matvecmult.c 6.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2010, 2011-2012 Université de Bordeaux 1
  4. * Copyright (C) 2010, 2011, 2012 Centre National de la Recherche Scientifique
  5. *
  6. * StarPU is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU Lesser General Public License as published by
  8. * the Free Software Foundation; either version 2.1 of the License, or (at
  9. * your option) any later version.
  10. *
  11. * StarPU is distributed in the hope that it will be useful, but
  12. * WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  14. *
  15. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  16. */
  17. #include <starpu.h>
  18. #include <pthread.h>
  19. #include <math.h>
  20. #define FPRINTF(ofile, fmt, args ...) do { if (!getenv("STARPU_SSILENT")) {fprintf(ofile, fmt, ##args); }} while(0)
  21. #ifdef STARPU_USE_OPENCL
  22. struct starpu_opencl_program opencl_code;
  23. void opencl_codelet(void *descr[], __attribute__ ((unused)) void *_args)
  24. {
  25. cl_kernel kernel;
  26. cl_command_queue queue;
  27. int id, devid, err, n;
  28. cl_mem matrix = (cl_mem)STARPU_MATRIX_GET_DEV_HANDLE(descr[0]);
  29. cl_mem vector = (cl_mem)STARPU_VECTOR_GET_DEV_HANDLE(descr[1]);
  30. cl_mem mult = (cl_mem)STARPU_VECTOR_GET_DEV_HANDLE(descr[2]);
  31. int nx = STARPU_MATRIX_GET_NX(descr[0]);
  32. int ny = STARPU_MATRIX_GET_NY(descr[0]);
  33. cl_event event;
  34. id = starpu_worker_get_id();
  35. devid = starpu_worker_get_devid(id);
  36. err = starpu_opencl_load_kernel(&kernel, &queue, &opencl_code, "matVecMult", devid);
  37. if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
  38. n=0;
  39. err = clSetKernelArg(kernel, n++, sizeof(matrix), &matrix);
  40. err |= clSetKernelArg(kernel, n++, sizeof(vector), &vector);
  41. err |= clSetKernelArg(kernel, n++, sizeof(nx), (void*)&nx);
  42. err |= clSetKernelArg(kernel, n++, sizeof(ny), (void*)&ny);
  43. err |= clSetKernelArg(kernel, n++, sizeof(mult), &mult);
  44. if (err) STARPU_OPENCL_REPORT_ERROR(err);
  45. {
  46. size_t global=nx*ny;
  47. err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, NULL, 0, NULL, &event);
  48. if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
  49. }
  50. clFinish(queue);
  51. starpu_opencl_collect_stats(event);
  52. clReleaseEvent(event);
  53. starpu_opencl_release_kernel(kernel);
  54. }
  55. #endif
  56. void fillArray(float* pfData, int iSize)
  57. {
  58. int i;
  59. const float fScale = 1.0f / (float)RAND_MAX;
  60. for (i = 0; i < iSize; ++i)
  61. {
  62. pfData[i] = fScale * rand();
  63. }
  64. }
  65. void printArray(float* pfData, int iSize)
  66. {
  67. int i;
  68. for (i = 0; i < iSize; ++i)
  69. {
  70. FPRINTF(stderr, "%f ", pfData[i]);
  71. }
  72. FPRINTF(stderr, "\n");
  73. }
  74. void matVecMult(const float *matrix, const float *vector, int width, int height, float *mult)
  75. {
  76. int i, j;
  77. for (i = 0; i < height; ++i)
  78. {
  79. double sum = 0;
  80. for (j = 0; j < width; ++j)
  81. {
  82. double a = matrix[i * width + j];
  83. double b = vector[j];
  84. sum += a * b;
  85. }
  86. mult[i] = (float)sum;
  87. }
  88. }
  89. int compareL2fe(const float* reference, const float* data, const unsigned int len, const float epsilon)
  90. {
  91. float error = 0;
  92. float ref = 0;
  93. unsigned int i;
  94. for(i = 0; i < len; ++i)
  95. {
  96. float diff = reference[i] - data[i];
  97. error += diff * diff;
  98. ref += reference[i] * reference[i];
  99. }
  100. float normRef = sqrtf(ref);
  101. if (fabs(ref) < 1e-7) return 1;
  102. float normError = sqrtf(error);
  103. error = normError / normRef;
  104. return error < epsilon ? 0 : 1;
  105. }
  106. int main(int argc, char **argv)
  107. {
  108. struct starpu_codelet cl = {};
  109. struct starpu_conf conf;
  110. starpu_conf_init(&conf);
  111. conf.ncpus = 0;
  112. conf.ncuda = 0;
  113. conf.nopencl = 1;
  114. /* int width=1100; */
  115. /* int height=244021; */
  116. int width=20;
  117. int height=4;
  118. float *matrix, *vector, *mult;
  119. float *correctResult;
  120. unsigned int mem_size_matrix, mem_size_vector, mem_size_mult;
  121. starpu_data_handle_t matrix_handle, vector_handle, mult_handle;
  122. int ret, submit;
  123. ret = starpu_init(&conf);
  124. if (STARPU_UNLIKELY(ret == -ENODEV))
  125. {
  126. FPRINTF(stderr, "This application requires an OpenCL worker.\n");
  127. starpu_shutdown();
  128. return 77;
  129. }
  130. mem_size_matrix = width * height * sizeof(float);
  131. matrix = (float*)malloc(mem_size_matrix);
  132. mem_size_vector = width * sizeof(float);
  133. vector = (float*)malloc(mem_size_vector);
  134. mem_size_mult = height * sizeof(float);
  135. mult = (float*)malloc(mem_size_mult);
  136. correctResult = (float*)malloc(mem_size_mult);
  137. assert(matrix);
  138. assert(vector);
  139. assert(mult);
  140. assert(correctResult);
  141. fillArray(matrix, width*height);
  142. fillArray(vector, width);
  143. fillArray(mult, height);
  144. matVecMult(matrix, vector, width, height, correctResult);
  145. starpu_matrix_data_register(&matrix_handle, 0, (uintptr_t)matrix, width, width, height, sizeof(float));
  146. starpu_vector_data_register(&vector_handle, 0, (uintptr_t)vector, width, sizeof(float));
  147. starpu_vector_data_register(&mult_handle, 0, (uintptr_t)mult, height, sizeof(float));
  148. #ifdef STARPU_USE_OPENCL
  149. ret = starpu_opencl_load_opencl_from_file("examples/matvecmult/matvecmult_kernel.cl", &opencl_code, NULL);
  150. STARPU_CHECK_RETURN_VALUE(ret, "starpu_opencl_load_opencl_from_file");
  151. #endif
  152. cl.where = STARPU_OPENCL;
  153. #ifdef STARPU_USE_OPENCL
  154. cl.opencl_funcs[0] = opencl_codelet;
  155. #endif
  156. cl.nbuffers = 3;
  157. cl.modes[0] = STARPU_R;
  158. cl.modes[1] = STARPU_R;
  159. cl.modes[2] = STARPU_RW;
  160. cl.model = NULL;
  161. struct starpu_task *task = starpu_task_create();
  162. task->cl = &cl;
  163. task->callback_func = NULL;
  164. task->handles[0] = matrix_handle;
  165. task->handles[1] = vector_handle;
  166. task->handles[2] = mult_handle;
  167. submit = starpu_task_submit(task);
  168. if (STARPU_UNLIKELY(submit == -ENODEV))
  169. {
  170. FPRINTF(stderr, "No worker may execute this task. This application requires an OpenCL worker.\n");
  171. }
  172. else
  173. {
  174. starpu_task_wait_for_all();
  175. }
  176. starpu_data_unregister(matrix_handle);
  177. starpu_data_unregister(vector_handle);
  178. starpu_data_unregister(mult_handle);
  179. if (STARPU_LIKELY(submit != -ENODEV))
  180. {
  181. int res = compareL2fe(correctResult, mult, height, 1e-6f);
  182. FPRINTF(stdout, "TEST %s\n\n", (res == 0) ? "PASSED" : "FAILED !!!");
  183. }
  184. #if 0
  185. printArray(matrix, width*height);
  186. printArray(vector, width);
  187. printArray(mult, height);
  188. #endif
  189. starpu_shutdown();
  190. return (submit == -ENODEV) ? 77 : 0;
  191. }