starpu_replay_sched.c 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2016-2021 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  4. * Copyright (C) 2017 Erwan Leria
  5. *
  6. * StarPU is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU Lesser General Public License as published by
  8. * the Free Software Foundation; either version 2.1 of the License, or (at
  9. * your option) any later version.
  10. *
  11. * StarPU is distributed in the hope that it will be useful, but
  12. * WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  14. *
  15. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  16. */
  17. /*
  18. * This reads a sched.rec file and mangles submitted tasks according to the hint
  19. * from that file.
  20. */
  21. #include <starpu.h>
  22. #include <unistd.h>
  23. #include <stdio.h>
  24. #include <math.h>
  25. #include <common/uthash.h>
  26. #include <common/list.h>
  27. #include <common/utils.h>
  28. #include <limits.h>
  29. /*
  30. sched.rec files look like this:
  31. SubmitOrder: 1234
  32. Priority: 12
  33. SpecificWorker: 1
  34. Workers: 0 1 2
  35. DependsOn: 1235
  36. Prefetch: 1234
  37. DependsOn: 1233
  38. MemoryNode: 1
  39. Parameters: 1
  40. */
  41. #define CPY(src, dst, n) memcpy(dst, src, n * sizeof(*dst))
  42. #if 0
  43. #define debug(fmt, ...) fprintf(stderr, fmt, ##__VA_ARGS__)
  44. #else
  45. #define debug(fmt, ...) (void)0
  46. #endif
  47. static unsigned long submitorder; /* Also use as prefetchtag */
  48. static int priority;
  49. static int eosw;
  50. static unsigned workerorder;
  51. static int memnode;
  52. /* FIXME: MAXs */
  53. static uint32_t workers[STARPU_NMAXWORKERS/32];
  54. static unsigned nworkers;
  55. static unsigned dependson[STARPU_NMAXBUFS];
  56. static unsigned ndependson;
  57. static unsigned params[STARPU_NMAXBUFS];
  58. static unsigned nparams;
  59. static enum sched_type
  60. {
  61. NormalTask,
  62. PrefetchTask,
  63. } sched_type;
  64. static struct starpu_codelet cl_prefetch =
  65. {
  66. .where = STARPU_NOWHERE,
  67. .nbuffers = 1,
  68. .modes = { STARPU_R },
  69. };
  70. static struct task
  71. {
  72. UT_hash_handle hh;
  73. unsigned long submitorder;
  74. int priority;
  75. int memnode;
  76. unsigned dependson[STARPU_NMAXBUFS];
  77. unsigned ndependson;
  78. struct starpu_task *depends_tasks[STARPU_NMAXBUFS];
  79. /* For real tasks */
  80. int eosw;
  81. unsigned workerorder;
  82. uint32_t workers[STARPU_NMAXWORKERS/32];
  83. unsigned nworkers;
  84. /* For prefetch tasks */
  85. unsigned params[STARPU_NMAXBUFS];
  86. unsigned nparams;
  87. struct starpu_task *pref_task; /* Actual prefetch task */
  88. } *mangled_tasks, *prefetch_tasks;
  89. LIST_TYPE(dep,
  90. struct task *task;
  91. unsigned i;
  92. );
  93. struct deps
  94. {
  95. UT_hash_handle hh;
  96. unsigned long submitorder;
  97. struct dep_list list;
  98. } *dependencies = NULL;
  99. static void reset(void)
  100. {
  101. submitorder = 0;
  102. priority = INT_MIN;
  103. eosw = -1;
  104. memset(&workers, 0, sizeof(workers));
  105. nworkers = 0;
  106. ndependson = 0;
  107. sched_type = NormalTask;
  108. nparams = 0;
  109. memnode = -1;
  110. workerorder = 0;
  111. }
  112. /* TODO : respecter l'ordre de soumission des tâches SubmitOrder */
  113. static void checkField(char * s)
  114. {
  115. /* Record various information */
  116. #define TEST(field) (!strncmp(s, field": ", strlen(field) + 2))
  117. if (TEST("SubmitOrder"))
  118. {
  119. s = s + strlen("SubmitOrder: ");
  120. submitorder = strtol(s, NULL, 10);
  121. }
  122. else if (TEST("Priority"))
  123. {
  124. s = s + strlen("Priority: ");
  125. priority = strtol(s, NULL, 10);
  126. }
  127. else if (TEST("SpecificWorker"))
  128. {
  129. s = s + strlen("SpecificWorker: ");
  130. eosw = strtol(s, NULL, 10);
  131. }
  132. else if (TEST("Workers"))
  133. {
  134. s = s + strlen("Workers: ");
  135. char * delim = " ";
  136. char * token = strtok(s, delim);
  137. int i = 0;
  138. while (token != NULL)
  139. {
  140. int k = strtol(token, NULL, 10);
  141. STARPU_ASSERT_MSG(k < STARPU_NMAXWORKERS, "%d is bigger than maximum %d\n", k, STARPU_NMAXWORKERS);
  142. workers[k/(sizeof(*workers)*8)] |= (1 << (k%(sizeof(*workers)*8)));
  143. i++;
  144. token = strtok(NULL, delim);
  145. }
  146. nworkers = i;
  147. }
  148. else if (TEST("DependsOn"))
  149. {
  150. /* NOTE : dependsons (in the sched.rec) should be the submit orders of the dependencies,
  151. otherwise it can occur an undefined behaviour
  152. (contrary to the tasks.rec where dependencies are jobids */
  153. unsigned i = 0;
  154. char * delim = " ";
  155. char * token = strtok(s+strlen("DependsOn: "), delim);
  156. while (token != NULL)
  157. {
  158. dependson[i] = strtol(token, NULL, 10);
  159. i++;
  160. token = strtok(NULL, delim);
  161. }
  162. ndependson = i;
  163. }
  164. else if (TEST("Prefetch"))
  165. {
  166. s = s + strlen("Prefetch: ");
  167. submitorder = strtol(s, NULL, 10);
  168. sched_type = PrefetchTask;
  169. }
  170. else if (TEST("Parameters"))
  171. {
  172. s = s + strlen("Parameters: ");
  173. char * delim = " ";
  174. char * token = strtok(s, delim);
  175. int i = 0;
  176. while (token != NULL)
  177. {
  178. params[i] = strtol(token, NULL, 10);
  179. i++;
  180. token = strtok(NULL, delim);
  181. }
  182. nparams = i;
  183. }
  184. else if (TEST("MemoryNode"))
  185. {
  186. s = s + strlen("MemoryNode: ");
  187. memnode = strtol(s, NULL, 10);
  188. }
  189. else if (TEST("Workerorder"))
  190. {
  191. s = s + strlen("Workerorder: ");
  192. workerorder = strtol(s, NULL, 10);
  193. }
  194. }
  195. void schedRecInit(const char * filename)
  196. {
  197. FILE * f = fopen(filename, "r");
  198. if(f == NULL)
  199. {
  200. fprintf(stderr,"unable to open file %s: %s\n", filename, strerror(errno));
  201. return;
  202. }
  203. size_t lnsize = 128;
  204. char *s;
  205. _STARPU_MALLOC(s, sizeof(*s) * lnsize);
  206. int eof = 0;
  207. reset();
  208. while(!eof && !feof(f))
  209. {
  210. char *ln;
  211. /* Get the line */
  212. if (!fgets(s, lnsize, f))
  213. {
  214. eof = 1;
  215. }
  216. while (!(ln = strchr(s, '\n')))
  217. {
  218. _STARPU_REALLOC(s, lnsize * 2);
  219. if (!fgets(s + lnsize-1, lnsize+1, f))
  220. {
  221. eof = 1;
  222. break;
  223. }
  224. lnsize *= 2;
  225. }
  226. if ((ln == s || eof) && submitorder)
  227. {
  228. /* Empty line, doit */
  229. struct task * task;
  230. unsigned i;
  231. _STARPU_MALLOC(task, sizeof(*task));
  232. task->submitorder = submitorder;
  233. task->priority = priority;
  234. task->memnode = memnode;
  235. CPY(dependson, task->dependson, ndependson);
  236. task->ndependson = ndependson;
  237. /* Also record submitorder of tasks that this one will need to depend on */
  238. for (i = 0; i < ndependson; i++)
  239. {
  240. struct dep *dep;
  241. struct starpu_task *starpu_task;
  242. _STARPU_MALLOC(dep, sizeof(*dep));
  243. dep->task = task;
  244. dep->i = i;
  245. struct deps *deps;
  246. HASH_FIND(hh, dependencies, &task->dependson[i], sizeof(submitorder), deps);
  247. if (!deps)
  248. {
  249. /* No task depends on this one yet, add a cell for it */
  250. _STARPU_MALLOC(deps, sizeof(*deps));
  251. dep_list_init(&deps->list);
  252. deps->submitorder = task->dependson[i];
  253. HASH_ADD(hh, dependencies, submitorder, sizeof(submitorder), deps);
  254. }
  255. dep_list_push_back(&deps->list, dep);
  256. /* Create the intermediate task */
  257. starpu_task = dep->task->depends_tasks[i] = starpu_task_create();
  258. starpu_task->cl = NULL;
  259. starpu_task->destroy = 0;
  260. starpu_task->no_submitorder = 1;
  261. }
  262. switch (sched_type)
  263. {
  264. case NormalTask:
  265. /* A new task to mangle, record what needs to be done */
  266. task->eosw = eosw;
  267. task->workerorder = workerorder;
  268. CPY(workers, task->workers, STARPU_NMAXWORKERS/32);
  269. task->nworkers = nworkers;
  270. STARPU_ASSERT(nparams == 0);
  271. debug("adding mangled task %lu\n", submitorder);
  272. HASH_ADD(hh, mangled_tasks, submitorder, sizeof(submitorder), task);
  273. break;
  274. case PrefetchTask:
  275. STARPU_ASSERT(memnode >= 0);
  276. STARPU_ASSERT(eosw == -1);
  277. STARPU_ASSERT(workerorder == 0);
  278. STARPU_ASSERT(nworkers == 0);
  279. CPY(params, task->params, nparams);
  280. task->nparams = nparams;
  281. /* TODO: more params */
  282. STARPU_ASSERT_MSG(nparams == 1, "only supports one parameter at a time");
  283. debug("adding prefetch task for %lu\n", submitorder);
  284. HASH_ADD(hh, prefetch_tasks, submitorder, sizeof(submitorder), task);
  285. break;
  286. default:
  287. STARPU_ASSERT(0);
  288. break;
  289. }
  290. reset();
  291. }
  292. else checkField(s);
  293. }
  294. fclose(f);
  295. free(s);
  296. }
  297. static void do_prefetch(void *arg)
  298. {
  299. unsigned node = (uintptr_t) arg;
  300. starpu_data_idle_prefetch_on_node(starpu_task_get_current()->handles[0], node, 1);
  301. }
  302. void applySchedRec(struct starpu_task *starpu_task, unsigned long submit_order)
  303. {
  304. struct task *task;
  305. struct deps *deps;
  306. int ret;
  307. HASH_FIND(hh, dependencies, &submit_order, sizeof(submit_order), deps);
  308. if (deps)
  309. {
  310. struct dep *dep;
  311. for (dep = dep_list_begin(&deps->list);
  312. dep != dep_list_end(&deps->list);
  313. dep = dep_list_next(dep))
  314. {
  315. debug("task %lu is %d-th dep for %lu\n", submit_order, dep->i, dep->task->submitorder);
  316. /* Some task will depend on this one, make the dependency */
  317. starpu_task_declare_deps_array(dep->task->depends_tasks[dep->i], 1, &starpu_task);
  318. ret = starpu_task_submit(dep->task->depends_tasks[dep->i]);
  319. STARPU_ASSERT(ret == 0);
  320. }
  321. }
  322. HASH_FIND(hh, prefetch_tasks, &submit_order, sizeof(submit_order), task);
  323. if (task)
  324. {
  325. /* We want to submit a prefetch for this task */
  326. debug("task %lu has a prefetch for parameter %d to node %d\n", submit_order, task->params[0], task->memnode);
  327. struct starpu_task *pref_task;
  328. pref_task = task->pref_task = starpu_task_create();
  329. pref_task->cl = &cl_prefetch;
  330. pref_task->destroy = 1;
  331. pref_task->no_submitorder = 1;
  332. pref_task->callback_arg = (void*)(uintptr_t) task->memnode;
  333. pref_task->callback_func = do_prefetch;
  334. /* TODO: more params */
  335. pref_task->handles[0] = starpu_task->handles[task->params[0]];
  336. /* Make it depend on intermediate tasks */
  337. if (task->ndependson)
  338. {
  339. debug("%u dependencies\n", task->ndependson);
  340. starpu_task_declare_deps_array(pref_task, task->ndependson, task->depends_tasks);
  341. }
  342. ret = starpu_task_submit(pref_task);
  343. STARPU_ASSERT(ret == 0);
  344. }
  345. HASH_FIND(hh, mangled_tasks, &submit_order, sizeof(submit_order), task);
  346. if (task == NULL)
  347. /* Nothing to do for this */
  348. return;
  349. debug("mangling task %lu\n", submit_order);
  350. if (task->eosw >= 0)
  351. {
  352. debug("execute on a specific worker %d\n", task->eosw);
  353. starpu_task->workerid = task->eosw;
  354. starpu_task->execute_on_a_specific_worker = 1;
  355. }
  356. if (task->workerorder > 0)
  357. {
  358. debug("workerorder %d\n", task->workerorder);
  359. starpu_task->workerorder = task->workerorder;
  360. }
  361. if (task->priority != INT_MIN)
  362. {
  363. debug("priority %d\n", task->priority);
  364. starpu_task->priority = task->priority;
  365. }
  366. if (task->nworkers)
  367. {
  368. debug("%u workers %x\n", task->nworkers, task->workers[0]);
  369. starpu_task->workerids_len = sizeof(task->workers) / sizeof(task->workers[0]);
  370. _STARPU_MALLOC(starpu_task->workerids, task->nworkers * sizeof(*starpu_task->workerids));
  371. CPY(task->workers, starpu_task->workerids, STARPU_NMAXWORKERS/32);
  372. }
  373. if (task->ndependson)
  374. {
  375. debug("%u dependencies\n", task->ndependson);
  376. starpu_task_declare_deps_array(starpu_task, task->ndependson, task->depends_tasks);
  377. }
  378. /* And now, let it go! */
  379. }