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.

trsm_kernel_RN.c 7.6 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319
  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 "common.h"
  39. static FLOAT dm1 = -1.;
  40. #ifdef CONJ
  41. #define GEMM_KERNEL GEMM_KERNEL_R
  42. #else
  43. #define GEMM_KERNEL GEMM_KERNEL_N
  44. #endif
  45. #if GEMM_DEFAULT_UNROLL_M == 1
  46. #define GEMM_UNROLL_M_SHIFT 0
  47. #endif
  48. #if GEMM_DEFAULT_UNROLL_M == 2
  49. #define GEMM_UNROLL_M_SHIFT 1
  50. #endif
  51. #if GEMM_DEFAULT_UNROLL_M == 4
  52. #define GEMM_UNROLL_M_SHIFT 2
  53. #endif
  54. #if GEMM_DEFAULT_UNROLL_M == 6
  55. #define GEMM_UNROLL_M_SHIFT 2
  56. #endif
  57. #if GEMM_DEFAULT_UNROLL_M == 8
  58. #define GEMM_UNROLL_M_SHIFT 3
  59. #endif
  60. #if GEMM_DEFAULT_UNROLL_M == 16
  61. #define GEMM_UNROLL_M_SHIFT 4
  62. #endif
  63. #if GEMM_DEFAULT_UNROLL_N == 1
  64. #define GEMM_UNROLL_N_SHIFT 0
  65. #endif
  66. #if GEMM_DEFAULT_UNROLL_N == 2
  67. #define GEMM_UNROLL_N_SHIFT 1
  68. #endif
  69. #if GEMM_DEFAULT_UNROLL_N == 4
  70. #define GEMM_UNROLL_N_SHIFT 2
  71. #endif
  72. #if GEMM_DEFAULT_UNROLL_N == 8
  73. #define GEMM_UNROLL_N_SHIFT 3
  74. #endif
  75. #if GEMM_DEFAULT_UNROLL_N == 16
  76. #define GEMM_UNROLL_N_SHIFT 4
  77. #endif
  78. #ifndef COMPLEX
  79. static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
  80. FLOAT aa, bb;
  81. int i, j, k;
  82. for (i = 0; i < n; i++) {
  83. bb = *(b + i);
  84. for (j = 0; j < m; j ++) {
  85. aa = *(c + j + i * ldc);
  86. aa *= bb;
  87. *a = aa;
  88. *(c + j + i * ldc) = aa;
  89. a ++;
  90. for (k = i + 1; k < n; k ++){
  91. *(c + j + k * ldc) -= aa * *(b + k);
  92. }
  93. }
  94. b += n;
  95. }
  96. }
  97. #else
  98. static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
  99. FLOAT aa1, aa2;
  100. FLOAT bb1, bb2;
  101. FLOAT cc1, cc2;
  102. int i, j, k;
  103. ldc *= 2;
  104. for (i = 0; i < n; i++) {
  105. bb1 = *(b + i * 2 + 0);
  106. bb2 = *(b + i * 2 + 1);
  107. for (j = 0; j < m; j ++) {
  108. aa1 = *(c + j * 2 + 0 + i * ldc);
  109. aa2 = *(c + j * 2 + 1 + i * ldc);
  110. #ifndef CONJ
  111. cc1 = aa1 * bb1 - aa2 * bb2;
  112. cc2 = aa1 * bb2 + aa2 * bb1;
  113. #else
  114. cc1 = aa1 * bb1 + aa2 * bb2;
  115. cc2 = -aa1 * bb2 + aa2 * bb1;
  116. #endif
  117. *(a + 0) = cc1;
  118. *(a + 1) = cc2;
  119. *(c + j * 2 + 0 + i * ldc) = cc1;
  120. *(c + j * 2 + 1 + i * ldc) = cc2;
  121. a += 2;
  122. for (k = i + 1; k < n; k ++){
  123. #ifndef CONJ
  124. *(c + j * 2 + 0 + k * ldc) -= cc1 * *(b + k * 2 + 0) - cc2 * *(b + k * 2 + 1);
  125. *(c + j * 2 + 1 + k * ldc) -= cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);
  126. #else
  127. *(c + j * 2 + 0 + k * ldc) -= cc1 * *(b + k * 2 + 0) + cc2 * *(b + k * 2 + 1);
  128. *(c + j * 2 + 1 + k * ldc) -= - cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);
  129. #endif
  130. }
  131. }
  132. b += n * 2;
  133. }
  134. }
  135. #endif
  136. int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1,
  137. #ifdef COMPLEX
  138. FLOAT dummy2,
  139. #endif
  140. FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG offset){
  141. FLOAT *aa, *cc;
  142. BLASLONG kk;
  143. BLASLONG i, j, jj;
  144. #if 0
  145. fprintf(stderr, "TRSM RN KERNEL m = %3ld n = %3ld k = %3ld offset = %3ld\n",
  146. m, n, k, offset);
  147. #endif
  148. jj = 0;
  149. j = (n >> GEMM_UNROLL_N_SHIFT);
  150. kk = -offset;
  151. while (j > 0) {
  152. aa = a;
  153. cc = c;
  154. i = (m >> GEMM_UNROLL_M_SHIFT);
  155. if (i > 0) {
  156. do {
  157. if (kk > 0) {
  158. GEMM_KERNEL(GEMM_UNROLL_M, GEMM_UNROLL_N, kk, dm1,
  159. #ifdef COMPLEX
  160. ZERO,
  161. #endif
  162. aa, b, cc, ldc);
  163. }
  164. solve(GEMM_UNROLL_M, GEMM_UNROLL_N,
  165. aa + kk * GEMM_UNROLL_M * COMPSIZE,
  166. b + kk * GEMM_UNROLL_N * COMPSIZE,
  167. cc, ldc);
  168. aa += GEMM_UNROLL_M * k * COMPSIZE;
  169. cc += GEMM_UNROLL_M * COMPSIZE;
  170. i --;
  171. } while (i > 0);
  172. }
  173. if (m & (GEMM_UNROLL_M - 1)) {
  174. i = (GEMM_UNROLL_M >> 1);
  175. while (i > 0) {
  176. if (m & i) {
  177. if (kk > 0) {
  178. GEMM_KERNEL(i, GEMM_UNROLL_N, kk, dm1,
  179. #ifdef COMPLEX
  180. ZERO,
  181. #endif
  182. aa, b, cc, ldc);
  183. }
  184. solve(i, GEMM_UNROLL_N,
  185. aa + kk * i * COMPSIZE,
  186. b + kk * GEMM_UNROLL_N * COMPSIZE,
  187. cc, ldc);
  188. aa += i * k * COMPSIZE;
  189. cc += i * COMPSIZE;
  190. }
  191. i >>= 1;
  192. }
  193. }
  194. kk += GEMM_UNROLL_N;
  195. b += GEMM_UNROLL_N * k * COMPSIZE;
  196. c += GEMM_UNROLL_N * ldc * COMPSIZE;
  197. j --;
  198. jj += GEMM_UNROLL_M;
  199. }
  200. if (n & (GEMM_UNROLL_N - 1)) {
  201. j = (GEMM_UNROLL_N >> 1);
  202. while (j > 0) {
  203. if (n & j) {
  204. aa = a;
  205. cc = c;
  206. i = (m >> GEMM_UNROLL_M_SHIFT);
  207. while (i > 0) {
  208. if (kk > 0) {
  209. GEMM_KERNEL(GEMM_UNROLL_M, j, kk, dm1,
  210. #ifdef COMPLEX
  211. ZERO,
  212. #endif
  213. aa,
  214. b,
  215. cc,
  216. ldc);
  217. }
  218. solve(GEMM_UNROLL_M, j,
  219. aa + kk * GEMM_UNROLL_M * COMPSIZE,
  220. b + kk * j * COMPSIZE, cc, ldc);
  221. aa += GEMM_UNROLL_M * k * COMPSIZE;
  222. cc += GEMM_UNROLL_M * COMPSIZE;
  223. i --;
  224. }
  225. if (m & (GEMM_UNROLL_M - 1)) {
  226. i = (GEMM_UNROLL_M >> 1);
  227. while (i > 0) {
  228. if (m & i) {
  229. if (kk > 0) {
  230. GEMM_KERNEL(i, j, kk, dm1,
  231. #ifdef COMPLEX
  232. ZERO,
  233. #endif
  234. aa,
  235. b,
  236. cc,
  237. ldc);
  238. }
  239. solve(i, j,
  240. aa + kk * i * COMPSIZE,
  241. b + kk * j * COMPSIZE, cc, ldc);
  242. aa += i * k * COMPSIZE;
  243. cc += i * COMPSIZE;
  244. }
  245. i >>= 1;
  246. }
  247. }
  248. b += j * k * COMPSIZE;
  249. c += j * ldc * COMPSIZE;
  250. kk += j;
  251. }
  252. j >>= 1;
  253. }
  254. }
  255. return 0;
  256. }