ring_sync_detached.c 4.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2010-2017 CNRS
  4. * Copyright (C) 2009,2010,2014-2018 Université de Bordeaux
  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. #include <starpu_mpi.h>
  18. #include "helper.h"
  19. #ifdef STARPU_QUICK_CHECK
  20. # define NITER 32
  21. #elif !defined(STARPU_LONG_CHECK)
  22. # define NITER 256
  23. #else
  24. # define NITER 2048
  25. #endif
  26. #ifdef STARPU_USE_CUDA
  27. extern void increment_cuda(void *descr[], void *_args);
  28. #endif
  29. void increment_cpu(void *descr[], void *_args)
  30. {
  31. (void)_args;
  32. int *tokenptr = (int *)STARPU_VECTOR_GET_PTR(descr[0]);
  33. (*tokenptr)++;
  34. }
  35. static struct starpu_codelet increment_cl =
  36. {
  37. #ifdef STARPU_USE_CUDA
  38. .cuda_funcs = {increment_cuda},
  39. #endif
  40. .cpu_funcs = {increment_cpu},
  41. .nbuffers = 1,
  42. .modes = {STARPU_RW},
  43. .model = &starpu_perfmodel_nop,
  44. };
  45. void increment_token(starpu_data_handle_t handle)
  46. {
  47. struct starpu_task *task = starpu_task_create();
  48. task->cl = &increment_cl;
  49. task->handles[0] = handle;
  50. task->synchronous = 1;
  51. int ret = starpu_task_submit(task);
  52. STARPU_CHECK_RETURN_VALUE(ret, "starpu_task_submit");
  53. }
  54. static starpu_pthread_mutex_t mutex = STARPU_PTHREAD_MUTEX_INITIALIZER;
  55. static starpu_pthread_cond_t cond = STARPU_PTHREAD_COND_INITIALIZER;
  56. void callback(void *arg)
  57. {
  58. unsigned *completed = arg;
  59. STARPU_PTHREAD_MUTEX_LOCK(&mutex);
  60. *completed = 1;
  61. STARPU_PTHREAD_COND_SIGNAL(&cond);
  62. STARPU_PTHREAD_MUTEX_UNLOCK(&mutex);
  63. }
  64. int main(int argc, char **argv)
  65. {
  66. int ret, rank, size;
  67. int token = 42;
  68. starpu_data_handle_t token_handle;
  69. int mpi_init;
  70. MPI_INIT_THREAD(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_init);
  71. ret = starpu_mpi_init_conf(&argc, &argv, mpi_init, MPI_COMM_WORLD, NULL);
  72. STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
  73. starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
  74. starpu_mpi_comm_size(MPI_COMM_WORLD, &size);
  75. if (size < 2 || (starpu_cpu_worker_get_count() + starpu_cuda_worker_get_count() == 0))
  76. {
  77. if (rank == 0)
  78. {
  79. if (size < 2)
  80. FPRINTF(stderr, "We need at least 2 processes.\n");
  81. else
  82. FPRINTF(stderr, "We need at least 1 CPU or CUDA worker.\n");
  83. }
  84. starpu_mpi_shutdown();
  85. if (!mpi_init)
  86. MPI_Finalize();
  87. return STARPU_TEST_SKIPPED;
  88. }
  89. starpu_vector_data_register(&token_handle, 0, (uintptr_t)&token, 1, sizeof(token));
  90. int nloops = NITER;
  91. int loop;
  92. int last_loop = nloops - 1;
  93. int last_rank = size - 1;
  94. for (loop = 0; loop < nloops; loop++)
  95. {
  96. int tag = loop*size + rank;
  97. if (loop == 0 && rank == 0)
  98. {
  99. starpu_data_acquire(token_handle, STARPU_W);
  100. token = 0;
  101. FPRINTF_MPI(stderr, "Start with token value %d\n", token);
  102. starpu_data_release(token_handle);
  103. }
  104. else
  105. {
  106. MPI_Status status;
  107. starpu_mpi_recv(token_handle, (rank+size-1)%size, tag, MPI_COMM_WORLD, &status);
  108. }
  109. increment_token(token_handle);
  110. if (loop == last_loop && rank == last_rank)
  111. {
  112. starpu_data_acquire(token_handle, STARPU_R);
  113. FPRINTF_MPI(stderr, "Finished : token value %d\n", token);
  114. starpu_data_release(token_handle);
  115. }
  116. else
  117. {
  118. int sent = 0;
  119. starpu_mpi_issend_detached(token_handle, (rank+1)%size, tag+1, MPI_COMM_WORLD, callback, &sent);
  120. STARPU_PTHREAD_MUTEX_LOCK(&mutex);
  121. while (!sent)
  122. STARPU_PTHREAD_COND_WAIT(&cond, &mutex);
  123. STARPU_PTHREAD_MUTEX_UNLOCK(&mutex);
  124. }
  125. }
  126. starpu_data_unregister(token_handle);
  127. starpu_mpi_shutdown();
  128. FPRINTF_MPI(stderr, "Final value for token %d\n", token);
  129. if (!mpi_init)
  130. MPI_Finalize();
  131. #ifndef STARPU_SIMGRID
  132. if (rank == last_rank)
  133. {
  134. FPRINTF(stderr, "[%d] token = %d == %d * %d ?\n", rank, token, nloops, size);
  135. STARPU_ASSERT(token == nloops*size);
  136. }
  137. #endif
  138. return 0;
  139. }