fblock_opencl.c 2.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263
  1. /*
  2. * StarPU
  3. * Copyright (C) Université Bordeaux 1, CNRS 2008-2010 (see AUTHORS file)
  4. *
  5. * This program 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. * This program 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 <starpu.h>
  17. #include <starpu_opencl.h>
  18. extern struct starpu_opencl_program codelet;
  19. void opencl_func(void *buffers[], void *cl_arg)
  20. {
  21. int id, devid, err;
  22. cl_kernel kernel;
  23. cl_command_queue queue;
  24. int *factor = cl_arg;
  25. int *block = (int *)STARPU_BLOCK_GET_PTR(buffers[0]);
  26. int nx = (int)STARPU_BLOCK_GET_NX(buffers[0]);
  27. int ny = (int)STARPU_BLOCK_GET_NY(buffers[0]);
  28. int nz = (int)STARPU_BLOCK_GET_NZ(buffers[0]);
  29. unsigned ldy = STARPU_BLOCK_GET_LDY(buffers[0]);
  30. unsigned ldz = STARPU_BLOCK_GET_LDZ(buffers[0]);
  31. id = starpu_worker_get_id();
  32. devid = starpu_worker_get_devid(id);
  33. err = starpu_opencl_load_kernel(&kernel, &queue, &codelet, "fblock_opencl", devid);
  34. if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
  35. err = 0;
  36. err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &block);
  37. err = clSetKernelArg(kernel, 1, sizeof(nx), &nx);
  38. err = clSetKernelArg(kernel, 2, sizeof(ny), &ny);
  39. err = clSetKernelArg(kernel, 3, sizeof(nz), &nz);
  40. err = clSetKernelArg(kernel, 4, sizeof(ldy), &ldy);
  41. err = clSetKernelArg(kernel, 5, sizeof(ldz), &ldz);
  42. err |= clSetKernelArg(kernel, 6, sizeof(*factor), factor);
  43. if (err) STARPU_OPENCL_REPORT_ERROR(err);
  44. {
  45. size_t global=nx*ny*nz;
  46. err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, NULL, 0, NULL, NULL);
  47. if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
  48. }
  49. clFinish(queue);
  50. starpu_opencl_release_kernel(kernel);
  51. }