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.

swap.S 8.5 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387
  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. #if defined(linux) || defined(__FreeBSD__)
  41. #ifndef __64BIT__
  42. #define N r3
  43. #define X r6
  44. #define INCX r7
  45. #define Y r8
  46. #define INCY r9
  47. #define PREA r4
  48. #define XX r10
  49. #define YY r11
  50. #else
  51. #define N r3
  52. #define X r7
  53. #define INCX r8
  54. #define Y r9
  55. #define INCY r10
  56. #define PREA r4
  57. #define XX r5
  58. #define YY r6
  59. #endif
  60. #endif
  61. #if defined(_AIX) || defined(__APPLE__)
  62. #if !defined(__64BIT__) && defined(DOUBLE)
  63. #define N r3
  64. #define X r8
  65. #define INCX r9
  66. #define Y r10
  67. #define INCY r4
  68. #define PREA r5
  69. #define XX r6
  70. #define YY r11
  71. #else
  72. #define N r3
  73. #define X r7
  74. #define INCX r8
  75. #define Y r9
  76. #define INCY r10
  77. #define PREA r4
  78. #define XX r5
  79. #define YY r6
  80. #endif
  81. #endif
  82. #define STACKSIZE 160
  83. PROLOGUE
  84. PROFCODE
  85. addi SP, SP, -STACKSIZE
  86. li r0, 0
  87. stfd f14, 0(SP)
  88. stfd f15, 8(SP)
  89. stfd f16, 16(SP)
  90. stfd f17, 24(SP)
  91. stfd f18, 32(SP)
  92. stfd f19, 40(SP)
  93. stfd f20, 48(SP)
  94. stfd f21, 56(SP)
  95. stfd f22, 64(SP)
  96. stfd f23, 72(SP)
  97. stfd f24, 80(SP)
  98. stfd f25, 88(SP)
  99. stfd f26, 96(SP)
  100. stfd f27, 104(SP)
  101. stfd f28, 112(SP)
  102. stfd f29, 120(SP)
  103. stfd f30, 128(SP)
  104. stfd f31, 136(SP)
  105. #if (defined(_AIX) || defined(__APPLE__)) && !defined(__64BIT__) && defined(DOUBLE)
  106. lwz INCY, FRAMESLOT(0) + STACKSIZE(SP)
  107. #endif
  108. slwi INCX, INCX, BASE_SHIFT
  109. slwi INCY, INCY, BASE_SHIFT
  110. #ifdef L1_DUALFETCH
  111. li PREA, (L1_PREFETCHSIZE) / 2
  112. #else
  113. li PREA, (L1_PREFETCHSIZE)
  114. #endif
  115. cmpwi cr0, N, 0
  116. ble- LL(999)
  117. cmpwi cr0, INCX, SIZE
  118. bne- cr0, LL(100)
  119. cmpwi cr0, INCY, SIZE
  120. bne- cr0, LL(100)
  121. srawi. r0, N, 4
  122. mtspr CTR, r0
  123. beq- cr0, LL(50)
  124. .align 4
  125. LL(10):
  126. LFD f0, 0 * SIZE(X)
  127. LFD f1, 1 * SIZE(X)
  128. LFD f2, 2 * SIZE(X)
  129. LFD f3, 3 * SIZE(X)
  130. LFD f16, 0 * SIZE(Y)
  131. LFD f17, 1 * SIZE(Y)
  132. LFD f18, 2 * SIZE(Y)
  133. LFD f19, 3 * SIZE(Y)
  134. LFD f4, 4 * SIZE(X)
  135. LFD f5, 5 * SIZE(X)
  136. LFD f6, 6 * SIZE(X)
  137. LFD f7, 7 * SIZE(X)
  138. LFD f20, 4 * SIZE(Y)
  139. LFD f21, 5 * SIZE(Y)
  140. LFD f22, 6 * SIZE(Y)
  141. LFD f23, 7 * SIZE(Y)
  142. LFD f8, 8 * SIZE(X)
  143. LFD f9, 9 * SIZE(X)
  144. LFD f10, 10 * SIZE(X)
  145. LFD f11, 11 * SIZE(X)
  146. LFD f24, 8 * SIZE(Y)
  147. LFD f25, 9 * SIZE(Y)
  148. LFD f26, 10 * SIZE(Y)
  149. LFD f27, 11 * SIZE(Y)
  150. LFD f12, 12 * SIZE(X)
  151. LFD f13, 13 * SIZE(X)
  152. LFD f14, 14 * SIZE(X)
  153. LFD f15, 15 * SIZE(X)
  154. LFD f28, 12 * SIZE(Y)
  155. LFD f29, 13 * SIZE(Y)
  156. LFD f30, 14 * SIZE(Y)
  157. LFD f31, 15 * SIZE(Y)
  158. STFD f16, 0 * SIZE(X)
  159. STFD f17, 1 * SIZE(X)
  160. STFD f18, 2 * SIZE(X)
  161. STFD f19, 3 * SIZE(X)
  162. STFD f0, 0 * SIZE(Y)
  163. STFD f1, 1 * SIZE(Y)
  164. STFD f2, 2 * SIZE(Y)
  165. STFD f3, 3 * SIZE(Y)
  166. STFD f20, 4 * SIZE(X)
  167. STFD f21, 5 * SIZE(X)
  168. STFD f22, 6 * SIZE(X)
  169. STFD f23, 7 * SIZE(X)
  170. STFD f4, 4 * SIZE(Y)
  171. STFD f5, 5 * SIZE(Y)
  172. STFD f6, 6 * SIZE(Y)
  173. STFD f7, 7 * SIZE(Y)
  174. STFD f24, 8 * SIZE(X)
  175. STFD f25, 9 * SIZE(X)
  176. STFD f26, 10 * SIZE(X)
  177. STFD f27, 11 * SIZE(X)
  178. STFD f8, 8 * SIZE(Y)
  179. STFD f9, 9 * SIZE(Y)
  180. STFD f10, 10 * SIZE(Y)
  181. STFD f11, 11 * SIZE(Y)
  182. STFD f28, 12 * SIZE(X)
  183. STFD f29, 13 * SIZE(X)
  184. STFD f30, 14 * SIZE(X)
  185. STFD f31, 15 * SIZE(X)
  186. STFD f12, 12 * SIZE(Y)
  187. STFD f13, 13 * SIZE(Y)
  188. STFD f14, 14 * SIZE(Y)
  189. STFD f15, 15 * SIZE(Y)
  190. addi X, X, 16 * SIZE
  191. addi Y, Y, 16 * SIZE
  192. dcbtst X, PREA
  193. #ifdef L1_DUALFETCH
  194. dcbtst Y, PREA
  195. #endif
  196. bdnz LL(10)
  197. .align 4
  198. LL(50):
  199. andi. r0, N, 15
  200. mtspr CTR, r0
  201. beq LL(999)
  202. .align 4
  203. LL(60):
  204. LFD f8, 0 * SIZE(X)
  205. LFD f9, 0 * SIZE(Y)
  206. STFD f9, 0 * SIZE(X)
  207. STFD f8, 0 * SIZE(Y)
  208. addi X, X, 1 * SIZE
  209. addi Y, Y, 1 * SIZE
  210. bdnz LL(60)
  211. b LL(999)
  212. .align 4
  213. LL(100):
  214. sub X, X, INCX
  215. sub Y, Y, INCY
  216. mr XX, X
  217. mr YY, Y
  218. srawi. r0, N, 4
  219. mtspr CTR, r0
  220. beq- LL(150)
  221. .align 4
  222. LL(110):
  223. LFDUX f0, X, INCX
  224. LFDUX f1, X, INCX
  225. LFDUX f2, X, INCX
  226. LFDUX f3, X, INCX
  227. LFDUX f16, Y, INCY
  228. LFDUX f17, Y, INCY
  229. LFDUX f18, Y, INCY
  230. LFDUX f19, Y, INCY
  231. LFDUX f4, X, INCX
  232. LFDUX f5, X, INCX
  233. LFDUX f6, X, INCX
  234. LFDUX f7, X, INCX
  235. LFDUX f20, Y, INCY
  236. LFDUX f21, Y, INCY
  237. LFDUX f22, Y, INCY
  238. LFDUX f23, Y, INCY
  239. LFDUX f8, X, INCX
  240. LFDUX f9, X, INCX
  241. LFDUX f10, X, INCX
  242. LFDUX f11, X, INCX
  243. LFDUX f24, Y, INCY
  244. LFDUX f25, Y, INCY
  245. LFDUX f26, Y, INCY
  246. LFDUX f27, Y, INCY
  247. LFDUX f12, X, INCX
  248. LFDUX f13, X, INCX
  249. LFDUX f14, X, INCX
  250. LFDUX f15, X, INCX
  251. LFDUX f28, Y, INCY
  252. LFDUX f29, Y, INCY
  253. LFDUX f30, Y, INCY
  254. LFDUX f31, Y, INCY
  255. STFDUX f16, XX, INCX
  256. STFDUX f17, XX, INCX
  257. STFDUX f18, XX, INCX
  258. STFDUX f19, XX, INCX
  259. STFDUX f0, YY, INCY
  260. STFDUX f1, YY, INCY
  261. STFDUX f2, YY, INCY
  262. STFDUX f3, YY, INCY
  263. STFDUX f20, XX, INCX
  264. STFDUX f21, XX, INCX
  265. STFDUX f22, XX, INCX
  266. STFDUX f23, XX, INCX
  267. STFDUX f4, YY, INCY
  268. STFDUX f5, YY, INCY
  269. STFDUX f6, YY, INCY
  270. STFDUX f7, YY, INCY
  271. STFDUX f24, XX, INCX
  272. STFDUX f25, XX, INCX
  273. STFDUX f26, XX, INCX
  274. STFDUX f27, XX, INCX
  275. STFDUX f8, YY, INCY
  276. STFDUX f9, YY, INCY
  277. STFDUX f10, YY, INCY
  278. STFDUX f11, YY, INCY
  279. STFDUX f28, XX, INCX
  280. STFDUX f29, XX, INCX
  281. STFDUX f30, XX, INCX
  282. STFDUX f31, XX, INCX
  283. STFDUX f12, YY, INCY
  284. STFDUX f13, YY, INCY
  285. STFDUX f14, YY, INCY
  286. STFDUX f15, YY, INCY
  287. bdnz LL(110)
  288. .align 4
  289. LL(150):
  290. andi. r0, N, 15
  291. mtspr CTR, r0
  292. beq LL(999)
  293. .align 4
  294. LL(160):
  295. LFDUX f8, X, INCX
  296. LFDUX f9, Y, INCY
  297. STFDUX f9, XX, INCX
  298. STFDUX f8, YY, INCY
  299. bdnz LL(160)
  300. .align 4
  301. LL(999):
  302. lfd f14, 0(SP)
  303. lfd f15, 8(SP)
  304. lfd f16, 16(SP)
  305. lfd f17, 24(SP)
  306. lfd f18, 32(SP)
  307. lfd f19, 40(SP)
  308. lfd f20, 48(SP)
  309. lfd f21, 56(SP)
  310. lfd f22, 64(SP)
  311. lfd f23, 72(SP)
  312. lfd f24, 80(SP)
  313. lfd f25, 88(SP)
  314. lfd f26, 96(SP)
  315. lfd f27, 104(SP)
  316. lfd f28, 112(SP)
  317. lfd f29, 120(SP)
  318. lfd f30, 128(SP)
  319. lfd f31, 136(SP)
  320. addi SP, SP, STACKSIZE
  321. blr
  322. EPILOGUE