|
145
|
1 /* Copyright (C) 2013-2020 Free Software Foundation, Inc.
|
|
111
|
2
|
|
|
3 Contributed by Mentor Embedded.
|
|
|
4
|
|
|
5 This file is part of the GNU Offloading and Multi Processing Library
|
|
|
6 (libgomp).
|
|
|
7
|
|
|
8 Libgomp is free software; you can redistribute it and/or modify it
|
|
|
9 under the terms of the GNU General Public License as published by
|
|
|
10 the Free Software Foundation; either version 3, or (at your option)
|
|
|
11 any later version.
|
|
|
12
|
|
|
13 Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
|
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
|
|
|
15 FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
|
16 more details.
|
|
|
17
|
|
|
18 Under Section 7 of GPL version 3, you are granted additional
|
|
|
19 permissions described in the GCC Runtime Library Exception, version
|
|
|
20 3.1, as published by the Free Software Foundation.
|
|
|
21
|
|
|
22 You should have received a copy of the GNU General Public License and
|
|
|
23 a copy of the GCC Runtime Library Exception along with this program;
|
|
|
24 see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
|
|
|
25 <http://www.gnu.org/licenses/>. */
|
|
|
26
|
|
|
27 /* This file handles OpenACC constructs. */
|
|
|
28
|
|
|
29 #include "openacc.h"
|
|
|
30 #include "libgomp.h"
|
|
|
31 #include "gomp-constants.h"
|
|
|
32 #include "oacc-int.h"
|
|
|
33 #ifdef HAVE_INTTYPES_H
|
|
|
34 # include <inttypes.h> /* For PRIu64. */
|
|
|
35 #endif
|
|
|
36 #include <string.h>
|
|
|
37 #include <stdarg.h>
|
|
|
38 #include <assert.h>
|
|
|
39
|
|
|
40
|
|
145
|
41 /* In the ABI, the GOACC_FLAGs are encoded as an inverted bitmask, so that we
|
|
|
42 continue to support the following two legacy values. */
|
|
|
43 _Static_assert (GOACC_FLAGS_UNMARSHAL (GOMP_DEVICE_ICV) == 0,
|
|
|
44 "legacy GOMP_DEVICE_ICV broken");
|
|
|
45 _Static_assert (GOACC_FLAGS_UNMARSHAL (GOMP_DEVICE_HOST_FALLBACK)
|
|
|
46 == GOACC_FLAG_HOST_FALLBACK,
|
|
|
47 "legacy GOMP_DEVICE_HOST_FALLBACK broken");
|
|
111
|
48
|
|
131
|
49
|
|
|
50 /* Handle the mapping pair that are presented when a
|
|
|
51 deviceptr clause is used with Fortran. */
|
|
|
52
|
|
|
53 static void
|
|
|
54 handle_ftn_pointers (size_t mapnum, void **hostaddrs, size_t *sizes,
|
|
|
55 unsigned short *kinds)
|
|
|
56 {
|
|
|
57 int i;
|
|
|
58
|
|
|
59 for (i = 0; i < mapnum; i++)
|
|
|
60 {
|
|
|
61 unsigned short kind1 = kinds[i] & 0xff;
|
|
|
62
|
|
|
63 /* Handle Fortran deviceptr clause. */
|
|
|
64 if (kind1 == GOMP_MAP_FORCE_DEVICEPTR)
|
|
|
65 {
|
|
|
66 unsigned short kind2;
|
|
|
67
|
|
|
68 if (i < (signed)mapnum - 1)
|
|
|
69 kind2 = kinds[i + 1] & 0xff;
|
|
|
70 else
|
|
|
71 kind2 = 0xffff;
|
|
|
72
|
|
|
73 if (sizes[i] == sizeof (void *))
|
|
|
74 continue;
|
|
|
75
|
|
|
76 /* At this point, we're dealing with a Fortran deviceptr.
|
|
|
77 If the next element is not what we're expecting, then
|
|
|
78 this is an instance of where the deviceptr variable was
|
|
|
79 not used within the region and the pointer was removed
|
|
|
80 by the gimplifier. */
|
|
|
81 if (kind2 == GOMP_MAP_POINTER
|
|
|
82 && sizes[i + 1] == 0
|
|
|
83 && hostaddrs[i] == *(void **)hostaddrs[i + 1])
|
|
|
84 {
|
|
|
85 kinds[i+1] = kinds[i];
|
|
|
86 sizes[i+1] = sizeof (void *);
|
|
|
87 }
|
|
|
88
|
|
|
89 /* Invalidate the entry. */
|
|
|
90 hostaddrs[i] = NULL;
|
|
|
91 }
|
|
|
92 }
|
|
111
|
93 }
|
|
|
94
|
|
|
95
|
|
145
|
96 /* Launch a possibly offloaded function with FLAGS. FN is the host fn
|
|
111
|
97 address. MAPNUM, HOSTADDRS, SIZES & KINDS describe the memory
|
|
|
98 blocks to be copied to/from the device. Varadic arguments are
|
|
|
99 keyed optional parameters terminated with a zero. */
|
|
|
100
|
|
|
101 void
|
|
145
|
102 GOACC_parallel_keyed (int flags_m, void (*fn) (void *),
|
|
111
|
103 size_t mapnum, void **hostaddrs, size_t *sizes,
|
|
|
104 unsigned short *kinds, ...)
|
|
|
105 {
|
|
145
|
106 int flags = GOACC_FLAGS_UNMARSHAL (flags_m);
|
|
|
107
|
|
111
|
108 va_list ap;
|
|
|
109 struct goacc_thread *thr;
|
|
|
110 struct gomp_device_descr *acc_dev;
|
|
|
111 struct target_mem_desc *tgt;
|
|
|
112 void **devaddrs;
|
|
|
113 unsigned int i;
|
|
|
114 struct splay_tree_key_s k;
|
|
|
115 splay_tree_key tgt_fn_key;
|
|
|
116 void (*tgt_fn);
|
|
|
117 int async = GOMP_ASYNC_SYNC;
|
|
|
118 unsigned dims[GOMP_DIM_MAX];
|
|
|
119 unsigned tag;
|
|
|
120
|
|
|
121 #ifdef HAVE_INTTYPES_H
|
|
|
122 gomp_debug (0, "%s: mapnum=%"PRIu64", hostaddrs=%p, size=%p, kinds=%p\n",
|
|
|
123 __FUNCTION__, (uint64_t) mapnum, hostaddrs, sizes, kinds);
|
|
|
124 #else
|
|
|
125 gomp_debug (0, "%s: mapnum=%lu, hostaddrs=%p, sizes=%p, kinds=%p\n",
|
|
|
126 __FUNCTION__, (unsigned long) mapnum, hostaddrs, sizes, kinds);
|
|
|
127 #endif
|
|
|
128 goacc_lazy_initialize ();
|
|
|
129
|
|
|
130 thr = goacc_thread ();
|
|
|
131 acc_dev = thr->dev;
|
|
|
132
|
|
145
|
133 bool profiling_p = GOACC_PROFILING_DISPATCH_P (true);
|
|
|
134
|
|
|
135 acc_prof_info prof_info;
|
|
|
136 if (profiling_p)
|
|
|
137 {
|
|
|
138 thr->prof_info = &prof_info;
|
|
|
139
|
|
|
140 prof_info.event_type = acc_ev_compute_construct_start;
|
|
|
141 prof_info.valid_bytes = _ACC_PROF_INFO_VALID_BYTES;
|
|
|
142 prof_info.version = _ACC_PROF_INFO_VERSION;
|
|
|
143 prof_info.device_type = acc_device_type (acc_dev->type);
|
|
|
144 prof_info.device_number = acc_dev->target_id;
|
|
|
145 prof_info.thread_id = -1;
|
|
|
146 prof_info.async = async;
|
|
|
147 prof_info.async_queue = prof_info.async;
|
|
|
148 prof_info.src_file = NULL;
|
|
|
149 prof_info.func_name = NULL;
|
|
|
150 prof_info.line_no = -1;
|
|
|
151 prof_info.end_line_no = -1;
|
|
|
152 prof_info.func_line_no = -1;
|
|
|
153 prof_info.func_end_line_no = -1;
|
|
|
154 }
|
|
|
155 acc_event_info compute_construct_event_info;
|
|
|
156 if (profiling_p)
|
|
|
157 {
|
|
|
158 compute_construct_event_info.other_event.event_type
|
|
|
159 = prof_info.event_type;
|
|
|
160 compute_construct_event_info.other_event.valid_bytes
|
|
|
161 = _ACC_OTHER_EVENT_INFO_VALID_BYTES;
|
|
|
162 compute_construct_event_info.other_event.parent_construct
|
|
|
163 = acc_construct_parallel;
|
|
|
164 compute_construct_event_info.other_event.implicit = 0;
|
|
|
165 compute_construct_event_info.other_event.tool_info = NULL;
|
|
|
166 }
|
|
|
167 acc_api_info api_info;
|
|
|
168 if (profiling_p)
|
|
|
169 {
|
|
|
170 thr->api_info = &api_info;
|
|
|
171
|
|
|
172 api_info.device_api = acc_device_api_none;
|
|
|
173 api_info.valid_bytes = _ACC_API_INFO_VALID_BYTES;
|
|
|
174 api_info.device_type = prof_info.device_type;
|
|
|
175 api_info.vendor = -1;
|
|
|
176 api_info.device_handle = NULL;
|
|
|
177 api_info.context_handle = NULL;
|
|
|
178 api_info.async_handle = NULL;
|
|
|
179 }
|
|
|
180
|
|
|
181 if (profiling_p)
|
|
|
182 goacc_profiling_dispatch (&prof_info, &compute_construct_event_info,
|
|
|
183 &api_info);
|
|
|
184
|
|
131
|
185 handle_ftn_pointers (mapnum, hostaddrs, sizes, kinds);
|
|
|
186
|
|
111
|
187 /* Host fallback if "if" clause is false or if the current device is set to
|
|
|
188 the host. */
|
|
145
|
189 if (flags & GOACC_FLAG_HOST_FALLBACK)
|
|
111
|
190 {
|
|
145
|
191 prof_info.device_type = acc_device_host;
|
|
|
192 api_info.device_type = prof_info.device_type;
|
|
111
|
193 goacc_save_and_set_bind (acc_device_host);
|
|
|
194 fn (hostaddrs);
|
|
|
195 goacc_restore_bind ();
|
|
145
|
196 goto out_prof;
|
|
111
|
197 }
|
|
|
198 else if (acc_device_type (acc_dev->type) == acc_device_host)
|
|
|
199 {
|
|
|
200 fn (hostaddrs);
|
|
145
|
201 goto out_prof;
|
|
111
|
202 }
|
|
|
203
|
|
|
204 /* Default: let the runtime choose. */
|
|
|
205 for (i = 0; i != GOMP_DIM_MAX; i++)
|
|
|
206 dims[i] = 0;
|
|
|
207
|
|
|
208 va_start (ap, kinds);
|
|
|
209 /* TODO: This will need amending when device_type is implemented. */
|
|
|
210 while ((tag = va_arg (ap, unsigned)) != 0)
|
|
|
211 {
|
|
|
212 if (GOMP_LAUNCH_DEVICE (tag))
|
|
|
213 gomp_fatal ("device_type '%d' offload parameters, libgomp is too old",
|
|
|
214 GOMP_LAUNCH_DEVICE (tag));
|
|
|
215
|
|
|
216 switch (GOMP_LAUNCH_CODE (tag))
|
|
|
217 {
|
|
|
218 case GOMP_LAUNCH_DIM:
|
|
|
219 {
|
|
|
220 unsigned mask = GOMP_LAUNCH_OP (tag);
|
|
|
221
|
|
|
222 for (i = 0; i != GOMP_DIM_MAX; i++)
|
|
|
223 if (mask & GOMP_DIM_MASK (i))
|
|
|
224 dims[i] = va_arg (ap, unsigned);
|
|
|
225 }
|
|
|
226 break;
|
|
|
227
|
|
|
228 case GOMP_LAUNCH_ASYNC:
|
|
|
229 {
|
|
|
230 /* Small constant values are encoded in the operand. */
|
|
|
231 async = GOMP_LAUNCH_OP (tag);
|
|
|
232
|
|
|
233 if (async == GOMP_LAUNCH_OP_MAX)
|
|
|
234 async = va_arg (ap, unsigned);
|
|
145
|
235
|
|
|
236 if (profiling_p)
|
|
|
237 {
|
|
|
238 prof_info.async = async;
|
|
|
239 prof_info.async_queue = prof_info.async;
|
|
|
240 }
|
|
|
241
|
|
111
|
242 break;
|
|
|
243 }
|
|
|
244
|
|
|
245 case GOMP_LAUNCH_WAIT:
|
|
|
246 {
|
|
|
247 unsigned num_waits = GOMP_LAUNCH_OP (tag);
|
|
145
|
248 goacc_wait (async, num_waits, &ap);
|
|
111
|
249 break;
|
|
|
250 }
|
|
|
251
|
|
|
252 default:
|
|
|
253 gomp_fatal ("unrecognized offload code '%d',"
|
|
|
254 " libgomp is too old", GOMP_LAUNCH_CODE (tag));
|
|
|
255 }
|
|
|
256 }
|
|
|
257 va_end (ap);
|
|
|
258
|
|
|
259 if (!(acc_dev->capabilities & GOMP_OFFLOAD_CAP_NATIVE_EXEC))
|
|
|
260 {
|
|
|
261 k.host_start = (uintptr_t) fn;
|
|
|
262 k.host_end = k.host_start + 1;
|
|
|
263 gomp_mutex_lock (&acc_dev->lock);
|
|
|
264 tgt_fn_key = splay_tree_lookup (&acc_dev->mem_map, &k);
|
|
|
265 gomp_mutex_unlock (&acc_dev->lock);
|
|
|
266
|
|
|
267 if (tgt_fn_key == NULL)
|
|
|
268 gomp_fatal ("target function wasn't mapped");
|
|
|
269
|
|
|
270 tgt_fn = (void (*)) tgt_fn_key->tgt_offset;
|
|
|
271 }
|
|
|
272 else
|
|
|
273 tgt_fn = (void (*)) fn;
|
|
|
274
|
|
145
|
275 acc_event_info enter_exit_data_event_info;
|
|
|
276 if (profiling_p)
|
|
|
277 {
|
|
|
278 prof_info.event_type = acc_ev_enter_data_start;
|
|
|
279 enter_exit_data_event_info.other_event.event_type
|
|
|
280 = prof_info.event_type;
|
|
|
281 enter_exit_data_event_info.other_event.valid_bytes
|
|
|
282 = _ACC_OTHER_EVENT_INFO_VALID_BYTES;
|
|
|
283 enter_exit_data_event_info.other_event.parent_construct
|
|
|
284 = compute_construct_event_info.other_event.parent_construct;
|
|
|
285 enter_exit_data_event_info.other_event.implicit = 1;
|
|
|
286 enter_exit_data_event_info.other_event.tool_info = NULL;
|
|
|
287 goacc_profiling_dispatch (&prof_info, &enter_exit_data_event_info,
|
|
|
288 &api_info);
|
|
|
289 }
|
|
111
|
290
|
|
145
|
291 goacc_aq aq = get_goacc_asyncqueue (async);
|
|
|
292
|
|
|
293 tgt = gomp_map_vars_async (acc_dev, aq, mapnum, hostaddrs, NULL, sizes, kinds,
|
|
|
294 true, GOMP_MAP_VARS_OPENACC);
|
|
|
295 if (profiling_p)
|
|
|
296 {
|
|
|
297 prof_info.event_type = acc_ev_enter_data_end;
|
|
|
298 enter_exit_data_event_info.other_event.event_type
|
|
|
299 = prof_info.event_type;
|
|
|
300 goacc_profiling_dispatch (&prof_info, &enter_exit_data_event_info,
|
|
|
301 &api_info);
|
|
|
302 }
|
|
|
303
|
|
111
|
304 devaddrs = gomp_alloca (sizeof (void *) * mapnum);
|
|
|
305 for (i = 0; i < mapnum; i++)
|
|
145
|
306 devaddrs[i] = (void *) gomp_map_val (tgt, hostaddrs, i);
|
|
|
307
|
|
|
308 if (aq == NULL)
|
|
|
309 acc_dev->openacc.exec_func (tgt_fn, mapnum, hostaddrs, devaddrs, dims,
|
|
|
310 tgt);
|
|
|
311 else
|
|
|
312 acc_dev->openacc.async.exec_func (tgt_fn, mapnum, hostaddrs, devaddrs,
|
|
|
313 dims, tgt, aq);
|
|
111
|
314
|
|
145
|
315 if (profiling_p)
|
|
|
316 {
|
|
|
317 prof_info.event_type = acc_ev_exit_data_start;
|
|
|
318 enter_exit_data_event_info.other_event.event_type = prof_info.event_type;
|
|
|
319 enter_exit_data_event_info.other_event.tool_info = NULL;
|
|
|
320 goacc_profiling_dispatch (&prof_info, &enter_exit_data_event_info,
|
|
|
321 &api_info);
|
|
|
322 }
|
|
111
|
323
|
|
|
324 /* If running synchronously, unmap immediately. */
|
|
145
|
325 if (aq == NULL)
|
|
111
|
326 gomp_unmap_vars (tgt, true);
|
|
|
327 else
|
|
145
|
328 gomp_unmap_vars_async (tgt, true, aq);
|
|
|
329
|
|
|
330 if (profiling_p)
|
|
131
|
331 {
|
|
145
|
332 prof_info.event_type = acc_ev_exit_data_end;
|
|
|
333 enter_exit_data_event_info.other_event.event_type = prof_info.event_type;
|
|
|
334 goacc_profiling_dispatch (&prof_info, &enter_exit_data_event_info,
|
|
|
335 &api_info);
|
|
131
|
336 }
|
|
111
|
337
|
|
145
|
338 out_prof:
|
|
|
339 if (profiling_p)
|
|
|
340 {
|
|
|
341 prof_info.event_type = acc_ev_compute_construct_end;
|
|
|
342 compute_construct_event_info.other_event.event_type
|
|
|
343 = prof_info.event_type;
|
|
|
344 goacc_profiling_dispatch (&prof_info, &compute_construct_event_info,
|
|
|
345 &api_info);
|
|
|
346
|
|
|
347 thr->prof_info = NULL;
|
|
|
348 thr->api_info = NULL;
|
|
|
349 }
|
|
111
|
350 }
|
|
|
351
|
|
145
|
352 /* Legacy entry point (GCC 5). Only provide host fallback execution. */
|
|
111
|
353
|
|
|
354 void
|
|
145
|
355 GOACC_parallel (int flags_m, void (*fn) (void *),
|
|
111
|
356 size_t mapnum, void **hostaddrs, size_t *sizes,
|
|
|
357 unsigned short *kinds,
|
|
|
358 int num_gangs, int num_workers, int vector_length,
|
|
|
359 int async, int num_waits, ...)
|
|
|
360 {
|
|
|
361 goacc_save_and_set_bind (acc_device_host);
|
|
|
362 fn (hostaddrs);
|
|
|
363 goacc_restore_bind ();
|
|
|
364 }
|
|
|
365
|
|
|
366 void
|
|
145
|
367 GOACC_data_start (int flags_m, size_t mapnum,
|
|
111
|
368 void **hostaddrs, size_t *sizes, unsigned short *kinds)
|
|
|
369 {
|
|
145
|
370 int flags = GOACC_FLAGS_UNMARSHAL (flags_m);
|
|
|
371
|
|
111
|
372 struct target_mem_desc *tgt;
|
|
|
373
|
|
|
374 #ifdef HAVE_INTTYPES_H
|
|
|
375 gomp_debug (0, "%s: mapnum=%"PRIu64", hostaddrs=%p, size=%p, kinds=%p\n",
|
|
|
376 __FUNCTION__, (uint64_t) mapnum, hostaddrs, sizes, kinds);
|
|
|
377 #else
|
|
|
378 gomp_debug (0, "%s: mapnum=%lu, hostaddrs=%p, sizes=%p, kinds=%p\n",
|
|
|
379 __FUNCTION__, (unsigned long) mapnum, hostaddrs, sizes, kinds);
|
|
|
380 #endif
|
|
|
381
|
|
|
382 goacc_lazy_initialize ();
|
|
|
383
|
|
|
384 struct goacc_thread *thr = goacc_thread ();
|
|
|
385 struct gomp_device_descr *acc_dev = thr->dev;
|
|
|
386
|
|
145
|
387 bool profiling_p = GOACC_PROFILING_DISPATCH_P (true);
|
|
|
388
|
|
|
389 acc_prof_info prof_info;
|
|
|
390 if (profiling_p)
|
|
|
391 {
|
|
|
392 thr->prof_info = &prof_info;
|
|
|
393
|
|
|
394 prof_info.event_type = acc_ev_enter_data_start;
|
|
|
395 prof_info.valid_bytes = _ACC_PROF_INFO_VALID_BYTES;
|
|
|
396 prof_info.version = _ACC_PROF_INFO_VERSION;
|
|
|
397 prof_info.device_type = acc_device_type (acc_dev->type);
|
|
|
398 prof_info.device_number = acc_dev->target_id;
|
|
|
399 prof_info.thread_id = -1;
|
|
|
400 prof_info.async = acc_async_sync; /* Always synchronous. */
|
|
|
401 prof_info.async_queue = prof_info.async;
|
|
|
402 prof_info.src_file = NULL;
|
|
|
403 prof_info.func_name = NULL;
|
|
|
404 prof_info.line_no = -1;
|
|
|
405 prof_info.end_line_no = -1;
|
|
|
406 prof_info.func_line_no = -1;
|
|
|
407 prof_info.func_end_line_no = -1;
|
|
|
408 }
|
|
|
409 acc_event_info enter_data_event_info;
|
|
|
410 if (profiling_p)
|
|
|
411 {
|
|
|
412 enter_data_event_info.other_event.event_type
|
|
|
413 = prof_info.event_type;
|
|
|
414 enter_data_event_info.other_event.valid_bytes
|
|
|
415 = _ACC_OTHER_EVENT_INFO_VALID_BYTES;
|
|
|
416 enter_data_event_info.other_event.parent_construct = acc_construct_data;
|
|
|
417 for (int i = 0; i < mapnum; ++i)
|
|
|
418 if ((kinds[i] & 0xff) == GOMP_MAP_USE_DEVICE_PTR
|
|
|
419 || (kinds[i] & 0xff) == GOMP_MAP_USE_DEVICE_PTR_IF_PRESENT)
|
|
|
420 {
|
|
|
421 /* If there is one such data mapping kind, then this is actually an
|
|
|
422 OpenACC 'host_data' construct. (GCC maps the OpenACC
|
|
|
423 'host_data' construct to the OpenACC 'data' construct.) Apart
|
|
|
424 from artificial test cases (such as an OpenACC 'host_data'
|
|
|
425 construct's (implicit) device initialization when there hasn't
|
|
|
426 been any device data be set up before...), there can't really
|
|
|
427 any meaningful events be generated from OpenACC 'host_data'
|
|
|
428 constructs, though. */
|
|
|
429 enter_data_event_info.other_event.parent_construct
|
|
|
430 = acc_construct_host_data;
|
|
|
431 break;
|
|
|
432 }
|
|
|
433 enter_data_event_info.other_event.implicit = 0;
|
|
|
434 enter_data_event_info.other_event.tool_info = NULL;
|
|
|
435 }
|
|
|
436 acc_api_info api_info;
|
|
|
437 if (profiling_p)
|
|
|
438 {
|
|
|
439 thr->api_info = &api_info;
|
|
|
440
|
|
|
441 api_info.device_api = acc_device_api_none;
|
|
|
442 api_info.valid_bytes = _ACC_API_INFO_VALID_BYTES;
|
|
|
443 api_info.device_type = prof_info.device_type;
|
|
|
444 api_info.vendor = -1;
|
|
|
445 api_info.device_handle = NULL;
|
|
|
446 api_info.context_handle = NULL;
|
|
|
447 api_info.async_handle = NULL;
|
|
|
448 }
|
|
|
449
|
|
|
450 if (profiling_p)
|
|
|
451 goacc_profiling_dispatch (&prof_info, &enter_data_event_info, &api_info);
|
|
|
452
|
|
111
|
453 /* Host fallback or 'do nothing'. */
|
|
|
454 if ((acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
|
|
145
|
455 || (flags & GOACC_FLAG_HOST_FALLBACK))
|
|
111
|
456 {
|
|
145
|
457 prof_info.device_type = acc_device_host;
|
|
|
458 api_info.device_type = prof_info.device_type;
|
|
111
|
459 tgt = gomp_map_vars (NULL, 0, NULL, NULL, NULL, NULL, true,
|
|
|
460 GOMP_MAP_VARS_OPENACC);
|
|
|
461 tgt->prev = thr->mapped_data;
|
|
|
462 thr->mapped_data = tgt;
|
|
|
463
|
|
145
|
464 goto out_prof;
|
|
111
|
465 }
|
|
|
466
|
|
|
467 gomp_debug (0, " %s: prepare mappings\n", __FUNCTION__);
|
|
|
468 tgt = gomp_map_vars (acc_dev, mapnum, hostaddrs, NULL, sizes, kinds, true,
|
|
|
469 GOMP_MAP_VARS_OPENACC);
|
|
|
470 gomp_debug (0, " %s: mappings prepared\n", __FUNCTION__);
|
|
|
471 tgt->prev = thr->mapped_data;
|
|
|
472 thr->mapped_data = tgt;
|
|
145
|
473
|
|
|
474 out_prof:
|
|
|
475 if (profiling_p)
|
|
|
476 {
|
|
|
477 prof_info.event_type = acc_ev_enter_data_end;
|
|
|
478 enter_data_event_info.other_event.event_type = prof_info.event_type;
|
|
|
479 goacc_profiling_dispatch (&prof_info, &enter_data_event_info, &api_info);
|
|
|
480
|
|
|
481 thr->prof_info = NULL;
|
|
|
482 thr->api_info = NULL;
|
|
|
483 }
|
|
111
|
484 }
|
|
|
485
|
|
|
486 void
|
|
|
487 GOACC_data_end (void)
|
|
|
488 {
|
|
|
489 struct goacc_thread *thr = goacc_thread ();
|
|
145
|
490 struct gomp_device_descr *acc_dev = thr->dev;
|
|
111
|
491 struct target_mem_desc *tgt = thr->mapped_data;
|
|
|
492
|
|
145
|
493 bool profiling_p = GOACC_PROFILING_DISPATCH_P (true);
|
|
|
494
|
|
|
495 acc_prof_info prof_info;
|
|
|
496 if (profiling_p)
|
|
|
497 {
|
|
|
498 thr->prof_info = &prof_info;
|
|
|
499
|
|
|
500 prof_info.event_type = acc_ev_exit_data_start;
|
|
|
501 prof_info.valid_bytes = _ACC_PROF_INFO_VALID_BYTES;
|
|
|
502 prof_info.version = _ACC_PROF_INFO_VERSION;
|
|
|
503 prof_info.device_type = acc_device_type (acc_dev->type);
|
|
|
504 prof_info.device_number = acc_dev->target_id;
|
|
|
505 prof_info.thread_id = -1;
|
|
|
506 prof_info.async = acc_async_sync; /* Always synchronous. */
|
|
|
507 prof_info.async_queue = prof_info.async;
|
|
|
508 prof_info.src_file = NULL;
|
|
|
509 prof_info.func_name = NULL;
|
|
|
510 prof_info.line_no = -1;
|
|
|
511 prof_info.end_line_no = -1;
|
|
|
512 prof_info.func_line_no = -1;
|
|
|
513 prof_info.func_end_line_no = -1;
|
|
|
514 }
|
|
|
515 acc_event_info exit_data_event_info;
|
|
|
516 if (profiling_p)
|
|
|
517 {
|
|
|
518 exit_data_event_info.other_event.event_type
|
|
|
519 = prof_info.event_type;
|
|
|
520 exit_data_event_info.other_event.valid_bytes
|
|
|
521 = _ACC_OTHER_EVENT_INFO_VALID_BYTES;
|
|
|
522 exit_data_event_info.other_event.parent_construct = acc_construct_data;
|
|
|
523 exit_data_event_info.other_event.implicit = 0;
|
|
|
524 exit_data_event_info.other_event.tool_info = NULL;
|
|
|
525 }
|
|
|
526 acc_api_info api_info;
|
|
|
527 if (profiling_p)
|
|
|
528 {
|
|
|
529 thr->api_info = &api_info;
|
|
|
530
|
|
|
531 api_info.device_api = acc_device_api_none;
|
|
|
532 api_info.valid_bytes = _ACC_API_INFO_VALID_BYTES;
|
|
|
533 api_info.device_type = prof_info.device_type;
|
|
|
534 api_info.vendor = -1;
|
|
|
535 api_info.device_handle = NULL;
|
|
|
536 api_info.context_handle = NULL;
|
|
|
537 api_info.async_handle = NULL;
|
|
|
538 }
|
|
|
539
|
|
|
540 if (profiling_p)
|
|
|
541 goacc_profiling_dispatch (&prof_info, &exit_data_event_info, &api_info);
|
|
|
542
|
|
111
|
543 gomp_debug (0, " %s: restore mappings\n", __FUNCTION__);
|
|
|
544 thr->mapped_data = tgt->prev;
|
|
|
545 gomp_unmap_vars (tgt, true);
|
|
|
546 gomp_debug (0, " %s: mappings restored\n", __FUNCTION__);
|
|
145
|
547
|
|
|
548 if (profiling_p)
|
|
|
549 {
|
|
|
550 prof_info.event_type = acc_ev_exit_data_end;
|
|
|
551 exit_data_event_info.other_event.event_type = prof_info.event_type;
|
|
|
552 goacc_profiling_dispatch (&prof_info, &exit_data_event_info, &api_info);
|
|
|
553
|
|
|
554 thr->prof_info = NULL;
|
|
|
555 thr->api_info = NULL;
|
|
|
556 }
|
|
111
|
557 }
|
|
|
558
|
|
|
559 void
|
|
145
|
560 GOACC_update (int flags_m, size_t mapnum,
|
|
|
561 void **hostaddrs, size_t *sizes, unsigned short *kinds,
|
|
|
562 int async, int num_waits, ...)
|
|
111
|
563 {
|
|
145
|
564 int flags = GOACC_FLAGS_UNMARSHAL (flags_m);
|
|
|
565
|
|
111
|
566 size_t i;
|
|
|
567
|
|
|
568 goacc_lazy_initialize ();
|
|
|
569
|
|
145
|
570 struct goacc_thread *thr = goacc_thread ();
|
|
|
571 struct gomp_device_descr *acc_dev = thr->dev;
|
|
|
572
|
|
|
573 bool profiling_p = GOACC_PROFILING_DISPATCH_P (true);
|
|
|
574
|
|
|
575 acc_prof_info prof_info;
|
|
|
576 if (profiling_p)
|
|
|
577 {
|
|
|
578 thr->prof_info = &prof_info;
|
|
|
579
|
|
|
580 prof_info.event_type = acc_ev_update_start;
|
|
|
581 prof_info.valid_bytes = _ACC_PROF_INFO_VALID_BYTES;
|
|
|
582 prof_info.version = _ACC_PROF_INFO_VERSION;
|
|
|
583 prof_info.device_type = acc_device_type (acc_dev->type);
|
|
|
584 prof_info.device_number = acc_dev->target_id;
|
|
|
585 prof_info.thread_id = -1;
|
|
|
586 prof_info.async = async;
|
|
|
587 prof_info.async_queue = prof_info.async;
|
|
|
588 prof_info.src_file = NULL;
|
|
|
589 prof_info.func_name = NULL;
|
|
|
590 prof_info.line_no = -1;
|
|
|
591 prof_info.end_line_no = -1;
|
|
|
592 prof_info.func_line_no = -1;
|
|
|
593 prof_info.func_end_line_no = -1;
|
|
|
594 }
|
|
|
595 acc_event_info update_event_info;
|
|
|
596 if (profiling_p)
|
|
|
597 {
|
|
|
598 update_event_info.other_event.event_type
|
|
|
599 = prof_info.event_type;
|
|
|
600 update_event_info.other_event.valid_bytes
|
|
|
601 = _ACC_OTHER_EVENT_INFO_VALID_BYTES;
|
|
|
602 update_event_info.other_event.parent_construct = acc_construct_update;
|
|
|
603 update_event_info.other_event.implicit = 0;
|
|
|
604 update_event_info.other_event.tool_info = NULL;
|
|
|
605 }
|
|
|
606 acc_api_info api_info;
|
|
|
607 if (profiling_p)
|
|
|
608 {
|
|
|
609 thr->api_info = &api_info;
|
|
|
610
|
|
|
611 api_info.device_api = acc_device_api_none;
|
|
|
612 api_info.valid_bytes = _ACC_API_INFO_VALID_BYTES;
|
|
|
613 api_info.device_type = prof_info.device_type;
|
|
|
614 api_info.vendor = -1;
|
|
|
615 api_info.device_handle = NULL;
|
|
|
616 api_info.context_handle = NULL;
|
|
|
617 api_info.async_handle = NULL;
|
|
|
618 }
|
|
|
619
|
|
|
620 if (profiling_p)
|
|
|
621 goacc_profiling_dispatch (&prof_info, &update_event_info, &api_info);
|
|
111
|
622
|
|
|
623 if ((acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
|
|
145
|
624 || (flags & GOACC_FLAG_HOST_FALLBACK))
|
|
|
625 {
|
|
|
626 prof_info.device_type = acc_device_host;
|
|
|
627 api_info.device_type = prof_info.device_type;
|
|
|
628
|
|
|
629 goto out_prof;
|
|
|
630 }
|
|
111
|
631
|
|
|
632 if (num_waits)
|
|
|
633 {
|
|
|
634 va_list ap;
|
|
|
635
|
|
|
636 va_start (ap, num_waits);
|
|
|
637 goacc_wait (async, num_waits, &ap);
|
|
|
638 va_end (ap);
|
|
|
639 }
|
|
|
640
|
|
131
|
641 bool update_device = false;
|
|
111
|
642 for (i = 0; i < mapnum; ++i)
|
|
|
643 {
|
|
|
644 unsigned char kind = kinds[i] & 0xff;
|
|
|
645
|
|
|
646 switch (kind)
|
|
|
647 {
|
|
|
648 case GOMP_MAP_POINTER:
|
|
|
649 case GOMP_MAP_TO_PSET:
|
|
|
650 break;
|
|
|
651
|
|
131
|
652 case GOMP_MAP_ALWAYS_POINTER:
|
|
|
653 if (update_device)
|
|
|
654 {
|
|
|
655 /* Save the contents of the host pointer. */
|
|
|
656 void *dptr = acc_deviceptr (hostaddrs[i-1]);
|
|
|
657 uintptr_t t = *(uintptr_t *) hostaddrs[i];
|
|
|
658
|
|
|
659 /* Update the contents of the host pointer to reflect
|
|
|
660 the value of the allocated device memory in the
|
|
|
661 previous pointer. */
|
|
|
662 *(uintptr_t *) hostaddrs[i] = (uintptr_t)dptr;
|
|
145
|
663 /* TODO: verify that we really cannot use acc_update_device_async
|
|
|
664 here. */
|
|
131
|
665 acc_update_device (hostaddrs[i], sizeof (uintptr_t));
|
|
|
666
|
|
|
667 /* Restore the host pointer. */
|
|
|
668 *(uintptr_t *) hostaddrs[i] = t;
|
|
|
669 update_device = false;
|
|
|
670 }
|
|
|
671 break;
|
|
|
672
|
|
|
673 case GOMP_MAP_TO:
|
|
|
674 if (!acc_is_present (hostaddrs[i], sizes[i]))
|
|
|
675 {
|
|
|
676 update_device = false;
|
|
|
677 break;
|
|
|
678 }
|
|
|
679 /* Fallthru */
|
|
111
|
680 case GOMP_MAP_FORCE_TO:
|
|
131
|
681 update_device = true;
|
|
145
|
682 acc_update_device_async (hostaddrs[i], sizes[i], async);
|
|
111
|
683 break;
|
|
|
684
|
|
131
|
685 case GOMP_MAP_FROM:
|
|
|
686 if (!acc_is_present (hostaddrs[i], sizes[i]))
|
|
|
687 {
|
|
|
688 update_device = false;
|
|
|
689 break;
|
|
|
690 }
|
|
|
691 /* Fallthru */
|
|
111
|
692 case GOMP_MAP_FORCE_FROM:
|
|
131
|
693 update_device = false;
|
|
145
|
694 acc_update_self_async (hostaddrs[i], sizes[i], async);
|
|
111
|
695 break;
|
|
|
696
|
|
|
697 default:
|
|
|
698 gomp_fatal (">>>> GOACC_update UNHANDLED kind 0x%.2x", kind);
|
|
|
699 break;
|
|
|
700 }
|
|
|
701 }
|
|
|
702
|
|
145
|
703 out_prof:
|
|
|
704 if (profiling_p)
|
|
|
705 {
|
|
|
706 prof_info.event_type = acc_ev_update_end;
|
|
|
707 update_event_info.other_event.event_type = prof_info.event_type;
|
|
|
708 goacc_profiling_dispatch (&prof_info, &update_event_info, &api_info);
|
|
|
709
|
|
|
710 thr->prof_info = NULL;
|
|
|
711 thr->api_info = NULL;
|
|
|
712 }
|
|
111
|
713 }
|
|
|
714
|
|
|
715
|
|
145
|
716 /* Legacy entry point (GCC 5). */
|
|
111
|
717
|
|
|
718 int
|
|
|
719 GOACC_get_num_threads (void)
|
|
|
720 {
|
|
|
721 return 1;
|
|
|
722 }
|
|
|
723
|
|
145
|
724 /* Legacy entry point (GCC 5). */
|
|
|
725
|
|
111
|
726 int
|
|
|
727 GOACC_get_thread_num (void)
|
|
|
728 {
|
|
|
729 return 0;
|
|
|
730 }
|
|
|
731
|
|
|
732 void
|
|
145
|
733 GOACC_declare (int flags_m, size_t mapnum,
|
|
111
|
734 void **hostaddrs, size_t *sizes, unsigned short *kinds)
|
|
|
735 {
|
|
|
736 int i;
|
|
|
737
|
|
|
738 for (i = 0; i < mapnum; i++)
|
|
|
739 {
|
|
|
740 unsigned char kind = kinds[i] & 0xff;
|
|
|
741
|
|
|
742 if (kind == GOMP_MAP_POINTER || kind == GOMP_MAP_TO_PSET)
|
|
|
743 continue;
|
|
|
744
|
|
|
745 switch (kind)
|
|
|
746 {
|
|
|
747 case GOMP_MAP_FORCE_ALLOC:
|
|
|
748 case GOMP_MAP_FORCE_FROM:
|
|
|
749 case GOMP_MAP_FORCE_TO:
|
|
|
750 case GOMP_MAP_POINTER:
|
|
131
|
751 case GOMP_MAP_RELEASE:
|
|
111
|
752 case GOMP_MAP_DELETE:
|
|
145
|
753 GOACC_enter_exit_data (flags_m, 1, &hostaddrs[i], &sizes[i],
|
|
131
|
754 &kinds[i], GOMP_ASYNC_SYNC, 0);
|
|
111
|
755 break;
|
|
|
756
|
|
|
757 case GOMP_MAP_FORCE_DEVICEPTR:
|
|
|
758 break;
|
|
|
759
|
|
|
760 case GOMP_MAP_ALLOC:
|
|
|
761 if (!acc_is_present (hostaddrs[i], sizes[i]))
|
|
145
|
762 GOACC_enter_exit_data (flags_m, 1, &hostaddrs[i], &sizes[i],
|
|
131
|
763 &kinds[i], GOMP_ASYNC_SYNC, 0);
|
|
111
|
764 break;
|
|
|
765
|
|
|
766 case GOMP_MAP_TO:
|
|
145
|
767 GOACC_enter_exit_data (flags_m, 1, &hostaddrs[i], &sizes[i],
|
|
131
|
768 &kinds[i], GOMP_ASYNC_SYNC, 0);
|
|
111
|
769
|
|
|
770 break;
|
|
|
771
|
|
|
772 case GOMP_MAP_FROM:
|
|
145
|
773 GOACC_enter_exit_data (flags_m, 1, &hostaddrs[i], &sizes[i],
|
|
131
|
774 &kinds[i], GOMP_ASYNC_SYNC, 0);
|
|
111
|
775 break;
|
|
|
776
|
|
|
777 case GOMP_MAP_FORCE_PRESENT:
|
|
|
778 if (!acc_is_present (hostaddrs[i], sizes[i]))
|
|
|
779 gomp_fatal ("[%p,%ld] is not mapped", hostaddrs[i],
|
|
|
780 (unsigned long) sizes[i]);
|
|
|
781 break;
|
|
|
782
|
|
|
783 default:
|
|
|
784 assert (0);
|
|
|
785 break;
|
|
|
786 }
|
|
|
787 }
|
|
|
788 }
|
