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.

dgemm_tcopy_macros_16_power8.S 13 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728
  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. * 2016/04/21 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=16
  36. **********************************************************************************************/
  37. #if defined(_AIX)
  38. define(`COPY_4x16', `
  39. #else
  40. .macro COPY_4x16
  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 vs40, o0, A1
  48. lxvd2x vs41, o16, A1
  49. lxvd2x vs42, o32, A1
  50. lxvd2x vs43, o48, A1
  51. addi A1, A1, 64
  52. lxvd2x vs48, o0, A2
  53. lxvd2x vs49, o16, A2
  54. lxvd2x vs50, o32, A2
  55. lxvd2x vs51, o48, A2
  56. addi A2, A2, 64
  57. lxvd2x vs4, o0, A3
  58. lxvd2x vs5, o16, A3
  59. lxvd2x vs6, o32, A3
  60. lxvd2x vs7, o48, A3
  61. addi A3, A3, 64
  62. lxvd2x vs36, o0, A0
  63. lxvd2x vs37, o16, A0
  64. lxvd2x vs38, o32, A0
  65. lxvd2x vs39, o48, A0
  66. addi A0, A0, 64
  67. lxvd2x vs44, o0, A1
  68. lxvd2x vs45, o16, A1
  69. lxvd2x vs46, o32, A1
  70. lxvd2x vs47, o48, A1
  71. addi A1, A1, 64
  72. lxvd2x vs12, o0, A2
  73. lxvd2x vs13, o16, A2
  74. lxvd2x vs2, o32, A2
  75. lxvd2x vs3, o48, A2
  76. addi A2, A2, 64
  77. lxvd2x vs8, o0, A3
  78. lxvd2x vs9, o16, A3
  79. lxvd2x vs10, o32, A3
  80. lxvd2x vs11, 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 vs12, o0, T1
  109. stxvd2x vs13, o16, T1
  110. stxvd2x vs2, o32, T1
  111. stxvd2x vs3, o48, T1
  112. addi T1, T1, 64
  113. stxvd2x vs4, o0, T1
  114. stxvd2x vs5, o16, T1
  115. stxvd2x vs6, o32, T1
  116. stxvd2x vs7, o48, T1
  117. addi T1, T1, 64
  118. stxvd2x vs8, o0, T1
  119. stxvd2x vs9, o16, T1
  120. stxvd2x vs10, o32, T1
  121. stxvd2x vs11, o48, T1
  122. #if defined(_AIX)
  123. ')
  124. #else
  125. .endm
  126. #endif
  127. /**********************************************************************************************
  128. * Macros for N=4 and M=8
  129. **********************************************************************************************/
  130. #if defined(_AIX)
  131. define(`COPY_4x8', `
  132. #else
  133. .macro COPY_4x8
  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=4
  182. **********************************************************************************************/
  183. #if defined(_AIX)
  184. define(`COPY_4x4', `
  185. #else
  186. .macro COPY_4x4
  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=2
  217. **********************************************************************************************/
  218. #if defined(_AIX)
  219. define(`COPY_4x2', `
  220. #else
  221. .macro COPY_4x2
  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=4 and M=1
  243. **********************************************************************************************/
  244. #if defined(_AIX)
  245. define(`COPY_4x1', `
  246. #else
  247. .macro COPY_4x1
  248. #endif
  249. lxsdx vs32, o0, A0
  250. addi A0, A0, 8
  251. lxsdx vs33, o0, A1
  252. addi A1, A1, 8
  253. lxsdx vs34, o0, A2
  254. addi A2, A2, 8
  255. lxsdx vs35, o0, A3
  256. addi A3, A3, 8
  257. mr T1, BO
  258. stxsdx vs32, o0, T1
  259. stxsdx vs33, o8, T1
  260. addi T1, T1, 16
  261. stxsdx vs34, o0, T1
  262. stxsdx vs35, o8, T1
  263. #if defined(_AIX)
  264. ')
  265. #else
  266. .endm
  267. #endif
  268. /**********************************************************************************************
  269. * Macros for N=2 and M=16
  270. **********************************************************************************************/
  271. #if defined(_AIX)
  272. define(`COPY_2x16', `
  273. #else
  274. .macro COPY_2x16
  275. #endif
  276. lxvd2x vs32, o0, A0
  277. lxvd2x vs33, o16, A0
  278. lxvd2x vs34, o32, A0
  279. lxvd2x vs35, o48, A0
  280. addi A0, A0, 64
  281. lxvd2x vs36, o0, A0
  282. lxvd2x vs37, o16, A0
  283. lxvd2x vs38, o32, A0
  284. lxvd2x vs39, o48, A0
  285. addi A0, A0, 64
  286. lxvd2x vs40, o0, A1
  287. lxvd2x vs41, o16, A1
  288. lxvd2x vs42, o32, A1
  289. lxvd2x vs43, o48, A1
  290. addi A1, A1, 64
  291. lxvd2x vs44, o0, A1
  292. lxvd2x vs45, o16, A1
  293. lxvd2x vs46, o32, A1
  294. lxvd2x vs47, o48, A1
  295. addi A1, A1, 64
  296. mr T1, BO
  297. stxvd2x vs32, o0, T1
  298. stxvd2x vs33, o16, T1
  299. stxvd2x vs34, o32, T1
  300. stxvd2x vs35, o48, T1
  301. addi T1, T1, 64
  302. stxvd2x vs36, o0, T1
  303. stxvd2x vs37, o16, T1
  304. stxvd2x vs38, o32, T1
  305. stxvd2x vs39, o48, T1
  306. addi T1, T1, 64
  307. stxvd2x vs40, o0, T1
  308. stxvd2x vs41, o16, T1
  309. stxvd2x vs42, o32, T1
  310. stxvd2x vs43, o48, T1
  311. addi T1, T1, 64
  312. stxvd2x vs44, o0, T1
  313. stxvd2x vs45, o16, T1
  314. stxvd2x vs46, o32, T1
  315. stxvd2x vs47, o48, T1
  316. #if defined(_AIX)
  317. ')
  318. #else
  319. .endm
  320. #endif
  321. /**********************************************************************************************
  322. * Macros for N=2 and M=8
  323. **********************************************************************************************/
  324. #if defined(_AIX)
  325. define(`COPY_2x8', `
  326. #else
  327. .macro COPY_2x8
  328. #endif
  329. lxvd2x vs32, o0, A0
  330. lxvd2x vs33, o16, A0
  331. lxvd2x vs34, o32, A0
  332. lxvd2x vs35, o48, A0
  333. addi A0, A0, 64
  334. lxvd2x vs36, o0, A1
  335. lxvd2x vs37, o16, A1
  336. lxvd2x vs38, o32, A1
  337. lxvd2x vs39, o48, A1
  338. addi A1, A1, 64
  339. mr T1, BO
  340. stxvd2x vs32, o0, T1
  341. stxvd2x vs33, o16, T1
  342. stxvd2x vs34, o32, T1
  343. stxvd2x vs35, o48, T1
  344. addi T1, T1, 64
  345. stxvd2x vs36, o0, T1
  346. stxvd2x vs37, o16, T1
  347. stxvd2x vs38, o32, T1
  348. stxvd2x vs39, o48, T1
  349. #if defined(_AIX)
  350. ')
  351. #else
  352. .endm
  353. #endif
  354. /**********************************************************************************************
  355. * Macros for N=2 and M=4
  356. **********************************************************************************************/
  357. #if defined(_AIX)
  358. define(`COPY_2x4', `
  359. #else
  360. .macro COPY_2x4
  361. #endif
  362. lxvd2x vs32, o0, A0
  363. lxvd2x vs33, o16, A0
  364. addi A0, A0, 32
  365. lxvd2x vs34, o0, A1
  366. lxvd2x vs35, o16, A1
  367. addi A1, A1, 32
  368. mr T1, BO
  369. stxvd2x vs32, o0, T1
  370. stxvd2x vs33, o16, T1
  371. stxvd2x vs34, o32, T1
  372. stxvd2x vs35, o48, T1
  373. #if defined(_AIX)
  374. ')
  375. #else
  376. .endm
  377. #endif
  378. /**********************************************************************************************
  379. * Macros for N=2 and M=2
  380. **********************************************************************************************/
  381. #if defined(_AIX)
  382. define(`COPY_2x2', `
  383. #else
  384. .macro COPY_2x2
  385. #endif
  386. lxvd2x vs32, o0, A0
  387. addi A0, A0, 16
  388. lxvd2x vs33, o0, A1
  389. addi A1, A1, 16
  390. mr T1, BO
  391. stxvd2x vs32, o0, T1
  392. stxvd2x vs33, o16, T1
  393. #if defined(_AIX)
  394. ')
  395. #else
  396. .endm
  397. #endif
  398. /**********************************************************************************************
  399. * Macros for N=2 and M=1
  400. **********************************************************************************************/
  401. #if defined(_AIX)
  402. define(`COPY_2x1', `
  403. #else
  404. .macro COPY_2x1
  405. #endif
  406. lxsdx vs32, o0, A0
  407. addi A0, A0, 8
  408. lxsdx vs33, o0, A1
  409. addi A1, A1, 8
  410. mr T1, BO
  411. stxsdx vs32, o0, T1
  412. stxsdx vs33, o8, T1
  413. #if defined(_AIX)
  414. ')
  415. #else
  416. .endm
  417. #endif
  418. /**********************************************************************************************
  419. * Macros for N=1 and M=16
  420. **********************************************************************************************/
  421. #if defined(_AIX)
  422. define(`COPY_1x16', `
  423. #else
  424. .macro COPY_1x16
  425. #endif
  426. lxvd2x vs32, o0, A0
  427. lxvd2x vs33, o16, A0
  428. lxvd2x vs34, o32, A0
  429. lxvd2x vs35, o48, A0
  430. addi A0, A0, 64
  431. lxvd2x vs36, o0, A0
  432. lxvd2x vs37, o16, A0
  433. lxvd2x vs38, o32, A0
  434. lxvd2x vs39, o48, A0
  435. addi A0, A0, 64
  436. mr T1, BO
  437. stxvd2x vs32, o0, T1
  438. stxvd2x vs33, o16, T1
  439. stxvd2x vs34, o32, T1
  440. stxvd2x vs35, o48, T1
  441. addi T1, T1, 64
  442. stxvd2x vs36, o0, T1
  443. stxvd2x vs37, o16, T1
  444. stxvd2x vs38, o32, T1
  445. stxvd2x vs39, o48, T1
  446. #if defined(_AIX)
  447. ')
  448. #else
  449. .endm
  450. #endif
  451. /**********************************************************************************************
  452. * Macros for N=1 and M=8
  453. **********************************************************************************************/
  454. #if defined(_AIX)
  455. define(`COPY_1x8', `
  456. #else
  457. .macro COPY_1x8
  458. #endif
  459. lxvd2x vs32, o0, A0
  460. lxvd2x vs33, o16, A0
  461. lxvd2x vs34, o32, A0
  462. lxvd2x vs35, o48, A0
  463. addi A0, A0, 64
  464. mr T1, BO
  465. stxvd2x vs32, o0, T1
  466. stxvd2x vs33, o16, T1
  467. stxvd2x vs34, o32, T1
  468. stxvd2x vs35, o48, T1
  469. #if defined(_AIX)
  470. ')
  471. #else
  472. .endm
  473. #endif
  474. /**********************************************************************************************
  475. * Macros for N=1 and M=4
  476. **********************************************************************************************/
  477. #if defined(_AIX)
  478. define(`COPY_1x4', `
  479. #else
  480. .macro COPY_1x4
  481. #endif
  482. lxvd2x vs32, o0, A0
  483. lxvd2x vs33, o16, A0
  484. addi A0, A0, 32
  485. mr T1, BO
  486. stxvd2x vs32, o0, T1
  487. stxvd2x vs33, o16, T1
  488. #if defined(_AIX)
  489. ')
  490. #else
  491. .endm
  492. #endif
  493. /**********************************************************************************************
  494. * Macros for N=1 and M=2
  495. **********************************************************************************************/
  496. #if defined(_AIX)
  497. define(`COPY_1x2', `
  498. #else
  499. .macro COPY_1x2
  500. #endif
  501. lxvd2x vs32, o0, A0
  502. addi A0, A0, 16
  503. mr T1, BO
  504. stxvd2x vs32, o0, T1
  505. #if defined(_AIX)
  506. ')
  507. #else
  508. .endm
  509. #endif
  510. /**********************************************************************************************
  511. * Macros for N=1 and M=1
  512. **********************************************************************************************/
  513. #if defined(_AIX)
  514. define(`COPY_1x1', `
  515. #else
  516. .macro COPY_1x1
  517. #endif
  518. lxsdx vs32, o0, A0
  519. addi A0, A0, 8
  520. mr T1, BO
  521. stxsdx vs32, o0, T1
  522. #if defined(_AIX)
  523. ')
  524. #else
  525. .endm
  526. #endif