cl_createkernel.c 5.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2010,2011 University of Bordeaux
  4. *
  5. * StarPU is free software; you can redistribute it and/or modify
  6. * it under the terms of the GNU Lesser General Public License as published by
  7. * the Free Software Foundation; either version 2.1 of the License, or (at
  8. * your option) any later version.
  9. *
  10. * StarPU is distributed in the hope that it will be useful, but
  11. * WITHOUT ANY WARRANTY; without even the implied warranty of
  12. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  13. *
  14. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  15. */
  16. #include "socl.h"
  17. static void soclCreateKernel_task(void *data) {
  18. struct _cl_kernel *k = (struct _cl_kernel *)data;
  19. int range = starpu_worker_get_range();
  20. cl_int err;
  21. DEBUG_MSG("[Device %d] Creating kernel...\n", starpu_worker_get_id());
  22. k->cl_kernels[range] = clCreateKernel(k->program->cl_programs[range], k->kernel_name, &err);
  23. if (err != CL_SUCCESS) {
  24. k->errcodes[range] = err;
  25. ERROR_STOP("[Device %d] Unable to create kernel. Aborting.\n", starpu_worker_get_id());
  26. return;
  27. }
  28. /* One worker creates argument structures */
  29. if (__sync_bool_compare_and_swap(&k->num_args, 0, 666)) {
  30. unsigned int i;
  31. cl_uint num_args;
  32. err = clGetKernelInfo(k->cl_kernels[range], CL_KERNEL_NUM_ARGS, sizeof(num_args), &num_args, NULL);
  33. if (err != CL_SUCCESS) {
  34. DEBUG_CL("clGetKernelInfo", err);
  35. ERROR_STOP("Unable to get kernel argument count. Aborting.\n");
  36. }
  37. k->num_args = num_args;
  38. DEBUG_MSG("Kernel has %d arguments\n", num_args);
  39. k->arg_size = (size_t*)malloc(sizeof(size_t) * num_args);
  40. k->arg_value = (void**)malloc(sizeof(void*) * num_args);
  41. k->arg_type = (enum kernel_arg_type*)malloc(sizeof(enum kernel_arg_type) * num_args);
  42. /* Settings default type to NULL */
  43. for (i=0; i<num_args; i++) {
  44. k->arg_value[i] = NULL;
  45. k->arg_type[i] = Null;
  46. }
  47. }
  48. }
  49. static void rk_task(void *data) {
  50. cl_kernel k = (cl_kernel)data;
  51. int range = starpu_worker_get_range();
  52. cl_int err = clReleaseKernel(k->cl_kernels[range]);
  53. if (err != CL_SUCCESS)
  54. DEBUG_CL("clReleaseKernel", err);
  55. }
  56. static void release_callback_kernel(void * e) {
  57. cl_kernel kernel = (cl_kernel)e;
  58. //Free args
  59. unsigned int j;
  60. for (j=0; j<kernel->num_args; j++) {
  61. switch (kernel->arg_type[j]) {
  62. case Null:
  63. break;
  64. case Buffer:
  65. gc_entity_unstore((cl_mem*)&kernel->arg_value[j]);
  66. break;
  67. case Immediate:
  68. free(kernel->arg_value[j]);
  69. break;
  70. }
  71. }
  72. if (kernel->arg_size != NULL)
  73. free(kernel->arg_size);
  74. if (kernel->arg_value != NULL)
  75. free(kernel->arg_value);
  76. if (kernel->arg_type != NULL)
  77. free(kernel->arg_type);
  78. //Release real kernels...
  79. starpu_execute_on_each_worker(rk_task, kernel, STARPU_OPENCL);
  80. gc_entity_unstore(&kernel->program);
  81. free(kernel->kernel_name);
  82. free(kernel->cl_kernels);
  83. free(kernel->errcodes);
  84. }
  85. CL_API_ENTRY cl_kernel CL_API_CALL
  86. soclCreateKernel(cl_program program,
  87. const char * kernel_name,
  88. cl_int * errcode_ret) CL_API_SUFFIX__VERSION_1_0
  89. {
  90. cl_kernel k;
  91. int device_count;
  92. if (program == NULL) {
  93. if (errcode_ret != NULL)
  94. *errcode_ret = CL_INVALID_PROGRAM;
  95. return NULL;
  96. }
  97. //TODO: check programs (see opencl specs)
  98. /* Create Kernel structure */
  99. k = (cl_kernel)gc_entity_alloc(sizeof(struct _cl_kernel), release_callback_kernel);
  100. if (k == NULL) {
  101. if (errcode_ret != NULL)
  102. *errcode_ret = CL_OUT_OF_HOST_MEMORY;
  103. return NULL;
  104. }
  105. gc_entity_store(&k->program, program);
  106. k->kernel_name = strdup(kernel_name);
  107. k->num_args = 0;
  108. k->arg_value = NULL;
  109. k->arg_size = NULL;
  110. #ifdef DEBUG
  111. static int id = 0;
  112. k->id = id++;
  113. #endif
  114. device_count = starpu_opencl_worker_get_count();
  115. k->cl_kernels = (cl_kernel*)malloc(device_count * sizeof(cl_kernel));
  116. k->errcodes = (cl_int*)malloc(device_count * sizeof(cl_int));
  117. {
  118. int i;
  119. for (i=0; i<device_count; i++) {
  120. k->cl_kernels[i] = NULL;
  121. k->errcodes[i] = -9999;
  122. }
  123. }
  124. /* Create kernel on each device */
  125. DEBUG_MSG("[Kernel %d] Create %d kernels (name \"%s\")\n", k->id, starpu_opencl_worker_get_count(), kernel_name);
  126. starpu_execute_on_each_worker(soclCreateKernel_task, k, STARPU_OPENCL);
  127. if (errcode_ret != NULL) {
  128. int i;
  129. *errcode_ret = CL_SUCCESS;
  130. for (i=0; i<device_count; i++) {
  131. switch (k->errcodes[i]) {
  132. #define CASE_RET(e) case e: *errcode_ret = e; return k;
  133. CASE_RET(CL_INVALID_PROGRAM)
  134. CASE_RET(CL_INVALID_PROGRAM_EXECUTABLE)
  135. CASE_RET(CL_INVALID_KERNEL_NAME)
  136. CASE_RET(CL_INVALID_KERNEL_DEFINITION)
  137. CASE_RET(CL_INVALID_VALUE)
  138. CASE_RET(CL_OUT_OF_RESOURCES)
  139. CASE_RET(CL_OUT_OF_HOST_MEMORY)
  140. #undef CASE_RET
  141. }
  142. }
  143. }
  144. return k;
  145. }