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_kernel_16x8_power8.S 11 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426
  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. /* Copyright 2009, 2010 The University of Texas at Austin. */
  36. /* All rights reserved. */
  37. /* */
  38. /* Redistribution and use in source and binary forms, with or */
  39. /* without modification, are permitted provided that the following */
  40. /* conditions are met: */
  41. /* */
  42. /* 1. Redistributions of source code must retain the above */
  43. /* copyright notice, this list of conditions and the following */
  44. /* disclaimer. */
  45. /* */
  46. /* 2. Redistributions in binary form must reproduce the above */
  47. /* copyright notice, this list of conditions and the following */
  48. /* disclaimer in the documentation and/or other materials */
  49. /* provided with the distribution. */
  50. /* */
  51. /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
  52. /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
  53. /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
  54. /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
  55. /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
  56. /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
  57. /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
  58. /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
  59. /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
  60. /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
  61. /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
  62. /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
  63. /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
  64. /* POSSIBILITY OF SUCH DAMAGE. */
  65. /* */
  66. /* The views and conclusions contained in the software and */
  67. /* documentation are those of the authors and should not be */
  68. /* interpreted as representing official policies, either expressed */
  69. /* or implied, of The University of Texas at Austin. */
  70. /*********************************************************************/
  71. #define ASSEMBLER
  72. #include "common.h"
  73. #include "def_vsx.h"
  74. #ifndef __64BIT__
  75. #define LOAD lwz
  76. #else
  77. #define LOAD ld
  78. #endif
  79. #ifdef __64BIT__
  80. #define STACKSIZE 32752
  81. #define ALPHA_SP 296+192(SP)
  82. #define FZERO 304+192(SP)
  83. #else
  84. #define STACKSIZE 440
  85. #define ALPHA_SP 224+200(SP)
  86. #define FZERO 232+200(SP)
  87. #endif
  88. #define M r3
  89. #define N r4
  90. #define K r5
  91. #if defined(linux) || defined(__FreeBSD__)
  92. #ifndef __64BIT__
  93. #define A r6
  94. #define B r7
  95. #define C r8
  96. #define LDC r9
  97. #define OFFSET r10
  98. #else
  99. #define A r7
  100. #define B r8
  101. #define C r9
  102. #define LDC r10
  103. #define OFFSET r6
  104. #endif
  105. #endif
  106. #if defined(_AIX) || defined(__APPLE__)
  107. #if !defined(__64BIT__) && defined(DOUBLE)
  108. #define A r8
  109. #define B r9
  110. #define C r10
  111. #define LDC r7
  112. #define OFFSET r6
  113. #else
  114. #define A r7
  115. #define B r8
  116. #define C r9
  117. #define LDC r10
  118. #define OFFSET r6
  119. #endif
  120. #endif
  121. #define alpha_r vs30
  122. #define alpha_vr vs31
  123. #define o0 0
  124. #define VECSAVE r11
  125. #define FRAMEPOINTER r12
  126. #define BBUFFER r14
  127. #define o4 r15
  128. #define o12 r16
  129. #define o8 r17
  130. #define L r18
  131. #define T1 r19
  132. #define KK r20
  133. #define BBO r21
  134. #define I r22
  135. #define J r23
  136. #define AO r24
  137. #define BO r25
  138. #define CO r26
  139. #define o16 r27
  140. #define o32 r28
  141. #define o48 r29
  142. #define PRE r30
  143. #define T2 r31
  144. #include "sgemm_macros_16x8_power8.S"
  145. #ifndef NEEDPARAM
  146. PROLOGUE
  147. PROFCODE
  148. mr FRAMEPOINTER, SP
  149. addi SP, SP, -STACKSIZE
  150. addi SP, SP, -STACKSIZE
  151. addi SP, SP, -STACKSIZE
  152. addi SP, SP, -STACKSIZE
  153. li r0, 0
  154. stfd f14, 0(SP)
  155. stfd f15, 8(SP)
  156. stfd f16, 16(SP)
  157. stfd f17, 24(SP)
  158. stfd f18, 32(SP)
  159. stfd f19, 40(SP)
  160. stfd f20, 48(SP)
  161. stfd f21, 56(SP)
  162. stfd f22, 64(SP)
  163. stfd f23, 72(SP)
  164. stfd f24, 80(SP)
  165. stfd f25, 88(SP)
  166. stfd f26, 96(SP)
  167. stfd f27, 104(SP)
  168. stfd f28, 112(SP)
  169. stfd f29, 120(SP)
  170. stfd f30, 128(SP)
  171. stfd f31, 136(SP)
  172. #ifdef __64BIT__
  173. std r31, 144(SP)
  174. std r30, 152(SP)
  175. std r29, 160(SP)
  176. std r28, 168(SP)
  177. std r27, 176(SP)
  178. std r26, 184(SP)
  179. std r25, 192(SP)
  180. std r24, 200(SP)
  181. std r23, 208(SP)
  182. std r22, 216(SP)
  183. std r21, 224(SP)
  184. std r20, 232(SP)
  185. std r19, 240(SP)
  186. std r18, 248(SP)
  187. std r17, 256(SP)
  188. std r16, 264(SP)
  189. std r15, 272(SP)
  190. std r14, 280(SP)
  191. addi r11, SP, 288
  192. #else
  193. stw r31, 144(SP)
  194. stw r30, 148(SP)
  195. stw r29, 152(SP)
  196. stw r28, 156(SP)
  197. stw r27, 160(SP)
  198. stw r26, 164(SP)
  199. stw r25, 168(SP)
  200. stw r24, 172(SP)
  201. stw r23, 176(SP)
  202. stw r22, 180(SP)
  203. stw r21, 184(SP)
  204. stw r20, 188(SP)
  205. stw r19, 192(SP)
  206. stw r18, 196(SP)
  207. stw r17, 200(SP)
  208. stw r16, 204(SP)
  209. stw r15, 208(SP)
  210. stw r14, 212(SP)
  211. addi r11, SP, 224
  212. #endif
  213. stvx v20, r11, r0
  214. addi r11, r11, 16
  215. stvx v21, r11, r0
  216. addi r11, r11, 16
  217. stvx v22, r11, r0
  218. addi r11, r11, 16
  219. stvx v23, r11, r0
  220. addi r11, r11, 16
  221. stvx v24, r11, r0
  222. addi r11, r11, 16
  223. stvx v25, r11, r0
  224. addi r11, r11, 16
  225. stvx v26, r11, r0
  226. addi r11, r11, 16
  227. stvx v27, r11, r0
  228. addi r11, r11, 16
  229. stvx v28, r11, r0
  230. addi r11, r11, 16
  231. stvx v29, r11, r0
  232. addi r11, r11, 16
  233. stvx v30, r11, r0
  234. addi r11, r11, 16
  235. stvx v31, r11, r0
  236. li r11,0
  237. // stfd f1, ALPHA_SP
  238. // stw r0, FZERO
  239. #if defined(_AIX) || defined(__APPLE__)
  240. #if !defined(__64BIT__) && defined(DOUBLE)
  241. lwz LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
  242. #endif
  243. #endif
  244. slwi LDC, LDC, 2
  245. #if defined(TRMMKERNEL)
  246. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  247. ld OFFSET, FRAMESLOT(0) + 0(FRAMEPOINTER)
  248. #endif
  249. #if defined(_AIX) || defined(__APPLE__)
  250. #ifdef __64BIT__
  251. ld OFFSET, FRAMESLOT(0) + 0(FRAMEPOINTER)
  252. #else
  253. #ifdef DOUBLE
  254. lwz OFFSET, FRAMESLOT(1) + 0(FRAMEPOINTER)
  255. #else
  256. lwz OFFSET, FRAMESLOT(0) + 0(FRAMEPOINTER)
  257. #endif
  258. #endif
  259. #endif
  260. #endif
  261. cmpwi cr0, M, 0
  262. ble L999_H1
  263. cmpwi cr0, N, 0
  264. ble L999_H1
  265. cmpwi cr0, K, 0
  266. ble L999_H1
  267. li PRE, 256
  268. li o4 , 4
  269. li o8 , 8
  270. li o12, 12
  271. li o16, 16
  272. li o32, 32
  273. li o48, 48
  274. addi BBUFFER, SP, 512+4096
  275. li T1, -4096
  276. and BBUFFER, BBUFFER, T1
  277. addi T1, SP, 300+192
  278. stxsspx f1, o0 , T1
  279. stxsspx f1, o4 , T1
  280. stxsspx f1, o8 , T1
  281. stxsspx f1, o12 , T1
  282. lxsspx alpha_r, o0, T1
  283. lxvw4x alpha_vr, o0, T1
  284. #include "sgemm_logic_16x8_power8.S"
  285. L999:
  286. addi r3, 0, 0
  287. lfd f14, 0(SP)
  288. lfd f15, 8(SP)
  289. lfd f16, 16(SP)
  290. lfd f17, 24(SP)
  291. lfd f18, 32(SP)
  292. lfd f19, 40(SP)
  293. lfd f20, 48(SP)
  294. lfd f21, 56(SP)
  295. lfd f22, 64(SP)
  296. lfd f23, 72(SP)
  297. lfd f24, 80(SP)
  298. lfd f25, 88(SP)
  299. lfd f26, 96(SP)
  300. lfd f27, 104(SP)
  301. lfd f28, 112(SP)
  302. lfd f29, 120(SP)
  303. lfd f30, 128(SP)
  304. lfd f31, 136(SP)
  305. #ifdef __64BIT__
  306. ld r31, 144(SP)
  307. ld r30, 152(SP)
  308. ld r29, 160(SP)
  309. ld r28, 168(SP)
  310. ld r27, 176(SP)
  311. ld r26, 184(SP)
  312. ld r25, 192(SP)
  313. ld r24, 200(SP)
  314. ld r23, 208(SP)
  315. ld r22, 216(SP)
  316. ld r21, 224(SP)
  317. ld r20, 232(SP)
  318. ld r19, 240(SP)
  319. ld r18, 248(SP)
  320. ld r17, 256(SP)
  321. ld r16, 264(SP)
  322. ld r15, 272(SP)
  323. ld r14, 280(SP)
  324. addi r11, SP, 288
  325. #else
  326. lwz r31, 144(SP)
  327. lwz r30, 148(SP)
  328. lwz r29, 152(SP)
  329. lwz r28, 156(SP)
  330. lwz r27, 160(SP)
  331. lwz r26, 164(SP)
  332. lwz r25, 168(SP)
  333. lwz r24, 172(SP)
  334. lwz r23, 176(SP)
  335. lwz r22, 180(SP)
  336. lwz r21, 184(SP)
  337. lwz r20, 188(SP)
  338. lwz r19, 192(SP)
  339. lwz r18, 196(SP)
  340. lwz r17, 200(SP)
  341. lwz r16, 204(SP)
  342. lwz r15, 208(SP)
  343. lwz r14, 212(SP)
  344. addi r11, SP, 224
  345. #endif
  346. lvx v20, r11, r3
  347. addi r11, r11, 16
  348. lvx v21, r11, r3
  349. addi r11, r11, 16
  350. lvx v22, r11, r3
  351. addi r11, r11, 16
  352. lvx v23, r11, r3
  353. addi r11, r11, 16
  354. lvx v24, r11, r3
  355. addi r11, r11, 16
  356. lvx v25, r11, r3
  357. addi r11, r11, 16
  358. lvx v26, r11, r3
  359. addi r11, r11, 16
  360. lvx v27, r11, r3
  361. addi r11, r11, 16
  362. lvx v28, r11, r3
  363. addi r11, r11, 16
  364. lvx v29, r11, r3
  365. addi r11, r11, 16
  366. lvx v30, r11, r3
  367. addi r11, r11, 16
  368. lvx v31, r11, r3
  369. li r11, 0
  370. addi SP, SP, STACKSIZE
  371. addi SP, SP, STACKSIZE
  372. addi SP, SP, STACKSIZE
  373. addi SP, SP, STACKSIZE
  374. blr
  375. EPILOGUE
  376. #endif