jobs.h 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2009-2015 Université de Bordeaux
  4. * Copyright (C) 2010, 2011, 2013, 2014, 2015 CNRS
  5. * Copyright (C) 2011 Télécom-SudParis
  6. * Copyright (C) 2014 INRIA
  7. *
  8. * StarPU is free software; you can redistribute it and/or modify
  9. * it under the terms of the GNU Lesser General Public License as published by
  10. * the Free Software Foundation; either version 2.1 of the License, or (at
  11. * your option) any later version.
  12. *
  13. * StarPU is distributed in the hope that it will be useful, but
  14. * WITHOUT ANY WARRANTY; without even the implied warranty of
  15. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  16. *
  17. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  18. */
  19. #ifndef __JOBS_H__
  20. #define __JOBS_H__
  21. #include <starpu.h>
  22. #include <semaphore.h>
  23. #include <stdio.h>
  24. #include <stdlib.h>
  25. #include <stdint.h>
  26. #include <string.h>
  27. #include <stdarg.h>
  28. #include <common/config.h>
  29. #ifdef HAVE_UNISTD_H
  30. #include <unistd.h>
  31. #endif
  32. #include <common/timing.h>
  33. #include <common/list.h>
  34. #include <common/fxt.h>
  35. #include <core/dependencies/tags.h>
  36. #include <datawizard/datawizard.h>
  37. #include <core/perfmodel/perfmodel.h>
  38. #include <core/errorcheck.h>
  39. #include <common/barrier.h>
  40. #include <common/utils.h>
  41. #ifdef STARPU_USE_CUDA
  42. #include <cuda.h>
  43. #endif
  44. struct _starpu_worker;
  45. /* codelet function */
  46. typedef void (*_starpu_cl_func_t)(void **, void *);
  47. #define _STARPU_CPU_MAY_PERFORM(j) ((j)->task->cl->where & STARPU_CPU)
  48. #define _STARPU_CUDA_MAY_PERFORM(j) ((j)->task->cl->where & STARPU_CUDA)
  49. #define _STARPU_OPENCL_MAY_PERFORM(j) ((j)->task->cl->where & STARPU_OPENCL)
  50. #define _STARPU_MIC_MAY_PERFORM(j) ((j)->task->cl->where & STARPU_MIC)
  51. #define _STARPU_SCC_MAY_PERFORM(j) ((j)->task->cl->where & STARPU_SCC)
  52. struct _starpu_data_descr
  53. {
  54. starpu_data_handle_t handle;
  55. enum starpu_data_access_mode mode;
  56. int node;
  57. };
  58. /* A job is the internal representation of a task. */
  59. LIST_TYPE(_starpu_job,
  60. /* Each job is attributed a unique id. */
  61. unsigned long job_id;
  62. /* The task associated to that job */
  63. struct starpu_task *task;
  64. /* These synchronization structures are used to wait for the job to be
  65. * available or terminated for instance. */
  66. starpu_pthread_mutex_t sync_mutex;
  67. starpu_pthread_cond_t sync_cond;
  68. /* To avoid deadlocks, we reorder the different buffers accessed to by
  69. * the task so that we always grab the rw-lock associated to the
  70. * handles in the same order. */
  71. struct _starpu_data_descr ordered_buffers[STARPU_NMAXBUFS];
  72. struct _starpu_task_wrapper_dlist dep_slots[STARPU_NMAXBUFS];
  73. struct _starpu_data_descr *dyn_ordered_buffers;
  74. struct _starpu_task_wrapper_dlist *dyn_dep_slots;
  75. /* If a tag is associated to the job, this points to the internal data
  76. * structure that describes the tag status. */
  77. struct _starpu_tag *tag;
  78. /* Maintain a list of all the completion groups that depend on the job.
  79. * */
  80. struct _starpu_cg_list job_successors;
  81. /* For tasks with cl==NULL but submitted with explicit data dependency,
  82. * the handle for this dependency, so as to remove the task from the
  83. * last_writer/readers */
  84. starpu_data_handle_t implicit_dep_handle;
  85. struct _starpu_task_wrapper_dlist implicit_dep_slot;
  86. /* Indicates whether the task associated to that job has already been
  87. * submitted to StarPU (1) or not (0) (using starpu_task_submit).
  88. * Becomes and stays 2 when the task is submitted several times.
  89. *
  90. * Protected by j->sync_mutex.
  91. */
  92. unsigned submitted:2;
  93. /* Indicates whether the task associated to this job is terminated or
  94. * not.
  95. *
  96. * Protected by j->sync_mutex.
  97. */
  98. unsigned terminated:2;
  99. #ifdef STARPU_OPENMP
  100. /* Job is a continuation or a regular task. */
  101. unsigned continuation;
  102. /* If 0, the prepared continuation is not resubmitted automatically
  103. * when going to sleep, if 1, the prepared continuation is immediately
  104. * resubmitted when going to sleep. */
  105. unsigned continuation_resubmit;
  106. /* Callback function called when:
  107. * - The continuation starpu task is ready to be submitted again if
  108. * continuation_resubmit = 0;
  109. * - The continuation starpu task has just been re-submitted if
  110. * continuation_resubmit = 1. */
  111. void (*continuation_callback_on_sleep)(void *arg);
  112. void *continuation_callback_on_sleep_arg;
  113. void (*omp_cleanup_callback)(void *arg);
  114. void *omp_cleanup_callback_arg;
  115. /* Job has been stopped at least once. */
  116. unsigned discontinuous;
  117. /* Cumulated execution time for discontinuous jobs */
  118. struct timespec cumulated_ts;
  119. /* Cumulated power consumption for discontinuous jobs */
  120. double cumulated_power_consumed;
  121. #endif
  122. /* The value of the footprint that identifies the job may be stored in
  123. * this structure. */
  124. uint32_t footprint;
  125. unsigned footprint_is_computed:1;
  126. /* Should that task appear in the debug tools ? (eg. the DAG generated
  127. * with dot) */
  128. unsigned exclude_from_dag:1;
  129. /* Is that task internal to StarPU? */
  130. unsigned internal:1;
  131. /* During the reduction of a handle, StarPU may have to submit tasks to
  132. * perform the reduction itself: those task should not be stalled while
  133. * other tasks are blocked until the handle has been properly reduced,
  134. * so we need a flag to differentiate them from "normal" tasks. */
  135. unsigned reduction_task:1;
  136. /* The implementation associated to the job */
  137. unsigned nimpl;
  138. /* Number of workers executing that task (>1 if the task is parallel)
  139. * */
  140. int task_size;
  141. /* In case we have assigned this job to a combined workerid */
  142. int combined_workerid;
  143. /* How many workers are currently running an alias of that job (for
  144. * parallel tasks only). */
  145. int active_task_alias_count;
  146. /* Used to record codelet start time instead of using a
  147. * local variable */
  148. struct timespec cl_start;
  149. struct bound_task *bound_task;
  150. /* Parallel workers may have to synchronize before/after the execution of a parallel task. */
  151. starpu_pthread_barrier_t before_work_barrier;
  152. starpu_pthread_barrier_t after_work_barrier;
  153. unsigned after_work_busy_barrier;
  154. #ifdef STARPU_DEBUG
  155. /* Linked-list of all jobs, for debugging */
  156. struct _starpu_job *prev_all;
  157. struct _starpu_job *next_all;
  158. #endif
  159. )
  160. /* Create an internal struct _starpu_job *structure to encapsulate the task. */
  161. struct _starpu_job* STARPU_ATTRIBUTE_MALLOC _starpu_job_create(struct starpu_task *task);
  162. /* Destroy the data structure associated to the job structure */
  163. void _starpu_job_destroy(struct _starpu_job *j);
  164. /* Test for the termination of the job */
  165. int _starpu_job_finished(struct _starpu_job *j);
  166. /* Wait for the termination of the job */
  167. void _starpu_wait_job(struct _starpu_job *j);
  168. #ifdef STARPU_OPENMP
  169. /* Test for the termination of the job */
  170. int _starpu_test_job_termination(struct _starpu_job *j);
  171. /* Prepare the job for accepting new dependencies before becoming a continuation. */
  172. void _starpu_job_prepare_for_continuation_ext(struct _starpu_job *j, unsigned continuation_resubmit,
  173. void (*continuation_callback_on_sleep)(void *arg), void *continuation_callback_on_sleep_arg);
  174. void _starpu_job_prepare_for_continuation(struct _starpu_job *j);
  175. void _starpu_job_set_omp_cleanup_callback(struct _starpu_job *j,
  176. void (*omp_cleanup_callback)(void *arg), void *omp_cleanup_callback_arg);
  177. #endif
  178. /* Specify that the task should not appear in the DAG generated by debug tools. */
  179. void _starpu_exclude_task_from_dag(struct starpu_task *task);
  180. /* try to submit job j, enqueue it if it's not schedulable yet. The job's sync mutex is supposed to be held already */
  181. unsigned _starpu_enforce_deps_and_schedule(struct _starpu_job *j);
  182. unsigned _starpu_enforce_deps_starting_from_task(struct _starpu_job *j);
  183. #ifdef STARPU_OPENMP
  184. /* When waking up a continuation, we only enforce new task dependencies */
  185. unsigned _starpu_reenforce_task_deps_and_schedule(struct _starpu_job *j);
  186. #endif
  187. /* Called at the submission of the job */
  188. void _starpu_handle_job_submission(struct _starpu_job *j);
  189. /* This function must be called after the execution of a job, this triggers all
  190. * job's dependencies and perform the callback function if any. */
  191. void _starpu_handle_job_termination(struct _starpu_job *j);
  192. /* Get the sum of the size of the data accessed by the job. */
  193. size_t _starpu_job_get_data_size(struct starpu_perfmodel *model, struct starpu_perfmodel_arch* arch, unsigned nimpl, struct _starpu_job *j);
  194. /* Get a task from the local pool of tasks that were explicitly attributed to
  195. * that worker. */
  196. struct starpu_task *_starpu_pop_local_task(struct _starpu_worker *worker);
  197. /* Put a task into the pool of tasks that are explicitly attributed to the
  198. * specified worker. If "back" is set, the task is put at the back of the list.
  199. * Considering the tasks are popped from the back, this value should be 0 to
  200. * enforce a FIFO ordering. */
  201. int _starpu_push_local_task(struct _starpu_worker *worker, struct starpu_task *task, int prio);
  202. #define _STARPU_JOB_GET_ORDERED_BUFFER_HANDLE(job, i) ((job->dyn_ordered_buffers) ? job->dyn_ordered_buffers[i].handle : job->ordered_buffers[i].handle)
  203. #define _STARPU_JOB_GET_ORDERED_BUFFER_MODE(job, i) ((job->dyn_ordered_buffers) ? job->dyn_ordered_buffers[i].mode : job->ordered_buffers[i].mode)
  204. #define _STARPU_JOB_GET_ORDERED_BUFFER_NODE(job, i) ((job->dyn_ordered_buffers) ? job->dyn_ordered_buffers[i].node : job->ordered_buffers[i].node)
  205. #define _STARPU_JOB_SET_ORDERED_BUFFER_HANDLE(job, handle, i) do { if (job->dyn_ordered_buffers) job->dyn_ordered_buffers[i].handle = (handle); else job->ordered_buffers[i].handle = (handle);} while(0)
  206. #define _STARPU_JOB_SET_ORDERED_BUFFER_MODE(job, __mode, i) do { if (job->dyn_ordered_buffers) job->dyn_ordered_buffers[i].mode = __mode; else job->ordered_buffers[i].mode = __mode;} while(0)
  207. #define _STARPU_JOB_SET_ORDERED_BUFFER_NODE(job, __node, i) do { if (job->dyn_ordered_buffers) job->dyn_ordered_buffers[i].node = __node; else job->ordered_buffers[i].node = __node;} while(0)
  208. #define _STARPU_JOB_SET_ORDERED_BUFFER(job, buffer, i) do { if (job->dyn_ordered_buffers) job->dyn_ordered_buffers[i] = buffer; else job->ordered_buffers[i] = buffer;} while(0)
  209. #define _STARPU_JOB_GET_ORDERED_BUFFERS(job) (job->dyn_ordered_buffers) ? job->dyn_ordered_buffers : job->ordered_buffers
  210. #define _STARPU_JOB_GET_DEP_SLOTS(job) (((job)->dyn_dep_slots) ? (job)->dyn_dep_slots : (job)->dep_slots)
  211. #endif // __JOBS_H__