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_tcopy_16.c 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387
  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, IFLOAT *a, BLASLONG lda, IFLOAT *b){
  41. BLASLONG i, j;
  42. IFLOAT *aoffset;
  43. IFLOAT *aoffset1, *aoffset2;
  44. IFLOAT *boffset;
  45. IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
  46. IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
  47. IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
  48. IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
  49. IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
  50. IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
  51. IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
  52. IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
  53. aoffset = a;
  54. boffset = b;
  55. #if 0
  56. fprintf(stderr, "m = %d n = %d\n", m, n);
  57. #endif
  58. j = (n >> 4);
  59. if (j > 0){
  60. do{
  61. aoffset1 = aoffset;
  62. aoffset2 = aoffset + lda;
  63. aoffset += 16;
  64. i = (m >> 1);
  65. if (i > 0){
  66. do{
  67. ctemp01 = *(aoffset1 + 0);
  68. ctemp02 = *(aoffset1 + 1);
  69. ctemp03 = *(aoffset1 + 2);
  70. ctemp04 = *(aoffset1 + 3);
  71. ctemp05 = *(aoffset1 + 4);
  72. ctemp06 = *(aoffset1 + 5);
  73. ctemp07 = *(aoffset1 + 6);
  74. ctemp08 = *(aoffset1 + 7);
  75. ctemp09 = *(aoffset1 + 8);
  76. ctemp10 = *(aoffset1 + 9);
  77. ctemp11 = *(aoffset1 + 10);
  78. ctemp12 = *(aoffset1 + 11);
  79. ctemp13 = *(aoffset1 + 12);
  80. ctemp14 = *(aoffset1 + 13);
  81. ctemp15 = *(aoffset1 + 14);
  82. ctemp16 = *(aoffset1 + 15);
  83. ctemp17 = *(aoffset2 + 0);
  84. ctemp18 = *(aoffset2 + 1);
  85. ctemp19 = *(aoffset2 + 2);
  86. ctemp20 = *(aoffset2 + 3);
  87. ctemp21 = *(aoffset2 + 4);
  88. ctemp22 = *(aoffset2 + 5);
  89. ctemp23 = *(aoffset2 + 6);
  90. ctemp24 = *(aoffset2 + 7);
  91. ctemp25 = *(aoffset2 + 8);
  92. ctemp26 = *(aoffset2 + 9);
  93. ctemp27 = *(aoffset2 + 10);
  94. ctemp28 = *(aoffset2 + 11);
  95. ctemp29 = *(aoffset2 + 12);
  96. ctemp30 = *(aoffset2 + 13);
  97. ctemp31 = *(aoffset2 + 14);
  98. ctemp32 = *(aoffset2 + 15);
  99. *(boffset + 0) = ctemp01;
  100. *(boffset + 1) = ctemp02;
  101. *(boffset + 2) = ctemp03;
  102. *(boffset + 3) = ctemp04;
  103. *(boffset + 4) = ctemp05;
  104. *(boffset + 5) = ctemp06;
  105. *(boffset + 6) = ctemp07;
  106. *(boffset + 7) = ctemp08;
  107. *(boffset + 8) = ctemp09;
  108. *(boffset + 9) = ctemp10;
  109. *(boffset + 10) = ctemp11;
  110. *(boffset + 11) = ctemp12;
  111. *(boffset + 12) = ctemp13;
  112. *(boffset + 13) = ctemp14;
  113. *(boffset + 14) = ctemp15;
  114. *(boffset + 15) = ctemp16;
  115. *(boffset + 16) = ctemp17;
  116. *(boffset + 17) = ctemp18;
  117. *(boffset + 18) = ctemp19;
  118. *(boffset + 19) = ctemp20;
  119. *(boffset + 20) = ctemp21;
  120. *(boffset + 21) = ctemp22;
  121. *(boffset + 22) = ctemp23;
  122. *(boffset + 23) = ctemp24;
  123. *(boffset + 24) = ctemp25;
  124. *(boffset + 25) = ctemp26;
  125. *(boffset + 26) = ctemp27;
  126. *(boffset + 27) = ctemp28;
  127. *(boffset + 28) = ctemp29;
  128. *(boffset + 29) = ctemp30;
  129. *(boffset + 30) = ctemp31;
  130. *(boffset + 31) = ctemp32;
  131. aoffset1 += 2 * lda;
  132. aoffset2 += 2 * lda;
  133. boffset += 32;
  134. i --;
  135. }while(i > 0);
  136. }
  137. if (m & 1){
  138. ctemp01 = *(aoffset1 + 0);
  139. ctemp02 = *(aoffset1 + 1);
  140. ctemp03 = *(aoffset1 + 2);
  141. ctemp04 = *(aoffset1 + 3);
  142. ctemp05 = *(aoffset1 + 4);
  143. ctemp06 = *(aoffset1 + 5);
  144. ctemp07 = *(aoffset1 + 6);
  145. ctemp08 = *(aoffset1 + 7);
  146. ctemp09 = *(aoffset1 + 8);
  147. ctemp10 = *(aoffset1 + 9);
  148. ctemp11 = *(aoffset1 + 10);
  149. ctemp12 = *(aoffset1 + 11);
  150. ctemp13 = *(aoffset1 + 12);
  151. ctemp14 = *(aoffset1 + 13);
  152. ctemp15 = *(aoffset1 + 14);
  153. ctemp16 = *(aoffset1 + 15);
  154. *(boffset + 0) = ctemp01;
  155. *(boffset + 1) = ctemp02;
  156. *(boffset + 2) = ctemp03;
  157. *(boffset + 3) = ctemp04;
  158. *(boffset + 4) = ctemp05;
  159. *(boffset + 5) = ctemp06;
  160. *(boffset + 6) = ctemp07;
  161. *(boffset + 7) = ctemp08;
  162. *(boffset + 8) = ctemp09;
  163. *(boffset + 9) = ctemp10;
  164. *(boffset + 10) = ctemp11;
  165. *(boffset + 11) = ctemp12;
  166. *(boffset + 12) = ctemp13;
  167. *(boffset + 13) = ctemp14;
  168. *(boffset + 14) = ctemp15;
  169. *(boffset + 15) = ctemp16;
  170. boffset += 16;
  171. }
  172. j--;
  173. }while(j > 0);
  174. } /* end of if(j > 0) */
  175. if (n & 8){
  176. aoffset1 = aoffset;
  177. aoffset2 = aoffset + lda;
  178. aoffset += 8;
  179. i = (m >> 1);
  180. if (i > 0){
  181. do{
  182. ctemp01 = *(aoffset1 + 0);
  183. ctemp02 = *(aoffset1 + 1);
  184. ctemp03 = *(aoffset1 + 2);
  185. ctemp04 = *(aoffset1 + 3);
  186. ctemp05 = *(aoffset1 + 4);
  187. ctemp06 = *(aoffset1 + 5);
  188. ctemp07 = *(aoffset1 + 6);
  189. ctemp08 = *(aoffset1 + 7);
  190. ctemp09 = *(aoffset2 + 0);
  191. ctemp10 = *(aoffset2 + 1);
  192. ctemp11 = *(aoffset2 + 2);
  193. ctemp12 = *(aoffset2 + 3);
  194. ctemp13 = *(aoffset2 + 4);
  195. ctemp14 = *(aoffset2 + 5);
  196. ctemp15 = *(aoffset2 + 6);
  197. ctemp16 = *(aoffset2 + 7);
  198. *(boffset + 0) = ctemp01;
  199. *(boffset + 1) = ctemp02;
  200. *(boffset + 2) = ctemp03;
  201. *(boffset + 3) = ctemp04;
  202. *(boffset + 4) = ctemp05;
  203. *(boffset + 5) = ctemp06;
  204. *(boffset + 6) = ctemp07;
  205. *(boffset + 7) = ctemp08;
  206. *(boffset + 8) = ctemp09;
  207. *(boffset + 9) = ctemp10;
  208. *(boffset + 10) = ctemp11;
  209. *(boffset + 11) = ctemp12;
  210. *(boffset + 12) = ctemp13;
  211. *(boffset + 13) = ctemp14;
  212. *(boffset + 14) = ctemp15;
  213. *(boffset + 15) = ctemp16;
  214. aoffset1 += 2 * lda;
  215. aoffset2 += 2 * lda;
  216. boffset += 16;
  217. i --;
  218. }while(i > 0);
  219. }
  220. if (m & 1){
  221. ctemp01 = *(aoffset1 + 0);
  222. ctemp02 = *(aoffset1 + 1);
  223. ctemp03 = *(aoffset1 + 2);
  224. ctemp04 = *(aoffset1 + 3);
  225. ctemp05 = *(aoffset1 + 4);
  226. ctemp06 = *(aoffset1 + 5);
  227. ctemp07 = *(aoffset1 + 6);
  228. ctemp08 = *(aoffset1 + 7);
  229. *(boffset + 0) = ctemp01;
  230. *(boffset + 1) = ctemp02;
  231. *(boffset + 2) = ctemp03;
  232. *(boffset + 3) = ctemp04;
  233. *(boffset + 4) = ctemp05;
  234. *(boffset + 5) = ctemp06;
  235. *(boffset + 6) = ctemp07;
  236. *(boffset + 7) = ctemp08;
  237. boffset += 8;
  238. }
  239. }
  240. if (n & 4){
  241. aoffset1 = aoffset;
  242. aoffset2 = aoffset + lda;
  243. aoffset += 4;
  244. i = (m >> 1);
  245. if (i > 0){
  246. do{
  247. ctemp01 = *(aoffset1 + 0);
  248. ctemp02 = *(aoffset1 + 1);
  249. ctemp03 = *(aoffset1 + 2);
  250. ctemp04 = *(aoffset1 + 3);
  251. ctemp05 = *(aoffset2 + 0);
  252. ctemp06 = *(aoffset2 + 1);
  253. ctemp07 = *(aoffset2 + 2);
  254. ctemp08 = *(aoffset2 + 3);
  255. *(boffset + 0) = ctemp01;
  256. *(boffset + 1) = ctemp02;
  257. *(boffset + 2) = ctemp03;
  258. *(boffset + 3) = ctemp04;
  259. *(boffset + 4) = ctemp05;
  260. *(boffset + 5) = ctemp06;
  261. *(boffset + 6) = ctemp07;
  262. *(boffset + 7) = ctemp08;
  263. aoffset1 += 2 * lda;
  264. aoffset2 += 2 * lda;
  265. boffset += 8;
  266. i --;
  267. }while(i > 0);
  268. }
  269. if (m & 1){
  270. ctemp01 = *(aoffset1 + 0);
  271. ctemp02 = *(aoffset1 + 1);
  272. ctemp03 = *(aoffset1 + 2);
  273. ctemp04 = *(aoffset1 + 3);
  274. *(boffset + 0) = ctemp01;
  275. *(boffset + 1) = ctemp02;
  276. *(boffset + 2) = ctemp03;
  277. *(boffset + 3) = ctemp04;
  278. boffset += 4;
  279. }
  280. }
  281. if (n & 2){
  282. aoffset1 = aoffset;
  283. aoffset2 = aoffset + lda;
  284. aoffset += 2;
  285. i = (m >> 1);
  286. if (i > 0){
  287. do{
  288. ctemp01 = *(aoffset1 + 0);
  289. ctemp02 = *(aoffset1 + 1);
  290. ctemp03 = *(aoffset2 + 0);
  291. ctemp04 = *(aoffset2 + 1);
  292. *(boffset + 0) = ctemp01;
  293. *(boffset + 1) = ctemp02;
  294. *(boffset + 2) = ctemp03;
  295. *(boffset + 3) = ctemp04;
  296. aoffset1 += 2 * lda;
  297. aoffset2 += 2 * lda;
  298. boffset += 4;
  299. i --;
  300. }while(i > 0);
  301. }
  302. if (m & 1){
  303. ctemp01 = *(aoffset1 + 0);
  304. ctemp02 = *(aoffset1 + 1);
  305. *(boffset + 0) = ctemp01;
  306. *(boffset + 1) = ctemp02;
  307. boffset += 2;
  308. }
  309. }
  310. if (n & 1){
  311. aoffset1 = aoffset;
  312. aoffset2 = aoffset + lda;
  313. i = (m >> 1);
  314. if (i > 0){
  315. do{
  316. ctemp01 = *(aoffset1 + 0);
  317. ctemp02 = *(aoffset2 + 0);
  318. *(boffset + 0) = ctemp01;
  319. *(boffset + 1) = ctemp02;
  320. aoffset1 += 2 * lda;
  321. aoffset2 += 2 * lda;
  322. boffset += 2;
  323. i --;
  324. }while(i > 0);
  325. }
  326. if (m & 1){
  327. ctemp01 = *(aoffset1 + 0);
  328. *(boffset + 0) = ctemp01;
  329. // boffset += 1;
  330. }
  331. }
  332. return 0;
  333. }