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_6.c 18 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584
  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, ii;
  42. BLASLONG X;
  43. FLOAT data01, data02, data05, data06;
  44. FLOAT *ao1, *ao2, *ao3, *ao4, *ao5, *ao6;
  45. js = n/6;
  46. if (js > 0){
  47. do {
  48. X = posX;
  49. if (posX <= posY) {
  50. ao1 = a + posX + (posY + 0) * lda;
  51. ao2 = a + posX + (posY + 1) * lda;
  52. ao3 = a + posX + (posY + 2) * lda;
  53. ao4 = a + posX + (posY + 3) * lda;
  54. ao5 = a + posX + (posY + 4) * lda;
  55. ao6 = a + posX + (posY + 5) * lda;
  56. } else {
  57. ao1 = a + posY + (posX + 0) * lda;
  58. ao2 = a + posY + (posX + 1) * lda;
  59. ao3 = a + posY + (posX + 2) * lda;
  60. ao4 = a + posY + (posX + 3) * lda;
  61. ao5 = a + posY + (posX + 4) * lda;
  62. ao6 = a + posY + (posX + 5) * lda;
  63. }
  64. i = m/6;
  65. if (i > 0) {
  66. do {
  67. if (X < posY) {
  68. for (ii = 0; ii < 6; ii++){
  69. b[ 0] = *(ao1 + 0);
  70. b[ 1] = *(ao2 + 0);
  71. b[ 2] = *(ao3 + 0);
  72. b[ 3] = *(ao4 + 0);
  73. b[ 4] = *(ao5 + 0);
  74. b[ 5] = *(ao6 + 0);
  75. ao1 ++;
  76. ao2 ++;
  77. ao3 ++;
  78. ao4 ++;
  79. ao5 ++;
  80. ao6 ++;
  81. b += 6;
  82. }
  83. } else if (X > posY) {
  84. // b[ 0] = ZERO;
  85. // b[ 1] = ZERO;
  86. // b[ 2] = ZERO;
  87. // b[ 3] = ZERO;
  88. // b[ 4] = ZERO;
  89. // b[ 5] = ZERO;
  90. // b[ 6] = ZERO;
  91. // b[ 7] = ZERO;
  92. // b[ 8] = ZERO;
  93. // b[ 9] = ZERO;
  94. // b[10] = ZERO;
  95. // b[11] = ZERO;
  96. // b[12] = ZERO;
  97. // b[13] = ZERO;
  98. // b[14] = ZERO;
  99. // b[15] = ZERO;
  100. // b[16] = ZERO;
  101. // b[17] = ZERO;
  102. // b[18] = ZERO;
  103. // b[19] = ZERO;
  104. // b[20] = ZERO;
  105. // b[21] = ZERO;
  106. // b[22] = ZERO;
  107. // b[23] = ZERO;
  108. // b[24] = ZERO;
  109. // b[25] = ZERO;
  110. // b[26] = ZERO;
  111. // b[27] = ZERO;
  112. // b[28] = ZERO;
  113. // b[29] = ZERO;
  114. // b[30] = ZERO;
  115. // b[31] = ZERO;
  116. // b[32] = ZERO;
  117. // b[33] = ZERO;
  118. // b[34] = ZERO;
  119. // b[35] = ZERO;
  120. ao1 += 6 * lda;
  121. ao2 += 6 * lda;
  122. ao3 += 6 * lda;
  123. ao4 += 6 * lda;
  124. ao5 += 6 * lda;
  125. ao6 += 6 * lda;
  126. b += 36;
  127. } else {
  128. #ifdef UNIT
  129. b[ 0] = ONE;
  130. #else
  131. b[ 0] = *(ao1 + 0);
  132. #endif
  133. b[ 1] = *(ao2 + 0);
  134. b[ 2] = *(ao3 + 0);
  135. b[ 3] = *(ao4 + 0);
  136. b[ 4] = *(ao5 + 0);
  137. b[ 5] = *(ao6 + 0);
  138. b[ 6] = ZERO;
  139. #ifdef UNIT
  140. b[ 7] = ONE;
  141. #else
  142. b[ 7] = *(ao2 + 1);
  143. #endif
  144. b[ 8] = *(ao3 + 1);
  145. b[ 9] = *(ao4 + 1);
  146. b[10] = *(ao5 + 1);
  147. b[11] = *(ao6 + 1);
  148. b[ 12] = ZERO;
  149. b[ 13] = ZERO;
  150. #ifdef UNIT
  151. b[ 14] = ONE;
  152. #else
  153. b[ 14] = *(ao3 + 2);
  154. #endif
  155. b[ 15] = *(ao4 + 2);
  156. b[ 16] = *(ao5 + 2);
  157. b[ 17] = *(ao6 + 2);
  158. b[ 18] = ZERO;
  159. b[ 19] = ZERO;
  160. b[ 20] = ZERO;
  161. #ifdef UNIT
  162. b[ 21] = ONE;
  163. #else
  164. b[ 21] = *(ao4 + 3);
  165. #endif
  166. b[ 22] = *(ao5 + 3);
  167. b[ 23] = *(ao6 + 3);
  168. b[ 24] = ZERO;
  169. b[ 25] = ZERO;
  170. b[ 26] = ZERO;
  171. b[ 27] = ZERO;
  172. #ifdef UNIT
  173. b[ 28] = ONE;
  174. #else
  175. b[ 28] = *(ao5 + 4);
  176. #endif
  177. b[ 29] = *(ao6 + 4);
  178. b[ 30] = ZERO;
  179. b[ 31] = ZERO;
  180. b[ 32] = ZERO;
  181. b[ 33] = ZERO;
  182. b[ 34] = ZERO;
  183. #ifdef UNIT
  184. b[ 35] = ONE;
  185. #else
  186. b[ 35] = *(ao6 + 5);
  187. #endif
  188. ao1 += 6 * lda;
  189. ao2 += 6 * lda;
  190. ao3 += 6 * lda;
  191. ao4 += 6 * lda;
  192. ao5 += 6 * lda;
  193. ao6 += 6 * lda;
  194. b += 36;
  195. }
  196. X += 6;
  197. i --;
  198. } while (i > 0);
  199. }
  200. i = m % 6;
  201. if (i) {
  202. if (X < posY) {
  203. for (ii = 0; ii < i; ii++){
  204. b[ 0] = *(ao1 + 0);
  205. b[ 1] = *(ao2 + 0);
  206. b[ 2] = *(ao3 + 0);
  207. b[ 3] = *(ao4 + 0);
  208. b[ 4] = *(ao5 + 0);
  209. b[ 5] = *(ao6 + 0);
  210. ao1 ++;
  211. ao2 ++;
  212. ao3 ++;
  213. ao4 ++;
  214. ao5 ++;
  215. ao6 ++;
  216. b += 6;
  217. }
  218. } else if (X > posY) {
  219. b += 6 * i;
  220. } else {
  221. #ifdef UNIT
  222. b[ 0] = ONE;
  223. #else
  224. b[ 0] = *(ao1 + 0);
  225. #endif
  226. b[ 1] = *(ao2 + 0);
  227. b[ 2] = *(ao3 + 0);
  228. b[ 3] = *(ao4 + 0);
  229. b[ 4] = *(ao5 + 0);
  230. b[ 5] = *(ao6 + 0);
  231. b += 6;
  232. if (i >= 2) {
  233. b[ 0] = ZERO;
  234. #ifdef UNIT
  235. b[ 1] = ONE;
  236. #else
  237. b[ 1] = *(ao2 + 1);
  238. #endif
  239. b[ 2] = *(ao3 + 1);
  240. b[ 3] = *(ao4 + 1);
  241. b[ 4] = *(ao5 + 1);
  242. b[ 5] = *(ao6 + 1);
  243. b += 6;
  244. }
  245. if (i >= 3) {
  246. b[ 0] = ZERO;
  247. b[ 1] = ZERO;
  248. #ifdef UNIT
  249. b[ 2] = ONE;
  250. #else
  251. b[ 2] = *(ao3 + 2);
  252. #endif
  253. b[ 3] = *(ao4 + 2);
  254. b[ 4] = *(ao5 + 2);
  255. b[ 5] = *(ao6 + 2);
  256. b += 6;
  257. }
  258. if (i >= 4) {
  259. b[ 0] = ZERO;
  260. b[ 1] = ZERO;
  261. b[ 2] = ZERO;
  262. #ifdef UNIT
  263. b[ 3] = ONE;
  264. #else
  265. b[ 3] = *(ao4 + 3);
  266. #endif
  267. b[ 4] = *(ao5 + 3);
  268. b[ 5] = *(ao6 + 3);
  269. b += 6;
  270. }
  271. if (i >= 5) {
  272. b[ 0] = ZERO;
  273. b[ 1] = ZERO;
  274. b[ 2] = ZERO;
  275. b[ 3] = ZERO;
  276. #ifdef UNIT
  277. b[ 4] = ONE;
  278. #else
  279. b[ 4] = *(ao5 + 4);
  280. #endif
  281. b[ 5] = *(ao6 + 4);
  282. b += 6;
  283. }
  284. }
  285. }
  286. posY += 6;
  287. js --;
  288. } while (js > 0);
  289. } /* End of main loop */
  290. if ((n % 6) & 4){
  291. X = posX;
  292. if (posX <= posY) {
  293. ao1 = a + posX + (posY + 0) * lda;
  294. ao2 = a + posX + (posY + 1) * lda;
  295. ao3 = a + posX + (posY + 2) * lda;
  296. ao4 = a + posX + (posY + 3) * lda;
  297. } else {
  298. ao1 = a + posY + (posX + 0) * lda;
  299. ao2 = a + posY + (posX + 1) * lda;
  300. ao3 = a + posY + (posX + 2) * lda;
  301. ao4 = a + posY + (posX + 3) * lda;
  302. }
  303. i = (m >> 1);
  304. if (i > 0) {
  305. do {
  306. if (X < posY) {
  307. for (ii = 0; ii < 2; ii++){
  308. b[ 0] = *(ao1 + 0);
  309. b[ 1] = *(ao2 + 0);
  310. b[ 2] = *(ao3 + 0);
  311. b[ 3] = *(ao4 + 0);
  312. ao1 ++;
  313. ao2 ++;
  314. ao3 ++;
  315. ao4 ++;
  316. b += 4;
  317. }
  318. } else if (X > posY) {
  319. ao1 += 2 * lda;
  320. ao2 += 2 * lda;
  321. ao3 += 2 * lda;
  322. ao4 += 2 * lda;
  323. b += 8;
  324. } else {
  325. #ifdef UNIT
  326. b[ 0] = ONE;
  327. #else
  328. b[ 0] = *(ao1 + 0);
  329. #endif
  330. b[ 1] = *(ao2 + 0);
  331. b[ 2] = *(ao3 + 0);
  332. b[ 3] = *(ao4 + 0);
  333. b[ 4] = ZERO;
  334. #ifdef UNIT
  335. b[ 5] = ONE;
  336. #else
  337. b[ 5] = *(ao2 + 1);
  338. #endif
  339. b[ 6] = *(ao3 + 1);
  340. b[ 7] = *(ao4 + 1);
  341. b[ 8] = ZERO;
  342. b[ 9] = ZERO;
  343. #ifdef UNIT
  344. b[ 10] = ONE;
  345. #else
  346. b[ 10] = *(ao3 + 2);
  347. #endif
  348. b[ 11] = *(ao4 + 2);
  349. b[ 12] = ZERO;
  350. b[ 13] = ZERO;
  351. b[ 14] = ZERO;
  352. #ifdef UNIT
  353. b[ 15] = ONE;
  354. #else
  355. b[ 15] = *(ao4 + 3);
  356. #endif
  357. ao1 += 4 * lda;
  358. ao2 += 4 * lda;
  359. ao3 += 4 * lda;
  360. ao4 += 4 * lda;
  361. b += 16;
  362. X += 4;
  363. i -= 2;
  364. continue;
  365. }
  366. X += 2;
  367. i --;
  368. } while (i > 0);
  369. }
  370. i = (m & 1);
  371. if (i) {
  372. if (X < posY) {
  373. for (ii = 0; ii < i; ii++){
  374. b[ 0] = *(ao1 + 0);
  375. b[ 1] = *(ao2 + 0);
  376. b[ 2] = *(ao3 + 0);
  377. b[ 3] = *(ao4 + 0);
  378. ao1 ++;
  379. ao2 ++;
  380. ao3 ++;
  381. ao4 ++;
  382. b += 4;
  383. }
  384. } else if (X > posY) {
  385. /* ao1 += i * lda;
  386. ao2 += i * lda;
  387. ao3 += i * lda;
  388. ao4 += i * lda; */
  389. b += 4 * i;
  390. } else {
  391. #ifdef UNIT
  392. b[ 0] = ONE;
  393. #else
  394. b[ 0] = *(ao1 + 0);
  395. #endif
  396. b[ 1] = *(ao2 + 0);
  397. b[ 2] = *(ao3 + 0);
  398. b[ 3] = *(ao4 + 0);
  399. b += 4;
  400. }
  401. }
  402. posY += 4;
  403. }
  404. if ((n % 6) & 2){
  405. X = posX;
  406. if (posX <= posY) {
  407. ao1 = a + posX + (posY + 0) * lda;
  408. ao2 = a + posX + (posY + 1) * lda;
  409. } else {
  410. ao1 = a + posY + (posX + 0) * lda;
  411. ao2 = a + posY + (posX + 1) * lda;
  412. }
  413. i = (m >> 1);
  414. if (i > 0) {
  415. do {
  416. if (X < posY) {
  417. data01 = *(ao1 + 0);
  418. data02 = *(ao1 + 1);
  419. data05 = *(ao2 + 0);
  420. data06 = *(ao2 + 1);
  421. b[ 0] = data01;
  422. b[ 1] = data05;
  423. b[ 2] = data02;
  424. b[ 3] = data06;
  425. ao1 += 2;
  426. ao2 += 2;
  427. b += 4;
  428. } else if (X > posY) {
  429. ao1 += 2 * lda;
  430. ao2 += 2 * lda;
  431. b += 4;
  432. } else {
  433. #ifdef UNIT
  434. data05 = *(ao2 + 0);
  435. b[ 0] = ONE;
  436. b[ 1] = data05;
  437. b[ 2] = ZERO;
  438. b[ 3] = ONE;
  439. #else
  440. data01 = *(ao1 + 0);
  441. data05 = *(ao2 + 0);
  442. data06 = *(ao2 + 1);
  443. b[ 0] = data01;
  444. b[ 1] = data05;
  445. b[ 2] = ZERO;
  446. b[ 3] = data06;
  447. #endif
  448. ao1 += 2 * lda;
  449. ao2 += 2 * lda;
  450. b += 4;
  451. }
  452. X += 2;
  453. i --;
  454. } while (i > 0);
  455. }
  456. i = (m & 1);
  457. if (i) {
  458. if (X < posY) {
  459. data01 = *(ao1 + 0);
  460. data05 = *(ao2 + 0);
  461. b[ 0] = data01;
  462. b[ 1] = data05;
  463. ao1 += 1;
  464. ao2 += 1;
  465. b += 2;
  466. } else if (X > posY) {
  467. ao1 += lda;
  468. ao2 += lda;
  469. b += 2;
  470. } else {
  471. #ifdef UNIT
  472. data05 = *(ao2 + 0);
  473. b[ 0] = ONE;
  474. b[ 1] = data05;
  475. #else
  476. data01 = *(ao1 + 0);
  477. data05 = *(ao2 + 0);
  478. b[ 0] = data01;
  479. b[ 1] = data05;
  480. #endif
  481. ao1 += lda;
  482. ao2 += lda;
  483. b += 2;
  484. }
  485. }
  486. posY += 2;
  487. }
  488. if ((n % 6) & 1){
  489. X = posX;
  490. if (posX <= posY) {
  491. ao1 = a + posX + (posY + 0) * lda;
  492. } else {
  493. ao1 = a + posY + (posX + 0) * lda;
  494. }
  495. i = m;
  496. if (m > 0) {
  497. do {
  498. if (X < posY) {
  499. data01 = *(ao1 + 0);
  500. b[ 0] = data01;
  501. ao1 += 1;
  502. b += 1;
  503. } else if (X > posY) {
  504. ao1 += lda;
  505. b += 1;
  506. } else {
  507. #ifdef UNIT
  508. b[ 0] = ONE;
  509. #else
  510. data01 = *(ao1 + 0);
  511. b[ 0] = data01;
  512. #endif
  513. ao1 += lda;
  514. b += 1;
  515. }
  516. X += 1;
  517. i --;
  518. } while (i > 0);
  519. }
  520. }
  521. return 0;
  522. }