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.

zgemm_tcopy_8.c 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361
  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 "common.h"
  40. int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){
  41. BLASLONG i, j;
  42. FLOAT *aoffset;
  43. FLOAT *aoffset1, *aoffset2;
  44. FLOAT *boffset;
  45. FLOAT ctemp01, ctemp02, ctemp03, ctemp04;
  46. FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
  47. FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
  48. FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
  49. FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
  50. FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
  51. FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
  52. FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
  53. aoffset = a;
  54. boffset = b;
  55. lda *= 2;
  56. #if 0
  57. fprintf(stderr, "M = %d N = %d\n", m, n);
  58. #endif
  59. j = (n >> 3);
  60. if (j > 0){
  61. do{
  62. aoffset1 = aoffset;
  63. aoffset2 = aoffset + lda;
  64. aoffset += 16;
  65. i = (m >> 1);
  66. if (i > 0){
  67. do{
  68. ctemp01 = *(aoffset1 + 0);
  69. ctemp02 = *(aoffset1 + 1);
  70. ctemp03 = *(aoffset1 + 2);
  71. ctemp04 = *(aoffset1 + 3);
  72. ctemp05 = *(aoffset1 + 4);
  73. ctemp06 = *(aoffset1 + 5);
  74. ctemp07 = *(aoffset1 + 6);
  75. ctemp08 = *(aoffset1 + 7);
  76. ctemp09 = *(aoffset1 + 8);
  77. ctemp10 = *(aoffset1 + 9);
  78. ctemp11 = *(aoffset1 + 10);
  79. ctemp12 = *(aoffset1 + 11);
  80. ctemp13 = *(aoffset1 + 12);
  81. ctemp14 = *(aoffset1 + 13);
  82. ctemp15 = *(aoffset1 + 14);
  83. ctemp16 = *(aoffset1 + 15);
  84. ctemp17 = *(aoffset2 + 0);
  85. ctemp18 = *(aoffset2 + 1);
  86. ctemp19 = *(aoffset2 + 2);
  87. ctemp20 = *(aoffset2 + 3);
  88. ctemp21 = *(aoffset2 + 4);
  89. ctemp22 = *(aoffset2 + 5);
  90. ctemp23 = *(aoffset2 + 6);
  91. ctemp24 = *(aoffset2 + 7);
  92. ctemp25 = *(aoffset2 + 8);
  93. ctemp26 = *(aoffset2 + 9);
  94. ctemp27 = *(aoffset2 + 10);
  95. ctemp28 = *(aoffset2 + 11);
  96. ctemp29 = *(aoffset2 + 12);
  97. ctemp30 = *(aoffset2 + 13);
  98. ctemp31 = *(aoffset2 + 14);
  99. ctemp32 = *(aoffset2 + 15);
  100. *(boffset + 0) = ctemp01;
  101. *(boffset + 1) = ctemp02;
  102. *(boffset + 2) = ctemp03;
  103. *(boffset + 3) = ctemp04;
  104. *(boffset + 4) = ctemp05;
  105. *(boffset + 5) = ctemp06;
  106. *(boffset + 6) = ctemp07;
  107. *(boffset + 7) = ctemp08;
  108. *(boffset + 8) = ctemp09;
  109. *(boffset + 9) = ctemp10;
  110. *(boffset + 10) = ctemp11;
  111. *(boffset + 11) = ctemp12;
  112. *(boffset + 12) = ctemp13;
  113. *(boffset + 13) = ctemp14;
  114. *(boffset + 14) = ctemp15;
  115. *(boffset + 15) = ctemp16;
  116. *(boffset + 16) = ctemp17;
  117. *(boffset + 17) = ctemp18;
  118. *(boffset + 18) = ctemp19;
  119. *(boffset + 19) = ctemp20;
  120. *(boffset + 20) = ctemp21;
  121. *(boffset + 21) = ctemp22;
  122. *(boffset + 22) = ctemp23;
  123. *(boffset + 23) = ctemp24;
  124. *(boffset + 24) = ctemp25;
  125. *(boffset + 25) = ctemp26;
  126. *(boffset + 26) = ctemp27;
  127. *(boffset + 27) = ctemp28;
  128. *(boffset + 28) = ctemp29;
  129. *(boffset + 29) = ctemp30;
  130. *(boffset + 30) = ctemp31;
  131. *(boffset + 31) = ctemp32;
  132. aoffset1 += 2 * lda;
  133. aoffset2 += 2 * lda;
  134. boffset += 32;
  135. i --;
  136. }while(i > 0);
  137. }
  138. if (m & 1){
  139. ctemp01 = *(aoffset1 + 0);
  140. ctemp02 = *(aoffset1 + 1);
  141. ctemp03 = *(aoffset1 + 2);
  142. ctemp04 = *(aoffset1 + 3);
  143. ctemp05 = *(aoffset1 + 4);
  144. ctemp06 = *(aoffset1 + 5);
  145. ctemp07 = *(aoffset1 + 6);
  146. ctemp08 = *(aoffset1 + 7);
  147. ctemp09 = *(aoffset1 + 8);
  148. ctemp10 = *(aoffset1 + 9);
  149. ctemp11 = *(aoffset1 + 10);
  150. ctemp12 = *(aoffset1 + 11);
  151. ctemp13 = *(aoffset1 + 12);
  152. ctemp14 = *(aoffset1 + 13);
  153. ctemp15 = *(aoffset1 + 14);
  154. ctemp16 = *(aoffset1 + 15);
  155. *(boffset + 0) = ctemp01;
  156. *(boffset + 1) = ctemp02;
  157. *(boffset + 2) = ctemp03;
  158. *(boffset + 3) = ctemp04;
  159. *(boffset + 4) = ctemp05;
  160. *(boffset + 5) = ctemp06;
  161. *(boffset + 6) = ctemp07;
  162. *(boffset + 7) = ctemp08;
  163. *(boffset + 8) = ctemp09;
  164. *(boffset + 9) = ctemp10;
  165. *(boffset + 10) = ctemp11;
  166. *(boffset + 11) = ctemp12;
  167. *(boffset + 12) = ctemp13;
  168. *(boffset + 13) = ctemp14;
  169. *(boffset + 14) = ctemp15;
  170. *(boffset + 15) = ctemp16;
  171. boffset += 16;
  172. }
  173. j--;
  174. }while(j > 0);
  175. } /* end of if(j > 0) */
  176. if (n & 4){
  177. aoffset1 = aoffset;
  178. aoffset2 = aoffset + lda;
  179. aoffset += 8;
  180. i = (m >> 1);
  181. if (i > 0){
  182. do{
  183. ctemp01 = *(aoffset1 + 0);
  184. ctemp02 = *(aoffset1 + 1);
  185. ctemp03 = *(aoffset1 + 2);
  186. ctemp04 = *(aoffset1 + 3);
  187. ctemp05 = *(aoffset1 + 4);
  188. ctemp06 = *(aoffset1 + 5);
  189. ctemp07 = *(aoffset1 + 6);
  190. ctemp08 = *(aoffset1 + 7);
  191. ctemp09 = *(aoffset2 + 0);
  192. ctemp10 = *(aoffset2 + 1);
  193. ctemp11 = *(aoffset2 + 2);
  194. ctemp12 = *(aoffset2 + 3);
  195. ctemp13 = *(aoffset2 + 4);
  196. ctemp14 = *(aoffset2 + 5);
  197. ctemp15 = *(aoffset2 + 6);
  198. ctemp16 = *(aoffset2 + 7);
  199. *(boffset + 0) = ctemp01;
  200. *(boffset + 1) = ctemp02;
  201. *(boffset + 2) = ctemp03;
  202. *(boffset + 3) = ctemp04;
  203. *(boffset + 4) = ctemp05;
  204. *(boffset + 5) = ctemp06;
  205. *(boffset + 6) = ctemp07;
  206. *(boffset + 7) = ctemp08;
  207. *(boffset + 8) = ctemp09;
  208. *(boffset + 9) = ctemp10;
  209. *(boffset + 10) = ctemp11;
  210. *(boffset + 11) = ctemp12;
  211. *(boffset + 12) = ctemp13;
  212. *(boffset + 13) = ctemp14;
  213. *(boffset + 14) = ctemp15;
  214. *(boffset + 15) = ctemp16;
  215. aoffset1 += 2 * lda;
  216. aoffset2 += 2 * lda;
  217. boffset += 16;
  218. i --;
  219. }while(i > 0);
  220. }
  221. if (m & 1){
  222. ctemp01 = *(aoffset1 + 0);
  223. ctemp02 = *(aoffset1 + 1);
  224. ctemp03 = *(aoffset1 + 2);
  225. ctemp04 = *(aoffset1 + 3);
  226. ctemp05 = *(aoffset1 + 4);
  227. ctemp06 = *(aoffset1 + 5);
  228. ctemp07 = *(aoffset1 + 6);
  229. ctemp08 = *(aoffset1 + 7);
  230. *(boffset + 0) = ctemp01;
  231. *(boffset + 1) = ctemp02;
  232. *(boffset + 2) = ctemp03;
  233. *(boffset + 3) = ctemp04;
  234. *(boffset + 4) = ctemp05;
  235. *(boffset + 5) = ctemp06;
  236. *(boffset + 6) = ctemp07;
  237. *(boffset + 7) = ctemp08;
  238. boffset += 8;
  239. }
  240. }
  241. if (n & 2){
  242. aoffset1 = aoffset;
  243. aoffset2 = aoffset + lda;
  244. aoffset += 4;
  245. i = (m >> 1);
  246. if (i > 0){
  247. do{
  248. ctemp01 = *(aoffset1 + 0);
  249. ctemp02 = *(aoffset1 + 1);
  250. ctemp03 = *(aoffset1 + 2);
  251. ctemp04 = *(aoffset1 + 3);
  252. ctemp05 = *(aoffset2 + 0);
  253. ctemp06 = *(aoffset2 + 1);
  254. ctemp07 = *(aoffset2 + 2);
  255. ctemp08 = *(aoffset2 + 3);
  256. *(boffset + 0) = ctemp01;
  257. *(boffset + 1) = ctemp02;
  258. *(boffset + 2) = ctemp03;
  259. *(boffset + 3) = ctemp04;
  260. *(boffset + 4) = ctemp05;
  261. *(boffset + 5) = ctemp06;
  262. *(boffset + 6) = ctemp07;
  263. *(boffset + 7) = ctemp08;
  264. aoffset1 += 2 * lda;
  265. aoffset2 += 2 * lda;
  266. boffset += 8;
  267. i --;
  268. }while(i > 0);
  269. }
  270. if (m & 1){
  271. ctemp01 = *(aoffset1 + 0);
  272. ctemp02 = *(aoffset1 + 1);
  273. ctemp03 = *(aoffset1 + 2);
  274. ctemp04 = *(aoffset1 + 3);
  275. *(boffset + 0) = ctemp01;
  276. *(boffset + 1) = ctemp02;
  277. *(boffset + 2) = ctemp03;
  278. *(boffset + 3) = ctemp04;
  279. boffset += 4;
  280. }
  281. }
  282. if (n & 1){
  283. aoffset1 = aoffset;
  284. aoffset2 = aoffset + lda;
  285. // aoffset += 2;
  286. i = (m >> 1);
  287. if (i > 0){
  288. do{
  289. ctemp01 = *(aoffset1 + 0);
  290. ctemp02 = *(aoffset1 + 1);
  291. ctemp03 = *(aoffset2 + 0);
  292. ctemp04 = *(aoffset2 + 1);
  293. *(boffset + 0) = ctemp01;
  294. *(boffset + 1) = ctemp02;
  295. *(boffset + 2) = ctemp03;
  296. *(boffset + 3) = ctemp04;
  297. aoffset1 += 2 * lda;
  298. aoffset2 += 2 * lda;
  299. boffset += 4;
  300. i --;
  301. }while(i > 0);
  302. }
  303. if (m & 1){
  304. ctemp01 = *(aoffset1 + 0);
  305. ctemp02 = *(aoffset1 + 1);
  306. *(boffset + 0) = ctemp01;
  307. *(boffset + 1) = ctemp02;
  308. // boffset += 2;
  309. }
  310. }
  311. return 0;
  312. }