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.

zgemm_tcopy_macros_8_power8.S 12 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631
  1. /***************************************************************************
  2. Copyright (c) 2013-2016, The OpenBLAS Project
  3. All rights reserved.
  4. Redistribution and use in source and binary forms, with or without
  5. modification, are permitted provided that the following conditions are
  6. met:
  7. 1. Redistributions of source code must retain the above copyright
  8. notice, this list of conditions and the following disclaimer.
  9. 2. Redistributions in binary form must reproduce the above copyright
  10. notice, this list of conditions and the following disclaimer in
  11. the documentation and/or other materials provided with the
  12. distribution.
  13. 3. Neither the name of the OpenBLAS project nor the names of
  14. its contributors may be used to endorse or promote products
  15. derived from this software without specific prior written permission.
  16. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  17. AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18. IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19. ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
  20. LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21. DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  22. SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  23. CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  24. OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
  25. USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  26. *****************************************************************************/
  27. /**************************************************************************************
  28. * 2016/04/22 Werner Saar (wernsaar@googlemail.com)
  29. * BLASTEST : OK
  30. * CTEST : OK
  31. * TEST : OK
  32. * LAPACK-TEST : OK
  33. **************************************************************************************/
  34. /**********************************************************************************************
  35. * Macros for N=4 and M=8
  36. **********************************************************************************************/
  37. #if defined(_AIX)
  38. define(`COPY_4x8', `
  39. #else
  40. .macro COPY_4x8
  41. #endif
  42. lxvd2x vs32, o0, A0
  43. lxvd2x vs33, o16, A0
  44. lxvd2x vs34, o32, A0
  45. lxvd2x vs35, o48, A0
  46. addi A0, A0, 64
  47. lxvd2x vs36, o0, A0
  48. lxvd2x vs37, o16, A0
  49. lxvd2x vs38, o32, A0
  50. lxvd2x vs39, o48, A0
  51. addi A0, A0, 64
  52. lxvd2x vs40, o0, A1
  53. lxvd2x vs41, o16, A1
  54. lxvd2x vs42, o32, A1
  55. lxvd2x vs43, o48, A1
  56. addi A1, A1, 64
  57. lxvd2x vs44, o0, A1
  58. lxvd2x vs45, o16, A1
  59. lxvd2x vs46, o32, A1
  60. lxvd2x vs47, o48, A1
  61. addi A1, A1, 64
  62. lxvd2x vs48, o0, A2
  63. lxvd2x vs49, o16, A2
  64. lxvd2x vs50, o32, A2
  65. lxvd2x vs51, o48, A2
  66. addi A2, A2, 64
  67. lxvd2x vs2, o0, A2
  68. lxvd2x vs3, o16, A2
  69. lxvd2x vs4, o32, A2
  70. lxvd2x vs5, o48, A2
  71. addi A2, A2, 64
  72. lxvd2x vs6, o0, A3
  73. lxvd2x vs7, o16, A3
  74. lxvd2x vs8, o32, A3
  75. lxvd2x vs9, o48, A3
  76. addi A3, A3, 64
  77. lxvd2x vs10, o0, A3
  78. lxvd2x vs11, o16, A3
  79. lxvd2x vs12, o32, A3
  80. lxvd2x vs13, o48, A3
  81. addi A3, A3, 64
  82. mr T1, BO
  83. stxvd2x vs32, o0, T1
  84. stxvd2x vs33, o16, T1
  85. stxvd2x vs34, o32, T1
  86. stxvd2x vs35, o48, T1
  87. addi T1, T1, 64
  88. stxvd2x vs36, o0, T1
  89. stxvd2x vs37, o16, T1
  90. stxvd2x vs38, o32, T1
  91. stxvd2x vs39, o48, T1
  92. addi T1, T1, 64
  93. stxvd2x vs40, o0, T1
  94. stxvd2x vs41, o16, T1
  95. stxvd2x vs42, o32, T1
  96. stxvd2x vs43, o48, T1
  97. addi T1, T1, 64
  98. stxvd2x vs44, o0, T1
  99. stxvd2x vs45, o16, T1
  100. stxvd2x vs46, o32, T1
  101. stxvd2x vs47, o48, T1
  102. addi T1, T1, 64
  103. stxvd2x vs48, o0, T1
  104. stxvd2x vs49, o16, T1
  105. stxvd2x vs50, o32, T1
  106. stxvd2x vs51, o48, T1
  107. addi T1, T1, 64
  108. stxvd2x vs2, o0, T1
  109. stxvd2x vs3, o16, T1
  110. stxvd2x vs4, o32, T1
  111. stxvd2x vs5, o48, T1
  112. addi T1, T1, 64
  113. stxvd2x vs6, o0, T1
  114. stxvd2x vs7, o16, T1
  115. stxvd2x vs8, o32, T1
  116. stxvd2x vs9, o48, T1
  117. addi T1, T1, 64
  118. stxvd2x vs10, o0, T1
  119. stxvd2x vs11, o16, T1
  120. stxvd2x vs12, o32, T1
  121. stxvd2x vs13, o48, T1
  122. #if defined(_AIX)
  123. ')
  124. #else
  125. .endm
  126. #endif
  127. /**********************************************************************************************
  128. * Macros for N=4 and M=4
  129. **********************************************************************************************/
  130. #if defined(_AIX)
  131. define(`COPY_4x4', `
  132. #else
  133. .macro COPY_4x4
  134. #endif
  135. lxvd2x vs32, o0, A0
  136. lxvd2x vs33, o16, A0
  137. lxvd2x vs34, o32, A0
  138. lxvd2x vs35, o48, A0
  139. addi A0, A0, 64
  140. lxvd2x vs36, o0, A1
  141. lxvd2x vs37, o16, A1
  142. lxvd2x vs38, o32, A1
  143. lxvd2x vs39, o48, A1
  144. addi A1, A1, 64
  145. lxvd2x vs40, o0, A2
  146. lxvd2x vs41, o16, A2
  147. lxvd2x vs42, o32, A2
  148. lxvd2x vs43, o48, A2
  149. addi A2, A2, 64
  150. lxvd2x vs44, o0, A3
  151. lxvd2x vs45, o16, A3
  152. lxvd2x vs46, o32, A3
  153. lxvd2x vs47, o48, A3
  154. addi A3, A3, 64
  155. mr T1, BO
  156. stxvd2x vs32, o0, T1
  157. stxvd2x vs33, o16, T1
  158. stxvd2x vs34, o32, T1
  159. stxvd2x vs35, o48, T1
  160. addi T1, T1, 64
  161. stxvd2x vs36, o0, T1
  162. stxvd2x vs37, o16, T1
  163. stxvd2x vs38, o32, T1
  164. stxvd2x vs39, o48, T1
  165. addi T1, T1, 64
  166. stxvd2x vs40, o0, T1
  167. stxvd2x vs41, o16, T1
  168. stxvd2x vs42, o32, T1
  169. stxvd2x vs43, o48, T1
  170. addi T1, T1, 64
  171. stxvd2x vs44, o0, T1
  172. stxvd2x vs45, o16, T1
  173. stxvd2x vs46, o32, T1
  174. stxvd2x vs47, o48, T1
  175. #if defined(_AIX)
  176. ')
  177. #else
  178. .endm
  179. #endif
  180. /**********************************************************************************************
  181. * Macros for N=4 and M=2
  182. **********************************************************************************************/
  183. #if defined(_AIX)
  184. define(`COPY_4x2', `
  185. #else
  186. .macro COPY_4x2
  187. #endif
  188. lxvd2x vs32, o0, A0
  189. lxvd2x vs33, o16, A0
  190. addi A0, A0, 32
  191. lxvd2x vs34, o0, A1
  192. lxvd2x vs35, o16, A1
  193. addi A1, A1, 32
  194. lxvd2x vs36, o0, A2
  195. lxvd2x vs37, o16, A2
  196. addi A2, A2, 32
  197. lxvd2x vs38, o0, A3
  198. lxvd2x vs39, o16, A3
  199. addi A3, A3, 32
  200. mr T1, BO
  201. stxvd2x vs32, o0, T1
  202. stxvd2x vs33, o16, T1
  203. stxvd2x vs34, o32, T1
  204. stxvd2x vs35, o48, T1
  205. addi T1, T1, 64
  206. stxvd2x vs36, o0, T1
  207. stxvd2x vs37, o16, T1
  208. stxvd2x vs38, o32, T1
  209. stxvd2x vs39, o48, T1
  210. #if defined(_AIX)
  211. ')
  212. #else
  213. .endm
  214. #endif
  215. /**********************************************************************************************
  216. * Macros for N=4 and M=1
  217. **********************************************************************************************/
  218. #if defined(_AIX)
  219. define(`COPY_4x1', `
  220. #else
  221. .macro COPY_4x1
  222. #endif
  223. lxvd2x vs32, o0, A0
  224. addi A0, A0, 16
  225. lxvd2x vs33, o0, A1
  226. addi A1, A1, 16
  227. lxvd2x vs34, o0, A2
  228. addi A2, A2, 16
  229. lxvd2x vs35, o0, A3
  230. addi A3, A3, 16
  231. mr T1, BO
  232. stxvd2x vs32, o0, T1
  233. stxvd2x vs33, o16, T1
  234. stxvd2x vs34, o32, T1
  235. stxvd2x vs35, o48, T1
  236. #if defined(_AIX)
  237. ')
  238. #else
  239. .endm
  240. #endif
  241. /**********************************************************************************************
  242. * Macros for N=2 and M=8
  243. **********************************************************************************************/
  244. #if defined(_AIX)
  245. define(`COPY_2x8', `
  246. #else
  247. .macro COPY_2x8
  248. #endif
  249. lxvd2x vs32, o0, A0
  250. lxvd2x vs33, o16, A0
  251. lxvd2x vs34, o32, A0
  252. lxvd2x vs35, o48, A0
  253. addi A0, A0, 64
  254. lxvd2x vs36, o0, A0
  255. lxvd2x vs37, o16, A0
  256. lxvd2x vs38, o32, A0
  257. lxvd2x vs39, o48, A0
  258. addi A0, A0, 64
  259. lxvd2x vs40, o0, A1
  260. lxvd2x vs41, o16, A1
  261. lxvd2x vs42, o32, A1
  262. lxvd2x vs43, o48, A1
  263. addi A1, A1, 64
  264. lxvd2x vs44, o0, A1
  265. lxvd2x vs45, o16, A1
  266. lxvd2x vs46, o32, A1
  267. lxvd2x vs47, o48, A1
  268. addi A1, A1, 64
  269. mr T1, BO
  270. stxvd2x vs32, o0, T1
  271. stxvd2x vs33, o16, T1
  272. stxvd2x vs34, o32, T1
  273. stxvd2x vs35, o48, T1
  274. addi T1, T1, 64
  275. stxvd2x vs36, o0, T1
  276. stxvd2x vs37, o16, T1
  277. stxvd2x vs38, o32, T1
  278. stxvd2x vs39, o48, T1
  279. addi T1, T1, 64
  280. stxvd2x vs40, o0, T1
  281. stxvd2x vs41, o16, T1
  282. stxvd2x vs42, o32, T1
  283. stxvd2x vs43, o48, T1
  284. addi T1, T1, 64
  285. stxvd2x vs44, o0, T1
  286. stxvd2x vs45, o16, T1
  287. stxvd2x vs46, o32, T1
  288. stxvd2x vs47, o48, T1
  289. #if defined(_AIX)
  290. ')
  291. #else
  292. .endm
  293. #endif
  294. /**********************************************************************************************
  295. * Macros for N=2 and M=4
  296. **********************************************************************************************/
  297. #if defined(_AIX)
  298. define(`COPY_2x4', `
  299. #else
  300. .macro COPY_2x4
  301. #endif
  302. lxvd2x vs32, o0, A0
  303. lxvd2x vs33, o16, A0
  304. lxvd2x vs34, o32, A0
  305. lxvd2x vs35, o48, A0
  306. addi A0, A0, 64
  307. lxvd2x vs36, o0, A1
  308. lxvd2x vs37, o16, A1
  309. lxvd2x vs38, o32, A1
  310. lxvd2x vs39, o48, A1
  311. addi A1, A1, 64
  312. mr T1, BO
  313. stxvd2x vs32, o0, T1
  314. stxvd2x vs33, o16, T1
  315. stxvd2x vs34, o32, T1
  316. stxvd2x vs35, o48, T1
  317. addi T1, T1, 64
  318. stxvd2x vs36, o0, T1
  319. stxvd2x vs37, o16, T1
  320. stxvd2x vs38, o32, T1
  321. stxvd2x vs39, o48, T1
  322. #if defined(_AIX)
  323. ')
  324. #else
  325. .endm
  326. #endif
  327. /**********************************************************************************************
  328. * Macros for N=2 and M=2
  329. **********************************************************************************************/
  330. #if defined(_AIX)
  331. define(`COPY_2x2', `
  332. #else
  333. .macro COPY_2x2
  334. #endif
  335. lxvd2x vs32, o0, A0
  336. lxvd2x vs33, o16, A0
  337. addi A0, A0, 32
  338. lxvd2x vs34, o0, A1
  339. lxvd2x vs35, o16, A1
  340. addi A1, A1, 32
  341. mr T1, BO
  342. stxvd2x vs32, o0, T1
  343. stxvd2x vs33, o16, T1
  344. stxvd2x vs34, o32, T1
  345. stxvd2x vs35, o48, T1
  346. #if defined(_AIX)
  347. ')
  348. #else
  349. .endm
  350. #endif
  351. /**********************************************************************************************
  352. * Macros for N=2 and M=1
  353. **********************************************************************************************/
  354. #if defined(_AIX)
  355. define(`COPY_2x1', `
  356. #else
  357. .macro COPY_2x1
  358. #endif
  359. lxvd2x vs32, o0, A0
  360. addi A0, A0, 16
  361. lxvd2x vs33, o0, A1
  362. addi A1, A1, 16
  363. mr T1, BO
  364. stxvd2x vs32, o0, T1
  365. stxvd2x vs33, o16, T1
  366. #if defined(_AIX)
  367. ')
  368. #else
  369. .endm
  370. #endif
  371. /**********************************************************************************************
  372. * Macros for N=1 and M=8
  373. **********************************************************************************************/
  374. #if defined(_AIX)
  375. define(`COPY_1x8', `
  376. #else
  377. .macro COPY_1x8
  378. #endif
  379. lxvd2x vs32, o0, A0
  380. lxvd2x vs33, o16, A0
  381. lxvd2x vs34, o32, A0
  382. lxvd2x vs35, o48, A0
  383. addi A0, A0, 64
  384. lxvd2x vs36, o0, A0
  385. lxvd2x vs37, o16, A0
  386. lxvd2x vs38, o32, A0
  387. lxvd2x vs39, o48, A0
  388. addi A0, A0, 64
  389. mr T1, BO
  390. stxvd2x vs32, o0, T1
  391. stxvd2x vs33, o16, T1
  392. stxvd2x vs34, o32, T1
  393. stxvd2x vs35, o48, T1
  394. addi T1, T1, 64
  395. stxvd2x vs36, o0, T1
  396. stxvd2x vs37, o16, T1
  397. stxvd2x vs38, o32, T1
  398. stxvd2x vs39, o48, T1
  399. #if defined(_AIX)
  400. ')
  401. #else
  402. .endm
  403. #endif
  404. /**********************************************************************************************
  405. * Macros for N=1 and M=4
  406. **********************************************************************************************/
  407. #if defined(_AIX)
  408. define(`COPY_1x4', `
  409. #else
  410. .macro COPY_1x4
  411. #endif
  412. lxvd2x vs32, o0, A0
  413. lxvd2x vs33, o16, A0
  414. lxvd2x vs34, o32, A0
  415. lxvd2x vs35, o48, A0
  416. addi A0, A0, 64
  417. mr T1, BO
  418. stxvd2x vs32, o0, T1
  419. stxvd2x vs33, o16, T1
  420. stxvd2x vs34, o32, T1
  421. stxvd2x vs35, o48, T1
  422. #if defined(_AIX)
  423. ')
  424. #else
  425. .endm
  426. #endif
  427. /**********************************************************************************************
  428. * Macros for N=1 and M=2
  429. **********************************************************************************************/
  430. #if defined(_AIX)
  431. define(`COPY_1x2', `
  432. #else
  433. .macro COPY_1x2
  434. #endif
  435. lxvd2x vs32, o0, A0
  436. lxvd2x vs33, o16, A0
  437. addi A0, A0, 32
  438. mr T1, BO
  439. stxvd2x vs32, o0, T1
  440. stxvd2x vs33, o16, T1
  441. #if defined(_AIX)
  442. ')
  443. #else
  444. .endm
  445. #endif
  446. /**********************************************************************************************
  447. * Macros for N=1 and M=1
  448. **********************************************************************************************/
  449. #if defined(_AIX)
  450. define(`COPY_1x1', `
  451. #else
  452. .macro COPY_1x1
  453. #endif
  454. lxvd2x vs32, o0, A0
  455. addi A0, A0, 16
  456. mr T1, BO
  457. stxvd2x vs32, o0, T1
  458. #if defined(_AIX)
  459. ')
  460. #else
  461. .endm
  462. #endif