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 16 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785
  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, mm;
  43. FLOAT data01, data02, data03, data04, data05, data06;
  44. FLOAT data07, data08, data09, data10, data11, data12;
  45. FLOAT data13, data14, data15, data16, data17, data18;
  46. FLOAT data19, data20, data21, data22, data23, data24;
  47. FLOAT data25, data26, data27, data28, data29, data30;
  48. FLOAT data31, data32, data33, data34, data35, data36;
  49. FLOAT *ao1, *ao2, *ao3, *ao4, *ao5, *ao6;
  50. //js = (n >> 2);
  51. js = n/6;
  52. if (js > 0){
  53. do {
  54. X = posX;
  55. if (posX <= posY) {
  56. ao1 = a + posX + (posY + 0) * lda;
  57. ao2 = a + posX + (posY + 1) * lda;
  58. ao3 = a + posX + (posY + 2) * lda;
  59. ao4 = a + posX + (posY + 3) * lda;
  60. ao5 = a + posX + (posY + 4) * lda;
  61. ao6 = a + posX + (posY + 5) * lda;
  62. } else {
  63. ao1 = a + posY + (posX + 0) * lda;
  64. ao2 = a + posY + (posX + 1) * lda;
  65. ao3 = a + posY + (posX + 2) * lda;
  66. ao4 = a + posY + (posX + 3) * lda;
  67. ao5 = a + posY + (posX + 4) * lda;
  68. ao6 = a + posY + (posX + 5) * lda;
  69. }
  70. i = m/6;
  71. if (i > 0) {
  72. do {
  73. if (X < posY) {
  74. data01 = *(ao1 + 0);
  75. data02 = *(ao1 + 1);
  76. data03 = *(ao1 + 2);
  77. data04 = *(ao1 + 3);
  78. data05 = *(ao1 + 4);
  79. data06 = *(ao1 + 5);
  80. data07 = *(ao2 + 0);
  81. data08 = *(ao2 + 1);
  82. data09 = *(ao2 + 2);
  83. data10 = *(ao2 + 3);
  84. data11 = *(ao2 + 4);
  85. data12 = *(ao2 + 5);
  86. data13 = *(ao3 + 0);
  87. data14 = *(ao3 + 1);
  88. data15 = *(ao3 + 2);
  89. data16 = *(ao3 + 3);
  90. data17 = *(ao3 + 4);
  91. data18 = *(ao3 + 5);
  92. data19 = *(ao4 + 0);
  93. data20 = *(ao4 + 1);
  94. data21 = *(ao4 + 2);
  95. data22 = *(ao4 + 3);
  96. data23 = *(ao4 + 4);
  97. data24 = *(ao4 + 5);
  98. data25 = *(ao5 + 0);
  99. data26 = *(ao5 + 1);
  100. data27 = *(ao5 + 2);
  101. data28 = *(ao5 + 3);
  102. data29 = *(ao5 + 4);
  103. data30 = *(ao5 + 5);
  104. data31 = *(ao6 + 0);
  105. data32 = *(ao6 + 1);
  106. data33 = *(ao6 + 2);
  107. data34 = *(ao6 + 3);
  108. data35 = *(ao6 + 4);
  109. data36 = *(ao6 + 5);
  110. b[ 0] = data01;
  111. b[ 1] = data07;
  112. b[ 2] = data13;
  113. b[ 3] = data19;
  114. b[ 4] = data25;
  115. b[ 5] = data31;
  116. b[ 6] = data02;
  117. b[ 7] = data08;
  118. b[ 8] = data14;
  119. b[ 9] = data20;
  120. b[10] = data26;
  121. b[11] = data32;
  122. b[12] = data03;
  123. b[13] = data09;
  124. b[14] = data15;
  125. b[15] = data21;
  126. b[16] = data27;
  127. b[17] = data33;
  128. b[18] = data04;
  129. b[19] = data10;
  130. b[20] = data16;
  131. b[21] = data22;
  132. b[22] = data28;
  133. b[23] = data34;
  134. b[24] = data05;
  135. b[25] = data11;
  136. b[26] = data17;
  137. b[27] = data23;
  138. b[28] = data29;
  139. b[29] = data35;
  140. b[30] = data06;
  141. b[31] = data12;
  142. b[32] = data18;
  143. b[33] = data24;
  144. b[34] = data30;
  145. b[35] = data36;
  146. ao1 += 6;
  147. ao2 += 6;
  148. ao3 += 6;
  149. ao4 += 6;
  150. ao5 += 6;
  151. ao6 += 6;
  152. b += 36;
  153. } else
  154. if (X > posY) {
  155. b[ 0] = ZERO;
  156. b[ 1] = ZERO;
  157. b[ 2] = ZERO;
  158. b[ 3] = ZERO;
  159. b[ 4] = ZERO;
  160. b[ 5] = ZERO;
  161. b[ 6] = ZERO;
  162. b[ 7] = ZERO;
  163. b[ 8] = ZERO;
  164. b[ 9] = ZERO;
  165. b[10] = ZERO;
  166. b[11] = ZERO;
  167. b[12] = ZERO;
  168. b[13] = ZERO;
  169. b[14] = ZERO;
  170. b[15] = ZERO;
  171. b[16] = ZERO;
  172. b[17] = ZERO;
  173. b[18] = ZERO;
  174. b[19] = ZERO;
  175. b[20] = ZERO;
  176. b[21] = ZERO;
  177. b[22] = ZERO;
  178. b[23] = ZERO;
  179. b[24] = ZERO;
  180. b[25] = ZERO;
  181. b[26] = ZERO;
  182. b[27] = ZERO;
  183. b[28] = ZERO;
  184. b[29] = ZERO;
  185. b[30] = ZERO;
  186. b[31] = ZERO;
  187. b[32] = ZERO;
  188. b[33] = ZERO;
  189. b[34] = ZERO;
  190. b[35] = ZERO;
  191. ao1 += 6 * lda;
  192. ao2 += 6 * lda;
  193. ao3 += 6 * lda;
  194. ao4 += 6 * lda;
  195. ao5 += 6 * lda;
  196. ao6 += 6 * lda;
  197. b += 36;
  198. } else {
  199. data01 = *(ao1 + 0);
  200. data07 = *(ao2 + 0);
  201. data13 = *(ao3 + 0);
  202. data19 = *(ao4 + 0);
  203. data25 = *(ao5 + 0);
  204. data31 = *(ao6 + 0);
  205. data08 = *(ao2 + 1);
  206. data14 = *(ao3 + 1);
  207. data20 = *(ao4 + 1);
  208. data26 = *(ao5 + 1);
  209. data32 = *(ao6 + 1);
  210. data15 = *(ao3 + 2);
  211. data21 = *(ao4 + 2);
  212. data27 = *(ao5 + 2);
  213. data33 = *(ao6 + 2);
  214. data22 = *(ao4 + 3);
  215. data28 = *(ao5 + 3);
  216. data34 = *(ao6 + 3);
  217. data29 = *(ao5 + 4);
  218. data35 = *(ao6 + 4);
  219. data36 = *(ao6 + 5);
  220. #ifdef UNIT
  221. b[ 0] = ONE;
  222. b[ 1] = data07;
  223. b[ 2] = data13;
  224. b[ 3] = data19;
  225. b[ 4] = data25;
  226. b[ 5] = data31;
  227. b[ 6] = ZERO;
  228. b[ 7] = ONE;
  229. b[ 8] = data14;
  230. b[ 9] = data20;
  231. b[10] = data26;
  232. b[11] = data32;
  233. b[12] = ZERO;
  234. b[13] = ZERO;
  235. b[14] = ONE;
  236. b[15] = data21;
  237. b[16] = data27;
  238. b[17] = data33;
  239. b[18] = ZERO;
  240. b[19] = ZERO;
  241. b[20] = ZERO;
  242. b[21] = ONE;
  243. b[22] = data28;
  244. b[23] = data34;
  245. b[24] = ZERO;
  246. b[25] = ZERO;
  247. b[26] = ZERO;
  248. b[27] = ZERO;
  249. b[28] = ONE;
  250. b[29] = data35;
  251. b[30] = ZERO;
  252. b[31] = ZERO;
  253. b[32] = ZERO;
  254. b[33] = ZERO;
  255. b[34] = ZERO;
  256. b[35] = ONE;
  257. #else
  258. b[ 0] = data01;
  259. b[ 1] = data07;
  260. b[ 2] = data13;
  261. b[ 3] = data19;
  262. b[ 4] = data25;
  263. b[ 5] = data31;
  264. b[ 6] = ZERO;
  265. b[ 7] = data08;
  266. b[ 8] = data14;
  267. b[ 9] = data20;
  268. b[10] = data26;
  269. b[11] = data32;
  270. b[12] = ZERO;
  271. b[13] = ZERO;
  272. b[14] = data15;
  273. b[15] = data21;
  274. b[16] = data27;
  275. b[17] = data33;
  276. b[18] = ZERO;
  277. b[19] = ZERO;
  278. b[20] = ZERO;
  279. b[21] = data22;
  280. b[22] = data28;
  281. b[23] = data34;
  282. b[24] = ZERO;
  283. b[25] = ZERO;
  284. b[26] = ZERO;
  285. b[27] = ZERO;
  286. b[28] = data29;
  287. b[29] = data35;
  288. b[30] = ZERO;
  289. b[31] = ZERO;
  290. b[32] = ZERO;
  291. b[33] = ZERO;
  292. b[34] = ZERO;
  293. b[35] = data36;
  294. #endif
  295. ao1 += 6;
  296. ao2 += 6;
  297. ao3 += 6;
  298. ao4 += 6;
  299. ao5 += 6;
  300. ao6 += 7;
  301. b += 36;
  302. }
  303. X += 6;
  304. i --;
  305. } while (i > 0);
  306. }
  307. mm = m - m/6;
  308. if (mm & 4) {
  309. if (X < posY) {
  310. data01 = *(ao1 + 0);
  311. data02 = *(ao1 + 1);
  312. data03 = *(ao1 + 2);
  313. data04 = *(ao1 + 3);
  314. data05 = *(ao2 + 0);
  315. data06 = *(ao2 + 1);
  316. data07 = *(ao2 + 2);
  317. data08 = *(ao2 + 3);
  318. data09 = *(ao3 + 0);
  319. data10 = *(ao3 + 1);
  320. data11 = *(ao3 + 2);
  321. data12 = *(ao3 + 3);
  322. data13 = *(ao4 + 0);
  323. data14 = *(ao4 + 1);
  324. data15 = *(ao4 + 2);
  325. data16 = *(ao4 + 3);
  326. b[ 0] = data01;
  327. b[ 1] = data05;
  328. b[ 2] = data09;
  329. b[ 3] = data13;
  330. b[ 4] = data02;
  331. b[ 5] = data06;
  332. b[ 6] = data10;
  333. b[ 7] = data14;
  334. b[ 8] = data03;
  335. b[ 9] = data07;
  336. b[10] = data11;
  337. b[11] = data15;
  338. b[12] = data04;
  339. b[13] = data08;
  340. b[14] = data12;
  341. b[15] = data16;
  342. ao1 += 4;
  343. ao2 += 4;
  344. ao3 += 4;
  345. ao4 += 4;
  346. b += 16;
  347. } else
  348. if (X > posY) {
  349. b[ 0] = ZERO;
  350. b[ 1] = ZERO;
  351. b[ 2] = ZERO;
  352. b[ 3] = ZERO;
  353. b[ 4] = ZERO;
  354. b[ 5] = ZERO;
  355. b[ 6] = ZERO;
  356. b[ 7] = ZERO;
  357. b[ 8] = ZERO;
  358. b[ 9] = ZERO;
  359. b[10] = ZERO;
  360. b[11] = ZERO;
  361. b[12] = ZERO;
  362. b[13] = ZERO;
  363. b[14] = ZERO;
  364. b[15] = ZERO;
  365. b[16] = ZERO;
  366. b[17] = ZERO;
  367. b[18] = ZERO;
  368. b[19] = ZERO;
  369. b[20] = ZERO;
  370. b[21] = ZERO;
  371. b[22] = ZERO;
  372. b[23] = ZERO;
  373. ao1 += 4 * lda;
  374. ao2 += 4 * lda;
  375. ao3 += 4 * lda;
  376. ao4 += 4 * lda;
  377. b += 16;
  378. } else {
  379. #ifdef UNIT
  380. data05 = *(ao2 + 0);
  381. data09 = *(ao3 + 0);
  382. data10 = *(ao3 + 1);
  383. data13 = *(ao4 + 0);
  384. data14 = *(ao4 + 1);
  385. data15 = *(ao4 + 2);
  386. b[ 0] = ONE;
  387. b[ 1] = data05;
  388. b[ 2] = data09;
  389. b[ 3] = data13;
  390. b[ 4] = ZERO;
  391. b[ 5] = ONE;
  392. b[ 6] = data10;
  393. b[ 7] = data14;
  394. b[ 8] = ZERO;
  395. b[ 9] = ZERO;
  396. b[10] = ONE;
  397. b[11] = data15;
  398. b[12] = ZERO;
  399. b[13] = ZERO;
  400. b[14] = ZERO;
  401. b[15] = ONE;
  402. #else
  403. data01 = *(ao1 + 0);
  404. data05 = *(ao2 + 0);
  405. data06 = *(ao2 + 1);
  406. data09 = *(ao3 + 0);
  407. data10 = *(ao3 + 1);
  408. data11 = *(ao3 + 2);
  409. data13 = *(ao4 + 0);
  410. data14 = *(ao4 + 1);
  411. data15 = *(ao4 + 2);
  412. data16 = *(ao4 + 3);
  413. b[ 0] = data01;
  414. b[ 1] = data05;
  415. b[ 2] = data09;
  416. b[ 3] = data13;
  417. b[ 4] = ZERO;
  418. b[ 5] = data06;
  419. b[ 6] = data10;
  420. b[ 7] = data14;
  421. b[ 8] = ZERO;
  422. b[ 9] = ZERO;
  423. b[10] = data11;
  424. b[11] = data15;
  425. b[12] = ZERO;
  426. b[13] = ZERO;
  427. b[14] = ZERO;
  428. b[15] = data16;
  429. #endif
  430. ao1 += 4;
  431. ao2 += 4;
  432. ao3 += 4;
  433. ao4 += 4;
  434. b += 16;
  435. }
  436. X += 4;
  437. }
  438. if (mm & 3) {
  439. if (X < posY) {
  440. if (mm & 2) {
  441. data01 = *(ao1 + 0);
  442. data02 = *(ao1 + 1);
  443. data03 = *(ao2 + 0);
  444. data04 = *(ao2 + 1);
  445. data05 = *(ao3 + 0);
  446. data06 = *(ao3 + 1);
  447. data07 = *(ao4 + 0);
  448. data08 = *(ao4 + 1);
  449. b[ 0] = data01;
  450. b[ 1] = data03;
  451. b[ 2] = data05;
  452. b[ 3] = data07;
  453. b[ 4] = data02;
  454. b[ 5] = data04;
  455. b[ 6] = data06;
  456. b[ 7] = data08;
  457. ao1 += 2;
  458. ao2 += 2;
  459. ao3 += 2;
  460. ao4 += 2;
  461. b += 8;
  462. }
  463. if (mm & 1) {
  464. data01 = *(ao1 + 0);
  465. data03 = *(ao2 + 0);
  466. data05 = *(ao3 + 0);
  467. data07 = *(ao4 + 0);
  468. b[ 0] = data01;
  469. b[ 1] = data03;
  470. b[ 2] = data05;
  471. b[ 3] = data07;
  472. ao1 += 1;
  473. ao2 += 1;
  474. ao3 += 1;
  475. ao4 += 1;
  476. b += 4;
  477. }
  478. } else
  479. if (X > posY) {
  480. if (m & 2) {
  481. ao1 += 2 * lda;
  482. ao2 += 2 * lda;
  483. b += 8;
  484. }
  485. if (m & 1) {
  486. ao1 += lda;
  487. b += 4;
  488. }
  489. } else {
  490. #ifdef UNIT
  491. data05 = *(ao2 + 0);
  492. data09 = *(ao3 + 0);
  493. data13 = *(ao4 + 0);
  494. if (i >= 2) {
  495. data10 = *(ao3 + 1);
  496. data14 = *(ao4 + 1);
  497. }
  498. if (i >= 3) {
  499. data15 = *(ao4 + 2);
  500. }
  501. b[ 0] = ONE;
  502. b[ 1] = data05;
  503. b[ 2] = data09;
  504. b[ 3] = data13;
  505. b += 4;
  506. if(i >= 2) {
  507. b[ 0] = ZERO;
  508. b[ 1] = ONE;
  509. b[ 2] = data10;
  510. b[ 3] = data14;
  511. b += 4;
  512. }
  513. if (i >= 3) {
  514. b[ 0] = ZERO;
  515. b[ 1] = ZERO;
  516. b[ 2] = ONE;
  517. b[ 3] = data15;
  518. b += 4;
  519. }
  520. #else
  521. data01 = *(ao1 + 0);
  522. data05 = *(ao2 + 0);
  523. data09 = *(ao3 + 0);
  524. data13 = *(ao4 + 0);
  525. if (i >= 2) {
  526. data06 = *(ao2 + 1);
  527. data10 = *(ao3 + 1);
  528. data14 = *(ao4 + 1);
  529. }
  530. if (i >= 3) {
  531. data11 = *(ao3 + 2);
  532. data15 = *(ao4 + 2);
  533. }
  534. b[ 0] = data01;
  535. b[ 1] = data05;
  536. b[ 2] = data09;
  537. b[ 3] = data13;
  538. b += 4;
  539. if(i >= 2) {
  540. b[ 0] = ZERO;
  541. b[ 1] = data06;
  542. b[ 2] = data10;
  543. b[ 3] = data14;
  544. b += 4;
  545. }
  546. if (i >= 3) {
  547. b[ 0] = ZERO;
  548. b[ 1] = ZERO;
  549. b[ 2] = data11;
  550. b[ 3] = data15;
  551. b += 4;
  552. }
  553. #endif
  554. }
  555. }
  556. posY += 4;
  557. js --;
  558. } while (js > 0);
  559. } /* End of main loop */
  560. if (n & 2){
  561. X = posX;
  562. if (posX <= posY) {
  563. ao1 = a + posX + (posY + 0) * lda;
  564. ao2 = a + posX + (posY + 1) * lda;
  565. } else {
  566. ao1 = a + posY + (posX + 0) * lda;
  567. ao2 = a + posY + (posX + 1) * lda;
  568. }
  569. i = (m >> 1);
  570. if (i > 0) {
  571. do {
  572. if (X < posY) {
  573. data01 = *(ao1 + 0);
  574. data02 = *(ao1 + 1);
  575. data05 = *(ao2 + 0);
  576. data06 = *(ao2 + 1);
  577. b[ 0] = data01;
  578. b[ 1] = data05;
  579. b[ 2] = data02;
  580. b[ 3] = data06;
  581. ao1 += 2;
  582. ao2 += 2;
  583. b += 4;
  584. } else
  585. if (X > posY) {
  586. ao1 += 2 * lda;
  587. ao2 += 2 * lda;
  588. b += 4;
  589. } else {
  590. #ifdef UNIT
  591. data05 = *(ao2 + 0);
  592. b[ 0] = ONE;
  593. b[ 1] = data05;
  594. b[ 2] = ZERO;
  595. b[ 3] = ONE;
  596. #else
  597. data01 = *(ao1 + 0);
  598. data05 = *(ao2 + 0);
  599. data06 = *(ao2 + 1);
  600. b[ 0] = data01;
  601. b[ 1] = data05;
  602. b[ 2] = ZERO;
  603. b[ 3] = data06;
  604. #endif
  605. ao1 += 2 * lda;
  606. ao2 += 2 * lda;
  607. b += 4;
  608. }
  609. X += 2;
  610. i --;
  611. } while (i > 0);
  612. }
  613. i = (m & 1);
  614. if (i) {
  615. if (X < posY) {
  616. data01 = *(ao1 + 0);
  617. data05 = *(ao2 + 0);
  618. b[ 0] = data01;
  619. b[ 1] = data05;
  620. ao1 += 1;
  621. ao2 += 1;
  622. b += 2;
  623. } else
  624. if (X > posY) {
  625. ao1 += lda;
  626. ao2 += lda;
  627. b += 2;
  628. } else {
  629. #ifdef UNIT
  630. data05 = *(ao2 + 0);
  631. b[ 0] = ONE;
  632. b[ 1] = data05;
  633. #else
  634. data01 = *(ao1 + 0);
  635. data05 = *(ao2 + 0);
  636. b[ 0] = data01;
  637. b[ 1] = data05;
  638. #endif
  639. ao1 += lda;
  640. ao2 += lda;
  641. b += 2;
  642. }
  643. }
  644. posY += 2;
  645. }
  646. if (n & 1){
  647. X = posX;
  648. if (posX <= posY) {
  649. ao1 = a + posX + (posY + 0) * lda;
  650. } else {
  651. ao1 = a + posY + (posX + 0) * lda;
  652. }
  653. i = m;
  654. if (m > 0) {
  655. do {
  656. if (X < posY) {
  657. data01 = *(ao1 + 0);
  658. b[ 0] = data01;
  659. ao1 += 1;
  660. b += 1;
  661. } else
  662. if (X > posY) {
  663. ao1 += lda;
  664. b += 1;
  665. } else {
  666. #ifdef UNIT
  667. b[ 0] = ONE;
  668. #else
  669. data01 = *(ao1 + 0);
  670. b[ 0] = data01;
  671. #endif
  672. ao1 += lda;
  673. b += 1;
  674. }
  675. X += 1;
  676. i --;
  677. } while (i > 0);
  678. }
  679. }
  680. return 0;
  681. }