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_kernel_16x4_power8.S 10 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416
  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 320
  81. #define STACKSIZE 512
  82. #define ALPHA_SP 296+192(SP)
  83. #define FZERO 304+192(SP)
  84. #else
  85. #define STACKSIZE 240
  86. #define STACKSIZE 440
  87. #define ALPHA_SP 224+200(SP)
  88. #define FZERO 232+200(SP)
  89. #endif
  90. #define M r3
  91. #define N r4
  92. #define K r5
  93. #if defined(linux) || defined(__FreeBSD__)
  94. #ifndef __64BIT__
  95. #define A r6
  96. #define B r7
  97. #define C r8
  98. #define LDC r9
  99. #define OFFSET r10
  100. #else
  101. #define A r7
  102. #define B r8
  103. #define C r9
  104. #define LDC r10
  105. #define OFFSET r6
  106. #endif
  107. #endif
  108. #if defined(_AIX) || defined(__APPLE__)
  109. #if !defined(__64BIT__) && defined(DOUBLE)
  110. #define A r8
  111. #define B r9
  112. #define C r10
  113. #define LDC r7
  114. #define OFFSET r6
  115. #else
  116. #define A r7
  117. #define B r8
  118. #define C r9
  119. #define LDC r10
  120. #define OFFSET r6
  121. #endif
  122. #endif
  123. #define alpha_r vs18
  124. #define o0 0
  125. #define T4 r12
  126. #define T3 r11
  127. #define o40 r12
  128. #define o56 r11
  129. #define o112 r14
  130. #define o8 r15
  131. #define o24 r16
  132. #define o64 r17
  133. #define L r18
  134. #define T1 r19
  135. #define o80 r20
  136. #define o96 r21
  137. #define I r22
  138. #define J r23
  139. #define AO r24
  140. #define BO r25
  141. #define CO r26
  142. #define o16 r27
  143. #define o32 r28
  144. #define o48 r29
  145. #define PRE r30
  146. #define T2 r31
  147. #include "dgemm_macros_16x4_power8.S"
  148. #ifndef NEEDPARAM
  149. PROLOGUE
  150. PROFCODE
  151. addi SP, SP, -STACKSIZE
  152. li r0, 0
  153. stfd f14, 0(SP)
  154. stfd f15, 8(SP)
  155. stfd f16, 16(SP)
  156. stfd f17, 24(SP)
  157. stfd f18, 32(SP)
  158. stfd f19, 40(SP)
  159. stfd f20, 48(SP)
  160. stfd f21, 56(SP)
  161. stfd f22, 64(SP)
  162. stfd f23, 72(SP)
  163. stfd f24, 80(SP)
  164. stfd f25, 88(SP)
  165. stfd f26, 96(SP)
  166. stfd f27, 104(SP)
  167. stfd f28, 112(SP)
  168. stfd f29, 120(SP)
  169. stfd f30, 128(SP)
  170. stfd f31, 136(SP)
  171. #ifdef __64BIT__
  172. std r31, 144(SP)
  173. std r30, 152(SP)
  174. std r29, 160(SP)
  175. std r28, 168(SP)
  176. std r27, 176(SP)
  177. std r26, 184(SP)
  178. std r25, 192(SP)
  179. std r24, 200(SP)
  180. std r23, 208(SP)
  181. std r22, 216(SP)
  182. std r21, 224(SP)
  183. std r20, 232(SP)
  184. std r19, 240(SP)
  185. std r18, 248(SP)
  186. std r17, 256(SP)
  187. std r16, 264(SP)
  188. std r15, 272(SP)
  189. std r14, 280(SP)
  190. addi r11,SP,288
  191. #else
  192. stw r31, 144(SP)
  193. stw r30, 148(SP)
  194. stw r29, 152(SP)
  195. stw r28, 156(SP)
  196. stw r27, 160(SP)
  197. stw r26, 164(SP)
  198. stw r25, 168(SP)
  199. stw r24, 172(SP)
  200. stw r23, 176(SP)
  201. stw r22, 180(SP)
  202. stw r21, 184(SP)
  203. stw r20, 188(SP)
  204. stw r19, 192(SP)
  205. stw r18, 196(SP)
  206. stw r17, 200(SP)
  207. stw r16, 204(SP)
  208. stw r15, 208(SP)
  209. stw r14, 212(SP)
  210. addi r11,SP,224
  211. #endif
  212. stvx v20, r11,r0
  213. addi r11,r11,16
  214. stvx v21, r11,r0
  215. addi r11,r11,16
  216. stvx v22, r11,r0
  217. addi r11,r11,16
  218. stvx v23, r11,r0
  219. addi r11,r11,16
  220. stvx v24, r11,r0
  221. addi r11,r11,16
  222. stvx v25, r11,r0
  223. addi r11,r11,16
  224. stvx v26, r11,r0
  225. addi r11,r11,16
  226. stvx v27, r11,r0
  227. addi r11,r11,16
  228. stvx v28, r11,r0
  229. addi r11,r11,16
  230. stvx v29, r11,r0
  231. addi r11,r11,16
  232. stvx v30, r11,r0
  233. addi r11,r11,16
  234. stvx v31, r11,r0
  235. li r11,0
  236. stfd f1, ALPHA_SP
  237. stw r0, FZERO
  238. #if defined(_AIX) || defined(__APPLE__)
  239. #if !defined(__64BIT__) && defined(DOUBLE)
  240. lwz LDC, FRAMESLOT(0) + STACKSIZE(SP)
  241. #endif
  242. #endif
  243. slwi LDC, LDC, BASE_SHIFT
  244. #if defined(TRMMKERNEL)
  245. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  246. ld OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  247. #endif
  248. #if defined(_AIX) || defined(__APPLE__)
  249. #ifdef __64BIT__
  250. ld OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  251. #else
  252. #ifdef DOUBLE
  253. lwz OFFSET, FRAMESLOT(1) + STACKSIZE(SP)
  254. #else
  255. lwz OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  256. #endif
  257. #endif
  258. #endif
  259. #endif
  260. cmpwi cr0, M, 0
  261. ble .L999_H1
  262. cmpwi cr0, N, 0
  263. ble .L999_H1
  264. cmpwi cr0, K, 0
  265. ble .L999_H1
  266. #ifdef __64BIT__
  267. addi T1, SP, 296+192
  268. #else
  269. addi T1, SP, 224+200
  270. #endif
  271. li PRE, 384
  272. li o8 , 8
  273. li o16, 16
  274. li o24, 24
  275. li o32, 32
  276. li o48, 48
  277. li o64, 64
  278. li o80, 80
  279. li o96, 96
  280. li o112, 112
  281. lxvdsx alpha_r, 0, T1
  282. #include "dgemm_logic_16x4_power8.S"
  283. .L999:
  284. addi r3, 0, 0
  285. lfd f14, 0(SP)
  286. lfd f15, 8(SP)
  287. lfd f16, 16(SP)
  288. lfd f17, 24(SP)
  289. lfd f18, 32(SP)
  290. lfd f19, 40(SP)
  291. lfd f20, 48(SP)
  292. lfd f21, 56(SP)
  293. lfd f22, 64(SP)
  294. lfd f23, 72(SP)
  295. lfd f24, 80(SP)
  296. lfd f25, 88(SP)
  297. lfd f26, 96(SP)
  298. lfd f27, 104(SP)
  299. lfd f28, 112(SP)
  300. lfd f29, 120(SP)
  301. lfd f30, 128(SP)
  302. lfd f31, 136(SP)
  303. #ifdef __64BIT__
  304. ld r31, 144(SP)
  305. ld r30, 152(SP)
  306. ld r29, 160(SP)
  307. ld r28, 168(SP)
  308. ld r27, 176(SP)
  309. ld r26, 184(SP)
  310. ld r25, 192(SP)
  311. ld r24, 200(SP)
  312. ld r23, 208(SP)
  313. ld r22, 216(SP)
  314. ld r21, 224(SP)
  315. ld r20, 232(SP)
  316. ld r19, 240(SP)
  317. ld r18, 248(SP)
  318. ld r17, 256(SP)
  319. ld r16, 264(SP)
  320. ld r15, 272(SP)
  321. ld r14, 280(SP)
  322. addi r11,SP,288
  323. #else
  324. lwz r31, 144(SP)
  325. lwz r30, 148(SP)
  326. lwz r29, 152(SP)
  327. lwz r28, 156(SP)
  328. lwz r27, 160(SP)
  329. lwz r26, 164(SP)
  330. lwz r25, 168(SP)
  331. lwz r24, 172(SP)
  332. lwz r23, 176(SP)
  333. lwz r22, 180(SP)
  334. lwz r21, 184(SP)
  335. lwz r20, 188(SP)
  336. lwz r19, 192(SP)
  337. lwz r18, 196(SP)
  338. lwz r17, 200(SP)
  339. lwz r16, 204(SP)
  340. lwz r15, 208(SP)
  341. lwz r14, 212(SP)
  342. addi r11,SP,224
  343. #endif
  344. lvx v20, r11,r3
  345. addi r11,r11,16
  346. lvx v21, r11,r3
  347. addi r11,r11,16
  348. lvx v22, r11,r3
  349. addi r11,r11,16
  350. lvx v23, r11,r3
  351. addi r11,r11,16
  352. lvx v24, r11,r3
  353. addi r11,r11,16
  354. lvx v25, r11,r3
  355. addi r11,r11,16
  356. lvx v26, r11,r3
  357. addi r11,r11,16
  358. lvx v27, r11,r3
  359. addi r11,r11,16
  360. lvx v28, r11,r3
  361. addi r11,r11,16
  362. lvx v29, r11,r3
  363. addi r11,r11,16
  364. lvx v30, r11,r3
  365. addi r11,r11,16
  366. lvx v31, r11,r3
  367. li r11,0
  368. addi SP, SP, STACKSIZE
  369. blr
  370. EPILOGUE
  371. #endif