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.

ztrmm_utcopy_4.c 14 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663
  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 data17, data18, data19, data20, data21, data22, data23, data24;
  46. FLOAT data25, data26, data27, data28, data29, data30, data31, data32;
  47. FLOAT *ao1, *ao2, *ao3, *ao4;
  48. lda += lda;
  49. js = (n >> 2);
  50. if (js > 0){
  51. do {
  52. X = posX;
  53. if (posX <= posY) {
  54. ao1 = a + posX * 2 + (posY + 0) * lda;
  55. ao2 = a + posX * 2 + (posY + 1) * lda;
  56. ao3 = a + posX * 2 + (posY + 2) * lda;
  57. ao4 = a + posX * 2 + (posY + 3) * lda;
  58. } else {
  59. ao1 = a + posY * 2 + (posX + 0) * lda;
  60. ao2 = a + posY * 2 + (posX + 1) * lda;
  61. ao3 = a + posY * 2 + (posX + 2) * lda;
  62. ao4 = a + posY * 2 + (posX + 3) * lda;
  63. }
  64. i = (m >> 2);
  65. if (i > 0) {
  66. do {
  67. if (X < posY) {
  68. ao1 += 8;
  69. ao2 += 8;
  70. ao3 += 8;
  71. ao4 += 8;
  72. b += 32;
  73. } else
  74. if (X > posY) {
  75. data01 = *(ao1 + 0);
  76. data02 = *(ao1 + 1);
  77. data03 = *(ao1 + 2);
  78. data04 = *(ao1 + 3);
  79. data05 = *(ao1 + 4);
  80. data06 = *(ao1 + 5);
  81. data07 = *(ao1 + 6);
  82. data08 = *(ao1 + 7);
  83. data09 = *(ao2 + 0);
  84. data10 = *(ao2 + 1);
  85. data11 = *(ao2 + 2);
  86. data12 = *(ao2 + 3);
  87. data13 = *(ao2 + 4);
  88. data14 = *(ao2 + 5);
  89. data15 = *(ao2 + 6);
  90. data16 = *(ao2 + 7);
  91. data17 = *(ao3 + 0);
  92. data18 = *(ao3 + 1);
  93. data19 = *(ao3 + 2);
  94. data20 = *(ao3 + 3);
  95. data21 = *(ao3 + 4);
  96. data22 = *(ao3 + 5);
  97. data23 = *(ao3 + 6);
  98. data24 = *(ao3 + 7);
  99. data25 = *(ao4 + 0);
  100. data26 = *(ao4 + 1);
  101. data27 = *(ao4 + 2);
  102. data28 = *(ao4 + 3);
  103. data29 = *(ao4 + 4);
  104. data30 = *(ao4 + 5);
  105. data31 = *(ao4 + 6);
  106. data32 = *(ao4 + 7);
  107. b[ 0] = data01;
  108. b[ 1] = data02;
  109. b[ 2] = data03;
  110. b[ 3] = data04;
  111. b[ 4] = data05;
  112. b[ 5] = data06;
  113. b[ 6] = data07;
  114. b[ 7] = data08;
  115. b[ 8] = data09;
  116. b[ 9] = data10;
  117. b[10] = data11;
  118. b[11] = data12;
  119. b[12] = data13;
  120. b[13] = data14;
  121. b[14] = data15;
  122. b[15] = data16;
  123. b[16] = data17;
  124. b[17] = data18;
  125. b[18] = data19;
  126. b[19] = data20;
  127. b[20] = data21;
  128. b[21] = data22;
  129. b[22] = data23;
  130. b[23] = data24;
  131. b[24] = data25;
  132. b[25] = data26;
  133. b[26] = data27;
  134. b[27] = data28;
  135. b[28] = data29;
  136. b[29] = data30;
  137. b[30] = data31;
  138. b[31] = data32;
  139. ao1 += 4 * lda;
  140. ao2 += 4 * lda;
  141. ao3 += 4 * lda;
  142. ao4 += 4 * lda;
  143. b += 32;
  144. } else {
  145. #ifdef UNIT
  146. data09 = *(ao2 + 0);
  147. data10 = *(ao2 + 1);
  148. data17 = *(ao3 + 0);
  149. data18 = *(ao3 + 1);
  150. data19 = *(ao3 + 2);
  151. data20 = *(ao3 + 3);
  152. data25 = *(ao4 + 0);
  153. data26 = *(ao4 + 1);
  154. data27 = *(ao4 + 2);
  155. data28 = *(ao4 + 3);
  156. data29 = *(ao4 + 4);
  157. data30 = *(ao4 + 5);
  158. b[ 0] = ONE;
  159. b[ 1] = ZERO;
  160. b[ 2] = ZERO;
  161. b[ 3] = ZERO;
  162. b[ 4] = ZERO;
  163. b[ 5] = ZERO;
  164. b[ 6] = ZERO;
  165. b[ 7] = ZERO;
  166. b[ 8] = data09;
  167. b[ 9] = data10;
  168. b[10] = ONE;
  169. b[11] = ZERO;
  170. b[12] = ZERO;
  171. b[13] = ZERO;
  172. b[14] = ZERO;
  173. b[15] = ZERO;
  174. b[16] = data17;
  175. b[17] = data18;
  176. b[18] = data19;
  177. b[19] = data20;
  178. b[20] = ONE;
  179. b[21] = ZERO;
  180. b[22] = ZERO;
  181. b[23] = ZERO;
  182. b[24] = data25;
  183. b[25] = data26;
  184. b[26] = data27;
  185. b[27] = data28;
  186. b[28] = data29;
  187. b[29] = data30;
  188. b[30] = ONE;
  189. b[31] = ZERO;
  190. #else
  191. data01 = *(ao1 + 0);
  192. data02 = *(ao1 + 1);
  193. data09 = *(ao2 + 0);
  194. data10 = *(ao2 + 1);
  195. data11 = *(ao2 + 2);
  196. data12 = *(ao2 + 3);
  197. data17 = *(ao3 + 0);
  198. data18 = *(ao3 + 1);
  199. data19 = *(ao3 + 2);
  200. data20 = *(ao3 + 3);
  201. data21 = *(ao3 + 4);
  202. data22 = *(ao3 + 5);
  203. data25 = *(ao4 + 0);
  204. data26 = *(ao4 + 1);
  205. data27 = *(ao4 + 2);
  206. data28 = *(ao4 + 3);
  207. data29 = *(ao4 + 4);
  208. data30 = *(ao4 + 5);
  209. data31 = *(ao4 + 6);
  210. data32 = *(ao4 + 7);
  211. b[ 0] = data01;
  212. b[ 1] = data02;
  213. b[ 2] = ZERO;
  214. b[ 3] = ZERO;
  215. b[ 4] = ZERO;
  216. b[ 5] = ZERO;
  217. b[ 6] = ZERO;
  218. b[ 7] = ZERO;
  219. b[ 8] = data09;
  220. b[ 9] = data10;
  221. b[10] = data11;
  222. b[11] = data12;
  223. b[12] = ZERO;
  224. b[13] = ZERO;
  225. b[14] = ZERO;
  226. b[15] = ZERO;
  227. b[16] = data17;
  228. b[17] = data18;
  229. b[18] = data19;
  230. b[19] = data20;
  231. b[20] = data21;
  232. b[21] = data22;
  233. b[22] = ZERO;
  234. b[23] = ZERO;
  235. b[24] = data25;
  236. b[25] = data26;
  237. b[26] = data27;
  238. b[27] = data28;
  239. b[28] = data29;
  240. b[29] = data30;
  241. b[30] = data31;
  242. b[31] = data32;
  243. #endif
  244. ao1 += 4 * lda;
  245. ao2 += 4 * lda;
  246. ao3 += 4 * lda;
  247. ao4 += 4 * lda;
  248. b += 32;
  249. }
  250. X += 4;
  251. i --;
  252. } while (i > 0);
  253. }
  254. i = (m & 3);
  255. if (i) {
  256. if (X < posY) {
  257. if (m & 2) {
  258. /* ao1 += 4;
  259. ao2 += 4;
  260. ao3 += 4;
  261. ao4 += 4; */
  262. b += 16;
  263. }
  264. if (m & 1) {
  265. /* ao1 += 2;
  266. ao2 += 2;
  267. ao3 += 2;
  268. ao4 += 2; */
  269. b += 8;
  270. }
  271. } else
  272. if (X > posY) {
  273. if (m & 2) {
  274. data01 = *(ao1 + 0);
  275. data02 = *(ao1 + 1);
  276. data03 = *(ao1 + 2);
  277. data04 = *(ao1 + 3);
  278. data05 = *(ao1 + 4);
  279. data06 = *(ao1 + 5);
  280. data07 = *(ao1 + 6);
  281. data08 = *(ao1 + 7);
  282. data09 = *(ao2 + 0);
  283. data10 = *(ao2 + 1);
  284. data11 = *(ao2 + 2);
  285. data12 = *(ao2 + 3);
  286. data13 = *(ao2 + 4);
  287. data14 = *(ao2 + 5);
  288. data15 = *(ao2 + 6);
  289. data16 = *(ao2 + 7);
  290. b[ 0] = data01;
  291. b[ 1] = data02;
  292. b[ 2] = data03;
  293. b[ 3] = data04;
  294. b[ 4] = data05;
  295. b[ 5] = data06;
  296. b[ 6] = data07;
  297. b[ 7] = data08;
  298. b[ 8] = data09;
  299. b[ 9] = data10;
  300. b[10] = data11;
  301. b[11] = data12;
  302. b[12] = data13;
  303. b[13] = data14;
  304. b[14] = data15;
  305. b[15] = data16;
  306. ao1 += 2 * lda;
  307. // ao2 += 2 * lda;
  308. b += 16;
  309. }
  310. if (m & 1) {
  311. data01 = *(ao1 + 0);
  312. data02 = *(ao1 + 1);
  313. data03 = *(ao1 + 2);
  314. data04 = *(ao1 + 3);
  315. data05 = *(ao1 + 4);
  316. data06 = *(ao1 + 5);
  317. data07 = *(ao1 + 6);
  318. data08 = *(ao1 + 7);
  319. b[ 0] = data01;
  320. b[ 1] = data02;
  321. b[ 2] = data03;
  322. b[ 3] = data04;
  323. b[ 4] = data05;
  324. b[ 5] = data06;
  325. b[ 6] = data07;
  326. b[ 7] = data08;
  327. // ao1 += lda;
  328. b += 8;
  329. }
  330. } else {
  331. #ifdef UNIT
  332. if (i >= 2) {
  333. data09 = *(ao2 + 0);
  334. data10 = *(ao2 + 1);
  335. }
  336. if (i >= 3) {
  337. data17 = *(ao3 + 0);
  338. data18 = *(ao3 + 1);
  339. data19 = *(ao3 + 2);
  340. data20 = *(ao3 + 3);
  341. }
  342. b[ 0] = ONE;
  343. b[ 1] = ZERO;
  344. b[ 2] = ZERO;
  345. b[ 3] = ZERO;
  346. b[ 4] = ZERO;
  347. b[ 5] = ZERO;
  348. b[ 6] = ZERO;
  349. b[ 7] = ZERO;
  350. b += 8;
  351. if (i >= 2) {
  352. b[ 0] = data09;
  353. b[ 1] = data10;
  354. b[ 2] = ONE;
  355. b[ 3] = ZERO;
  356. b[ 4] = ZERO;
  357. b[ 5] = ZERO;
  358. b[ 6] = ZERO;
  359. b[ 7] = ZERO;
  360. b += 8;
  361. }
  362. if (i >= 3) {
  363. b[ 0] = data17;
  364. b[ 1] = data18;
  365. b[ 2] = data19;
  366. b[ 3] = data20;
  367. b[ 4] = ONE;
  368. b[ 5] = ZERO;
  369. b[ 6] = ZERO;
  370. b[ 7] = ZERO;
  371. b += 8;
  372. }
  373. #else
  374. data01 = *(ao1 + 0);
  375. data02 = *(ao1 + 1);
  376. if (i >= 2) {
  377. data09 = *(ao2 + 0);
  378. data10 = *(ao2 + 1);
  379. data11 = *(ao2 + 2);
  380. data12 = *(ao2 + 3);
  381. }
  382. if (i >= 3) {
  383. data17 = *(ao3 + 0);
  384. data18 = *(ao3 + 1);
  385. data19 = *(ao3 + 2);
  386. data20 = *(ao3 + 3);
  387. data21 = *(ao3 + 4);
  388. data22 = *(ao3 + 5);
  389. }
  390. b[ 0] = data01;
  391. b[ 1] = data02;
  392. b[ 2] = ZERO;
  393. b[ 3] = ZERO;
  394. b[ 4] = ZERO;
  395. b[ 5] = ZERO;
  396. b[ 6] = ZERO;
  397. b[ 7] = ZERO;
  398. b += 8;
  399. if (i >= 2) {
  400. b[ 0] = data09;
  401. b[ 1] = data10;
  402. b[ 2] = data11;
  403. b[ 3] = data12;
  404. b[ 4] = ZERO;
  405. b[ 5] = ZERO;
  406. b[ 6] = ZERO;
  407. b[ 7] = ZERO;
  408. b += 8;
  409. }
  410. if (i >= 3) {
  411. b[ 0] = data17;
  412. b[ 1] = data18;
  413. b[ 2] = data19;
  414. b[ 3] = data20;
  415. b[ 4] = data21;
  416. b[ 5] = data22;
  417. b[ 6] = ZERO;
  418. b[ 7] = ZERO;
  419. b += 8;
  420. }
  421. #endif
  422. }
  423. }
  424. posY += 4;
  425. js --;
  426. } while (js > 0);
  427. } /* End of main loop */
  428. if (n & 2){
  429. X = posX;
  430. if (posX <= posY) {
  431. ao1 = a + posX * 2 + (posY + 0) * lda;
  432. ao2 = a + posX * 2 + (posY + 1) * lda;
  433. } else {
  434. ao1 = a + posY * 2 + (posX + 0) * lda;
  435. ao2 = a + posY * 2 + (posX + 1) * lda;
  436. }
  437. i = (m >> 1);
  438. if (i > 0) {
  439. do {
  440. if (X < posY) {
  441. ao1 += 4;
  442. ao2 += 4;
  443. b += 8;
  444. } else
  445. if (X > posY) {
  446. data01 = *(ao1 + 0);
  447. data02 = *(ao1 + 1);
  448. data03 = *(ao1 + 2);
  449. data04 = *(ao1 + 3);
  450. data09 = *(ao2 + 0);
  451. data10 = *(ao2 + 1);
  452. data11 = *(ao2 + 2);
  453. data12 = *(ao2 + 3);
  454. b[ 0] = data01;
  455. b[ 1] = data02;
  456. b[ 2] = data03;
  457. b[ 3] = data04;
  458. b[ 4] = data09;
  459. b[ 5] = data10;
  460. b[ 6] = data11;
  461. b[ 7] = data12;
  462. ao1 += 2 * lda;
  463. ao2 += 2 * lda;
  464. b += 8;
  465. } else {
  466. #ifdef UNIT
  467. data09 = *(ao2 + 0);
  468. data10 = *(ao2 + 1);
  469. b[ 0] = ONE;
  470. b[ 1] = ZERO;
  471. b[ 2] = ZERO;
  472. b[ 3] = ZERO;
  473. b[ 4] = data09;
  474. b[ 5] = data10;
  475. b[ 6] = ONE;
  476. b[ 7] = ZERO;
  477. #else
  478. data01 = *(ao1 + 0);
  479. data02 = *(ao1 + 1);
  480. data09 = *(ao2 + 0);
  481. data10 = *(ao2 + 1);
  482. data11 = *(ao2 + 2);
  483. data12 = *(ao2 + 3);
  484. b[ 0] = data01;
  485. b[ 1] = data02;
  486. b[ 2] = ZERO;
  487. b[ 3] = ZERO;
  488. b[ 4] = data09;
  489. b[ 5] = data10;
  490. b[ 6] = data11;
  491. b[ 7] = data12;
  492. #endif
  493. ao1 += 2 * lda;
  494. ao2 += 2 * lda;
  495. b += 8;
  496. }
  497. X += 2;
  498. i --;
  499. } while (i > 0);
  500. }
  501. i = (m & 1);
  502. if (i) {
  503. if (X < posY) {
  504. b += 4;
  505. } else
  506. if (X > posY) {
  507. data01 = *(ao1 + 0);
  508. data02 = *(ao1 + 1);
  509. data03 = *(ao1 + 2);
  510. data04 = *(ao1 + 3);
  511. b[ 0] = data01;
  512. b[ 1] = data02;
  513. b[ 2] = data03;
  514. b[ 3] = data04;
  515. b += 4;
  516. } else {
  517. #ifdef UNIT
  518. b[ 0] = ONE;
  519. b[ 1] = ZERO;
  520. b[ 2] = ZERO;
  521. b[ 3] = ZERO;
  522. #else
  523. data01 = *(ao1 + 0);
  524. data02 = *(ao1 + 1);
  525. b[ 0] = data01;
  526. b[ 1] = data02;
  527. b[ 2] = ZERO;
  528. b[ 3] = ZERO;
  529. #endif
  530. b += 4;
  531. }
  532. }
  533. posY += 2;
  534. }
  535. if (n & 1){
  536. X = posX;
  537. if (posX <= posY) {
  538. ao1 = a + posX * 2 + (posY + 0) * lda;
  539. } else {
  540. ao1 = a + posY * 2 + (posX + 0) * lda;
  541. }
  542. i = m;
  543. if (m > 0) {
  544. do {
  545. if (X < posY) {
  546. b += 2;
  547. ao1 += 2;
  548. } else
  549. if (X > posY) {
  550. data01 = *(ao1 + 0);
  551. data02 = *(ao1 + 1);
  552. b[ 0] = data01;
  553. b[ 1] = data02;
  554. ao1 += lda;
  555. b += 2;
  556. } else {
  557. #ifdef UNIT
  558. b[ 0] = ONE;
  559. b[ 1] = ZERO;
  560. #else
  561. data01 = *(ao1 + 0);
  562. data02 = *(ao1 + 1);
  563. b[ 0] = data01;
  564. b[ 1] = data02;
  565. #endif
  566. b += 2;
  567. ao1 += lda;
  568. }
  569. X += 1;
  570. i --;
  571. } while (i > 0);
  572. }
  573. }
  574. return 0;
  575. }