cl_enqueuecopybuffer.c 3.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2011-2012 Inria
  4. * Copyright (C) 2012,2016-2017 CNRS
  5. * Copyright (C) 2010-2011,2013-2014,2017 Université de Bordeaux
  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 "socl.h"
  19. static void soclEnqueueCopyBuffer_opencl_task(void *descr[], void *args)
  20. {
  21. int wid;
  22. cl_command_queue cq;
  23. cl_event ev;
  24. command_copy_buffer cmd = (command_copy_buffer)args;
  25. cl_event event = command_event_get(cmd);
  26. event->prof_start = _socl_nanotime();
  27. gc_entity_release(event);
  28. wid = starpu_worker_get_id_check();
  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, cmd->src_offset, cmd->dst_offset, cmd->cb, 0, NULL, &ev);
  33. clWaitForEvents(1, &ev);
  34. clReleaseEvent(ev);
  35. gc_entity_release_cmd(cmd);
  36. }
  37. static void soclEnqueueCopyBuffer_cpu_task(void *descr[], void *args)
  38. {
  39. command_copy_buffer cmd = (command_copy_buffer)args;
  40. cl_event ev = command_event_get(cmd);
  41. ev->prof_start = _socl_nanotime();
  42. gc_entity_release(ev);
  43. char * src = (void*)STARPU_VARIABLE_GET_PTR(descr[0]);
  44. char * dst = (void*)STARPU_VARIABLE_GET_PTR(descr[1]);
  45. memcpy(dst+cmd->dst_offset, src+cmd->src_offset, cmd->cb);
  46. gc_entity_release_cmd(cmd);
  47. }
  48. static struct starpu_perfmodel copy_buffer_perfmodel =
  49. {
  50. .type = STARPU_HISTORY_BASED,
  51. .symbol = "SOCL_COPY_BUFFER"
  52. };
  53. static struct starpu_codelet codelet_copybuffer =
  54. {
  55. .where = STARPU_CPU | STARPU_OPENCL,
  56. .model = &copy_buffer_perfmodel,
  57. .cpu_funcs = { &soclEnqueueCopyBuffer_cpu_task },
  58. .opencl_funcs = { &soclEnqueueCopyBuffer_opencl_task },
  59. .modes = {STARPU_R, STARPU_RW},
  60. .nbuffers = 2
  61. };
  62. cl_int command_copy_buffer_submit(command_copy_buffer cmd)
  63. {
  64. struct starpu_task * task = task_create(CL_COMMAND_COPY_BUFFER);
  65. task->handles[0] = cmd->src_buffer->handle;
  66. task->handles[1] = cmd->dst_buffer->handle;
  67. task->cl = &codelet_copybuffer;
  68. /* Execute the task on a specific worker? */
  69. if (cmd->_command.event->cq->device != NULL)
  70. {
  71. task->execute_on_a_specific_worker = 1;
  72. task->workerid = cmd->_command.event->cq->device->worker_id;
  73. }
  74. gc_entity_store_cmd(&task->cl_arg, cmd);
  75. task->cl_arg_size = sizeof(*cmd);
  76. cmd->dst_buffer->scratch = 0;
  77. task_submit(task, cmd);
  78. return CL_SUCCESS;
  79. }
  80. CL_API_ENTRY cl_int CL_API_CALL
  81. soclEnqueueCopyBuffer(cl_command_queue cq,
  82. cl_mem src_buffer,
  83. cl_mem dst_buffer,
  84. size_t src_offset,
  85. size_t dst_offset,
  86. size_t cb,
  87. cl_uint num_events,
  88. const cl_event * events,
  89. cl_event * event) CL_API_SUFFIX__VERSION_1_0
  90. {
  91. command_copy_buffer cmd = command_copy_buffer_create(src_buffer, dst_buffer, src_offset, dst_offset, cb);
  92. cl_event ev = command_event_get(cmd);
  93. command_queue_enqueue(cq, cmd, num_events, events);
  94. RETURN_EVENT(ev, event);
  95. return CL_SUCCESS;
  96. }