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.

amax_ppc440.S 7.3 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332
  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 N r3
  41. #define X r4
  42. #define INCX r5
  43. #define PREX r8
  44. #define STACKSIZE 160
  45. PROLOGUE
  46. PROFCODE
  47. addi SP, SP, -STACKSIZE
  48. stfd f14, 0(SP)
  49. stfd f15, 8(SP)
  50. stfd f16, 16(SP)
  51. stfd f17, 24(SP)
  52. stfd f18, 32(SP)
  53. stfd f19, 40(SP)
  54. stfd f20, 48(SP)
  55. stfd f21, 56(SP)
  56. stfd f22, 64(SP)
  57. stfd f23, 72(SP)
  58. stfd f24, 80(SP)
  59. stfd f25, 88(SP)
  60. stfd f26, 96(SP)
  61. stfd f27, 104(SP)
  62. stfd f28, 112(SP)
  63. stfd f29, 120(SP)
  64. stfd f30, 128(SP)
  65. stfd f31, 136(SP)
  66. #ifdef F_INTERFACE
  67. LDINT N, 0(N)
  68. LDINT INCX, 0(INCX)
  69. #endif
  70. slwi INCX, INCX, BASE_SHIFT
  71. sub X, X, INCX
  72. cmpwi cr0, N, 0
  73. ble- LL(9999)
  74. cmpwi cr0, INCX, 0
  75. ble- LL(9999)
  76. LFDUX f1, X, INCX
  77. fabs f0, f1
  78. li PREX, 3 * 16 * SIZE
  79. fabs f2, f1
  80. fabs f3, f1
  81. fabs f4, f1
  82. fabs f5, f1
  83. fabs f6, f1
  84. fabs f7, f1
  85. fabs f1, f1
  86. subi N, N, 1
  87. srawi. r0, N, 4
  88. mtspr CTR, r0
  89. beq- LL(150)
  90. LFDUX f24, X, INCX
  91. LFDUX f25, X, INCX
  92. LFDUX f26, X, INCX
  93. LFDUX f27, X, INCX
  94. LFDUX f28, X, INCX
  95. LFDUX f29, X, INCX
  96. LFDUX f30, X, INCX
  97. LFDUX f31, X, INCX
  98. fabs f8, f24
  99. LFDUX f24, X, INCX
  100. fabs f9, f25
  101. LFDUX f25, X, INCX
  102. fabs f10, f26
  103. LFDUX f26, X, INCX
  104. fabs f11, f27
  105. LFDUX f27, X, INCX
  106. fabs f12, f28
  107. LFDUX f28, X, INCX
  108. fabs f13, f29
  109. LFDUX f29, X, INCX
  110. fabs f14, f30
  111. LFDUX f30, X, INCX
  112. fabs f15, f31
  113. LFDUX f31, X, INCX
  114. bdz LL(120)
  115. .align 4
  116. LL(110):
  117. fsub f16, f0, f8
  118. #ifdef PPCG4
  119. dcbt X, PREX
  120. #endif
  121. fsub f17, f1, f9
  122. fsub f18, f2, f10
  123. fsub f19, f3, f11
  124. fsub f20, f4, f12
  125. fsub f21, f5, f13
  126. fsub f22, f6, f14
  127. fsub f23, f7, f15
  128. fsel f0, f16, f0, f8
  129. fabs f8, f24
  130. LFDUX f24, X, INCX
  131. fsel f1, f17, f1, f9
  132. fabs f9, f25
  133. LFDUX f25, X, INCX
  134. fsel f2, f18, f2, f10
  135. fabs f10, f26
  136. LFDUX f26, X, INCX
  137. fsel f3, f19, f3, f11
  138. fabs f11, f27
  139. LFDUX f27, X, INCX
  140. fsel f4, f20, f4, f12
  141. #ifdef PPCG4
  142. dcbt X, PREX
  143. #endif
  144. fabs f12, f28
  145. LFDUX f28, X, INCX
  146. fsel f5, f21, f5, f13
  147. fabs f13, f29
  148. LFDUX f29, X, INCX
  149. fsel f6, f22, f6, f14
  150. fabs f14, f30
  151. LFDUX f30, X, INCX
  152. fsel f7, f23, f7, f15
  153. fabs f15, f31
  154. LFDUX f31, X, INCX
  155. fsub f16, f0, f8
  156. #ifdef PPCG4
  157. dcbt X, PREX
  158. #endif
  159. fsub f17, f1, f9
  160. fsub f18, f2, f10
  161. fsub f19, f3, f11
  162. fsub f20, f4, f12
  163. fsub f21, f5, f13
  164. fsub f22, f6, f14
  165. fsub f23, f7, f15
  166. fsel f0, f16, f0, f8
  167. fabs f8, f24
  168. LFDUX f24, X, INCX
  169. fsel f1, f17, f1, f9
  170. fabs f9, f25
  171. LFDUX f25, X, INCX
  172. fsel f2, f18, f2, f10
  173. fabs f10, f26
  174. LFDUX f26, X, INCX
  175. fsel f3, f19, f3, f11
  176. fabs f11, f27
  177. LFDUX f27, X, INCX
  178. fsel f4, f20, f4, f12
  179. #ifdef PPCG4
  180. dcbt X, PREX
  181. #endif
  182. fabs f12, f28
  183. LFDUX f28, X, INCX
  184. fsel f5, f21, f5, f13
  185. fabs f13, f29
  186. LFDUX f29, X, INCX
  187. fsel f6, f22, f6, f14
  188. fabs f14, f30
  189. LFDUX f30, X, INCX
  190. fsel f7, f23, f7, f15
  191. fabs f15, f31
  192. LFDUX f31, X, INCX
  193. bdnz LL(110)
  194. .align 4
  195. LL(120):
  196. fsub f16, f0, f8
  197. fsub f17, f1, f9
  198. fsub f18, f2, f10
  199. fsub f19, f3, f11
  200. fsub f20, f4, f12
  201. fsub f21, f5, f13
  202. fsub f22, f6, f14
  203. fsub f23, f7, f15
  204. fsel f0, f16, f0, f8
  205. fabs f8, f24
  206. fsel f1, f17, f1, f9
  207. fabs f9, f25
  208. fsel f2, f18, f2, f10
  209. fabs f10, f26
  210. fsel f3, f19, f3, f11
  211. fabs f11, f27
  212. fsel f4, f20, f4, f12
  213. fabs f12, f28
  214. fsel f5, f21, f5, f13
  215. fabs f13, f29
  216. fsel f6, f22, f6, f14
  217. fabs f14, f30
  218. fsel f7, f23, f7, f15
  219. fabs f15, f31
  220. fsub f16, f0, f8
  221. fsub f17, f1, f9
  222. fsub f18, f2, f10
  223. fsub f19, f3, f11
  224. fsub f20, f4, f12
  225. fsub f21, f5, f13
  226. fsub f22, f6, f14
  227. fsub f23, f7, f15
  228. fsel f0, f16, f0, f8
  229. fsel f1, f17, f1, f9
  230. fsel f2, f18, f2, f10
  231. fsel f3, f19, f3, f11
  232. fsel f4, f20, f4, f12
  233. fsel f5, f21, f5, f13
  234. fsel f6, f22, f6, f14
  235. fsel f7, f23, f7, f15
  236. .align 4
  237. LL(150):
  238. andi. r0, N, 15
  239. mtspr CTR, r0
  240. beq LL(999)
  241. .align 4
  242. LL(160):
  243. LFDUX f8, X, INCX
  244. fabs f8, f8
  245. fsub f16, f1, f8
  246. fsel f1, f16, f1, f8
  247. bdnz LL(160)
  248. .align 4
  249. LL(999):
  250. fsub f8, f0, f1
  251. fsub f9, f2, f3
  252. fsub f10, f4, f5
  253. fsub f11, f6, f7
  254. fsel f0, f8, f0, f1
  255. fsel f2, f9, f2, f3
  256. fsel f4, f10, f4, f5
  257. fsel f6, f11, f6, f7
  258. fsub f8, f0, f2
  259. fsub f9, f4, f6
  260. fsel f0, f8, f0, f2
  261. fsel f4, f9, f4, f6
  262. fsub f8, f0, f4
  263. fsel f1, f8, f0, f4
  264. .align 4
  265. LL(9999):
  266. lfd f14, 0(SP)
  267. lfd f15, 8(SP)
  268. lfd f16, 16(SP)
  269. lfd f17, 24(SP)
  270. lfd f18, 32(SP)
  271. lfd f19, 40(SP)
  272. lfd f20, 48(SP)
  273. lfd f21, 56(SP)
  274. lfd f22, 64(SP)
  275. lfd f23, 72(SP)
  276. lfd f24, 80(SP)
  277. lfd f25, 88(SP)
  278. lfd f26, 96(SP)
  279. lfd f27, 104(SP)
  280. lfd f28, 112(SP)
  281. lfd f29, 120(SP)
  282. lfd f30, 128(SP)
  283. lfd f31, 136(SP)
  284. addi SP, SP, STACKSIZE
  285. blr
  286. EPILOGUE