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_uncopy_4.c 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496
  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, data02, data03, data04;
  43. FLOAT data05, data06, data07, data08;
  44. FLOAT data09, data10, data11, data12;
  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. data09 = *(a2 + 0);
  68. data10 = *(a2 + 1);
  69. #ifndef UNIT
  70. data11 = *(a2 + 2);
  71. data12 = *(a2 + 3);
  72. #endif
  73. data17 = *(a3 + 0);
  74. data18 = *(a3 + 1);
  75. data19 = *(a3 + 2);
  76. data20 = *(a3 + 3);
  77. #ifndef UNIT
  78. data21 = *(a3 + 4);
  79. data22 = *(a3 + 5);
  80. #endif
  81. data25 = *(a4 + 0);
  82. data26 = *(a4 + 1);
  83. data27 = *(a4 + 2);
  84. data28 = *(a4 + 3);
  85. data29 = *(a4 + 4);
  86. data30 = *(a4 + 5);
  87. #ifndef UNIT
  88. data31 = *(a4 + 6);
  89. data32 = *(a4 + 7);
  90. #endif
  91. compinv(b + 0, data01, data02);
  92. *(b + 2) = data09;
  93. *(b + 3) = data10;
  94. *(b + 4) = data17;
  95. *(b + 5) = data18;
  96. *(b + 6) = data25;
  97. *(b + 7) = data26;
  98. compinv(b + 10, data11, data12);
  99. *(b + 12) = data19;
  100. *(b + 13) = data20;
  101. *(b + 14) = data27;
  102. *(b + 15) = data28;
  103. compinv(b + 20, data21, data22);
  104. *(b + 22) = data29;
  105. *(b + 23) = data30;
  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. data09 = *(a2 + 0);
  189. data10 = *(a2 + 1);
  190. #ifndef UNIT
  191. data11 = *(a2 + 2);
  192. data12 = *(a2 + 3);
  193. #endif
  194. data17 = *(a3 + 0);
  195. data18 = *(a3 + 1);
  196. data19 = *(a3 + 2);
  197. data20 = *(a3 + 3);
  198. data25 = *(a4 + 0);
  199. data26 = *(a4 + 1);
  200. data27 = *(a4 + 2);
  201. data28 = *(a4 + 3);
  202. compinv(b + 0, data01, data02);
  203. *(b + 2) = data09;
  204. *(b + 3) = data10;
  205. *(b + 4) = data17;
  206. *(b + 5) = data18;
  207. *(b + 6) = data25;
  208. *(b + 7) = data26;
  209. compinv(b + 10, data11, data12);
  210. *(b + 12) = data19;
  211. *(b + 13) = data20;
  212. *(b + 14) = data27;
  213. *(b + 15) = data28;
  214. }
  215. if (ii < jj) {
  216. data01 = *(a1 + 0);
  217. data02 = *(a1 + 1);
  218. data03 = *(a1 + 2);
  219. data04 = *(a1 + 3);
  220. data09 = *(a2 + 0);
  221. data10 = *(a2 + 1);
  222. data11 = *(a2 + 2);
  223. data12 = *(a2 + 3);
  224. data17 = *(a3 + 0);
  225. data18 = *(a3 + 1);
  226. data19 = *(a3 + 2);
  227. data20 = *(a3 + 3);
  228. data25 = *(a4 + 0);
  229. data26 = *(a4 + 1);
  230. data27 = *(a4 + 2);
  231. data28 = *(a4 + 3);
  232. *(b + 0) = data01;
  233. *(b + 1) = data02;
  234. *(b + 2) = data09;
  235. *(b + 3) = data10;
  236. *(b + 4) = data17;
  237. *(b + 5) = data18;
  238. *(b + 6) = data25;
  239. *(b + 7) = data26;
  240. *(b + 8) = data03;
  241. *(b + 9) = data04;
  242. *(b + 10) = data11;
  243. *(b + 11) = data12;
  244. *(b + 12) = data19;
  245. *(b + 13) = data20;
  246. *(b + 14) = data27;
  247. *(b + 15) = data28;
  248. }
  249. a1 += 4;
  250. a2 += 4;
  251. a3 += 4;
  252. a4 += 4;
  253. b += 16;
  254. ii += 2;
  255. }
  256. if (m & 1) {
  257. if (ii == jj) {
  258. #ifndef UNIT
  259. data01 = *(a1 + 0);
  260. data02 = *(a1 + 1);
  261. #endif
  262. data09 = *(a2 + 0);
  263. data10 = *(a2 + 1);
  264. data17 = *(a3 + 0);
  265. data18 = *(a3 + 1);
  266. data25 = *(a4 + 0);
  267. data26 = *(a4 + 1);
  268. compinv(b + 0, data01, data02);
  269. *(b + 2) = data09;
  270. *(b + 3) = data10;
  271. *(b + 4) = data17;
  272. *(b + 5) = data18;
  273. *(b + 6) = data25;
  274. *(b + 7) = data26;
  275. }
  276. if (ii < jj) {
  277. data01 = *(a1 + 0);
  278. data02 = *(a1 + 1);
  279. data09 = *(a2 + 0);
  280. data10 = *(a2 + 1);
  281. data17 = *(a3 + 0);
  282. data18 = *(a3 + 1);
  283. data25 = *(a4 + 0);
  284. data26 = *(a4 + 1);
  285. *(b + 0) = data01;
  286. *(b + 1) = data02;
  287. *(b + 2) = data09;
  288. *(b + 3) = data10;
  289. *(b + 4) = data17;
  290. *(b + 5) = data18;
  291. *(b + 6) = data25;
  292. *(b + 7) = data26;
  293. }
  294. a1 += 2;
  295. a2 += 2;
  296. a3 += 2;
  297. a4 += 2;
  298. b += 8;
  299. ii += 1;
  300. }
  301. a += 4 * lda;
  302. jj += 4;
  303. j --;
  304. }
  305. if (n & 2) {
  306. a1 = a + 0 * lda;
  307. a2 = a + 1 * lda;
  308. ii = 0;
  309. i = (m >> 1);
  310. while (i > 0) {
  311. if (ii == jj) {
  312. #ifndef UNIT
  313. data01 = *(a1 + 0);
  314. data02 = *(a1 + 1);
  315. #endif
  316. data09 = *(a2 + 0);
  317. data10 = *(a2 + 1);
  318. #ifndef UNIT
  319. data11 = *(a2 + 2);
  320. data12 = *(a2 + 3);
  321. #endif
  322. compinv(b + 0, data01, data02);
  323. *(b + 2) = data09;
  324. *(b + 3) = data10;
  325. compinv(b + 6, data11, data12);
  326. }
  327. if (ii < jj) {
  328. data01 = *(a1 + 0);
  329. data02 = *(a1 + 1);
  330. data03 = *(a1 + 2);
  331. data04 = *(a1 + 3);
  332. data09 = *(a2 + 0);
  333. data10 = *(a2 + 1);
  334. data11 = *(a2 + 2);
  335. data12 = *(a2 + 3);
  336. *(b + 0) = data01;
  337. *(b + 1) = data02;
  338. *(b + 2) = data09;
  339. *(b + 3) = data10;
  340. *(b + 4) = data03;
  341. *(b + 5) = data04;
  342. *(b + 6) = data11;
  343. *(b + 7) = data12;
  344. }
  345. a1 += 4;
  346. a2 += 4;
  347. b += 8;
  348. i --;
  349. ii += 2;
  350. }
  351. if (m & 1) {
  352. if (ii == jj) {
  353. #ifndef UNIT
  354. data01 = *(a1 + 0);
  355. data02 = *(a1 + 1);
  356. #endif
  357. data09 = *(a2 + 0);
  358. data10 = *(a2 + 1);
  359. compinv(b + 0, data01, data02);
  360. *(b + 2) = data09;
  361. *(b + 3) = data10;
  362. }
  363. if (ii < jj) {
  364. data01 = *(a1 + 0);
  365. data02 = *(a1 + 1);
  366. data09 = *(a2 + 0);
  367. data10 = *(a2 + 1);
  368. *(b + 0) = data01;
  369. *(b + 1) = data02;
  370. *(b + 2) = data09;
  371. *(b + 3) = data10;
  372. }
  373. a1 += 2;
  374. a2 += 2;
  375. b += 4;
  376. ii += 1;
  377. }
  378. a += 2 *lda;
  379. jj += 2;
  380. }
  381. if (n & 1) {
  382. a1 = a + 0 * lda;
  383. ii = 0;
  384. i = m;
  385. while (i > 0) {
  386. if (ii == jj) {
  387. #ifndef UNIT
  388. data01 = *(a1 + 0);
  389. data02 = *(a1 + 1);
  390. #endif
  391. compinv(b + 0, data01, data02);
  392. }
  393. if (ii < jj) {
  394. data01 = *(a1 + 0);
  395. data02 = *(a1 + 1);
  396. *(b + 0) = data01;
  397. *(b + 1) = data02;
  398. }
  399. a1 += 2;
  400. b += 2;
  401. i --;
  402. ii += 1;
  403. }
  404. a += lda;
  405. jj += 1;
  406. }
  407. return 0;
  408. }