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.

zswap.S 9.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414
  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 r5
  49. #define YY r10
  50. #else
  51. #define N r3
  52. #define X r8
  53. #define INCX r9
  54. #define Y r10
  55. #define INCY r4
  56. #define PREA r5
  57. #define XX r6
  58. #define YY r7
  59. #endif
  60. #endif
  61. #if defined(_AIX) || defined(__APPLE__)
  62. #if !defined(__64BIT__) && defined(DOUBLE)
  63. #define N r3
  64. #define X r10
  65. #define INCX r4
  66. #define Y r5
  67. #define INCY r6
  68. #define PREA r7
  69. #define XX r8
  70. #define YY r9
  71. #else
  72. #define N r3
  73. #define X r8
  74. #define INCX r9
  75. #define Y r10
  76. #define INCY r4
  77. #define PREA r5
  78. #define XX r6
  79. #define YY r7
  80. #endif
  81. #endif
  82. #define INCXM1 r11
  83. #define INCYM1 r12
  84. #define STACKSIZE 160
  85. PROLOGUE
  86. PROFCODE
  87. addi SP, SP, -STACKSIZE
  88. li r0, 0
  89. stfd f14, 0(SP)
  90. stfd f15, 8(SP)
  91. stfd f16, 16(SP)
  92. stfd f17, 24(SP)
  93. stfd f18, 32(SP)
  94. stfd f19, 40(SP)
  95. stfd f20, 48(SP)
  96. stfd f21, 56(SP)
  97. stfd f22, 64(SP)
  98. stfd f23, 72(SP)
  99. stfd f24, 80(SP)
  100. stfd f25, 88(SP)
  101. stfd f26, 96(SP)
  102. stfd f27, 104(SP)
  103. stfd f28, 112(SP)
  104. stfd f29, 120(SP)
  105. stfd f30, 128(SP)
  106. stfd f31, 136(SP)
  107. #if (defined(linux) || defined(__FreeBSD__)) && defined(__64BIT__)
  108. ld INCY, FRAMESLOT(0) + STACKSIZE(SP)
  109. #endif
  110. #if defined(_AIX) || defined(__APPLE__)
  111. #ifdef __64BIT__
  112. ld INCY, FRAMESLOT(0) + STACKSIZE(SP)
  113. #else
  114. #ifdef DOUBLE
  115. lwz INCX, FRAMESLOT(0) + STACKSIZE(SP)
  116. lwz Y, FRAMESLOT(1) + STACKSIZE(SP)
  117. lwz INCY, FRAMESLOT(2) + STACKSIZE(SP)
  118. #else
  119. lwz INCY, FRAMESLOT(0) + STACKSIZE(SP)
  120. #endif
  121. #endif
  122. #endif
  123. slwi INCX, INCX, ZBASE_SHIFT
  124. slwi INCY, INCY, ZBASE_SHIFT
  125. subi INCXM1, INCX, SIZE
  126. subi INCYM1, INCY, SIZE
  127. #ifdef L1_DUALFETCH
  128. li PREA, (L1_PREFETCHSIZE) / 2
  129. #else
  130. li PREA, (L1_PREFETCHSIZE)
  131. #endif
  132. cmpwi cr0, N, 0
  133. ble- LL(999)
  134. cmpwi cr0, INCX, 2 * SIZE
  135. bne- cr0, LL(100)
  136. cmpwi cr0, INCY, 2 * SIZE
  137. bne- cr0, LL(100)
  138. srawi. r0, N, 3
  139. mtspr CTR, r0
  140. beq- cr0, LL(50)
  141. .align 4
  142. LL(10):
  143. LFD f0, 0 * SIZE(X)
  144. LFD f1, 1 * SIZE(X)
  145. LFD f2, 2 * SIZE(X)
  146. LFD f3, 3 * SIZE(X)
  147. LFD f16, 0 * SIZE(Y)
  148. LFD f17, 1 * SIZE(Y)
  149. LFD f18, 2 * SIZE(Y)
  150. LFD f19, 3 * SIZE(Y)
  151. LFD f4, 4 * SIZE(X)
  152. LFD f5, 5 * SIZE(X)
  153. LFD f6, 6 * SIZE(X)
  154. LFD f7, 7 * SIZE(X)
  155. LFD f20, 4 * SIZE(Y)
  156. LFD f21, 5 * SIZE(Y)
  157. LFD f22, 6 * SIZE(Y)
  158. LFD f23, 7 * SIZE(Y)
  159. LFD f8, 8 * SIZE(X)
  160. LFD f9, 9 * SIZE(X)
  161. LFD f10, 10 * SIZE(X)
  162. LFD f11, 11 * SIZE(X)
  163. LFD f24, 8 * SIZE(Y)
  164. LFD f25, 9 * SIZE(Y)
  165. LFD f26, 10 * SIZE(Y)
  166. LFD f27, 11 * SIZE(Y)
  167. LFD f12, 12 * SIZE(X)
  168. LFD f13, 13 * SIZE(X)
  169. LFD f14, 14 * SIZE(X)
  170. LFD f15, 15 * SIZE(X)
  171. LFD f28, 12 * SIZE(Y)
  172. LFD f29, 13 * SIZE(Y)
  173. LFD f30, 14 * SIZE(Y)
  174. LFD f31, 15 * SIZE(Y)
  175. STFD f16, 0 * SIZE(X)
  176. STFD f17, 1 * SIZE(X)
  177. STFD f18, 2 * SIZE(X)
  178. STFD f19, 3 * SIZE(X)
  179. STFD f0, 0 * SIZE(Y)
  180. STFD f1, 1 * SIZE(Y)
  181. STFD f2, 2 * SIZE(Y)
  182. STFD f3, 3 * SIZE(Y)
  183. STFD f20, 4 * SIZE(X)
  184. STFD f21, 5 * SIZE(X)
  185. STFD f22, 6 * SIZE(X)
  186. STFD f23, 7 * SIZE(X)
  187. STFD f4, 4 * SIZE(Y)
  188. STFD f5, 5 * SIZE(Y)
  189. STFD f6, 6 * SIZE(Y)
  190. STFD f7, 7 * SIZE(Y)
  191. STFD f24, 8 * SIZE(X)
  192. STFD f25, 9 * SIZE(X)
  193. STFD f26, 10 * SIZE(X)
  194. STFD f27, 11 * SIZE(X)
  195. STFD f8, 8 * SIZE(Y)
  196. STFD f9, 9 * SIZE(Y)
  197. STFD f10, 10 * SIZE(Y)
  198. STFD f11, 11 * SIZE(Y)
  199. STFD f28, 12 * SIZE(X)
  200. STFD f29, 13 * SIZE(X)
  201. STFD f30, 14 * SIZE(X)
  202. STFD f31, 15 * SIZE(X)
  203. STFD f12, 12 * SIZE(Y)
  204. STFD f13, 13 * SIZE(Y)
  205. STFD f14, 14 * SIZE(Y)
  206. STFD f15, 15 * SIZE(Y)
  207. addi X, X, 16 * SIZE
  208. addi Y, Y, 16 * SIZE
  209. dcbtst X, PREA
  210. #ifdef L1_DUALFETCH
  211. dcbtst Y, PREA
  212. #endif
  213. bdnz LL(10)
  214. .align 4
  215. LL(50):
  216. andi. r0, N, 7
  217. mtspr CTR, r0
  218. beq LL(999)
  219. .align 4
  220. LL(60):
  221. LFD f0, 0 * SIZE(X)
  222. LFD f1, 1 * SIZE(X)
  223. LFD f2, 0 * SIZE(Y)
  224. LFD f3, 1 * SIZE(Y)
  225. STFD f2, 0 * SIZE(X)
  226. STFD f3, 1 * SIZE(X)
  227. STFD f0, 0 * SIZE(Y)
  228. STFD f1, 1 * SIZE(Y)
  229. addi X, X, 2 * SIZE
  230. addi Y, Y, 2 * SIZE
  231. bdnz LL(60)
  232. b LL(999)
  233. .align 4
  234. LL(100):
  235. sub X, X, INCXM1
  236. sub Y, Y, INCYM1
  237. mr XX, X
  238. mr YY, Y
  239. srawi. r0, N, 3
  240. mtspr CTR, r0
  241. beq- LL(150)
  242. .align 4
  243. LL(110):
  244. LFDX f0, X, INCXM1
  245. LFDUX f1, X, INCX
  246. LFDX f2, X, INCXM1
  247. LFDUX f3, X, INCX
  248. LFDX f16, Y, INCYM1
  249. LFDUX f17, Y, INCY
  250. LFDX f18, Y, INCYM1
  251. LFDUX f19, Y, INCY
  252. LFDX f4, X, INCXM1
  253. LFDUX f5, X, INCX
  254. LFDX f6, X, INCXM1
  255. LFDUX f7, X, INCX
  256. LFDX f20, Y, INCYM1
  257. LFDUX f21, Y, INCY
  258. LFDX f22, Y, INCYM1
  259. LFDUX f23, Y, INCY
  260. LFDX f8, X, INCXM1
  261. LFDUX f9, X, INCX
  262. LFDX f10, X, INCXM1
  263. LFDUX f11, X, INCX
  264. LFDX f24, Y, INCYM1
  265. LFDUX f25, Y, INCY
  266. LFDX f26, Y, INCYM1
  267. LFDUX f27, Y, INCY
  268. LFDX f12, X, INCXM1
  269. LFDUX f13, X, INCX
  270. LFDX f14, X, INCXM1
  271. LFDUX f15, X, INCX
  272. LFDX f28, Y, INCYM1
  273. LFDUX f29, Y, INCY
  274. LFDX f30, Y, INCYM1
  275. LFDUX f31, Y, INCY
  276. STFDX f16, XX, INCXM1
  277. STFDUX f17, XX, INCX
  278. STFDX f18, XX, INCXM1
  279. STFDUX f19, XX, INCX
  280. STFDX f0, YY, INCYM1
  281. STFDUX f1, YY, INCY
  282. STFDX f2, YY, INCYM1
  283. STFDUX f3, YY, INCY
  284. STFDX f20, XX, INCXM1
  285. STFDUX f21, XX, INCX
  286. STFDX f22, XX, INCXM1
  287. STFDUX f23, XX, INCX
  288. STFDX f4, YY, INCYM1
  289. STFDUX f5, YY, INCY
  290. STFDX f6, YY, INCYM1
  291. STFDUX f7, YY, INCY
  292. STFDX f24, XX, INCXM1
  293. STFDUX f25, XX, INCX
  294. STFDX f26, XX, INCXM1
  295. STFDUX f27, XX, INCX
  296. STFDX f8, YY, INCYM1
  297. STFDUX f9, YY, INCY
  298. STFDX f10, YY, INCYM1
  299. STFDUX f11, YY, INCY
  300. STFDX f28, XX, INCXM1
  301. STFDUX f29, XX, INCX
  302. STFDX f30, XX, INCXM1
  303. STFDUX f31, XX, INCX
  304. STFDX f12, YY, INCYM1
  305. STFDUX f13, YY, INCY
  306. STFDX f14, YY, INCYM1
  307. STFDUX f15, YY, INCY
  308. bdnz LL(110)
  309. .align 4
  310. LL(150):
  311. andi. r0, N, 7
  312. mtspr CTR, r0
  313. beq LL(999)
  314. .align 4
  315. LL(160):
  316. LFDX f0, X, INCXM1
  317. LFDUX f1, X, INCX
  318. LFDX f2, Y, INCYM1
  319. LFDUX f3, Y, INCY
  320. STFDX f2, XX, INCXM1
  321. STFDUX f3, XX, INCX
  322. STFDX f0, YY, INCYM1
  323. STFDUX f1, YY, INCY
  324. bdnz LL(160)
  325. .align 4
  326. LL(999):
  327. lfd f14, 0(SP)
  328. lfd f15, 8(SP)
  329. lfd f16, 16(SP)
  330. lfd f17, 24(SP)
  331. lfd f18, 32(SP)
  332. lfd f19, 40(SP)
  333. lfd f20, 48(SP)
  334. lfd f21, 56(SP)
  335. lfd f22, 64(SP)
  336. lfd f23, 72(SP)
  337. lfd f24, 80(SP)
  338. lfd f25, 88(SP)
  339. lfd f26, 96(SP)
  340. lfd f27, 104(SP)
  341. lfd f28, 112(SP)
  342. lfd f29, 120(SP)
  343. lfd f30, 128(SP)
  344. lfd f31, 136(SP)
  345. addi SP, SP, STACKSIZE
  346. blr
  347. EPILOGUE