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_ncopy_16.c 9.9 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332
  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. IFLOAT *aoffset;
  43. IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4;
  44. IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8;
  45. IFLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12;
  46. IFLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16;
  47. IFLOAT *boffset;
  48. IFLOAT ctemp01, ctemp02, ctemp03, ctemp04;
  49. IFLOAT ctemp05, ctemp06, ctemp07, ctemp08;
  50. IFLOAT ctemp09, ctemp10, ctemp11, ctemp12;
  51. IFLOAT ctemp13, ctemp14, ctemp15, ctemp16;
  52. IFLOAT ctemp17, ctemp18, ctemp19, ctemp20;
  53. IFLOAT ctemp21, ctemp22, ctemp23, ctemp24;
  54. IFLOAT ctemp25, ctemp26, ctemp27, ctemp28;
  55. IFLOAT ctemp29, ctemp30, ctemp31, ctemp32;
  56. aoffset = a;
  57. boffset = b;
  58. lda *= 2;
  59. j = (n >> 4);
  60. if (j > 0){
  61. do{
  62. aoffset1 = aoffset;
  63. aoffset2 = aoffset1 + lda;
  64. aoffset3 = aoffset2 + lda;
  65. aoffset4 = aoffset3 + lda;
  66. aoffset5 = aoffset4 + lda;
  67. aoffset6 = aoffset5 + lda;
  68. aoffset7 = aoffset6 + lda;
  69. aoffset8 = aoffset7 + lda;
  70. aoffset9 = aoffset8 + lda;
  71. aoffset10 = aoffset9 + lda;
  72. aoffset11 = aoffset10 + lda;
  73. aoffset12 = aoffset11 + lda;
  74. aoffset13 = aoffset12 + lda;
  75. aoffset14 = aoffset13 + lda;
  76. aoffset15 = aoffset14 + lda;
  77. aoffset16 = aoffset15 + lda;
  78. aoffset += 16 * lda;
  79. i = m;
  80. if (i > 0){
  81. do{
  82. ctemp01 = *(aoffset1 + 0);
  83. ctemp02 = *(aoffset1 + 1);
  84. ctemp03 = *(aoffset2 + 0);
  85. ctemp04 = *(aoffset2 + 1);
  86. ctemp05 = *(aoffset3 + 0);
  87. ctemp06 = *(aoffset3 + 1);
  88. ctemp07 = *(aoffset4 + 0);
  89. ctemp08 = *(aoffset4 + 1);
  90. ctemp09 = *(aoffset5 + 0);
  91. ctemp10 = *(aoffset5 + 1);
  92. ctemp11 = *(aoffset6 + 0);
  93. ctemp12 = *(aoffset6 + 1);
  94. ctemp13 = *(aoffset7 + 0);
  95. ctemp14 = *(aoffset7 + 1);
  96. ctemp15 = *(aoffset8 + 0);
  97. ctemp16 = *(aoffset8 + 1);
  98. ctemp17 = *(aoffset9 + 0);
  99. ctemp18 = *(aoffset9 + 1);
  100. ctemp19 = *(aoffset10 + 0);
  101. ctemp20 = *(aoffset10 + 1);
  102. ctemp21 = *(aoffset11 + 0);
  103. ctemp22 = *(aoffset11 + 1);
  104. ctemp23 = *(aoffset12 + 0);
  105. ctemp24 = *(aoffset12 + 1);
  106. ctemp25 = *(aoffset13 + 0);
  107. ctemp26 = *(aoffset13 + 1);
  108. ctemp27 = *(aoffset14 + 0);
  109. ctemp28 = *(aoffset14 + 1);
  110. ctemp29 = *(aoffset15 + 0);
  111. ctemp30 = *(aoffset15 + 1);
  112. ctemp31 = *(aoffset16 + 0);
  113. ctemp32 = *(aoffset16 + 1);
  114. *(boffset + 0) = ctemp01;
  115. *(boffset + 1) = ctemp02;
  116. *(boffset + 2) = ctemp03;
  117. *(boffset + 3) = ctemp04;
  118. *(boffset + 4) = ctemp05;
  119. *(boffset + 5) = ctemp06;
  120. *(boffset + 6) = ctemp07;
  121. *(boffset + 7) = ctemp08;
  122. *(boffset + 8) = ctemp09;
  123. *(boffset + 9) = ctemp10;
  124. *(boffset + 10) = ctemp11;
  125. *(boffset + 11) = ctemp12;
  126. *(boffset + 12) = ctemp13;
  127. *(boffset + 13) = ctemp14;
  128. *(boffset + 14) = ctemp15;
  129. *(boffset + 15) = ctemp16;
  130. *(boffset + 16) = ctemp17;
  131. *(boffset + 17) = ctemp18;
  132. *(boffset + 18) = ctemp19;
  133. *(boffset + 19) = ctemp20;
  134. *(boffset + 20) = ctemp21;
  135. *(boffset + 21) = ctemp22;
  136. *(boffset + 22) = ctemp23;
  137. *(boffset + 23) = ctemp24;
  138. *(boffset + 24) = ctemp25;
  139. *(boffset + 25) = ctemp26;
  140. *(boffset + 26) = ctemp27;
  141. *(boffset + 27) = ctemp28;
  142. *(boffset + 28) = ctemp29;
  143. *(boffset + 29) = ctemp30;
  144. *(boffset + 30) = ctemp31;
  145. *(boffset + 31) = ctemp32;
  146. aoffset1 += 2;
  147. aoffset2 += 2;
  148. aoffset3 += 2;
  149. aoffset4 += 2;
  150. aoffset5 += 2;
  151. aoffset6 += 2;
  152. aoffset7 += 2;
  153. aoffset8 += 2;
  154. aoffset9 += 2;
  155. aoffset10 += 2;
  156. aoffset11 += 2;
  157. aoffset12 += 2;
  158. aoffset13 += 2;
  159. aoffset14 += 2;
  160. aoffset15 += 2;
  161. aoffset16 += 2;
  162. boffset += 32;
  163. i --;
  164. }while(i > 0);
  165. }
  166. j--;
  167. }while(j > 0);
  168. } /* end of if(j > 0) */
  169. if (n & 8){
  170. aoffset1 = aoffset;
  171. aoffset2 = aoffset1 + lda;
  172. aoffset3 = aoffset2 + lda;
  173. aoffset4 = aoffset3 + lda;
  174. aoffset5 = aoffset4 + lda;
  175. aoffset6 = aoffset5 + lda;
  176. aoffset7 = aoffset6 + lda;
  177. aoffset8 = aoffset7 + lda;
  178. aoffset += 8 * lda;
  179. i = m;
  180. if (i > 0){
  181. do{
  182. ctemp01 = *(aoffset1 + 0);
  183. ctemp02 = *(aoffset1 + 1);
  184. ctemp03 = *(aoffset2 + 0);
  185. ctemp04 = *(aoffset2 + 1);
  186. ctemp05 = *(aoffset3 + 0);
  187. ctemp06 = *(aoffset3 + 1);
  188. ctemp07 = *(aoffset4 + 0);
  189. ctemp08 = *(aoffset4 + 1);
  190. ctemp09 = *(aoffset5 + 0);
  191. ctemp10 = *(aoffset5 + 1);
  192. ctemp11 = *(aoffset6 + 0);
  193. ctemp12 = *(aoffset6 + 1);
  194. ctemp13 = *(aoffset7 + 0);
  195. ctemp14 = *(aoffset7 + 1);
  196. ctemp15 = *(aoffset8 + 0);
  197. ctemp16 = *(aoffset8 + 1);
  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;
  215. aoffset2 += 2;
  216. aoffset3 += 2;
  217. aoffset4 += 2;
  218. aoffset5 += 2;
  219. aoffset6 += 2;
  220. aoffset7 += 2;
  221. aoffset8 += 2;
  222. boffset += 16;
  223. i --;
  224. }while(i > 0);
  225. }
  226. }
  227. if (n & 4){
  228. aoffset1 = aoffset;
  229. aoffset2 = aoffset1 + lda;
  230. aoffset3 = aoffset2 + lda;
  231. aoffset4 = aoffset3 + lda;
  232. aoffset += 4 * lda;
  233. i = m;
  234. if (i > 0){
  235. do{
  236. ctemp01 = *(aoffset1 + 0);
  237. ctemp02 = *(aoffset1 + 1);
  238. ctemp03 = *(aoffset2 + 0);
  239. ctemp04 = *(aoffset2 + 1);
  240. ctemp05 = *(aoffset3 + 0);
  241. ctemp06 = *(aoffset3 + 1);
  242. ctemp07 = *(aoffset4 + 0);
  243. ctemp08 = *(aoffset4 + 1);
  244. *(boffset + 0) = ctemp01;
  245. *(boffset + 1) = ctemp02;
  246. *(boffset + 2) = ctemp03;
  247. *(boffset + 3) = ctemp04;
  248. *(boffset + 4) = ctemp05;
  249. *(boffset + 5) = ctemp06;
  250. *(boffset + 6) = ctemp07;
  251. *(boffset + 7) = ctemp08;
  252. aoffset1 += 2;
  253. aoffset2 += 2;
  254. aoffset3 += 2;
  255. aoffset4 += 2;
  256. boffset += 8;
  257. i --;
  258. }while(i > 0);
  259. }
  260. } /* end of if(j > 0) */
  261. if (n & 2){
  262. aoffset1 = aoffset;
  263. aoffset2 = aoffset1 + lda;
  264. aoffset += 2 * lda;
  265. i = m;
  266. if (i > 0){
  267. do{
  268. ctemp01 = *(aoffset1 + 0);
  269. ctemp02 = *(aoffset1 + 1);
  270. ctemp03 = *(aoffset2 + 0);
  271. ctemp04 = *(aoffset2 + 1);
  272. *(boffset + 0) = ctemp01;
  273. *(boffset + 1) = ctemp02;
  274. *(boffset + 2) = ctemp03;
  275. *(boffset + 3) = ctemp04;
  276. aoffset1 += 2;
  277. aoffset2 += 2;
  278. boffset += 4;
  279. i --;
  280. }while(i > 0);
  281. }
  282. } /* end of if(j > 0) */
  283. if (n & 1){
  284. aoffset1 = aoffset;
  285. i = m;
  286. if (i > 0){
  287. do{
  288. ctemp01 = *(aoffset1 + 0);
  289. ctemp02 = *(aoffset1 + 1);
  290. *(boffset + 0) = ctemp01;
  291. *(boffset + 1) = ctemp02;
  292. aoffset1 += 2;
  293. boffset += 2;
  294. i --;
  295. }while(i > 0);
  296. }
  297. } /* end of if(j > 0) */
  298. return 0;
  299. }