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.

zhemm_ltcopy_8.c 12 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480
  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, offset;
  42. FLOAT data01, data02, data03, data04, data05, data06, data07, data08;
  43. FLOAT data09, data10, data11, data12, data13, data14, data15, data16;
  44. FLOAT *ao1, *ao2, *ao3, *ao4, *ao5, *ao6, *ao7, *ao8;
  45. lda *= 2;
  46. js = (n >> 3);
  47. while (js > 0){
  48. offset = posX - posY;
  49. if (offset > 0) ao1 = a + (posX + 0) * 2 + posY * lda; else ao1 = a + posY * 2 + (posX + 0) * lda;
  50. if (offset > -1) ao2 = a + (posX + 1) * 2 + posY * lda; else ao2 = a + posY * 2 + (posX + 1) * lda;
  51. if (offset > -2) ao3 = a + (posX + 2) * 2 + posY * lda; else ao3 = a + posY * 2 + (posX + 2) * lda;
  52. if (offset > -3) ao4 = a + (posX + 3) * 2 + posY * lda; else ao4 = a + posY * 2 + (posX + 3) * lda;
  53. if (offset > -4) ao5 = a + (posX + 4) * 2 + posY * lda; else ao5 = a + posY * 2 + (posX + 4) * lda;
  54. if (offset > -5) ao6 = a + (posX + 5) * 2 + posY * lda; else ao6 = a + posY * 2 + (posX + 5) * lda;
  55. if (offset > -6) ao7 = a + (posX + 6) * 2 + posY * lda; else ao7 = a + posY * 2 + (posX + 6) * lda;
  56. if (offset > -7) ao8 = a + (posX + 7) * 2 + posY * lda; else ao8 = a + posY * 2 + (posX + 7) * lda;
  57. i = m;
  58. while (i > 0) {
  59. data01 = *(ao1 + 0);
  60. data02 = *(ao1 + 1);
  61. data03 = *(ao2 + 0);
  62. data04 = *(ao2 + 1);
  63. data05 = *(ao3 + 0);
  64. data06 = *(ao3 + 1);
  65. data07 = *(ao4 + 0);
  66. data08 = *(ao4 + 1);
  67. data09 = *(ao5 + 0);
  68. data10 = *(ao5 + 1);
  69. data11 = *(ao6 + 0);
  70. data12 = *(ao6 + 1);
  71. data13 = *(ao7 + 0);
  72. data14 = *(ao7 + 1);
  73. data15 = *(ao8 + 0);
  74. data16 = *(ao8 + 1);
  75. if (offset > 0) ao1 += lda; else ao1 += 2;
  76. if (offset > -1) ao2 += lda; else ao2 += 2;
  77. if (offset > -2) ao3 += lda; else ao3 += 2;
  78. if (offset > -3) ao4 += lda; else ao4 += 2;
  79. if (offset > -4) ao5 += lda; else ao5 += 2;
  80. if (offset > -5) ao6 += lda; else ao6 += 2;
  81. if (offset > -6) ao7 += lda; else ao7 += 2;
  82. if (offset > -7) ao8 += lda; else ao8 += 2;
  83. if (offset > 0) {
  84. b[ 0] = data01;
  85. b[ 1] = data02;
  86. b[ 2] = data03;
  87. b[ 3] = data04;
  88. b[ 4] = data05;
  89. b[ 5] = data06;
  90. b[ 6] = data07;
  91. b[ 7] = data08;
  92. b[ 8] = data09;
  93. b[ 9] = data10;
  94. b[10] = data11;
  95. b[11] = data12;
  96. b[12] = data13;
  97. b[13] = data14;
  98. b[14] = data15;
  99. b[15] = data16;
  100. } else
  101. if (offset < -7) {
  102. b[ 0] = data01;
  103. b[ 1] = -data02;
  104. b[ 2] = data03;
  105. b[ 3] = -data04;
  106. b[ 4] = data05;
  107. b[ 5] = -data06;
  108. b[ 6] = data07;
  109. b[ 7] = -data08;
  110. b[ 8] = data09;
  111. b[ 9] = -data10;
  112. b[10] = data11;
  113. b[11] = -data12;
  114. b[12] = data13;
  115. b[13] = -data14;
  116. b[14] = data15;
  117. b[15] = -data16;
  118. } else {
  119. switch (offset) {
  120. case 0 :
  121. b[ 0] = data01;
  122. b[ 1] = ZERO;
  123. b[ 2] = data03;
  124. b[ 3] = data04;
  125. b[ 4] = data05;
  126. b[ 5] = data06;
  127. b[ 6] = data07;
  128. b[ 7] = data08;
  129. b[ 8] = data09;
  130. b[ 9] = data10;
  131. b[10] = data11;
  132. b[11] = data12;
  133. b[12] = data13;
  134. b[13] = data14;
  135. b[14] = data15;
  136. b[15] = data16;
  137. break;
  138. case -1 :
  139. b[ 0] = data01;
  140. b[ 1] = -data02;
  141. b[ 2] = data03;
  142. b[ 3] = ZERO;
  143. b[ 4] = data05;
  144. b[ 5] = data06;
  145. b[ 6] = data07;
  146. b[ 7] = data08;
  147. b[ 8] = data09;
  148. b[ 9] = data10;
  149. b[10] = data11;
  150. b[11] = data12;
  151. b[12] = data13;
  152. b[13] = data14;
  153. b[14] = data15;
  154. b[15] = data16;
  155. break;
  156. case -2 :
  157. b[ 0] = data01;
  158. b[ 1] = -data02;
  159. b[ 2] = data03;
  160. b[ 3] = -data04;
  161. b[ 4] = data05;
  162. b[ 5] = ZERO;
  163. b[ 6] = data07;
  164. b[ 7] = data08;
  165. b[ 8] = data09;
  166. b[ 9] = data10;
  167. b[10] = data11;
  168. b[11] = data12;
  169. b[12] = data13;
  170. b[13] = data14;
  171. b[14] = data15;
  172. b[15] = data16;
  173. break;
  174. case -3 :
  175. b[ 0] = data01;
  176. b[ 1] = -data02;
  177. b[ 2] = data03;
  178. b[ 3] = -data04;
  179. b[ 4] = data05;
  180. b[ 5] = -data06;
  181. b[ 6] = data07;
  182. b[ 7] = ZERO;
  183. b[ 8] = data09;
  184. b[ 9] = data10;
  185. b[10] = data11;
  186. b[11] = data12;
  187. b[12] = data13;
  188. b[13] = data14;
  189. b[14] = data15;
  190. b[15] = data16;
  191. break;
  192. case -4 :
  193. b[ 0] = data01;
  194. b[ 1] = -data02;
  195. b[ 2] = data03;
  196. b[ 3] = -data04;
  197. b[ 4] = data05;
  198. b[ 5] = -data06;
  199. b[ 6] = data07;
  200. b[ 7] = -data08;
  201. b[ 8] = data09;
  202. b[ 9] = ZERO;
  203. b[10] = data11;
  204. b[11] = data12;
  205. b[12] = data13;
  206. b[13] = data14;
  207. b[14] = data15;
  208. b[15] = data16;
  209. break;
  210. case -5 :
  211. b[ 0] = data01;
  212. b[ 1] = -data02;
  213. b[ 2] = data03;
  214. b[ 3] = -data04;
  215. b[ 4] = data05;
  216. b[ 5] = -data06;
  217. b[ 6] = data07;
  218. b[ 7] = -data08;
  219. b[ 8] = data09;
  220. b[ 9] = -data10;
  221. b[10] = data11;
  222. b[11] = ZERO;
  223. b[12] = data13;
  224. b[13] = data14;
  225. b[14] = data15;
  226. b[15] = data16;
  227. break;
  228. case -6 :
  229. b[ 0] = data01;
  230. b[ 1] = -data02;
  231. b[ 2] = data03;
  232. b[ 3] = -data04;
  233. b[ 4] = data05;
  234. b[ 5] = -data06;
  235. b[ 6] = data07;
  236. b[ 7] = -data08;
  237. b[ 8] = data09;
  238. b[ 9] = -data10;
  239. b[10] = data11;
  240. b[11] = -data12;
  241. b[12] = data13;
  242. b[13] = ZERO;
  243. b[14] = data15;
  244. b[15] = data16;
  245. break;
  246. case -7 :
  247. b[ 0] = data01;
  248. b[ 1] = -data02;
  249. b[ 2] = data03;
  250. b[ 3] = -data04;
  251. b[ 4] = data05;
  252. b[ 5] = -data06;
  253. b[ 6] = data07;
  254. b[ 7] = -data08;
  255. b[ 8] = data09;
  256. b[ 9] = -data10;
  257. b[10] = data11;
  258. b[11] = -data12;
  259. b[12] = data13;
  260. b[13] = -data14;
  261. b[14] = data15;
  262. b[15] = ZERO;
  263. break;
  264. }
  265. }
  266. b += 16;
  267. offset --;
  268. i --;
  269. }
  270. posX += 8;
  271. js --;
  272. }
  273. if (n & 4) {
  274. offset = posX - posY;
  275. if (offset > 0) ao1 = a + (posX + 0) * 2 + posY * lda; else ao1 = a + posY * 2 + (posX + 0) * lda;
  276. if (offset > -1) ao2 = a + (posX + 1) * 2 + posY * lda; else ao2 = a + posY * 2 + (posX + 1) * lda;
  277. if (offset > -2) ao3 = a + (posX + 2) * 2 + posY * lda; else ao3 = a + posY * 2 + (posX + 2) * lda;
  278. if (offset > -3) ao4 = a + (posX + 3) * 2 + posY * lda; else ao4 = a + posY * 2 + (posX + 3) * lda;
  279. i = m;
  280. while (i > 0) {
  281. data01 = *(ao1 + 0);
  282. data02 = *(ao1 + 1);
  283. data03 = *(ao2 + 0);
  284. data04 = *(ao2 + 1);
  285. data05 = *(ao3 + 0);
  286. data06 = *(ao3 + 1);
  287. data07 = *(ao4 + 0);
  288. data08 = *(ao4 + 1);
  289. if (offset > 0) ao1 += lda; else ao1 += 2;
  290. if (offset > -1) ao2 += lda; else ao2 += 2;
  291. if (offset > -2) ao3 += lda; else ao3 += 2;
  292. if (offset > -3) ao4 += lda; else ao4 += 2;
  293. if (offset > 0) {
  294. b[ 0] = data01;
  295. b[ 1] = data02;
  296. b[ 2] = data03;
  297. b[ 3] = data04;
  298. b[ 4] = data05;
  299. b[ 5] = data06;
  300. b[ 6] = data07;
  301. b[ 7] = data08;
  302. } else
  303. if (offset < -3) {
  304. b[ 0] = data01;
  305. b[ 1] = -data02;
  306. b[ 2] = data03;
  307. b[ 3] = -data04;
  308. b[ 4] = data05;
  309. b[ 5] = -data06;
  310. b[ 6] = data07;
  311. b[ 7] = -data08;
  312. } else {
  313. switch (offset) {
  314. case 0 :
  315. b[ 0] = data01;
  316. b[ 1] = ZERO;
  317. b[ 2] = data03;
  318. b[ 3] = data04;
  319. b[ 4] = data05;
  320. b[ 5] = data06;
  321. b[ 6] = data07;
  322. b[ 7] = data08;
  323. break;
  324. case -1 :
  325. b[ 0] = data01;
  326. b[ 1] = -data02;
  327. b[ 2] = data03;
  328. b[ 3] = ZERO;
  329. b[ 4] = data05;
  330. b[ 5] = data06;
  331. b[ 6] = data07;
  332. b[ 7] = data08;
  333. break;
  334. case -2 :
  335. b[ 0] = data01;
  336. b[ 1] = -data02;
  337. b[ 2] = data03;
  338. b[ 3] = -data04;
  339. b[ 4] = data05;
  340. b[ 5] = ZERO;
  341. b[ 6] = data07;
  342. b[ 7] = data08;
  343. break;
  344. case -3 :
  345. b[ 0] = data01;
  346. b[ 1] = -data02;
  347. b[ 2] = data03;
  348. b[ 3] = -data04;
  349. b[ 4] = data05;
  350. b[ 5] = -data06;
  351. b[ 6] = data07;
  352. b[ 7] = ZERO;
  353. break;
  354. }
  355. }
  356. b += 8;
  357. offset --;
  358. i --;
  359. }
  360. posX += 4;
  361. }
  362. if (n & 2) {
  363. offset = posX - posY;
  364. if (offset > 0) ao1 = a + (posX + 0) * 2 + posY * lda; else ao1 = a + posY * 2 + (posX + 0) * lda;
  365. if (offset > -1) ao2 = a + (posX + 1) * 2 + posY * lda; else ao2 = a + posY * 2 + (posX + 1) * lda;
  366. i = m;
  367. while (i > 0) {
  368. data01 = *(ao1 + 0);
  369. data02 = *(ao1 + 1);
  370. data03 = *(ao2 + 0);
  371. data04 = *(ao2 + 1);
  372. if (offset > 0) ao1 += lda; else ao1 += 2;
  373. if (offset > -1) ao2 += lda; else ao2 += 2;
  374. if (offset > 0) {
  375. b[ 0] = data01;
  376. b[ 1] = data02;
  377. b[ 2] = data03;
  378. b[ 3] = data04;
  379. } else
  380. if (offset < -1) {
  381. b[ 0] = data01;
  382. b[ 1] = -data02;
  383. b[ 2] = data03;
  384. b[ 3] = -data04;
  385. } else {
  386. switch (offset) {
  387. case 0 :
  388. b[ 0] = data01;
  389. b[ 1] = ZERO;
  390. b[ 2] = data03;
  391. b[ 3] = data04;
  392. break;
  393. case -1 :
  394. b[ 0] = data01;
  395. b[ 1] = -data02;
  396. b[ 2] = data03;
  397. b[ 3] = ZERO;
  398. break;
  399. }
  400. }
  401. b += 4;
  402. offset --;
  403. i --;
  404. }
  405. posX += 2;
  406. }
  407. if (n & 1) {
  408. offset = posX - posY;
  409. if (offset > 0) ao1 = a + (posX + 0) * 2 + posY * lda; else ao1 = a + posY * 2 + (posX + 0) * lda;
  410. i = m;
  411. while (i > 0) {
  412. data01 = *(ao1 + 0);
  413. data02 = *(ao1 + 1);
  414. if (offset > 0) ao1 += lda; else ao1 += 2;
  415. if (offset > 0) {
  416. b[ 0] = data01;
  417. b[ 1] = data02;
  418. } else
  419. if (offset < 0) {
  420. b[ 0] = data01;
  421. b[ 1] = -data02;
  422. } else {
  423. b[ 0] = data01;
  424. b[ 1] = ZERO;
  425. }
  426. b += 2;
  427. offset --;
  428. i --;
  429. }
  430. }
  431. return 0;
  432. }