You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

gemm.c 13 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484
  1. /*********************************************************************/
  2. /* Copyright 2009, 2010 The University of Texas at Austin. */
  3. /* All rights reserved. */
  4. /* */
  5. /* Redistribution and use in source and binary forms, with or */
  6. /* without modification, are permitted provided that the following */
  7. /* conditions are met: */
  8. /* */
  9. /* 1. Redistributions of source code must retain the above */
  10. /* copyright notice, this list of conditions and the following */
  11. /* disclaimer. */
  12. /* */
  13. /* 2. Redistributions in binary form must reproduce the above */
  14. /* copyright notice, this list of conditions and the following */
  15. /* disclaimer in the documentation and/or other materials */
  16. /* provided with the distribution. */
  17. /* */
  18. /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
  19. /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
  20. /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
  21. /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
  22. /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
  23. /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
  24. /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
  25. /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
  26. /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
  27. /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
  28. /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
  29. /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
  30. /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
  31. /* POSSIBILITY OF SUCH DAMAGE. */
  32. /* */
  33. /* The views and conclusions contained in the software and */
  34. /* documentation are those of the authors and should not be */
  35. /* interpreted as representing official policies, either expressed */
  36. /* or implied, of The University of Texas at Austin. */
  37. /*********************************************************************/
  38. #include <stdio.h>
  39. #include <stdlib.h>
  40. #include "common.h"
  41. #ifdef FUNCTION_PROFILE
  42. #include "functable.h"
  43. #endif
  44. #ifndef COMPLEX
  45. #ifdef XDOUBLE
  46. #define ERROR_NAME "QGEMM "
  47. #elif defined(DOUBLE)
  48. #define ERROR_NAME "DGEMM "
  49. #else
  50. #define ERROR_NAME "SGEMM "
  51. #endif
  52. #else
  53. #ifndef GEMM3M
  54. #ifdef XDOUBLE
  55. #define ERROR_NAME "XGEMM "
  56. #elif defined(DOUBLE)
  57. #define ERROR_NAME "ZGEMM "
  58. #else
  59. #define ERROR_NAME "CGEMM "
  60. #endif
  61. #else
  62. #ifdef XDOUBLE
  63. #define ERROR_NAME "XGEMM3M "
  64. #elif defined(DOUBLE)
  65. #define ERROR_NAME "ZGEMM3M "
  66. #else
  67. #define ERROR_NAME "CGEMM3M "
  68. #endif
  69. #endif
  70. #endif
  71. #ifndef GEMM_MULTITHREAD_THRESHOLD
  72. #define GEMM_MULTITHREAD_THRESHOLD 4
  73. #endif
  74. static int (*gemm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
  75. #ifndef GEMM3M
  76. GEMM_NN, GEMM_TN, GEMM_RN, GEMM_CN,
  77. GEMM_NT, GEMM_TT, GEMM_RT, GEMM_CT,
  78. GEMM_NR, GEMM_TR, GEMM_RR, GEMM_CR,
  79. GEMM_NC, GEMM_TC, GEMM_RC, GEMM_CC,
  80. #if defined(SMP) && !defined(USE_SIMPLE_THREADED_LEVEL3)
  81. GEMM_THREAD_NN, GEMM_THREAD_TN, GEMM_THREAD_RN, GEMM_THREAD_CN,
  82. GEMM_THREAD_NT, GEMM_THREAD_TT, GEMM_THREAD_RT, GEMM_THREAD_CT,
  83. GEMM_THREAD_NR, GEMM_THREAD_TR, GEMM_THREAD_RR, GEMM_THREAD_CR,
  84. GEMM_THREAD_NC, GEMM_THREAD_TC, GEMM_THREAD_RC, GEMM_THREAD_CC,
  85. #endif
  86. #else
  87. GEMM3M_NN, GEMM3M_TN, GEMM3M_RN, GEMM3M_CN,
  88. GEMM3M_NT, GEMM3M_TT, GEMM3M_RT, GEMM3M_CT,
  89. GEMM3M_NR, GEMM3M_TR, GEMM3M_RR, GEMM3M_CR,
  90. GEMM3M_NC, GEMM3M_TC, GEMM3M_RC, GEMM3M_CC,
  91. #if defined(SMP) && !defined(USE_SIMPLE_THREADED_LEVEL3)
  92. GEMM3M_THREAD_NN, GEMM3M_THREAD_TN, GEMM3M_THREAD_RN, GEMM3M_THREAD_CN,
  93. GEMM3M_THREAD_NT, GEMM3M_THREAD_TT, GEMM3M_THREAD_RT, GEMM3M_THREAD_CT,
  94. GEMM3M_THREAD_NR, GEMM3M_THREAD_TR, GEMM3M_THREAD_RR, GEMM3M_THREAD_CR,
  95. GEMM3M_THREAD_NC, GEMM3M_THREAD_TC, GEMM3M_THREAD_RC, GEMM3M_THREAD_CC,
  96. #endif
  97. #endif
  98. };
  99. #ifndef CBLAS
  100. void NAME(char *TRANSA, char *TRANSB,
  101. blasint *M, blasint *N, blasint *K,
  102. FLOAT *alpha,
  103. FLOAT *a, blasint *ldA,
  104. FLOAT *b, blasint *ldB,
  105. FLOAT *beta,
  106. FLOAT *c, blasint *ldC){
  107. blas_arg_t args;
  108. int transa, transb, nrowa, nrowb;
  109. blasint info;
  110. char transA, transB;
  111. FLOAT *buffer;
  112. FLOAT *sa, *sb;
  113. #ifdef SMP
  114. int nthreads_max;
  115. int nthreads_avail;
  116. double MNK;
  117. #ifndef COMPLEX
  118. #ifdef XDOUBLE
  119. int mode = BLAS_XDOUBLE | BLAS_REAL;
  120. #elif defined(DOUBLE)
  121. int mode = BLAS_DOUBLE | BLAS_REAL;
  122. #else
  123. int mode = BLAS_SINGLE | BLAS_REAL;
  124. #endif
  125. #else
  126. #ifdef XDOUBLE
  127. int mode = BLAS_XDOUBLE | BLAS_COMPLEX;
  128. #elif defined(DOUBLE)
  129. int mode = BLAS_DOUBLE | BLAS_COMPLEX;
  130. #else
  131. int mode = BLAS_SINGLE | BLAS_COMPLEX;
  132. #endif
  133. #endif
  134. #endif
  135. #if defined(SMP) && !defined(NO_AFFINITY) && !defined(USE_SIMPLE_THREADED_LEVEL3)
  136. int nodes;
  137. #endif
  138. PRINT_DEBUG_NAME;
  139. args.m = *M;
  140. args.n = *N;
  141. args.k = *K;
  142. args.a = (void *)a;
  143. args.b = (void *)b;
  144. args.c = (void *)c;
  145. args.lda = *ldA;
  146. args.ldb = *ldB;
  147. args.ldc = *ldC;
  148. args.alpha = (void *)alpha;
  149. args.beta = (void *)beta;
  150. transA = *TRANSA;
  151. transB = *TRANSB;
  152. TOUPPER(transA);
  153. TOUPPER(transB);
  154. transa = -1;
  155. transb = -1;
  156. if (transA == 'N') transa = 0;
  157. if (transA == 'T') transa = 1;
  158. #ifndef COMPLEX
  159. if (transA == 'R') transa = 0;
  160. if (transA == 'C') transa = 1;
  161. #else
  162. if (transA == 'R') transa = 2;
  163. if (transA == 'C') transa = 3;
  164. #endif
  165. if (transB == 'N') transb = 0;
  166. if (transB == 'T') transb = 1;
  167. #ifndef COMPLEX
  168. if (transB == 'R') transb = 0;
  169. if (transB == 'C') transb = 1;
  170. #else
  171. if (transB == 'R') transb = 2;
  172. if (transB == 'C') transb = 3;
  173. #endif
  174. nrowa = args.m;
  175. if (transa & 1) nrowa = args.k;
  176. nrowb = args.k;
  177. if (transb & 1) nrowb = args.n;
  178. info = 0;
  179. if (args.ldc < args.m) info = 13;
  180. if (args.ldb < nrowb) info = 10;
  181. if (args.lda < nrowa) info = 8;
  182. if (args.k < 0) info = 5;
  183. if (args.n < 0) info = 4;
  184. if (args.m < 0) info = 3;
  185. if (transb < 0) info = 2;
  186. if (transa < 0) info = 1;
  187. if (info){
  188. BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
  189. return;
  190. }
  191. #else
  192. void CNAME(enum CBLAS_ORDER order, enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANSPOSE TransB,
  193. blasint m, blasint n, blasint k,
  194. #ifndef COMPLEX
  195. FLOAT alpha,
  196. FLOAT *a, blasint lda,
  197. FLOAT *b, blasint ldb,
  198. FLOAT beta,
  199. FLOAT *c, blasint ldc) {
  200. #else
  201. void *valpha,
  202. void *va, blasint lda,
  203. void *vb, blasint ldb,
  204. void *vbeta,
  205. void *vc, blasint ldc) {
  206. FLOAT *alpha = (FLOAT*) valpha;
  207. FLOAT *beta = (FLOAT*) vbeta;
  208. FLOAT *a = (FLOAT*) va;
  209. FLOAT *b = (FLOAT*) vb;
  210. FLOAT *c = (FLOAT*) vc;
  211. #endif
  212. blas_arg_t args;
  213. int transa, transb;
  214. blasint nrowa, nrowb, info;
  215. XFLOAT *buffer;
  216. XFLOAT *sa, *sb;
  217. #ifdef SMP
  218. int nthreads_max;
  219. int nthreads_avail;
  220. double MNK;
  221. #ifndef COMPLEX
  222. #ifdef XDOUBLE
  223. int mode = BLAS_XDOUBLE | BLAS_REAL;
  224. #elif defined(DOUBLE)
  225. int mode = BLAS_DOUBLE | BLAS_REAL;
  226. #else
  227. int mode = BLAS_SINGLE | BLAS_REAL;
  228. #endif
  229. #else
  230. #ifdef XDOUBLE
  231. int mode = BLAS_XDOUBLE | BLAS_COMPLEX;
  232. #elif defined(DOUBLE)
  233. int mode = BLAS_DOUBLE | BLAS_COMPLEX;
  234. #else
  235. int mode = BLAS_SINGLE | BLAS_COMPLEX;
  236. #endif
  237. #endif
  238. #endif
  239. #if defined(SMP) && !defined(NO_AFFINITY) && !defined(USE_SIMPLE_THREADED_LEVEL3)
  240. int nodes;
  241. #endif
  242. PRINT_DEBUG_CNAME;
  243. #ifndef COMPLEX
  244. args.alpha = (void *)&alpha;
  245. args.beta = (void *)&beta;
  246. #else
  247. args.alpha = (void *)alpha;
  248. args.beta = (void *)beta;
  249. #endif
  250. transa = -1;
  251. transb = -1;
  252. info = 0;
  253. if (order == CblasColMajor) {
  254. args.m = m;
  255. args.n = n;
  256. args.k = k;
  257. args.a = (void *)a;
  258. args.b = (void *)b;
  259. args.c = (void *)c;
  260. args.lda = lda;
  261. args.ldb = ldb;
  262. args.ldc = ldc;
  263. if (TransA == CblasNoTrans) transa = 0;
  264. if (TransA == CblasTrans) transa = 1;
  265. #ifndef COMPLEX
  266. if (TransA == CblasConjNoTrans) transa = 0;
  267. if (TransA == CblasConjTrans) transa = 1;
  268. #else
  269. if (TransA == CblasConjNoTrans) transa = 2;
  270. if (TransA == CblasConjTrans) transa = 3;
  271. #endif
  272. if (TransB == CblasNoTrans) transb = 0;
  273. if (TransB == CblasTrans) transb = 1;
  274. #ifndef COMPLEX
  275. if (TransB == CblasConjNoTrans) transb = 0;
  276. if (TransB == CblasConjTrans) transb = 1;
  277. #else
  278. if (TransB == CblasConjNoTrans) transb = 2;
  279. if (TransB == CblasConjTrans) transb = 3;
  280. #endif
  281. nrowa = args.m;
  282. if (transa & 1) nrowa = args.k;
  283. nrowb = args.k;
  284. if (transb & 1) nrowb = args.n;
  285. info = -1;
  286. if (args.ldc < args.m) info = 13;
  287. if (args.ldb < nrowb) info = 10;
  288. if (args.lda < nrowa) info = 8;
  289. if (args.k < 0) info = 5;
  290. if (args.n < 0) info = 4;
  291. if (args.m < 0) info = 3;
  292. if (transb < 0) info = 2;
  293. if (transa < 0) info = 1;
  294. }
  295. if (order == CblasRowMajor) {
  296. args.m = n;
  297. args.n = m;
  298. args.k = k;
  299. args.a = (void *)b;
  300. args.b = (void *)a;
  301. args.c = (void *)c;
  302. args.lda = ldb;
  303. args.ldb = lda;
  304. args.ldc = ldc;
  305. if (TransB == CblasNoTrans) transa = 0;
  306. if (TransB == CblasTrans) transa = 1;
  307. #ifndef COMPLEX
  308. if (TransB == CblasConjNoTrans) transa = 0;
  309. if (TransB == CblasConjTrans) transa = 1;
  310. #else
  311. if (TransB == CblasConjNoTrans) transa = 2;
  312. if (TransB == CblasConjTrans) transa = 3;
  313. #endif
  314. if (TransA == CblasNoTrans) transb = 0;
  315. if (TransA == CblasTrans) transb = 1;
  316. #ifndef COMPLEX
  317. if (TransA == CblasConjNoTrans) transb = 0;
  318. if (TransA == CblasConjTrans) transb = 1;
  319. #else
  320. if (TransA == CblasConjNoTrans) transb = 2;
  321. if (TransA == CblasConjTrans) transb = 3;
  322. #endif
  323. nrowa = args.m;
  324. if (transa & 1) nrowa = args.k;
  325. nrowb = args.k;
  326. if (transb & 1) nrowb = args.n;
  327. info = -1;
  328. if (args.ldc < args.m) info = 13;
  329. if (args.ldb < nrowb) info = 10;
  330. if (args.lda < nrowa) info = 8;
  331. if (args.k < 0) info = 5;
  332. if (args.n < 0) info = 4;
  333. if (args.m < 0) info = 3;
  334. if (transb < 0) info = 2;
  335. if (transa < 0) info = 1;
  336. }
  337. if (info >= 0) {
  338. BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
  339. return;
  340. }
  341. #endif
  342. if ((args.m == 0) || (args.n == 0)) return;
  343. #if 0
  344. fprintf(stderr, "m = %4d n = %d k = %d lda = %4d ldb = %4d ldc = %4d\n",
  345. args.m, args.n, args.k, args.lda, args.ldb, args.ldc);
  346. #endif
  347. IDEBUG_START;
  348. FUNCTION_PROFILE_START();
  349. buffer = (XFLOAT *)blas_memory_alloc(0);
  350. sa = (XFLOAT *)((BLASLONG)buffer +GEMM_OFFSET_A);
  351. sb = (XFLOAT *)(((BLASLONG)sa + ((GEMM_P * GEMM_Q * COMPSIZE * SIZE + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
  352. #ifdef SMP
  353. mode |= (transa << BLAS_TRANSA_SHIFT);
  354. mode |= (transb << BLAS_TRANSB_SHIFT);
  355. nthreads_max = num_cpu_avail(3);
  356. nthreads_avail = nthreads_max;
  357. #ifndef COMPLEX
  358. MNK = (double) args.m * (double) args.n * (double) args.k;
  359. if ( MNK <= (65536.0 * (double) GEMM_MULTITHREAD_THRESHOLD) )
  360. nthreads_max = 1;
  361. #else
  362. MNK = (double) args.m * (double) args.n * (double) args.k;
  363. if ( MNK <= (8192.0 * (double) GEMM_MULTITHREAD_THRESHOLD) )
  364. nthreads_max = 1;
  365. #endif
  366. args.common = NULL;
  367. if ( nthreads_max > nthreads_avail )
  368. args.nthreads = nthreads_avail;
  369. else
  370. args.nthreads = nthreads_max;
  371. if (args.nthreads == 1) {
  372. #endif
  373. (gemm[(transb << 2) | transa])(&args, NULL, NULL, sa, sb, 0);
  374. #ifdef SMP
  375. } else {
  376. #ifndef USE_SIMPLE_THREADED_LEVEL3
  377. #ifndef NO_AFFINITY
  378. nodes = get_num_nodes();
  379. if ((nodes > 1) && get_node_equal()) {
  380. args.nthreads /= nodes;
  381. gemm_thread_mn(mode, &args, NULL, NULL, gemm[16 | (transb << 2) | transa], sa, sb, nodes);
  382. } else {
  383. #endif
  384. (gemm[16 | (transb << 2) | transa])(&args, NULL, NULL, sa, sb, 0);
  385. #else
  386. GEMM_THREAD(mode, &args, NULL, NULL, gemm[(transb << 2) | transa], sa, sb, args.nthreads);
  387. #endif
  388. #ifndef USE_SIMPLE_THREADED_LEVEL3
  389. #ifndef NO_AFFINITY
  390. }
  391. #endif
  392. #endif
  393. #endif
  394. #ifdef SMP
  395. }
  396. #endif
  397. blas_memory_free(buffer);
  398. FUNCTION_PROFILE_END(COMPSIZE * COMPSIZE, args.m * args.k + args.k * args.n + args.m * args.n, 2 * args.m * args.n * args.k);
  399. IDEBUG_END;
  400. return;
  401. }