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.

zlaswp_k_2.c 14 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660
  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. #ifndef MINUS
  41. #define a2 (a1 + 2)
  42. #else
  43. #define a2 (a1 - 2)
  44. #endif
  45. int CNAME(BLASLONG n, BLASLONG k1, BLASLONG k2, FLOAT dummy1, FLOAT dummy4,
  46. FLOAT *a, BLASLONG lda,
  47. FLOAT *dummy2, BLASLONG dumy3, blasint *ipiv, BLASLONG incx){
  48. BLASLONG i, j, ip1, ip2, rows;
  49. blasint *piv;
  50. FLOAT *a1;
  51. FLOAT *b1, *b2;
  52. FLOAT A1, A2, B1, B2, A3, A4, B3, B4;
  53. FLOAT A5, A6, B5, B6, A7, A8, B7, B8;
  54. a -= 2;
  55. lda *= 2;
  56. k1 --;
  57. ipiv += k1;
  58. #ifdef MINUS
  59. ipiv -= (k2 - k1 - 1) * incx;
  60. #endif
  61. if (n <= 0) return 0;
  62. rows = k2-k1;
  63. if (rows <=0) return 0;
  64. if (rows == 1) {
  65. //Only have 1 row
  66. ip1 = *ipiv * 2;
  67. #ifndef MINUS
  68. a1 = a + (k1 + 1) * 2;
  69. #else
  70. a1 = a + k2 * 2;
  71. #endif
  72. b1 = a + ip1;
  73. if(a1 == b1) return 0;
  74. for(j=0; j<n; j++){
  75. A1 = *(a1 + 0);
  76. A2 = *(a1 + 1);
  77. B1 = *(b1 + 0);
  78. B2 = *(b1 + 1);
  79. *(a1 + 0) = B1;
  80. *(a1 + 1) = B2;
  81. *(b1 + 0) = A1;
  82. *(b1 + 1) = A2;
  83. a1 += lda;
  84. b1 += lda;
  85. }
  86. return 0;
  87. }
  88. j = (n >> 1);
  89. if (j > 0) {
  90. do {
  91. piv = ipiv;
  92. #ifndef MINUS
  93. a1 = a + (k1 + 1) * 2;
  94. #else
  95. a1 = a + k2 * 2;
  96. #endif
  97. ip1 = *piv * 2;
  98. piv += incx;
  99. ip2 = *piv * 2;
  100. piv += incx;
  101. b1 = a + ip1;
  102. b2 = a + ip2;
  103. i = (rows >> 1);
  104. i--;
  105. //Loop pipeline
  106. //Main Loop
  107. while (i > 0) {
  108. #ifdef CORE2
  109. #ifndef MINUS
  110. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(b1));
  111. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(b1 + lda));
  112. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(a1));
  113. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(a1 + lda));
  114. #else
  115. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(b1));
  116. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(b1 + lda));
  117. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(a1));
  118. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(a1 + lda));
  119. #endif
  120. #endif
  121. A1 = *(a1 + 0);
  122. A2 = *(a1 + 1);
  123. A3 = *(a2 + 0);
  124. A4 = *(a2 + 1);
  125. A5 = *(a1 + 0 + lda);
  126. A6 = *(a1 + 1 + lda);
  127. A7 = *(a2 + 0 + lda);
  128. A8 = *(a2 + 1 + lda);
  129. B1 = *(b1 + 0);
  130. B2 = *(b1 + 1);
  131. B3 = *(b2 + 0);
  132. B4 = *(b2 + 1);
  133. B5 = *(b1 + 0 + lda);
  134. B6 = *(b1 + 1 + lda);
  135. B7 = *(b2 + 0 + lda);
  136. B8 = *(b2 + 1 + lda);
  137. ip1 = *piv * 2;
  138. piv += incx;
  139. ip2 = *piv * 2;
  140. piv += incx;
  141. if (b1 == a1) {
  142. if (b2 == a1) {
  143. *(a1 + 0) = A3;
  144. *(a1 + 1) = A4;
  145. *(a2 + 0) = A1;
  146. *(a2 + 1) = A2;
  147. *(a1 + 0 + lda) = A7;
  148. *(a1 + 1 + lda) = A8;
  149. *(a2 + 0 + lda) = A5;
  150. *(a2 + 1 + lda) = A6;
  151. } else
  152. if (b2 != a2) {
  153. *(a2 + 0) = B3;
  154. *(a2 + 1) = B4;
  155. *(b2 + 0) = A3;
  156. *(b2 + 1) = A4;
  157. *(a2 + 0 + lda) = B7;
  158. *(a2 + 1 + lda) = B8;
  159. *(b2 + 0 + lda) = A7;
  160. *(b2 + 1 + lda) = A8;
  161. }
  162. } else
  163. if (b1 == a2) {
  164. if (b2 != a1) {
  165. if (b2 == a2) {
  166. *(a1 + 0) = A3;
  167. *(a1 + 1) = A4;
  168. *(a2 + 0) = A1;
  169. *(a2 + 1) = A2;
  170. *(a1 + 0 + lda) = A7;
  171. *(a1 + 1 + lda) = A8;
  172. *(a2 + 0 + lda) = A5;
  173. *(a2 + 1 + lda) = A6;
  174. } else {
  175. *(a1 + 0) = A3;
  176. *(a1 + 1) = A4;
  177. *(a2 + 0) = B3;
  178. *(a2 + 1) = B4;
  179. *(b2 + 0) = A1;
  180. *(b2 + 1) = A2;
  181. *(a1 + 0 + lda) = A7;
  182. *(a1 + 1 + lda) = A8;
  183. *(a2 + 0 + lda) = B7;
  184. *(a2 + 1 + lda) = B8;
  185. *(b2 + 0 + lda) = A5;
  186. *(b2 + 1 + lda) = A6;
  187. }
  188. }
  189. } else {
  190. if (b2 == a1) {
  191. *(a1 + 0) = A3;
  192. *(a1 + 1) = A4;
  193. *(a2 + 0) = B1;
  194. *(a2 + 1) = B2;
  195. *(b1 + 0) = A1;
  196. *(b1 + 1) = A2;
  197. *(a1 + 0 + lda) = A7;
  198. *(a1 + 1 + lda) = A8;
  199. *(a2 + 0 + lda) = B5;
  200. *(a2 + 1 + lda) = B6;
  201. *(b1 + 0 + lda) = A5;
  202. *(b1 + 1 + lda) = A6;
  203. } else
  204. if (b2 == a2) {
  205. *(a1 + 0) = B1;
  206. *(a1 + 1) = B2;
  207. *(b1 + 0) = A1;
  208. *(b1 + 1) = A2;
  209. *(a1 + 0 + lda) = B5;
  210. *(a1 + 1 + lda) = B6;
  211. *(b1 + 0 + lda) = A5;
  212. *(b1 + 1 + lda) = A6;
  213. } else
  214. if (b2 == b1) {
  215. *(a1 + 0) = B1;
  216. *(a1 + 1) = B2;
  217. *(a2 + 0) = A1;
  218. *(a2 + 1) = A2;
  219. *(b1 + 0) = A3;
  220. *(b1 + 1) = A4;
  221. *(a1 + 0 + lda) = B5;
  222. *(a1 + 1 + lda) = B6;
  223. *(a2 + 0 + lda) = A5;
  224. *(a2 + 1 + lda) = A6;
  225. *(b1 + 0 + lda) = A7;
  226. *(b1 + 1 + lda) = A8;
  227. } else {
  228. *(a1 + 0) = B1;
  229. *(a1 + 1) = B2;
  230. *(a2 + 0) = B3;
  231. *(a2 + 1) = B4;
  232. *(b1 + 0) = A1;
  233. *(b1 + 1) = A2;
  234. *(b2 + 0) = A3;
  235. *(b2 + 1) = A4;
  236. *(a1 + 0 + lda) = B5;
  237. *(a1 + 1 + lda) = B6;
  238. *(a2 + 0 + lda) = B7;
  239. *(a2 + 1 + lda) = B8;
  240. *(b1 + 0 + lda) = A5;
  241. *(b1 + 1 + lda) = A6;
  242. *(b2 + 0 + lda) = A7;
  243. *(b2 + 1 + lda) = A8;
  244. }
  245. }
  246. b1 = a + ip1;
  247. b2 = a + ip2;
  248. #ifndef MINUS
  249. a1 += 4;
  250. #else
  251. a1 -= 4;
  252. #endif
  253. i --;
  254. }
  255. //Loop Ending
  256. A1 = *(a1 + 0);
  257. A2 = *(a1 + 1);
  258. A3 = *(a2 + 0);
  259. A4 = *(a2 + 1);
  260. A5 = *(a1 + 0 + lda);
  261. A6 = *(a1 + 1 + lda);
  262. A7 = *(a2 + 0 + lda);
  263. A8 = *(a2 + 1 + lda);
  264. B1 = *(b1 + 0);
  265. B2 = *(b1 + 1);
  266. B3 = *(b2 + 0);
  267. B4 = *(b2 + 1);
  268. B5 = *(b1 + 0 + lda);
  269. B6 = *(b1 + 1 + lda);
  270. B7 = *(b2 + 0 + lda);
  271. B8 = *(b2 + 1 + lda);
  272. if (b1 == a1) {
  273. if (b2 == a1) {
  274. *(a1 + 0) = A3;
  275. *(a1 + 1) = A4;
  276. *(a2 + 0) = A1;
  277. *(a2 + 1) = A2;
  278. *(a1 + 0 + lda) = A7;
  279. *(a1 + 1 + lda) = A8;
  280. *(a2 + 0 + lda) = A5;
  281. *(a2 + 1 + lda) = A6;
  282. } else
  283. if (b2 != a2) {
  284. *(a2 + 0) = B3;
  285. *(a2 + 1) = B4;
  286. *(b2 + 0) = A3;
  287. *(b2 + 1) = A4;
  288. *(a2 + 0 + lda) = B7;
  289. *(a2 + 1 + lda) = B8;
  290. *(b2 + 0 + lda) = A7;
  291. *(b2 + 1 + lda) = A8;
  292. }
  293. } else
  294. if (b1 == a2) {
  295. if (b2 != a1) {
  296. if (b2 == a2) {
  297. *(a1 + 0) = A3;
  298. *(a1 + 1) = A4;
  299. *(a2 + 0) = A1;
  300. *(a2 + 1) = A2;
  301. *(a1 + 0 + lda) = A7;
  302. *(a1 + 1 + lda) = A8;
  303. *(a2 + 0 + lda) = A5;
  304. *(a2 + 1 + lda) = A6;
  305. } else {
  306. *(a1 + 0) = A3;
  307. *(a1 + 1) = A4;
  308. *(a2 + 0) = B3;
  309. *(a2 + 1) = B4;
  310. *(b2 + 0) = A1;
  311. *(b2 + 1) = A2;
  312. *(a1 + 0 + lda) = A7;
  313. *(a1 + 1 + lda) = A8;
  314. *(a2 + 0 + lda) = B7;
  315. *(a2 + 1 + lda) = B8;
  316. *(b2 + 0 + lda) = A5;
  317. *(b2 + 1 + lda) = A6;
  318. }
  319. }
  320. } else {
  321. if (b2 == a1) {
  322. *(a1 + 0) = A3;
  323. *(a1 + 1) = A4;
  324. *(a2 + 0) = B1;
  325. *(a2 + 1) = B2;
  326. *(b1 + 0) = A1;
  327. *(b1 + 1) = A2;
  328. *(a1 + 0 + lda) = A7;
  329. *(a1 + 1 + lda) = A8;
  330. *(a2 + 0 + lda) = B5;
  331. *(a2 + 1 + lda) = B6;
  332. *(b1 + 0 + lda) = A5;
  333. *(b1 + 1 + lda) = A6;
  334. } else
  335. if (b2 == a2) {
  336. *(a1 + 0) = B1;
  337. *(a1 + 1) = B2;
  338. *(b1 + 0) = A1;
  339. *(b1 + 1) = A2;
  340. *(a1 + 0 + lda) = B5;
  341. *(a1 + 1 + lda) = B6;
  342. *(b1 + 0 + lda) = A5;
  343. *(b1 + 1 + lda) = A6;
  344. } else
  345. if (b2 == b1) {
  346. *(a1 + 0) = B1;
  347. *(a1 + 1) = B2;
  348. *(a2 + 0) = A1;
  349. *(a2 + 1) = A2;
  350. *(b1 + 0) = A3;
  351. *(b1 + 1) = A4;
  352. *(a1 + 0 + lda) = B5;
  353. *(a1 + 1 + lda) = B6;
  354. *(a2 + 0 + lda) = A5;
  355. *(a2 + 1 + lda) = A6;
  356. *(b1 + 0 + lda) = A7;
  357. *(b1 + 1 + lda) = A8;
  358. } else {
  359. *(a1 + 0) = B1;
  360. *(a1 + 1) = B2;
  361. *(a2 + 0) = B3;
  362. *(a2 + 1) = B4;
  363. *(b1 + 0) = A1;
  364. *(b1 + 1) = A2;
  365. *(b2 + 0) = A3;
  366. *(b2 + 1) = A4;
  367. *(a1 + 0 + lda) = B5;
  368. *(a1 + 1 + lda) = B6;
  369. *(a2 + 0 + lda) = B7;
  370. *(a2 + 1 + lda) = B8;
  371. *(b1 + 0 + lda) = A5;
  372. *(b1 + 1 + lda) = A6;
  373. *(b2 + 0 + lda) = A7;
  374. *(b2 + 1 + lda) = A8;
  375. }
  376. }
  377. #ifndef MINUS
  378. a1 += 4;
  379. #else
  380. a1 -= 4;
  381. #endif
  382. //Remain
  383. i = (rows & 1);
  384. if (i > 0) {
  385. ip1 = *piv * 2;
  386. b1 = a + ip1;
  387. A1 = *(a1 + 0);
  388. A2 = *(a1 + 1);
  389. A3 = *(a1 + 0 + lda);
  390. A4 = *(a1 + 1 + lda);
  391. B1 = *(b1 + 0);
  392. B2 = *(b1 + 1);
  393. B3 = *(b1 + 0 + lda);
  394. B4 = *(b1 + 1 + lda);
  395. *(a1 + 0) = B1;
  396. *(a1 + 1) = B2;
  397. *(a1 + 0 + lda) = B3;
  398. *(a1 + 1 + lda) = B4;
  399. *(b1 + 0) = A1;
  400. *(b1 + 1) = A2;
  401. *(b1 + 0 + lda) = A3;
  402. *(b1 + 1 + lda) = A4;
  403. }
  404. a += 2 * lda;
  405. j --;
  406. } while (j > 0);
  407. }
  408. if (n & 1) {
  409. piv = ipiv;
  410. #ifndef MINUS
  411. a1 = a + (k1 + 1) * 2;
  412. #else
  413. a1 = a + k2 * 2;
  414. #endif
  415. ip1 = *piv * 2;
  416. piv += incx;
  417. ip2 = *piv * 2;
  418. piv += incx;
  419. b1 = a + ip1;
  420. b2 = a + ip2;
  421. i = (rows >> 1);
  422. i--;
  423. //Loop pipeline
  424. //Main Loop
  425. while (i > 0) {
  426. A1 = *(a1 + 0);
  427. A2 = *(a1 + 1);
  428. A3 = *(a2 + 0);
  429. A4 = *(a2 + 1);
  430. B1 = *(b1 + 0);
  431. B2 = *(b1 + 1);
  432. B3 = *(b2 + 0);
  433. B4 = *(b2 + 1);
  434. ip1 = *piv * 2;
  435. piv += incx;
  436. ip2 = *piv * 2;
  437. piv += incx;
  438. if (b1 == a1) {
  439. if (b2 == a1) {
  440. *(a1 + 0) = A3;
  441. *(a1 + 1) = A4;
  442. *(a2 + 0) = A1;
  443. *(a2 + 1) = A2;
  444. } else
  445. if (b2 != a2) {
  446. *(a2 + 0) = B3;
  447. *(a2 + 1) = B4;
  448. *(b2 + 0) = A3;
  449. *(b2 + 1) = A4;
  450. }
  451. } else
  452. if (b1 == a2) {
  453. if (b2 != a1) {
  454. if (b2 == a2) {
  455. *(a1 + 0) = A3;
  456. *(a1 + 1) = A4;
  457. *(a2 + 0) = A1;
  458. *(a2 + 1) = A2;
  459. } else {
  460. *(a1 + 0) = A3;
  461. *(a1 + 1) = A4;
  462. *(a2 + 0) = B3;
  463. *(a2 + 1) = B4;
  464. *(b2 + 0) = A1;
  465. *(b2 + 1) = A2;
  466. }
  467. }
  468. } else {
  469. if (b2 == a1) {
  470. *(a1 + 0) = A3;
  471. *(a1 + 1) = A4;
  472. *(a2 + 0) = B1;
  473. *(a2 + 1) = B2;
  474. *(b1 + 0) = A1;
  475. *(b1 + 1) = A2;
  476. } else
  477. if (b2 == a2) {
  478. *(a1 + 0) = B1;
  479. *(a1 + 1) = B2;
  480. *(b1 + 0) = A1;
  481. *(b1 + 1) = A2;
  482. } else
  483. if (b2 == b1) {
  484. *(a1 + 0) = B1;
  485. *(a1 + 1) = B2;
  486. *(a2 + 0) = A1;
  487. *(a2 + 1) = A2;
  488. *(b1 + 0) = A3;
  489. *(b1 + 1) = A4;
  490. } else {
  491. *(a1 + 0) = B1;
  492. *(a1 + 1) = B2;
  493. *(a2 + 0) = B3;
  494. *(a2 + 1) = B4;
  495. *(b1 + 0) = A1;
  496. *(b1 + 1) = A2;
  497. *(b2 + 0) = A3;
  498. *(b2 + 1) = A4;
  499. }
  500. }
  501. b1 = a + ip1;
  502. b2 = a + ip2;
  503. #ifndef MINUS
  504. a1 += 4;
  505. #else
  506. a1 -= 4;
  507. #endif
  508. i --;
  509. }
  510. //Loop Ending
  511. A1 = *(a1 + 0);
  512. A2 = *(a1 + 1);
  513. A3 = *(a2 + 0);
  514. A4 = *(a2 + 1);
  515. B1 = *(b1 + 0);
  516. B2 = *(b1 + 1);
  517. B3 = *(b2 + 0);
  518. B4 = *(b2 + 1);
  519. if (b1 == a1) {
  520. if (b2 == a1) {
  521. *(a1 + 0) = A3;
  522. *(a1 + 1) = A4;
  523. *(a2 + 0) = A1;
  524. *(a2 + 1) = A2;
  525. } else
  526. if (b2 != a2) {
  527. *(a2 + 0) = B3;
  528. *(a2 + 1) = B4;
  529. *(b2 + 0) = A3;
  530. *(b2 + 1) = A4;
  531. }
  532. } else
  533. if (b1 == a2) {
  534. if (b2 != a1) {
  535. if (b2 == a2) {
  536. *(a1 + 0) = A3;
  537. *(a1 + 1) = A4;
  538. *(a2 + 0) = A1;
  539. *(a2 + 1) = A2;
  540. } else {
  541. *(a1 + 0) = A3;
  542. *(a1 + 1) = A4;
  543. *(a2 + 0) = B3;
  544. *(a2 + 1) = B4;
  545. *(b2 + 0) = A1;
  546. *(b2 + 1) = A2;
  547. }
  548. }
  549. } else {
  550. if (b2 == a1) {
  551. *(a1 + 0) = A3;
  552. *(a1 + 1) = A4;
  553. *(a2 + 0) = B1;
  554. *(a2 + 1) = B2;
  555. *(b1 + 0) = A1;
  556. *(b1 + 1) = A2;
  557. } else
  558. if (b2 == a2) {
  559. *(a1 + 0) = B1;
  560. *(a1 + 1) = B2;
  561. *(b1 + 0) = A1;
  562. *(b1 + 1) = A2;
  563. } else
  564. if (b2 == b1) {
  565. *(a1 + 0) = B1;
  566. *(a1 + 1) = B2;
  567. *(a2 + 0) = A1;
  568. *(a2 + 1) = A2;
  569. *(b1 + 0) = A3;
  570. *(b1 + 1) = A4;
  571. } else {
  572. *(a1 + 0) = B1;
  573. *(a1 + 1) = B2;
  574. *(a2 + 0) = B3;
  575. *(a2 + 1) = B4;
  576. *(b1 + 0) = A1;
  577. *(b1 + 1) = A2;
  578. *(b2 + 0) = A3;
  579. *(b2 + 1) = A4;
  580. }
  581. }
  582. #ifndef MINUS
  583. a1 += 4;
  584. #else
  585. a1 -= 4;
  586. #endif
  587. //Remain
  588. i = (rows & 1);
  589. if (i > 0) {
  590. ip1 = *piv * 2;
  591. b1 = a + ip1;
  592. A1 = *(a1 + 0);
  593. A2 = *(a1 + 1);
  594. B1 = *(b1 + 0);
  595. B2 = *(b1 + 1);
  596. *(a1 + 0) = B1;
  597. *(a1 + 1) = B2;
  598. *(b1 + 0) = A1;
  599. *(b1 + 1) = A2;
  600. }
  601. }
  602. return 0;
  603. }