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.

cgemm_kernel_8x4_power8.S 12 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462
  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. * 2016/04/04 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 32196
  81. #define ALPHA_R_SP 296+196(SP)
  82. #define ALPHA_I_SP 304+196(SP)
  83. #define FZERO 312+196(SP)
  84. #else
  85. #define STACKSIZE 456
  86. #define ALPHA_R_SP 224+200(SP)
  87. #define ALPHA_I_SP 232+200(SP)
  88. #define FZERO 240+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 r8
  102. #define B r9
  103. #define C r10
  104. #define LDC r6
  105. #define OFFSET r7
  106. #endif
  107. #endif
  108. #if defined(_AIX) || defined(__APPLE__)
  109. #if !defined(__64BIT__) && defined(DOUBLE)
  110. #define A r10
  111. #define B r6
  112. #define C r7
  113. #define LDC r8
  114. #define OFFSET r9
  115. #else
  116. #define A r8
  117. #define B r9
  118. #define C r10
  119. #define LDC r6
  120. #define OFFSET r7
  121. #endif
  122. #endif
  123. #define o0 0
  124. #define alpha_dr vs28
  125. #define alpha_di vs29
  126. #define alpha_sr vs30
  127. #define alpha_si vs31
  128. #define FRAMEPOINTER r12
  129. #define VECSAVE r11
  130. #define BBUFFER r14
  131. #define L r15
  132. #define o12 r16
  133. #define o4 r17
  134. #define T2 r19
  135. #define BBO r20
  136. #define o8 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 T1 r31
  147. #ifndef NEEDPARAM
  148. PROLOGUE
  149. PROFCODE
  150. mr FRAMEPOINTER, SP
  151. addi SP, SP, -STACKSIZE
  152. addi SP, SP, -STACKSIZE
  153. addi SP, SP, -STACKSIZE
  154. addi SP, SP, -STACKSIZE
  155. li r0, 0
  156. stfd f14, 0(SP)
  157. stfd f15, 8(SP)
  158. stfd f16, 16(SP)
  159. stfd f17, 24(SP)
  160. stfd f18, 32(SP)
  161. stfd f19, 40(SP)
  162. stfd f20, 48(SP)
  163. stfd f21, 56(SP)
  164. stfd f22, 64(SP)
  165. stfd f23, 72(SP)
  166. stfd f24, 80(SP)
  167. stfd f25, 88(SP)
  168. stfd f26, 96(SP)
  169. stfd f27, 104(SP)
  170. stfd f28, 112(SP)
  171. stfd f29, 120(SP)
  172. stfd f30, 128(SP)
  173. stfd f31, 136(SP)
  174. #ifdef __64BIT__
  175. std r31, 144(SP)
  176. std r30, 152(SP)
  177. std r29, 160(SP)
  178. std r28, 168(SP)
  179. std r27, 176(SP)
  180. std r26, 184(SP)
  181. std r25, 192(SP)
  182. std r24, 200(SP)
  183. std r23, 208(SP)
  184. std r22, 216(SP)
  185. std r21, 224(SP)
  186. std r20, 232(SP)
  187. std r19, 240(SP)
  188. std r18, 248(SP)
  189. std r17, 256(SP)
  190. std r16, 264(SP)
  191. std r15, 272(SP)
  192. std r14, 280(SP)
  193. addi r11, SP, 288
  194. #else
  195. stw r31, 144(SP)
  196. stw r30, 148(SP)
  197. stw r29, 152(SP)
  198. stw r28, 156(SP)
  199. stw r27, 160(SP)
  200. stw r26, 164(SP)
  201. stw r25, 168(SP)
  202. stw r24, 172(SP)
  203. stw r23, 176(SP)
  204. stw r22, 180(SP)
  205. stw r21, 184(SP)
  206. stw r20, 188(SP)
  207. stw r19, 192(SP)
  208. stw r18, 196(SP)
  209. stw r17, 200(SP)
  210. stw r16, 204(SP)
  211. stw r15, 208(SP)
  212. stw r14, 212(SP)
  213. addi r11, SP, 224
  214. #endif
  215. stvx v20, r11, r0
  216. addi r11, r11, 16
  217. stvx v21, r11, r0
  218. addi r11, r11, 16
  219. stvx v22, r11, r0
  220. addi r11, r11, 16
  221. stvx v23, r11, r0
  222. addi r11, r11, 16
  223. stvx v24, r11, r0
  224. addi r11, r11, 16
  225. stvx v25, r11, r0
  226. addi r11, r11, 16
  227. stvx v26, r11, r0
  228. addi r11, r11, 16
  229. stvx v27, r11, r0
  230. addi r11, r11, 16
  231. stvx v28, r11, r0
  232. addi r11, r11, 16
  233. stvx v29, r11, r0
  234. addi r11, r11, 16
  235. stvx v30, r11, r0
  236. addi r11, r11, 16
  237. stvx v31, r11, r0
  238. li r11, 0
  239. stfs f1, ALPHA_R_SP
  240. stfs f2, ALPHA_I_SP
  241. // stw r0, FZERO
  242. #if defined(linux) || defined(__FreeBSD__)
  243. #ifdef __64BIT__
  244. ld LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
  245. #endif
  246. #endif
  247. #if defined(_AIX) || defined(__APPLE__)
  248. #ifdef __64BIT__
  249. ld LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
  250. #else
  251. #ifdef DOUBLE
  252. lwz B, FRAMESLOT(0) + 0(FRAMEPOINTER)
  253. lwz C, FRAMESLOT(1) + 0(FRAMEPOINTER)
  254. lwz LDC, FRAMESLOT(2) + 0(FRAMEPOINTER)
  255. #else
  256. lwz LDC, FRAMESLOT(0) + 0(FRAMEPOINTER)
  257. #endif
  258. #endif
  259. #endif
  260. #ifdef TRMMKERNEL
  261. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  262. ld OFFSET, FRAMESLOT(1) + 0(FRAMEPOINTER)
  263. #endif
  264. #if defined(_AIX) || defined(__APPLE__)
  265. #ifdef __64BIT__
  266. ld OFFSET, FRAMESLOT(1) + 0(FRAMEPOINTER)
  267. #else
  268. #ifdef DOUBLE
  269. lwz OFFSET, FRAMESLOT(3) + 0(FRAMEPOINTER)
  270. #else
  271. lwz OFFSET, FRAMESLOT(1) + 0(FRAMEPOINTER)
  272. #endif
  273. #endif
  274. #endif
  275. #if defined(TRMMKERNEL) && !defined(LEFT)
  276. neg KK, OFFSET
  277. #endif
  278. #endif
  279. #include "cgemm_macros_8x4_power8.S"
  280. cmpwi cr0, M, 0
  281. ble L999_H1
  282. cmpwi cr0, N, 0
  283. ble L999_H1
  284. cmpwi cr0, K, 0
  285. ble L999_H1
  286. slwi LDC, LDC, ZBASE_SHIFT
  287. li PRE, 384
  288. li o4 , 4
  289. li o8 , 8
  290. li o12 , 12
  291. li o16 , 16
  292. li o32 , 32
  293. li o48 , 48
  294. addi BBUFFER, SP, 512+4096
  295. li T1, -4096
  296. and BBUFFER, BBUFFER, T1
  297. #ifdef __64BIT__
  298. addi T1 , SP, 296+196
  299. #else
  300. addi T1 , SP, 224+200
  301. #endif
  302. stxsspx vs1, 0, T1
  303. lxsspx alpha_dr, 0, T1
  304. stxsspx vs2, o8 , T1
  305. lxsspx alpha_di, o8, T1
  306. addi T1, SP, 360
  307. li T2, 0
  308. stw T2, 0(T1)
  309. stw T2, 4(T1)
  310. stw T2, 8(T1)
  311. stxsspx alpha_dr, o12, T1
  312. lxvw4x alpha_sr, o0 , T1
  313. addi T1, T1, 16
  314. stw T2, 0(T1)
  315. stw T2, 4(T1)
  316. stw T2, 8(T1)
  317. stxsspx alpha_di, o12, T1
  318. lxvw4x alpha_si, o0 , T1
  319. .align 5
  320. #include "cgemm_logic_8x4_power8.S"
  321. L999:
  322. addi r3, 0, 0
  323. lfd f14, 0(SP)
  324. lfd f15, 8(SP)
  325. lfd f16, 16(SP)
  326. lfd f17, 24(SP)
  327. lfd f18, 32(SP)
  328. lfd f19, 40(SP)
  329. lfd f20, 48(SP)
  330. lfd f21, 56(SP)
  331. lfd f22, 64(SP)
  332. lfd f23, 72(SP)
  333. lfd f24, 80(SP)
  334. lfd f25, 88(SP)
  335. lfd f26, 96(SP)
  336. lfd f27, 104(SP)
  337. lfd f28, 112(SP)
  338. lfd f29, 120(SP)
  339. lfd f30, 128(SP)
  340. lfd f31, 136(SP)
  341. #ifdef __64BIT__
  342. ld r31, 144(SP)
  343. ld r30, 152(SP)
  344. ld r29, 160(SP)
  345. ld r28, 168(SP)
  346. ld r27, 176(SP)
  347. ld r26, 184(SP)
  348. ld r25, 192(SP)
  349. ld r24, 200(SP)
  350. ld r23, 208(SP)
  351. ld r22, 216(SP)
  352. ld r21, 224(SP)
  353. ld r20, 232(SP)
  354. ld r19, 240(SP)
  355. ld r18, 248(SP)
  356. ld r17, 256(SP)
  357. ld r16, 264(SP)
  358. ld r15, 272(SP)
  359. ld r14, 280(SP)
  360. addi r11, SP, 288
  361. #else
  362. lwz r31, 144(SP)
  363. lwz r30, 148(SP)
  364. lwz r29, 152(SP)
  365. lwz r28, 156(SP)
  366. lwz r27, 160(SP)
  367. lwz r26, 164(SP)
  368. lwz r25, 168(SP)
  369. lwz r24, 172(SP)
  370. lwz r23, 176(SP)
  371. lwz r22, 180(SP)
  372. lwz r21, 184(SP)
  373. lwz r20, 188(SP)
  374. lwz r19, 192(SP)
  375. lwz r18, 196(SP)
  376. lwz r17, 200(SP)
  377. lwz r16, 204(SP)
  378. lwz r15, 208(SP)
  379. lwz r14, 212(SP)
  380. addi r11, SP, 224
  381. #endif
  382. lvx v20, r11, r0
  383. addi r11, r11, 16
  384. lvx v21, r11, r0
  385. addi r11, r11, 16
  386. lvx v22, r11, r0
  387. addi r11, r11, 16
  388. lvx v23, r11, r0
  389. addi r11, r11, 16
  390. lvx v24, r11, r0
  391. addi r11, r11, 16
  392. lvx v25, r11, r0
  393. addi r11, r11, 16
  394. lvx v26, r11, r0
  395. addi r11, r11, 16
  396. lvx v27, r11, r0
  397. addi r11, r11, 16
  398. lvx v28, r11, r0
  399. addi r11, r11, 16
  400. lvx v29, r11, r0
  401. addi r11, r11, 16
  402. lvx v30, r11, r0
  403. addi r11, r11, 16
  404. lvx v31, r11, r0
  405. li r11, 0
  406. addi SP, SP, STACKSIZE
  407. addi SP, SP, STACKSIZE
  408. addi SP, SP, STACKSIZE
  409. addi SP, SP, STACKSIZE
  410. blr
  411. EPILOGUE
  412. #endif