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

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661
  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. #ifndef MINUS
  58. ipiv += k1;
  59. #else
  60. ipiv -= (k2 - 1) * incx;
  61. #endif
  62. if (n <= 0) return 0;
  63. rows = k2-k1;
  64. if (rows <=0) return 0;
  65. if (rows == 1) {
  66. //Only have 1 row
  67. ip1 = *ipiv * 2;
  68. #ifndef MINUS
  69. a1 = a + (k1 + 1) * 2;
  70. #else
  71. a1 = a + k2 * 2;
  72. #endif
  73. b1 = a + ip1;
  74. if(a1 == b1) return 0;
  75. for(j=0; j<n; j++){
  76. A1 = *(a1 + 0);
  77. A2 = *(a1 + 1);
  78. B1 = *(b1 + 0);
  79. B2 = *(b1 + 1);
  80. *(a1 + 0) = B1;
  81. *(a1 + 1) = B2;
  82. *(b1 + 0) = A1;
  83. *(b1 + 1) = A2;
  84. a1 += lda;
  85. b1 += lda;
  86. }
  87. return 0;
  88. }
  89. j = (n >> 1);
  90. if (j > 0) {
  91. do {
  92. piv = ipiv;
  93. #ifndef MINUS
  94. a1 = a + (k1 + 1) * 2;
  95. #else
  96. a1 = a + k2 * 2;
  97. #endif
  98. ip1 = *piv * 2;
  99. piv += incx;
  100. ip2 = *piv * 2;
  101. piv += incx;
  102. b1 = a + ip1;
  103. b2 = a + ip2;
  104. i = (rows >> 1);
  105. i--;
  106. //Loop pipeline
  107. //Main Loop
  108. while (i > 0) {
  109. #ifdef CORE2
  110. #ifndef MINUS
  111. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(b1));
  112. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(b1 + lda));
  113. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(a1));
  114. asm volatile("prefetcht0 1 * 64(%0)\n" : : "r"(a1 + lda));
  115. #else
  116. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(b1));
  117. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(b1 + lda));
  118. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(a1));
  119. asm volatile("prefetcht0 -1 * 64(%0)\n" : : "r"(a1 + lda));
  120. #endif
  121. #endif
  122. A1 = *(a1 + 0);
  123. A2 = *(a1 + 1);
  124. A3 = *(a2 + 0);
  125. A4 = *(a2 + 1);
  126. A5 = *(a1 + 0 + lda);
  127. A6 = *(a1 + 1 + lda);
  128. A7 = *(a2 + 0 + lda);
  129. A8 = *(a2 + 1 + lda);
  130. B1 = *(b1 + 0);
  131. B2 = *(b1 + 1);
  132. B3 = *(b2 + 0);
  133. B4 = *(b2 + 1);
  134. B5 = *(b1 + 0 + lda);
  135. B6 = *(b1 + 1 + lda);
  136. B7 = *(b2 + 0 + lda);
  137. B8 = *(b2 + 1 + lda);
  138. ip1 = *piv * 2;
  139. piv += incx;
  140. ip2 = *piv * 2;
  141. piv += incx;
  142. if (b1 == a1) {
  143. if (b2 == a1) {
  144. *(a1 + 0) = A3;
  145. *(a1 + 1) = A4;
  146. *(a2 + 0) = A1;
  147. *(a2 + 1) = A2;
  148. *(a1 + 0 + lda) = A7;
  149. *(a1 + 1 + lda) = A8;
  150. *(a2 + 0 + lda) = A5;
  151. *(a2 + 1 + lda) = A6;
  152. } else
  153. if (b2 != a2) {
  154. *(a2 + 0) = B3;
  155. *(a2 + 1) = B4;
  156. *(b2 + 0) = A3;
  157. *(b2 + 1) = A4;
  158. *(a2 + 0 + lda) = B7;
  159. *(a2 + 1 + lda) = B8;
  160. *(b2 + 0 + lda) = A7;
  161. *(b2 + 1 + lda) = A8;
  162. }
  163. } else
  164. if (b1 == a2) {
  165. if (b2 != a1) {
  166. if (b2 == a2) {
  167. *(a1 + 0) = A3;
  168. *(a1 + 1) = A4;
  169. *(a2 + 0) = A1;
  170. *(a2 + 1) = A2;
  171. *(a1 + 0 + lda) = A7;
  172. *(a1 + 1 + lda) = A8;
  173. *(a2 + 0 + lda) = A5;
  174. *(a2 + 1 + lda) = A6;
  175. } else {
  176. *(a1 + 0) = A3;
  177. *(a1 + 1) = A4;
  178. *(a2 + 0) = B3;
  179. *(a2 + 1) = B4;
  180. *(b2 + 0) = A1;
  181. *(b2 + 1) = A2;
  182. *(a1 + 0 + lda) = A7;
  183. *(a1 + 1 + lda) = A8;
  184. *(a2 + 0 + lda) = B7;
  185. *(a2 + 1 + lda) = B8;
  186. *(b2 + 0 + lda) = A5;
  187. *(b2 + 1 + lda) = A6;
  188. }
  189. }
  190. } else {
  191. if (b2 == a1) {
  192. *(a1 + 0) = A3;
  193. *(a1 + 1) = A4;
  194. *(a2 + 0) = B1;
  195. *(a2 + 1) = B2;
  196. *(b1 + 0) = A1;
  197. *(b1 + 1) = A2;
  198. *(a1 + 0 + lda) = A7;
  199. *(a1 + 1 + lda) = A8;
  200. *(a2 + 0 + lda) = B5;
  201. *(a2 + 1 + lda) = B6;
  202. *(b1 + 0 + lda) = A5;
  203. *(b1 + 1 + lda) = A6;
  204. } else
  205. if (b2 == a2) {
  206. *(a1 + 0) = B1;
  207. *(a1 + 1) = B2;
  208. *(b1 + 0) = A1;
  209. *(b1 + 1) = A2;
  210. *(a1 + 0 + lda) = B5;
  211. *(a1 + 1 + lda) = B6;
  212. *(b1 + 0 + lda) = A5;
  213. *(b1 + 1 + lda) = A6;
  214. } else
  215. if (b2 == b1) {
  216. *(a1 + 0) = B1;
  217. *(a1 + 1) = B2;
  218. *(a2 + 0) = A1;
  219. *(a2 + 1) = A2;
  220. *(b1 + 0) = A3;
  221. *(b1 + 1) = A4;
  222. *(a1 + 0 + lda) = B5;
  223. *(a1 + 1 + lda) = B6;
  224. *(a2 + 0 + lda) = A5;
  225. *(a2 + 1 + lda) = A6;
  226. *(b1 + 0 + lda) = A7;
  227. *(b1 + 1 + lda) = A8;
  228. } else {
  229. *(a1 + 0) = B1;
  230. *(a1 + 1) = B2;
  231. *(a2 + 0) = B3;
  232. *(a2 + 1) = B4;
  233. *(b1 + 0) = A1;
  234. *(b1 + 1) = A2;
  235. *(b2 + 0) = A3;
  236. *(b2 + 1) = A4;
  237. *(a1 + 0 + lda) = B5;
  238. *(a1 + 1 + lda) = B6;
  239. *(a2 + 0 + lda) = B7;
  240. *(a2 + 1 + lda) = B8;
  241. *(b1 + 0 + lda) = A5;
  242. *(b1 + 1 + lda) = A6;
  243. *(b2 + 0 + lda) = A7;
  244. *(b2 + 1 + lda) = A8;
  245. }
  246. }
  247. b1 = a + ip1;
  248. b2 = a + ip2;
  249. #ifndef MINUS
  250. a1 += 4;
  251. #else
  252. a1 -= 4;
  253. #endif
  254. i --;
  255. }
  256. //Loop Ending
  257. A1 = *(a1 + 0);
  258. A2 = *(a1 + 1);
  259. A3 = *(a2 + 0);
  260. A4 = *(a2 + 1);
  261. A5 = *(a1 + 0 + lda);
  262. A6 = *(a1 + 1 + lda);
  263. A7 = *(a2 + 0 + lda);
  264. A8 = *(a2 + 1 + lda);
  265. B1 = *(b1 + 0);
  266. B2 = *(b1 + 1);
  267. B3 = *(b2 + 0);
  268. B4 = *(b2 + 1);
  269. B5 = *(b1 + 0 + lda);
  270. B6 = *(b1 + 1 + lda);
  271. B7 = *(b2 + 0 + lda);
  272. B8 = *(b2 + 1 + lda);
  273. if (b1 == a1) {
  274. if (b2 == a1) {
  275. *(a1 + 0) = A3;
  276. *(a1 + 1) = A4;
  277. *(a2 + 0) = A1;
  278. *(a2 + 1) = A2;
  279. *(a1 + 0 + lda) = A7;
  280. *(a1 + 1 + lda) = A8;
  281. *(a2 + 0 + lda) = A5;
  282. *(a2 + 1 + lda) = A6;
  283. } else
  284. if (b2 != a2) {
  285. *(a2 + 0) = B3;
  286. *(a2 + 1) = B4;
  287. *(b2 + 0) = A3;
  288. *(b2 + 1) = A4;
  289. *(a2 + 0 + lda) = B7;
  290. *(a2 + 1 + lda) = B8;
  291. *(b2 + 0 + lda) = A7;
  292. *(b2 + 1 + lda) = A8;
  293. }
  294. } else
  295. if (b1 == a2) {
  296. if (b2 != a1) {
  297. if (b2 == a2) {
  298. *(a1 + 0) = A3;
  299. *(a1 + 1) = A4;
  300. *(a2 + 0) = A1;
  301. *(a2 + 1) = A2;
  302. *(a1 + 0 + lda) = A7;
  303. *(a1 + 1 + lda) = A8;
  304. *(a2 + 0 + lda) = A5;
  305. *(a2 + 1 + lda) = A6;
  306. } else {
  307. *(a1 + 0) = A3;
  308. *(a1 + 1) = A4;
  309. *(a2 + 0) = B3;
  310. *(a2 + 1) = B4;
  311. *(b2 + 0) = A1;
  312. *(b2 + 1) = A2;
  313. *(a1 + 0 + lda) = A7;
  314. *(a1 + 1 + lda) = A8;
  315. *(a2 + 0 + lda) = B7;
  316. *(a2 + 1 + lda) = B8;
  317. *(b2 + 0 + lda) = A5;
  318. *(b2 + 1 + lda) = A6;
  319. }
  320. }
  321. } else {
  322. if (b2 == a1) {
  323. *(a1 + 0) = A3;
  324. *(a1 + 1) = A4;
  325. *(a2 + 0) = B1;
  326. *(a2 + 1) = B2;
  327. *(b1 + 0) = A1;
  328. *(b1 + 1) = A2;
  329. *(a1 + 0 + lda) = A7;
  330. *(a1 + 1 + lda) = A8;
  331. *(a2 + 0 + lda) = B5;
  332. *(a2 + 1 + lda) = B6;
  333. *(b1 + 0 + lda) = A5;
  334. *(b1 + 1 + lda) = A6;
  335. } else
  336. if (b2 == a2) {
  337. *(a1 + 0) = B1;
  338. *(a1 + 1) = B2;
  339. *(b1 + 0) = A1;
  340. *(b1 + 1) = A2;
  341. *(a1 + 0 + lda) = B5;
  342. *(a1 + 1 + lda) = B6;
  343. *(b1 + 0 + lda) = A5;
  344. *(b1 + 1 + lda) = A6;
  345. } else
  346. if (b2 == b1) {
  347. *(a1 + 0) = B1;
  348. *(a1 + 1) = B2;
  349. *(a2 + 0) = A1;
  350. *(a2 + 1) = A2;
  351. *(b1 + 0) = A3;
  352. *(b1 + 1) = A4;
  353. *(a1 + 0 + lda) = B5;
  354. *(a1 + 1 + lda) = B6;
  355. *(a2 + 0 + lda) = A5;
  356. *(a2 + 1 + lda) = A6;
  357. *(b1 + 0 + lda) = A7;
  358. *(b1 + 1 + lda) = A8;
  359. } else {
  360. *(a1 + 0) = B1;
  361. *(a1 + 1) = B2;
  362. *(a2 + 0) = B3;
  363. *(a2 + 1) = B4;
  364. *(b1 + 0) = A1;
  365. *(b1 + 1) = A2;
  366. *(b2 + 0) = A3;
  367. *(b2 + 1) = A4;
  368. *(a1 + 0 + lda) = B5;
  369. *(a1 + 1 + lda) = B6;
  370. *(a2 + 0 + lda) = B7;
  371. *(a2 + 1 + lda) = B8;
  372. *(b1 + 0 + lda) = A5;
  373. *(b1 + 1 + lda) = A6;
  374. *(b2 + 0 + lda) = A7;
  375. *(b2 + 1 + lda) = A8;
  376. }
  377. }
  378. #ifndef MINUS
  379. a1 += 4;
  380. #else
  381. a1 -= 4;
  382. #endif
  383. //Remain
  384. i = (rows & 1);
  385. if (i > 0) {
  386. ip1 = *piv * 2;
  387. b1 = a + ip1;
  388. A1 = *(a1 + 0);
  389. A2 = *(a1 + 1);
  390. A3 = *(a1 + 0 + lda);
  391. A4 = *(a1 + 1 + lda);
  392. B1 = *(b1 + 0);
  393. B2 = *(b1 + 1);
  394. B3 = *(b1 + 0 + lda);
  395. B4 = *(b1 + 1 + lda);
  396. *(a1 + 0) = B1;
  397. *(a1 + 1) = B2;
  398. *(a1 + 0 + lda) = B3;
  399. *(a1 + 1 + lda) = B4;
  400. *(b1 + 0) = A1;
  401. *(b1 + 1) = A2;
  402. *(b1 + 0 + lda) = A3;
  403. *(b1 + 1 + lda) = A4;
  404. }
  405. a += 2 * lda;
  406. j --;
  407. } while (j > 0);
  408. }
  409. if (n & 1) {
  410. piv = ipiv;
  411. #ifndef MINUS
  412. a1 = a + (k1 + 1) * 2;
  413. #else
  414. a1 = a + k2 * 2;
  415. #endif
  416. ip1 = *piv * 2;
  417. piv += incx;
  418. ip2 = *piv * 2;
  419. piv += incx;
  420. b1 = a + ip1;
  421. b2 = a + ip2;
  422. i = (rows >> 1);
  423. i--;
  424. //Loop pipeline
  425. //Main Loop
  426. while (i > 0) {
  427. A1 = *(a1 + 0);
  428. A2 = *(a1 + 1);
  429. A3 = *(a2 + 0);
  430. A4 = *(a2 + 1);
  431. B1 = *(b1 + 0);
  432. B2 = *(b1 + 1);
  433. B3 = *(b2 + 0);
  434. B4 = *(b2 + 1);
  435. ip1 = *piv * 2;
  436. piv += incx;
  437. ip2 = *piv * 2;
  438. piv += incx;
  439. if (b1 == a1) {
  440. if (b2 == a1) {
  441. *(a1 + 0) = A3;
  442. *(a1 + 1) = A4;
  443. *(a2 + 0) = A1;
  444. *(a2 + 1) = A2;
  445. } else
  446. if (b2 != a2) {
  447. *(a2 + 0) = B3;
  448. *(a2 + 1) = B4;
  449. *(b2 + 0) = A3;
  450. *(b2 + 1) = A4;
  451. }
  452. } else
  453. if (b1 == a2) {
  454. if (b2 != a1) {
  455. if (b2 == a2) {
  456. *(a1 + 0) = A3;
  457. *(a1 + 1) = A4;
  458. *(a2 + 0) = A1;
  459. *(a2 + 1) = A2;
  460. } else {
  461. *(a1 + 0) = A3;
  462. *(a1 + 1) = A4;
  463. *(a2 + 0) = B3;
  464. *(a2 + 1) = B4;
  465. *(b2 + 0) = A1;
  466. *(b2 + 1) = A2;
  467. }
  468. }
  469. } else {
  470. if (b2 == a1) {
  471. *(a1 + 0) = A3;
  472. *(a1 + 1) = A4;
  473. *(a2 + 0) = B1;
  474. *(a2 + 1) = B2;
  475. *(b1 + 0) = A1;
  476. *(b1 + 1) = A2;
  477. } else
  478. if (b2 == a2) {
  479. *(a1 + 0) = B1;
  480. *(a1 + 1) = B2;
  481. *(b1 + 0) = A1;
  482. *(b1 + 1) = A2;
  483. } else
  484. if (b2 == b1) {
  485. *(a1 + 0) = B1;
  486. *(a1 + 1) = B2;
  487. *(a2 + 0) = A1;
  488. *(a2 + 1) = A2;
  489. *(b1 + 0) = A3;
  490. *(b1 + 1) = A4;
  491. } else {
  492. *(a1 + 0) = B1;
  493. *(a1 + 1) = B2;
  494. *(a2 + 0) = B3;
  495. *(a2 + 1) = B4;
  496. *(b1 + 0) = A1;
  497. *(b1 + 1) = A2;
  498. *(b2 + 0) = A3;
  499. *(b2 + 1) = A4;
  500. }
  501. }
  502. b1 = a + ip1;
  503. b2 = a + ip2;
  504. #ifndef MINUS
  505. a1 += 4;
  506. #else
  507. a1 -= 4;
  508. #endif
  509. i --;
  510. }
  511. //Loop Ending
  512. A1 = *(a1 + 0);
  513. A2 = *(a1 + 1);
  514. A3 = *(a2 + 0);
  515. A4 = *(a2 + 1);
  516. B1 = *(b1 + 0);
  517. B2 = *(b1 + 1);
  518. B3 = *(b2 + 0);
  519. B4 = *(b2 + 1);
  520. if (b1 == a1) {
  521. if (b2 == a1) {
  522. *(a1 + 0) = A3;
  523. *(a1 + 1) = A4;
  524. *(a2 + 0) = A1;
  525. *(a2 + 1) = A2;
  526. } else
  527. if (b2 != a2) {
  528. *(a2 + 0) = B3;
  529. *(a2 + 1) = B4;
  530. *(b2 + 0) = A3;
  531. *(b2 + 1) = A4;
  532. }
  533. } else
  534. if (b1 == a2) {
  535. if (b2 != a1) {
  536. if (b2 == a2) {
  537. *(a1 + 0) = A3;
  538. *(a1 + 1) = A4;
  539. *(a2 + 0) = A1;
  540. *(a2 + 1) = A2;
  541. } else {
  542. *(a1 + 0) = A3;
  543. *(a1 + 1) = A4;
  544. *(a2 + 0) = B3;
  545. *(a2 + 1) = B4;
  546. *(b2 + 0) = A1;
  547. *(b2 + 1) = A2;
  548. }
  549. }
  550. } else {
  551. if (b2 == a1) {
  552. *(a1 + 0) = A3;
  553. *(a1 + 1) = A4;
  554. *(a2 + 0) = B1;
  555. *(a2 + 1) = B2;
  556. *(b1 + 0) = A1;
  557. *(b1 + 1) = A2;
  558. } else
  559. if (b2 == a2) {
  560. *(a1 + 0) = B1;
  561. *(a1 + 1) = B2;
  562. *(b1 + 0) = A1;
  563. *(b1 + 1) = A2;
  564. } else
  565. if (b2 == b1) {
  566. *(a1 + 0) = B1;
  567. *(a1 + 1) = B2;
  568. *(a2 + 0) = A1;
  569. *(a2 + 1) = A2;
  570. *(b1 + 0) = A3;
  571. *(b1 + 1) = A4;
  572. } else {
  573. *(a1 + 0) = B1;
  574. *(a1 + 1) = B2;
  575. *(a2 + 0) = B3;
  576. *(a2 + 1) = B4;
  577. *(b1 + 0) = A1;
  578. *(b1 + 1) = A2;
  579. *(b2 + 0) = A3;
  580. *(b2 + 1) = A4;
  581. }
  582. }
  583. #ifndef MINUS
  584. a1 += 4;
  585. #else
  586. a1 -= 4;
  587. #endif
  588. //Remain
  589. i = (rows & 1);
  590. if (i > 0) {
  591. ip1 = *piv * 2;
  592. b1 = a + ip1;
  593. A1 = *(a1 + 0);
  594. A2 = *(a1 + 1);
  595. B1 = *(b1 + 0);
  596. B2 = *(b1 + 1);
  597. *(a1 + 0) = B1;
  598. *(a1 + 1) = B2;
  599. *(b1 + 0) = A1;
  600. *(b1 + 1) = A2;
  601. }
  602. }
  603. return 0;
  604. }