 0) &&
(image->matte != MagickFalse))
{
if (pixel.opacity < *minima)
*minima=(double) pixel.opacity;
if (pixel.opacity > *maxima)
*maxima=(double) pixel.opacity;
}
if (((channel & IndexChannel) != 0) &&
(image->colorspace == CMYKColorspace))
{
if ((double) GetPixelIndex(indexes+x) < *minima)
*minima=(double) GetPixelIndex(indexes+x);
if ((double) GetPixelIndex(indexes+x) > *maxima)
*maxima=(double) GetPixelIndex(indexes+x);
}
p++;
}
}
return(y == (ssize_t) image->rows ? MagickTrue : MagickFalse);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% G e t I m a g e C h a n n e l S t a t i s t i c s %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% GetImageChannelStatistics() returns statistics for each channel in the
% image. The statistics include the channel depth, its minima, maxima, mean,
% standard deviation, kurtosis and skewness. You can access the red channel
% mean, for example, like this:
%
% channel_statistics=GetImageChannelStatistics(image,exception);
% red_mean=channel_statistics[RedChannel].mean;
%
% Use MagickRelinquishMemory() to free the statistics buffer.
%
% The format of the GetImageChannelStatistics method is:
%
% ChannelStatistics *GetImageChannelStatistics(const Image *image,
% ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport ChannelStatistics *GetImageChannelStatistics(const Image *image,
ExceptionInfo *exception)
{
ChannelStatistics
*channel_statistics;
double
area;
MagickStatusType
status;
QuantumAny
range;
register ssize_t
i;
size_t
channels,
depth,
length;
ssize_t
y;
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
length=CompositeChannels+1UL;
channel_statistics=(ChannelStatistics *) AcquireQuantumMemory(length,
sizeof(*channel_statistics));
if (channel_statistics == (ChannelStatistics *) NULL)
ThrowFatalException(ResourceLimitFatalError,"MemoryAllocationFailed");
(void) ResetMagickMemory(channel_statistics,0,length*
sizeof(*channel_statistics));
for (i=0; i <= (ssize_t) CompositeChannels; i++)
{
channel_statistics[i].depth=1;
channel_statistics[i].maxima=(-MagickHuge);
channel_statistics[i].minima=MagickHuge;
}
for (y=0; y < (ssize_t) image->rows; y++)
{
register const IndexPacket
*restrict indexes;
register const PixelPacket
*restrict p;
register ssize_t
x;
p=GetVirtualPixels(image,0,y,image->columns,1,exception);
if (p == (const PixelPacket *) NULL)
break;
indexes=GetVirtualIndexQueue(image);
for (x=0; x < (ssize_t) image->columns; )
{
if (channel_statistics[RedChannel].depth != MAGICKCORE_QUANTUM_DEPTH)
{
depth=channel_statistics[RedChannel].depth;
range=GetQuantumRange(depth);
status=GetPixelRed(p) != ScaleAnyToQuantum(ScaleQuantumToAny(
GetPixelRed(p),range),range) ? MagickTrue : MagickFalse;
if (status != MagickFalse)
{
channel_statistics[RedChannel].depth++;
continue;
}
}
if (channel_statistics[GreenChannel].depth != MAGICKCORE_QUANTUM_DEPTH)
{
depth=channel_statistics[GreenChannel].depth;
range=GetQuantumRange(depth);
status=GetPixelGreen(p) != ScaleAnyToQuantum(ScaleQuantumToAny(
GetPixelGreen(p),range),range) ? MagickTrue : MagickFalse;
if (status != MagickFalse)
{
channel_statistics[GreenChannel].depth++;
continue;
}
}
if (channel_statistics[BlueChannel].depth != MAGICKCORE_QUANTUM_DEPTH)
{
depth=channel_statistics[BlueChannel].depth;
range=GetQuantumRange(depth);
status=GetPixelBlue(p) != ScaleAnyToQuantum(ScaleQuantumToAny(
GetPixelBlue(p),range),range) ? MagickTrue : MagickFalse;
if (status != MagickFalse)
{
channel_statistics[BlueChannel].depth++;
continue;
}
}
if (image->matte != MagickFalse)
{
if (channel_statistics[OpacityChannel].depth != MAGICKCORE_QUANTUM_DEPTH)
{
depth=channel_statistics[OpacityChannel].depth;
range=GetQuantumRange(depth);
status=GetPixelOpacity(p) != ScaleAnyToQuantum(ScaleQuantumToAny(
GetPixelOpacity(p),range),range) ? MagickTrue : MagickFalse;
if (status != MagickFalse)
{
channel_statistics[OpacityChannel].depth++;
continue;
}
}
}
if (image->colorspace == CMYKColorspace)
{
if (channel_statistics[BlackChannel].depth != MAGICKCORE_QUANTUM_DEPTH)
{
depth=channel_statistics[BlackChannel].depth;
range=GetQuantumRange(depth);
status=GetPixelIndex(indexes+x) != ScaleAnyToQuantum(
ScaleQuantumToAny(GetPixelIndex(indexes+x),range),range) ?
MagickTrue : MagickFalse;
if (status != MagickFalse)
{
channel_statistics[BlackChannel].depth++;
continue;
}
}
}
if ((double) GetPixelRed(p) < channel_statistics[RedChannel].minima)
channel_statistics[RedChannel].minima=(double) GetPixelRed(p);
if ((double) GetPixelRed(p) > channel_statistics[RedChannel].maxima)
channel_statistics[RedChannel].maxima=(double) GetPixelRed(p);
channel_statistics[RedChannel].sum+=GetPixelRed(p);
channel_statistics[RedChannel].sum_squared+=(double) GetPixelRed(p)*
GetPixelRed(p);
channel_statistics[RedChannel].sum_cubed+=(double)
GetPixelRed(p)*GetPixelRed(p)*GetPixelRed(p);
channel_statistics[RedChannel].sum_fourth_power+=(double)
GetPixelRed(p)*GetPixelRed(p)*GetPixelRed(p)*GetPixelRed(p);
if ((double) GetPixelGreen(p) < channel_statistics[GreenChannel].minima)
channel_statistics[GreenChannel].minima=(double) GetPixelGreen(p);
if ((double) GetPixelGreen(p) > channel_statistics[GreenChannel].maxima)
channel_statistics[GreenChannel].maxima=(double) GetPixelGreen(p);
channel_statistics[GreenChannel].sum+=GetPixelGreen(p);
channel_statistics[GreenChannel].sum_squared+=(double) GetPixelGreen(p)*
GetPixelGreen(p);
channel_statistics[GreenChannel].sum_cubed+=(double) GetPixelGreen(p)*
GetPixelGreen(p)*GetPixelGreen(p);
channel_statistics[GreenChannel].sum_fourth_power+=(double)
GetPixelGreen(p)*GetPixelGreen(p)*GetPixelGreen(p)*GetPixelGreen(p);
if ((double) GetPixelBlue(p) < channel_statistics[BlueChannel].minima)
channel_statistics[BlueChannel].minima=(double) GetPixelBlue(p);
if ((double) GetPixelBlue(p) > channel_statistics[BlueChannel].maxima)
channel_statistics[BlueChannel].maxima=(double) GetPixelBlue(p);
channel_statistics[BlueChannel].sum+=GetPixelBlue(p);
channel_statistics[BlueChannel].sum_squared+=(double) GetPixelBlue(p)*
GetPixelBlue(p);
channel_statistics[BlueChannel].sum_cubed+=(double) GetPixelBlue(p)*
GetPixelBlue(p)*GetPixelBlue(p);
channel_statistics[BlueChannel].sum_fourth_power+=(double)
GetPixelBlue(p)*GetPixelBlue(p)*GetPixelBlue(p)*GetPixelBlue(p);
if (image->matte != MagickFalse)
{
if ((double) GetPixelOpacity(p) < channel_statistics[OpacityChannel].minima)
channel_statistics[OpacityChannel].minima=(double)
GetPixelOpacity(p);
if ((double) GetPixelOpacity(p) > channel_statistics[OpacityChannel].maxima)
channel_statistics[OpacityChannel].maxima=(double)
GetPixelOpacity(p);
channel_statistics[OpacityChannel].sum+=GetPixelOpacity(p);
channel_statistics[OpacityChannel].sum_squared+=(double)
GetPixelOpacity(p)*GetPixelOpacity(p);
channel_statistics[OpacityChannel].sum_cubed+=(double)
GetPixelOpacity(p)*GetPixelOpacity(p)*GetPixelOpacity(p);
channel_statistics[OpacityChannel].sum_fourth_power+=(double)
GetPixelOpacity(p)*GetPixelOpacity(p)*GetPixelOpacity(p)*
GetPixelOpacity(p);
}
if (image->colorspace == CMYKColorspace)
{
if ((double) GetPixelIndex(indexes+x) < channel_statistics[BlackChannel].minima)
channel_statistics[BlackChannel].minima=(double)
GetPixelIndex(indexes+x);
if ((double) GetPixelIndex(indexes+x) > channel_statistics[BlackChannel].maxima)
channel_statistics[BlackChannel].maxima=(double)
GetPixelIndex(indexes+x);
channel_statistics[BlackChannel].sum+=GetPixelIndex(indexes+x);
channel_statistics[BlackChannel].sum_squared+=(double)
GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x);
channel_statistics[BlackChannel].sum_cubed+=(double)
GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x);
channel_statistics[BlackChannel].sum_fourth_power+=(double)
GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x)*
GetPixelIndex(indexes+x)*GetPixelIndex(indexes+x);
}
x++;
p++;
}
}
area=(double) image->columns*image->rows;
for (i=0; i < (ssize_t) CompositeChannels; i++)
{
channel_statistics[i].sum/=area;
channel_statistics[i].sum_squared/=area;
channel_statistics[i].sum_cubed/=area;
channel_statistics[i].sum_fourth_power/=area;
channel_statistics[i].mean=channel_statistics[i].sum;
channel_statistics[i].variance=channel_statistics[i].sum_squared;
channel_statistics[i].standard_deviation=sqrt(
channel_statistics[i].variance-(channel_statistics[i].mean*
channel_statistics[i].mean));
}
for (i=0; i < (ssize_t) CompositeChannels; i++)
{
channel_statistics[CompositeChannels].depth=(size_t) EvaluateMax((double)
channel_statistics[CompositeChannels].depth,(double)
channel_statistics[i].depth);
channel_statistics[CompositeChannels].minima=MagickMin(
channel_statistics[CompositeChannels].minima,
channel_statistics[i].minima);
channel_statistics[CompositeChannels].maxima=EvaluateMax(
channel_statistics[CompositeChannels].maxima,
channel_statistics[i].maxima);
channel_statistics[CompositeChannels].sum+=channel_statistics[i].sum;
channel_statistics[CompositeChannels].sum_squared+=
channel_statistics[i].sum_squared;
channel_statistics[CompositeChannels].sum_cubed+=
channel_statistics[i].sum_cubed;
channel_statistics[CompositeChannels].sum_fourth_power+=
channel_statistics[i].sum_fourth_power;
channel_statistics[CompositeChannels].mean+=channel_statistics[i].mean;
channel_statistics[CompositeChannels].variance+=
channel_statistics[i].variance-channel_statistics[i].mean*
channel_statistics[i].mean;
channel_statistics[CompositeChannels].standard_deviation+=
channel_statistics[i].variance-channel_statistics[i].mean*
channel_statistics[i].mean;
}
channels=3;
if (image->matte != MagickFalse)
channels++;
if (image->colorspace == CMYKColorspace)
channels++;
channel_statistics[CompositeChannels].sum/=channels;
channel_statistics[CompositeChannels].sum_squared/=channels;
channel_statistics[CompositeChannels].sum_cubed/=channels;
channel_statistics[CompositeChannels].sum_fourth_power/=channels;
channel_statistics[CompositeChannels].mean/=channels;
channel_statistics[CompositeChannels].variance/=channels;
channel_statistics[CompositeChannels].standard_deviation=
sqrt(channel_statistics[CompositeChannels].standard_deviation/channels);
channel_statistics[CompositeChannels].kurtosis/=channels;
channel_statistics[CompositeChannels].skewness/=channels;
for (i=0; i <= (ssize_t) CompositeChannels; i++)
{
if (channel_statistics[i].standard_deviation == 0.0)
continue;
channel_statistics[i].skewness=(channel_statistics[i].sum_cubed-
3.0*channel_statistics[i].mean*channel_statistics[i].sum_squared+
2.0*channel_statistics[i].mean*channel_statistics[i].mean*
channel_statistics[i].mean)/(channel_statistics[i].standard_deviation*
channel_statistics[i].standard_deviation*
channel_statistics[i].standard_deviation);
channel_statistics[i].kurtosis=(channel_statistics[i].sum_fourth_power-
4.0*channel_statistics[i].mean*channel_statistics[i].sum_cubed+
6.0*channel_statistics[i].mean*channel_statistics[i].mean*
channel_statistics[i].sum_squared-3.0*channel_statistics[i].mean*
channel_statistics[i].mean*1.0*channel_statistics[i].mean*
channel_statistics[i].mean)/(channel_statistics[i].standard_deviation*
channel_statistics[i].standard_deviation*
channel_statistics[i].standard_deviation*
channel_statistics[i].standard_deviation)-3.0;
}
return(channel_statistics);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% P o l y n o m i a l I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% PolynomialImage() returns a new image where each pixel is the sum of the
% pixels in the image sequence after applying its corresponding terms
% (coefficient and degree pairs).
%
% The format of the PolynomialImage method is:
%
% Image *PolynomialImage(const Image *images,const size_t number_terms,
% const double *terms,ExceptionInfo *exception)
% Image *PolynomialImageChannel(const Image *images,
% const size_t number_terms,const ChannelType channel,
% const double *terms,ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o images: the image sequence.
%
% o channel: the channel.
%
% o number_terms: the number of terms in the list. The actual list length
% is 2 x number_terms + 1 (the constant).
%
% o terms: the list of polynomial coefficients and degree pairs and a
% constant.
%
% o exception: return any errors or warnings in this structure.
%
*/
MagickExport Image *PolynomialImage(const Image *images,
const size_t number_terms,const double *terms,ExceptionInfo *exception)
{
Image
*polynomial_image;
polynomial_image=PolynomialImageChannel(images,DefaultChannels,number_terms,
terms,exception);
return(polynomial_image);
}
MagickExport Image *PolynomialImageChannel(const Image *images,
const ChannelType channel,const size_t number_terms,const double *terms,
ExceptionInfo *exception)
{
#define PolynomialImageTag "Polynomial/Image"
CacheView
*polynomial_view;
Image
*image;
MagickBooleanType
status;
MagickOffsetType
progress;
MagickPixelPacket
**restrict polynomial_pixels,
zero;
size_t
number_images;
ssize_t
y;
assert(images != (Image *) NULL);
assert(images->signature == MagickSignature);
if (images->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",images->filename);
assert(exception != (ExceptionInfo *) NULL);
assert(exception->signature == MagickSignature);
image=CloneImage(images,images->columns,images->rows,MagickTrue,exception);
if (image == (Image *) NULL)
return((Image *) NULL);
if (SetImageStorageClass(image,DirectClass) == MagickFalse)
{
InheritException(exception,&image->exception);
image=DestroyImage(image);
return((Image *) NULL);
}
number_images=GetImageListLength(images);
polynomial_pixels=AcquirePixelThreadSet(images,number_images);
if (polynomial_pixels == (MagickPixelPacket **) NULL)
{
image=DestroyImage(image);
(void) ThrowMagickException(exception,GetMagickModule(),
ResourceLimitError,"MemoryAllocationFailed","`%s'",images->filename);
return((Image *) NULL);
}
/*
Polynomial image pixels.
*/
status=MagickTrue;
progress=0;
GetMagickPixelPacket(images,&zero);
polynomial_view=AcquireAuthenticCacheView(image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,image,image->rows,1)
#endif
for (y=0; y < (ssize_t) image->rows; y++)
{
CacheView
*image_view;
const Image
*next;
const int
id = GetOpenMPThreadId();
register IndexPacket
*restrict polynomial_indexes;
register MagickPixelPacket
*polynomial_pixel;
register PixelPacket
*restrict q;
register ssize_t
i,
x;
if (status == MagickFalse)
continue;
q=QueueCacheViewAuthenticPixels(polynomial_view,0,y,image->columns,1,
exception);
if (q == (PixelPacket *) NULL)
{
status=MagickFalse;
continue;
}
polynomial_indexes=GetCacheViewAuthenticIndexQueue(polynomial_view);
polynomial_pixel=polynomial_pixels[id];
for (x=0; x < (ssize_t) image->columns; x++)
polynomial_pixel[x]=zero;
next=images;
for (i=0; i < (ssize_t) number_images; i++)
{
register const IndexPacket
*indexes;
register const PixelPacket
*p;
if (i >= (ssize_t) number_terms)
break;
image_view=AcquireVirtualCacheView(next,exception);
p=GetCacheViewVirtualPixels(image_view,0,y,image->columns,1,exception);
if (p == (const PixelPacket *) NULL)
{
image_view=DestroyCacheView(image_view);
break;
}
indexes=GetCacheViewVirtualIndexQueue(image_view);
for (x=0; x < (ssize_t) image->columns; x++)
{
double
coefficient,
degree;
coefficient=terms[i << 1];
degree=terms[(i << 1)+1];
polynomial_pixel[x].red+=coefficient*pow(QuantumScale*p->red,degree);
polynomial_pixel[x].green+=coefficient*pow(QuantumScale*p->green,
degree);
polynomial_pixel[x].blue+=coefficient*pow(QuantumScale*p->blue,degree);
polynomial_pixel[x].opacity+=coefficient*pow(QuantumScale*p->opacity,
degree);
if (image->colorspace == CMYKColorspace)
polynomial_pixel[x].index+=coefficient*pow(QuantumScale*indexes[x],
degree);
p++;
}
image_view=DestroyCacheView(image_view);
next=GetNextImageInList(next);
}
for (x=0; x < (ssize_t) image->columns; x++)
{
SetPixelRed(q,ClampToQuantum(QuantumRange*polynomial_pixel[x].red));
SetPixelGreen(q,ClampToQuantum(QuantumRange*polynomial_pixel[x].green));
SetPixelBlue(q,ClampToQuantum(QuantumRange*polynomial_pixel[x].blue));
if (image->matte == MagickFalse)
SetPixelOpacity(q,ClampToQuantum(QuantumRange*
polynomial_pixel[x].opacity));
else
SetPixelAlpha(q,ClampToQuantum(QuantumRange*
polynomial_pixel[x].opacity));
if (image->colorspace == CMYKColorspace)
SetPixelIndex(polynomial_indexes+x,ClampToQuantum(QuantumRange*
polynomial_pixel[x].index));
q++;
}
if (SyncCacheViewAuthenticPixels(polynomial_view,exception) == MagickFalse)
status=MagickFalse;
if (images->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_PolynomialImages)
#endif
proceed=SetImageProgress(images,PolynomialImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
polynomial_view=DestroyCacheView(polynomial_view);
polynomial_pixels=DestroyPixelThreadSet(polynomial_pixels);
if (status == MagickFalse)
image=DestroyImage(image);
return(image);
}
/*
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% %
% %
% S t a t i s t i c I m a g e %
% %
% %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% StatisticImage() makes each pixel the min / max / median / mode / etc. of
% the neighborhood of the specified width and height.
%
% The format of the StatisticImage method is:
%
% Image *StatisticImage(const Image *image,const StatisticType type,
% const size_t width,const size_t height,ExceptionInfo *exception)
% Image *StatisticImageChannel(const Image *image,
% const ChannelType channel,const StatisticType type,
% const size_t width,const size_t height,ExceptionInfo *exception)
%
% A description of each parameter follows:
%
% o image: the image.
%
% o channel: the image channel.
%
% o type: the statistic type (median, mode, etc.).
%
% o width: the width of the pixel neighborhood.
%
% o height: the height of the pixel neighborhood.
%
% o exception: return any errors or warnings in this structure.
%
*/
#define ListChannels 5
typedef struct _ListNode
{
size_t
next[9],
count,
signature;
} ListNode;
typedef struct _SkipList
{
ssize_t
level;
ListNode
*nodes;
} SkipList;
typedef struct _PixelList
{
size_t
length,
seed,
signature;
SkipList
lists[ListChannels];
} PixelList;
static PixelList *DestroyPixelList(PixelList *pixel_list)
{
register ssize_t
i;
if (pixel_list == (PixelList *) NULL)
return((PixelList *) NULL);
for (i=0; i < ListChannels; i++)
if (pixel_list->lists[i].nodes != (ListNode *) NULL)
pixel_list->lists[i].nodes=(ListNode *) RelinquishMagickMemory(
pixel_list->lists[i].nodes);
pixel_list=(PixelList *) RelinquishMagickMemory(pixel_list);
return(pixel_list);
}
static PixelList **DestroyPixelListThreadSet(PixelList **pixel_list)
{
register ssize_t
i;
assert(pixel_list != (PixelList **) NULL);
for (i=0; i < (ssize_t) GetMagickResourceLimit(ThreadResource); i++)
if (pixel_list[i] != (PixelList *) NULL)
pixel_list[i]=DestroyPixelList(pixel_list[i]);
pixel_list=(PixelList **) RelinquishMagickMemory(pixel_list);
return(pixel_list);
}
static PixelList *AcquirePixelList(const size_t width,const size_t height)
{
PixelList
*pixel_list;
register ssize_t
i;
pixel_list=(PixelList *) AcquireMagickMemory(sizeof(*pixel_list));
if (pixel_list == (PixelList *) NULL)
return(pixel_list);
(void) ResetMagickMemory((void *) pixel_list,0,sizeof(*pixel_list));
pixel_list->length=width*height;
for (i=0; i < ListChannels; i++)
{
pixel_list->lists[i].nodes=(ListNode *) AcquireQuantumMemory(65537UL,
sizeof(*pixel_list->lists[i].nodes));
if (pixel_list->lists[i].nodes == (ListNode *) NULL)
return(DestroyPixelList(pixel_list));
(void) ResetMagickMemory(pixel_list->lists[i].nodes,0,65537UL*
sizeof(*pixel_list->lists[i].nodes));
}
pixel_list->signature=MagickSignature;
return(pixel_list);
}
static PixelList **AcquirePixelListThreadSet(const size_t width,
const size_t height)
{
PixelList
**pixel_list;
register ssize_t
i;
size_t
number_threads;
number_threads=(size_t) GetMagickResourceLimit(ThreadResource);
pixel_list=(PixelList **) AcquireQuantumMemory(number_threads,
sizeof(*pixel_list));
if (pixel_list == (PixelList **) NULL)
return((PixelList **) NULL);
(void) ResetMagickMemory(pixel_list,0,number_threads*sizeof(*pixel_list));
for (i=0; i < (ssize_t) number_threads; i++)
{
pixel_list[i]=AcquirePixelList(width,height);
if (pixel_list[i] == (PixelList *) NULL)
return(DestroyPixelListThreadSet(pixel_list));
}
return(pixel_list);
}
static void AddNodePixelList(PixelList *pixel_list,const ssize_t channel,
const size_t color)
{
register SkipList
*list;
register ssize_t
level;
size_t
search,
update[9];
/*
Initialize the node.
*/
list=pixel_list->lists+channel;
list->nodes[color].signature=pixel_list->signature;
list->nodes[color].count=1;
/*
Determine where it belongs in the list.
*/
search=65536UL;
for (level=list->level; level >= 0; level--)
{
while (list->nodes[search].next[level] < color)
search=list->nodes[search].next[level];
update[level]=search;
}
/*
Generate a pseudo-random level for this node.
*/
for (level=0; ; level++)
{
pixel_list->seed=(pixel_list->seed*42893621L)+1L;
if ((pixel_list->seed & 0x300) != 0x300)
break;
}
if (level > 8)
level=8;
if (level > (list->level+2))
level=list->level+2;
/*
If we're raising the list's level, link back to the root node.
*/
while (level > list->level)
{
list->level++;
update[list->level]=65536UL;
}
/*
Link the node into the skip-list.
*/
do
{
list->nodes[color].next[level]=list->nodes[update[level]].next[level];
list->nodes[update[level]].next[level]=color;
} while (level-- > 0);
}
static void GetMaximumPixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
register SkipList
*list;
register ssize_t
channel;
size_t
color,
maximum;
ssize_t
count;
unsigned short
channels[ListChannels];
/*
Find the maximum value for each of the color.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
count=0;
maximum=list->nodes[color].next[0];
do
{
color=list->nodes[color].next[0];
if (color > maximum)
maximum=color;
count+=list->nodes[color].count;
} while (count < (ssize_t) pixel_list->length);
channels[channel]=(unsigned short) maximum;
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetMeanPixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
MagickRealType
sum;
register SkipList
*list;
register ssize_t
channel;
size_t
color;
ssize_t
count;
unsigned short
channels[ListChannels];
/*
Find the mean value for each of the color.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
count=0;
sum=0.0;
do
{
color=list->nodes[color].next[0];
sum+=(MagickRealType) list->nodes[color].count*color;
count+=list->nodes[color].count;
} while (count < (ssize_t) pixel_list->length);
sum/=pixel_list->length;
channels[channel]=(unsigned short) sum;
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetMedianPixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
register SkipList
*list;
register ssize_t
channel;
size_t
color;
ssize_t
count;
unsigned short
channels[ListChannels];
/*
Find the median value for each of the color.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
count=0;
do
{
color=list->nodes[color].next[0];
count+=list->nodes[color].count;
} while (count <= (ssize_t) (pixel_list->length >> 1));
channels[channel]=(unsigned short) color;
}
GetMagickPixelPacket((const Image *) NULL,pixel);
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetMinimumPixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
register SkipList
*list;
register ssize_t
channel;
size_t
color,
minimum;
ssize_t
count;
unsigned short
channels[ListChannels];
/*
Find the minimum value for each of the color.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
count=0;
color=65536UL;
minimum=list->nodes[color].next[0];
do
{
color=list->nodes[color].next[0];
if (color < minimum)
minimum=color;
count+=list->nodes[color].count;
} while (count < (ssize_t) pixel_list->length);
channels[channel]=(unsigned short) minimum;
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetModePixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
register SkipList
*list;
register ssize_t
channel;
size_t
color,
max_count,
mode;
ssize_t
count;
unsigned short
channels[5];
/*
Make each pixel the 'predominant color' of the specified neighborhood.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
mode=color;
max_count=list->nodes[mode].count;
count=0;
do
{
color=list->nodes[color].next[0];
if (list->nodes[color].count > max_count)
{
mode=color;
max_count=list->nodes[mode].count;
}
count+=list->nodes[color].count;
} while (count < (ssize_t) pixel_list->length);
channels[channel]=(unsigned short) mode;
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetNonpeakPixelList(PixelList *pixel_list,MagickPixelPacket *pixel)
{
register SkipList
*list;
register ssize_t
channel;
size_t
color,
next,
previous;
ssize_t
count;
unsigned short
channels[5];
/*
Finds the non peak value for each of the colors.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
next=list->nodes[color].next[0];
count=0;
do
{
previous=color;
color=next;
next=list->nodes[color].next[0];
count+=list->nodes[color].count;
} while (count <= (ssize_t) (pixel_list->length >> 1));
if ((previous == 65536UL) && (next != 65536UL))
color=next;
else
if ((previous != 65536UL) && (next == 65536UL))
color=previous;
channels[channel]=(unsigned short) color;
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static void GetStandardDeviationPixelList(PixelList *pixel_list,
MagickPixelPacket *pixel)
{
MagickRealType
sum,
sum_squared;
register SkipList
*list;
register ssize_t
channel;
size_t
color;
ssize_t
count;
unsigned short
channels[ListChannels];
/*
Find the standard-deviation value for each of the color.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
color=65536L;
count=0;
sum=0.0;
sum_squared=0.0;
do
{
register ssize_t
i;
color=list->nodes[color].next[0];
sum+=(MagickRealType) list->nodes[color].count*color;
for (i=0; i < (ssize_t) list->nodes[color].count; i++)
sum_squared+=((MagickRealType) color)*((MagickRealType) color);
count+=list->nodes[color].count;
} while (count < (ssize_t) pixel_list->length);
sum/=pixel_list->length;
sum_squared/=pixel_list->length;
channels[channel]=(unsigned short) sqrt(sum_squared-(sum*sum));
}
pixel->red=(MagickRealType) ScaleShortToQuantum(channels[0]);
pixel->green=(MagickRealType) ScaleShortToQuantum(channels[1]);
pixel->blue=(MagickRealType) ScaleShortToQuantum(channels[2]);
pixel->opacity=(MagickRealType) ScaleShortToQuantum(channels[3]);
pixel->index=(MagickRealType) ScaleShortToQuantum(channels[4]);
}
static inline void InsertPixelList(const Image *image,const PixelPacket *pixel,
const IndexPacket *indexes,PixelList *pixel_list)
{
size_t
signature;
unsigned short
index;
index=ScaleQuantumToShort(GetPixelRed(pixel));
signature=pixel_list->lists[0].nodes[index].signature;
if (signature == pixel_list->signature)
pixel_list->lists[0].nodes[index].count++;
else
AddNodePixelList(pixel_list,0,index);
index=ScaleQuantumToShort(GetPixelGreen(pixel));
signature=pixel_list->lists[1].nodes[index].signature;
if (signature == pixel_list->signature)
pixel_list->lists[1].nodes[index].count++;
else
AddNodePixelList(pixel_list,1,index);
index=ScaleQuantumToShort(GetPixelBlue(pixel));
signature=pixel_list->lists[2].nodes[index].signature;
if (signature == pixel_list->signature)
pixel_list->lists[2].nodes[index].count++;
else
AddNodePixelList(pixel_list,2,index);
index=ScaleQuantumToShort(GetPixelOpacity(pixel));
signature=pixel_list->lists[3].nodes[index].signature;
if (signature == pixel_list->signature)
pixel_list->lists[3].nodes[index].count++;
else
AddNodePixelList(pixel_list,3,index);
if (image->colorspace == CMYKColorspace)
index=ScaleQuantumToShort(GetPixelIndex(indexes));
signature=pixel_list->lists[4].nodes[index].signature;
if (signature == pixel_list->signature)
pixel_list->lists[4].nodes[index].count++;
else
AddNodePixelList(pixel_list,4,index);
}
static inline MagickRealType MagickAbsoluteValue(const MagickRealType x)
{
if (x < 0)
return(-x);
return(x);
}
static void ResetPixelList(PixelList *pixel_list)
{
int
level;
register ListNode
*root;
register SkipList
*list;
register ssize_t
channel;
/*
Reset the skip-list.
*/
for (channel=0; channel < 5; channel++)
{
list=pixel_list->lists+channel;
root=list->nodes+65536UL;
list->level=0;
for (level=0; level < 9; level++)
root->next[level]=65536UL;
}
pixel_list->seed=pixel_list->signature++;
}
MagickExport Image *StatisticImage(const Image *image,const StatisticType type,
const size_t width,const size_t height,ExceptionInfo *exception)
{
Image
*statistic_image;
statistic_image=StatisticImageChannel(image,DefaultChannels,type,width,
height,exception);
return(statistic_image);
}
MagickExport Image *StatisticImageChannel(const Image *image,
const ChannelType channel,const StatisticType type,const size_t width,
const size_t height,ExceptionInfo *exception)
{
#define StatisticImageTag "Statistic/Image"
CacheView
*image_view,
*statistic_view;
Image
*statistic_image;
MagickBooleanType
status;
MagickOffsetType
progress;
PixelList
**restrict pixel_list;
size_t
neighbor_height,
neighbor_width;
ssize_t
y;
/*
Initialize statistics image attributes.
*/
assert(image != (Image *) NULL);
assert(image->signature == MagickSignature);
if (image->debug != MagickFalse)
(void) LogMagickEvent(TraceEvent,GetMagickModule(),"%s",image->filename);
assert(exception != (ExceptionInfo *) NULL);
assert(exception->signature == MagickSignature);
statistic_image=CloneImage(image,image->columns,image->rows,MagickTrue,
exception);
if (statistic_image == (Image *) NULL)
return((Image *) NULL);
if (SetImageStorageClass(statistic_image,DirectClass) == MagickFalse)
{
InheritException(exception,&statistic_image->exception);
statistic_image=DestroyImage(statistic_image);
return((Image *) NULL);
}
neighbor_width=width == 0 ? GetOptimalKernelWidth2D((double) width,0.5) :
width;
neighbor_height=height == 0 ? GetOptimalKernelWidth2D((double) height,0.5) :
height;
pixel_list=AcquirePixelListThreadSet(neighbor_width,neighbor_height);
if (pixel_list == (PixelList **) NULL)
{
statistic_image=DestroyImage(statistic_image);
ThrowImageException(ResourceLimitError,"MemoryAllocationFailed");
}
/*
Make each pixel the min / max / median / mode / etc. of the neighborhood.
*/
status=MagickTrue;
progress=0;
image_view=AcquireVirtualCacheView(image,exception);
statistic_view=AcquireAuthenticCacheView(statistic_image,exception);
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp parallel for schedule(static,4) shared(progress,status) \
magick_threads(image,statistic_image,statistic_image->rows,1)
#endif
for (y=0; y < (ssize_t) statistic_image->rows; y++)
{
const int
id = GetOpenMPThreadId();
register const IndexPacket
*restrict indexes;
register const PixelPacket
*restrict p;
register IndexPacket
*restrict statistic_indexes;
register PixelPacket
*restrict q;
register ssize_t
x;
if (status == MagickFalse)
continue;
p=GetCacheViewVirtualPixels(image_view,-((ssize_t) neighbor_width/2L),y-
(ssize_t) (neighbor_height/2L),image->columns+neighbor_width,
neighbor_height,exception);
q=QueueCacheViewAuthenticPixels(statistic_view,0,y,statistic_image->columns, 1,exception);
if ((p == (const PixelPacket *) NULL) || (q == (PixelPacket *) NULL))
{
status=MagickFalse;
continue;
}
indexes=GetCacheViewVirtualIndexQueue(image_view);
statistic_indexes=GetCacheViewAuthenticIndexQueue(statistic_view);
for (x=0; x < (ssize_t) statistic_image->columns; x++)
{
MagickPixelPacket
pixel;
register const IndexPacket
*restrict s;
register const PixelPacket
*restrict r;
register ssize_t
u,
v;
r=p;
s=indexes+x;
ResetPixelList(pixel_list[id]);
for (v=0; v < (ssize_t) neighbor_height; v++)
{
for (u=0; u < (ssize_t) neighbor_width; u++)
InsertPixelList(image,r+u,s+u,pixel_list[id]);
r+=image->columns+neighbor_width;
s+=image->columns+neighbor_width;
}
GetMagickPixelPacket(image,&pixel);
SetMagickPixelPacket(image,p+neighbor_width*neighbor_height/2,indexes+
neighbor_width*neighbor_height/2+x,&pixel);
switch (type)
{
case GradientStatistic:
{
MagickPixelPacket
maximum,
minimum;
GetMinimumPixelList(pixel_list[id],&pixel);
minimum=pixel;
GetMaximumPixelList(pixel_list[id],&pixel);
maximum=pixel;
pixel.red=MagickAbsoluteValue(maximum.red-minimum.red);
pixel.green=MagickAbsoluteValue(maximum.green-minimum.green);
pixel.blue=MagickAbsoluteValue(maximum.blue-minimum.blue);
pixel.opacity=MagickAbsoluteValue(maximum.opacity-minimum.opacity);
if (image->colorspace == CMYKColorspace)
pixel.index=MagickAbsoluteValue(maximum.index-minimum.index);
break;
}
case MaximumStatistic:
{
GetMaximumPixelList(pixel_list[id],&pixel);
break;
}
case MeanStatistic:
{
GetMeanPixelList(pixel_list[id],&pixel);
break;
}
case MedianStatistic:
default:
{
GetMedianPixelList(pixel_list[id],&pixel);
break;
}
case MinimumStatistic:
{
GetMinimumPixelList(pixel_list[id],&pixel);
break;
}
case ModeStatistic:
{
GetModePixelList(pixel_list[id],&pixel);
break;
}
case NonpeakStatistic:
{
GetNonpeakPixelList(pixel_list[id],&pixel);
break;
}
case StandardDeviationStatistic:
{
GetStandardDeviationPixelList(pixel_list[id],&pixel);
break;
}
}
if ((channel & RedChannel) != 0)
SetPixelRed(q,ClampToQuantum(pixel.red));
if ((channel & GreenChannel) != 0)
SetPixelGreen(q,ClampToQuantum(pixel.green));
if ((channel & BlueChannel) != 0)
SetPixelBlue(q,ClampToQuantum(pixel.blue));
if ((channel & OpacityChannel) != 0)
SetPixelOpacity(q,ClampToQuantum(pixel.opacity));
if (((channel & IndexChannel) != 0) &&
(image->colorspace == CMYKColorspace))
SetPixelIndex(statistic_indexes+x,ClampToQuantum(pixel.index));
p++;
q++;
}
if (SyncCacheViewAuthenticPixels(statistic_view,exception) == MagickFalse)
status=MagickFalse;
if (image->progress_monitor != (MagickProgressMonitor) NULL)
{
MagickBooleanType
proceed;
#if defined(MAGICKCORE_OPENMP_SUPPORT)
#pragma omp critical (MagickCore_StatisticImage)
#endif
proceed=SetImageProgress(image,StatisticImageTag,progress++,
image->rows);
if (proceed == MagickFalse)
status=MagickFalse;
}
}
statistic_view=DestroyCacheView(statistic_view);
image_view=DestroyCacheView(image_view);
pixel_list=DestroyPixelListThreadSet(pixel_list);
if (status == MagickFalse)
statistic_image=DestroyImage(statistic_image);
return(statistic_image);
}
|
triplet_iw.c | /* Copyright (C) 2016 Atsushi Togo */
/* All rights reserved. */
/* This file is part of phonopy. */
/* Redistribution and use in source and binary forms, with or without */
/* modification, are permitted provided that the following conditions */
/* are met: */
/* * Redistributions of source code must retain the above copyright */
/* notice, this list of conditions and the following disclaimer. */
/* * Redistributions in binary form must reproduce the above copyright */
/* notice, this list of conditions and the following disclaimer in */
/* the documentation and/or other materials provided with the */
/* distribution. */
/* * Neither the name of the phonopy project nor the names of its */
/* contributors may be used to endorse or promote products derived */
/* from this software without specific prior written permission. */
/* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS */
/* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT */
/* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS */
/* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE */
/* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, */
/* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; */
/* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER */
/* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT */
/* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN */
/* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
/* POSSIBILITY OF SUCH DAMAGE. */
#include <math.h>
#include "grgrid.h"
#include "phonoc_utils.h"
#include "triplet.h"
#include "triplet_iw.h"
#include "tetrahedron_method.h"
static void set_freq_vertices(double freq_vertices[3][24][4],
const double *frequencies1,
const double *frequencies2,
const long vertices[2][24][4],
const long num_band1,
const long num_band2,
const long b1,
const long b2,
const long tp_type);
static long set_g(double g[3],
const double f0,
const double freq_vertices[3][24][4],
const long max_i);
static long in_tetrahedra(const double f0, const double freq_vertices[24][4]);
static void get_triplet_tetrahedra_vertices(
long vertices[2][24][4],
const long tp_relative_grid_address[2][24][4][3],
const long triplet[3],
const ConstBZGrid *bzgrid);
static void
get_neighboring_grid_points_type1(long *neighboring_grid_points,
const long grid_point,
const long (*relative_grid_address)[3],
const long num_relative_grid_address,
const ConstBZGrid *bzgrid);
static void
get_neighboring_grid_points_type2(long *neighboring_grid_points,
const long grid_point,
const long (*relative_grid_address)[3],
const long num_relative_grid_address,
const ConstBZGrid *bzgrid);
void
tpi_get_integration_weight(double *iw,
char *iw_zero,
const double *frequency_points,
const long num_band0,
const long tp_relative_grid_address[2][24][4][3],
const long triplets[3],
const long num_triplets,
const ConstBZGrid *bzgrid,
const double *frequencies1,
const long num_band1,
const double *frequencies2,
const long num_band2,
const long tp_type,
const long openmp_per_bands)
{
long max_i, j, b1, b2, b12, num_band_prod, adrs_shift;
long vertices[2][24][4];
double g[3];
double freq_vertices[3][24][4];
get_triplet_tetrahedra_vertices(vertices,
tp_relative_grid_address,
triplets,
bzgrid);
num_band_prod = num_triplets * num_band0 * num_band1 * num_band2;
/* tp_type: Type of integration weights stored */
/* */
/* g0 -> \delta(f0 - (-f1 + f2)) */
/* g1 -> \delta(f0 - (f1 - f2)) */
/* g2 -> \delta(f0 - (f1 + f2)) */
/* */
/* tp_type = 2: (g[2], g[0] - g[1]) mainly for ph-ph */
/* tp_type = 3: (g[2], g[0] - g[1], g[0] + g[1] + g[2]) mainly for ph-ph */
/* tp_type = 4: (g[0]) mainly for el-ph phonon decay, */
/* f0: ph, f1: el_i, f2: el_f */
if ((tp_type == 2) || (tp_type == 3)) {
max_i = 3;
}
if (tp_type == 4) {
max_i = 1;
}
#pragma omp parallel for private(j, b1, b2, adrs_shift, g, freq_vertices) if (openmp_per_bands)
for (b12 = 0; b12 < num_band1 * num_band2; b12++) {
b1 = b12 / num_band2;
b2 = b12 % num_band2;
set_freq_vertices(freq_vertices, frequencies1, frequencies2,
vertices, num_band1, num_band2, b1, b2, tp_type);
for (j = 0; j < num_band0; j++) {
adrs_shift = j * num_band1 * num_band2 + b1 * num_band2 + b2;
iw_zero[adrs_shift] = set_g(g, frequency_points[j], freq_vertices, max_i);
if (tp_type == 2) {
iw[adrs_shift] = g[2];
adrs_shift += num_band_prod;
iw[adrs_shift] = g[0] - g[1];
}
if (tp_type == 3) {
iw[adrs_shift] = g[2];
adrs_shift += num_band_prod;
iw[adrs_shift] = g[0] - g[1];
adrs_shift += num_band_prod;
iw[adrs_shift] = g[0] + g[1] + g[2];
}
if (tp_type == 4) {
iw[adrs_shift] = g[0];
}
}
}
}
void tpi_get_integration_weight_with_sigma(double *iw,
char *iw_zero,
const double sigma,
const double cutoff,
const double *frequency_points,
const long num_band0,
const long triplet[3],
const long const_adrs_shift,
const double *frequencies,
const long num_band,
const long tp_type,
const long openmp_per_bands)
{
long j, b12, b1, b2, adrs_shift;
double f0, f1, f2, g0, g1, g2;
#pragma omp parallel for private(j, b1, b2, f0, f1, f2, g0, g1, g2, adrs_shift) if (openmp_per_bands)
for (b12 = 0; b12 < num_band * num_band; b12++) {
b1 = b12 / num_band;
b2 = b12 % num_band;
f1 = frequencies[triplet[1] * num_band + b1];
f2 = frequencies[triplet[2] * num_band + b2];
for (j = 0; j < num_band0; j++) {
f0 = frequency_points[j];
adrs_shift = j * num_band * num_band + b1 * num_band + b2;
if ((tp_type == 2) || (tp_type == 3)) {
if (cutoff > 0 &&
fabs(f0 + f1 - f2) > cutoff &&
fabs(f0 - f1 + f2) > cutoff &&
fabs(f0 - f1 - f2) > cutoff) {
iw_zero[adrs_shift] = 1;
g0 = 0;
g1 = 0;
g2 = 0;
} else {
iw_zero[adrs_shift] = 0;
g0 = phonoc_gaussian(f0 + f1 - f2, sigma);
g1 = phonoc_gaussian(f0 - f1 + f2, sigma);
g2 = phonoc_gaussian(f0 - f1 - f2, sigma);
}
if (tp_type == 2) {
iw[adrs_shift] = g2;
adrs_shift += const_adrs_shift;
iw[adrs_shift] = g0 - g1;
}
if (tp_type == 3) {
iw[adrs_shift] = g2;
adrs_shift += const_adrs_shift;
iw[adrs_shift] = g0 - g1;
adrs_shift += const_adrs_shift;
iw[adrs_shift] = g0 + g1 + g2;
}
}
if (tp_type == 4) {
if (cutoff > 0 && fabs(f0 + f1 - f2) > cutoff) {
iw_zero[adrs_shift] = 1;
iw[adrs_shift] = 0;
} else {
iw_zero[adrs_shift] = 0;
iw[adrs_shift] = phonoc_gaussian(f0 + f1 - f2, sigma);
}
}
}
}
}
void
tpi_get_neighboring_grid_points(long *neighboring_grid_points,
const long grid_point,
const long (*relative_grid_address)[3],
const long num_relative_grid_address,
const ConstBZGrid *bzgrid)
{
if (bzgrid->type == 1) {
get_neighboring_grid_points_type1(neighboring_grid_points,
grid_point,
relative_grid_address,
num_relative_grid_address,
bzgrid);
} else {
get_neighboring_grid_points_type2(neighboring_grid_points,
grid_point,
relative_grid_address,
num_relative_grid_address,
bzgrid);
}
}
static void set_freq_vertices(double freq_vertices[3][24][4],
const double *frequencies1,
const double *frequencies2,
const long vertices[2][24][4],
const long num_band1,
const long num_band2,
const long b1,
const long b2,
const long tp_type)
{
long i, j;
double f1, f2;
for (i = 0; i < 24; i++) {
for (j = 0; j < 4; j++) {
f1 = frequencies1[vertices[0][i][j] * num_band1 + b1];
f2 = frequencies2[vertices[1][i][j] * num_band2 + b2];
if ((tp_type == 2) || (tp_type == 3)) {
if (f1 < 0) {f1 = 0;}
if (f2 < 0) {f2 = 0;}
freq_vertices[0][i][j] = -f1 + f2;
freq_vertices[1][i][j] = f1 - f2;
freq_vertices[2][i][j] = f1 + f2;
} else {
freq_vertices[0][i][j] = -f1 + f2;
}
}
}
}
/* Integration weight g is calculated. */
/* iw_zero = 1 means g[0] to g[max_i - 1] are all zero. */
/* max_i depends on what we compute, e.g., ph-ph lifetime, */
/* ph-ph collision matrix, and el-ph relaxation time. */
/* iw_zero is definitely determined by in_tetrahedra in case that */
/* f0 is out of the tetrahedra. */
/* iw_zero=1 information can be used to omit to compute particles */
/* interaction strength that is often heaviest part in throughout */
/* calculation. */
static long set_g(double g[3],
const double f0,
const double freq_vertices[3][24][4],
const long max_i)
{
long i, iw_zero;
iw_zero = 1;
for (i = 0; i < max_i; i++) {
if (in_tetrahedra(f0, freq_vertices[i])) {
g[i] = thm_get_integration_weight(f0, freq_vertices[i], 'I');
iw_zero = 0;
} else {
g[i] = 0;
}
}
return iw_zero;
}
static long in_tetrahedra(const double f0, const double freq_vertices[24][4])
{
long i, j;
double fmin, fmax;
fmin = freq_vertices[0][0];
fmax = freq_vertices[0][0];
for (i = 0; i < 24; i++) {
for (j = 0; j < 4; j++) {
if (fmin > freq_vertices[i][j]) {
fmin = freq_vertices[i][j];
}
if (fmax < freq_vertices[i][j]) {
fmax = freq_vertices[i][j];
}
}
}
if (fmin > f0 || fmax < f0) {
return 0;
} else {
return 1;
}
}
static void get_triplet_tetrahedra_vertices(long vertices[2][24][4],
const long tp_relative_grid_address[2][24][4][3],
const long triplet[3],
const ConstBZGrid *bzgrid)
{
long i, j;
for (i = 0; i < 2; i++) {
for (j = 0; j < 24; j++) {
tpi_get_neighboring_grid_points(vertices[i][j],
triplet[i + 1],
tp_relative_grid_address[i][j],
4,
bzgrid);
}
}
}
static void
get_neighboring_grid_points_type1(long *neighboring_grid_points,
const long grid_point,
const long (*relative_grid_address)[3],
const long num_relative_grid_address,
const ConstBZGrid *bzgrid)
{
long bzmesh[3], bz_address[3];
long i, j, bz_gp, prod_bz_mesh;
for (i = 0; i < 3; i++) {
bzmesh[i] = bzgrid->D_diag[i] * 2;
}
prod_bz_mesh = bzmesh[0] * bzmesh[1] * bzmesh[2];
for (i = 0; i < num_relative_grid_address; i++) {
for (j = 0; j < 3; j++) {
bz_address[j] = bzgrid->addresses[grid_point][j]
+ relative_grid_address[i][j];
}
bz_gp = bzgrid->gp_map[grg_get_grid_index(bz_address, bzmesh)];
if (bz_gp == prod_bz_mesh) {
neighboring_grid_points[i] =
grg_get_grid_index(bz_address, bzgrid->D_diag);
} else {
neighboring_grid_points[i] = bz_gp;
}
}
}
static void
get_neighboring_grid_points_type2(long *neighboring_grid_points,
const long grid_point,
const long (*relative_grid_address)[3],
const long num_relative_grid_address,
const ConstBZGrid *bzgrid)
{
long bz_address[3];
long i, j, gp;
for (i = 0; i < num_relative_grid_address; i++) {
for (j = 0; j < 3; j++) {
bz_address[j] = bzgrid->addresses[grid_point][j]
+ relative_grid_address[i][j];
}
gp = grg_get_grid_index(bz_address, bzgrid->D_diag);
neighboring_grid_points[i] = bzgrid->gp_map[gp];
if (bzgrid->gp_map[gp + 1] - bzgrid->gp_map[gp] > 1) {
for (j = bzgrid->gp_map[gp]; j < bzgrid->gp_map[gp + 1]; j++) {
if (bz_address[0] == bzgrid->addresses[j][0]
&& bz_address[1] == bzgrid->addresses[j][1]
&& bz_address[2] == bzgrid->addresses[j][2]) {
neighboring_grid_points[i] = j;
break;
}
}
}
}
}
|
cones.c | #include "cones.h"
#define CONE_RATE (2)
#define CONE_TOL (1e-7)
#define EXP_CONE_MAX_ITERS (100)
#ifdef LAPACK_LIB_FOUND
/* underscore for blas / lapack, single or double precision */
#ifdef NOBLASUNDERSCORE
#ifndef FLOAT
#define BLAS(x) d ## x
#else
#define BLAS(x) s ## x
#endif
#else
#ifndef FLOAT
#define BLAS(x) d ## x ## _
#else
#define BLAS(x) s ## x ## _
#endif
#endif
#ifdef MATLAB_MEX_FILE
typedef ptrdiff_t blasint;
#elif defined BLAS64
#include <stdint.h>
typedef int64_t blasint;
#else
typedef int blasint;
#endif
void BLAS(syevr)(char* jobz, char* range, char* uplo, blasint* n, pfloat* a, blasint* lda, pfloat* vl,
pfloat* vu, blasint* il, blasint* iu, pfloat* abstol, blasint* m, pfloat* w, pfloat* z, blasint* ldz,
blasint* isuppz, pfloat* work, blasint* lwork, blasint* iwork, blasint* liwork, blasint* info);
void BLAS(syr)(const char *uplo, const blasint *n, const pfloat *alpha, const pfloat *x, const blasint *incx,
pfloat *a, const blasint *lda);
void BLAS(axpy)(const blasint *n, const pfloat *alpha, const pfloat *dx, const blasint *incx, pfloat *dy,
const blasint *incy);
/* private data to help cone projection step */
static struct ConeData_t {
/* workspace for eigenvector decompositions: */
pfloat * Xs, *Z, *e, *work;
blasint *iwork, lwork, liwork;
}c;
#endif
static timer coneTimer;
static pfloat totalConeTime;
/*
* boundaries will contain array of indices of rows of A corresponding to
* cone boundaries, boundaries[0] is starting index for cones of size strictly larger than 1
* returns length of boundaries array, boundaries malloc-ed here so should be freed
*/
idxint getConeBoundaries(Cone * k, idxint ** boundaries) {
idxint i, count = 0;
idxint len = 1 + k->qsize + k->ssize + k->ed + k->ep;
idxint * b = scs_malloc(sizeof(idxint) * len);
b[count] = k->f + k->l;
count += 1;
if (k->qsize > 0) {
memcpy(&b[count], k->q, k->qsize * sizeof(idxint));
}
count += k->qsize;
for (i = 0; i < k->ssize; ++i) {
b[count + i] = k->s[i] * k->s[i];
}
count += k->ssize;
for (i = 0; i < k->ep + k->ed; ++i) {
b[count + i] = 3;
}
count += k->ep + k->ed;
*boundaries = b;
return len;
}
idxint getFullConeDims(Cone * k) {
idxint i, c = 0;
if (k->f)
c += k->f;
if (k->l)
c += k->l;
if (k->qsize && k->q) {
for (i = 0; i < k->qsize; ++i) {
c += k->q[i];
}
}
if (k->ssize && k->s) {
for (i = 0; i < k->ssize; ++i) {
c += k->s[i] * k->s[i];
}
}
if (k->ed)
c += 3 * k->ed;
if (k->ep)
c += 3 * k->ep;
return c;
}
idxint validateCones(Data * d, Cone * k) {
idxint i;
if (k->f && k->f < 0) {
scs_printf("free cone error\n");
return -1;
}
if (k->l && k->l < 0) {
scs_printf("lp cone error\n");
return -1;
}
if (k->qsize && k->q) {
for (i = 0; i < k->qsize; ++i) {
if (k->q[i] < 0) {
scs_printf("soc cone error\n");
return -1;
}
}
}
if (k->ssize && k->s) {
for (i = 0; i < k->ssize; ++i) {
if (k->s[i] < 0) {
scs_printf("sd cone error\n");
return -1;
}
}
}
if (k->ed && k->ed < 0) {
scs_printf("ep cone error\n");
return -1;
}
if (k->ep && k->ep < 0) {
scs_printf("ed cone error\n");
return -1;
}
if (getFullConeDims(k) != d->m) {
scs_printf("cone dimensions %i not equal to num rows in A = m = %i\n", (int) getFullConeDims(k), (int) d->m);
return -1;
}
return 0;
}
char * getConeSummary(Info * info) {
char * str = scs_malloc(sizeof(char) * 64);
sprintf(str, "\tCones: avg projection time: %1.2es\n", totalConeTime / (info->iter + 1) / 1e3);
totalConeTime = 0.0;
return str;
}
void finishCone() {
#ifdef LAPACK_LIB_FOUND
if (c.Xs)
scs_free(c.Xs);
if (c.Z)
scs_free(c.Z);
if (c.e)
scs_free(c.e);
if (c.work)
scs_free(c.work);
if (c.iwork)
scs_free(c.iwork);
#endif
}
char * getConeHeader(Cone * k) {
char * tmp = scs_malloc(sizeof(char) * 512);
idxint i, socVars, socBlks, sdVars, sdBlks, expPvars, expDvars;
sprintf(tmp, "Cones:");
if (k->f) {
sprintf(tmp + strlen(tmp), "\tprimal zero / dual free vars: %i\n", (int) k->f);
}
if (k->l) {
sprintf(tmp + strlen(tmp), "\tlinear vars: %i\n", (int) k->l);
}
socVars = 0;
socBlks = 0;
if (k->qsize && k->q) {
socBlks = k->qsize;
for (i = 0; i < k->qsize; i++) {
socVars += k->q[i];
}
sprintf(tmp + strlen(tmp), "\tsoc vars: %i, soc blks: %i\n", (int) socVars, (int) socBlks);
}
sdVars = 0;
sdBlks = 0;
if (k->ssize && k->s) {
sdBlks = k->ssize;
for (i = 0; i < k->ssize; i++) {
sdVars += k->s[i] * k->s[i];
}
sprintf(tmp + strlen(tmp), "\tsd vars: %i, sd blks: %i\n", (int) sdVars, (int) sdBlks);
}
if (k->ep || k->ed) {
expPvars = k->ep ? 3 * k->ep : 0;
expDvars = k->ed ? 3 * k->ed : 0;
sprintf(tmp + strlen(tmp), "\texp vars: %i, dual exp vars: %i\n", (int) expPvars, (int) expDvars);
}
return tmp;
}
idxint isSimpleSemiDefiniteCone(idxint * s, idxint ssize) {
idxint i;
for (i = 0; i < ssize; i++) {
if (s[i] >= 3) {
return 0; /* false */
}
}
return 1; /* true */
}
pfloat expNewtonOneD(pfloat rho, pfloat y_hat, pfloat z_hat) {
pfloat t = MAX(-z_hat, 1e-6);
pfloat f, fp;
idxint i;
for (i = 0; i < EXP_CONE_MAX_ITERS; ++i) {
f = t * (t + z_hat) / rho / rho - y_hat / rho + log(t / rho) + 1;
fp = (2 * t + z_hat) / rho / rho + 1 / t;
t = t - f / fp;
if (t <= -z_hat) {
return 0;
} else if (t <= 0) {
return z_hat;
} else if ( ABS(f) < CONE_TOL) {
break;
}
}
return t + z_hat;
}
void expSolveForXWithRho(pfloat * v, pfloat * x, pfloat rho) {
x[2] = expNewtonOneD(rho, v[1], v[2]);
x[1] = (x[2] - v[2]) * x[2] / rho;
x[0] = v[0] - rho;
}
pfloat expCalcGrad(pfloat * v, pfloat * x, pfloat rho) {
expSolveForXWithRho(v, x, rho);
if (x[1] <= 1e-12) {
return x[0];
} else {
return x[0] + x[1] * log(x[1] / x[2]);
}
}
void expGetRhoUb(pfloat * v, pfloat * x, pfloat * ub, pfloat * lb) {
*lb = 0;
*ub = 0.125;
while (expCalcGrad(v, x, *ub) > 0) {
*lb = *ub;
(*ub) *= 2;
}
}
/* project onto the exponential cone, v has dimension *exactly* 3 */
static idxint projExpCone(pfloat * v, idxint iter) {
idxint i;
pfloat ub, lb, rho, g, x[3];
pfloat r = v[0], s = v[1], t = v[2];
pfloat tol = CONE_TOL; /* iter < 0 ? CONE_TOL : MAX(CONE_TOL, 1 / POWF((iter + 1), CONE_RATE)); */
/* v in cl(Kexp) */
if ((s * exp(r / s) <= t && s > 0) || (r <= 0 && s == 0 && t >= 0)) {
return 0;
}
/* -v in Kexp^* */
if ((-r < 0 && r * exp(s / r) <= -exp(1) * t) || (-r == 0 && -s >= 0 && -t >= 0)) {
memset(v, 0, 3 * sizeof(pfloat));
return 0;
}
/* special case with analytical solution */
if (r < 0 && s < 0) {
v[1] = 0.0;
v[2] = MAX(v[2], 0);
return 0;
}
/* iterative procedure to find projection, bisects on dual variable: */
expGetRhoUb(v, x, &ub, &lb); /* get starting upper and lower bounds */
for (i = 0; i < EXP_CONE_MAX_ITERS; ++i) {
rho = (ub + lb) / 2; /* halfway between upper and lower bounds */
g = expCalcGrad(v, x, rho); /* calculates gradient wrt dual var */
if (g > 0) {
lb = rho;
} else {
ub = rho;
}
if (ub - lb < tol) {
break;
}
}
/*
#ifdef EXTRAVERBOSE
scs_printf("exponential cone proj iters %i\n", i);
#endif
*/
v[0] = x[0];
v[1] = x[1];
v[2] = x[2];
return 0;
}
idxint initCone(Cone * k) {
#ifdef LAPACK_LIB_FOUND
idxint i;
blasint nMax = 0;
pfloat eigTol = 1e-8;
blasint negOne = -1;
blasint m = 0;
blasint info;
pfloat wkopt;
c.Xs = NULL;
c.Z = NULL;
c.e = NULL;
c.work = NULL;
c.iwork = NULL;
#endif
totalConeTime = 0.0;
#ifdef EXTRAVERBOSE
scs_printf("initCone\n");
#ifdef MATLAB_MEX_FILE
mexEvalString("drawnow;");
#endif
#endif
if (k->ssize && k->s) {
if (isSimpleSemiDefiniteCone(k->s, k->ssize)) {
return 0;
}
#ifdef LAPACK_LIB_FOUND
/* eigenvector decomp workspace */
for (i = 0; i < k->ssize; ++i) {
if (k->s[i] > nMax) {
nMax = (blasint) k->s[i];
}
}
c.Xs = scs_calloc(nMax * nMax, sizeof(pfloat));
c.Z = scs_calloc(nMax * nMax, sizeof(pfloat));
c.e = scs_calloc(nMax, sizeof(pfloat));
BLAS(syevr)("Vectors", "All", "Upper", &nMax, c.Xs, &nMax, NULL, NULL, NULL, NULL,
&eigTol, &m, c.e, c.Z, &nMax, NULL, &wkopt, &negOne, &(c.liwork), &negOne, &info);
if (info != 0) {
scs_printf("FATAL: syevr failure, info = %i\n", (int) info);
return -1;
}
c.lwork = (blasint) (wkopt + 0.01); /* 0.01 for int casting safety */
c.work = scs_malloc(c.lwork * sizeof(pfloat));
c.iwork = scs_malloc(c.liwork * sizeof(blasint));
if (!c.Xs || !c.Z || !c.e || !c.work || !c.iwork) {
return -1;
}
#else
scs_printf("FATAL: Cannot solve SDPs with > 2x2 matrices without linked blas+lapack libraries\n");
scs_printf("Edit scs.mk to point to blas+lapack libray locations\n");
return -1;
#endif
}
#ifdef EXTRAVERBOSE
scs_printf("initCone complete\n");
#ifdef MATLAB_MEX_FILE
mexEvalString("drawnow;");
#endif
#endif
return 0;
}
idxint project2By2Sdc(pfloat *X) {
pfloat a, b, d, l1, l2, x1, x2, rad;
a = X[0];
b = 0.5 * (X[1] + X[2]);
d = X[3];
rad = SQRTF((a - d) * (a - d) + 4 * b * b);
/* l1 >= l2 always, since rad >= 0 */
l1 = 0.5 * (a + d + rad);
l2 = 0.5 * (a + d - rad);
if (l2 >= 0) { /* both positive, just symmetrize */
X[1] = b;
X[2] = b;
return 0;
}
if (l1 <= 0) { /* both negative, set to 0 */
X[0] = 0;
X[1] = 0;
X[2] = 0;
X[3] = 0;
return 0;
}
/* l1 pos, l2 neg */
x1 = 1 / SQRTF(1 + (l1 - a) * (l1 - a) / b / b);
x2 = x1 * (l1 - a) / b;
X[0] = l1 * x1 * x1;
X[1] = l1 * x1 * x2;
X[2] = X[1];
X[3] = l1 * x2 * x2;
return 0;
}
static idxint projSemiDefiniteCone(pfloat *X, idxint n, idxint iter) {
/* project onto the positive semi-definite cone */
#ifdef LAPACK_LIB_FOUND
idxint i, j;
blasint one = 1;
blasint m = 0;
blasint nb = (blasint) n;
pfloat * Xs = c.Xs;
pfloat * Z = c.Z;
pfloat * e = c.e;
pfloat * work = c.work;
blasint * iwork = c.iwork;
blasint lwork = c.lwork;
blasint liwork = c.liwork;
pfloat eigTol = CONE_TOL; /* iter < 0 ? CONE_TOL : MAX(CONE_TOL, 1 / POWF(iter + 1, CONE_RATE)); */
pfloat onef = 1.0;
pfloat zero = 0.0;
blasint info;
pfloat vupper;
#endif
if (n == 0) {
return 0;
}
if (n == 1) {
if (X[0] < 0.0) {
X[0] = 0.0;
}
return 0;
}
if (n == 2) {
return project2By2Sdc(X);
}
#ifdef LAPACK_LIB_FOUND
memcpy(Xs, X, n * n * sizeof(pfloat));
/* Xs = X + X', save div by 2 for eigen-recomp */
for (i = 0; i < n; ++i) {
BLAS(axpy)(&nb, &onef, &(X[i]), &nb, &(Xs[i * n]), &one);
}
vupper = MAX(calcNorm(Xs, n * n), 0.001);
/* Solve eigenproblem, reuse workspaces */
BLAS(syevr)("Vectors", "VInterval", "Upper", &nb, Xs, &nb, &zero, &vupper,
NULL, NULL, &eigTol, &m, e, Z, &nb, NULL, work, &lwork, iwork, &liwork, &info);
if (info != 0) {
scs_printf("WARN: LAPACK syevr error, info = %i, attempting to continue...\n", info);
#ifdef EXTRAVERBOSE
scs_printf("syevr input parameter dump:\n");
scs_printf("nb = %li\n", (long) nb);
scs_printf("lwork = %li\n", (long) lwork);
scs_printf("liwork = %li\n", (long) liwork);
scs_printf("vupper = %f\n", vupper);
scs_printf("eigTol = %e\n", eigTol);
printArray(Xs, n * n, "Xs");
printArray(X, n * n, "X");
printArray(e, m, "e");
printArray(Z, m * n, "Z");
#endif
/* return -1; */
}
memset(X, 0, n * n * sizeof(pfloat));
for (i = 0; i < m; ++i) {
pfloat a = e[i] / 2;
BLAS(syr)("Lower", &nb, &a, &(Z[i * n]), &one, X, &nb);
}
/* fill in upper half */
for (i = 0; i < n; ++i) {
for (j = i + 1; j < n; ++j) {
X[i + j * n] = X[j + i * n];
}
}
#else
scs_printf("FAILURE: solving SDP with > 2x2 matrices, but no blas/lapack libraries were linked!\n");
scs_printf("scs will return nonsense!\n");
scaleArray(X, NAN, n);
return -1;
#endif
return 0;
}
/* outward facing cone projection routine, iter is outer algorithm iteration, if iter < 0 then iter is ignored
warm_start contains guess of projection (can be set to NULL) */
idxint projDualCone(pfloat *x, Cone * k, const pfloat * warm_start, idxint iter) {
idxint i;
idxint count = (k->f ? k->f : 0);
#ifdef EXTRAVERBOSE
timer projTimer;
tic(&projTimer);
#endif
tic(&coneTimer);
if (k->l) {
/* project onto positive orthant */
for (i = count; i < count + k->l; ++i) {
if (x[i] < 0.0)
x[i] = 0.0;
/*x[i] = (x[i] < 0.0) ? 0.0 : x[i]; */
}
count += k->l;
#ifdef EXTRAVERBOSE
scs_printf("pos orthant proj time: %1.2es\n", tocq(&projTimer) / 1e3);
tic(&projTimer);
#endif
}
if (k->qsize && k->q) {
/* project onto SOC */
for (i = 0; i < k->qsize; ++i) {
if (k->q[i] == 0) {
continue;
}
if (k->q[i] == 1) {
if (x[count] < 0.0)
x[count] = 0.0;
} else {
pfloat v1 = x[count];
pfloat s = calcNorm(&(x[count + 1]), k->q[i] - 1);
pfloat alpha = (s + v1) / 2.0;
if (s <= v1) { /* do nothing */
} else if (s <= -v1) {
memset(&(x[count]), 0, k->q[i] * sizeof(pfloat));
} else {
x[count] = alpha;
scaleArray(&(x[count + 1]), alpha / s, k->q[i] - 1);
}
}
count += k->q[i];
}
#ifdef EXTRAVERBOSE
scs_printf("SOC proj time: %1.2es\n", tocq(&projTimer) / 1e3);
tic(&projTimer);
#endif
}
if (k->ssize && k->s) {
/* project onto PSD cone */
for (i = 0; i < k->ssize; ++i) {
if (k->s[i] == 0) {
continue;
}
if (projSemiDefiniteCone(&(x[count]), k->s[i], iter) < 0) return -1;
count += (k->s[i]) * (k->s[i]);
}
#ifdef EXTRAVERBOSE
scs_printf("SD proj time: %1.2es\n", tocq(&projTimer) / 1e3);
tic(&projTimer);
#endif
}
if (k->ep) {
pfloat r, s, t;
idxint idx;
/*
* exponential cone is not self dual, if s \in K
* then y \in K^* and so if K is the primal cone
* here we project onto K^*, via Moreau
* \Pi_C^*(y) = y + \Pi_C(-y)
*/
scaleArray(&(x[count]), -1, 3 * k->ep); /* x = -x; */
#ifdef OPENMP
#pragma omp parallel for private(r,s,t,idx)
#endif
for (i = 0; i < k->ep; ++i) {
idx = count + 3 * i;
r = x[idx];
s = x[idx + 1];
t = x[idx + 2];
if (projExpCone(&(x[idx]), iter) < 0) return -1;
x[idx] -= r;
x[idx + 1] -= s;
x[idx + 2] -= t;
}
count += 3 * k->ep;
#ifdef EXTRAVERBOSE
scs_printf("EP proj time: %1.2es\n", tocq(&projTimer) / 1e3);
tic(&projTimer);
#endif
}
if (k->ed) {
/* exponential cone: */
#ifdef OPENMP
#pragma omp parallel for
#endif
for (i = 0; i < k->ed; ++i) {
if (projExpCone(&(x[count + 3 * i]), iter) < 0) return -1;
}
count += 3 * k->ed;
#ifdef EXTRAVERBOSE
scs_printf("ED proj time: %1.2es\n", tocq(&projTimer) / 1e3);
tic(&projTimer);
#endif
}
/* project onto OTHER cones */
totalConeTime += tocq(&coneTimer);
return 0;
}
|
triang.c |
// See the Cormen book for details of the following algorithm
// Accelerating Minimum Cost Polygon Triangulation Code with the TRACO Compiler, Palkowski Bielecki FedCsis 2018
// https://annals-csis.org/Volume_17/drp/pdf/8.pdf
#include<stdio.h>
#include<limits.h>
#include <math.h>
#include <omp.h>
#define min(a,b) (((a)<(b))?(a):(b))
#define MIN(a,b) (((a)<(b))?(a):(b))
#define max(a,b) (((a)>(b))?(a):(b))
#define MAX(a,b) (((a)>(b))?(a):(b))
#define floord(n,d) floor(((double)(n))/((double)(d)))
#define ceild(n,d) ceil(((double)(n))/((double)(d)))
int N = 1500, DIM = 1502;
#include "mem.h"
#define pluto 3
#define pluto2 6
#define traco 2
#define tstile 4
int **points;
// A utility function to find distance between two points in a plane
double dist(int * p1, int * p2)
{
return sqrt((p1[0] - p2[0])*(p1[0] - p2[0]) +
(p1[1] - p2[1])*(p1[1] - p2[1]));
}
// A utility function to find cost of a triangle. The cost is considered
// as perimeter (sum of lengths of all edges) of the triangle
double cost(int i, int j, int k)
{
int* p1 = points[i];
int* p2 = points[j];
int *p3 = points[k];
return dist(p1, p2) + dist(p2, p3) + dist(p3, p1);
}
// Matrix Ai has dimension p[i-1] x p[i] for i = 1..n
double mcTDP(int kind)
{
double** table = memd();
int i, j, k, gap, q;
/* m[i,j] = Minimum number of scalar multiplications needed
to compute the matrix A[i]A[i+1]...A[j] = A[i..j] where
dimension of A[i] is p[i-1] x p[i] */
double start = omp_get_wtime();
// L is chain length.
if(kind==-1){
}
if(kind==1){
printf("original\n");
for (gap = 0; gap < N; gap++)
{
for (j = gap; j < N; j++) // i = j - gap
{
if (gap < 2)
table[j-gap][j] = 0.0;
else
{
table[j-gap][j] = INT_MAX;
for (k = j-gap+1; k < j; k++)
{
table[j-gap][j] = MIN(table[j-gap][j], table[j-gap][k] + table[k][j] + cost(j-gap,j,k));
}
}
}
}
}
if(kind==pluto) // sprawdzic bez maxfuse
{
int t1, t2, t3, t4, t5;
int lb, ub, lbp, ubp, lb2, ub2;
register int lbv, ubv;
printf("pluto\n");
if (N >= 1) {
for (t1=0;t1<=floord(N-1,8);t1++) {
lbp=ceild(t1,2);
ubp=min(floord(N-1,16),t1);
#pragma omp parallel for private(lbv,ubv,t3,t4,t5,t2)
for (t2=lbp;t2<=ubp;t2++) {
if (t1 == t2) {
for (t3=0;t3<=min(1,N-1);t3++) {
for (t4=max(16*t1,t3);t4<=min(N-1,16*t1+15);t4++) {
table[t4-t3][t4] = 0.0;;
}
}
}
for (t3=max(2,16*t1-16*t2);t3<=min(N-1,16*t1-16*t2+15);t3++) {
for (t4=max(16*t2,t3);t4<=min(N-1,16*t2+15);t4++) {
table[t4-t3][t4] = INT_MAX;
for (t5=-t3+t4+1;t5<=t4-1;t5++) {
table[t4-t3][t4] = MIN(table[t4-t3][t4], table[t4-t3][t5] + table[t5][t4] + cost(t4-t3,t4,t5));
}
}
}
}
}
}
}
if(kind==traco)
{
int c1,c3,c4,c5,c9,c11;
int t1, t2, t3, t4, t5,t6;
int lb, ub, lbp, ubp, lb2, ub2;
register int lbv, ubv;
printf("traco\n");
if (N >= 1) {
lbp=0;
ubp=floord(N-1,16);
#pragma omp parallel for private(lbv,ubv,t3,t4,t5,t6)
for (t2=lbp;t2<=ubp;t2++) {
for (t3=t2;t3<=floord(N-1,16);t3++) {
if (t2 == 0) {
for (t4=0;t4<=min(1,N-1);t4++) {
lbv=max(16*t3,t4);
ubv=min(N-1,16*t3+15);
#pragma ivdep
#pragma vector always
for (t5=lbv;t5<=ubv;t5++) {
table[t5-t4][t5] = 0.0;;
}
}
}
for (t4=max(2,16*t2);t4<=min(N-1,16*t2+15);t4++) {
lbv=max(16*t3,t4);
ubv=min(N-1,16*t3+15);
#pragma ivdep
#pragma vector always
for (t5=lbv;t5<=ubv;t5++) {
table[t5-t4][t5] = INT_MAX;
}
}
}
}
// 1 x 32 x 16
for( c1 = 4; c1 < 2 * N - 1; c1 += 1)
#pragma omp parallel for schedule(dynamic, 1) private(c3,c5,c9,c11) shared(c1)
for( c3 = -((c1 - 1) % 2) + 1; c3 <= min(c1 - 4, (2 * N - c1 - 2) / 31); c3 += 2)
for( c5 = 0; c5 <= (c1 - c3 - 4) / 32; c5 += 1)
for( c9 = (c1 + 31 * c3) / 2; c9 <= min(N - 1, ((c1 + 31 * c3) / 2) + 15); c9 += 1)
for( c11 = ((-c1 + c3) / 2) + 16 * c5 + c9 + 1; c11 <= min(c9 - 1, ((-c1 + c3) / 2) + 16 * c5 + c9 + 16); c11 += 1)
table[c9-((c1-c3)/2)][c9] = MIN(table[c9-((c1-c3)/2)][c9], table[c9-((c1-c3)/2)][c11] + table[c11][c9] + cost(c9-((c1-c3)/2),c9,c11));
}
}
if(kind==tstile)
{
int c0,c1,c3,c4,c5,c6,c8,c9,c10, c11;
int t1, t2, t3, t4, t5,t6;
int lb, ub, lbp, ubp, lb2, ub2;
register int lbv, ubv;
printf("tstile\n");
if (N >= 1) {
lbp=0;
ubp=floord(N-1,16);
#pragma omp parallel for private(lbv,ubv,t3,t4,t5,t6)
for (t2=lbp;t2<=ubp;t2++) {
for (t3=t2;t3<=floord(N-1,16);t3++) {
if (t2 == 0) {
for (t4=0;t4<=min(1,N-1);t4++) {
lbv=max(16*t3,t4);
ubv=min(N-1,16*t3+15);
#pragma ivdep
#pragma vector always
for (t5=lbv;t5<=ubv;t5++) {
table[t5-t4][t5] = 0.0;;
}
}
}
for (t4=max(2,16*t2);t4<=min(N-1,16*t2+15);t4++) {
lbv=max(16*t3,t4);
ubv=min(N-1,16*t3+15);
#pragma ivdep
#pragma vector always
for (t5=lbv;t5<=ubv;t5++) {
table[t5-t4][t5] = INT_MAX;
}
}
}
}
// 16x16x16
for( c0 = 0; c0 <= floord(N - 1, 16); c0 += 1)
#pragma omp parallel for schedule(dynamic, 1) private(c1,c3,c4,c6,c10) shared(c0)
for( c1 = max(0, c0 - (N + 13) / 16 + 1); c1 <= c0; c1 += 1)
for( c3 = max(16 * c0 + 16 * c1, 16 * c0 - 16 * c1 + 4); c3 <= min(min(2 * N - 16 * c0 + 16 * c1 - 2, N + 16 * c1 + 14), 16 * c0 + 16 * c1 + 45); c3 += 1)
for( c4 = max(c0 - c1, -2 * c1 + (c3 + 3) / 16 - 2); c4 <= min(min((N - 2) / 16, -c1 + (c3 - 1) / 16), -c1 + (16 * c0 + 16 * c1 + c3 + 13) / 32); c4 += 1)
for( c6 = max(max(max(max(2, 16 * c1), -N + c3 + 1), -8 * c0 + 8 * c1 + c3 / 2 - 7), -8 * c4 + (c3 + 1) / 2 - 7); c6 <= min(min(16 * c1 + 15, c3 - 16 * c4 - 1), -8 * c0 + 8 * c1 + c3 / 2); c6 += 1)
for( c10 = max(16 * c4, c3 - 2 * c6 + 1); c10 <= min(16 * c4 + 15, c3 - c6 - 1); c10 += 1)
table[(c3-c6)-c6][(c3-c6)] = MIN(table[(c3-c6)-c6][(c3-c6)], table[(c3-c6)-c6][c10] + table[c10][(c3-c6)] + cost((c3-c6)-c6,(c3-c6),c10));
} // if
} // kind
double stop = omp_get_wtime();
printf("%.4f\n",stop - start);
return table[0][N-1];
}
int main(int argc, char *argv[]){
int num_proc=1, i;
if(argc > 1)
num_proc = atoi(argv[1]);
omp_set_num_threads(num_proc);
int kind=1;
if(argc > 2)
N = atoi(argv[2]);
DIM = N+2;
if(argc > 3)
kind = atoi(argv[3]);
points = (int **) malloc(DIM * sizeof(int*));
for(i=0; i<DIM; i++)
points[i] = (int *) malloc(2 * sizeof(int));
mcTDP(kind);
return 0;
}
|
multi_bspline_create.c | /////////////////////////////////////////////////////////////////////////////
// einspline: a library for creating and evaluating B-splines //
// Copyright (C) 2007 Kenneth P. Esler, Jr. //
// //
// This program is free software; you can redistribute it and/or modify //
// it under the terms of the GNU General Public License as published by //
// the Free Software Foundation; either version 2 of the License, or //
// (at your option) any later version. //
// //
// This program is distributed in the hope that it will be useful, //
// but WITHOUT ANY WARRANTY; without even the implied warranty of //
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the //
// GNU General Public License for more details. //
// //
// You should have received a copy of the GNU General Public License //
// along with this program; if not, write to the Free Software //
// Foundation, Inc., 51 Franklin Street, Fifth Floor, //
// Boston, MA 02110-1301 USA //
/////////////////////////////////////////////////////////////////////////////
#include "multi_bspline_create.h"
#ifndef _XOPEN_SOURCE
#define _XOPEN_SOURCE 600
#endif
#ifndef __USE_XOPEN2K
#define __USE_XOPEN2K
#endif
#include <stdlib.h>
#include <stdio.h>
#include <inttypes.h>
int posix_memalign(void **memptr, size_t alignment, size_t size);
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
//// Helper functions for spline creation ////
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
void init_sse_data();
void
find_coefs_1d_d (Ugrid grid, BCtype_d bc,
double *data, intptr_t dstride,
double *coefs, intptr_t cstride);
void
solve_deriv_interp_1d_s (float bands[], float coefs[],
int M, int cstride);
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
void
solve_periodic_interp_1d_s (float bands[], float coefs[],
int M, int cstride);
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
void
solve_antiperiodic_interp_1d_s (float bands[], float coefs[],
int M, int cstride);
void
find_coefs_1d_s (Ugrid grid, BCtype_s bc,
float *data, intptr_t dstride,
float *coefs, intptr_t cstride);
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
//// Single-Precision, Real Creation Routines ////
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
multi_UBspline_1d_s*
create_multi_UBspline_1d_s (Ugrid x_grid, BCtype_s xBC, int num_splines)
{
// Create new spline
multi_UBspline_1d_s* restrict spline = malloc (sizeof(multi_UBspline_1d_s));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_1d_s.\n");
abort();
}
spline->spcode = MULTI_U1D;
spline->tcode = SINGLE_REAL;
spline->xBC = xBC; spline->x_grid = x_grid;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int Nx;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC) {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num);
Nx = Mx+3;
}
else {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num-1);
Nx = Mx+2;
}
int N = num_splines;
#ifdef HAVE_SSE
if (N % 4)
N += 4 - (N % 4);
#endif
spline->x_stride = N;
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (sizeof(float)*Nx*N);
#else
posix_memalign ((void**)&spline->coefs, 64, (sizeof(float)*Nx*N));
#endif
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficient in create_multi_UBspline_1d_s.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_1d_s (multi_UBspline_1d_s *spline, int num,
float *data)
{
float *coefs = spline->coefs + num;
int xs = spline->x_stride;
find_coefs_1d_s (spline->x_grid, spline->xBC, data, 1,
coefs, xs);
}
multi_UBspline_2d_s*
create_multi_UBspline_2d_s (Ugrid x_grid, Ugrid y_grid,
BCtype_s xBC, BCtype_s yBC, int num_splines)
{
// Create new spline
multi_UBspline_2d_s* restrict spline = malloc (sizeof(multi_UBspline_2d_s));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_2d_s.\n");
abort();
}
spline->spcode = MULTI_U2D;
spline->tcode = SINGLE_REAL;
spline->xBC = xBC;
spline->yBC = yBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int My = y_grid.num;
int Nx, Ny;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
int N = num_splines;
#ifdef HAVE_SSE
if (N % 4)
N += 4 - (N % 4);
#endif
spline->x_stride = Ny*N;
spline->y_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc ((size_t)sizeof(float)*Nx*Ny*N);
#else
posix_memalign ((void**)&spline->coefs, 64,
sizeof(float)*Nx*Ny*N);
#endif
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_2d_s.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_2d_s (multi_UBspline_2d_s* spline, int num, float *data)
{
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Nx, Ny;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
float *coefs = spline->coefs + num;
int ys = spline->y_stride;
// First, solve in the X-direction
for (int iy=0; iy<My; iy++) {
intptr_t doffset = iy;
intptr_t coffset = iy*ys;
find_coefs_1d_s (spline->x_grid, spline->xBC, data+doffset, (intptr_t)My,
coefs+coffset, (intptr_t)Ny*ys);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++) {
intptr_t doffset = ix*Ny*ys;
intptr_t coffset = ix*Ny*ys;
find_coefs_1d_s (spline->y_grid, spline->yBC, coefs+doffset, (intptr_t)ys,
coefs+coffset, (intptr_t)ys);
}
}
multi_UBspline_3d_s*
create_multi_UBspline_3d_s (Ugrid x_grid, Ugrid y_grid, Ugrid z_grid,
BCtype_s xBC, BCtype_s yBC, BCtype_s zBC,
int num_splines)
{
// Create new spline
multi_UBspline_3d_s* restrict spline = malloc (sizeof(multi_UBspline_3d_s));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_3d_s.\n");
abort();
}
spline->spcode = MULTI_U3D;
spline->tcode = SINGLE_REAL;
spline->xBC = xBC;
spline->yBC = yBC;
spline->zBC = zBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num; int My = y_grid.num; int Mz = z_grid.num;
int Nx, Ny, Nz;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
if (zBC.lCode == PERIODIC || zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
z_grid.delta = (z_grid.end - z_grid.start)/(double)(Nz-3);
z_grid.delta_inv = 1.0/z_grid.delta;
spline->z_grid = z_grid;
int N = num_splines;
#ifdef HAVE_SSE
if (N % 4)
N += 4 - (N % 4);
#endif
spline->x_stride = Ny*Nz*N;
spline->y_stride = Nz*N;
spline->z_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (sizeof(float)*Nx*Ny*Nz*N);
#else
posix_memalign ((void**)&spline->coefs, 64,
((size_t)sizeof(float)*Nx*Ny*Nz*N));
#endif
spline->coefs_size=(size_t)Nx*(size_t)Ny*(size_t)Nz*(size_t)N;
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_3d_s.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_3d_s (multi_UBspline_3d_s* spline, int num, float *data)
{
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
float *coefs = spline->coefs + num;
intptr_t zs = spline->z_stride;
// First, solve in the X-direction
#pragma omp parallel for
for (int iy=0; iy<My; iy++)
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = iy*Mz+iz;
intptr_t coffset = (iy*Nz+iz)*zs;
find_coefs_1d_s (spline->x_grid, spline->xBC, data+doffset, (intptr_t)(My*Mz),
coefs+coffset, (intptr_t)(Ny*Nz)*zs);
}
// Now, solve in the Y-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = (ix*Ny*Nz + iz)*zs;
intptr_t coffset = (ix*Ny*Nz + iz)*zs;
find_coefs_1d_s (spline->y_grid, spline->yBC, coefs+doffset, (intptr_t)Nz*zs,
coefs+coffset, (intptr_t)Nz*zs);
}
// Now, solve in the Z-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = ((ix*Ny+iy)*Nz)*zs;
intptr_t coffset = ((ix*Ny+iy)*Nz)*zs;
find_coefs_1d_s (spline->z_grid, spline->zBC, coefs+doffset,
zs, coefs+coffset, zs);
}
}
void
set_multi_UBspline_3d_s_d(multi_UBspline_3d_s* spline, int num, double *data)
{
BCtype_d xBC, yBC, zBC;
xBC.lCode=spline->xBC.lCode; xBC.rCode=spline->xBC.rCode;
yBC.lCode=spline->yBC.lCode; yBC.rCode=spline->yBC.rCode;
zBC.lCode=spline->zBC.lCode; zBC.rCode=spline->zBC.rCode;
xBC.lVal=spline->xBC.lVal; xBC.rVal=spline->xBC.rVal;
yBC.lVal=spline->yBC.lVal; yBC.rVal=spline->yBC.rVal;
zBC.lVal=spline->zBC.lVal; zBC.rVal=spline->zBC.rVal;
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
double *spline_tmp = malloc(sizeof(double)*Nx*Ny*Nz);
// First, solve in the X-direction
#pragma omp parallel for
for (int iy=0; iy<My; iy++)
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = iy*Mz+iz;
intptr_t coffset = iy*Nz+iz;
find_coefs_1d_d (spline->x_grid, xBC, data+doffset, My*Mz, spline_tmp+coffset, Ny*Nz);
}
// Now, solve in the Y-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = ix*Ny*Nz + iz;
intptr_t coffset = ix*Ny*Nz + iz;
find_coefs_1d_d (spline->y_grid, yBC, spline_tmp+doffset, Nz, spline_tmp+coffset, Nz);
}
// Now, solve in the Z-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = (ix*Ny+iy)*Nz;
intptr_t coffset = (ix*Ny+iy)*Nz;
find_coefs_1d_d (spline->z_grid, zBC, spline_tmp+doffset, 1, spline_tmp+coffset, 1);
}
{
// const double* restrict i_ptr=spline_tmp;
#pragma omp parallel for
for(int ix=0; ix<Nx; ++ix)
{
const double* restrict i_ptr=spline_tmp+ix*Ny*Nz;
for(int iy=0; iy<Ny; ++iy)
for(int iz=0; iz<Nz; ++iz)
spline->coefs[ix*spline->x_stride +
iy*spline->y_stride +
iz*spline->z_stride + num] = (float)(*i_ptr++);
}
}
free (spline_tmp);
}
/////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
//// Single-Precision, Complex Creation Routines ////
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
multi_UBspline_1d_c*
create_multi_UBspline_1d_c (Ugrid x_grid, BCtype_c xBC, int num_splines)
{
// Create new spline
multi_UBspline_1d_c* restrict spline = malloc (sizeof(multi_UBspline_1d_c));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_1d_c.\n");
abort();
}
spline->spcode = MULTI_U1D;
spline->tcode = SINGLE_COMPLEX;
spline->xBC = xBC;
spline->num_splines = num_splines;
// Setup internal variables
int M = x_grid.num;
int N;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC) {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num);
N = M+3;
}
else {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num-1);
N = M+2;
}
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
spline->x_stride = num_splines;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (2*sizeof(float)*N*num_splines);
#else
posix_memalign ((void**)&spline->coefs, 64, 2*sizeof(float)*N*num_splines);
#endif
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_1d_c.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_1d_c (multi_UBspline_1d_c* spline, int num, complex_float *data)
{
complex_float *coefs = spline->coefs + num;
BCtype_s xBC_r, xBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
int xs = spline->x_stride;
// Real part
find_coefs_1d_s (spline->x_grid, xBC_r,
(float*)data, (intptr_t)2, (float*)coefs, (intptr_t)2*xs);
// Imaginarty part
find_coefs_1d_s (spline->x_grid, xBC_i,
((float*)data)+1, (intptr_t)2, ((float*)coefs+1), (intptr_t)2*xs);
}
multi_UBspline_2d_c*
create_multi_UBspline_2d_c (Ugrid x_grid, Ugrid y_grid,
BCtype_c xBC, BCtype_c yBC, int num_splines)
{
// Create new spline
multi_UBspline_2d_c* restrict spline = malloc (sizeof(multi_UBspline_2d_c));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_2d_c.\n");
abort();
}
spline->spcode = MULTI_U2D;
spline->tcode = SINGLE_COMPLEX;
spline->xBC = xBC;
spline->yBC = yBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int My = y_grid.num;
int Nx, Ny;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
int N = num_splines;
#ifdef HAVE_SSE
if (N % 2)
N++;
#endif
spline->x_stride = Ny*N;
spline->y_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (2*sizeof(float)*Nx*Ny*N);
spline->lapl2 = malloc (4*sizeof(float)*N);
#else
posix_memalign ((void**)&spline->coefs, 64,
2*sizeof(float)*Nx*Ny*N);
posix_memalign ((void**)&spline->lapl2, 64,
4*sizeof(float)*N);
#endif
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs || !spline->lapl2) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_2d_c.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_2d_c (multi_UBspline_2d_c* spline, int num, complex_float *data)
{
// Setup internal variables
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Nx, Ny;
complex_float* coefs = spline->coefs + num;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
BCtype_s xBC_r, xBC_i, yBC_r, yBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
yBC_r.lCode = spline->yBC.lCode; yBC_r.rCode = spline->yBC.rCode;
yBC_r.lVal = spline->yBC.lVal_r; yBC_r.rVal = spline->yBC.rVal_r;
yBC_i.lCode = spline->yBC.lCode; yBC_i.rCode = spline->yBC.rCode;
yBC_i.lVal = spline->yBC.lVal_i; yBC_i.rVal = spline->yBC.rVal_i;
int ys = spline->y_stride;
// First, solve in the X-direction
for (int iy=0; iy<My; iy++) {
intptr_t doffset = (2*iy);
intptr_t coffset = (2*iy)*ys;
// Real part
find_coefs_1d_s (spline->x_grid, xBC_r, ((float*)data)+doffset, (intptr_t)2*My,
(float*)coefs+coffset, (intptr_t)2*Ny*ys);
// Imag part
find_coefs_1d_s (spline->x_grid, xBC_i, ((float*)data)+doffset+1, (intptr_t)2*My,
((float*)coefs)+coffset+1, (intptr_t)2*Ny*ys);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++) {
intptr_t doffset = (2*ix*Ny)*ys;
intptr_t coffset = (2*ix*Ny)*ys;
// Real part
find_coefs_1d_s (spline->y_grid, yBC_r, ((float*)coefs)+doffset,
(intptr_t)2*ys, ((float*)coefs)+coffset, (intptr_t)2*ys);
// Imag part
find_coefs_1d_s (spline->y_grid, yBC_i, ((float*)coefs)+doffset+1,
(intptr_t)2*ys, ((float*)coefs)+coffset+1, (intptr_t)2*ys);
}
}
multi_UBspline_3d_c*
create_multi_UBspline_3d_c (Ugrid x_grid, Ugrid y_grid, Ugrid z_grid,
BCtype_c xBC, BCtype_c yBC, BCtype_c zBC,
int num_splines)
{
// Create new spline
multi_UBspline_3d_c* restrict spline = malloc (sizeof(multi_UBspline_3d_c));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_3d_c.\n");
abort();
}
spline->spcode = MULTI_U3D;
spline->tcode = SINGLE_COMPLEX;
spline->xBC = xBC;
spline->yBC = yBC;
spline->zBC = zBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num; int My = y_grid.num; int Mz = z_grid.num;
int Nx, Ny, Nz;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
if (zBC.lCode == PERIODIC || zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
z_grid.delta = (z_grid.end - z_grid.start)/(double)(Nz-3);
z_grid.delta_inv = 1.0/z_grid.delta;
spline->z_grid = z_grid;
int N = spline->num_splines;
#ifdef HAVE_SSE
if (N % 2)
N++;
#endif
spline->x_stride = Ny*Nz*N;
spline->y_stride = Nz*N;
spline->z_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc ((size_t)2*sizeof(float)*Nx*Ny*Nz*N);
spline->lapl3 = malloc (6*sizeof(float)*N);
#else
posix_memalign ((void**)&spline->coefs, 64, (size_t)2*sizeof(float)*Nx*Ny*Nz*N);
posix_memalign ((void**)&spline->lapl3, 64, 6*sizeof(float)*N);
#endif
spline->coefs_size=(size_t)Nx*(size_t)Ny*(size_t)Nz*(size_t)N;
#ifdef HAVE_SSE
init_sse_data();
#endif
if (!spline->coefs || !spline->lapl3) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_3d_c.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_3d_c (multi_UBspline_3d_c* spline, int num, complex_float *data)
{
// Setup internal variables
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
BCtype_s xBC_r, xBC_i, yBC_r, yBC_i, zBC_r, zBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
yBC_r.lCode = spline->yBC.lCode; yBC_r.rCode = spline->yBC.rCode;
yBC_r.lVal = spline->yBC.lVal_r; yBC_r.rVal = spline->yBC.rVal_r;
yBC_i.lCode = spline->yBC.lCode; yBC_i.rCode = spline->yBC.rCode;
yBC_i.lVal = spline->yBC.lVal_i; yBC_i.rVal = spline->yBC.rVal_i;
zBC_r.lCode = spline->zBC.lCode; zBC_r.rCode = spline->zBC.rCode;
zBC_r.lVal = spline->zBC.lVal_r; zBC_r.rVal = spline->zBC.rVal_r;
zBC_i.lCode = spline->zBC.lCode; zBC_i.rCode = spline->zBC.rCode;
zBC_i.lVal = spline->zBC.lVal_i; zBC_i.rVal = spline->zBC.rVal_i;
complex_float *coefs = spline->coefs + num;
int zs = spline->z_stride;
// First, solve in the X-direction
for (int iy=0; iy<My; iy++)
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = 2*(iy*Mz+iz);
intptr_t coffset = 2*(iy*Nz+iz)*zs;
// Real part
find_coefs_1d_s (spline->x_grid, xBC_r,
((float*)data)+doffset, (intptr_t)2*My*Mz,
((float*)coefs)+coffset, (intptr_t)2*Ny*Nz*zs);
// Imag part
find_coefs_1d_s (spline->x_grid, xBC_i,
((float*)data)+doffset+1, (intptr_t)2*My*Mz,
((float*)coefs)+coffset+1, (intptr_t)2*Ny*Nz*zs);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++)
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = 2*(ix*Ny*Nz + iz)*zs;
intptr_t coffset = 2*(ix*Ny*Nz + iz)*zs;
// Real part
find_coefs_1d_s (spline->y_grid, yBC_r,
((float*)coefs)+doffset, (intptr_t)2*Nz*zs,
((float*)coefs)+coffset, (intptr_t)2*Nz*zs);
// Imag part
find_coefs_1d_s (spline->y_grid, yBC_i,
((float*)coefs)+doffset+1, (intptr_t)2*Nz*zs,
((float*)coefs)+coffset+1, (intptr_t)2*Nz*zs);
}
// Now, solve in the Z-direction
for (int ix=0; ix<Nx; ix++)
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = 2*((ix*Ny+iy)*Nz)*zs;
intptr_t coffset = 2*((ix*Ny+iy)*Nz)*zs;
// Real part
find_coefs_1d_s (spline->z_grid, zBC_r,
((float*)coefs)+doffset, (intptr_t)2*zs,
((float*)coefs)+coffset, (intptr_t)2*zs);
// Imag part
find_coefs_1d_s (spline->z_grid, zBC_i,
((float*)coefs)+doffset+1, (intptr_t)2*zs,
((float*)coefs)+coffset+1, (intptr_t)2*zs);
}
}
void
set_multi_UBspline_3d_c_z (multi_UBspline_3d_c* spline, int num, complex_double *data)
{
// Setup internal variables
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
BCtype_d xBC_r, xBC_i, yBC_r, yBC_i, zBC_r, zBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = (double)spline->xBC.lVal_r; xBC_r.rVal = (double)spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = (double)spline->xBC.lVal_i; xBC_i.rVal = (double)spline->xBC.rVal_i;
yBC_r.lCode = spline->yBC.lCode; yBC_r.rCode = spline->yBC.rCode;
yBC_r.lVal = (double)spline->yBC.lVal_r; yBC_r.rVal = (double)spline->yBC.rVal_r;
yBC_i.lCode = spline->yBC.lCode; yBC_i.rCode = spline->yBC.rCode;
yBC_i.lVal = (double)spline->yBC.lVal_i; yBC_i.rVal = (double)spline->yBC.rVal_i;
zBC_r.lCode = spline->zBC.lCode; zBC_r.rCode = spline->zBC.rCode;
zBC_r.lVal = (double)spline->zBC.lVal_r; zBC_r.rVal = (double)spline->zBC.rVal_r;
zBC_i.lCode = spline->zBC.lCode; zBC_i.rCode = spline->zBC.rCode;
zBC_i.lVal = (double)spline->zBC.lVal_i; zBC_i.rVal = (double)spline->zBC.rVal_i;
complex_double *spline_tmp = malloc(2*sizeof(double)*Nx*Ny*Nz);
// First, solve in the X-direction
for (int iy=0; iy<My; iy++)
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = 2*(iy*Mz+iz);
intptr_t coffset = 2*(iy*Nz+iz);
// Real part
find_coefs_1d_d (spline->x_grid, xBC_r, ((double*)data)+doffset, 2*My*Mz,
((double*)spline_tmp)+coffset, 2*Ny*Nz);
// Imag part
find_coefs_1d_d (spline->x_grid, xBC_i, ((double*)data)+doffset+1, 2*My*Mz,
((double*)spline_tmp)+coffset+1, 2*Ny*Nz);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++)
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = 2*(ix*Ny*Nz + iz);
intptr_t coffset = 2*(ix*Ny*Nz + iz);
// Real part
find_coefs_1d_d (spline->y_grid, yBC_r, ((double*)spline_tmp)+doffset, 2*Nz,
((double*)spline_tmp)+coffset, 2*Nz);
// Imag part
find_coefs_1d_d (spline->y_grid, yBC_i, ((double*)spline_tmp)+doffset+1, 2*Nz,
((double*)spline_tmp)+coffset+1, 2*Nz);
}
// Now, solve in the Z-direction
for (int ix=0; ix<Nx; ix++)
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = 2*((ix*Ny+iy)*Nz);
intptr_t coffset = 2*((ix*Ny+iy)*Nz);
// Real part
find_coefs_1d_d (spline->z_grid, zBC_r, ((double*)spline_tmp)+doffset, 2,
((double*)spline_tmp)+coffset, 2);
// Imag part
find_coefs_1d_d (spline->z_grid, zBC_i, ((double*)spline_tmp)+doffset+1, 2,
((double*)spline_tmp)+coffset+1, 2);
}
{
const complex_double* restrict i_ptr=spline_tmp;
for(int ix=0; ix<Nx; ++ix)
for(int iy=0; iy<Ny; ++iy)
for(int iz=0; iz<Nz; ++iz)
spline->coefs[ix*spline->x_stride +
iy*spline->y_stride +
iz*spline->z_stride + num] = (complex_float)(*i_ptr++);
}
free(spline_tmp);
}
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
//// Double-Precision, Real Creation Routines ////
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
void
solve_deriv_interp_1d_d (double bands[], double coefs[],
int M, int cstride);
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
void
solve_periodic_interp_1d_d (double bands[], double coefs[],
int M, intptr_t cstride);
void
find_coefs_1d_d (Ugrid grid, BCtype_d bc,
double *data, intptr_t dstride,
double *coefs, intptr_t cstride);
multi_UBspline_1d_d*
create_multi_UBspline_1d_d (Ugrid x_grid, BCtype_d xBC, int num_splines)
{
// Create new spline
multi_UBspline_1d_d* restrict spline = malloc (sizeof(multi_UBspline_1d_d));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_1d_d.\n");
abort();
}
spline->spcode = MULTI_U1D;
spline->tcode = DOUBLE_REAL;
spline->xBC = xBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int Nx;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC) {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num);
Nx = Mx+3;
}
else {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num-1);
Nx = Mx+2;
}
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
int N = num_splines;
#ifdef HAVE_SSE2
// We must pad to keep data aligned for SSE operations
if (N & 1)
N++;
#endif
spline->x_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (sizeof(double)*Nx*N);
#else
posix_memalign ((void**)&spline->coefs, 64, sizeof(double)*Nx*N);
#endif
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_1d_d.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_1d_d (multi_UBspline_1d_d* spline, int num, double *data)
{
double *coefs = spline->coefs + num;
int xs = spline->x_stride;
find_coefs_1d_d (spline->x_grid, spline->xBC, data, 1, coefs, xs);
}
void
set_multi_UBspline_1d_d_BC (multi_UBspline_1d_d* spline, int num, double *data,
BCtype_d xBC)
{
double *coefs = spline->coefs + num;
int xs = spline->x_stride;
find_coefs_1d_d (spline->x_grid, xBC, data, 1, coefs, xs);
}
multi_UBspline_2d_d*
create_multi_UBspline_2d_d (Ugrid x_grid, Ugrid y_grid,
BCtype_d xBC, BCtype_d yBC, int num_splines)
{
// Create new spline
multi_UBspline_2d_d* restrict spline = malloc (sizeof(multi_UBspline_2d_d));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_2d_d.\n");
abort();
}
spline->spcode = MULTI_U2D;
spline->tcode = DOUBLE_REAL;
spline->xBC = xBC;
spline->yBC = yBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int My = y_grid.num;
int Nx, Ny;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
int N = num_splines;
#ifdef HAVE_SSE2
// We must pad to keep data align for SSE operations
if (num_splines & 1)
N++;
#endif
spline->x_stride = Ny*N;
spline->y_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (sizeof(double)*Nx*Ny*N);
#else
posix_memalign ((void**)&spline->coefs, 64, (sizeof(double)*Nx*Ny*N));
#endif
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_2d_d.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_2d_d (multi_UBspline_2d_d* spline, int num, double *data)
{
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Nx, Ny;
double *coefs = spline->coefs + num;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
int ys = spline->y_stride;
// First, solve in the X-direction
for (int iy=0; iy<My; iy++) {
intptr_t doffset = iy;
intptr_t coffset = iy*ys;
find_coefs_1d_d (spline->x_grid, spline->xBC,
data+doffset, (intptr_t)My,
coefs+coffset, (intptr_t)Ny*ys);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++) {
intptr_t doffset = ix*Ny*ys;
intptr_t coffset = ix*Ny*ys;
find_coefs_1d_d (spline->y_grid, spline->yBC,
coefs+doffset, (intptr_t)ys,
coefs+coffset, (intptr_t)ys);
}
}
multi_UBspline_3d_d*
create_multi_UBspline_3d_d (Ugrid x_grid, Ugrid y_grid, Ugrid z_grid,
BCtype_d xBC, BCtype_d yBC, BCtype_d zBC,
int num_splines)
{
// Create new spline
multi_UBspline_3d_d* restrict spline;
#ifdef HAVE_POSIX_MEMALIGN
posix_memalign ((void**)&spline, 64, (size_t)sizeof(multi_UBspline_3d_d));
#else
spline = malloc (sizeof(multi_UBspline_3d_d));
#endif
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_3d_d.\n");
abort();
}
spline->spcode = MULTI_U3D;
spline->tcode = DOUBLE_REAL;
spline->xBC = xBC;
spline->yBC = yBC;
spline->zBC = zBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num; int My = y_grid.num; int Mz = z_grid.num;
int Nx, Ny, Nz;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
if (zBC.lCode == PERIODIC || zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
z_grid.delta = (z_grid.end - z_grid.start)/(double)(Nz-3);
z_grid.delta_inv = 1.0/z_grid.delta;
spline->z_grid = z_grid;
int N = num_splines;
#if defined HAVE_SSE2 || defined HAVE_VSX
// We must pad to keep data align for SSE operations
if (N & 1)
N++;
#endif
spline->x_stride = Ny*Nz*N;
spline->y_stride = Nz*N;
spline->z_stride = N;
#ifdef HAVE_POSIX_MEMALIGN
posix_memalign ((void**)&spline->coefs, 64, ((size_t)sizeof(double)*Nx*Ny*Nz*N));
#else
spline->coefs = malloc ((size_t)sizeof(double)*Nx*Ny*Nz*N);
#endif
spline->coefs_size=(size_t)Nx*(size_t)Ny*(size_t)Nz*(size_t)N;
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_3d_d.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_3d_d (multi_UBspline_3d_d* spline, int num, double *data)
{
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
double *coefs = spline->coefs + num;
intptr_t zs = spline->z_stride;
// First, solve in the X-direction
#pragma omp parallel for
for (int iy=0; iy<My; iy++)
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = iy*Mz+iz;
intptr_t coffset = (iy*Nz+iz)*zs;
find_coefs_1d_d (spline->x_grid, spline->xBC,
data+doffset, (intptr_t)My*Mz,
coefs+coffset, (intptr_t)Ny*Nz*zs);
}
// Now, solve in the Y-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = (ix*Ny*Nz + iz)*zs;
intptr_t coffset = (ix*Ny*Nz + iz)*zs;
find_coefs_1d_d (spline->y_grid, spline->yBC,
coefs+doffset, (intptr_t)Nz*zs,
coefs+coffset, (intptr_t)Nz*zs);
}
// Now, solve in the Z-direction
#pragma omp parallel for
for (int ix=0; ix<Nx; ix++)
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = (ix*Ny+iy)*Nz*zs;
intptr_t coffset = (ix*Ny+iy)*Nz*zs;
find_coefs_1d_d (spline->z_grid, spline->zBC,
coefs+doffset, (intptr_t)zs,
coefs+coffset, (intptr_t)zs);
}
}
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
//// Double-Precision, Complex Creation Routines ////
////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////
// On input, bands should be filled with:
// row 0 : abcdInitial from boundary conditions
// rows 1:M: basis functions in first 3 cols, data in last
// row M+1 : abcdFinal from boundary conditions
// cstride gives the stride between values in coefs.
// On exit, coefs with contain interpolating B-spline coefs
multi_UBspline_1d_z*
create_multi_UBspline_1d_z (Ugrid x_grid, BCtype_z xBC, int num_splines)
{
// Create new spline
multi_UBspline_1d_z* restrict spline = malloc (sizeof(multi_UBspline_1d_z));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_1d_z.\n");
abort();
}
spline->spcode = MULTI_U1D;
spline->tcode = DOUBLE_COMPLEX;
spline->xBC = xBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int Nx;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC) {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num);
Nx = Mx+3;
}
else {
x_grid.delta = (x_grid.end-x_grid.start)/(double)(x_grid.num-1);
Nx = Mx+2;
}
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
spline->x_stride = num_splines;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (2*sizeof(double)*Nx*num_splines);
#else
posix_memalign ((void**)&spline->coefs, 64, 2*sizeof(double)*Nx*num_splines);
#endif
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_1d_z.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_1d_z (multi_UBspline_1d_z* spline, int num, complex_double *data)
{
int Mx = spline->x_grid.num;
int Nx;
complex_double *coefs = spline->coefs + num;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
BCtype_d xBC_r, xBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
int xs = spline->x_stride;
// Real part
find_coefs_1d_d (spline->x_grid, xBC_r,
(double*)data, (intptr_t)2,
((double*)coefs), (intptr_t)2*xs);
// Imaginary part
find_coefs_1d_d (spline->x_grid, xBC_i,
((double*)data)+1, (intptr_t)2,
((double*)coefs)+1, (intptr_t)2*xs);
}
void
set_multi_UBspline_1d_z_BC (multi_UBspline_1d_z *spline, int num,
complex_double *data, BCtype_z xBC)
{
int Mx = spline->x_grid.num;
int Nx;
complex_double *coefs = spline->coefs + num;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
BCtype_d xBC_r, xBC_i;
xBC_r.lCode = xBC.lCode; xBC_r.rCode = xBC.rCode;
xBC_r.lVal = xBC.lVal_r; xBC_r.rVal = xBC.rVal_r;
xBC_i.lCode = xBC.lCode; xBC_i.rCode = xBC.rCode;
xBC_i.lVal = xBC.lVal_i; xBC_i.rVal = xBC.rVal_i;
int xs = spline->x_stride;
// Real part
find_coefs_1d_d (spline->x_grid, xBC_r,
(double*)data, (intptr_t)2,
((double*)coefs), (intptr_t)2*xs);
// Imaginary part
find_coefs_1d_d (spline->x_grid, xBC_i,
((double*)data)+1, (intptr_t)2,
((double*)coefs)+1, (intptr_t)2*xs);
}
multi_UBspline_2d_z*
create_multi_UBspline_2d_z (Ugrid x_grid, Ugrid y_grid,
BCtype_z xBC, BCtype_z yBC, int num_splines)
{
// Create new spline
multi_UBspline_2d_z* restrict spline = malloc (sizeof(multi_UBspline_2d_z));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_2d_z.\n");
abort();
}
spline->spcode = MULTI_U2D;
spline->tcode = DOUBLE_COMPLEX;
spline->xBC = xBC;
spline->yBC = yBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num;
int My = y_grid.num;
int Nx, Ny;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
spline->x_stride = Ny*num_splines;
spline->y_stride = num_splines;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc (2*sizeof(double)*Nx*Ny*num_splines);
spline->lapl2 = malloc (4*sizeof(double)*num_splines);
#else
posix_memalign ((void**)&spline->coefs, 64, 2*sizeof(double)*Nx*Ny*num_splines);
posix_memalign ((void**)&spline->lapl2, 64, 4*sizeof(double)*num_splines);
#endif
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs || !spline->lapl2) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_2d_z.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_2d_z (multi_UBspline_2d_z* spline, int num,
complex_double *data)
{
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Nx, Ny;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
BCtype_d xBC_r, xBC_i, yBC_r, yBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
yBC_r.lCode = spline->yBC.lCode; yBC_r.rCode = spline->yBC.rCode;
yBC_r.lVal = spline->yBC.lVal_r; yBC_r.rVal = spline->yBC.rVal_r;
yBC_i.lCode = spline->yBC.lCode; yBC_i.rCode = spline->yBC.rCode;
yBC_i.lVal = spline->yBC.lVal_i; yBC_i.rVal = spline->yBC.rVal_i;
complex_double *coefs = spline->coefs + num;
int ys = spline->y_stride;
// First, solve in the X-direction
for (int iy=0; iy<My; iy++) {
intptr_t doffset = 2*iy;
intptr_t coffset = 2*iy*ys;
// Real part
find_coefs_1d_d (spline->x_grid, xBC_r,
((double*)data+doffset), (intptr_t)2*My,
(double*)coefs+coffset, (intptr_t)2*Ny*ys);
// Imag part
find_coefs_1d_d (spline->x_grid, xBC_i,
((double*)data)+doffset+1, (intptr_t)2*My,
((double*)coefs)+coffset+1, (intptr_t)2*Ny*ys);
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++) {
intptr_t doffset = 2*ix*Ny*ys;
intptr_t coffset = 2*ix*Ny*ys;
// Real part
find_coefs_1d_d (spline->y_grid, yBC_r,
((double*)coefs)+doffset, (intptr_t)2*ys,
(double*)coefs+coffset, (intptr_t)2*ys);
// Imag part
find_coefs_1d_d (spline->y_grid, yBC_i,
(double*)coefs+doffset+1, (intptr_t)2*ys,
((double*)coefs)+coffset+1, (intptr_t)2*ys);
}
}
multi_UBspline_3d_z*
create_multi_UBspline_3d_z (Ugrid x_grid, Ugrid y_grid, Ugrid z_grid,
BCtype_z xBC, BCtype_z yBC, BCtype_z zBC,
int num_splines)
{
// Create new spline
multi_UBspline_3d_z* restrict spline = malloc (sizeof(multi_UBspline_3d_z));
if (!spline) {
fprintf (stderr, "Out of memory allocating spline in create_multi_UBspline_3d_z.\n");
abort();
}
spline->spcode = MULTI_U3D;
spline->tcode = DOUBLE_COMPLEX;
spline->xBC = xBC;
spline->yBC = yBC;
spline->zBC = zBC;
spline->num_splines = num_splines;
// Setup internal variables
int Mx = x_grid.num; int My = y_grid.num; int Mz = z_grid.num;
int Nx, Ny, Nz;
if (xBC.lCode == PERIODIC || xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
x_grid.delta = (x_grid.end - x_grid.start)/(double)(Nx-3);
x_grid.delta_inv = 1.0/x_grid.delta;
spline->x_grid = x_grid;
if (yBC.lCode == PERIODIC || yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
y_grid.delta = (y_grid.end - y_grid.start)/(double)(Ny-3);
y_grid.delta_inv = 1.0/y_grid.delta;
spline->y_grid = y_grid;
if (zBC.lCode == PERIODIC || zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
z_grid.delta = (z_grid.end - z_grid.start)/(double)(Nz-3);
z_grid.delta_inv = 1.0/z_grid.delta;
spline->z_grid = z_grid;
int N = num_splines;
#ifdef HAVE_SSE2
if (N & 3)
N += 4-(N & 3);
#endif
spline->x_stride = (intptr_t)Ny*(intptr_t)Nz*N;
spline->y_stride = Nz*N;
spline->z_stride = N;
#ifndef HAVE_POSIX_MEMALIGN
spline->coefs = malloc ((size_t)2*sizeof(double)*Nx*Ny*Nz*N);
spline->lapl3 = malloc (6*sizeof(double)*N);
#else
posix_memalign ((void**)&spline->coefs, 64, (size_t)2*sizeof(double)*Nx*Ny*Nz*N);
posix_memalign ((void**)&spline->lapl3, 64, 6*sizeof(double)*N);
#endif
spline->coefs_size=(size_t)Nx*(size_t)Ny*(size_t)Nz*(size_t)N;
#ifdef HAVE_SSE2
init_sse_data();
#endif
if (!spline->coefs || !spline->lapl3) {
fprintf (stderr, "Out of memory allocating spline coefficients in create_multi_UBspline_3d_z.\n");
abort();
}
return spline;
}
void
set_multi_UBspline_3d_z (multi_UBspline_3d_z* spline, int num, complex_double *data)
{
// Setup internal variables
int Mx = spline->x_grid.num;
int My = spline->y_grid.num;
int Mz = spline->z_grid.num;
int Nx, Ny, Nz;
if (spline->xBC.lCode == PERIODIC || spline->xBC.lCode == ANTIPERIODIC)
Nx = Mx+3;
else
Nx = Mx+2;
if (spline->yBC.lCode == PERIODIC || spline->yBC.lCode == ANTIPERIODIC)
Ny = My+3;
else
Ny = My+2;
if (spline->zBC.lCode == PERIODIC || spline->zBC.lCode == ANTIPERIODIC)
Nz = Mz+3;
else
Nz = Mz+2;
BCtype_d xBC_r, xBC_i, yBC_r, yBC_i, zBC_r, zBC_i;
xBC_r.lCode = spline->xBC.lCode; xBC_r.rCode = spline->xBC.rCode;
xBC_r.lVal = spline->xBC.lVal_r; xBC_r.rVal = spline->xBC.rVal_r;
xBC_i.lCode = spline->xBC.lCode; xBC_i.rCode = spline->xBC.rCode;
xBC_i.lVal = spline->xBC.lVal_i; xBC_i.rVal = spline->xBC.rVal_i;
yBC_r.lCode = spline->yBC.lCode; yBC_r.rCode = spline->yBC.rCode;
yBC_r.lVal = spline->yBC.lVal_r; yBC_r.rVal = spline->yBC.rVal_r;
yBC_i.lCode = spline->yBC.lCode; yBC_i.rCode = spline->yBC.rCode;
yBC_i.lVal = spline->yBC.lVal_i; yBC_i.rVal = spline->yBC.rVal_i;
zBC_r.lCode = spline->zBC.lCode; zBC_r.rCode = spline->zBC.rCode;
zBC_r.lVal = spline->zBC.lVal_r; zBC_r.rVal = spline->zBC.rVal_r;
zBC_i.lCode = spline->zBC.lCode; zBC_i.rCode = spline->zBC.rCode;
zBC_i.lVal = spline->zBC.lVal_i; zBC_i.rVal = spline->zBC.rVal_i;
complex_double *coefs = spline->coefs + num;
int N = spline->num_splines;
int zs = spline->z_stride;
// First, solve in the X-direction
#pragma omp parallel
{
for (int iy=0; iy<My; iy++)
{
#pragma omp for
for (int iz=0; iz<Mz; iz++) {
intptr_t doffset = 2*(iy*Mz+iz);
intptr_t coffset = 2*(iy*Nz+iz)*zs;
// Real part
find_coefs_1d_d (spline->x_grid, xBC_r,
((double*)data)+doffset, (intptr_t)2*My*Mz,
((double*)coefs)+coffset, (intptr_t)2*Ny*Nz*zs);
// Imag part
find_coefs_1d_d (spline->x_grid, xBC_i,
((double*)data)+doffset+1, (intptr_t)2*My*Mz,
((double*)coefs)+coffset+1, (intptr_t)2*Ny*Nz*zs);
}
}
// Now, solve in the Y-direction
for (int ix=0; ix<Nx; ix++)
{
#pragma omp for
for (int iz=0; iz<Nz; iz++) {
intptr_t doffset = 2*(ix*Ny*Nz + iz)*zs;
intptr_t coffset = 2*(ix*Ny*Nz + iz)*zs;
// Real part
find_coefs_1d_d (spline->y_grid, yBC_r,
((double*)coefs)+doffset, (intptr_t)2*Nz*zs,
((double*)coefs)+coffset, (intptr_t)2*Nz*zs);
// Imag part
find_coefs_1d_d (spline->y_grid, yBC_i,
((double*)coefs)+doffset+1, (intptr_t)2*Nz*zs,
((double*)coefs)+coffset+1, (intptr_t)2*Nz*zs);
}
}
// Now, solve in the Z-direction
for (int ix=0; ix<Nx; ix++)
{
#pragma omp for
for (int iy=0; iy<Ny; iy++) {
intptr_t doffset = 2*((ix*Ny+iy)*Nz)*zs;
intptr_t coffset = 2*((ix*Ny+iy)*Nz)*zs;
// Real part
find_coefs_1d_d (spline->z_grid, zBC_r,
((double*)coefs)+doffset, (intptr_t)2*zs,
((double*)coefs)+coffset, (intptr_t)2*zs);
// Imag part
find_coefs_1d_d (spline->z_grid, zBC_i,
((double*)coefs)+doffset+1, (intptr_t)2*zs,
((double*)coefs)+coffset+1, (intptr_t)2*zs);
}
}
}
}
void
destroy_multi_UBspline (Bspline *spline)
{
free (spline->coefs);
free (spline);
}
|
sampler.h | // Copyright (c) 2015-2017 Contibutors.
// Author: Yafei Zhang (zhangyafeikimi@gmail.com)
//
// sampler and inference
//
#ifndef SAMPLER_H_
#define SAMPLER_H_
#include <math.h>
#include <string>
#include <vector>
#include "alias.h"
#include "model.h"
#include "table.h"
#include "x.h"
/************************************************************************/
/* Sampler */
/************************************************************************/
template <class Tables>
class Sampler : public Model<Tables> {
protected:
typedef Model<Tables> BaseType;
using BaseType::docs_;
using BaseType::words_;
using BaseType::M_;
using BaseType::V_;
using BaseType::K_;
using BaseType::topics_count_;
using BaseType::docs_topics_count_;
using BaseType::words_topics_count_;
using BaseType::hp_alpha_;
using BaseType::hp_sum_alpha_;
using BaseType::hp_beta_;
using BaseType::hp_sum_beta_;
using BaseType::random_;
using BaseType::Init;
// hyper parameters optimizations
int hp_opt_;
int hp_opt_interval_;
double hp_opt_alpha_shape_;
double hp_opt_alpha_scale_;
int hp_opt_alpha_iteration_;
int hp_opt_beta_iteration_;
// hp_opt_docs_topic_count_hist_[k][n]:
// # of documents in which topic "k" occurs "n" times.
std::vector<std::vector<int> > hp_opt_docs_topic_count_hist_;
// hp_opt_doc_len_hist_[n]:
// # of documents whose length are "n".
std::vector<int> hp_opt_doc_len_hist_;
// hp_opt_word_topic_count_hist_[n]:
// # of words which are assigned to a topic "n" times.
std::vector<int> hp_opt_word_topic_count_hist_;
// hp_opt_topic_len_hist_a[n]:
// # of topics which occurs "n" times.
std::vector<int> hp_opt_topic_len_hist_a;
// iteration variables
int total_iteration_;
int burnin_iteration_;
int log_likelihood_interval_;
int iteration_;
public:
typedef Tables TablesType;
typedef typename TablesType::TableType TableType;
Sampler()
: hp_opt_(0),
hp_opt_interval_(0),
hp_opt_alpha_shape_(0.0),
hp_opt_alpha_scale_(0.0),
hp_opt_alpha_iteration_(0),
hp_opt_beta_iteration_(0),
total_iteration_(0),
burnin_iteration_(0),
log_likelihood_interval_(0),
iteration_(0) {}
int& hp_opt() { return hp_opt_; }
int& hp_opt_interval() { return hp_opt_interval_; }
double& hp_opt_alpha_shape() { return hp_opt_alpha_shape_; }
double& hp_opt_alpha_scale() { return hp_opt_alpha_scale_; }
int& hp_opt_alpha_iteration() { return hp_opt_alpha_iteration_; }
int& hp_opt_beta_iteration() { return hp_opt_beta_iteration_; }
int& total_iteration() { return total_iteration_; }
int& burnin_iteration() { return burnin_iteration_; }
int& log_likelihood_interval() { return log_likelihood_interval_; }
virtual double LogLikelihood() const;
virtual void Train();
virtual void PreSampleCorpus();
virtual void PostSampleCorpus();
virtual void SampleCorpus();
virtual void PreSampleDocument(int m);
virtual void PostSampleDocument(int m);
virtual void SampleDocument(int m);
virtual void SampleDocument(Word* word, int doc_length,
TableType* doc_topics_count);
void HPOpt_Init();
void HPOpt_Optimize();
void HPOpt_OptimizeAlpha();
void HPOpt_PrepareOptimizeBeta();
void HPOpt_OptimizeBeta();
void HPOpt_PostSampleDocument(int m);
bool HPOpt_Enabled() const {
if (hp_opt_ && iteration_ > burnin_iteration_ &&
(iteration_ % hp_opt_interval_) == 0) {
return true;
}
return false;
}
};
template <class Tables>
double Sampler<Tables>::LogLikelihood() const {
double sum = 0.0;
#if defined _OPENMP
#pragma omp parallel for schedule(static) reduction(+ : sum)
#endif
for (int m = 0; m < M_; m++) {
const int N = docs_[m + 1] - docs_[m];
const Word* word = &words_[docs_[m]];
const auto& doc_topics_count = docs_topics_count_[m];
for (int n = 0; n < N; n++, word++) {
const int v = word->v;
const auto& word_topics_count = words_topics_count_[v];
double word_sum = 0.0;
for (int k = 0; k < K_; k++) {
const double phi_kv = (word_topics_count[k] + hp_beta_) /
(topics_count_[k] + hp_sum_beta_);
word_sum += (doc_topics_count[k] + hp_alpha_[k]) * phi_kv;
}
word_sum /= (N + hp_sum_alpha_);
sum += log(word_sum);
}
}
return sum;
}
template <class Tables>
void Sampler<Tables>::Train() {
INFO("Training begins.");
Init();
for (iteration_ = 1; iteration_ <= total_iteration_; iteration_++) {
INFO("Iteration %d begins.", iteration_);
PreSampleCorpus();
SampleCorpus();
PostSampleCorpus();
if ((iteration_ > burnin_iteration_) &&
(iteration_ % log_likelihood_interval_ == 0)) {
INFO("Calculating LogLikelihood.");
const double llh = LogLikelihood();
INFO("LogLikelihood(total/word)=%lg/%lg.", llh, llh / words_.size());
}
}
INFO("Training ended.");
}
template <class Tables>
void Sampler<Tables>::PreSampleCorpus() {
HPOpt_Init();
}
template <class Tables>
void Sampler<Tables>::PostSampleCorpus() {
HPOpt_Optimize();
}
template <class Tables>
void Sampler<Tables>::SampleCorpus() {
for (int m = 0; m < M_; m++) {
PreSampleDocument(m);
SampleDocument(m);
PostSampleDocument(m);
}
}
template <class Tables>
void Sampler<Tables>::PreSampleDocument(int m) {}
template <class Tables>
void Sampler<Tables>::PostSampleDocument(int m) {
HPOpt_PostSampleDocument(m);
}
template <class Tables>
void Sampler<Tables>::SampleDocument(int m) {
const int N = docs_[m + 1] - docs_[m];
Word* word = &words_[docs_[m]];
auto& doc_topics_count = docs_topics_count_[m];
SampleDocument(word, N, &doc_topics_count);
}
template <class Tables>
void Sampler<Tables>::SampleDocument(Word* word, int doc_length,
TableType* doc_topics_count) {}
template <class Tables>
void Sampler<Tables>::HPOpt_Init() {
if (!HPOpt_Enabled()) {
return;
}
INFO("Hyper optimization will be carried out in this iteration.");
hp_opt_docs_topic_count_hist_.clear();
hp_opt_docs_topic_count_hist_.resize(K_);
hp_opt_doc_len_hist_.clear();
hp_opt_word_topic_count_hist_.clear();
hp_opt_topic_len_hist_a.clear();
}
template <class Tables>
void Sampler<Tables>::HPOpt_Optimize() {
if (!HPOpt_Enabled()) {
return;
}
if (hp_opt_alpha_iteration_ > 0) {
INFO("Hyper optimizing alpha.");
HPOpt_OptimizeAlpha();
}
if (hp_opt_beta_iteration_ > 0) {
INFO("Hyper optimizing beta.");
HPOpt_PrepareOptimizeBeta();
HPOpt_OptimizeBeta();
}
}
template <class Tables>
void Sampler<Tables>::HPOpt_OptimizeAlpha() {
for (int i = 0; i < hp_opt_alpha_iteration_; i++) {
double denom = 0;
double diff_digamma = 0;
for (int j = 1, size = static_cast<int>(hp_opt_doc_len_hist_.size());
j < size; j++) {
diff_digamma += 1.0 / (j - 1 + hp_sum_alpha_);
denom += hp_opt_doc_len_hist_[j] * diff_digamma;
}
denom -= 1.0 / hp_opt_alpha_scale_;
hp_sum_alpha_ = 0;
for (int k = 0,
size = static_cast<int>(hp_opt_docs_topic_count_hist_.size());
k < size; k++) {
double num = 0;
double alpha_k = hp_alpha_[k];
const auto& docs_topic_k_count_hist = hp_opt_docs_topic_count_hist_[k];
diff_digamma = 0;
for (int j = 1,
size = static_cast<int>(hp_opt_docs_topic_count_hist_[k].size());
j < size; j++) {
diff_digamma += 1.0 / (j - 1 + alpha_k);
num += docs_topic_k_count_hist[j] * diff_digamma;
}
alpha_k = (alpha_k * num + hp_opt_alpha_shape_) / denom;
hp_alpha_[k] = alpha_k;
hp_sum_alpha_ += alpha_k;
}
}
}
template <class Tables>
void Sampler<Tables>::HPOpt_PrepareOptimizeBeta() {
for (int m = 0; m < M_; m++) {
const auto& doc_topics_count = docs_topics_count_[m];
for (int k = 0; k < K_; k++) {
const int count = doc_topics_count[k];
if (count == 0) {
continue;
}
if (static_cast<int>(hp_opt_word_topic_count_hist_.size()) <= count) {
hp_opt_word_topic_count_hist_.resize(count + 1);
}
hp_opt_word_topic_count_hist_[count]++;
}
}
for (int k = 0; k < K_; k++) {
const int count = topics_count_[k];
if (count == 0) {
continue;
}
if (static_cast<int>(hp_opt_topic_len_hist_a.size()) <= count) {
hp_opt_topic_len_hist_a.resize(count + 1);
}
hp_opt_topic_len_hist_a[count]++;
}
}
template <class Tables>
void Sampler<Tables>::HPOpt_OptimizeBeta() {
for (int i = 0; i < hp_opt_beta_iteration_; i++) {
double num = 0;
double diff_digamma = 0;
for (int j = 1,
size = static_cast<int>(hp_opt_word_topic_count_hist_.size());
j < size; j++) {
diff_digamma += 1.0 / (j - 1 + hp_beta_);
num += diff_digamma * hp_opt_word_topic_count_hist_[j];
}
double denom = 0;
diff_digamma = 0;
for (int j = 1, size = static_cast<int>(hp_opt_topic_len_hist_a.size());
j < size; j++) {
diff_digamma += 1.0 / (j - 1 + hp_sum_beta_);
denom += diff_digamma * hp_opt_topic_len_hist_a[j];
}
hp_sum_beta_ = hp_beta_ * num / denom;
hp_beta_ = hp_sum_beta_ / V_;
}
}
template <class Tables>
void Sampler<Tables>::HPOpt_PostSampleDocument(int m) {
if (!HPOpt_Enabled()) {
return;
}
if (hp_opt_alpha_iteration_ > 0) {
const int N = docs_[m + 1] - docs_[m];
const auto& doc_topics_count = docs_topics_count_[m];
for (int k = 0; k < K_; k++) {
const int count = doc_topics_count[k];
if (count == 0) {
continue;
}
auto& docs_topic_k_count_hist = hp_opt_docs_topic_count_hist_[k];
if (static_cast<int>(docs_topic_k_count_hist.size()) <= count) {
docs_topic_k_count_hist.resize(count + 1);
}
docs_topic_k_count_hist[count]++;
}
if (N > 0) {
if (static_cast<int>(hp_opt_doc_len_hist_.size()) <= N) {
hp_opt_doc_len_hist_.resize(N + 1);
}
hp_opt_doc_len_hist_[N]++;
}
}
}
/************************************************************************/
/* GibbsSampler */
/************************************************************************/
class GibbsSampler : public Sampler<HashTables> {
private:
std::vector<double> word_topic_cdf_; // cached
public:
GibbsSampler() {}
virtual void Init() override;
virtual void SampleDocument(Word* word, int doc_length,
TableType* doc_topics_count) override;
};
/************************************************************************/
/* SparseLDASampler */
/************************************************************************/
class SparseLDASampler : public Sampler<SparseTables> {
private:
double smooth_sum_;
double doc_sum_;
double word_sum_;
std::vector<double> smooth_pdf_;
std::vector<double> doc_pdf_;
std::vector<double> word_pdf_;
std::vector<double> cache_;
public:
SparseLDASampler() {}
virtual void Init() override;
virtual void PostSampleCorpus() override;
virtual void PostSampleDocument(int m) override;
virtual void SampleDocument(int m) override;
private:
void RemoveOrAddWordTopic(int m, int v, int k, int remove);
int SampleDocumentWord(int m, int v);
void PrepareSmoothBucket();
void PrepareDocBucket(int m);
void PrepareWordBucket(int v);
};
/************************************************************************/
/* AliasLDASampler */
/************************************************************************/
class AliasLDASampler : public Sampler<HashTables> {
private:
std::vector<double> p_pdf_;
std::vector<double> q_sums_; // for each word v
std::vector<std::vector<int> > q_samples_; // for each word v
std::vector<double> q_pdf_;
AliasBuilder q_alias_table_;
AliasD q_alias_;
int mh_step_;
public:
AliasLDASampler() : mh_step_(0) {}
int& mh_step() { return mh_step_; }
virtual void Init() override;
virtual void SampleDocument(Word* word, int doc_length,
TableType* doc_topics_count) override;
};
/************************************************************************/
/* LightLDASampler */
/************************************************************************/
class LightLDASampler : public Sampler<HashTables> {
private:
AliasBuilder hp_alpha_alias_table_;
AliasD hp_alpha_alias_;
AliasBuilder word_alias_table_;
AliasD word_alias_;
std::vector<double> word_topics_pdf_;
std::vector<std::vector<int> > words_topic_samples_;
int mh_step_;
int enable_word_proposal_;
int enable_doc_proposal_;
public:
LightLDASampler()
: mh_step_(0), enable_word_proposal_(1), enable_doc_proposal_(1) {}
int& mh_step() { return mh_step_; }
int& enable_word_proposal() { return enable_word_proposal_; }
int& enable_doc_proposal() { return enable_doc_proposal_; }
virtual void Init() override;
virtual void PostSampleCorpus() override;
virtual void SampleDocument(Word* word, int doc_length,
TableType* doc_topics_count) override;
private:
int SampleWithWord(int v);
int SampleWithDoc(Word* word, int doc_length, int v);
};
#endif // SAMPLER_H_
|
my-alloc.h | #ifndef _MY_ALLOC_H
#define _MY_ALLOC_H
#include <stdio.h>
#include <stdlib.h>
#include <omp.h>
#include <assert.h>
#include <stdarg.h>
#include <string.h>
#include "../common/stats.h"
#ifdef NDEBUG
#define MYALLOC_DISABLE_CRT
#define MYALLOC_DISABLE_ALERT
#endif
#ifndef MYALLOC_DISABLE_CRT
#define MYALLOC_ENABLE_CRT
#endif
#ifndef MYALLOC_DISABLE_ALERT
#define MYALLOC_ENABLE_ALERT
#endif
#define MYALLOC_COUNT_IT 0x1
#define MYALLOC_WARN_MAX 0x2
#define MYALLOC_ERR_MAX 0x4
#define MYALLOC_WARN_FAIL 0x8
#define MYALLOC_ERR_FAIL 0xF
extern bool my_alloc_initialized;
#ifdef MYALLOC_ENABLE_CRT
extern size_t max_mem;
extern size_t crt_mem;
extern bool warned_max;
#endif
#ifdef MYALLOC_ENABLE_ALERT
extern size_t alert_mem;
#endif
extern bool warned_fail;
static inline void
my_alloc_init(size_t _max_mem, size_t _alert_mem)
{
assert(sizeof(size_t) == sizeof(void *));
#ifdef MYALLOC_ENABLE_CRT
max_mem = _max_mem;
crt_mem = 0;
warned_max = false;
#endif
#ifdef MYALLOC_ENABLE_ALERT
alert_mem = _alert_mem;
#endif
warned_fail = false;
my_alloc_initialized = true;
}
/*
static inline void *
my_malloc_long(size_t size, count_t * counter, int options, char const * msg, ...)
{
#ifndef NDEBUG
va_list fmtargs;
#endif
void * res;
#pragma omp critical (my_alloc)
{
if (options & MYALLOC_COUNT_IT) {
if (crt_mem + size > max_mem) {
if ((options & MYALLOC_WARN_MAX) && !warned_max) {
warned_max = true;
fprintf(stderr, "my-alloc warning: exceeding maximum memory"
#ifdef NDEBUG
"\n");
#else
": ");
va_start(fmtargs, msg);
vfprintf(stderr, msg, fmtargs);
va_end(fmtargs);
#endif
}
if (options & MYALLOC_ERR_MAX) {
fprintf(stderr, "my-alloc error: exceeding maximum memory"
#ifdef NDEBUG
"\n");
#else
": ");
va_start(fmtargs, msg);
vfprintf(stderr, msg, fmtargs);
va_end(fmtargs);
#endif
exit(1);
}
}
}
res = malloc(size);
if (res == NULL) {
if ((options & MYALLOC_WARN_FAIL) && !warned_fail) {
warned_fail = true;
fprintf(stderr, "my-alloc warning: malloc failed"
#ifdef NDEBUG
"\n");
#else
": ");
va_start(fmtargs, msg);
vfprintf(stderr, msg, fmtargs);
va_end(fmtargs);
#endif
}
if (options & MYALLOC_ERR_FAIL) {
fprintf(stderr, "my-alloc error: malloc failed"
#ifdef NDEBUG
"\n");
#else
": ");
va_start(fmtargs, msg);
vfprintf(stderr, msg, fmtargs);
va_end(fmtargs);
#endif
exit(1);
}
} else {
if (options & MYALLOC_COUNT_IT)
crt_mem += size;
if (counter != NULL)
count_add(counter, size);
}
}
return res;
}
*/
static inline void *
my_malloc(size_t size, count_t * counter, char const * msg, ...)
{
#ifndef NDEBUG
va_list fmtargs;
#endif
void * res;
assert(my_alloc_initialized);
//assert(size > 0);
#ifdef DEBUG_MY_ALLOC
{
fprintf(stderr, "my_malloc: +%lld: ", (long long)size);
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
}
#endif
#ifdef MYALLOC_ENABLE_CRT
#pragma omp critical (my_alloc)
{
if (crt_mem + size > max_mem) {
if (!warned_max) {
warned_max = true;
#ifdef NDEBUG
fprintf(stderr, "my_malloc warning: exceeding maximum memory\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_malloc warning: exceeding maximum memory: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
}
#endif
#ifdef MYALLOC_ENABLE_ALERT
if (size > alert_mem) {
#ifdef NDEBUG
fprintf(stderr, "my_malloc alert: size=%lld\n", (long long)size);
#else
char new_msg[strlen(msg) + 1 + 200];
sprintf(new_msg, "my_malloc alert: size=%lld: %s\n", (long long)size, msg);
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
#endif
res = malloc(size);
if (size > 0 && res == NULL) {
// spit error message and crash
#ifdef NDEBUG
fprintf(stderr, "my_malloc error: malloc failed\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_malloc warning: malloc failed: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
exit(1);
}
#ifdef MYALLOC_ENABLE_CRT
else {
crt_mem += size;
if (counter != NULL)
count_add(counter, (int64_t)size);
}
} // end of critical section
#endif
return res;
}
static inline void *
my_calloc(size_t size, count_t * counter, char const * msg, ...)
{
#ifndef NDEBUG
va_list fmtargs;
#endif
void * res;
assert(my_alloc_initialized);
#ifdef DEBUG_MY_ALLOC
{
fprintf(stderr, "my_calloc: +%lld: ", (long long)size);
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
}
#endif
#ifdef MYALLOC_ENABLE_CRT
#pragma omp critical (my_alloc)
{
if (crt_mem + size > max_mem) {
if (!warned_max) {
warned_max = true;
#ifdef NDEBUG
fprintf(stderr, "my_calloc warning: exceeding maximum memory\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_calloc warning: exceeding maximum memory: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
}
#endif
#ifdef MYALLOC_ENABLE_ALERT
if (size > alert_mem) {
#ifdef NDEBUG
fprintf(stderr, "my_calloc alert: size=%lld\n", (long long)size);
#else
char new_msg[strlen(msg) + 1 + 200];
sprintf(new_msg, "my_calloc alert: size=%lld: %s\n", (long long)size, msg);
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
#endif
res = calloc(size, 1);
if (size > 0 && res == NULL) {
// spit error message and crash
#ifdef NDEBUG
fprintf(stderr, "my_calloc error: calloc failed\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_calloc warning: calloc failed: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
exit(1);
}
#ifdef MYALLOC_ENABLE_CRT
else {
crt_mem += size;
if (counter != NULL)
count_add(counter, (int64_t)size);
}
} // end of critical section
#endif
return res;
}
static inline void *
my_realloc(void * p, size_t size, size_t old_size, count_t * counter, char const * msg, ...)
{
#ifndef NDEBUG
va_list fmtargs;
#endif
void * res;
assert(my_alloc_initialized);
#ifdef DEBUG_MY_ALLOC
{
fprintf(stderr, "my_realloc: -%lld +%lld: ", (long long)old_size, (long long)size);
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
}
#endif
#ifdef MYALLOC_ENABLE_CRT
#pragma omp critical (my_alloc)
{
if ((crt_mem - old_size) + size > max_mem) {
if (!warned_max) {
warned_max = true;
#ifdef NDEBUG
fprintf(stderr, "my_realloc warning: exceeding maximum memory\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_realloc warning: exceeding maximum memory: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
}
#endif
#ifdef MYALLOC_ENABLE_ALERT
if ((long long)size - (long long)old_size > (long long)alert_mem) {
#ifdef NDEBUG
fprintf(stderr, "my_realloc alert: size=%lld\n", (long long)size - (long long)old_size);
#else
char new_msg[strlen(msg) + 1 + 200];
sprintf(new_msg, "my_realloc alert: size=%lld: %s\n", (long long)size - (long long)old_size, msg);
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
}
#endif
res = realloc(p, size);
if (size > 0 && res == NULL) {
// spit error message and crash
#ifdef NDEBUG
fprintf(stderr, "my_realloc error: realloc failed\n");
#else
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, "my_realloc warning: realloc failed: ");
strcat(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
#endif
exit(1);
}
#ifdef MYALLOC_ENABLE_CRT
else {
crt_mem -= old_size;
crt_mem += size;
if (counter != NULL)
count_add(counter, (int64_t)size - (int64_t)old_size);
}
}
#endif
return res;
}
static inline void
my_free(void * p, size_t size, count_t * counter, char const * msg = NULL, ...)
{
#ifndef NDEBUG
va_list fmtargs;
#endif
#ifdef DEBUG_MY_ALLOC
{
fprintf(stderr, "my_free: -%lld: ", (long long)size);
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
}
#endif
#ifdef MYALLOC_ENABLE_CRT
#pragma omp critical (my_alloc)
{
#ifndef NDEBUG
if (size > crt_mem) {
fprintf(stderr, "my_free: crashing: p=%p size=%lld crt_mem=%lld counter.crt=%lld counter.max=%lld: ",
p, (long long)size, (long long)crt_mem, (long long)counter->crt, (long long) counter->max);
char new_msg[strlen(msg) + 1 + 200];
strcpy(new_msg, msg);
strcat(new_msg, "\n");
va_start(fmtargs, msg);
vfprintf(stderr, new_msg, fmtargs);
va_end(fmtargs);
}
#endif
assert(size <= crt_mem);
#endif
free(p);
#ifdef MYALLOC_ENABLE_CRT
crt_mem -= size;
if (counter != NULL)
count_add(counter, -((int64_t)size));
}
#endif
}
#endif
|
cloudsc_validate.c | /*
* (C) Copyright 1988- ECMWF.
*
* This software is licensed under the terms of the Apache Licence Version 2.0
* which can be obtained at http://www.apache.org/licenses/LICENSE-2.0.
* In applying this licence, ECMWF does not waive the privileges and immunities
* granted to it by virtue of its status as an intergovernmental organisation
* nor does it submit to any jurisdiction.
*/
#include "cloudsc_validate.h"
#include <float.h>
#include <math.h>
#define min(a, b) (((a) < (b)) ? (a) : (b))
#define max(a, b) (((a) > (b)) ? (a) : (b))
void print_error(const char *name, double zminval, double zmaxval, double zmaxerr,
double zerrsum, double zsum, double zavgpgp, int ndim)
{
double zrelerr, zeps = DBL_EPSILON;
int iopt = 0;
if (zerrsum < zeps) {
zrelerr = 0.0;
iopt = 1;
} else if (zsum < zeps) {
zrelerr = zerrsum / (1.0 + zsum);
iopt = 2;
} else {
zrelerr = zerrsum / zsum;
iopt = 3;
}
//-- If you get 4 exclamation marks next to your error output,
// then it is likely that some uninitialized variables exists or
// some other screw-up -- watch out this !!!!
char *clwarn;
clwarn = (zrelerr > 10.0 * zeps) ? " !!!!" : " ";
zrelerr = 100.0 * zrelerr;
printf(" %+20s %dD%d %20.13le %20.13le %20.13le %20.13le %20.13le %s\n",
name, ndim, iopt, zminval, zmaxval, zmaxerr, zavgpgp, zrelerr, clwarn);
}
void validate_1d(const char *name, double * v_ref, double * v_field, int nlon, int ngptot, int nblocks)
{
/* Computes and prints errors in the "L2 norm sense" */
int b, bsize, jk;
double zminval, zmaxval, zdiff, zmaxerr, zerrsum, zsum, zrelerr, zavgpgp;
double (*field)[nlon] = (double (*)[nlon]) v_field;
double (*reference)[nlon] = (double (*)[nlon]) v_ref;
zminval = +DBL_MAX;
zmaxval = -DBL_MAX;
zmaxerr = 0.0;
zerrsum = 0.0;
zsum = 0.0;
#pragma omp parallel for default(shared) private(b, bsize, jk) \
reduction(min:zminval) reduction(max:zmaxval,zmaxerr) reduction(+:zerrsum,zsum)
for (b = 0; b < nblocks; b++) {
bsize = min(nlon, ngptot - b*nlon); // field block size
for (jk = 0; jk < bsize; jk++) {
zminval = fmin(zminval, field[b][jk]);
zmaxval = fmax(zmaxval, field[b][jk]);
// Difference against reference result in one-norm sense
zdiff = fabs(field[b][jk] - reference[b][jk]);
zmaxerr = fmax(zmaxerr, zdiff);
zerrsum = zerrsum + zdiff;
zsum = zsum + abs(reference[b][jk]);
}
}
zavgpgp = zerrsum / (double) ngptot;
print_error(name, zminval, zmaxval, zmaxerr, zerrsum, zsum, zavgpgp, 2);
}
void validate_2d(const char *name, double *v_ref, double *v_field, int nlon, int nlev, int ngptot, int nblocks)
{
/* Computes and prints errors in the "L2 norm sense" */
int b, bsize, jl, jk;
double zminval, zmaxval, zdiff, zmaxerr, zerrsum, zsum, zrelerr, zavgpgp;
double (*field)[nlev][nlon] = (double (*)[nlev][nlon]) v_field;
double (*reference)[nlev][nlon] = (double (*)[nlev][nlon]) v_ref;
zminval = +DBL_MAX;
zmaxval = -DBL_MAX;
zmaxerr = 0.0;
zerrsum = 0.0;
zsum = 0.0;
#pragma omp parallel for default(shared) private(b, bsize, jl, jk) \
reduction(min:zminval) reduction(max:zmaxval,zmaxerr) reduction(+:zerrsum,zsum)
for (b = 0; b < nblocks; b++) {
bsize = min(nlon, ngptot - b*nlon); // field block size
for (jl = 0; jl < nlev; jl++) {
for (jk = 0; jk < bsize; jk++) {
zminval = fmin(zminval, field[b][jl][jk]);
zmaxval = fmax(zmaxval, field[b][jl][jk]);
// Difference against reference result in one-norm sense
zdiff = fabs(field[b][jl][jk] - reference[b][jl][jk]);
zmaxerr = fmax(zmaxerr, zdiff);
zerrsum = zerrsum + zdiff;
zsum = zsum + abs(reference[b][jl][jk]);
}
}
}
zavgpgp = zerrsum / (double) ngptot;
print_error(name, zminval, zmaxval, zmaxerr, zerrsum, zsum, zavgpgp, 2);
}
void validate_3d(const char *name, double *v_ref, double *v_field, int nlon,
int nlev, int nclv, int ngptot, int nblocks)
{
/* Computes and prints errors in the "L2 norm sense" */
int b, bsize, jl, jk, jm;
double zminval, zmaxval, zdiff, zmaxerr, zerrsum, zsum, zrelerr, zavgpgp;
double (*field)[nclv][nlev][nlon] = (double (*)[nclv][nlev][nlon]) v_field;
double (*reference)[nclv][nlev][nlon] = (double (*)[nclv][nlev][nlon]) v_ref;
zminval = +DBL_MAX;
zmaxval = -DBL_MAX;
zmaxerr = 0.0;
zerrsum = 0.0;
zsum = 0.0;
#pragma omp parallel for default(shared) private(b, bsize, jl, jk, jm) \
reduction(min:zminval) reduction(max:zmaxval,zmaxerr) reduction(+:zerrsum,zsum)
for (b = 0; b < nblocks; b++) {
bsize = min(nlon, ngptot - b*nlon); // field block size
for (jm = 0; jm < nclv; jm++) {
for (jl = 0; jl < nlev; jl++) {
for (jk = 0; jk < bsize; jk++) {
zminval = fmin(zminval, field[b][jm][jl][jk]);
zmaxval = fmax(zmaxval, field[b][jm][jl][jk]);
// Difference against reference result in one-norm sense
zdiff = fabs(field[b][jm][jl][jk] - reference[b][jm][jl][jk]);
zmaxerr = fmax(zmaxerr, zdiff);
zerrsum = zerrsum + zdiff;
zsum = zsum + abs(reference[b][jm][jl][jk]);
}
}
}
}
zavgpgp = zerrsum / (double) ngptot;
print_error(name, zminval, zmaxval, zmaxerr, zerrsum, zsum, zavgpgp, 2);
}
int cloudsc_validate(const int nlon, const int nlev, const int nclv, const int ngptot, const int nproma,
double *plude, double *pcovptot, double *prainfrac_toprfz, double *pfsqlf, double *pfsqif,
double *pfcqlng, double *pfcqnng, double *pfsqrf, double *pfsqsf, double *pfcqrng, double *pfcqsng,
double *pfsqltur, double *pfsqitur, double *pfplsl, double *pfplsn, double *pfhpsl, double *pfhpsn,
double *tend_loc_a, double *tend_loc_q, double *tend_loc_t, double *tend_loc_cld)
{
const int nblocks = (ngptot / nproma) + min(ngptot % nproma, 1);
double *ref_plude = (double*) malloc( sizeof(double) * nblocks*nlev*nproma );
double *ref_pcovptot = (double*) malloc( sizeof(double) * nblocks*nlev*nproma );
double *ref_prainfrac_toprfz = (double*) malloc( sizeof(double) * nblocks*nproma );
double *ref_pfsqlf = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfsqif = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfcqlng = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfcqnng = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfsqrf = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfsqsf = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfcqrng = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfcqsng = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfsqltur = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfsqitur = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfplsl = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfplsn = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfhpsl = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_pfhpsn = (double*) malloc( sizeof(double) * nblocks*(nlev+1)*nproma );
double *ref_tend_loc_a = (double*) malloc( sizeof(double) * nblocks*nlev*nproma );
double *ref_tend_loc_q = (double*) malloc( sizeof(double) * nblocks*nlev*nproma );
double *ref_tend_loc_t = (double*) malloc( sizeof(double) * nblocks*nlev*nproma );
double *ref_tend_loc_cld = (double*) malloc( sizeof(double) * nblocks*nclv*nlev*nproma );
load_reference(nlon, nlev, nclv, ngptot, nproma,
ref_plude, ref_pcovptot, ref_prainfrac_toprfz, ref_pfsqlf, ref_pfsqif,
ref_pfcqlng, ref_pfcqnng, ref_pfsqrf, ref_pfsqsf, ref_pfcqrng, ref_pfcqsng,
ref_pfsqltur, ref_pfsqitur, ref_pfplsl, ref_pfplsn, ref_pfhpsl, ref_pfhpsn,
ref_tend_loc_a, ref_tend_loc_q, ref_tend_loc_t, ref_tend_loc_cld);
printf(" %+20s %s %+20s %+20s %+20s %+20s %+20s\n",
"Variable", "Dim", "MinValue", "MaxValue", "AbsMaxErr", "AvgAbsErr/GP", "MaxRelErr-%");
validate_2d("PLUDE", ref_plude, plude, nproma, nlev, ngptot, nblocks);
validate_2d("PCOVPTOT", ref_pcovptot, pcovptot, nproma, nlev, ngptot, nblocks);
validate_1d("PRAINFRAC_TOPRFZ", ref_prainfrac_toprfz, prainfrac_toprfz, nproma, ngptot, nblocks);
validate_2d("PFSQLF", ref_pfsqlf, pfsqlf, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFSQIF", ref_pfsqif, pfsqif, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFCQLNG", ref_pfcqlng, pfcqlng, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFCQNNG", ref_pfcqnng, pfcqnng, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFSQRF", ref_pfsqrf, pfsqrf, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFSQSF", ref_pfsqsf, pfsqsf, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFCQRNG", ref_pfcqrng, pfcqrng, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFCQSNG", ref_pfcqsng, pfcqsng, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFSQLTUR", ref_pfsqltur, pfsqltur, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFSQITUR", ref_pfsqitur, pfsqitur, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFPLSL", ref_pfplsl, pfplsl, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFPLSN", ref_pfplsn, pfplsn, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFHPSL", ref_pfhpsl, pfhpsl, nproma, nlev+1, ngptot, nblocks);
validate_2d("PFHPSN", ref_pfhpsn, pfhpsn, nproma, nlev+1, ngptot, nblocks);
validate_2d("TENDENCY_LOC%A", ref_tend_loc_a, tend_loc_a, nproma, nlev, ngptot, nblocks);
validate_2d("TENDENCY_LOC%Q", ref_tend_loc_q, tend_loc_q, nproma, nlev, ngptot, nblocks);
validate_2d("TENDENCY_LOC%T", ref_tend_loc_t, tend_loc_t, nproma, nlev, ngptot, nblocks);
validate_3d("TENDENCY_LOC%CLD", ref_tend_loc_cld, tend_loc_cld, nproma, nlev, nclv, ngptot, nblocks);
free(ref_plude);
free(ref_pcovptot);
free(ref_prainfrac_toprfz);
free(ref_pfsqlf);
free(ref_pfsqif);
free(ref_pfcqlng);
free(ref_pfcqnng);
free(ref_pfsqrf);
free(ref_pfsqsf);
free(ref_pfcqrng);
free(ref_pfcqsng);
free(ref_pfsqltur);
free(ref_pfsqitur);
free(ref_pfplsl);
free(ref_pfplsn);
free(ref_pfhpsl);
free(ref_pfhpsn);
free(ref_tend_loc_a);
free(ref_tend_loc_q);
free(ref_tend_loc_t);
free(ref_tend_loc_cld);
}
|
cryptocontext.h | /**
* @file cryptocontext.h -- Control for encryption operations.
* @author TPOC: palisade@njit.edu
*
* @section LICENSE
*
* Copyright (c) 2017, New Jersey Institute of Technology (NJIT)
* All rights reserved.
* Redistribution and use in source and binary forms, with or without modification,
* are permitted provided that the following conditions are met:
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright notice, this
* list of conditions and the following disclaimer in the documentation and/or other
* materials provided with the distribution.
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
* ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
* OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
* IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifndef SRC_DEMO_PRE_CRYPTOCONTEXT_H_
#define SRC_DEMO_PRE_CRYPTOCONTEXT_H_
#include "palisade.h"
#include "cryptocontexthelper.h"
#include "cryptotiming.h"
namespace lbcrypto {
template<typename Element>
class CryptoContextFactory;
template<typename Element>
class CryptoContextImpl;
template<typename Element>
using CryptoContext = shared_ptr<CryptoContextImpl<Element>>;
/**
* @brief CryptoContextImpl
*
* A CryptoContextImpl is the object used to access the PALISADE library
*
* All PALISADE functionality is accessed by way of an instance of a CryptoContextImpl; we say that various objects are
* "created in" a context, and can only be used in the context in which they were created
*
* All PALISADE methods are accessed through CryptoContextImpl methods. Guards are implemented to make certain that
* only valid objects that have been created in the context are used
*
* Contexts are created using the CryptoContextFactory, and can be serialized and recovered from a serialization
*/
template<typename Element>
class CryptoContextImpl : public Serializable {
friend class CryptoContextFactory<Element>;
private:
shared_ptr<LPCryptoParameters<Element>> params; /*!< crypto parameters used for this context */
shared_ptr<LPPublicKeyEncryptionScheme<Element>> scheme; /*!< algorithm used; accesses all crypto methods */
static std::map<string,std::vector<LPEvalKey<Element>>> evalMultKeyMap; /*!< cached evalmult keys, by secret key UID */
static std::map<string,shared_ptr<std::map<usint,LPEvalKey<Element>>>> evalSumKeyMap; /*!< cached evalsum keys, by secret key UID */
bool doTiming;
vector<TimingInfo>* timeSamples;
/**
* Private methods to compare two contexts; this is only used internally and is not generally available
* @param a - shared pointer in the object
* @param b - this object, usually
* @return true if the shared pointer is a pointer to "this"
*/
friend bool operator==(const CryptoContext<Element>& a, const CryptoContext<Element>& b) {
if( a->params.get() != b->params.get() ) return false;
return true;
}
friend bool operator!=(const CryptoContext<Element>& a, const CryptoContext<Element>& b) {
return !( a == b );
}
/**
* Private methods to compare two contexts; this is only used internally and is not generally available
* @param a - shared pointer in the object
* @param b - this object, usually
* @return true if the shared pointer is a pointer to "this"
*/
friend bool operator==(const CryptoContextImpl<Element>& a, const CryptoContextImpl<Element>& b) {
if( a.params.get() != b.params.get() ) return false;
return true;
}
friend bool operator!=(const CryptoContextImpl<Element>& a, const CryptoContextImpl<Element>& b) {
return !( a == b );
}
/**
* TypeCheck makes sure that an operation between two ciphertexts is permitted
* @param a
* @param b
*/
void TypeCheck(const Ciphertext<Element> a, const Ciphertext<Element> b) const {
if( a == NULL || b == NULL )
PALISADE_THROW( type_error, "Null Ciphertext");
if( a->GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Ciphertext was not created in this CryptoContextImpl");
if( a->GetCryptoContext() != b->GetCryptoContext() )
PALISADE_THROW( type_error, "Ciphertexts were not created in the same CryptoContextImpl");
if( a->GetKeyTag() != b->GetKeyTag() )
PALISADE_THROW( type_error, "Ciphertexts were not encrypted with same keys" );
if( a->GetEncodingType() != b->GetEncodingType() )
PALISADE_THROW( type_error, "Ciphertext encoding types do not match");
}
/**
* TypeCheck makes sure that an operation between a ciphertext and a plaintext is permitted
* @param a
* @param b
*/
void TypeCheck(const Ciphertext<Element> a, const Plaintext b) const {
if( a == NULL )
PALISADE_THROW( type_error, "Null Ciphertext");
if( b == NULL )
PALISADE_THROW( type_error, "Null Plaintext");
if( a->GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Ciphertext was not created in this CryptoContextImpl");
if( a->GetEncodingType() != b->GetEncodingType() )
PALISADE_THROW( type_error, "Ciphertext and Plaintext encoding types do not match");
}
/**
* TypeCheck makes sure that an operation between two ciphertexts is permitted
* @param a
* @param b
*/
void TypeCheck(const RationalCiphertext<Element>& a, const RationalCiphertext<Element>& b) const {
if( a.GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Ciphertext was not created in this CryptoContextImpl");
if( a.GetCryptoContext() != b.GetCryptoContext() )
PALISADE_THROW( type_error, "Ciphertexts were not created in the same CryptoContextImpl");
if( a.GetKeyTag() != b.GetKeyTag() )
PALISADE_THROW( type_error, "Ciphertexts were not encrypted with same keys" );
if( a.GetNumerator()->GetEncodingType() != b.GetNumerator()->GetEncodingType() )
PALISADE_THROW( type_error, "Ciphertext encoding types do not match");
}
/**
* TypeCheck makes sure that an operation between a ciphertext and a plaintext is permitted
* @param a
* @param b
*/
void TypeCheck(const RationalCiphertext<Element>& a, const Plaintext b) const {
if( b == NULL )
PALISADE_THROW( type_error, "Null Plaintext");
if( a.GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Ciphertext was not created in this CryptoContextImpl");
if( a.GetNumerator()->GetEncodingType() != b->GetEncodingType() )
PALISADE_THROW( type_error, "Ciphertext and Plaintext encoding types do not match");
}
bool Mismatched(const CryptoContext<Element> a) const {
if( a.get() != this ) {
return true;
}
return false;
}
public:
/**
* CryptoContextImpl constructor from pointers to parameters and scheme
* @param params - pointer to CryptoParameters
* @param scheme - pointer to Crypto Scheme
*/
CryptoContextImpl(LPCryptoParameters<Element> *params = 0, LPPublicKeyEncryptionScheme<Element> *scheme = 0) {
this->params.reset(params);
this->scheme.reset(scheme);
this->doTiming = false;
this->timeSamples = 0;
}
/**
* CryptoContextImpl constructor from shared pointers to parameters and scheme
* @param params - shared pointer to CryptoParameters
* @param scheme - sharedpointer to Crypto Scheme
*/
CryptoContextImpl(shared_ptr<LPCryptoParameters<Element>> params, shared_ptr<LPPublicKeyEncryptionScheme<Element>> scheme) {
this->params = params;
this->scheme = scheme;
this->doTiming = false;
this->timeSamples = 0;
}
/**
* Copy constructor
* @param c - source
*/
CryptoContextImpl(const CryptoContextImpl<Element>& c) {
params = c.params;
scheme = c.scheme;
doTiming = c.doTiming;
timeSamples = c.timeSamples;
}
/**
* Assignment
* @param rhs - assigning from
* @return this
*/
CryptoContextImpl<Element>& operator=(const CryptoContextImpl<Element>& rhs) {
params = rhs.params;
scheme = rhs.scheme;
doTiming = rhs.doTiming;
timeSamples = rhs.timeSamples;
return *this;
}
/**
* A CryptoContextImpl is only valid if the shared pointers are both valid
*/
operator bool() const { return bool(params) && bool(scheme); }
// TIMING METHODS
/**
* StartTiming method activates timing of CryptoMethods
*
* @param timeSamples points to a vector in which timing samples will be stored
*/
void StartTiming(vector<TimingInfo>* timeSamples) {
this->timeSamples = timeSamples;
doTiming = true;
}
/*
* StopTiming - turns off timing
*/
void StopTiming() {
doTiming = false;
}
/**
* ResumeTiming - re-enables timing with existing TimingInfo vector
*/
void ResumeTiming() {
doTiming = true;
}
/**
* ResetTiming - erases measurements
*/
void ResetTiming() {
this->timeSamples->clear();
}
// SERIALIZATION METHODS
/**
* Serialize the CryptoContextImpl
*
* @param serObj - rapidJson object for the serializaion
* @return true on success
*/
bool Serialize(Serialized* serObj) const;
/**
* Deserialize the context AND initialize the algorithm
*
* @param serObj
* @return true on success
*/
bool Deserialize(const Serialized& serObj) {
throw std::logic_error("Deserialize by using CryptoContextFactory::DeserializeAndCreateContext");
}
/**
* SerializeEvalMultKey for all EvalMult keys
* method will serialize each CryptoContextImpl only once
*
* @param serObj - serialization
* @return true on success
*/
static bool SerializeEvalMultKey(Serialized* serObj);
/**
* SerializeEvalMultKey for a single EvalMult key
* method will serialize entire key AND cryptocontext
*
* @param serObj - serialization
* @param id for key to serialize
* @return true on success (false on failure or key id not found)
*/
static bool SerializeEvalMultKey(Serialized* serObj, const string& id);
/**
* SerializeEvalMultKey for all EvalMultKeys made in a given context
* method will serialize the context only once
*
* @param serObj - serialization
* @param cc whose keys should be serialized
* @return true on success (false on failure or no keys found)
*/
static bool SerializeEvalMultKey(Serialized* serObj, const CryptoContext<Element> cc);
/**
* DeserializeEvalMultKey deserialize all keys in the serialization
* deserialized keys silently replace any existing matching keys
* deserialization will create CryptoContextImpl if necessary
*
* @param serObj - serialization
* @return true on success
*/
static bool DeserializeEvalMultKey(const Serialized& serObj);
/**
* ClearEvalMultKeys - flush EvalMultKey cache
*/
static void ClearEvalMultKeys();
/**
* ClearEvalMultKeys - flush EvalMultKey cache for a given id
* @param id
*/
static void ClearEvalMultKeys(const string& id);
/**
* ClearEvalMultKeys - flush EvalMultKey cache for a given context
* @param cc
*/
static void ClearEvalMultKeys(const CryptoContext<Element> cc);
/**
* InsertEvalMultKey - add the given vector of keys to the map, replacing the existing vector if there
* @param vectorToInsert
*/
static void InsertEvalMultKey(const std::vector<LPEvalKey<Element>>& vectorToInsert);
/**
* SerializeEvalSumKey for all EvalSum keys
* method will serialize each CryptoContextImpl only once
*
* @param serObj - serialization
* @return true on success
*/
static bool SerializeEvalSumKey(Serialized* serObj);
/**
* SerializeEvalSumKey for a single EvalSum key
* method will serialize entire key AND cryptocontext
*
* @param serObj - serialization
* @param id for key to serialize
* @return true on success (false on failure or key id not found)
*/
static bool SerializeEvalSumKey(Serialized* serObj, const string& id);
/**
* SerializeEvalSumKey for all EvalSumKeys made in a given context
* method will serialize the context only once
*
* @param serObj - serialization
* @param cc whose keys should be serialized
* @return true on success (false on failure or no keys found)
*/
static bool SerializeEvalSumKey(Serialized* serObj, const CryptoContext<Element> cc);
/**
* DeserializeEvalSumKey deserialize all keys in the serialization
* deserialized keys silently replace any existing matching keys
* deserialization will create CryptoContextImpl if necessary
*
* @param serObj - serialization
* @return true on success
*/
static bool DeserializeEvalSumKey(const Serialized& serObj);
/**
* ClearEvalSumKeys - flush EvalSumKey cache
*/
static void ClearEvalSumKeys();
/**
* ClearEvalSumKeys - flush EvalSumKey cache for a given id
* @param id
*/
static void ClearEvalSumKeys(const string& id);
/**
* ClearEvalSumKeys - flush EvalSumKey cache for a given context
* @param cc
*/
static void ClearEvalSumKeys(const CryptoContext<Element> cc);
/**
* InsertEvalSumKey - add the given map of keys to the map, replacing the existing map if there
* @param mapToInsert
*/
static void InsertEvalSumKey(const shared_ptr<std::map<usint,LPEvalKey<Element>>> mapToInsert);
// TURN FEATURES ON
/**
* Enable a particular feature for use with this CryptoContextImpl
* @param feature - the feature that should be enabled
*/
void Enable(PKESchemeFeature feature) { scheme->Enable(feature); }
/**
* Enable several features at once
* @param featureMask - bitwise or of several PKESchemeFeatures
*/
void Enable(usint featureMask) { scheme->Enable(featureMask); }
// GETTERS
/**
* Getter for Scheme
* @return scheme
*/
const shared_ptr<LPPublicKeyEncryptionScheme<Element>> GetEncryptionAlgorithm() const { return scheme; }
/**
* Getter for CryptoParams
* @return params
*/
const shared_ptr<LPCryptoParameters<Element>> GetCryptoParameters() const { return params; }
/**
* Getter for element params
* @return
*/
const shared_ptr<typename Element::Params> GetElementParams() const { return params->GetElementParams(); }
/**
* Getter for encoding params
* @return
*/
const EncodingParams GetEncodingParams() const { return params->GetEncodingParams(); }
/**
* Get the cyclotomic order used for this context
*
* @return
*/
const usint GetCyclotomicOrder() const { return params->GetElementParams()->GetCyclotomicOrder(); }
/**
* Get the ring dimension used for this context
*
* @return
*/
const usint GetRingDimension() const { return params->GetElementParams()->GetRingDimension(); }
/**
* Get the ciphertext modulus used for this context
*
* @return
*/
const typename Element::Integer& GetModulus() const { return params->GetElementParams()->GetModulus(); }
/**
* Get the ciphertext modulus used for this context
*
* @return
*/
const typename Element::Integer& GetRootOfUnity() const { return params->GetElementParams()->GetRootOfUnity(); }
/**
* KeyGen generates a key pair using this algorithm's KeyGen method
* @return a public/secret key pair
*/
LPKeyPair<Element> KeyGen() {
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->KeyGen(CryptoContextFactory<Element>::GetContextForPointer(this), false);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpKeyGen, currentDateTime() - start) );
}
return r;
}
/**
* KeyGen generates a Multiparty key pair using this algorithm's KeyGen method from two keys
* @param pk first public key used to coordinate the creation of later public keys.
* @return a public/secret key pair
*/
LPKeyPair<Element> MultipartyKeyGen(
const LPPublicKey<Element> pk, bool makeSparse=false, bool pre=false) {
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->MultipartyKeyGen(CryptoContextFactory<Element>::GetContextForPointer(this), pk, makeSparse, pre);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpMultiPartyKeyGenKey, currentDateTime() - start) );
}
return r;
}
/**
* KeyGen generates a Multiparty key pair using a vector of secret keys
* @param secretKeys a vector of the secret keys to be used for multiparty computation.
* @return a public/secret key pair
*/
LPKeyPair<Element> MultipartyKeyGen(
const vector<LPPrivateKey<Element>>& secretKeys) {
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->MultipartyKeyGen(CryptoContextFactory<Element>::GetContextForPointer(this), secretKeys, false);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpMultiPartyKeyGenKeyvec, currentDateTime() - start) );
}
return r;
}
/**
* Lead Multiparty Decryption method for PALISADE multiparty operations.
* This should be performed by exactly one of the clients.
* All other clients should perform the MultipartyDecryptMain operation.
* @param privateKey the secret key of the lead decryption client
* @param ciphertext vector of encrypted ciphertext
* @return vector of partially decrypted ciphertexts
*/
vector<Ciphertext<Element>> MultipartyDecryptLead(
const LPPrivateKey<Element> privateKey,
const vector<Ciphertext<Element>>& ciphertext) const
{
if( privateKey == NULL || Mismatched(privateKey->GetCryptoContext()) )
throw std::logic_error("Information passed to MultipartyDecryptLead was not generated with this crypto context");
vector<Ciphertext<Element>> newCiphertext;
double start = 0;
if( doTiming ) start = currentDateTime();
for( size_t i = 0; i < ciphertext.size(); i++ ) {
if( ciphertext[i] == NULL || Mismatched(ciphertext[i]->GetCryptoContext()) )
throw std::logic_error("A ciphertext passed to MultipartyDecryptLead was not generated with this crypto context");
newCiphertext.push_back( GetEncryptionAlgorithm()->MultipartyDecryptLead(privateKey, ciphertext[i]) );
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpMultiPartyDecryptLead, currentDateTime() - start) );
}
return newCiphertext;
}
/**
* Multiparty decryption method for PALISADE multiparty operations.
* The lead multiparty decryption operation should be performed by exactly one of the clients.
* All other clients should perform this MultipartyDecryptMain operation.
* @param privateKey - for decryption
* @param ciphertext - vector of encrypted ciphertext
* @return vector of partially decrypted ciphertexts
*/
vector<Ciphertext<Element>> MultipartyDecryptMain(
const LPPrivateKey<Element> privateKey,
const vector<Ciphertext<Element>>& ciphertext) const
{
if( privateKey == NULL || Mismatched(privateKey->GetCryptoContext()) )
throw std::logic_error("Information passed to MultipartyDecryptMain was not generated with this crypto context");
vector<Ciphertext<Element>> newCiphertext;
double start = 0;
if( doTiming ) start = currentDateTime();
for( size_t i = 0; i < ciphertext.size(); i++ ) {
if( ciphertext[i] == NULL || Mismatched(ciphertext[i]->GetCryptoContext()) )
throw std::logic_error("A ciphertext passed to MultipartyDecryptMain was not generated with this crypto context");
newCiphertext.push_back( GetEncryptionAlgorithm()->MultipartyDecryptMain(privateKey, ciphertext[i]) );
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpMultiPartyDecryptMain, currentDateTime() - start) );
}
return newCiphertext;
}
/**
* Final multiparty decryption method to fuse the partially decrypted ciphertexts into a decrypted plaintext.
* The lead multiparty decryption operation should be performed by exactly one of the clients.
* All other clients should perform the MultipartyDecryptMain operation.
* @param partialCiphertextVec - vector of partially decrypted ciphertexts.
* @param plaintext - pointer to destination for the result of decryption
* @param doPadding - true if input plaintext was padded; causes unpadding on last piece of ciphertext
* @return size of plaintext
*/
DecryptResult MultipartyDecryptFusion(
const vector<Ciphertext<Element>>& partialCiphertextVec,
Plaintext *plaintext) const
{
DecryptResult result;
//Make sure we're processing ciphertexts.
size_t last_ciphertext = partialCiphertextVec.size();
if ( last_ciphertext < 1 )
return result;
double start = 0;
if( doTiming ) start = currentDateTime();
for( size_t i = 0; i < last_ciphertext; i++ ) {
if (partialCiphertextVec[i] == NULL || Mismatched(partialCiphertextVec[i]->GetCryptoContext()))
throw std::logic_error("A ciphertext passed to MultipartyDecryptFusion was not generated with this crypto context");
if (partialCiphertextVec[i]->GetEncodingType() != partialCiphertextVec[0]->GetEncodingType())
throw std::logic_error("Ciphertexts passed to MultipartyDecryptFusion have mismatched encoding types");
}
// determine which type of plaintext that you need to decrypt into
Plaintext decrypted = GetPlaintextForDecrypt(partialCiphertextVec[0]->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
result = GetEncryptionAlgorithm()->MultipartyDecryptFusion(partialCiphertextVec, &decrypted->GetElement<NativePoly>());
if (result.isValid == false) return result;
decrypted->Decode();
*plaintext = decrypted;
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpMultiPartyDecryptFusion, currentDateTime() - start) );
}
return result;
}
/**
* SparseKeyGen generates a key pair with special structure, and without full entropy,
* for use in special cases like Ring Reduction
* @return a public/secret key pair
*/
LPKeyPair<Element> SparseKeyGen() {
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->KeyGen(CryptoContextFactory<Element>::GetContextForPointer(this), true);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpSparseKeyGen, currentDateTime() - start) );
}
return r;
}
/**
* ReKeyGen produces an Eval Key that PALISADE can use for Proxy Re Encryption
* @param newKey (public)
* @param oldKey (private)
* @return new evaluation key
*/
LPEvalKey<Element> ReKeyGen(
const LPPublicKey<Element> newKey,
const LPPrivateKey<Element> oldKey) const {
if( newKey == NULL || oldKey == NULL ||
Mismatched(newKey->GetCryptoContext()) ||
Mismatched(oldKey->GetCryptoContext()) )
throw std::logic_error("Keys passed to ReKeyGen were not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->ReKeyGen(newKey, oldKey);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpReKeyGenPubPri, currentDateTime() - start) );
}
return r;
}
/**
* ReKeyGen produces an Eval Key that PALISADE can use for Proxy Re Encryption
* @param newKey (private)
* @param oldKey (private)
* @return new evaluation key
*/
LPEvalKey<Element> ReKeyGen(
const LPPrivateKey<Element> newKey,
const LPPrivateKey<Element> oldKey) const {
if (newKey == NULL || oldKey == NULL ||
Mismatched(newKey->GetCryptoContext()) ||
Mismatched(oldKey->GetCryptoContext()) )
throw std::logic_error("Keys passed to ReKeyGen were not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->ReKeyGen(newKey, oldKey);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpReKeyGenPriPri, currentDateTime() - start) );
}
return r;
}
/**
* EvalMultKeyGen creates a key that can be used with the PALISADE EvalMult operator
* @param key
* @return new evaluation key
*/
void EvalMultKeyGen(const LPPrivateKey<Element> key);
/**
* EvalMultsKeyGen creates a vector evalmult keys that can be used with the PALISADE EvalMult operator
* 1st key (for s^2) is used for multiplication of ciphertexts of depth 1
* 2nd key (for s^3) is used for multiplication of ciphertexts of depth 2, etc.
*
* @param key
* @return a vector of evaluation keys
*/
void EvalMultKeysGen(const LPPrivateKey<Element> key);
/**
* GetEvalMultKeyVector fetches the eval mult keys for a given KeyID
* @param keyID
* @return key vector from ID
*/
static const vector<LPEvalKey<Element>>& GetEvalMultKeyVector(const string& keyID);
/**
* GetEvalMultKeys
* @return map of all the keys
*/
static const std::map<string,std::vector<LPEvalKey<Element>>>& GetAllEvalMultKeys();
/**
* KeySwitchGen creates a key that can be used with the PALISADE KeySwitch operation
* @param key1
* @param key2
* @return new evaluation key
*/
LPEvalKey<Element> KeySwitchGen(
const LPPrivateKey<Element> key1, const LPPrivateKey<Element> key2) const {
if( key1 == NULL || key2 == NULL ||
Mismatched(key1->GetCryptoContext()) ||
Mismatched(key2->GetCryptoContext()) )
throw std::logic_error("Keys passed to KeySwitchGen were not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto r = GetEncryptionAlgorithm()->KeySwitchGen(key1, key2);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpKeySwitchGen, currentDateTime() - start) );
}
return r;
}
/**
* Encrypt a plaintext using a given public key
* @param publicKey
* @param plaintext
* @return ciphertext (or null on failure)
*/
Ciphertext<Element> Encrypt(
const LPPublicKey<Element> publicKey,
Plaintext plaintext)
{
if( publicKey == NULL )
throw std::logic_error("null key passed to Encrypt");
if( plaintext == NULL )
throw std::logic_error("null plaintext passed to Encrypt");
if( Mismatched(publicKey->GetCryptoContext()) )
throw std::logic_error("key passed to Encrypt was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
Ciphertext<Element> ciphertext = GetEncryptionAlgorithm()->Encrypt(publicKey, plaintext->GetEncodedElement<Element>());
if (ciphertext) {
ciphertext->SetEncodingType( plaintext->GetEncodingType() );
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEncryptPub, currentDateTime() - start) );
}
return ciphertext;
}
/**
* Encrypt a plaintext using a given private key
* @param privateKey
* @param plaintext
* @return ciphertext (or null on failure)
*/
Ciphertext<Element> Encrypt(
const LPPrivateKey<Element> privateKey,
Plaintext plaintext) const
{
if( privateKey == NULL || Mismatched(privateKey->GetCryptoContext()) )
throw std::logic_error("key passed to Encrypt was not generated with this crypto context");
if( plaintext == NULL )
throw std::logic_error("null plaintext passed to Encrypt");
double start = 0;
if( doTiming ) start = currentDateTime();
Ciphertext<Element> ciphertext = GetEncryptionAlgorithm()->Encrypt(privateKey, plaintext->GetEncodedElement<Element>());
if (ciphertext) {
ciphertext->SetEncodingType( plaintext->GetEncodingType() );
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEncryptPriv, currentDateTime() - start) );
}
return ciphertext;
}
/**
* Encrypt a matrix of Plaintext
* @param publicKey - for encryption
* @param plaintext - to encrypt
* @param doEncryption encrypts if true, embeds (encodes) the plaintext into cryptocontext if false
* @return a vector of pointers to Ciphertexts created by encrypting the plaintext
*/
shared_ptr<Matrix<RationalCiphertext<Element>>> EncryptMatrix(
const LPPublicKey<Element> publicKey,
Matrix<Plaintext> &plaintext)
{
if (publicKey == NULL || Mismatched(publicKey->GetCryptoContext()))
throw std::logic_error("key passed to EncryptMatrix was not generated with this crypto context");
auto zeroAlloc = [=]() { return make_unique<RationalCiphertext<Element>>(publicKey->GetCryptoContext(), true); };
shared_ptr<Matrix<RationalCiphertext<Element>>> cipherResults(new Matrix<RationalCiphertext<Element>>
(zeroAlloc, plaintext.GetRows(), plaintext.GetCols()));
double start = 0;
if( doTiming ) start = currentDateTime();
for (size_t row = 0; row < plaintext.GetRows(); row++)
{
for (size_t col = 0; col < plaintext.GetCols(); col++)
{
if( plaintext(row,col)->Encode() == false )
return 0;
Ciphertext<Element> ciphertext = GetEncryptionAlgorithm()->Encrypt(publicKey, plaintext(row,col)->GetElement<Element>());
if (ciphertext) {
ciphertext->SetEncodingType( plaintext(row,col)->GetEncodingType() );
}
(*cipherResults)(row, col).SetNumerator(ciphertext);
}
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEncryptMatrixPlain, currentDateTime() - start) );
}
return cipherResults;
}
/**
* Perform an encryption by reading plaintext from a stream, serializing each piece of ciphertext,
* and writing the serializations to an output stream
* @param publicKey - the encryption key in use
* @param instream - where to read the input from
* @param ostream - where to write the serialization to
* @param doEncryption encrypts if true, embeds (encodes) the plaintext into cryptocontext if false
* @return
*/
void EncryptStream(
const LPPublicKey<Element> publicKey,
std::istream& instream,
std::ostream& outstream) const
{
// NOTE timing this operation is not supported
if( publicKey == NULL || Mismatched(publicKey->GetCryptoContext()) )
throw std::logic_error("key passed to EncryptStream was not generated with this crypto context");
bool padded = false;
Plaintext px;
size_t chunkSize = this->GetRingDimension();
char *ptxt = new char[chunkSize];
while (instream.good()) {
instream.read(ptxt, chunkSize);
size_t nRead = instream.gcount();
if (nRead <= 0 && padded)
break;
px = this->MakeStringPlaintext(std::string(ptxt,nRead));
if (nRead < chunkSize) {
padded = true;
}
Ciphertext<Element> ciphertext = GetEncryptionAlgorithm()->Encrypt(publicKey, px->GetEncodedElement<Element>());
if (!ciphertext) {
break;
}
ciphertext->SetEncodingType( px->GetEncodingType() );
Serialized cS;
if (ciphertext->Serialize(&cS)) {
if (!SerializableHelper::SerializationToStream(cS, outstream)) {
break;
}
}
else {
break;
}
}
delete [] ptxt;
return;
}
// PLAINTEXT FACTORY METHODS
/**
* MakeScalarPlaintext constructs a ScalarEncoding in this context
* @param value
* @param isSigned
* @return plaintext
*/
Plaintext MakeScalarPlaintext(int64_t value) const {
return Plaintext( new ScalarEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
/**
* MakeStringPlaintext constructs a StringEncoding in this context
* @param str
* @return plaintext
*/
Plaintext MakeStringPlaintext(const string& str) const {
return Plaintext( new StringEncoding( this->GetElementParams(), this->GetEncodingParams(), str ) );
}
/**
* MakeIntegerPlaintext constructs an IntegerEncoding in this context
* @param value
* @return plaintext
*/
Plaintext MakeIntegerPlaintext(int64_t value) const {
return Plaintext( new IntegerEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
/**
* MakeCoefPackedPlaintext constructs a CoefPackedEncoding in this context
* @param value
* @param isSigned
* @return plaintext
*/
Plaintext MakeCoefPackedPlaintext(const vector<int64_t>& value) const {
return Plaintext( new CoefPackedEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
/**
* MakeCoefPackedPlaintext constructs a CoefPackedEncoding in this context
* @param value
* @param isSigned
* @return plaintext
*/
Plaintext MakeCoefPackedPlaintext(const std::initializer_list<int64_t>& value) const {
return Plaintext( new CoefPackedEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
/**
* MakePackedPlaintext constructs a PackedEncoding in this context
* @param value
* @return plaintext
*/
Plaintext MakePackedPlaintext(const vector<uint64_t>& value) const {
return Plaintext( new PackedEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
/**
* MakePackedPlaintext constructs a PackedEncoding in this context
* @param value
* @return plaintext
*/
Plaintext MakePackedPlaintext(const std::initializer_list<uint64_t>& value) const {
return Plaintext( new PackedEncoding( this->GetElementParams(), this->GetEncodingParams(), value ) );
}
private:
static Plaintext
GetPlaintextForDecrypt(PlaintextEncodings pte, shared_ptr<typename Element::Params> evp, EncodingParams ep) {
Plaintext pt;
shared_ptr<typename NativePoly::Params> vp(
new typename NativePoly::Params(evp->GetCyclotomicOrder(), ep->GetPlaintextModulus(), 1) );
switch(pte) {
case Unknown:
throw std::logic_error("Unknown plaintext encoding type in GetPlaintextForDecrypt");
break;
case Scalar:
pt.reset( new ScalarEncoding(vp,ep) );
break;
case Integer:
pt.reset( new IntegerEncoding(vp,ep) );
break;
case CoefPacked:
pt.reset( new CoefPackedEncoding(vp,ep) );
break;
case Packed:
pt.reset( new PackedEncoding(vp,ep) );
break;
case String:
pt.reset( new StringEncoding(vp,ep) );
break;
}
return pt;
}
public:
/**
* Decrypt a single ciphertext into the appropriate plaintext
*
* @param privateKey - decryption key
* @param ciphertext - ciphertext to decrypt
* @param plaintext - resulting plaintext object pointer is here
* @return
*/
DecryptResult Decrypt(
const LPPrivateKey<Element> privateKey,
const Ciphertext<Element> ciphertext,
Plaintext* plaintext)
{
if( privateKey == NULL || Mismatched(privateKey->GetCryptoContext()) )
throw std::logic_error("Information passed to Decrypt was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
// determine which type of plaintext that you need to decrypt into
Plaintext decrypted = GetPlaintextForDecrypt(ciphertext->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
DecryptResult result = GetEncryptionAlgorithm()->Decrypt(privateKey, ciphertext, &decrypted->GetElement<NativePoly>());
if (result.isValid == false) return result;
decrypted->Decode();
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpDecrypt, currentDateTime() - start) );
}
*plaintext = decrypted;
return result;
}
/**
* Decrypt method for a matrix of ciphertexts
* @param privateKey - for decryption
* @param ciphertext - matrix of encrypted ciphertexts
* @param plaintext - pointer to the destination martrix of plaintexts
* @return size of plaintext
*/
DecryptResult DecryptMatrix(
const LPPrivateKey<Element> privateKey,
const shared_ptr<Matrix<RationalCiphertext<Element>>> ciphertext,
shared_ptr<Matrix<Plaintext>> *numerator,
shared_ptr<Matrix<Plaintext>> *denominator) const
{
// edge case
if ((ciphertext->GetCols()== 0) && (ciphertext->GetRows() == 0))
return DecryptResult();
if (privateKey == NULL || Mismatched(privateKey->GetCryptoContext()))
throw std::runtime_error("Information passed to DecryptMatrix was not generated with this crypto context");
const Ciphertext<Element> ctN = (*ciphertext)(0, 0).GetNumerator();
// need to build matrices for the result
Plaintext ptx = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
auto zeroPackingAlloc = [=]() { return lbcrypto::make_unique<Plaintext>(ptx); };
*numerator = shared_ptr<Matrix<Plaintext>>( new Matrix<Plaintext>(zeroPackingAlloc, ciphertext->GetRows(), ciphertext->GetCols()) );
*denominator = shared_ptr<Matrix<Plaintext>>( new Matrix<Plaintext>(zeroPackingAlloc, ciphertext->GetRows(), ciphertext->GetCols()) );
double start = 0;
if( doTiming ) start = currentDateTime();
for (size_t row = 0; row < ciphertext->GetRows(); row++)
{
for (size_t col = 0; col < ciphertext->GetCols(); col++)
{
if (Mismatched((*ciphertext)(row, col).GetCryptoContext()))
throw std::runtime_error("A ciphertext passed to DecryptMatrix was not generated with this crypto context");
const Ciphertext<Element> ctN = (*ciphertext)(row, col).GetNumerator();
// determine which type of plaintext that you need to decrypt into
Plaintext decryptedNumerator = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
DecryptResult resultN = GetEncryptionAlgorithm()->Decrypt(privateKey, ctN, &decryptedNumerator->GetElement<NativePoly>());
if (resultN.isValid == false) return resultN;
(**numerator)(row,col) = decryptedNumerator;
(**numerator)(row,col)->Decode();
Plaintext decryptedDenominator = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
if( (*ciphertext)(row,col).GetIntegerFlag() == true ) {
decryptedDenominator->GetElement<Poly>().SetValuesToZero();
decryptedDenominator->GetElement<Poly>().at(0) = 1;
}
else {
const Ciphertext<Element> ctD = (*ciphertext)(row, col).GetDenominator();
DecryptResult resultD = GetEncryptionAlgorithm()->Decrypt(privateKey, ctD, &decryptedDenominator->GetElement<NativePoly>());
if (resultD.isValid == false) return resultD;
(**denominator)(row,col) = decryptedDenominator;
}
(**denominator)(row, col)->Decode();
}
}
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpDecryptMatrixPlain, currentDateTime() - start) );
}
return DecryptResult((**numerator)((*numerator)->GetRows()-1,(*numerator)->GetCols()-1)->GetLength());
}
/**
* Decrypt method for numerators in a matrix of ciphertexts (packed encoding)
* @param privateKey - for decryption
* @param ciphertext - matrix of encrypted ciphertexts
* @param plaintext - pointer to the destination martrix of plaintexts
* @return size of plaintext
*/
DecryptResult DecryptMatrixNumerator(
const LPPrivateKey<Element> privateKey,
const shared_ptr<Matrix<RationalCiphertext<Element>>> ciphertext,
shared_ptr<Matrix<Plaintext>> *numerator) const
{
// edge case
if ((ciphertext->GetCols() == 0) && (ciphertext->GetRows() == 0))
return DecryptResult();
if (privateKey == NULL || Mismatched(privateKey->GetCryptoContext()))
throw std::runtime_error("Information passed to DecryptMatrix was not generated with this crypto context");
double start = 0;
if (doTiming) start = currentDateTime();
//force all precomputations to take place in advance
if( Mismatched((*ciphertext)(0, 0).GetCryptoContext()) )
throw std::runtime_error("A ciphertext passed to DecryptMatrix was not generated with this crypto context");
const Ciphertext<Element> ctN = (*ciphertext)(0, 0).GetNumerator();
// need to build a numerator matrix for the result
Plaintext ptx = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
auto zeroPackingAlloc = [=]() { return lbcrypto::make_unique<Plaintext>(ptx); };
*numerator = shared_ptr<Matrix<Plaintext>>( new Matrix<Plaintext>(zeroPackingAlloc, ciphertext->GetRows(), ciphertext->GetCols()) );
Plaintext decryptedNumerator = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
DecryptResult resultN = GetEncryptionAlgorithm()->Decrypt(privateKey, ctN, &decryptedNumerator->GetElement<NativePoly>());
if (resultN.isValid == false) return resultN;
(**numerator)(0, 0) = decryptedNumerator;
(**numerator)(0, 0)->Decode();
for (size_t row = 0; row < ciphertext->GetRows(); row++)
{
#pragma omp parallel for
for (size_t col = 0; col < ciphertext->GetCols(); col++)
{
if (row + col > 0)
{
if( Mismatched((*ciphertext)(row, col).GetCryptoContext()) )
throw std::runtime_error("A ciphertext passed to DecryptMatrix was not generated with this crypto context");
const Ciphertext<Element> ctN = (*ciphertext)(row, col).GetNumerator();
Plaintext decryptedNumerator = GetPlaintextForDecrypt(ctN->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
GetEncryptionAlgorithm()->Decrypt(privateKey, ctN, &decryptedNumerator->GetElement<NativePoly>());
(**numerator)(row, col) = decryptedNumerator;
(**numerator)(row, col)->Decode();
}
}
}
if (doTiming) {
timeSamples->push_back(TimingInfo(OpDecryptMatrixPacked, currentDateTime() - start));
}
return DecryptResult((**numerator)((*numerator)->GetRows() - 1, (*numerator)->GetCols() - 1)->GetLength());
}
/**
* read instream for a sequence of serialized ciphertext; deserialize it, decrypt it, and write it to outstream
* @param privateKey - reference to the decryption key
* @param instream - input stream with sequence of serialized ciphertexts
* @param outstream - output stream for plaintext
* @return total bytes processed
*/
size_t DecryptStream(
const LPPrivateKey<Element> privateKey,
std::istream& instream,
std::ostream& outstream)
{
// NOTE timing this operation is not supported
if( privateKey == NULL || Mismatched(privateKey->GetCryptoContext()) )
throw std::logic_error("Information passed to DecryptStream was not generated with this crypto context");
Serialized serObj;
size_t tot = 0;
bool firstTime = true;
Plaintext pte[2];
bool whichArray = false;
while( SerializableHelper::StreamToSerialization(instream, &serObj) ) {
Ciphertext<Element> ct;
if( (ct = deserializeCiphertext(serObj)) != NULL ) {
if( ct->GetEncodingType() != String ) {
throw std::logic_error("Library can only stream string encodings");
}
pte[whichArray] = GetPlaintextForDecrypt(ct->GetEncodingType(), this->GetElementParams(), this->GetEncodingParams());
DecryptResult res = GetEncryptionAlgorithm()->Decrypt(privateKey, ct, &pte[whichArray]->GetElement<NativePoly>());
if( !res.isValid )
return tot;
tot += res.messageLength;
pte[whichArray]->Decode();
if( !firstTime ) {
outstream << pte[!whichArray]->GetStringValue();
}
firstTime = false;
whichArray = !whichArray;
}
else
return tot;
}
outstream << pte[!whichArray]->GetStringValue();
return tot;
}
/**
* ReEncrypt - Proxy Re Encryption mechanism for PALISADE
* @param evalKey - evaluation key from the PRE keygen method
* @param ciphertext - vector of shared pointers to encrypted Ciphertext
* @return vector of shared pointers to re-encrypted ciphertexts
*/
Ciphertext<Element> ReEncrypt(
LPEvalKey<Element> evalKey,
Ciphertext<Element> ciphertext) const
{
if( evalKey == NULL || Mismatched(evalKey->GetCryptoContext()) )
throw std::logic_error("Information passed to ReEncrypt was not generated with this crypto context");
if( ciphertext == NULL || Mismatched(ciphertext->GetCryptoContext()) )
throw std::logic_error("The ciphertext passed to ReEncrypt was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
Ciphertext<Element> newCiphertext = GetEncryptionAlgorithm()->ReEncrypt(evalKey, ciphertext);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpReEncrypt, currentDateTime() - start) );
}
return newCiphertext;
}
/**
* read instream for a serialized ciphertext. deserialize, re-encrypt, serialize, and write to outstream
* @param evalKey - reference to the re-encryption key
* @param instream - input stream with sequence of serialized ciphertext
* @param outstream - output stream with sequence of serialized re-encrypted ciphertext
*/
void ReEncryptStream(
const LPEvalKey<Element> evalKey,
std::istream& instream,
std::ostream& outstream)
{
// NOTE timing this operation is not supported
if( evalKey == NULL || Mismatched(evalKey->GetCryptoContext()) )
throw std::logic_error("Information passed to ReEncryptStream was not generated with this crypto context");
Serialized serObj;
while( SerializableHelper::StreamToSerialization(instream, &serObj) ) {
Ciphertext<Element> ct;
ct = deserializeCiphertext(serObj);
if( ct ) {
Ciphertext<Element> reCt = ReEncrypt(evalKey, ct);
Serialized serReObj;
if( reCt->Serialize(&serReObj) ) {
SerializableHelper::SerializationToStream(serReObj, outstream);
}
else {
return;
}
}
else {
return;
}
}
}
/**
* EvalAdd - PALISADE EvalAdd method for a pair of ciphertexts
* @param ct1
* @param ct2
* @return new ciphertext for ct1 + ct2
*/
Ciphertext<Element>
EvalAdd(const Ciphertext<Element> ct1, const Ciphertext<Element> ct2) const
{
TypeCheck(ct1, ct2);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalAdd(ct1, ct2);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAdd, currentDateTime() - start) );
}
return rv;
}
/**
* EvalAddMatrix - PALISADE EvalAdd method for a pair of matrices of ciphertexts
* @param ct1
* @param ct2
* @return new matrix for ct1 + ct2
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalAddMatrix(const shared_ptr<Matrix<RationalCiphertext<Element>>> ct1, const shared_ptr<Matrix<RationalCiphertext<Element>>> ct2) const
{
TypeCheck((*ct1)(0,0), (*ct2)(0,0)); // TODO only checking one; when Matrix is refactored, this should be revisited
double start = 0;
if( doTiming ) start = currentDateTime();
Matrix<RationalCiphertext<Element>> rv = *ct1 + *ct2;
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAddMatrix, currentDateTime() - start) );
}
shared_ptr<Matrix<RationalCiphertext<Element>>> a(new Matrix<RationalCiphertext<Element>>(rv));
return a;
}
/**
* EvalSub - PALISADE EvalSub method for a pair of ciphertexts
* @param ct1
* @param ct2
* @return new ciphertext for ct1 - ct2
*/
Ciphertext<Element>
EvalSub(const Ciphertext<Element> ct1, const Ciphertext<Element> ct2) const
{
TypeCheck(ct1, ct2);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalSub(ct1, ct2);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalSub, currentDateTime() - start) );
}
return rv;
}
/**
* EvalSubMatrix - PALISADE EvalSub method for a pair of matrices of ciphertexts
* @param ct1
* @param ct2
* @return new matrix for ct1 + ct2
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalSubMatrix(const shared_ptr<Matrix<RationalCiphertext<Element>>> ct1, const shared_ptr<Matrix<RationalCiphertext<Element>>> ct2) const
{
TypeCheck((*ct1)(0,0), (*ct2)(0,0)); // TODO only checking one; when Matrix is refactored, this should be revisited
double start = 0;
if( doTiming ) start = currentDateTime();
Matrix<RationalCiphertext<Element>> rv = *ct1 - *ct2;
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalSubMatrix, currentDateTime() - start) );
}
shared_ptr<Matrix<RationalCiphertext<Element>>> a(new Matrix<RationalCiphertext<Element>>(rv));
return a;
}
/**
* EvalAdd - PALISADE EvalAdd method for a ciphertext and plaintext
* @param ciphertext
* @param plaintext
* @return new ciphertext for ciphertext + plaintext
*/
Ciphertext<Element>
EvalAdd(const Ciphertext<Element> ciphertext, const Plaintext plaintext) const
{
TypeCheck(ciphertext, plaintext);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalAdd(ciphertext, plaintext);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAddPlain, currentDateTime() - start) );
}
return rv;
}
/**
* EvalSubPlain - PALISADE EvalSub method for a ciphertext and plaintext
* @param ciphertext
* @param plaintext
* @return new ciphertext for ciphertext - plaintext
*/
Ciphertext<Element>
EvalSub(const Ciphertext<Element> ciphertext, const Plaintext plaintext) const
{
TypeCheck(ciphertext, plaintext);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalSub(ciphertext, plaintext);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalSubPlain, currentDateTime() - start) );
}
return rv;
}
/**
* EvalMult - PALISADE EvalMult method for a pair of ciphertexts - with key switching
* @param ct1
* @param ct2
* @return new ciphertext for ct1 * ct2
*/
Ciphertext<Element>
EvalMult(const Ciphertext<Element> ct1, const Ciphertext<Element> ct2) const
{
TypeCheck(ct1, ct2);
auto ek = GetEvalMultKeyVector(ct1->GetKeyTag());
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalMult(ct1, ct2, ek[0]);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* EvalMult - PALISADE EvalMult method for a pair of ciphertexts - no key switching (relinearization)
* @param ct1
* @param ct2
* @return new ciphertext for ct1 * ct2
*/
Ciphertext<Element>
EvalMultNoRelin(const Ciphertext<Element> ct1, const Ciphertext<Element> ct2) const
{
TypeCheck(ct1, ct2);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalMult(ct1, ct2);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* EvalMultMany - PALISADE function for evaluating multiplication on ciphertext followed by relinearization operation (at the end).
* It computes the multiplication in a binary tree manner. Also, it reduces the number of
* elements in the ciphertext to two after each multiplication.
* Currently it assumes that the consecutive two input arguments have
* total depth smaller than the supported depth. Otherwise, it throws an error.
*
* @param cipherTextList is the ciphertext list.
*
* @return new ciphertext.
*/
Ciphertext<Element> EvalMultMany(const vector<Ciphertext<Element>>& ct) const{
const auto ek = GetEvalMultKeyVector(ct[0]->GetKeyTag());
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalMultMany(ct, ek);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* Function for evaluating multiplication on ciphertext followed by relinearization operation.
* Currently it assumes that the input arguments have total depth smaller than the supported depth. Otherwise, it throws an error.
*
* @param ct1 first input ciphertext.
* @param ct2 second input ciphertext.
*
* @return new ciphertext
*/
Ciphertext<Element> EvalMultAndRelinearize(const Ciphertext<Element> ct1,
const Ciphertext<Element> ct2) const{
const auto ek = GetEvalMultKeyVector(ct1->GetKeyTag());
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalMultAndRelinearize(ct1, ct2, ek);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* EvalMult - PALISADE EvalMult method for plaintext * ciphertext
* @param pt2
* @param ct1
* @return new ciphertext for ct1 * pt2
*/
Ciphertext<Element>
EvalMult(const Plaintext pt2, const Ciphertext<Element> ct1) const
{
return EvalMult(ct1, pt2);
}
/**
* EvalMult - PALISADE EvalMult method for plaintext * ciphertext
* @param ct1
* @param pt2
* @return new ciphertext for ct1 * pt2
*/
Ciphertext<Element>
EvalMult(const Ciphertext<Element> ct1, const Plaintext pt2) const
{
TypeCheck(ct1, pt2);
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalMult(ct1, pt2);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* EvalMultMatrix - PALISADE EvalMult method for two matrices of ciphertext
* @param ct1
* @param ct2
* @return new matrix for ct1 * ct2
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalMultMatrix(const shared_ptr<Matrix<RationalCiphertext<Element>>> ct1, const shared_ptr<Matrix<RationalCiphertext<Element>>> ct2) const
{
TypeCheck((*ct1)(0,0), (*ct2)(0,0)); // TODO only checking one; when Matrix is refactored, this should be revisited
double start = 0;
if( doTiming ) start = currentDateTime();
Matrix<RationalCiphertext<Element>> rv = *ct1 * *ct2;
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalMultMatrix, currentDateTime() - start) );
}
shared_ptr<Matrix<RationalCiphertext<Element>>> a(new Matrix<RationalCiphertext<Element>>(rv));
return a;
}
/**
* EvalSub - PALISADE Negate method for a ciphertext
* @param ct
* @return new ciphertext -ct
*/
Ciphertext<Element>
EvalNegate(const Ciphertext<Element> ct) const
{
if (ct == NULL || Mismatched(ct->GetCryptoContext()) )
throw std::logic_error("Information passed to EvalNegate was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalNegate(ct);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalNeg, currentDateTime() - start) );
}
return rv;
}
/**
* EvalSub - PALISADE Negate method for a ciphertext
* @param ct
* @return new ciphertext -ct
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalNegateMatrix(const shared_ptr<Matrix<RationalCiphertext<Element>>> ct) const
{
if (ct == NULL || Mismatched((*ct)(0,0).GetCryptoContext()) )
throw std::logic_error("Information passed to EvalNegateMatrix was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
shared_ptr<Matrix<RationalCiphertext<Element>>> m(
new Matrix<RationalCiphertext<Element>>(ct->GetAllocator(), ct->GetRows(), ct->GetCols()));
for( size_t r = 0; r < m->GetRows(); r++ )
for( size_t c = 0; c < m->GetCols(); c++ )
(*m)(r,c) = -((*ct)(r,c));
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalNegMatrix, currentDateTime() - start) );
}
return m;
}
/**
* Generate automophism keys for a given private key
*
* @param publicKey original public key.
* @param origPrivateKey original private key.
* @param indexList list of automorphism indices to be computed
* @return returns the evaluation keys; index 0 of the vector corresponds to plaintext index 2, index 1 to plaintex index 3, etc.
*/
shared_ptr<std::map<usint, LPEvalKey<Element>>> EvalAutomorphismKeyGen(const LPPublicKey<Element> publicKey,
const LPPrivateKey<Element> origPrivateKey, const std::vector<usint> &indexList) const {
if( publicKey == NULL || origPrivateKey == NULL )
PALISADE_THROW( type_error, "Null Keys");
if( publicKey->GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Key was not created in this CryptoContextImpl");
if( publicKey->GetCryptoContext() != origPrivateKey->GetCryptoContext() )
PALISADE_THROW( type_error, "Keys were not created in the same CryptoContextImpl");
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalAutomorphismKeyGen(publicKey, origPrivateKey, indexList);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAutomorphismKeyGen, currentDateTime() - start) );
}
return rv;
}
/**
* Function for evaluating automorphism of ciphertext at index i
*
* @param ciphertext the input ciphertext.
* @param i automorphism index
* @param &evalKeys - reference to the vector of evaluation keys generated by EvalAutomorphismKeyGen.
* @return resulting ciphertext
*/
Ciphertext<Element> EvalAutomorphism(const Ciphertext<Element> ciphertext, usint i,
const std::map<usint, LPEvalKey<Element>> &evalKeys) const {
auto mf = evalKeys.begin();
if( mf == evalKeys.end() )
PALISADE_THROW( type_error, "Empty key map");
auto tk = mf->second;
if( ciphertext == NULL || tk == NULL )
PALISADE_THROW( type_error, "Null inputs");
if( ciphertext->GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Ciphertext was not created in this CryptoContextImpl");
if( ciphertext->GetCryptoContext() != tk->GetCryptoContext() )
PALISADE_THROW( type_error, "Items were not created in the same CryptoContextImpl");
if( ciphertext->GetKeyTag() != tk->GetKeyTag() )
PALISADE_THROW( type_error, "Items were not encrypted with same keys" );
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalAutomorphism(ciphertext, i, evalKeys);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAutomorphismI, currentDateTime() - start) );
}
return rv;
}
/**
* Generate automophism keys for a given private key; Uses the private key for encryption
*
* @param privateKey private key.
* @param indexList list of automorphism indices to be computed
* @return returns the evaluation keys
*/
shared_ptr<std::map<usint, LPEvalKey<Element>>> EvalAutomorphismKeyGen(const LPPrivateKey<Element> privateKey,
const std::vector<usint> &indexList) const {
if( privateKey == NULL )
PALISADE_THROW( type_error, "Null input");
if( privateKey->GetCryptoContext().get() != this )
PALISADE_THROW( type_error, "Key was not created in this CryptoContextImpl");
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalAutomorphismKeyGen(privateKey, indexList);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpEvalAutomorphismK, currentDateTime() - start) );
}
return rv;
}
/**
* EvalSumKeyGen Generates the key map to be used by evalsum
*
* @param privateKey private key.
* @param publicKey public key (used in NTRU schemes).
*/
void EvalSumKeyGen(
const LPPrivateKey<Element> privateKey,
const LPPublicKey<Element> publicKey = nullptr);
/**
* GetEvalSumKey returns the map
*
* @return the EvalSum key map
*/
static const std::map<usint, LPEvalKey<Element>>& GetEvalSumKeyMap(const string& id);
static const std::map<string,shared_ptr<std::map<usint, LPEvalKey<Element>>>>& GetAllEvalSumKeys();
/**
* Function for evaluating a sum of all components
*
* @param ciphertext the input ciphertext.
* @param batchSize size of the batch
* @return resulting ciphertext
*/
Ciphertext<Element> EvalSum(const Ciphertext<Element> ciphertext, usint batchSize) const;
/**
* Evaluates inner product in batched encoding
*
* @param ciphertext1 first vector.
* @param ciphertext2 second vector.
* @param batchSize size of the batch to be summed up
* @return resulting ciphertext
*/
Ciphertext<Element> EvalInnerProduct(const Ciphertext<Element> ciphertext1, const Ciphertext<Element> ciphertext2, usint batchSize) const;
/**
* Evaluates inner product in batched encoding
*
* @param ciphertext1 first vector.
* @param ciphertext2 second vector.
* @param batchSize size of the batch to be summed up
* @return resulting ciphertext
*/
Ciphertext<Element> EvalInnerProduct(const Ciphertext<Element> ciphertext1, const Plaintext ciphertext2, usint batchSize) const;
/**
* EvalCrossCorrelation - Computes the sliding sum of inner products (known as
* as cross-correlation, sliding inner product, or sliding dot product in
* image processing
* @param x - first vector of row vectors
* @param y - second vector of row vectors
* @param batchSize - batch size for packed encoding
* @param indexStart - starting index in the vectors of row vectors
* @param length - length of the slice in the vectors of row vectors; default is 0 meaning to use the full length of the vector
* @return sum(x_i*y_i), i.e., a sum of inner products
*/
Ciphertext<Element>
EvalCrossCorrelation(const shared_ptr<Matrix<RationalCiphertext<Element>>> x,
const shared_ptr<Matrix<RationalCiphertext<Element>>> y, usint batchSize,
usint indexStart = 0, usint length = 0) const;
/**
* EvalLinRegressBatched- Computes the parameter vector for linear regression using the least squares method
* Supported only in batched mode; currently works only for two regressors
* @param x - matrix of regressors
* @param y - vector of dependent variables
* @return the parameter vector using (x^T x)^{-1} x^T y (using least squares method)
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalLinRegressBatched(const shared_ptr<Matrix<RationalCiphertext<Element>>> x,
const shared_ptr<Matrix<RationalCiphertext<Element>>> y, usint batchSize) const;
/**
* EvalLinRegression - Computes the parameter vector for linear regression using the least squares method
* @param x - matrix of regressors
* @param y - vector of dependent variables
* @return the parameter vector using (x^T x)^{-1} x^T y (using least squares method)
*/
shared_ptr<Matrix<RationalCiphertext<Element>>>
EvalLinRegression(const shared_ptr<Matrix<RationalCiphertext<Element>>> x,
const shared_ptr<Matrix<RationalCiphertext<Element>>> y) const
{
TypeCheck((*x)(0,0), (*y)(0,0)); // TODO only checking one; when Matrix is refactored, this should be revisited
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->EvalLinRegression(x, y);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpLinRegression, currentDateTime() - start) );
}
return rv;
}
/**
* KeySwitch - PALISADE KeySwitch method
* @param keySwitchHint - reference to KeySwitchHint
* @param ciphertext - vector of ciphertext
* @return new CiphertextImpl after applying key switch
*/
Ciphertext<Element> KeySwitch(
const LPEvalKey<Element> keySwitchHint,
const Ciphertext<Element> ciphertext) const
{
if( keySwitchHint == NULL || Mismatched(keySwitchHint->GetCryptoContext()) )
throw std::logic_error("Key passed to KeySwitch was not generated with this crypto context");
if( ciphertext == NULL || Mismatched(ciphertext->GetCryptoContext()) )
throw std::logic_error("Ciphertext passed to KeySwitch was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->KeySwitch(keySwitchHint, ciphertext);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpKeySwitch, currentDateTime() - start) );
}
return rv;
}
/**
* ModReduce - PALISADE ModReduce method
* @param ciphertext - vector of ciphertext
* @return vector of mod reduced ciphertext
*/
Ciphertext<Element> ModReduce(Ciphertext<Element> ciphertext) const {
if( ciphertext == NULL || Mismatched(ciphertext->GetCryptoContext()) )
throw std::logic_error("Information passed to ModReduce was not generated with this crypto context");
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->ModReduce(ciphertext);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpModReduce, currentDateTime() - start) );
}
return rv;
}
/**
* ModReduce - PALISADE ModReduce method
* @param ciphertext - vector of ciphertext
* @return vector of mod reduced ciphertext
*/
RationalCiphertext<Element> ModReduceRational(RationalCiphertext<Element> ciphertext) const {
double start = 0;
if( doTiming ) start = currentDateTime();
Ciphertext<Element> n = GetEncryptionAlgorithm()->ModReduce(ciphertext.GetNumerator());
Ciphertext<Element> d = GetEncryptionAlgorithm()->ModReduce(ciphertext.GetDenominator());
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpModReduce, currentDateTime() - start) );
}
return RationalCiphertext<Element>(n,d);
}
/**
* ModReduce - PALISADE ModReduce method
* @param ciphertext - vector of ciphertext
* @return vector of mod reduced ciphertext
*/
shared_ptr<Matrix<RationalCiphertext<Element>>> ModReduceMatrix(shared_ptr<Matrix<RationalCiphertext<Element>>> ciphertext) const {
// needs context check
double start = 0;
if( doTiming ) start = currentDateTime();
shared_ptr<Matrix<RationalCiphertext<Element>>> m(
new Matrix<RationalCiphertext<Element>>(ciphertext->GetAllocator(), ciphertext->GetRows(), ciphertext->GetCols()));
for( size_t r = 0; r < m->GetRows(); r++ )
for( size_t c = 0; c < m->GetCols(); c++ )
(*m)(r,c) = ModReduceRational((*ciphertext)(r,c));
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpModReduceMatrix, currentDateTime() - start) );
}
return m;
}
/**
* LevelReduce - PALISADE LevelReduce method
* @param cipherText1
* @param linearKeySwitchHint
* @return vector of level reduced ciphertext
*/
Ciphertext<Element> LevelReduce(const Ciphertext<Element> cipherText1,
const LPEvalKeyNTRU<Element> linearKeySwitchHint) const {
if( cipherText1 == NULL || linearKeySwitchHint == NULL ||
Mismatched(cipherText1->GetCryptoContext()) ||
Mismatched(linearKeySwitchHint->GetCryptoContext()) ) {
throw std::logic_error("Information passed to LevelReduce was not generated with this crypto context");
}
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->LevelReduce(cipherText1, linearKeySwitchHint);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpLevelReduce, currentDateTime() - start) );
}
return rv;
}
/**
* RingReduce - PALISADE RingReduce method
* @param ciphertext - vector of ciphertext
* @param keySwitchHint - the keySwitchHint from original private key to sparse private key
* @return vector of ring-reduced ciphertexts
*/
Ciphertext<Element> RingReduce(
Ciphertext<Element> ciphertext,
const LPEvalKey<Element> keySwitchHint) const
{
if( keySwitchHint == NULL ||
Mismatched(keySwitchHint->GetCryptoContext()) )
throw std::logic_error("Key passed to RingReduce was not generated with this crypto context");
if( ciphertext == NULL || Mismatched(ciphertext->GetCryptoContext()) )
throw std::logic_error("Ciphertext passed to RingReduce was not generated with this crypto context");
Ciphertext<Element> newCiphertext;
double start = 0;
if( doTiming ) start = currentDateTime();
newCiphertext = GetEncryptionAlgorithm()->RingReduce(ciphertext, keySwitchHint);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpRingReduce, currentDateTime() - start) );
}
return newCiphertext;
}
/**
* ComposedEvalMult - PALISADE composed evalmult
* @param ciphertext1 - vector for first cipher text
* @param ciphertext2 - vector for second cipher text
* @param quadKeySwitchHint - is the quadratic key switch hint from original private key to the quadratic key
* return vector of resulting ciphertext
*/
Ciphertext<Element> ComposedEvalMult(
const Ciphertext<Element> ciphertext1,
const Ciphertext<Element> ciphertext2) const
{
if( ciphertext1 == NULL || ciphertext2 == NULL || ciphertext1->GetKeyTag() != ciphertext2->GetKeyTag() ||
Mismatched(ciphertext1->GetCryptoContext()) )
throw std::logic_error("Ciphertexts passed to ComposedEvalMult were not generated with this crypto context");
auto ek = GetEvalMultKeyVector(ciphertext1->GetKeyTag());
double start = 0;
if( doTiming ) start = currentDateTime();
auto rv = GetEncryptionAlgorithm()->ComposedEvalMult(ciphertext1, ciphertext2, ek[0]);
if( doTiming ) {
timeSamples->push_back( TimingInfo(OpComposedEvalMult, currentDateTime() - start) );
}
return rv;
}
/**
* Deserialize into a Public Key
* @param serObj
* @return deserialized object
*/
static LPPublicKey<Element> deserializePublicKey(const Serialized& serObj);
/**
* Deserialize into a Private Key
* @param serObj
* @return deserialized object
*/
static LPPrivateKey<Element> deserializeSecretKey(const Serialized& serObj);
/**
* Deserialize into a Ciphertext
* @param serObj
* @return deserialized object
*/
static Ciphertext<Element> deserializeCiphertext(const Serialized& serObj);
/**
* Deserialize into an Eval Key in a given context
* @param serObj
* @return deserialized object
*/
static LPEvalKey<Element> deserializeEvalKey(const Serialized& serObj);
/**
* Deserialize into an Eval Key
* @param serObj
* @return deserialized object
*/
static LPEvalKey<Element> deserializeEvalKeyInContext(const Serialized& serObj, CryptoContext<Element> cc);
};
/**
* @brief CryptoObject
*
* A class to aid in referring to the crypto context that an object belongs to
*/
template<typename Element>
class CryptoObject {
protected:
CryptoContext<Element> context; /*!< crypto context this object belongs to */
string keyTag; /*!< tag used to find the evaluation key needed for SHE/FHE operations */
public:
CryptoObject(CryptoContext<Element> cc = 0, const string& tag = "") : context(cc), keyTag(tag) {}
CryptoObject(const CryptoObject& rhs) {
context = rhs.context;
keyTag = rhs.keyTag;
}
CryptoObject(const CryptoObject&& rhs) {
context = std::move(rhs.context);
keyTag = std::move(rhs.keyTag);
}
virtual ~CryptoObject() {}
const CryptoObject& operator=(const CryptoObject& rhs) {
this->context = rhs.context;
this->keyTag = rhs.keyTag;
return *this;
}
const CryptoObject& operator=(const CryptoObject&& rhs) {
this->context = std::move(rhs.context);
this->keyTag = std::move(rhs.keyTag);
return *this;
}
bool operator==(const CryptoObject& rhs) const {
return context.get() == rhs.context.get() &&
keyTag == rhs.keyTag;
}
CryptoContext<Element> GetCryptoContext() const { return context; }
const shared_ptr<LPCryptoParameters<Element>> GetCryptoParameters() const { return context->GetCryptoParameters(); }
const EncodingParams GetEncodingParameters() const { return context->GetCryptoParameters()->GetEncodingParams(); }
const string GetKeyTag() const { return keyTag; }
void SetKeyTag(const string& tag) { keyTag = tag; }
/**
* SerializeCryptoObject serializes this header into a Serialized
* @param serObj is used to store the serialized result.
* @return true if successfully serialized
*/
bool SerializeCryptoObject(Serialized* serObj, bool includeContext = true) const;
/**
* DeserializeCryptoObject Populates this header from the deserialization of the Serialized
* @param serObj contains the serialized object
* @return true on success
*/
bool DeserializeCryptoObject(const Serialized& serObj, bool includesContext = true);
};
/**
* @brief CryptoContextFactory
*
* A class that contains static methods to generate new crypto contexts from user parameters
*
*/
template<typename Element>
class CryptoContextFactory {
static vector<CryptoContext<Element>> AllContexts;
public:
static void ReleaseAllContexts();
static int GetContextCount();
static CryptoContext<Element> GetSingleContext();
static CryptoContext<Element> GetContext(
shared_ptr<LPCryptoParameters<Element>> params,
shared_ptr<LPPublicKeyEncryptionScheme<Element>> scheme);
static CryptoContext<Element> GetContextForPointer(
CryptoContextImpl<Element>* cc);
static const vector<CryptoContext<Element>>& GetAllContexts() { return AllContexts; }
/**
* construct a PALISADE CryptoContextImpl for the LTV Scheme
* @param plaintextmodulus
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param depth
* @return new context
*/
static CryptoContext<Element> genCryptoContextLTV(shared_ptr<typename Element::Params> params,
const PlaintextModulus plaintextmodulus,
usint relinWindow, float stDev, int depth = 1, int assuranceMeasure = 9, float securityLevel = 1.006);
/**
* construct a PALISADE CryptoContextImpl for the LTV Scheme
* @param encodingParams
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param depth
* @return new context
*/
static CryptoContext<Element> genCryptoContextLTV(shared_ptr<typename Element::Params> params,
EncodingParams encodingParams,
usint relinWindow, float stDev, int depth = 1, int assuranceMeasure = 9, float securityLevel = 1.006);
/**
* construct a PALISADE CryptoContextImpl for the LTV Scheme using the scheme's ParamsGen methods
* @param plaintextModulus
* @param securityLevel
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextLTV(
const PlaintextModulus plaintextModulus, float securityLevel, usint relinWindow, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches);
/**
* construct a PALISADE CryptoContextImpl for the LTV Scheme using the scheme's ParamsGen methods
* @param encodingParams
* @param securityLevel
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextLTV(
EncodingParams encodingParams, float securityLevel, usint relinWindow, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches);
/**
* construct a PALISADE CryptoContextImpl for the BFV Scheme
* @param plaintextmodulus
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param delta
* @param mode
* @param bigmodulus
* @param bigrootofunity
* @param depth
* @param assuranceMeasure
* @param securityLevel
* @param bigmodulusarb
* @param bigrootofunityarb
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFV(shared_ptr<typename Element::Params> params,
const PlaintextModulus plaintextmodulus,
usint relinWindow, float stDev, const std::string& delta,
MODE mode = RLWE, const std::string& bigmodulus = "0", const std::string& bigrootofunity = "0",
int depth = 0, int assuranceMeasure = 0, float securityLevel = 0,
const std::string& bigmodulusarb = "0", const std::string& bigrootofunityarb = "0", int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BFV Scheme
* @param encodingParams
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param delta
* @param mode
* @param bigmodulus
* @param bigrootofunity
* @param depth
* @param assuranceMeasure
* @param securityLevel
* @param bigmodulusarb
* @param bigrootofunityarb
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFV(shared_ptr<typename Element::Params> params,
EncodingParams encodingParams,
usint relinWindow, float stDev, const std::string& delta,
MODE mode = RLWE, const std::string& bigmodulus = "0", const std::string& bigrootofunity = "0",
int depth = 0, int assuranceMeasure = 0, float securityLevel = 0,
const std::string& bigmodulusarb = "0", const std::string& bigrootofunityarb = "0", int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BFV Scheme using the scheme's ParamsGen methods
* @param plaintextModulus
* @param securityLevel
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFV(
const PlaintextModulus plaintextModulus, float securityLevel, usint relinWindow, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches, MODE mode = OPTIMIZED, int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BFV Scheme using the scheme's ParamsGen methods
* @param encodingParams
* @param securityLevel
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFV(
EncodingParams encodingParams, float securityLevel, usint relinWindow, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches, MODE mode = OPTIMIZED, int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BFVrns Scheme using the scheme's ParamsGen methods
* @param plaintextModulus
* @param securityLevel
* @param distribution parameter
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFVrns(
const PlaintextModulus plaintextModulus, float securityLevel, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches, MODE mode = OPTIMIZED, int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BFVrns Scheme using the scheme's ParamsGen methods
* @param encodingParams
* @param securityLevel
* @param distribution parameter
* @param numAdds
* @param numMults
* @param numKeyswitches
* @return new context
*/
static CryptoContext<Element> genCryptoContextBFVrns(
EncodingParams encodingParams, float securityLevel, float dist,
unsigned int numAdds, unsigned int numMults, unsigned int numKeyswitches, MODE mode = OPTIMIZED, int maxDepth = 2);
/**
* construct a PALISADE CryptoContextImpl for the BGV Scheme
* @param plaintextmodulus
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param mode
* @return new context
*/
static CryptoContext<Element> genCryptoContextBGV(shared_ptr<typename Element::Params> params,
const PlaintextModulus plaintextmodulus,
usint relinWindow, float stDev,
MODE mode = RLWE, int depth = 1);
/**
* construct a PALISADE CryptoContextImpl for the BGV Scheme
* @param encodingParams
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param mode
* @return new context
*/
static CryptoContext<Element> genCryptoContextBGV(shared_ptr<typename Element::Params> params,
EncodingParams encodingParams,
usint relinWindow, float stDev,
MODE mode = RLWE, int depth = 1);
/**
* construct a PALISADE CryptoContextImpl for the StehleSteinfeld Scheme
* @param plaintextmodulus
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param stDevStSt
* @return new context
*/
static CryptoContext<Element> genCryptoContextStehleSteinfeld(shared_ptr<typename Element::Params> params,
const PlaintextModulus plaintextmodulus,
usint relinWindow, float stDev, float stDevStSt, int depth = 1, int assuranceMeasure = 9, float securityLevel = 1.006);
/**
* construct a PALISADE CryptoContextImpl for the StehleSteinfeld Scheme
* @param encodingParams
* @param ringdim
* @param modulus
* @param rootOfUnity
* @param relinWindow
* @param stDev
* @param stDevStSt
* @return new context
*/
static CryptoContext<Element> genCryptoContextStehleSteinfeld(shared_ptr<typename Element::Params> params,
EncodingParams encodingParams,
usint relinWindow, float stDev, float stDevStSt, int depth = 1, int assuranceMeasure = 9, float securityLevel = 1.006);
/**
* construct a PALISADE CryptoContextImpl for the Null Scheme
* @param plaintext modulus
* @return
*/
static CryptoContext<Element> genCryptoContextNull(unsigned int m, const PlaintextModulus ptModulus);
/**
* construct a PALISADE CryptoContextImpl for the Null Scheme
* @param encodingParams
* @return
*/
static CryptoContext<Element> genCryptoContextNull(unsigned int m, EncodingParams encodingParams);
/**
* Create a PALISADE CryptoContextImpl from a serialization
* @param serObj
* @return new context
*/
static CryptoContext<Element> DeserializeAndCreateContext(const Serialized& serObj);
};
}
#endif /* SRC_DEMO_PRE_CRYPTOCONTEXT_H_ */
|
GB_unop__isfinite_bool_fc64.c | //------------------------------------------------------------------------------
// GB_unop: hard-coded functions for each built-in unary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2021, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_atomics.h"
#include "GB_unop__include.h"
// C=unop(A) is defined by the following types and operators:
// op(A) function: GB (_unop_apply__isfinite_bool_fc64)
// op(A') function: GB (_unop_tran__isfinite_bool_fc64)
// C type: bool
// A type: GxB_FC64_t
// cast: GxB_FC64_t cij = (aij)
// unaryop: cij = GB_cisfinite (aij)
#define GB_ATYPE \
GxB_FC64_t
#define GB_CTYPE \
bool
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
GxB_FC64_t aij = Ax [pA]
#define GB_CX(p) Cx [p]
// unary operator
#define GB_OP(z, x) \
z = GB_cisfinite (x) ;
// casting
#define GB_CAST(z, aij) \
GxB_FC64_t z = (aij) ;
// cij = op (aij)
#define GB_CAST_OP(pC,pA) \
{ \
/* aij = Ax [pA] */ \
GxB_FC64_t aij = Ax [pA] ; \
/* Cx [pC] = op (cast (aij)) */ \
GxB_FC64_t z = (aij) ; \
Cx [pC] = GB_cisfinite (z) ; \
}
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_ISFINITE || GxB_NO_BOOL || GxB_NO_FC64)
//------------------------------------------------------------------------------
// Cx = op (cast (Ax)): apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_apply__isfinite_bool_fc64)
(
bool *Cx, // Cx and Ax may be aliased
const GxB_FC64_t *Ax,
const int8_t *restrict Ab, // A->b if A is bitmap
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
if (Ab == NULL)
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
GxB_FC64_t aij = Ax [p] ;
GxB_FC64_t z = (aij) ;
Cx [p] = GB_cisfinite (z) ;
}
}
else
{
// bitmap case, no transpose; A->b already memcpy'd into C->b
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!Ab [p]) continue ;
GxB_FC64_t aij = Ax [p] ;
GxB_FC64_t z = (aij) ;
Cx [p] = GB_cisfinite (z) ;
}
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (cast (A')): transpose, typecast, and apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_tran__isfinite_bool_fc64)
(
GrB_Matrix C,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
smooth_periodic_function.h | // Copyright (c) 2013-2017 Anton Kozhevnikov, Thomas Schulthess
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without modification, are permitted provided that
// the following conditions are met:
//
// 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the
// following disclaimer.
// 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions
// and the following disclaimer in the documentation and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED
// WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
// PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
// ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
// OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
/** \file smooth_periodic_function.h
*
* \brief Contains declaration and implementation of sirius::Smooth_periodic_function and
* sirius::Smooth_periodic_function_gradient classes.
*/
#ifndef __SMOOTH_PERIODIC_FUNCTION_H__
#define __SMOOTH_PERIODIC_FUNCTION_H__
namespace sirius {
/// Representation of a smooth (Fourier-transformable) periodic function.
/** The class is designed to handle periodic functions such as density or potential, defined on a regular FFT grid.
* The following functionality is expected:
* - access to real-space values
* - access to plane-wave coefficients
* - distribution of plane-wave coefficients over entire communicator
* - Fourier transformation using FFT communicator
* - gather PW coefficients into global array
*/
template <typename T>
class Smooth_periodic_function
{
protected:
/// FFT driver.
FFT3D* fft_{nullptr};
/// Distribution of G-vectors.
Gvec_partition const* gvecp_{nullptr};
/// Function on the regular real-space grid.
mdarray<T, 1> f_rg_;
/// Local set of plane-wave expansion coefficients.
mdarray<double_complex, 1> f_pw_local_;
/// Storage of the PW coefficients for the FFT transformation.
mdarray<double_complex, 1> f_pw_fft_;
/// Gather plane-wave coefficients for the subsequent FFT call.
inline void gather_f_pw_fft()
{
gvecp_->gather_pw_fft(f_pw_local_.at<CPU>(), f_pw_fft_.at<CPU>());
}
public:
/// Default constructor.
Smooth_periodic_function()
{
}
Smooth_periodic_function(FFT3D& fft__, Gvec_partition const& gvecp__)
: fft_(&fft__)
, gvecp_(&gvecp__)
{
f_rg_ = mdarray<T, 1>(fft_->local_size(), memory_t::host, "Smooth_periodic_function.f_rg_");
f_pw_fft_ = mdarray<double_complex, 1>(gvecp_->gvec_count_fft(), memory_t::host, "Smooth_periodic_function.f_pw_fft_");
f_pw_local_ = mdarray<double_complex, 1>(gvecp_->gvec().count(), memory_t::host, "Smooth_periodic_function.f_pw_local_");
}
inline void zero()
{
f_rg_.zero();
}
inline T& f_rg(int ir__)
{
return f_rg_(ir__);
}
inline T const& f_rg(int ir__) const
{
return f_rg_(ir__);
}
inline mdarray<T, 1>& f_rg()
{
return f_rg_;
}
inline mdarray<T, 1> const& f_rg() const
{
return f_rg_;
}
inline double_complex& f_pw_local(int ig__)
{
return f_pw_local_(ig__);
}
inline double_complex const& f_pw_local(int ig__) const
{
return f_pw_local_(ig__);
}
inline double_complex& f_pw_fft(int ig__)
{
return f_pw_fft_(ig__);
}
/// Return plane-wave coefficient for G=0 component.
inline double_complex f_0() const
{
double_complex z;
if (gvecp_->gvec().comm().rank() == 0) {
z = f_pw_local_(0);
}
gvecp_->gvec().comm().bcast(&z, 1, 0);
return z;
}
FFT3D& fft()
{
assert(fft_ != nullptr);
return *fft_;
}
FFT3D const& fft() const
{
assert(fft_ != nullptr);
return *fft_;
}
Gvec const& gvec() const
{
assert(gvecp_ != nullptr);
return gvecp_->gvec();
}
Gvec_partition const& gvec_partition() const
{
return *gvecp_;
}
void fft_transform(int direction__)
{
PROFILE("sirius::Smooth_periodic_function::fft_transform");
assert(gvecp_ != nullptr);
switch (direction__) {
case 1: {
gather_f_pw_fft();
fft_->transform<1>(f_pw_fft_.at<CPU>());
fft_->output(f_rg_.template at<CPU>());
break;
}
case -1: {
fft_->input(f_rg_.template at<CPU>());
fft_->transform<-1>(f_pw_fft_.at<CPU>());
int count = gvecp_->gvec_fft_slab().counts[gvecp_->comm_ortho_fft().rank()];
int offset = gvecp_->gvec_fft_slab().offsets[gvecp_->comm_ortho_fft().rank()];
std::memcpy(f_pw_local_.at<CPU>(), f_pw_fft_.at<CPU>(offset), count * sizeof(double_complex));
break;
}
default: {
TERMINATE("wrong fft direction");
}
}
}
inline std::vector<double_complex> gather_f_pw()
{
PROFILE("sirius::Smooth_periodic_function::gather_f_pw");
std::vector<double_complex> fpw(gvecp_->gvec().num_gvec());
gvec().comm().allgather(&f_pw_local_[0], fpw.data(), gvec().offset(), gvec().count());
return std::move(fpw);
}
inline void scatter_f_pw(std::vector<double_complex> const& f_pw__)
{
std::copy(&f_pw__[gvecp_->gvec().offset()], &f_pw__[gvecp_->gvec().offset()] + gvecp_->gvec().count(), &f_pw_local_(0));
}
void add(Smooth_periodic_function<T> const& g__)
{
#pragma omp parallel for schedule(static)
for (int irloc = 0; irloc < this->fft_->local_size(); irloc++) {
this->f_rg_(irloc) += g__.f_rg(irloc);
}
}
inline T checksum_rg() const
{
T cs = this->f_rg_.checksum();
this->fft_->comm().allreduce(&cs, 1);
return cs;
}
/// Compute inner product <f|g>
T inner(Smooth_periodic_function<T> const& g__) const
{
PROFILE("sirius::Periodic_function::inner");
assert(this->fft_ == g__.fft_);
T result_rg{0};
#pragma omp parallel
{
T rt{0};
#pragma omp for schedule(static)
for (int irloc = 0; irloc < this->fft_->local_size(); irloc++) {
rt += type_wrapper<T>::bypass(std::conj(this->f_rg(irloc))) * g__.f_rg(irloc);
}
#pragma omp critical
result_rg += rt;
}
double omega = std::pow(twopi, 3) / std::abs(this->gvec().lattice_vectors().det());
result_rg *= (omega / this->fft_->size());
this->fft_->comm().allreduce(&result_rg, 1);
return result_rg;
}
inline uint64_t hash_f_pw() const
{
auto h = f_pw_local_.hash();
gvecp_->gvec().comm().bcast(&h, 1, 0);
for (int r = 1; r < gvecp_->gvec().comm().size(); r++) {
h = f_pw_local_.hash(h);
gvecp_->gvec().comm().bcast(&h, 1, r);
}
return h;
}
inline uint64_t hash_f_rg() const
{
auto h = f_rg_.hash();
fft_->comm().bcast(&h, 1, 0);
for (int r = 1; r < fft_->comm().size(); r++) {
h = f_rg_.hash(h);
fft_->comm().bcast(&h, 1, r);
}
return h;
}
};
/// Gradient of the smooth periodic function.
template<typename T>
class Smooth_periodic_function_gradient
{
private:
/// FFT driver.
FFT3D* fft_{nullptr};
/// Distribution of G-vectors.
Gvec_partition const* gvecp_{nullptr};
std::array<Smooth_periodic_function<T>, 3> grad_f_;
public:
Smooth_periodic_function_gradient()
{
}
Smooth_periodic_function_gradient(FFT3D& fft__, Gvec_partition const& gvecp__)
: fft_(&fft__)
, gvecp_(&gvecp__)
{
for (int x: {0, 1, 2}) {
grad_f_[x] = Smooth_periodic_function<T>(fft__, gvecp__);
}
}
Smooth_periodic_function<T>& operator[](const int idx__)
{
return grad_f_[idx__];
}
FFT3D& fft() const
{
assert(fft_ != nullptr);
return *fft_;
}
Gvec_partition const& gvec_partition() const
{
assert(gvecp_ != nullptr);
return *gvecp_;
}
};
/// Gradient of the function in the plane-wave domain.
inline Smooth_periodic_function_gradient<double> gradient(Smooth_periodic_function<double>& f__)
{
Smooth_periodic_function_gradient<double> g(f__.fft(), f__.gvec_partition());
#pragma omp parallel for schedule(static)
for (int igloc = 0; igloc < f__.gvec().count(); igloc++) {
int ig = f__.gvec().offset() + igloc;
auto G = f__.gvec().gvec_cart(ig);
for (int x: {0, 1, 2}) {
g[x].f_pw_local(igloc) = f__.f_pw_local(igloc) * double_complex(0, G[x]);
}
}
return std::move(g);
}
/// Laplacian of the function in the plane-wave domain.
inline Smooth_periodic_function<double> laplacian(Smooth_periodic_function<double>& f__)
{
Smooth_periodic_function<double> g(f__.fft(), f__.gvec_partition());
#pragma omp parallel for schedule(static)
for (int igloc = 0; igloc < f__.gvec().count(); igloc++) {
int ig = f__.gvec().offset() + igloc;
auto G = f__.gvec().gvec_cart(ig);
g.f_pw_local(igloc) = f__.f_pw_local(igloc) * double_complex(-std::pow(G.length(), 2), 0);
}
return std::move(g);
}
template <typename T>
inline Smooth_periodic_function<T> dot(Smooth_periodic_function_gradient<T>& grad_f__,
Smooth_periodic_function_gradient<T>& grad_g__)
{
assert(&grad_f__.fft() == &grad_g__.fft());
assert(&grad_f__.gvec_partition() == &grad_g__.gvec_partition());
Smooth_periodic_function<T> result(grad_f__.fft(), grad_f__.gvec_partition());
#pragma omp parallel for schedule(static)
for (int ir = 0; ir < grad_f__.fft().local_size(); ir++) {
double d{0};
for (int x: {0, 1, 2}) {
d += grad_f__[x].f_rg(ir) * grad_g__[x].f_rg(ir);
}
result.f_rg(ir) = d;
}
return std::move(result);
}
} // namespace sirius
#endif // __SMOOTH_PERIODIC_FUNCTION_H__
|
DRB013-nowait-orig-yes.c | /*
Copyright (c) 2017, Lawrence Livermore National Security, LLC.
Produced at the Lawrence Livermore National Laboratory
Written by Chunhua Liao, Pei-Hung Lin, Joshua Asplund,
Markus Schordan, and Ian Karlin
(email: liao6@llnl.gov, lin32@llnl.gov, asplund1@llnl.gov,
schordan1@llnl.gov, karlin1@llnl.gov)
LLNL-CODE-732144
All rights reserved.
This file is part of DataRaceBench. For details, see
https://github.com/LLNL/dataracebench. Please also see the LICENSE file
for our additional BSD notice.
Redistribution and use in source and binary forms, with
or without modification, are permitted provided that the following
conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the disclaimer below.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the disclaimer (as noted below)
in the documentation and/or other materials provided with the
distribution.
* Neither the name of the LLNS/LLNL nor the names of its contributors
may be used to endorse or promote products derived from this
software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL LAWRENCE LIVERMORE NATIONAL
SECURITY, LLC, THE U.S. DEPARTMENT OF ENERGY OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
This example is extracted from a paper:
Ma etc. Symbolic Analysis of Concurrency Errors in OpenMP Programs, ICPP 2013
Some threads may finish the for loop early and execute errors = dt[9]+1
while another thread may still be simultaneously executing
the for worksharing region by writing to d[9], causing data races.
Data race pair: a[i]@72:7 vs. a[9]@75:13.
*/
#include <stdio.h>
int main()
{
int i,error;
int len = 1000;
int a[len], b=5;
for (i=0; i<len; i++)
a[i]= i;
#pragma omp parallel shared(b, error)
{
#pragma omp for nowait
for(i = 0; i < len; i++)
a[i] = b + a[i]*5;
#pragma omp single
error = a[9] + 1;
}
printf ("error = %d\n", error);
return 0;
}
|
explicit_solver_strategy.h | //
// Authors:
// Miguel Angel Celigueta maceli@cimne.upc.edu
// Miquel Santasusana msantasusana@cimne.upc.edu
//
#if !defined(KRATOS_EXPLICIT_SOLVER_STRATEGY)
#define KRATOS_EXPLICIT_SOLVER_STRATEGY
// Project includes
#include "utilities/timer.h"
#include "custom_elements/Particle_Contact_Element.h"
#include "includes/variables.h"
#include "includes/deprecated_variables.h"
/* System includes */
#include <limits>
#include <iostream>
#include <iomanip>
#include <time.h>
/* External includes */
#ifdef _OPENMP
#include <omp.h>
#endif
#define CUSTOMTIMER 0 // ACTIVATES AND DISABLES ::TIMER:::::
#include "includes/define.h"
#include "utilities/openmp_utils.h"
#include "includes/model_part.h"
#include "solving_strategies/strategies/implicit_solving_strategy.h"
#include "solving_strategies/schemes/scheme.h"
#include "custom_strategies/schemes/dem_integration_scheme.h"
#include "custom_utilities/create_and_destroy.h"
#include "custom_utilities/dem_fem_utilities.h"
#include "custom_utilities/GeometryFunctions.h"
#include "custom_utilities/inlet.h"
#include "custom_elements/cluster3D.h"
#include "custom_elements/rigid_body_element.h"
////Cfeng
#include "custom_utilities/dem_fem_search.h"
#include "custom_utilities/discrete_particle_configure.h"
#include "custom_utilities/rigid_face_geometrical_object_configure.h"
#ifdef USING_CGAL
#include <CGAL/spatial_sort.h>
#endif
/* Timer defines */
#ifdef CUSTOMTIMER
#define KRATOS_TIMER_START(t) Timer::Start(t);
#define KRATOS_TIMER_STOP(t) Timer::Stop(t);
#else
#define KRATOS_TIMER_START(t)
#define KRATOS_TIMER_STOP(t)
#endif
namespace Kratos {
class ExplicitSolverSettings {
public:
KRATOS_CLASS_POINTER_DEFINITION(ExplicitSolverSettings);
ExplicitSolverSettings() {
}
~ExplicitSolverSettings() {
}
ModelPart* r_model_part;
ModelPart* contact_model_part;
ModelPart* fem_model_part;
ModelPart* cluster_model_part;
ModelPart* inlet_model_part;
};
class KRATOS_API(DEM_APPLICATION) ExplicitSolverStrategy {
public:
typedef ModelPart::NodesContainerType NodesArrayType;
typedef ModelPart::ElementsContainerType ElementsArrayType;
typedef ElementsArrayType::iterator ElementsIterator;
typedef ModelPart::ConditionsContainerType ConditionsArrayType;
typedef ModelPart::NodesContainerType::ContainerType NodesContainerType;
typedef ModelPart::ElementsContainerType::ContainerType ElementsContainerType;
typedef ModelPart::ConditionsContainerType::ContainerType ConditionsContainerType;
typedef SpatialSearch::ResultElementsContainerType ResultElementsContainerType;
typedef SpatialSearch::VectorResultElementsContainerType VectorResultElementsContainerType;
typedef SpatialSearch::RadiusArrayType RadiusArrayType;
typedef SpatialSearch::DistanceType DistanceType;
typedef SpatialSearch::VectorDistanceType VectorDistanceType;
typedef SpatialSearch::ResultConditionsContainerType ResultConditionsContainerType;
typedef SpatialSearch::VectorResultConditionsContainerType VectorResultConditionsContainerType;
typedef PointerVectorSet<Properties, IndexedObject> PropertiesContainerType;
typedef PropertiesContainerType::iterator PropertiesIterator;
typedef DiscreteParticleConfigure<3> ElementConfigureType;
typedef RigidFaceGeometricalObjectConfigure<3> RigidFaceGeometricalConfigureType;
typedef Variable<double> ComponentOf3ComponentsVariableType;
/// Pointer definition of ExplicitSolverStrategy
KRATOS_CLASS_POINTER_DEFINITION(ExplicitSolverStrategy);
ExplicitSolverStrategy() {
}
ExplicitSolverStrategy(ExplicitSolverSettings& settings,
const double max_delta_time,
const int n_step_search,
const double safety_factor,
const int delta_option,
ParticleCreatorDestructor::Pointer p_creator_destructor,
DEM_FEM_Search::Pointer p_dem_fem_search,
SpatialSearch::Pointer pSpSearch,
Parameters strategy_parameters) {
mParameters = strategy_parameters;
mDeltaOption = delta_option;
mpParticleCreatorDestructor = p_creator_destructor;
mpDemFemSearch = p_dem_fem_search;
mpSpSearch = pSpSearch;
//Also checks old flag name for backward compatibility issues.
if(mParameters["do_search_dem_neighbours"].GetBool()) {
mDoSearchNeighbourElements = true;
} else mDoSearchNeighbourElements = false;
p_creator_destructor->SetDoSearchNeighbourElements(mDoSearchNeighbourElements);
if(mParameters["do_search_fem_neighbours"].GetBool()) mDoSearchNeighbourFEMElements = true;
else mDoSearchNeighbourFEMElements = false;
mMaxTimeStep = max_delta_time;
mNStepSearch = n_step_search;
mSafetyFactor = safety_factor;
mpDem_model_part = &(*(settings.r_model_part));
KRATOS_ERROR_IF(mpDem_model_part == NULL) << "Undefined settings.r_model_part in ExplicitSolverStrategy constructor" << std::endl;
mpContact_model_part = &(*(settings.contact_model_part));
KRATOS_ERROR_IF(mpContact_model_part == NULL) << "Undefined settings.contact_model_part in ExplicitSolverStrategy constructor" << std::endl;
mpFem_model_part = &(*(settings.fem_model_part));
KRATOS_ERROR_IF(mpFem_model_part == NULL) << "Undefined settings.fem_model_part in ExplicitSolverStrategy constructor" << std::endl;
mpCluster_model_part = &(*(settings.cluster_model_part));
KRATOS_ERROR_IF(mpCluster_model_part == NULL) << "Undefined settings.cluster_model_part in ExplicitSolverStrategy constructor" << std::endl;
mpInlet_model_part = &(*(settings.inlet_model_part));
KRATOS_ERROR_IF(mpInlet_model_part == NULL) << "Undefined settings.inlet_model_part in ExplicitSolverStrategy constructor" << std::endl;
if(mParameters["RemoveBallsInitiallyTouchingWalls"].GetBool()) mRemoveBallsInitiallyTouchingWallsOption = true;
else mRemoveBallsInitiallyTouchingWallsOption = false;
}
/// Destructor.
virtual ~ExplicitSolverStrategy() {
//Timer::SetOuputFile("TimesPartialRelease");
//Timer::PrintTimingInformation();
}
struct LessX {
bool operator()(const SphericParticle* p, const SphericParticle* q) const {return p->GetGeometry()[0].Coordinates()[0] < q->GetGeometry()[0].Coordinates()[0];}
};
struct LessY {
bool operator()(const SphericParticle* p, const SphericParticle* q) const {return p->GetGeometry()[0].Coordinates()[1] < q->GetGeometry()[0].Coordinates()[1];}
};
struct LessZ {
bool operator()(const SphericParticle* p, const SphericParticle* q) const {return p->GetGeometry()[0].Coordinates()[2] < q->GetGeometry()[0].Coordinates()[2];}
};
struct SpatialSortingTraits {
typedef SphericParticle* Point_2;
typedef LessX Less_x_2;
typedef LessY Less_y_2;
typedef LessZ Less_z_2;
Less_x_2 less_x_2_object() const {return Less_x_2();}
Less_y_2 less_y_2_object() const {return Less_y_2();}
Less_z_2 less_z_2_object() const { return Less_z_2();}
};
#ifdef USING_CGAL
void ReorderParticles() {
SpatialSortingTraits sst;
CGAL::spatial_sort(mListOfSphericParticles.begin(), mListOfSphericParticles.end(), sst);
}
#endif
template <class T>
void RebuildListOfSphericParticles(ElementsArrayType& pElements, std::vector<T*>& rCustomListOfParticles){
KRATOS_TRY
rCustomListOfParticles.resize(pElements.size());
#pragma omp parallel for
for (int k = 0; k < (int)pElements.size(); k++){
ElementsArrayType::iterator particle_pointer_it = pElements.ptr_begin() + k;
T* spheric_particle = dynamic_cast<T*>(&(*particle_pointer_it));
rCustomListOfParticles[k] = spheric_particle;
}
return;
KRATOS_CATCH("")
}
void RebuildListOfDiscontinuumSphericParticles() {
RebuildListOfSphericParticles<SphericParticle>(GetModelPart().GetCommunicator().LocalMesh().Elements(), mListOfSphericParticles);
}
void RebuildPropertiesProxyPointers(std::vector<SphericParticle*>& rCustomListOfSphericParticles);
void SendProcessInfoToClustersModelPart();
void UpdateMaxIdOfCreatorDestructor();
void RepairPointersToNormalProperties(std::vector<SphericParticle*>& rCustomListOfSphericParticles);
virtual void Initialize();
virtual void AttachSpheresToStickyWalls();
virtual void DisplayThreadInfo();
virtual void CalculateMaxTimeStep();
double CalculateMaxInletTimeStep();
virtual void InitializeClusters();
virtual void GetClustersForce();
virtual void GetRigidBodyElementsForce();
virtual double SolveSolutionStep();
void SearchDEMOperations(ModelPart& r_model_part, bool has_mpi = true);
void SearchFEMOperations(ModelPart& r_model_part, bool has_mpi = true) ;
virtual void ForceOperations(ModelPart& r_model_part);
void InitialTimeStepCalculation(); //TODO: remove this one
void GetForce();
void FastGetForce();
virtual void PerformTimeIntegrationOfMotion(int StepFlag = 0);
void InitializeSolutionStep();
virtual void BoundingBoxUtility(bool is_time_to_mark_and_remove = true);
virtual void FinalizeSolutionStep();
void InitializeElements();
void InitializeDEMElements();
void InitializeFEMElements();
//void InitializeRigidBodyElements();
void InitializeFEMWallsAsRigidBodyElements(ModelPart::SubModelPartsContainerType::iterator& sub_model_part);
void MarkToDeleteAllSpheresInitiallyIndentedWithFEM(ModelPart& rSpheresModelPart);
void ComputeNodalArea();
void ComputeNormalPressureVectorField();
virtual void CalculateConditionsRHSAndAdd();
void ClearFEMForces();
void CalculateNodalPressuresAndStressesOnWalls();
void SetFlagAndVariableToNodes(const Kratos::Flags& r_flag_name, ComponentOf3ComponentsVariableType& r_variable_to_set, const double value, NodesArrayType& r_nodes_array);
void SetVariableToNodes(ComponentOf3ComponentsVariableType& r_variable_to_set, const double value, NodesArrayType& r_nodes_array);
void ResetPrescribedMotionFlagsRespectingImposedDofs();
void ApplyPrescribedBoundaryConditions();
void ApplyInitialConditions();
void SetSearchRadiiOnAllParticles(ModelPart& r_model_part, const double added_search_distance = 0.0, const double amplification = 1.0);
void SetNormalRadiiOnAllParticles(ModelPart& r_model_part);
void SetSearchRadiiWithFemOnAllParticles(ModelPart& r_model_part, const double added_search_distance = 0.0, const double amplification = 1.0);
virtual void SearchNeighbours();
virtual void ComputeNewNeighboursHistoricalData();
virtual void CreateContactElements();
void InitializeContactElements();
// void ContactInitializeSolutionStep();
void PrepareContactElementsForPrinting();
virtual void ComputeNewRigidFaceNeighboursHistoricalData();
virtual void SearchRigidFaceNeighbours();
void CheckHierarchyWithCurrentNeighbours();
/* This should work only with one iteration, but it with mpi does not */
void CalculateInitialMaxIndentations(const ProcessInfo& r_process_info);
void PrepareContactModelPart(ModelPart& r_model_part, ModelPart& mcontacts_model_part);
void PrepareElementsForPrinting();
void SynchronizeHistoricalVariables(ModelPart& r_model_part);
void SynchronizeRHS(ModelPart& r_model_part);
void CleanEnergies();
void Check_MPI(bool& has_mpi);
ModelPart& GetModelPart() { return (*mpDem_model_part);}
ModelPart& GetFemModelPart() { return (*mpFem_model_part);}
ModelPart& GetContactModelPart() { return (*mpContact_model_part);}
ModelPart& GetClusterModelPart() { return (*mpCluster_model_part);}
ModelPart& GetInletModelPart() { return (*mpInlet_model_part);}
ModelPart& GetRigidBodyModelPart() { return (*mpRigidBody_model_part);}
VectorResultElementsContainerType& GetResults() { return (mResults);}
VectorDistanceType& GetResultsDistances() { return (mResultsDistances);}
RadiusArrayType& GetArrayOfAmplifiedRadii() { return (mArrayOfAmplifiedRadii);}
int& GetNStepSearch() { return (mNStepSearch);}
int& GetSearchControl() { return mSearchControl;}
int& GetNumberOfThreads() { return (mNumberOfThreads);}
double& GetMaxTimeStep() { return (mMaxTimeStep);}
double& GetSafetyFactor() { return (mSafetyFactor);}
int& GetDeltaOption() { return (mDeltaOption);}
ParticleCreatorDestructor::Pointer& GetParticleCreatorDestructor() { return (mpParticleCreatorDestructor);}
SpatialSearch::Pointer& GetSpSearch() { return (mpSpSearch);}
VectorResultConditionsContainerType& GetRigidFaceResults() { return (mRigidFaceResults);}
VectorDistanceType& GetRigidFaceResultsDistances() { return (mRigidFaceResultsDistances);}
DEM_FEM_Search::Pointer& GetDemFemSearch() { return (mpDemFemSearch);}
virtual ElementsArrayType& GetElements(ModelPart& r_model_part) { return r_model_part.GetCommunicator().LocalMesh().Elements();}
virtual ElementsArrayType& GetAllElements(ModelPart& r_model_part) {
return r_model_part.Elements();
}
protected:
Parameters mParameters;
bool mRemoveBallsInitiallyTouchingWallsOption;
VectorResultElementsContainerType mResults;
VectorDistanceType mResultsDistances;
RadiusArrayType mArrayOfAmplifiedRadii;
int mNStepSearch;
int mSearchControl;
int mNumberOfThreads;
double mMaxTimeStep;
double mSafetyFactor;
int mDeltaOption;
ParticleCreatorDestructor::Pointer mpParticleCreatorDestructor;
DEM_FEM_Search::Pointer mpDemFemSearch;
SpatialSearch::Pointer mpSpSearch;
bool mDoSearchNeighbourElements;
bool mDoSearchNeighbourFEMElements;
VectorResultConditionsContainerType mRigidFaceResults;
VectorDistanceType mRigidFaceResultsDistances;
ModelPart *mpFem_model_part;
ModelPart *mpDem_model_part;
ModelPart *mpInlet_model_part;
ModelPart *mpContact_model_part;
ModelPart *mpCluster_model_part;
ModelPart *mpRigidBody_model_part;
std::vector<SphericParticle*> mListOfSphericParticles;
std::vector<SphericParticle*> mListOfGhostSphericParticles;
}; // Class ExplicitSolverStrategy
} // namespace Kratos.
#endif // KRATOS_EXPLICIT_SOLVER_STRATEGY defined
|
speed_omp.c | #include <sys/time.h>
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include <err.h>
#include <omp.h>
#include <assert.h>
#include "feslite.h"
#include "cycleclock.h"
/* Measure raw multi-threaded-thread speed of all kernels in the library */
int n = 32;
int T = -1;
void bench_kernel(int kernel)
{
const char *name = feslite_kernel_name(kernel);
if (!feslite_kernel_is_available(kernel)) {
printf("[%s] is not available on this machine\n", name);
return;
}
int m = feslite_kernel_batch_size(kernel);
printf("kernel %d [%s] : %d lane(s)...\n", kernel, name, m);
fflush(stdout);
srand48(1337);
/* initalize m random related systems */
u32 Fq[496];
for (int i = 0; i < 496; i++)
Fq[i] = lrand48();
double start_wt = omp_get_wtime();
u64 start = Now();
int count = 256;
#pragma omp parallel
{
assert(omp_get_num_threads() == T);
u32 Fl[34 * m];
for (int i = 0; i < 34 * m; i++)
Fl[i] = 0;
for (int i = 0; i < (n+1) * m; i++)
Fl[i] = lrand48();
u32 buffer[count * m];
int size[m];
feslite_kernel_solve(kernel, n, m, Fq, Fl, count, buffer, size);
}
u64 stop = Now();
double stop_wt = omp_get_wtime();
double freq = 2.6; // in GHz
double cycles = stop - start;
double seconds = stop_wt - start_wt;
double rate = cycles / m / (1ll << n) / 2;
printf("\t---> %.2f s\n", stop_wt - start_wt);
printf("\t---> %.2f cycles/candidate || %.2f candidate/cycle/thread\n", rate, 1/rate);
printf("\t---> %.2f cycles/3-steps/core (at %.1fGHz)\n", (seconds * freq * 1e9 / 2) / 1431655765.3, freq);
printf("\t---> 2^%.2f candidate/s on all threads\n", log2(m) + n + log2(T) - log2(seconds));
}
int main(int argc, char **argv)
{
T = omp_get_max_threads();
printf("INFO : running with %d threads\n", T);
int nkernels = feslite_num_kernels();
if (argc > 1) {
int kernel = feslite_kernel_find_by_name(argv[1]);
if (kernel < 0)
errx(1, "bad kernel name (use ./list)");
bench_kernel(kernel);
} else {
for (int kernel = 0; kernel < nkernels; kernel++)
bench_kernel(kernel);
}
return EXIT_SUCCESS;
} |
caesar_decode.c | #include <unistd.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <strings.h>
#include <time.h>
#include <omp.h>
#include "mpi.h"
#define CHARS 100000
// caesar decode paralelizado
int main(int argc, char * argv[]) {
// variables para MPI
int myid, numprocs;
char hostname[MPI_MAX_PROCESSOR_NAME];
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
// variables
int data = CHARS/numprocs;
char local[data];
double start, stop;
int maxdata = data;
// aqui es donde cambiamos la llave para encriptar
int llave = 3;
char *path_crypt;
FILE *crypt_file;
char mensaje_enc[CHARS], temp1;
int x;
if (myid == 0) {
// path para el archivo a leer
path_crypt = "texto_encriptado.txt";
crypt_file = fopen(path_crypt, "r");
//mensaje por si no encuentra le archivo
if (crypt_file == NULL) {
printf("ERROR, no se pudo abrir el archivo txt\n");
exit(-1);
}
fread(mensaje_enc, CHARS, 1, crypt_file);
}
//dividimos el mensaje
MPI_Scatter(mensaje_enc, data, MPI_CHAR, local, data, MPI_CHAR, 0, MPI_COMM_WORLD);
start = omp_get_wtime();
// paralelizamos el algoritmo para desencriptar el mensaje
#pragma omp parallel for
for(x=0; x < maxdata; x++) {
temp1 = local[x];
if(temp1 >= 'a' && temp1 <= 'z'){
temp1 = temp1 - llave;
if(temp1 > 'z'){
temp1 = temp1 + 'a' - 'z' - 1;
}
if(temp1 < 'a'){
temp1 = temp1 + 'z' - 'a' + 1;
}
local[x] = temp1;
} else if (temp1 >= 'A' && temp1 <= 'Z'){
temp1 = temp1 - llave;
if(temp1 > 'Z'){
temp1 = temp1 + 'A' - 'Z' - 1;
}
if(temp1 < 'A'){
temp1 = temp1 + 'Z' - 'A' + 1;
}
local[x] = temp1;
}
}
// recolectamos el mensaje
MPI_Gather(local, data, MPI_CHAR, mensaje_enc, data, MPI_CHAR, 0, MPI_COMM_WORLD);
if (myid == 0){
printf("Mensaje desencriptado: %s\n", mensaje_enc);
fclose(crypt_file);
stop = omp_get_wtime();
printf("Tiempo de ejecución de la sección paralela = %f \n", stop-start);
}
MPI_Finalize();
return 0;
} |
NetCDFMesh.h | /**
* @file
* This file is part of PUMGen
*
* For conditions of distribution and use, please see the copyright
* notice in the file 'COPYING' at the root directory of this package
* and the copyright notice at https://github.com/SeisSol/PUMGen
*
* @copyright 2017 Technical University of Munich
* @author Sebastian Rettenberger <sebastian.rettenberger@tum.de>
*/
#ifndef NETCDF_MESH_H
#define NETCDF_MESH_H
#include <mpi.h>
#include <netcdf.h>
#include <netcdf_par.h>
#include <PCU.h>
#include <apfConvert.h>
#include <apfMDS.h>
#include <apfMesh2.h>
#include <gmi_null.h>
#include "utils/logger.h"
#include "MeshInput.h"
#include "NetCDFPartition.h"
#include "ParallelVertexFilter.h"
/**
* Read PUMGen generated mesh files
*/
class NetCDFMesh : public MeshInput {
public:
NetCDFMesh(const char* meshFile, MPI_Comm comm = MPI_COMM_WORLD) {
int rank = 0;
int nProcs = 1;
MPI_Comm_rank(comm, &rank);
MPI_Comm_size(comm, &nProcs);
gmi_register_null();
gmi_model* model = gmi_load(".null");
m_mesh = apf::makeEmptyMdsMesh(model, 3, false);
int ncFile;
checkNcError(nc_open_par(meshFile, NC_MPIIO, comm, MPI_INFO_NULL, &ncFile));
// Get number of partitions
int ncDimPart;
checkNcError(nc_inq_dimid(ncFile, "partitions", &ncDimPart));
size_t nPartitions;
checkNcError(nc_inq_dimlen(ncFile, ncDimPart, &nPartitions));
// Local partitions
unsigned int nMaxLocalPart = (nPartitions + nProcs - 1) / nProcs;
unsigned int nLocalPart = nMaxLocalPart;
if (nPartitions < (rank + 1) * nMaxLocalPart)
nLocalPart = std::max(0, static_cast<int>(nPartitions - rank * nMaxLocalPart));
MPI_Comm commIO;
MPI_Comm_split(MPI_COMM_WORLD, (nLocalPart > 0 ? 0 : MPI_UNDEFINED), 0, &commIO);
// Reopen netCDF file with correct communicator
checkNcError(nc_close(ncFile));
if (nLocalPart > 0)
checkNcError(nc_open_par(meshFile, NC_MPIIO, commIO, MPI_INFO_NULL, &ncFile));
PCU_Switch_Comm(commIO);
unsigned int nElements = 0;
unsigned int nVertices = 0;
int* elements = 0L;
double* vertices = 0L;
int* boundaries = 0L;
int* groups = 0L;
if (nLocalPart > 0) {
// Create netCDF variables
int ncVarElemSize;
checkNcError(nc_inq_varid(ncFile, "element_size", &ncVarElemSize));
collectiveAccess(ncFile, ncVarElemSize);
int ncVarElemVertices;
checkNcError(nc_inq_varid(ncFile, "element_vertices", &ncVarElemVertices));
collectiveAccess(ncFile, ncVarElemVertices);
int ncVarElemBoundaries;
checkNcError(nc_inq_varid(ncFile, "element_boundaries", &ncVarElemBoundaries));
collectiveAccess(ncFile, ncVarElemBoundaries);
int ncVarElemGroup;
bool useGroups = true;
if (nc_inq_varid(ncFile, "element_group", &ncVarElemGroup) != NC_NOERR) {
useGroups = false;
logWarning() << "No group found, using group 0 for all elements";
} else {
collectiveAccess(ncFile, ncVarElemGroup);
}
int ncVarVrtxSize;
checkNcError(nc_inq_varid(ncFile, "vertex_size", &ncVarVrtxSize));
collectiveAccess(ncFile, ncVarVrtxSize);
int ncVarVrtxCoords;
checkNcError(nc_inq_varid(ncFile, "vertex_coordinates", &ncVarVrtxCoords));
collectiveAccess(ncFile, ncVarVrtxCoords);
Partition* partitions = new Partition[nLocalPart];
// Read elements
logInfo(rank) << "Reading netCDF file";
for (unsigned int i = 0; i < nMaxLocalPart; i++) {
unsigned int j = i % nLocalPart;
size_t start[3] = {j + rank * nMaxLocalPart, 0, 0};
// Element size
unsigned int size;
checkNcError(nc_get_var1_uint(ncFile, ncVarElemSize, start, &size));
partitions[j].setElemSize(size);
size_t count[3] = {1, size, 4};
// Elements
checkNcError(
nc_get_vara_int(ncFile, ncVarElemVertices, start, count, partitions[j].elements()));
// Boundaries and group
checkNcError(
nc_get_vara_int(ncFile, ncVarElemBoundaries, start, count, partitions[j].boundaries()));
if (useGroups)
checkNcError(
nc_get_vara_int(ncFile, ncVarElemGroup, start, count, partitions[j].groups()));
// Vertex size
checkNcError(nc_get_var1_uint(ncFile, ncVarVrtxSize, start, &size));
partitions[j].setVrtxSize(size);
// Vertices
count[1] = size;
count[2] = 3;
checkNcError(
nc_get_vara_double(ncFile, ncVarVrtxCoords, start, count, partitions[j].vertices()));
}
checkNcError(nc_close(ncFile));
for (unsigned int i = 0; i < nLocalPart; i++) {
nElements += partitions[i].nElements();
nVertices += partitions[i].nVertices();
}
// Copy to the buffer
unsigned int* elementsLocal = new unsigned int[nElements * 4];
elements = new int[nElements * 4];
vertices = new double[nVertices * 3];
boundaries = new int[nElements * 4];
groups = new int[nElements];
unsigned int elementOffset = 0;
unsigned int vertexOffset = 0;
for (unsigned int i = 0; i < nLocalPart; i++) {
#ifdef _OPENMP
#pragma omp parallel
#endif
for (unsigned int j = 0; j < partitions[i].nElements() * 4; j++)
elementsLocal[elementOffset * 4 + j] = partitions[i].elements()[j] + vertexOffset;
memcpy(&vertices[vertexOffset * 3], partitions[i].vertices(),
partitions[i].nVertices() * 3 * sizeof(double));
partitions[i].convertBoundary();
memcpy(&boundaries[elementOffset * 4], partitions[i].boundaries(),
partitions[i].nElements() * 4 * sizeof(int));
memcpy(&groups[elementOffset], partitions[i].groups(),
partitions[i].nElements() * sizeof(int));
elementOffset += partitions[i].nElements();
vertexOffset += partitions[i].nVertices();
}
logInfo(rank) << "Running vertex filter";
ParallelVertexFilter filter(commIO);
filter.filter(nVertices, vertices);
// Create filtered vertex list
delete[] vertices;
nVertices = filter.numLocalVertices();
vertices = new double[nVertices * 3];
memcpy(vertices, filter.localVertices(), nVertices * 3 * sizeof(double));
logInfo(rank) << "Converting local to global vertex identifier";
#ifdef _OPENMP
#pragma omp parallel
#endif
for (unsigned int i = 0; i < nElements * 4; i++)
elements[i] = filter.globalIds()[elementsLocal[i]];
delete[] partitions;
}
logInfo(rank) << "Constructing the mesh";
apf::GlobalToVert vertMap;
apf::construct(m_mesh, elements, nElements, apf::Mesh::TET, vertMap);
delete[] elements;
apf::alignMdsRemotes(m_mesh);
apf::deriveMdsModel(m_mesh);
logInfo(rank) << "Set coordinates in APF";
apf::setCoords(m_mesh, vertices, nVertices, vertMap);
delete[] vertices;
// Set boundaries
apf::MeshTag* boundaryTag = m_mesh->createIntTag("boundary condition", 1);
apf::MeshIterator* it = m_mesh->begin(3);
unsigned int i = 0;
while (apf::MeshEntity* element = m_mesh->iterate(it)) {
apf::Adjacent adjacent;
m_mesh->getAdjacent(element, 2, adjacent);
for (unsigned int j = 0; j < 4; j++) {
if (!boundaries[i * 4 + j])
continue;
m_mesh->setIntTag(adjacent[j], boundaryTag, &boundaries[i * 4 + j]);
}
i++;
}
m_mesh->end(it);
delete[] boundaries;
// Set groups
apf::MeshTag* groupTag = m_mesh->createIntTag("group", 1);
it = m_mesh->begin(3);
i = 0;
while (apf::MeshEntity* element = m_mesh->iterate(it)) {
m_mesh->setIntTag(element, groupTag, &groups[i]);
i++;
}
m_mesh->end(it);
delete[] groups;
PCU_Switch_Comm(MPI_COMM_WORLD);
}
private:
/**
* Switch to collective access for a netCDf variable
*/
static void collectiveAccess(int ncFile, int ncVar) {
checkNcError(nc_var_par_access(ncFile, ncVar, NC_COLLECTIVE));
}
static void checkNcError(int error) {
if (error != NC_NOERR)
logError() << "Error while reading netCDF file:" << nc_strerror(error);
}
};
#endif // NETCDF_MESH_H
|
GB_is_diagonal.c | //------------------------------------------------------------------------------
// GB_is_diagonal: check if A is a diagonal matrix
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// Returns true if A is a square diagonal matrix, with all diagonal entries
// present. All pending tuples are ignored. Zombies are treated as entries.
#include "GB_mxm.h"
#include "GB_atomics.h"
bool GB_is_diagonal // true if A is diagonal
(
const GrB_Matrix A, // input matrix to examine
GB_Context Context
)
{
//--------------------------------------------------------------------------
// check inputs
//--------------------------------------------------------------------------
ASSERT (A != NULL) ;
ASSERT_MATRIX_OK (A, "A check diag", GB0) ;
ASSERT (!GB_ZOMBIES (A)) ;
ASSERT (GB_JUMBLED_OK (A)) ;
ASSERT (!GB_PENDING (A)) ;
//--------------------------------------------------------------------------
// trivial cases
//--------------------------------------------------------------------------
int64_t n = GB_NROWS (A) ;
int64_t ncols = GB_NCOLS (A) ;
if (n != ncols)
{
// A is rectangular
return (false) ;
}
if (GB_IS_BITMAP (A))
{
// never treat bitmaps as diagonal
return (false) ;
}
if (GB_IS_FULL (A))
{
// A is full, and is diagonal only if 1-by-1, but always return
// false so that GB_AxB_rowscale and GB_AxB_colscale are not used
// by GB_reduce_to_vector.
return (false) ;
}
int64_t anz = GB_nnz (A) ;
int64_t nvec = A->nvec ;
if (n != anz || n != nvec)
{
// A must have exactly n entries in n vectors. A can be sparse or
// hypersparse. If hypersparse, all vectors must be present, so
// Ap has size n+1 whether sparse or hypersparse.
return (false) ;
}
//--------------------------------------------------------------------------
// determine the number of threads to use
//--------------------------------------------------------------------------
// Break the work into lots of tasks so the early-exit can be exploited.
GB_GET_NTHREADS_MAX (nthreads_max, chunk, Context) ;
int nthreads = GB_nthreads (n, chunk, nthreads_max) ;
int ntasks = (nthreads == 1) ? 1 : (256 * nthreads) ;
ntasks = GB_IMIN (ntasks, n) ;
ntasks = GB_IMAX (ntasks, 1) ;
//--------------------------------------------------------------------------
// examine each vector of A
//--------------------------------------------------------------------------
const int64_t *restrict Ap = A->p ;
const int64_t *restrict Ai = A->i ;
int diagonal = true ;
int tid ;
#pragma omp parallel for num_threads(nthreads) schedule(dynamic,1)
for (tid = 0 ; tid < ntasks ; tid++)
{
//----------------------------------------------------------------------
// check for early exit
//----------------------------------------------------------------------
int diag = true ;
{
GB_ATOMIC_READ
diag = diagonal ;
}
if (!diag) continue ;
//----------------------------------------------------------------------
// check if vectors jstart:jend-1 are diagonal
//----------------------------------------------------------------------
int64_t jstart, jend ;
GB_PARTITION (jstart, jend, n, tid, ntasks) ;
for (int64_t j = jstart ; diag && j < jend ; j++)
{
int64_t p = Ap [j] ;
int64_t ajnz = Ap [j+1] - p ;
if (ajnz != 1)
{
// A(:,j) must have exactly one entry
diag = false ;
}
int64_t i = Ai [p] ;
if (i != j)
{
// the single entry must be A(i,i)
diag = false ;
}
}
//----------------------------------------------------------------------
// early exit: tell all other tasks to halt
//----------------------------------------------------------------------
if (!diag)
{
GB_ATOMIC_WRITE
diagonal = false ;
}
}
//--------------------------------------------------------------------------
// return result
//--------------------------------------------------------------------------
return ((bool) diagonal) ;
}
|
target.c | #include <stdio.h>
#include <stdlib.h>
int main (int argc, char** argv) {
int i;
#pragma omp target
#pragma omp parallel for
for (i = 0; i < 2; i++) {
printf("Test 1.\n");
}
#pragma omp target
#pragma omp parallel for
for (i = 0; i < 6; i++) {
printf("Test 2.\n");
}
return 0;
}
|
DRB046-doall2-orig-no.c | /*
Copyright (c) 2017, Lawrence Livermore National Security, LLC.
Produced at the Lawrence Livermore National Laboratory
Written by Chunhua Liao, Pei-Hung Lin, Joshua Asplund,
Markus Schordan, and Ian Karlin
(email: liao6@llnl.gov, lin32@llnl.gov, asplund1@llnl.gov,
schordan1@llnl.gov, karlin1@llnl.gov)
LLNL-CODE-732144
All rights reserved.
This file is part of DataRaceBench. For details, see
https://github.com/LLNL/dataracebench. Please also see the LICENSE file
for our additional BSD notice.
Redistribution and use in source and binary forms, with
or without modification, are permitted provided that the following
conditions are met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the disclaimer below.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the disclaimer (as noted below)
in the documentation and/or other materials provided with the
distribution.
* Neither the name of the LLNS/LLNL nor the names of its contributors
may be used to endorse or promote products derived from this
software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL LAWRENCE LIVERMORE NATIONAL
SECURITY, LLC, THE U.S. DEPARTMENT OF ENERGY OR CONTRIBUTORS BE
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
THE POSSIBILITY OF SUCH DAMAGE.
*/
/*
Two-dimensional array computation:
Only one loop is associated with the omp for construct.
The inner loop's loop iteration variable needs an explicit private() clause,
otherwise it will be shared by default.
*/
#include <stdio.h>
#include <stdlib.h>
int a[100][100];
int main()
{
int i,j;
#pragma omp parallel for private(i ,j )
for (i=0;i<100;i++)
#pragma omp parallel for private(j )
for (j=0;j<100;j++)
a[i][j]= i * 200 + j;
#pragma omp parallel for private(i ,j )
for (i=0;i<100;i++)
#pragma omp parallel for private(j )
for (j=0;j<100;j++)
a[i][j]=a[i][j]+1;
for (i=0;i<100;i++)
for (j=0;j<100;j++)
printf("%d", a[i][j]);
return 0;
}
|
parallel_allocate_predefined_modifier.c | #include <stdio.h>
#ifdef _OPENMP
#include <omp.h>
#endif
int main()
{
int a, b, c;
#pragma omp parallel allocate (omp_cgroup_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_default_mem_alloc:a, b) allocate (a)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_const_mem_alloc:a, b) allocate (c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_large_cap_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_high_bw_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_low_lat_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_pteam_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
#pragma omp parallel allocate (omp_thread_mem_alloc:a, b) allocate (a, c)
{
printf("This is for testing parser and AST construction, which could be only syntax correct.\n");
}
return 0;
}
|
libmsr_write_test.c | /**
* @author Asim YarKhan (updated)
* @author Vince Weaver (original version)
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "papi.h"
#include "msr_core.h"
#include "msr_rapl.h"
#define MAX_EVENTS 128
char events[MAX_EVENTS][BUFSIZ];
char filenames[MAX_EVENTS][BUFSIZ];
int ompcpuloadprimes( int limit )
{
int num, primes=0;
#pragma omp parallel for schedule(dynamic) reduction(+ : primes)
for (num = 1; num <= limit; num++) {
int i = 2;
while(i <= num) {
if(num % i == 0)
break;
i++;
}
if(i == num)
primes++;
}
return primes;
}
int main (int argc, char **argv)
{
int retval,cid,rapl_cid=-1,numcmp;
int EventSet = PAPI_NULL;
long long values[MAX_EVENTS];
int i,code,enum_retval;
const PAPI_component_info_t *cmpinfo = NULL;
long long start_time,write_start_time,write_end_time,read_start_time,read_end_time;
char event_name[BUFSIZ];
union { long long ll; double dbl; } event_value_union;
static int num_events=0;
FILE *fileout;
/* PAPI Initialization */
retval = PAPI_library_init( PAPI_VER_CURRENT );
if ( retval != PAPI_VER_CURRENT ) {
fprintf(stderr,"PAPI_library_init failed\n");
exit(1);
}
/* Find the libmsr component */
numcmp = PAPI_num_components();
for(cid=0; cid<numcmp; cid++) {
if ( (cmpinfo = PAPI_get_component_info(cid)) == NULL) {
fprintf(stderr,"PAPI_get_component_info failed\n");
exit(1);
}
if (strstr(cmpinfo->name,"libmsr")) {
rapl_cid=cid;
printf("Found libmsr component at cid %d\n", rapl_cid);
if (cmpinfo->disabled) {
fprintf(stderr,"No libmsr events found: %s\n", cmpinfo->disabled_reason);
exit(1);
}
break;
}
}
/* Component not found */
if (cid==numcmp) {
fprintf(stderr,"No libmsr component found\n");
exit(1);
}
/* Find events in the component */
code = PAPI_NATIVE_MASK;
enum_retval = PAPI_enum_cmp_event( &code, PAPI_ENUM_FIRST, cid );
while ( enum_retval == PAPI_OK ) {
retval = PAPI_event_code_to_name( code, event_name );
if ( retval != PAPI_OK ) {
printf("Error translating %#x\n",code);
exit(1);
}
printf("Found: %s\n",event_name);
strncpy(events[num_events],event_name,BUFSIZ);
sprintf(filenames[num_events],"results.%s",event_name);
num_events++;
if (num_events==MAX_EVENTS) {
printf("Too many events! %d\n",num_events);
exit(1);
}
enum_retval = PAPI_enum_cmp_event( &code, PAPI_ENUM_EVENTS, cid );
}
if (num_events==0) {
printf("Error! No libmsr events found!\n");
exit(1);
}
/* Open output file */
char fileoutname[]="libmsr_write_test_output.txt";
fileout=fopen( fileoutname ,"w" );
if ( fileout==NULL) { fprintf( stderr,"Could not open %s\n",fileoutname ); exit(1); }
/* Create EventSet */
retval = PAPI_create_eventset( &EventSet );
if (retval != PAPI_OK) {
fprintf(stderr,"Error creating eventset!\n");
}
for(i=0;i<num_events;i++) {
retval = PAPI_add_named_event( EventSet, events[i] );
if (retval != PAPI_OK) fprintf(stderr,"Error adding event %s\n",events[i]);
}
start_time=PAPI_get_real_nsec();
/* Grab the initial values for the events */
retval = PAPI_start( EventSet);
if (retval != PAPI_OK) { fprintf(stderr,"PAPI_start() failed\n"); exit(1); }
/* Initial checking read */
retval = PAPI_read( EventSet, values);
if (retval != PAPI_OK) { fprintf(stderr,"PAPI_read() failed\n"); exit(1); }
/* Write a header line */
fprintf( fileout, "ACTION TIME-STAMP TIME-FOR-UNIT-WORK TIME-OVERHEAD-RW\t" );
for(i=0; i<num_events; i++)
fprintf( fileout, "%s ", events[i]+9 );
fprintf( fileout, "\n" );
/* Read the initial values */
retval = PAPI_read( EventSet, values);
if (retval != PAPI_OK) { fprintf(stderr,"PAPI_read() failed\n"); exit(1); }
fprintf( fileout, "INIT %8.3f %8.3f ", ((double)(PAPI_get_real_nsec()-start_time))/1.0e9, 0.0 );
fprintf( fileout, "%8.3e ", 0.0);
for(i=0; i<num_events; i++) {
event_value_union.ll = values[i];
fprintf( fileout, "%8.3f ", event_value_union.dbl );
}
fprintf( fileout, "\n" );
int rpt=0;
int limit1base=10;
int limit2base=10;
while(rpt++<200) {
//printf("rpt %d\n", rpt);
if ( rpt % 10 == 0 ) {
for (i=0; i<num_events; i++) {
event_value_union.ll = values[i];
if ( !strcmp( events[i], "libmsr:::PKG_POWER_LIMIT_1:PACKAGE0" )) event_value_union.dbl=limit1base+(rpt/2);
else if ( !strcmp( events[i], "libmsr:::PKG_TIME_WINDOW_POWER_LIMIT_1:PACKAGE0" )) event_value_union.dbl=1.0;
else if ( !strcmp( events[i], "libmsr:::PKG_POWER_LIMIT_2:PACKAGE0" )) event_value_union.dbl=limit2base+(rpt/2);
else if ( !strcmp( events[i], "libmsr:::PKG_TIME_WINDOW_POWER_LIMIT_2:PACKAGE0" )) event_value_union.dbl=1.0;
else if ( !strcmp( events[i], "libmsr:::PKG_POWER_LIMIT_1:PACKAGE1" )) event_value_union.dbl=limit1base+(rpt/2);
else if ( !strcmp( events[i], "libmsr:::PKG_TIME_WINDOW_POWER_LIMIT_1:PACKAGE1" )) event_value_union.dbl=1.0;
else if ( !strcmp( events[i], "libmsr:::PKG_POWER_LIMIT_2:PACKAGE1" )) event_value_union.dbl=limit2base+(rpt/2);
else if ( !strcmp( events[i], "libmsr:::PKG_TIME_WINDOW_POWER_LIMIT_2:PACKAGE1" )) event_value_union.dbl=1.0;
else event_value_union.dbl=PAPI_NULL;
values[i]=event_value_union.ll;
}
write_start_time=PAPI_get_real_nsec();
retval = PAPI_write( EventSet, values );
write_end_time=PAPI_get_real_nsec();
if (retval != PAPI_OK) { fprintf(stderr,"PAPI_write() failed\n"); exit(1); }
fprintf( fileout, "SET %8.3f %8.3f ", ((double)(PAPI_get_real_nsec()-start_time))/1.0e9, 0.0 );
fprintf( fileout, "%8.3e ", ((double)(write_end_time-write_start_time))/1.0e9 );
for(i=0; i<num_events; i++) {
event_value_union.ll = values[i];
fprintf( fileout, "%8.3f ", event_value_union.dbl );
}
fprintf( fileout, "\n" );
}
/* DO SOME WORK TO USE ENERGY */
//usleep(100000);
double work_start_time=PAPI_get_real_nsec();
ompcpuloadprimes( 100000 );
double work_time=PAPI_get_real_nsec()-work_start_time;
//printf("primescount %d\n", primescount);
/* Read and output the values */
read_start_time=PAPI_get_real_nsec();
retval = PAPI_read( EventSet, values );
read_end_time=PAPI_get_real_nsec();
if (retval != PAPI_OK) { fprintf(stderr,"PAPI_read() failed\n"); exit(1); }
fprintf( fileout, "READ %8.3f %8.3f ", ((double)(PAPI_get_real_nsec()-start_time))/1.0e9, work_time/1.0e9 );
fprintf( fileout, "%8.3e ", ((double)(read_end_time-read_start_time))/1.0e9 );
for(i=0; i<num_events; i++) {
event_value_union.ll = values[i];
fprintf( fileout, "%8.3f ", event_value_union.dbl );
}
fprintf( fileout, "\n" );
}
retval = PAPI_stop( EventSet, values);
return 0;
}
|
vect-simd-clone-7.c | /* { dg-require-effective-target vect_simd_clones } */
/* { dg-additional-options "-fopenmp-simd" } */
/* { dg-additional-options "-mavx" { target avx_runtime } } */
#include "tree-vect.h"
#ifndef N
#define N 1024
#endif
int a[N];
long long int b[N];
short c[N];
#pragma omp declare simd
#pragma omp declare simd uniform(b) linear(c:3)
__attribute__((noinline)) short
foo (int a, long long int b, int c)
{
return a + b + c;
}
__attribute__((noinline, noclone)) void
bar (int x)
{
int i;
if (x == 0)
{
#pragma omp simd
for (i = 0; i < N; i++)
c[i] = foo (a[i], b[i], c[i]);
}
else
{
#pragma omp simd
for (i = 0; i < N; i++)
c[i] = foo (a[i], x, i * 3);
}
}
__attribute__((noinline, noclone)) void
baz (void)
{
int i;
for (i = 0; i < N; i++)
{
a[i] = 2 * i;
b[i] = -7 * i + 6;
c[i] = (i & 31) << 4;
}
}
int
main ()
{
int i;
check_vect ();
baz ();
bar (0);
for (i = 0; i < N; i++)
if (a[i] != 2 * i || b[i] != 6 - 7 * i
|| c[i] != 6 - 5 * i + ((i & 31) << 4))
abort ();
else
a[i] = c[i];
bar (17);
for (i = 0; i < N; i++)
if (a[i] != 6 - 5 * i + ((i & 31) << 4)
|| b[i] != 6 - 7 * i
|| c[i] != 23 - 2 * i + ((i & 31) << 4))
abort ();
return 0;
}
|
GB_unop__identity_uint16_int64.c | //------------------------------------------------------------------------------
// GB_unop: hard-coded functions for each built-in unary operator
//------------------------------------------------------------------------------
// SuiteSparse:GraphBLAS, Timothy A. Davis, (c) 2017-2022, All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
//------------------------------------------------------------------------------
// If this file is in the Generated2/ folder, do not edit it
// (it is auto-generated from Generator/*).
#include "GB.h"
#ifndef GBCOMPACT
#include "GB_control.h"
#include "GB_atomics.h"
#include "GB_unop__include.h"
// C=unop(A) is defined by the following types and operators:
// op(A) function: GB (_unop_apply__identity_uint16_int64)
// op(A') function: GB (_unop_tran__identity_uint16_int64)
// C type: uint16_t
// A type: int64_t
// cast: uint16_t cij = (uint16_t) aij
// unaryop: cij = aij
#define GB_ATYPE \
int64_t
#define GB_CTYPE \
uint16_t
// aij = Ax [pA]
#define GB_GETA(aij,Ax,pA) \
int64_t aij = Ax [pA]
#define GB_CX(p) Cx [p]
// unary operator
#define GB_OP(z, x) \
z = x ;
// casting
#define GB_CAST(z, aij) \
uint16_t z = (uint16_t) aij ;
// cij = op (aij)
#define GB_CAST_OP(pC,pA) \
{ \
/* aij = Ax [pA] */ \
int64_t aij = Ax [pA] ; \
/* Cx [pC] = op (cast (aij)) */ \
uint16_t z = (uint16_t) aij ; \
Cx [pC] = z ; \
}
// disable this operator and use the generic case if these conditions hold
#define GB_DISABLE \
(GxB_NO_IDENTITY || GxB_NO_UINT16 || GxB_NO_INT64)
//------------------------------------------------------------------------------
// Cx = op (cast (Ax)): apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_apply__identity_uint16_int64)
(
uint16_t *Cx, // Cx and Ax may be aliased
const int64_t *Ax,
const int8_t *restrict Ab, // A->b if A is bitmap
int64_t anz,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
int64_t p ;
if (Ab == NULL)
{
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
int64_t aij = Ax [p] ;
uint16_t z = (uint16_t) aij ;
Cx [p] = z ;
}
}
else
{
// bitmap case, no transpose; A->b already memcpy'd into C->b
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (p = 0 ; p < anz ; p++)
{
if (!Ab [p]) continue ;
int64_t aij = Ax [p] ;
uint16_t z = (uint16_t) aij ;
Cx [p] = z ;
}
}
return (GrB_SUCCESS) ;
#endif
}
//------------------------------------------------------------------------------
// C = op (cast (A')): transpose, typecast, and apply a unary operator
//------------------------------------------------------------------------------
GrB_Info GB (_unop_tran__identity_uint16_int64)
(
GrB_Matrix C,
const GrB_Matrix A,
int64_t *restrict *Workspaces,
const int64_t *restrict A_slice,
int nworkspaces,
int nthreads
)
{
#if GB_DISABLE
return (GrB_NO_VALUE) ;
#else
#include "GB_unop_transpose.c"
return (GrB_SUCCESS) ;
#endif
}
#endif
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.