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.

ztrsm_lncopy_4.c 9.4 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459
  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, BLASLONG offset, FLOAT *b){
  41. BLASLONG i, ii, j, jj;
  42. FLOAT data01 = 0.0, data02 = 0.0, data03, data04;
  43. FLOAT data05, data06, data07, data08;
  44. FLOAT data09, data10, data11 = 0.0, data12 = 0.0;
  45. FLOAT data13, data14, data15, data16;
  46. FLOAT data17, data18, data19, data20;
  47. FLOAT data21, data22, data23, data24;
  48. FLOAT data25, data26, data27, data28;
  49. FLOAT data29, data30, data31, data32;
  50. FLOAT *a1, *a2, *a3, *a4;
  51. lda *= 2;
  52. jj = offset;
  53. j = (n >> 2);
  54. while (j > 0){
  55. a1 = a + 0 * lda;
  56. a2 = a + 1 * lda;
  57. a3 = a + 2 * lda;
  58. a4 = a + 3 * lda;
  59. ii = 0;
  60. i = (m >> 2);
  61. while (i > 0) {
  62. if (ii == jj) {
  63. #ifndef UNIT
  64. data01 = *(a1 + 0);
  65. data02 = *(a1 + 1);
  66. #endif
  67. data03 = *(a1 + 2);
  68. data04 = *(a1 + 3);
  69. data05 = *(a1 + 4);
  70. data06 = *(a1 + 5);
  71. data07 = *(a1 + 6);
  72. data08 = *(a1 + 7);
  73. #ifndef UNIT
  74. data11 = *(a2 + 2);
  75. data12 = *(a2 + 3);
  76. #endif
  77. data13 = *(a2 + 4);
  78. data14 = *(a2 + 5);
  79. data15 = *(a2 + 6);
  80. data16 = *(a2 + 7);
  81. #ifndef UNIT
  82. data21 = *(a3 + 4);
  83. data22 = *(a3 + 5);
  84. #endif
  85. data23 = *(a3 + 6);
  86. data24 = *(a3 + 7);
  87. #ifndef UNIT
  88. data31 = *(a4 + 6);
  89. data32 = *(a4 + 7);
  90. #endif
  91. compinv(b + 0, data01, data02);
  92. *(b + 8) = data03;
  93. *(b + 9) = data04;
  94. compinv(b + 10, data11, data12);
  95. *(b + 16) = data05;
  96. *(b + 17) = data06;
  97. *(b + 18) = data13;
  98. *(b + 19) = data14;
  99. compinv(b + 20, data21, data22);
  100. *(b + 24) = data07;
  101. *(b + 25) = data08;
  102. *(b + 26) = data15;
  103. *(b + 27) = data16;
  104. *(b + 28) = data23;
  105. *(b + 29) = data24;
  106. compinv(b + 30, data31, data32);
  107. }
  108. if (ii > jj) {
  109. data01 = *(a1 + 0);
  110. data02 = *(a1 + 1);
  111. data03 = *(a1 + 2);
  112. data04 = *(a1 + 3);
  113. data05 = *(a1 + 4);
  114. data06 = *(a1 + 5);
  115. data07 = *(a1 + 6);
  116. data08 = *(a1 + 7);
  117. data09 = *(a2 + 0);
  118. data10 = *(a2 + 1);
  119. data11 = *(a2 + 2);
  120. data12 = *(a2 + 3);
  121. data13 = *(a2 + 4);
  122. data14 = *(a2 + 5);
  123. data15 = *(a2 + 6);
  124. data16 = *(a2 + 7);
  125. data17 = *(a3 + 0);
  126. data18 = *(a3 + 1);
  127. data19 = *(a3 + 2);
  128. data20 = *(a3 + 3);
  129. data21 = *(a3 + 4);
  130. data22 = *(a3 + 5);
  131. data23 = *(a3 + 6);
  132. data24 = *(a3 + 7);
  133. data25 = *(a4 + 0);
  134. data26 = *(a4 + 1);
  135. data27 = *(a4 + 2);
  136. data28 = *(a4 + 3);
  137. data29 = *(a4 + 4);
  138. data30 = *(a4 + 5);
  139. data31 = *(a4 + 6);
  140. data32 = *(a4 + 7);
  141. *(b + 0) = data01;
  142. *(b + 1) = data02;
  143. *(b + 2) = data09;
  144. *(b + 3) = data10;
  145. *(b + 4) = data17;
  146. *(b + 5) = data18;
  147. *(b + 6) = data25;
  148. *(b + 7) = data26;
  149. *(b + 8) = data03;
  150. *(b + 9) = data04;
  151. *(b + 10) = data11;
  152. *(b + 11) = data12;
  153. *(b + 12) = data19;
  154. *(b + 13) = data20;
  155. *(b + 14) = data27;
  156. *(b + 15) = data28;
  157. *(b + 16) = data05;
  158. *(b + 17) = data06;
  159. *(b + 18) = data13;
  160. *(b + 19) = data14;
  161. *(b + 20) = data21;
  162. *(b + 21) = data22;
  163. *(b + 22) = data29;
  164. *(b + 23) = data30;
  165. *(b + 24) = data07;
  166. *(b + 25) = data08;
  167. *(b + 26) = data15;
  168. *(b + 27) = data16;
  169. *(b + 28) = data23;
  170. *(b + 29) = data24;
  171. *(b + 30) = data31;
  172. *(b + 31) = data32;
  173. }
  174. a1 += 8;
  175. a2 += 8;
  176. a3 += 8;
  177. a4 += 8;
  178. b += 32;
  179. i --;
  180. ii += 4;
  181. }
  182. if (m & 2) {
  183. if (ii == jj) {
  184. #ifndef UNIT
  185. data01 = *(a1 + 0);
  186. data02 = *(a1 + 1);
  187. #endif
  188. data03 = *(a1 + 2);
  189. data04 = *(a1 + 3);
  190. #ifndef UNIT
  191. data11 = *(a2 + 2);
  192. data12 = *(a2 + 3);
  193. #endif
  194. compinv(b + 0, data01, data02);
  195. *(b + 4) = data03;
  196. *(b + 5) = data04;
  197. compinv(b + 6, data11, data12);
  198. }
  199. if (ii > jj) {
  200. data01 = *(a1 + 0);
  201. data02 = *(a1 + 1);
  202. data03 = *(a1 + 2);
  203. data04 = *(a1 + 3);
  204. data09 = *(a2 + 0);
  205. data10 = *(a2 + 1);
  206. data11 = *(a2 + 2);
  207. data12 = *(a2 + 3);
  208. data17 = *(a3 + 0);
  209. data18 = *(a3 + 1);
  210. data19 = *(a3 + 2);
  211. data20 = *(a3 + 3);
  212. data25 = *(a4 + 0);
  213. data26 = *(a4 + 1);
  214. data27 = *(a4 + 2);
  215. data28 = *(a4 + 3);
  216. *(b + 0) = data01;
  217. *(b + 1) = data02;
  218. *(b + 2) = data09;
  219. *(b + 3) = data10;
  220. *(b + 4) = data17;
  221. *(b + 5) = data18;
  222. *(b + 6) = data25;
  223. *(b + 7) = data26;
  224. *(b + 8) = data03;
  225. *(b + 9) = data04;
  226. *(b + 10) = data11;
  227. *(b + 11) = data12;
  228. *(b + 12) = data19;
  229. *(b + 13) = data20;
  230. *(b + 14) = data27;
  231. *(b + 15) = data28;
  232. }
  233. a1 += 4;
  234. a2 += 4;
  235. a3 += 4;
  236. a4 += 4;
  237. b += 16;
  238. ii += 2;
  239. }
  240. if (m & 1) {
  241. if (ii == jj) {
  242. #ifndef UNIT
  243. data01 = *(a1 + 0);
  244. data02 = *(a1 + 1);
  245. #endif
  246. compinv(b + 0, data01, data02);
  247. }
  248. if (ii > jj) {
  249. data01 = *(a1 + 0);
  250. data02 = *(a1 + 1);
  251. data09 = *(a2 + 0);
  252. data10 = *(a2 + 1);
  253. data17 = *(a3 + 0);
  254. data18 = *(a3 + 1);
  255. data25 = *(a4 + 0);
  256. data26 = *(a4 + 1);
  257. *(b + 0) = data01;
  258. *(b + 1) = data02;
  259. *(b + 2) = data09;
  260. *(b + 3) = data10;
  261. *(b + 4) = data17;
  262. *(b + 5) = data18;
  263. *(b + 6) = data25;
  264. *(b + 7) = data26;
  265. }
  266. /* a1 += 2;
  267. a2 += 2;
  268. a3 += 2;
  269. a4 += 2; */
  270. b += 8;
  271. // ii += 1;
  272. }
  273. a += 4 * lda;
  274. jj += 4;
  275. j --;
  276. }
  277. if (n & 2) {
  278. a1 = a + 0 * lda;
  279. a2 = a + 1 * lda;
  280. ii = 0;
  281. i = (m >> 1);
  282. while (i > 0) {
  283. if (ii == jj) {
  284. #ifndef UNIT
  285. data01 = *(a1 + 0);
  286. data02 = *(a1 + 1);
  287. #endif
  288. data03 = *(a1 + 2);
  289. data04 = *(a1 + 3);
  290. #ifndef UNIT
  291. data11 = *(a2 + 2);
  292. data12 = *(a2 + 3);
  293. #endif
  294. compinv(b + 0, data01, data02);
  295. *(b + 4) = data03;
  296. *(b + 5) = data04;
  297. compinv(b + 6, data11, data12);
  298. }
  299. if (ii > jj) {
  300. data01 = *(a1 + 0);
  301. data02 = *(a1 + 1);
  302. data03 = *(a1 + 2);
  303. data04 = *(a1 + 3);
  304. data09 = *(a2 + 0);
  305. data10 = *(a2 + 1);
  306. data11 = *(a2 + 2);
  307. data12 = *(a2 + 3);
  308. *(b + 0) = data01;
  309. *(b + 1) = data02;
  310. *(b + 2) = data09;
  311. *(b + 3) = data10;
  312. *(b + 4) = data03;
  313. *(b + 5) = data04;
  314. *(b + 6) = data11;
  315. *(b + 7) = data12;
  316. }
  317. a1 += 4;
  318. a2 += 4;
  319. b += 8;
  320. i --;
  321. ii += 2;
  322. }
  323. if (m & 1) {
  324. if (ii == jj) {
  325. #ifndef UNIT
  326. data01 = *(a1 + 0);
  327. data02 = *(a1 + 1);
  328. #endif
  329. compinv(b + 0, data01, data02);
  330. }
  331. if (ii > jj) {
  332. data01 = *(a1 + 0);
  333. data02 = *(a1 + 1);
  334. data09 = *(a2 + 0);
  335. data10 = *(a2 + 1);
  336. *(b + 0) = data01;
  337. *(b + 1) = data02;
  338. *(b + 2) = data09;
  339. *(b + 3) = data10;
  340. }
  341. /* a1 += 2;
  342. a2 += 2; */
  343. b += 4;
  344. // ii += 1;
  345. }
  346. a += 2 * lda;
  347. jj += 2;
  348. }
  349. if (n & 1) {
  350. a1 = a + 0 * lda;
  351. ii = 0;
  352. i = m;
  353. while (i > 0) {
  354. if (ii == jj) {
  355. #ifndef UNIT
  356. data01 = *(a1 + 0);
  357. data02 = *(a1 + 1);
  358. #endif
  359. compinv(b + 0, data01, data02);
  360. }
  361. if (ii > jj) {
  362. data01 = *(a1 + 0);
  363. data02 = *(a1 + 1);
  364. *(b + 0) = data01;
  365. *(b + 1) = data02;
  366. }
  367. a1 += 2;
  368. b += 2;
  369. i --;
  370. ii += 1;
  371. }
  372. // a += lda;
  373. // jj += 1;
  374. }
  375. return 0;
  376. }