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.

sgemm_tcopy_macros_16_power8.S 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536
  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. lxvw4x vs32, o0, A0
  43. lxvw4x vs33, o16, A0
  44. lxvw4x vs34, o32, A0
  45. lxvw4x vs35, o48, A0
  46. lxvw4x vs36, o0, A1
  47. lxvw4x vs37, o16, A1
  48. lxvw4x vs38, o32, A1
  49. lxvw4x vs39, o48, A1
  50. lxvw4x vs40, o0, A2
  51. lxvw4x vs41, o16, A2
  52. lxvw4x vs42, o32, A2
  53. lxvw4x vs43, o48, A2
  54. lxvw4x vs44, o0, A3
  55. lxvw4x vs45, o16, A3
  56. lxvw4x vs46, o32, A3
  57. lxvw4x vs47, o48, A3
  58. mr T1, BO
  59. stxvw4x vs32, o0, T1
  60. stxvw4x vs33, o16, T1
  61. stxvw4x vs34, o32, T1
  62. stxvw4x vs35, o48, T1
  63. addi T1, T1, 64
  64. stxvw4x vs36, o0, T1
  65. stxvw4x vs37, o16, T1
  66. stxvw4x vs38, o32, T1
  67. stxvw4x vs39, o48, T1
  68. addi T1, T1, 64
  69. stxvw4x vs40, o0, T1
  70. stxvw4x vs41, o16, T1
  71. stxvw4x vs42, o32, T1
  72. stxvw4x vs43, o48, T1
  73. addi T1, T1, 64
  74. stxvw4x vs44, o0, T1
  75. stxvw4x vs45, o16, T1
  76. stxvw4x vs46, o32, T1
  77. stxvw4x vs47, o48, T1
  78. #if defined(_AIX)
  79. ')
  80. #else
  81. .endm
  82. #endif
  83. /**********************************************************************************************
  84. * Macros for N=4 and M=8
  85. **********************************************************************************************/
  86. #if defined(_AIX)
  87. define(`COPY_4x8', `
  88. #else
  89. .macro COPY_4x8
  90. #endif
  91. lxvw4x vs32, o0, A0
  92. lxvw4x vs33, o16, A0
  93. lxvw4x vs34, o0, A1
  94. lxvw4x vs35, o16, A1
  95. lxvw4x vs36, o0, A2
  96. lxvw4x vs37, o16, A2
  97. lxvw4x vs38, o0, A3
  98. lxvw4x vs39, o16, A3
  99. mr T1, BO
  100. stxvw4x vs32, o0, T1
  101. stxvw4x vs33, o16, T1
  102. stxvw4x vs34, o32, T1
  103. stxvw4x vs35, o48, T1
  104. addi T1, T1, 64
  105. stxvw4x vs36, o0, T1
  106. stxvw4x vs37, o16, T1
  107. stxvw4x vs38, o32, T1
  108. stxvw4x vs39, o48, T1
  109. #if defined(_AIX)
  110. ')
  111. #else
  112. .endm
  113. #endif
  114. /**********************************************************************************************
  115. * Macros for N=4 and M=4
  116. **********************************************************************************************/
  117. #if defined(_AIX)
  118. define(`COPY_4x4', `
  119. #else
  120. .macro COPY_4x4
  121. #endif
  122. lxvw4x vs32, o0, A0
  123. lxvw4x vs33, o0, A1
  124. lxvw4x vs34, o0, A2
  125. lxvw4x vs35, o0, A3
  126. mr T1, BO
  127. stxvw4x vs32, o0, T1
  128. stxvw4x vs33, o16, T1
  129. stxvw4x vs34, o32, T1
  130. stxvw4x vs35, o48, T1
  131. #if defined(_AIX)
  132. ')
  133. #else
  134. .endm
  135. #endif
  136. /**********************************************************************************************
  137. * Macros for N=4 and M=2
  138. **********************************************************************************************/
  139. #if defined(_AIX)
  140. define(`COPY_4x2', `
  141. #else
  142. .macro COPY_4x2
  143. #endif
  144. lxsspx vs32, o0, A0
  145. lxsspx vs33, o4, A0
  146. lxsspx vs34, o0, A1
  147. lxsspx vs35, o4, A1
  148. lxsspx vs36, o0, A2
  149. lxsspx vs37, o4, A2
  150. lxsspx vs38, o0, A3
  151. lxsspx vs39, o4, A3
  152. mr T1, BO
  153. stxsspx vs32, o0, T1
  154. stxsspx vs33, o4, T1
  155. addi T1, T1, 8
  156. stxsspx vs34, o0, T1
  157. stxsspx vs35, o4, T1
  158. addi T1, T1, 8
  159. stxsspx vs36, o0, T1
  160. stxsspx vs37, o4, T1
  161. addi T1, T1, 8
  162. stxsspx vs38, o0, T1
  163. stxsspx vs39, o4, T1
  164. #if defined(_AIX)
  165. ')
  166. #else
  167. .endm
  168. #endif
  169. /**********************************************************************************************
  170. * Macros for N=4 and M=1
  171. **********************************************************************************************/
  172. #if defined(_AIX)
  173. define(`COPY_4x1', `
  174. #else
  175. .macro COPY_4x1
  176. #endif
  177. lxsspx vs32, o0, A0
  178. lxsspx vs33, o0, A1
  179. lxsspx vs34, o0, A2
  180. lxsspx vs35, o0, A3
  181. mr T1, BO
  182. stxsspx vs32, o0, T1
  183. stxsspx vs33, o4, T1
  184. addi T1, T1, 8
  185. stxsspx vs34, o0, T1
  186. stxsspx vs35, o4, T1
  187. #if defined(_AIX)
  188. ')
  189. #else
  190. .endm
  191. #endif
  192. /**********************************************************************************************
  193. * Macros for N=2 and M=16
  194. **********************************************************************************************/
  195. #if defined(_AIX)
  196. define(`COPY_2x16', `
  197. #else
  198. .macro COPY_2x16
  199. #endif
  200. lxvw4x vs32, o0, A0
  201. lxvw4x vs33, o16, A0
  202. lxvw4x vs34, o32, A0
  203. lxvw4x vs35, o48, A0
  204. lxvw4x vs36, o0, A1
  205. lxvw4x vs37, o16, A1
  206. lxvw4x vs38, o32, A1
  207. lxvw4x vs39, o48, A1
  208. mr T1, BO
  209. stxvw4x vs32, o0, T1
  210. stxvw4x vs33, o16, T1
  211. stxvw4x vs34, o32, T1
  212. stxvw4x vs35, o48, T1
  213. addi T1, T1, 64
  214. stxvw4x vs36, o0, T1
  215. stxvw4x vs37, o16, T1
  216. stxvw4x vs38, o32, T1
  217. stxvw4x vs39, o48, T1
  218. #if defined(_AIX)
  219. ')
  220. #else
  221. .endm
  222. #endif
  223. /**********************************************************************************************
  224. * Macros for N=2 and M=8
  225. **********************************************************************************************/
  226. #if defined(_AIX)
  227. define(`COPY_2x8', `
  228. #else
  229. .macro COPY_2x8
  230. #endif
  231. lxvw4x vs32, o0, A0
  232. lxvw4x vs33, o16, A0
  233. lxvw4x vs34, o0, A1
  234. lxvw4x vs35, o16, A1
  235. mr T1, BO
  236. stxvw4x vs32, o0, T1
  237. stxvw4x vs33, o16, T1
  238. stxvw4x vs34, o32, T1
  239. stxvw4x vs35, o48, T1
  240. #if defined(_AIX)
  241. ')
  242. #else
  243. .endm
  244. #endif
  245. /**********************************************************************************************
  246. * Macros for N=2 and M=4
  247. **********************************************************************************************/
  248. #if defined(_AIX)
  249. define(`COPY_2x4', `
  250. #else
  251. .macro COPY_2x4
  252. #endif
  253. lxvw4x vs32, o0, A0
  254. lxvw4x vs33, o0, A1
  255. mr T1, BO
  256. stxvw4x vs32, o0, T1
  257. stxvw4x vs33, o16, T1
  258. #if defined(_AIX)
  259. ')
  260. #else
  261. .endm
  262. #endif
  263. /**********************************************************************************************
  264. * Macros for N=2 and M=2
  265. **********************************************************************************************/
  266. #if defined(_AIX)
  267. define(`COPY_2x2', `
  268. #else
  269. .macro COPY_2x2
  270. #endif
  271. lxsspx vs32, o0, A0
  272. lxsspx vs33, o4, A0
  273. lxsspx vs34, o0, A1
  274. lxsspx vs35, o4, A1
  275. mr T1, BO
  276. stxsspx vs32, o0, T1
  277. stxsspx vs33, o4, T1
  278. addi T1, T1, 8
  279. stxsspx vs34, o0, T1
  280. stxsspx vs35, o4, T1
  281. #if defined(_AIX)
  282. ')
  283. #else
  284. .endm
  285. #endif
  286. /**********************************************************************************************
  287. * Macros for N=2 and M=1
  288. **********************************************************************************************/
  289. #if defined(_AIX)
  290. define(`COPY_2x1', `
  291. #else
  292. .macro COPY_2x1
  293. #endif
  294. lxsspx vs32, o0, A0
  295. lxsspx vs33, o0, A1
  296. mr T1, BO
  297. stxsspx vs32, o0, T1
  298. stxsspx vs33, o4, T1
  299. #if defined(_AIX)
  300. ')
  301. #else
  302. .endm
  303. #endif
  304. /**********************************************************************************************
  305. * Macros for N=1 and M=16
  306. **********************************************************************************************/
  307. #if defined(_AIX)
  308. define(`COPY_1x16', `
  309. #else
  310. .macro COPY_1x16
  311. #endif
  312. lxvw4x vs32, o0, A0
  313. lxvw4x vs33, o16, A0
  314. lxvw4x vs34, o32, A0
  315. lxvw4x vs35, o48, A0
  316. mr T1, BO
  317. stxvw4x vs32, o0, T1
  318. stxvw4x vs33, o16, T1
  319. stxvw4x vs34, o32, T1
  320. stxvw4x vs35, o48, T1
  321. #if defined(_AIX)
  322. ')
  323. #else
  324. .endm
  325. #endif
  326. /**********************************************************************************************
  327. * Macros for N=1 and M=8
  328. **********************************************************************************************/
  329. #if defined(_AIX)
  330. define(`COPY_1x8', `
  331. #else
  332. .macro COPY_1x8
  333. #endif
  334. lxvw4x vs32, o0, A0
  335. lxvw4x vs33, o16, A0
  336. mr T1, BO
  337. stxvw4x vs32, o0, T1
  338. stxvw4x vs33, o16, T1
  339. #if defined(_AIX)
  340. ')
  341. #else
  342. .endm
  343. #endif
  344. /**********************************************************************************************
  345. * Macros for N=1 and M=4
  346. **********************************************************************************************/
  347. #if defined(_AIX)
  348. define(`COPY_1x4', `
  349. #else
  350. .macro COPY_1x4
  351. #endif
  352. lxvw4x vs32, o0, A0
  353. mr T1, BO
  354. stxvw4x vs32, o0, T1
  355. #if defined(_AIX)
  356. ')
  357. #else
  358. .endm
  359. #endif
  360. /**********************************************************************************************
  361. * Macros for N=1 and M=2
  362. **********************************************************************************************/
  363. #if defined(_AIX)
  364. define(`COPY_1x2', `
  365. #else
  366. .macro COPY_1x2
  367. #endif
  368. lxsspx vs32, o0, A0
  369. lxsspx vs33, o4, A0
  370. mr T1, BO
  371. stxsspx vs32, o0, T1
  372. stxsspx vs33, o4, T1
  373. #if defined(_AIX)
  374. ')
  375. #else
  376. .endm
  377. #endif
  378. /**********************************************************************************************
  379. * Macros for N=1 and M=1
  380. **********************************************************************************************/
  381. #if defined(_AIX)
  382. define(`COPY_1x1', `
  383. #else
  384. .macro COPY_1x1
  385. #endif
  386. lxsspx vs32, o0, A0
  387. mr T1, BO
  388. stxsspx vs32, o0, T1
  389. #if defined(_AIX)
  390. ')
  391. #else
  392. .endm
  393. #endif