cl_enqueuecopybuffer.c 3.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134
  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. struct arg_copybuffer {
  18. size_t src_offset, dst_offset;
  19. cl_mem src_buffer, dst_buffer;
  20. size_t cb;
  21. };
  22. static void soclEnqueueCopyBuffer_opencl_task(void *descr[], void *args) {
  23. struct arg_copybuffer *arg;
  24. int wid;
  25. cl_command_queue cq;
  26. cl_event ev;
  27. arg = (struct arg_copybuffer*)args;
  28. wid = starpu_worker_get_id();
  29. starpu_opencl_get_queue(wid, &cq);
  30. cl_mem src = (cl_mem)STARPU_VARIABLE_GET_PTR(descr[0]);
  31. cl_mem dst = (cl_mem)STARPU_VARIABLE_GET_PTR(descr[1]);
  32. clEnqueueCopyBuffer(cq, src,dst, arg->src_offset, arg->dst_offset, arg->cb, 0, NULL, &ev);
  33. clWaitForEvents(1, &ev);
  34. clReleaseEvent(ev);
  35. gc_entity_unstore(&arg->src_buffer);
  36. gc_entity_unstore(&arg->dst_buffer);
  37. free(arg);
  38. }
  39. static void soclEnqueueCopyBuffer_cpu_task(void *descr[], void *args) {
  40. struct arg_copybuffer *arg;
  41. arg = (struct arg_copybuffer*)args;
  42. void * src = (void*)STARPU_VARIABLE_GET_PTR(descr[0]);
  43. void * dst = (void*)STARPU_VARIABLE_GET_PTR(descr[1]);
  44. memcpy(dst+arg->dst_offset, src+arg->src_offset, arg->cb);
  45. gc_entity_unstore(&arg->src_buffer);
  46. gc_entity_unstore(&arg->dst_buffer);
  47. free(arg);
  48. }
  49. static struct starpu_perfmodel copy_buffer_perfmodel = {
  50. .type = STARPU_HISTORY_BASED,
  51. .symbol = "SOCL_COPY_BUFFER"
  52. };
  53. static struct starpu_codelet codelet_copybuffer = {
  54. .where = STARPU_CPU | STARPU_OPENCL,
  55. .model = &copy_buffer_perfmodel,
  56. .cpu_funcs = { &soclEnqueueCopyBuffer_cpu_task, NULL },
  57. .opencl_funcs = { &soclEnqueueCopyBuffer_opencl_task, NULL },
  58. .modes = {STARPU_R, STARPU_RW},
  59. .nbuffers = 2
  60. };
  61. cl_int command_copy_buffer_submit(command_copy_buffer cmd) {
  62. /* Aliases */
  63. cl_mem src_buffer = cmd->src_buffer;
  64. cl_mem dst_buffer = cmd->dst_buffer;
  65. size_t src_offset = cmd->src_offset;
  66. size_t dst_offset = cmd->dst_offset;
  67. size_t cb = cmd->cb;
  68. struct starpu_task *task;
  69. struct arg_copybuffer *arg;
  70. task = task_create(CL_COMMAND_COPY_BUFFER);
  71. task->handles[0] = src_buffer->handle;
  72. task->handles[1] = dst_buffer->handle;
  73. task->cl = &codelet_copybuffer;
  74. /* Execute the task on a specific worker? */
  75. if (cmd->_command.cq->device != NULL) {
  76. task->execute_on_a_specific_worker = 1;
  77. task->workerid = cmd->_command.cq->device->worker_id;
  78. }
  79. arg = (struct arg_copybuffer*)malloc(sizeof(struct arg_copybuffer));
  80. arg->src_offset = src_offset;
  81. arg->dst_offset = dst_offset;
  82. arg->cb = cb;
  83. gc_entity_store(&arg->src_buffer, src_buffer);
  84. gc_entity_store(&arg->dst_buffer, dst_buffer);
  85. task->cl_arg = arg;
  86. task->cl_arg_size = sizeof(struct arg_copybuffer);
  87. dst_buffer->scratch = 0;
  88. task_submit(task, cmd);
  89. return CL_SUCCESS;
  90. }
  91. CL_API_ENTRY cl_int CL_API_CALL
  92. soclEnqueueCopyBuffer(cl_command_queue cq,
  93. cl_mem src_buffer,
  94. cl_mem dst_buffer,
  95. size_t src_offset,
  96. size_t dst_offset,
  97. size_t cb,
  98. cl_uint num_events,
  99. const cl_event * events,
  100. cl_event * event) CL_API_SUFFIX__VERSION_1_0
  101. {
  102. command_copy_buffer cmd = command_copy_buffer_create(src_buffer, dst_buffer, src_offset, dst_offset, cb);
  103. command_queue_enqueue(cq, cmd, num_events, events);
  104. RETURN_EVENT(cmd, event);
  105. return CL_SUCCESS;
  106. }