]>
Commit | Line | Data |
---|---|---|
d77de738 ML |
1 | \input texinfo @c -*-texinfo-*- |
2 | ||
3 | @c %**start of header | |
4 | @setfilename libgomp.info | |
5 | @settitle GNU libgomp | |
6 | @c %**end of header | |
7 | ||
8 | ||
9 | @copying | |
74d5206f | 10 | Copyright @copyright{} 2006-2023 Free Software Foundation, Inc. |
d77de738 ML |
11 | |
12 | Permission is granted to copy, distribute and/or modify this document | |
13 | under the terms of the GNU Free Documentation License, Version 1.3 or | |
14 | any later version published by the Free Software Foundation; with the | |
15 | Invariant Sections being ``Funding Free Software'', the Front-Cover | |
16 | texts being (a) (see below), and with the Back-Cover Texts being (b) | |
17 | (see below). A copy of the license is included in the section entitled | |
18 | ``GNU Free Documentation License''. | |
19 | ||
20 | (a) The FSF's Front-Cover Text is: | |
21 | ||
22 | A GNU Manual | |
23 | ||
24 | (b) The FSF's Back-Cover Text is: | |
25 | ||
26 | You have freedom to copy and modify this GNU Manual, like GNU | |
27 | software. Copies published by the Free Software Foundation raise | |
28 | funds for GNU development. | |
29 | @end copying | |
30 | ||
31 | @ifinfo | |
32 | @dircategory GNU Libraries | |
33 | @direntry | |
34 | * libgomp: (libgomp). GNU Offloading and Multi Processing Runtime Library. | |
35 | @end direntry | |
36 | ||
37 | This manual documents libgomp, the GNU Offloading and Multi Processing | |
38 | Runtime library. This is the GNU implementation of the OpenMP and | |
39 | OpenACC APIs for parallel and accelerator programming in C/C++ and | |
40 | Fortran. | |
41 | ||
42 | Published by the Free Software Foundation | |
43 | 51 Franklin Street, Fifth Floor | |
44 | Boston, MA 02110-1301 USA | |
45 | ||
46 | @insertcopying | |
47 | @end ifinfo | |
48 | ||
49 | ||
50 | @setchapternewpage odd | |
51 | ||
52 | @titlepage | |
53 | @title GNU Offloading and Multi Processing Runtime Library | |
54 | @subtitle The GNU OpenMP and OpenACC Implementation | |
55 | @page | |
56 | @vskip 0pt plus 1filll | |
57 | @comment For the @value{version-GCC} Version* | |
58 | @sp 1 | |
59 | Published by the Free Software Foundation @* | |
60 | 51 Franklin Street, Fifth Floor@* | |
61 | Boston, MA 02110-1301, USA@* | |
62 | @sp 1 | |
63 | @insertcopying | |
64 | @end titlepage | |
65 | ||
66 | @summarycontents | |
67 | @contents | |
68 | @page | |
69 | ||
70 | ||
71 | @node Top, Enabling OpenMP | |
72 | @top Introduction | |
73 | @cindex Introduction | |
74 | ||
75 | This manual documents the usage of libgomp, the GNU Offloading and | |
76 | Multi Processing Runtime Library. This includes the GNU | |
77 | implementation of the @uref{https://www.openmp.org, OpenMP} Application | |
78 | Programming Interface (API) for multi-platform shared-memory parallel | |
79 | programming in C/C++ and Fortran, and the GNU implementation of the | |
80 | @uref{https://www.openacc.org, OpenACC} Application Programming | |
81 | Interface (API) for offloading of code to accelerator devices in C/C++ | |
82 | and Fortran. | |
83 | ||
84 | Originally, libgomp implemented the GNU OpenMP Runtime Library. Based | |
85 | on this, support for OpenACC and offloading (both OpenACC and OpenMP | |
86 | 4's target construct) has been added later on, and the library's name | |
87 | changed to GNU Offloading and Multi Processing Runtime Library. | |
88 | ||
89 | ||
90 | ||
91 | @comment | |
92 | @comment When you add a new menu item, please keep the right hand | |
93 | @comment aligned to the same column. Do not use tabs. This provides | |
94 | @comment better formatting. | |
95 | @comment | |
96 | @menu | |
97 | * Enabling OpenMP:: How to enable OpenMP for your applications. | |
98 | * OpenMP Implementation Status:: List of implemented features by OpenMP version | |
99 | * OpenMP Runtime Library Routines: Runtime Library Routines. | |
100 | The OpenMP runtime application programming | |
101 | interface. | |
102 | * OpenMP Environment Variables: Environment Variables. | |
103 | Influencing OpenMP runtime behavior with | |
104 | environment variables. | |
105 | * Enabling OpenACC:: How to enable OpenACC for your | |
106 | applications. | |
107 | * OpenACC Runtime Library Routines:: The OpenACC runtime application | |
108 | programming interface. | |
109 | * OpenACC Environment Variables:: Influencing OpenACC runtime behavior with | |
110 | environment variables. | |
111 | * CUDA Streams Usage:: Notes on the implementation of | |
112 | asynchronous operations. | |
113 | * OpenACC Library Interoperability:: OpenACC library interoperability with the | |
114 | NVIDIA CUBLAS library. | |
115 | * OpenACC Profiling Interface:: | |
116 | * OpenMP-Implementation Specifics:: Notes specifics of this OpenMP | |
117 | implementation | |
118 | * Offload-Target Specifics:: Notes on offload-target specific internals | |
119 | * The libgomp ABI:: Notes on the external ABI presented by libgomp. | |
120 | * Reporting Bugs:: How to report bugs in the GNU Offloading and | |
121 | Multi Processing Runtime Library. | |
122 | * Copying:: GNU general public license says | |
123 | how you can copy and share libgomp. | |
124 | * GNU Free Documentation License:: | |
125 | How you can copy and share this manual. | |
126 | * Funding:: How to help assure continued work for free | |
127 | software. | |
128 | * Library Index:: Index of this documentation. | |
129 | @end menu | |
130 | ||
131 | ||
132 | @c --------------------------------------------------------------------- | |
133 | @c Enabling OpenMP | |
134 | @c --------------------------------------------------------------------- | |
135 | ||
136 | @node Enabling OpenMP | |
137 | @chapter Enabling OpenMP | |
138 | ||
139 | To activate the OpenMP extensions for C/C++ and Fortran, the compile-time | |
140 | flag @command{-fopenmp} must be specified. This enables the OpenMP directive | |
141 | @code{#pragma omp} in C/C++ and @code{!$omp} directives in free form, | |
142 | @code{c$omp}, @code{*$omp} and @code{!$omp} directives in fixed form, | |
143 | @code{!$} conditional compilation sentinels in free form and @code{c$}, | |
144 | @code{*$} and @code{!$} sentinels in fixed form, for Fortran. The flag also | |
145 | arranges for automatic linking of the OpenMP runtime library | |
146 | (@ref{Runtime Library Routines}). | |
147 | ||
148 | A complete description of all OpenMP directives may be found in the | |
149 | @uref{https://www.openmp.org, OpenMP Application Program Interface} manuals. | |
150 | See also @ref{OpenMP Implementation Status}. | |
151 | ||
152 | ||
153 | @c --------------------------------------------------------------------- | |
154 | @c OpenMP Implementation Status | |
155 | @c --------------------------------------------------------------------- | |
156 | ||
157 | @node OpenMP Implementation Status | |
158 | @chapter OpenMP Implementation Status | |
159 | ||
160 | @menu | |
161 | * OpenMP 4.5:: Feature completion status to 4.5 specification | |
162 | * OpenMP 5.0:: Feature completion status to 5.0 specification | |
163 | * OpenMP 5.1:: Feature completion status to 5.1 specification | |
164 | * OpenMP 5.2:: Feature completion status to 5.2 specification | |
c16e85d7 | 165 | * OpenMP Technical Report 11:: Feature completion status to first 6.0 preview |
d77de738 ML |
166 | @end menu |
167 | ||
168 | The @code{_OPENMP} preprocessor macro and Fortran's @code{openmp_version} | |
169 | parameter, provided by @code{omp_lib.h} and the @code{omp_lib} module, have | |
170 | the value @code{201511} (i.e. OpenMP 4.5). | |
171 | ||
172 | @node OpenMP 4.5 | |
173 | @section OpenMP 4.5 | |
174 | ||
175 | The OpenMP 4.5 specification is fully supported. | |
176 | ||
177 | @node OpenMP 5.0 | |
178 | @section OpenMP 5.0 | |
179 | ||
180 | @unnumberedsubsec New features listed in Appendix B of the OpenMP specification | |
181 | @c This list is sorted as in OpenMP 5.1's B.3 not as in OpenMP 5.0's B.2 | |
182 | ||
183 | @multitable @columnfractions .60 .10 .25 | |
184 | @headitem Description @tab Status @tab Comments | |
185 | @item Array shaping @tab N @tab | |
186 | @item Array sections with non-unit strides in C and C++ @tab N @tab | |
187 | @item Iterators @tab Y @tab | |
188 | @item @code{metadirective} directive @tab N @tab | |
189 | @item @code{declare variant} directive | |
190 | @tab P @tab @emph{simd} traits not handled correctly | |
2cd0689a | 191 | @item @var{target-offload-var} ICV and @code{OMP_TARGET_OFFLOAD} |
d77de738 | 192 | env variable @tab Y @tab |
2cd0689a | 193 | @item Nested-parallel changes to @var{max-active-levels-var} ICV @tab Y @tab |
d77de738 | 194 | @item @code{requires} directive @tab P |
8c2fc744 | 195 | @tab complete but no non-host device provides @code{unified_shared_memory} |
d77de738 | 196 | @item @code{teams} construct outside an enclosing target region @tab Y @tab |
85da0b40 TB |
197 | @item Non-rectangular loop nests @tab P |
198 | @tab Full support for C/C++, partial for Fortran | |
199 | (@uref{https://gcc.gnu.org/PR110735,PR110735}) | |
d77de738 ML |
200 | @item @code{!=} as relational-op in canonical loop form for C/C++ @tab Y @tab |
201 | @item @code{nonmonotonic} as default loop schedule modifier for worksharing-loop | |
202 | constructs @tab Y @tab | |
203 | @item Collapse of associated loops that are imperfectly nested loops @tab N @tab | |
204 | @item Clauses @code{if}, @code{nontemporal} and @code{order(concurrent)} in | |
205 | @code{simd} construct @tab Y @tab | |
206 | @item @code{atomic} constructs in @code{simd} @tab Y @tab | |
207 | @item @code{loop} construct @tab Y @tab | |
208 | @item @code{order(concurrent)} clause @tab Y @tab | |
209 | @item @code{scan} directive and @code{in_scan} modifier for the | |
210 | @code{reduction} clause @tab Y @tab | |
211 | @item @code{in_reduction} clause on @code{task} constructs @tab Y @tab | |
212 | @item @code{in_reduction} clause on @code{target} constructs @tab P | |
213 | @tab @code{nowait} only stub | |
214 | @item @code{task_reduction} clause with @code{taskgroup} @tab Y @tab | |
215 | @item @code{task} modifier to @code{reduction} clause @tab Y @tab | |
216 | @item @code{affinity} clause to @code{task} construct @tab Y @tab Stub only | |
217 | @item @code{detach} clause to @code{task} construct @tab Y @tab | |
218 | @item @code{omp_fulfill_event} runtime routine @tab Y @tab | |
219 | @item @code{reduction} and @code{in_reduction} clauses on @code{taskloop} | |
220 | and @code{taskloop simd} constructs @tab Y @tab | |
221 | @item @code{taskloop} construct cancelable by @code{cancel} construct | |
222 | @tab Y @tab | |
223 | @item @code{mutexinoutset} @emph{dependence-type} for @code{depend} clause | |
224 | @tab Y @tab | |
225 | @item Predefined memory spaces, memory allocators, allocator traits | |
13c3e29d | 226 | @tab Y @tab See also @ref{Memory allocation} |
d77de738 ML |
227 | @item Memory management routines @tab Y @tab |
228 | @item @code{allocate} directive @tab N @tab | |
229 | @item @code{allocate} clause @tab P @tab Initial support | |
230 | @item @code{use_device_addr} clause on @code{target data} @tab Y @tab | |
f84fdb13 | 231 | @item @code{ancestor} modifier on @code{device} clause @tab Y @tab |
d77de738 ML |
232 | @item Implicit declare target directive @tab Y @tab |
233 | @item Discontiguous array section with @code{target update} construct | |
234 | @tab N @tab | |
235 | @item C/C++'s lvalue expressions in @code{to}, @code{from} | |
236 | and @code{map} clauses @tab N @tab | |
237 | @item C/C++'s lvalue expressions in @code{depend} clauses @tab Y @tab | |
238 | @item Nested @code{declare target} directive @tab Y @tab | |
239 | @item Combined @code{master} constructs @tab Y @tab | |
240 | @item @code{depend} clause on @code{taskwait} @tab Y @tab | |
241 | @item Weak memory ordering clauses on @code{atomic} and @code{flush} construct | |
242 | @tab Y @tab | |
243 | @item @code{hint} clause on the @code{atomic} construct @tab Y @tab Stub only | |
244 | @item @code{depobj} construct and depend objects @tab Y @tab | |
245 | @item Lock hints were renamed to synchronization hints @tab Y @tab | |
246 | @item @code{conditional} modifier to @code{lastprivate} clause @tab Y @tab | |
247 | @item Map-order clarifications @tab P @tab | |
248 | @item @code{close} @emph{map-type-modifier} @tab Y @tab | |
249 | @item Mapping C/C++ pointer variables and to assign the address of | |
250 | device memory mapped by an array section @tab P @tab | |
251 | @item Mapping of Fortran pointer and allocatable variables, including pointer | |
252 | and allocatable components of variables | |
253 | @tab P @tab Mapping of vars with allocatable components unsupported | |
254 | @item @code{defaultmap} extensions @tab Y @tab | |
255 | @item @code{declare mapper} directive @tab N @tab | |
256 | @item @code{omp_get_supported_active_levels} routine @tab Y @tab | |
257 | @item Runtime routines and environment variables to display runtime thread | |
258 | affinity information @tab Y @tab | |
259 | @item @code{omp_pause_resource} and @code{omp_pause_resource_all} runtime | |
260 | routines @tab Y @tab | |
261 | @item @code{omp_get_device_num} runtime routine @tab Y @tab | |
262 | @item OMPT interface @tab N @tab | |
263 | @item OMPD interface @tab N @tab | |
264 | @end multitable | |
265 | ||
266 | @unnumberedsubsec Other new OpenMP 5.0 features | |
267 | ||
268 | @multitable @columnfractions .60 .10 .25 | |
269 | @headitem Description @tab Status @tab Comments | |
270 | @item Supporting C++'s range-based for loop @tab Y @tab | |
271 | @end multitable | |
272 | ||
273 | ||
274 | @node OpenMP 5.1 | |
275 | @section OpenMP 5.1 | |
276 | ||
277 | @unnumberedsubsec New features listed in Appendix B of the OpenMP specification | |
278 | ||
279 | @multitable @columnfractions .60 .10 .25 | |
280 | @headitem Description @tab Status @tab Comments | |
281 | @item OpenMP directive as C++ attribute specifiers @tab Y @tab | |
282 | @item @code{omp_all_memory} reserved locator @tab Y @tab | |
283 | @item @emph{target_device trait} in OpenMP Context @tab N @tab | |
284 | @item @code{target_device} selector set in context selectors @tab N @tab | |
285 | @item C/C++'s @code{declare variant} directive: elision support of | |
286 | preprocessed code @tab N @tab | |
287 | @item @code{declare variant}: new clauses @code{adjust_args} and | |
288 | @code{append_args} @tab N @tab | |
289 | @item @code{dispatch} construct @tab N @tab | |
290 | @item device-specific ICV settings with environment variables @tab Y @tab | |
eda38850 | 291 | @item @code{assume} and @code{assumes} directives @tab Y @tab |
d77de738 ML |
292 | @item @code{nothing} directive @tab Y @tab |
293 | @item @code{error} directive @tab Y @tab | |
294 | @item @code{masked} construct @tab Y @tab | |
295 | @item @code{scope} directive @tab Y @tab | |
296 | @item Loop transformation constructs @tab N @tab | |
297 | @item @code{strict} modifier in the @code{grainsize} and @code{num_tasks} | |
298 | clauses of the @code{taskloop} construct @tab Y @tab | |
b2e1c49b TB |
299 | @item @code{align} clause in @code{allocate} directive @tab N @tab |
300 | @item @code{align} modifier in @code{allocate} clause @tab Y @tab | |
d77de738 ML |
301 | @item @code{thread_limit} clause to @code{target} construct @tab Y @tab |
302 | @item @code{has_device_addr} clause to @code{target} construct @tab Y @tab | |
303 | @item Iterators in @code{target update} motion clauses and @code{map} | |
304 | clauses @tab N @tab | |
305 | @item Indirect calls to the device version of a procedure or function in | |
306 | @code{target} regions @tab N @tab | |
307 | @item @code{interop} directive @tab N @tab | |
308 | @item @code{omp_interop_t} object support in runtime routines @tab N @tab | |
309 | @item @code{nowait} clause in @code{taskwait} directive @tab Y @tab | |
310 | @item Extensions to the @code{atomic} directive @tab Y @tab | |
311 | @item @code{seq_cst} clause on a @code{flush} construct @tab Y @tab | |
312 | @item @code{inoutset} argument to the @code{depend} clause @tab Y @tab | |
313 | @item @code{private} and @code{firstprivate} argument to @code{default} | |
314 | clause in C and C++ @tab Y @tab | |
4ede915d | 315 | @item @code{present} argument to @code{defaultmap} clause @tab Y @tab |
d77de738 ML |
316 | @item @code{omp_set_num_teams}, @code{omp_set_teams_thread_limit}, |
317 | @code{omp_get_max_teams}, @code{omp_get_teams_thread_limit} runtime | |
318 | routines @tab Y @tab | |
319 | @item @code{omp_target_is_accessible} runtime routine @tab Y @tab | |
320 | @item @code{omp_target_memcpy_async} and @code{omp_target_memcpy_rect_async} | |
321 | runtime routines @tab Y @tab | |
322 | @item @code{omp_get_mapped_ptr} runtime routine @tab Y @tab | |
323 | @item @code{omp_calloc}, @code{omp_realloc}, @code{omp_aligned_alloc} and | |
324 | @code{omp_aligned_calloc} runtime routines @tab Y @tab | |
325 | @item @code{omp_alloctrait_key_t} enum: @code{omp_atv_serialized} added, | |
326 | @code{omp_atv_default} changed @tab Y @tab | |
327 | @item @code{omp_display_env} runtime routine @tab Y @tab | |
328 | @item @code{ompt_scope_endpoint_t} enum: @code{ompt_scope_beginend} @tab N @tab | |
329 | @item @code{ompt_sync_region_t} enum additions @tab N @tab | |
330 | @item @code{ompt_state_t} enum: @code{ompt_state_wait_barrier_implementation} | |
331 | and @code{ompt_state_wait_barrier_teams} @tab N @tab | |
332 | @item @code{ompt_callback_target_data_op_emi_t}, | |
333 | @code{ompt_callback_target_emi_t}, @code{ompt_callback_target_map_emi_t} | |
334 | and @code{ompt_callback_target_submit_emi_t} @tab N @tab | |
335 | @item @code{ompt_callback_error_t} type @tab N @tab | |
336 | @item @code{OMP_PLACES} syntax extensions @tab Y @tab | |
337 | @item @code{OMP_NUM_TEAMS} and @code{OMP_TEAMS_THREAD_LIMIT} environment | |
338 | variables @tab Y @tab | |
339 | @end multitable | |
340 | ||
341 | @unnumberedsubsec Other new OpenMP 5.1 features | |
342 | ||
343 | @multitable @columnfractions .60 .10 .25 | |
344 | @headitem Description @tab Status @tab Comments | |
345 | @item Support of strictly structured blocks in Fortran @tab Y @tab | |
346 | @item Support of structured block sequences in C/C++ @tab Y @tab | |
347 | @item @code{unconstrained} and @code{reproducible} modifiers on @code{order} | |
348 | clause @tab Y @tab | |
349 | @item Support @code{begin/end declare target} syntax in C/C++ @tab Y @tab | |
350 | @item Pointer predetermined firstprivate getting initialized | |
351 | to address of matching mapped list item per 5.1, Sect. 2.21.7.2 @tab N @tab | |
352 | @item For Fortran, diagnose placing declarative before/between @code{USE}, | |
353 | @code{IMPORT}, and @code{IMPLICIT} as invalid @tab N @tab | |
eda38850 | 354 | @item Optional comma between directive and clause in the @code{#pragma} form @tab Y @tab |
c16e85d7 TB |
355 | @item @code{indirect} clause in @code{declare target} @tab N @tab |
356 | @item @code{device_type(nohost)}/@code{device_type(host)} for variables @tab N @tab | |
4ede915d TB |
357 | @item @code{present} modifier to the @code{map}, @code{to} and @code{from} |
358 | clauses @tab Y @tab | |
d77de738 ML |
359 | @end multitable |
360 | ||
361 | ||
362 | @node OpenMP 5.2 | |
363 | @section OpenMP 5.2 | |
364 | ||
365 | @unnumberedsubsec New features listed in Appendix B of the OpenMP specification | |
366 | ||
367 | @multitable @columnfractions .60 .10 .25 | |
368 | @headitem Description @tab Status @tab Comments | |
2cd0689a | 369 | @item @code{omp_in_explicit_task} routine and @var{explicit-task-var} ICV |
d77de738 ML |
370 | @tab Y @tab |
371 | @item @code{omp}/@code{ompx}/@code{omx} sentinels and @code{omp_}/@code{ompx_} | |
372 | namespaces @tab N/A | |
373 | @tab warning for @code{ompx/omx} sentinels@footnote{The @code{ompx} | |
374 | sentinel as C/C++ pragma and C++ attributes are warned for with | |
375 | @code{-Wunknown-pragmas} (implied by @code{-Wall}) and @code{-Wattributes} | |
376 | (enabled by default), respectively; for Fortran free-source code, there is | |
377 | a warning enabled by default and, for fixed-source code, the @code{omx} | |
378 | sentinel is warned for with with @code{-Wsurprising} (enabled by | |
379 | @code{-Wall}). Unknown clauses are always rejected with an error.} | |
091b6dbc | 380 | @item Clauses on @code{end} directive can be on directive @tab Y @tab |
d77de738 ML |
381 | @item Deprecation of no-argument @code{destroy} clause on @code{depobj} |
382 | @tab N @tab | |
383 | @item @code{linear} clause syntax changes and @code{step} modifier @tab Y @tab | |
384 | @item Deprecation of minus operator for reductions @tab N @tab | |
385 | @item Deprecation of separating @code{map} modifiers without comma @tab N @tab | |
386 | @item @code{declare mapper} with iterator and @code{present} modifiers | |
387 | @tab N @tab | |
388 | @item If a matching mapped list item is not found in the data environment, the | |
b25ea7ab | 389 | pointer retains its original value @tab Y @tab |
d77de738 ML |
390 | @item New @code{enter} clause as alias for @code{to} on declare target directive |
391 | @tab Y @tab | |
392 | @item Deprecation of @code{to} clause on declare target directive @tab N @tab | |
393 | @item Extended list of directives permitted in Fortran pure procedures | |
2df7e451 | 394 | @tab Y @tab |
d77de738 ML |
395 | @item New @code{allocators} directive for Fortran @tab N @tab |
396 | @item Deprecation of @code{allocate} directive for Fortran | |
397 | allocatables/pointers @tab N @tab | |
398 | @item Optional paired @code{end} directive with @code{dispatch} @tab N @tab | |
399 | @item New @code{memspace} and @code{traits} modifiers for @code{uses_allocators} | |
400 | @tab N @tab | |
401 | @item Deprecation of traits array following the allocator_handle expression in | |
402 | @code{uses_allocators} @tab N @tab | |
403 | @item New @code{otherwise} clause as alias for @code{default} on metadirectives | |
404 | @tab N @tab | |
405 | @item Deprecation of @code{default} clause on metadirectives @tab N @tab | |
406 | @item Deprecation of delimited form of @code{declare target} @tab N @tab | |
407 | @item Reproducible semantics changed for @code{order(concurrent)} @tab N @tab | |
408 | @item @code{allocate} and @code{firstprivate} clauses on @code{scope} | |
409 | @tab Y @tab | |
410 | @item @code{ompt_callback_work} @tab N @tab | |
9f80367e | 411 | @item Default map-type for the @code{map} clause in @code{target enter/exit data} |
d77de738 ML |
412 | @tab Y @tab |
413 | @item New @code{doacross} clause as alias for @code{depend} with | |
414 | @code{source}/@code{sink} modifier @tab Y @tab | |
415 | @item Deprecation of @code{depend} with @code{source}/@code{sink} modifier | |
416 | @tab N @tab | |
417 | @item @code{omp_cur_iteration} keyword @tab Y @tab | |
418 | @end multitable | |
419 | ||
420 | @unnumberedsubsec Other new OpenMP 5.2 features | |
421 | ||
422 | @multitable @columnfractions .60 .10 .25 | |
423 | @headitem Description @tab Status @tab Comments | |
424 | @item For Fortran, optional comma between directive and clause @tab N @tab | |
425 | @item Conforming device numbers and @code{omp_initial_device} and | |
426 | @code{omp_invalid_device} enum/PARAMETER @tab Y @tab | |
2cd0689a | 427 | @item Initial value of @var{default-device-var} ICV with |
18c8b56c | 428 | @code{OMP_TARGET_OFFLOAD=mandatory} @tab Y @tab |
d77de738 ML |
429 | @item @emph{interop_types} in any position of the modifier list for the @code{init} clause |
430 | of the @code{interop} construct @tab N @tab | |
431 | @end multitable | |
432 | ||
433 | ||
c16e85d7 TB |
434 | @node OpenMP Technical Report 11 |
435 | @section OpenMP Technical Report 11 | |
436 | ||
437 | Technical Report (TR) 11 is the first preview for OpenMP 6.0. | |
438 | ||
439 | @unnumberedsubsec New features listed in Appendix B of the OpenMP specification | |
440 | @multitable @columnfractions .60 .10 .25 | |
441 | @item Features deprecated in versions 5.2, 5.1 and 5.0 were removed | |
442 | @tab N/A @tab Backward compatibility | |
443 | @item The @code{decl} attribute was added to the C++ attribute syntax | |
444 | @tab N @tab | |
445 | @item @code{_ALL} suffix to the device-scope environment variables | |
446 | @tab P @tab Host device number wrongly accepted | |
447 | @item For Fortran, @emph{locator list} can be also function reference with | |
448 | data pointer result @tab N @tab | |
449 | @item Ref-count change for @code{use_device_ptr}/@code{use_device_addr} | |
450 | @tab N @tab | |
451 | @item Implicit reduction identifiers of C++ classes | |
452 | @tab N @tab | |
453 | @item Change of the @emph{map-type} property from @emph{ultimate} to | |
454 | @emph{default} @tab N @tab | |
455 | @item Concept of @emph{assumed-size arrays} in C and C++ | |
456 | @tab N @tab | |
457 | @item Mapping of @emph{assumed-size arrays} in C, C++ and Fortran | |
458 | @tab N @tab | |
459 | @item @code{groupprivate} directive @tab N @tab | |
460 | @item @code{local} clause to declare target directive @tab N @tab | |
461 | @item @code{part_size} allocator trait @tab N @tab | |
462 | @item @code{pin_device}, @code{preferred_device} and @code{target_access} | |
463 | allocator traits | |
464 | @tab N @tab | |
465 | @item @code{access} allocator trait changes @tab N @tab | |
466 | @item Extension of @code{interop} operation of @code{append_args}, allowing all | |
467 | modifiers of the @code{init} clause | |
9f80367e | 468 | @tab N @tab |
c16e85d7 TB |
469 | @item @code{interop} clause to @code{dispatch} @tab N @tab |
470 | @item @code{apply} code to loop-transforming constructs @tab N @tab | |
471 | @item @code{omp_curr_progress_width} identifier @tab N @tab | |
472 | @item @code{safesync} clause to the @code{parallel} construct @tab N @tab | |
473 | @item @code{omp_get_max_progress_width} runtime routine @tab N @tab | |
8da7476c | 474 | @item @code{strict} modifier keyword to @code{num_threads} @tab N @tab |
c16e85d7 TB |
475 | @item @code{memscope} clause to @code{atomic} and @code{flush} @tab N @tab |
476 | @item Routines for obtaining memory spaces/allocators for shared/device memory | |
477 | @tab N @tab | |
478 | @item @code{omp_get_memspace_num_resources} routine @tab N @tab | |
479 | @item @code{omp_get_submemspace} routine @tab N @tab | |
480 | @item @code{ompt_get_buffer_limits} OMPT routine @tab N @tab | |
481 | @item Extension of @code{OMP_DEFAULT_DEVICE} and new | |
482 | @code{OMP_AVAILABLE_DEVICES} environment vars @tab N @tab | |
483 | @item Supporting increments with abstract names in @code{OMP_PLACES} @tab N @tab | |
484 | @end multitable | |
485 | ||
486 | @unnumberedsubsec Other new TR 11 features | |
487 | @multitable @columnfractions .60 .10 .25 | |
488 | @item Relaxed Fortran restrictions to the @code{aligned} clause @tab N @tab | |
489 | @item Mapping lambda captures @tab N @tab | |
490 | @item For Fortran, atomic compare with storing the comparison result | |
491 | @tab N @tab | |
c16e85d7 TB |
492 | @end multitable |
493 | ||
494 | ||
495 | ||
d77de738 ML |
496 | @c --------------------------------------------------------------------- |
497 | @c OpenMP Runtime Library Routines | |
498 | @c --------------------------------------------------------------------- | |
499 | ||
500 | @node Runtime Library Routines | |
501 | @chapter OpenMP Runtime Library Routines | |
502 | ||
503 | The runtime routines described here are defined by Section 3 of the OpenMP | |
504 | specification in version 4.5. The routines are structured in following | |
505 | three parts: | |
506 | ||
507 | @menu | |
508 | Control threads, processors and the parallel environment. They have C | |
509 | linkage, and do not throw exceptions. | |
510 | ||
511 | * omp_get_active_level:: Number of active parallel regions | |
512 | * omp_get_ancestor_thread_num:: Ancestor thread ID | |
513 | * omp_get_cancellation:: Whether cancellation support is enabled | |
514 | * omp_get_default_device:: Get the default device for target regions | |
515 | * omp_get_device_num:: Get device that current thread is running on | |
516 | * omp_get_dynamic:: Dynamic teams setting | |
517 | * omp_get_initial_device:: Device number of host device | |
518 | * omp_get_level:: Number of parallel regions | |
519 | * omp_get_max_active_levels:: Current maximum number of active regions | |
520 | * omp_get_max_task_priority:: Maximum task priority value that can be set | |
521 | * omp_get_max_teams:: Maximum number of teams for teams region | |
522 | * omp_get_max_threads:: Maximum number of threads of parallel region | |
523 | * omp_get_nested:: Nested parallel regions | |
524 | * omp_get_num_devices:: Number of target devices | |
525 | * omp_get_num_procs:: Number of processors online | |
526 | * omp_get_num_teams:: Number of teams | |
527 | * omp_get_num_threads:: Size of the active team | |
0b9bd33d | 528 | * omp_get_proc_bind:: Whether threads may be moved between CPUs |
d77de738 ML |
529 | * omp_get_schedule:: Obtain the runtime scheduling method |
530 | * omp_get_supported_active_levels:: Maximum number of active regions supported | |
531 | * omp_get_team_num:: Get team number | |
532 | * omp_get_team_size:: Number of threads in a team | |
533 | * omp_get_teams_thread_limit:: Maximum number of threads imposed by teams | |
534 | * omp_get_thread_limit:: Maximum number of threads | |
535 | * omp_get_thread_num:: Current thread ID | |
536 | * omp_in_parallel:: Whether a parallel region is active | |
537 | * omp_in_final:: Whether in final or included task region | |
538 | * omp_is_initial_device:: Whether executing on the host device | |
539 | * omp_set_default_device:: Set the default device for target regions | |
540 | * omp_set_dynamic:: Enable/disable dynamic teams | |
541 | * omp_set_max_active_levels:: Limits the number of active parallel regions | |
542 | * omp_set_nested:: Enable/disable nested parallel regions | |
543 | * omp_set_num_teams:: Set upper teams limit for teams region | |
544 | * omp_set_num_threads:: Set upper team size limit | |
545 | * omp_set_schedule:: Set the runtime scheduling method | |
546 | * omp_set_teams_thread_limit:: Set upper thread limit for teams construct | |
547 | ||
548 | Initialize, set, test, unset and destroy simple and nested locks. | |
549 | ||
550 | * omp_init_lock:: Initialize simple lock | |
551 | * omp_set_lock:: Wait for and set simple lock | |
552 | * omp_test_lock:: Test and set simple lock if available | |
553 | * omp_unset_lock:: Unset simple lock | |
554 | * omp_destroy_lock:: Destroy simple lock | |
555 | * omp_init_nest_lock:: Initialize nested lock | |
556 | * omp_set_nest_lock:: Wait for and set simple lock | |
557 | * omp_test_nest_lock:: Test and set nested lock if available | |
558 | * omp_unset_nest_lock:: Unset nested lock | |
559 | * omp_destroy_nest_lock:: Destroy nested lock | |
560 | ||
561 | Portable, thread-based, wall clock timer. | |
562 | ||
563 | * omp_get_wtick:: Get timer precision. | |
564 | * omp_get_wtime:: Elapsed wall clock time. | |
565 | ||
566 | Support for event objects. | |
567 | ||
568 | * omp_fulfill_event:: Fulfill and destroy an OpenMP event. | |
569 | @end menu | |
570 | ||
571 | ||
572 | ||
573 | @node omp_get_active_level | |
574 | @section @code{omp_get_active_level} -- Number of parallel regions | |
575 | @table @asis | |
576 | @item @emph{Description}: | |
577 | This function returns the nesting level for the active parallel blocks, | |
578 | which enclose the calling call. | |
579 | ||
580 | @item @emph{C/C++} | |
581 | @multitable @columnfractions .20 .80 | |
582 | @item @emph{Prototype}: @tab @code{int omp_get_active_level(void);} | |
583 | @end multitable | |
584 | ||
585 | @item @emph{Fortran}: | |
586 | @multitable @columnfractions .20 .80 | |
587 | @item @emph{Interface}: @tab @code{integer function omp_get_active_level()} | |
588 | @end multitable | |
589 | ||
590 | @item @emph{See also}: | |
591 | @ref{omp_get_level}, @ref{omp_get_max_active_levels}, @ref{omp_set_max_active_levels} | |
592 | ||
593 | @item @emph{Reference}: | |
594 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.20. | |
595 | @end table | |
596 | ||
597 | ||
598 | ||
599 | @node omp_get_ancestor_thread_num | |
600 | @section @code{omp_get_ancestor_thread_num} -- Ancestor thread ID | |
601 | @table @asis | |
602 | @item @emph{Description}: | |
603 | This function returns the thread identification number for the given | |
604 | nesting level of the current thread. For values of @var{level} outside | |
605 | zero to @code{omp_get_level} -1 is returned; if @var{level} is | |
606 | @code{omp_get_level} the result is identical to @code{omp_get_thread_num}. | |
607 | ||
608 | @item @emph{C/C++} | |
609 | @multitable @columnfractions .20 .80 | |
610 | @item @emph{Prototype}: @tab @code{int omp_get_ancestor_thread_num(int level);} | |
611 | @end multitable | |
612 | ||
613 | @item @emph{Fortran}: | |
614 | @multitable @columnfractions .20 .80 | |
615 | @item @emph{Interface}: @tab @code{integer function omp_get_ancestor_thread_num(level)} | |
616 | @item @tab @code{integer level} | |
617 | @end multitable | |
618 | ||
619 | @item @emph{See also}: | |
620 | @ref{omp_get_level}, @ref{omp_get_thread_num}, @ref{omp_get_team_size} | |
621 | ||
622 | @item @emph{Reference}: | |
623 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.18. | |
624 | @end table | |
625 | ||
626 | ||
627 | ||
628 | @node omp_get_cancellation | |
629 | @section @code{omp_get_cancellation} -- Whether cancellation support is enabled | |
630 | @table @asis | |
631 | @item @emph{Description}: | |
632 | This function returns @code{true} if cancellation is activated, @code{false} | |
633 | otherwise. Here, @code{true} and @code{false} represent their language-specific | |
634 | counterparts. Unless @env{OMP_CANCELLATION} is set true, cancellations are | |
635 | deactivated. | |
636 | ||
637 | @item @emph{C/C++}: | |
638 | @multitable @columnfractions .20 .80 | |
639 | @item @emph{Prototype}: @tab @code{int omp_get_cancellation(void);} | |
640 | @end multitable | |
641 | ||
642 | @item @emph{Fortran}: | |
643 | @multitable @columnfractions .20 .80 | |
644 | @item @emph{Interface}: @tab @code{logical function omp_get_cancellation()} | |
645 | @end multitable | |
646 | ||
647 | @item @emph{See also}: | |
648 | @ref{OMP_CANCELLATION} | |
649 | ||
650 | @item @emph{Reference}: | |
651 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.9. | |
652 | @end table | |
653 | ||
654 | ||
655 | ||
656 | @node omp_get_default_device | |
657 | @section @code{omp_get_default_device} -- Get the default device for target regions | |
658 | @table @asis | |
659 | @item @emph{Description}: | |
660 | Get the default device for target regions without device clause. | |
661 | ||
662 | @item @emph{C/C++}: | |
663 | @multitable @columnfractions .20 .80 | |
664 | @item @emph{Prototype}: @tab @code{int omp_get_default_device(void);} | |
665 | @end multitable | |
666 | ||
667 | @item @emph{Fortran}: | |
668 | @multitable @columnfractions .20 .80 | |
669 | @item @emph{Interface}: @tab @code{integer function omp_get_default_device()} | |
670 | @end multitable | |
671 | ||
672 | @item @emph{See also}: | |
673 | @ref{OMP_DEFAULT_DEVICE}, @ref{omp_set_default_device} | |
674 | ||
675 | @item @emph{Reference}: | |
676 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.30. | |
677 | @end table | |
678 | ||
679 | ||
680 | ||
681 | @node omp_get_device_num | |
682 | @section @code{omp_get_device_num} -- Return device number of current device | |
683 | @table @asis | |
684 | @item @emph{Description}: | |
685 | This function returns a device number that represents the device that the | |
686 | current thread is executing on. For OpenMP 5.0, this must be equal to the | |
687 | value returned by the @code{omp_get_initial_device} function when called | |
688 | from the host. | |
689 | ||
690 | @item @emph{C/C++} | |
691 | @multitable @columnfractions .20 .80 | |
692 | @item @emph{Prototype}: @tab @code{int omp_get_device_num(void);} | |
693 | @end multitable | |
694 | ||
695 | @item @emph{Fortran}: | |
696 | @multitable @columnfractions .20 .80 | |
697 | @item @emph{Interface}: @tab @code{integer function omp_get_device_num()} | |
698 | @end multitable | |
699 | ||
700 | @item @emph{See also}: | |
701 | @ref{omp_get_initial_device} | |
702 | ||
703 | @item @emph{Reference}: | |
704 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.2.37. | |
705 | @end table | |
706 | ||
707 | ||
708 | ||
709 | @node omp_get_dynamic | |
710 | @section @code{omp_get_dynamic} -- Dynamic teams setting | |
711 | @table @asis | |
712 | @item @emph{Description}: | |
713 | This function returns @code{true} if enabled, @code{false} otherwise. | |
714 | Here, @code{true} and @code{false} represent their language-specific | |
715 | counterparts. | |
716 | ||
717 | The dynamic team setting may be initialized at startup by the | |
718 | @env{OMP_DYNAMIC} environment variable or at runtime using | |
719 | @code{omp_set_dynamic}. If undefined, dynamic adjustment is | |
720 | disabled by default. | |
721 | ||
722 | @item @emph{C/C++}: | |
723 | @multitable @columnfractions .20 .80 | |
724 | @item @emph{Prototype}: @tab @code{int omp_get_dynamic(void);} | |
725 | @end multitable | |
726 | ||
727 | @item @emph{Fortran}: | |
728 | @multitable @columnfractions .20 .80 | |
729 | @item @emph{Interface}: @tab @code{logical function omp_get_dynamic()} | |
730 | @end multitable | |
731 | ||
732 | @item @emph{See also}: | |
733 | @ref{omp_set_dynamic}, @ref{OMP_DYNAMIC} | |
734 | ||
735 | @item @emph{Reference}: | |
736 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.8. | |
737 | @end table | |
738 | ||
739 | ||
740 | ||
741 | @node omp_get_initial_device | |
742 | @section @code{omp_get_initial_device} -- Return device number of initial device | |
743 | @table @asis | |
744 | @item @emph{Description}: | |
745 | This function returns a device number that represents the host device. | |
746 | For OpenMP 5.1, this must be equal to the value returned by the | |
747 | @code{omp_get_num_devices} function. | |
748 | ||
749 | @item @emph{C/C++} | |
750 | @multitable @columnfractions .20 .80 | |
751 | @item @emph{Prototype}: @tab @code{int omp_get_initial_device(void);} | |
752 | @end multitable | |
753 | ||
754 | @item @emph{Fortran}: | |
755 | @multitable @columnfractions .20 .80 | |
756 | @item @emph{Interface}: @tab @code{integer function omp_get_initial_device()} | |
757 | @end multitable | |
758 | ||
759 | @item @emph{See also}: | |
760 | @ref{omp_get_num_devices} | |
761 | ||
762 | @item @emph{Reference}: | |
763 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.35. | |
764 | @end table | |
765 | ||
766 | ||
767 | ||
768 | @node omp_get_level | |
769 | @section @code{omp_get_level} -- Obtain the current nesting level | |
770 | @table @asis | |
771 | @item @emph{Description}: | |
772 | This function returns the nesting level for the parallel blocks, | |
773 | which enclose the calling call. | |
774 | ||
775 | @item @emph{C/C++} | |
776 | @multitable @columnfractions .20 .80 | |
777 | @item @emph{Prototype}: @tab @code{int omp_get_level(void);} | |
778 | @end multitable | |
779 | ||
780 | @item @emph{Fortran}: | |
781 | @multitable @columnfractions .20 .80 | |
782 | @item @emph{Interface}: @tab @code{integer function omp_level()} | |
783 | @end multitable | |
784 | ||
785 | @item @emph{See also}: | |
786 | @ref{omp_get_active_level} | |
787 | ||
788 | @item @emph{Reference}: | |
789 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.17. | |
790 | @end table | |
791 | ||
792 | ||
793 | ||
794 | @node omp_get_max_active_levels | |
795 | @section @code{omp_get_max_active_levels} -- Current maximum number of active regions | |
796 | @table @asis | |
797 | @item @emph{Description}: | |
798 | This function obtains the maximum allowed number of nested, active parallel regions. | |
799 | ||
800 | @item @emph{C/C++} | |
801 | @multitable @columnfractions .20 .80 | |
802 | @item @emph{Prototype}: @tab @code{int omp_get_max_active_levels(void);} | |
803 | @end multitable | |
804 | ||
805 | @item @emph{Fortran}: | |
806 | @multitable @columnfractions .20 .80 | |
807 | @item @emph{Interface}: @tab @code{integer function omp_get_max_active_levels()} | |
808 | @end multitable | |
809 | ||
810 | @item @emph{See also}: | |
811 | @ref{omp_set_max_active_levels}, @ref{omp_get_active_level} | |
812 | ||
813 | @item @emph{Reference}: | |
814 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.16. | |
815 | @end table | |
816 | ||
817 | ||
818 | @node omp_get_max_task_priority | |
819 | @section @code{omp_get_max_task_priority} -- Maximum priority value | |
820 | that can be set for tasks. | |
821 | @table @asis | |
822 | @item @emph{Description}: | |
823 | This function obtains the maximum allowed priority number for tasks. | |
824 | ||
825 | @item @emph{C/C++} | |
826 | @multitable @columnfractions .20 .80 | |
827 | @item @emph{Prototype}: @tab @code{int omp_get_max_task_priority(void);} | |
828 | @end multitable | |
829 | ||
830 | @item @emph{Fortran}: | |
831 | @multitable @columnfractions .20 .80 | |
832 | @item @emph{Interface}: @tab @code{integer function omp_get_max_task_priority()} | |
833 | @end multitable | |
834 | ||
835 | @item @emph{Reference}: | |
836 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.29. | |
837 | @end table | |
838 | ||
839 | ||
840 | @node omp_get_max_teams | |
841 | @section @code{omp_get_max_teams} -- Maximum number of teams of teams region | |
842 | @table @asis | |
843 | @item @emph{Description}: | |
844 | Return the maximum number of teams used for the teams region | |
845 | that does not use the clause @code{num_teams}. | |
846 | ||
847 | @item @emph{C/C++}: | |
848 | @multitable @columnfractions .20 .80 | |
849 | @item @emph{Prototype}: @tab @code{int omp_get_max_teams(void);} | |
850 | @end multitable | |
851 | ||
852 | @item @emph{Fortran}: | |
853 | @multitable @columnfractions .20 .80 | |
854 | @item @emph{Interface}: @tab @code{integer function omp_get_max_teams()} | |
855 | @end multitable | |
856 | ||
857 | @item @emph{See also}: | |
858 | @ref{omp_set_num_teams}, @ref{omp_get_num_teams} | |
859 | ||
860 | @item @emph{Reference}: | |
861 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.4. | |
862 | @end table | |
863 | ||
864 | ||
865 | ||
866 | @node omp_get_max_threads | |
867 | @section @code{omp_get_max_threads} -- Maximum number of threads of parallel region | |
868 | @table @asis | |
869 | @item @emph{Description}: | |
870 | Return the maximum number of threads used for the current parallel region | |
871 | that does not use the clause @code{num_threads}. | |
872 | ||
873 | @item @emph{C/C++}: | |
874 | @multitable @columnfractions .20 .80 | |
875 | @item @emph{Prototype}: @tab @code{int omp_get_max_threads(void);} | |
876 | @end multitable | |
877 | ||
878 | @item @emph{Fortran}: | |
879 | @multitable @columnfractions .20 .80 | |
880 | @item @emph{Interface}: @tab @code{integer function omp_get_max_threads()} | |
881 | @end multitable | |
882 | ||
883 | @item @emph{See also}: | |
884 | @ref{omp_set_num_threads}, @ref{omp_set_dynamic}, @ref{omp_get_thread_limit} | |
885 | ||
886 | @item @emph{Reference}: | |
887 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.3. | |
888 | @end table | |
889 | ||
890 | ||
891 | ||
892 | @node omp_get_nested | |
893 | @section @code{omp_get_nested} -- Nested parallel regions | |
894 | @table @asis | |
895 | @item @emph{Description}: | |
896 | This function returns @code{true} if nested parallel regions are | |
897 | enabled, @code{false} otherwise. Here, @code{true} and @code{false} | |
898 | represent their language-specific counterparts. | |
899 | ||
900 | The state of nested parallel regions at startup depends on several | |
901 | environment variables. If @env{OMP_MAX_ACTIVE_LEVELS} is defined | |
902 | and is set to greater than one, then nested parallel regions will be | |
903 | enabled. If not defined, then the value of the @env{OMP_NESTED} | |
904 | environment variable will be followed if defined. If neither are | |
905 | defined, then if either @env{OMP_NUM_THREADS} or @env{OMP_PROC_BIND} | |
906 | are defined with a list of more than one value, then nested parallel | |
907 | regions are enabled. If none of these are defined, then nested parallel | |
908 | regions are disabled by default. | |
909 | ||
910 | Nested parallel regions can be enabled or disabled at runtime using | |
911 | @code{omp_set_nested}, or by setting the maximum number of nested | |
912 | regions with @code{omp_set_max_active_levels} to one to disable, or | |
913 | above one to enable. | |
914 | ||
2cd0689a TB |
915 | Note that the @code{omp_get_nested} API routine was deprecated |
916 | in the OpenMP specification 5.2 in favor of @code{omp_get_max_active_levels}. | |
917 | ||
d77de738 ML |
918 | @item @emph{C/C++}: |
919 | @multitable @columnfractions .20 .80 | |
920 | @item @emph{Prototype}: @tab @code{int omp_get_nested(void);} | |
921 | @end multitable | |
922 | ||
923 | @item @emph{Fortran}: | |
924 | @multitable @columnfractions .20 .80 | |
925 | @item @emph{Interface}: @tab @code{logical function omp_get_nested()} | |
926 | @end multitable | |
927 | ||
928 | @item @emph{See also}: | |
2cd0689a | 929 | @ref{omp_get_max_active_levels}, @ref{omp_set_nested}, |
d77de738 ML |
930 | @ref{OMP_MAX_ACTIVE_LEVELS}, @ref{OMP_NESTED} |
931 | ||
932 | @item @emph{Reference}: | |
933 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.11. | |
934 | @end table | |
935 | ||
936 | ||
937 | ||
938 | @node omp_get_num_devices | |
939 | @section @code{omp_get_num_devices} -- Number of target devices | |
940 | @table @asis | |
941 | @item @emph{Description}: | |
942 | Returns the number of target devices. | |
943 | ||
944 | @item @emph{C/C++}: | |
945 | @multitable @columnfractions .20 .80 | |
946 | @item @emph{Prototype}: @tab @code{int omp_get_num_devices(void);} | |
947 | @end multitable | |
948 | ||
949 | @item @emph{Fortran}: | |
950 | @multitable @columnfractions .20 .80 | |
951 | @item @emph{Interface}: @tab @code{integer function omp_get_num_devices()} | |
952 | @end multitable | |
953 | ||
954 | @item @emph{Reference}: | |
955 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.31. | |
956 | @end table | |
957 | ||
958 | ||
959 | ||
960 | @node omp_get_num_procs | |
961 | @section @code{omp_get_num_procs} -- Number of processors online | |
962 | @table @asis | |
963 | @item @emph{Description}: | |
964 | Returns the number of processors online on that device. | |
965 | ||
966 | @item @emph{C/C++}: | |
967 | @multitable @columnfractions .20 .80 | |
968 | @item @emph{Prototype}: @tab @code{int omp_get_num_procs(void);} | |
969 | @end multitable | |
970 | ||
971 | @item @emph{Fortran}: | |
972 | @multitable @columnfractions .20 .80 | |
973 | @item @emph{Interface}: @tab @code{integer function omp_get_num_procs()} | |
974 | @end multitable | |
975 | ||
976 | @item @emph{Reference}: | |
977 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.5. | |
978 | @end table | |
979 | ||
980 | ||
981 | ||
982 | @node omp_get_num_teams | |
983 | @section @code{omp_get_num_teams} -- Number of teams | |
984 | @table @asis | |
985 | @item @emph{Description}: | |
986 | Returns the number of teams in the current team region. | |
987 | ||
988 | @item @emph{C/C++}: | |
989 | @multitable @columnfractions .20 .80 | |
990 | @item @emph{Prototype}: @tab @code{int omp_get_num_teams(void);} | |
991 | @end multitable | |
992 | ||
993 | @item @emph{Fortran}: | |
994 | @multitable @columnfractions .20 .80 | |
995 | @item @emph{Interface}: @tab @code{integer function omp_get_num_teams()} | |
996 | @end multitable | |
997 | ||
998 | @item @emph{Reference}: | |
999 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.32. | |
1000 | @end table | |
1001 | ||
1002 | ||
1003 | ||
1004 | @node omp_get_num_threads | |
1005 | @section @code{omp_get_num_threads} -- Size of the active team | |
1006 | @table @asis | |
1007 | @item @emph{Description}: | |
1008 | Returns the number of threads in the current team. In a sequential section of | |
1009 | the program @code{omp_get_num_threads} returns 1. | |
1010 | ||
1011 | The default team size may be initialized at startup by the | |
1012 | @env{OMP_NUM_THREADS} environment variable. At runtime, the size | |
1013 | of the current team may be set either by the @code{NUM_THREADS} | |
1014 | clause or by @code{omp_set_num_threads}. If none of the above were | |
1015 | used to define a specific value and @env{OMP_DYNAMIC} is disabled, | |
1016 | one thread per CPU online is used. | |
1017 | ||
1018 | @item @emph{C/C++}: | |
1019 | @multitable @columnfractions .20 .80 | |
1020 | @item @emph{Prototype}: @tab @code{int omp_get_num_threads(void);} | |
1021 | @end multitable | |
1022 | ||
1023 | @item @emph{Fortran}: | |
1024 | @multitable @columnfractions .20 .80 | |
1025 | @item @emph{Interface}: @tab @code{integer function omp_get_num_threads()} | |
1026 | @end multitable | |
1027 | ||
1028 | @item @emph{See also}: | |
1029 | @ref{omp_get_max_threads}, @ref{omp_set_num_threads}, @ref{OMP_NUM_THREADS} | |
1030 | ||
1031 | @item @emph{Reference}: | |
1032 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.2. | |
1033 | @end table | |
1034 | ||
1035 | ||
1036 | ||
1037 | @node omp_get_proc_bind | |
0b9bd33d | 1038 | @section @code{omp_get_proc_bind} -- Whether threads may be moved between CPUs |
d77de738 ML |
1039 | @table @asis |
1040 | @item @emph{Description}: | |
1041 | This functions returns the currently active thread affinity policy, which is | |
1042 | set via @env{OMP_PROC_BIND}. Possible values are @code{omp_proc_bind_false}, | |
1043 | @code{omp_proc_bind_true}, @code{omp_proc_bind_primary}, | |
1044 | @code{omp_proc_bind_master}, @code{omp_proc_bind_close} and @code{omp_proc_bind_spread}, | |
1045 | where @code{omp_proc_bind_master} is an alias for @code{omp_proc_bind_primary}. | |
1046 | ||
1047 | @item @emph{C/C++}: | |
1048 | @multitable @columnfractions .20 .80 | |
1049 | @item @emph{Prototype}: @tab @code{omp_proc_bind_t omp_get_proc_bind(void);} | |
1050 | @end multitable | |
1051 | ||
1052 | @item @emph{Fortran}: | |
1053 | @multitable @columnfractions .20 .80 | |
1054 | @item @emph{Interface}: @tab @code{integer(kind=omp_proc_bind_kind) function omp_get_proc_bind()} | |
1055 | @end multitable | |
1056 | ||
1057 | @item @emph{See also}: | |
1058 | @ref{OMP_PROC_BIND}, @ref{OMP_PLACES}, @ref{GOMP_CPU_AFFINITY}, | |
1059 | ||
1060 | @item @emph{Reference}: | |
1061 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.22. | |
1062 | @end table | |
1063 | ||
1064 | ||
1065 | ||
1066 | @node omp_get_schedule | |
1067 | @section @code{omp_get_schedule} -- Obtain the runtime scheduling method | |
1068 | @table @asis | |
1069 | @item @emph{Description}: | |
1070 | Obtain the runtime scheduling method. The @var{kind} argument will be | |
1071 | set to the value @code{omp_sched_static}, @code{omp_sched_dynamic}, | |
1072 | @code{omp_sched_guided} or @code{omp_sched_auto}. The second argument, | |
1073 | @var{chunk_size}, is set to the chunk size. | |
1074 | ||
1075 | @item @emph{C/C++} | |
1076 | @multitable @columnfractions .20 .80 | |
1077 | @item @emph{Prototype}: @tab @code{void omp_get_schedule(omp_sched_t *kind, int *chunk_size);} | |
1078 | @end multitable | |
1079 | ||
1080 | @item @emph{Fortran}: | |
1081 | @multitable @columnfractions .20 .80 | |
1082 | @item @emph{Interface}: @tab @code{subroutine omp_get_schedule(kind, chunk_size)} | |
1083 | @item @tab @code{integer(kind=omp_sched_kind) kind} | |
1084 | @item @tab @code{integer chunk_size} | |
1085 | @end multitable | |
1086 | ||
1087 | @item @emph{See also}: | |
1088 | @ref{omp_set_schedule}, @ref{OMP_SCHEDULE} | |
1089 | ||
1090 | @item @emph{Reference}: | |
1091 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.13. | |
1092 | @end table | |
1093 | ||
1094 | ||
1095 | @node omp_get_supported_active_levels | |
1096 | @section @code{omp_get_supported_active_levels} -- Maximum number of active regions supported | |
1097 | @table @asis | |
1098 | @item @emph{Description}: | |
1099 | This function returns the maximum number of nested, active parallel regions | |
1100 | supported by this implementation. | |
1101 | ||
1102 | @item @emph{C/C++} | |
1103 | @multitable @columnfractions .20 .80 | |
1104 | @item @emph{Prototype}: @tab @code{int omp_get_supported_active_levels(void);} | |
1105 | @end multitable | |
1106 | ||
1107 | @item @emph{Fortran}: | |
1108 | @multitable @columnfractions .20 .80 | |
1109 | @item @emph{Interface}: @tab @code{integer function omp_get_supported_active_levels()} | |
1110 | @end multitable | |
1111 | ||
1112 | @item @emph{See also}: | |
1113 | @ref{omp_get_max_active_levels}, @ref{omp_set_max_active_levels} | |
1114 | ||
1115 | @item @emph{Reference}: | |
1116 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.2.15. | |
1117 | @end table | |
1118 | ||
1119 | ||
1120 | ||
1121 | @node omp_get_team_num | |
1122 | @section @code{omp_get_team_num} -- Get team number | |
1123 | @table @asis | |
1124 | @item @emph{Description}: | |
1125 | Returns the team number of the calling thread. | |
1126 | ||
1127 | @item @emph{C/C++}: | |
1128 | @multitable @columnfractions .20 .80 | |
1129 | @item @emph{Prototype}: @tab @code{int omp_get_team_num(void);} | |
1130 | @end multitable | |
1131 | ||
1132 | @item @emph{Fortran}: | |
1133 | @multitable @columnfractions .20 .80 | |
1134 | @item @emph{Interface}: @tab @code{integer function omp_get_team_num()} | |
1135 | @end multitable | |
1136 | ||
1137 | @item @emph{Reference}: | |
1138 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.33. | |
1139 | @end table | |
1140 | ||
1141 | ||
1142 | ||
1143 | @node omp_get_team_size | |
1144 | @section @code{omp_get_team_size} -- Number of threads in a team | |
1145 | @table @asis | |
1146 | @item @emph{Description}: | |
1147 | This function returns the number of threads in a thread team to which | |
1148 | either the current thread or its ancestor belongs. For values of @var{level} | |
1149 | outside zero to @code{omp_get_level}, -1 is returned; if @var{level} is zero, | |
1150 | 1 is returned, and for @code{omp_get_level}, the result is identical | |
1151 | to @code{omp_get_num_threads}. | |
1152 | ||
1153 | @item @emph{C/C++}: | |
1154 | @multitable @columnfractions .20 .80 | |
1155 | @item @emph{Prototype}: @tab @code{int omp_get_team_size(int level);} | |
1156 | @end multitable | |
1157 | ||
1158 | @item @emph{Fortran}: | |
1159 | @multitable @columnfractions .20 .80 | |
1160 | @item @emph{Interface}: @tab @code{integer function omp_get_team_size(level)} | |
1161 | @item @tab @code{integer level} | |
1162 | @end multitable | |
1163 | ||
1164 | @item @emph{See also}: | |
1165 | @ref{omp_get_num_threads}, @ref{omp_get_level}, @ref{omp_get_ancestor_thread_num} | |
1166 | ||
1167 | @item @emph{Reference}: | |
1168 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.19. | |
1169 | @end table | |
1170 | ||
1171 | ||
1172 | ||
1173 | @node omp_get_teams_thread_limit | |
1174 | @section @code{omp_get_teams_thread_limit} -- Maximum number of threads imposed by teams | |
1175 | @table @asis | |
1176 | @item @emph{Description}: | |
1177 | Return the maximum number of threads that will be able to participate in | |
1178 | each team created by a teams construct. | |
1179 | ||
1180 | @item @emph{C/C++}: | |
1181 | @multitable @columnfractions .20 .80 | |
1182 | @item @emph{Prototype}: @tab @code{int omp_get_teams_thread_limit(void);} | |
1183 | @end multitable | |
1184 | ||
1185 | @item @emph{Fortran}: | |
1186 | @multitable @columnfractions .20 .80 | |
1187 | @item @emph{Interface}: @tab @code{integer function omp_get_teams_thread_limit()} | |
1188 | @end multitable | |
1189 | ||
1190 | @item @emph{See also}: | |
1191 | @ref{omp_set_teams_thread_limit}, @ref{OMP_TEAMS_THREAD_LIMIT} | |
1192 | ||
1193 | @item @emph{Reference}: | |
1194 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.6. | |
1195 | @end table | |
1196 | ||
1197 | ||
1198 | ||
1199 | @node omp_get_thread_limit | |
1200 | @section @code{omp_get_thread_limit} -- Maximum number of threads | |
1201 | @table @asis | |
1202 | @item @emph{Description}: | |
1203 | Return the maximum number of threads of the program. | |
1204 | ||
1205 | @item @emph{C/C++}: | |
1206 | @multitable @columnfractions .20 .80 | |
1207 | @item @emph{Prototype}: @tab @code{int omp_get_thread_limit(void);} | |
1208 | @end multitable | |
1209 | ||
1210 | @item @emph{Fortran}: | |
1211 | @multitable @columnfractions .20 .80 | |
1212 | @item @emph{Interface}: @tab @code{integer function omp_get_thread_limit()} | |
1213 | @end multitable | |
1214 | ||
1215 | @item @emph{See also}: | |
1216 | @ref{omp_get_max_threads}, @ref{OMP_THREAD_LIMIT} | |
1217 | ||
1218 | @item @emph{Reference}: | |
1219 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.14. | |
1220 | @end table | |
1221 | ||
1222 | ||
1223 | ||
1224 | @node omp_get_thread_num | |
1225 | @section @code{omp_get_thread_num} -- Current thread ID | |
1226 | @table @asis | |
1227 | @item @emph{Description}: | |
1228 | Returns a unique thread identification number within the current team. | |
1229 | In a sequential parts of the program, @code{omp_get_thread_num} | |
1230 | always returns 0. In parallel regions the return value varies | |
1231 | from 0 to @code{omp_get_num_threads}-1 inclusive. The return | |
1232 | value of the primary thread of a team is always 0. | |
1233 | ||
1234 | @item @emph{C/C++}: | |
1235 | @multitable @columnfractions .20 .80 | |
1236 | @item @emph{Prototype}: @tab @code{int omp_get_thread_num(void);} | |
1237 | @end multitable | |
1238 | ||
1239 | @item @emph{Fortran}: | |
1240 | @multitable @columnfractions .20 .80 | |
1241 | @item @emph{Interface}: @tab @code{integer function omp_get_thread_num()} | |
1242 | @end multitable | |
1243 | ||
1244 | @item @emph{See also}: | |
1245 | @ref{omp_get_num_threads}, @ref{omp_get_ancestor_thread_num} | |
1246 | ||
1247 | @item @emph{Reference}: | |
1248 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.4. | |
1249 | @end table | |
1250 | ||
1251 | ||
1252 | ||
1253 | @node omp_in_parallel | |
1254 | @section @code{omp_in_parallel} -- Whether a parallel region is active | |
1255 | @table @asis | |
1256 | @item @emph{Description}: | |
1257 | This function returns @code{true} if currently running in parallel, | |
1258 | @code{false} otherwise. Here, @code{true} and @code{false} represent | |
1259 | their language-specific counterparts. | |
1260 | ||
1261 | @item @emph{C/C++}: | |
1262 | @multitable @columnfractions .20 .80 | |
1263 | @item @emph{Prototype}: @tab @code{int omp_in_parallel(void);} | |
1264 | @end multitable | |
1265 | ||
1266 | @item @emph{Fortran}: | |
1267 | @multitable @columnfractions .20 .80 | |
1268 | @item @emph{Interface}: @tab @code{logical function omp_in_parallel()} | |
1269 | @end multitable | |
1270 | ||
1271 | @item @emph{Reference}: | |
1272 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.6. | |
1273 | @end table | |
1274 | ||
1275 | ||
1276 | @node omp_in_final | |
1277 | @section @code{omp_in_final} -- Whether in final or included task region | |
1278 | @table @asis | |
1279 | @item @emph{Description}: | |
1280 | This function returns @code{true} if currently running in a final | |
1281 | or included task region, @code{false} otherwise. Here, @code{true} | |
1282 | and @code{false} represent their language-specific counterparts. | |
1283 | ||
1284 | @item @emph{C/C++}: | |
1285 | @multitable @columnfractions .20 .80 | |
1286 | @item @emph{Prototype}: @tab @code{int omp_in_final(void);} | |
1287 | @end multitable | |
1288 | ||
1289 | @item @emph{Fortran}: | |
1290 | @multitable @columnfractions .20 .80 | |
1291 | @item @emph{Interface}: @tab @code{logical function omp_in_final()} | |
1292 | @end multitable | |
1293 | ||
1294 | @item @emph{Reference}: | |
1295 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.21. | |
1296 | @end table | |
1297 | ||
1298 | ||
1299 | ||
1300 | @node omp_is_initial_device | |
1301 | @section @code{omp_is_initial_device} -- Whether executing on the host device | |
1302 | @table @asis | |
1303 | @item @emph{Description}: | |
1304 | This function returns @code{true} if currently running on the host device, | |
1305 | @code{false} otherwise. Here, @code{true} and @code{false} represent | |
1306 | their language-specific counterparts. | |
1307 | ||
1308 | @item @emph{C/C++}: | |
1309 | @multitable @columnfractions .20 .80 | |
1310 | @item @emph{Prototype}: @tab @code{int omp_is_initial_device(void);} | |
1311 | @end multitable | |
1312 | ||
1313 | @item @emph{Fortran}: | |
1314 | @multitable @columnfractions .20 .80 | |
1315 | @item @emph{Interface}: @tab @code{logical function omp_is_initial_device()} | |
1316 | @end multitable | |
1317 | ||
1318 | @item @emph{Reference}: | |
1319 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.34. | |
1320 | @end table | |
1321 | ||
1322 | ||
1323 | ||
1324 | @node omp_set_default_device | |
1325 | @section @code{omp_set_default_device} -- Set the default device for target regions | |
1326 | @table @asis | |
1327 | @item @emph{Description}: | |
1328 | Set the default device for target regions without device clause. The argument | |
1329 | shall be a nonnegative device number. | |
1330 | ||
1331 | @item @emph{C/C++}: | |
1332 | @multitable @columnfractions .20 .80 | |
1333 | @item @emph{Prototype}: @tab @code{void omp_set_default_device(int device_num);} | |
1334 | @end multitable | |
1335 | ||
1336 | @item @emph{Fortran}: | |
1337 | @multitable @columnfractions .20 .80 | |
1338 | @item @emph{Interface}: @tab @code{subroutine omp_set_default_device(device_num)} | |
1339 | @item @tab @code{integer device_num} | |
1340 | @end multitable | |
1341 | ||
1342 | @item @emph{See also}: | |
1343 | @ref{OMP_DEFAULT_DEVICE}, @ref{omp_get_default_device} | |
1344 | ||
1345 | @item @emph{Reference}: | |
1346 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.29. | |
1347 | @end table | |
1348 | ||
1349 | ||
1350 | ||
1351 | @node omp_set_dynamic | |
1352 | @section @code{omp_set_dynamic} -- Enable/disable dynamic teams | |
1353 | @table @asis | |
1354 | @item @emph{Description}: | |
1355 | Enable or disable the dynamic adjustment of the number of threads | |
1356 | within a team. The function takes the language-specific equivalent | |
1357 | of @code{true} and @code{false}, where @code{true} enables dynamic | |
1358 | adjustment of team sizes and @code{false} disables it. | |
1359 | ||
1360 | @item @emph{C/C++}: | |
1361 | @multitable @columnfractions .20 .80 | |
1362 | @item @emph{Prototype}: @tab @code{void omp_set_dynamic(int dynamic_threads);} | |
1363 | @end multitable | |
1364 | ||
1365 | @item @emph{Fortran}: | |
1366 | @multitable @columnfractions .20 .80 | |
1367 | @item @emph{Interface}: @tab @code{subroutine omp_set_dynamic(dynamic_threads)} | |
1368 | @item @tab @code{logical, intent(in) :: dynamic_threads} | |
1369 | @end multitable | |
1370 | ||
1371 | @item @emph{See also}: | |
1372 | @ref{OMP_DYNAMIC}, @ref{omp_get_dynamic} | |
1373 | ||
1374 | @item @emph{Reference}: | |
1375 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.7. | |
1376 | @end table | |
1377 | ||
1378 | ||
1379 | ||
1380 | @node omp_set_max_active_levels | |
1381 | @section @code{omp_set_max_active_levels} -- Limits the number of active parallel regions | |
1382 | @table @asis | |
1383 | @item @emph{Description}: | |
1384 | This function limits the maximum allowed number of nested, active | |
1385 | parallel regions. @var{max_levels} must be less or equal to | |
1386 | the value returned by @code{omp_get_supported_active_levels}. | |
1387 | ||
1388 | @item @emph{C/C++} | |
1389 | @multitable @columnfractions .20 .80 | |
1390 | @item @emph{Prototype}: @tab @code{void omp_set_max_active_levels(int max_levels);} | |
1391 | @end multitable | |
1392 | ||
1393 | @item @emph{Fortran}: | |
1394 | @multitable @columnfractions .20 .80 | |
1395 | @item @emph{Interface}: @tab @code{subroutine omp_set_max_active_levels(max_levels)} | |
1396 | @item @tab @code{integer max_levels} | |
1397 | @end multitable | |
1398 | ||
1399 | @item @emph{See also}: | |
1400 | @ref{omp_get_max_active_levels}, @ref{omp_get_active_level}, | |
1401 | @ref{omp_get_supported_active_levels} | |
1402 | ||
1403 | @item @emph{Reference}: | |
1404 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.15. | |
1405 | @end table | |
1406 | ||
1407 | ||
1408 | ||
1409 | @node omp_set_nested | |
1410 | @section @code{omp_set_nested} -- Enable/disable nested parallel regions | |
1411 | @table @asis | |
1412 | @item @emph{Description}: | |
1413 | Enable or disable nested parallel regions, i.e., whether team members | |
1414 | are allowed to create new teams. The function takes the language-specific | |
1415 | equivalent of @code{true} and @code{false}, where @code{true} enables | |
1416 | dynamic adjustment of team sizes and @code{false} disables it. | |
1417 | ||
1418 | Enabling nested parallel regions will also set the maximum number of | |
1419 | active nested regions to the maximum supported. Disabling nested parallel | |
1420 | regions will set the maximum number of active nested regions to one. | |
1421 | ||
2cd0689a TB |
1422 | Note that the @code{omp_set_nested} API routine was deprecated |
1423 | in the OpenMP specification 5.2 in favor of @code{omp_set_max_active_levels}. | |
1424 | ||
d77de738 ML |
1425 | @item @emph{C/C++}: |
1426 | @multitable @columnfractions .20 .80 | |
1427 | @item @emph{Prototype}: @tab @code{void omp_set_nested(int nested);} | |
1428 | @end multitable | |
1429 | ||
1430 | @item @emph{Fortran}: | |
1431 | @multitable @columnfractions .20 .80 | |
1432 | @item @emph{Interface}: @tab @code{subroutine omp_set_nested(nested)} | |
1433 | @item @tab @code{logical, intent(in) :: nested} | |
1434 | @end multitable | |
1435 | ||
1436 | @item @emph{See also}: | |
1437 | @ref{omp_get_nested}, @ref{omp_set_max_active_levels}, | |
1438 | @ref{OMP_MAX_ACTIVE_LEVELS}, @ref{OMP_NESTED} | |
1439 | ||
1440 | @item @emph{Reference}: | |
1441 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.10. | |
1442 | @end table | |
1443 | ||
1444 | ||
1445 | ||
1446 | @node omp_set_num_teams | |
1447 | @section @code{omp_set_num_teams} -- Set upper teams limit for teams construct | |
1448 | @table @asis | |
1449 | @item @emph{Description}: | |
1450 | Specifies the upper bound for number of teams created by the teams construct | |
1451 | which does not specify a @code{num_teams} clause. The | |
1452 | argument of @code{omp_set_num_teams} shall be a positive integer. | |
1453 | ||
1454 | @item @emph{C/C++}: | |
1455 | @multitable @columnfractions .20 .80 | |
1456 | @item @emph{Prototype}: @tab @code{void omp_set_num_teams(int num_teams);} | |
1457 | @end multitable | |
1458 | ||
1459 | @item @emph{Fortran}: | |
1460 | @multitable @columnfractions .20 .80 | |
1461 | @item @emph{Interface}: @tab @code{subroutine omp_set_num_teams(num_teams)} | |
1462 | @item @tab @code{integer, intent(in) :: num_teams} | |
1463 | @end multitable | |
1464 | ||
1465 | @item @emph{See also}: | |
1466 | @ref{OMP_NUM_TEAMS}, @ref{omp_get_num_teams}, @ref{omp_get_max_teams} | |
1467 | ||
1468 | @item @emph{Reference}: | |
1469 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.3. | |
1470 | @end table | |
1471 | ||
1472 | ||
1473 | ||
1474 | @node omp_set_num_threads | |
1475 | @section @code{omp_set_num_threads} -- Set upper team size limit | |
1476 | @table @asis | |
1477 | @item @emph{Description}: | |
1478 | Specifies the number of threads used by default in subsequent parallel | |
1479 | sections, if those do not specify a @code{num_threads} clause. The | |
1480 | argument of @code{omp_set_num_threads} shall be a positive integer. | |
1481 | ||
1482 | @item @emph{C/C++}: | |
1483 | @multitable @columnfractions .20 .80 | |
1484 | @item @emph{Prototype}: @tab @code{void omp_set_num_threads(int num_threads);} | |
1485 | @end multitable | |
1486 | ||
1487 | @item @emph{Fortran}: | |
1488 | @multitable @columnfractions .20 .80 | |
1489 | @item @emph{Interface}: @tab @code{subroutine omp_set_num_threads(num_threads)} | |
1490 | @item @tab @code{integer, intent(in) :: num_threads} | |
1491 | @end multitable | |
1492 | ||
1493 | @item @emph{See also}: | |
1494 | @ref{OMP_NUM_THREADS}, @ref{omp_get_num_threads}, @ref{omp_get_max_threads} | |
1495 | ||
1496 | @item @emph{Reference}: | |
1497 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.1. | |
1498 | @end table | |
1499 | ||
1500 | ||
1501 | ||
1502 | @node omp_set_schedule | |
1503 | @section @code{omp_set_schedule} -- Set the runtime scheduling method | |
1504 | @table @asis | |
1505 | @item @emph{Description}: | |
1506 | Sets the runtime scheduling method. The @var{kind} argument can have the | |
1507 | value @code{omp_sched_static}, @code{omp_sched_dynamic}, | |
1508 | @code{omp_sched_guided} or @code{omp_sched_auto}. Except for | |
1509 | @code{omp_sched_auto}, the chunk size is set to the value of | |
1510 | @var{chunk_size} if positive, or to the default value if zero or negative. | |
1511 | For @code{omp_sched_auto} the @var{chunk_size} argument is ignored. | |
1512 | ||
1513 | @item @emph{C/C++} | |
1514 | @multitable @columnfractions .20 .80 | |
1515 | @item @emph{Prototype}: @tab @code{void omp_set_schedule(omp_sched_t kind, int chunk_size);} | |
1516 | @end multitable | |
1517 | ||
1518 | @item @emph{Fortran}: | |
1519 | @multitable @columnfractions .20 .80 | |
1520 | @item @emph{Interface}: @tab @code{subroutine omp_set_schedule(kind, chunk_size)} | |
1521 | @item @tab @code{integer(kind=omp_sched_kind) kind} | |
1522 | @item @tab @code{integer chunk_size} | |
1523 | @end multitable | |
1524 | ||
1525 | @item @emph{See also}: | |
1526 | @ref{omp_get_schedule} | |
1527 | @ref{OMP_SCHEDULE} | |
1528 | ||
1529 | @item @emph{Reference}: | |
1530 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.2.12. | |
1531 | @end table | |
1532 | ||
1533 | ||
1534 | ||
1535 | @node omp_set_teams_thread_limit | |
1536 | @section @code{omp_set_teams_thread_limit} -- Set upper thread limit for teams construct | |
1537 | @table @asis | |
1538 | @item @emph{Description}: | |
1539 | Specifies the upper bound for number of threads that will be available | |
1540 | for each team created by the teams construct which does not specify a | |
1541 | @code{thread_limit} clause. The argument of | |
1542 | @code{omp_set_teams_thread_limit} shall be a positive integer. | |
1543 | ||
1544 | @item @emph{C/C++}: | |
1545 | @multitable @columnfractions .20 .80 | |
1546 | @item @emph{Prototype}: @tab @code{void omp_set_teams_thread_limit(int thread_limit);} | |
1547 | @end multitable | |
1548 | ||
1549 | @item @emph{Fortran}: | |
1550 | @multitable @columnfractions .20 .80 | |
1551 | @item @emph{Interface}: @tab @code{subroutine omp_set_teams_thread_limit(thread_limit)} | |
1552 | @item @tab @code{integer, intent(in) :: thread_limit} | |
1553 | @end multitable | |
1554 | ||
1555 | @item @emph{See also}: | |
1556 | @ref{OMP_TEAMS_THREAD_LIMIT}, @ref{omp_get_teams_thread_limit}, @ref{omp_get_thread_limit} | |
1557 | ||
1558 | @item @emph{Reference}: | |
1559 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 3.4.5. | |
1560 | @end table | |
1561 | ||
1562 | ||
1563 | ||
1564 | @node omp_init_lock | |
1565 | @section @code{omp_init_lock} -- Initialize simple lock | |
1566 | @table @asis | |
1567 | @item @emph{Description}: | |
1568 | Initialize a simple lock. After initialization, the lock is in | |
1569 | an unlocked state. | |
1570 | ||
1571 | @item @emph{C/C++}: | |
1572 | @multitable @columnfractions .20 .80 | |
1573 | @item @emph{Prototype}: @tab @code{void omp_init_lock(omp_lock_t *lock);} | |
1574 | @end multitable | |
1575 | ||
1576 | @item @emph{Fortran}: | |
1577 | @multitable @columnfractions .20 .80 | |
1578 | @item @emph{Interface}: @tab @code{subroutine omp_init_lock(svar)} | |
1579 | @item @tab @code{integer(omp_lock_kind), intent(out) :: svar} | |
1580 | @end multitable | |
1581 | ||
1582 | @item @emph{See also}: | |
1583 | @ref{omp_destroy_lock} | |
1584 | ||
1585 | @item @emph{Reference}: | |
1586 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.1. | |
1587 | @end table | |
1588 | ||
1589 | ||
1590 | ||
1591 | @node omp_set_lock | |
1592 | @section @code{omp_set_lock} -- Wait for and set simple lock | |
1593 | @table @asis | |
1594 | @item @emph{Description}: | |
1595 | Before setting a simple lock, the lock variable must be initialized by | |
1596 | @code{omp_init_lock}. The calling thread is blocked until the lock | |
1597 | is available. If the lock is already held by the current thread, | |
1598 | a deadlock occurs. | |
1599 | ||
1600 | @item @emph{C/C++}: | |
1601 | @multitable @columnfractions .20 .80 | |
1602 | @item @emph{Prototype}: @tab @code{void omp_set_lock(omp_lock_t *lock);} | |
1603 | @end multitable | |
1604 | ||
1605 | @item @emph{Fortran}: | |
1606 | @multitable @columnfractions .20 .80 | |
1607 | @item @emph{Interface}: @tab @code{subroutine omp_set_lock(svar)} | |
1608 | @item @tab @code{integer(omp_lock_kind), intent(inout) :: svar} | |
1609 | @end multitable | |
1610 | ||
1611 | @item @emph{See also}: | |
1612 | @ref{omp_init_lock}, @ref{omp_test_lock}, @ref{omp_unset_lock} | |
1613 | ||
1614 | @item @emph{Reference}: | |
1615 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.4. | |
1616 | @end table | |
1617 | ||
1618 | ||
1619 | ||
1620 | @node omp_test_lock | |
1621 | @section @code{omp_test_lock} -- Test and set simple lock if available | |
1622 | @table @asis | |
1623 | @item @emph{Description}: | |
1624 | Before setting a simple lock, the lock variable must be initialized by | |
1625 | @code{omp_init_lock}. Contrary to @code{omp_set_lock}, @code{omp_test_lock} | |
1626 | does not block if the lock is not available. This function returns | |
1627 | @code{true} upon success, @code{false} otherwise. Here, @code{true} and | |
1628 | @code{false} represent their language-specific counterparts. | |
1629 | ||
1630 | @item @emph{C/C++}: | |
1631 | @multitable @columnfractions .20 .80 | |
1632 | @item @emph{Prototype}: @tab @code{int omp_test_lock(omp_lock_t *lock);} | |
1633 | @end multitable | |
1634 | ||
1635 | @item @emph{Fortran}: | |
1636 | @multitable @columnfractions .20 .80 | |
1637 | @item @emph{Interface}: @tab @code{logical function omp_test_lock(svar)} | |
1638 | @item @tab @code{integer(omp_lock_kind), intent(inout) :: svar} | |
1639 | @end multitable | |
1640 | ||
1641 | @item @emph{See also}: | |
1642 | @ref{omp_init_lock}, @ref{omp_set_lock}, @ref{omp_set_lock} | |
1643 | ||
1644 | @item @emph{Reference}: | |
1645 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.6. | |
1646 | @end table | |
1647 | ||
1648 | ||
1649 | ||
1650 | @node omp_unset_lock | |
1651 | @section @code{omp_unset_lock} -- Unset simple lock | |
1652 | @table @asis | |
1653 | @item @emph{Description}: | |
1654 | A simple lock about to be unset must have been locked by @code{omp_set_lock} | |
1655 | or @code{omp_test_lock} before. In addition, the lock must be held by the | |
1656 | thread calling @code{omp_unset_lock}. Then, the lock becomes unlocked. If one | |
1657 | or more threads attempted to set the lock before, one of them is chosen to, | |
1658 | again, set the lock to itself. | |
1659 | ||
1660 | @item @emph{C/C++}: | |
1661 | @multitable @columnfractions .20 .80 | |
1662 | @item @emph{Prototype}: @tab @code{void omp_unset_lock(omp_lock_t *lock);} | |
1663 | @end multitable | |
1664 | ||
1665 | @item @emph{Fortran}: | |
1666 | @multitable @columnfractions .20 .80 | |
1667 | @item @emph{Interface}: @tab @code{subroutine omp_unset_lock(svar)} | |
1668 | @item @tab @code{integer(omp_lock_kind), intent(inout) :: svar} | |
1669 | @end multitable | |
1670 | ||
1671 | @item @emph{See also}: | |
1672 | @ref{omp_set_lock}, @ref{omp_test_lock} | |
1673 | ||
1674 | @item @emph{Reference}: | |
1675 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.5. | |
1676 | @end table | |
1677 | ||
1678 | ||
1679 | ||
1680 | @node omp_destroy_lock | |
1681 | @section @code{omp_destroy_lock} -- Destroy simple lock | |
1682 | @table @asis | |
1683 | @item @emph{Description}: | |
1684 | Destroy a simple lock. In order to be destroyed, a simple lock must be | |
1685 | in the unlocked state. | |
1686 | ||
1687 | @item @emph{C/C++}: | |
1688 | @multitable @columnfractions .20 .80 | |
1689 | @item @emph{Prototype}: @tab @code{void omp_destroy_lock(omp_lock_t *lock);} | |
1690 | @end multitable | |
1691 | ||
1692 | @item @emph{Fortran}: | |
1693 | @multitable @columnfractions .20 .80 | |
1694 | @item @emph{Interface}: @tab @code{subroutine omp_destroy_lock(svar)} | |
1695 | @item @tab @code{integer(omp_lock_kind), intent(inout) :: svar} | |
1696 | @end multitable | |
1697 | ||
1698 | @item @emph{See also}: | |
1699 | @ref{omp_init_lock} | |
1700 | ||
1701 | @item @emph{Reference}: | |
1702 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.3. | |
1703 | @end table | |
1704 | ||
1705 | ||
1706 | ||
1707 | @node omp_init_nest_lock | |
1708 | @section @code{omp_init_nest_lock} -- Initialize nested lock | |
1709 | @table @asis | |
1710 | @item @emph{Description}: | |
1711 | Initialize a nested lock. After initialization, the lock is in | |
1712 | an unlocked state and the nesting count is set to zero. | |
1713 | ||
1714 | @item @emph{C/C++}: | |
1715 | @multitable @columnfractions .20 .80 | |
1716 | @item @emph{Prototype}: @tab @code{void omp_init_nest_lock(omp_nest_lock_t *lock);} | |
1717 | @end multitable | |
1718 | ||
1719 | @item @emph{Fortran}: | |
1720 | @multitable @columnfractions .20 .80 | |
1721 | @item @emph{Interface}: @tab @code{subroutine omp_init_nest_lock(nvar)} | |
1722 | @item @tab @code{integer(omp_nest_lock_kind), intent(out) :: nvar} | |
1723 | @end multitable | |
1724 | ||
1725 | @item @emph{See also}: | |
1726 | @ref{omp_destroy_nest_lock} | |
1727 | ||
1728 | @item @emph{Reference}: | |
1729 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.1. | |
1730 | @end table | |
1731 | ||
1732 | ||
1733 | @node omp_set_nest_lock | |
1734 | @section @code{omp_set_nest_lock} -- Wait for and set nested lock | |
1735 | @table @asis | |
1736 | @item @emph{Description}: | |
1737 | Before setting a nested lock, the lock variable must be initialized by | |
1738 | @code{omp_init_nest_lock}. The calling thread is blocked until the lock | |
1739 | is available. If the lock is already held by the current thread, the | |
1740 | nesting count for the lock is incremented. | |
1741 | ||
1742 | @item @emph{C/C++}: | |
1743 | @multitable @columnfractions .20 .80 | |
1744 | @item @emph{Prototype}: @tab @code{void omp_set_nest_lock(omp_nest_lock_t *lock);} | |
1745 | @end multitable | |
1746 | ||
1747 | @item @emph{Fortran}: | |
1748 | @multitable @columnfractions .20 .80 | |
1749 | @item @emph{Interface}: @tab @code{subroutine omp_set_nest_lock(nvar)} | |
1750 | @item @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar} | |
1751 | @end multitable | |
1752 | ||
1753 | @item @emph{See also}: | |
1754 | @ref{omp_init_nest_lock}, @ref{omp_unset_nest_lock} | |
1755 | ||
1756 | @item @emph{Reference}: | |
1757 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.4. | |
1758 | @end table | |
1759 | ||
1760 | ||
1761 | ||
1762 | @node omp_test_nest_lock | |
1763 | @section @code{omp_test_nest_lock} -- Test and set nested lock if available | |
1764 | @table @asis | |
1765 | @item @emph{Description}: | |
1766 | Before setting a nested lock, the lock variable must be initialized by | |
1767 | @code{omp_init_nest_lock}. Contrary to @code{omp_set_nest_lock}, | |
1768 | @code{omp_test_nest_lock} does not block if the lock is not available. | |
1769 | If the lock is already held by the current thread, the new nesting count | |
1770 | is returned. Otherwise, the return value equals zero. | |
1771 | ||
1772 | @item @emph{C/C++}: | |
1773 | @multitable @columnfractions .20 .80 | |
1774 | @item @emph{Prototype}: @tab @code{int omp_test_nest_lock(omp_nest_lock_t *lock);} | |
1775 | @end multitable | |
1776 | ||
1777 | @item @emph{Fortran}: | |
1778 | @multitable @columnfractions .20 .80 | |
1779 | @item @emph{Interface}: @tab @code{logical function omp_test_nest_lock(nvar)} | |
1780 | @item @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar} | |
1781 | @end multitable | |
1782 | ||
1783 | ||
1784 | @item @emph{See also}: | |
1785 | @ref{omp_init_lock}, @ref{omp_set_lock}, @ref{omp_set_lock} | |
1786 | ||
1787 | @item @emph{Reference}: | |
1788 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.6. | |
1789 | @end table | |
1790 | ||
1791 | ||
1792 | ||
1793 | @node omp_unset_nest_lock | |
1794 | @section @code{omp_unset_nest_lock} -- Unset nested lock | |
1795 | @table @asis | |
1796 | @item @emph{Description}: | |
1797 | A nested lock about to be unset must have been locked by @code{omp_set_nested_lock} | |
1798 | or @code{omp_test_nested_lock} before. In addition, the lock must be held by the | |
1799 | thread calling @code{omp_unset_nested_lock}. If the nesting count drops to zero, the | |
1800 | lock becomes unlocked. If one ore more threads attempted to set the lock before, | |
1801 | one of them is chosen to, again, set the lock to itself. | |
1802 | ||
1803 | @item @emph{C/C++}: | |
1804 | @multitable @columnfractions .20 .80 | |
1805 | @item @emph{Prototype}: @tab @code{void omp_unset_nest_lock(omp_nest_lock_t *lock);} | |
1806 | @end multitable | |
1807 | ||
1808 | @item @emph{Fortran}: | |
1809 | @multitable @columnfractions .20 .80 | |
1810 | @item @emph{Interface}: @tab @code{subroutine omp_unset_nest_lock(nvar)} | |
1811 | @item @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar} | |
1812 | @end multitable | |
1813 | ||
1814 | @item @emph{See also}: | |
1815 | @ref{omp_set_nest_lock} | |
1816 | ||
1817 | @item @emph{Reference}: | |
1818 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.5. | |
1819 | @end table | |
1820 | ||
1821 | ||
1822 | ||
1823 | @node omp_destroy_nest_lock | |
1824 | @section @code{omp_destroy_nest_lock} -- Destroy nested lock | |
1825 | @table @asis | |
1826 | @item @emph{Description}: | |
1827 | Destroy a nested lock. In order to be destroyed, a nested lock must be | |
1828 | in the unlocked state and its nesting count must equal zero. | |
1829 | ||
1830 | @item @emph{C/C++}: | |
1831 | @multitable @columnfractions .20 .80 | |
1832 | @item @emph{Prototype}: @tab @code{void omp_destroy_nest_lock(omp_nest_lock_t *);} | |
1833 | @end multitable | |
1834 | ||
1835 | @item @emph{Fortran}: | |
1836 | @multitable @columnfractions .20 .80 | |
1837 | @item @emph{Interface}: @tab @code{subroutine omp_destroy_nest_lock(nvar)} | |
1838 | @item @tab @code{integer(omp_nest_lock_kind), intent(inout) :: nvar} | |
1839 | @end multitable | |
1840 | ||
1841 | @item @emph{See also}: | |
1842 | @ref{omp_init_lock} | |
1843 | ||
1844 | @item @emph{Reference}: | |
1845 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.3.3. | |
1846 | @end table | |
1847 | ||
1848 | ||
1849 | ||
1850 | @node omp_get_wtick | |
1851 | @section @code{omp_get_wtick} -- Get timer precision | |
1852 | @table @asis | |
1853 | @item @emph{Description}: | |
1854 | Gets the timer precision, i.e., the number of seconds between two | |
1855 | successive clock ticks. | |
1856 | ||
1857 | @item @emph{C/C++}: | |
1858 | @multitable @columnfractions .20 .80 | |
1859 | @item @emph{Prototype}: @tab @code{double omp_get_wtick(void);} | |
1860 | @end multitable | |
1861 | ||
1862 | @item @emph{Fortran}: | |
1863 | @multitable @columnfractions .20 .80 | |
1864 | @item @emph{Interface}: @tab @code{double precision function omp_get_wtick()} | |
1865 | @end multitable | |
1866 | ||
1867 | @item @emph{See also}: | |
1868 | @ref{omp_get_wtime} | |
1869 | ||
1870 | @item @emph{Reference}: | |
1871 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.4.2. | |
1872 | @end table | |
1873 | ||
1874 | ||
1875 | ||
1876 | @node omp_get_wtime | |
1877 | @section @code{omp_get_wtime} -- Elapsed wall clock time | |
1878 | @table @asis | |
1879 | @item @emph{Description}: | |
1880 | Elapsed wall clock time in seconds. The time is measured per thread, no | |
1881 | guarantee can be made that two distinct threads measure the same time. | |
1882 | Time is measured from some "time in the past", which is an arbitrary time | |
1883 | guaranteed not to change during the execution of the program. | |
1884 | ||
1885 | @item @emph{C/C++}: | |
1886 | @multitable @columnfractions .20 .80 | |
1887 | @item @emph{Prototype}: @tab @code{double omp_get_wtime(void);} | |
1888 | @end multitable | |
1889 | ||
1890 | @item @emph{Fortran}: | |
1891 | @multitable @columnfractions .20 .80 | |
1892 | @item @emph{Interface}: @tab @code{double precision function omp_get_wtime()} | |
1893 | @end multitable | |
1894 | ||
1895 | @item @emph{See also}: | |
1896 | @ref{omp_get_wtick} | |
1897 | ||
1898 | @item @emph{Reference}: | |
1899 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 3.4.1. | |
1900 | @end table | |
1901 | ||
1902 | ||
1903 | ||
1904 | @node omp_fulfill_event | |
1905 | @section @code{omp_fulfill_event} -- Fulfill and destroy an OpenMP event | |
1906 | @table @asis | |
1907 | @item @emph{Description}: | |
1908 | Fulfill the event associated with the event handle argument. Currently, it | |
1909 | is only used to fulfill events generated by detach clauses on task | |
1910 | constructs - the effect of fulfilling the event is to allow the task to | |
1911 | complete. | |
1912 | ||
1913 | The result of calling @code{omp_fulfill_event} with an event handle other | |
1914 | than that generated by a detach clause is undefined. Calling it with an | |
1915 | event handle that has already been fulfilled is also undefined. | |
1916 | ||
1917 | @item @emph{C/C++}: | |
1918 | @multitable @columnfractions .20 .80 | |
1919 | @item @emph{Prototype}: @tab @code{void omp_fulfill_event(omp_event_handle_t event);} | |
1920 | @end multitable | |
1921 | ||
1922 | @item @emph{Fortran}: | |
1923 | @multitable @columnfractions .20 .80 | |
1924 | @item @emph{Interface}: @tab @code{subroutine omp_fulfill_event(event)} | |
1925 | @item @tab @code{integer (kind=omp_event_handle_kind) :: event} | |
1926 | @end multitable | |
1927 | ||
1928 | @item @emph{Reference}: | |
1929 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 3.5.1. | |
1930 | @end table | |
1931 | ||
1932 | ||
1933 | ||
1934 | @c --------------------------------------------------------------------- | |
1935 | @c OpenMP Environment Variables | |
1936 | @c --------------------------------------------------------------------- | |
1937 | ||
1938 | @node Environment Variables | |
1939 | @chapter OpenMP Environment Variables | |
1940 | ||
1941 | The environment variables which beginning with @env{OMP_} are defined by | |
2cd0689a TB |
1942 | section 4 of the OpenMP specification in version 4.5 or in a later version |
1943 | of the specification, while those beginning with @env{GOMP_} are GNU extensions. | |
1944 | Most @env{OMP_} environment variables have an associated internal control | |
1945 | variable (ICV). | |
1946 | ||
1947 | For any OpenMP environment variable that sets an ICV and is neither | |
1948 | @code{OMP_DEFAULT_DEVICE} nor has global ICV scope, associated | |
1949 | device-specific environment variables exist. For them, the environment | |
1950 | variable without suffix affects the host. The suffix @code{_DEV_} followed | |
1951 | by a non-negative device number less that the number of available devices sets | |
1952 | the ICV for the corresponding device. The suffix @code{_DEV} sets the ICV | |
1953 | of all non-host devices for which a device-specific corresponding environment | |
1954 | variable has not been set while the @code{_ALL} suffix sets the ICV of all | |
1955 | host and non-host devices for which a more specific corresponding environment | |
1956 | variable is not set. | |
d77de738 ML |
1957 | |
1958 | @menu | |
73a0d3bf TB |
1959 | * OMP_ALLOCATOR:: Set the default allocator |
1960 | * OMP_AFFINITY_FORMAT:: Set the format string used for affinity display | |
d77de738 | 1961 | * OMP_CANCELLATION:: Set whether cancellation is activated |
73a0d3bf | 1962 | * OMP_DISPLAY_AFFINITY:: Display thread affinity information |
d77de738 ML |
1963 | * OMP_DISPLAY_ENV:: Show OpenMP version and environment variables |
1964 | * OMP_DEFAULT_DEVICE:: Set the device used in target regions | |
1965 | * OMP_DYNAMIC:: Dynamic adjustment of threads | |
1966 | * OMP_MAX_ACTIVE_LEVELS:: Set the maximum number of nested parallel regions | |
1967 | * OMP_MAX_TASK_PRIORITY:: Set the maximum task priority value | |
1968 | * OMP_NESTED:: Nested parallel regions | |
1969 | * OMP_NUM_TEAMS:: Specifies the number of teams to use by teams region | |
1970 | * OMP_NUM_THREADS:: Specifies the number of threads to use | |
0b9bd33d JJ |
1971 | * OMP_PROC_BIND:: Whether threads may be moved between CPUs |
1972 | * OMP_PLACES:: Specifies on which CPUs the threads should be placed | |
d77de738 ML |
1973 | * OMP_STACKSIZE:: Set default thread stack size |
1974 | * OMP_SCHEDULE:: How threads are scheduled | |
1975 | * OMP_TARGET_OFFLOAD:: Controls offloading behaviour | |
1976 | * OMP_TEAMS_THREAD_LIMIT:: Set the maximum number of threads imposed by teams | |
1977 | * OMP_THREAD_LIMIT:: Set the maximum number of threads | |
1978 | * OMP_WAIT_POLICY:: How waiting threads are handled | |
1979 | * GOMP_CPU_AFFINITY:: Bind threads to specific CPUs | |
1980 | * GOMP_DEBUG:: Enable debugging output | |
1981 | * GOMP_STACKSIZE:: Set default thread stack size | |
1982 | * GOMP_SPINCOUNT:: Set the busy-wait spin count | |
1983 | * GOMP_RTEMS_THREAD_POOLS:: Set the RTEMS specific thread pools | |
1984 | @end menu | |
1985 | ||
1986 | ||
73a0d3bf TB |
1987 | @node OMP_ALLOCATOR |
1988 | @section @env{OMP_ALLOCATOR} -- Set the default allocator | |
1989 | @cindex Environment Variable | |
1990 | @table @asis | |
2cd0689a TB |
1991 | @item @emph{ICV:} @var{available-devices-var} |
1992 | @item @emph{Scope:} data environment | |
73a0d3bf TB |
1993 | @item @emph{Description}: |
1994 | Sets the default allocator that is used when no allocator has been specified | |
1995 | in the @code{allocate} or @code{allocator} clause or if an OpenMP memory | |
1996 | routine is invoked with the @code{omp_null_allocator} allocator. | |
1997 | If unset, @code{omp_default_mem_alloc} is used. | |
1998 | ||
1999 | The value can either be a predefined allocator or a predefined memory space | |
2000 | or a predefined memory space followed by a colon and a comma-separated list | |
2001 | of memory trait and value pairs, separated by @code{=}. | |
2002 | ||
2cd0689a TB |
2003 | Note: The corresponding device environment variables are currently not |
2004 | supported. Therefore, the non-host @var{def-allocator-var} ICVs are always | |
2005 | initialized to @code{omp_default_mem_alloc}. However, on all devices, | |
2006 | the @code{omp_set_default_allocator} API routine can be used to change | |
2007 | value. | |
2008 | ||
73a0d3bf | 2009 | @multitable @columnfractions .45 .45 |
a85a106c | 2010 | @headitem Predefined allocators @tab Associated predefined memory spaces |
73a0d3bf TB |
2011 | @item omp_default_mem_alloc @tab omp_default_mem_space |
2012 | @item omp_large_cap_mem_alloc @tab omp_large_cap_mem_space | |
2013 | @item omp_const_mem_alloc @tab omp_const_mem_space | |
2014 | @item omp_high_bw_mem_alloc @tab omp_high_bw_mem_space | |
2015 | @item omp_low_lat_mem_alloc @tab omp_low_lat_mem_space | |
2016 | @item omp_cgroup_mem_alloc @tab -- | |
2017 | @item omp_pteam_mem_alloc @tab -- | |
2018 | @item omp_thread_mem_alloc @tab -- | |
2019 | @end multitable | |
2020 | ||
a85a106c TB |
2021 | The predefined allocators use the default values for the traits, |
2022 | as listed below. Except that the last three allocators have the | |
2023 | @code{access} trait set to @code{cgroup}, @code{pteam}, and | |
2024 | @code{thread}, respectively. | |
2025 | ||
2026 | @multitable @columnfractions .25 .40 .25 | |
2027 | @headitem Trait @tab Allowed values @tab Default value | |
73a0d3bf TB |
2028 | @item @code{sync_hint} @tab @code{contended}, @code{uncontended}, |
2029 | @code{serialized}, @code{private} | |
a85a106c | 2030 | @tab @code{contended} |
73a0d3bf | 2031 | @item @code{alignment} @tab Positive integer being a power of two |
a85a106c | 2032 | @tab 1 byte |
73a0d3bf TB |
2033 | @item @code{access} @tab @code{all}, @code{cgroup}, |
2034 | @code{pteam}, @code{thread} | |
a85a106c | 2035 | @tab @code{all} |
73a0d3bf | 2036 | @item @code{pool_size} @tab Positive integer |
a85a106c | 2037 | @tab See @ref{Memory allocation} |
73a0d3bf TB |
2038 | @item @code{fallback} @tab @code{default_mem_fb}, @code{null_fb}, |
2039 | @code{abort_fb}, @code{allocator_fb} | |
a85a106c | 2040 | @tab See below |
73a0d3bf | 2041 | @item @code{fb_data} @tab @emph{unsupported as it needs an allocator handle} |
a85a106c | 2042 | @tab (none) |
73a0d3bf | 2043 | @item @code{pinned} @tab @code{true}, @code{false} |
a85a106c | 2044 | @tab @code{false} |
73a0d3bf TB |
2045 | @item @code{partition} @tab @code{environment}, @code{nearest}, |
2046 | @code{blocked}, @code{interleaved} | |
a85a106c | 2047 | @tab @code{environment} |
73a0d3bf TB |
2048 | @end multitable |
2049 | ||
a85a106c TB |
2050 | For the @code{fallback} trait, the default value is @code{null_fb} for the |
2051 | @code{omp_default_mem_alloc} allocator and any allocator that is associated | |
2052 | with device memory; for all other other allocators, it is @code{default_mem_fb} | |
2053 | by default. | |
2054 | ||
73a0d3bf TB |
2055 | Examples: |
2056 | @smallexample | |
2057 | OMP_ALLOCATOR=omp_high_bw_mem_alloc | |
2058 | OMP_ALLOCATOR=omp_large_cap_mem_space | |
2059 | OMP_ALLOCATR=omp_low_lat_mem_space:pinned=true,partition=nearest | |
2060 | @end smallexample | |
2061 | ||
a85a106c TB |
2062 | @item @emph{See also}: |
2063 | @ref{Memory allocation} | |
73a0d3bf TB |
2064 | |
2065 | @item @emph{Reference}: | |
2066 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.21 | |
2067 | @end table | |
2068 | ||
2069 | ||
2070 | ||
2071 | @node OMP_AFFINITY_FORMAT | |
2072 | @section @env{OMP_AFFINITY_FORMAT} -- Set the format string used for affinity display | |
2073 | @cindex Environment Variable | |
2074 | @table @asis | |
2cd0689a TB |
2075 | @item @emph{ICV:} @var{affinity-format-var} |
2076 | @item @emph{Scope:} device | |
73a0d3bf TB |
2077 | @item @emph{Description}: |
2078 | Sets the format string used when displaying OpenMP thread affinity information. | |
2079 | Special values are output using @code{%} followed by an optional size | |
2080 | specification and then either the single-character field type or its long | |
2081 | name enclosed in curly braces; using @code{%%} will display a literal percent. | |
2082 | The size specification consists of an optional @code{0.} or @code{.} followed | |
450b05ce | 2083 | by a positive integer, specifying the minimal width of the output. With |
73a0d3bf TB |
2084 | @code{0.} and numerical values, the output is padded with zeros on the left; |
2085 | with @code{.}, the output is padded by spaces on the left; otherwise, the | |
2086 | output is padded by spaces on the right. If unset, the value is | |
2087 | ``@code{level %L thread %i affinity %A}''. | |
2088 | ||
2089 | Supported field types are: | |
2090 | ||
2091 | @multitable @columnfractions .10 .25 .60 | |
2092 | @item t @tab team_num @tab value returned by @code{omp_get_team_num} | |
2093 | @item T @tab num_teams @tab value returned by @code{omp_get_num_teams} | |
2094 | @item L @tab nesting_level @tab value returned by @code{omp_get_level} | |
2095 | @item n @tab thread_num @tab value returned by @code{omp_get_thread_num} | |
2096 | @item N @tab num_threads @tab value returned by @code{omp_get_num_threads} | |
2097 | @item a @tab ancestor_tnum | |
2098 | @tab value returned by | |
2099 | @code{omp_get_ancestor_thread_num(omp_get_level()-1)} | |
2100 | @item H @tab host @tab name of the host that executes the thread | |
450b05ce TB |
2101 | @item P @tab process_id @tab process identifier |
2102 | @item i @tab native_thread_id @tab native thread identifier | |
73a0d3bf TB |
2103 | @item A @tab thread_affinity |
2104 | @tab comma separated list of integer values or ranges, representing the | |
2105 | processors on which a process might execute, subject to affinity | |
2106 | mechanisms | |
2107 | @end multitable | |
2108 | ||
2109 | For instance, after setting | |
2110 | ||
2111 | @smallexample | |
2112 | OMP_AFFINITY_FORMAT="%0.2a!%n!%.4L!%N;%.2t;%0.2T;%@{team_num@};%@{num_teams@};%A" | |
2113 | @end smallexample | |
2114 | ||
2115 | with either @code{OMP_DISPLAY_AFFINITY} being set or when calling | |
2116 | @code{omp_display_affinity} with @code{NULL} or an empty string, the program | |
2117 | might display the following: | |
2118 | ||
2119 | @smallexample | |
2120 | 00!0! 1!4; 0;01;0;1;0-11 | |
2121 | 00!3! 1!4; 0;01;0;1;0-11 | |
2122 | 00!2! 1!4; 0;01;0;1;0-11 | |
2123 | 00!1! 1!4; 0;01;0;1;0-11 | |
2124 | @end smallexample | |
2125 | ||
2126 | @item @emph{See also}: | |
2127 | @ref{OMP_DISPLAY_AFFINITY} | |
2128 | ||
2129 | @item @emph{Reference}: | |
2130 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.14 | |
2131 | @end table | |
2132 | ||
2133 | ||
2134 | ||
d77de738 ML |
2135 | @node OMP_CANCELLATION |
2136 | @section @env{OMP_CANCELLATION} -- Set whether cancellation is activated | |
2137 | @cindex Environment Variable | |
2138 | @table @asis | |
2cd0689a TB |
2139 | @item @emph{ICV:} @var{cancel-var} |
2140 | @item @emph{Scope:} global | |
d77de738 ML |
2141 | @item @emph{Description}: |
2142 | If set to @code{TRUE}, the cancellation is activated. If set to @code{FALSE} or | |
2143 | if unset, cancellation is disabled and the @code{cancel} construct is ignored. | |
2144 | ||
2145 | @item @emph{See also}: | |
2146 | @ref{omp_get_cancellation} | |
2147 | ||
2148 | @item @emph{Reference}: | |
2149 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.11 | |
2150 | @end table | |
2151 | ||
2152 | ||
2153 | ||
73a0d3bf TB |
2154 | @node OMP_DISPLAY_AFFINITY |
2155 | @section @env{OMP_DISPLAY_AFFINITY} -- Display thread affinity information | |
2156 | @cindex Environment Variable | |
2157 | @table @asis | |
2cd0689a TB |
2158 | @item @emph{ICV:} @var{display-affinity-var} |
2159 | @item @emph{Scope:} global | |
73a0d3bf TB |
2160 | @item @emph{Description}: |
2161 | If set to @code{FALSE} or if unset, affinity displaying is disabled. | |
2162 | If set to @code{TRUE}, the runtime will display affinity information about | |
2163 | OpenMP threads in a parallel region upon entering the region and every time | |
2164 | any change occurs. | |
2165 | ||
2166 | @item @emph{See also}: | |
2167 | @ref{OMP_AFFINITY_FORMAT} | |
2168 | ||
2169 | @item @emph{Reference}: | |
2170 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.13 | |
2171 | @end table | |
2172 | ||
2173 | ||
2174 | ||
2175 | ||
d77de738 ML |
2176 | @node OMP_DISPLAY_ENV |
2177 | @section @env{OMP_DISPLAY_ENV} -- Show OpenMP version and environment variables | |
2178 | @cindex Environment Variable | |
2179 | @table @asis | |
2cd0689a TB |
2180 | @item @emph{ICV:} none |
2181 | @item @emph{Scope:} not applicable | |
d77de738 ML |
2182 | @item @emph{Description}: |
2183 | If set to @code{TRUE}, the OpenMP version number and the values | |
2184 | associated with the OpenMP environment variables are printed to @code{stderr}. | |
2185 | If set to @code{VERBOSE}, it additionally shows the value of the environment | |
2186 | variables which are GNU extensions. If undefined or set to @code{FALSE}, | |
2187 | this information will not be shown. | |
2188 | ||
2189 | ||
2190 | @item @emph{Reference}: | |
2191 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.12 | |
2192 | @end table | |
2193 | ||
2194 | ||
2195 | ||
2196 | @node OMP_DEFAULT_DEVICE | |
2197 | @section @env{OMP_DEFAULT_DEVICE} -- Set the device used in target regions | |
2198 | @cindex Environment Variable | |
2199 | @table @asis | |
2cd0689a TB |
2200 | @item @emph{ICV:} @var{default-device-var} |
2201 | @item @emph{Scope:} data environment | |
d77de738 ML |
2202 | @item @emph{Description}: |
2203 | Set to choose the device which is used in a @code{target} region, unless the | |
2204 | value is overridden by @code{omp_set_default_device} or by a @code{device} | |
2205 | clause. The value shall be the nonnegative device number. If no device with | |
2206 | the given device number exists, the code is executed on the host. If unset, | |
18c8b56c TB |
2207 | @env{OMP_TARGET_OFFLOAD} is @code{mandatory} and no non-host devices are |
2208 | available, it is set to @code{omp_invalid_device}. Otherwise, if unset, | |
d77de738 ML |
2209 | device number 0 will be used. |
2210 | ||
2211 | ||
2212 | @item @emph{See also}: | |
2213 | @ref{omp_get_default_device}, @ref{omp_set_default_device}, | |
2214 | ||
2215 | @item @emph{Reference}: | |
2216 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.13 | |
2217 | @end table | |
2218 | ||
2219 | ||
2220 | ||
2221 | @node OMP_DYNAMIC | |
2222 | @section @env{OMP_DYNAMIC} -- Dynamic adjustment of threads | |
2223 | @cindex Environment Variable | |
2224 | @table @asis | |
2cd0689a TB |
2225 | @item @emph{ICV:} @var{dyn-var} |
2226 | @item @emph{Scope:} global | |
d77de738 ML |
2227 | @item @emph{Description}: |
2228 | Enable or disable the dynamic adjustment of the number of threads | |
2229 | within a team. The value of this environment variable shall be | |
2230 | @code{TRUE} or @code{FALSE}. If undefined, dynamic adjustment is | |
2231 | disabled by default. | |
2232 | ||
2233 | @item @emph{See also}: | |
2234 | @ref{omp_set_dynamic} | |
2235 | ||
2236 | @item @emph{Reference}: | |
2237 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.3 | |
2238 | @end table | |
2239 | ||
2240 | ||
2241 | ||
2242 | @node OMP_MAX_ACTIVE_LEVELS | |
2243 | @section @env{OMP_MAX_ACTIVE_LEVELS} -- Set the maximum number of nested parallel regions | |
2244 | @cindex Environment Variable | |
2245 | @table @asis | |
2cd0689a TB |
2246 | @item @emph{ICV:} @var{max-active-levels-var} |
2247 | @item @emph{Scope:} data environment | |
d77de738 ML |
2248 | @item @emph{Description}: |
2249 | Specifies the initial value for the maximum number of nested parallel | |
2250 | regions. The value of this variable shall be a positive integer. | |
2251 | If undefined, then if @env{OMP_NESTED} is defined and set to true, or | |
2252 | if @env{OMP_NUM_THREADS} or @env{OMP_PROC_BIND} are defined and set to | |
2253 | a list with more than one item, the maximum number of nested parallel | |
2254 | regions will be initialized to the largest number supported, otherwise | |
2255 | it will be set to one. | |
2256 | ||
2257 | @item @emph{See also}: | |
2cd0689a TB |
2258 | @ref{omp_set_max_active_levels}, @ref{OMP_NESTED}, @ref{OMP_PROC_BIND}, |
2259 | @ref{OMP_NUM_THREADS} | |
2260 | ||
d77de738 ML |
2261 | |
2262 | @item @emph{Reference}: | |
2263 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.9 | |
2264 | @end table | |
2265 | ||
2266 | ||
2267 | ||
2268 | @node OMP_MAX_TASK_PRIORITY | |
2269 | @section @env{OMP_MAX_TASK_PRIORITY} -- Set the maximum priority | |
2270 | number that can be set for a task. | |
2271 | @cindex Environment Variable | |
2272 | @table @asis | |
2cd0689a TB |
2273 | @item @emph{ICV:} @var{max-task-priority-var} |
2274 | @item @emph{Scope:} global | |
d77de738 ML |
2275 | @item @emph{Description}: |
2276 | Specifies the initial value for the maximum priority value that can be | |
2277 | set for a task. The value of this variable shall be a non-negative | |
2278 | integer, and zero is allowed. If undefined, the default priority is | |
2279 | 0. | |
2280 | ||
2281 | @item @emph{See also}: | |
2282 | @ref{omp_get_max_task_priority} | |
2283 | ||
2284 | @item @emph{Reference}: | |
2285 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.14 | |
2286 | @end table | |
2287 | ||
2288 | ||
2289 | ||
2290 | @node OMP_NESTED | |
2291 | @section @env{OMP_NESTED} -- Nested parallel regions | |
2292 | @cindex Environment Variable | |
2293 | @cindex Implementation specific setting | |
2294 | @table @asis | |
2cd0689a TB |
2295 | @item @emph{ICV:} @var{max-active-levels-var} |
2296 | @item @emph{Scope:} data environment | |
d77de738 ML |
2297 | @item @emph{Description}: |
2298 | Enable or disable nested parallel regions, i.e., whether team members | |
2299 | are allowed to create new teams. The value of this environment variable | |
2300 | shall be @code{TRUE} or @code{FALSE}. If set to @code{TRUE}, the number | |
2301 | of maximum active nested regions supported will by default be set to the | |
2302 | maximum supported, otherwise it will be set to one. If | |
2303 | @env{OMP_MAX_ACTIVE_LEVELS} is defined, its setting will override this | |
2304 | setting. If both are undefined, nested parallel regions are enabled if | |
2305 | @env{OMP_NUM_THREADS} or @env{OMP_PROC_BINDS} are defined to a list with | |
2306 | more than one item, otherwise they are disabled by default. | |
2307 | ||
2cd0689a TB |
2308 | Note that the @code{OMP_NESTED} environment variable was deprecated in |
2309 | the OpenMP specification 5.2 in favor of @code{OMP_MAX_ACTIVE_LEVELS}. | |
2310 | ||
d77de738 | 2311 | @item @emph{See also}: |
2cd0689a TB |
2312 | @ref{omp_set_max_active_levels}, @ref{omp_set_nested}, |
2313 | @ref{OMP_MAX_ACTIVE_LEVELS} | |
d77de738 ML |
2314 | |
2315 | @item @emph{Reference}: | |
2316 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.6 | |
2317 | @end table | |
2318 | ||
2319 | ||
2320 | ||
2321 | @node OMP_NUM_TEAMS | |
2322 | @section @env{OMP_NUM_TEAMS} -- Specifies the number of teams to use by teams region | |
2323 | @cindex Environment Variable | |
2324 | @table @asis | |
2cd0689a TB |
2325 | @item @emph{ICV:} @var{nteams-var} |
2326 | @item @emph{Scope:} device | |
d77de738 ML |
2327 | @item @emph{Description}: |
2328 | Specifies the upper bound for number of teams to use in teams regions | |
2329 | without explicit @code{num_teams} clause. The value of this variable shall | |
2330 | be a positive integer. If undefined it defaults to 0 which means | |
2331 | implementation defined upper bound. | |
2332 | ||
2333 | @item @emph{See also}: | |
2334 | @ref{omp_set_num_teams} | |
2335 | ||
2336 | @item @emph{Reference}: | |
2337 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 6.23 | |
2338 | @end table | |
2339 | ||
2340 | ||
2341 | ||
2342 | @node OMP_NUM_THREADS | |
2343 | @section @env{OMP_NUM_THREADS} -- Specifies the number of threads to use | |
2344 | @cindex Environment Variable | |
2345 | @cindex Implementation specific setting | |
2346 | @table @asis | |
2cd0689a TB |
2347 | @item @emph{ICV:} @var{nthreads-var} |
2348 | @item @emph{Scope:} data environment | |
d77de738 ML |
2349 | @item @emph{Description}: |
2350 | Specifies the default number of threads to use in parallel regions. The | |
2351 | value of this variable shall be a comma-separated list of positive integers; | |
2352 | the value specifies the number of threads to use for the corresponding nested | |
2353 | level. Specifying more than one item in the list will automatically enable | |
2354 | nesting by default. If undefined one thread per CPU is used. | |
2355 | ||
2cd0689a TB |
2356 | When a list with more than value is specified, it also affects the |
2357 | @var{max-active-levels-var} ICV as described in @ref{OMP_MAX_ACTIVE_LEVELS}. | |
2358 | ||
d77de738 | 2359 | @item @emph{See also}: |
2cd0689a | 2360 | @ref{omp_set_num_threads}, @ref{OMP_MAX_ACTIVE_LEVELS} |
d77de738 ML |
2361 | |
2362 | @item @emph{Reference}: | |
2363 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.2 | |
2364 | @end table | |
2365 | ||
2366 | ||
2367 | ||
2368 | @node OMP_PROC_BIND | |
0b9bd33d | 2369 | @section @env{OMP_PROC_BIND} -- Whether threads may be moved between CPUs |
d77de738 ML |
2370 | @cindex Environment Variable |
2371 | @table @asis | |
2cd0689a TB |
2372 | @item @emph{ICV:} @var{bind-var} |
2373 | @item @emph{Scope:} data environment | |
d77de738 ML |
2374 | @item @emph{Description}: |
2375 | Specifies whether threads may be moved between processors. If set to | |
0b9bd33d | 2376 | @code{TRUE}, OpenMP threads should not be moved; if set to @code{FALSE} |
d77de738 ML |
2377 | they may be moved. Alternatively, a comma separated list with the |
2378 | values @code{PRIMARY}, @code{MASTER}, @code{CLOSE} and @code{SPREAD} can | |
2379 | be used to specify the thread affinity policy for the corresponding nesting | |
2380 | level. With @code{PRIMARY} and @code{MASTER} the worker threads are in the | |
2381 | same place partition as the primary thread. With @code{CLOSE} those are | |
2382 | kept close to the primary thread in contiguous place partitions. And | |
2383 | with @code{SPREAD} a sparse distribution | |
2384 | across the place partitions is used. Specifying more than one item in the | |
2385 | list will automatically enable nesting by default. | |
2386 | ||
2cd0689a TB |
2387 | When a list is specified, it also affects the @var{max-active-levels-var} ICV |
2388 | as described in @ref{OMP_MAX_ACTIVE_LEVELS}. | |
2389 | ||
d77de738 ML |
2390 | When undefined, @env{OMP_PROC_BIND} defaults to @code{TRUE} when |
2391 | @env{OMP_PLACES} or @env{GOMP_CPU_AFFINITY} is set and @code{FALSE} otherwise. | |
2392 | ||
2393 | @item @emph{See also}: | |
2cd0689a TB |
2394 | @ref{omp_get_proc_bind}, @ref{GOMP_CPU_AFFINITY}, @ref{OMP_PLACES}, |
2395 | @ref{OMP_MAX_ACTIVE_LEVELS} | |
d77de738 ML |
2396 | |
2397 | @item @emph{Reference}: | |
2398 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.4 | |
2399 | @end table | |
2400 | ||
2401 | ||
2402 | ||
2403 | @node OMP_PLACES | |
0b9bd33d | 2404 | @section @env{OMP_PLACES} -- Specifies on which CPUs the threads should be placed |
d77de738 ML |
2405 | @cindex Environment Variable |
2406 | @table @asis | |
2cd0689a TB |
2407 | @item @emph{ICV:} @var{place-partition-var} |
2408 | @item @emph{Scope:} implicit tasks | |
d77de738 ML |
2409 | @item @emph{Description}: |
2410 | The thread placement can be either specified using an abstract name or by an | |
2411 | explicit list of the places. The abstract names @code{threads}, @code{cores}, | |
2412 | @code{sockets}, @code{ll_caches} and @code{numa_domains} can be optionally | |
2413 | followed by a positive number in parentheses, which denotes the how many places | |
2414 | shall be created. With @code{threads} each place corresponds to a single | |
2415 | hardware thread; @code{cores} to a single core with the corresponding number of | |
2416 | hardware threads; with @code{sockets} the place corresponds to a single | |
2417 | socket; with @code{ll_caches} to a set of cores that shares the last level | |
2418 | cache on the device; and @code{numa_domains} to a set of cores for which their | |
2419 | closest memory on the device is the same memory and at a similar distance from | |
2420 | the cores. The resulting placement can be shown by setting the | |
2421 | @env{OMP_DISPLAY_ENV} environment variable. | |
2422 | ||
2423 | Alternatively, the placement can be specified explicitly as comma-separated | |
2424 | list of places. A place is specified by set of nonnegative numbers in curly | |
2425 | braces, denoting the hardware threads. The curly braces can be omitted | |
2426 | when only a single number has been specified. The hardware threads | |
2427 | belonging to a place can either be specified as comma-separated list of | |
2428 | nonnegative thread numbers or using an interval. Multiple places can also be | |
2429 | either specified by a comma-separated list of places or by an interval. To | |
2430 | specify an interval, a colon followed by the count is placed after | |
2431 | the hardware thread number or the place. Optionally, the length can be | |
2432 | followed by a colon and the stride number -- otherwise a unit stride is | |
2433 | assumed. Placing an exclamation mark (@code{!}) directly before a curly | |
2434 | brace or numbers inside the curly braces (excluding intervals) will | |
2435 | exclude those hardware threads. | |
2436 | ||
2437 | For instance, the following specifies the same places list: | |
2438 | @code{"@{0,1,2@}, @{3,4,6@}, @{7,8,9@}, @{10,11,12@}"}; | |
2439 | @code{"@{0:3@}, @{3:3@}, @{7:3@}, @{10:3@}"}; and @code{"@{0:2@}:4:3"}. | |
2440 | ||
2441 | If @env{OMP_PLACES} and @env{GOMP_CPU_AFFINITY} are unset and | |
2442 | @env{OMP_PROC_BIND} is either unset or @code{false}, threads may be moved | |
2443 | between CPUs following no placement policy. | |
2444 | ||
2445 | @item @emph{See also}: | |
2446 | @ref{OMP_PROC_BIND}, @ref{GOMP_CPU_AFFINITY}, @ref{omp_get_proc_bind}, | |
2447 | @ref{OMP_DISPLAY_ENV} | |
2448 | ||
2449 | @item @emph{Reference}: | |
2450 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.5 | |
2451 | @end table | |
2452 | ||
2453 | ||
2454 | ||
2455 | @node OMP_STACKSIZE | |
2456 | @section @env{OMP_STACKSIZE} -- Set default thread stack size | |
2457 | @cindex Environment Variable | |
2458 | @table @asis | |
2cd0689a TB |
2459 | @item @emph{ICV:} @var{stacksize-var} |
2460 | @item @emph{Scope:} device | |
d77de738 ML |
2461 | @item @emph{Description}: |
2462 | Set the default thread stack size in kilobytes, unless the number | |
2463 | is suffixed by @code{B}, @code{K}, @code{M} or @code{G}, in which | |
2464 | case the size is, respectively, in bytes, kilobytes, megabytes | |
2465 | or gigabytes. This is different from @code{pthread_attr_setstacksize} | |
2466 | which gets the number of bytes as an argument. If the stack size cannot | |
2467 | be set due to system constraints, an error is reported and the initial | |
2468 | stack size is left unchanged. If undefined, the stack size is system | |
2469 | dependent. | |
2470 | ||
2cd0689a TB |
2471 | @item @emph{See also}: |
2472 | @ref{GOMP_STACKSIZE} | |
2473 | ||
d77de738 ML |
2474 | @item @emph{Reference}: |
2475 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.7 | |
2476 | @end table | |
2477 | ||
2478 | ||
2479 | ||
2480 | @node OMP_SCHEDULE | |
2481 | @section @env{OMP_SCHEDULE} -- How threads are scheduled | |
2482 | @cindex Environment Variable | |
2483 | @cindex Implementation specific setting | |
2484 | @table @asis | |
2cd0689a TB |
2485 | @item @emph{ICV:} @var{run-sched-var} |
2486 | @item @emph{Scope:} data environment | |
d77de738 ML |
2487 | @item @emph{Description}: |
2488 | Allows to specify @code{schedule type} and @code{chunk size}. | |
2489 | The value of the variable shall have the form: @code{type[,chunk]} where | |
2490 | @code{type} is one of @code{static}, @code{dynamic}, @code{guided} or @code{auto} | |
2491 | The optional @code{chunk} size shall be a positive integer. If undefined, | |
2492 | dynamic scheduling and a chunk size of 1 is used. | |
2493 | ||
2494 | @item @emph{See also}: | |
2495 | @ref{omp_set_schedule} | |
2496 | ||
2497 | @item @emph{Reference}: | |
2498 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Sections 2.7.1.1 and 4.1 | |
2499 | @end table | |
2500 | ||
2501 | ||
2502 | ||
2503 | @node OMP_TARGET_OFFLOAD | |
2504 | @section @env{OMP_TARGET_OFFLOAD} -- Controls offloading behaviour | |
2505 | @cindex Environment Variable | |
2506 | @cindex Implementation specific setting | |
2507 | @table @asis | |
2cd0689a TB |
2508 | @item @emph{ICV:} @var{target-offload-var} |
2509 | @item @emph{Scope:} global | |
d77de738 ML |
2510 | @item @emph{Description}: |
2511 | Specifies the behaviour with regard to offloading code to a device. This | |
2512 | variable can be set to one of three values - @code{MANDATORY}, @code{DISABLED} | |
2513 | or @code{DEFAULT}. | |
2514 | ||
2515 | If set to @code{MANDATORY}, the program will terminate with an error if | |
2516 | the offload device is not present or is not supported. If set to | |
2517 | @code{DISABLED}, then offloading is disabled and all code will run on the | |
2518 | host. If set to @code{DEFAULT}, the program will try offloading to the | |
2519 | device first, then fall back to running code on the host if it cannot. | |
2520 | ||
2521 | If undefined, then the program will behave as if @code{DEFAULT} was set. | |
2522 | ||
2523 | @item @emph{Reference}: | |
2524 | @uref{https://www.openmp.org, OpenMP specification v5.0}, Section 6.17 | |
2525 | @end table | |
2526 | ||
2527 | ||
2528 | ||
2529 | @node OMP_TEAMS_THREAD_LIMIT | |
2530 | @section @env{OMP_TEAMS_THREAD_LIMIT} -- Set the maximum number of threads imposed by teams | |
2531 | @cindex Environment Variable | |
2532 | @table @asis | |
2cd0689a TB |
2533 | @item @emph{ICV:} @var{teams-thread-limit-var} |
2534 | @item @emph{Scope:} device | |
d77de738 ML |
2535 | @item @emph{Description}: |
2536 | Specifies an upper bound for the number of threads to use by each contention | |
2537 | group created by a teams construct without explicit @code{thread_limit} | |
2538 | clause. The value of this variable shall be a positive integer. If undefined, | |
2539 | the value of 0 is used which stands for an implementation defined upper | |
2540 | limit. | |
2541 | ||
2542 | @item @emph{See also}: | |
2543 | @ref{OMP_THREAD_LIMIT}, @ref{omp_set_teams_thread_limit} | |
2544 | ||
2545 | @item @emph{Reference}: | |
2546 | @uref{https://www.openmp.org, OpenMP specification v5.1}, Section 6.24 | |
2547 | @end table | |
2548 | ||
2549 | ||
2550 | ||
2551 | @node OMP_THREAD_LIMIT | |
2552 | @section @env{OMP_THREAD_LIMIT} -- Set the maximum number of threads | |
2553 | @cindex Environment Variable | |
2554 | @table @asis | |
2cd0689a TB |
2555 | @item @emph{ICV:} @var{thread-limit-var} |
2556 | @item @emph{Scope:} data environment | |
d77de738 ML |
2557 | @item @emph{Description}: |
2558 | Specifies the number of threads to use for the whole program. The | |
2559 | value of this variable shall be a positive integer. If undefined, | |
2560 | the number of threads is not limited. | |
2561 | ||
2562 | @item @emph{See also}: | |
2563 | @ref{OMP_NUM_THREADS}, @ref{omp_get_thread_limit} | |
2564 | ||
2565 | @item @emph{Reference}: | |
2566 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.10 | |
2567 | @end table | |
2568 | ||
2569 | ||
2570 | ||
2571 | @node OMP_WAIT_POLICY | |
2572 | @section @env{OMP_WAIT_POLICY} -- How waiting threads are handled | |
2573 | @cindex Environment Variable | |
2574 | @table @asis | |
2575 | @item @emph{Description}: | |
2576 | Specifies whether waiting threads should be active or passive. If | |
2577 | the value is @code{PASSIVE}, waiting threads should not consume CPU | |
2578 | power while waiting; while the value is @code{ACTIVE} specifies that | |
2579 | they should. If undefined, threads wait actively for a short time | |
2580 | before waiting passively. | |
2581 | ||
2582 | @item @emph{See also}: | |
2583 | @ref{GOMP_SPINCOUNT} | |
2584 | ||
2585 | @item @emph{Reference}: | |
2586 | @uref{https://www.openmp.org, OpenMP specification v4.5}, Section 4.8 | |
2587 | @end table | |
2588 | ||
2589 | ||
2590 | ||
2591 | @node GOMP_CPU_AFFINITY | |
2592 | @section @env{GOMP_CPU_AFFINITY} -- Bind threads to specific CPUs | |
2593 | @cindex Environment Variable | |
2594 | @table @asis | |
2595 | @item @emph{Description}: | |
2596 | Binds threads to specific CPUs. The variable should contain a space-separated | |
2597 | or comma-separated list of CPUs. This list may contain different kinds of | |
2598 | entries: either single CPU numbers in any order, a range of CPUs (M-N) | |
2599 | or a range with some stride (M-N:S). CPU numbers are zero based. For example, | |
2600 | @code{GOMP_CPU_AFFINITY="0 3 1-2 4-15:2"} will bind the initial thread | |
2601 | to CPU 0, the second to CPU 3, the third to CPU 1, the fourth to | |
2602 | CPU 2, the fifth to CPU 4, the sixth through tenth to CPUs 6, 8, 10, 12, | |
2603 | and 14 respectively and then start assigning back from the beginning of | |
2604 | the list. @code{GOMP_CPU_AFFINITY=0} binds all threads to CPU 0. | |
2605 | ||
2606 | There is no libgomp library routine to determine whether a CPU affinity | |
2607 | specification is in effect. As a workaround, language-specific library | |
2608 | functions, e.g., @code{getenv} in C or @code{GET_ENVIRONMENT_VARIABLE} in | |
2609 | Fortran, may be used to query the setting of the @code{GOMP_CPU_AFFINITY} | |
2610 | environment variable. A defined CPU affinity on startup cannot be changed | |
2611 | or disabled during the runtime of the application. | |
2612 | ||
2613 | If both @env{GOMP_CPU_AFFINITY} and @env{OMP_PROC_BIND} are set, | |
2614 | @env{OMP_PROC_BIND} has a higher precedence. If neither has been set and | |
2615 | @env{OMP_PROC_BIND} is unset, or when @env{OMP_PROC_BIND} is set to | |
2616 | @code{FALSE}, the host system will handle the assignment of threads to CPUs. | |
2617 | ||
2618 | @item @emph{See also}: | |
2619 | @ref{OMP_PLACES}, @ref{OMP_PROC_BIND} | |
2620 | @end table | |
2621 | ||
2622 | ||
2623 | ||
2624 | @node GOMP_DEBUG | |
2625 | @section @env{GOMP_DEBUG} -- Enable debugging output | |
2626 | @cindex Environment Variable | |
2627 | @table @asis | |
2628 | @item @emph{Description}: | |
2629 | Enable debugging output. The variable should be set to @code{0} | |
2630 | (disabled, also the default if not set), or @code{1} (enabled). | |
2631 | ||
2632 | If enabled, some debugging output will be printed during execution. | |
2633 | This is currently not specified in more detail, and subject to change. | |
2634 | @end table | |
2635 | ||
2636 | ||
2637 | ||
2638 | @node GOMP_STACKSIZE | |
2639 | @section @env{GOMP_STACKSIZE} -- Set default thread stack size | |
2640 | @cindex Environment Variable | |
2641 | @cindex Implementation specific setting | |
2642 | @table @asis | |
2643 | @item @emph{Description}: | |
2644 | Set the default thread stack size in kilobytes. This is different from | |
2645 | @code{pthread_attr_setstacksize} which gets the number of bytes as an | |
2646 | argument. If the stack size cannot be set due to system constraints, an | |
2647 | error is reported and the initial stack size is left unchanged. If undefined, | |
2648 | the stack size is system dependent. | |
2649 | ||
2650 | @item @emph{See also}: | |
2651 | @ref{OMP_STACKSIZE} | |
2652 | ||
2653 | @item @emph{Reference}: | |
2654 | @uref{https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00493.html, | |
2655 | GCC Patches Mailinglist}, | |
2656 | @uref{https://gcc.gnu.org/ml/gcc-patches/2006-06/msg00496.html, | |
2657 | GCC Patches Mailinglist} | |
2658 | @end table | |
2659 | ||
2660 | ||
2661 | ||
2662 | @node GOMP_SPINCOUNT | |
2663 | @section @env{GOMP_SPINCOUNT} -- Set the busy-wait spin count | |
2664 | @cindex Environment Variable | |
2665 | @cindex Implementation specific setting | |
2666 | @table @asis | |
2667 | @item @emph{Description}: | |
2668 | Determines how long a threads waits actively with consuming CPU power | |
2669 | before waiting passively without consuming CPU power. The value may be | |
2670 | either @code{INFINITE}, @code{INFINITY} to always wait actively or an | |
2671 | integer which gives the number of spins of the busy-wait loop. The | |
2672 | integer may optionally be followed by the following suffixes acting | |
2673 | as multiplication factors: @code{k} (kilo, thousand), @code{M} (mega, | |
2674 | million), @code{G} (giga, billion), or @code{T} (tera, trillion). | |
2675 | If undefined, 0 is used when @env{OMP_WAIT_POLICY} is @code{PASSIVE}, | |
2676 | 300,000 is used when @env{OMP_WAIT_POLICY} is undefined and | |
2677 | 30 billion is used when @env{OMP_WAIT_POLICY} is @code{ACTIVE}. | |
2678 | If there are more OpenMP threads than available CPUs, 1000 and 100 | |
2679 | spins are used for @env{OMP_WAIT_POLICY} being @code{ACTIVE} or | |
2680 | undefined, respectively; unless the @env{GOMP_SPINCOUNT} is lower | |
2681 | or @env{OMP_WAIT_POLICY} is @code{PASSIVE}. | |
2682 | ||
2683 | @item @emph{See also}: | |
2684 | @ref{OMP_WAIT_POLICY} | |
2685 | @end table | |
2686 | ||
2687 | ||
2688 | ||
2689 | @node GOMP_RTEMS_THREAD_POOLS | |
2690 | @section @env{GOMP_RTEMS_THREAD_POOLS} -- Set the RTEMS specific thread pools | |
2691 | @cindex Environment Variable | |
2692 | @cindex Implementation specific setting | |
2693 | @table @asis | |
2694 | @item @emph{Description}: | |
2695 | This environment variable is only used on the RTEMS real-time operating system. | |
2696 | It determines the scheduler instance specific thread pools. The format for | |
2697 | @env{GOMP_RTEMS_THREAD_POOLS} is a list of optional | |
2698 | @code{<thread-pool-count>[$<priority>]@@<scheduler-name>} configurations | |
2699 | separated by @code{:} where: | |
2700 | @itemize @bullet | |
2701 | @item @code{<thread-pool-count>} is the thread pool count for this scheduler | |
2702 | instance. | |
2703 | @item @code{$<priority>} is an optional priority for the worker threads of a | |
2704 | thread pool according to @code{pthread_setschedparam}. In case a priority | |
2705 | value is omitted, then a worker thread will inherit the priority of the OpenMP | |
2706 | primary thread that created it. The priority of the worker thread is not | |
2707 | changed after creation, even if a new OpenMP primary thread using the worker has | |
2708 | a different priority. | |
2709 | @item @code{@@<scheduler-name>} is the scheduler instance name according to the | |
2710 | RTEMS application configuration. | |
2711 | @end itemize | |
2712 | In case no thread pool configuration is specified for a scheduler instance, | |
2713 | then each OpenMP primary thread of this scheduler instance will use its own | |
2714 | dynamically allocated thread pool. To limit the worker thread count of the | |
2715 | thread pools, each OpenMP primary thread must call @code{omp_set_num_threads}. | |
2716 | @item @emph{Example}: | |
2717 | Lets suppose we have three scheduler instances @code{IO}, @code{WRK0}, and | |
2718 | @code{WRK1} with @env{GOMP_RTEMS_THREAD_POOLS} set to | |
2719 | @code{"1@@WRK0:3$4@@WRK1"}. Then there are no thread pool restrictions for | |
2720 | scheduler instance @code{IO}. In the scheduler instance @code{WRK0} there is | |
2721 | one thread pool available. Since no priority is specified for this scheduler | |
2722 | instance, the worker thread inherits the priority of the OpenMP primary thread | |
2723 | that created it. In the scheduler instance @code{WRK1} there are three thread | |
2724 | pools available and their worker threads run at priority four. | |
2725 | @end table | |
2726 | ||
2727 | ||
2728 | ||
2729 | @c --------------------------------------------------------------------- | |
2730 | @c Enabling OpenACC | |
2731 | @c --------------------------------------------------------------------- | |
2732 | ||
2733 | @node Enabling OpenACC | |
2734 | @chapter Enabling OpenACC | |
2735 | ||
2736 | To activate the OpenACC extensions for C/C++ and Fortran, the compile-time | |
2737 | flag @option{-fopenacc} must be specified. This enables the OpenACC directive | |
2738 | @code{#pragma acc} in C/C++ and @code{!$acc} directives in free form, | |
2739 | @code{c$acc}, @code{*$acc} and @code{!$acc} directives in fixed form, | |
2740 | @code{!$} conditional compilation sentinels in free form and @code{c$}, | |
2741 | @code{*$} and @code{!$} sentinels in fixed form, for Fortran. The flag also | |
2742 | arranges for automatic linking of the OpenACC runtime library | |
2743 | (@ref{OpenACC Runtime Library Routines}). | |
2744 | ||
2745 | See @uref{https://gcc.gnu.org/wiki/OpenACC} for more information. | |
2746 | ||
2747 | A complete description of all OpenACC directives accepted may be found in | |
2748 | the @uref{https://www.openacc.org, OpenACC} Application Programming | |
2749 | Interface manual, version 2.6. | |
2750 | ||
2751 | ||
2752 | ||
2753 | @c --------------------------------------------------------------------- | |
2754 | @c OpenACC Runtime Library Routines | |
2755 | @c --------------------------------------------------------------------- | |
2756 | ||
2757 | @node OpenACC Runtime Library Routines | |
2758 | @chapter OpenACC Runtime Library Routines | |
2759 | ||
2760 | The runtime routines described here are defined by section 3 of the OpenACC | |
2761 | specifications in version 2.6. | |
2762 | They have C linkage, and do not throw exceptions. | |
2763 | Generally, they are available only for the host, with the exception of | |
2764 | @code{acc_on_device}, which is available for both the host and the | |
2765 | acceleration device. | |
2766 | ||
2767 | @menu | |
2768 | * acc_get_num_devices:: Get number of devices for the given device | |
2769 | type. | |
2770 | * acc_set_device_type:: Set type of device accelerator to use. | |
2771 | * acc_get_device_type:: Get type of device accelerator to be used. | |
2772 | * acc_set_device_num:: Set device number to use. | |
2773 | * acc_get_device_num:: Get device number to be used. | |
2774 | * acc_get_property:: Get device property. | |
2775 | * acc_async_test:: Tests for completion of a specific asynchronous | |
2776 | operation. | |
2777 | * acc_async_test_all:: Tests for completion of all asynchronous | |
2778 | operations. | |
2779 | * acc_wait:: Wait for completion of a specific asynchronous | |
2780 | operation. | |
2781 | * acc_wait_all:: Waits for completion of all asynchronous | |
2782 | operations. | |
2783 | * acc_wait_all_async:: Wait for completion of all asynchronous | |
2784 | operations. | |
2785 | * acc_wait_async:: Wait for completion of asynchronous operations. | |
2786 | * acc_init:: Initialize runtime for a specific device type. | |
2787 | * acc_shutdown:: Shuts down the runtime for a specific device | |
2788 | type. | |
2789 | * acc_on_device:: Whether executing on a particular device | |
2790 | * acc_malloc:: Allocate device memory. | |
2791 | * acc_free:: Free device memory. | |
2792 | * acc_copyin:: Allocate device memory and copy host memory to | |
2793 | it. | |
2794 | * acc_present_or_copyin:: If the data is not present on the device, | |
2795 | allocate device memory and copy from host | |
2796 | memory. | |
2797 | * acc_create:: Allocate device memory and map it to host | |
2798 | memory. | |
2799 | * acc_present_or_create:: If the data is not present on the device, | |
2800 | allocate device memory and map it to host | |
2801 | memory. | |
2802 | * acc_copyout:: Copy device memory to host memory. | |
2803 | * acc_delete:: Free device memory. | |
2804 | * acc_update_device:: Update device memory from mapped host memory. | |
2805 | * acc_update_self:: Update host memory from mapped device memory. | |
2806 | * acc_map_data:: Map previously allocated device memory to host | |
2807 | memory. | |
2808 | * acc_unmap_data:: Unmap device memory from host memory. | |
2809 | * acc_deviceptr:: Get device pointer associated with specific | |
2810 | host address. | |
2811 | * acc_hostptr:: Get host pointer associated with specific | |
2812 | device address. | |
2813 | * acc_is_present:: Indicate whether host variable / array is | |
2814 | present on device. | |
2815 | * acc_memcpy_to_device:: Copy host memory to device memory. | |
2816 | * acc_memcpy_from_device:: Copy device memory to host memory. | |
2817 | * acc_attach:: Let device pointer point to device-pointer target. | |
2818 | * acc_detach:: Let device pointer point to host-pointer target. | |
2819 | ||
2820 | API routines for target platforms. | |
2821 | ||
2822 | * acc_get_current_cuda_device:: Get CUDA device handle. | |
2823 | * acc_get_current_cuda_context::Get CUDA context handle. | |
2824 | * acc_get_cuda_stream:: Get CUDA stream handle. | |
2825 | * acc_set_cuda_stream:: Set CUDA stream handle. | |
2826 | ||
2827 | API routines for the OpenACC Profiling Interface. | |
2828 | ||
2829 | * acc_prof_register:: Register callbacks. | |
2830 | * acc_prof_unregister:: Unregister callbacks. | |
2831 | * acc_prof_lookup:: Obtain inquiry functions. | |
2832 | * acc_register_library:: Library registration. | |
2833 | @end menu | |
2834 | ||
2835 | ||
2836 | ||
2837 | @node acc_get_num_devices | |
2838 | @section @code{acc_get_num_devices} -- Get number of devices for given device type | |
2839 | @table @asis | |
2840 | @item @emph{Description} | |
2841 | This function returns a value indicating the number of devices available | |
2842 | for the device type specified in @var{devicetype}. | |
2843 | ||
2844 | @item @emph{C/C++}: | |
2845 | @multitable @columnfractions .20 .80 | |
2846 | @item @emph{Prototype}: @tab @code{int acc_get_num_devices(acc_device_t devicetype);} | |
2847 | @end multitable | |
2848 | ||
2849 | @item @emph{Fortran}: | |
2850 | @multitable @columnfractions .20 .80 | |
2851 | @item @emph{Interface}: @tab @code{integer function acc_get_num_devices(devicetype)} | |
2852 | @item @tab @code{integer(kind=acc_device_kind) devicetype} | |
2853 | @end multitable | |
2854 | ||
2855 | @item @emph{Reference}: | |
2856 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
2857 | 3.2.1. | |
2858 | @end table | |
2859 | ||
2860 | ||
2861 | ||
2862 | @node acc_set_device_type | |
2863 | @section @code{acc_set_device_type} -- Set type of device accelerator to use. | |
2864 | @table @asis | |
2865 | @item @emph{Description} | |
2866 | This function indicates to the runtime library which device type, specified | |
2867 | in @var{devicetype}, to use when executing a parallel or kernels region. | |
2868 | ||
2869 | @item @emph{C/C++}: | |
2870 | @multitable @columnfractions .20 .80 | |
2871 | @item @emph{Prototype}: @tab @code{acc_set_device_type(acc_device_t devicetype);} | |
2872 | @end multitable | |
2873 | ||
2874 | @item @emph{Fortran}: | |
2875 | @multitable @columnfractions .20 .80 | |
2876 | @item @emph{Interface}: @tab @code{subroutine acc_set_device_type(devicetype)} | |
2877 | @item @tab @code{integer(kind=acc_device_kind) devicetype} | |
2878 | @end multitable | |
2879 | ||
2880 | @item @emph{Reference}: | |
2881 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
2882 | 3.2.2. | |
2883 | @end table | |
2884 | ||
2885 | ||
2886 | ||
2887 | @node acc_get_device_type | |
2888 | @section @code{acc_get_device_type} -- Get type of device accelerator to be used. | |
2889 | @table @asis | |
2890 | @item @emph{Description} | |
2891 | This function returns what device type will be used when executing a | |
2892 | parallel or kernels region. | |
2893 | ||
2894 | This function returns @code{acc_device_none} if | |
2895 | @code{acc_get_device_type} is called from | |
2896 | @code{acc_ev_device_init_start}, @code{acc_ev_device_init_end} | |
2897 | callbacks of the OpenACC Profiling Interface (@ref{OpenACC Profiling | |
2898 | Interface}), that is, if the device is currently being initialized. | |
2899 | ||
2900 | @item @emph{C/C++}: | |
2901 | @multitable @columnfractions .20 .80 | |
2902 | @item @emph{Prototype}: @tab @code{acc_device_t acc_get_device_type(void);} | |
2903 | @end multitable | |
2904 | ||
2905 | @item @emph{Fortran}: | |
2906 | @multitable @columnfractions .20 .80 | |
2907 | @item @emph{Interface}: @tab @code{function acc_get_device_type(void)} | |
2908 | @item @tab @code{integer(kind=acc_device_kind) acc_get_device_type} | |
2909 | @end multitable | |
2910 | ||
2911 | @item @emph{Reference}: | |
2912 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
2913 | 3.2.3. | |
2914 | @end table | |
2915 | ||
2916 | ||
2917 | ||
2918 | @node acc_set_device_num | |
2919 | @section @code{acc_set_device_num} -- Set device number to use. | |
2920 | @table @asis | |
2921 | @item @emph{Description} | |
2922 | This function will indicate to the runtime which device number, | |
2923 | specified by @var{devicenum}, associated with the specified device | |
2924 | type @var{devicetype}. | |
2925 | ||
2926 | @item @emph{C/C++}: | |
2927 | @multitable @columnfractions .20 .80 | |
2928 | @item @emph{Prototype}: @tab @code{acc_set_device_num(int devicenum, acc_device_t devicetype);} | |
2929 | @end multitable | |
2930 | ||
2931 | @item @emph{Fortran}: | |
2932 | @multitable @columnfractions .20 .80 | |
2933 | @item @emph{Interface}: @tab @code{subroutine acc_set_device_num(devicenum, devicetype)} | |
2934 | @item @tab @code{integer devicenum} | |
2935 | @item @tab @code{integer(kind=acc_device_kind) devicetype} | |
2936 | @end multitable | |
2937 | ||
2938 | @item @emph{Reference}: | |
2939 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
2940 | 3.2.4. | |
2941 | @end table | |
2942 | ||
2943 | ||
2944 | ||
2945 | @node acc_get_device_num | |
2946 | @section @code{acc_get_device_num} -- Get device number to be used. | |
2947 | @table @asis | |
2948 | @item @emph{Description} | |
2949 | This function returns which device number associated with the specified device | |
2950 | type @var{devicetype}, will be used when executing a parallel or kernels | |
2951 | region. | |
2952 | ||
2953 | @item @emph{C/C++}: | |
2954 | @multitable @columnfractions .20 .80 | |
2955 | @item @emph{Prototype}: @tab @code{int acc_get_device_num(acc_device_t devicetype);} | |
2956 | @end multitable | |
2957 | ||
2958 | @item @emph{Fortran}: | |
2959 | @multitable @columnfractions .20 .80 | |
2960 | @item @emph{Interface}: @tab @code{function acc_get_device_num(devicetype)} | |
2961 | @item @tab @code{integer(kind=acc_device_kind) devicetype} | |
2962 | @item @tab @code{integer acc_get_device_num} | |
2963 | @end multitable | |
2964 | ||
2965 | @item @emph{Reference}: | |
2966 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
2967 | 3.2.5. | |
2968 | @end table | |
2969 | ||
2970 | ||
2971 | ||
2972 | @node acc_get_property | |
2973 | @section @code{acc_get_property} -- Get device property. | |
2974 | @cindex acc_get_property | |
2975 | @cindex acc_get_property_string | |
2976 | @table @asis | |
2977 | @item @emph{Description} | |
2978 | These routines return the value of the specified @var{property} for the | |
2979 | device being queried according to @var{devicenum} and @var{devicetype}. | |
2980 | Integer-valued and string-valued properties are returned by | |
2981 | @code{acc_get_property} and @code{acc_get_property_string} respectively. | |
2982 | The Fortran @code{acc_get_property_string} subroutine returns the string | |
2983 | retrieved in its fourth argument while the remaining entry points are | |
2984 | functions, which pass the return value as their result. | |
2985 | ||
2986 | Note for Fortran, only: the OpenACC technical committee corrected and, hence, | |
2987 | modified the interface introduced in OpenACC 2.6. The kind-value parameter | |
2988 | @code{acc_device_property} has been renamed to @code{acc_device_property_kind} | |
2989 | for consistency and the return type of the @code{acc_get_property} function is | |
2990 | now a @code{c_size_t} integer instead of a @code{acc_device_property} integer. | |
2991 | The parameter @code{acc_device_property} will continue to be provided, | |
2992 | but might be removed in a future version of GCC. | |
2993 | ||
2994 | @item @emph{C/C++}: | |
2995 | @multitable @columnfractions .20 .80 | |
2996 | @item @emph{Prototype}: @tab @code{size_t acc_get_property(int devicenum, acc_device_t devicetype, acc_device_property_t property);} | |
2997 | @item @emph{Prototype}: @tab @code{const char *acc_get_property_string(int devicenum, acc_device_t devicetype, acc_device_property_t property);} | |
2998 | @end multitable | |
2999 | ||
3000 | @item @emph{Fortran}: | |
3001 | @multitable @columnfractions .20 .80 | |
3002 | @item @emph{Interface}: @tab @code{function acc_get_property(devicenum, devicetype, property)} | |
3003 | @item @emph{Interface}: @tab @code{subroutine acc_get_property_string(devicenum, devicetype, property, string)} | |
3004 | @item @tab @code{use ISO_C_Binding, only: c_size_t} | |
3005 | @item @tab @code{integer devicenum} | |
3006 | @item @tab @code{integer(kind=acc_device_kind) devicetype} | |
3007 | @item @tab @code{integer(kind=acc_device_property_kind) property} | |
3008 | @item @tab @code{integer(kind=c_size_t) acc_get_property} | |
3009 | @item @tab @code{character(*) string} | |
3010 | @end multitable | |
3011 | ||
3012 | @item @emph{Reference}: | |
3013 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3014 | 3.2.6. | |
3015 | @end table | |
3016 | ||
3017 | ||
3018 | ||
3019 | @node acc_async_test | |
3020 | @section @code{acc_async_test} -- Test for completion of a specific asynchronous operation. | |
3021 | @table @asis | |
3022 | @item @emph{Description} | |
3023 | This function tests for completion of the asynchronous operation specified | |
3024 | in @var{arg}. In C/C++, a non-zero value will be returned to indicate | |
3025 | the specified asynchronous operation has completed. While Fortran will return | |
3026 | a @code{true}. If the asynchronous operation has not completed, C/C++ returns | |
3027 | a zero and Fortran returns a @code{false}. | |
3028 | ||
3029 | @item @emph{C/C++}: | |
3030 | @multitable @columnfractions .20 .80 | |
3031 | @item @emph{Prototype}: @tab @code{int acc_async_test(int arg);} | |
3032 | @end multitable | |
3033 | ||
3034 | @item @emph{Fortran}: | |
3035 | @multitable @columnfractions .20 .80 | |
3036 | @item @emph{Interface}: @tab @code{function acc_async_test(arg)} | |
3037 | @item @tab @code{integer(kind=acc_handle_kind) arg} | |
3038 | @item @tab @code{logical acc_async_test} | |
3039 | @end multitable | |
3040 | ||
3041 | @item @emph{Reference}: | |
3042 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3043 | 3.2.9. | |
3044 | @end table | |
3045 | ||
3046 | ||
3047 | ||
3048 | @node acc_async_test_all | |
3049 | @section @code{acc_async_test_all} -- Tests for completion of all asynchronous operations. | |
3050 | @table @asis | |
3051 | @item @emph{Description} | |
3052 | This function tests for completion of all asynchronous operations. | |
3053 | In C/C++, a non-zero value will be returned to indicate all asynchronous | |
3054 | operations have completed. While Fortran will return a @code{true}. If | |
3055 | any asynchronous operation has not completed, C/C++ returns a zero and | |
3056 | Fortran returns a @code{false}. | |
3057 | ||
3058 | @item @emph{C/C++}: | |
3059 | @multitable @columnfractions .20 .80 | |
3060 | @item @emph{Prototype}: @tab @code{int acc_async_test_all(void);} | |
3061 | @end multitable | |
3062 | ||
3063 | @item @emph{Fortran}: | |
3064 | @multitable @columnfractions .20 .80 | |
3065 | @item @emph{Interface}: @tab @code{function acc_async_test()} | |
3066 | @item @tab @code{logical acc_get_device_num} | |
3067 | @end multitable | |
3068 | ||
3069 | @item @emph{Reference}: | |
3070 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3071 | 3.2.10. | |
3072 | @end table | |
3073 | ||
3074 | ||
3075 | ||
3076 | @node acc_wait | |
3077 | @section @code{acc_wait} -- Wait for completion of a specific asynchronous operation. | |
3078 | @table @asis | |
3079 | @item @emph{Description} | |
3080 | This function waits for completion of the asynchronous operation | |
3081 | specified in @var{arg}. | |
3082 | ||
3083 | @item @emph{C/C++}: | |
3084 | @multitable @columnfractions .20 .80 | |
3085 | @item @emph{Prototype}: @tab @code{acc_wait(arg);} | |
3086 | @item @emph{Prototype (OpenACC 1.0 compatibility)}: @tab @code{acc_async_wait(arg);} | |
3087 | @end multitable | |
3088 | ||
3089 | @item @emph{Fortran}: | |
3090 | @multitable @columnfractions .20 .80 | |
3091 | @item @emph{Interface}: @tab @code{subroutine acc_wait(arg)} | |
3092 | @item @tab @code{integer(acc_handle_kind) arg} | |
3093 | @item @emph{Interface (OpenACC 1.0 compatibility)}: @tab @code{subroutine acc_async_wait(arg)} | |
3094 | @item @tab @code{integer(acc_handle_kind) arg} | |
3095 | @end multitable | |
3096 | ||
3097 | @item @emph{Reference}: | |
3098 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3099 | 3.2.11. | |
3100 | @end table | |
3101 | ||
3102 | ||
3103 | ||
3104 | @node acc_wait_all | |
3105 | @section @code{acc_wait_all} -- Waits for completion of all asynchronous operations. | |
3106 | @table @asis | |
3107 | @item @emph{Description} | |
3108 | This function waits for the completion of all asynchronous operations. | |
3109 | ||
3110 | @item @emph{C/C++}: | |
3111 | @multitable @columnfractions .20 .80 | |
3112 | @item @emph{Prototype}: @tab @code{acc_wait_all(void);} | |
3113 | @item @emph{Prototype (OpenACC 1.0 compatibility)}: @tab @code{acc_async_wait_all(void);} | |
3114 | @end multitable | |
3115 | ||
3116 | @item @emph{Fortran}: | |
3117 | @multitable @columnfractions .20 .80 | |
3118 | @item @emph{Interface}: @tab @code{subroutine acc_wait_all()} | |
3119 | @item @emph{Interface (OpenACC 1.0 compatibility)}: @tab @code{subroutine acc_async_wait_all()} | |
3120 | @end multitable | |
3121 | ||
3122 | @item @emph{Reference}: | |
3123 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3124 | 3.2.13. | |
3125 | @end table | |
3126 | ||
3127 | ||
3128 | ||
3129 | @node acc_wait_all_async | |
3130 | @section @code{acc_wait_all_async} -- Wait for completion of all asynchronous operations. | |
3131 | @table @asis | |
3132 | @item @emph{Description} | |
3133 | This function enqueues a wait operation on the queue @var{async} for any | |
3134 | and all asynchronous operations that have been previously enqueued on | |
3135 | any queue. | |
3136 | ||
3137 | @item @emph{C/C++}: | |
3138 | @multitable @columnfractions .20 .80 | |
3139 | @item @emph{Prototype}: @tab @code{acc_wait_all_async(int async);} | |
3140 | @end multitable | |
3141 | ||
3142 | @item @emph{Fortran}: | |
3143 | @multitable @columnfractions .20 .80 | |
3144 | @item @emph{Interface}: @tab @code{subroutine acc_wait_all_async(async)} | |
3145 | @item @tab @code{integer(acc_handle_kind) async} | |
3146 | @end multitable | |
3147 | ||
3148 | @item @emph{Reference}: | |
3149 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3150 | 3.2.14. | |
3151 | @end table | |
3152 | ||
3153 | ||
3154 | ||
3155 | @node acc_wait_async | |
3156 | @section @code{acc_wait_async} -- Wait for completion of asynchronous operations. | |
3157 | @table @asis | |
3158 | @item @emph{Description} | |
3159 | This function enqueues a wait operation on queue @var{async} for any and all | |
3160 | asynchronous operations enqueued on queue @var{arg}. | |
3161 | ||
3162 | @item @emph{C/C++}: | |
3163 | @multitable @columnfractions .20 .80 | |
3164 | @item @emph{Prototype}: @tab @code{acc_wait_async(int arg, int async);} | |
3165 | @end multitable | |
3166 | ||
3167 | @item @emph{Fortran}: | |
3168 | @multitable @columnfractions .20 .80 | |
3169 | @item @emph{Interface}: @tab @code{subroutine acc_wait_async(arg, async)} | |
3170 | @item @tab @code{integer(acc_handle_kind) arg, async} | |
3171 | @end multitable | |
3172 | ||
3173 | @item @emph{Reference}: | |
3174 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3175 | 3.2.12. | |
3176 | @end table | |
3177 | ||
3178 | ||
3179 | ||
3180 | @node acc_init | |
3181 | @section @code{acc_init} -- Initialize runtime for a specific device type. | |
3182 | @table @asis | |
3183 | @item @emph{Description} | |
3184 | This function initializes the runtime for the device type specified in | |
3185 | @var{devicetype}. | |
3186 | ||
3187 | @item @emph{C/C++}: | |
3188 | @multitable @columnfractions .20 .80 | |
3189 | @item @emph{Prototype}: @tab @code{acc_init(acc_device_t devicetype);} | |
3190 | @end multitable | |
3191 | ||
3192 | @item @emph{Fortran}: | |
3193 | @multitable @columnfractions .20 .80 | |
3194 | @item @emph{Interface}: @tab @code{subroutine acc_init(devicetype)} | |
3195 | @item @tab @code{integer(acc_device_kind) devicetype} | |
3196 | @end multitable | |
3197 | ||
3198 | @item @emph{Reference}: | |
3199 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3200 | 3.2.7. | |
3201 | @end table | |
3202 | ||
3203 | ||
3204 | ||
3205 | @node acc_shutdown | |
3206 | @section @code{acc_shutdown} -- Shuts down the runtime for a specific device type. | |
3207 | @table @asis | |
3208 | @item @emph{Description} | |
3209 | This function shuts down the runtime for the device type specified in | |
3210 | @var{devicetype}. | |
3211 | ||
3212 | @item @emph{C/C++}: | |
3213 | @multitable @columnfractions .20 .80 | |
3214 | @item @emph{Prototype}: @tab @code{acc_shutdown(acc_device_t devicetype);} | |
3215 | @end multitable | |
3216 | ||
3217 | @item @emph{Fortran}: | |
3218 | @multitable @columnfractions .20 .80 | |
3219 | @item @emph{Interface}: @tab @code{subroutine acc_shutdown(devicetype)} | |
3220 | @item @tab @code{integer(acc_device_kind) devicetype} | |
3221 | @end multitable | |
3222 | ||
3223 | @item @emph{Reference}: | |
3224 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3225 | 3.2.8. | |
3226 | @end table | |
3227 | ||
3228 | ||
3229 | ||
3230 | @node acc_on_device | |
3231 | @section @code{acc_on_device} -- Whether executing on a particular device | |
3232 | @table @asis | |
3233 | @item @emph{Description}: | |
3234 | This function returns whether the program is executing on a particular | |
3235 | device specified in @var{devicetype}. In C/C++ a non-zero value is | |
3236 | returned to indicate the device is executing on the specified device type. | |
3237 | In Fortran, @code{true} will be returned. If the program is not executing | |
3238 | on the specified device type C/C++ will return a zero, while Fortran will | |
3239 | return @code{false}. | |
3240 | ||
3241 | @item @emph{C/C++}: | |
3242 | @multitable @columnfractions .20 .80 | |
3243 | @item @emph{Prototype}: @tab @code{acc_on_device(acc_device_t devicetype);} | |
3244 | @end multitable | |
3245 | ||
3246 | @item @emph{Fortran}: | |
3247 | @multitable @columnfractions .20 .80 | |
3248 | @item @emph{Interface}: @tab @code{function acc_on_device(devicetype)} | |
3249 | @item @tab @code{integer(acc_device_kind) devicetype} | |
3250 | @item @tab @code{logical acc_on_device} | |
3251 | @end multitable | |
3252 | ||
3253 | ||
3254 | @item @emph{Reference}: | |
3255 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3256 | 3.2.17. | |
3257 | @end table | |
3258 | ||
3259 | ||
3260 | ||
3261 | @node acc_malloc | |
3262 | @section @code{acc_malloc} -- Allocate device memory. | |
3263 | @table @asis | |
3264 | @item @emph{Description} | |
3265 | This function allocates @var{len} bytes of device memory. It returns | |
3266 | the device address of the allocated memory. | |
3267 | ||
3268 | @item @emph{C/C++}: | |
3269 | @multitable @columnfractions .20 .80 | |
3270 | @item @emph{Prototype}: @tab @code{d_void* acc_malloc(size_t len);} | |
3271 | @end multitable | |
3272 | ||
3273 | @item @emph{Reference}: | |
3274 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3275 | 3.2.18. | |
3276 | @end table | |
3277 | ||
3278 | ||
3279 | ||
3280 | @node acc_free | |
3281 | @section @code{acc_free} -- Free device memory. | |
3282 | @table @asis | |
3283 | @item @emph{Description} | |
3284 | Free previously allocated device memory at the device address @code{a}. | |
3285 | ||
3286 | @item @emph{C/C++}: | |
3287 | @multitable @columnfractions .20 .80 | |
3288 | @item @emph{Prototype}: @tab @code{acc_free(d_void *a);} | |
3289 | @end multitable | |
3290 | ||
3291 | @item @emph{Reference}: | |
3292 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3293 | 3.2.19. | |
3294 | @end table | |
3295 | ||
3296 | ||
3297 | ||
3298 | @node acc_copyin | |
3299 | @section @code{acc_copyin} -- Allocate device memory and copy host memory to it. | |
3300 | @table @asis | |
3301 | @item @emph{Description} | |
3302 | In C/C++, this function allocates @var{len} bytes of device memory | |
3303 | and maps it to the specified host address in @var{a}. The device | |
3304 | address of the newly allocated device memory is returned. | |
3305 | ||
3306 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3307 | a contiguous array section. The second form @var{a} specifies a | |
3308 | variable or array element and @var{len} specifies the length in bytes. | |
3309 | ||
3310 | @item @emph{C/C++}: | |
3311 | @multitable @columnfractions .20 .80 | |
3312 | @item @emph{Prototype}: @tab @code{void *acc_copyin(h_void *a, size_t len);} | |
3313 | @item @emph{Prototype}: @tab @code{void *acc_copyin_async(h_void *a, size_t len, int async);} | |
3314 | @end multitable | |
3315 | ||
3316 | @item @emph{Fortran}: | |
3317 | @multitable @columnfractions .20 .80 | |
3318 | @item @emph{Interface}: @tab @code{subroutine acc_copyin(a)} | |
3319 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3320 | @item @emph{Interface}: @tab @code{subroutine acc_copyin(a, len)} | |
3321 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3322 | @item @tab @code{integer len} | |
3323 | @item @emph{Interface}: @tab @code{subroutine acc_copyin_async(a, async)} | |
3324 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3325 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3326 | @item @emph{Interface}: @tab @code{subroutine acc_copyin_async(a, len, async)} | |
3327 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3328 | @item @tab @code{integer len} | |
3329 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3330 | @end multitable | |
3331 | ||
3332 | @item @emph{Reference}: | |
3333 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3334 | 3.2.20. | |
3335 | @end table | |
3336 | ||
3337 | ||
3338 | ||
3339 | @node acc_present_or_copyin | |
3340 | @section @code{acc_present_or_copyin} -- If the data is not present on the device, allocate device memory and copy from host memory. | |
3341 | @table @asis | |
3342 | @item @emph{Description} | |
3343 | This function tests if the host data specified by @var{a} and of length | |
3344 | @var{len} is present or not. If it is not present, then device memory | |
3345 | will be allocated and the host memory copied. The device address of | |
3346 | the newly allocated device memory is returned. | |
3347 | ||
3348 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3349 | a contiguous array section. The second form @var{a} specifies a variable or | |
3350 | array element and @var{len} specifies the length in bytes. | |
3351 | ||
3352 | Note that @code{acc_present_or_copyin} and @code{acc_pcopyin} exist for | |
3353 | backward compatibility with OpenACC 2.0; use @ref{acc_copyin} instead. | |
3354 | ||
3355 | @item @emph{C/C++}: | |
3356 | @multitable @columnfractions .20 .80 | |
3357 | @item @emph{Prototype}: @tab @code{void *acc_present_or_copyin(h_void *a, size_t len);} | |
3358 | @item @emph{Prototype}: @tab @code{void *acc_pcopyin(h_void *a, size_t len);} | |
3359 | @end multitable | |
3360 | ||
3361 | @item @emph{Fortran}: | |
3362 | @multitable @columnfractions .20 .80 | |
3363 | @item @emph{Interface}: @tab @code{subroutine acc_present_or_copyin(a)} | |
3364 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3365 | @item @emph{Interface}: @tab @code{subroutine acc_present_or_copyin(a, len)} | |
3366 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3367 | @item @tab @code{integer len} | |
3368 | @item @emph{Interface}: @tab @code{subroutine acc_pcopyin(a)} | |
3369 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3370 | @item @emph{Interface}: @tab @code{subroutine acc_pcopyin(a, len)} | |
3371 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3372 | @item @tab @code{integer len} | |
3373 | @end multitable | |
3374 | ||
3375 | @item @emph{Reference}: | |
3376 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3377 | 3.2.20. | |
3378 | @end table | |
3379 | ||
3380 | ||
3381 | ||
3382 | @node acc_create | |
3383 | @section @code{acc_create} -- Allocate device memory and map it to host memory. | |
3384 | @table @asis | |
3385 | @item @emph{Description} | |
3386 | This function allocates device memory and maps it to host memory specified | |
3387 | by the host address @var{a} with a length of @var{len} bytes. In C/C++, | |
3388 | the function returns the device address of the allocated device memory. | |
3389 | ||
3390 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3391 | a contiguous array section. The second form @var{a} specifies a variable or | |
3392 | array element and @var{len} specifies the length in bytes. | |
3393 | ||
3394 | @item @emph{C/C++}: | |
3395 | @multitable @columnfractions .20 .80 | |
3396 | @item @emph{Prototype}: @tab @code{void *acc_create(h_void *a, size_t len);} | |
3397 | @item @emph{Prototype}: @tab @code{void *acc_create_async(h_void *a, size_t len, int async);} | |
3398 | @end multitable | |
3399 | ||
3400 | @item @emph{Fortran}: | |
3401 | @multitable @columnfractions .20 .80 | |
3402 | @item @emph{Interface}: @tab @code{subroutine acc_create(a)} | |
3403 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3404 | @item @emph{Interface}: @tab @code{subroutine acc_create(a, len)} | |
3405 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3406 | @item @tab @code{integer len} | |
3407 | @item @emph{Interface}: @tab @code{subroutine acc_create_async(a, async)} | |
3408 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3409 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3410 | @item @emph{Interface}: @tab @code{subroutine acc_create_async(a, len, async)} | |
3411 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3412 | @item @tab @code{integer len} | |
3413 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3414 | @end multitable | |
3415 | ||
3416 | @item @emph{Reference}: | |
3417 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3418 | 3.2.21. | |
3419 | @end table | |
3420 | ||
3421 | ||
3422 | ||
3423 | @node acc_present_or_create | |
3424 | @section @code{acc_present_or_create} -- If the data is not present on the device, allocate device memory and map it to host memory. | |
3425 | @table @asis | |
3426 | @item @emph{Description} | |
3427 | This function tests if the host data specified by @var{a} and of length | |
3428 | @var{len} is present or not. If it is not present, then device memory | |
3429 | will be allocated and mapped to host memory. In C/C++, the device address | |
3430 | of the newly allocated device memory is returned. | |
3431 | ||
3432 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3433 | a contiguous array section. The second form @var{a} specifies a variable or | |
3434 | array element and @var{len} specifies the length in bytes. | |
3435 | ||
3436 | Note that @code{acc_present_or_create} and @code{acc_pcreate} exist for | |
3437 | backward compatibility with OpenACC 2.0; use @ref{acc_create} instead. | |
3438 | ||
3439 | @item @emph{C/C++}: | |
3440 | @multitable @columnfractions .20 .80 | |
3441 | @item @emph{Prototype}: @tab @code{void *acc_present_or_create(h_void *a, size_t len)} | |
3442 | @item @emph{Prototype}: @tab @code{void *acc_pcreate(h_void *a, size_t len)} | |
3443 | @end multitable | |
3444 | ||
3445 | @item @emph{Fortran}: | |
3446 | @multitable @columnfractions .20 .80 | |
3447 | @item @emph{Interface}: @tab @code{subroutine acc_present_or_create(a)} | |
3448 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3449 | @item @emph{Interface}: @tab @code{subroutine acc_present_or_create(a, len)} | |
3450 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3451 | @item @tab @code{integer len} | |
3452 | @item @emph{Interface}: @tab @code{subroutine acc_pcreate(a)} | |
3453 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3454 | @item @emph{Interface}: @tab @code{subroutine acc_pcreate(a, len)} | |
3455 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3456 | @item @tab @code{integer len} | |
3457 | @end multitable | |
3458 | ||
3459 | @item @emph{Reference}: | |
3460 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3461 | 3.2.21. | |
3462 | @end table | |
3463 | ||
3464 | ||
3465 | ||
3466 | @node acc_copyout | |
3467 | @section @code{acc_copyout} -- Copy device memory to host memory. | |
3468 | @table @asis | |
3469 | @item @emph{Description} | |
3470 | This function copies mapped device memory to host memory which is specified | |
3471 | by host address @var{a} for a length @var{len} bytes in C/C++. | |
3472 | ||
3473 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3474 | a contiguous array section. The second form @var{a} specifies a variable or | |
3475 | array element and @var{len} specifies the length in bytes. | |
3476 | ||
3477 | @item @emph{C/C++}: | |
3478 | @multitable @columnfractions .20 .80 | |
3479 | @item @emph{Prototype}: @tab @code{acc_copyout(h_void *a, size_t len);} | |
3480 | @item @emph{Prototype}: @tab @code{acc_copyout_async(h_void *a, size_t len, int async);} | |
3481 | @item @emph{Prototype}: @tab @code{acc_copyout_finalize(h_void *a, size_t len);} | |
3482 | @item @emph{Prototype}: @tab @code{acc_copyout_finalize_async(h_void *a, size_t len, int async);} | |
3483 | @end multitable | |
3484 | ||
3485 | @item @emph{Fortran}: | |
3486 | @multitable @columnfractions .20 .80 | |
3487 | @item @emph{Interface}: @tab @code{subroutine acc_copyout(a)} | |
3488 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3489 | @item @emph{Interface}: @tab @code{subroutine acc_copyout(a, len)} | |
3490 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3491 | @item @tab @code{integer len} | |
3492 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_async(a, async)} | |
3493 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3494 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3495 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_async(a, len, async)} | |
3496 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3497 | @item @tab @code{integer len} | |
3498 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3499 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize(a)} | |
3500 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3501 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize(a, len)} | |
3502 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3503 | @item @tab @code{integer len} | |
3504 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize_async(a, async)} | |
3505 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3506 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3507 | @item @emph{Interface}: @tab @code{subroutine acc_copyout_finalize_async(a, len, async)} | |
3508 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3509 | @item @tab @code{integer len} | |
3510 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3511 | @end multitable | |
3512 | ||
3513 | @item @emph{Reference}: | |
3514 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3515 | 3.2.22. | |
3516 | @end table | |
3517 | ||
3518 | ||
3519 | ||
3520 | @node acc_delete | |
3521 | @section @code{acc_delete} -- Free device memory. | |
3522 | @table @asis | |
3523 | @item @emph{Description} | |
3524 | This function frees previously allocated device memory specified by | |
3525 | the device address @var{a} and the length of @var{len} bytes. | |
3526 | ||
3527 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3528 | a contiguous array section. The second form @var{a} specifies a variable or | |
3529 | array element and @var{len} specifies the length in bytes. | |
3530 | ||
3531 | @item @emph{C/C++}: | |
3532 | @multitable @columnfractions .20 .80 | |
3533 | @item @emph{Prototype}: @tab @code{acc_delete(h_void *a, size_t len);} | |
3534 | @item @emph{Prototype}: @tab @code{acc_delete_async(h_void *a, size_t len, int async);} | |
3535 | @item @emph{Prototype}: @tab @code{acc_delete_finalize(h_void *a, size_t len);} | |
3536 | @item @emph{Prototype}: @tab @code{acc_delete_finalize_async(h_void *a, size_t len, int async);} | |
3537 | @end multitable | |
3538 | ||
3539 | @item @emph{Fortran}: | |
3540 | @multitable @columnfractions .20 .80 | |
3541 | @item @emph{Interface}: @tab @code{subroutine acc_delete(a)} | |
3542 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3543 | @item @emph{Interface}: @tab @code{subroutine acc_delete(a, len)} | |
3544 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3545 | @item @tab @code{integer len} | |
3546 | @item @emph{Interface}: @tab @code{subroutine acc_delete_async(a, async)} | |
3547 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3548 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3549 | @item @emph{Interface}: @tab @code{subroutine acc_delete_async(a, len, async)} | |
3550 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3551 | @item @tab @code{integer len} | |
3552 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3553 | @item @emph{Interface}: @tab @code{subroutine acc_delete_finalize(a)} | |
3554 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3555 | @item @emph{Interface}: @tab @code{subroutine acc_delete_finalize(a, len)} | |
3556 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3557 | @item @tab @code{integer len} | |
3558 | @item @emph{Interface}: @tab @code{subroutine acc_delete_async_finalize(a, async)} | |
3559 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3560 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3561 | @item @emph{Interface}: @tab @code{subroutine acc_delete_async_finalize(a, len, async)} | |
3562 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3563 | @item @tab @code{integer len} | |
3564 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3565 | @end multitable | |
3566 | ||
3567 | @item @emph{Reference}: | |
3568 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3569 | 3.2.23. | |
3570 | @end table | |
3571 | ||
3572 | ||
3573 | ||
3574 | @node acc_update_device | |
3575 | @section @code{acc_update_device} -- Update device memory from mapped host memory. | |
3576 | @table @asis | |
3577 | @item @emph{Description} | |
3578 | This function updates the device copy from the previously mapped host memory. | |
3579 | The host memory is specified with the host address @var{a} and a length of | |
3580 | @var{len} bytes. | |
3581 | ||
3582 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3583 | a contiguous array section. The second form @var{a} specifies a variable or | |
3584 | array element and @var{len} specifies the length in bytes. | |
3585 | ||
3586 | @item @emph{C/C++}: | |
3587 | @multitable @columnfractions .20 .80 | |
3588 | @item @emph{Prototype}: @tab @code{acc_update_device(h_void *a, size_t len);} | |
3589 | @item @emph{Prototype}: @tab @code{acc_update_device(h_void *a, size_t len, async);} | |
3590 | @end multitable | |
3591 | ||
3592 | @item @emph{Fortran}: | |
3593 | @multitable @columnfractions .20 .80 | |
3594 | @item @emph{Interface}: @tab @code{subroutine acc_update_device(a)} | |
3595 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3596 | @item @emph{Interface}: @tab @code{subroutine acc_update_device(a, len)} | |
3597 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3598 | @item @tab @code{integer len} | |
3599 | @item @emph{Interface}: @tab @code{subroutine acc_update_device_async(a, async)} | |
3600 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3601 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3602 | @item @emph{Interface}: @tab @code{subroutine acc_update_device_async(a, len, async)} | |
3603 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3604 | @item @tab @code{integer len} | |
3605 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3606 | @end multitable | |
3607 | ||
3608 | @item @emph{Reference}: | |
3609 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3610 | 3.2.24. | |
3611 | @end table | |
3612 | ||
3613 | ||
3614 | ||
3615 | @node acc_update_self | |
3616 | @section @code{acc_update_self} -- Update host memory from mapped device memory. | |
3617 | @table @asis | |
3618 | @item @emph{Description} | |
3619 | This function updates the host copy from the previously mapped device memory. | |
3620 | The host memory is specified with the host address @var{a} and a length of | |
3621 | @var{len} bytes. | |
3622 | ||
3623 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3624 | a contiguous array section. The second form @var{a} specifies a variable or | |
3625 | array element and @var{len} specifies the length in bytes. | |
3626 | ||
3627 | @item @emph{C/C++}: | |
3628 | @multitable @columnfractions .20 .80 | |
3629 | @item @emph{Prototype}: @tab @code{acc_update_self(h_void *a, size_t len);} | |
3630 | @item @emph{Prototype}: @tab @code{acc_update_self_async(h_void *a, size_t len, int async);} | |
3631 | @end multitable | |
3632 | ||
3633 | @item @emph{Fortran}: | |
3634 | @multitable @columnfractions .20 .80 | |
3635 | @item @emph{Interface}: @tab @code{subroutine acc_update_self(a)} | |
3636 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3637 | @item @emph{Interface}: @tab @code{subroutine acc_update_self(a, len)} | |
3638 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3639 | @item @tab @code{integer len} | |
3640 | @item @emph{Interface}: @tab @code{subroutine acc_update_self_async(a, async)} | |
3641 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3642 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3643 | @item @emph{Interface}: @tab @code{subroutine acc_update_self_async(a, len, async)} | |
3644 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3645 | @item @tab @code{integer len} | |
3646 | @item @tab @code{integer(acc_handle_kind) :: async} | |
3647 | @end multitable | |
3648 | ||
3649 | @item @emph{Reference}: | |
3650 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3651 | 3.2.25. | |
3652 | @end table | |
3653 | ||
3654 | ||
3655 | ||
3656 | @node acc_map_data | |
3657 | @section @code{acc_map_data} -- Map previously allocated device memory to host memory. | |
3658 | @table @asis | |
3659 | @item @emph{Description} | |
3660 | This function maps previously allocated device and host memory. The device | |
3661 | memory is specified with the device address @var{d}. The host memory is | |
3662 | specified with the host address @var{h} and a length of @var{len}. | |
3663 | ||
3664 | @item @emph{C/C++}: | |
3665 | @multitable @columnfractions .20 .80 | |
3666 | @item @emph{Prototype}: @tab @code{acc_map_data(h_void *h, d_void *d, size_t len);} | |
3667 | @end multitable | |
3668 | ||
3669 | @item @emph{Reference}: | |
3670 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3671 | 3.2.26. | |
3672 | @end table | |
3673 | ||
3674 | ||
3675 | ||
3676 | @node acc_unmap_data | |
3677 | @section @code{acc_unmap_data} -- Unmap device memory from host memory. | |
3678 | @table @asis | |
3679 | @item @emph{Description} | |
3680 | This function unmaps previously mapped device and host memory. The latter | |
3681 | specified by @var{h}. | |
3682 | ||
3683 | @item @emph{C/C++}: | |
3684 | @multitable @columnfractions .20 .80 | |
3685 | @item @emph{Prototype}: @tab @code{acc_unmap_data(h_void *h);} | |
3686 | @end multitable | |
3687 | ||
3688 | @item @emph{Reference}: | |
3689 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3690 | 3.2.27. | |
3691 | @end table | |
3692 | ||
3693 | ||
3694 | ||
3695 | @node acc_deviceptr | |
3696 | @section @code{acc_deviceptr} -- Get device pointer associated with specific host address. | |
3697 | @table @asis | |
3698 | @item @emph{Description} | |
3699 | This function returns the device address that has been mapped to the | |
3700 | host address specified by @var{h}. | |
3701 | ||
3702 | @item @emph{C/C++}: | |
3703 | @multitable @columnfractions .20 .80 | |
3704 | @item @emph{Prototype}: @tab @code{void *acc_deviceptr(h_void *h);} | |
3705 | @end multitable | |
3706 | ||
3707 | @item @emph{Reference}: | |
3708 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3709 | 3.2.28. | |
3710 | @end table | |
3711 | ||
3712 | ||
3713 | ||
3714 | @node acc_hostptr | |
3715 | @section @code{acc_hostptr} -- Get host pointer associated with specific device address. | |
3716 | @table @asis | |
3717 | @item @emph{Description} | |
3718 | This function returns the host address that has been mapped to the | |
3719 | device address specified by @var{d}. | |
3720 | ||
3721 | @item @emph{C/C++}: | |
3722 | @multitable @columnfractions .20 .80 | |
3723 | @item @emph{Prototype}: @tab @code{void *acc_hostptr(d_void *d);} | |
3724 | @end multitable | |
3725 | ||
3726 | @item @emph{Reference}: | |
3727 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3728 | 3.2.29. | |
3729 | @end table | |
3730 | ||
3731 | ||
3732 | ||
3733 | @node acc_is_present | |
3734 | @section @code{acc_is_present} -- Indicate whether host variable / array is present on device. | |
3735 | @table @asis | |
3736 | @item @emph{Description} | |
3737 | This function indicates whether the specified host address in @var{a} and a | |
3738 | length of @var{len} bytes is present on the device. In C/C++, a non-zero | |
3739 | value is returned to indicate the presence of the mapped memory on the | |
3740 | device. A zero is returned to indicate the memory is not mapped on the | |
3741 | device. | |
3742 | ||
3743 | In Fortran, two (2) forms are supported. In the first form, @var{a} specifies | |
3744 | a contiguous array section. The second form @var{a} specifies a variable or | |
3745 | array element and @var{len} specifies the length in bytes. If the host | |
3746 | memory is mapped to device memory, then a @code{true} is returned. Otherwise, | |
3747 | a @code{false} is return to indicate the mapped memory is not present. | |
3748 | ||
3749 | @item @emph{C/C++}: | |
3750 | @multitable @columnfractions .20 .80 | |
3751 | @item @emph{Prototype}: @tab @code{int acc_is_present(h_void *a, size_t len);} | |
3752 | @end multitable | |
3753 | ||
3754 | @item @emph{Fortran}: | |
3755 | @multitable @columnfractions .20 .80 | |
3756 | @item @emph{Interface}: @tab @code{function acc_is_present(a)} | |
3757 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3758 | @item @tab @code{logical acc_is_present} | |
3759 | @item @emph{Interface}: @tab @code{function acc_is_present(a, len)} | |
3760 | @item @tab @code{type, dimension(:[,:]...) :: a} | |
3761 | @item @tab @code{integer len} | |
3762 | @item @tab @code{logical acc_is_present} | |
3763 | @end multitable | |
3764 | ||
3765 | @item @emph{Reference}: | |
3766 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3767 | 3.2.30. | |
3768 | @end table | |
3769 | ||
3770 | ||
3771 | ||
3772 | @node acc_memcpy_to_device | |
3773 | @section @code{acc_memcpy_to_device} -- Copy host memory to device memory. | |
3774 | @table @asis | |
3775 | @item @emph{Description} | |
3776 | This function copies host memory specified by host address of @var{src} to | |
3777 | device memory specified by the device address @var{dest} for a length of | |
3778 | @var{bytes} bytes. | |
3779 | ||
3780 | @item @emph{C/C++}: | |
3781 | @multitable @columnfractions .20 .80 | |
3782 | @item @emph{Prototype}: @tab @code{acc_memcpy_to_device(d_void *dest, h_void *src, size_t bytes);} | |
3783 | @end multitable | |
3784 | ||
3785 | @item @emph{Reference}: | |
3786 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3787 | 3.2.31. | |
3788 | @end table | |
3789 | ||
3790 | ||
3791 | ||
3792 | @node acc_memcpy_from_device | |
3793 | @section @code{acc_memcpy_from_device} -- Copy device memory to host memory. | |
3794 | @table @asis | |
3795 | @item @emph{Description} | |
3796 | This function copies host memory specified by host address of @var{src} from | |
3797 | device memory specified by the device address @var{dest} for a length of | |
3798 | @var{bytes} bytes. | |
3799 | ||
3800 | @item @emph{C/C++}: | |
3801 | @multitable @columnfractions .20 .80 | |
3802 | @item @emph{Prototype}: @tab @code{acc_memcpy_from_device(d_void *dest, h_void *src, size_t bytes);} | |
3803 | @end multitable | |
3804 | ||
3805 | @item @emph{Reference}: | |
3806 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3807 | 3.2.32. | |
3808 | @end table | |
3809 | ||
3810 | ||
3811 | ||
3812 | @node acc_attach | |
3813 | @section @code{acc_attach} -- Let device pointer point to device-pointer target. | |
3814 | @table @asis | |
3815 | @item @emph{Description} | |
3816 | This function updates a pointer on the device from pointing to a host-pointer | |
3817 | address to pointing to the corresponding device data. | |
3818 | ||
3819 | @item @emph{C/C++}: | |
3820 | @multitable @columnfractions .20 .80 | |
3821 | @item @emph{Prototype}: @tab @code{acc_attach(h_void **ptr);} | |
3822 | @item @emph{Prototype}: @tab @code{acc_attach_async(h_void **ptr, int async);} | |
3823 | @end multitable | |
3824 | ||
3825 | @item @emph{Reference}: | |
3826 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3827 | 3.2.34. | |
3828 | @end table | |
3829 | ||
3830 | ||
3831 | ||
3832 | @node acc_detach | |
3833 | @section @code{acc_detach} -- Let device pointer point to host-pointer target. | |
3834 | @table @asis | |
3835 | @item @emph{Description} | |
3836 | This function updates a pointer on the device from pointing to a device-pointer | |
3837 | address to pointing to the corresponding host data. | |
3838 | ||
3839 | @item @emph{C/C++}: | |
3840 | @multitable @columnfractions .20 .80 | |
3841 | @item @emph{Prototype}: @tab @code{acc_detach(h_void **ptr);} | |
3842 | @item @emph{Prototype}: @tab @code{acc_detach_async(h_void **ptr, int async);} | |
3843 | @item @emph{Prototype}: @tab @code{acc_detach_finalize(h_void **ptr);} | |
3844 | @item @emph{Prototype}: @tab @code{acc_detach_finalize_async(h_void **ptr, int async);} | |
3845 | @end multitable | |
3846 | ||
3847 | @item @emph{Reference}: | |
3848 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3849 | 3.2.35. | |
3850 | @end table | |
3851 | ||
3852 | ||
3853 | ||
3854 | @node acc_get_current_cuda_device | |
3855 | @section @code{acc_get_current_cuda_device} -- Get CUDA device handle. | |
3856 | @table @asis | |
3857 | @item @emph{Description} | |
3858 | This function returns the CUDA device handle. This handle is the same | |
3859 | as used by the CUDA Runtime or Driver API's. | |
3860 | ||
3861 | @item @emph{C/C++}: | |
3862 | @multitable @columnfractions .20 .80 | |
3863 | @item @emph{Prototype}: @tab @code{void *acc_get_current_cuda_device(void);} | |
3864 | @end multitable | |
3865 | ||
3866 | @item @emph{Reference}: | |
3867 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3868 | A.2.1.1. | |
3869 | @end table | |
3870 | ||
3871 | ||
3872 | ||
3873 | @node acc_get_current_cuda_context | |
3874 | @section @code{acc_get_current_cuda_context} -- Get CUDA context handle. | |
3875 | @table @asis | |
3876 | @item @emph{Description} | |
3877 | This function returns the CUDA context handle. This handle is the same | |
3878 | as used by the CUDA Runtime or Driver API's. | |
3879 | ||
3880 | @item @emph{C/C++}: | |
3881 | @multitable @columnfractions .20 .80 | |
3882 | @item @emph{Prototype}: @tab @code{void *acc_get_current_cuda_context(void);} | |
3883 | @end multitable | |
3884 | ||
3885 | @item @emph{Reference}: | |
3886 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3887 | A.2.1.2. | |
3888 | @end table | |
3889 | ||
3890 | ||
3891 | ||
3892 | @node acc_get_cuda_stream | |
3893 | @section @code{acc_get_cuda_stream} -- Get CUDA stream handle. | |
3894 | @table @asis | |
3895 | @item @emph{Description} | |
3896 | This function returns the CUDA stream handle for the queue @var{async}. | |
3897 | This handle is the same as used by the CUDA Runtime or Driver API's. | |
3898 | ||
3899 | @item @emph{C/C++}: | |
3900 | @multitable @columnfractions .20 .80 | |
3901 | @item @emph{Prototype}: @tab @code{void *acc_get_cuda_stream(int async);} | |
3902 | @end multitable | |
3903 | ||
3904 | @item @emph{Reference}: | |
3905 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3906 | A.2.1.3. | |
3907 | @end table | |
3908 | ||
3909 | ||
3910 | ||
3911 | @node acc_set_cuda_stream | |
3912 | @section @code{acc_set_cuda_stream} -- Set CUDA stream handle. | |
3913 | @table @asis | |
3914 | @item @emph{Description} | |
3915 | This function associates the stream handle specified by @var{stream} with | |
3916 | the queue @var{async}. | |
3917 | ||
3918 | This cannot be used to change the stream handle associated with | |
3919 | @code{acc_async_sync}. | |
3920 | ||
3921 | The return value is not specified. | |
3922 | ||
3923 | @item @emph{C/C++}: | |
3924 | @multitable @columnfractions .20 .80 | |
3925 | @item @emph{Prototype}: @tab @code{int acc_set_cuda_stream(int async, void *stream);} | |
3926 | @end multitable | |
3927 | ||
3928 | @item @emph{Reference}: | |
3929 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3930 | A.2.1.4. | |
3931 | @end table | |
3932 | ||
3933 | ||
3934 | ||
3935 | @node acc_prof_register | |
3936 | @section @code{acc_prof_register} -- Register callbacks. | |
3937 | @table @asis | |
3938 | @item @emph{Description}: | |
3939 | This function registers callbacks. | |
3940 | ||
3941 | @item @emph{C/C++}: | |
3942 | @multitable @columnfractions .20 .80 | |
3943 | @item @emph{Prototype}: @tab @code{void acc_prof_register (acc_event_t, acc_prof_callback, acc_register_t);} | |
3944 | @end multitable | |
3945 | ||
3946 | @item @emph{See also}: | |
3947 | @ref{OpenACC Profiling Interface} | |
3948 | ||
3949 | @item @emph{Reference}: | |
3950 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3951 | 5.3. | |
3952 | @end table | |
3953 | ||
3954 | ||
3955 | ||
3956 | @node acc_prof_unregister | |
3957 | @section @code{acc_prof_unregister} -- Unregister callbacks. | |
3958 | @table @asis | |
3959 | @item @emph{Description}: | |
3960 | This function unregisters callbacks. | |
3961 | ||
3962 | @item @emph{C/C++}: | |
3963 | @multitable @columnfractions .20 .80 | |
3964 | @item @emph{Prototype}: @tab @code{void acc_prof_unregister (acc_event_t, acc_prof_callback, acc_register_t);} | |
3965 | @end multitable | |
3966 | ||
3967 | @item @emph{See also}: | |
3968 | @ref{OpenACC Profiling Interface} | |
3969 | ||
3970 | @item @emph{Reference}: | |
3971 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3972 | 5.3. | |
3973 | @end table | |
3974 | ||
3975 | ||
3976 | ||
3977 | @node acc_prof_lookup | |
3978 | @section @code{acc_prof_lookup} -- Obtain inquiry functions. | |
3979 | @table @asis | |
3980 | @item @emph{Description}: | |
3981 | Function to obtain inquiry functions. | |
3982 | ||
3983 | @item @emph{C/C++}: | |
3984 | @multitable @columnfractions .20 .80 | |
3985 | @item @emph{Prototype}: @tab @code{acc_query_fn acc_prof_lookup (const char *);} | |
3986 | @end multitable | |
3987 | ||
3988 | @item @emph{See also}: | |
3989 | @ref{OpenACC Profiling Interface} | |
3990 | ||
3991 | @item @emph{Reference}: | |
3992 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
3993 | 5.3. | |
3994 | @end table | |
3995 | ||
3996 | ||
3997 | ||
3998 | @node acc_register_library | |
3999 | @section @code{acc_register_library} -- Library registration. | |
4000 | @table @asis | |
4001 | @item @emph{Description}: | |
4002 | Function for library registration. | |
4003 | ||
4004 | @item @emph{C/C++}: | |
4005 | @multitable @columnfractions .20 .80 | |
4006 | @item @emph{Prototype}: @tab @code{void acc_register_library (acc_prof_reg, acc_prof_reg, acc_prof_lookup_func);} | |
4007 | @end multitable | |
4008 | ||
4009 | @item @emph{See also}: | |
4010 | @ref{OpenACC Profiling Interface}, @ref{ACC_PROFLIB} | |
4011 | ||
4012 | @item @emph{Reference}: | |
4013 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
4014 | 5.3. | |
4015 | @end table | |
4016 | ||
4017 | ||
4018 | ||
4019 | @c --------------------------------------------------------------------- | |
4020 | @c OpenACC Environment Variables | |
4021 | @c --------------------------------------------------------------------- | |
4022 | ||
4023 | @node OpenACC Environment Variables | |
4024 | @chapter OpenACC Environment Variables | |
4025 | ||
4026 | The variables @env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM} | |
4027 | are defined by section 4 of the OpenACC specification in version 2.0. | |
4028 | The variable @env{ACC_PROFLIB} | |
4029 | is defined by section 4 of the OpenACC specification in version 2.6. | |
4030 | The variable @env{GCC_ACC_NOTIFY} is used for diagnostic purposes. | |
4031 | ||
4032 | @menu | |
4033 | * ACC_DEVICE_TYPE:: | |
4034 | * ACC_DEVICE_NUM:: | |
4035 | * ACC_PROFLIB:: | |
4036 | * GCC_ACC_NOTIFY:: | |
4037 | @end menu | |
4038 | ||
4039 | ||
4040 | ||
4041 | @node ACC_DEVICE_TYPE | |
4042 | @section @code{ACC_DEVICE_TYPE} | |
4043 | @table @asis | |
4044 | @item @emph{Reference}: | |
4045 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
4046 | 4.1. | |
4047 | @end table | |
4048 | ||
4049 | ||
4050 | ||
4051 | @node ACC_DEVICE_NUM | |
4052 | @section @code{ACC_DEVICE_NUM} | |
4053 | @table @asis | |
4054 | @item @emph{Reference}: | |
4055 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
4056 | 4.2. | |
4057 | @end table | |
4058 | ||
4059 | ||
4060 | ||
4061 | @node ACC_PROFLIB | |
4062 | @section @code{ACC_PROFLIB} | |
4063 | @table @asis | |
4064 | @item @emph{See also}: | |
4065 | @ref{acc_register_library}, @ref{OpenACC Profiling Interface} | |
4066 | ||
4067 | @item @emph{Reference}: | |
4068 | @uref{https://www.openacc.org, OpenACC specification v2.6}, section | |
4069 | 4.3. | |
4070 | @end table | |
4071 | ||
4072 | ||
4073 | ||
4074 | @node GCC_ACC_NOTIFY | |
4075 | @section @code{GCC_ACC_NOTIFY} | |
4076 | @table @asis | |
4077 | @item @emph{Description}: | |
4078 | Print debug information pertaining to the accelerator. | |
4079 | @end table | |
4080 | ||
4081 | ||
4082 | ||
4083 | @c --------------------------------------------------------------------- | |
4084 | @c CUDA Streams Usage | |
4085 | @c --------------------------------------------------------------------- | |
4086 | ||
4087 | @node CUDA Streams Usage | |
4088 | @chapter CUDA Streams Usage | |
4089 | ||
4090 | This applies to the @code{nvptx} plugin only. | |
4091 | ||
4092 | The library provides elements that perform asynchronous movement of | |
4093 | data and asynchronous operation of computing constructs. This | |
4094 | asynchronous functionality is implemented by making use of CUDA | |
4095 | streams@footnote{See "Stream Management" in "CUDA Driver API", | |
4096 | TRM-06703-001, Version 5.5, for additional information}. | |
4097 | ||
4098 | The primary means by that the asynchronous functionality is accessed | |
4099 | is through the use of those OpenACC directives which make use of the | |
4100 | @code{async} and @code{wait} clauses. When the @code{async} clause is | |
4101 | first used with a directive, it creates a CUDA stream. If an | |
4102 | @code{async-argument} is used with the @code{async} clause, then the | |
4103 | stream is associated with the specified @code{async-argument}. | |
4104 | ||
4105 | Following the creation of an association between a CUDA stream and the | |
4106 | @code{async-argument} of an @code{async} clause, both the @code{wait} | |
4107 | clause and the @code{wait} directive can be used. When either the | |
4108 | clause or directive is used after stream creation, it creates a | |
4109 | rendezvous point whereby execution waits until all operations | |
4110 | associated with the @code{async-argument}, that is, stream, have | |
4111 | completed. | |
4112 | ||
4113 | Normally, the management of the streams that are created as a result of | |
4114 | using the @code{async} clause, is done without any intervention by the | |
4115 | caller. This implies the association between the @code{async-argument} | |
4116 | and the CUDA stream will be maintained for the lifetime of the program. | |
4117 | However, this association can be changed through the use of the library | |
4118 | function @code{acc_set_cuda_stream}. When the function | |
4119 | @code{acc_set_cuda_stream} is called, the CUDA stream that was | |
4120 | originally associated with the @code{async} clause will be destroyed. | |
4121 | Caution should be taken when changing the association as subsequent | |
4122 | references to the @code{async-argument} refer to a different | |
4123 | CUDA stream. | |
4124 | ||
4125 | ||
4126 | ||
4127 | @c --------------------------------------------------------------------- | |
4128 | @c OpenACC Library Interoperability | |
4129 | @c --------------------------------------------------------------------- | |
4130 | ||
4131 | @node OpenACC Library Interoperability | |
4132 | @chapter OpenACC Library Interoperability | |
4133 | ||
4134 | @section Introduction | |
4135 | ||
4136 | The OpenACC library uses the CUDA Driver API, and may interact with | |
4137 | programs that use the Runtime library directly, or another library | |
4138 | based on the Runtime library, e.g., CUBLAS@footnote{See section 2.26, | |
4139 | "Interactions with the CUDA Driver API" in | |
4140 | "CUDA Runtime API", Version 5.5, and section 2.27, "VDPAU | |
4141 | Interoperability", in "CUDA Driver API", TRM-06703-001, Version 5.5, | |
4142 | for additional information on library interoperability.}. | |
4143 | This chapter describes the use cases and what changes are | |
4144 | required in order to use both the OpenACC library and the CUBLAS and Runtime | |
4145 | libraries within a program. | |
4146 | ||
4147 | @section First invocation: NVIDIA CUBLAS library API | |
4148 | ||
4149 | In this first use case (see below), a function in the CUBLAS library is called | |
4150 | prior to any of the functions in the OpenACC library. More specifically, the | |
4151 | function @code{cublasCreate()}. | |
4152 | ||
4153 | When invoked, the function initializes the library and allocates the | |
4154 | hardware resources on the host and the device on behalf of the caller. Once | |
4155 | the initialization and allocation has completed, a handle is returned to the | |
4156 | caller. The OpenACC library also requires initialization and allocation of | |
4157 | hardware resources. Since the CUBLAS library has already allocated the | |
4158 | hardware resources for the device, all that is left to do is to initialize | |
4159 | the OpenACC library and acquire the hardware resources on the host. | |
4160 | ||
4161 | Prior to calling the OpenACC function that initializes the library and | |
4162 | allocate the host hardware resources, you need to acquire the device number | |
4163 | that was allocated during the call to @code{cublasCreate()}. The invoking of the | |
4164 | runtime library function @code{cudaGetDevice()} accomplishes this. Once | |
4165 | acquired, the device number is passed along with the device type as | |
4166 | parameters to the OpenACC library function @code{acc_set_device_num()}. | |
4167 | ||
4168 | Once the call to @code{acc_set_device_num()} has completed, the OpenACC | |
4169 | library uses the context that was created during the call to | |
4170 | @code{cublasCreate()}. In other words, both libraries will be sharing the | |
4171 | same context. | |
4172 | ||
4173 | @smallexample | |
4174 | /* Create the handle */ | |
4175 | s = cublasCreate(&h); | |
4176 | if (s != CUBLAS_STATUS_SUCCESS) | |
4177 | @{ | |
4178 | fprintf(stderr, "cublasCreate failed %d\n", s); | |
4179 | exit(EXIT_FAILURE); | |
4180 | @} | |
4181 | ||
4182 | /* Get the device number */ | |
4183 | e = cudaGetDevice(&dev); | |
4184 | if (e != cudaSuccess) | |
4185 | @{ | |
4186 | fprintf(stderr, "cudaGetDevice failed %d\n", e); | |
4187 | exit(EXIT_FAILURE); | |
4188 | @} | |
4189 | ||
4190 | /* Initialize OpenACC library and use device 'dev' */ | |
4191 | acc_set_device_num(dev, acc_device_nvidia); | |
4192 | ||
4193 | @end smallexample | |
4194 | @center Use Case 1 | |
4195 | ||
4196 | @section First invocation: OpenACC library API | |
4197 | ||
4198 | In this second use case (see below), a function in the OpenACC library is | |
eda38850 | 4199 | called prior to any of the functions in the CUBLAS library. More specifically, |
d77de738 ML |
4200 | the function @code{acc_set_device_num()}. |
4201 | ||
4202 | In the use case presented here, the function @code{acc_set_device_num()} | |
4203 | is used to both initialize the OpenACC library and allocate the hardware | |
4204 | resources on the host and the device. In the call to the function, the | |
4205 | call parameters specify which device to use and what device | |
4206 | type to use, i.e., @code{acc_device_nvidia}. It should be noted that this | |
4207 | is but one method to initialize the OpenACC library and allocate the | |
4208 | appropriate hardware resources. Other methods are available through the | |
4209 | use of environment variables and these will be discussed in the next section. | |
4210 | ||
4211 | Once the call to @code{acc_set_device_num()} has completed, other OpenACC | |
4212 | functions can be called as seen with multiple calls being made to | |
4213 | @code{acc_copyin()}. In addition, calls can be made to functions in the | |
4214 | CUBLAS library. In the use case a call to @code{cublasCreate()} is made | |
4215 | subsequent to the calls to @code{acc_copyin()}. | |
4216 | As seen in the previous use case, a call to @code{cublasCreate()} | |
4217 | initializes the CUBLAS library and allocates the hardware resources on the | |
4218 | host and the device. However, since the device has already been allocated, | |
4219 | @code{cublasCreate()} will only initialize the CUBLAS library and allocate | |
4220 | the appropriate hardware resources on the host. The context that was created | |
4221 | as part of the OpenACC initialization is shared with the CUBLAS library, | |
4222 | similarly to the first use case. | |
4223 | ||
4224 | @smallexample | |
4225 | dev = 0; | |
4226 | ||
4227 | acc_set_device_num(dev, acc_device_nvidia); | |
4228 | ||
4229 | /* Copy the first set to the device */ | |
4230 | d_X = acc_copyin(&h_X[0], N * sizeof (float)); | |
4231 | if (d_X == NULL) | |
4232 | @{ | |
4233 | fprintf(stderr, "copyin error h_X\n"); | |
4234 | exit(EXIT_FAILURE); | |
4235 | @} | |
4236 | ||
4237 | /* Copy the second set to the device */ | |
4238 | d_Y = acc_copyin(&h_Y1[0], N * sizeof (float)); | |
4239 | if (d_Y == NULL) | |
4240 | @{ | |
4241 | fprintf(stderr, "copyin error h_Y1\n"); | |
4242 | exit(EXIT_FAILURE); | |
4243 | @} | |
4244 | ||
4245 | /* Create the handle */ | |
4246 | s = cublasCreate(&h); | |
4247 | if (s != CUBLAS_STATUS_SUCCESS) | |
4248 | @{ | |
4249 | fprintf(stderr, "cublasCreate failed %d\n", s); | |
4250 | exit(EXIT_FAILURE); | |
4251 | @} | |
4252 | ||
4253 | /* Perform saxpy using CUBLAS library function */ | |
4254 | s = cublasSaxpy(h, N, &alpha, d_X, 1, d_Y, 1); | |
4255 | if (s != CUBLAS_STATUS_SUCCESS) | |
4256 | @{ | |
4257 | fprintf(stderr, "cublasSaxpy failed %d\n", s); | |
4258 | exit(EXIT_FAILURE); | |
4259 | @} | |
4260 | ||
4261 | /* Copy the results from the device */ | |
4262 | acc_memcpy_from_device(&h_Y1[0], d_Y, N * sizeof (float)); | |
4263 | ||
4264 | @end smallexample | |
4265 | @center Use Case 2 | |
4266 | ||
4267 | @section OpenACC library and environment variables | |
4268 | ||
4269 | There are two environment variables associated with the OpenACC library | |
4270 | that may be used to control the device type and device number: | |
4271 | @env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM}, respectively. These two | |
4272 | environment variables can be used as an alternative to calling | |
4273 | @code{acc_set_device_num()}. As seen in the second use case, the device | |
4274 | type and device number were specified using @code{acc_set_device_num()}. | |
4275 | If however, the aforementioned environment variables were set, then the | |
4276 | call to @code{acc_set_device_num()} would not be required. | |
4277 | ||
4278 | ||
4279 | The use of the environment variables is only relevant when an OpenACC function | |
4280 | is called prior to a call to @code{cudaCreate()}. If @code{cudaCreate()} | |
4281 | is called prior to a call to an OpenACC function, then you must call | |
4282 | @code{acc_set_device_num()}@footnote{More complete information | |
4283 | about @env{ACC_DEVICE_TYPE} and @env{ACC_DEVICE_NUM} can be found in | |
4284 | sections 4.1 and 4.2 of the @uref{https://www.openacc.org, OpenACC} | |
4285 | Application Programming Interface”, Version 2.6.} | |
4286 | ||
4287 | ||
4288 | ||
4289 | @c --------------------------------------------------------------------- | |
4290 | @c OpenACC Profiling Interface | |
4291 | @c --------------------------------------------------------------------- | |
4292 | ||
4293 | @node OpenACC Profiling Interface | |
4294 | @chapter OpenACC Profiling Interface | |
4295 | ||
4296 | @section Implementation Status and Implementation-Defined Behavior | |
4297 | ||
4298 | We're implementing the OpenACC Profiling Interface as defined by the | |
4299 | OpenACC 2.6 specification. We're clarifying some aspects here as | |
4300 | @emph{implementation-defined behavior}, while they're still under | |
4301 | discussion within the OpenACC Technical Committee. | |
4302 | ||
4303 | This implementation is tuned to keep the performance impact as low as | |
4304 | possible for the (very common) case that the Profiling Interface is | |
4305 | not enabled. This is relevant, as the Profiling Interface affects all | |
4306 | the @emph{hot} code paths (in the target code, not in the offloaded | |
4307 | code). Users of the OpenACC Profiling Interface can be expected to | |
4308 | understand that performance will be impacted to some degree once the | |
4309 | Profiling Interface has gotten enabled: for example, because of the | |
4310 | @emph{runtime} (libgomp) calling into a third-party @emph{library} for | |
4311 | every event that has been registered. | |
4312 | ||
4313 | We're not yet accounting for the fact that @cite{OpenACC events may | |
4314 | occur during event processing}. | |
4315 | We just handle one case specially, as required by CUDA 9.0 | |
4316 | @command{nvprof}, that @code{acc_get_device_type} | |
4317 | (@ref{acc_get_device_type})) may be called from | |
4318 | @code{acc_ev_device_init_start}, @code{acc_ev_device_init_end} | |
4319 | callbacks. | |
4320 | ||
4321 | We're not yet implementing initialization via a | |
4322 | @code{acc_register_library} function that is either statically linked | |
4323 | in, or dynamically via @env{LD_PRELOAD}. | |
4324 | Initialization via @code{acc_register_library} functions dynamically | |
4325 | loaded via the @env{ACC_PROFLIB} environment variable does work, as | |
4326 | does directly calling @code{acc_prof_register}, | |
4327 | @code{acc_prof_unregister}, @code{acc_prof_lookup}. | |
4328 | ||
4329 | As currently there are no inquiry functions defined, calls to | |
4330 | @code{acc_prof_lookup} will always return @code{NULL}. | |
4331 | ||
4332 | There aren't separate @emph{start}, @emph{stop} events defined for the | |
4333 | event types @code{acc_ev_create}, @code{acc_ev_delete}, | |
4334 | @code{acc_ev_alloc}, @code{acc_ev_free}. It's not clear if these | |
4335 | should be triggered before or after the actual device-specific call is | |
4336 | made. We trigger them after. | |
4337 | ||
4338 | Remarks about data provided to callbacks: | |
4339 | ||
4340 | @table @asis | |
4341 | ||
4342 | @item @code{acc_prof_info.event_type} | |
4343 | It's not clear if for @emph{nested} event callbacks (for example, | |
4344 | @code{acc_ev_enqueue_launch_start} as part of a parent compute | |
4345 | construct), this should be set for the nested event | |
4346 | (@code{acc_ev_enqueue_launch_start}), or if the value of the parent | |
4347 | construct should remain (@code{acc_ev_compute_construct_start}). In | |
4348 | this implementation, the value will generally correspond to the | |
4349 | innermost nested event type. | |
4350 | ||
4351 | @item @code{acc_prof_info.device_type} | |
4352 | @itemize | |
4353 | ||
4354 | @item | |
4355 | For @code{acc_ev_compute_construct_start}, and in presence of an | |
4356 | @code{if} clause with @emph{false} argument, this will still refer to | |
4357 | the offloading device type. | |
4358 | It's not clear if that's the expected behavior. | |
4359 | ||
4360 | @item | |
4361 | Complementary to the item before, for | |
4362 | @code{acc_ev_compute_construct_end}, this is set to | |
4363 | @code{acc_device_host} in presence of an @code{if} clause with | |
4364 | @emph{false} argument. | |
4365 | It's not clear if that's the expected behavior. | |
4366 | ||
4367 | @end itemize | |
4368 | ||
4369 | @item @code{acc_prof_info.thread_id} | |
4370 | Always @code{-1}; not yet implemented. | |
4371 | ||
4372 | @item @code{acc_prof_info.async} | |
4373 | @itemize | |
4374 | ||
4375 | @item | |
4376 | Not yet implemented correctly for | |
4377 | @code{acc_ev_compute_construct_start}. | |
4378 | ||
4379 | @item | |
4380 | In a compute construct, for host-fallback | |
4381 | execution/@code{acc_device_host} it will always be | |
4382 | @code{acc_async_sync}. | |
4383 | It's not clear if that's the expected behavior. | |
4384 | ||
4385 | @item | |
4386 | For @code{acc_ev_device_init_start} and @code{acc_ev_device_init_end}, | |
4387 | it will always be @code{acc_async_sync}. | |
4388 | It's not clear if that's the expected behavior. | |
4389 | ||
4390 | @end itemize | |
4391 | ||
4392 | @item @code{acc_prof_info.async_queue} | |
4393 | There is no @cite{limited number of asynchronous queues} in libgomp. | |
4394 | This will always have the same value as @code{acc_prof_info.async}. | |
4395 | ||
4396 | @item @code{acc_prof_info.src_file} | |
4397 | Always @code{NULL}; not yet implemented. | |
4398 | ||
4399 | @item @code{acc_prof_info.func_name} | |
4400 | Always @code{NULL}; not yet implemented. | |
4401 | ||
4402 | @item @code{acc_prof_info.line_no} | |
4403 | Always @code{-1}; not yet implemented. | |
4404 | ||
4405 | @item @code{acc_prof_info.end_line_no} | |
4406 | Always @code{-1}; not yet implemented. | |
4407 | ||
4408 | @item @code{acc_prof_info.func_line_no} | |
4409 | Always @code{-1}; not yet implemented. | |
4410 | ||
4411 | @item @code{acc_prof_info.func_end_line_no} | |
4412 | Always @code{-1}; not yet implemented. | |
4413 | ||
4414 | @item @code{acc_event_info.event_type}, @code{acc_event_info.*.event_type} | |
4415 | Relating to @code{acc_prof_info.event_type} discussed above, in this | |
4416 | implementation, this will always be the same value as | |
4417 | @code{acc_prof_info.event_type}. | |
4418 | ||
4419 | @item @code{acc_event_info.*.parent_construct} | |
4420 | @itemize | |
4421 | ||
4422 | @item | |
4423 | Will be @code{acc_construct_parallel} for all OpenACC compute | |
4424 | constructs as well as many OpenACC Runtime API calls; should be the | |
4425 | one matching the actual construct, or | |
4426 | @code{acc_construct_runtime_api}, respectively. | |
4427 | ||
4428 | @item | |
4429 | Will be @code{acc_construct_enter_data} or | |
4430 | @code{acc_construct_exit_data} when processing variable mappings | |
4431 | specified in OpenACC @emph{declare} directives; should be | |
4432 | @code{acc_construct_declare}. | |
4433 | ||
4434 | @item | |
4435 | For implicit @code{acc_ev_device_init_start}, | |
4436 | @code{acc_ev_device_init_end}, and explicit as well as implicit | |
4437 | @code{acc_ev_alloc}, @code{acc_ev_free}, | |
4438 | @code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end}, | |
4439 | @code{acc_ev_enqueue_download_start}, and | |
4440 | @code{acc_ev_enqueue_download_end}, will be | |
4441 | @code{acc_construct_parallel}; should reflect the real parent | |
4442 | construct. | |
4443 | ||
4444 | @end itemize | |
4445 | ||
4446 | @item @code{acc_event_info.*.implicit} | |
4447 | For @code{acc_ev_alloc}, @code{acc_ev_free}, | |
4448 | @code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end}, | |
4449 | @code{acc_ev_enqueue_download_start}, and | |
4450 | @code{acc_ev_enqueue_download_end}, this currently will be @code{1} | |
4451 | also for explicit usage. | |
4452 | ||
4453 | @item @code{acc_event_info.data_event.var_name} | |
4454 | Always @code{NULL}; not yet implemented. | |
4455 | ||
4456 | @item @code{acc_event_info.data_event.host_ptr} | |
4457 | For @code{acc_ev_alloc}, and @code{acc_ev_free}, this is always | |
4458 | @code{NULL}. | |
4459 | ||
4460 | @item @code{typedef union acc_api_info} | |
4461 | @dots{} as printed in @cite{5.2.3. Third Argument: API-Specific | |
4462 | Information}. This should obviously be @code{typedef @emph{struct} | |
4463 | acc_api_info}. | |
4464 | ||
4465 | @item @code{acc_api_info.device_api} | |
4466 | Possibly not yet implemented correctly for | |
4467 | @code{acc_ev_compute_construct_start}, | |
4468 | @code{acc_ev_device_init_start}, @code{acc_ev_device_init_end}: | |
4469 | will always be @code{acc_device_api_none} for these event types. | |
4470 | For @code{acc_ev_enter_data_start}, it will be | |
4471 | @code{acc_device_api_none} in some cases. | |
4472 | ||
4473 | @item @code{acc_api_info.device_type} | |
4474 | Always the same as @code{acc_prof_info.device_type}. | |
4475 | ||
4476 | @item @code{acc_api_info.vendor} | |
4477 | Always @code{-1}; not yet implemented. | |
4478 | ||
4479 | @item @code{acc_api_info.device_handle} | |
4480 | Always @code{NULL}; not yet implemented. | |
4481 | ||
4482 | @item @code{acc_api_info.context_handle} | |
4483 | Always @code{NULL}; not yet implemented. | |
4484 | ||
4485 | @item @code{acc_api_info.async_handle} | |
4486 | Always @code{NULL}; not yet implemented. | |
4487 | ||
4488 | @end table | |
4489 | ||
4490 | Remarks about certain event types: | |
4491 | ||
4492 | @table @asis | |
4493 | ||
4494 | @item @code{acc_ev_device_init_start}, @code{acc_ev_device_init_end} | |
4495 | @itemize | |
4496 | ||
4497 | @item | |
4498 | @c See 'DEVICE_INIT_INSIDE_COMPUTE_CONSTRUCT' in | |
4499 | @c 'libgomp.oacc-c-c++-common/acc_prof-kernels-1.c', | |
4500 | @c 'libgomp.oacc-c-c++-common/acc_prof-parallel-1.c'. | |
4501 | When a compute construct triggers implicit | |
4502 | @code{acc_ev_device_init_start} and @code{acc_ev_device_init_end} | |
4503 | events, they currently aren't @emph{nested within} the corresponding | |
4504 | @code{acc_ev_compute_construct_start} and | |
4505 | @code{acc_ev_compute_construct_end}, but they're currently observed | |
4506 | @emph{before} @code{acc_ev_compute_construct_start}. | |
4507 | It's not clear what to do: the standard asks us provide a lot of | |
4508 | details to the @code{acc_ev_compute_construct_start} callback, without | |
4509 | (implicitly) initializing a device before? | |
4510 | ||
4511 | @item | |
4512 | Callbacks for these event types will not be invoked for calls to the | |
4513 | @code{acc_set_device_type} and @code{acc_set_device_num} functions. | |
4514 | It's not clear if they should be. | |
4515 | ||
4516 | @end itemize | |
4517 | ||
4518 | @item @code{acc_ev_enter_data_start}, @code{acc_ev_enter_data_end}, @code{acc_ev_exit_data_start}, @code{acc_ev_exit_data_end} | |
4519 | @itemize | |
4520 | ||
4521 | @item | |
4522 | Callbacks for these event types will also be invoked for OpenACC | |
4523 | @emph{host_data} constructs. | |
4524 | It's not clear if they should be. | |
4525 | ||
4526 | @item | |
4527 | Callbacks for these event types will also be invoked when processing | |
4528 | variable mappings specified in OpenACC @emph{declare} directives. | |
4529 | It's not clear if they should be. | |
4530 | ||
4531 | @end itemize | |
4532 | ||
4533 | @end table | |
4534 | ||
4535 | Callbacks for the following event types will be invoked, but dispatch | |
4536 | and information provided therein has not yet been thoroughly reviewed: | |
4537 | ||
4538 | @itemize | |
4539 | @item @code{acc_ev_alloc} | |
4540 | @item @code{acc_ev_free} | |
4541 | @item @code{acc_ev_update_start}, @code{acc_ev_update_end} | |
4542 | @item @code{acc_ev_enqueue_upload_start}, @code{acc_ev_enqueue_upload_end} | |
4543 | @item @code{acc_ev_enqueue_download_start}, @code{acc_ev_enqueue_download_end} | |
4544 | @end itemize | |
4545 | ||
4546 | During device initialization, and finalization, respectively, | |
4547 | callbacks for the following event types will not yet be invoked: | |
4548 | ||
4549 | @itemize | |
4550 | @item @code{acc_ev_alloc} | |
4551 | @item @code{acc_ev_free} | |
4552 | @end itemize | |
4553 | ||
4554 | Callbacks for the following event types have not yet been implemented, | |
4555 | so currently won't be invoked: | |
4556 | ||
4557 | @itemize | |
4558 | @item @code{acc_ev_device_shutdown_start}, @code{acc_ev_device_shutdown_end} | |
4559 | @item @code{acc_ev_runtime_shutdown} | |
4560 | @item @code{acc_ev_create}, @code{acc_ev_delete} | |
4561 | @item @code{acc_ev_wait_start}, @code{acc_ev_wait_end} | |
4562 | @end itemize | |
4563 | ||
4564 | For the following runtime library functions, not all expected | |
4565 | callbacks will be invoked (mostly concerning implicit device | |
4566 | initialization): | |
4567 | ||
4568 | @itemize | |
4569 | @item @code{acc_get_num_devices} | |
4570 | @item @code{acc_set_device_type} | |
4571 | @item @code{acc_get_device_type} | |
4572 | @item @code{acc_set_device_num} | |
4573 | @item @code{acc_get_device_num} | |
4574 | @item @code{acc_init} | |
4575 | @item @code{acc_shutdown} | |
4576 | @end itemize | |
4577 | ||
4578 | Aside from implicit device initialization, for the following runtime | |
4579 | library functions, no callbacks will be invoked for shared-memory | |
4580 | offloading devices (it's not clear if they should be): | |
4581 | ||
4582 | @itemize | |
4583 | @item @code{acc_malloc} | |
4584 | @item @code{acc_free} | |
4585 | @item @code{acc_copyin}, @code{acc_present_or_copyin}, @code{acc_copyin_async} | |
4586 | @item @code{acc_create}, @code{acc_present_or_create}, @code{acc_create_async} | |
4587 | @item @code{acc_copyout}, @code{acc_copyout_async}, @code{acc_copyout_finalize}, @code{acc_copyout_finalize_async} | |
4588 | @item @code{acc_delete}, @code{acc_delete_async}, @code{acc_delete_finalize}, @code{acc_delete_finalize_async} | |
4589 | @item @code{acc_update_device}, @code{acc_update_device_async} | |
4590 | @item @code{acc_update_self}, @code{acc_update_self_async} | |
4591 | @item @code{acc_map_data}, @code{acc_unmap_data} | |
4592 | @item @code{acc_memcpy_to_device}, @code{acc_memcpy_to_device_async} | |
4593 | @item @code{acc_memcpy_from_device}, @code{acc_memcpy_from_device_async} | |
4594 | @end itemize | |
4595 | ||
4596 | @c --------------------------------------------------------------------- | |
4597 | @c OpenMP-Implementation Specifics | |
4598 | @c --------------------------------------------------------------------- | |
4599 | ||
4600 | @node OpenMP-Implementation Specifics | |
4601 | @chapter OpenMP-Implementation Specifics | |
4602 | ||
4603 | @menu | |
2cd0689a | 4604 | * Implementation-defined ICV Initialization:: |
d77de738 | 4605 | * OpenMP Context Selectors:: |
450b05ce | 4606 | * Memory allocation:: |
d77de738 ML |
4607 | @end menu |
4608 | ||
2cd0689a TB |
4609 | @node Implementation-defined ICV Initialization |
4610 | @section Implementation-defined ICV Initialization | |
4611 | @cindex Implementation specific setting | |
4612 | ||
4613 | @multitable @columnfractions .30 .70 | |
4614 | @item @var{affinity-format-var} @tab See @ref{OMP_AFFINITY_FORMAT}. | |
4615 | @item @var{def-allocator-var} @tab See @ref{OMP_ALLOCATOR}. | |
4616 | @item @var{max-active-levels-var} @tab See @ref{OMP_MAX_ACTIVE_LEVELS}. | |
4617 | @item @var{dyn-var} @tab See @ref{OMP_DYNAMIC}. | |
4618 | @item @var{nthreads-var} @tab See @code{OMP_NUM_THREADS}. | |
4619 | @item @var{num-devices-var} @tab Number of non-host devices found | |
4620 | by GCC's run-time library | |
4621 | @item @var{num-procs-var} @tab The number of CPU cores on the | |
4622 | initial device, except that affinity settings might lead to a | |
4623 | smaller number. On non-host devices, the value of the | |
4624 | @var{nthreads-var} ICV. | |
4625 | @item @var{place-partition-var} @tab See @ref{OMP_PLACES}. | |
4626 | @item @var{run-sched-var} @tab See @ref{OMP_SCHEDULE}. | |
4627 | @item @var{stacksize-var} @tab See @ref{OMP_STACKSIZE}. | |
4628 | @item @var{thread-limit-var} @tab See @ref{OMP_TEAMS_THREAD_LIMIT} | |
4629 | @item @var{wait-policy-var} @tab See @ref{OMP_WAIT_POLICY} and | |
4630 | @ref{GOMP_SPINCOUNT} | |
4631 | @end multitable | |
4632 | ||
d77de738 ML |
4633 | @node OpenMP Context Selectors |
4634 | @section OpenMP Context Selectors | |
4635 | ||
4636 | @code{vendor} is always @code{gnu}. References are to the GCC manual. | |
4637 | ||
4638 | @multitable @columnfractions .60 .10 .25 | |
4639 | @headitem @code{arch} @tab @code{kind} @tab @code{isa} | |
4640 | @item @code{x86}, @code{x86_64}, @code{i386}, @code{i486}, | |
4641 | @code{i586}, @code{i686}, @code{ia32} | |
4642 | @tab @code{host} | |
4643 | @tab See @code{-m...} flags in ``x86 Options'' (without @code{-m}) | |
4644 | @item @code{amdgcn}, @code{gcn} | |
4645 | @tab @code{gpu} | |
e0b95c2e TB |
4646 | @tab See @code{-march=} in ``AMD GCN Options''@footnote{Additionally, |
4647 | @code{gfx803} is supported as an alias for @code{fiji}.} | |
d77de738 ML |
4648 | @item @code{nvptx} |
4649 | @tab @code{gpu} | |
4650 | @tab See @code{-march=} in ``Nvidia PTX Options'' | |
4651 | @end multitable | |
4652 | ||
450b05ce TB |
4653 | @node Memory allocation |
4654 | @section Memory allocation | |
d77de738 | 4655 | |
a85a106c TB |
4656 | For the available predefined allocators and, as applicable, their associated |
4657 | predefined memory spaces and for the available traits and their default values, | |
4658 | see @ref{OMP_ALLOCATOR}. Predefined allocators without an associated memory | |
4659 | space use the @code{omp_default_mem_space} memory space. | |
4660 | ||
8c2fc744 TB |
4661 | For the memory spaces, the following applies: |
4662 | @itemize | |
4663 | @item @code{omp_default_mem_space} is supported | |
4664 | @item @code{omp_const_mem_space} maps to @code{omp_default_mem_space} | |
4665 | @item @code{omp_low_lat_mem_space} maps to @code{omp_default_mem_space} | |
4666 | @item @code{omp_large_cap_mem_space} maps to @code{omp_default_mem_space}, | |
4667 | unless the memkind library is available | |
4668 | @item @code{omp_high_bw_mem_space} maps to @code{omp_default_mem_space}, | |
4669 | unless the memkind library is available | |
4670 | @end itemize | |
4671 | ||
d77de738 ML |
4672 | On Linux systems, where the @uref{https://github.com/memkind/memkind, memkind |
4673 | library} (@code{libmemkind.so.0}) is available at runtime, it is used when | |
4674 | creating memory allocators requesting | |
4675 | ||
4676 | @itemize | |
4677 | @item the memory space @code{omp_high_bw_mem_space} | |
4678 | @item the memory space @code{omp_large_cap_mem_space} | |
450b05ce | 4679 | @item the @code{partition} trait @code{interleaved}; note that for |
8c2fc744 | 4680 | @code{omp_large_cap_mem_space} the allocation will not be interleaved |
d77de738 ML |
4681 | @end itemize |
4682 | ||
450b05ce TB |
4683 | On Linux systems, where the @uref{https://github.com/numactl/numactl, numa |
4684 | library} (@code{libnuma.so.1}) is available at runtime, it used when creating | |
4685 | memory allocators requesting | |
4686 | ||
4687 | @itemize | |
4688 | @item the @code{partition} trait @code{nearest}, except when both the | |
4689 | libmemkind library is available and the memory space is either | |
4690 | @code{omp_large_cap_mem_space} or @code{omp_high_bw_mem_space} | |
4691 | @end itemize | |
4692 | ||
4693 | Note that the numa library will round up the allocation size to a multiple of | |
4694 | the system page size; therefore, consider using it only with large data or | |
4695 | by sharing allocations via the @code{pool_size} trait. Furthermore, the Linux | |
4696 | kernel does not guarantee that an allocation will always be on the nearest NUMA | |
4697 | node nor that after reallocation the same node will be used. Note additionally | |
4698 | that, on Linux, the default setting of the memory placement policy is to use the | |
4699 | current node; therefore, unless the memory placement policy has been overridden, | |
4700 | the @code{partition} trait @code{environment} (the default) will be effectively | |
4701 | a @code{nearest} allocation. | |
4702 | ||
a85a106c | 4703 | Additional notes regarding the traits: |
8c2fc744 TB |
4704 | @itemize |
4705 | @item The @code{pinned} trait is unsupported. | |
a85a106c TB |
4706 | @item The default for the @code{pool_size} trait is no pool and for every |
4707 | (re)allocation the associated library routine is called, which might | |
4708 | internally use a memory pool. | |
8c2fc744 TB |
4709 | @item For the @code{partition} trait, the partition part size will be the same |
4710 | as the requested size (i.e. @code{interleaved} or @code{blocked} has no | |
4711 | effect), except for @code{interleaved} when the memkind library is | |
450b05ce TB |
4712 | available. Furthermore, for @code{nearest} and unless the numa library |
4713 | is available, the memory might not be on the same NUMA node as thread | |
4714 | that allocated the memory; on Linux, this is in particular the case when | |
4715 | the memory placement policy is set to preferred. | |
8c2fc744 TB |
4716 | @item The @code{access} trait has no effect such that memory is always |
4717 | accessible by all threads. | |
4718 | @item The @code{sync_hint} trait has no effect. | |
4719 | @end itemize | |
d77de738 ML |
4720 | |
4721 | @c --------------------------------------------------------------------- | |
4722 | @c Offload-Target Specifics | |
4723 | @c --------------------------------------------------------------------- | |
4724 | ||
4725 | @node Offload-Target Specifics | |
4726 | @chapter Offload-Target Specifics | |
4727 | ||
4728 | The following sections present notes on the offload-target specifics | |
4729 | ||
4730 | @menu | |
4731 | * AMD Radeon:: | |
4732 | * nvptx:: | |
4733 | @end menu | |
4734 | ||
4735 | @node AMD Radeon | |
4736 | @section AMD Radeon (GCN) | |
4737 | ||
4738 | On the hardware side, there is the hierarchy (fine to coarse): | |
4739 | @itemize | |
4740 | @item work item (thread) | |
4741 | @item wavefront | |
4742 | @item work group | |
81476bc4 | 4743 | @item compute unit (CU) |
d77de738 ML |
4744 | @end itemize |
4745 | ||
4746 | All OpenMP and OpenACC levels are used, i.e. | |
4747 | @itemize | |
4748 | @item OpenMP's simd and OpenACC's vector map to work items (thread) | |
4749 | @item OpenMP's threads (``parallel'') and OpenACC's workers map | |
4750 | to wavefronts | |
4751 | @item OpenMP's teams and OpenACC's gang use a threadpool with the | |
4752 | size of the number of teams or gangs, respectively. | |
4753 | @end itemize | |
4754 | ||
4755 | The used sizes are | |
4756 | @itemize | |
4757 | @item Number of teams is the specified @code{num_teams} (OpenMP) or | |
81476bc4 MV |
4758 | @code{num_gangs} (OpenACC) or otherwise the number of CU. It is limited |
4759 | by two times the number of CU. | |
d77de738 ML |
4760 | @item Number of wavefronts is 4 for gfx900 and 16 otherwise; |
4761 | @code{num_threads} (OpenMP) and @code{num_workers} (OpenACC) | |
4762 | overrides this if smaller. | |
4763 | @item The wavefront has 102 scalars and 64 vectors | |
4764 | @item Number of workitems is always 64 | |
4765 | @item The hardware permits maximally 40 workgroups/CU and | |
4766 | 16 wavefronts/workgroup up to a limit of 40 wavefronts in total per CU. | |
4767 | @item 80 scalars registers and 24 vector registers in non-kernel functions | |
4768 | (the chosen procedure-calling API). | |
4769 | @item For the kernel itself: as many as register pressure demands (number of | |
4770 | teams and number of threads, scaled down if registers are exhausted) | |
4771 | @end itemize | |
4772 | ||
4773 | The implementation remark: | |
4774 | @itemize | |
4775 | @item I/O within OpenMP target regions and OpenACC parallel/kernels is supported | |
4776 | using the C library @code{printf} functions and the Fortran | |
4777 | @code{print}/@code{write} statements. | |
243fa488 | 4778 | @item Reverse offload regions (i.e. @code{target} regions with |
f84fdb13 TB |
4779 | @code{device(ancestor:1)}) are processed serially per @code{target} region |
4780 | such that the next reverse offload region is only executed after the previous | |
4781 | one returned. | |
f1af7d65 | 4782 | @item OpenMP code that has a @code{requires} directive with |
f84fdb13 TB |
4783 | @code{unified_shared_memory} will remove any GCN device from the list of |
4784 | available devices (``host fallback''). | |
2e3dd14d TB |
4785 | @item The available stack size can be changed using the @code{GCN_STACK_SIZE} |
4786 | environment variable; the default is 32 kiB per thread. | |
d77de738 ML |
4787 | @end itemize |
4788 | ||
4789 | ||
4790 | ||
4791 | @node nvptx | |
4792 | @section nvptx | |
4793 | ||
4794 | On the hardware side, there is the hierarchy (fine to coarse): | |
4795 | @itemize | |
4796 | @item thread | |
4797 | @item warp | |
4798 | @item thread block | |
4799 | @item streaming multiprocessor | |
4800 | @end itemize | |
4801 | ||
4802 | All OpenMP and OpenACC levels are used, i.e. | |
4803 | @itemize | |
4804 | @item OpenMP's simd and OpenACC's vector map to threads | |
4805 | @item OpenMP's threads (``parallel'') and OpenACC's workers map to warps | |
4806 | @item OpenMP's teams and OpenACC's gang use a threadpool with the | |
4807 | size of the number of teams or gangs, respectively. | |
4808 | @end itemize | |
4809 | ||
4810 | The used sizes are | |
4811 | @itemize | |
4812 | @item The @code{warp_size} is always 32 | |
4813 | @item CUDA kernel launched: @code{dim=@{#teams,1,1@}, blocks=@{#threads,warp_size,1@}}. | |
81476bc4 MV |
4814 | @item The number of teams is limited by the number of blocks the device can |
4815 | host simultaneously. | |
d77de738 ML |
4816 | @end itemize |
4817 | ||
4818 | Additional information can be obtained by setting the environment variable to | |
4819 | @code{GOMP_DEBUG=1} (very verbose; grep for @code{kernel.*launch} for launch | |
4820 | parameters). | |
4821 | ||
4822 | GCC generates generic PTX ISA code, which is just-in-time compiled by CUDA, | |
4823 | which caches the JIT in the user's directory (see CUDA documentation; can be | |
4824 | tuned by the environment variables @code{CUDA_CACHE_@{DISABLE,MAXSIZE,PATH@}}. | |
4825 | ||
4826 | Note: While PTX ISA is generic, the @code{-mptx=} and @code{-march=} commandline | |
eda38850 | 4827 | options still affect the used PTX ISA code and, thus, the requirements on |
d77de738 ML |
4828 | CUDA version and hardware. |
4829 | ||
4830 | The implementation remark: | |
4831 | @itemize | |
4832 | @item I/O within OpenMP target regions and OpenACC parallel/kernels is supported | |
4833 | using the C library @code{printf} functions. Note that the Fortran | |
4834 | @code{print}/@code{write} statements are not supported, yet. | |
4835 | @item Compilation OpenMP code that contains @code{requires reverse_offload} | |
4836 | requires at least @code{-march=sm_35}, compiling for @code{-march=sm_30} | |
4837 | is not supported. | |
eda38850 TB |
4838 | @item For code containing reverse offload (i.e. @code{target} regions with |
4839 | @code{device(ancestor:1)}), there is a slight performance penalty | |
4840 | for @emph{all} target regions, consisting mostly of shutdown delay | |
4841 | Per device, reverse offload regions are processed serially such that | |
4842 | the next reverse offload region is only executed after the previous | |
4843 | one returned. | |
f1af7d65 TB |
4844 | @item OpenMP code that has a @code{requires} directive with |
4845 | @code{unified_shared_memory} will remove any nvptx device from the | |
eda38850 | 4846 | list of available devices (``host fallback''). |
2cd0689a TB |
4847 | @item The default per-warp stack size is 128 kiB; see also @code{-msoft-stack} |
4848 | in the GCC manual. | |
d77de738 ML |
4849 | @end itemize |
4850 | ||
4851 | ||
4852 | @c --------------------------------------------------------------------- | |
4853 | @c The libgomp ABI | |
4854 | @c --------------------------------------------------------------------- | |
4855 | ||
4856 | @node The libgomp ABI | |
4857 | @chapter The libgomp ABI | |
4858 | ||
4859 | The following sections present notes on the external ABI as | |
4860 | presented by libgomp. Only maintainers should need them. | |
4861 | ||
4862 | @menu | |
4863 | * Implementing MASTER construct:: | |
4864 | * Implementing CRITICAL construct:: | |
4865 | * Implementing ATOMIC construct:: | |
4866 | * Implementing FLUSH construct:: | |
4867 | * Implementing BARRIER construct:: | |
4868 | * Implementing THREADPRIVATE construct:: | |
4869 | * Implementing PRIVATE clause:: | |
4870 | * Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses:: | |
4871 | * Implementing REDUCTION clause:: | |
4872 | * Implementing PARALLEL construct:: | |
4873 | * Implementing FOR construct:: | |
4874 | * Implementing ORDERED construct:: | |
4875 | * Implementing SECTIONS construct:: | |
4876 | * Implementing SINGLE construct:: | |
4877 | * Implementing OpenACC's PARALLEL construct:: | |
4878 | @end menu | |
4879 | ||
4880 | ||
4881 | @node Implementing MASTER construct | |
4882 | @section Implementing MASTER construct | |
4883 | ||
4884 | @smallexample | |
4885 | if (omp_get_thread_num () == 0) | |
4886 | block | |
4887 | @end smallexample | |
4888 | ||
4889 | Alternately, we generate two copies of the parallel subfunction | |
4890 | and only include this in the version run by the primary thread. | |
4891 | Surely this is not worthwhile though... | |
4892 | ||
4893 | ||
4894 | ||
4895 | @node Implementing CRITICAL construct | |
4896 | @section Implementing CRITICAL construct | |
4897 | ||
4898 | Without a specified name, | |
4899 | ||
4900 | @smallexample | |
4901 | void GOMP_critical_start (void); | |
4902 | void GOMP_critical_end (void); | |
4903 | @end smallexample | |
4904 | ||
4905 | so that we don't get COPY relocations from libgomp to the main | |
4906 | application. | |
4907 | ||
4908 | With a specified name, use omp_set_lock and omp_unset_lock with | |
4909 | name being transformed into a variable declared like | |
4910 | ||
4911 | @smallexample | |
4912 | omp_lock_t gomp_critical_user_<name> __attribute__((common)) | |
4913 | @end smallexample | |
4914 | ||
4915 | Ideally the ABI would specify that all zero is a valid unlocked | |
4916 | state, and so we wouldn't need to initialize this at | |
4917 | startup. | |
4918 | ||
4919 | ||
4920 | ||
4921 | @node Implementing ATOMIC construct | |
4922 | @section Implementing ATOMIC construct | |
4923 | ||
4924 | The target should implement the @code{__sync} builtins. | |
4925 | ||
4926 | Failing that we could add | |
4927 | ||
4928 | @smallexample | |
4929 | void GOMP_atomic_enter (void) | |
4930 | void GOMP_atomic_exit (void) | |
4931 | @end smallexample | |
4932 | ||
4933 | which reuses the regular lock code, but with yet another lock | |
4934 | object private to the library. | |
4935 | ||
4936 | ||
4937 | ||
4938 | @node Implementing FLUSH construct | |
4939 | @section Implementing FLUSH construct | |
4940 | ||
4941 | Expands to the @code{__sync_synchronize} builtin. | |
4942 | ||
4943 | ||
4944 | ||
4945 | @node Implementing BARRIER construct | |
4946 | @section Implementing BARRIER construct | |
4947 | ||
4948 | @smallexample | |
4949 | void GOMP_barrier (void) | |
4950 | @end smallexample | |
4951 | ||
4952 | ||
4953 | @node Implementing THREADPRIVATE construct | |
4954 | @section Implementing THREADPRIVATE construct | |
4955 | ||
4956 | In _most_ cases we can map this directly to @code{__thread}. Except | |
4957 | that OMP allows constructors for C++ objects. We can either | |
4958 | refuse to support this (how often is it used?) or we can | |
4959 | implement something akin to .ctors. | |
4960 | ||
4961 | Even more ideally, this ctor feature is handled by extensions | |
4962 | to the main pthreads library. Failing that, we can have a set | |
4963 | of entry points to register ctor functions to be called. | |
4964 | ||
4965 | ||
4966 | ||
4967 | @node Implementing PRIVATE clause | |
4968 | @section Implementing PRIVATE clause | |
4969 | ||
4970 | In association with a PARALLEL, or within the lexical extent | |
4971 | of a PARALLEL block, the variable becomes a local variable in | |
4972 | the parallel subfunction. | |
4973 | ||
4974 | In association with FOR or SECTIONS blocks, create a new | |
4975 | automatic variable within the current function. This preserves | |
4976 | the semantic of new variable creation. | |
4977 | ||
4978 | ||
4979 | ||
4980 | @node Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses | |
4981 | @section Implementing FIRSTPRIVATE LASTPRIVATE COPYIN and COPYPRIVATE clauses | |
4982 | ||
4983 | This seems simple enough for PARALLEL blocks. Create a private | |
4984 | struct for communicating between the parent and subfunction. | |
4985 | In the parent, copy in values for scalar and "small" structs; | |
4986 | copy in addresses for others TREE_ADDRESSABLE types. In the | |
4987 | subfunction, copy the value into the local variable. | |
4988 | ||
4989 | It is not clear what to do with bare FOR or SECTION blocks. | |
4990 | The only thing I can figure is that we do something like: | |
4991 | ||
4992 | @smallexample | |
4993 | #pragma omp for firstprivate(x) lastprivate(y) | |
4994 | for (int i = 0; i < n; ++i) | |
4995 | body; | |
4996 | @end smallexample | |
4997 | ||
4998 | which becomes | |
4999 | ||
5000 | @smallexample | |
5001 | @{ | |
5002 | int x = x, y; | |
5003 | ||
5004 | // for stuff | |
5005 | ||
5006 | if (i == n) | |
5007 | y = y; | |
5008 | @} | |
5009 | @end smallexample | |
5010 | ||
5011 | where the "x=x" and "y=y" assignments actually have different | |
5012 | uids for the two variables, i.e. not something you could write | |
5013 | directly in C. Presumably this only makes sense if the "outer" | |
5014 | x and y are global variables. | |
5015 | ||
5016 | COPYPRIVATE would work the same way, except the structure | |
5017 | broadcast would have to happen via SINGLE machinery instead. | |
5018 | ||
5019 | ||
5020 | ||
5021 | @node Implementing REDUCTION clause | |
5022 | @section Implementing REDUCTION clause | |
5023 | ||
5024 | The private struct mentioned in the previous section should have | |
5025 | a pointer to an array of the type of the variable, indexed by the | |
5026 | thread's @var{team_id}. The thread stores its final value into the | |
5027 | array, and after the barrier, the primary thread iterates over the | |
5028 | array to collect the values. | |
5029 | ||
5030 | ||
5031 | @node Implementing PARALLEL construct | |
5032 | @section Implementing PARALLEL construct | |
5033 | ||
5034 | @smallexample | |
5035 | #pragma omp parallel | |
5036 | @{ | |
5037 | body; | |
5038 | @} | |
5039 | @end smallexample | |
5040 | ||
5041 | becomes | |
5042 | ||
5043 | @smallexample | |
5044 | void subfunction (void *data) | |
5045 | @{ | |
5046 | use data; | |
5047 | body; | |
5048 | @} | |
5049 | ||
5050 | setup data; | |
5051 | GOMP_parallel_start (subfunction, &data, num_threads); | |
5052 | subfunction (&data); | |
5053 | GOMP_parallel_end (); | |
5054 | @end smallexample | |
5055 | ||
5056 | @smallexample | |
5057 | void GOMP_parallel_start (void (*fn)(void *), void *data, unsigned num_threads) | |
5058 | @end smallexample | |
5059 | ||
5060 | The @var{FN} argument is the subfunction to be run in parallel. | |
5061 | ||
5062 | The @var{DATA} argument is a pointer to a structure used to | |
5063 | communicate data in and out of the subfunction, as discussed | |
5064 | above with respect to FIRSTPRIVATE et al. | |
5065 | ||
5066 | The @var{NUM_THREADS} argument is 1 if an IF clause is present | |
5067 | and false, or the value of the NUM_THREADS clause, if | |
5068 | present, or 0. | |
5069 | ||
5070 | The function needs to create the appropriate number of | |
5071 | threads and/or launch them from the dock. It needs to | |
5072 | create the team structure and assign team ids. | |
5073 | ||
5074 | @smallexample | |
5075 | void GOMP_parallel_end (void) | |
5076 | @end smallexample | |
5077 | ||
5078 | Tears down the team and returns us to the previous @code{omp_in_parallel()} state. | |
5079 | ||
5080 | ||
5081 | ||
5082 | @node Implementing FOR construct | |
5083 | @section Implementing FOR construct | |
5084 | ||
5085 | @smallexample | |
5086 | #pragma omp parallel for | |
5087 | for (i = lb; i <= ub; i++) | |
5088 | body; | |
5089 | @end smallexample | |
5090 | ||
5091 | becomes | |
5092 | ||
5093 | @smallexample | |
5094 | void subfunction (void *data) | |
5095 | @{ | |
5096 | long _s0, _e0; | |
5097 | while (GOMP_loop_static_next (&_s0, &_e0)) | |
5098 | @{ | |
5099 | long _e1 = _e0, i; | |
5100 | for (i = _s0; i < _e1; i++) | |
5101 | body; | |
5102 | @} | |
5103 | GOMP_loop_end_nowait (); | |
5104 | @} | |
5105 | ||
5106 | GOMP_parallel_loop_static (subfunction, NULL, 0, lb, ub+1, 1, 0); | |
5107 | subfunction (NULL); | |
5108 | GOMP_parallel_end (); | |
5109 | @end smallexample | |
5110 | ||
5111 | @smallexample | |
5112 | #pragma omp for schedule(runtime) | |
5113 | for (i = 0; i < n; i++) | |
5114 | body; | |
5115 | @end smallexample | |
5116 | ||
5117 | becomes | |
5118 | ||
5119 | @smallexample | |
5120 | @{ | |
5121 | long i, _s0, _e0; | |
5122 | if (GOMP_loop_runtime_start (0, n, 1, &_s0, &_e0)) | |
5123 | do @{ | |
5124 | long _e1 = _e0; | |
5125 | for (i = _s0, i < _e0; i++) | |
5126 | body; | |
5127 | @} while (GOMP_loop_runtime_next (&_s0, _&e0)); | |
5128 | GOMP_loop_end (); | |
5129 | @} | |
5130 | @end smallexample | |
5131 | ||
5132 | Note that while it looks like there is trickiness to propagating | |
5133 | a non-constant STEP, there isn't really. We're explicitly allowed | |
5134 | to evaluate it as many times as we want, and any variables involved | |
5135 | should automatically be handled as PRIVATE or SHARED like any other | |
5136 | variables. So the expression should remain evaluable in the | |
5137 | subfunction. We can also pull it into a local variable if we like, | |
5138 | but since its supposed to remain unchanged, we can also not if we like. | |
5139 | ||
5140 | If we have SCHEDULE(STATIC), and no ORDERED, then we ought to be | |
5141 | able to get away with no work-sharing context at all, since we can | |
5142 | simply perform the arithmetic directly in each thread to divide up | |
5143 | the iterations. Which would mean that we wouldn't need to call any | |
5144 | of these routines. | |
5145 | ||
5146 | There are separate routines for handling loops with an ORDERED | |
5147 | clause. Bookkeeping for that is non-trivial... | |
5148 | ||
5149 | ||
5150 | ||
5151 | @node Implementing ORDERED construct | |
5152 | @section Implementing ORDERED construct | |
5153 | ||
5154 | @smallexample | |
5155 | void GOMP_ordered_start (void) | |
5156 | void GOMP_ordered_end (void) | |
5157 | @end smallexample | |
5158 | ||
5159 | ||
5160 | ||
5161 | @node Implementing SECTIONS construct | |
5162 | @section Implementing SECTIONS construct | |
5163 | ||
5164 | A block as | |
5165 | ||
5166 | @smallexample | |
5167 | #pragma omp sections | |
5168 | @{ | |
5169 | #pragma omp section | |
5170 | stmt1; | |
5171 | #pragma omp section | |
5172 | stmt2; | |
5173 | #pragma omp section | |
5174 | stmt3; | |
5175 | @} | |
5176 | @end smallexample | |
5177 | ||
5178 | becomes | |
5179 | ||
5180 | @smallexample | |
5181 | for (i = GOMP_sections_start (3); i != 0; i = GOMP_sections_next ()) | |
5182 | switch (i) | |
5183 | @{ | |
5184 | case 1: | |
5185 | stmt1; | |
5186 | break; | |
5187 | case 2: | |
5188 | stmt2; | |
5189 | break; | |
5190 | case 3: | |
5191 | stmt3; | |
5192 | break; | |
5193 | @} | |
5194 | GOMP_barrier (); | |
5195 | @end smallexample | |
5196 | ||
5197 | ||
5198 | @node Implementing SINGLE construct | |
5199 | @section Implementing SINGLE construct | |
5200 | ||
5201 | A block like | |
5202 | ||
5203 | @smallexample | |
5204 | #pragma omp single | |
5205 | @{ | |
5206 | body; | |
5207 | @} | |
5208 | @end smallexample | |
5209 | ||
5210 | becomes | |
5211 | ||
5212 | @smallexample | |
5213 | if (GOMP_single_start ()) | |
5214 | body; | |
5215 | GOMP_barrier (); | |
5216 | @end smallexample | |
5217 | ||
5218 | while | |
5219 | ||
5220 | @smallexample | |
5221 | #pragma omp single copyprivate(x) | |
5222 | body; | |
5223 | @end smallexample | |
5224 | ||
5225 | becomes | |
5226 | ||
5227 | @smallexample | |
5228 | datap = GOMP_single_copy_start (); | |
5229 | if (datap == NULL) | |
5230 | @{ | |
5231 | body; | |
5232 | data.x = x; | |
5233 | GOMP_single_copy_end (&data); | |
5234 | @} | |
5235 | else | |
5236 | x = datap->x; | |
5237 | GOMP_barrier (); | |
5238 | @end smallexample | |
5239 | ||
5240 | ||
5241 | ||
5242 | @node Implementing OpenACC's PARALLEL construct | |
5243 | @section Implementing OpenACC's PARALLEL construct | |
5244 | ||
5245 | @smallexample | |
5246 | void GOACC_parallel () | |
5247 | @end smallexample | |
5248 | ||
5249 | ||
5250 | ||
5251 | @c --------------------------------------------------------------------- | |
5252 | @c Reporting Bugs | |
5253 | @c --------------------------------------------------------------------- | |
5254 | ||
5255 | @node Reporting Bugs | |
5256 | @chapter Reporting Bugs | |
5257 | ||
5258 | Bugs in the GNU Offloading and Multi Processing Runtime Library should | |
5259 | be reported via @uref{https://gcc.gnu.org/bugzilla/, Bugzilla}. Please add | |
5260 | "openacc", or "openmp", or both to the keywords field in the bug | |
5261 | report, as appropriate. | |
5262 | ||
5263 | ||
5264 | ||
5265 | @c --------------------------------------------------------------------- | |
5266 | @c GNU General Public License | |
5267 | @c --------------------------------------------------------------------- | |
5268 | ||
5269 | @include gpl_v3.texi | |
5270 | ||
5271 | ||
5272 | ||
5273 | @c --------------------------------------------------------------------- | |
5274 | @c GNU Free Documentation License | |
5275 | @c --------------------------------------------------------------------- | |
5276 | ||
5277 | @include fdl.texi | |
5278 | ||
5279 | ||
5280 | ||
5281 | @c --------------------------------------------------------------------- | |
5282 | @c Funding Free Software | |
5283 | @c --------------------------------------------------------------------- | |
5284 | ||
5285 | @include funding.texi | |
5286 | ||
5287 | @c --------------------------------------------------------------------- | |
5288 | @c Index | |
5289 | @c --------------------------------------------------------------------- | |
5290 | ||
5291 | @node Library Index | |
5292 | @unnumbered Library Index | |
5293 | ||
5294 | @printindex cp | |
5295 | ||
5296 | @bye |