starpu_task_insert_utils.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2011, 2013-2015 Université Bordeaux
  4. * Copyright (C) 2011-2015 CNRS
  5. * Copyright (C) 2011, 2014 INRIA
  6. *
  7. * StarPU is free software; you can redistribute it and/or modify
  8. * it under the terms of the GNU Lesser General Public License as published by
  9. * the Free Software Foundation; either version 2.1 of the License, or (at
  10. * your option) any later version.
  11. *
  12. * StarPU is distributed in the hope that it will be useful, but
  13. * WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  15. *
  16. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  17. */
  18. #include <util/starpu_task_insert_utils.h>
  19. #include <common/config.h>
  20. #include <common/utils.h>
  21. #include <core/task.h>
  22. /* Deal with callbacks. The unpack function may be called multiple times when
  23. * we have a parallel task, and we should not free the cl_arg parameter from
  24. * the callback function. */
  25. struct _starpu_task_insert_cb_wrapper
  26. {
  27. _starpu_callback_func_t callback_func;
  28. void *callback_arg;
  29. };
  30. static
  31. void _starpu_task_insert_callback_wrapper(void *_cl_arg_wrapper)
  32. {
  33. struct _starpu_task_insert_cb_wrapper *cl_arg_wrapper = (struct _starpu_task_insert_cb_wrapper *) _cl_arg_wrapper;
  34. /* Execute the callback specified by the application */
  35. if (cl_arg_wrapper->callback_func)
  36. cl_arg_wrapper->callback_func(cl_arg_wrapper->callback_arg);
  37. }
  38. int _starpu_codelet_pack_args(void **arg_buffer, size_t *arg_buffer_size, va_list varg_list)
  39. {
  40. int arg_type;
  41. int nargs = 0;
  42. char *_arg_buffer = NULL; // We would like a void* but we use a char* to allow pointer arithmetic
  43. size_t _arg_buffer_size = 0;
  44. size_t current_offset = sizeof(nargs);
  45. while((arg_type = va_arg(varg_list, int)) != 0)
  46. {
  47. if (arg_type & STARPU_R || arg_type & STARPU_W || arg_type & STARPU_SCRATCH || arg_type & STARPU_REDUX)
  48. {
  49. (void)va_arg(varg_list, starpu_data_handle_t);
  50. }
  51. else if (arg_type==STARPU_DATA_ARRAY)
  52. {
  53. (void)va_arg(varg_list, starpu_data_handle_t*);
  54. (void)va_arg(varg_list, int);
  55. }
  56. else if (arg_type==STARPU_DATA_MODE_ARRAY)
  57. {
  58. (void)va_arg(varg_list, struct starpu_data_descr*);
  59. (void)va_arg(varg_list, int);
  60. }
  61. else if (arg_type==STARPU_VALUE)
  62. {
  63. /* We have a constant value: this should be followed by a pointer to the cst value and the size of the constant */
  64. void *ptr = va_arg(varg_list, void *);
  65. size_t ptr_size = va_arg(varg_list, size_t);
  66. nargs++;
  67. if (current_offset > _arg_buffer_size)
  68. {
  69. if (_arg_buffer_size == 0) _arg_buffer_size = 1024; else _arg_buffer_size *= 2;
  70. _arg_buffer = realloc(_arg_buffer, _arg_buffer_size);
  71. }
  72. memcpy(_arg_buffer+current_offset, (void *)&ptr_size, sizeof(ptr_size));
  73. current_offset += sizeof(ptr_size);
  74. memcpy(_arg_buffer+current_offset, ptr, ptr_size);
  75. current_offset += ptr_size;
  76. STARPU_ASSERT(current_offset <= _arg_buffer_size);
  77. }
  78. else if (arg_type==STARPU_CALLBACK)
  79. {
  80. (void)va_arg(varg_list, _starpu_callback_func_t);
  81. }
  82. else if (arg_type==STARPU_CALLBACK_WITH_ARG)
  83. {
  84. va_arg(varg_list, _starpu_callback_func_t);
  85. va_arg(varg_list, void *);
  86. }
  87. else if (arg_type==STARPU_CALLBACK_ARG)
  88. {
  89. (void)va_arg(varg_list, void *);
  90. }
  91. else if (arg_type==STARPU_PROLOGUE_CALLBACK)
  92. {
  93. va_arg(varg_list, _starpu_callback_func_t);
  94. }
  95. else if (arg_type==STARPU_PROLOGUE_CALLBACK_ARG)
  96. {
  97. (void)va_arg(varg_list, void *);
  98. }
  99. else if (arg_type==STARPU_PROLOGUE_CALLBACK_POP)
  100. {
  101. va_arg(varg_list, _starpu_callback_func_t);
  102. }
  103. else if (arg_type==STARPU_PROLOGUE_CALLBACK_POP_ARG)
  104. {
  105. (void)va_arg(varg_list, void *);
  106. }
  107. else if (arg_type==STARPU_PRIORITY)
  108. {
  109. (void)va_arg(varg_list, int);
  110. }
  111. else if (arg_type==STARPU_EXECUTE_ON_NODE)
  112. {
  113. (void)va_arg(varg_list, int);
  114. }
  115. else if (arg_type==STARPU_EXECUTE_ON_DATA)
  116. {
  117. (void)va_arg(varg_list, starpu_data_handle_t);
  118. }
  119. else if (arg_type==STARPU_EXECUTE_ON_WORKER)
  120. {
  121. va_arg(varg_list, int);
  122. }
  123. else if (arg_type==STARPU_WORKER_ORDER)
  124. {
  125. va_arg(varg_list, unsigned);
  126. }
  127. else if (arg_type==STARPU_SCHED_CTX)
  128. {
  129. (void)va_arg(varg_list, unsigned);
  130. }
  131. else if (arg_type==STARPU_HYPERVISOR_TAG)
  132. {
  133. (void)va_arg(varg_list, int);
  134. }
  135. else if (arg_type==STARPU_POSSIBLY_PARALLEL)
  136. {
  137. (void)va_arg(varg_list, unsigned);
  138. }
  139. else if (arg_type==STARPU_FLOPS)
  140. {
  141. (void)va_arg(varg_list, double);
  142. }
  143. else if (arg_type==STARPU_TAG || arg_type==STARPU_TAG_ONLY)
  144. {
  145. (void)va_arg(varg_list, starpu_tag_t);
  146. }
  147. else if (arg_type==STARPU_NODE_SELECTION_POLICY)
  148. {
  149. (void)va_arg(varg_list, int);
  150. }
  151. else
  152. {
  153. STARPU_ABORT_MSG("Unrecognized argument %d\n", arg_type);
  154. }
  155. }
  156. if (nargs)
  157. {
  158. memcpy(_arg_buffer, (int *)&nargs, sizeof(nargs));
  159. }
  160. else
  161. {
  162. free(_arg_buffer);
  163. _arg_buffer = NULL;
  164. }
  165. *arg_buffer = _arg_buffer;
  166. *arg_buffer_size = _arg_buffer_size;
  167. return 0;
  168. }
  169. static
  170. void _starpu_task_insert_check_nb_buffers(struct starpu_codelet *cl, struct starpu_task **task, int *allocated_buffers, int current_buffer)
  171. {
  172. if (current_buffer >= STARPU_NMAXBUFS)
  173. {
  174. if (*allocated_buffers == 0)
  175. {
  176. int i;
  177. struct starpu_codelet *cl2 = (*task)->cl;
  178. *allocated_buffers = STARPU_NMAXBUFS * 2;
  179. (*task)->dyn_handles = malloc(*allocated_buffers * sizeof(starpu_data_handle_t));
  180. for(i=0 ; i<current_buffer ; i++)
  181. {
  182. (*task)->dyn_handles[i] = (*task)->handles[i];
  183. }
  184. if (cl2->nbuffers == STARPU_VARIABLE_NBUFFERS || !cl2->dyn_modes)
  185. {
  186. (*task)->dyn_modes = malloc(*allocated_buffers * sizeof(enum starpu_data_access_mode));
  187. for(i=0 ; i<current_buffer ; i++)
  188. {
  189. (*task)->dyn_modes[i] = (*task)->modes[i];
  190. }
  191. }
  192. }
  193. else if (current_buffer >= *allocated_buffers)
  194. {
  195. *allocated_buffers *= 2;
  196. (*task)->dyn_handles = realloc((*task)->dyn_handles, *allocated_buffers * sizeof(starpu_data_handle_t));
  197. if (cl->nbuffers == STARPU_VARIABLE_NBUFFERS || !cl->dyn_modes)
  198. (*task)->dyn_modes = realloc((*task)->dyn_modes, *allocated_buffers * sizeof(enum starpu_data_access_mode));
  199. }
  200. }
  201. }
  202. void _starpu_task_insert_create(struct starpu_codelet *cl, struct starpu_task **task, va_list varg_list)
  203. {
  204. int arg_type;
  205. char *arg_buffer_ = NULL;
  206. size_t arg_buffer_size_ = 0;
  207. size_t current_offset = sizeof(int);
  208. int current_buffer;
  209. int nargs = 0;
  210. int allocated_buffers = 0;
  211. struct _starpu_task_insert_cb_wrapper *cl_arg_wrapper = (struct _starpu_task_insert_cb_wrapper *) malloc(sizeof(struct _starpu_task_insert_cb_wrapper));
  212. STARPU_ASSERT(cl_arg_wrapper);
  213. cl_arg_wrapper->callback_func = NULL;
  214. struct _starpu_task_insert_cb_wrapper *prologue_cl_arg_wrapper = (struct _starpu_task_insert_cb_wrapper *) malloc(sizeof(struct _starpu_task_insert_cb_wrapper));
  215. STARPU_ASSERT(prologue_cl_arg_wrapper);
  216. prologue_cl_arg_wrapper->callback_func = NULL;
  217. struct _starpu_task_insert_cb_wrapper *prologue_pop_cl_arg_wrapper = (struct _starpu_task_insert_cb_wrapper *) malloc(sizeof(struct _starpu_task_insert_cb_wrapper));
  218. STARPU_ASSERT(prologue_pop_cl_arg_wrapper);
  219. prologue_pop_cl_arg_wrapper->callback_func = NULL;
  220. (*task)->cl = cl;
  221. current_buffer = 0;
  222. while((arg_type = va_arg(varg_list, int)) != 0)
  223. {
  224. if (arg_type & STARPU_R || arg_type & STARPU_W || arg_type & STARPU_SCRATCH || arg_type & STARPU_REDUX)
  225. {
  226. /* We have an access mode : we expect to find a handle */
  227. starpu_data_handle_t handle = va_arg(varg_list, starpu_data_handle_t);
  228. enum starpu_data_access_mode mode = (enum starpu_data_access_mode) arg_type & ~STARPU_SSEND;
  229. STARPU_ASSERT(cl != NULL);
  230. _starpu_task_insert_check_nb_buffers(cl, task, &allocated_buffers, current_buffer);
  231. STARPU_TASK_SET_HANDLE((*task), handle, current_buffer);
  232. if (cl->nbuffers == STARPU_VARIABLE_NBUFFERS || (cl->nbuffers > STARPU_NMAXBUFS && !cl->dyn_modes))
  233. STARPU_TASK_SET_MODE(*task, mode, current_buffer);
  234. else if (STARPU_CODELET_GET_MODE(cl, current_buffer))
  235. {
  236. STARPU_ASSERT_MSG(STARPU_CODELET_GET_MODE(cl, current_buffer) == mode,
  237. "The codelet <%s> defines the access mode %d for the buffer %d which is different from the mode %d given to starpu_task_insert\n",
  238. cl->name, STARPU_CODELET_GET_MODE(cl, current_buffer),
  239. current_buffer, mode);
  240. }
  241. else
  242. {
  243. #ifdef STARPU_DEVEL
  244. # warning shall we print a warning to the user
  245. /* Morse uses it to avoid having to set it in the codelet structure */
  246. #endif
  247. STARPU_CODELET_SET_MODE(cl, mode, current_buffer);
  248. }
  249. current_buffer++;
  250. }
  251. else if (arg_type == STARPU_DATA_ARRAY)
  252. {
  253. // Expect to find a array of handles and its size
  254. starpu_data_handle_t *handles = va_arg(varg_list, starpu_data_handle_t *);
  255. int nb_handles = va_arg(varg_list, int);
  256. STARPU_ASSERT(cl != NULL);
  257. int i;
  258. for(i=0 ; i<nb_handles ; i++)
  259. {
  260. _starpu_task_insert_check_nb_buffers(cl, task, &allocated_buffers, current_buffer);
  261. STARPU_TASK_SET_HANDLE((*task), handles[i], current_buffer);
  262. current_buffer++;
  263. }
  264. }
  265. else if (arg_type==STARPU_DATA_MODE_ARRAY)
  266. {
  267. // Expect to find a array of descr and its size
  268. struct starpu_data_descr *descrs = va_arg(varg_list, struct starpu_data_descr *);
  269. int nb_descrs = va_arg(varg_list, int);
  270. STARPU_ASSERT(cl != NULL);
  271. int i;
  272. for(i=0 ; i<nb_descrs ; i++)
  273. {
  274. _starpu_task_insert_check_nb_buffers(cl, task, &allocated_buffers, current_buffer);
  275. STARPU_TASK_SET_HANDLE((*task), descrs[i].handle, current_buffer);
  276. if ((*task)->dyn_modes)
  277. {
  278. (*task)->dyn_modes[i] = descrs[i].mode;
  279. }
  280. else if (cl->nbuffers == STARPU_VARIABLE_NBUFFERS || (cl->nbuffers > STARPU_NMAXBUFS && !cl->dyn_modes))
  281. STARPU_TASK_SET_MODE(*task, descrs[i].mode, current_buffer);
  282. else if (STARPU_CODELET_GET_MODE(cl, current_buffer))
  283. {
  284. STARPU_ASSERT_MSG(STARPU_CODELET_GET_MODE(cl, current_buffer) == descrs[i].mode,
  285. "The codelet <%s> defines the access mode %d for the buffer %d which is different from the mode %d given to starpu_task_insert\n",
  286. cl->name, STARPU_CODELET_GET_MODE(cl, current_buffer),
  287. current_buffer, descrs[i].mode);
  288. }
  289. else
  290. {
  291. STARPU_CODELET_SET_MODE(cl, descrs[i].mode, current_buffer);
  292. }
  293. current_buffer++;
  294. }
  295. }
  296. else if (arg_type==STARPU_VALUE)
  297. {
  298. void *ptr = va_arg(varg_list, void *);
  299. size_t ptr_size = va_arg(varg_list, size_t);
  300. nargs++;
  301. if (current_offset > arg_buffer_size_)
  302. {
  303. if (arg_buffer_size_ == 0) arg_buffer_size_ = 1024; else arg_buffer_size_ *= 2;
  304. arg_buffer_ = realloc(arg_buffer_, arg_buffer_size_);
  305. }
  306. memcpy(arg_buffer_+current_offset, (void *)&ptr_size, sizeof(ptr_size));
  307. current_offset += sizeof(ptr_size);
  308. memcpy(arg_buffer_+current_offset, ptr, ptr_size);
  309. current_offset += ptr_size;
  310. STARPU_ASSERT(current_offset <= arg_buffer_size_);
  311. }
  312. else if (arg_type==STARPU_CALLBACK)
  313. {
  314. void (*callback_func)(void *);
  315. callback_func = va_arg(varg_list, _starpu_callback_func_t);
  316. cl_arg_wrapper->callback_func = callback_func;
  317. }
  318. else if (arg_type==STARPU_CALLBACK_WITH_ARG)
  319. {
  320. void (*callback_func)(void *);
  321. void *callback_arg;
  322. callback_func = va_arg(varg_list, _starpu_callback_func_t);
  323. callback_arg = va_arg(varg_list, void *);
  324. cl_arg_wrapper->callback_func = callback_func;
  325. cl_arg_wrapper->callback_arg = callback_arg;
  326. }
  327. else if (arg_type==STARPU_CALLBACK_ARG)
  328. {
  329. void *callback_arg = va_arg(varg_list, void *);
  330. cl_arg_wrapper->callback_arg = callback_arg;
  331. }
  332. else if (arg_type==STARPU_PROLOGUE_CALLBACK)
  333. {
  334. void (*callback_func)(void *);
  335. callback_func = va_arg(varg_list, _starpu_callback_func_t);
  336. prologue_cl_arg_wrapper->callback_func = callback_func;
  337. }
  338. else if (arg_type==STARPU_PROLOGUE_CALLBACK_ARG)
  339. {
  340. void *callback_arg = va_arg(varg_list, void *);
  341. prologue_cl_arg_wrapper->callback_arg = callback_arg;
  342. }
  343. else if (arg_type==STARPU_PROLOGUE_CALLBACK_POP)
  344. {
  345. void (*callback_func)(void *);
  346. callback_func = va_arg(varg_list, _starpu_callback_func_t);
  347. prologue_pop_cl_arg_wrapper->callback_func = callback_func;
  348. }
  349. else if (arg_type==STARPU_PROLOGUE_CALLBACK_POP_ARG)
  350. {
  351. void *callback_arg = va_arg(varg_list, void *);
  352. prologue_pop_cl_arg_wrapper->callback_arg = callback_arg;
  353. }
  354. else if (arg_type==STARPU_PRIORITY)
  355. {
  356. /* Followed by a priority level */
  357. int prio = va_arg(varg_list, int);
  358. (*task)->priority = prio;
  359. }
  360. else if (arg_type==STARPU_EXECUTE_ON_NODE)
  361. {
  362. (void)va_arg(varg_list, int);
  363. }
  364. else if (arg_type==STARPU_EXECUTE_ON_DATA)
  365. {
  366. (void)va_arg(varg_list, starpu_data_handle_t);
  367. }
  368. else if (arg_type==STARPU_EXECUTE_ON_WORKER)
  369. {
  370. int worker = va_arg(varg_list, int);
  371. if (worker != -1)
  372. {
  373. (*task)->workerid = worker;
  374. (*task)->execute_on_a_specific_worker = 1;
  375. }
  376. }
  377. else if (arg_type==STARPU_WORKER_ORDER)
  378. {
  379. unsigned order = va_arg(varg_list, unsigned);
  380. if (order != 0)
  381. {
  382. STARPU_ASSERT_MSG((*task)->execute_on_a_specific_worker, "worker order only makes sense if a workerid is provided");
  383. (*task)->workerorder = order;
  384. }
  385. }
  386. else if (arg_type==STARPU_SCHED_CTX)
  387. {
  388. unsigned sched_ctx = va_arg(varg_list, unsigned);
  389. (*task)->sched_ctx = sched_ctx;
  390. }
  391. else if (arg_type==STARPU_HYPERVISOR_TAG)
  392. {
  393. int hypervisor_tag = va_arg(varg_list, int);
  394. (*task)->hypervisor_tag = hypervisor_tag;
  395. }
  396. else if (arg_type==STARPU_POSSIBLY_PARALLEL)
  397. {
  398. unsigned possibly_parallel = va_arg(varg_list, unsigned);
  399. (*task)->possibly_parallel = possibly_parallel;
  400. }
  401. else if (arg_type==STARPU_FLOPS)
  402. {
  403. double flops = va_arg(varg_list, double);
  404. (*task)->flops = flops;
  405. }
  406. else if (arg_type==STARPU_TAG)
  407. {
  408. starpu_tag_t tag = va_arg(varg_list, starpu_tag_t);
  409. (*task)->tag_id = tag;
  410. (*task)->use_tag = 1;
  411. }
  412. else if (arg_type==STARPU_TAG_ONLY)
  413. {
  414. starpu_tag_t tag = va_arg(varg_list, starpu_tag_t);
  415. (*task)->tag_id = tag;
  416. }
  417. else if (arg_type==STARPU_NODE_SELECTION_POLICY)
  418. {
  419. (void)va_arg(varg_list, int);
  420. }
  421. else
  422. {
  423. STARPU_ABORT_MSG("Unrecognized argument %d\n", arg_type);
  424. }
  425. }
  426. if (cl)
  427. {
  428. if (cl->nbuffers == STARPU_VARIABLE_NBUFFERS)
  429. {
  430. (*task)->nbuffers = current_buffer;
  431. }
  432. else
  433. {
  434. STARPU_ASSERT_MSG(current_buffer == cl->nbuffers, "Incoherent number of buffers between cl (%d) and number of parameters (%d)", cl->nbuffers, current_buffer);
  435. }
  436. }
  437. if (nargs)
  438. {
  439. memcpy(arg_buffer_, (int *)&nargs, sizeof(nargs));
  440. (*task)->cl_arg = arg_buffer_;
  441. (*task)->cl_arg_size = arg_buffer_size_;
  442. }
  443. else
  444. {
  445. free(arg_buffer_);
  446. arg_buffer_ = NULL;
  447. }
  448. /* The callback will free the argument stack and execute the
  449. * application's callback, if any. */
  450. (*task)->callback_func = _starpu_task_insert_callback_wrapper;
  451. (*task)->callback_arg = cl_arg_wrapper;
  452. (*task)->callback_arg_free = 1;
  453. (*task)->prologue_callback_func = _starpu_task_insert_callback_wrapper;
  454. (*task)->prologue_callback_arg = prologue_cl_arg_wrapper;
  455. (*task)->prologue_callback_arg_free = 1;
  456. (*task)->prologue_callback_pop_func = _starpu_task_insert_callback_wrapper;
  457. (*task)->prologue_callback_pop_arg = prologue_pop_cl_arg_wrapper;
  458. (*task)->prologue_callback_pop_arg_free = 1;
  459. }