repo
stringlengths
1
152
file
stringlengths
14
221
code
stringlengths
501
25k
file_length
int64
501
25k
avg_line_length
float64
20
99.5
max_line_length
int64
21
134
extension_type
stringclasses
2 values
null
pytorch-main/caffe2/quantization/server/activation_distribution_observer.h
#pragma once #include "caffe2/core/observer.h" #include "caffe2/core/operator.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/dynamic_histogram.h" #include <memory> #include <set> #include <vector> namespace caffe2 { class OutputMinMaxObserver final : public ObserverBase<OperatorBase> { public: explicit OutputMinMaxObserver(OperatorBase* op); ~OutputMinMaxObserver() override; struct TensorInfo { explicit TensorInfo(const std::string& name) : min(std::numeric_limits<float>::max()), max(std::numeric_limits<float>::lowest()), total_min(std::numeric_limits<float>::max()), total_max(std::numeric_limits<float>::lowest()), name(name) {} void Update(float cur_min, float cur_max) { min = std::min(min, cur_min); max = std::max(max, cur_max); total_min = std::min(total_min, cur_min); total_max = std::max(total_max, cur_max); } float min, max; float total_min, total_max; std::string name; }; struct OperatorInfo { std::vector<TensorInfo> tensor_infos; std::string type; }; // OutputMinMaxObserver is assumed to be used together with // OutputMinMaxNetObserver and the information shared via shared_ptr to be // prepared for the case when OutputMinMaxObserver is destroyed before // OutputMinMaxNetObserver std::shared_ptr<OperatorInfo> GetInfo() { return info_; } private: void Stop() override; std::shared_ptr<OperatorInfo> info_; bool warning_printed_ = false; }; // class OutputMinMaxObserver class OutputMinMaxNetObserver final : public NetObserver { public: /// @params dump_freq Print out only once in destructor if -1. // Otherwise, print out every dum_freq invocations explicit OutputMinMaxNetObserver( NetBase* subject, const std::string& out_file_name, int dump_freq = -1, string delimiter = " "); ~OutputMinMaxNetObserver() override; private: void Stop() override; void DumpAndReset_( const std::string& out_file_name, bool print_total_min_max = false); int dump_freq_, cnt_; const std::string out_file_name_; std::string delimiter_; std::vector<std::shared_ptr<OutputMinMaxObserver::OperatorInfo>> min_max_infos_; }; /** * Given min/max, collect histogram */ class HistogramObserver final : public ObserverBase<OperatorBase> { public: struct Info { std::vector<dnnlowp::DynamicHistogram> histograms; std::vector<dnnlowp::DynamicHistogram> total_histograms; OutputMinMaxObserver::OperatorInfo min_max_info; }; explicit HistogramObserver(OperatorBase* op, std::shared_ptr<Info> info); private: void Stop() override; std::shared_ptr<Info> info_; bool warning_printed_ = false; }; // class HistogramObserver /** * Given min/max, collect histogram of the max value of each column of tensor */ class OutputColumnMaxHistogramObserver final : public ObserverBase<OperatorBase> { public: explicit OutputColumnMaxHistogramObserver( OperatorBase* op, const std::string& col_max_blob_name, int nbins, std::shared_ptr<HistogramObserver::Info> info); private: void Stop() override; std::string col_max_blob_name_; int nbins_; std::shared_ptr<HistogramObserver::Info> info_; bool warning_printed_ = false; int col_max_blob_idx_ = -1; int num_columns_ = -1; }; // class OutputColumnMaxHistogramObserver class HistogramNetObserver final : public NetObserver { public: /** * @params mul_nets true if we expect multiple nets with the same name so * we include extra information in the file name to * distinghuish them * @params dump_freq if not -1 we dump histogram every dump_freq invocation * of the net */ explicit HistogramNetObserver( NetBase* subject, const std::string& out_file_name, int nbins, int dump_freq = -1, bool mul_nets = false, string op_filter = "", string delimiter = " "); ~HistogramNetObserver() override; void DumpHistogramFile() { DumpAndReset_(out_file_name_, false); } private: void Stop() override; void DumpAndReset_( const std::string& out_file_name, bool print_total_min_max = false); int dump_freq_, cnt_; /** If multiple nets exist and are attached with the observers, the histogram * files for the nets will be appended with netbase addresses. */ bool mul_nets_; string net_name_; string op_filter_; string delimiter_; const std::string out_file_name_; std::vector<std::shared_ptr<HistogramObserver::Info>> hist_infos_; }; class OutputColumnMaxHistogramNetObserver final : public NetObserver { public: explicit OutputColumnMaxHistogramNetObserver( NetBase* subject, const std::string& out_file_name, const std::vector<std::string>& observe_column_max_for_blobs, int nbins, int dump_freq = -1, bool mul_nets = false, string delimiter = " "); ~OutputColumnMaxHistogramNetObserver() override; void DumpOutputColumnMaxHistogramFile() { DumpAndReset_(out_file_name_, false); } private: void Stop() override; void DumpAndReset_( const std::string& out_file_name, bool print_total_min_max = false); int dump_freq_, cnt_; bool mul_nets_; const std::string out_file_name_; std::string delimiter_; std::unordered_set<std::string> col_max_blob_names_; // {op_idx: {output_index: col_hists}} std::unordered_map< int, std::unordered_map<int, std::shared_ptr<HistogramObserver::Info>>> hist_infos_; }; /** * Set quantization parameters of operators based on min/max * collected from OutputMinMaxObserver */ class RegisterQuantizationParamsNetObserver final : public NetObserver { public: explicit RegisterQuantizationParamsNetObserver( NetBase* subject, const std::string& min_max_file_name, bool is_weight = false, const std::string& qparams_output_file_name = ""); }; /** * Set quantization parameters of operators based on min/max * collected from OutputMinMaxObserver */ class RegisterQuantizationParamsWithHistogramNetObserver final : public NetObserver { public: explicit RegisterQuantizationParamsWithHistogramNetObserver( NetBase* subject, const std::string& histogram_file_name, bool is_weight = false, const std::string& qparams_output_file_name = ""); }; #ifdef _MSC_VER struct tm* localtime_r(time_t* _clock, struct tm* _result) { struct tm* candidate_result = localtime(_clock); if (candidate_result) { *(_result) = *candidate_result; } return candidate_result; } #endif } // namespace caffe2
6,736
27.306723
79
h
null
pytorch-main/caffe2/quantization/server/batch_matmul_dnnlowp_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 "caffe2/operators/batch_matmul_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" #include "fbgemm/Fbgemm.h" namespace caffe2 { template <typename T> class BatchMatMulDNNLowPOp final : public DNNLowPOp<T, BatchMatMulOp<CPUContext>> { public: BatchMatMulDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, BatchMatMulOp<CPUContext>); private: bool trans_a_; bool trans_b_; bool broadcast_{false}; bool is_B_constant_{false}; std::vector<std::int8_t> B_quantized_; std::vector<std::unique_ptr<fbgemm::PackBMatrix<std::int8_t>>> Bq_packed_; std::vector<std::uint8_t> A_pack_buf_; std::vector<std::int32_t> row_offsets_, column_offsets_; std::vector<dnnlowp::TensorQuantizationParams> B_qparams_; std::vector<dnnlowp::RequantizationParams> requantization_params_; std::vector<std::int32_t> Y_int32_; bool first_invocation_{true}; }; // BatchMatMulDNNLowPOp } // namespace caffe2
1,674
30.018519
76
h
null
pytorch-main/caffe2/quantization/server/batch_permutation_dnnlowp_op.h
#pragma once #include "caffe2/operators/copy_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { // FIXME using BatchPermutationFP32Op = CopyOp<CPUContext, CPUContext, CPUContext>; template <typename T> class BatchPermutationDNNLowPOp final : public DNNLowPOp<T, BatchPermutationFP32Op> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, BatchPermutationFP32Op); BatchPermutationDNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : BaseType(operator_def, ws) {} bool RunOnDevice() override; private: INPUT_TAGS(INPUT, INDICES); OUTPUT_TAGS(OUTPUT); }; } // namespace caffe2
671
22.172414
75
h
null
pytorch-main/caffe2/quantization/server/caffe2_dnnlowp_utils.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/utils/eigen_utils.h" namespace dnnlowp { /** * Let consumers of op know that qparams the quantization parameter used * for output_index'th output of op. */ void PropagateOutputTensorQuantizationParams( caffe2::OperatorBase* op, int output_index, const TensorQuantizationParams& qparams); /** * If input_index'th input is already quantized, return quantization parameter * used for the input tensor (should've been set by * PropagateOutputTensorQuantizationParams when the producer was invoked). * If the input tensor is not quantized, return the quantization parameter * chosen by qfactory based on the distribution of the input tensor */ TensorQuantizationParams GetInputTensorQuantizationParamsOf( caffe2::OperatorBase* op, int input_index, const QuantizationFactory* qfactory, bool is_weight = false); void SetStaticQuantizationParams( caffe2::OperatorBase* op, int output_index, const TensorQuantizationParams& qparams); /** * @return true if op's outputs should use static quantization (i.e. op has * Y_scale and optionally Y_zero_offset argument). */ bool HasStaticQuantization( const caffe2::OperatorBase* op, int output_index = 0); /** * Get output_index'th quantization parameter. * Should be used only when UseStaticQuantization is true */ TensorQuantizationParams GetStaticQuantizationParamsOf( const caffe2::OperatorBase* op, int output_index); /** * Quantize input_index'th input if it's not already quantized. * a vector temp should be passed to store quantized results. * * @return array of quantized values */ template <typename T> const T* QuantizeInputIfNeeded( caffe2::OperatorBase* op, int input_index, const TensorQuantizationParams& qparams, std::vector<T>& temp); template <typename T> const T* RowWiseQuantizeInputIfNeeded( caffe2::OperatorBase* op, int input_index, const std::vector<TensorQuantizationParams>& qparams, std::vector<T>& temp); struct QuantizationErrorStats { float sum_sq{0}, sum_err_sq{0}; float max_abs_err{0}; // actual and reference values that resulted in max_abs_err float max_err_actual{0}, max_err_ref{0}; int measure_cnt{0}; }; void MeasureQuantizationError( const float* actual, const float* ref, size_t len, QuantizationErrorStats* stat); void ReportQuantizationError( const caffe2::OperatorBase* op, const QuantizationErrorStats& stat); /** * Get QuantizationFactory based on the arguments of op */ std::unique_ptr<QuantizationFactory> GetQuantizationFactoryOf( const caffe2::OperatorBase* op); void AdjustOutputTensorQuantizationParamsWithFollowedBy( caffe2::OperatorBase* op, const std::string& followed_by); void ParseDNNLowPOperatorArguments( caffe2::OperatorBase* op, bool* dequantize_output = nullptr, bool* measure_quantization_error = nullptr, std::string* followed_by = nullptr); caffe2::NetDef AddScaleZeroOffsetArgumentsWithHistogram( caffe2::NetDef net_def, const std::string& histogram_file_name); } // namespace dnnlowp
3,209
27.918919
78
h
null
pytorch-main/caffe2/quantization/server/channel_shuffle_dnnlowp_op.h
#pragma once #include "caffe2/operators/channel_shuffle_op.h" #include "caffe2/operators/conv_pool_op_base.h" #include "caffe2/quantization/server/conv_pool_dnnlowp_op_base.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { namespace { template <class Context> using ChannelShuffleFp32Op = ChannelShuffleOp<float, Context>; } // namespace template <typename T> class ChannelShuffleDNNLowPOp final : public DNNLowPOp<T, ChannelShuffleFp32Op<CPUContext>> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ChannelShuffleFp32Op<CPUContext>); ChannelShuffleDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; bool RunOnDeviceWithOrderNCHW(); bool RunOnDeviceWithOrderNHWC(); private: const StorageOrder order_; const int group_; }; } // namespace caffe2
923
23.315789
75
h
null
pytorch-main/caffe2/quantization/server/compute_equalization_scale.h
// Copyright 2004-present Facebook. All Rights Reserved. #pragma once #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" namespace caffe2 { class ComputeEqualizationScaleOp final : public Operator<CPUContext> { public: ComputeEqualizationScaleOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws) {} bool RunOnDevice() override; }; // class ComputeEqualizationScaleOp } // namespace caffe2
503
25.526316
76
h
null
pytorch-main/caffe2/quantization/server/concat_dnnlowp_op.h
#pragma once #include "caffe2/operators/concat_split_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { template <typename T> class ConcatDNNLowPOp final : public DNNLowPOp<T, ConcatOp<CPUContext>> { public: ConcatDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ConcatOp<CPUContext>); private: void GetQuantizationParameters_(); int axis_; int add_axis_; // Input: a number of tensors. Output: Y, split // The split are stored in CPU. std::vector<dnnlowp::RequantizationParams> requantization_params_; }; // class ConcatDNNLowPOp } // namespace caffe2
722
23.931034
73
h
null
pytorch-main/caffe2/quantization/server/conv_dnnlowp_acc16_op.h
#pragma once #include "caffe2/quantization/server/conv_dnnlowp_op.h" #include "fbgemm/Fbgemm.h" namespace caffe2 { /** * Quantized Conv operator with 16-bit accumulation. * We'll encounter saturation but this will be faster in Intel CPUs */ template <bool ReluFused = false> class ConvDNNLowPAcc16Op final : public ConvDNNLowPOp<std::uint8_t, ReluFused> { public: USE_CONV_POOL_BASE_FUNCTIONS(CPUContext); ConvDNNLowPAcc16Op(const OperatorDef& operator_def, Workspace* ws); using BaseType = ConvDNNLowPOp<std::uint8_t, ReluFused>; using BaseType::BIAS; using BaseType::col_buffer_; using BaseType::FILTER; using BaseType::in_qparams_; using BaseType::INPUT; using BaseType::InputTensorCPU_; using BaseType::out_qparams_; using BaseType::OutputTensorCPU_; using BaseType::row_offsets_; using BaseType::W_quantized_; using BaseType::X_pack_buf_; using BaseType::Y_int32_; private: bool RunOnDeviceWithOrderNCHW() override; bool RunOnDeviceWithOrderNHWC() override; bool GetQuantizationParameters_(); template <typename PackAMatrix, fbgemm::QuantizationGranularity Q_GRAN> void DispatchFBGEMM_( PackAMatrix& packA, const std::uint8_t* col_buffer_data, vector<std::int32_t>* Y_int32, uint8_t* Y_uint8_data); void ConvOutlier_( const std::uint8_t* col_buffer, vector<std::int32_t>* Y_int32); bool Acc16() const override { return !fallback_to_32_bit_accumulation_; } std::shared_ptr<fbgemm::PackBMatrix<std::int8_t, std::int16_t>> Wq_acc16_packed_; // Wq outlier in CSC format std::shared_ptr<fbgemm::CompressedSparseColumn> Wq_outlier_; // Threshold to decide whether a weight is outlier. // For example, if nbits_in_non_outlier_ == 7, w is an outlier if w < -64 or // w >= 64. // nbits_in_non_outlier_ == 0 means everything is outlier. // nbits_in_non_outlier_ == 8 means nothing is outlier. int nbits_in_non_outlier_; int copy_to_32bit_frequency_; bool first_invocation_{true}; // If outlier matrix is not sparse enough, using 16-bit accumulation won't // give speedup due to too much overhead of sparse matrix multiplication or // sparse convolution anyway, so fallback to 32-bit accumulation bool fallback_to_32_bit_accumulation_{false}; }; // class ConvDNNLowPAcc16Op } // namespace caffe2
2,332
30.106667
80
h
null
pytorch-main/caffe2/quantization/server/conv_dnnlowp_op.h
#pragma once #include <fbgemm/Fbgemm.h> #include "caffe2/operators/conv_op.h" #include "caffe2/operators/conv_pool_op_base.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/conv_pool_dnnlowp_op_base.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/op_wrapper.h" namespace caffe2 { using ConvFp32Op = ConvOp<float, CPUContext>; // Convolutional layer computed in integer with quantization template <typename T, bool ReluFused = false> class ConvDNNLowPOp : public ConvPoolDNNLowPOpBase<T, ConvFp32Op> { public: USE_CONV_POOL_BASE_FUNCTIONS(CPUContext); USE_CONV_POOL_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ConvFp32Op); ConvDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); ~ConvDNNLowPOp() override; protected: bool RunOnDeviceWithOrderNCHW() override; bool RunOnDeviceWithOrderNHWC() override; bool GetQuantizationParameters_(); /** * @return true if convolution is basically a GEMM point-wise (e.g., 1x1) * convolution, no stride/dilation/pad */ bool IsConvGEMM_() const; bool NoIm2ColNHWC_(); int KernelDim_(); const T* Im2ColNHWC_(Tensor* col_buffer); dnnlowp::TensorQuantizationParams& FilterQuantizationParams(int group_id); dnnlowp::RequantizationParams& RequantizationParams(int group_id); static void PartitionGroupedNHWCConv_( int* group_begin, int* group_end, int* i_begin, int* i_end, int num_groups, int m, int nthreads, int thread_id); virtual bool Acc16() const { return false; } Tensor col_buffer_{CPU}; Tensor img_shape_device_{CPU}; Tensor col_buffer_shape_device_{CPU}; // Input: X, W, b // Output: Y INPUT_TAGS(INPUT, FILTER, BIAS); // x86 only provides SIMD instructions that multiply a signed integer with an // unsigned integer. We use signed for weights. using T_signed = typename std::make_signed<T>::type; // used in slow path for T != uint8_t std::vector<T_signed> W_quantized_; // pre-computed biases and offsets std::shared_ptr<std::vector<std::int32_t>> column_offsets_; std::vector<std::int32_t> row_offsets_; const std::int32_t* b_quantized_data_{nullptr}; std::vector<std::uint8_t> X_pack_buf_; void RunOnDeviceEpilogueNCHW_( const T* col_buffer_data, std::int32_t* Y_int32, T* Y_data, std::size_t i_offset, int group_id); void RunOnDeviceEpilogueNHWC_( const T* col_buffer_data, std::int32_t* Y_int32); std::vector<std::int32_t> Y_int32_; std::vector<dnnlowp::TensorQuantizationParams> filter_qparams_; std::vector<std::int32_t> filter_zero_points_; std::vector<float> requantization_multipliers_; bool quantize_groupwise_; private: void QuantizeWeight_(); void PreComputeRowColumnOffsets_(); void QuantizeBias_(); bool TakeDepthWise3x3FastPath_(); bool TakeDepthWise3x3x3FastPath_(); bool TakeGConvFastPath_(); template <typename PackAMatrix, fbgemm::QuantizationGranularity Q_GRAN> void DispatchFBGEMM_( PackAMatrix& packA, vector<std::int32_t>* Y_int32, uint8_t* Y_uint8_data); void ConvNHWCCore_(const T* col_buffer_data, vector<std::int32_t>* Y_int32); fbgemm::conv_param_t<> GetConvParam_(); fbgemm::conv_param_t<3> GetConv3DParam_(); std::vector<dnnlowp::RequantizationParams> requantization_params_; // used in fast path for T == uint8_t std::shared_ptr<fbgemm::PackBMatrix<std::int8_t>> Wq_packed_; // For depthwise conv std::shared_ptr<fbgemm::PackedDepthWiseConvMatrix> Wq_depthwise_packed_; // For small gconv std::shared_ptr<fbgemm::PackWeightMatrixForGConv<std::int8_t>> Wq_gconv_packed_; std::shared_ptr< fbgemm::PackWeightMatrixForGConv<std::int8_t, std::int32_t, 3>> Wq_gconv3d_packed_; // pre-computed biases and offsets std::shared_ptr<std::vector<std::int32_t>> b_quantized_; float in_qparams_scale_old_{0}; std::int32_t in_qparams_zero_point_old_{0}; }; // class ConvDNNLowPOp } // namespace caffe2
4,060
28.427536
79
h
null
pytorch-main/caffe2/quantization/server/conv_pool_dnnlowp_op_base.h
#pragma once #ifdef _OPENMP #include <omp.h> #endif #include "caffe2/core/tensor_int8.h" #include "caffe2/operators/conv_op_shared.h" #include "caffe2/operators/conv_pool_op_base.h" #include "caffe2/quantization/server/fbgemm_pack_blob.h" #include "caffe2/quantization/server/op_wrapper.h" #ifdef _OPENMP C10_DECLARE_int(caffe2_omp_num_threads); #endif C10_DECLARE_bool(caffe2_dnnlowp_shared_int32_buffer); C10_DECLARE_bool(caffe2_force_shared_col_buffer); namespace caffe2 { // TODO: code duplication with dnnlowp_op.h template <typename T, typename FP32_OP> class ConvPoolDNNLowPOpBase : public ConvPoolOpBase<CPUContext> { static_assert(std::is_integral<T>::value, "Integral required."); public: USE_CONV_POOL_BASE_FUNCTIONS(CPUContext); ConvPoolDNNLowPOpBase(const OperatorDef& operator_def, Workspace* ws) : ConvPoolOpBase<CPUContext>(operator_def, ws), in_qparams_(InputSize()), qfactory_(dnnlowp::GetQuantizationFactoryOf(this)) { #ifdef _OPENMP if (FLAGS_caffe2_omp_num_threads > 0) { omp_set_num_threads(FLAGS_caffe2_omp_num_threads); } #endif if (this->debug_def().engine() == "DNNLOWP_16" || this->debug_def().engine() == "DNNLOWP_ROWWISE_16") { LOG(WARNING) << this->debug_def().engine() << " is an experimental feature mostly for testing accuracy with " "fixed-point precision higher than 8 and performance is very slow"; } } virtual ~ConvPoolDNNLowPOpBase() { if (measure_quantization_error_) { dnnlowp::ReportQuantizationError(this, quantization_error_stats_); LOG(WARNING) << this->debug_def().output(0) << " with type " << this->debug_def().type() << " has output qparams : " << "scale " << out_qparams_.scale << " offset " << out_qparams_.zero_point << "; "; } } protected: const TensorCPU& InputTensorCPU_(int idx) { if (InputIsType<int8::Int8TensorCPU>(idx)) { return this->Input<int8::Int8TensorCPU>(idx).t; } else if (InputIsType<Int8ConvDNNLowPPackedWeightBlob>(idx)) { return this->Input<Int8ConvDNNLowPPackedWeightBlob>(idx).original_tensor; } else { return Input(idx); } } TensorCPU* OutputTensorCPU_(int idx) { return &Outputs()[idx]->template GetMutable<int8::Int8TensorCPU>()->t; } Tensor* OutputTensorCPU_(int idx, at::IntArrayRef dims, at::TensorOptions options) { auto* t = &Outputs()[idx]->template GetMutable<int8::Int8TensorCPU>()->t; ReinitializeTensor(t, dims, options.device(CPU)); return t; } T* GetQuantizedOutputData_() { return OutputTensorCPU_(0)->template mutable_data<T>(); } void MeasureQuantizationError_() { if (!measure_quantization_error_ || !Fp32Op_()) { return; } const float* actual = nullptr; vector<float> actual_temp; if (OutputTensorCPU_(0)->template IsType<float>()) { actual = OutputTensorCPU_(0)->template data<float>(); } else { actual_temp.resize(OutputTensorCPU_(0)->numel()); fbgemm::Dequantize<T>( OutputTensorCPU_(0)->template data<T>(), actual_temp.data(), OutputTensorCPU_(0)->numel(), out_qparams_); actual = actual_temp.data(); } TensorCPU* float_tensor = Fp32Op_()->Get()->Output(0); float* ref = float_tensor->template mutable_data<float>(); if (followed_by_ == "Relu" || debug_def().type() == "ConvRelu" || debug_def().type() == "Int8ConvRelu") { for (int i = 0; i < OutputTensorCPU_(0)->numel(); ++i) { ref[i] = std::max(0.f, ref[i]); } } dnnlowp::MeasureQuantizationError( actual, ref, OutputTensorCPU_(0)->numel(), &quantization_error_stats_); } void RunOnDeviceEpilogue_() { dnnlowp::PropagateOutputTensorQuantizationParams(this, 0, out_qparams_); MeasureQuantizationError_(); } void ParseDNNLowPOperatorArguments_() { if (!arguments_parsed_) { bool dequantize_output; dnnlowp::ParseDNNLowPOperatorArguments( this, &dequantize_output, &measure_quantization_error_, &followed_by_); CAFFE_ENFORCE_EQ( dequantize_output, false, "Conv DNNLOWP operators don't support dequantize_output"); arguments_parsed_ = true; } } void GetOutputQuantizationParams_() { using namespace dnnlowp; ParseDNNLowPOperatorArguments_(); if (HasStaticQuantization(this)) { out_qparams_ = GetStaticQuantizationParamsOf(this, 0); if (measure_quantization_error_) { // To measure quantization error, run ref fp32 impl. // This doesn't really belong here but we need to run the reference fp32 // implementation before quantized computation of some inplace operators // will overwrite their inputs. Fp32Op_()->DequantizeInput(); Fp32Op_()->Get()->RunOnDevice(); } } else { // TODO: this is only needed when dequantize_output_ == false but leave // as it is now because some code relies on out_qparams_ initialized even // though it never actually uses it. Fp32Op_()->DequantizeInput(); Fp32Op_()->Get()->RunOnDevice(); out_qparams_ = Fp32Op_()->GetOutputQuantizationParams(qfactory_.get()); } } OpWrapper<FP32_OP, T>* Fp32Op_() { if (!fp32_op_) { fp32_op_.reset(new OpWrapper<FP32_OP, T>(this, qfactory_.get())); } return fp32_op_.get(); } void CreateSharedInt32Buffer_() { auto* mutexPtr = ws_->CreateBlob("__CAFFE2_DNNLOWP_SHARED_INT32_BUFFER_CPU_MUTEX__") ->GetMutable<std::unique_ptr<std::mutex>>(); mutexPtr->reset(new std::mutex()); ws_->CreateBlob("__CAFFE2_DNNLOWP_SHARED_INT32_BUFFER_CPU__"); } void RunWithSharedBuffer_( Tensor* col_buffer, vector<int32_t>* Y_int32, std::function< void(Tensor* col_buffer_shared, vector<int32_t>* Y_int32_shared)> f) { auto f2 = [this, Y_int32, f](Tensor* col_buffer_shared) { if (FLAGS_caffe2_dnnlowp_shared_int32_buffer) { auto* mutexBlob = ws_->GetBlob("__CAFFE2_DNNLOWP_SHARED_INT32_BUFFER_CPU_MUTEX__"); CAFFE_ENFORCE(mutexBlob, "Must call CreateSharedInt32Buffer() first"); auto* mutexPtr = mutexBlob->GetMutable<std::unique_ptr<std::mutex>>(); std::lock_guard<std::mutex> g(**mutexPtr); auto* Y_int32_shared = ws_->GetBlob("__CAFFE2_DNNLOWP_SHARED_INT32_BUFFER_CPU__") ->template GetMutable<vector<int32_t>>(); f(col_buffer_shared, Y_int32_shared); } else { f(col_buffer_shared, Y_int32); } }; if (FLAGS_caffe2_force_shared_col_buffer || this->shared_buffer_) { runWithSharedBuffer<CPUContext>(this->ws_, f2); } else { f2(col_buffer); } } bool measure_quantization_error_{false}; std::string followed_by_; std::vector<dnnlowp::TensorQuantizationParams> in_qparams_; dnnlowp::TensorQuantizationParams out_qparams_; std::unique_ptr<OpWrapper<FP32_OP, T>> fp32_op_; std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; std::vector<T> out_temp_; // Buffer to store quantized output temporarily // when we output dequantized values. dnnlowp::QuantizationErrorStats quantization_error_stats_; bool arguments_parsed_{false}; }; #define USE_CONV_POOL_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, FP32_OP) \ /* using override */ using BaseType = ConvPoolDNNLowPOpBase<T, FP32_OP>; \ /* using override */ using BaseType::GetOutputQuantizationParams_; \ /* using override */ using BaseType::GetQuantizedOutputData_; \ /* using override */ using BaseType::Fp32Op_; \ /* using override */ using BaseType::InputTensorCPU_; \ /* using override */ using BaseType::MeasureQuantizationError_; \ /* using override */ using BaseType::OutputTensorCPU_; \ /* using override */ using BaseType::RunOnDeviceEpilogue_; \ /* using override */ using BaseType::followed_by_; \ /* using override */ using BaseType::in_qparams_; \ /* using override */ using BaseType::measure_quantization_error_; \ /* using override */ using BaseType::out_qparams_; \ /* using override */ using BaseType::qfactory_; } // namespace caffe2
8,443
34.037344
86
h
null
pytorch-main/caffe2/quantization/server/conv_relu_op.h
#pragma once #include "caffe2/operators/conv_op.h" #include "caffe2/operators/conv_pool_op_base.h" namespace caffe2 { template <typename T, class Context> class ConvReluOp final : public ConvPoolOpBase<Context> { public: ConvReluOp(const OperatorDef& operator_def, Workspace* ws) : ConvPoolOpBase<Context>(operator_def, ws) { for (auto name : operator_def.input()) { local_input_blobs_.push_back(local_ws_.CreateBlob(name)); TORCH_CHECK_NOTNULL(local_input_blobs_.back()); } local_op_.reset(new ConvOp<T, Context>(operator_def, &local_ws_)); for (auto name : operator_def.output()) { local_output_blobs_.push_back(local_ws_.GetBlob(name)); TORCH_CHECK_NOTNULL(local_output_blobs_.back()); } } ~ConvReluOp() override {} bool RunOnDeviceWithOrderNCHW() override; bool RunOnDeviceWithOrderNHWC() override; private: Workspace local_ws_; std::vector<Blob*> local_input_blobs_; std::vector<Blob*> local_output_blobs_; std::unique_ptr<ConvOp<T, Context>> local_op_; }; // class ConvReluOp } // namespace caffe2
1,080
29.027778
70
h
null
pytorch-main/caffe2/quantization/server/dnnlowp.h
#pragma once #include <algorithm> #include <cassert> #include <cmath> #include <cstdint> #include <limits> #ifdef __x86_64__ #include <immintrin.h> #endif #include <fbgemm/QuantUtils.h> #include "caffe2/quantization/server/dynamic_histogram.h" #include "caffe2/utils/cpuid.h" namespace dnnlowp { using fbgemm::RequantizationParams; using fbgemm::TensorQuantizationParams; // Represents a quantization scheme that provides quantization parameter based // on distribution of data to be quantized. class QuantizationFactory { public: enum QuantizationKind { // A simple quantization scheme that determines quantization parameter by // just looking at min/max. MIN_MAX_QUANTIZATION, // Minimizes L2 norm of quantization error L2_MIN_QUANTIZATION, // fast search to remove histogram outliers and approximate L2 min L2_MIN_QUANTIZATION_APPROX, // Minimizes Kullback-Leibler divergence KL_MIN_QUANTIZATION, // Take 99 percentail (only works with sparsity preserving quantization) P99_QUANTIZATION, L1_MIN_QUANTIZATION, }; /// Get the default factory whose policy is determined by gflags static QuantizationFactory* GetDefaultInstance(); /// Choose quantization scale and zero_point that maps /// floating-point range [min, max] to the integer range of the specified /// precision TensorQuantizationParams ChooseQuantizationParams( float min, float max, int precision, bool preserve_sparsity, bool is_signed = false) const { TensorQuantizationParams qparams = fbgemm::ChooseQuantizationParams( min, max, is_signed ? -(1 << (precision - 1)) : 0, is_signed ? ((1 << (precision - 1)) - 1) : (1 << precision) - 1, preserve_sparsity, force_scale_power_of_two_); qparams.precision = precision; return qparams; } /// Choose quantization scale and zero_point that maps /// floating-point range [min, max] to the default integer range of /// this quantization factory TensorQuantizationParams ChooseQuantizationParams(float min, float max, bool is_weight = false) const { return ChooseQuantizationParams( min, max, is_weight ? GetWeightPrecision() : GetActivationPrecision(), is_weight ? GetPreserveWeightSparsity() : GetPreserveActivationSparsity()); } /// Choose quantization based on the values in an array to optimize the /// quantization errors ignoring a few outliers TensorQuantizationParams ChooseQuantizationParams( const float* values, int len, QuantizationKind kind, int precision, bool preserve_sparsity) const; TensorQuantizationParams ChooseQuantizationParams( const float* values, int len, bool is_weight = false) const; /// Choose quantization based on histogram of values to optimize the /// quantization errors ignoring a few outliers TensorQuantizationParams ChooseQuantizationParams( const Histogram& hist, QuantizationKind kind, int precision, bool preserve_sparsity, bool is_weight = false) const; TensorQuantizationParams ChooseQuantizationParams( const Histogram& hist, bool is_weight = false) const; // Given a real_multiplier, produces a pair (quantized_multiplier, // right_shift) where quantized_multiplier is an int32 representing a // fixed-point value (in practice we only produce positive values) and // right_shift is an amount to shift right by, so that the floating-point // multiplication of some int32 input value by real_multiplier, // // return static_cast<int32>(int32_value * real_multiplier); // // is best approximated by the integer-arithmetic-only code // // return RoundingRightShift( // Multiplication(int32_value, quantized_multiplier), // right_shift); // // Note: all this code only needs to run offline to generate the quantized // neural network workload, not at runtime on the device on which quantized // neural networks need to run. So it's not performance-critical at all. RequantizationParams ChooseRequantizationMultiplier( float real_multiplier, TensorQuantizationParams target_qparams) const; int GetActivationPrecision() const { return activation_precision_; } int GetWeightPrecision() const { return weight_precision_; } int GetEltwiseQuantizePrecision() const { return eltwise_quantize_precision_; } bool GetPreserveActivationSparsity() const { return preserve_activation_sparsity_; } bool GetPreserveWeightSparsity() const { return preserve_weight_sparsity_; } QuantizationKind GetActivationKind() const { return activation_kind_; } QuantizationKind GetWeightKind() const { return weight_kind_; } void SetWeightP99Threshold(float threshold) { weight_p99_threshold_ = threshold; } void SetActivationP99Threshold(float threshold) { activation_p99_threshold_ = threshold; } explicit QuantizationFactory( int activation_precision = 8, // precision used for activations in main operations like matmul int weight_precision = 8, // precision used for weights int requantization_multiplier_precision = 32, // precision used for the requantization multiplier int eltwise_quantize_precision = 16, // precision used for element-wise addition bool preserve_activation_sparsity = false, // preserve zeros in quantization bool preserve_weight_sparsity = false, // preserve zeros in quantization bool force_scale_power_of_two = false, // restrict scaling to a power of two QuantizationKind activation_kind = MIN_MAX_QUANTIZATION, QuantizationKind weight_kind = MIN_MAX_QUANTIZATION, float weight_p99_threshold = 0.99, // P99 percentage to select out from the full histogram for weights float activation_p99_threshold = 0.99 // P99 percentage to select out from the full histogram for activations ); private: int activation_precision_; int weight_precision_; int requantization_multiplier_precision_; int eltwise_quantize_precision_; bool preserve_activation_sparsity_; bool preserve_weight_sparsity_; bool force_scale_power_of_two_; QuantizationKind activation_kind_, weight_kind_; float weight_p99_threshold_; float activation_p99_threshold_; }; // class QuantizationFactory /** * Parse a string to QuantizationKind */ QuantizationFactory::QuantizationKind StringToKind(const std::string& s); std::vector<float> adjust_hist_to_include_zero(const Histogram& hist, float* min, float* max); } // namespace dnnlowp
6,678
31.580488
80
h
null
pytorch-main/caffe2/quantization/server/dnnlowp_op.h
#pragma once #ifdef _OPENMP #include <omp.h> #endif #include "caffe2/core/operator.h" #include "caffe2/core/tensor_int8.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/fbgemm_pack_blob.h" #include "caffe2/quantization/server/op_wrapper.h" #include "caffe2/quantization/server/sigmoid.h" #include "caffe2/quantization/server/tanh.h" #ifdef _OPENMP C10_DECLARE_int(caffe2_omp_num_threads); #endif namespace caffe2 { /** * @brief A convenient base class for C2 operators with DNNLOWP engine. * DNNLOWP ops give flexibility on the type of input/output blobs. * For example, some inputs can be the usual fp32 tensor and they will be * quantized before doing actual computation. * Otherwise, the inputs should be pre-quantized Int8TensorCPU. * A few constraints: when the weight is pre-quantized if and only if the * bias is also pre-quantized. * * static quantization vs. dynamic quantization * When Y_scale and Y_zero_point (optional with default = 0) arg is set, * and dequantize_output is false, we do static quantization, meaning * we're using the same pre-computed scale and zero_point for the output * activation tensor. * Otherwise, we do dynamic quantization by looking at the min/max of * output activation tensor for each batch. * Y_scale and Y_zero_point arguments are used for static quantization. * scale and zero_point of Int8TensorCPU is used for carrying * quantization information across operators both in static and dynamic * quantization. This means scale and zero_point of Int8TensorCPU is * valid only for the current batch and will be reset in the next batch * when dynamic quantization is used. * * C2 operators with DNNLOWP engine have the following arguments: * - dequantize_output (default=false): when true, output is dequantized * as fp32. Useful when we're only quantizing individual operators * rather than doing end-to-end quantization. Conv operators don't support dequantize_output option as an exception because doing so complicate the implementation significantly and having a separate Dequantize operator doesn't add much overhead because Conv ops are usually used in deep networks where regions of quantization are long chains. * - followed_by (default=null): can be relu, sigmoid, or tanh. When * specified, the current operator is only followed by relu, sigmoid, * or tanh, and this fact can be used for more accurate output * quantization. * - measure_quantization_error (default=false): when true, L2 error * with respect to the baseline C2 operator in fp32 is reported. * WARNING: turning this option will make performance very slow and * this option is intended for debugging accuracy issues. * * For the following quantization method related options, please refer * to caffe2/quantization/server/dnnlowp.cc for more details. * * - activation_quantization_precision (default=8) * - weight_quantization_precision (default=8) * - requantization_multiplier_precision (default=32) * - eltwise_quantization_precision (default=16) * - force_scale_power_of_two (default=0) * - preserve_activation_sparsity (default=0) * - preserve_weight_sparsity (default=0) * - activation_quantization_kind (default=min_max) * - weight_quantization_kind (default=min_max) */ template <typename T, typename FP32_OP> class DNNLowPOp : public Operator<CPUContext> { static_assert(std::is_integral<T>::value, "Integral required."); public: USE_OPERATOR_FUNCTIONS(CPUContext); DNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), in_qparams_(InputSize()), qfactory_(dnnlowp::GetQuantizationFactoryOf(this)) { #ifdef _OPENMP if (FLAGS_caffe2_omp_num_threads > 0) { omp_set_num_threads(FLAGS_caffe2_omp_num_threads); } #endif if (this->debug_def().engine() == "DNNLOWP_16" || this->debug_def().engine() == "DNNLOWP_ROWWISE_16") { LOG(WARNING) << this->debug_def().engine() << " is an experimental feature mostly for testing accuracy with " "fixed-point precision higher than 8 and performance is very slow"; } } virtual ~DNNLowPOp() { if (measure_quantization_error_) { dnnlowp::ReportQuantizationError(this, quantization_error_stats_); } } protected: const TensorCPU& InputTensorCPU_(int idx) { if (InputIsType<int8::Int8TensorCPU>(idx)) { return this->Input<int8::Int8TensorCPU>(idx).t; } else if (InputIsType<Int8FCDNNLowPPackedWeightBlob>(idx)) { return this->Input<Int8FCDNNLowPPackedWeightBlob>(idx).original_tensor; } else { return Input(idx); } } TensorCPU* OutputTensorCPU_(int idx) { if (dequantize_output_) { return Output(idx); } else { return &Outputs()[idx]->template GetMutable<int8::Int8TensorCPU>()->t; } } Tensor* OutputTensorCPU_(int idx, at::IntArrayRef dims, at::TensorOptions options) { if (dequantize_output_) { return Output(idx, dims, options.device(CPU)); } else { auto* t = &Outputs()[idx]->template GetMutable<int8::Int8TensorCPU>()->t; ReinitializeTensor(t, dims, options.device(CPU)); return t; } } T* GetQuantizedOutputData_() { if (dequantize_output_) { out_temp_.resize(Output(0)->numel()); return out_temp_.data(); } else { return OutputTensorCPU_(0)->template mutable_data<T>(); } } void MeasureQuantizationError_() { if (!measure_quantization_error_ || !Fp32Op_()) { return; } const float* actual = nullptr; std::vector<float> actual_temp; if (OutputTensorCPU_(0)->template IsType<float>()) { actual = OutputTensorCPU_(0)->template data<float>(); std::string op_type = this->debug_def().type(); bool relu_fused = op_type.length() >= 4 && op_type.compare(op_type.length() - 4, 4, "Relu") == 0; if (GetSingleArgument<std::string>("followed_by", "") == "Relu" && !relu_fused) { // If dequantize_output_ is true and relu is not fused, // dnnlowp op won't clip negative values. Do it here. actual_temp.resize(OutputTensorCPU_(0)->numel()); for (int i = 0; i < Output(0)->numel(); ++i) { actual_temp[i] = std::max(0.f, actual[i]); } actual = actual_temp.data(); } } else { actual_temp.resize(OutputTensorCPU_(0)->numel()); fbgemm::Dequantize<T>( OutputTensorCPU_(0)->template data<T>(), actual_temp.data(), OutputTensorCPU_(0)->numel(), out_qparams_); actual = actual_temp.data(); } float* ref = Fp32Op_()->Get()->Output(0)->template mutable_data<float>(); if (followed_by_ == "Relu") { for (int i = 0; i < OutputTensorCPU_(0)->numel(); ++i) { ref[i] = std::max(0.f, ref[i]); } } dnnlowp::MeasureQuantizationError( actual, ref, OutputTensorCPU_(0)->numel(), &quantization_error_stats_); } void RunOnDeviceEpilogue_() { if (dequantize_output_) { fbgemm::Dequantize<T>( out_temp_.data(), OutputTensorCPU_(0)->template mutable_data<float>(), OutputTensorCPU_(0)->numel(), out_qparams_); } else { dnnlowp::PropagateOutputTensorQuantizationParams(this, 0, out_qparams_); } MeasureQuantizationError_(); } void ParseDNNLowPOperatorArguments_() { // Ideally, this should be done in constructor but any modification of // arguments in ParseDNNLowPOperatorArguments will be ignored if we call // this from constructor. // Make sure all derived classes call this "early enough" so that they // use correct parameters. if (!arguments_parsed_) { dnnlowp::ParseDNNLowPOperatorArguments( this, &dequantize_output_, &measure_quantization_error_, &followed_by_); arguments_parsed_ = true; } } void GetOutputQuantizationParams_( dnnlowp::TensorQuantizationParams* out_qparams_overwrite = nullptr) { using namespace dnnlowp; ParseDNNLowPOperatorArguments_(); if (HasStaticQuantization(this)) { if (out_qparams_overwrite != nullptr) { out_qparams_ = *out_qparams_overwrite; } else { out_qparams_ = GetStaticQuantizationParamsOf(this, 0); } if (measure_quantization_error_) { // To measure quantization error, run ref fp32 impl. // This doesn't really belong here but we need to run the reference fp32 // implementation before quantized computation of some inplace operators // will overwrite their inputs. Fp32Op_()->DequantizeInput(); Fp32Op_()->Get()->RunOnDevice(); } } else { // TODO: this is only needed when dequantize_output_ == false but leave // as it is now because some code relies on out_qparams_ initialized even // though it never actually uses it. Fp32Op_()->DequantizeInput(); Fp32Op_()->Get()->RunOnDevice(); if (out_qparams_overwrite != nullptr) { out_qparams_ = *out_qparams_overwrite; } else { out_qparams_ = Fp32Op_()->GetOutputQuantizationParams(qfactory_.get()); } } } OpWrapper<FP32_OP, T>* Fp32Op_() { if (!fp32_op_) { fp32_op_.reset(new OpWrapper<FP32_OP, T>(this, qfactory_.get())); } return fp32_op_.get(); } bool dequantize_output_{false}, measure_quantization_error_{false}; std::string followed_by_; std::vector<dnnlowp::TensorQuantizationParams> in_qparams_; dnnlowp::TensorQuantizationParams out_qparams_; std::unique_ptr<OpWrapper<FP32_OP, T>> fp32_op_; std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; std::vector<T> out_temp_; // Buffer to store quantized output temporarily // when we output dequantized values. dnnlowp::QuantizationErrorStats quantization_error_stats_; bool arguments_parsed_{false}; }; #define USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, FP32_OP) \ /* using override */ using BaseType = DNNLowPOp<T, FP32_OP>; \ /* using override */ using BaseType::GetOutputQuantizationParams_; \ /* using override */ using BaseType::GetQuantizedOutputData_; \ /* using override */ using BaseType::Fp32Op_; \ /* using override */ using BaseType::InputTensorCPU_; \ /* using override */ using BaseType::MeasureQuantizationError_; \ /* using override */ using BaseType::OutputTensorCPU_; \ /* using override */ using BaseType::RunOnDeviceEpilogue_; \ /* using override */ using BaseType::dequantize_output_; \ /* using override */ using BaseType::followed_by_; \ /* using override */ using BaseType::in_qparams_; \ /* using override */ using BaseType::measure_quantization_error_; \ /* using override */ using BaseType::out_qparams_; \ /* using override */ using BaseType::qfactory_; inline int dnnlowp_get_num_threads() { #ifdef _OPENMP return omp_get_num_threads(); #else return 1; #endif } inline int dnnlowp_get_max_threads() { #ifdef _OPENMP return omp_get_max_threads(); #else return 1; #endif } inline int dnnlowp_get_thread_num() { #ifdef _OPENMP return omp_get_thread_num(); #else return 0; #endif } } // namespace caffe2
11,769
35.896552
80
h
null
pytorch-main/caffe2/quantization/server/dnnlowp_partition.h
#pragma once #include <algorithm> #include <cstddef> #include <utility> namespace caffe2 { std::pair<size_t, size_t> Get1DPartition(size_t work, int nthreads, int tid, int work_align = 1); /** * 1D-partition m x n 2D work. * First try partitioning m if m >= nthreads. * Otherwise, each row is partitioned by multiple threads. * In this case, each thread only works on a single row. * Optionally, we can force the number of columns assigned per thread is a * multiple of n_align. */ void Get1DPartitionOf2D( int m, int n, int nthreads, int thread_id, int* m_begin, int* m_end, int* n_begin, int* n_end, int n_align = 1); } // namespace caffe2
690
20.59375
74
h
null
pytorch-main/caffe2/quantization/server/dynamic_histogram.h
#pragma once #include <memory> #include <vector> namespace dnnlowp { /** * bin_width = (max - min)/nbins * ith bin (zero-based indexing) contains [i*bin_width, (i+1)*bin_width) * with an exception that (nbins - 1)th bin contains * [(nbins-1)*bin_width, nbins*bin_width] * */ class Histogram { public: Histogram(int nbins, float min, float max) : min_(min), max_(max), histogram_(nbins) {} Histogram(float min, float max, const std::vector<uint64_t>& bins) : min_(min), max_(max), histogram_(bins) {} void Add(float f, uint64_t cnt = 1); /** * This version collects histogram with single thread */ void Add(const float* f, int len); float Min() const { return min_; } float Max() const { return max_; } const std::vector<uint64_t>* GetHistogram() const { return &histogram_; } private: float min_, max_; std::vector<uint64_t> histogram_; }; /// An equi-width histogram where the spread of bins change over time when /// we see new min or max values. class DynamicHistogram { public: DynamicHistogram(int nbins); void Add(float f); void Add(const float* f, int len); /// Indicate we're not dynamically adjusting histogram bins any more and /// return the current static histogram. const Histogram* Finalize(); private: /// Dynamic histogram is implemented by the series of static histograms /// and expands from the old histogram to new histogram when /// we see a new extremum. /// An invariant: the beginning of the first bin of histograms_[i] exactly /// matches with the beginning of a bin in histograms_[i+1]. The end of the /// last bin of histograms_[i] exactly matches with the end of a bin in /// histograms_[i+1]. std::unique_ptr<Histogram> histogram_; int nbins_; float min_, max_; std::unique_ptr<Histogram> final_histogram_; }; // class DynamicHistogram } // namespace dnnlowp
1,901
25.054795
77
h
null
pytorch-main/caffe2/quantization/server/elementwise_dnnlowp_op.h
#pragma once #include "caffe2/core/tensor_int8.h" #include "caffe2/operators/elementwise_ops.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp_op.h" #include "caffe2/quantization/server/sigmoid.h" namespace caffe2 { template <typename T, class Functor> class UnaryElementwiseWithArgsDNNLowPOp : public Operator<CPUContext> { public: USE_OPERATOR_FUNCTIONS(CPUContext); UnaryElementwiseWithArgsDNNLowPOp( const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), functor_() {} bool RunOnDevice() override { if (!arguments_parsed_) { dnnlowp::ParseDNNLowPOperatorArguments(this); dnnlowp::SetStaticQuantizationParams( this, 0, functor_.GetOutputQuantizationParams()); arguments_parsed_ = true; } auto& input = this->template Input<int8::Int8TensorCPU>(0).t; auto& output = Outputs()[0]->template GetMutable<int8::Int8TensorCPU>()->t; output.ResizeLike(input); functor_( input.size(), input.template data<T>(), output.template mutable_data<T>()); dnnlowp::PropagateOutputTensorQuantizationParams( this, 0, functor_.GetOutputQuantizationParams()); return true; } private: Functor functor_; bool arguments_parsed_{false}; }; template <typename T, typename FP32_OP> class BinaryElementwiseDNNLowPOp : public DNNLowPOp<T, FP32_OP> { public: USE_OPERATOR_FUNCTIONS(CPUContext); BinaryElementwiseDNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : DNNLowPOp<T, FP32_OP>(operator_def, ws), OP_SINGLE_ARG(bool, "broadcast", enable_broadcast_, 0), OP_SINGLE_ARG(int, "axis", axis_, -1), OP_SINGLE_ARG(string, "axis_str", axis_str_, ""), OP_SINGLE_ARG(string, "order", order_, "NCHW") { // Figure out the correct axis to use. if (enable_broadcast_) { if (axis_ != -1) { // Get axis from an explicit axis argument. CAFFE_ENFORCE_EQ( axis_str_.size(), 0, "Args axis and axis_str cannot be used simultaneously."); } else if (axis_str_.size()) { // Get the axis index semantically. CAFFE_ENFORCE_EQ( axis_str_.size(), 1, "Unsupported axis string", axis_str_); size_t semantic_axis_ = order_.find(axis_str_); CAFFE_ENFORCE_NE( semantic_axis_, string::npos, "Unrecognizable axis string ", axis_str_, " from order string ", order_); axis_ = semantic_axis_; } } else { CAFFE_ENFORCE( axis_ == -1 && axis_str_.size() == 0, "Do not specify axis or axis_str if broadcast is not enabled."); } } protected: bool enable_broadcast_; int axis_; string axis_str_; string order_; dnnlowp::RequantizationParams requantization_params_; }; // BinaryElementwiseDNNLowPOp // For arithmetic operators, Eigen provides a good way to vectorize even // when broadcasting. #define DECLARE_EIGEN_FUNCTOR(name, eigen_op, input_type, output_type) \ struct Eigen##name##Functor { \ template <int b_is_scalar, typename T, typename R> \ inline void Run(size_t n, const T* a, const T* b, R* out, CPUContext*) { \ if (b_is_scalar) { \ EigenVectorArrayMap<R>(out, n) = \ eigen_op((ConstEigenVectorArrayMap<T>(a, n)), (b[0])); \ } else { \ EigenVectorArrayMap<R>(out, n) = eigen_op( \ (ConstEigenVectorArrayMap<T>(a, n)), \ (ConstEigenVectorArrayMap<T>(b, n))); \ } \ } \ template <typename T, typename R> \ void RunWithBroadcast( \ const T* a, \ const T* b, \ R* out, \ size_t pre, \ size_t n, \ CPUContext*) { \ EigenArrayMap<R>(out, n, pre) = eigen_op( \ (ConstEigenArrayMap<T>(a, n, pre).colwise()), \ (ConstEigenVectorArrayMap<T>(b, n))); \ } \ template <typename T, typename R> \ void RunWithBroadcast2( \ const T* a, \ const T* b, \ R* out, \ size_t pre, \ size_t n, \ size_t post, \ CPUContext*) { \ for (const auto i : c10::irange(pre)) { \ EigenArrayMap<R>(out + i * n * post, post, n) = eigen_op( \ (ConstEigenArrayMap<T>(a + i * n * post, post, n).rowwise()), \ (Eigen::Map<const Eigen::Array<T, 1, Eigen::Dynamic>>(b, n))); \ } \ } \ }; } // namespace caffe2
6,252
44.311594
86
h
null
pytorch-main/caffe2/quantization/server/elementwise_linear_dnnlowp_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 "caffe2/operators/elementwise_linear_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { using ElementwiseLinearFp32Op = ElementwiseLinearOp<float, CPUContext>; template <typename T> class ElementwiseLinearDNNLowPOp final : public DNNLowPOp<T, ElementwiseLinearFp32Op> { public: ElementwiseLinearDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ElementwiseLinearFp32Op); private: bool GetQuantizationParameters_(); int axis_; dnnlowp::RequantizationParams requantization_params_; std::vector<T> a_quantized_; }; } // namespace caffe2
1,339
27.510638
77
h
null
pytorch-main/caffe2/quantization/server/fb_fc_packed_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 <fbgemm/FbgemmFP16.h> #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/operators/fully_connected_op.h" #include "caffe2/utils/conversions.h" #include "caffe2/utils/math.h" namespace caffe2 { /** * C2 wrapper for fp16 gemm * * Suppose your predict_net has an FC operator in fp32 as follows: * op { * input: "x" * input: "w" * input: "b" * output: "y" * type: "FC" * } * ... * external_input: "w" * * To use FbFCPacked operator with fp16 fbgemm, in init_net * ... # an operator that generates w * op { * input: "w" * output: "w_packed" * type: "FbGemmPack" * } * ... * external_output: "w_packed" * * in predict_net: * op { * input: "x" * input: "w_packed" * input: "b" * output: "y" * type: "FbFCPacked" * } * ... * external_input: "w_packed" */ template < class Context, class Engine = DefaultEngine, typename T_W = fbgemm::float16> class FbFCPackedOperator final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FbFCPackedOperator(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), axis_(this->template GetSingleArgument<int32_t>("axis", 1)), axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)) {} ~FbFCPackedOperator() override {} // template on X, B, and Y. template <typename T_X, typename T_B, typename T_Y> bool DoRunWithType() { const auto& X = Input(0); const auto& b = Input(2); CAFFE_ENFORCE(b.dim() == 1, b.dim()); // batch size const auto canonical_axis = X.canonical_axis_index(axis_); const int M = X.size_to_dim(canonical_axis); const int N = b.numel(); // Load the packed matrix auto* W = OperatorBase::Input<caffe2::unique_ptr<fbgemm::PackedGemmMatrixFP16>>(1) .get(); const int K = W->numRows(); if (!W->packed()) { if (!packed_w_) { std::vector<float> src_mat(W->matSize()); for (int i = 0; i < W->matSize(); ++i) { src_mat[i] = fbgemm::cpu_half2float(W->pmat()[i]); } packed_w_ = std::make_unique<fbgemm::PackedGemmMatrixFP16>( fbgemm::matrix_op_t::Transpose, W->numRows(), W->numCols(), 1.0, src_mat.data()); } W = packed_w_.get(); } auto dimErrorString = [&]() { return c10::str( "Dimension mismatch: ", "X: ", X.sizes(), ", W: ", std::vector<int>({K, W->numCols()}), ", b: ", b.sizes(), ", axis: ", axis_, ", M: ", M, ", N: ", N, ", K: ", K); }; // Error checking CAFFE_ENFORCE(M == X.numel() / K, dimErrorString()); CAFFE_ENFORCE(K == X.size_from_dim(canonical_axis), dimErrorString()); CAFFE_ENFORCE(N == W->numCols(), dimErrorString()); Y_shape_cache_ = X.sizes().vec(); // This is an invariant of canonical_axis, so we can DCHECK. TORCH_DCHECK_LE(canonical_axis + 1, Y_shape_cache_.size()); Y_shape_cache_.resize(canonical_axis + 1); Y_shape_cache_[canonical_axis] = N; auto* Y = Output(0, Y_shape_cache_, at::dtype<T_Y>()); if (X.numel() == 0) { // skip the rest of the computation if X is empty Y->template mutable_data<T_Y>(); return true; } // Call the fp16 gemm interface fbgemm::cblas_gemm_compute( fbgemm::matrix_op_t::NoTranspose, M, X.template data<T_X>(), *W, 0.f, Y->template mutable_data<T_Y>()); // Add bias term, accumulation is still in fp32. TensorProto::DataType math_type = TensorProto_DataType_FLOAT; if (bias_multiplier_.numel() != M) { // If the helper bias multiplier is not M, reshape and fill it with one. bias_multiplier_.Resize(M); math::Set<T_B, Context>( M, convert::To<float, T_B>(1), bias_multiplier_.template mutable_data<T_B>(), &context_); } math::Gemm<T_B, Context, Engine>( CblasNoTrans, CblasNoTrans, M, N, 1, 1, bias_multiplier_.template data<T_B>(), b.template data<T_B>(), 1, Y->template mutable_data<T_Y>(), &context_, math_type); return true; } bool RunOnDevice() override { return DoRunWithType< float, // X float, // B float>(); // Y } protected: size_t axis_{1}; size_t axis_w_{1}; // A local vector to cache the output shape so we don't need to recreate // a vector object every time we run Run(). vector<int64_t> Y_shape_cache_; Tensor bias_multiplier_{Context::GetDeviceType()}; caffe2::unique_ptr<fbgemm::PackedGemmMatrixFP16> packed_w_{nullptr}; }; class PackedGemmMatrixFP16ShapeFunctions : public ExternalTensorFunctionsBase { public: explicit PackedGemmMatrixFP16ShapeFunctions() : ExternalTensorFunctionsBase() {} ~PackedGemmMatrixFP16ShapeFunctions() override {} bool isQuantized() const override { return false; } bool IsSameMetaType(TypeIdentifier id) override; void SetupExternalTensorDescriptor( const Blob* blob, std::vector<std::vector<uint64_t>>* shapes, std::vector<std::vector<float>>* all_scales, std::vector<std::vector<int32_t>>* all_offsets, ExternalTensorDescriptor* desc) override; void LoadInfoOfBlob( const Blob* /* unused */, std::vector<float>* /* unused */, std::vector<float>* /* unused */, uint32_t* /* unused */) override {} TypeIdentifier GetTypeMetaId() override; TypeMeta GetExternalTensorType(const void* c) override; vector<int64_t> GetExternalTensorInfo( const void* c, size_t* capacity, DeviceOption* device) override; }; } // namespace caffe2
6,518
27.845133
80
h
null
pytorch-main/caffe2/quantization/server/fbgemm_fp16_pack_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 <fbgemm/FbgemmFP16.h> #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/utils/conversions.h" #include "caffe2/utils/math.h" namespace caffe2 { template < class Context, class Engine = DefaultEngine, bool TransposeWeight = true, typename TPacked = fbgemm::float16> class FbGemmPackOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FbGemmPackOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), axis_(this->template GetSingleArgument<int32_t>("axis_w", 1)), no_packing_( this->template GetSingleArgument<int32_t>("no_packing", 0)) {} ~FbGemmPackOp() override {} bool RunOnDevice() override { const auto& X = Input(0); const auto canonical_axis = X.canonical_axis_index(axis_); const auto N = X.size_to_dim(canonical_axis); const auto K = X.size_from_dim(canonical_axis); fbgemm::PackedGemmMatrixFP16* resultPtr; if (TransposeWeight) { resultPtr = new fbgemm::PackedGemmMatrixFP16( fbgemm::matrix_op_t::Transpose, K, N, 1.0f, /*alpha*/ X.template data<float>()); } else { resultPtr = new fbgemm::PackedGemmMatrixFP16( fbgemm::matrix_op_t::NoTranspose, N, K, 1.0f, /*alpha*/ X.template data<float>()); } if (no_packing_) { C10_LOG_FIRST_N(WARNING, 10) << "no_packing will be deprecated soon"; vector<fbgemm::float16> src_mat(resultPtr->matSize()); fbgemm::float16* pmat = resultPtr->pmat(); memcpy( src_mat.data(), pmat, resultPtr->matSize() * sizeof(fbgemm::float16)); resultPtr->unpackFromSrc(fbgemm::matrix_op_t::Transpose, src_mat.data()); } auto* Y = this->template Output<unique_ptr<fbgemm::PackedGemmMatrixFP16>>(0); Y->reset(resultPtr); return true; } protected: size_t axis_{1}; // Do not pack the layout, for testing only bool no_packing_; }; } // namespace caffe2
2,695
29.636364
80
h
null
pytorch-main/caffe2/quantization/server/fbgemm_pack_blob.h
#pragma once #include <memory> #include <fbgemm/Fbgemm.h> #include <caffe2/core/tensor.h> #include "caffe2/quantization/server/dnnlowp.h" namespace caffe2 { /** * Packed weight matrix for DNNLOWP Int8FC operator */ struct Int8FCDNNLowPPackedWeightBlob { std::vector<dnnlowp::TensorQuantizationParams> qparams; std::shared_ptr<std::vector<std::int32_t>> column_offsets; // The original tensor before packing but only with meta information Tensor original_tensor{CPU}; std::shared_ptr<std::vector<std::int32_t>> bias; // Only for 32-bit accumulation std::shared_ptr<fbgemm::PackBMatrix<std::int8_t>> W; // Only for 16-bit accumulation // Dense matrix holding common values std::shared_ptr<fbgemm::PackBMatrix<std::int8_t, std::int16_t>> W_acc16; // Sparse matrix holding outliers std::shared_ptr<fbgemm::CompressedSparseColumn> W_outlier; int nbits_in_non_outlier; }; /** * Packed weight matrix for DNNLOWP Int8Conv operator */ struct Int8ConvDNNLowPPackedWeightBlob : public Int8FCDNNLowPPackedWeightBlob { // Only for 32-bit accumulation std::shared_ptr<fbgemm::PackedDepthWiseConvMatrix> W_depthwise; std::shared_ptr<fbgemm::PackWeightMatrixForGConv<std::int8_t>> W_gconv; std::shared_ptr< fbgemm::PackWeightMatrixForGConv<std::int8_t, std::int32_t, 3>> W_gconv3d; }; } // namespace caffe2
1,352
27.1875
79
h
null
pytorch-main/caffe2/quantization/server/fbgemm_pack_matrix_cache.h
#pragma once #include "fbgemm/Fbgemm.h" namespace caffe2 { /** * If there's an existing packed matrix for the same matrix, reuse it. * Create a new one otherwise. This can save memory usage if many threads are * sharing the same weight. */ template <typename ACC_T> std::shared_ptr<fbgemm::PackBMatrix<int8_t, ACC_T>> GetOrCreateFbgemmPackBMatrix( fbgemm::matrix_op_t trans, std::int32_t m, std::int32_t n, const void* orig_data, const std::int8_t* quantized_data, std::int32_t ld); } // namespace caffe2
537
22.391304
77
h
null
pytorch-main/caffe2/quantization/server/fbgemm_pack_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/operators/conv_op.h" #include "caffe2/quantization/server/conv_pool_dnnlowp_op_base.h" #include "caffe2/quantization/server/fbgemm_pack_blob.h" #include "caffe2/quantization/server/fully_connected_dnnlowp_op.h" namespace caffe2 { using FCFp32Op = FullyConnectedOp<CPUContext>; void QuantizeConvBias( const Blob& blob, int M, const dnnlowp::TensorQuantizationParams& in_qparams, const vector<dnnlowp::TensorQuantizationParams>& filter_qparams, std::vector<int32_t>& b_quantized, bool use_fp16=false, bool round_nearest_even=true); class FullyConnectedDNNLowPPackWeightOp final : public DNNLowPOp<std::uint8_t, FCFp32Op> { public: FullyConnectedDNNLowPPackWeightOp( const OperatorDef& operator_def, Workspace* ws); USE_OPERATOR_FUNCTIONS(CPUContext); bool RunOnDevice() override; private: int axis_w_; bool quantize_channelwise_; int nbits_in_non_outlier_; // only for DNNLOWP_ACC16 bool save_unpacked_weights_; INPUT_TAGS(FILTER, BIAS); }; using ConvFp32Op = ConvOp<float, CPUContext>; /** * Pack a weight matrix that can be used by DNNLOWP Int8Conv operators. * DNNLOWP operators can pack matrix on demand during their first invocations * but calling this operator to pre-pack can have benefits like saving memory * space when multiple operators are sharing the same weight. * This operator should be a part of init net to be called once to populate * packed blob to be used by Int8Conv DNNLOWP operators in the predictor net * * This operator optionally can also pre-quantize bias. * Then, we should also provide the scale of input activation tensor as in_scale * argument. */ class ConvDNNLowPPackWeightOp final : public ConvPoolDNNLowPOpBase<std::uint8_t, ConvFp32Op> { public: USE_CONV_POOL_BASE_FUNCTIONS(CPUContext); USE_CONV_POOL_DNNLOWP_OPERATOR_BASE_FUNCTIONS(std::uint8_t, ConvFp32Op); ConvDNNLowPPackWeightOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; private: bool TakeDepthWise3x3FastPath_(); bool TakeDepthWise3x3x3FastPath_(); bool TakeGConvFastPath_(); fbgemm::conv_param_t<> GetConvParam_(); fbgemm::conv_param_t<3> GetConv3DParam_(); // Save quantized weights right after quantization before layout packing for // performance purpose bool save_unpacked_weights_; bool quantize_groupwise_; int nbits_in_non_outlier_; // only for DNNLOWP_ACC16 INPUT_TAGS(FILTER, BIAS); }; // Helper functions for packing weights that can be used by // ConvDNNLowPAcc16PackWeightOp, ConvDNNLowPOp, and ConvDNNLowPAcc16Op template <typename T> void QuantizeWeight( const Blob& blob, int kernel_dim, int M, vector<dnnlowp::TensorQuantizationParams>& qparams, vector<typename std::make_signed<T>::type>& w_quantized, dnnlowp::QuantizationFactory* qfactory); template <typename T> void ComputeColumnOffsets( int num_rows, int num_cols, const T* W, const vector<dnnlowp::TensorQuantizationParams>& qparams, vector<int32_t>& col_offsets); int CountOutliers( int groups, int kernel_dim, int M, int nbits_in_non_outlier, vector<std::int8_t>& W_quantized); /** * @param W_quantized input quantized weight that is not packed yet */ fbgemm::CompressedSparseColumn* ExtractOutlierMatrix( int groups, int kernel_dim, int M, int nbits_in_non_outlier, vector<std::int8_t>& W_quantized); /* * Set up used onnxifi data type constexpr * Should always be synced with onnxifi.h */ constexpr uint64_t kONNXIFI_DATATYPE_UINT8 = 2; constexpr uint64_t kONNXIFI_DATATYPE_INT32 = 6; constexpr uint64_t kONNXIFI_DATATYPE_INT8 = 3; class Int8ConvDNNLowpPackedWeightBlobShapeFunctions : public ExternalTensorFunctionsBase { public: explicit Int8ConvDNNLowpPackedWeightBlobShapeFunctions() : ExternalTensorFunctionsBase() {} ~Int8ConvDNNLowpPackedWeightBlobShapeFunctions() override {} bool isQuantized() const override { return true; } bool IsSameMetaType(TypeIdentifier id) override; void SetupExternalTensorDescriptor( const Blob* blob, std::vector<std::vector<uint64_t>>* shapes, std::vector<std::vector<float>>* all_scales, std::vector<std::vector<int32_t>>* all_offsets, ExternalTensorDescriptor* desc) override; void LoadInfoOfBlob( const Blob* blob, std::vector<float>* scale, std::vector<float>* offset, uint32_t* axis) override; TypeIdentifier GetTypeMetaId() override; TypeMeta GetExternalTensorType(const void* c) override; vector<int64_t> GetExternalTensorInfo( const void* c, size_t* capacity, DeviceOption* device) override; }; class Int8FCDNNLowpPackedWeightBlobShapeFunctions : public ExternalTensorFunctionsBase { public: explicit Int8FCDNNLowpPackedWeightBlobShapeFunctions() : ExternalTensorFunctionsBase() {} ~Int8FCDNNLowpPackedWeightBlobShapeFunctions() override {} bool isQuantized() const override { return true; } bool IsSameMetaType(TypeIdentifier id) override; void SetupExternalTensorDescriptor( const Blob* blob, std::vector<std::vector<uint64_t>>* shapes, std::vector<std::vector<float>>* all_scales, std::vector<std::vector<int32_t>>* all_offsets, ExternalTensorDescriptor* desc) override; void LoadInfoOfBlob( const Blob* blob, std::vector<float>* scale, std::vector<float>* offset, uint32_t* axis) override; TypeIdentifier GetTypeMetaId() override; TypeMeta GetExternalTensorType(const void* c) override; vector<int64_t> GetExternalTensorInfo( const void* c, size_t* capacity, DeviceOption* device) override; }; } // namespace caffe2
5,763
30.67033
90
h
null
pytorch-main/caffe2/quantization/server/fully_connected_dnnlowp_acc16_op.h
#pragma once #include "caffe2/quantization/server/fully_connected_dnnlowp_op.h" namespace caffe2 { /** * Quantized FC operator with 16-bit accumulation. * We'll encounter saturation but this will be faster in Intel CPUs */ class FullyConnectedDNNLowPAcc16Op final : public FullyConnectedDNNLowPOp<std::uint8_t> { public: FullyConnectedDNNLowPAcc16Op(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); using BaseType = FullyConnectedDNNLowPOp<std::uint8_t>; using BaseType::dequantize_output_; using BaseType::in_qparams_; using BaseType::InputTensorCPU_; using BaseType::out_qparams_; using BaseType::OutputTensorCPU_; using BaseType::W_quantized_; private: std::shared_ptr<fbgemm::PackBMatrix<std::int8_t, std::int16_t>> Wq_acc16_packed_; // Wq outlier in CSC format std::shared_ptr<fbgemm::CompressedSparseColumn> Wq_outlier_; int nbits_in_non_outlier_; int copy_to_32bit_frequency_; }; // class FullyConnectedDNNLowPAcc16Op } // namespace caffe2
1,065
27.052632
79
h
null
pytorch-main/caffe2/quantization/server/fully_connected_dnnlowp_op.h
#pragma once #include <fbgemm/Fbgemm.h> #include "caffe2/operators/fully_connected_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { template <typename T, bool ReluFused = false> class FullyConnectedDNNLowPOp : public DNNLowPOp<T, FullyConnectedOp<CPUContext>> { public: FullyConnectedDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, FullyConnectedOp<CPUContext>); protected: bool GetQuantizationParameters_(float X_scale_=-1.0, int X_zero_point_=0); std::size_t axis_{1}; std::size_t axis_w_{1}; float X_scale_{-1.0}; int X_zero_point_{0}; vector<std::int64_t> Y_shape_cache_; std::vector<dnnlowp::RequantizationParams> requantization_params_; bool requantization_param_selected_{false}; // x86 only provides SIMD instructions that multiply a signed integer with an // unsigned integer. We use signed for weights. using T_signed = typename std::make_signed<T>::type; // used in fast path for T == uint8_t std::shared_ptr<fbgemm::PackBMatrix<std::int8_t>> Wq_packed_; std::vector<std::uint8_t> X_pack_buf_; std::vector<std::int32_t> Y_int32_; std::vector<dnnlowp::TensorQuantizationParams> filter_qparams_; std::vector<float> filter_scales_; std::vector<std::int32_t> filter_zero_points_; std::vector<float> requantization_multipliers_; bool quantize_channelwise_; // used in slow path for T != uint8_t std::vector<T_signed> W_quantized_; // pre-computed biases and offsets std::shared_ptr<std::vector<std::int32_t>> b_quantized_; const std::int32_t* b_quantized_data_{nullptr}; std::vector<std::int32_t> row_offsets_; std::shared_ptr<std::vector<std::int32_t>> column_offsets_; // Dequantized bias populated when input bias is quantized and // dequantized_output_ == true std::vector<float> b_dequantized_; const float* b_dequantized_data_{nullptr}; bool is_weight_constant_{true}; float in_qparams0_scale_old_ = 0; std::int32_t in_qparams0_zero_point_old_ = 0; }; // class FullyConnectedDNNLowPOp } // namespace caffe2
2,160
30.779412
79
h
null
pytorch-main/caffe2/quantization/server/fully_connected_fake_lowp_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 #ifdef __x86_64__ #include <immintrin.h> #endif #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/utils/conversions.h" #include "caffe2/utils/math.h" namespace caffe2 { // convert to float16 reducing mantissa, preserving exponent void fp32_to_bfp16(const float* source, size_t size, float* dest); // convert to float24 reducing mantissa, preserving exponent void fp32_to_bfp24(const float* source, size_t size, float* dest); // convert to float14 reducing mantissa, preserving exponent void fp32_to_bfp14(const float* source, size_t size, float* dest); void fp32_to_bfp16_scalar(const float* source, size_t size, float* dest); // convert to IEEE float16 void fp32_to_fp16(const float* source, size_t size, float* dest); // fp32 -> int32 -> += 1<< 15 -> fp32 -> truncation void fp32_to_bfp16_round(const float* source, size_t size, float* dest); // This is Caffe's InnerProductOp, with a name that fits its purpose better. template < void (*Q)(const float*, size_t, float*), class Context, class Engine = DefaultEngine, bool TransposeWeight = true> class FullyConnectedFakeLowpFPOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FullyConnectedFakeLowpFPOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), axis_(this->template GetSingleArgument<int32_t>("axis", 1)), axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)), float16_compute_( this->template GetSingleArgument<bool>("float16_compute", false)) {} ~FullyConnectedFakeLowpFPOp() override {} template < typename T_X, typename T_W, typename T_B, typename T_Y, typename MATH> bool DoRunWithType(); bool RunOnDevice() override { return DoRunWithType< float, // X float, // W float, // B float, // Y float>(); // Math } protected: size_t axis_{1}; size_t axis_w_{1}; // A local vector to cache the output shape so we don't need to recreate // a vector object every time we run Run(). vector<int64_t> Y_shape_cache_; Tensor bias_multiplier_; bool float16_compute_; }; template < void (*Q)(const float*, size_t, float*), class Context, class Engine = DefaultEngine, bool TransposeWeight = true> class FullyConnectedGradientFakeLowpFPOp : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FullyConnectedGradientFakeLowpFPOp( const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), axis_(this->template GetSingleArgument<int32_t>("axis", 1)), axis_w_(this->template GetSingleArgument<int32_t>("axis_w", 1)), float16_compute_( this->template GetSingleArgument<bool>("float16_compute", false)) {} ~FullyConnectedGradientFakeLowpFPOp() override {} template < typename T_X, typename T_W, typename T_DY, typename T_B, typename T_DX, typename T_DW, typename T_DB, typename MATH> bool DoRunWithType(); bool RunOnDevice() override { return DoRunWithType< float, // X float, // W float, // dY float, // B float, // dX float, // dW float, // dB float>(); // Math } protected: size_t axis_{1}; size_t axis_w_{1}; Tensor bias_multiplier_; bool float16_compute_; }; } // namespace caffe2
4,101
28.3
80
h
null
pytorch-main/caffe2/quantization/server/group_norm_dnnlowp_op.h
#pragma once #include <vector> #include "caffe2/operators/group_norm_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { using GroupNormFP32Op = GroupNormOp<float, CPUContext>; template <typename T> class GroupNormDNNLowPOp final : public DNNLowPOp<T, GroupNormFP32Op> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, GroupNormFP32Op); GroupNormDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; private: bool GetQuantizationParameters(); void QuantizeGamma(); void QuantizeGammaImpl(); void QuantizeBeta(); bool RunOnDeviceWithOrderNCHW(); bool RunOnDeviceWithOrderNHWC(); void QuantizedGroupMomentsNCHW( int N, int G, int K, int HxW, const T* X, int32_t* mu, int32_t* rsig); void QuantizedGroupMomentsNHWC( int N, int G, int K, int HxW, const T* X, int32_t* mu, int32_t* rsig); void DequantizedGroupMomentsNCHW( int N, int G, int K, int HxW, const T* X, float* mu, float* rsig); void DequantizedGroupMomentsNHWC( int N, int G, int K, int HxW, const T* X, float* mu, float* rsig); void ComputeQuantizedInvStd( int N, const float* var, float* rsig, int32_t* rsig_quantized); void ComputeQuantizedFusedParams( int N, int G, int K, const int32_t* mu, const int32_t* rsig, const int32_t* gamma, const int32_t* beta, int32_t* scale, int32_t* bias); void ComputeDequantizedFusedParams( int N, int G, int K, const float* mu, const float* rsig, const float* gamma, const float* beta, float* scale, float* bias); void AffineBatchChannelQuantizedNCHW( int N, int C, int HxW, const T* X, const int32_t* scale, const int32_t* bias, T* Y); void AffineBatchChannelQuantizedNHWC( int N, int C, int HxW, const T* X, const int32_t* scale, const int32_t* bias, T* Y); void AffineBatchChannelDequantizedNCHW( int N, int C, int HxW, const float* X, const float* scale, const float* bias, float* Y); void AffineBatchChannelDequantizedNHWC( int N, int C, int HxW, const float* X, const float* scale, const float* bias, float* Y); const bool is_test_; const int group_; const float epsilon_; const StorageOrder order_; const bool is_param_constant_; std::vector<int32_t> mu_quantized_; std::vector<int32_t> rsig_quantized_; std::vector<float> mu_dequantized_; std::vector<float> rsig_dequantized_; dnnlowp::TensorQuantizationParams rsig_qparams_; std::vector<int32_t> gamma_quantized_; std::vector<int32_t> beta_quantized_; std::vector<float> gamma_dequantized_; std::vector<float> beta_dequantized_; const int32_t* gamma_quantized_data_ = nullptr; const int32_t* beta_quantized_data_ = nullptr; const float* gamma_dequantized_data_ = nullptr; const float* beta_dequantized_data_ = nullptr; std::vector<int32_t> scale_quantized_; std::vector<int32_t> bias_quantized_; std::vector<float> scale_dequantized_; std::vector<float> bias_dequantized_; dnnlowp::TensorQuantizationParams internal_qparams_; std::vector<float> X_dequantized_; std::vector<int32_t> Y_int32_; float cached_X_qparams_scale_ = 0.0f; // Input: X, gamma, beta // Output: Y, mu, inv_sig INPUT_TAGS(INPUT, GAMMA, BETA); OUTPUT_TAGS(OUTPUT, MU, INV_SIGMA); }; namespace internal { template <typename T> void VectorMomentsAVX2(const int N, const T* src, int64_t* sum, int64_t* sumsq); void ComputeQuantizedFusedParamsAVX2( const int N, const int G, const int K, const int32_t X_zero_point, const int32_t* mu, const int32_t* rsig, const int32_t* gamma, int32_t* scale, int32_t* bias); template <typename T> void AffineBatchChannelAndRequantizeNCHWAVX2( const int N, const int C, const int HxW, const dnnlowp::RequantizationParams& params, const T* X, const int32_t* scale, const int32_t* bias, T* Y); template <typename T> void AffineBatchChannelAndRequantizeNHWCAVX2( const int N, const int C, const int HxW, const dnnlowp::RequantizationParams& params, const T* X, const int32_t* scale, const int32_t* bias, T* Y); } // namespace internal } // namespace caffe2
4,610
20.546729
80
h
null
pytorch-main/caffe2/quantization/server/im2col_dnnlowp.h
#pragma once #ifdef _OPENMP #include <omp.h> #endif #include "caffe2/core/operator.h" #include "caffe2/utils/math.h" #include "caffe2/utils/math/utils.h" namespace caffe2 { namespace math { template <typename T> static void Im2ColNCHW( const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int dilation_h, const int dilation_w, const int pad_t, const int pad_l, const int pad_b, const int pad_r, const int stride_h, const int stride_w, const T* data_im, T* data_col, CPUContext* /*context*/, const T& zero_point = 0) { const int output_h = (height + pad_b + pad_t - (dilation_h * (kernel_h - 1) + 1)) / stride_h + 1; const int output_w = (width + pad_l + pad_r - (dilation_w * (kernel_w - 1) + 1)) / stride_w + 1; // Fast path for zero padding and no dilation // From Torch, THNN_(unfolded_copy) if (dilation_h == 1 && dilation_w == 1 && pad_l == 0 && pad_r == 0 && pad_t == 0 && pad_b == 0) { for (auto k = 0; k < channels * kernel_h * kernel_w; k++) { const auto nip = k / (kernel_h * kernel_w); const auto rest = k % (kernel_h * kernel_w); const auto kh = rest / kernel_w; const auto kw = rest % kernel_w; auto* dst = data_col + nip * (kernel_h * kernel_w * output_h * output_w) + kh * (kernel_w * output_h * output_w) + kw * (output_h * output_w); const auto* src = data_im + nip * (height * width); for (const auto y : c10::irange(output_h)) { const auto iy = y * stride_h + kh; const auto ix = kw; if (stride_w == 1) { memcpy( dst + (y * output_w), src + (iy * width + ix), sizeof(T) * output_w); } else { for (const auto x : c10::irange(output_w)) { memcpy( dst + (y * output_w + x), src + (iy * width + ix + x * stride_w), sizeof(T)); } } } } return; } // Fast path for equal padding if (pad_l == pad_r && pad_t == pad_b) { // From Intel, https://github.com/BVLC/caffe/pull/3536 const int pad_h = pad_t; const int pad_w = pad_l; const int channel_size = height * width; for (int channel = channels; channel--; data_im += channel_size) { for (const auto kernel_row : c10::irange(kernel_h)) { for (const auto kernel_col : c10::irange(kernel_w)) { int input_row = -pad_h + kernel_row * dilation_h; for (int output_rows = output_h; output_rows; output_rows--) { if (!utils::IsAGeZeroAndALtB(input_row, height)) { for (int output_cols = output_w; output_cols; output_cols--) { *(data_col++) = zero_point; } } else { int input_col = -pad_w + kernel_col * dilation_w; for (int output_col = output_w; output_col; output_col--) { if (utils::IsAGeZeroAndALtB(input_col, width)) { *(data_col++) = data_im[input_row * width + input_col]; } else { *(data_col++) = zero_point; } input_col += stride_w; } } input_row += stride_h; } } } } return; } // Baseline const int dkernel_h = dilation_h * (kernel_h - 1) + 1; const int dkernel_w = dilation_w * (kernel_w - 1) + 1; int height_col = (height + pad_t + pad_b - dkernel_h) / stride_h + 1; int width_col = (width + pad_l + pad_r - dkernel_w) / stride_w + 1; int channels_col = channels * kernel_h * kernel_w; for (const auto c : c10::irange(channels_col)) { int w_offset = c % kernel_w; int h_offset = (c / kernel_w) % kernel_h; int c_im = c / kernel_h / kernel_w; for (const auto h : c10::irange(height_col)) { for (const auto w : c10::irange(width_col)) { int h_pad = h * stride_h - pad_t + h_offset * dilation_h; int w_pad = w * stride_w - pad_l + w_offset * dilation_w; if (h_pad >= 0 && h_pad < height && w_pad >= 0 && w_pad < width) data_col[(c * height_col + h) * width_col + w] = data_im[(c_im * height + h_pad) * width + w_pad]; else data_col[(c * height_col + h) * width_col + w] = zero_point; } } } } template <typename T> static void Im2ColNdNCHW( const int N, const int /* img_size*/, const int col_size, const int* img_shape, const int* col_shape, const int* kernel_shape, const int* stride, const int* dilation, const int* pad, const T* X_data, T* Y_data, CPUContext* /* context */, const T& zero_point = 0) { const int outer_size = col_shape[0]; const int inner_size = col_size / outer_size; const int kernel_size = std::accumulate( kernel_shape, kernel_shape + N, 1, std::multiplies<int>()); std::vector<int> d_offset(N, 0); std::vector<int> d_iter(N, 0); for (const auto i : c10::irange(outer_size)) { // Loop over spatial axes in reverse order to compute a per-axis offset. int offset = i; for (int d_i = N - 1; d_i >= 0; --d_i) { d_offset[d_i] = offset % kernel_shape[d_i]; offset /= kernel_shape[d_i]; } for (const auto j : c10::irange(inner_size)) { // Loop over spatial axes in forward order to compute the indices in the // image and column, and whether the index lies in the padding. const int col_index = i * inner_size + j; int img_index = i / kernel_size; bool is_padding = false; for (const auto d_i : c10::irange(N)) { const int d_img = d_iter[d_i] * stride[d_i] - pad[d_i] + d_offset[d_i] * dilation[d_i]; is_padding |= d_img < 0 || d_img >= img_shape[d_i + 1]; img_index = img_index * img_shape[d_i + 1] + d_img; } Y_data[col_index] = is_padding ? zero_point : X_data[img_index]; utils::IncreaseIndexInDims(N, col_shape + 1, d_iter.data()); } } } /** * The layout of the result is N H W G R S C/G. * Note that groups are pulled out to an outer dimension so that we can use * GEMMs efficiently. */ template <typename T> static void Im2ColNHWC( const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int dilation_h, const int dilation_w, const int pad_t, const int pad_l, const int pad_b, const int pad_r, const int stride_h, const int stride_w, const T* data_im, T* data_col, CPUContext* /*context*/, const int groups, const T& zero_point) { const int dkernel_h = dilation_h * (kernel_h - 1) + 1; const int dkernel_w = dilation_w * (kernel_w - 1) + 1; int height_col = (height + pad_t + pad_b - dkernel_h) / stride_h + 1; int width_col = (width + pad_l + pad_r - dkernel_w) / stride_w + 1; #ifdef _OPENMP #pragma omp parallel for if (!omp_in_parallel()) #endif for (int h = 0; h < height_col; ++h) { int h_pad = -pad_t + h * stride_h; T* data_col_temp = data_col + h * width_col * kernel_h * kernel_w * channels; int w_pad = -pad_l; for (C10_UNUSED const auto w : c10::irange(width_col)) { int r = 0; for (int ih = h_pad; ih < h_pad + dkernel_h; ih += dilation_h, ++r) { int s = 0; for (int iw = w_pad; iw < w_pad + dkernel_w; iw += dilation_w, ++s) { if (ih >= 0 && ih < height && iw >= 0 && iw < width) { for (const auto g : c10::irange(groups)) { memcpy( data_col_temp + ((g * kernel_h + r) * kernel_w + s) * (channels / groups), data_im + (ih * width + iw) * channels + g * (channels / groups), sizeof(T) * (channels / groups)); } } else { // This should be simply padded with zero. for (const auto g : c10::irange(groups)) { for (int i = 0; i < channels / groups; ++i) { data_col_temp [(((g * kernel_h + r) * kernel_w) + s) * (channels / groups) + i] = zero_point; } } } } // for each iw } // for each ih data_col_temp += kernel_h * kernel_w * channels; w_pad += stride_w; } // for each output pixel } // for each image row } /** * The layout of the result is N T H W G Q R S C/G. * Note that groups are pulled out to an outer dimension so that we can use * GEMMs efficiently. */ template <typename T> static void Im2Col3DNHWC( const int channels, const int num_frames, const int height, const int width, const int kernel_t, const int kernel_h, const int kernel_w, const int dilation_t, const int dilation_h, const int dilation_w, const int pad_p, // previous frame const int pad_t, // top const int pad_l, // left const int pad_n, // next frame const int pad_b, // bottom const int pad_r, // right const int stride_t, const int stride_h, const int stride_w, const T* data_im, T* data_col, CPUContext* /*context*/, const int groups, const T& zero_point) { const int dkernel_t = dilation_t * (kernel_t - 1) + 1; const int dkernel_h = dilation_h * (kernel_h - 1) + 1; const int dkernel_w = dilation_w * (kernel_w - 1) + 1; int frame_col = (num_frames + pad_p + pad_n - dkernel_t) / stride_t + 1; int height_col = (height + pad_t + pad_b - dkernel_h) / stride_h + 1; int width_col = (width + pad_l + pad_r - dkernel_w) / stride_w + 1; #ifdef _OPENMP #pragma omp parallel for if (!omp_in_parallel()) #endif for (int t = 0; t < frame_col; ++t) { int t_pad = -pad_p + t * stride_t; for (const auto h : c10::irange(height_col)) { int h_pad = -pad_t + h * stride_h; T* data_col_temp = data_col + (t * height_col + h) * width_col * kernel_t * kernel_h * kernel_w * channels; for (const auto w : c10::irange(width_col)) { int w_pad = -pad_l + w * stride_w; int q = 0; for (int it = t_pad; it < t_pad + dkernel_t; it += dilation_t, ++q) { int r = 0; for (int ih = h_pad; ih < h_pad + dkernel_h; ih += dilation_h, ++r) { int s = 0; for (int iw = w_pad; iw < w_pad + dkernel_w; iw += dilation_w, ++s) { if (it >= 0 && it < num_frames && ih >= 0 && ih < height && iw >= 0 && iw < width) { for (const auto g : c10::irange(groups)) { memcpy( data_col_temp + (((g * kernel_t + q) * kernel_h + r) * kernel_w + s) * (channels / groups), data_im + ((it * height + ih) * width + iw) * channels + g * (channels / groups), sizeof(T) * (channels / groups)); } } else { // This should be simply padded with zero. for (const auto g : c10::irange(groups)) { for (int i = 0; i < channels / groups; ++i) { data_col_temp [((((g * kernel_t + q) * kernel_h + r) * kernel_w) + s) * (channels / groups) + i] = zero_point; } } } } // for each iw } // for each ih } // for each it data_col_temp += kernel_t * kernel_h * kernel_w * channels; } // for each output pixel } // for each image row } // for each frame } } // namespace math } // namespace caffe2
11,848
33.344928
80
h
null
pytorch-main/caffe2/quantization/server/int8_gen_quant_params.h
// Copyright 2004-present Facebook. All Rights Reserved. #pragma once #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" namespace caffe2 { using namespace std; using dnnlowp::TensorQuantizationParams; struct Int8QuantSchemeBlob { public: Int8QuantSchemeBlob(std::string quantization_kind, bool preserve_sparsity) : quantization_kind_(quantization_kind), preserve_sparsity_(preserve_sparsity) {} std::string quantization_kind_; bool preserve_sparsity_; }; struct Int8QuantParamsBlob { public: Int8QuantParamsBlob(float scale, int zero_point) { qparam.scale = scale; qparam.zero_point = zero_point; } dnnlowp::TensorQuantizationParams qparam; }; template <class Context, class Engine = DefaultEngine> class Int8GenQuantParamsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; Int8GenQuantParamsOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) {} bool RunOnDevice() override { // Generate Int8 quant params based on the input data (last N samples of the // activations) and the quant scheme const auto& input_data = Input(0); const auto* quant_scheme = this->template Input<unique_ptr<Int8QuantSchemeBlob>>(1).get(); CAFFE_ENFORCE(input_data.dim() > 0); CAFFE_ENFORCE(quant_scheme); std::string quant_kind = quant_scheme->quantization_kind_; bool preserve_sparsity = quant_scheme->preserve_sparsity_; dnnlowp::QuantizationFactory* qfactory = dnnlowp::QuantizationFactory::GetDefaultInstance(); TensorQuantizationParams qparam = qfactory->ChooseQuantizationParams( input_data.template data<float>(), input_data.numel(), dnnlowp::StringToKind(quant_kind), 8, preserve_sparsity); auto* output_qparam = this->template Output<unique_ptr<Int8QuantParamsBlob>>(0); output_qparam->reset( new Int8QuantParamsBlob(qparam.scale, qparam.zero_point)); return true; } }; // class Int8GenQuantParamsOp } // namespace caffe2
2,113
33.096774
80
h
null
pytorch-main/caffe2/quantization/server/int8_gen_quant_params_min_max.h
// Copyright 2004-present Facebook. All Rights Reserved. #pragma once #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/int8_gen_quant_params.h" #include <glog/logging.h> namespace caffe2 { using namespace std; using dnnlowp::TensorQuantizationParams; template <class Context, class Engine = DefaultEngine> class Int8GenQuantParamsMinMaxOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; Int8GenQuantParamsMinMaxOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) {} bool RunOnDevice() override { // Generate Int8 quant params based on the input data (last N samples of the // activations) and the quant scheme const float min = OperatorBase::Input<Tensor>(0, CPU).template data<float>()[0]; const float max = OperatorBase::Input<Tensor>(1, CPU).template data<float>()[0]; bool preserve_sparsity = false; if (InputSize() == 3){ const auto* quant_scheme = this->template Input<unique_ptr<Int8QuantSchemeBlob>>(2).get(); preserve_sparsity = quant_scheme->preserve_sparsity_; } dnnlowp::QuantizationFactory* qfactory = dnnlowp::QuantizationFactory::GetDefaultInstance(); TensorQuantizationParams qparam = qfactory->ChooseQuantizationParams( min, max, 8, preserve_sparsity); auto* output_qparam = this->template Output<unique_ptr<Int8QuantParamsBlob>>(0); output_qparam->reset( new Int8QuantParamsBlob(qparam.scale, qparam.zero_point)); LOG_EVERY_N(INFO, 1) << "scale and bias are " << qparam.scale << "," << qparam.zero_point; return true; } }; // class Int8GenQuantParamsOp } // namespace caffe2
1,827
34.843137
94
h
null
pytorch-main/caffe2/quantization/server/int8_quant_scheme_blob_fill.h
// Copyright 2004-present Facebook. All Rights Reserved. #pragma once #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/int8_gen_quant_params.h" namespace caffe2 { using namespace std; template <class Context, class Engine = DefaultEngine> class Int8QuantSchemeBlobFillOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; Int8QuantSchemeBlobFillOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) {} bool RunOnDevice() override { std::string quantization_kind = this->template GetSingleArgument<std::string>( "quantization_kind", "MIN_MAX_QUANTIZATION"); bool preserve_sparsity = this->template GetSingleArgument<bool>("preserve_sparsity", false); auto* output_qscheme = this->template Output<unique_ptr<Int8QuantSchemeBlob>>(0); output_qscheme->reset( new Int8QuantSchemeBlob(quantization_kind, preserve_sparsity)); return true; } }; // class Int8QuantSchemeBlobFillOp } // namespace caffe2
1,135
32.411765
75
h
null
pytorch-main/caffe2/quantization/server/lstm_unit_dnnlowp_op.h
#pragma once #include "caffe2/operators/lstm_unit_op.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/op_wrapper.h" #include "caffe2/quantization/server/sigmoid.h" namespace caffe2 { template <typename T> class LSTMUnitDNNLowPOp final : public LSTMUnitOp<CPUContext> { static_assert(std::is_integral<T>::value, "Integral required."); public: LSTMUnitDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); ~LSTMUnitDNNLowPOp() override; bool RunOnDevice() override; private: const TensorCPU& InputTensorCPU_(int idx); TensorCPU* OutputTensorCPU_(int idx); bool GetQuantizationParameters_(); OpWrapper<LSTMUnitOp<CPUContext>, T>* Fp32Op_(); bool drop_states_; dnnlowp::Sigmoid<T> sigmoid_; dnnlowp::Tanh<T> tanh_; dnnlowp::TensorQuantizationParams H_in_qparams_, C_in_qparams_, G_in_qparams_, H_out_qparams_, C_out_qparams_; std::unique_ptr<OpWrapper<LSTMUnitOp<CPUContext>, T>> fp32_op_; bool dequantize_output_{false}, measure_quantization_error_{false}; std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; dnnlowp::QuantizationErrorStats cell_quantization_error_stats_, hidden_quantization_error_stats_; bool arguments_parsed_{false}; }; // class LSTMUnitDNNLowPOp } // namespace caffe2
1,357
29.177778
80
h
null
pytorch-main/caffe2/quantization/server/mmio.h
#pragma once #include <cstdio> #include <set> #include <string> #include <type_traits> namespace caffe2 { template <typename T> void StoreMatrixInMatrixMarketFormat( int m, int n, const T* a, const std::string& matrix_name) { using namespace std; static set<string> dumped_matrix_names; string name(matrix_name); string::size_type pos = name.rfind('/'); if (pos != string::npos) { name = name.substr(pos + 1); } if (dumped_matrix_names.find(name) == dumped_matrix_names.end()) { dumped_matrix_names.insert(name); FILE* fp = fopen((matrix_name + ".mtx").c_str(), "w"); if (!fp) { return; } if (is_integral<T>::value) { fprintf(fp, "%%%%MatrixMarket matrix array integer general\n"); } else { fprintf(fp, "%%%%MatrixMarket matrix array real general\n"); } fprintf(fp, "%d %d\n", m, n); // matrix market array format uses column-major order for (const auto j : c10::irange(n)) { for (const auto i : c10::irange(m)) { if (is_integral<T>::value) { // NOLINTNEXTLINE(clang-analyzer-core.NullDereference) fprintf(fp, "%d\n", static_cast<int>(a[j * m + i])); } else { fprintf(fp, "%f\n", static_cast<float>(a[j * m + i])); } } } fclose(fp); } } } // namespace caffe2
1,335
23.290909
69
h
null
pytorch-main/caffe2/quantization/server/op_wrapper.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/core/tensor_int8.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" namespace caffe2 { /** * Wrap a floating-point operator with quantized inputs with type T. * This class is to measure quantization error against fp32 reference. */ template <typename OpType, typename T> class OpWrapper { public: OpWrapper(OperatorBase* op, dnnlowp::QuantizationFactory* qfactory) : op_(op), qfactory_(qfactory) { for (auto name : op->debug_def().input()) { local_input_blobs_.push_back(local_ws_.CreateBlob(name)); TORCH_CHECK_NOTNULL(local_input_blobs_.back()); } OperatorDef def = op->debug_def(); local_op_.reset(new OpType(def, &local_ws_)); for (auto name : def.output()) { local_output_blobs_.push_back(local_ws_.GetBlob(name)); TORCH_CHECK_NOTNULL(local_output_blobs_.back()); } } void DequantizeInput() { const OperatorDef& def = op_->debug_def(); CPUContext context(def.device_option()); for (int i = 0; i < op_->InputSize(); ++i) { if (op_->InputIsType<int8::Int8TensorCPU>(i)) { const TensorCPU& qtensor = op_->Input<int8::Int8TensorCPU>(i).t; TensorCPU* float_tensor = BlobGetMutableTensor(local_input_blobs_[i], CPU); // FIXME: doesn't work for bias so we shouldn't quantize bias before // model loading when we're running a shadow operator in fp32 for // example for measuring quantization error. float_tensor->ResizeLike(qtensor); fbgemm::Dequantize<T>( qtensor.data<T>(), float_tensor->template mutable_data<float>(), qtensor.numel(), dnnlowp::GetInputTensorQuantizationParamsOf(op_, i, qfactory_)); } else { local_input_blobs_[i]->ShareExternal( const_cast<void*>(op_->Inputs()[i]->GetRaw()), op_->Inputs()[i]->meta()); } } } OpType* Get() { return local_op_.get(); } dnnlowp::TensorQuantizationParams GetOutputQuantizationParams( dnnlowp::QuantizationFactory* qfactory, int index = 0) { using namespace dnnlowp; float min, max; auto& out_tensor = local_output_blobs_[index]->template Get<TensorCPU>(); fbgemm::FindMinMax( out_tensor.template data<float>(), &min, &max, out_tensor.numel()); if (op_->OperatorBase::GetSingleArgument<std::string>("followed_by", "") == "Relu") { min = std::max(0.0f, min); max = std::max(0.0f, max); } return qfactory->ChooseQuantizationParams(min, max); } private: OperatorBase* op_; /* container quantized op */ Workspace local_ws_; std::vector<Blob*> local_input_blobs_; std::vector<Blob*> local_output_blobs_; std::unique_ptr<OpType> local_op_; /* contained fp32 reference op */ dnnlowp::QuantizationFactory* qfactory_; }; } // namespace caffe2
2,959
32.258427
79
h
null
pytorch-main/caffe2/quantization/server/quantization_error_minimization.h
#pragma once #include "caffe2/quantization/server/dnnlowp.h" namespace dnnlowp { class QuantizationErrorMinimization { public: virtual TensorQuantizationParams ChooseQuantizationParams( const Histogram& hist, bool preserve_sparsity = false, int precision = 8) = 0; virtual ~QuantizationErrorMinimization(){}; }; class NormMinimization : public QuantizationErrorMinimization { public: enum Kind { L1, L2, }; NormMinimization(Kind kind) : kind_(kind) {} /** * Faster approximate search */ TensorQuantizationParams NonlinearQuantizationParamsSearch( const Histogram& hist, bool preserve_sparsity = false, int precision = 8); TensorQuantizationParams ChooseQuantizationParams( const Histogram& hist, bool preserve_sparsity = false, int precision = 8) override; protected: Kind kind_; }; class L1ErrorMinimization : public NormMinimization { public: L1ErrorMinimization() : NormMinimization(L1) {} }; class P99 : public QuantizationErrorMinimization { public: float threshold_; P99(float p99_threshold = 0.99) : threshold_(p99_threshold) {} TensorQuantizationParams ChooseQuantizationParams( const Histogram& hist, bool preserve_sparsity = true, int precision = 8) override; }; // class P99QuantizationFactory } // namespace dnnlowp
1,356
22.396552
64
h
null
pytorch-main/caffe2/quantization/server/quantize_dnnlowp_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" namespace caffe2 { template <typename T> class QuantizeDNNLowPOp final : public Operator<CPUContext> { public: USE_OPERATOR_FUNCTIONS(CPUContext); QuantizeDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; private: std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; bool arguments_parsed_{false}; }; // class QuantizeDNNLowPOp } // namespace caffe2
518
22.590909
68
h
null
pytorch-main/caffe2/quantization/server/relu_dnnlowp_op.h
#pragma once #include "caffe2/operators/relu_op.h" #include "caffe2/core/tensor_int8.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" namespace caffe2 { template <typename T> class ReluDNNLowPOp final : public Operator<CPUContext> { public: USE_OPERATOR_FUNCTIONS(CPUContext); ReluDNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), qfactory_(dnnlowp::GetQuantizationFactoryOf(this)) {} bool RunOnDevice() override; private: std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; }; namespace internal { template <typename T> void ReluAVX2(const int N, const int zero_point, const T* X, T* Y); } // namespace internal } // namespace caffe2
739
22.125
67
h
null
pytorch-main/caffe2/quantization/server/resize_nearest_3d_dnnlowp_op.h
#pragma once #include "caffe2/operators/resize_3d_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { using ResizeNearest3DFP32Op = ResizeNearest3DOp<float, CPUContext>; template <typename T> class ResizeNearest3DDNNLowPOp final : public DNNLowPOp<T, ResizeNearest3DFP32Op> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ResizeNearest3DFP32Op); ResizeNearest3DDNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : BaseType(operator_def, ws), temporal_scale_( this->template GetSingleArgument<float>("temporal_scale", 1)), width_scale_(this->template GetSingleArgument<float>("width_scale", 1)), height_scale_( this->template GetSingleArgument<float>("height_scale", 1)) { CAFFE_ENFORCE_GT(temporal_scale_, 0); CAFFE_ENFORCE_GT(width_scale_, 0); CAFFE_ENFORCE_GT(height_scale_, 0); const auto& order = StringToStorageOrder( this->template GetSingleArgument<std::string>("order", "NHWC")); CAFFE_ENFORCE_EQ(order, StorageOrder::NHWC); } bool RunOnDevice() override; private: float temporal_scale_; float width_scale_; float height_scale_; }; } // namespace caffe2
1,243
28.619048
80
h
null
pytorch-main/caffe2/quantization/server/resize_nearest_dnnlowp_op.h
#pragma once #include "caffe2/operators/resize_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { using ResizeNearestFP32Op = ResizeNearestOp<float, CPUContext>; template <typename T> class ResizeNearestDNNLowPOp final : public DNNLowPOp<T, ResizeNearestFP32Op> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, ResizeNearestFP32Op); ResizeNearestDNNLowPOp(const OperatorDef& operator_def, Workspace* ws) : BaseType(operator_def, ws), width_scale_(this->template GetSingleArgument<float>("width_scale", 1)), height_scale_( this->template GetSingleArgument<float>("height_scale", 1)) { CAFFE_ENFORCE_GT(width_scale_, 0); CAFFE_ENFORCE_GT(height_scale_, 0); const auto& order = StringToStorageOrder( this->template GetSingleArgument<std::string>("order", "NHWC")); CAFFE_ENFORCE_EQ(order, StorageOrder::NHWC); } bool RunOnDevice() override; private: float width_scale_; float height_scale_; }; } // namespace caffe2
1,057
27.594595
80
h
null
pytorch-main/caffe2/quantization/server/sigmoid.h
#pragma once #include "caffe2/quantization/server/tanh.h" namespace dnnlowp { /** * sigmoid(x) = (tanh(x/2) + 1)/2 * Quantized sigmoid is computed as tanh under the hood, we just use different * input/output quantization parameters. */ template <typename T> class Sigmoid { public: Sigmoid(double max_abs_err_ = Tanh<T>::DEFAULT_MAX_ABS_ERR); T Compute(T x) const; TensorQuantizationParams GetInputQuantizationParams() const { return in_qparams_; } TensorQuantizationParams GetOutputQuantizationParams() const { return out_qparams_; } private: const int num_in_bits_ = Tanh<T>::DEFAULT_NUM_IN_BITS; const int num_out_bits_ = Tanh<T>::DEFAULT_NUM_OUT_BITS; Tanh<T> tanh_; TensorQuantizationParams in_qparams_, out_qparams_; }; // class Sigmoid } // namespace dnnlowp
807
22.764706
78
h
null
pytorch-main/caffe2/quantization/server/spatial_batch_norm_dnnlowp_op.h
#pragma once #include "caffe2/operators/spatial_batch_norm_op.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { /** * Note this implementation assumes SCALE, BIAS, EST_MEAN, and EST_VAR inputs * are still in fp32, so is epsilon argument */ template <typename T, bool ReluFused = false> class SpatialBNDNNLowPOp final : public DNNLowPOp<T, SpatialBNOp<CPUContext>> { public: USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, SpatialBNOp<CPUContext>); SpatialBNDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); ~SpatialBNDNNLowPOp() override = default; bool RunOnDevice() override; private: void ComputeFusedParam_( const int C, const float* scale, const float* bias, const float* mean, const float* var, float* alpha, float* beta); double epsilon_; const StorageOrder order_; Tensor alpha_; Tensor beta_; INPUT_TAGS(INPUT, SCALE, BIAS, EST_MEAN, EST_VAR); OUTPUT_TAGS(OUTPUT); }; namespace internal { template <typename T> void SpatialBNNHWCAVX2( const int N, const int C, const int HxW, const int in_zero_point, const int out_zero_point, const T* X, const float* alpha, const float* beta, T* Y, bool relu_fused); } // namespace internal } // namespace caffe2
1,340
21.35
79
h
null
pytorch-main/caffe2/quantization/server/tanh.h
#pragma once #include "caffe2/quantization/server/dnnlowp.h" #include <cmath> #include <vector> namespace dnnlowp { /** * We use the 3-region approach described in "Efficient VLSI Implementation of * Neural Networks with Hyperbolic Tangent Activation Function", IEEE * Transactions on Very Large Scale Integration Systems, Zamanlooy and * Mirhassani. * The pass region (x < x_pq) is approximated as x. * The saturation region (x >= x_sq) is approximated as 1. * The processing region (x_pq <= x < x_sq) is divided into sub-ranges and the * average value of tanh(x) is used per sub-range. */ template <typename T> class Tanh { public: Tanh(double max_abs_err = DEFAULT_MAX_ABS_ERR); T Compute(T x) const; TensorQuantizationParams GetInputQuantizationParams() const { return in_qparams_; } TensorQuantizationParams GetOutputQuantizationParams() const { return out_qparams_; } int GetPassRegionEnd() const { return x_pq_index_; } float GetPassRegionEndDequantized() const { return fbgemm::Dequantize<T>( static_cast<uint8_t>(x_pq_index_ + in_qparams_.zero_point), in_qparams_); } float GetSaturationRegionBegin() const { return fbgemm::Dequantize<T>( static_cast<T>((1 << num_in_bits_) - 1), in_qparams_); } static constexpr double DEFAULT_MAX_ABS_ERR = 0.02; static constexpr int DEFAULT_NUM_IN_BITS = 8; static constexpr int DEFAULT_NUM_OUT_BITS = 8; private: const double max_abs_err_; const int num_in_bits_ = DEFAULT_NUM_IN_BITS; const int num_out_bits_ = DEFAULT_NUM_OUT_BITS; int x_pq_index_; std::vector<T> processing_region_lut_; TensorQuantizationParams in_qparams_, out_qparams_; }; // class TanhApproximation } // namespace dnnlowp
1,751
26.375
78
h
null
pytorch-main/caffe2/quantization/server/utility_dnnlowp_ops.h
#pragma once #include "caffe2/operators/utility_ops.h" #include "caffe2/quantization/server/caffe2_dnnlowp_utils.h" #include "caffe2/quantization/server/dnnlowp.h" #include "caffe2/quantization/server/dnnlowp_op.h" namespace caffe2 { template <typename T, bool ReluFused = false> class SumDNNLowPOp final : public DNNLowPOp<T, SumOp<CPUContext>> { public: SumDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); bool RunOnDevice() override; USE_OPERATOR_FUNCTIONS(CPUContext); USE_DNNLOWP_OPERATOR_BASE_FUNCTIONS(T, SumOp<CPUContext>); private: bool GetQuantizationParameters_(); dnnlowp::TensorQuantizationParams intermediate_qparams_; dnnlowp::RequantizationParams out_requantization_params_; }; // class SumDNNLowPOp template <typename T> class GatherDNNLowPOp final : public GatherOp<CPUContext> { static_assert(std::is_integral<T>::value, "Integral required."); public: GatherDNNLowPOp(const OperatorDef& operator_def, Workspace* ws); ~GatherDNNLowPOp() override; bool RunOnDevice() override; template <typename Index> bool DoRunWithType() { // If we endup using it on GPU doing O(N) memcpy is probably not best :) // TODO: implement prefetching if it starts mattering (TF does it) auto& data = (this->template Input<int8::Int8TensorCPU>(DATA)).t; auto& indices = Input(INDICES); auto* output = &Outputs()[0]->template GetMutable<int8::Int8TensorCPU>()->t; CAFFE_ENFORCE_GE(data.ndim(), 1, "DATA should be at least 1-D"); auto shape = indices.sizes().vec(); shape.insert(shape.end(), data.sizes().begin() + 1, data.sizes().end()); output->Resize(shape); int block_size = data.size_from_dim(1); auto block_bytesize = data.size_from_dim(1) * data.dtype().itemsize(); int N = indices.numel(); auto src_base = static_cast<const char*>(data.raw_data()); const Index* idxs = indices.template data<Index>(); auto out = static_cast<char*>(output->raw_mutable_data(data.dtype())); for (const auto i : c10::irange(N)) { auto idx = idxs[i]; CAFFE_ENFORCE( 0 <= idx && idx < data.size(0), "INDICES element is out of DATA bounds, id=", idx, " data_dim=", data.size(0)); auto src = src_base + idx * block_bytesize; context_.CopyItemsSameDevice( data.dtype(), block_size, src, out + block_bytesize * i); } return true; } USE_OPERATOR_FUNCTIONS(CPUContext); private: OpWrapper<GatherOp<CPUContext>, T>* Fp32Op_() { if (!fp32_op_) { fp32_op_.reset( new OpWrapper<GatherOp<CPUContext>, T>(this, qfactory_.get())); } return fp32_op_.get(); } std::unique_ptr<OpWrapper<GatherOp<CPUContext>, T>> fp32_op_; bool dequantize_output_{false}, measure_quantization_error_{false}; std::unique_ptr<dnnlowp::QuantizationFactory> qfactory_; dnnlowp::QuantizationErrorStats quantization_error_stats_; bool arguments_parsed_{false}; }; // class GatherDNNLowPOp namespace internal { template <typename T, bool ReluFused> void ElementWiseSumAVX2( const T* input0, const T* input1, T* output, int len, float a_scale, int32_t a_zero_point, float b_scale, int32_t b_zero_point, float c_scale, int32_t c_zero_points); } } // namespace caffe2
3,310
28.828829
80
h
null
pytorch-main/caffe2/queue/blobs_queue.h
#pragma once #include <atomic> #include <condition_variable> #include <memory> #include <mutex> #include <queue> #include "caffe2/core/blob_stats.h" #include "caffe2/core/logging.h" #include "caffe2/core/stats.h" #include "caffe2/core/tensor.h" #include "caffe2/core/workspace.h" namespace caffe2 { // A thread-safe, bounded, blocking queue. // Modelled as a circular buffer. // Containing blobs are owned by the workspace. // On read, we swap out the underlying data for the blob passed in for blobs class TORCH_API BlobsQueue : public std::enable_shared_from_this<BlobsQueue> { public: BlobsQueue( Workspace* ws, const std::string& queueName, size_t capacity, size_t numBlobs, bool enforceUniqueName, const std::vector<std::string>& fieldNames = {}); ~BlobsQueue() { close(); } bool blockingRead( const std::vector<Blob*>& inputs, float timeout_secs = 0.0f); bool tryWrite(const std::vector<Blob*>& inputs); bool blockingWrite(const std::vector<Blob*>& inputs); void close(); size_t getNumBlobs() const { return numBlobs_; } private: bool canWrite(); void doWrite(const std::vector<Blob*>& inputs); std::atomic<bool> closing_{false}; size_t numBlobs_; std::mutex mutex_; // protects all variables in the class. std::condition_variable cv_; int64_t reader_{0}; int64_t writer_{0}; std::vector<std::vector<Blob*>> queue_; const std::string name_; struct QueueStats { CAFFE_STAT_CTOR(QueueStats); CAFFE_EXPORTED_STAT(queue_balance); CAFFE_EXPORTED_STAT(queue_dequeued_records); CAFFE_DETAILED_EXPORTED_STAT(queue_dequeued_bytes); CAFFE_AVG_EXPORTED_STAT(read_time_ns); CAFFE_AVG_EXPORTED_STAT(write_time_ns); } stats_; }; } // namespace caffe2
1,779
24.070423
78
h
null
pytorch-main/caffe2/queue/blobs_queue_db.h
#pragma once #include <chrono> #include <string> #include "caffe2/core/db.h" #include "caffe2/core/logging.h" #include "caffe2/core/stats.h" #include "caffe2/queue/blobs_queue.h" namespace caffe2 { namespace db { namespace { const std::string& GetStringFromBlob(Blob* blob) { if (blob->template IsType<string>()) { return blob->template Get<string>(); } else if (blob->template IsType<Tensor>()) { return *blob->template Get<Tensor>().template data<string>(); } else { CAFFE_THROW("Unsupported Blob type"); } } } // namespace class BlobsQueueDBCursor : public Cursor { public: explicit BlobsQueueDBCursor( std::shared_ptr<BlobsQueue> queue, int key_blob_index, int value_blob_index, float timeout_secs) : queue_(queue), key_blob_index_(key_blob_index), value_blob_index_(value_blob_index), timeout_secs_(timeout_secs), inited_(false), valid_(false) { LOG(INFO) << "BlobsQueueDBCursor constructed"; CAFFE_ENFORCE(queue_ != nullptr, "queue is null"); CAFFE_ENFORCE(value_blob_index_ >= 0, "value_blob_index < 0"); } ~BlobsQueueDBCursor() override {} void Seek(const string& /* unused */) override { CAFFE_THROW("Seek is not supported."); } bool SupportsSeek() override { return false; } void SeekToFirst() override { // not applicable } void Next() override { unique_ptr<Blob> blob = make_unique<Blob>(); vector<Blob*> blob_vector{blob.get()}; auto success = queue_->blockingRead(blob_vector, timeout_secs_); if (!success) { LOG(ERROR) << "Timed out reading from BlobsQueue or it is closed"; valid_ = false; return; } if (key_blob_index_ >= 0) { key_ = GetStringFromBlob(blob_vector[key_blob_index_]); } value_ = GetStringFromBlob(blob_vector[value_blob_index_]); valid_ = true; } string key() override { if (!inited_) { Next(); inited_ = true; } return key_; } string value() override { if (!inited_) { Next(); inited_ = true; } return value_; } bool Valid() override { return valid_; } private: std::shared_ptr<BlobsQueue> queue_; int key_blob_index_; int value_blob_index_; float timeout_secs_; bool inited_; string key_; string value_; bool valid_; }; class BlobsQueueDB : public DB { public: BlobsQueueDB( const string& source, Mode mode, std::shared_ptr<BlobsQueue> queue, int key_blob_index = -1, int value_blob_index = 0, float timeout_secs = 0.0) : DB(source, mode), queue_(queue), key_blob_index_(key_blob_index), value_blob_index_(value_blob_index), timeout_secs_(timeout_secs) { LOG(INFO) << "BlobsQueueDB constructed"; } ~BlobsQueueDB() override { Close(); } void Close() override {} unique_ptr<Cursor> NewCursor() override { return make_unique<BlobsQueueDBCursor>( queue_, key_blob_index_, value_blob_index_, timeout_secs_); } unique_ptr<Transaction> NewTransaction() override { CAFFE_THROW("Not implemented."); } private: std::shared_ptr<BlobsQueue> queue_; int key_blob_index_; int value_blob_index_; float timeout_secs_; }; } // namespace db } // namespace caffe2
3,304
21.636986
72
h
null
pytorch-main/caffe2/queue/queue_ops.h
#pragma once #include <memory> #include "blobs_queue.h" #include "caffe2/core/operator.h" #include "caffe2/utils/math.h" #include <c10/util/irange.h> namespace caffe2 { template <typename Context> class CreateBlobsQueueOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; CreateBlobsQueueOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), ws_(ws), name(operator_def.output().Get(0)) {} bool RunOnDevice() override { const auto capacity = GetSingleArgument("capacity", 1); const auto numBlobs = GetSingleArgument("num_blobs", 1); const auto enforceUniqueName = GetSingleArgument("enforce_unique_name", false); const auto fieldNames = OperatorBase::template GetRepeatedArgument<std::string>("field_names"); CAFFE_ENFORCE_EQ(this->OutputSize(), 1); auto queuePtr = Operator<Context>::Outputs()[0] ->template GetMutable<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE(queuePtr); *queuePtr = std::make_shared<BlobsQueue>( ws_, name, capacity, numBlobs, enforceUniqueName, fieldNames); return true; } private: Workspace* ws_{nullptr}; const std::string name; }; template <typename Context> class EnqueueBlobsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; using Operator<Context>::Operator; bool RunOnDevice() override { CAFFE_ENFORCE(InputSize() > 1); auto queue = Operator<Context>::Inputs()[0] ->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE( queue && static_cast<size_t>(OutputSize()) == queue->getNumBlobs()); return queue->blockingWrite(this->Outputs()); } private: }; template <typename Context> class DequeueBlobsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; DequeueBlobsOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) { timeout_secs_ = OperatorBase::GetSingleArgument<float>("timeout_secs", 0); } bool RunOnDevice() override { CAFFE_ENFORCE(InputSize() == 1); auto queue = OperatorBase::Inputs()[0]->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE( queue && static_cast<size_t>(OutputSize()) == queue->getNumBlobs()); return queue->blockingRead(this->Outputs(), timeout_secs_); } private: float timeout_secs_; }; template <typename Context> class CloseBlobsQueueOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; using Operator<Context>::Operator; bool RunOnDevice() override { CAFFE_ENFORCE_EQ(InputSize(), 1); auto queue = OperatorBase::Inputs()[0]->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE(queue); queue->close(); return true; } private: }; template <typename Context> class SafeEnqueueBlobsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; using Operator<Context>::Operator; bool RunOnDevice() override { auto queue = Operator<Context>::Inputs()[0] ->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE(queue); auto size = queue->getNumBlobs(); CAFFE_ENFORCE( static_cast<size_t>(OutputSize()) == size + 1, "Expected " + c10::to_string(size + 1) + ", " + " got: " + c10::to_string(size)); bool status = queue->blockingWrite(this->Outputs()); Output(size)->Resize(); math::Set<bool, Context>( 1, !status, Output(size)->template mutable_data<bool>(), &context_); return true; } void Cancel() override { auto queue = Operator<Context>::Inputs()[0] ->template Get<std::shared_ptr<BlobsQueue>>(); queue->close(); } }; template <typename Context> class SafeDequeueBlobsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; using Operator<Context>::Operator; SafeDequeueBlobsOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), numRecords_(OperatorBase::GetSingleArgument<int>("num_records", 1)) { CAFFE_ENFORCE_GT(numRecords_, 0); } bool dequeueMany(std::shared_ptr<BlobsQueue>& queue) { auto size = queue->getNumBlobs(); if (blobs_.size() != size) { blobs_.resize(size); blobPtrs_.resize(size); for (auto col : c10::irange(size)) { blobPtrs_.at(col) = &blobs_.at(col); } } const int kTensorGrowthPct = 40; for (const auto i : c10::irange(numRecords_)) { if (!queue->blockingRead(blobPtrs_)) { // if we read at least one record, status is still true return i > 0; } for (auto col : c10::irange(size)) { auto* out = this->Output(col); const auto& in = blobPtrs_.at(col)->template Get<Tensor>(); if (i == 0) { out->CopyFrom(in); } else { auto oldSize = out->numel(); CAFFE_ENFORCE( in.dim() > 0, "Empty tensor to dequeue at column ", col, " within ", size, " total columns"); out->Extend(in.sizes()[0], kTensorGrowthPct); auto* dst = (char*)out->raw_mutable_data() + oldSize * in.dtype().itemsize(); context_.template CopyItems<Context, Context>( in.meta(), in.numel(), in.raw_data(), dst); } } } return true; } bool dequeueOne(std::shared_ptr<BlobsQueue>& queue) { return queue->blockingRead(this->Outputs()); } bool RunOnDevice() override { CAFFE_ENFORCE(InputSize() == 1); auto queue = Operator<Context>::Inputs()[0] ->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE(queue); auto size = queue->getNumBlobs(); CAFFE_ENFORCE_EQ(OutputSize(), size + 1); bool status = numRecords_ > 1 ? dequeueMany(queue) : dequeueOne(queue); Output(size)->Resize(); math::Set<bool, Context>( 1, !status, Output(size)->template mutable_data<bool>(), &context_); return true; } void Cancel() override { auto queue = Operator<Context>::Inputs()[0] ->template Get<std::shared_ptr<BlobsQueue>>(); queue->close(); } private: int numRecords_; std::vector<Blob> blobs_; std::vector<Blob*> blobPtrs_; }; template <typename Context> class WeightedSampleDequeueBlobsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; WeightedSampleDequeueBlobsOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), table_idx_blob_( OperatorBase::GetSingleArgument<int>("table_idx_blob", -1)) { CAFFE_ENFORCE_LT(table_idx_blob_, OutputSize() - 1); vector<float> weights = OperatorBase::GetRepeatedArgument<float>("weights"); if (weights.empty()) { weights.resize(InputSize(), 1.0f); } CAFFE_ENFORCE_EQ(InputSize(), weights.size()); float sum = accumulate(weights.begin(), weights.end(), 0.0f); CAFFE_ENFORCE(sum > 0.0f, "Sum of weights must be positive"); cumProbs_.resize(weights.size()); for (auto i : c10::irange(weights.size())) { cumProbs_[i] = weights[i] / sum; CAFFE_ENFORCE_GE( cumProbs_[i], 0.0f, "Each probability must be non-negative"); } std::partial_sum(cumProbs_.begin(), cumProbs_.end(), cumProbs_.begin()); // Put last value to be 1.0001 to avoid numerical issues. cumProbs_.back() = 1.0001f; LOG(INFO) << "Dequeue weights: " << weights; LOG(INFO) << "cumProbs: " << cumProbs_; } bool RunOnDevice() override { float r; math::RandUniform<float, Context>(1, 0.0f, 1.0f, &r, &context_); auto lb = lower_bound(cumProbs_.begin(), cumProbs_.end(), r); CAFFE_ENFORCE(lb != cumProbs_.end(), "Cannot find ", r, " in cumProbs_."); const int32_t idx = lb - cumProbs_.begin(); auto queue = Operator<Context>::Inputs()[idx] ->template Get<std::shared_ptr<BlobsQueue>>(); CAFFE_ENFORCE(queue); auto size = queue->getNumBlobs(); CAFFE_ENFORCE_EQ(OutputSize(), size + 1); bool status = queue->blockingRead(this->Outputs()); if (table_idx_blob_ >= 0) { auto* table_idx_blob_out = Output(table_idx_blob_, {1}, at::dtype<int32_t>()); int32_t* data = table_idx_blob_out->template mutable_data<int32_t>(); data[0] = idx; } Output(size)->Resize(); math::Set<bool, Context>( 1, !status, Output(size)->template mutable_data<bool>(), &context_); return true; } private: vector<float> cumProbs_; int table_idx_blob_; }; } // namespace caffe2
8,735
30.2
80
h
null
pytorch-main/caffe2/queue/rebatching_queue_ops.h
#pragma once #include "rebatching_queue.h" #include "c10/util/irange.h" namespace caffe2 { using RebatchingQueuePtr = std::unique_ptr<RebatchingQueue>; class CreateRebatchingQueueOp : public Operator<CPUContext> { public: CreateRebatchingQueueOp(const OperatorDef& operator_def, Workspace* ws) : Operator(operator_def, ws) {} bool RunOnDevice() override { *OperatorBase::Output<RebatchingQueuePtr>(0) = RebatchingQueuePtr(new RebatchingQueue( OperatorBase::GetSingleArgument<int>("capacity", 1), OperatorBase::GetSingleArgument<int>("num_blobs", 1))); return true; } }; class EnqueueRebatchingQueueOp : public Operator<CPUContext> { public: EnqueueRebatchingQueueOp(const OperatorDef& operator_def, Workspace* ws) : Operator(operator_def, ws), enqueueBatch_( OperatorBase::GetSingleArgument<bool>("enqueue_batch", false)) {} bool RunOnDevice() override { auto& queue = Inputs()[0]->template Get<RebatchingQueuePtr>(); CHECK(queue); CAFFE_ENFORCE_EQ(InputSize(), queue->numBlobs() + 1); std::vector<const Tensor*> inputTensors; inputTensors.reserve(InputSize() - 1); for (const auto i : c10::irange(1, InputSize())) { inputTensors.push_back(&Input(i)); } return enqueueBatch_ ? queue->enqueueMany(context_, inputTensors) : queue->enqueueOne(context_, inputTensors); } private: const bool enqueueBatch_; }; class DequeueRebatchingQueueOp : public Operator<CPUContext> { public: DequeueRebatchingQueueOp(const OperatorDef& operator_def, Workspace* ws) : Operator(operator_def, ws), numElements_(OperatorBase::GetSingleArgument<int>("num_elements", 1)) {} bool RunOnDevice() override { auto& queue = Inputs()[0]->template Get<RebatchingQueuePtr>(); CHECK(queue); std::vector<Tensor*> outputTensors; outputTensors.reserve(OutputSize()); for (const auto i : c10::irange(OutputSize())) { outputTensors.push_back(Output(i)); } return queue->dequeue(context_, numElements_, outputTensors); } private: int numElements_; }; class CloseRebatchingQueueOp : public Operator<CPUContext> { public: CloseRebatchingQueueOp(const OperatorDef& operator_def, Workspace* ws) : Operator(operator_def, ws) {} bool RunOnDevice() override { CAFFE_ENFORCE_EQ(InputSize(), 1); auto& queue = Inputs()[0]->template Get<RebatchingQueuePtr>(); CAFFE_ENFORCE(queue); queue->close(); return true; } }; } // namespace caffe2
2,549
28.651163
80
h
null
pytorch-main/caffe2/serialize/file_adapter.h
#pragma once #include <fstream> #include <memory> #include <c10/macros/Macros.h> #include "caffe2/serialize/istream_adapter.h" #include "caffe2/serialize/read_adapter_interface.h" namespace caffe2 { namespace serialize { class TORCH_API FileAdapter final : public ReadAdapterInterface { public: C10_DISABLE_COPY_AND_ASSIGN(FileAdapter); explicit FileAdapter(const std::string& file_name); size_t size() const override; size_t read(uint64_t pos, void* buf, size_t n, const char* what = "") const override; ~FileAdapter() override; private: // An RAII Wrapper for a FILE pointer. Closes on destruction. struct RAIIFile { FILE* fp_; explicit RAIIFile(const std::string& file_name); ~RAIIFile(); }; RAIIFile file_; // The size of the opened file in bytes uint64_t size_; }; } // namespace serialize } // namespace caffe2
866
22.432432
71
h
null
pytorch-main/caffe2/serialize/inline_container.h
#pragma once #include <cerrno> #include <cstdio> #include <cstring> #include <fstream> #include <istream> #include <mutex> #include <ostream> #include <unordered_set> #include <c10/core/Allocator.h> #include <c10/core/Backend.h> #include "caffe2/serialize/istream_adapter.h" #include "caffe2/serialize/read_adapter_interface.h" #include "caffe2/serialize/versions.h" extern "C" { typedef struct mz_zip_archive mz_zip_archive; } // PyTorch containers are a special zip archive with the following layout // archive_name.zip contains: // archive_name/ // version # a file with a single decimal number written in ascii, // # used to establish the version of the archive format // model.json # overall model description, this is a json output of // # ModelDef from torch.proto // # the following names are by convention only, model.json will // # refer to these files by full names // tensors/ // 0 # flat storage for tensor data, meta-data about shapes, etc. is // # in model.json // 1 // ... // # code entries will only exist for modules that have methods attached // code/ // archive_name.py # serialized torch script code (python syntax, using // PythonPrint) archive_name_my_submodule.py # submodules have separate // files // // The PyTorchStreamWriter also ensures additional useful properties for these // files // 1. All files are stored uncompressed. // 2. All files in the archive are aligned to 64 byte boundaries such that // it is possible to mmap the entire file and get an aligned pointer to // tensor data. // 3. We universally write in ZIP64 format for consistency. // The PyTorchStreamReader also provides additional properties: // 1. It can read zip files that are created with common // zip tools. This means that even though our writer doesn't compress files, // the reader can still read files that were compressed. // 2. It provides a getRecordOffset function which returns the offset into the // raw file where file data lives. If the file was written with // PyTorchStreamWriter it is guaranteed to be 64 byte aligned. // PyTorchReader/Writer handle checking the version number on the archive format // and ensure that all files are written to a archive_name directory so they // unzip cleanly. // When developing this format we want to pay particular attention to the // following use cases: // // -- Reading -- // 1) Reading with full random access // a) Reading with file api's such as fread() // b) mmaping the file and jumping around the mapped region // 2) Reading with 1-pass sequential access // -> A reader will need to build up a data structure of parsed structures // as it reads // // -- Writing -- // 1) Writing with full random access // 2) Writing with 1-pass sequential access // -> We must take care not to require updating values that have already // been written. We place the variable-length index at the end and do // not put any indicies into the header to fulfill this constraint. // The model.json, which contains all the metadata information, // should be written as the last file. One reason is that the size of tensor // data is usually stable. As long as the shape and type of the tensor do not // change, the size of the data won't change. On the other sied, the size of the // serialized model is likely to change, so we store it as the last record, and // we don't need to move previous records when updating the model data. // The zip format is sufficiently flexible to handle the above use-case. // it puts its central directory at the end of the archive and we write // model.json as the last file when writing after we have accumulated all // other information. namespace caffe2 { namespace serialize { static constexpr const char* kSerializationIdRecordName = ".data/serialization_id"; class TORCH_API PyTorchStreamReader final { public: explicit PyTorchStreamReader(const std::string& file_name); explicit PyTorchStreamReader(std::istream* in); explicit PyTorchStreamReader(std::shared_ptr<ReadAdapterInterface> in); // return dataptr, size std::tuple<at::DataPtr, size_t> getRecord(const std::string& name); // inplace memory writing size_t getRecord(const std::string& name, void* dst, size_t n); size_t getRecord( const std::string& name, void* dst, size_t n, size_t chunk_size, const std::function<void(void*, const void*, size_t)>& memcpy_func); size_t getRecordOffset(const std::string& name); bool hasRecord(const std::string& name); std::vector<std::string> getAllRecords(); ~PyTorchStreamReader(); uint64_t version() const { return version_; } const std::string& serializationId() { return serialization_id_; } void setShouldLoadDebugSymbol(bool should_load_debug_symbol) { load_debug_symbol_ = should_load_debug_symbol; } private: void init(); size_t read(uint64_t pos, char* buf, size_t n); void valid(const char* what, const char* info = ""); size_t getRecordID(const std::string& name); friend size_t istream_read_func(void* pOpaque, uint64_t file_ofs, void* pBuf, size_t n); std::unique_ptr<mz_zip_archive> ar_; std::string archive_name_; std::string archive_name_plus_slash_; std::shared_ptr<ReadAdapterInterface> in_; int64_t version_; std::mutex reader_lock_; bool load_debug_symbol_ = true; std::string serialization_id_; }; class TORCH_API PyTorchStreamWriter final { public: explicit PyTorchStreamWriter(const std::string& archive_name); explicit PyTorchStreamWriter( const std::function<size_t(const void*, size_t)> writer_func); void setMinVersion(const uint64_t version); void writeRecord( const std::string& name, const void* data, size_t size, bool compress = false); void writeEndOfFile(); const std::unordered_set<std::string>& getAllWrittenRecords(); bool finalized() const { return finalized_; } const std::string& archiveName() { return archive_name_; } const std::string& serializationId() { return serialization_id_; } ~PyTorchStreamWriter(); private: void setup(const std::string& file_name); void valid(const char* what, const char* info = ""); void writeSerializationId(); size_t current_pos_ = 0; std::unordered_set<std::string> files_written_; std::unique_ptr<mz_zip_archive> ar_; std::string archive_name_; std::string archive_name_plus_slash_; std::string padding_; std::ofstream file_stream_; std::function<size_t(const void*, size_t)> writer_func_; uint64_t combined_uncomp_crc32_ = 0; std::string serialization_id_; // This number will be updated when the model has operators // that have valid upgraders. uint64_t version_ = kMinProducedFileFormatVersion; bool finalized_ = false; bool err_seen_ = false; friend size_t ostream_write_func( void* pOpaque, uint64_t file_ofs, const void* pBuf, size_t n); }; namespace detail { // Writer-specific constants constexpr uint64_t kFieldAlignment = 64; // Returns a record to be appended to the local user extra data entry in order // to make data beginning aligned at kFieldAlignment bytes boundary. size_t getPadding( size_t cursor, size_t filename_size, size_t size, std::string& padding_buf); } // namespace detail } // namespace serialize } // namespace caffe2
7,501
32.945701
83
h
null
pytorch-main/caffe2/serialize/istream_adapter.h
#pragma once #include <istream> #include "c10/macros/Macros.h" #include "caffe2/serialize/read_adapter_interface.h" namespace caffe2 { namespace serialize { // this is a reader implemented by std::istream class TORCH_API IStreamAdapter final : public ReadAdapterInterface { public: C10_DISABLE_COPY_AND_ASSIGN(IStreamAdapter); explicit IStreamAdapter(std::istream* istream); size_t size() const override; size_t read(uint64_t pos, void* buf, size_t n, const char* what = "") const override; ~IStreamAdapter() override; private: std::istream* istream_; void validate(const char* what) const; }; } // namespace serialize } // namespace caffe2
669
22.928571
71
h
null
pytorch-main/caffe2/serialize/read_adapter_interface.h
#pragma once #include <cstddef> #include <cstdint> #include "c10/macros/Macros.h" namespace caffe2 { namespace serialize { // this is the interface for the (file/stream/memory) reader in // PyTorchStreamReader. with this interface, we can extend the support // besides standard istream class TORCH_API ReadAdapterInterface { public: virtual size_t size() const = 0; virtual size_t read(uint64_t pos, void* buf, size_t n, const char* what = "") const = 0; virtual ~ReadAdapterInterface(); }; } // namespace serialize } // namespace caffe2
556
22.208333
79
h
null
pytorch-main/caffe2/serialize/versions.h
#pragma once #include <cstdint> namespace caffe2 { namespace serialize { constexpr uint64_t kMinSupportedFileFormatVersion = 0x1L; constexpr uint64_t kMaxSupportedFileFormatVersion = 0xAL; // Versions (i.e. why was the version number bumped?) // Note [Dynamic Versions and torch.jit.save vs. torch.save] // // Our versioning scheme has a "produced file format version" which // describes how an archive is to be read. The version written in an archive // is at least this current produced file format version, but may be greater // if it includes certain symbols. We refer to these conditional versions // as "dynamic," since they are identified at runtime. // // Dynamic versioning is useful when an operator's semantics are updated. // When using torch.jit.save we want those semantics to be preserved. If // we bumped the produced file format version on every change, however, // then older versions of PyTorch couldn't read even simple archives, like // a single tensor, from newer versions of PyTorch. Instead, we // assign dynamic versions to these changes that override the // produced file format version as needed. That is, when the semantics // of torch.div changed it was assigned dynamic version 4, and when // torch.jit.saving modules that use torch.div those archives also have // (at least) version 4. This prevents earlier versions of PyTorch // from accidentally performing the wrong kind of division. Modules // that don't use torch.div or other operators with dynamic versions // can write the produced file format version, and these programs will // run as expected on earlier versions of PyTorch. // // While torch.jit.save attempts to preserve operator semantics, // torch.save does not. torch.save is analogous to pickling Python, so // a function that uses torch.div will have different behavior if torch.saved // and torch.loaded across PyTorch versions. From a technical perspective, // torch.save ignores dynamic versioning. // 1. Initial version // 2. Removed op_version_set version numbers // 3. Added type tags to pickle serialization of container types // 4. (Dynamic) Stopped integer division using torch.div // (a versioned symbol preserves the historic behavior of versions 1--3) // 5. (Dynamic) Stops torch.full inferring a floating point dtype // when given bool or integer fill values. // 6. Write version string to `./data/version` instead of `version`. // [12/15/2021] // kProducedFileFormatVersion is set to 7 from 3 due to a different // interpretation of what file format version is. // Whenever there is new upgrader introduced, // this number should be bumped. // The reasons that version is bumped in the past: // 1. aten::div is changed at version 4 // 2. aten::full is changed at version 5 // 3. torch.package uses version 6 // 4. Introduce new upgrader design and set the version number to 7 // mark this change // -------------------------------------------------- // We describe new operator version bump reasons here: // 1) [01/24/2022] // We bump the version number to 8 to update aten::linspace // and aten::linspace.out to error out when steps is not // provided. (see: https://github.com/pytorch/pytorch/issues/55951) // 2) [01/30/2022] // Bump the version number to 9 to update aten::logspace and // and aten::logspace.out to error out when steps is not // provided. (see: https://github.com/pytorch/pytorch/issues/55951) // 3) [02/11/2022] // Bump the version number to 10 to update aten::gelu and // and aten::gelu.out to support the new approximate kwarg. // (see: https://github.com/pytorch/pytorch/pull/61439) constexpr uint64_t kProducedFileFormatVersion = 0xAL; // Absolute minimum version we will write packages. This // means that every package from now on will always be // greater than this number. constexpr uint64_t kMinProducedFileFormatVersion = 0x3L; // The version we write when the archive contains bytecode. // It must be higher or eq to kProducedFileFormatVersion. // Because torchscript changes is likely introduce bytecode change. // If kProducedFileFormatVersion is increased, kProducedBytecodeVersion // should be increased too. The relationship is: // kMaxSupportedFileFormatVersion >= (most likely ==) kProducedBytecodeVersion // >= kProducedFileFormatVersion // If a format change is forward compatible (still readable by older // executables), we will not increment the version number, to minimize the // risk of breaking existing clients. TODO: A better way would be to allow // the caller that creates a model to specify a maximum version that its // clients can accept. // Versions: // 0x1L: Initial version // 0x2L: (Comment missing) // 0x3L: (Comment missing) // 0x4L: (update) Added schema to function tuple. Forward-compatible change. // 0x5L: (update) Update bytecode is sharing constant tensor files from // torchscript, and only serialize extra tensors that are not in the // torchscript constant table. Also update tensor storage schema adapting to // the unify format, the root key of tensor storage is updated from {index} to // {the_pointer_value_the_tensor.storage}, for example: // `140245072983168.storage` Forward-compatibility change. // 0x6L: Implicit opereator versioning using number of specified argument. // Refer to the summary of https://github.com/pytorch/pytorch/pull/56845 for // details. // 0x7L: Enable support for operators with default arguments plus out // arguments. Refer. See https://github.com/pytorch/pytorch/pull/63651 for // details. // 0x8L: Emit promoted operators as instructions. See // https://github.com/pytorch/pytorch/pull/71662 for details. // 0x9L: Change serialization format from pickle to format This version is to // serve migration. v8 pickle and v9 flatbuffer are the same. Refer to the // summary of https://github.com/pytorch/pytorch/pull/75201 for more details. constexpr uint64_t kProducedBytecodeVersion = 0x8L; // static_assert( // kProducedBytecodeVersion >= kProducedFileFormatVersion, // "kProducedBytecodeVersion must be higher or equal to // kProducedFileFormatVersion."); // Introduce kMinSupportedBytecodeVersion and kMaxSupportedBytecodeVersion // for limited backward/forward compatibility support of bytecode. If // kMinSupportedBytecodeVersion <= model_version <= kMaxSupportedBytecodeVersion // (in loader), we should support this model_version. For example, we provide a // wrapper to handle an updated operator. constexpr uint64_t kMinSupportedBytecodeVersion = 0x4L; constexpr uint64_t kMaxSupportedBytecodeVersion = 0x9L; } // namespace serialize } // namespace caffe2
6,648
48.619403
80
h
null
pytorch-main/caffe2/sgd/adadelta_op.h
#include "caffe2/core/operator.h" #include "c10/util/irange.h" namespace caffe2 { namespace { template <typename Context> void AdadeltaUpdate( int N, const float* w, const float* g, const float* h, const float* d, const float epsilon, const float decay, const float* lr, float* nw, float* nh, float* nd, Context* /*context*/) { for (const auto i : c10::irange(N)) { float gi = g[i]; float di = d[i]; float hi = nh[i] = decay * h[i] + (1.0f - decay) * gi * gi; float ng = (std::sqrt(di + epsilon) / std::sqrt(hi + epsilon)) * gi; nw[i] = w[i] + lr[0] * ng; nd[i] = decay * di + (1.0f - decay) * ng * ng; } } } // namespace template <class Context> class AdadeltaOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; AdadeltaOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), OP_SINGLE_ARG(float, "epsilon", epsilon_, 1e-5f), OP_SINGLE_ARG(float, "decay", decay_, 0.95f) {} bool RunOnDevice() override { CAFFE_ENFORCE(Input(GRAD).numel() == Input(MOMENT_GRAD).numel()); CAFFE_ENFORCE(Input(GRAD).numel() == Input(MOMENT_DELTA).numel()); CAFFE_ENFORCE(Input(GRAD).numel() == Input(PARAM).numel()); CAFFE_ENFORCE_GE(epsilon_, 0.0f); CAFFE_ENFORCE_GT(decay_, 0.0f); CAFFE_ENFORCE_LT(decay_, 1.0f); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_GRAD)->ResizeLike(Input(MOMENT_GRAD)); Output(OUTPUT_MOMENT_DELTA)->ResizeLike(Input(MOMENT_DELTA)); AdadeltaUpdate<Context>( Input(GRAD).numel(), Input(PARAM).template data<float>(), Input(GRAD).template data<float>(), Input(MOMENT_GRAD).template data<float>(), Input(MOMENT_DELTA).template data<float>(), epsilon_, decay_, Input(LR).template data<float>(), Output(OUTPUT_PARAM)->template mutable_data<float>(), Output(OUTPUT_MOMENT_GRAD)->template mutable_data<float>(), Output(OUTPUT_MOMENT_DELTA)->template mutable_data<float>(), &context_); return true; } protected: const float epsilon_; const float decay_; INPUT_TAGS(PARAM, MOMENT_GRAD, MOMENT_DELTA, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_GRAD, OUTPUT_MOMENT_DELTA); }; template <class Context> class SparseAdadeltaOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; SparseAdadeltaOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), OP_SINGLE_ARG(float, "epsilon", epsilon_, 1e-5f), OP_SINGLE_ARG(float, "decay", decay_, 0.95f) {} bool RunOnDevice() override { // Enforce shapes CAFFE_ENFORCE_EQ(Input(PARAM).numel(), Input(MOMENT_GRAD).numel()); CAFFE_ENFORCE_EQ(Input(PARAM).numel(), Input(MOMENT_DELTA).numel()); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); // Enforce domain constraints for attributes CAFFE_ENFORCE_GE(epsilon_, 0.0f); CAFFE_ENFORCE_GT(decay_, 0.0f); CAFFE_ENFORCE_LT(decay_, 1.0f); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<float>(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = Input(GRAD).template data<float>(); const auto* paramIn = Input(PARAM).template data<float>(); const auto* momentIn = Input(MOMENT_GRAD).template data<float>(); const auto* momentDeltaIn = Input(MOMENT_DELTA).template data<float>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<float>(); auto* momentOut = Output(OUTPUT_MOMENT_GRAD)->template mutable_data<float>(); auto* momentDeltaOut = Output(OUTPUT_MOMENT_DELTA)->template mutable_data<float>(); auto n = Input(INDICES).numel(); if (n == 0) { return true; } auto block_size = Input(GRAD).numel() / n; for (const auto i : c10::irange(n)) { auto idx = indices[i]; if (block_size == 1) { float gi = gradIn[i]; float di = momentDeltaIn[idx]; float hi = momentOut[idx] = decay_ * momentIn[idx] + (1.0f - decay_) * gi * gi; float ng = (std::sqrt(di + epsilon_) / std::sqrt(hi + epsilon_)) * gi; paramOut[idx] = paramIn[idx] + lr[0] * ng; momentDeltaOut[idx] = decay_ * di + (1.0f - decay_) * ng * ng; } else { auto offsetI = i * block_size; auto offsetIdx = idx * block_size; #ifndef NDEBUG CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input i:", i, " and block size:", block_size); CAFFE_ENFORCE_GE( Input(GRAD).numel(), block_size + offsetI, this->debug_def().input(GRAD), ", out of bound idx, idx:", idx, " for input i:", i); #endif AdadeltaUpdate( block_size, paramIn + offsetIdx, gradIn + offsetI, momentIn + offsetIdx, momentDeltaIn + offsetIdx, epsilon_, decay_, lr, paramOut + offsetIdx, momentOut + offsetIdx, momentDeltaOut + offsetIdx, &context_); } } return true; } protected: const float epsilon_; const float decay_; INPUT_TAGS(PARAM, MOMENT_GRAD, MOMENT_DELTA, INDICES, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_GRAD, OUTPUT_MOMENT_DELTA); }; } // namespace caffe2
5,897
31.054348
78
h
null
pytorch-main/caffe2/sgd/adagrad_fused.h
#pragma once #include "caffe2/sgd/adagrad_op.h" #include "caffe2/sgd/math_lp.h" namespace caffe2 { namespace { template < typename Tdata, // embedding and momentum types typename T, // everything else typename TLengths, typename adagradT, bool is_mean = false> class SparseAdagradFusedWithSparseLengthsSumGradientOp final : public Operator<CPUContext> { public: SparseAdagradFusedWithSparseLengthsSumGradientOp( const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)) { VLOG(1) << "gradient optimization operator in use: " << "SparseAdagradFusedWithSparseLengthsSumGradientOp" << " weight_decay_=" << weight_decay_; const T decay = this->template GetSingleArgument<T>("decay", 1.0); CAFFE_ENFORCE_EQ( decay, 1.0, "Decay is not supported for SparseSimdAdagradOp"); } bool RunOnDevice() override { return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<T>(); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_1)->ResizeLike(Input(MOMENT_1)); auto& segmentGradsInput = Input(GRAD); auto& lengthsInput = Input(LENGTHS); CAFFE_ENFORCE_EQ(lengthsInput.dim(), 1, "LENGTHS must be a vector"); auto numSegments = lengthsInput.size(0); CAFFE_ENFORCE_GT(segmentGradsInput.dim(), 0); CAFFE_ENFORCE_EQ(numSegments, segmentGradsInput.size(0)); const auto* lengths = lengthsInput.template data<TLengths>(); auto n = Input(INDICES).numel(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = segmentGradsInput.template data<T>(); const auto* paramIn = Input(PARAM).template data<Tdata>(); const auto* momentIn = Input(MOMENT_1).template data<Tdata>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<Tdata>(); auto* momentOut = Output(OUTPUT_MOMENT_1)->template mutable_data<Tdata>(); if (numSegments == 0) { return true; } auto block_size = segmentGradsInput.size_from_dim(1); // Enforce: // input(embedding/momentum) == outputs(embedding/momentum) CAFFE_ENFORCE_EQ( Input(PARAM).numel(), Input(MOMENT_1).numel(), "Input Param size: ", Input(PARAM).numel(), " Input Moment size: ", Input(MOMENT_1).numel()); int dataIndex = 0; if (is_mean) { grad_buffer_.ResizeLike(Input(GRAD)); } auto* grad_buffer_data = is_mean ? grad_buffer_.template mutable_data<T>() : NULL; if (is_mean) { for (const auto rangeIndex : c10::irange(numSegments)) { for (const auto tmpIndex : c10::irange(block_size)) { auto offsetI = rangeIndex * block_size; grad_buffer_data[offsetI + tmpIndex] = lengths[rangeIndex] > 0 ? gradIn[offsetI + tmpIndex] / lengths[rangeIndex] : gradIn[offsetI + tmpIndex]; } } } for (const auto rangeIndex : c10::irange(numSegments)) { for (auto start = dataIndex; dataIndex < start + lengths[rangeIndex]; ++dataIndex) { std::size_t idx = indices[dataIndex]; auto offsetI = rangeIndex * block_size; auto offsetIdx = idx * block_size; // Enforce: // access within range // gradient access within range CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input dataIndex:", dataIndex, " and block size:", block_size, " max size:", Input(PARAM).numel()); if (block_size == 1) { float gi = std::fma( weight_decay_, paramIn[idx], is_mean ? grad_buffer_data[offsetI] : gradIn[offsetI]); float hi = momentOut[idx] = momentIn[idx] + gi * gi; paramOut[idx] = paramIn[idx] + lr[0] * gi / (std::sqrt(hi) + epsilon_); } else { // prefetching const int prefdist_T0 = 16; int i_pref = (dataIndex < n - prefdist_T0) ? dataIndex + prefdist_T0 : dataIndex; std::size_t idx_pref = indices[i_pref]; kernel_( block_size, paramIn + offsetIdx, &paramIn[idx_pref * block_size], is_mean ? grad_buffer_data + offsetI : gradIn + offsetI, momentIn + offsetIdx, &momentIn[idx_pref * block_size], paramOut + offsetIdx, &paramOut[idx_pref * block_size], momentOut + offsetIdx, &momentOut[idx_pref * block_size], epsilon_, lr[0], weight_decay_); } } } CAFFE_ENFORCE_EQ(dataIndex, n); return true; } protected: T epsilon_; T weight_decay_; adagradT kernel_; Tensor grad_buffer_{CPU}; INPUT_TAGS(PARAM, MOMENT_1, INDICES, GRAD, LR, LENGTHS); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1); }; template <typename Tdata, typename T, typename TLengths, typename adagradT> class SparseAdagradFusedWithSparseLengthsWeightedSumGradientOp final : public Operator<CPUContext> { public: SparseAdagradFusedWithSparseLengthsWeightedSumGradientOp( const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)) { VLOG(1) << "gradient optimization operator in use: " << "SparseAdagradFusedWithSparseLengthsWeightedSumGradientOp"; const T decay = this->template GetSingleArgument<T>("decay", 1.0); CAFFE_ENFORCE_EQ( decay, 1.0, "Decay is not supported for SparseSimdAdagradOp"); } bool RunOnDevice() override { return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<T>(); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_1)->ResizeLike(Input(MOMENT_1)); auto& segmentGradsInput = Input(GRAD); auto& lengthsInput = Input(LENGTHS); CAFFE_ENFORCE_EQ(lengthsInput.dim(), 1, "LENGTHS must be a vector"); auto numSegments = lengthsInput.size(0); CAFFE_ENFORCE_GT(segmentGradsInput.dim(), 0); CAFFE_ENFORCE_EQ(numSegments, segmentGradsInput.size(0)); const auto* lengths = lengthsInput.template data<TLengths>(); auto n = Input(INDICES).numel(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = segmentGradsInput.template data<T>(); const auto* paramIn = Input(PARAM).template data<Tdata>(); const auto* momentIn = Input(MOMENT_1).template data<Tdata>(); const auto* auxParamIn = Input(AUX_PARAM).template data<T>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<Tdata>(); auto* momentOut = Output(OUTPUT_MOMENT_1)->template mutable_data<Tdata>(); Output(AUX_GRAD)->Resize(n); auto* auxGrad = Output(AUX_GRAD)->template mutable_data<T>(); if (numSegments == 0) { return true; } auto block_size = segmentGradsInput.size_from_dim(1); // Enforce: // input(embedding/momentum) == outputs(embedding/momentum) CAFFE_ENFORCE_EQ( Input(PARAM).numel(), Input(MOMENT_1).numel(), "Input Param size: ", Input(PARAM).numel(), " Input Moment size: ", Input(MOMENT_1).numel()); // Cannot fuse this loop with the loop below because paramIn is updated // by the second loop. Specifically, there could be dataIndex1 != dataIndex2 // s.t. indices[dataIndex1] == indices[dataIndex2], and fusing these two // loops would violate dependencies w.r.t. // paramIn[indices[dataIndex1]:block_size] The approximate version. // (RowWiseSparseSimdAdagradFusedWithSparseLengthsWeightedSumGradientApproxOp) // ignores this dependency and fuses these two loops. std::vector<T> temp_grad(block_size); int dataIndex = 0; for (const auto rangeIndex : c10::irange(numSegments)) { for (auto start = dataIndex; dataIndex < start + lengths[rangeIndex]; ++dataIndex) { std::size_t idx = indices[dataIndex]; auto offsetI = rangeIndex * block_size; auto offsetIdx = idx * block_size; // Enforce: // access within range // gradient access within range CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input dataIndex:", dataIndex, " and block size:", block_size, " max size:", Input(PARAM).numel()); internal::dot<T, Tdata, T>( block_size, gradIn + offsetI, paramIn + offsetIdx, auxGrad + dataIndex, &context_); } } CAFFE_ENFORCE_EQ(dataIndex, n); dataIndex = 0; for (const auto rangeIndex : c10::irange(numSegments)) { for (auto start = dataIndex; dataIndex < start + lengths[rangeIndex]; ++dataIndex) { std::size_t idx = indices[dataIndex]; auto offsetI = rangeIndex * block_size; auto offsetIdx = idx * block_size; auto localOffset = dataIndex - start; for (const auto i : c10::irange(block_size)) { temp_grad[i] = auxParamIn[localOffset] * gradIn[offsetI + i]; } if (block_size == 1) { float gi = std::fma(weight_decay_, paramIn[idx], temp_grad[0]); float hi = momentOut[idx] = momentIn[idx] + gi * gi; paramOut[idx] = paramIn[idx] + lr[0] * gi / (std::sqrt(hi) + epsilon_); } else { // prefetching const int prefdist_T0 = 16; int i_pref = (dataIndex < n - prefdist_T0) ? dataIndex + prefdist_T0 : dataIndex; std::size_t idx_pref = indices[i_pref]; kernel_( block_size, paramIn + offsetIdx, &paramIn[idx_pref * block_size], temp_grad.data(), momentIn + offsetIdx, &momentIn[idx_pref * block_size], paramOut + offsetIdx, &paramOut[idx_pref * block_size], momentOut + offsetIdx, &momentOut[idx_pref * block_size], epsilon_, lr[0], weight_decay_); } } } return true; } protected: T epsilon_; T weight_decay_; adagradT kernel_; INPUT_TAGS(PARAM, MOMENT_1, AUX_PARAM, INDICES, GRAD, LR, LENGTHS); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1, AUX_GRAD); }; template < typename Tdata, // embedding and momentum types typename T, // everything else typename TLengths, typename adagradT> class SparseAdagradFusedWithSparseLengthsWeightedSumGradientApproxOp final : public Operator<CPUContext> { public: SparseAdagradFusedWithSparseLengthsWeightedSumGradientApproxOp( const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)) { VLOG(1) << "gradient optimization operator in use: " << "SparseAdagradFusedWithSparseLengthsWeightedSumGradientApproxOp"; const T decay = this->template GetSingleArgument<T>("decay", 1.0); CAFFE_ENFORCE_EQ( decay, 1.0, "Decay is not supported for SparseSimdAdagradOp"); } bool RunOnDevice() override { return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<T>(); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_1)->ResizeLike(Input(MOMENT_1)); auto& segmentGradsInput = Input(GRAD); auto& lengthsInput = Input(LENGTHS); CAFFE_ENFORCE_EQ(lengthsInput.dim(), 1, "LENGTHS must be a vector"); auto numSegments = lengthsInput.size(0); CAFFE_ENFORCE_GT(segmentGradsInput.dim(), 0); CAFFE_ENFORCE_EQ(numSegments, segmentGradsInput.size(0)); const auto* lengths = lengthsInput.template data<TLengths>(); auto n = Input(INDICES).numel(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = segmentGradsInput.template data<T>(); const auto* paramIn = Input(PARAM).template data<Tdata>(); const auto* momentIn = Input(MOMENT_1).template data<Tdata>(); const auto* auxParamIn = Input(AUX_PARAM).template data<T>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<Tdata>(); auto* momentOut = Output(OUTPUT_MOMENT_1)->template mutable_data<Tdata>(); Output(AUX_GRAD)->Resize(n); auto* auxGrad = Output(AUX_GRAD)->template mutable_data<T>(); if (numSegments == 0) { return true; } auto block_size = segmentGradsInput.size_from_dim(1); // Enforce: // input(embedding/momentum) == outputs(embedding/momentum) CAFFE_ENFORCE_EQ( Input(PARAM).numel(), Input(MOMENT_1).numel(), "Input Param size: ", Input(PARAM).numel(), " Input Moment size: ", Input(MOMENT_1).numel()); std::vector<T> temp_grad(block_size); int dataIndex = 0; for (const auto rangeIndex : c10::irange(numSegments)) { for (auto start = dataIndex; dataIndex < start + lengths[rangeIndex]; ++dataIndex) { std::size_t idx = indices[dataIndex]; auto offsetI = rangeIndex * block_size; auto offsetIdx = idx * block_size; auto localOffset = dataIndex - start; // Enforce: // access within range // gradient access within range CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input dataIndex:", dataIndex, " and block size:", block_size, " max size:", Input(PARAM).numel()); internal::dot<T, Tdata, T>( block_size, gradIn + offsetI, paramIn + offsetIdx, auxGrad + dataIndex, &context_); for (const auto i : c10::irange(block_size)) { temp_grad[i] = auxParamIn[localOffset] * gradIn[offsetI + i]; } if (block_size == 1) { float gi = std::fma(weight_decay_, paramIn[idx], temp_grad[0]); float hi = momentOut[idx] = momentIn[idx] + gi * gi; paramOut[idx] = paramIn[idx] + lr[0] * gi / (std::sqrt(hi) + epsilon_); } else { // prefetching const int prefdist_T0 = 16; int i_pref = (dataIndex < n - prefdist_T0) ? dataIndex + prefdist_T0 : dataIndex; std::size_t idx_pref = indices[i_pref]; kernel_( block_size, paramIn + offsetIdx, &paramIn[idx_pref * block_size], temp_grad.data(), momentIn + offsetIdx, &momentIn[idx_pref * block_size], paramOut + offsetIdx, &paramOut[idx_pref * block_size], momentOut + offsetIdx, &momentOut[idx_pref * block_size], epsilon_, lr[0], weight_decay_); } } } CAFFE_ENFORCE_EQ(dataIndex, n); return true; } protected: T epsilon_; T weight_decay_; adagradT kernel_; INPUT_TAGS(PARAM, MOMENT_1, AUX_PARAM, INDICES, GRAD, LR, LENGTHS); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1, AUX_GRAD); }; } // namespace } // namespace caffe2
16,586
32.374245
82
h
null
pytorch-main/caffe2/sgd/adagrad_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/perfkernels/adagrad.h" #if defined(USE_FBGEMM) && !defined(__NVCC__) #include "fbgemm/FbgemmEmbedding.h" #endif namespace caffe2 { template <typename Context> void adagrad_update( int N, const float* w, const float* g, const float* h, float* nw, float* nh, float epsilon, float decay, const float* lr, Context* /*context*/, float weight_decay = 0.f) { return adagrad_update( N, w, g, h, nw, nh, epsilon, decay, lr[0], weight_decay); } template <typename Context> void adagrad_update_output_effective_lr( int N, const float* paramIn, const float* gradIn, const float* momentIn, float* paramOut, float* momentOut, float* effectiveLROut, float epsilon, float decay, const float* lr, Context* /*context*/, float weight_decay = 0.f) { for (const auto i : c10::irange(N)) { float grad = std::fma(weight_decay, paramIn[i], gradIn[i]); float moment = momentOut[i] = decay * momentIn[i] + grad * grad; float effective_lr = effectiveLROut[i] = lr[0] / (std::sqrt(moment) + epsilon); paramOut[i] = paramIn[i] + effective_lr * grad; } } template <typename Context> void adagrad_update_output_effective_lr_and_update( int N, const float* paramIn, const float* gradIn, const float* momentIn, float* paramOut, float* momentOut, float* effectiveLROut, float* updateOut, float epsilon, float decay, const float* lr, Context* /*context*/, float weight_decay = 0.f) { for (const auto i : c10::irange(N)) { float grad = std::fma(weight_decay, paramIn[i], gradIn[i]); float moment = momentOut[i] = decay * momentIn[i] + grad * grad; float effective_lr = effectiveLROut[i] = lr[0] / (std::sqrt(moment) + epsilon); float update = updateOut[i] = effective_lr * grad; paramOut[i] = paramIn[i] + update; } } template <class Context> class AdagradOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; AdagradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)), decay_(this->template GetSingleArgument<float>("decay", 1.0f)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)) { VLOG(1) << "gradient optimization operator in use: " << "AdagradOp" << " weight_decay_=" << weight_decay_; } bool RunOnDevice() override { CAFFE_ENFORCE_EQ( Input(GRAD).numel(), Input(MOMENT_1).numel(), "PARAM size: ", Input(PARAM).numel(), ", GRAD size: ", Input(GRAD).numel(), ", MOMENT_1 size: ", Input(MOMENT_1).numel(), ", LR size: ", Input(LR).numel()); CAFFE_ENFORCE_EQ(Input(GRAD).numel(), Input(PARAM).numel()); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_1)->ResizeLike(Input(MOMENT_1)); if (OutputSize() == 2) { adagrad_update<Context>( Input(GRAD).numel(), Input(PARAM).template data<float>(), Input(GRAD).template data<float>(), Input(MOMENT_1).template data<float>(), Output(OUTPUT_PARAM)->template mutable_data<float>(), Output(OUTPUT_MOMENT_1)->template mutable_data<float>(), epsilon_, decay_, Input(LR).template data<float>(), &context_, weight_decay_); } else if (OutputSize() == 3) { Output(OUTPUT_EFFECTIVE_LR)->ResizeLike(Input(GRAD)); adagrad_update_output_effective_lr<Context>( Input(GRAD).numel(), Input(PARAM).template data<float>(), Input(GRAD).template data<float>(), Input(MOMENT_1).template data<float>(), Output(OUTPUT_PARAM)->template mutable_data<float>(), Output(OUTPUT_MOMENT_1)->template mutable_data<float>(), Output(OUTPUT_EFFECTIVE_LR)->template mutable_data<float>(), epsilon_, decay_, Input(LR).template data<float>(), &context_, weight_decay_); } else { Output(OUTPUT_EFFECTIVE_LR)->ResizeLike(Input(GRAD)); Output(OUTPUT_UPDATE)->ResizeLike(Input(GRAD)); adagrad_update_output_effective_lr_and_update<Context>( Input(GRAD).numel(), Input(PARAM).template data<float>(), Input(GRAD).template data<float>(), Input(MOMENT_1).template data<float>(), Output(OUTPUT_PARAM)->template mutable_data<float>(), Output(OUTPUT_MOMENT_1)->template mutable_data<float>(), Output(OUTPUT_EFFECTIVE_LR)->template mutable_data<float>(), Output(OUTPUT_UPDATE)->template mutable_data<float>(), epsilon_, decay_, Input(LR).template data<float>(), &context_, weight_decay_); } return true; } protected: float epsilon_; float decay_; float weight_decay_; INPUT_TAGS(PARAM, MOMENT_1, GRAD, LR); OUTPUT_TAGS( OUTPUT_PARAM, OUTPUT_MOMENT_1, OUTPUT_EFFECTIVE_LR, OUTPUT_UPDATE); }; class SparseAdagradOp final : public Operator<CPUContext> { public: SparseAdagradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)) { VLOG(1) << "gradient optimization operator in use: " << "SparseAdagradOp" << " weight_decay_=" << weight_decay_; const float decay = this->template GetSingleArgument<float>("decay", 1.0); CAFFE_ENFORCE_EQ( decay, 1.0, "Decay is not supported for SparseSimdAdagradOp"); } bool RunOnDevice() override { // Enforce shapes // input(embedding/momentum) == outputs(embedding/momentum) CAFFE_ENFORCE_EQ( Input(PARAM).numel(), Input(MOMENT_1).numel(), "Input Param size: ", Input(PARAM).numel(), " Input Moment size: ", Input(MOMENT_1).numel()); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<float>(); auto n = Input(INDICES).numel(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = Input(GRAD).template data<float>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<float>(); auto* momentOut = Output(OUTPUT_MOMENT_1)->template mutable_data<float>(); if (n == 0) { return true; } auto block_size = Input(GRAD).numel() / n; // input(grad) is compatible with size of indexes CAFFE_ENFORCE_EQ( Input(GRAD).numel() % n, 0, "Incorrect gradient size:", Input(GRAD).numel(), " size of indexes:", n); #if defined(USE_FBGEMM) && !defined(__NVCC__) VLOG(1) << "using fbgemm::GenerateSparseAdaGrad in SparseAdagradOp"; if (block_size != last_block_size_) { last_block_size_ = block_size; if (std::is_same<SIndex, std::int32_t>::value) { kernel_i32_ = fbgemm::GenerateSparseAdaGrad<std::int32_t>( block_size, /*rowwise=*/false, /*prefetch=*/16, weight_decay_ != 0.0f); } else { CAFFE_ENFORCE((std::is_same<SIndex, std::int64_t>::value)); kernel_i64_ = fbgemm::GenerateSparseAdaGrad<std::int64_t>( block_size, /*rowwise=*/false, /*prefetch=*/16, weight_decay_ != 0.0f); } } int num_rows_processed; if (std::is_same<SIndex, std::int32_t>::value) { num_rows_processed = kernel_i32_( n, Input(PARAM).numel(), paramOut, gradIn, momentOut, reinterpret_cast<const std::int32_t*>(indices), epsilon_, lr[0], weight_decay_, /*counter=*/nullptr, /*counter_halflife=*/0); } else { num_rows_processed = kernel_i64_( n, Input(PARAM).numel(), paramOut, gradIn, momentOut, reinterpret_cast<const std::int64_t*>(indices), epsilon_, lr[0], weight_decay_, /*counter=*/nullptr, /*counter_halflife=*/0); } if (num_rows_processed < n) { CAFFE_ENFORCE_GE( Input(PARAM).numel(), (indices[num_rows_processed] + 1) * block_size, this->debug_def().input(PARAM), ", out of bound, idx:", indices[num_rows_processed], " for input i:", num_rows_processed, " and block_size:", block_size, " max size:", Input(PARAM).numel()); return false; } else { return true; } #endif VLOG(1) << "using internal::adagrad_update_prefetch_inlined in SparseAdagradOp"; const auto* paramIn = Input(PARAM).template data<float>(); const auto* momentIn = Input(MOMENT_1).template data<float>(); std::vector<float> grad(block_size); for (const auto i : c10::irange(n)) { auto idx = indices[i]; auto offsetI = i * block_size; auto offsetIdx = idx * block_size; // Enforce: // access within range // gradient access within range CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input i:", i, " and block size:", block_size, " max size:", Input(PARAM).numel()); if (block_size == 1) { float gi = std::fma(weight_decay_, paramIn[idx], gradIn[i]); float hi = momentOut[idx] = momentIn[idx] + gi * gi; paramOut[idx] = paramIn[idx] + lr[0] * gi / (std::sqrt(hi) + epsilon_); } else { // prefetching const int prefdist_T0 = 16; int i_pref = (i < n - prefdist_T0) ? i + prefdist_T0 : i; std::size_t idx_pref = indices[i_pref]; internal::adagrad_update_prefetch_inlined( block_size, paramIn + offsetIdx, &paramIn[idx_pref * block_size], gradIn + offsetI, momentIn + offsetIdx, &momentIn[idx_pref * block_size], paramOut + offsetIdx, &paramOut[idx_pref * block_size], momentOut + offsetIdx, &momentOut[idx_pref * block_size], epsilon_, lr[0], weight_decay_); } } return true; } protected: float epsilon_; const float weight_decay_; #if defined(USE_FBGEMM) && !defined(__NVCC__) fbgemm::SparseAdaGradSignature<std::int32_t>::Type kernel_i32_; fbgemm::SparseAdaGradSignature<std::int64_t>::Type kernel_i64_; std::int64_t last_block_size_{-1}; #endif INPUT_TAGS(PARAM, MOMENT_1, INDICES, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1); }; template <class Context> class RowWiseSparseAdagradOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; RowWiseSparseAdagradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.f)), counter_halflife_( this->template GetSingleArgument<int64_t>("counter_halflife", -1)) { VLOG(1) << "gradient optimization operator in use: " << "RowWiseSparseAdagradOp" << " weight_decay_=" << weight_decay_ << " counter_halflife=" << counter_halflife_; } bool RunOnDevice() override { // Enforce shapes CAFFE_ENFORCE_EQ(Input(PARAM).sizes()[0], Input(MOMENT_1).numel()); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<float>(); auto* param = Output(OUTPUT_PARAM)->template mutable_data<float>(); auto* moment = Output(OUTPUT_MOMENT_1)->template mutable_data<float>(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = Input(GRAD).template data<float>(); const auto* count = counter_halflife_ == -1 ? nullptr : Input(COUNTER).template data<double>(); auto n = Input(INDICES).numel(); if (n == 0) { return true; } auto block_size = Input(GRAD).numel() / n; // Enforce: // Input(embedding/momentum) == outputs(embedding/momentum) CAFFE_ENFORCE_EQ( Input(PARAM).numel() / block_size, Input(MOMENT_1).numel(), "Input Param size: ", Input(PARAM).numel(), " Block size: ", block_size, " Input Moment size: ", Input(MOMENT_1).numel()); // input(grad) is compatible with size of indexes CAFFE_ENFORCE_EQ( Input(GRAD).numel() % n, 0, "Incorrect gradient size:", Input(GRAD).numel(), " size of indexes:", n); #if defined(USE_FBGEMM) && !defined(__NVCC__) VLOG(1) << "using fbgemm::GenerateSparseAdaGrad in RowWiseSparseAdagradOp"; if (block_size != last_block_size_) { last_block_size_ = block_size; if (std::is_same<SIndex, std::int32_t>::value) { kernel_i32_ = fbgemm::GenerateSparseAdaGrad<std::int32_t>( block_size, /*rowwise=*/true, /*prefetch=*/16, weight_decay_ != 0.0f); } else { CAFFE_ENFORCE((std::is_same<SIndex, std::int64_t>::value)); kernel_i64_ = fbgemm::GenerateSparseAdaGrad<std::int64_t>( block_size, /*rowwise=*/true, /*prefetch=*/16, weight_decay_ != 0.0f); } } int num_rows_processed; if (std::is_same<SIndex, std::int32_t>::value) { num_rows_processed = kernel_i32_( n, Input(PARAM).numel(), param, gradIn, moment, reinterpret_cast<const std::int32_t*>(indices), epsilon_, lr[0], weight_decay_, (counter_halflife_ > 0) ? count : nullptr, counter_halflife_); } else { num_rows_processed = kernel_i64_( n, Input(PARAM).numel(), param, gradIn, moment, reinterpret_cast<const std::int64_t*>(indices), epsilon_, lr[0], weight_decay_, (counter_halflife_ > 0) ? count : nullptr, counter_halflife_); } if (num_rows_processed < n) { // Enforce: // access within range CAFFE_ENFORCE_GE( Input(PARAM).numel(), (indices[num_rows_processed] + 1) * block_size, this->debug_def().input(PARAM), ", out of bound, idx:", indices[num_rows_processed], " for input i:", num_rows_processed, " and block size:", block_size, " max size:", Input(PARAM).numel()); return false; } else { return true; } #else VLOG(1) << "using plain adagrad updates in RowWiseSparseAdagradOp"; for (const auto i : c10::irange(n)) { auto idx = indices[i]; float freq = (counter_halflife_ > 0 && count[idx] > 0) ? counter_halflife_ / count[idx] : 1.0; if (block_size == 1) { float gi = std::fma(weight_decay_ * freq, param[idx], gradIn[i]); float hi = moment[idx] = moment[idx] + gi * gi; param[idx] = param[idx] + lr[0] * gi / (std::sqrt(hi) + epsilon_); } else { auto offsetI = i * block_size; auto offsetIdx = idx * block_size; #ifndef NDEBUG CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input i:", i, " and block size:", block_size); CAFFE_ENFORCE_GE( Input(GRAD).numel(), block_size + offsetI, this->debug_def().input(GRAD), ", out of bound idx, idx:", idx, " for input i:", i); #endif float* w = param + offsetIdx; const float* g = gradIn + offsetI; float* h = moment + idx; float hs = 0.; for (const auto j : c10::irange(block_size)) { float gj = std::fma(weight_decay_ * freq, w[j], g[j]); hs += gj * gj; } float hi = h[0] = h[0] + hs / block_size; float step = lr[0] / (std::sqrt(hi) + epsilon_); for (const auto j : c10::irange(block_size)) { float gj = std::fma(weight_decay_ * freq, w[j], g[j]); w[j] = w[j] + gj * step; } } } return true; #endif // !USE_FBGEMM } protected: float epsilon_; const float weight_decay_; const int64_t counter_halflife_; #if defined(USE_FBGEMM) && !defined(__NVCC__) fbgemm::SparseAdaGradSignature<std::int32_t>::Type kernel_i32_; fbgemm::SparseAdaGradSignature<std::int64_t>::Type kernel_i64_; std::int64_t last_block_size_{-1}; #endif INPUT_TAGS(PARAM, MOMENT_1, INDICES, GRAD, LR, COUNTER); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1); }; } // namespace caffe2
18,035
30.366957
80
h
null
pytorch-main/caffe2/sgd/clip_tensor_op.h
#ifndef CAFFE2_OPERATORS_CLIP_TENSOR_OP_H_ #define CAFFE2_OPERATORS_CLIP_TENSOR_OP_H_ #include <vector> #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/core/tensor.h" #include "caffe2/utils/math.h" namespace caffe2 { template <typename Context> class ClipTensorByScalingOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; ClipTensorByScalingOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) { threshold_ = this->template GetSingleArgument<float>("threshold", 0.0); CAFFE_ENFORCE_GT(threshold_, 0, "Threshold must be greater than 0"); } bool RunOnDevice() override { const auto& input_tensor = Input(0); CAFFE_ENFORCE_GT(input_tensor.numel(), 0); const auto& val = Input(1); CAFFE_ENFORCE_EQ(val.numel(), 1); const auto* input_tensor_data = input_tensor.template data<float>(); const auto* val_data = val.template data<float>(); auto* clipped = Output(0, input_tensor.sizes(), at::dtype<float>()); float* clipped_tensor_data = clipped->template mutable_data<float>(); if (InputSize() > 2) { const auto& additional_threshold = Input(2); CAFFE_ENFORCE_EQ(additional_threshold.numel(), 1); threshold_ *= *(additional_threshold.template data<float>()); } if (*val_data > threshold_) { float ratio = threshold_ / *val_data; math::Scale<float, float, Context>( clipped->numel(), ratio, input_tensor_data, clipped_tensor_data, &context_); } else { if (input_tensor_data != clipped_tensor_data) { clipped->CopyFrom(input_tensor, /*async*/ true); } } return true; } private: float threshold_; }; } // namespace caffe2 #endif // CAFFE2_OPERATORS_CLIP_TENSOR_OP_H_
1,853
26.671642
75
h
null
pytorch-main/caffe2/sgd/decay_adagrad_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/utils/eigen_utils.h" namespace caffe2 { template <typename Context> void decay_adagrad_compute( int N, const float* w, const float* g, const float* m, const float* v, float* nw, float* nm, float* nv, float beta1, float beta2, float eps_hat, float weight_decay, float c, const float* lr, Context* /*context*/) { ConstEigenVectorArrayMap<float> w_arr(w, N); ConstEigenVectorArrayMap<float> g_arr(g, N); ConstEigenVectorArrayMap<float> m_arr(m, N); ConstEigenVectorArrayMap<float> v_arr(v, N); EigenVectorArrayMap<float> nw_arr(nw, N); EigenVectorArrayMap<float> nm_arr(nm, N); EigenVectorArrayMap<float> nv_arr(nv, N); nm_arr = m_arr * beta1 + g_arr * (1.0f - beta1); nv_arr = v_arr + g_arr.square(); nw_arr = w_arr + *lr * (nm_arr / c / (nv_arr.sqrt() + eps_hat) + weight_decay * w_arr); } template <typename T, class Context> class DecayAdagradOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; DecayAdagradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), beta1_(this->template GetSingleArgument<float>("beta1", 0.9f)), beta2_(this->template GetSingleArgument<float>("beta2", 0.999f)), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)), weight_decay_(this->template GetSingleArgument<float>("weight_decay", 0.0f)), bias_correction_first_(this->template GetSingleArgument<bool>("bias_correction_first", true)) {} bool RunOnDevice() override { // Iter live on the CPU CAFFE_ENFORCE(OperatorBase::InputIsTensorType(ITER, CPU)); CAFFE_ENFORCE(Input(LR).numel() == 1); CAFFE_ENFORCE(Input(GRAD).numel() == Input(PARAM).numel()); CAFFE_ENFORCE(Input(GRAD).numel() == Input(MOMENT_1).numel()); CAFFE_ENFORCE(Input(GRAD).numel() == Input(MOMENT_2).numel()); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT_1)->ResizeLike(Input(MOMENT_1)); Output(OUTPUT_MOMENT_2)->ResizeLike(Input(MOMENT_2)); const auto iter = OperatorBase::Input<Tensor>(ITER, CPU).template data<int64_t>()[0]; const auto t = iter + 1; const auto c = (bias_correction_first_)? (T(1.) - std::pow(beta1_, t)) : 1.0; decay_adagrad_compute<Context>( Input(GRAD).numel(), Input(PARAM).template data<T>(), Input(GRAD).template data<T>(), Input(MOMENT_1).template data<T>(), Input(MOMENT_2).template data<T>(), Output(OUTPUT_PARAM)->template mutable_data<T>(), Output(OUTPUT_MOMENT_1)->template mutable_data<T>(), Output(OUTPUT_MOMENT_2)->template mutable_data<T>(), beta1_, beta2_, epsilon_, weight_decay_, c, Input(LR).template data<T>(), &context_); return true; } protected: T beta1_{0.9}; T beta2_{0.999}; T epsilon_{1e-8}; T weight_decay_{0.0}; bool bias_correction_first_{true}; INPUT_TAGS(PARAM, MOMENT_1, MOMENT_2, GRAD, LR, ITER); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT_1, OUTPUT_MOMENT_2); }; } // namespace caffe2
3,215
32.5
104
h
null
pytorch-main/caffe2/sgd/fp16_momentum_sgd_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/core/timer.h" namespace caffe2 { template <class Context> void fp16_momentum_sgd_update( int N, const at::Half* g, const at::Half* m, at::Half* ng, at::Half* nm, const float* lr, float momentum, bool nesterov, float weight_decay, bool fp32_update, at::Half* param, Context* /*context*/) {} template <typename T, class Context> class FP16MomentumSGDUpdateOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FP16MomentumSGDUpdateOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), momentum_(this->template GetSingleArgument<float>("momentum", 0.0)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.0)), nesterov_(this->template GetSingleArgument<int>("nesterov", 0)), // when set, fp32_update will read in the fp16 data but // perform all the compute in fp32 precision. fp32_update_(this->template GetSingleArgument<int>("fp32_update", 0)) {} bool RunOnDevice() override { auto device_type = Context::GetDeviceType(); // Iter live on the CPU CAFFE_ENFORCE(OperatorBase::InputIsTensorType(GRAD, device_type)); CAFFE_ENFORCE(OperatorBase::InputIsTensorType(MOMENTUM, device_type)); CAFFE_ENFORCE(Input(LR).size() == 1); CAFFE_ENFORCE(Input(GRAD).size() == Input(MOMENTUM).size()); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_MOMENTUM)->ResizeLike(Input(MOMENTUM)); fp16_momentum_sgd_update<Context>( Input(GRAD).size(), Input(GRAD).template data<T>(), Input(MOMENTUM).template data<T>(), Output(OUTPUT_GRAD)->template mutable_data<T>(), Output(OUTPUT_MOMENTUM)->template mutable_data<T>(), Input(LR).template data<float>(), momentum_, nesterov_, weight_decay_, fp32_update_, Output(OUTPUT_PARAM)->template mutable_data<T>(), &context_); return true; } protected: float momentum_{0.9}; float weight_decay_{0.0}; bool nesterov_; bool fp32_update_; INPUT_TAGS(GRAD, MOMENTUM, LR, PARAM); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MOMENTUM, OUTPUT_PARAM); }; }
2,292
30.410959
80
h
null
pytorch-main/caffe2/sgd/fp32_momentum_sgd_op.h
#pragma once #include "caffe2/core/operator.h" #include "caffe2/core/timer.h" namespace caffe2 { template <class Context> void fp32_momentum_sgd_update( int N, const float* g, const float* m, float* ng, float* nm, const float* lr, float momentum, bool nesterov, float weight_decay, float* param, Context* /*context*/) {} template <typename T, class Context> class FP32MomentumSGDUpdateOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FP32MomentumSGDUpdateOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), momentum_(this->template GetSingleArgument<float>("momentum", 0.0)), weight_decay_( this->template GetSingleArgument<float>("weight_decay", 0.0)), nesterov_(this->template GetSingleArgument<int>("nesterov", 0)) {} bool RunOnDevice() override { auto device_type = Context::GetDeviceType(); // Iter live on the CPU CAFFE_ENFORCE(OperatorBase::InputIsTensorType(GRAD, device_type)); CAFFE_ENFORCE(OperatorBase::InputIsTensorType(MOMENTUM, device_type)); CAFFE_ENFORCE(Input(LR).size() == 1); CAFFE_ENFORCE(Input(GRAD).size() == Input(MOMENTUM).size()); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_MOMENTUM)->ResizeLike(Input(MOMENTUM)); fp32_momentum_sgd_update<Context>( Input(GRAD).size(), Input(GRAD).template data<T>(), Input(MOMENTUM).template data<T>(), Output(OUTPUT_GRAD)->template mutable_data<T>(), Output(OUTPUT_MOMENTUM)->template mutable_data<T>(), Input(LR).template data<float>(), momentum_, nesterov_, weight_decay_, Output(OUTPUT_PARAM)->template mutable_data<T>(), &context_); return true; } protected: float momentum_{0.9}; float weight_decay_{0.0}; bool nesterov_; INPUT_TAGS(GRAD, MOMENTUM, LR, PARAM); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MOMENTUM, OUTPUT_PARAM); }; }
2,015
29.089552
76
h
null
pytorch-main/caffe2/sgd/ftrl_op.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { template <typename T> struct FtrlParams { explicit FtrlParams(OperatorBase* op) : alphaInv(1.0 / op->GetSingleArgument<float>("alpha", 0.005f)), beta(op->GetSingleArgument<float>("beta", 1.0f)), lambda1(op->GetSingleArgument<float>("lambda1", 0.001f)), lambda2(op->GetSingleArgument<float>("lambda2", 0.001f)) {} T alphaInv; T beta; T lambda1; T lambda2; }; // TODO(dzhulgakov): implement GPU version if necessary template <typename T, class Context> class FtrlOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; FtrlOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), params_(this) { CAFFE_ENFORCE( !HasArgument("alpha") || ALPHA >= InputSize(), "Cannot specify alpha by both input and argument"); } bool RunOnDevice() override; protected: FtrlParams<T> params_; INPUT_TAGS(VAR, N_Z, GRAD, ALPHA); OUTPUT_TAGS(OUTPUT_VAR, OUTPUT_N_Z); }; template <typename T> class SparseFtrlOp final : public Operator<CPUContext> { public: SparseFtrlOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), params_(this) { CAFFE_ENFORCE( !HasArgument("alpha") || ALPHA >= InputSize(), "Cannot specify alpha by both input and argument"); } bool RunOnDevice() override { // run time learning rate override if (ALPHA < InputSize()) { CAFFE_ENFORCE_EQ(Input(ALPHA).numel(), 1, "alpha should be real-valued"); params_.alphaInv = 1.0 / *(Input(ALPHA).template data<T>()); } // Use run-time polymorphism auto& indices = Input(INDICES); if (indices.template IsType<int32_t>()) { DoRun<int32_t>(); } else if (indices.template IsType<int64_t>()) { DoRun<int64_t>(); } else { LOG(FATAL) << "Unsupported type of INDICES in SparseFtrlOp: " << indices.dtype().name(); } return true; } protected: FtrlParams<T> params_; INPUT_TAGS(VAR, N_Z, INDICES, GRAD, ALPHA); OUTPUT_TAGS(OUTPUT_VAR, OUTPUT_N_Z); private: template <typename SIndex> void DoRun(); }; }
2,219
27.101266
79
h
null
pytorch-main/caffe2/sgd/gftrl_op.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { template <typename T> struct GFtrlParams { explicit GFtrlParams(OperatorBase* op) : alphaInv(1.0 / op->GetSingleArgument<float>("alpha", 0.005f)), beta(op->GetSingleArgument<float>("beta", 1.0f)), lambda1(op->GetSingleArgument<float>("lambda1", 0.001f)), lambda2(op->GetSingleArgument<float>("lambda2", 0.001f)) {} T alphaInv; T beta; T lambda1; T lambda2; }; template <typename T, class Context> class GFtrlOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; GFtrlOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), params_(this) { CAFFE_ENFORCE( !HasArgument("alpha") || ALPHA >= InputSize(), "Cannot specify alpha by both input and argument"); } bool RunOnDevice() override; protected: GFtrlParams<T> params_; INPUT_TAGS(VAR, N_Z, GRAD, ALPHA); OUTPUT_TAGS(OUTPUT_VAR, OUTPUT_N_Z); }; } // namespace caffe2
1,028
25.384615
70
h
null
pytorch-main/caffe2/sgd/iter_op.h
#ifndef CAFFE2_SGD_ITER_OP_H_ #define CAFFE2_SGD_ITER_OP_H_ #include <limits> #include <mutex> #include "caffe2/core/blob_serialization.h" #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/core/stats.h" namespace caffe2 { inline void IncrementIter(TensorCPU* output) { CAFFE_ENFORCE_EQ( output->numel(), 1, "The output of IterOp exists, but not of the right size."); int64_t* iter = output->template mutable_data<int64_t>(); CAFFE_ENFORCE(*iter >= 0, "Previous iteration number is negative."); CAFFE_ENFORCE( *iter < std::numeric_limits<int64_t>::max(), "Overflow will happen!"); (*iter)++; } // IterOp runs an iteration counter. I cannot think of a case where we would // need to access the iter variable on device, so this will always produce a // tensor on the CPU side. If the blob already exists and is a tensor<int64_t> // object, we will simply increment it (this emulates the case when we want to // resume training). Otherwise we will have the iter starting with 0. template <class Context> class IterOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; IterOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws) {} bool RunOnDevice() override { if (InputSize() == 0) { VLOG(1) << "[Input size is zero]"; if (!OperatorBase::OutputIsTensorType(0, CPU)) { // This is the first run; set the iter to start with 0. LOG(ERROR) << "You are using an old definition of IterOp that will " "be deprecated soon. More specifically, IterOp now " "requires an explicit in-place input and output."; VLOG(1) << "Initializing iter counter."; auto* output = OperatorBase::OutputTensor( 0, {1}, at::dtype<int64_t>().device(CPU)); output->template mutable_data<int64_t>()[0] = 0; } } IncrementIter(OperatorBase::Output<Tensor>(0, CPU)); return true; } }; template <class Context> class AtomicIterOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; AtomicIterOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), stats_(std::string("atomic_iter/stats/") + operator_def.input(1)) {} bool RunOnDevice() override { auto& mutex = OperatorBase::Input<std::unique_ptr<std::mutex>>(0); std::lock_guard<std::mutex> lg(*mutex); IncrementIter(OperatorBase::Output<Tensor>(0, CPU)); // NOLINTNEXTLINE(clang-diagnostic-unused-variable) CAFFE_EVENT(stats_, num_iter); return true; } private: struct AtomicIterOpStats { CAFFE_STAT_CTOR(AtomicIterOpStats); CAFFE_EXPORTED_STAT(num_iter); } stats_; }; class MutexSerializer : public BlobSerializerBase { public: /** * Serializes a std::unique_ptr<std::mutex>. Note that this blob has to * contain std::unique_ptr<std::mutex>, otherwise this function produces a * fatal error. */ void Serialize( const void* pointer, TypeMeta typeMeta, const string& name, BlobSerializerBase::SerializationAcceptor acceptor) override; }; class MutexDeserializer : public BlobDeserializerBase { public: void Deserialize(const BlobProto& proto, Blob* blob) override; }; } // namespace caffe2 #endif // CAFFE2_SGD_ITER_OP_H_
3,379
30.886792
78
h
null
pytorch-main/caffe2/sgd/lars_op.h
#ifndef CAFFE2_OPERATORS_LARS_OP_H_ #define CAFFE2_OPERATORS_LARS_OP_H_ #include "caffe2/core/context.h" #include "caffe2/core/logging.h" #include "caffe2/core/operator.h" #include "caffe2/utils/math.h" namespace caffe2 { template <typename T, class Context> class LarsOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; LarsOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), offset_(this->template GetSingleArgument<float>("offset", 0.5)), lr_min_(this->template GetSingleArgument<float>("lr_min", 0.02)) {} bool RunOnDevice() override { auto& X = Input(0); auto& dX = Input(1); CAFFE_ENFORCE( dX.numel() == X.numel(), "Gradient size doesn't match parameter size."); CAFFE_ENFORCE_GE(offset_, 0); CAFFE_ENFORCE_GE(lr_min_, 0); auto& wd = Input(2); auto& trust = Input(3); auto& lr_max = Input(4); auto* lr_rescaled = Output(0, vector<int64_t>{1}, at::dtype<T>()); ReinitializeTensor(&X_norm_tensor_, {1}, at::dtype<T>().device(Context::GetDeviceType())); T* X_norm_ = X_norm_tensor_.template mutable_data<T>(); ReinitializeTensor(&dX_norm_tensor_, {1}, at::dtype<T>().device(Context::GetDeviceType())); T* dX_norm_ = dX_norm_tensor_.template mutable_data<T>(); ComputeNorms( dX.numel(), X.template data<T>(), dX.template data<T>(), X_norm_, dX_norm_); ComputeLearningRate( wd.template data<T>(), trust.template data<T>(), lr_max.template data<T>(), offset_, lr_min_, X_norm_, dX_norm_, lr_rescaled->template mutable_data<T>()); return true; } private: // Compute the l2 norm of X_data and dX_data void ComputeNorms( int64_t N, const T* X_data, const T* dX_data, T* X_norm, T* dX_norm) { math::SumSqr(N, X_data, X_norm, &context_); math::Sqrt(1, X_norm, X_norm, &context_); math::SumSqr(N, dX_data, dX_norm, &context_); math::Sqrt(1, dX_norm, dX_norm, &context_); } // Compute the learning rate and apply clipping void ComputeLearningRate( const T* wd, const T* trust, const T* lr_max, T offset, T lr_min, T* X_norm, T* dX_norm, T* lr_rescaled); T offset_; T lr_min_; Tensor X_norm_tensor_; Tensor dX_norm_tensor_; }; } // namespace caffe2 #endif // CAFFE2_OPERATORS_LARS_OP_H_
2,479
25.382979
95
h
null
pytorch-main/caffe2/sgd/learning_rate_adaption_op.h
#pragma once #include <cfloat> #include <cmath> #include "caffe2/core/context.h" #include "caffe2/core/operator.h" #include "caffe2/utils/math.h" namespace caffe2 { template <typename Context> void lr_update( int n, const float* grad, const float* effgrad, const float* lr, float* nlr, float lr_alpha, bool normalized_lr_adaption, Context* /*context*/) { float x = 0; float y = 0, z = 0; const float kEps = 1e-12f; for (const auto i : c10::irange(n)) { x += grad[i] * effgrad[i]; if (normalized_lr_adaption) { y += grad[i] * grad[i]; z += effgrad[i] * effgrad[i]; } } if (normalized_lr_adaption) { y = fmax(std::sqrt(y), kEps); z = fmax(std::sqrt(z), kEps); nlr[0] = lr[0] * (1 - lr_alpha * x / (y * z)); } else { nlr[0] = lr[0] - lr_alpha * x; } } template <typename T, class Context> class LearningRateAdaptionOp final : public Operator<Context> { public: LearningRateAdaptionOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), lr_alpha_(this->template GetSingleArgument<float>("lr_alpha", 0.01f)), normalized_lr_adaption_(this->template GetSingleArgument<bool>( "normalized_lr_adaption", true)) {} USE_OPERATOR_CONTEXT_FUNCTIONS; bool RunOnDevice() override { CAFFE_ENFORCE(Input(LR).numel() == 1); CAFFE_ENFORCE(Input(GRAD).numel() == Input(EFFGRAD).numel()); Output(OUTPUT_LR)->ResizeLike(Input(LR)); lr_update<Context>( Input(GRAD).numel(), Input(GRAD).template data<T>(), Input(EFFGRAD).template data<T>(), Input(LR).template data<T>(), Output(OUTPUT_LR)->template mutable_data<T>(), lr_alpha_, normalized_lr_adaption_, &context_); return true; } protected: T lr_alpha_{1e-2}; bool normalized_lr_adaption_{true}; INPUT_TAGS(LR, GRAD, EFFGRAD); OUTPUT_TAGS(OUTPUT_LR); }; } // namespace caffe2
1,981
25.426667
78
h
null
pytorch-main/caffe2/sgd/learning_rate_functors.h
#ifndef CAFFE2_SGD_LEARNING_RATE_FUNCTORS_H_ #define CAFFE2_SGD_LEARNING_RATE_FUNCTORS_H_ #include <cmath> #include <list> #include <map> #ifdef _MSC_VER #ifndef _USE_MATH_DEFINES #define _USE_MATH_DEFINES #endif #include <math.h> #endif // _MSC_VER #include "caffe2/core/context.h" #include "caffe2/core/operator.h" namespace caffe2 { // LearningRateFunctor is a functor that when fed with an iter number, produces // the learning rate for the corresponding iteration. template <typename T> class LearningRateFunctor { public: virtual ~LearningRateFunctor() {} virtual T operator()(const int64_t iter) const = 0; }; // Fixed: not changing the learning rate at all. template <typename T> class FixedLearningRate : public LearningRateFunctor<T> { public: T operator()(const int64_t /*iter*/) const override { return 1.; } }; // Alter: alternatate learning rate with active_period and inactive_period. // update for a duration of active_period and then stop for a duration of // inactive_period if active_first, and vice versa template <typename T> class AlternateLearningRate : public LearningRateFunctor<T> { public: AlternateLearningRate( const int64_t active_period, const int64_t inactive_period, const bool active_first) : active_period_(active_period), inactive_period_(inactive_period), active_first_(active_first) {} T operator()(const int64_t iter) const override { if (iter % (active_period_ + inactive_period_) < (active_first_ ? active_period_ : inactive_period_)) { return active_first_ ? 1. : 0.; } else { return active_first_ ? 0. : 1.; }; }; int64_t active_period_; int64_t inactive_period_; bool active_first_; }; // Step: return gamma ^ (floor(iter / step)) template <typename T> class StepLearningRate : public LearningRateFunctor<T> { public: StepLearningRate(const int stepsize, const T gamma) : stepsize_(stepsize), gamma_(gamma) {} T operator()(const int64_t iter) const override { return std::pow(gamma_, static_cast<T>(iter / stepsize_)); } int stepsize_; T gamma_; }; // Exp: return gamma ^ iter template <typename T> class ExpLearningRate : public LearningRateFunctor<T> { public: explicit ExpLearningRate(const T gamma) : gamma_(gamma) {} T operator()(const int64_t iter) const override { return std::pow(gamma_, static_cast<T>(iter)); } T gamma_; }; // Gate: return multiplier_1 if before num_iter, else multiplier_2 template <typename T> class GateLearningRate : public LearningRateFunctor<T> { public: GateLearningRate( const T multiplier_1, const T multiplier_2, const int64_t num_iter) : multiplier_1_(multiplier_1), multiplier_2_(multiplier_2), num_iter_(num_iter) {} T operator()(const int64_t iter) const override { if (iter >= int64_t(num_iter_)) { return T(multiplier_2_); } return T(multiplier_1_); } T multiplier_1_; T multiplier_2_; uint64_t num_iter_; }; // Inv: return (1 + gamma * iter) ^ (-power) template <typename T> class InvLearningRate : public LearningRateFunctor<T> { public: InvLearningRate(const T gamma, const T power) : gamma_(gamma), power_(power) {} T operator()(const int64_t iter) const override { return std::pow(T(1) + gamma_ * iter, -power_); } T gamma_; T power_; }; // Poly: return (1 - iter/max_iter) ^ (power) template <typename T> class PolyLearningRate : public LearningRateFunctor<T> { public: PolyLearningRate(const T power, const int64_t max_iter) : power_(power), max_iter_(max_iter) {} T operator()(const int64_t iter) const override { return std::pow(1 - T(iter) / T(max_iter_), power_); } T power_; uint64_t max_iter_; }; // LinearWarmup: return max(iter/num_iter, 1) template <typename T> class LinearWarmupLearningRate : public LearningRateFunctor<T> { public: LinearWarmupLearningRate(const T start_multiplier, const int64_t num_iter) : start_multiplier_(start_multiplier), num_iter_(num_iter) {} T operator()(const int64_t iter) const override { if (iter >= int64_t(num_iter_)) { return 1.; } return start_multiplier_ + (1. - start_multiplier_) * T(iter) / T(num_iter_); } T start_multiplier_; uint64_t num_iter_; }; // ConstantWarmup: return scale when iter < num_iter, and 1 otherwise template <typename T> class ConstantWarmupLearningRate : public LearningRateFunctor<T> { public: ConstantWarmupLearningRate(const T multiplier, const int64_t num_iter) : multiplier_(multiplier), num_iter_(num_iter) {} T operator()(const int64_t iter) const override { if (iter >= int64_t(num_iter_)) { return 1.; } return T(multiplier_); } T multiplier_; uint64_t num_iter_; }; // ConstantWarmup: return scale when iter < num_iter, and 1 otherwise template <typename T> class PieceWarmupLearningRate : public LearningRateFunctor<T> { public: PieceWarmupLearningRate( const T m1, const int64_t n1, const T m2, const int64_t n2, const T m3) : m1_(m1), m2_(m2), m3_(m3), n1_(n1), n2_(n2){}; T operator()(const int64_t iter) const override { if (iter < int64_t(n1_)) { return m1_; } else if (iter < int64_t(n2_)) { return m2_; } return m3_; } T m1_, m2_, m3_; uint64_t n1_, n2_; }; // hill: the learning rate changes according to following 3 stages // 1) linear warmup (increasing) at first num_iter steps from start_multiplier // 2) inverse shrink (decreasing) afterwards (gamma, power) // 3) lower bounded by end_multiplier template <typename T> class HillLearningRate : public LearningRateFunctor<T> { public: HillLearningRate( const int64_t num_iter, const T start_multiplier, const T gamma, const T power, const T end_multiplier) : linear_warmup_lr_(start_multiplier, num_iter), inv_lr_(gamma, power), num_iter_(num_iter), end_multiplier_(end_multiplier) {} T operator()(const int64_t iter) const override { if (iter < num_iter_) { return linear_warmup_lr_(iter); } else { return std::max(end_multiplier_, inv_lr_(iter - num_iter_)); } } LinearWarmupLearningRate<T> linear_warmup_lr_; InvLearningRate<T> inv_lr_; int64_t num_iter_; T end_multiplier_; }; // slope: the learning rate changes according to 2 stages // 1) constantWarmup with multiplier_1 // 2) linearly shink to multiplier_2: // max{ // multiplier_1 + (iter - num_iter_1) * (multiplier_2 - multiplier_1) / (num_iter_2 - num_iter_1), // multiplier_2 // } template <typename T> class SlopeLearningRate : public LearningRateFunctor<T> { public: SlopeLearningRate( const int64_t num_iter_1, const T multiplier_1, const T num_iter_2, const T multiplier_2) : num_iter_1_(num_iter_1), multiplier_1_(multiplier_1), num_iter_2_(num_iter_2), multiplier_2_(multiplier_2) {} T operator()(const int64_t iter) const override { if (iter < num_iter_1_) { return multiplier_1_; } else { return std::max( multiplier_2_, multiplier_1_ + (iter - num_iter_1_) * (multiplier_2_ - multiplier_1_) / (num_iter_2_ - num_iter_1_) ); } } int64_t num_iter_1_; T multiplier_1_; int64_t num_iter_2_; T multiplier_2_; }; template <typename T> class CompositeLearningRateItem { public: CompositeLearningRateItem( int64_t num_iter, float lr_scale, LearningRateFunctor<T>* policy) : num_iter_(num_iter), lr_scale_(lr_scale), policy_(policy) {} int64_t num_iter_; float lr_scale_; LearningRateFunctor<T>* policy_; }; // composite: the learning policy changes according to current iteration # template <typename T> class CompositeLearningRate : public LearningRateFunctor<T> { public: CompositeLearningRate( const std::list<CompositeLearningRateItem<T>>& sub_policies) { TORCH_DCHECK_GT(sub_policies.size(), 0); int64_t num_iter_start = 1; for (auto it = sub_policies.begin(); it != sub_policies.end(); ++it) { TORCH_DCHECK_GT(it->num_iter_, 0); sub_policies_[num_iter_start].reset(it->policy_); sub_policy_lr_scales_[num_iter_start] = it->lr_scale_; num_iter_start += it->num_iter_; } } T operator()(const int64_t iter) const override { auto sub_policy = sub_policies_.upper_bound(iter); DCHECK(sub_policy != sub_policies_.begin()); --sub_policy; auto sub_policy_lr_scale = sub_policy_lr_scales_.upper_bound(iter); DCHECK(sub_policy_lr_scale != sub_policy_lr_scales_.begin()); --sub_policy_lr_scale; return ((*sub_policy->second)(iter)) * (sub_policy_lr_scale->second); } private: std::map<int64_t, std::unique_ptr<LearningRateFunctor<T>>> sub_policies_; std::map<int64_t, float> sub_policy_lr_scales_; }; // Cyclical: return a learning rate with period 2 * stepsize and // lower bound base_lr, upper bound max_lr. // See https://arxiv.org/pdf/1506.01186.pdf template <typename T> class CyclicalLearningRate : public LearningRateFunctor<T> { public: CyclicalLearningRate( const T base_lr, const T max_lr, const int stepsize, const T decay) : base_lr_(base_lr), max_lr_(max_lr), stepsize_(stepsize), decay_(decay) {} T operator()(const int64_t iter) const override { int64_t cycle = static_cast<int>((iter / (2 * stepsize_)) + 1); T x = std::abs(static_cast<T>(iter) / stepsize_ - 2 * cycle + 1); return 1 + (T(std::abs(max_lr_)) / T(std::abs(base_lr_)) - 1) * std::max(T(0.0), (1 - x)) * std::pow(decay_, static_cast<int>(iter / (2 * stepsize_))); } T base_lr_; T max_lr_; int stepsize_; T decay_; }; // Cosine: return a learning rate with a cosine schedule // lower bound min_lr, upper bound max_lr. // See https://arxiv.org/pdf/1608.03983.pdf template <typename T> class CosineLearningRate : public LearningRateFunctor<T> { public: CosineLearningRate( const T min_lr, const T max_lr, const int64_t period, const T t_mult, const T lr_shrink) : min_lr_(min_lr), max_lr_(max_lr), period_(period), t_mult_(t_mult), lr_shrink_(lr_shrink) {} T operator()(const int64_t iter) const override { T i, t_i, t_curr; if (t_mult_ != 1.0) { // the period is changed every time i = floor( log(1 - double(iter) / double(period_) * (1.0 - t_mult_)) / log(t_mult_)); t_i = pow(t_mult_, i) * period_; t_curr = iter - (1.0 - pow(t_mult_, i)) / (1.0 - t_mult_) * period_; } else { // fixed period i = floor(double(iter) / double(period_)); t_i = period_; t_curr = iter - t_i * i; } T lr_shrink = pow(lr_shrink_, i); T min_lr = min_lr_ * lr_shrink; T max_lr = max_lr_ * lr_shrink; T final_lr = min_lr + 0.5 * (max_lr - min_lr) * (1 + cos(M_PI * t_curr / t_i)); return final_lr; } T min_lr_; T max_lr_; int64_t period_; T t_mult_; T lr_shrink_; }; // constantThenLinearWarmup: first use a constant multiplier // and then ramp up to the global lr template <typename T> class ConstantThenLinearWarmupLearningRate : public LearningRateFunctor<T> { public: ConstantThenLinearWarmupLearningRate( const T start_warmup_multiplier, const int64_t constant_warmup_num_iter, const int64_t linear_warmup_num_iter) : constant_warmup_num_iter_(constant_warmup_num_iter), linear_warmup_num_iter_(linear_warmup_num_iter), constant_warmup_lr_(start_warmup_multiplier, constant_warmup_num_iter), linear_warmup_lr_(start_warmup_multiplier, linear_warmup_num_iter) {} T operator()(const int64_t iter) const override { if (iter < constant_warmup_num_iter_) { return constant_warmup_lr_(iter); } else if (iter < constant_warmup_num_iter_ + linear_warmup_num_iter_) { return linear_warmup_lr_(iter - constant_warmup_num_iter_); } else { return 1.0; } } int64_t constant_warmup_num_iter_; int64_t linear_warmup_num_iter_; ConstantWarmupLearningRate<T> constant_warmup_lr_; LinearWarmupLearningRate<T> linear_warmup_lr_; }; // CompositeCosineLearningRate: first use a constant multiplier // and then ramp up to the global lr, and then use a cosine learning rate template <typename T> class CompositeCosineLearningRate : public LearningRateFunctor<T> { public: CompositeCosineLearningRate( const T start_warmup_multiplier, const int64_t constant_warmup_num_iter, const int64_t linear_warmup_num_iter, const T cosine_min_lr, const T cosine_max_lr, const int64_t cosine_period, const T consine_t_mult, const T cosine_lr_shrink) : constant_warmup_num_iter_(constant_warmup_num_iter), linear_warmup_num_iter_(linear_warmup_num_iter), constant_then_linear_warmup_lr_( start_warmup_multiplier, constant_warmup_num_iter, linear_warmup_num_iter), cosine_lr_( cosine_min_lr, cosine_max_lr, cosine_period, consine_t_mult, cosine_lr_shrink) {} T operator()(const int64_t iter) const override { if (iter < constant_warmup_num_iter_ + linear_warmup_num_iter_) { return constant_then_linear_warmup_lr_(iter); } return cosine_lr_( iter - constant_warmup_num_iter_ - linear_warmup_num_iter_); } int64_t constant_warmup_num_iter_; int64_t linear_warmup_num_iter_; ConstantThenLinearWarmupLearningRate<T> constant_then_linear_warmup_lr_; CosineLearningRate<T> cosine_lr_; }; // CompositeCyclicalLearningRate: first use a constant multiplier // and then ramp up to the global lr, and then use a cyclical learning rate template <typename T> class CompositeCyclicalLearningRate : public LearningRateFunctor<T> { public: CompositeCyclicalLearningRate( const T base_lr, const T start_warmup_multiplier, const int64_t constant_warmup_num_iter, const int64_t linear_warmup_num_iter, const T cyclical_max_lr, const int cyclical_step_size, const T cyclical_decay) : constant_warmup_num_iter_(constant_warmup_num_iter), linear_warmup_num_iter_(linear_warmup_num_iter), constant_then_linear_warmup_lr_( start_warmup_multiplier, constant_warmup_num_iter, linear_warmup_num_iter), cyclical_lr_( base_lr, cyclical_max_lr, cyclical_step_size, cyclical_decay) {} T operator()(const int64_t iter) const override { if (iter < constant_warmup_num_iter_ + linear_warmup_num_iter_) { return constant_then_linear_warmup_lr_(iter); } return cyclical_lr_( iter - constant_warmup_num_iter_ - linear_warmup_num_iter_); } int64_t constant_warmup_num_iter_; int64_t linear_warmup_num_iter_; ConstantThenLinearWarmupLearningRate<T> constant_then_linear_warmup_lr_; CyclicalLearningRate<T> cyclical_lr_; }; } // namespace caffe2 #endif // CAFFE2_SGD_LEARNING_RATE_FUNCTORS_H_
15,158
29.873727
108
h
null
pytorch-main/caffe2/sgd/learning_rate_op.h
#ifndef CAFFE2_SGD_LEARNING_RATE_OP_H_ #define CAFFE2_SGD_LEARNING_RATE_OP_H_ #include <cfloat> #include <cmath> #include "caffe2/core/context.h" #include "caffe2/core/export_caffe2_op_to_c10.h" #include <c10/util/irange.h> #include "caffe2/core/operator.h" #include "caffe2/sgd/learning_rate_functors.h" C10_DECLARE_EXPORT_CAFFE2_OP_TO_C10(LearningRate); namespace caffe2 { template <typename T, class Context> class LearningRateOp final : public Operator<Context> { public: template <class... Args> LearningRateOp(Args&&... args) : Operator<Context>(std::forward<Args>(args)...), functor_(nullptr), base_lr_(this->template GetSingleArgument<float>("base_lr", FLT_MAX)) { CAFFE_ENFORCE_NE(base_lr_, FLT_MAX, "Base learning rate must be set."); const string policy = this->template GetSingleArgument<string>("policy", ""); CAFFE_ENFORCE(policy.size(), "Must specify a learning rate policy."); functor_.reset(createLearningRateFunctor(policy)); } USE_OPERATOR_CONTEXT_FUNCTIONS; bool RunOnDevice() override { int64_t iter = OperatorBase::Input<Tensor>(0, CPU).template data<int64_t>()[0]; T learning_rate = base_lr_ * (*functor_)(iter); // Write to output. auto* output = Output(0); output->Resize(vector<int64_t>()); context_.template CopyFromCPU<T>( 1, &learning_rate, Output(0)->template mutable_data<T>()); return true; } private: unique_ptr<LearningRateFunctor<T>> functor_; T base_lr_; LearningRateFunctor<T>* createLearningRateFunctor( const string& policy, const string& arg_prefix = "") { if (policy == "fixed") { return new FixedLearningRate<T>(); } else if (policy == "alter") { bool active_first = this->template GetSingleArgument<bool>( arg_prefix + "active_first", true); int64_t active_period = this->template GetSingleArgument<int64_t>( arg_prefix + "active_period", -1); int64_t inactive_period = this->template GetSingleArgument<int64_t>( arg_prefix + "inactive_period", -1); TORCH_DCHECK_GE(active_period, 0); TORCH_DCHECK_GE(inactive_period, 0); return new AlternateLearningRate<T>( active_period, inactive_period, active_first); } else if (policy == "hill") { int64_t num_iter = this->template GetSingleArgument<int64_t>(arg_prefix + "num_iter", 0); TORCH_DCHECK_GT(num_iter, 0); T start_multiplier = this->template GetSingleArgument<float>( arg_prefix + "start_multiplier", 0.); TORCH_DCHECK_GE(start_multiplier, 0); // start_multiplier in range [0, 1] TORCH_DCHECK_LE(start_multiplier, 1); T gamma = this->template GetSingleArgument<float>(arg_prefix + "gamma", 0); TORCH_DCHECK_GT(gamma, 0); T power = this->template GetSingleArgument<float>(arg_prefix + "power", 0); TORCH_DCHECK_GT(power, 0); T end_multiplier = this->template GetSingleArgument<float>( arg_prefix + "end_multiplier", 0); TORCH_DCHECK_GE(end_multiplier, 0); // end_multiplier in range [0, 1] TORCH_DCHECK_LE(end_multiplier, 1); return new HillLearningRate<T>( num_iter, start_multiplier, gamma, power, end_multiplier); } else if (policy == "slope") { int64_t num_iter_1 = this->template GetSingleArgument<int64_t>( arg_prefix + "num_iter_1", 0); TORCH_DCHECK_GT(num_iter_1, 0); T multiplier_1 = this->template GetSingleArgument<float>( arg_prefix + "multiplier_1", 0.); int64_t num_iter_2 = this->template GetSingleArgument<int64_t>( arg_prefix + "num_iter_2", 0); TORCH_DCHECK_GT(num_iter_1, 0); T multiplier_2 = this->template GetSingleArgument<float>( arg_prefix + "multiplier_2", 0.); TORCH_DCHECK_GT(num_iter_2, num_iter_1); return new SlopeLearningRate<T>( num_iter_1, multiplier_1, num_iter_2, multiplier_2); } else if (policy == "step") { int stepsize = this->template GetSingleArgument<int>(arg_prefix + "stepsize", 0); T gamma = this->template GetSingleArgument<float>(arg_prefix + "gamma", 0); TORCH_DCHECK_GT(stepsize, 0); TORCH_DCHECK_GT(gamma, 0); return new StepLearningRate<T>(stepsize, gamma); } else if (policy == "exp") { T gamma = this->template GetSingleArgument<float>(arg_prefix + "gamma", 0); TORCH_DCHECK_GT(gamma, 0); return new ExpLearningRate<T>(gamma); } else if (policy == "gate") { T multiplier_1 = this->template GetSingleArgument<float>( arg_prefix + "multiplier_1", 1); T multiplier_2 = this->template GetSingleArgument<float>( arg_prefix + "multiplier_2", 1); int num_iter = this->template GetSingleArgument<int>(arg_prefix + "num_iter", 0); // no constraint on the range of multiplier_1 and multiplier_2 return new GateLearningRate<T>(multiplier_1, multiplier_2, num_iter); } else if (policy == "inv") { T gamma = this->template GetSingleArgument<float>(arg_prefix + "gamma", 0); T power = this->template GetSingleArgument<float>(arg_prefix + "power", 0); TORCH_DCHECK_GT(gamma, 0); TORCH_DCHECK_GT(power, 0); return new InvLearningRate<T>(gamma, power); } else if (policy == "poly") { int max_iter = this->template GetSingleArgument<int>(arg_prefix + "max_iter", -1); T power = this->template GetSingleArgument<float>(arg_prefix + "power", 0); TORCH_DCHECK_GT(power, 0); return new PolyLearningRate<T>(power, max_iter); } else if (policy == "linearWarmup") { T start_multiplier = this->template GetSingleArgument<float>( arg_prefix + "start_multiplier", 0.); int num_iter = this->template GetSingleArgument<int>(arg_prefix + "num_iter", 0); TORCH_DCHECK_GE(start_multiplier, 0); return new LinearWarmupLearningRate<T>(start_multiplier, num_iter); } else if (policy == "constantWarmup") { T multiplier = this->template GetSingleArgument<float>( arg_prefix + "multiplier", 0.5); int num_iter = this->template GetSingleArgument<int>(arg_prefix + "num_iter", 0); TORCH_DCHECK_GT(multiplier, 0); return new ConstantWarmupLearningRate<T>(multiplier, num_iter); } else if (policy == "pieceWarmup") { T m1 = this->template GetSingleArgument<float>(arg_prefix + "m1", 0.5); int64_t n1 = this->template GetSingleArgument<int64_t>(arg_prefix + "n1", 0); T m2 = this->template GetSingleArgument<float>(arg_prefix + "m2", 0.5); int64_t n2 = this->template GetSingleArgument<int64_t>(arg_prefix + "n2", 0); T m3 = this->template GetSingleArgument<float>(arg_prefix + "m3", 0.5); return new PieceWarmupLearningRate<T>(m1, n1, m2, n2, m3); } else if (policy == "composite") { std::vector<int> sub_policy_num_iters = this->template GetRepeatedArgument<int>("sub_policy_num_iters"); std::list<CompositeLearningRateItem<T>> sub_policies; CAFFE_ENFORCE_GT( sub_policy_num_iters.size(), 0, "Must specify at least one sub learning rate policy."); for (const auto i : c10::irange(sub_policy_num_iters.size())) { CAFFE_ENFORCE_GT( sub_policy_num_iters[i], 0, "The number of iterations for sub learning rate policy should be positive."); std::stringstream sub_policy_arg_prefix; sub_policy_arg_prefix << "sub_policy_" << i << "_"; const string sub_policy_arg_prefix_str = sub_policy_arg_prefix.str(); const string sub_policy = this->template GetSingleArgument<string>( sub_policy_arg_prefix_str + "policy", ""); if (sub_policy == "composite") { CAFFE_THROW( "Defining composite LR policy as a subpolicy of composite LR " "policy is not allowed."); } const float scale_lr = this->template GetSingleArgument<float>( sub_policy_arg_prefix_str + "lr_scale", 1.0); sub_policies.push_back(CompositeLearningRateItem<T>( sub_policy_num_iters[i], scale_lr, createLearningRateFunctor(sub_policy, sub_policy_arg_prefix_str))); } return new CompositeLearningRate<T>(sub_policies); } else if (policy == "cyclical") { T max_lr = this->template GetSingleArgument<float>(arg_prefix + "max_lr", 0.005); int stepsize = this->template GetSingleArgument<int>(arg_prefix + "stepsize", 0); T decay = this->template GetSingleArgument<float>(arg_prefix + "decay", 1.0); TORCH_DCHECK_GT(stepsize, 0); TORCH_DCHECK_GE(max_lr, base_lr_); return new CyclicalLearningRate<T>(base_lr_, max_lr, stepsize, decay); } else if (policy == "constantThenLinearWarmup") { T start_warmup_multiplier = this->template GetSingleArgument<float>( arg_prefix + "start_warmup_multiplier", 0.1); int64_t constant_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "constant_warmup_num_iter", 10000000); int64_t linear_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "linear_warmup_num_iter", 10000000); return new ConstantThenLinearWarmupLearningRate<T>( start_warmup_multiplier, constant_warmup_num_iter, linear_warmup_num_iter); } else if (policy == "compositeCyclical") { T start_warmup_multiplier = this->template GetSingleArgument<float>( arg_prefix + "start_warmup_multiplier", 0.1); int64_t constant_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "constant_warmup_num_iter", 10000000); int64_t linear_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "linear_warmup_num_iter", 10000000); T cyclical_max_lr = this->template GetSingleArgument<float>( arg_prefix + "cyclical_max_lr", 0.05); int cyclical_step_size = this->template GetSingleArgument<int>( arg_prefix + "cyclical_step_size", 1000000); T cyclical_decay = this->template GetSingleArgument<float>( arg_prefix + "cyclical_decay", 1.0); TORCH_DCHECK_GE(cyclical_max_lr, base_lr_); return new CompositeCyclicalLearningRate<T>( base_lr_, start_warmup_multiplier, constant_warmup_num_iter, linear_warmup_num_iter, cyclical_max_lr, cyclical_step_size, cyclical_decay); } else if (policy == "cosine") { T max_lr = this->template GetSingleArgument<float>(arg_prefix + "max_lr", 0.5); T min_lr = this->template GetSingleArgument<float>(arg_prefix + "min_lr", 0.1); int64_t period = this->template GetSingleArgument<int>(arg_prefix + "period", 50); T t_mult = this->template GetSingleArgument<float>(arg_prefix + "t_mult", 1.0); T lr_shrink = this->template GetSingleArgument<float>( arg_prefix + "lr_shrink", 0.99); TORCH_DCHECK_GE(max_lr, min_lr); return new CosineLearningRate<T>( min_lr, max_lr, period, t_mult, lr_shrink); } else if (policy == "compositeCosine") { T start_warmup_multiplier = this->template GetSingleArgument<float>( arg_prefix + "start_warmup_multiplier", 0.1); int64_t constant_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "constant_warmup_num_iter", 10000000); int64_t linear_warmup_num_iter = this->template GetSingleArgument<int64_t>( arg_prefix + "linear_warmup_num_iter", 10000000); T cosine_max_lr = this->template GetSingleArgument<float>( arg_prefix + "cosine_max_lr", 0.5); T cosine_min_lr = this->template GetSingleArgument<float>( arg_prefix + "cosine_min_lr", 0.1); int64_t cosine_period = this->template GetSingleArgument<int>( arg_prefix + "cosine_period", 50); T cosine_t_mult = this->template GetSingleArgument<float>( arg_prefix + "cosine_t_mult", 1.0); T cosine_lr_shrink = this->template GetSingleArgument<float>( arg_prefix + "cosine_lr_shrink", 0.99); TORCH_DCHECK_GE(cosine_max_lr, cosine_min_lr); return new CompositeCosineLearningRate<T>( start_warmup_multiplier, constant_warmup_num_iter, linear_warmup_num_iter, cosine_min_lr, cosine_max_lr, cosine_period, cosine_t_mult, cosine_lr_shrink); } else { CAFFE_THROW("Unknown learning rate policy: ", policy); return NULL; } } }; } // namespace caffe2 #endif // CAFFE2_SGD_LEARNING_RATE_OP_H_
12,880
44.355634
89
h
null
pytorch-main/caffe2/sgd/momentum_sgd_op.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { template <typename Context> void momentum_sgd_update( const int N, const float* g, const float* m, float* ng, float* nm, const float* lr, const float momentum, const bool nesterov, float* param, Context* /*context*/) { const float LR = lr[0]; for (const auto i : c10::irange(N)) { if (!nesterov) { const float adjusted_gradient = LR * g[i] + momentum * m[i]; nm[i] = adjusted_gradient; ng[i] = adjusted_gradient; } else { const float mi = m[i]; const float mi_new = momentum * mi + LR * g[i]; nm[i] = mi_new; ng[i] = (1 + momentum) * mi_new - momentum * mi; } if (param) { param[i] -= ng[i]; } } } template <typename T, class Context> class MomentumSGDOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; MomentumSGDOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), momentum_(this->template GetSingleArgument<T>("momentum", 0.0)), nesterov_(this->template GetSingleArgument<bool>("nesterov", false)) {} bool RunOnDevice() override { auto device_type = Context::GetDeviceType(); // Iter live on the CPU CAFFE_ENFORCE(OperatorBase::InputIsTensorType(GRAD, device_type)); CAFFE_ENFORCE(OperatorBase::InputIsTensorType(MOMENTUM, device_type)); CAFFE_ENFORCE(Input(LR).numel() == 1); CAFFE_ENFORCE(Input(GRAD).numel() == Input(MOMENTUM).numel()); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_MOMENTUM)->ResizeLike(Input(MOMENTUM)); momentum_sgd_update<Context>( Input(GRAD).numel(), Input(GRAD).template data<T>(), Input(MOMENTUM).template data<T>(), Output(OUTPUT_GRAD)->template mutable_data<T>(), Output(OUTPUT_MOMENTUM)->template mutable_data<T>(), Input(LR).template data<T>(), momentum_, nesterov_, NULL, &context_); return true; } protected: T momentum_{0.9}; bool nesterov_; INPUT_TAGS(GRAD, MOMENTUM, LR); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MOMENTUM); }; template <typename T, class Context> class MomentumSGDUpdateOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; MomentumSGDUpdateOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), momentum_(this->template GetSingleArgument<T>("momentum", 0.0)), nesterov_(this->template GetSingleArgument<bool>("nesterov", false)) {} bool RunOnDevice() override { auto device_type = Context::GetDeviceType(); // Iter live on the CPU CAFFE_ENFORCE(OperatorBase::InputIsTensorType(GRAD, device_type)); CAFFE_ENFORCE(OperatorBase::InputIsTensorType(MOMENTUM, device_type)); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ(Input(GRAD).numel(), Input(MOMENTUM).numel()); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_MOMENTUM)->ResizeLike(Input(MOMENTUM)); momentum_sgd_update<Context>( Input(GRAD).numel(), Input(GRAD).template data<T>(), Input(MOMENTUM).template data<T>(), Output(OUTPUT_GRAD)->template mutable_data<T>(), Output(OUTPUT_MOMENTUM)->template mutable_data<T>(), Input(LR).template data<T>(), momentum_, nesterov_, Output(OUTPUT_PARAM)->template mutable_data<T>(), &context_); return true; } protected: T momentum_{0.9}; bool nesterov_; INPUT_TAGS(GRAD, MOMENTUM, LR, PARAM); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MOMENTUM, OUTPUT_PARAM); }; template <typename T, class Context> class SparseMomentumSGDUpdateOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; SparseMomentumSGDUpdateOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), momentum_(this->template GetSingleArgument<T>("momentum", 0.0)), nesterov_(this->template GetSingleArgument<bool>("nesterov", false)) {} bool RunOnDevice() override { // Resize [potentially] out-of-place blobs Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); // Enforce shapes CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ(Input(PARAM).numel(), Input(MOMENTUM).numel()); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { auto block_size = Input(PARAM).numel() / Input(PARAM).size(0); auto n = Input(GRAD).numel() / block_size; const auto* gradIn = Input(GRAD).template data<T>(); const auto* momentumIn = Input(MOMENTUM).template data<T>(); const auto* lr = Input(LR).template data<T>(); // const auto* paramIn = Input(PARAM).template data<T>(); const auto* indices = Input(INDICES).template data<SIndex>(); auto* gradOut = Output(OUTPUT_GRAD)->template mutable_data<T>(); auto* momentumOut = Output(OUTPUT_MOMENTUM)->template mutable_data<T>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<T>(); for (const auto i : c10::irange(n)) { auto idx = indices[i]; auto offsetI = i * block_size; auto offsetIdx = idx * block_size; CAFFE_ENFORCE(offsetIdx + block_size <= Input(PARAM).numel()); CAFFE_ENFORCE(offsetI + block_size <= Input(GRAD).numel()); momentum_sgd_update<Context>( block_size, gradIn + offsetI, momentumIn + offsetIdx, gradOut + offsetI, momentumOut + offsetIdx, lr, momentum_, nesterov_, paramOut + offsetIdx, &context_); } return true; } protected: T momentum_; bool nesterov_; INPUT_TAGS(GRAD, MOMENTUM, LR, PARAM, INDICES); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MOMENTUM, OUTPUT_PARAM); }; } // namespace caffe2
6,049
31.352941
79
h
null
pytorch-main/caffe2/sgd/rmsprop_op.h
#pragma once #include "caffe2/core/common_omp.h" #include "caffe2/core/operator.h" namespace caffe2 { template <typename Context> void rmsprop_update( int N, const float* g, const float* ms, const float* mom, float* ng, float* nms, float* nmom, float decay, float momentum, float epsilon, const float* lr, Context* context); template <typename T, class Context> class RmsPropOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; RmsPropOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), decay_(this->template GetSingleArgument<float>("decay", 0.9f)), momentum_(this->template GetSingleArgument<float>("momentum", 0.0f)), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)) {} bool RunOnDevice() override { CAFFE_ENFORCE(Input(LR).numel() == 1); CAFFE_ENFORCE(Input(GRAD).numel() == Input(MEAN_SQUARES).numel()); CAFFE_ENFORCE(Input(GRAD).numel() == Input(OUTPUT_MOMENTUM).numel()); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_GRAD)->ResizeLike(Input(GRAD)); Output(OUTPUT_MEAN_SQUARES)->ResizeLike(Input(MEAN_SQUARES)); Output(OUTPUT_MOMENTUM)->ResizeLike(Input(MOMENTUM)); rmsprop_update<Context>( Input(GRAD).numel(), Input(GRAD).template data<T>(), Input(MEAN_SQUARES).template data<T>(), Input(MOMENTUM).template data<T>(), Output(OUTPUT_GRAD)->template mutable_data<T>(), Output(OUTPUT_MEAN_SQUARES)->template mutable_data<T>(), Output(OUTPUT_MOMENTUM)->template mutable_data<T>(), decay_, momentum_, epsilon_, Input(LR).template data<T>(), &context_); return true; } protected: T decay_{0.9}; T momentum_{0.0}; T epsilon_{1e-8}; INPUT_TAGS(GRAD, MEAN_SQUARES, MOMENTUM, LR); OUTPUT_TAGS(OUTPUT_GRAD, OUTPUT_MEAN_SQUARES, OUTPUT_MOMENTUM); }; }
1,980
29.953125
78
h
null
pytorch-main/caffe2/sgd/rowwise_counter.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { class RowWiseCounterOp final : public Operator<CPUContext> { public: RowWiseCounterOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws), counter_halflife_( this->template GetSingleArgument<int64_t>("counter_halflife", -1)), counter_neg_log_rho_(0.0) { if (counter_halflife_ > 0) { counter_neg_log_rho_ = std::log(2.0) / counter_halflife_; } } bool RunOnDevice() override { CAFFE_ENFORCE_EQ(Input(PREV_ITER).numel(), Input(COUNTER).numel()); CAFFE_ENFORCE_EQ(Input(ITER).numel(), 1); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { auto* prev_iter = Output(OUTPUT_PREV_ITER)->template mutable_data<int64_t>(); auto* counter = Output(OUTPUT_COUNTER)->template mutable_data<double>(); const int64_t curr_iter = Input(ITER).template data<int64_t>()[0]; const auto* indices = Input(INDICES).template data<SIndex>(); auto n = Input(INDICES).numel(); if (n == 0) { return true; } if (counter_halflife_ <= 0) { return true; } for (const auto i : c10::irange(n)) { const std::size_t idx = indices[i]; CAFFE_ENFORCE_GE( Input(COUNTER).numel(), idx, this->debug_def().input(COUNTER), ", out of bound, idx:", idx, " for input i:", i, " max size:", Input(COUNTER).numel()); const int64_t iter_delta = std::max<int64_t>(0, curr_iter - prev_iter[idx]); counter[idx] = 1.0 + std::exp(-iter_delta * counter_neg_log_rho_) * counter[idx]; prev_iter[idx] = std::max<int64_t>(curr_iter, prev_iter[idx]); } return true; } protected: int64_t counter_halflife_; double counter_neg_log_rho_; INPUT_TAGS(PREV_ITER, COUNTER, INDICES, ITER); OUTPUT_TAGS(OUTPUT_PREV_ITER, OUTPUT_COUNTER); }; } // namespace caffe2
2,090
27.643836
79
h
null
pytorch-main/caffe2/sgd/storm_op.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { template <typename Context> void storm_update( const int N, const float* paramIn, const float* momentIn, const float* gradSqSumIn, const float* gradIn, const float* lr, float* paramOut, float* momentOut, float* gradSqSumOut, const float momentum, const float beta, Context* /*context*/) { float gradSqSumTmp = 0.0; for (const auto i : c10::irange(N)) { const float gi = gradIn[i]; gradSqSumTmp += gi * gi; } gradSqSumOut[0] = gradSqSumIn[0] + gradSqSumTmp; const float nlr = lr[0] * std::pow(beta + gradSqSumOut[0], -1.0 / 3.0); const float alpha = momentum * nlr * nlr; for (const auto i : c10::irange(N)) { const float gi = gradIn[i]; const float mi = momentIn[i]; float new_mi = momentOut[i] = gi + (1.0 - alpha) * (mi - gi); paramOut[i] = paramIn[i] + nlr * new_mi; } } template <class Context> class StormOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; StormOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), OP_SINGLE_ARG(float, "momentum", momentum_, 10.0), OP_SINGLE_ARG(float, "beta", beta_, 0.1) {} bool RunOnDevice() override { // Enforce shapes CAFFE_ENFORCE_EQ(Input(GRAD).numel(), Input(PARAM).numel()); CAFFE_ENFORCE_EQ(Input(GRAD).numel(), Input(MOMENT).numel()); CAFFE_ENFORCE_EQ(Input(GRADSQSUM).numel(), 1); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); // Resize [potentially] out-of-place blobs Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_MOMENT)->ResizeLike(Input(MOMENT)); Output(OUTPUT_GRAGSQSUM)->ResizeLike(Input(GRADSQSUM)); storm_update<Context>( Input(GRAD).numel(), Input(PARAM).template data<float>(), Input(MOMENT).template data<float>(), Input(GRADSQSUM).template data<float>(), Input(GRAD).template data<float>(), Input(LR).template data<float>(), Output(OUTPUT_PARAM)->template mutable_data<float>(), Output(OUTPUT_MOMENT)->template mutable_data<float>(), Output(OUTPUT_GRAGSQSUM)->template mutable_data<float>(), momentum_, beta_, &context_); return true; } protected: const float momentum_; const float beta_; INPUT_TAGS(PARAM, MOMENT, GRADSQSUM, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT, OUTPUT_GRAGSQSUM); }; template <class Context> class SparseStormOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; SparseStormOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), OP_SINGLE_ARG(float, "momentum", momentum_, 10.0), OP_SINGLE_ARG(float, "beta", beta_, 0.1) {} bool RunOnDevice() override { // Enforce shapes CAFFE_ENFORCE_EQ(Input(PARAM).numel(), Input(MOMENT).numel()); CAFFE_ENFORCE_EQ(Input(GRADSQSUM).numel(), 1); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* paramIn = Input(PARAM).template data<float>(); const auto* momentIn = Input(MOMENT).template data<float>(); const auto* gradSqSumIn = Input(GRADSQSUM).template data<float>(); const auto* gradIn = Input(GRAD).template data<float>(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* lr = Input(LR).template data<float>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<float>(); auto* momentOut = Output(OUTPUT_MOMENT)->template mutable_data<float>(); auto* gradSqSumOut = Output(OUTPUT_GRAGSQSUM)->template mutable_data<float>(); auto n = Input(INDICES).numel(); if (n == 0) { return true; } float gradSqSumTmp = 0.0; for (const auto i : c10::irange(Input(GRAD).numel())) { const float gi = gradIn[i]; gradSqSumTmp += gi * gi; } gradSqSumOut[0] = gradSqSumIn[0] + gradSqSumTmp; const float nlr = lr[0] * std::pow(beta_ + gradSqSumOut[0], -1.0 / 3.0); const float alpha = momentum_ * nlr * nlr; const auto block_size = Input(GRAD).numel() / n; for (const auto i : c10::irange(n)) { auto idx = indices[i]; if (block_size == 1) { const float gi = gradIn[i]; const float mi = momentIn[idx]; float new_mi = momentOut[idx] = gi + (1.0 - alpha) * (mi - gi); paramOut[idx] = paramIn[idx] + nlr * new_mi; } else { auto offsetI = i * block_size; auto offsetIdx = idx * block_size; #ifndef NDEBUG CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input i:", i, " and block size:", block_size); CAFFE_ENFORCE_GE( Input(GRAD).numel(), block_size + offsetI, this->debug_def().input(GRAD), ", out of bound idx, idx:", idx, " for input i:", i); #endif for (const auto j : c10::irange(block_size)) { const float gi = gradIn[offsetI + j]; const float mi = momentIn[offsetIdx + j]; float new_mi = momentOut[offsetIdx + j] = gi + (1.0 - alpha) * (mi - gi); paramOut[offsetIdx + j] = paramIn[offsetIdx + j] + nlr * new_mi; } } } return true; } protected: const float momentum_; const float beta_; INPUT_TAGS(PARAM, MOMENT, GRADSQSUM, GRAD, INDICES, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_MOMENT, OUTPUT_GRAGSQSUM); }; } // namespace caffe2
5,950
31.167568
76
h
null
pytorch-main/caffe2/sgd/weight_scale_op.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 "caffe2/core/operator.h" #include <stdlib.h> #include <time.h> namespace caffe2 { template <typename T, class Context> void weight_scale_update( int N, const T* w, const T scale, int64_t iter, int64_t stepsize, int64_t update_upper_bound, T* nw, Context* context) { const auto w_size = N * sizeof(float); if (iter % stepsize != 0 || iter >= update_upper_bound) { memcpy(nw, w, w_size); return; } // perform the weight scaling caffe2::math::Scale<T, T, Context>(N, scale, w, nw, context); } template <class Context> class WeightScaleOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; WeightScaleOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), stepsize_(OperatorBase::GetSingleArgument<int64_t>( "stepsize", std::numeric_limits<int64_t>::max())), update_upper_bound_(OperatorBase::GetSingleArgument<int64_t>( "upper_bound_iter", std::numeric_limits<int64_t>::max())), scale_(this->template GetSingleArgument<float>("scale", 1.0f)) {} bool RunOnDevice() override { Output(OUTPUT_WEIGHTS)->ResizeLike(Input(WEIGHTS)); return DispatchHelper<TensorTypes<float>>::call(this, Input(WEIGHTS)); } template <typename T> bool DoRunWithType() { const auto iter = OperatorBase::Input<Tensor>(ITER, CPU).template data<int64_t>()[0] + 1; weight_scale_update<T, Context>( Input(WEIGHTS).size(), Input(WEIGHTS).template data<T>(), scale_, iter, stepsize_, update_upper_bound_, Output(OUTPUT_WEIGHTS)->template mutable_data<T>(), &context_); return true; } protected: int64_t stepsize_; int64_t update_upper_bound_; float scale_; INPUT_TAGS(WEIGHTS, ITER); OUTPUT_TAGS(OUTPUT_WEIGHTS); }; } // namespace caffe2
2,552
27.366667
79
h
null
pytorch-main/caffe2/sgd/wngrad_op.h
#pragma once #include "caffe2/core/operator.h" namespace caffe2 { template <typename Context> void wngrad_update( int N, const float* w, const float* g, const float* h, float* nw, float* nh, float epsilon, const float* lr, Context* /*context*/) { for (const auto i : c10::irange(N)) { float gi = g[i]; nw[i] = w[i] + lr[0] * gi / (h[0] + epsilon); } float nhTmp = 0.0; for (const auto i : c10::irange(N)) { float gi = g[i]; nhTmp += gi * gi; } nhTmp /= (h[0] + epsilon); nh[0] = h[0] + nhTmp; } template <typename Context> void wngrad_update_output_effective_lr( int N, const float* paramIn, const float* gradIn, const float* seqBIn, float* paramOut, float* seqBOut, float* effectiveLROut, float epsilon, const float* lr, Context* /*context*/) { effectiveLROut[0] = lr[0] / (seqBIn[0] + epsilon); float seqBTmp = 0.0; for (const auto i : c10::irange(N)) { float gi = gradIn[i]; seqBTmp += gi * gi; } seqBTmp /= (seqBIn[0] + epsilon); seqBOut[0] = seqBIn[0] + seqBTmp; for (const auto i : c10::irange(N)) { float grad = gradIn[i]; paramOut[i] = paramIn[i] + effectiveLROut[0] * grad; } } template <typename Context> void wngrad_update_output_effective_lr_and_update( int N, const float* paramIn, const float* gradIn, const float* seqBIn, float* paramOut, float* seqBOut, float* effectiveLROut, float* updateOut, float epsilon, const float* lr, Context* /*context*/) { effectiveLROut[0] = lr[0] / (seqBIn[0] + epsilon); float seqBTmp = 0.0; for (const auto i : c10::irange(N)) { float gi = gradIn[i]; seqBTmp += gi * gi; } seqBTmp /= (seqBIn[0] + epsilon); seqBOut[0] = seqBIn[0] + seqBTmp; for (const auto i : c10::irange(N)) { float grad = gradIn[i]; float update = updateOut[i] = effectiveLROut[0] * grad; paramOut[i] = paramIn[i] + update; } } template <typename T, class Context> class WngradOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; WngradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), epsilon_(this->template GetSingleArgument<T>("epsilon", 1e-5f)) {} bool RunOnDevice() override { CAFFE_ENFORCE_EQ( Input(GRAD).numel(), Input(PARAM).numel(), "PARAM size: ", Input(PARAM).numel(), ", GRAD size: ", Input(GRAD).numel(), ", SEQ_B size: ", Input(SEQ_B).numel(), ", LR size: ", Input(LR).numel()); Output(OUTPUT_PARAM)->ResizeLike(Input(PARAM)); Output(OUTPUT_SEQ_B)->ResizeLike(Input(SEQ_B)); if (OutputSize() == 2) { wngrad_update<Context>( Input(GRAD).numel(), Input(PARAM).template data<T>(), Input(GRAD).template data<T>(), Input(SEQ_B).template data<T>(), Output(OUTPUT_PARAM)->template mutable_data<T>(), Output(OUTPUT_SEQ_B)->template mutable_data<T>(), epsilon_, Input(LR).template data<T>(), &context_); } else if (OutputSize() == 3) { Output(OUTPUT_EFFECTIVE_LR)->ResizeLike(Input(SEQ_B)); wngrad_update_output_effective_lr<Context>( Input(GRAD).numel(), Input(PARAM).template data<T>(), Input(GRAD).template data<T>(), Input(SEQ_B).template data<T>(), Output(OUTPUT_PARAM)->template mutable_data<T>(), Output(OUTPUT_SEQ_B)->template mutable_data<T>(), Output(OUTPUT_EFFECTIVE_LR)->template mutable_data<T>(), epsilon_, Input(LR).template data<T>(), &context_); } else { Output(OUTPUT_EFFECTIVE_LR)->ResizeLike(Input(SEQ_B)); Output(OUTPUT_UPDATE)->ResizeLike(Input(GRAD)); wngrad_update_output_effective_lr_and_update<Context>( Input(GRAD).numel(), Input(PARAM).template data<T>(), Input(GRAD).template data<T>(), Input(SEQ_B).template data<T>(), Output(OUTPUT_PARAM)->template mutable_data<T>(), Output(OUTPUT_SEQ_B)->template mutable_data<T>(), Output(OUTPUT_EFFECTIVE_LR)->template mutable_data<T>(), Output(OUTPUT_UPDATE)->template mutable_data<T>(), epsilon_, Input(LR).template data<T>(), &context_); } return true; } protected: T epsilon_; INPUT_TAGS(PARAM, SEQ_B, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_SEQ_B, OUTPUT_EFFECTIVE_LR, OUTPUT_UPDATE); }; template <typename T, class Context> class SparseWngradOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; SparseWngradOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), epsilon_(this->template GetSingleArgument<float>("epsilon", 1e-5f)) {} bool RunOnDevice() override { // Enforce shapes CAFFE_ENFORCE_EQ(Input(SEQ_B).numel(), 1); CAFFE_ENFORCE_EQ(Input(LR).numel(), 1); CAFFE_ENFORCE_EQ( Input(PARAM).size_from_dim(1), Input(GRAD).size_from_dim(Input(INDICES).dim())); return DispatchHelper<TensorTypes<int32_t, int64_t>>::call( this, Input(INDICES)); } template <typename SIndex> bool DoRunWithType() { const auto* lr = Input(LR).template data<T>(); const auto* indices = Input(INDICES).template data<SIndex>(); const auto* gradIn = Input(GRAD).template data<T>(); const auto* paramIn = Input(PARAM).template data<T>(); const auto* seqBIn = Input(SEQ_B).template data<T>(); auto* paramOut = Output(OUTPUT_PARAM)->template mutable_data<T>(); auto* seqBOut = Output(OUTPUT_SEQ_B)->template mutable_data<T>(); auto n = Input(INDICES).numel(); if (n == 0) { return true; } auto block_size = Input(GRAD).numel() / n; for (const auto i : c10::irange(n)) { auto idx = indices[i]; if (block_size == 1) { float gi = gradIn[i]; paramOut[idx] = paramIn[idx] + lr[0] * gi / (seqBIn[0] + epsilon_); } else { auto offsetI = i * block_size; auto offsetIdx = idx * block_size; #ifndef NDEBUG CAFFE_ENFORCE_GE( Input(PARAM).numel(), block_size + offsetIdx, this->debug_def().input(PARAM), ", out of bound, idx:", idx, " for input i:", i, " and block size:", block_size); CAFFE_ENFORCE_GE( Input(GRAD).numel(), block_size + offsetI, this->debug_def().input(GRAD), ", out of bound idx, idx:", idx, " for input i:", i); #endif for (const auto j : c10::irange(block_size)) { float gi = gradIn[offsetI + j]; paramOut[offsetIdx + j] = paramIn[offsetIdx + j] + lr[0] * gi / (seqBIn[0] + epsilon_); } } } float seqBTmp = 0.0; for (const auto i : c10::irange(Input(GRAD).numel())) { float gi = gradIn[i]; seqBTmp += gi * gi; } seqBTmp /= seqBIn[0]; seqBOut[0] = seqBTmp + seqBIn[0]; return true; } protected: T epsilon_; INPUT_TAGS(PARAM, SEQ_B, INDICES, GRAD, LR); OUTPUT_TAGS(OUTPUT_PARAM, OUTPUT_SEQ_B); }; } // namespace caffe2
7,335
28.461847
78
h
null
pytorch-main/caffe2/sgd/yellowfin_op.h
// YellowFin: An automatic tuner for momentum SGD // (https://arxiv.org/abs/1706.03471) // The YellowFinOp tunes learning rate and momentum and performs momentum SGD // steps. The learning rate and momentum are separate for any matrix of // parameters. #pragma once #include <cmath> #include <cstring> #include "caffe2/core/operator.h" #include "caffe2/utils/math.h" namespace caffe2 { template <typename T, class Context> class YellowFinOp final : public Operator<Context> { public: USE_OPERATOR_CONTEXT_FUNCTIONS; YellowFinOp(const OperatorDef& operator_def, Workspace* ws) : Operator<Context>(operator_def, ws), curv_win_width_( this->template GetSingleArgument<int>("curv_win_width", 20)), nesterov_(this->template GetSingleArgument<int>("nesterov", false)), zero_debias_( this->template GetSingleArgument<bool>("zero_debias", true)), epsilon_(this->template GetSingleArgument<T>("epsilon", 1e-6f)), beta_(this->template GetSingleArgument<T>("beta", 0.999f)) {} protected: // GetLrMu and MomentumSgdUpdate have different implementations for GPU and // CPU. All other methods are generic. void GetLrMu(); void MomentumSgdUpdate(); void AfterApply() { // g MovingAverage(D_, grad_, g_avg_, g_avg_out_, g_deb_); // g2 math::Mul(D_, grad_, grad_, aux_vector_, &context_); MovingAverage(D_, aux_vector_, g2_avg_, g2_avg_out_, g2_deb_); // g_norm2 math::Dot(D_, grad_, grad_, g_norm2_, &context_); math::Maximum(1, epsilon_, g_norm2_, g_norm2_, &context_); MovingAverage(1, g_norm2_, g_norm2_avg_, g_norm2_avg_out_, g_norm2_deb_); // g_norm math::Sqrt(1, g_norm2_, g_norm_, &context_); MovingAverage(1, g_norm_, g_norm_avg_, g_norm_avg_out_, g_norm_deb_); math::Maximum(1, epsilon_, g_norm_deb_, g_norm_deb_, &context_); // Curvature range: g_norm2_min, g_norm2_max math::CopyVector(curv_win_width_, curv_win_, curv_win_out_, &context_); T* curv_win_cell = curv_win_out_ + (iter_ - 1) % curv_win_width_; math::Log(1, g_norm2_, curv_win_cell, &context_); int valid_end = std::min(curv_win_width_, iter_); math::ReduceMin( valid_end, curv_win_out_, g_norm2_min_, &scratch_tensor_, &context_); math::ReduceMax( valid_end, curv_win_out_, g_norm2_max_, &scratch_tensor_, &context_); MovingAverage( 1, g_norm2_min_, g_norm2_min_avg_, g_norm2_min_avg_out_, g_norm2_min_deb_); MovingAverage( 1, g_norm2_max_, g_norm2_max_avg_, g_norm2_max_avg_out_, g_norm2_max_deb_); math::Exp(1, g_norm2_min_deb_, g_norm2_min_deb_, &context_); math::Exp(1, g_norm2_max_deb_, g_norm2_max_deb_, &context_); math::Maximum(1, epsilon_, g_norm2_min_deb_, g_norm2_min_deb_, &context_); math::Maximum(1, epsilon_, g_norm2_max_deb_, g_norm2_max_deb_, &context_); // Gradient variance math::Dot(D_, g_deb_, g_deb_, aux_scalar_, &context_); math::Sub(1, g_norm2_deb_, aux_scalar_, variance_, &context_); math::Maximum(1, epsilon_, variance_, variance_, &context_); // Distance to opt math::Div(1, g_norm_avg_out_, g_norm2_avg_out_, distance_, &context_); MovingAverage( 1, distance_, distance_avg_, distance_avg_out_, distance_deb_); if (iter_ > 1) { GetLrMu(); } } void MovingAverage( const int N, const T* elt, const T* avg, T* new_avg, T* debias_avg) { const T one = 1; math::Scale(N, beta_, avg, new_avg, &context_); math::Axpy(N, one - beta_, elt, new_avg, &context_); math::Scale(N, debias_factor_, new_avg, debias_avg, &context_); } T ZeroDebiasFactor() { if (zero_debias_) { const T one = 1; return one / (one - std::pow(beta_, iter_)); } else { return 1; } } public: bool RunOnDevice() override { // Iter live on the CPU #define CAFFE2_YF_READ_INPUT(INPUT_NAME, VAR_NAME) \ const auto& VAR_NAME##_tensor = Input(INPUT_NAME); \ VAR_NAME##_ = VAR_NAME##_tensor.template data<T>(); CAFFE2_YF_READ_INPUT(PARAM, param) CAFFE2_YF_READ_INPUT(MOMENT, moment) CAFFE2_YF_READ_INPUT(LR_AVG, lr_avg) CAFFE2_YF_READ_INPUT(MU_AVG, mu_avg) CAFFE2_YF_READ_INPUT(CURV_WIN, curv_win) CAFFE2_YF_READ_INPUT(G_AVG, g_avg) CAFFE2_YF_READ_INPUT(G2_AVG, g2_avg) CAFFE2_YF_READ_INPUT(SCALARS_MEMORY, scalars_memory) CAFFE2_YF_READ_INPUT(GRAD, grad) #undef CAFFE2_YF_READ_OUTPUT CAFFE_ENFORCE(OperatorBase::InputIsTensorType(ITER, CPU)); CAFFE_ENFORCE_EQ(lr_avg_tensor.numel(), 1); CAFFE_ENFORCE_EQ(mu_avg_tensor.numel(), 1); CAFFE_ENFORCE_EQ(param_tensor.dim(), moment_tensor.dim()); CAFFE_ENFORCE_EQ(param_tensor.dim(), g_avg_tensor.dim()); CAFFE_ENFORCE_EQ(param_tensor.dim(), g2_avg_tensor.dim()); CAFFE_ENFORCE_EQ(param_tensor.dim(), grad_tensor.dim()); for (const auto i : c10::irange(param_tensor.dim())) { CAFFE_ENFORCE_EQ(param_tensor.dim32(i), moment_tensor.dim32(i)); CAFFE_ENFORCE_EQ(param_tensor.dim32(i), g_avg_tensor.dim32(i)); CAFFE_ENFORCE_EQ(param_tensor.dim32(i), g2_avg_tensor.dim32(i)); CAFFE_ENFORCE_EQ(param_tensor.dim32(i), grad_tensor.dim32(i)); } iter_ = OperatorBase::Input<Tensor>(ITER, CPU).template data<int64_t>()[0]; D_ = param_tensor.numel(); // Input data - persistent memory for internal scalars // Note: Memory for these scalars is being allocated during initialization // of the network. If you want to add / remove a scalar, make a // suitable change of memory size in the initialization. const T* memory_it = scalars_memory_ - 1; g_norm_avg_ = ++memory_it; g_norm2_avg_ = ++memory_it; g_norm2_min_avg_ = ++memory_it; g_norm2_max_avg_ = ++memory_it; distance_avg_ = ++memory_it; // Output data #define CAFFE2_YF_READ_OUTPUT(OUTPUT_NAME, VAR_NAME) \ auto VAR_NAME##_out_tensor = \ Output(OUTPUT_##OUTPUT_NAME, VAR_NAME##_tensor.sizes(), at::dtype<T>()); \ VAR_NAME##_out_ = VAR_NAME##_out_tensor->template mutable_data<T>(); CAFFE2_YF_READ_OUTPUT(PARAM, param) CAFFE2_YF_READ_OUTPUT(MOMENT, moment) CAFFE2_YF_READ_OUTPUT(LR_AVG, lr_avg) CAFFE2_YF_READ_OUTPUT(MU_AVG, mu_avg) CAFFE2_YF_READ_OUTPUT(CURV_WIN, curv_win) CAFFE2_YF_READ_OUTPUT(G_AVG, g_avg) CAFFE2_YF_READ_OUTPUT(G2_AVG, g2_avg) CAFFE2_YF_READ_OUTPUT(SCALARS_MEMORY, scalars_memory) #undef CAFFE2_YF_READ_OUTPUT T* out_memory_it = scalars_memory_out_ - 1; g_norm_avg_out_ = ++out_memory_it; g_norm2_avg_out_ = ++out_memory_it; g_norm2_min_avg_out_ = ++out_memory_it; g_norm2_max_avg_out_ = ++out_memory_it; distance_avg_out_ = ++out_memory_it; #define CAFFE2_YF_INIT_VECTOR(NAME) \ ReinitializeTensor(&NAME##_tensor_, {D_}, at::dtype<T>().device(Context::GetDeviceType())); \ NAME##_ = NAME##_tensor_.template mutable_data<T>(); CAFFE2_YF_INIT_VECTOR(aux_vector) CAFFE2_YF_INIT_VECTOR(g_deb) CAFFE2_YF_INIT_VECTOR(g2_deb) CAFFE2_YF_INIT_VECTOR(g_deb2) #undef CAFFE2_YF_INIT_VECTOR #define CAFFE2_YF_INIT_SCALAR(NAME) \ ReinitializeTensor(&NAME##_tensor_, {1}, at::dtype<T>().device(Context::GetDeviceType())); \ NAME##_ = NAME##_tensor_.template mutable_data<T>(); CAFFE2_YF_INIT_SCALAR(aux_scalar) CAFFE2_YF_INIT_SCALAR(distance) CAFFE2_YF_INIT_SCALAR(distance_deb) CAFFE2_YF_INIT_SCALAR(g_norm) CAFFE2_YF_INIT_SCALAR(g_norm_deb) CAFFE2_YF_INIT_SCALAR(g_norm2) CAFFE2_YF_INIT_SCALAR(g_norm2_max) CAFFE2_YF_INIT_SCALAR(g_norm2_max_deb) CAFFE2_YF_INIT_SCALAR(g_norm2_min) CAFFE2_YF_INIT_SCALAR(g_norm2_min_deb) CAFFE2_YF_INIT_SCALAR(g_norm2_deb) CAFFE2_YF_INIT_SCALAR(lr) CAFFE2_YF_INIT_SCALAR(lr_deb) CAFFE2_YF_INIT_SCALAR(mu_deb) CAFFE2_YF_INIT_SCALAR(mu) CAFFE2_YF_INIT_SCALAR(variance) #undef CAFFE2_YF_INIT_SCALAR debias_factor_ = ZeroDebiasFactor(); MomentumSgdUpdate(); AfterApply(); return true; } protected: int curv_win_width_; bool nesterov_; bool zero_debias_; T epsilon_; T beta_; T debias_factor_; int D_; // Temporary memory on device, listed all variables used in calculations #define CAFFE2_YF_DEFINE_TENSOR(NAME) \ Tensor NAME##_tensor_; \ T* NAME##_; CAFFE2_YF_DEFINE_TENSOR(aux_vector) CAFFE2_YF_DEFINE_TENSOR(g_deb) CAFFE2_YF_DEFINE_TENSOR(g2_deb) CAFFE2_YF_DEFINE_TENSOR(g_deb2) CAFFE2_YF_DEFINE_TENSOR(aux_scalar) CAFFE2_YF_DEFINE_TENSOR(distance) CAFFE2_YF_DEFINE_TENSOR(distance_deb) CAFFE2_YF_DEFINE_TENSOR(g_norm) CAFFE2_YF_DEFINE_TENSOR(g_norm_deb) CAFFE2_YF_DEFINE_TENSOR(g_norm2) CAFFE2_YF_DEFINE_TENSOR(g_norm2_deb) CAFFE2_YF_DEFINE_TENSOR(g_norm2_max) CAFFE2_YF_DEFINE_TENSOR(g_norm2_max_deb) CAFFE2_YF_DEFINE_TENSOR(g_norm2_min) CAFFE2_YF_DEFINE_TENSOR(g_norm2_min_deb) CAFFE2_YF_DEFINE_TENSOR(lr) CAFFE2_YF_DEFINE_TENSOR(lr_deb) CAFFE2_YF_DEFINE_TENSOR(mu) CAFFE2_YF_DEFINE_TENSOR(mu_deb) CAFFE2_YF_DEFINE_TENSOR(variance) Tensor scratch_tensor_{Context::GetDeviceType()}; #undef CAFFE2_YF_DEFINE_TENSOR // Input tensors' data const T* param_; const T* moment_; const T* lr_avg_; const T* mu_avg_; const T* curv_win_; const T* g_avg_; const T* g2_avg_; const T* scalars_memory_; const T* grad_; int iter_; // Scalar data from scalars_memory_ input tensor const T* g_norm_avg_; const T* g_norm2_avg_; const T* g_norm2_min_avg_; const T* g_norm2_max_avg_; const T* distance_avg_; // Output tensors' data T* param_out_; T* moment_out_; T* lr_avg_out_; T* mu_avg_out_; T* curv_win_out_; T* g_avg_out_; T* g2_avg_out_; T* scalars_memory_out_; // Scalar data from scalars_memory_ output tensor T* g_norm_avg_out_; T* g_norm2_avg_out_; T* g_norm2_min_avg_out_; T* g_norm2_max_avg_out_; T* distance_avg_out_; INPUT_TAGS( PARAM, MOMENT, LR_AVG, MU_AVG, CURV_WIN, G_AVG, G2_AVG, SCALARS_MEMORY, GRAD, ITER); OUTPUT_TAGS( OUTPUT_PARAM, OUTPUT_MOMENT, OUTPUT_LR_AVG, OUTPUT_MU_AVG, OUTPUT_CURV_WIN, OUTPUT_G_AVG, OUTPUT_G2_AVG, OUTPUT_SCALARS_MEMORY); }; } // namespace caffe2
10,292
30.965839
98
h
null
pytorch-main/caffe2/share/contrib/zstd/quant_decomp_zstd_op.h
#ifndef QUANT_DECOMP_OP_H_ #define QUANT_DECOMP_OP_H_ #include "caffe2/core/context.h" #include "caffe2/core/operator.h" namespace caffe2 { // Decompress a set of tensors compressed using zstd, // see quant_decomp_op_test.py for how to compress class QuantDecompZstdOp final : public Operator<CPUContext> { public: USE_OPERATOR_FUNCTIONS(CPUContext); QuantDecompZstdOp(const OperatorDef& operator_def, Workspace* ws) : Operator<CPUContext>(operator_def, ws) {} ~QuantDecompZstdOp() {} bool RunOnDevice() override; }; } // namespace caffe2 #endif // QUANT_DECOMP_OP_H_
589
23.583333
67
h
null
pytorch-main/caffe2/transforms/common_subexpression_elimination.h
#pragma once #include "caffe2/core/common.h" #include "caffe2/core/transform.h" #include "caffe2/proto/caffe2_pb.h" #include "caffe2/utils/proto_utils.h" namespace caffe2 { /** * Common Subexpression Elimination * * This transforms looks for specific operators (denoted by allowed_ops_), * and removes unnecessary repetition of that operator. * * Consider some operator of X, that reads from blob b_ written to by W. * X_a and X_b read the output of X. However, another operator Y, is the same * type as X, has the same arguments as X, and reads from the same input b_, * written to by W. It's output is the same as X. Y_a, Y_b, and Y_c read from Y. * * Then, we can eliminate the common subexpressions X and Y, and merge them to * Z, where X_a, X_b, Y_a, Y_b, and Y_c all read from Z. * * * TODO(benz): Fix the error to not match nodes that write to external output. */ class TORCH_API CommonSubexpressionEliminationTransform : public Transform { public: CommonSubexpressionEliminationTransform() { SetPatternMatchType(SORTED_WRT_EXECUTION_ORDER); } protected: bool PatternRule( const transform::Graph& g, const std::vector<int>& subgraph, int idx) override; bool ValidatorRule( const transform::Graph& g, const std::vector<int>& subgraph) override; bool ReplaceRule(const std::vector<int>& subgraph, transform::Graph* g_ptr) override; private: bool IsAllowed(string op_type) { return allowed_ops_.count(op_type); } std::set<string> allowed_ops_ = {"LearningRate", "FC"}; }; } // namespace caffe2
1,584
28.90566
80
h
null
pytorch-main/caffe2/transforms/conv_to_nnpack_transform.h
#pragma once #include "caffe2/core/common.h" #include "caffe2/proto/caffe2_pb.h" #include "caffe2/transforms/single_op_transform.h" #include "caffe2/utils/proto_utils.h" namespace caffe2 { class TORCH_API ConvToNNPackTransform : public SingleOpTransform { protected: // Specify what the op needs to be to match the pattern. bool MatchOperator(const OperatorDef& op) override { return ( op.type() == "Conv" && op.device_option().device_type() == PROTO_CPU && op.engine() != "NNPACK"); } // Specify how the operator should be replaced. void ReplaceOperator(OperatorDef* op) override { op->set_engine("NNPACK"); } }; } // namespace caffe2
678
25.115385
79
h
null
pytorch-main/caffe2/transforms/pattern_net_transform.h
#pragma once #include "caffe2/core/common.h" #include "caffe2/core/transform.h" #include "caffe2/proto/caffe2_pb.h" #include "caffe2/utils/proto_utils.h" namespace caffe2 { /** * PatternNetTransform allows you to create transforms using a simple * interface. * * Simply provide a Pattern NetDef and a Replace NetDef, * and this Transform will find subgraphs which fit the pattern net, * and replace it with the replace net. */ class TORCH_API PatternNetTransform : public Transform { public: PatternNetTransform(const NetDef& pattern_net, const NetDef& replace_net) : p_(transform::Graph(pattern_net)), r_(transform::Graph(replace_net)) { // external input and output must match! CAFFE_ENFORCE( p_.external_input() == r_.external_input(), "External inputs do not match!"); CAFFE_ENFORCE( p_.external_output() == r_.external_output(), "External outputs do not match!"); ordered_ops_ = GetPatternTraversalOrder(p_); inverse_ops_.resize(ordered_ops_.size()); for (const auto i : c10::irange(ordered_ops_.size())) { inverse_ops_[ordered_ops_[i]] = i; } } void EnableArgumentMatching() { argument_match_ = true; } void DisableArgumentMatching() { argument_match_ = false; } protected: /** * We want to the final result of subgraph to match the PatternNet in the * order of ordered_ops, operator by operator. * * [[[ ie. g.node(subgraph[i]) should match p.node(ordered_ops[i]) ]]] * * PatternRule for PatternNetTransform does the following: * * When trying to insert node idx into subgraph[p_idx], * we need to see if the edges between index and the * subgraph match the edges between p[ordered_ops[idx]] * and p[ordered_ops[0]...ordered_ops[p_idx-1]]. */ bool PatternRule( const transform::Graph& g, const std::vector<int>& subgraph, int idx) override; /** * ValidatorRule for PatternNetTransform does the following: * * Checks if the size of subgraph and p.size() are the same. That's it! */ bool ValidatorRule( const transform::Graph& g, const std::vector<int>& subgraph) override; /** * ReplaceRule for PatternNet Transform does the following: * * 1) Figure out edge renamings for edges going into/out of the subgraph. * That is, for each blob in the pattern graph, what is it called in the * matched subgraph? * * 2) Remove the matched subgraph. * * 3) Append the replace graph's operators to the graph's operators, and use * the renamings to rename the blob names. * * 4) Create all the children/parent relationships within the replaced graph, * and stitch together the inputs and outputs into the rest of the graph, * matching the removed subgraph. */ bool ReplaceRule(const std::vector<int>& subgraph, transform::Graph* g_ptr) override; private: /** * This returns a permutation of the Pattern Net's operators. * The permutation satisfies this property: * - For any index i, order(i) is a neighbor of some node from * {order(1), ..., order(i-1)}. * * Why is this important? Consider the following case: * PatternNet: 0 ---> 2 <--- 1 * * When we have matched onto [0], and trying to add [1] to our subgraph, * we cannot, since PatternMatch only considers neighbors of the current * subgraph as a candidate next node. * * Therefore, we must present the subgraph in an order such that each node is * a neighbor of its prefix subgraph. One ordering for the above example is * [0, 2, 1]. */ std::vector<int> GetPatternTraversalOrder(const transform::Graph& g); // Graph of Pattern NetDef transform::Graph p_; // The Traversal Order of the Pattern Net's Operators // This is a permutation of the numbers from {0, ..., p.size()-1} std::vector<int> ordered_ops_; // The Inverse of the Traversal Order of the Pattern Net's Operators // That is, inverse_ops[ordered_ops[i]] == i is always true. std::vector<int> inverse_ops_; // Graph of Replace NetDef transform::Graph r_; // This flag determines if the transform will match operator arguments. bool argument_match_ = false; const string TransformBlobWrapper(const string& blob_name) { return "transform/" + blob_name + "_" + c10::to_string(ssa_id_); } int ssa_id_ = 0; }; } // namespace caffe2
4,399
31.835821
79
h
null
pytorch-main/caffe2/transforms/single_op_transform.h
#pragma once #include "caffe2/core/common.h" #include "caffe2/core/transform.h" #include "caffe2/proto/caffe2_pb.h" #include "caffe2/utils/proto_utils.h" namespace caffe2 { /** * Single Op Transform Base class * * A transform which is applied to a single node, in place. * * Transforms which derive from SingleOpTransform need to override: * ReplaceOperator and MatchOperator. */ class TORCH_API SingleOpTransform : public Transform { protected: bool PatternRule( const transform::Graph& g, const std::vector<int>& subgraph, int idx) override; bool ValidatorRule( const transform::Graph& g, const std::vector<int>& subgraph) override; bool ReplaceRule(const std::vector<int>& subgraph, transform::Graph* g_ptr) override; // Specify what the op needs to be to match the pattern. virtual bool MatchOperator(const OperatorDef& op) = 0; // Specify how the operator should be replaced. virtual void ReplaceOperator(OperatorDef* op) = 0; }; } // namespace caffe2
1,021
25.894737
77
h
null
pytorch-main/caffe2/utils/bench_utils.h
/** * Copyright (c) 2016-present, Facebook, Inc. * * 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 CAFFE2_UTILS_BENCH_UTILS_H_ #define CAFFE2_UTILS_BENCH_UTILS_H_ #include <stdint.h> #include <c10/macros/Export.h> namespace caffe2 { TORCH_API uint32_t wipe_cache(); } // namespace caffe2 #endif // CAFFE2_UTILS_BENCH_UTILS_H_
854
26.580645
75
h
null
pytorch-main/caffe2/utils/cast.h
#pragma once #include <caffe2/utils/proto_utils.h> namespace caffe2 { namespace cast { inline TensorProto_DataType GetCastDataType(const ArgumentHelper& helper, std::string arg) { TensorProto_DataType to; if (helper.HasSingleArgumentOfType<string>(arg)) { string s = helper.GetSingleArgument<string>(arg, "float"); std::transform(s.begin(), s.end(), s.begin(), ::toupper); #ifndef CAFFE2_USE_LITE_PROTO CAFFE_ENFORCE(TensorProto_DataType_Parse(s, &to), "Unknown 'to' argument: ", s); #else // Manually implement in the lite proto case. #define X(t) \ if (s == #t) { \ return TensorProto_DataType_##t; \ } X(FLOAT); X(INT32); X(BYTE); X(STRING); X(BOOL); X(UINT8); X(INT8); X(UINT16); X(INT16); X(INT64); X(FLOAT16); X(DOUBLE); #undef X CAFFE_THROW("Unhandled type argument: ", s); #endif } else { to = static_cast<TensorProto_DataType>( helper.GetSingleArgument<int>(arg, TensorProto_DataType_FLOAT)); } return to; } }; // namespace cast }; // namespace caffe2
1,110
21.22
92
h
null
pytorch-main/caffe2/utils/cpu_neon.h
#ifndef CAFFE2_UTILS_CPU_NEON_H_ #define CAFFE2_UTILS_CPU_NEON_H_ // Provides a variety of ARM NEON-specific utility functions #if defined(__ARM_NEON__) || defined(__ARM_NEON) #include <arm_neon.h> namespace caffe2 { template <typename T> inline bool isPointerAligned(T* p, size_t align) { return (reinterpret_cast<uintptr_t>(p) % align == 0); } inline float32x4_t vert_sum_f32(float32x4_t v0, float32x4_t v1, float32x4_t v2, float32x4_t v3) { v0 = vaddq_f32(v0, v1); v2 = vaddq_f32(v2, v3); return vaddq_f32(v0, v2); } inline float horizontal_sum_f32(float32x4_t v0, float32x4_t v1, float32x4_t v2, float32x4_t v3) { v0 = vert_sum_f32(v0, v1, v2, v3); float32x2_t v = vadd_f32(vget_high_f32(v0), vget_low_f32(v0)); return vget_lane_f32(vpadd_f32(v, v), 0); } // Load/store functions that assume alignment inline float32x4_t vld1q_f32_aligned(const float* p) { return vld1q_f32((const float*) __builtin_assume_aligned(p, sizeof(float32x4_t))); } inline void vst1q_f32_aligned(float* p, float32x4_t v) { vst1q_f32((float*) __builtin_assume_aligned(p, sizeof(float32x4_t)), v); } inline void vst4_u8_aligned(uint8_t* p, uint8x8x4_t v) { vst4_u8((uint8_t*) __builtin_assume_aligned(p, sizeof(uint8x8x4_t)), v); } } // namespace caffe2 #endif // defined(__ARM_NEON__) || defined(__ARM_NEON) #endif // CAFFE2_UTILS_CPU_NEON_H_
1,578
28.240741
74
h
null
pytorch-main/caffe2/utils/eigen_utils.h
// Copyright 2004-present Facebook. All Rights Reserved. #ifndef CAFFE2_OPERATORS_UTILS_EIGEN_H_ #define CAFFE2_OPERATORS_UTILS_EIGEN_H_ #include "Eigen/Core" #include "Eigen/Dense" #include <c10/util/Logging.h> #include <c10/util/irange.h> namespace caffe2 { // Common Eigen types that we will often use template <typename T> using EigenMatrixMap = Eigen::Map<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>>; template <typename T> using EigenArrayMap = Eigen::Map<Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>>; template <typename T> using EigenVectorMap = Eigen::Map<Eigen::Matrix<T, Eigen::Dynamic, 1>>; template <typename T> using EigenVectorArrayMap = Eigen::Map<Eigen::Array<T, Eigen::Dynamic, 1>>; template <typename T> using ConstEigenMatrixMap = Eigen::Map<const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>>; template <typename T> using ConstEigenArrayMap = Eigen::Map<const Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>>; template <typename T> using ConstEigenVectorMap = Eigen::Map<const Eigen::Matrix<T, Eigen::Dynamic, 1>>; template <typename T> using ConstEigenVectorArrayMap = Eigen::Map<const Eigen::Array<T, Eigen::Dynamic, 1>>; using EigenOuterStride = Eigen::OuterStride<Eigen::Dynamic>; using EigenInnerStride = Eigen::InnerStride<Eigen::Dynamic>; using EigenStride = Eigen::Stride<Eigen::Dynamic, Eigen::Dynamic>; template <typename T> using EigenOuterStridedMatrixMap = Eigen:: Map<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenOuterStride>; template <typename T> using EigenOuterStridedArrayMap = Eigen:: Map<Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenOuterStride>; template <typename T> using ConstEigenOuterStridedMatrixMap = Eigen::Map< const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenOuterStride>; template <typename T> using ConstEigenOuterStridedArrayMap = Eigen::Map< const Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenOuterStride>; template <typename T> using EigenStridedMatrixMap = Eigen:: Map<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenStride>; template <typename T> using EigenStridedArrayMap = Eigen::Map<Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenStride>; template <typename T> using ConstEigenStridedMatrixMap = Eigen:: Map<const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenStride>; template <typename T> using ConstEigenStridedArrayMap = Eigen:: Map<const Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>, 0, EigenStride>; // 1-d array template <typename T> using EArrXt = Eigen::Array<T, Eigen::Dynamic, 1>; using EArrXf = Eigen::ArrayXf; using EArrXd = Eigen::ArrayXd; using EArrXi = Eigen::ArrayXi; using EArrXb = EArrXt<bool>; using EArrXI32 = EArrXt<int32_t>; using EArrXU16 = EArrXt<uint16_t>; using EArrXU8 = EArrXt<uint8_t>; using EArr3U8 = Eigen::Array<uint8_t, 3, 1>; // 2-d array, column major template <typename T> using EArrXXt = Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic>; using EArrXXf = Eigen::ArrayXXf; using EArrXXI32 = EArrXXt<int32_t>; using EArrXXU16 = EArrXXt<uint16_t>; using EArrXXU8 = EArrXXt<uint8_t>; using EArrXXi = EArrXXt<int>; // 2-d array, row major template <typename T> using ERArrXXt = Eigen::Array<T, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>; using ERArrXXf = ERArrXXt<float>; using ERArrXXI32t = ERArrXXt<int32_t>; using ERArrXXU16t = ERArrXXt<uint16_t>; using ERArrXXU8t = ERArrXXt<uint8_t>; using ERArrXXi = ERArrXXt<int>; using ERArrXXi64t = ERArrXXt<int64_t>; using ERArrXXi32t = ERArrXXt<int32_t>; // 1-d vector template <typename T> using EVecXt = Eigen::Matrix<T, Eigen::Dynamic, 1>; using EVecXd = Eigen::VectorXd; using EVecXf = Eigen::VectorXf; // 1-d row vector using ERVecXd = Eigen::RowVectorXd; using ERVecXf = Eigen::RowVectorXf; // 2-d matrix, column major template <typename T> using EMatXt = Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic>; using EMatXd = Eigen::MatrixXd; using EMatXf = Eigen::MatrixXf; using EMatXU8 = EMatXt<uint8_t>; using EMatXU16 = EMatXt<uint16_t>; // 2-d matrix, row major template <typename T> using ERMatXt = Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>; using ERMatXd = ERMatXt<double>; using ERMatXf = ERMatXt<float>; using ERMatXU8 = ERMatXt<uint8_t>; namespace utils { template <typename T> Eigen::Map<const EArrXt<T>> AsEArrXt(const std::vector<T>& arr) { return {arr.data(), static_cast<int>(arr.size())}; } template <typename T> Eigen::Map<EArrXt<T>> AsEArrXt(std::vector<T>& arr) { return {arr.data(), static_cast<int>(arr.size())}; } // return a sub array of 'array' based on indices 'indices' template <class Derived, class Derived1, class Derived2> void GetSubArray( const Eigen::ArrayBase<Derived>& array, const Eigen::ArrayBase<Derived1>& indices, Eigen::ArrayBase<Derived2>* out_array) { CAFFE_ENFORCE_EQ(array.cols(), 1); // using T = typename Derived::Scalar; out_array->derived().resize(indices.size()); for (const auto i : c10::irange(indices.size())) { TORCH_DCHECK_LT(indices[i], array.size()); (*out_array)[i] = array[indices[i]]; } } // return a sub array of 'array' based on indices 'indices' template <class Derived, class Derived1> EArrXt<typename Derived::Scalar> GetSubArray( const Eigen::ArrayBase<Derived>& array, const Eigen::ArrayBase<Derived1>& indices) { using T = typename Derived::Scalar; EArrXt<T> ret(indices.size()); GetSubArray(array, indices, &ret); return ret; } // return a sub array of 'array' based on indices 'indices' template <class Derived> EArrXt<typename Derived::Scalar> GetSubArray( const Eigen::ArrayBase<Derived>& array, const std::vector<int>& indices) { return GetSubArray(array, AsEArrXt(indices)); } // return 2d sub array of 'array' based on row indices 'row_indices' template <class Derived, class Derived1, class Derived2> void GetSubArrayRows( const Eigen::ArrayBase<Derived>& array2d, const Eigen::ArrayBase<Derived1>& row_indices, Eigen::ArrayBase<Derived2>* out_array) { out_array->derived().resize(row_indices.size(), array2d.cols()); for (const auto i : c10::irange(row_indices.size())) { TORCH_DCHECK_LT(row_indices[i], array2d.size()); out_array->row(i) = array2d.row(row_indices[i]).template cast<typename Derived2::Scalar>(); } } // return indices of 1d array for elements evaluated to true template <class Derived> std::vector<int> GetArrayIndices(const Eigen::ArrayBase<Derived>& array) { std::vector<int> ret; for (const auto i : c10::irange(array.size())) { if (array[i]) { ret.push_back(i); } } return ret; } } // namespace utils } // namespace caffe2 #endif
6,718
31.616505
80
h
null
pytorch-main/caffe2/utils/filler.h
#ifndef CAFFE2_FILLER_H_ #define CAFFE2_FILLER_H_ #include <sstream> #include "caffe2/core/logging.h" #include "caffe2/core/tensor.h" #include "caffe2/utils/math.h" namespace caffe2 { // TODO: replace filler distribution enum with a better abstraction enum FillerDistribution { FD_UNIFORM, FD_FIXEDSUM, FD_SYNTHETIC }; class TensorFiller { public: template <class Type, class Context> void Fill(Tensor* tensor, Context* context) const { CAFFE_ENFORCE(context, "context is null"); CAFFE_ENFORCE(tensor, "tensor is null"); auto min = (min_ < (double)std::numeric_limits<Type>::min()) ? std::numeric_limits<Type>::min() : static_cast<Type>(min_); auto max = (max_ > (double)std::numeric_limits<Type>::max()) ? std::numeric_limits<Type>::max() : static_cast<Type>(max_); CAFFE_ENFORCE_LE(min, max); Tensor temp_tensor(shape_, Context::GetDeviceType()); std::swap(*tensor, temp_tensor); Type* data = tensor->template mutable_data<Type>(); // select distribution switch (dist_) { case FD_UNIFORM: { math::RandUniform<Type, Context>( tensor->numel(), min, max, data, context); break; } case FD_FIXEDSUM: { auto fixed_sum = static_cast<Type>(fixed_sum_); CAFFE_ENFORCE_LE(min * tensor->numel(), fixed_sum); CAFFE_ENFORCE_GE(max * tensor->numel(), fixed_sum); math::RandFixedSum<Type, Context>( tensor->numel(), min, max, fixed_sum_, data, context); break; } case FD_SYNTHETIC: { math::RandSyntheticData<Type, Context>( tensor->numel(), min, max, data, context); break; } } } TensorFiller& Dist(FillerDistribution dist) { dist_ = dist; return *this; } template <class Type> TensorFiller& Min(Type min) { min_ = (double)min; return *this; } template <class Type> TensorFiller& Max(Type max) { max_ = (double)max; return *this; } template <class Type> TensorFiller& FixedSum(Type fixed_sum) { dist_ = FD_FIXEDSUM; fixed_sum_ = (double)fixed_sum; return *this; } // A helper function to construct the lengths vector for sparse features // We try to pad least one index per batch unless the total_length is 0 template <class Type> TensorFiller& SparseLengths(Type total_length) { return FixedSum(total_length) .Min(std::min(static_cast<Type>(1), total_length)) .Max(total_length); } // a helper function to construct the segments vector for sparse features template <class Type> TensorFiller& SparseSegments(Type max_segment) { CAFFE_ENFORCE(dist_ != FD_FIXEDSUM); return Min(0).Max(max_segment).Dist(FD_SYNTHETIC); } TensorFiller& Shape(const std::vector<int64_t>& shape) { shape_ = shape; return *this; } template <class Type> TensorFiller(const std::vector<int64_t>& shape, Type fixed_sum) : shape_(shape), dist_(FD_FIXEDSUM), fixed_sum_((double)fixed_sum) {} TensorFiller(const std::vector<int64_t>& shape) : shape_(shape), dist_(FD_UNIFORM), fixed_sum_(0) {} TensorFiller() : TensorFiller(std::vector<int64_t>()) {} std::string DebugString() const { std::stringstream stream; stream << "shape = [" << shape_ << "]; min = " << min_ << "; max = " << max_; switch (dist_) { case FD_FIXEDSUM: stream << "; dist = FD_FIXEDSUM"; break; case FD_SYNTHETIC: stream << "; dist = FD_SYNTHETIC"; break; default: stream << "; dist = FD_UNIFORM"; break; } return stream.str(); } private: std::vector<int64_t> shape_; // TODO: type is unknown until a user starts to fill data; // cast everything to double for now. double min_ = 0.0; double max_ = 1.0; FillerDistribution dist_; double fixed_sum_; }; } // namespace caffe2 #endif // CAFFE2_FILLER_H_
3,927
26.858156
75
h
null
pytorch-main/caffe2/utils/fixed_divisor.h
#ifndef CAFFE2_UTILS_FIXED_DIVISOR_H_ #define CAFFE2_UTILS_FIXED_DIVISOR_H_ #include <cstdint> #include <cstdio> #include <cstdlib> // See Note [hip-clang differences to hcc] #if defined(__CUDA_ARCH__) || defined(__HIP_ARCH__) || defined(__HIP__) || \ (defined(__clang__) && defined(__CUDA__)) #define FIXED_DIVISOR_DECL inline __host__ __device__ #else #define FIXED_DIVISOR_DECL inline #endif namespace caffe2 { // Utility class for quickly calculating quotients and remainders for // a known integer divisor template <typename T> class FixedDivisor {}; // Works for any positive divisor, 1 to INT_MAX. One 64-bit // multiplication and one 64-bit shift is used to calculate the // result. template <> class FixedDivisor<std::int32_t> { public: FixedDivisor() = default; explicit FixedDivisor(const std::int32_t d) : d_(d) { #if !defined(USE_ROCM) CalcSignedMagic(); #endif // USE_ROCM } FIXED_DIVISOR_DECL std::int32_t d() const { return d_; } #if !defined(USE_ROCM) FIXED_DIVISOR_DECL std::uint64_t magic() const { return magic_; } FIXED_DIVISOR_DECL int shift() const { return shift_; } #endif // USE_ROCM /// Calculates `q = n / d`. FIXED_DIVISOR_DECL std::int32_t Div(const std::int32_t n) const { #if defined(USE_ROCM) return n / d_; #else // USE_ROCM // In lieu of a mulhi instruction being available, perform the // work in uint64 return (int32_t)((magic_ * (uint64_t)n) >> shift_); #endif // USE_ROCM } /// Calculates `r = n % d`. FIXED_DIVISOR_DECL std::int32_t Mod(const std::int32_t n) const { return n - d_ * Div(n); } /// Calculates `q = n / d` and `r = n % d` together. FIXED_DIVISOR_DECL void DivMod(const std::int32_t n, std::int32_t* q, int32_t* r) const { *q = Div(n); *r = n - d_ * *q; } private: #if !defined(USE_ROCM) // Calculates magic multiplicative value and shift amount for calculating `q = // n / d` for signed 32-bit integers. // Implementation taken from Hacker's Delight section 10. void CalcSignedMagic() { if (d_ == 1) { magic_ = UINT64_C(0x1) << 32; shift_ = 32; return; } const std::uint32_t two31 = UINT32_C(0x80000000); const std::uint32_t ad = std::abs(d_); const std::uint32_t t = two31 + ((uint32_t)d_ >> 31); const std::uint32_t anc = t - 1 - t % ad; // Absolute value of nc. std::uint32_t p = 31; // Init. p. std::uint32_t q1 = two31 / anc; // Init. q1 = 2**p/|nc|. std::uint32_t r1 = two31 - q1 * anc; // Init. r1 = rem(2**p, |nc|). std::uint32_t q2 = two31 / ad; // Init. q2 = 2**p/|d|. std::uint32_t r2 = two31 - q2 * ad; // Init. r2 = rem(2**p, |d|). std::uint32_t delta = 0; do { ++p; q1 <<= 1; // Update q1 = 2**p/|nc|. r1 <<= 1; // Update r1 = rem(2**p, |nc|). if (r1 >= anc) { // (Must be an unsigned ++q1; // comparison here). r1 -= anc; } q2 <<= 1; // Update q2 = 2**p/|d|. r2 <<= 1; // Update r2 = rem(2**p, |d|). if (r2 >= ad) { // (Must be an unsigned ++q2; // comparison here). r2 -= ad; } delta = ad - r2; } while (q1 < delta || (q1 == delta && r1 == 0)); std::int32_t magic = q2 + 1; if (d_ < 0) { magic = -magic; } shift_ = p; magic_ = (std::uint64_t)(std::uint32_t)magic; } #endif // USE_ROCM std::int32_t d_ = 1; #if !defined(USE_ROCM) std::uint64_t magic_; int shift_; #endif // USE_ROCM }; } // namespace caffe2 #endif // CAFFE2_UTILS_FIXED_DIVISOR_H_
3,532
25.56391
80
h
null
pytorch-main/caffe2/utils/knob_patcher.h
#pragma once #include <memory> #include <c10/util/string_view.h> namespace caffe2 { /** * Patch the value of a knob during a unit test. * * This forces the knob to the specified value for as long as the KnobPatcher * object exists. When the KnobPatcher object is destroyed the knob will revert * to its previous value. */ class KnobPatcher { public: KnobPatcher(c10::string_view name, bool value); ~KnobPatcher(); KnobPatcher(KnobPatcher&&) noexcept; KnobPatcher& operator=(KnobPatcher&&) noexcept; KnobPatcher(const KnobPatcher&) = delete; KnobPatcher& operator=(const KnobPatcher&) = delete; private: class PatchState; std::unique_ptr<PatchState> state_; }; } // namespace caffe2
715
20.69697
80
h
null
pytorch-main/caffe2/utils/map_utils.h
#pragma once namespace caffe2 { // Get value from map given key. Return supplied default value if not found // This is a stripped down version from folly: // https://github.com/facebook/folly/blob/5a07e203d79324b68d69f294fa38e43b9671e9b1/folly/MapUtil.h#L35-L45 template < class Map, typename Key = typename Map::key_type, typename Value = typename Map::mapped_type> typename Map::mapped_type get_default(const Map& map, const Key& key, Value&& dflt) { using M = typename Map::mapped_type; auto pos = map.find(key); return (pos != map.end()) ? (pos->second) : M(std::forward<Value>(dflt)); } } // namespace caffe2
635
30.8
106
h
null
pytorch-main/caffe2/utils/math-detail.h
#ifndef CAFFE2_UTILS_MATH_DETAIL_H_ #define CAFFE2_UTILS_MATH_DETAIL_H_ namespace caffe2 { class CPUContext; namespace math { namespace detail { // proxy to a class because of partial specialization limitations for functions template<typename T, class Context, int FixedSize> struct ScaleImpl { inline void operator()( const int N, const float alpha, const T* x, T* y, Context* context) { Scale(N, alpha, x, y, context); } }; // Put light-weight implementations in .h file to enable inlining template<typename T> struct ScaleImpl<T, CPUContext, 1> { inline void operator()( const int N, const float alpha, const T* x, T* y, CPUContext* /*context*/) { TORCH_DCHECK_EQ(N, 1); *y = *x * alpha; } }; template<typename T, class Context, int FixedSize> struct AxpyImpl { inline void operator()( const int N, const float alpha, const T* x, T* y, Context* context) { Axpy(N, alpha, x, y, context); } }; // Put light-weight implementations in .h file to enable inlining template<typename T> struct AxpyImpl<T, CPUContext, 1> { inline void operator()( const int N, const float alpha, const T* x, T* y, CPUContext* /*context*/) { TORCH_DCHECK_EQ(N, 1); *y += *x * alpha; } }; } // namespace detail template <typename T, class Context, int FixedSize> inline void ScaleFixedSize( const int N, const float alpha, const T* x, T* y, Context* context) { detail::ScaleImpl<T, Context, FixedSize>()(N, alpha, x, y, context); } template <typename T, class Context, int FixedSize> inline void AxpyFixedSize( const int N, const float alpha, const T* x, T* y, Context* context) { detail::AxpyImpl<T, Context, FixedSize>()(N, alpha, x, y, context); } } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_DETAIL_H_
1,928
20.197802
79
h
null
pytorch-main/caffe2/utils/math.h
#ifndef CAFFE2_UTILS_MATH_H_ #define CAFFE2_UTILS_MATH_H_ // This is a simple translation from the old Caffe math interfaces. We aim to // still keep it simple, so all platforms would be able to support it fairly // easily. // We include the cblas header here so that we can obtain the macros from cblas. extern "C" { #include "caffe2/utils/cblas.h" } #ifdef CAFFE2_USE_ACCELERATE #include <Accelerate/Accelerate.h> #endif // CAFFE2_USE_ACCELERATE #include "caffe2/core/common.h" #include "caffe2/core/types.h" #include "caffe2/utils/math/broadcast.h" #include "caffe2/utils/math/elementwise.h" #include "caffe2/utils/math/reduce.h" #include "caffe2/utils/math/transpose.h" #include "caffe2/utils/math/utils.h" namespace caffe2 { // TODO: Change dims related arguments to int64_t? class Tensor; // An empty class as a placeholder for a math function that has no specific // engine specified. class TORCH_API DefaultEngine {}; namespace math { #define C10_DECLARE_COMPARE_OP(Comp) \ template <typename T, class Context, bool kBroadcast1st = false> \ void Rowwise##Comp( \ const int rows, \ const int cols, \ const T* A, \ const T* B, \ bool* C, \ Context* context); \ \ template <typename T, class Context, bool kBroadcast1st = false> \ void Colwise##Comp( \ const int rows, \ const int cols, \ const T* A, \ const T* B, \ bool* C, \ Context* context); \ \ template <typename T, class Context> \ void Comp( \ const int A_ndim, \ const int* A_dims, \ const int B_ndim, \ const int* B_dims, \ const T* A, \ const T* B, \ bool* C, \ Context* context); C10_DECLARE_COMPARE_OP(EQ) C10_DECLARE_COMPARE_OP(NE) C10_DECLARE_COMPARE_OP(LT) C10_DECLARE_COMPARE_OP(LE) C10_DECLARE_COMPARE_OP(GT) C10_DECLARE_COMPARE_OP(GE) #undef C10_DECLARE_COMPARE_OP #define C10_DECLARE_BINARY_OP(Func) \ template <typename T, class Context, bool kBroadcast1st = false> \ void Rowwise##Func( \ const int rows, \ const int cols, \ const T* A, \ const T* B, \ T* C, \ Context* context); \ \ template <typename T, class Context, bool kBroadcast1st = false> \ void Colwise##Func( \ const int rows, \ const int cols, \ const T* A, \ const T* B, \ T* C, \ Context* context); \ \ template <typename T, class Context> \ void Func( \ const int A_ndim, \ const int* A_dims, \ const int B_ndim, \ const int* B_dims, \ const T* A, \ const T* B, \ T* C, \ Context* context); C10_DECLARE_BINARY_OP(Add) C10_DECLARE_BINARY_OP(Sub) C10_DECLARE_BINARY_OP(Mul) C10_DECLARE_BINARY_OP(Div) C10_DECLARE_BINARY_OP(And) C10_DECLARE_BINARY_OP(Or) C10_DECLARE_BINARY_OP(Xor) C10_DECLARE_BINARY_OP(BitwiseAnd) C10_DECLARE_BINARY_OP(BitwiseOr) C10_DECLARE_BINARY_OP(BitwiseXor) #undef C10_DECLARE_BINARY_OP // Broadcasts X with X_dims to Y with Y_dims. template <typename T, class Context> TORCH_API void Broadcast( const int X_ndim, const int* X_dims, const int Y_ndim, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Computes inv_std from variance. template <typename T, class Context> TORCH_API void InvStd( const int N, const T epsilon, const T* var, T* inv_std, Context* context); // Adds batch sub-tensors elementwise to output. Stripe is the stripe length // and N is the number of elements to add (size of Y). template <typename T, class Context> TORCH_API void AddStripedBatch( const int N, const T* first, T* y, const int stripe, const int batch, Context* context); // Compute the row-wise max of a N*D matrix X, and write it to a N // dimensional vector y. template <typename T, class Context> TORCH_API void RowwiseMax(const int N, const int D, const T* x, T* y, Context* context); // Compute the column-wise max of a N*D matrix X, and write it to a D // dimensional vector y. template <typename T, class Context> TORCH_API void ColwiseMax(const int N, const int D, const T* x, T* y, Context* context); // Elemwise maximum of vector x and scalar alpha. y[i] = max(x[i], alpha) template <typename T, class Context> TORCH_API void Maximum(const int N, const float alpha, const T* x, T* y, Context* context); // Decaf gemm provides a simpler interface to the gemm functions, with the // limitation that the data has to be contiguous in memory. template <typename T, class Context, class Engine = DefaultEngine> TORCH_API void Gemm( const CBLAS_TRANSPOSE trans_A, const CBLAS_TRANSPOSE trans_B, const int M, const int N, const int K, const float alpha, const T* A, const T* B, const float beta, T* C, Context* context, TensorProto::DataType math_type = TensorProto_DataType_FLOAT); // We also provide a gemm that has explicit lda, ldb and ldc specified. // In most cases you probably want to use the function above, though. template <typename T, class Context, class Engine = DefaultEngine> TORCH_API void GemmEx( const CBLAS_TRANSPOSE trans_A, const CBLAS_TRANSPOSE trans_B, const int M, const int N, const int K, const T alpha, const T* A, const int lda, const T* B, const int ldb, const T beta, T* C, const int ldc, Context* context); // GemmBatched provides a simple abstraction into library routines template <typename T, class Context, class Engine = DefaultEngine> TORCH_API void GemmBatched( const CBLAS_TRANSPOSE trans_A, const CBLAS_TRANSPOSE trans_B, const int batch_size, const int M, const int N, const int K, const float alpha, const T** A, const T** B, const float beta, T** C, Context* context, TensorProto::DataType math_type = TensorProto_DataType_FLOAT); template <typename T, class Context, class Engine = DefaultEngine> TORCH_API void GemmStridedBatched( const CBLAS_TRANSPOSE trans_A, const CBLAS_TRANSPOSE trans_B, const int batch_size, const int M, const int N, const int K, const float alpha, const T* A, const int A_stride, const T* B, const int B_stride, const float beta, T* C, const int C_stride, Context* context, TensorProto::DataType math_type = TensorProto_DataType_FLOAT); // Gemv always takes in a M*N matrix A, and depending on whether we set TransA // to Trans, the output is: // CblasNoTrans: x is an N dim vector and y is an M dim vector. // CblasTrans: x is an M dim vector and y is an N dim vector. template <typename T, class Context, class Engine = DefaultEngine> TORCH_API void Gemv( const CBLAS_TRANSPOSE trans_A, const int M, const int N, const float alpha, const T* A, const T* x, const float beta, T* y, Context* context, TensorProto::DataType math_type = TensorProto_DataType_FLOAT); template <typename T, class Context> TORCH_API void RandUniform(const size_t n, const T a, const T b, T* r, Context* context); // Generate n values that sum up to a fixed sum // and subject to a restriction a <= x <= b for each x generated template <typename T, class Context> TORCH_API void RandFixedSum( const size_t n, const T a, const T b, const T sum, T* r, Context* context); template <typename T, class Context> TORCH_API void RandUniformUnique( const size_t n, const T a, const T b, T* r, const size_t m, const T* avoid, Context* context); // Generate n values from synthetic data distribution, // define by unique accesses and stack distances template <typename T, class Context> TORCH_API void RandSyntheticData(const size_t n, const T a, const T b, T* r, Context* context); template <typename T, class Context> TORCH_API void RandGaussian(const size_t n, const T mean, const T std, T* r, Context* context); // Dot matrix of vector a and b, and writes the result to a single value y. template <typename T, class Context> TORCH_API void Dot(const int N, const T* a, const T* b, T* y, Context* context); // Sum of vector x, and writes the result to a single value y. template <typename T, class Context> TORCH_API void Sum( const int N, const T* x, T* y, Context* context, Tensor* scratch_ptr = nullptr); // Sum of squares of vector x, and writes the result to a single value y. template <typename T, class Context> TORCH_API void SumSqr( const int N, const T* x, T* y, Context* context, Tensor* scratch_ptr = nullptr); // Select does index selection of the rows a N*D matrix x, and gives the N // dimensional vector y that contains the selected data. template <typename T, class Context> TORCH_API void Select( const int N, const int D, const T* x, const int* idx, T* y, Context* context); // groups must be 1 for GPU // For NHWC order with groups > 1, the result will be layout in // NHW G RS C/G order to make data within the same group to be contiguous. // For NCHW order, groups doesn't make any difference because we're doing Im2Col // for each N and C is the slowest moving dimension among CHW. template <typename T, class Context, StorageOrder kOrder> TORCH_API void Im2Col( const int channels, const int height, const int width, const int kernel_h, const int kernel_w, const int dilation_h, const int dilation_w, const int pad_t, const int pad_l, const int pad_b, const int pad_r, const int stride_h, const int stride_w, const T* img_data, T* col_data, Context* context, const int groups = 1); // groups must be 1 for GPU template <typename T, class Context, StorageOrder kOrder> TORCH_API void Im2ColNd( const int N, const int img_size, const int col_size, const int* img_shape, const int* col_shape, const int* kernel_shape, const int* stride, const int* dilation, const int* pad, const T* img_data, T* col_data, Context* context, const int groups = 1); // groups must be 1 for GPU // For NHWC order with groups > 1, the result will be layout in // NHW G RS C/G order to make data within the same group to be contiguous. // For NCHW order, groups doesn't make any difference because we're doing Im2Col // for each N and C is the slowest moving dimension among CHW. template <typename T, class Context, StorageOrder kOrder> TORCH_API void Col2Im( const int channels, const int height, const int width, const int patch_h, const int patch_w, const int dilation_h, const int dilation_w, const int pad_t, const int pad_l, const int pad_b, const int pad_r, const int stride_h, const int stride_w, const T* col_data, T* img_data, Context* context, const int groups = 1); // groups must be 1 for GPU // For NHWC order with groups > 1, the result will be layout in // NHW G RS C/G order to make data within the same group to be contiguous. // For NCHW order, groups doesn't make any difference because we're doing Im2Col // for each N and C is the slowest moving dimension among CHW. template <typename T, class Context, StorageOrder kOrder> TORCH_API void Col2ImNd( const int N, const int img_size, const int col_size, const int* img_shape, const int* col_shape, const int* kernel_shape, const int* stride, const int* dilation, const int* pad, const T* col_data, T* img_data, Context* context, const int groups = 1); // Applies a per-channel bias value to each channel of the input // image. image_size is H * W template <typename T, class Context> TORCH_API void BiasCHW( const T* bias, const T* bias_multiplier, const int bias_channels, const int image_size, T* image, Context* context); template <class Context> TORCH_API void CopyMatrix( const size_t item_size, const int M, const int N, const void* A, const int lda, void* B, const int ldb, Context* context, TypeMeta::Copy copy = nullptr); template <typename T, class Context> TORCH_API void CopyMatrix( const int M, const int N, const T* A, const int lda, T* B, const int ldb, Context* context); template <typename T, class Context> TORCH_API void CopyMatrix( const int M, const int N, const T* A, const int A_outer_stride, const int A_inner_stride, T* B, const int B_outer_stride, const int B_inner_stride, Context* context); template <typename T, class Context> TORCH_API void CopyVector(const int N, const T* A, T* B, Context* context); } // namespace math } // namespace caffe2 #include "caffe2/utils/math-detail.h" #endif // CAFFE2_UTILS_MATH_H_
15,251
31.589744
80
h
null
pytorch-main/caffe2/utils/murmur_hash3.h
//----------------------------------------------------------------------------- // MurmurHash3 was written by Austin Appleby, and is placed in the public // domain. The author hereby disclaims copyright to this source code. #pragma once //----------------------------------------------------------------------------- // Platform-specific functions and macros // Microsoft Visual Studio #if defined(_MSC_VER) && (_MSC_VER < 1600) typedef unsigned char uint8_t; typedef unsigned int uint32_t; typedef unsigned __int64 uint64_t; // Other compilers #else // defined(_MSC_VER) #include <stdint.h> #endif // !defined(_MSC_VER) namespace caffe2 { void MurmurHash3_x86_32(const void* key, int len, uint32_t seed, void* out); void MurmurHash3_x86_128(const void* key, int len, uint32_t seed, void* out); void MurmurHash3_x64_128(const void* key, int len, uint32_t seed, void* out); } // namespace caffe2
909
25
79
h
null
pytorch-main/caffe2/utils/proto_utils.h
#ifndef CAFFE2_UTILS_PROTO_UTILS_H_ #define CAFFE2_UTILS_PROTO_UTILS_H_ #ifdef CAFFE2_USE_LITE_PROTO #include <google/protobuf/message_lite.h> #else // CAFFE2_USE_LITE_PROTO #include <google/protobuf/message.h> #endif // !CAFFE2_USE_LITE_PROTO #include <c10/util/Logging.h> #include <c10/util/string_view.h> #include <c10/util/irange.h> #include "caffe2/utils/proto_wrap.h" #include "caffe2/proto/caffe2_pb.h" #ifndef C10_ANDROID #define CAFFE2_ENABLE_REDUCED_STRINGS_IN_ARGUMENT_LOOKUP #define CAFFE2_ARG_MAP_FIND(map, key) map.find(key) #else #define CAFFE2_ARG_MAP_FIND(map, key) map.find(std::string(key)) #endif namespace caffe2 { using std::string; using ::google::protobuf::MessageLite; // A wrapper function to return device name string for use in blob serialization // / deserialization. This should have one to one correspondence with // caffe2/proto/caffe2.proto: enum DeviceType. // // Note that we can't use DeviceType_Name, because that is only available in // protobuf-full, and some platforms (like mobile) may want to use // protobuf-lite instead. TORCH_API std::string DeviceTypeName(const int32_t& d); TORCH_API int DeviceId(const DeviceOption& option); // Returns if the two DeviceOptions are pointing to the same device. TORCH_API bool IsSameDevice(const DeviceOption& lhs, const DeviceOption& rhs); TORCH_API bool IsCPUDeviceType(int device_type); TORCH_API bool IsGPUDeviceType(int device_type); // Common interfaces that reads file contents into a string. TORCH_API bool ReadStringFromFile(const char* filename, string* str); TORCH_API bool WriteStringToFile(const string& str, const char* filename); // Common interfaces that are supported by both lite and full protobuf. TORCH_API bool ReadProtoFromBinaryFile(const char* filename, MessageLite* proto); inline bool ReadProtoFromBinaryFile(const string filename, MessageLite* proto) { return ReadProtoFromBinaryFile(filename.c_str(), proto); } TORCH_API void WriteProtoToBinaryFile(const MessageLite& proto, const char* filename); inline void WriteProtoToBinaryFile(const MessageLite& proto, const string& filename) { return WriteProtoToBinaryFile(proto, filename.c_str()); } #ifdef CAFFE2_USE_LITE_PROTO namespace TextFormat { inline bool ParseFromString(const string& spec, MessageLite* proto) { LOG(FATAL) << "If you are running lite version, you should not be " << "calling any text-format protobuffers."; return false; } } // namespace TextFormat TORCH_API string ProtoDebugString(const MessageLite& proto); TORCH_API bool ParseProtoFromLargeString(const string& str, MessageLite* proto); // Text format MessageLite wrappers: these functions do nothing but just // allowing things to compile. It will produce a runtime error if you are using // MessageLite but still want text support. inline bool ReadProtoFromTextFile( const char* /*filename*/, MessageLite* /*proto*/) { LOG(FATAL) << "If you are running lite version, you should not be " << "calling any text-format protobuffers."; return false; // Just to suppress compiler warning. } inline bool ReadProtoFromTextFile(const string filename, MessageLite* proto) { return ReadProtoFromTextFile(filename.c_str(), proto); } inline void WriteProtoToTextFile( const MessageLite& /*proto*/, const char* /*filename*/, bool throwIfError = true) { LOG(FATAL) << "If you are running lite version, you should not be " << "calling any text-format protobuffers."; } inline void WriteProtoToTextFile(const MessageLite& proto, const string& filename, bool throwIfError = true) { return WriteProtoToTextFile(proto, filename.c_str(), throwIfError); } inline bool ReadProtoFromFile(const char* filename, MessageLite* proto) { return (ReadProtoFromBinaryFile(filename, proto) || ReadProtoFromTextFile(filename, proto)); } inline bool ReadProtoFromFile(const string& filename, MessageLite* proto) { return ReadProtoFromFile(filename.c_str(), proto); } #else // CAFFE2_USE_LITE_PROTO using ::google::protobuf::Message; namespace TextFormat { TORCH_API bool ParseFromString(const string& spec, Message* proto); } // namespace TextFormat TORCH_API string ProtoDebugString(const Message& proto); TORCH_API bool ParseProtoFromLargeString(const string& str, Message* proto); TORCH_API bool ReadProtoFromTextFile(const char* filename, Message* proto); inline bool ReadProtoFromTextFile(const string filename, Message* proto) { return ReadProtoFromTextFile(filename.c_str(), proto); } TORCH_API void WriteProtoToTextFile(const Message& proto, const char* filename, bool throwIfError = true); inline void WriteProtoToTextFile(const Message& proto, const string& filename, bool throwIfError = true) { return WriteProtoToTextFile(proto, filename.c_str(), throwIfError); } // Read Proto from a file, letting the code figure out if it is text or binary. inline bool ReadProtoFromFile(const char* filename, Message* proto) { return (ReadProtoFromBinaryFile(filename, proto) || ReadProtoFromTextFile(filename, proto)); } inline bool ReadProtoFromFile(const string& filename, Message* proto) { return ReadProtoFromFile(filename.c_str(), proto); } #endif // CAFFE2_USE_LITE_PROTO template < class IterableInputs = std::initializer_list<string>, class IterableOutputs = std::initializer_list<string>, class IterableArgs = std::initializer_list<Argument>> OperatorDef CreateOperatorDef( const string& type, const string& name, const IterableInputs& inputs, const IterableOutputs& outputs, const IterableArgs& args, const DeviceOption& device_option = DeviceOption(), const string& engine = "") { OperatorDef def; def.set_type(type); def.set_name(name); for (const string& in : inputs) { def.add_input(in); } for (const string& out : outputs) { def.add_output(out); } for (const Argument& arg : args) { def.add_arg()->CopyFrom(arg); } if (device_option.has_device_type()) { def.mutable_device_option()->CopyFrom(device_option); } if (engine.size()) { def.set_engine(engine); } return def; } // A simplified version compared to the full CreateOperator, if you do not need // to specify args. template < class IterableInputs = std::initializer_list<string>, class IterableOutputs = std::initializer_list<string>> inline OperatorDef CreateOperatorDef( const string& type, const string& name, const IterableInputs& inputs, const IterableOutputs& outputs, const DeviceOption& device_option = DeviceOption(), const string& engine = "") { return CreateOperatorDef( type, name, inputs, outputs, std::vector<Argument>(), device_option, engine); } TORCH_API bool HasOutput(const OperatorDef& op, const std::string& output); TORCH_API bool HasInput(const OperatorDef& op, const std::string& input); /** * @brief A helper class to index into arguments. * * This helper helps us to more easily index into a set of arguments * that are present in the operator. To save memory, the argument helper * does not copy the operator def, so one would need to make sure that the * lifetime of the OperatorDef object outlives that of the ArgumentHelper. */ class C10_EXPORT ArgumentHelper { public: template <typename Def> static bool HasArgument(const Def& def, c10::string_view name) { return ArgumentHelper(def).HasArgument(name); } template <typename Def, typename T> static T GetSingleArgument( const Def& def, c10::string_view name, const T& default_value) { return ArgumentHelper(def).GetSingleArgument<T>(name, default_value); } template <typename Def, typename T> static bool HasSingleArgumentOfType(const Def& def, c10::string_view name) { return ArgumentHelper(def).HasSingleArgumentOfType<T>(name); } template <typename Def, typename T> static std::vector<T> GetRepeatedArgument( const Def& def, c10::string_view name, const std::vector<T>& default_value = std::vector<T>()) { return ArgumentHelper(def).GetRepeatedArgument<T>(name, default_value); } template <typename Def, typename MessageType> static MessageType GetMessageArgument(const Def& def, c10::string_view name) { return ArgumentHelper(def).GetMessageArgument<MessageType>(name); } template <typename Def, typename MessageType> static std::vector<MessageType> GetRepeatedMessageArgument( const Def& def, c10::string_view name) { return ArgumentHelper(def).GetRepeatedMessageArgument<MessageType>(name); } template <typename Def> static bool RemoveArgument(Def& def, int index) { if (index >= def.arg_size()) { return false; } if (index < def.arg_size() - 1) { def.mutable_arg()->SwapElements(index, def.arg_size() - 1); } def.mutable_arg()->RemoveLast(); return true; } explicit ArgumentHelper(const OperatorDef& def); explicit ArgumentHelper(const NetDef& netdef); bool HasArgument(c10::string_view name) const; template <typename T> T GetSingleArgument(c10::string_view name, const T& default_value) const; template <typename T> bool HasSingleArgumentOfType(c10::string_view name) const; template <typename T> std::vector<T> GetRepeatedArgument( c10::string_view name, const std::vector<T>& default_value = std::vector<T>()) const; template <typename MessageType> MessageType GetMessageArgument(c10::string_view name) const { auto it = CAFFE2_ARG_MAP_FIND(arg_map_, name); CAFFE_ENFORCE(it != arg_map_.end(), "Cannot find parameter named ", name); MessageType message; if (it->second.has_s()) { CAFFE_ENFORCE( message.ParseFromString(it->second.s()), "Failed to parse content from the string"); } else { VLOG(1) << "Return empty message for parameter " << name; } return message; } template <typename MessageType> std::vector<MessageType> GetRepeatedMessageArgument(c10::string_view name) const { auto it = CAFFE2_ARG_MAP_FIND(arg_map_, name); CAFFE_ENFORCE(it != arg_map_.end(), "Cannot find parameter named ", name); std::vector<MessageType> messages(it->second.strings_size()); for (int i = 0; i < messages.size(); ++i) { CAFFE_ENFORCE( messages[i].ParseFromString(it->second.strings(i)), "Failed to parse content from the string"); } return messages; } private: std::map<string, Argument #ifdef CAFFE2_ENABLE_REDUCED_STRINGS_IN_ARGUMENT_LOOKUP , std::less<> #endif > arg_map_; }; // **** Arguments Utils ***** // Helper methods to get an argument from OperatorDef or NetDef given argument // name. Throws if argument does not exist. TORCH_API const Argument& GetArgument(const OperatorDef& def, c10::string_view name); TORCH_API const Argument& GetArgument(const NetDef& def, c10::string_view name); // Helper methods to get an argument from OperatorDef or NetDef given argument // name. Returns nullptr if argument does not exist. TORCH_API const Argument* GetArgumentPtr(const OperatorDef& def, c10::string_view name); TORCH_API const Argument* GetArgumentPtr(const NetDef& def, c10::string_view name); // Helper methods to query a boolean argument flag from OperatorDef or NetDef // given argument name. If argument does not exist, return default value. // Throws if argument exists but the type is not boolean. TORCH_API bool GetFlagArgument( const OperatorDef& def, c10::string_view name, bool default_value = false); TORCH_API bool GetFlagArgument( const NetDef& def, c10::string_view name, bool default_value = false); TORCH_API Argument* GetMutableArgument( const string& name, const bool create_if_missing, OperatorDef* def); TORCH_API Argument* GetMutableArgument( const string& name, const bool create_if_missing, NetDef* def); template <typename T> TORCH_API Argument MakeArgument(const string& name, const T& value); template <typename T, typename Def> inline void AddArgument(const string& name, const T& value, Def* def) { GetMutableArgument(name, true, def)->CopyFrom(MakeArgument(name, value)); } // **** End Arguments Utils ***** bool inline operator==(const DeviceOption& dl, const DeviceOption& dr) { return IsSameDevice(dl, dr); } // Given a net, modify the external inputs/outputs if necessary so that // the following conditions are met // - No duplicate external inputs // - No duplicate external outputs // - Going through list of ops in order, all op inputs must be outputs // from other ops, or registered as external inputs. // - All external outputs must be outputs of some operators. TORCH_API void cleanupExternalInputsAndOutputs(NetDef* net); } // namespace caffe2 namespace std { template <> struct hash<caffe2::DeviceOption> { typedef caffe2::DeviceOption argument_type; typedef std::size_t result_type; result_type operator()(argument_type const& device_option) const { std::string serialized; CAFFE_ENFORCE(device_option.SerializeToString(&serialized)); return std::hash<std::string>{}(serialized); } }; } // namespace std #endif // CAFFE2_UTILS_PROTO_UTILS_H_
13,283
33.59375
106
h
null
pytorch-main/caffe2/utils/signal_handler.h
#pragma once #include <c10/util/signal_handler.h> namespace caffe2 { #if defined(C10_SUPPORTS_FATAL_SIGNAL_HANDLERS) class TORCH_API C2FatalSignalHandler : public c10::FatalSignalHandler { public: void fatalSignalHandlerPostProcess() override; static C2FatalSignalHandler& getInstance(); private: explicit C2FatalSignalHandler(); }; // This works by setting up certain fatal signal handlers. Previous fatal // signal handlers will still be called when the signal is raised. Defaults // to being off. TORCH_API void setPrintStackTracesOnFatalSignal(bool print); TORCH_API bool printStackTracesOnFatalSignal(); #endif // defined(C10_SUPPORTS_FATAL_SIGNAL_HANDLER) } // namespace caffe2
698
26.96
75
h
null
pytorch-main/caffe2/utils/simple_queue.h
#ifndef CAFFE2_UTILS_SIMPLE_QUEUE_H_ #define CAFFE2_UTILS_SIMPLE_QUEUE_H_ #include <condition_variable> // NOLINT #include <mutex> // NOLINT #include <queue> #include <c10/util/Logging.h> namespace caffe2 { // This is a very simple queue that Yangqing wrote when bottlefeeding the baby, // so don't take it seriously. What it does is a minimal thread-safe queue that // allows me to run network as a DAG. // // A usual work pattern looks like this: one or multiple producers push jobs // into this queue, and one or multiple workers pops jobs from this queue. If // nothing is in the queue but NoMoreJobs() is not called yet, the pop calls // will wait. If NoMoreJobs() has been called, pop calls will return false, // which serves as a message to the workers that they should exit. template <typename T> class SimpleQueue { public: SimpleQueue() : no_more_jobs_(false) {} // Pops a value and writes it to the value pointer. If there is nothing in the // queue, this will wait till a value is inserted to the queue. If there are // no more jobs to pop, the function returns false. Otherwise, it returns // true. bool Pop(T* value) { std::unique_lock<std::mutex> mutex_lock(mutex_); while (queue_.size() == 0 && !no_more_jobs_) cv_.wait(mutex_lock); if (queue_.size() == 0 && no_more_jobs_) return false; *value = queue_.front(); queue_.pop(); return true; } int size() { std::unique_lock<std::mutex> mutex_lock(mutex_); return queue_.size(); } // Push pushes a value to the queue. void Push(const T& value) { { std::lock_guard<std::mutex> mutex_lock(mutex_); CAFFE_ENFORCE(!no_more_jobs_, "Cannot push to a closed queue."); queue_.push(value); } cv_.notify_one(); } // NoMoreJobs() marks the close of this queue. It also notifies all waiting // Pop() calls so that they either check out remaining jobs, or return false. // After NoMoreJobs() is called, this queue is considered closed - no more // Push() functions are allowed, and once existing items are all checked out // by the Pop() functions, any more Pop() function will immediately return // false with nothing set to the value. void NoMoreJobs() { { std::lock_guard<std::mutex> mutex_lock(mutex_); no_more_jobs_ = true; } cv_.notify_all(); } private: std::mutex mutex_; std::condition_variable cv_; std::queue<T> queue_; bool no_more_jobs_{}; // We do not allow copy constructors. SimpleQueue(const SimpleQueue& /*src*/) {} }; } // namespace caffe2 #endif // CAFFE2_UTILS_SIMPLE_QUEUE_H_
2,602
31.5375
80
h
null
pytorch-main/caffe2/utils/smart_tensor_printer.h
#pragma once #include "caffe2/core/tensor.h" namespace caffe2 { // This is a wrapper around the TensorPrinter that doesn't require the user to // explicit specify the type of the tensor while calling the Print() method. // It also supports a convenience function with a default constructed printer as // a static method. class TORCH_API SmartTensorPrinter { public: // The proliferation of constructors is to give the feature parity with // TensorPrinter // yet not repeat the default arguments explicitly in case they change in the // future. SmartTensorPrinter() = default; explicit SmartTensorPrinter(const std::string& tensor_name); SmartTensorPrinter( const std::string& tensor_name, const std::string& file_name); SmartTensorPrinter( const std::string& tensor_name, const std::string& file_name, int limit); void Print(const Tensor& tensor); void PrintMeta(const Tensor& tensor) { tensorPrinter_.PrintMeta(tensor); } // Uses a default constructed SmartTensorPrinter static void PrintTensor(const Tensor& tensor); // Uses a default constructed SmartTensorPrinter void PrintTensorMeta(const Tensor& tensor) { DefaultTensorPrinter().PrintMeta(tensor); } private: // Returns a thread local default constructed TensorPrinter static SmartTensorPrinter& DefaultTensorPrinter(); TensorPrinter tensorPrinter_; }; }
1,402
26.509804
80
h
null
pytorch-main/caffe2/utils/string_utils.h
#pragma once #include <algorithm> #include <memory> #include <string> #include <vector> #include <c10/macros/Export.h> namespace caffe2 { TORCH_API std::vector<std::string> split(char separator, const std::string& string, bool ignore_empty = false); TORCH_API std::string trim(const std::string& str); TORCH_API size_t editDistance( const std::string& s1, const std::string& s2, size_t max_distance = 0); TORCH_API inline bool StartsWith( const std::string& str, const std::string& prefix) { return str.length() >= prefix.length() && std::mismatch(prefix.begin(), prefix.end(), str.begin()).first == prefix.end(); } TORCH_API inline bool EndsWith( const std::string& full, const std::string& ending) { if (full.length() >= ending.length()) { return ( 0 == full.compare(full.length() - ending.length(), ending.length(), ending)); } else { return false; } } TORCH_API int32_t editDistanceHelper( const char* s1, size_t s1_len, const char* s2, size_t s2_len, std::vector<size_t>& current, std::vector<size_t>& previous, std::vector<size_t>& previous1, size_t max_distance); } // namespace caffe2
1,206
22.211538
80
h
null
pytorch-main/caffe2/utils/zmq_helper.h
#ifndef CAFFE2_UTILS_ZMQ_HELPER_H_ #define CAFFE2_UTILS_ZMQ_HELPER_H_ #include <zmq.h> #include "caffe2/core/logging.h" namespace caffe2 { class ZmqContext { public: explicit ZmqContext(int io_threads) : ptr_(zmq_ctx_new()) { CAFFE_ENFORCE(ptr_ != nullptr, "Failed to create zmq context."); int rc = zmq_ctx_set(ptr_, ZMQ_IO_THREADS, io_threads); CAFFE_ENFORCE_EQ(rc, 0); rc = zmq_ctx_set(ptr_, ZMQ_MAX_SOCKETS, ZMQ_MAX_SOCKETS_DFLT); CAFFE_ENFORCE_EQ(rc, 0); } ~ZmqContext() { int rc = zmq_ctx_destroy(ptr_); CAFFE_ENFORCE_EQ(rc, 0); } void* ptr() { return ptr_; } private: void* ptr_; C10_DISABLE_COPY_AND_ASSIGN(ZmqContext); }; class ZmqMessage { public: ZmqMessage() { int rc = zmq_msg_init(&msg_); CAFFE_ENFORCE_EQ(rc, 0); } ~ZmqMessage() { int rc = zmq_msg_close(&msg_); CAFFE_ENFORCE_EQ(rc, 0); } zmq_msg_t* msg() { return &msg_; } void* data() { return zmq_msg_data(&msg_); } size_t size() { return zmq_msg_size(&msg_); } private: zmq_msg_t msg_; C10_DISABLE_COPY_AND_ASSIGN(ZmqMessage); }; class ZmqSocket { public: explicit ZmqSocket(int type) : context_(1), ptr_(zmq_socket(context_.ptr(), type)) { CAFFE_ENFORCE(ptr_ != nullptr, "Failed to create zmq socket."); } ~ZmqSocket() { int rc = zmq_close(ptr_); CAFFE_ENFORCE_EQ(rc, 0); } void Bind(const string& addr) { int rc = zmq_bind(ptr_, addr.c_str()); CAFFE_ENFORCE_EQ(rc, 0); } void Unbind(const string& addr) { int rc = zmq_unbind(ptr_, addr.c_str()); CAFFE_ENFORCE_EQ(rc, 0); } void Connect(const string& addr) { int rc = zmq_connect(ptr_, addr.c_str()); CAFFE_ENFORCE_EQ(rc, 0); } void Disconnect(const string& addr) { int rc = zmq_disconnect(ptr_, addr.c_str()); CAFFE_ENFORCE_EQ(rc, 0); } int Send(const string& msg, int flags) { int nbytes = zmq_send(ptr_, msg.c_str(), msg.size(), flags); if (nbytes) { return nbytes; } else if (zmq_errno() == EAGAIN) { return 0; } else { LOG(FATAL) << "Cannot send zmq message. Error number: " << zmq_errno(); return 0; } } int SendTillSuccess(const string& msg, int flags) { CAFFE_ENFORCE(msg.size(), "You cannot send an empty message."); int nbytes = 0; do { nbytes = Send(msg, flags); } while (nbytes == 0); return nbytes; } int Recv(ZmqMessage* msg) { int nbytes = zmq_msg_recv(msg->msg(), ptr_, 0); if (nbytes >= 0) { return nbytes; } else if (zmq_errno() == EAGAIN || zmq_errno() == EINTR) { return 0; } else { LOG(FATAL) << "Cannot receive zmq message. Error number: " << zmq_errno(); return 0; } } int RecvTillSuccess(ZmqMessage* msg) { int nbytes = 0; do { nbytes = Recv(msg); } while (nbytes == 0); return nbytes; } private: ZmqContext context_; void* ptr_; }; } // namespace caffe2 #endif // CAFFE2_UTILS_ZMQ_HELPER_H_
3,020
20.891304
68
h
null
pytorch-main/caffe2/utils/math/elementwise.h
#ifndef CAFFE2_UTILS_MATH_ELEMENTWISE_H_ #define CAFFE2_UTILS_MATH_ELEMENTWISE_H_ #include "caffe2/core/common.h" #include "caffe2/core/types.h" namespace caffe2 { namespace math { template <typename T, class Context> TORCH_API void Exp(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Log(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Log1p(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Sin(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Asin(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Cos(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Acos(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Tan(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Atan(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Sinh(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Cosh(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void SinCos(int N, const T* X, T* S, T* C, Context* context); template <typename T, class Context> TORCH_API void Tanh(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Abs(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Sqr(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Sqrt(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Rsqrt(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Cube(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Cbrt(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Neg(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Sign(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Not(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Powx(int N, const T* A, const T b, T* Y, Context* context); template <typename T, class Context> TORCH_API void Inv(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Erf(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void CdfNorm(int N, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void Set(std::int64_t N, T alpha, T* X, Context* context); template <typename TAlpha, typename TData, class Context> TORCH_API void Scale(std::int64_t N, TAlpha alpha, const TData* X, TData* Y, Context* context); // Different from the Scale function above, if alpha is passed in as a pointer, // we will assume that it lives on the Context device, for example on GPU. template <typename TAlpha, typename TData, class Context> TORCH_API void Scale( std::int64_t N, const TAlpha* alpha, const TData* X, TData* Y, Context* context); template <typename T, class Context> TORCH_API void Add(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Sub(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Mul(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Div(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Min(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Max(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void And(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Or(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void Xor(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void BitwiseAnd(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void BitwiseOr(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void BitwiseXor(int N, const T* A, const T* B, T* C, Context* context); template <typename T, class Context> TORCH_API void EQ(int N, const T* A, const T* B, bool* C, Context* context); template <typename T, class Context> TORCH_API void NE(int N, const T* A, const T* B, bool* C, Context* context); template <typename T, class Context> TORCH_API void LT(int N, const T* A, const T* B, bool* C, Context* context); template <typename T, class Context> TORCH_API void LE(int N, const T* A, const T* B, bool* C, Context* context); template <typename T, class Context> TORCH_API void GT(int N, const T* A, const T* B, bool* C, Context* context); template <typename T, class Context> TORCH_API void GE(int N, const T* A, const T* B, bool* C, Context* context); template <typename TAlpha, typename TData, class Context> TORCH_API void Axpy(std::int64_t N, TAlpha alpha, const TData* X, TData* Y, Context* context); // Different from the Axpy function above, if alpha is passed in // as a pointer, we will assume that it lives on the Context device, // for example on GPU. template <typename TAlpha, typename TData, class Context> TORCH_API void Axpy( std::int64_t N, const TAlpha* alpha, const TData* X, TData* Y, Context* context); template <typename TAlpha, typename TData, class Context> TORCH_API void Axpby( std::int64_t N, TAlpha alpha, const TData* X, TAlpha beta, TData* Y, Context* context); template <typename TAlpha, typename TData, class Context> TORCH_API void Axpby( std::int64_t N, const TAlpha* alpha, const TData* X, const TAlpha* beta, TData* Y, Context* context); } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_ELEMENTWISE_H_
6,455
39.099379
80
h
null
pytorch-main/caffe2/utils/math/half_utils.h
#ifndef CAFFE2_UTILS_MATH_HALF_UTILS_H_ #define CAFFE2_UTILS_MATH_HALF_UTILS_H_ #include "caffe2/core/common.h" #include "caffe2/core/types.h" #include "caffe2/utils/conversions.h" #include "caffe2/utils/math/utils.h" namespace caffe2 { namespace math { namespace utils { struct HalfAddFunctor { MATH_UTILS_DECL at::Half operator()(const at::Half a, const at::Half b) const { return convert::To<float, at::Half>( convert::To<at::Half, float>(a) + convert::To<at::Half, float>(b)); } }; struct HalfSubFunctor { MATH_UTILS_DECL at::Half operator()(const at::Half a, const at::Half b) const { return convert::To<float, at::Half>( convert::To<at::Half, float>(a) - convert::To<at::Half, float>(b)); } }; struct HalfMulFunctor { MATH_UTILS_DECL at::Half operator()(const at::Half a, const at::Half b) const { return convert::To<float, at::Half>( convert::To<at::Half, float>(a) * convert::To<at::Half, float>(b)); } }; struct HalfDivFunctor { MATH_UTILS_DECL at::Half operator()(const at::Half a, const at::Half b) const { return convert::To<float, at::Half>( convert::To<at::Half, float>(a) / convert::To<at::Half, float>(b)); } }; } // namespace utils } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_HALF_UTILS_H_
1,329
25.6
75
h
null
pytorch-main/caffe2/utils/math/reduce.h
#ifndef CAFFE2_UTILS_MATH_REDUCE_H_ #define CAFFE2_UTILS_MATH_REDUCE_H_ #include "caffe2/core/common.h" #include "caffe2/core/types.h" namespace caffe2 { class Tensor; namespace math { template <typename T, class Context> TORCH_API void ReduceMin(const int N, const T* X, T* y, Tensor* scratch_ptr, Context* context); template <typename T, class Context> TORCH_API void ReduceMax(const int N, const T* X, T* y, Tensor* scratch_ptr, Context* context); // In all of the reduce functions, X_dims and Y_dims should have ndim elements. // Each dimension of Y_dims must match the corresponding dimension of X_dims or // must be equal to 1. The dimensions equal to 1 indicate the dimensions of X to // be reduced. // Y = alpha * ReduceMin(X) template <typename T, class Context> TORCH_API void ReduceMin( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Y = alpha * ReduceMax(X) template <typename T, class Context> TORCH_API void ReduceMax( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Y = alpha * ReduceSum(X) template <typename T, class Context> TORCH_API void ReduceSum( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Y = alpha * ReduceMean(X) template <typename T, class Context> TORCH_API void ReduceMean( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Y = alpha * ReduceL1(X) template <typename T, class Context> TORCH_API void ReduceL1( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Y = alpha * ReduceL2(X) template <typename T, class Context> TORCH_API void ReduceL2( const int ndim, const int* X_dims, const int* Y_dims, const T alpha, const T* X, T* Y, Context* context, bool allow_broadcast_fastpath=false); // Computes mean and variance over axes. template <typename T, class Context> TORCH_API void Moments( const int ndims, const int* X_dims, const int* Y_dims, const T* X, T* mean, T* var, Context* context, bool allow_broadcast_fastpath=false); } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_REDUCE_H_
2,672
22.243478
80
h
null
pytorch-main/caffe2/utils/math/transpose.h
#ifndef CAFFE2_UTILS_MATH_TRANSPOSE_H_ #define CAFFE2_UTILS_MATH_TRANSPOSE_H_ #include "caffe2/core/common.h" #include "caffe2/core/types.h" namespace caffe2 { namespace math { // Transpose tensor X with dims by axes and write the result to tensor Y. template <typename TIndex, typename TData, class Context> TORCH_API void Transpose( int ndim, const TIndex* dims, const int* axes, const TData* X, TData* Y, Context* context); template <typename T, class Context> TORCH_API void NCHW2NHWC(int N, int C, int HxW, const T* X, T* Y, Context* context); template <typename T, class Context> TORCH_API void NHWC2NCHW(int N, int C, int HxW, const T* X, T* Y, Context* context); } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_TRANSPOSE_H_
785
23.5625
73
h
null
pytorch-main/caffe2/utils/math/utils.h
#ifndef CAFFE2_UTILS_MATH_UTILS_H_ #define CAFFE2_UTILS_MATH_UTILS_H_ #include <vector> #include "caffe2/core/common.h" #if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__) || \ defined(__HIP__) || (defined(__clang__) && defined(__CUDA__)) #define MATH_UTILS_DECL inline __host__ __device__ #else #define MATH_UTILS_DECL inline #endif namespace caffe2 { namespace math { namespace utils { template <typename T> MATH_UTILS_DECL T Not(const T x) { return !x; } template <typename T> MATH_UTILS_DECL T Sign(const T x) { return x > 0 ? T(1) : (x < 0 ? T(-1) : T(0)); } template <typename T> MATH_UTILS_DECL T Negate(const T x) { return -x; } template <typename T> MATH_UTILS_DECL T Inv(const T x) { return T(1) / x; } template <typename T> MATH_UTILS_DECL T Square(const T x) { return x * x; } template <typename T> MATH_UTILS_DECL T Cube(const T x) { return x * x * x; } // Function uses casting from int to unsigned to compare if value of // parameter a is greater or equal to zero and lower than value of // parameter b. The b parameter is of type signed and is always // positive, // therefore its value is always lower than 0x800... where casting // negative value of a parameter converts it to value higher than // 0x800... // The casting allows to use one condition instead of two. MATH_UTILS_DECL bool IsAGeZeroAndALtB(const int a, const int b) { return static_cast<unsigned int>(a) < static_cast<unsigned int>(b); } // Increase the index digits by one based on dims. template <typename TIndex> TORCH_API void IncreaseIndexInDims(int ndim, const TIndex* dims, TIndex* index); // Get index value from dims and index digits. template <typename TIndex> TORCH_API TIndex GetIndexFromDims(const int n, const TIndex* dims, const TIndex* index); // Checks if the input permutation is an identity permutation; TORCH_API bool IsIdentityPermutation(const int n, const int* perm); TORCH_API bool CheckReduceDims(const int ndim, const int* X_dims, const int* Y_dims); TORCH_API bool IsRowwiseReduce( const int ndim, const int* X_dims, const int* Y_dims, int* rows, int* cols); TORCH_API bool IsColwiseReduce( const int ndim, const int* X_dims, const int* Y_dims, int* rows, int* cols); TORCH_API bool IsBothEndsReduce( const int ndim, const int* X_dims, const int* Y_dims, int* pre, int* mid, int* nxt); // Computest the broadcast binary operation dims. template <typename TIndex> TORCH_API void ComputeBroadcastBinaryOpDims( const int A_ndim, const TIndex* A_dims, const int B_ndim, const TIndex* B_dims, TIndex* A_broadcast_dims, TIndex* B_broadcast_dims, TIndex* C_broadcast_dims); TORCH_API bool IsRowwiseBroadcastBinaryOp( const int ndim, const int* A_dims, const int* B_dims, int* rows, int* cols, bool* broadcast_1st); TORCH_API bool IsColwiseBroadcastBinaryOp( const int ndim, const int* A_dims, const int* B_dims, int* rows, int* cols, bool* broadcast_1st); TORCH_API bool IsBothEndsBroadcastBinaryOp( const int ndim, const int* A_dims, const int* B_dims, int* pre, int* mid, int* nxt, bool* broadcast_1st); TORCH_API bool IsBatchTranspose2D(const int ndim, const int* axes); TORCH_API void ComputeTransposeAxesForReduceOp( const int num_dims, const int num_reduce_axes, const int* reduce_axes, int* transpose_axes); TORCH_API void ComputeTransposeAxesForReduceOp(const int ndim, const int* dims, int* axes); template <typename TIndex> TORCH_API void ComputeTransposedStrides( int ndim, const TIndex* dims, const int* axes, TIndex* strides); } // namespace utils // Calculates ceil(a / b). User must be careful to ensure that there // is no overflow or underflow in the calculation. template <typename T> constexpr T DivUp(const T a, const T b) { return (a + b - T(1)) / b; } // Rounds a up to the next highest multiple of b. User must be careful // to ensure that there is no overflow or underflow in the calculation // of divUp. template <typename T> constexpr T RoundUp(const T a, const T b) { return DivUp<T>(a, b) * b; } // Returns log2(n) for a positive integer type template <typename T> constexpr int IntegerLog2(T n, int p = 0) { return (n <= 1) ? p : IntegerLog2(n / 2, p + 1); } // Returns the next highest power-of-2 for an integer type template <typename T> constexpr T IntegerNextHighestPowerOf2(T v) { return (IntegerIsPowerOf2(v) ? T(2) * v : (T(1) << (IntegerLog2(v) + 1))); } } // namespace math } // namespace caffe2 #endif // CAFFE2_UTILS_MATH_UTILS_H_
4,643
23.834225
76
h
null
pytorch-main/caffe2/utils/threadpool/ThreadPool.h
#ifndef CAFFE2_UTILS_THREADPOOL_H_ #define CAFFE2_UTILS_THREADPOOL_H_ #include "ThreadPoolCommon.h" #include <atomic> #include <functional> #include <memory> #include <mutex> #include <vector> #include "caffe2/core/common.h" // // A work-stealing threadpool loosely based off of pthreadpool // namespace caffe2 { struct Task; class WorkersPool; constexpr size_t kCacheLineSize = 64; // A threadpool with the given number of threads. // NOTE: the kCacheLineSize alignment is present only for cache // performance, and is not strictly enforced (for example, when // the object is created on the heap). Thus, in order to avoid // misaligned intrinsics, no SSE instructions shall be involved in // the ThreadPool implementation. // Note: alignas is disabled because some compilers do not deal with // TORCH_API and alignas annotations at the same time. class TORCH_API /*alignas(kCacheLineSize)*/ ThreadPool { public: static ThreadPool* createThreadPool(int numThreads); static std::unique_ptr<ThreadPool> defaultThreadPool(); virtual ~ThreadPool() = default; // Returns the number of threads currently in use virtual int getNumThreads() const = 0; virtual void setNumThreads(size_t numThreads) = 0; // Sets the minimum work size (range) for which to invoke the // threadpool; work sizes smaller than this will just be run on the // main (calling) thread void setMinWorkSize(size_t size) { std::lock_guard<std::mutex> guard(executionMutex_); minWorkSize_ = size; } size_t getMinWorkSize() const { return minWorkSize_; } virtual void run(const std::function<void(int, size_t)>& fn, size_t range) = 0; // Run an arbitrary function in a thread-safe manner accessing the Workers // Pool virtual void withPool(const std::function<void(WorkersPool*)>& fn) = 0; protected: static size_t defaultNumThreads_; mutable std::mutex executionMutex_; size_t minWorkSize_; }; size_t getDefaultNumThreads(); } // namespace caffe2 #endif // CAFFE2_UTILS_THREADPOOL_H_
2,014
28.202899
81
h