Split (2)

train · 301 rows

text stringlengths 129 1.02k
e data, and bytes speciﬁes the data size in bytes In Fortran, two forms are supported In the ﬁrst, data_arg is a variable or contiguous array section In the second, data_arg is a variable or array element and bytes is the length in bytes A bytes value of zero is treated as a value of one if data_arg is not a null point...
ized false is used for a condition that evaluates to zero in C or C++, or false in Fortran 17 Organization of this document The rest of this document is organized as follows: Chapter 2 Directives, describes the C, C++, and Fortran directives us ed to delineate accelerator regions and augment information available to th...
ruct is used to make the address of data in device memory available in host code Restrictions Avarin ause_device clause must be the name of a variable or array At least one use_device clause must appear At most one ifclause may appear In Fortran, the condition must evaluate to a scalar logica l value; in C or C++, t...
y clauses as described in Section 262 For each reduction var, a private copy is created for each parallel gang and initialized for that o perator At the end of the region, the values for each gang are combined using the r eduction operator, and the result combined with the value of the original varand stored in the ori...
ata section is not Anacc_error_invalid_async error is issued if async_arg is not a valid async- argument value See Section 522 3221 acc map data Summary Theacc_map_data routine maps previously allocated space in the current device memory to the speciﬁed host data Format C or C++: void acc_map_data(h_void *data_arg, d_...
or, or a null pointer size_t runtime_info - A value, such as an error code, from the underlying device runtime or driver, if one is available and appropriate Other Events For any event that does not use the acc_data_event_info ,acc_launch_event_info , or acc_error_event_info struct, the second argument to the callback...
vector_length(1) , deﬁned the serial construct as executing with a single gang of a single worker with a vector length of one See Section 252 Consolidated compute construct restrictions into a new section to improve read ability See Section 254 Clariﬁed that a default clause may appear at most once on a compute const...
a_stream routine returns the NVIDIA CUDA stream handle in use for the current device for the asynchronous activity queue associated with the async argument This argument must be an async-argument as deﬁned in Section 216 Asynchronous Behavior Format C or C++: void*acc_get_cuda_stream ( int async ); accsetcuda stream Su...
as for the compute construct events 518 Enqueue Kernel Launch Thelaunch proﬁling event names are acc_ev_enqueue_launch_start acc_ev_enqueue_launch_end Theacc_ev_enqueue_launch_start event is triggered just before an accelerator compu- tation is enqueued for execution on a device, and acc_ev_enqueue_launch_end is trig-...
ator programming languages such as CUDA or OpenCL, in which data movement be tween the memories can dominate user code In the OpenACC model, data movement between the memo ries can be implicit and managed by the compiler, based on directives from the programmer However, the programmer must be aware of the potentially...
A call to acc_callback_register with a value of acc_toggle as the third argument will enable those callbacks acc_callback_unregister(acc_ev_enqueue_upload_start , prof_data, acc_toggle); // prof_data is disabled acc_callback_register(acc_ev_enqueue_upload_start, prof_data, acc_toggle); // prof_data is re-enabled A ...
ition from vector-single mode to vector-partitioned mode Similar to the worker clause, the vector clause ﬁrst activates additional vector-level parallelism and then distribu tes the loop iterations across those vector lanes The operations will execute using vectors of the length speciﬁed or chosen for the parallel regi...
ethod of parallel execution (single-instruction, multiple-data) where the s ame instruc- tion is applied to multiple data elements simultaneously SIMD operation – avector operation implemented with SIMD instructions Structured block – in C or C++, an executable statement, possibly compound, with a single entry at the t...
loop construct, thegang clause behaves as follows It speciﬁes that the iterations of the associate d loop or loops are to be executed in parallel by distributing the iterations among the gangs alon g the associated dimension created by the compute construct The associated dimension is the va lue of thedim argument, if ...
k routines in registration order, from the head to the tail Adding a new callback routine for such an event will cause the run time to invoke that newly registered callback routine for this instance of the event Both the run time and the library must implement and expect this behavior If an existing callback routine is...
ased to zero, the pointer is detached by initiating an update for the pointer varin device memory to have the same value as the corresponding pointer in local memory The update may complete a synchronously, depending on other clauses on the directive The implementation schedules p ointer updates before any data copies ...
tion is performed 2713 detach clause Thedetach clause may appear on exitdata directives Each varargument to a detach clause must be a C or C++ pointer or a Fortran variable or array with the pointer orallocatable attribute For each varinvar-list , ifvaris in shared memory, no action is taken; if varis not in shared me...
lel/coarray features OpenMP Application Program Interface, version 50, November NVIDIA CUDATMC Programming Guide , version 1111, October The OpenCL Speciﬁcation , version 22, Khronos OpenCL Working Group, July INCITS INCLUSIVE TERMINOLOGY GUIDELINES , version 20210607, InterNational Com- mittee for Information Tech...
cc_async_test(wait_arg) logical function acc_async_test_device(wait_arg, dev_n um) logical function acc_async_test_all() logical function acc_async_test_all_device(dev_num) integer(acc_handle_kind) :: wait_arg integer :: dev_num Description wait_arg must be an async-argument as deﬁned in Section 216 Asynchronous Behavi...
dynamically-allocated dimensions For statically siz ed dimensions, all dimensions except the ﬁrst must specify the whole extent to preserve the c ontiguous data restriction, discussed below For dynamically allocated dimensions, the implementation will alloc ate pointers in device memory corresponding to the pointers in...
At exit from a region, the structured reference counter is used On an exitdata directive, the dynamic reference counter is used If the appropriate reference counter for varis zero, no action is taken Otherwise, a detach action is performed if varis a pointer reference, and the reference counter is updated if var i...
tion The behavior of the program is undeﬁned if any of the pointer members are attached when an update of a composite variable is performed 27 Data Clauses Data clauses may appear on the parallel construct,serial construct,kernels construct, data construct, the enter data andexit data directives, and declare directives...
function prototype is allowe d and applies to the function or the C++ lambda in scope with that name See Section A34 for recom- mended diagnostics for a routine directive with a name In Fortran the syntax of the routine directive is: !$acc routine clause-list !$acc routine( name)clause-list In Fortran, the routine dir...
o an iterative loop, data which is used only during a particular loop iteration With respect to a more general region of code, data which is used within the region but is not initialized prior to the region and is re-initialized prior to any use after the re gion Procedure – in C or C++, a function or C++ lambda; in Fo...
mon block names may now appear in many data clauses link clause for the declare directive the behavior of the declare directive for global data the behavior of a data clause with a C or C++ pointer variable has been clariﬁ ed predeﬁned data attributes support for multidimensional dynamic C/C++ arrays tile andaut...
y starting at zero This is not necessarily the same as the OpenMP thread number ssize_tasync - The async-value used for operations associated with this event; see Sec- tion 216 Asynchronous Behavior ssize_t async_queue - The actual activity queue onto which the async ﬁeld gets mapped; see Section 216 Asynchronous Beh...
the resolution of the name can be confus- ing Implementations should then issue a compile-time warning diagnostic even thou gh the application is conforming For example: void g(); // routine directive applies namespace NS { #pragma acc routine(g) seq // compile-time warning void g(); // routine directive does not apply...
unmap_data(h_void data_arg); Fortran: subroutine acc_unmap_data(data_arg) type(),dimension( *) :: data_arg Description A call to the acc_unmap_data routine is similar to a call to acc_delete , except the device memory is not deallocated data_arg is a host address The behavior of the acc_unmap_data routine is: Ifdata...
nohost clause appears Ignore any bind clause when compiling the procedure for the multicore host CPU Disallow a bind clause to appear after a device_type(host) clause A2 API Routines for Target Platforms These runtime routines allow access to the interface between the OpenACC ru ntime API and the underlying target p...
present decrement action with the dynamic reference counter is performed Ifvaris a pointer reference, a detach action is performed If both structured and dynamic reference counters are zero, a delete action is performed Anexitdata directive with a delete clause and with or without a finalize clause is func- tionally eq...
se, or for a call to an acc_copyout API routine (Section 3219) See those sections for details Acopyout action for a varoccurs only when varis present in device memory Acopyout action for a var: performs an immediate detach action for any pointer in var; initiates a copy of the data for varfrom device memory to the co...
s and CPUs also support SIMD or vector operations within ea ch execution unit The execution model exposes these multiple levels of parallelism on a device and the programmer is required to understand the difference between, for example, a fully par allel loop and a loop that is vectorizable but requires synchronization...
c parallel copy(x) { // gang-shared x undefined #pragma acc loop auto gang reduction(+:x) for (int i = 0; i < I; ++i) { // complex loop body x += arr[i]; // gang or loop-private x modified } // gang-shared x undefined }// gang-shared x updated for gang or gang/seq reduction // x = arr sum possibly times number of gangs...
ed in Section 2143 This routine tells the runtime which typ e of device to use among those available and sets the value of acc-current-device-type-var for the current thread to dev_type Restrictions If some compute regions are compiled to only use one device type, the result o f calling this routine with a different ...
ich the encountering accelerator thr ead is executing In that case, the target device shares memory with the encountering thread 261 Variables with Predetermined Data Attributes The loop variable in a C for statement or Fortran dostatement that is associated with a loop directive is predetermined to be private to each ...
ed fro m local memory to device memory This action enters a data lifetime for those vars, and will make the data available for present clauses on constructs within the data lifetime Dynamic reference counters are incremented for this data, as described in Section 267 Refer ence Counters Pointers in device memory may be...
es of nonconformity are more feasible to diagnose at run time (eg, see Section 15) Where implemen- tations are not able to diagnose nonconformity reliably (eg, an independent clause on a loop with data-dependent loop iterations), they might offer no diagnostics, or th ey might diagnose only subcases In order to suppor...
_type If the program attempts to execute a compute region on a device or to access a ny data in the memory of a device that was shut down, the behavior is undeﬁned If the program attempts to shut down the acc_device_host device type, the behavior is undeﬁned Errors Anacc_error_device_type_unavailable error is issue...
eger for both 32-bit and 64-bit binaries 521 First Argument: General Information The ﬁrst argument is a pointer to the acc_callback_info struct type: typedef struct acc_prof_info { acc_event_t event_type; int valid_bytes; int version; acc_device_t device_type; int device_number; int thread_id; ssize_t async; ssize_t as...
nqueue Data Update (Upload and Download) 5110 Wait 5111 Error Event 52 Callbacks Signature 521 First Argument: General Information 522 Second Argument: Event-Speciﬁc Information ...
nt informa- tion available to the compilers, including speciﬁcation of data local to an acceler ator, guidance on mapping of loops for parallel execution, and similar performance-related d etails 11 Scope This OpenACC API document covers only user-directed parallel and ac celerator programming, where the user speciﬁes...
In this mode, only one tool library is supported 533 Runtime Dynamic Library Loading A common case is to build the tools library as a dynamic library (shared object fo r Linux or OS/X, DLL for Windows) In that case, you can have the OpenACC runtime load th e library during initial- ization This allows you to enable ru...
type for compute or da ta regions in the cur- rent thread and sets the value of acc-current-device-num-var todev_num If the value of dev_num is negative, the runtime will revert to its default behavior, which is implementation -deﬁned If the value of the dev_type is zero, the selected device number will be used for al...
acc_wait_all to await directive with no wait-argument acc_wait_all_device to await(devnum:dev_num) directive wait_arg must be an async-argument as deﬁned in Section 216 Asynchronous Behavior dev_num must be a valid device number of the current device type The behavior of the acc_wait routines is: If there is no dev_...
s device_handle , context_handle andasync_handle to the respective device API type The following device APIs are deﬁned in the interface below Any implementation-deﬁned device A PI type must have a value greater than acc_device_api_implementation_defined typedef enum acc_device_api { acc_device_api_none = 0, /* no dev...
clause Thecreate clause may appear on structured data and compute constructs, on declare direc- tives, and on enterdata directives The clause may optionally have a zero modiﬁer For each varinvar-list , ifvaris in shared memory, no action is taken; if varis not in shared memory, thecreate clause behaves as follows: At ...
326 acc getproperty 327 acc init 328 acc shutdown 329 acc async test 3210 acc wait 3211 acc wait async ...
on is taken Anupdate directive with the device clause is equivalent to a call to the acc_update_device routine, described in Section 3220 if clause Theifclause is optional; when there is no ifclause, the implementation will generate code to perform the updates unconditionally When an ifclause appears, the implementatio...
mary This construct deﬁnes a region of the program that is to be executed seq uentially on the current device The behavior of the serial construct is the same as that of the parallel construct except that it always executes with a single gang of a single worker with a v ector length of one Note: Theserial construct may...
ence equal to or greater than binop , or by using parentheses around expr or subexpressions of expr For forms that allow multiple occurrences of x, the number of times that xis evaluated is unspeciﬁed In Fortran the syntax of the atomic constructs is: !$acc atomic read capture-statement [!$acc end atomic ] or !$acc at...
rator rou- tine If no explicit routine directive applies to a procedure that is called or whose address is acces sed in a compute region, and the procedure’s deﬁnition appears in the progra m unit being compiled, then the implementation applies an implicit routine directive with a seq clause to that procedure A C++ lam...
It is likely a programming error to use the independent clause on a loop if any iteration writes to a variable or array element that any other iteration also writes or rea ds, except for vars which appear in a reduction clause or which are modiﬁed in an atomic region The implementation may be restricted in the levels ...
(OpenACC API) for writ- ing parallel programs in C, C++, and Fortran that run identiﬁed regions in p arallel on multicore CPUs or attached accelerators The method described provides a model f or parallel programming that is portable across operating systems and various types of multicore CPU s and accelerators The dir...
data_arg, size_t bytes); void acc_update_self (h_void data_arg, size_t bytes); void acc_update_device_async(h_void data_arg, size_t bytes, int async_arg); void acc_update_self_async (h_void data_arg, size_t bytes, int async_arg); Fortran: subroutine acc_update_device(data_arg [, bytes ]) subroutine acc_update_se...
at event until the event itself is enabled Initially, all events are enabled acc_callback_unregister(acc_ev_enqueue_upload_start , prof_data, acc_toggle); // prof_data is disabled acc_callback_unregister(acc_ev_enqueue_upload_start , NULL, acc_toggle); // acc_ev_enqueue_upload_start callbacks are disabled acc_ca...
most one gang clause may appear on a loop directive The region of a loop with a gang(dim:d) clause may not contain a loop construct with a gang(dim:e) clause where e >= d unless it appears within a nested compute region 293 worker clause When the parent compute construct is a parallel construct, or on an orphaned loop...
rallel or !$acc parallel [clause-list ] block construct [!$acc end parallel ] where clause is one of the following: async [(int-expr)] wait [(int-expr-list )] num_gangs( int-expr-list ) num_workers( int-expr) vector_length( int-expr) device_type( device-type-list ) if( condition) self [(condition)] reduction( operator:...
cc_error_present is used for an error related to data being unexpectedly present at runtime acc_error_invalid_argument is used when an API routine is called with a invalid argument value, other than those described above acc_error_invalid_async is used when an API routine is called with an invalid async-argument , ...
al memory The update may complete asynchronously, depending on other clau ses on the directive The implementation schedules pointer updates before any data copies due to copyout actions that are performed for the same directive 273 Data Clause Errors An error is issued for a varthat appears in a copy ,copyin ,copyout ...
acc data [clause-list ]new-line structured block and in Fortran, the syntax is !$acc data [clause-list ] structured block !$acc end data or !$acc data [clause-list ] block construct [!$acc end data ] where clause is one of the following: if( condition) async [(int-expr)] wait [(wait-argument )] device_type( device-type...
C and C++, modifying pointers in pointer arrays during the data lifetime, eithe r on the host or on the device, may result in undeﬁned behavior If a subarray appears in a data clause, the implementation may choose to alloca te memory for only that subarray on the accelerator In Fortran, array pointers may appear, but...
counter is updated: Theacc_copyout andacc_delete ) routines perform a present decrement action with the dynamic reference counter Theacc_copyout_finalize oracc_delete_finalize routines set the dy- namic reference counter to zero If both reference counters are then zero: Theacc_copyout routines perform a copyout ...
the code of the construct’s structured block or block construct, and after that, the associated operations that occur upon exit of the construct may be processed asynchronously w hile the local thread continues executing the code following the construct In this section and throughout the speciﬁcation, the term wait-ar...
ar datatype such as array and composite datatypes In Fo rtran, aggregate datatypes include arrays, derived types, character types In C, aggregate datatypes include arrays, targets of pointers, structs, and unions In C++, aggregate d atatypes include arrays, targets of pointers, classes, structs, and unions Aggregate v...
ot known intend_line_no - For an OpenACC construct, this contains the line number of the end of the construct A negative or zero value means the line number is not know n intfunc_line_no - The line number of the ﬁrst line of the function named in func_name A negative or zero value means the line number is not known ...
varis not in shared memory, thepresent clause behaves as follows: At entry to the region: Anattach action is performed if varis a pointer reference, and a present increment action with the structured reference counter is performed if varis not a null pointer At exit from the region: If the structured reference ...
Contents 1 Introduction 11 Scope 12 Execution Model 13 Memory Model 14 Language Interoperability 15 Runtime Errors 16 Conventions used in thi...
l s hort, long or long long at- tribute), enum, ﬂoat, double, long double, Complex (with optional ﬂoat or long attribute), or any pointer datatype In C++, scalar datatypes are char (signed or unsigne d), wchar t, int (signed or unsigned, with optional short, long or long long attribute), enum, bool, ﬂoat, double, long ...
The integer parameter openacc_version with a value yyyymm where yyyy andmmare the year and month designations of the version of the Accelerator programming mo del supported This value matches the value of the preprocessor variable _OPENACC Interfaces for all routines in the chapter Integer parameters to deﬁne intege...
peration – a single operation or sequence of operations applied uniformly to each ele ment of an array Visible data clause – with respect to a compute construct, any data clause on the compute constru ct, a lexically containing data construct, or a visible declare directive See Section 262 Visibledefault clause – with...
,serial , or kernels construct The meaning is identical to explicitly specifying a parallel ,serial , or kernels construct containing a loop construct Any clause that is allowed on a parallel or loop construct is allowed on the parallelloop construct; any clause allowed on a serial or loop construct is allowed on a ser...
ed across the threads associated with the workers an d vector lanes of that gang A procedure called from a compute construct will be annotated as seq,vector ,worker , or gang , as described Section 215 Procedure Calls in Compute Regions Variables declared in seq routine are private to the thread that made the call Vari...
ev_create = 6, acc_ev_delete = 7, acc_ev_alloc = 8, acc_ev_free = 9, acc_ev_enter_data_start = 10, acc_ev_enter_data_end = 11, acc_ev_exit_data_start = 12, acc_ev_exit_data_end = 13, acc_ev_update_start = 14, acc_ev_update_end = 15, acc_ev_compute_construct_start = 16, acc_ev_compute_construct_end = 17, acc_ev_enqueue_...
m_devices routine returns the number of available devices of the given type Format C or C++: int acc_get_num_devices(acc_device_t dev_type); Fortran: integer function acc_get_num_devices(dev_type) integer(acc_device_kind) :: dev_type Description Theacc_get_num_devices routine returns the number of available devices of ...
ode is typedef enum acc_error_t { acc_error_none = 0, acc_error_other = 1, acc_error_system = 2, acc_error_execution = 3, acc_error_device_init = 4, acc_error_device_shutdown = 5, acc_error_device_unavailable = 6, acc_error_device_type_unavailable = 7, acc_error_wrong_device_type = 8, acc_error_out_of_memory = 9, acc_e...
the variable is not private to a scope lexically enclosed within the compute construct See Section 262 false – a condition that evaluates to zero in C or C++, or false in Fortran GPU – a Graphics Processing Unit; one type of accelerator GPGPU – General Purpose computation on Graphics Processing Units Host – the main C...
a constructs See Section 265 The behavior of data clauses and data API routines with a null pointer in the c lause or as a routine argument is deﬁned See Sections 275-2711, 281, and 3216-3 230 Precision issues with the loop trip count calculation were clariﬁed See Sectio n 29 Text in Section 216 was moved and reorga...
gest repre sentable value for that data type, respectively At a minimum, the supported data types include Fortran logical as well as the numerical data types in C (eg, _Bool ,char ,int,float ,double ,float _Complex , double _Complex ), C++ (eg, bool ,char ,wchar_t ,int,float ,double ), and Fortran (eg,integer ,real ,do...
modiﬁer was added to the copyin data clause and cache directive See Sections 277 and 210 The term local device was deﬁned; see Section 12 Execution Model and the Glossary The term varis used more consistently throughout the speciﬁcation to mean a variable name, array name, subarray speciﬁcation, array element, compo...
sync_arg) subroutine acc_delete_async (data_arg [, bytes ], async_arg) subroutine acc_copyout_finalize_async(data_arg [, bytes ], & async_arg) subroutine acc_delete_finalize_async (data_arg [, bytes ], & async_arg) type(*), dimension() :: data_arg integer :: bytes integer(acc_handle_kind) :: async_arg Description A c...
choose an appropriate vector length The re gion of a loop with the vector clause may not contain a loop with a gang ,worker , orvector clause unless within a nested compute region All vector lanes will complete execution of their assigned iterations before an y vector lane proceeds beyond the end of the loop 295 seq c...
block name enclosed within slashes if that common block name also appears in a declare directivelink clause In all cases, the compiler will allocate and manage a copy of the varin the memory of the current device, creating a visible device copy of that var, for data not in shared memory OpenACC supports accelerators w...
If the reduction varis an array element or a subarray, accessing the elements of the array outside the speciﬁed index range results in unspeciﬁed behavior The reduction varmay not be a member of a composite variable If the reduction varis a composite variable, each member of the composite variable must be a supported...
ument , the associated device and activity queues are as speciﬁed in the wait-argument ; see Section 216 If there is no argument to thewait clause, the associated device is the current device and associated activ ity queues are all activity queues The associated operations may not be launched or exec uted until all ope...
is allowed on the parallel andkernels constructs On a parallel construct, it may have one, two, or three arguments The values of the integ er expressions deﬁne the number of parallel gangs along dimensions one, two, and three that will e xecute the parallel region If it has fewer than three arguments, the missing valu...
operation unconditionally When an ifclause appears, the implementation will generate code to conditionally perform the wait operation only when the condition evaluates to true Await directive is functionally equivalent to a call to one of the acc_wait ,acc_wait_async , acc_wait_all , oracc_wait_all_async runtime API ro...
ere are two or more host thre ads executing and sharing the same device, asynchronous operations on any thread with the same async-value will be enqueued onto the same activity queue If the threads are not synchronized with res pect to each other, the operations may be enqueued in either order and therefore may execute...
copyout ,create ,no_create ,present ,deviceptr , attach , ordefault clause must appear on a data construct Only theasync andwait clauses may follow a device_type clause if clause Theifclause is optional; when there is no ifclause, the compiler will generate code to allocate space in the current device memory and move ...
ﬁned A set default_async directive is functionally equivalent to a call to the acc_set_default_async runtime API routine, as described in Sec- tion 3214 device num clause Thedevice_num clause speciﬁes the device number to set as the default device for accele rator regions and changes the value of acc-current-device-num...
he clause that all references to that variable must lie within the region being cached; see Section 210 Cache Directive Text was added to the private andreduction clauses on a combined construct to clarify that they act like private andreduction on theloop , notprivate andreduction on theparallel orreduction on theker...
es of derived type may appear, including a subarray of a mem- ber Members of subarrays of derived type may not appear In Fortran, if array or subarray notation is used for a derived type memb er, array or subarray notation may not be used for a parent of that derived type member See Section 2171 Optional Arguments f...
name( x, expr-list) x = intrinsic procedure name(expr-list, x ) In the preceding statements: xandv(as applicable) are both scalar variables of intrinsic type xmust not be an allocatable variable During the execution of an atomic region, multiple syntactic occurrences of xmust designate the same storage location No...
are available in the header ﬁle acc_callbackh , which is delivered with the OpenACC implementation For backward compatibility with previous versions o f OpenACC, the implementation also delivers the same information in acc_profh Event names are preﬁxed withacc_ev_ The ordering of events must reﬂect the order in whic...
ith the structured reference counter is performed if varis not a null pointer If both structured and dynamic reference counters are zero, a delete action is performed The errors in Section 273 Data Clause Errors may be issued for this clause For compatibility with OpenACC 20, present_or_create andpcreate are alternate...
cc ispresent 3226 acc memcpy todevice 3227 acc memcpy from device 3228 acc memcpy device 3229 acc attach and acc detach 3230 acc memcpy d2d 4 Envi...
synchronous to the local thread Nested OpenACC constructs, directiv es, and calls to runtime li- brary routines do not inherit the async clause from this construct, and the programmer must take care to not accidentally introduce race conditions related to asynchronou s data transfers wait clause Thewait clause is opti...
d for this clause For compatibility with OpenACC 20, present_or_copyin andpcopyin are alternate names forcopyin Anenterdata directive with a copyin clause is functionally equivalent to a call to the acc_copyin API routine, as described in Section 3218 278 copyout clause Thecopyout clause may appear on structured data...
ice(d_void data_dev_dest, h_voiddata_host_src, size_t bytes); void acc_memcpy_to_device_async(d_void data_dev_dest, h_voiddata_host_src, size_t bytes, int async_arg); Fortran: subroutine acc_memcpy_to_device(data_dev_dest, data_host_src, bytes) subroutine acc_memcpy_to_device_async(data_dev_dest, data_host_src, byt...
theacc_device_t type or return values from OpenACC Runtime API routines ACC DEVICE TYPE These implementations should use the case-insensitive name radeon for the environment variable ACC_DEVICE_TYPE device type clause argument An implementation should use the case-insensitive name radeon as the argument to the device...
i- ables declared inside the construct in a declare clause Deﬁning the behavior of additional parallelism constructs in the base langua ges when used inside a compute construct or accelerator routine Optimization directives or clauses, such as an unroll directive or clause Extended reductions Fortran bindings for ...

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.