insert_task_sent_cache.c 4.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2011-2018 CNRS
  4. * Copyright (C) 2013-2015,2017,2018 Université de Bordeaux
  5. * Copyright (C) 2013 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 <starpu.h>
  19. #include <starpu_mpi.h>
  20. #include <math.h>
  21. #include "helper.h"
  22. #if !defined(STARPU_HAVE_SETENV)
  23. #warning setenv is not defined. Skipping test
  24. int main(void)
  25. {
  26. return STARPU_TEST_SKIPPED;
  27. }
  28. #else
  29. void func_cpu(void *descr[], void *_args)
  30. {
  31. (void)descr;
  32. (void)_args;
  33. }
  34. struct starpu_codelet mycodelet =
  35. {
  36. .cpu_funcs = {func_cpu},
  37. .nbuffers = 2,
  38. .modes = {STARPU_RW, STARPU_R},
  39. .model = &starpu_perfmodel_nop,
  40. };
  41. #define N 1000
  42. /* Returns the MPI node number where data indexes index is */
  43. int my_distrib(int x)
  44. {
  45. return x;
  46. }
  47. void test_cache(int rank, char *enabled, size_t *comm_amount)
  48. {
  49. int i;
  50. int ret;
  51. unsigned *v[2];
  52. starpu_data_handle_t data_handles[2];
  53. setenv("STARPU_MPI_CACHE", enabled, 1);
  54. ret = starpu_mpi_init_conf(NULL, NULL, 0, MPI_COMM_WORLD, NULL);
  55. STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_init_conf");
  56. for(i = 0; i < 2; i++)
  57. {
  58. int j;
  59. v[i] = malloc(N * sizeof(unsigned));
  60. for(j=0 ; j<N ; j++)
  61. {
  62. v[i][j] = 12;
  63. }
  64. }
  65. for(i = 0; i < 2; i++)
  66. {
  67. int mpi_rank = my_distrib(i);
  68. if (mpi_rank == rank)
  69. {
  70. //FPRINTF(stderr, "[%d] Owning data[%d][%d]\n", rank, x, y);
  71. starpu_vector_data_register(&data_handles[i], STARPU_MAIN_RAM, (uintptr_t)v[i], N, sizeof(unsigned));
  72. }
  73. else
  74. {
  75. /* I don't own this index, but will need it for my computations */
  76. //FPRINTF(stderr, "[%d] Neighbour of data[%d][%d]\n", rank, x, y);
  77. starpu_vector_data_register(&data_handles[i], -1, (uintptr_t)NULL, N, sizeof(unsigned));
  78. }
  79. starpu_mpi_data_register(data_handles[i], i, mpi_rank);
  80. }
  81. for(i = 0; i < 5; i++)
  82. {
  83. ret = starpu_mpi_task_insert(MPI_COMM_WORLD, &mycodelet, STARPU_RW, data_handles[0], STARPU_R, data_handles[1], 0);
  84. STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_task_insert");
  85. }
  86. for(i = 0; i < 5; i++)
  87. {
  88. ret = starpu_mpi_task_insert(MPI_COMM_WORLD, &mycodelet, STARPU_RW, data_handles[1], STARPU_R, data_handles[0], 0);
  89. STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_task_insert");
  90. }
  91. for(i = 0; i < 5; i++)
  92. {
  93. starpu_mpi_cache_flush(MPI_COMM_WORLD, data_handles[0]);
  94. }
  95. for(i = 0; i < 5; i++)
  96. {
  97. ret = starpu_mpi_task_insert(MPI_COMM_WORLD, &mycodelet, STARPU_RW, data_handles[1], STARPU_R, data_handles[0], 0);
  98. STARPU_CHECK_RETURN_VALUE(ret, "starpu_mpi_task_insert");
  99. }
  100. starpu_task_wait_for_all();
  101. for(i = 0; i < 2; i++)
  102. {
  103. starpu_data_unregister(data_handles[i]);
  104. free(v[i]);
  105. }
  106. starpu_mpi_comm_amounts_retrieve(comm_amount);
  107. starpu_mpi_shutdown();
  108. }
  109. int main(int argc, char **argv)
  110. {
  111. int rank, size;
  112. int result=0;
  113. size_t *comm_amount_with_cache;
  114. size_t *comm_amount_without_cache;
  115. MPI_INIT_THREAD_real(&argc, &argv, MPI_THREAD_SERIALIZED);
  116. starpu_mpi_comm_rank(MPI_COMM_WORLD, &rank);
  117. starpu_mpi_comm_size(MPI_COMM_WORLD, &size);
  118. setenv("STARPU_COMM_STATS", "1", 1);
  119. comm_amount_with_cache = malloc(size * sizeof(size_t));
  120. comm_amount_without_cache = malloc(size * sizeof(size_t));
  121. test_cache(rank, "0", comm_amount_with_cache);
  122. test_cache(rank, "1", comm_amount_without_cache);
  123. if (rank == 0 || rank == 1)
  124. {
  125. int dst = (rank == 0) ? 1 : 0;
  126. result = (comm_amount_with_cache[dst] == comm_amount_without_cache[dst] * 5);
  127. FPRINTF_MPI(stderr, "Communication cache mechanism is %sworking\n", result?"":"NOT ");
  128. }
  129. else
  130. {
  131. result = 1;
  132. }
  133. free(comm_amount_without_cache);
  134. free(comm_amount_with_cache);
  135. MPI_Finalize();
  136. return !result;
  137. }
  138. #endif