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.

imin_ppc440.S 8.6 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. #define RET r3
  41. #define X r4
  42. #define INCX r5
  43. #define N r6
  44. #define NN r7
  45. #define XX r8
  46. #define PRE r9
  47. #define FZERO f1
  48. #define STACKSIZE 160
  49. PROLOGUE
  50. PROFCODE
  51. addi SP, SP, -STACKSIZE
  52. li r0, 0
  53. stfd f14, 0(SP)
  54. stfd f15, 8(SP)
  55. stfd f16, 16(SP)
  56. stfd f17, 24(SP)
  57. stfd f18, 32(SP)
  58. stfd f19, 40(SP)
  59. stfd f20, 48(SP)
  60. stfd f21, 56(SP)
  61. stfd f22, 64(SP)
  62. stfd f23, 72(SP)
  63. stfd f24, 80(SP)
  64. stfd f25, 88(SP)
  65. stfd f26, 96(SP)
  66. stfd f27, 104(SP)
  67. stfd f28, 112(SP)
  68. stfd f29, 120(SP)
  69. stfd f30, 128(SP)
  70. stfd f31, 136(SP)
  71. stw r0, 144(SP)
  72. lfs FZERO,144(SP)
  73. #ifdef F_INTERFACE
  74. LDINT N, 0(r3)
  75. LDINT INCX, 0(INCX)
  76. #else
  77. mr N, r3
  78. #endif
  79. li RET, 0
  80. li PRE, 3 * 16 * SIZE
  81. slwi INCX, INCX, BASE_SHIFT
  82. sub X, X, INCX
  83. mr NN, N
  84. mr XX, X
  85. cmpwi cr0, N, 0
  86. ble- LL(9999)
  87. cmpwi cr0, INCX, 0
  88. ble- LL(9999)
  89. LFDUX f1, X, INCX
  90. fmr f0, f1
  91. fmr f2, f1
  92. fmr f3, f1
  93. fmr f4, f1
  94. subi N, N, 1
  95. fmr f5, f1
  96. srawi. r0, N, 4
  97. fmr f6, f1
  98. mtspr CTR, r0
  99. fmr f7, f1
  100. beq- LL(150)
  101. LFDUX f16, X, INCX
  102. LFDUX f17, X, INCX
  103. LFDUX f18, X, INCX
  104. LFDUX f19, X, INCX
  105. LFDUX f20, X, INCX
  106. LFDUX f21, X, INCX
  107. LFDUX f22, X, INCX
  108. LFDUX f23, X, INCX
  109. LFDUX f24, X, INCX
  110. fsub f8, f0, f16
  111. LFDUX f25, X, INCX
  112. fsub f9, f1, f17
  113. LFDUX f26, X, INCX
  114. fsub f10, f2, f18
  115. LFDUX f27, X, INCX
  116. fsub f11, f3, f19
  117. LFDUX f28, X, INCX
  118. fsub f12, f4, f20
  119. LFDUX f29, X, INCX
  120. fsub f13, f5, f21
  121. LFDUX f30, X, INCX
  122. fsub f14, f6, f22
  123. LFDUX f31, X, INCX
  124. fsub f15, f7, f23
  125. bdz LL(120)
  126. .align 4
  127. LL(110):
  128. fsel f0, f8, f16, f0
  129. LFDUX f16, X, INCX
  130. fsub f8, f0, f24
  131. #ifdef PPCG4
  132. dcbt X, PRE
  133. #endif
  134. fsel f1, f9, f17, f1
  135. LFDUX f17, X, INCX
  136. fsub f9, f1, f25
  137. fsel f2, f10, f18, f2
  138. LFDUX f18, X, INCX
  139. fsub f10, f2, f26
  140. fsel f3, f11, f19, f3
  141. LFDUX f19, X, INCX
  142. fsub f11, f3, f27
  143. fsel f4, f12, f20, f4
  144. LFDUX f20, X, INCX
  145. fsub f12, f4, f28
  146. #if defined(PPCG4) && defined(DOUBLE)
  147. dcbt X, PRE
  148. #endif
  149. fsel f5, f13, f21, f5
  150. LFDUX f21, X, INCX
  151. fsub f13, f5, f29
  152. fsel f6, f14, f22, f6
  153. LFDUX f22, X, INCX
  154. fsub f14, f6, f30
  155. fsel f7, f15, f23, f7
  156. LFDUX f23, X, INCX
  157. fsub f15, f7, f31
  158. fsel f0, f8, f24, f0
  159. LFDUX f24, X, INCX
  160. fsub f8, f0, f16
  161. #ifdef PPCG4
  162. dcbt X, PRE
  163. #endif
  164. fsel f1, f9, f25, f1
  165. LFDUX f25, X, INCX
  166. fsub f9, f1, f17
  167. fsel f2, f10, f26, f2
  168. LFDUX f26, X, INCX
  169. fsub f10, f2, f18
  170. fsel f3, f11, f27, f3
  171. LFDUX f27, X, INCX
  172. fsub f11, f3, f19
  173. fsel f4, f12, f28, f4
  174. LFDUX f28, X, INCX
  175. fsub f12, f4, f20
  176. #if defined(PPCG4) && defined(DOUBLE)
  177. dcbt X, PRE
  178. #endif
  179. fsel f5, f13, f29, f5
  180. LFDUX f29, X, INCX
  181. fsub f13, f5, f21
  182. fsel f6, f14, f30, f6
  183. LFDUX f30, X, INCX
  184. fsub f14, f6, f22
  185. fsel f7, f15, f31, f7
  186. LFDUX f31, X, INCX
  187. fsub f15, f7, f23
  188. bdnz LL(110)
  189. .align 4
  190. LL(120):
  191. fsel f0, f8, f16, f0
  192. fsub f8, f0, f24
  193. fsel f1, f9, f17, f1
  194. fsub f9, f1, f25
  195. fsel f2, f10, f18, f2
  196. fsub f10, f2, f26
  197. fsel f3, f11, f19, f3
  198. fsub f11, f3, f27
  199. fsel f4, f12, f20, f4
  200. fsub f12, f4, f28
  201. fsel f5, f13, f21, f5
  202. fsub f13, f5, f29
  203. fsel f6, f14, f22, f6
  204. fsub f14, f6, f30
  205. fsel f7, f15, f23, f7
  206. fsub f15, f7, f31
  207. fsel f0, f8, f24, f0
  208. fsel f1, f9, f25, f1
  209. fsel f2, f10, f26, f2
  210. fsel f3, f11, f27, f3
  211. fsel f4, f12, f28, f4
  212. fsel f5, f13, f29, f5
  213. fsel f6, f14, f30, f6
  214. fsel f7, f15, f31, f7
  215. .align 4
  216. LL(150):
  217. andi. r0, N, 15
  218. mtspr CTR, r0
  219. beq LL(999)
  220. .align 4
  221. LL(160):
  222. LFDUX f8, X, INCX
  223. fsub f16, f1, f8
  224. fsel f1, f16, f8, f1
  225. bdnz LL(160)
  226. .align 4
  227. LL(999):
  228. fsub f8, f0, f1
  229. fsub f9, f2, f3
  230. fsub f10, f4, f5
  231. fsub f11, f6, f7
  232. fsel f0, f8, f1, f0
  233. fsel f2, f9, f3, f2
  234. fsel f4, f10, f5, f4
  235. fsel f6, f11, f7, f6
  236. fsub f8, f0, f2
  237. fsub f9, f4, f6
  238. fsel f0, f8, f2, f0
  239. fsel f4, f9, f6, f4
  240. fsub f8, f0, f4
  241. fsel f1, f8, f4, f0
  242. .align 4
  243. LL(1000):
  244. srawi. r0, NN, 3
  245. mtspr CTR, r0
  246. beq- LL(1150)
  247. LFDUX f8, XX, INCX
  248. LFDUX f9, XX, INCX
  249. LFDUX f10, XX, INCX
  250. LFDUX f11, XX, INCX
  251. LFDUX f12, XX, INCX
  252. LFDUX f13, XX, INCX
  253. LFDUX f14, XX, INCX
  254. LFDUX f15, XX, INCX
  255. bdz LL(1120)
  256. .align 4
  257. LL(1110):
  258. addi RET, RET, 1
  259. fcmpu cr0, f1, f8
  260. LFDUX f8, XX, INCX
  261. beq cr0, LL(9999)
  262. addi RET, RET, 1
  263. fcmpu cr0, f1, f9
  264. LFDUX f9, XX, INCX
  265. beq cr0, LL(9999)
  266. addi RET, RET, 1
  267. fcmpu cr0, f1, f10
  268. LFDUX f10, XX, INCX
  269. beq cr0, LL(9999)
  270. addi RET, RET, 1
  271. fcmpu cr0, f1, f11
  272. LFDUX f11, XX, INCX
  273. beq cr0, LL(9999)
  274. addi RET, RET, 1
  275. fcmpu cr0, f1, f12
  276. LFDUX f12, XX, INCX
  277. beq cr0, LL(9999)
  278. addi RET, RET, 1
  279. fcmpu cr0, f1, f13
  280. LFDUX f13, XX, INCX
  281. beq cr0, LL(9999)
  282. addi RET, RET, 1
  283. fcmpu cr0, f1, f14
  284. LFDUX f14, XX, INCX
  285. beq cr0, LL(9999)
  286. addi RET, RET, 1
  287. fcmpu cr0, f1, f15
  288. LFDUX f15, XX, INCX
  289. beq cr0, LL(9999)
  290. bdnz LL(1110)
  291. .align 4
  292. LL(1120):
  293. addi RET, RET, 1
  294. fcmpu cr0, f1, f8
  295. beq cr0, LL(9999)
  296. addi RET, RET, 1
  297. fcmpu cr0, f1, f9
  298. beq cr0, LL(9999)
  299. addi RET, RET, 1
  300. fcmpu cr0, f1, f10
  301. beq cr0, LL(9999)
  302. addi RET, RET, 1
  303. fcmpu cr0, f1, f11
  304. beq cr0, LL(9999)
  305. addi RET, RET, 1
  306. fcmpu cr0, f1, f12
  307. beq cr0, LL(9999)
  308. addi RET, RET, 1
  309. fcmpu cr0, f1, f13
  310. beq cr0, LL(9999)
  311. addi RET, RET, 1
  312. fcmpu cr0, f1, f14
  313. beq cr0, LL(9999)
  314. addi RET, RET, 1
  315. fcmpu cr0, f1, f15
  316. beq cr0, LL(9999)
  317. .align 4
  318. LL(1150):
  319. andi. r0, NN, 7
  320. mtspr CTR, r0
  321. beq LL(9999)
  322. .align 4
  323. LL(1160):
  324. LFDUX f8, XX, INCX
  325. addi RET, RET, 1
  326. fcmpu cr0, f1, f8
  327. beq cr0, LL(9999)
  328. bdnz LL(1160)
  329. .align 4
  330. LL(9999):
  331. lfd f14, 0(SP)
  332. lfd f15, 8(SP)
  333. lfd f16, 16(SP)
  334. lfd f17, 24(SP)
  335. lfd f18, 32(SP)
  336. lfd f19, 40(SP)
  337. lfd f20, 48(SP)
  338. lfd f21, 56(SP)
  339. lfd f22, 64(SP)
  340. lfd f23, 72(SP)
  341. lfd f24, 80(SP)
  342. lfd f25, 88(SP)
  343. lfd f26, 96(SP)
  344. lfd f27, 104(SP)
  345. lfd f28, 112(SP)
  346. lfd f29, 120(SP)
  347. lfd f30, 128(SP)
  348. lfd f31, 136(SP)
  349. addi SP, SP, STACKSIZE
  350. blr
  351. EPILOGUE