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.

dtrsm_kernel_LT_16x4_power8.S 7.8 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349
  1. /*********************************************************************/
  2. /* Copyright 2009, 2010 The University of Texas at Austin. */
  3. /* All rights reserved. */
  4. /* */
  5. /* Redistribution and use in source and binary forms, with or */
  6. /* without modification, are permitted provided that the following */
  7. /* conditions are met: */
  8. /* */
  9. /* 1. Redistributions of source code must retain the above */
  10. /* copyright notice, this list of conditions and the following */
  11. /* disclaimer. */
  12. /* */
  13. /* 2. Redistributions in binary form must reproduce the above */
  14. /* copyright notice, this list of conditions and the following */
  15. /* disclaimer in the documentation and/or other materials */
  16. /* provided with the distribution. */
  17. /* */
  18. /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
  19. /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
  20. /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
  21. /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
  22. /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
  23. /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
  24. /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
  25. /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
  26. /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
  27. /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
  28. /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
  29. /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
  30. /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
  31. /* POSSIBILITY OF SUCH DAMAGE. */
  32. /* */
  33. /* The views and conclusions contained in the software and */
  34. /* documentation are those of the authors and should not be */
  35. /* interpreted as representing official policies, either expressed */
  36. /* or implied, of The University of Texas at Austin. */
  37. /*********************************************************************/
  38. #define ASSEMBLER
  39. #include "common.h"
  40. #include "def_vsx.h"
  41. #ifndef __64BIT__
  42. #define LOAD lwz
  43. #else
  44. #define LOAD ld
  45. #endif
  46. #ifdef __64BIT__
  47. #define STACKSIZE 320
  48. #define STACKSIZE 520
  49. #define ALPHA 296+200(SP)
  50. #define FZERO 304+200(SP)
  51. #else
  52. #define STACKSIZE 240
  53. #define ALPHA 224(SP)
  54. #define FZERO 232(SP)
  55. #endif
  56. #define M r3
  57. #define N r4
  58. #define K r5
  59. #if defined(linux) || defined(__FreeBSD__)
  60. #ifndef __64BIT__
  61. #define A r6
  62. #define B r7
  63. #define C r8
  64. #define LDC r9
  65. #define OFFSET r10
  66. #else
  67. #define A r7
  68. #define B r8
  69. #define C r9
  70. #define LDC r10
  71. #define OFFSET r6
  72. #endif
  73. #endif
  74. #if defined(_AIX) || defined(__APPLE__)
  75. #if !defined(__64BIT__) && defined(DOUBLE)
  76. #define A r8
  77. #define B r9
  78. #define C r10
  79. #define LDC r7
  80. #define OFFSET r6
  81. #else
  82. #define A r7
  83. #define B r8
  84. #define C r9
  85. #define LDC r10
  86. #define OFFSET r6
  87. #endif
  88. #endif
  89. #define o0 0
  90. #define PRE r15
  91. #define T4 r16
  92. #define L r17
  93. #define T3 r18
  94. #define T2 r19
  95. #define KK r20
  96. #define I r21
  97. #define J r22
  98. #define AO r23
  99. #define BO r24
  100. #define CO r25
  101. #define o8 r26
  102. #define o16 r27
  103. #define o24 r28
  104. #define o32 r29
  105. #define o48 r30
  106. #define T1 r31
  107. #define VECSAVE r11
  108. #include "dtrsm_macros_LT_16x4_power8.S"
  109. #ifndef NEEDPARAM
  110. PROLOGUE
  111. PROFCODE
  112. addi SP, SP, -STACKSIZE
  113. li r0, 0
  114. stfd f14, 0(SP)
  115. stfd f15, 8(SP)
  116. stfd f16, 16(SP)
  117. stfd f17, 24(SP)
  118. stfd f18, 32(SP)
  119. stfd f19, 40(SP)
  120. stfd f20, 48(SP)
  121. stfd f21, 56(SP)
  122. stfd f22, 64(SP)
  123. stfd f23, 72(SP)
  124. stfd f24, 80(SP)
  125. stfd f25, 88(SP)
  126. stfd f26, 96(SP)
  127. stfd f27, 104(SP)
  128. stfd f28, 112(SP)
  129. stfd f29, 120(SP)
  130. stfd f30, 128(SP)
  131. stfd f31, 136(SP)
  132. #ifdef __64BIT__
  133. std r31, 144(SP)
  134. std r30, 152(SP)
  135. std r29, 160(SP)
  136. std r28, 168(SP)
  137. std r27, 176(SP)
  138. std r26, 184(SP)
  139. std r25, 192(SP)
  140. std r24, 200(SP)
  141. std r23, 208(SP)
  142. std r22, 216(SP)
  143. std r21, 224(SP)
  144. std r20, 232(SP)
  145. std r19, 240(SP)
  146. std r18, 248(SP)
  147. std r17, 256(SP)
  148. std r16, 264(SP)
  149. std r15, 272(SP)
  150. addi r11,SP,288
  151. #else
  152. stw r31, 144(SP)
  153. stw r30, 148(SP)
  154. stw r29, 152(SP)
  155. stw r28, 156(SP)
  156. stw r27, 160(SP)
  157. stw r26, 164(SP)
  158. stw r25, 168(SP)
  159. stw r24, 172(SP)
  160. stw r23, 176(SP)
  161. stw r22, 180(SP)
  162. stw r21, 184(SP)
  163. stw r20, 188(SP)
  164. stw r19, 192(SP)
  165. stw r18, 196(SP)
  166. addi r11,SP,208
  167. #endif
  168. stvx v20, r11,r0
  169. addi r11,r11,16
  170. stvx v21, r11,r0
  171. addi r11,r11,16
  172. stvx v22, r11,r0
  173. addi r11,r11,16
  174. stvx v23, r11,r0
  175. addi r11,r11,16
  176. stvx v24, r11,r0
  177. addi r11,r11,16
  178. stvx v25, r11,r0
  179. addi r11,r11,16
  180. stvx v26, r11,r0
  181. addi r11,r11,16
  182. stvx v27, r11,r0
  183. addi r11,r11,16
  184. stvx v28, r11,r0
  185. addi r11,r11,16
  186. stvx v29, r11,r0
  187. addi r11,r11,16
  188. stvx v30, r11,r0
  189. addi r11,r11,16
  190. stvx v31, r11,r0
  191. li r11,0
  192. #if defined(_AIX) || defined(__APPLE__)
  193. #if !defined(__64BIT__) && defined(DOUBLE)
  194. lwz LDC, FRAMESLOT(0) + STACKSIZE(SP)
  195. #endif
  196. #endif
  197. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  198. ld OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  199. #endif
  200. #if defined(_AIX) || defined(__APPLE__)
  201. #ifdef __64BIT__
  202. ld OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  203. #else
  204. #ifdef DOUBLE
  205. lwz OFFSET, FRAMESLOT(1) + STACKSIZE(SP)
  206. #else
  207. lwz OFFSET, FRAMESLOT(0) + STACKSIZE(SP)
  208. #endif
  209. #endif
  210. #endif
  211. cmpwi cr0, M, 0
  212. ble L999
  213. cmpwi cr0, N, 0
  214. ble L999
  215. cmpwi cr0, K, 0
  216. ble L999
  217. slwi LDC, LDC, BASE_SHIFT
  218. li o8, 8
  219. li o16, 16
  220. li o24, 24
  221. li o32, 32
  222. li o48, 48
  223. li PRE, 384
  224. mr KK, OFFSET
  225. #include "dtrsm_logic_LT_16x4_power8.S"
  226. L999:
  227. addi r3, 0, 0
  228. lfd f14, 0(SP)
  229. lfd f15, 8(SP)
  230. lfd f16, 16(SP)
  231. lfd f17, 24(SP)
  232. lfd f18, 32(SP)
  233. lfd f19, 40(SP)
  234. lfd f20, 48(SP)
  235. lfd f21, 56(SP)
  236. lfd f22, 64(SP)
  237. lfd f23, 72(SP)
  238. lfd f24, 80(SP)
  239. lfd f25, 88(SP)
  240. lfd f26, 96(SP)
  241. lfd f27, 104(SP)
  242. lfd f28, 112(SP)
  243. lfd f29, 120(SP)
  244. lfd f30, 128(SP)
  245. lfd f31, 136(SP)
  246. #ifdef __64BIT__
  247. ld r31, 144(SP)
  248. ld r30, 152(SP)
  249. ld r29, 160(SP)
  250. ld r28, 168(SP)
  251. ld r27, 176(SP)
  252. ld r26, 184(SP)
  253. ld r25, 192(SP)
  254. ld r24, 200(SP)
  255. ld r23, 208(SP)
  256. ld r22, 216(SP)
  257. ld r21, 224(SP)
  258. ld r20, 232(SP)
  259. ld r19, 240(SP)
  260. ld r18, 248(SP)
  261. ld r17, 256(SP)
  262. ld r16, 264(SP)
  263. ld r15, 272(SP)
  264. addi r11,SP,288
  265. #else
  266. lwz r31, 144(SP)
  267. lwz r30, 148(SP)
  268. lwz r29, 152(SP)
  269. lwz r28, 156(SP)
  270. lwz r27, 160(SP)
  271. lwz r26, 164(SP)
  272. lwz r25, 168(SP)
  273. lwz r24, 172(SP)
  274. lwz r23, 176(SP)
  275. lwz r22, 180(SP)
  276. lwz r21, 184(SP)
  277. lwz r20, 188(SP)
  278. lwz r19, 192(SP)
  279. lwz r18, 196(SP)
  280. addi r11,SP,208
  281. #endif
  282. lvx v20, r11,r3
  283. addi r11,r11,16
  284. lvx v21, r11,r3
  285. addi r11,r11,16
  286. lvx v22, r11,r3
  287. addi r11,r11,16
  288. lvx v23, r11,r3
  289. addi r11,r11,16
  290. lvx v24, r11,r3
  291. addi r11,r11,16
  292. lvx v25, r11,r3
  293. addi r11,r11,16
  294. lvx v26, r11,r3
  295. addi r11,r11,16
  296. lvx v27, r11,r3
  297. addi r11,r11,16
  298. lvx v28, r11,r3
  299. addi r11,r11,16
  300. lvx v29, r11,r3
  301. addi r11,r11,16
  302. lvx v30, r11,r3
  303. addi r11,r11,16
  304. lvx v31, r11,r3
  305. li r11,0
  306. addi SP, SP, STACKSIZE
  307. blr
  308. EPILOGUE
  309. #endif