fblock.c 5.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2010,2011,2013-2015 Université de Bordeaux
  4. * Copyright (C) 2012,2013 Inria
  5. * Copyright (C) 2010-2013,2015,2017 CNRS
  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. /*
  19. * This examplifies how to use partitioning filters. We here just split a 3D
  20. * matrix into 3D slices (along the X axis), and run a dumb kernel on them.
  21. */
  22. #include <starpu.h>
  23. #define NX 5
  24. #define NY 4
  25. #define NZ 3
  26. #define PARTS 2
  27. #define FPRINTF(ofile, fmt, ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ## __VA_ARGS__); }} while(0)
  28. extern void cpu_func(void *buffers[], void *cl_arg);
  29. #ifdef STARPU_USE_CUDA
  30. extern void cuda_func(void *buffers[], void *cl_arg);
  31. #endif
  32. #ifdef STARPU_USE_OPENCL
  33. extern void opencl_func(void *buffers[], void *cl_arg);
  34. #endif
  35. void print_block(int *block, int nx, int ny, int nz, unsigned ldy, unsigned ldz)
  36. {
  37. int i, j, k;
  38. FPRINTF(stderr, "block=%p nx=%d ny=%d nz=%d ldy=%u ldz=%u\n", block, nx, ny, nz, ldy, ldz);
  39. for(k=0 ; k<nz ; k++)
  40. {
  41. for(j=0 ; j<ny ; j++)
  42. {
  43. for(i=0 ; i<nx ; i++)
  44. {
  45. FPRINTF(stderr, "%2d ", block[(k*ldz)+(j*ldy)+i]);
  46. }
  47. FPRINTF(stderr,"\n");
  48. }
  49. FPRINTF(stderr,"\n");
  50. }
  51. FPRINTF(stderr,"\n");
  52. }
  53. void print_data(starpu_data_handle_t block_handle)
  54. {
  55. int *block = (int *)starpu_block_get_local_ptr(block_handle);
  56. int nx = starpu_block_get_nx(block_handle);
  57. int ny = starpu_block_get_ny(block_handle);
  58. int nz = starpu_block_get_nz(block_handle);
  59. unsigned ldy = starpu_block_get_local_ldy(block_handle);
  60. unsigned ldz = starpu_block_get_local_ldz(block_handle);
  61. print_block(block, nx, ny, nz, ldy, ldz);
  62. }
  63. #ifdef STARPU_USE_OPENCL
  64. struct starpu_opencl_program opencl_program;
  65. #endif
  66. int main(void)
  67. {
  68. int *block,n=0;
  69. int i, j, k;
  70. int ret;
  71. block = (int*)malloc(NX*NY*NZ*sizeof(block[0]));
  72. assert(block);
  73. for(k=0 ; k<NZ ; k++)
  74. {
  75. for(j=0 ; j<NY ; j++)
  76. {
  77. for(i=0 ; i<NX ; i++)
  78. {
  79. block[(k*NX*NY)+(j*NX)+i] = n++;
  80. }
  81. }
  82. }
  83. starpu_data_handle_t handle;
  84. struct starpu_codelet cl =
  85. {
  86. .cpu_funcs = {cpu_func},
  87. .cpu_funcs_name = {"cpu_func"},
  88. #ifdef STARPU_USE_CUDA
  89. .cuda_funcs = {cuda_func},
  90. .cuda_flags = {STARPU_CUDA_ASYNC},
  91. #endif
  92. #ifdef STARPU_USE_OPENCL
  93. .opencl_funcs = {opencl_func},
  94. .opencl_flags = {STARPU_OPENCL_ASYNC},
  95. #endif
  96. .nbuffers = 1,
  97. .modes = {STARPU_RW},
  98. .name = "block_scal"
  99. };
  100. ret = starpu_init(NULL);
  101. if (ret == -ENODEV)
  102. exit(77);
  103. STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
  104. #ifdef STARPU_USE_OPENCL
  105. ret = starpu_opencl_load_opencl_from_file("examples/filters/fblock_opencl_kernel.cl", &opencl_program, NULL);
  106. STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
  107. #endif
  108. /* Declare data to StarPU */
  109. starpu_block_data_register(&handle, STARPU_MAIN_RAM, (uintptr_t)block, NX, NX*NY, NX, NY, NZ, sizeof(int));
  110. FPRINTF(stderr, "IN Block\n");
  111. print_data(handle);
  112. /* Partition the block in PARTS sub-blocks */
  113. struct starpu_data_filter f =
  114. {
  115. .filter_func = starpu_block_filter_block,
  116. .nchildren = PARTS
  117. };
  118. starpu_data_partition(handle, &f);
  119. FPRINTF(stderr,"Nb of partitions : %d\n",starpu_data_get_nb_children(handle));
  120. for(i=0 ; i<starpu_data_get_nb_children(handle) ; i++)
  121. {
  122. starpu_data_handle_t sblock = starpu_data_get_sub_data(handle, 1, i);
  123. FPRINTF(stderr, "Sub block %d\n", i);
  124. print_data(sblock);
  125. }
  126. /* Submit a task on each sub-block */
  127. for(i=0 ; i<starpu_data_get_nb_children(handle) ; i++)
  128. {
  129. int multiplier=i;
  130. struct starpu_task *task = starpu_task_create();
  131. FPRINTF(stderr,"Dealing with sub-block %d\n", i);
  132. task->cl = &cl;
  133. task->synchronous = 1;
  134. task->callback_func = NULL;
  135. task->handles[0] = starpu_data_get_sub_data(handle, 1, i);
  136. task->cl_arg = &multiplier;
  137. task->cl_arg_size = sizeof(multiplier);
  138. ret = starpu_task_submit(task);
  139. if (ret)
  140. {
  141. FPRINTF(stderr, "Error when submitting task\n");
  142. exit(ret);
  143. }
  144. }
  145. /* Unpartition the data, unregister it from StarPU and shutdown */
  146. starpu_data_unpartition(handle, STARPU_MAIN_RAM);
  147. print_data(handle);
  148. starpu_data_unregister(handle);
  149. /* Print result block */
  150. FPRINTF(stderr, "OUT Block\n");
  151. print_block(block, NX, NY, NZ, NX, NX*NY);
  152. free(block);
  153. starpu_shutdown();
  154. return 0;
  155. }