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.

trmm_uncopy_4.c 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489
  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 posX, BLASLONG posY, FLOAT *b){
  41. BLASLONG i, js;
  42. BLASLONG X;
  43. FLOAT data01, data02, data03, data04, data05, data06, data07, data08;
  44. FLOAT data09, data10, data11, data12, data13, data14, data15, data16;
  45. FLOAT *ao1, *ao2, *ao3, *ao4;
  46. js = (n >> 2);
  47. if (js > 0){
  48. do {
  49. X = posX;
  50. if (posX <= posY) {
  51. ao1 = a + posX + (posY + 0) * lda;
  52. ao2 = a + posX + (posY + 1) * lda;
  53. ao3 = a + posX + (posY + 2) * lda;
  54. ao4 = a + posX + (posY + 3) * lda;
  55. } else {
  56. ao1 = a + posY + (posX + 0) * lda;
  57. ao2 = a + posY + (posX + 1) * lda;
  58. ao3 = a + posY + (posX + 2) * lda;
  59. ao4 = a + posY + (posX + 3) * lda;
  60. }
  61. i = (m >> 2);
  62. if (i > 0) {
  63. do {
  64. if (X < posY) {
  65. data01 = *(ao1 + 0);
  66. data02 = *(ao1 + 1);
  67. data03 = *(ao1 + 2);
  68. data04 = *(ao1 + 3);
  69. data05 = *(ao2 + 0);
  70. data06 = *(ao2 + 1);
  71. data07 = *(ao2 + 2);
  72. data08 = *(ao2 + 3);
  73. data09 = *(ao3 + 0);
  74. data10 = *(ao3 + 1);
  75. data11 = *(ao3 + 2);
  76. data12 = *(ao3 + 3);
  77. data13 = *(ao4 + 0);
  78. data14 = *(ao4 + 1);
  79. data15 = *(ao4 + 2);
  80. data16 = *(ao4 + 3);
  81. b[ 0] = data01;
  82. b[ 1] = data05;
  83. b[ 2] = data09;
  84. b[ 3] = data13;
  85. b[ 4] = data02;
  86. b[ 5] = data06;
  87. b[ 6] = data10;
  88. b[ 7] = data14;
  89. b[ 8] = data03;
  90. b[ 9] = data07;
  91. b[10] = data11;
  92. b[11] = data15;
  93. b[12] = data04;
  94. b[13] = data08;
  95. b[14] = data12;
  96. b[15] = data16;
  97. ao1 += 4;
  98. ao2 += 4;
  99. ao3 += 4;
  100. ao4 += 4;
  101. b += 16;
  102. } else
  103. if (X > posY) {
  104. ao1 += 4 * lda;
  105. ao2 += 4 * lda;
  106. ao3 += 4 * lda;
  107. ao4 += 4 * lda;
  108. b += 16;
  109. } else {
  110. #ifdef UNIT
  111. data05 = *(ao2 + 0);
  112. data09 = *(ao3 + 0);
  113. data10 = *(ao3 + 1);
  114. data13 = *(ao4 + 0);
  115. data14 = *(ao4 + 1);
  116. data15 = *(ao4 + 2);
  117. b[ 0] = ONE;
  118. b[ 1] = data05;
  119. b[ 2] = data09;
  120. b[ 3] = data13;
  121. b[ 4] = ZERO;
  122. b[ 5] = ONE;
  123. b[ 6] = data10;
  124. b[ 7] = data14;
  125. b[ 8] = ZERO;
  126. b[ 9] = ZERO;
  127. b[10] = ONE;
  128. b[11] = data15;
  129. b[12] = ZERO;
  130. b[13] = ZERO;
  131. b[14] = ZERO;
  132. b[15] = ONE;
  133. #else
  134. data01 = *(ao1 + 0);
  135. data05 = *(ao2 + 0);
  136. data06 = *(ao2 + 1);
  137. data09 = *(ao3 + 0);
  138. data10 = *(ao3 + 1);
  139. data11 = *(ao3 + 2);
  140. data13 = *(ao4 + 0);
  141. data14 = *(ao4 + 1);
  142. data15 = *(ao4 + 2);
  143. data16 = *(ao4 + 3);
  144. b[ 0] = data01;
  145. b[ 1] = data05;
  146. b[ 2] = data09;
  147. b[ 3] = data13;
  148. b[ 4] = ZERO;
  149. b[ 5] = data06;
  150. b[ 6] = data10;
  151. b[ 7] = data14;
  152. b[ 8] = ZERO;
  153. b[ 9] = ZERO;
  154. b[10] = data11;
  155. b[11] = data15;
  156. b[12] = ZERO;
  157. b[13] = ZERO;
  158. b[14] = ZERO;
  159. b[15] = data16;
  160. #endif
  161. ao1 += 4;
  162. ao2 += 4;
  163. ao3 += 4;
  164. ao4 += 4;
  165. b += 16;
  166. }
  167. X += 4;
  168. i --;
  169. } while (i > 0);
  170. }
  171. i = (m & 3);
  172. if (i) {
  173. if (X < posY) {
  174. if (m & 2) {
  175. data01 = *(ao1 + 0);
  176. data02 = *(ao1 + 1);
  177. data03 = *(ao2 + 0);
  178. data04 = *(ao2 + 1);
  179. data05 = *(ao3 + 0);
  180. data06 = *(ao3 + 1);
  181. data07 = *(ao4 + 0);
  182. data08 = *(ao4 + 1);
  183. b[ 0] = data01;
  184. b[ 1] = data03;
  185. b[ 2] = data05;
  186. b[ 3] = data07;
  187. b[ 4] = data02;
  188. b[ 5] = data04;
  189. b[ 6] = data06;
  190. b[ 7] = data08;
  191. ao1 += 2;
  192. ao2 += 2;
  193. ao3 += 2;
  194. ao4 += 2;
  195. b += 8;
  196. }
  197. if (m & 1) {
  198. data01 = *(ao1 + 0);
  199. data03 = *(ao2 + 0);
  200. data05 = *(ao3 + 0);
  201. data07 = *(ao4 + 0);
  202. b[ 0] = data01;
  203. b[ 1] = data03;
  204. b[ 2] = data05;
  205. b[ 3] = data07;
  206. /* ao1 += 1;
  207. ao2 += 1;
  208. ao3 += 1;
  209. ao4 += 1; */
  210. b += 4;
  211. }
  212. } else
  213. if (X > posY) {
  214. if (m & 2) {
  215. /* ao1 += 2 * lda;
  216. ao2 += 2 * lda; */
  217. b += 8;
  218. }
  219. if (m & 1) {
  220. // ao1 += lda;
  221. b += 4;
  222. }
  223. } else {
  224. #ifdef UNIT
  225. data05 = *(ao2 + 0);
  226. data09 = *(ao3 + 0);
  227. data13 = *(ao4 + 0);
  228. if (i >= 2) {
  229. data10 = *(ao3 + 1);
  230. data14 = *(ao4 + 1);
  231. }
  232. if (i >= 3) {
  233. data15 = *(ao4 + 2);
  234. }
  235. b[ 0] = ONE;
  236. b[ 1] = data05;
  237. b[ 2] = data09;
  238. b[ 3] = data13;
  239. b += 4;
  240. if(i >= 2) {
  241. b[ 0] = ZERO;
  242. b[ 1] = ONE;
  243. b[ 2] = data10;
  244. b[ 3] = data14;
  245. b += 4;
  246. }
  247. if (i >= 3) {
  248. b[ 0] = ZERO;
  249. b[ 1] = ZERO;
  250. b[ 2] = ONE;
  251. b[ 3] = data15;
  252. b += 4;
  253. }
  254. #else
  255. data01 = *(ao1 + 0);
  256. data05 = *(ao2 + 0);
  257. data09 = *(ao3 + 0);
  258. data13 = *(ao4 + 0);
  259. if (i >= 2) {
  260. data06 = *(ao2 + 1);
  261. data10 = *(ao3 + 1);
  262. data14 = *(ao4 + 1);
  263. }
  264. if (i >= 3) {
  265. data11 = *(ao3 + 2);
  266. data15 = *(ao4 + 2);
  267. }
  268. b[ 0] = data01;
  269. b[ 1] = data05;
  270. b[ 2] = data09;
  271. b[ 3] = data13;
  272. b += 4;
  273. if(i >= 2) {
  274. b[ 0] = ZERO;
  275. b[ 1] = data06;
  276. b[ 2] = data10;
  277. b[ 3] = data14;
  278. b += 4;
  279. }
  280. if (i >= 3) {
  281. b[ 0] = ZERO;
  282. b[ 1] = ZERO;
  283. b[ 2] = data11;
  284. b[ 3] = data15;
  285. b += 4;
  286. }
  287. #endif
  288. }
  289. }
  290. posY += 4;
  291. js --;
  292. } while (js > 0);
  293. } /* End of main loop */
  294. if (n & 2){
  295. X = posX;
  296. if (posX <= posY) {
  297. ao1 = a + posX + (posY + 0) * lda;
  298. ao2 = a + posX + (posY + 1) * lda;
  299. } else {
  300. ao1 = a + posY + (posX + 0) * lda;
  301. ao2 = a + posY + (posX + 1) * lda;
  302. }
  303. i = (m >> 1);
  304. if (i > 0) {
  305. do {
  306. if (X < posY) {
  307. data01 = *(ao1 + 0);
  308. data02 = *(ao1 + 1);
  309. data05 = *(ao2 + 0);
  310. data06 = *(ao2 + 1);
  311. b[ 0] = data01;
  312. b[ 1] = data05;
  313. b[ 2] = data02;
  314. b[ 3] = data06;
  315. ao1 += 2;
  316. ao2 += 2;
  317. b += 4;
  318. } else
  319. if (X > posY) {
  320. ao1 += 2 * lda;
  321. ao2 += 2 * lda;
  322. b += 4;
  323. } else {
  324. #ifdef UNIT
  325. data05 = *(ao2 + 0);
  326. b[ 0] = ONE;
  327. b[ 1] = data05;
  328. b[ 2] = ZERO;
  329. b[ 3] = ONE;
  330. #else
  331. data01 = *(ao1 + 0);
  332. data05 = *(ao2 + 0);
  333. data06 = *(ao2 + 1);
  334. b[ 0] = data01;
  335. b[ 1] = data05;
  336. b[ 2] = ZERO;
  337. b[ 3] = data06;
  338. #endif
  339. ao1 += 2 * lda;
  340. ao2 += 2 * lda;
  341. b += 4;
  342. }
  343. X += 2;
  344. i --;
  345. } while (i > 0);
  346. }
  347. i = (m & 1);
  348. if (i) {
  349. if (X < posY) {
  350. data01 = *(ao1 + 0);
  351. data05 = *(ao2 + 0);
  352. b[ 0] = data01;
  353. b[ 1] = data05;
  354. /* ao1 += 1;
  355. ao2 += 1; */
  356. b += 2;
  357. } else
  358. if (X > posY) {
  359. /* ao1 += lda;
  360. ao2 += lda; */
  361. b += 2;
  362. } else {
  363. #ifdef UNIT
  364. data05 = *(ao2 + 0);
  365. b[ 0] = ONE;
  366. b[ 1] = data05;
  367. #else
  368. data01 = *(ao1 + 0);
  369. data05 = *(ao2 + 0);
  370. b[ 0] = data01;
  371. b[ 1] = data05;
  372. #endif
  373. /* ao1 += lda;
  374. ao2 += lda; */
  375. b += 2;
  376. }
  377. }
  378. posY += 2;
  379. }
  380. if (n & 1){
  381. X = posX;
  382. if (posX <= posY) {
  383. ao1 = a + posX + (posY + 0) * lda;
  384. } else {
  385. ao1 = a + posY + (posX + 0) * lda;
  386. }
  387. i = m;
  388. if (m > 0) {
  389. do {
  390. if (X < posY) {
  391. data01 = *(ao1 + 0);
  392. b[ 0] = data01;
  393. ao1 += 1;
  394. b += 1;
  395. } else
  396. if (X > posY) {
  397. ao1 += lda;
  398. b += 1;
  399. } else {
  400. #ifdef UNIT
  401. b[ 0] = ONE;
  402. #else
  403. data01 = *(ao1 + 0);
  404. b[ 0] = data01;
  405. #endif
  406. ao1 += lda;
  407. b += 1;
  408. }
  409. X += 1;
  410. i --;
  411. } while (i > 0);
  412. }
  413. }
  414. return 0;
  415. }