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.

ctrmm_kernel_8x4_power8.S 12 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454
  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 400
  81. #define STACKSIZE 592
  82. #define ALPHA_R_SP 304+192(SP)
  83. #define ALPHA_I_SP 312+192(SP)
  84. #else
  85. #define STACKSIZE 256
  86. #define STACKSIZE 452
  87. #define ALPHA_R_SP 224+196(SP)
  88. #define ALPHA_I_SP 232+196(SP)
  89. #define FZERO 240+196(SP)
  90. #endif
  91. #define M r3
  92. #define N r4
  93. #define K r5
  94. #if defined(linux) || defined(__FreeBSD__)
  95. #ifndef __64BIT__
  96. #define A r6
  97. #define B r7
  98. #define C r8
  99. #define LDC r9
  100. #define OFFSET r10
  101. #else
  102. #define A r8
  103. #define B r9
  104. #define C r10
  105. #define LDC r6
  106. #define OFFSET r7
  107. #endif
  108. #endif
  109. #if defined(_AIX) || defined(__APPLE__)
  110. #if !defined(__64BIT__) && defined(DOUBLE)
  111. #define A r10
  112. #define B r6
  113. #define C r7
  114. #define LDC r8
  115. #define OFFSET r9
  116. #else
  117. #define A r8
  118. #define B r9
  119. #define C r10
  120. #define LDC r6
  121. #define OFFSET r7
  122. #endif
  123. #endif
  124. #define o0 0
  125. #define alpha_dr vs28
  126. #define alpha_di vs29
  127. #define alpha_sr vs30
  128. #define alpha_si vs31
  129. #define VECSAVE r11
  130. #define o12 r12
  131. #define KKK r13
  132. #define K1 r14
  133. #define L r15
  134. #define o16 r16
  135. #define NOTUSED r17
  136. #define T2 r19
  137. #define KK r20
  138. #define o8 r21
  139. #define I r22
  140. #define J r23
  141. #define AO r24
  142. #define BO r25
  143. #define CO r26
  144. #define o4 r27
  145. #define o32 r28
  146. #define o48 r29
  147. #define PRE r30
  148. #define T1 r31
  149. #ifndef NEEDPARAM
  150. PROLOGUE
  151. PROFCODE
  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. std r13, 288(SP)
  192. std r12, 296(SP)
  193. addi r11, SP, 304
  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. stw r13, 216(SP)
  214. addi r11, SP, 224
  215. #endif
  216. stvx v20, r11, r0
  217. addi r11, r11, 16
  218. stvx v21, r11, r0
  219. addi r11, r11, 16
  220. stvx v22, r11, r0
  221. addi r11, r11, 16
  222. stvx v23, r11, r0
  223. addi r11, r11, 16
  224. stvx v24, r11, r0
  225. addi r11, r11, 16
  226. stvx v25, r11, r0
  227. addi r11, r11, 16
  228. stvx v26, r11, r0
  229. addi r11, r11, 16
  230. stvx v27, r11, r0
  231. addi r11, r11, 16
  232. stvx v28, r11, r0
  233. addi r11, r11, 16
  234. stvx v29, r11, r0
  235. addi r11, r11, 16
  236. stvx v30, r11, r0
  237. addi r11, r11, 16
  238. stvx v31, r11, r0
  239. li r11, 0
  240. stfs f1, ALPHA_R_SP
  241. stfs f2, ALPHA_I_SP
  242. // stw r0, FZERO
  243. #if defined(linux) || defined(__FreeBSD__)
  244. #ifdef __64BIT__
  245. ld LDC, FRAMESLOT(0) + STACKSIZE(SP)
  246. #endif
  247. #endif
  248. #if defined(_AIX) || defined(__APPLE__)
  249. #ifdef __64BIT__
  250. ld LDC, FRAMESLOT(0) + STACKSIZE(SP)
  251. #else
  252. #ifdef DOUBLE
  253. lwz B, FRAMESLOT(0) + STACKSIZE(SP)
  254. lwz C, FRAMESLOT(1) + STACKSIZE(SP)
  255. lwz LDC, FRAMESLOT(2) + STACKSIZE(SP)
  256. #else
  257. lwz LDC, FRAMESLOT(0) + STACKSIZE(SP)
  258. #endif
  259. #endif
  260. #endif
  261. #ifdef TRMMKERNEL
  262. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  263. ld OFFSET, FRAMESLOT(1) + STACKSIZE(SP)
  264. #endif
  265. #if defined(_AIX) || defined(__APPLE__)
  266. #ifdef __64BIT__
  267. ld OFFSET, FRAMESLOT(1) + STACKSIZE(SP)
  268. #else
  269. #ifdef DOUBLE
  270. lwz OFFSET, FRAMESLOT(3) + STACKSIZE(SP)
  271. #else
  272. lwz OFFSET, FRAMESLOT(1) + STACKSIZE(SP)
  273. #endif
  274. #endif
  275. #endif
  276. #if defined(TRMMKERNEL) && !defined(LEFT)
  277. neg KK, OFFSET
  278. #endif
  279. #endif
  280. #include "ctrmm_macros_8x4_power8.S"
  281. cmpwi cr0, M, 0
  282. ble L999_H1
  283. cmpwi cr0, N, 0
  284. ble L999_H1
  285. cmpwi cr0, K, 0
  286. ble L999_H1
  287. slwi LDC, LDC, ZBASE_SHIFT
  288. li PRE, 384
  289. li o4 , 4
  290. li o8 , 8
  291. li o12 , 12
  292. li o16 , 16
  293. li o32 , 32
  294. li o48 , 48
  295. #ifdef __64BIT__
  296. addi T1, SP, 304+192
  297. #else
  298. addi T1, SP, 224+196
  299. #endif
  300. lxsspx alpha_dr, 0, T1
  301. lxsspx alpha_di, o8, T1
  302. addi T1, SP, 360
  303. li T2, 0
  304. stw T2, 0(T1)
  305. stw T2, 4(T1)
  306. stw T2, 8(T1)
  307. stxsspx alpha_dr, o12, T1
  308. lxvw4x alpha_sr, o0 , T1
  309. addi T1, T1, 16
  310. stw T2, 0(T1)
  311. stw T2, 4(T1)
  312. stw T2, 8(T1)
  313. stxsspx alpha_di, o12, T1
  314. lxvw4x alpha_si, o0 , T1
  315. .align 5
  316. #include "ctrmm_logic_8x4_power8.S"
  317. L999:
  318. addi r3, 0, 0
  319. lfd f14, 0(SP)
  320. lfd f15, 8(SP)
  321. lfd f16, 16(SP)
  322. lfd f17, 24(SP)
  323. lfd f18, 32(SP)
  324. lfd f19, 40(SP)
  325. lfd f20, 48(SP)
  326. lfd f21, 56(SP)
  327. lfd f22, 64(SP)
  328. lfd f23, 72(SP)
  329. lfd f24, 80(SP)
  330. lfd f25, 88(SP)
  331. lfd f26, 96(SP)
  332. lfd f27, 104(SP)
  333. lfd f28, 112(SP)
  334. lfd f29, 120(SP)
  335. lfd f30, 128(SP)
  336. lfd f31, 136(SP)
  337. #ifdef __64BIT__
  338. ld r31, 144(SP)
  339. ld r30, 152(SP)
  340. ld r29, 160(SP)
  341. ld r28, 168(SP)
  342. ld r27, 176(SP)
  343. ld r26, 184(SP)
  344. ld r25, 192(SP)
  345. ld r24, 200(SP)
  346. ld r23, 208(SP)
  347. ld r22, 216(SP)
  348. ld r21, 224(SP)
  349. ld r20, 232(SP)
  350. ld r19, 240(SP)
  351. ld r18, 248(SP)
  352. ld r17, 256(SP)
  353. ld r16, 264(SP)
  354. ld r15, 272(SP)
  355. ld r14, 280(SP)
  356. ld r13, 288(SP)
  357. ld r12, 296(SP)
  358. addi r11, SP, 304
  359. #else
  360. lwz r31, 144(SP)
  361. lwz r30, 148(SP)
  362. lwz r29, 152(SP)
  363. lwz r28, 156(SP)
  364. lwz r27, 160(SP)
  365. lwz r26, 164(SP)
  366. lwz r25, 168(SP)
  367. lwz r24, 172(SP)
  368. lwz r23, 176(SP)
  369. lwz r22, 180(SP)
  370. lwz r21, 184(SP)
  371. lwz r20, 188(SP)
  372. lwz r19, 192(SP)
  373. lwz r18, 196(SP)
  374. lwz r17, 200(SP)
  375. lwz r16, 204(SP)
  376. lwz r15, 208(SP)
  377. lwz r14, 212(SP)
  378. lwz r13, 216(SP)
  379. addi r11, SP, 224
  380. #endif
  381. lvx v20, r11, r3
  382. addi r11, r11, 16
  383. lvx v21, r11, r3
  384. addi r11, r11, 16
  385. lvx v22, r11, r3
  386. addi r11, r11, 16
  387. lvx v23, r11, r3
  388. addi r11, r11, 16
  389. lvx v24, r11, r3
  390. addi r11, r11, 16
  391. lvx v25, r11, r3
  392. addi r11, r11, 16
  393. lvx v26, r11, r3
  394. addi r11, r11, 16
  395. lvx v27, r11, r3
  396. addi r11, r11, 16
  397. lvx v28, r11, r3
  398. addi r11, r11, 16
  399. lvx v29, r11, r3
  400. addi r11, r11, 16
  401. lvx v30, r11, r3
  402. addi r11, r11, 16
  403. lvx v31, r11, r3
  404. li r11, 0
  405. addi SP, SP, STACKSIZE
  406. blr
  407. EPILOGUE
  408. #endif