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.

amin_ppc440.S 7.4 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333
  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 INC1 r6
  44. #define PREX r8
  45. #define STACKSIZE 160
  46. PROLOGUE
  47. PROFCODE
  48. addi SP, SP, -STACKSIZE
  49. li r0, 0
  50. stfd f14, 0(SP)
  51. stfd f15, 8(SP)
  52. stfd f16, 16(SP)
  53. stfd f17, 24(SP)
  54. stfd f18, 32(SP)
  55. stfd f19, 40(SP)
  56. stfd f20, 48(SP)
  57. stfd f21, 56(SP)
  58. stfd f22, 64(SP)
  59. stfd f23, 72(SP)
  60. stfd f24, 80(SP)
  61. stfd f25, 88(SP)
  62. stfd f26, 96(SP)
  63. stfd f27, 104(SP)
  64. stfd f28, 112(SP)
  65. stfd f29, 120(SP)
  66. stfd f30, 128(SP)
  67. stfd f31, 136(SP)
  68. #ifdef F_INTERFACE
  69. LDINT N, 0(N)
  70. LDINT INCX, 0(INCX)
  71. #endif
  72. slwi INCX, INCX, BASE_SHIFT
  73. sub X, X, INCX
  74. li INC1, SIZE
  75. cmpwi cr0, N, 0
  76. ble- LL(9999)
  77. cmpwi cr0, INCX, 0
  78. ble- LL(9999)
  79. LFDUX f1, X, INCX
  80. fabs f0, f1
  81. li PREX, 3 * 16 * SIZE
  82. fabs f2, f1
  83. fabs f3, f1
  84. fabs f4, f1
  85. fabs f5, f1
  86. subi N, N, 1
  87. fabs f6, f1
  88. srawi. r0, N, 4
  89. fabs f7, f1
  90. mtspr CTR, r0
  91. fabs f1, f1
  92. beq- LL(150)
  93. LFDUX f24, X, INCX
  94. LFDUX f25, X, INCX
  95. LFDUX f26, X, INCX
  96. LFDUX f27, X, INCX
  97. LFDUX f28, X, INCX
  98. LFDUX f29, X, INCX
  99. LFDUX f30, X, INCX
  100. LFDUX f31, X, INCX
  101. fabs f8, f24
  102. LFDUX f24, X, INCX
  103. fabs f9, f25
  104. LFDUX f25, X, INCX
  105. fabs f10, f26
  106. LFDUX f26, X, INCX
  107. fabs f11, f27
  108. LFDUX f27, X, INCX
  109. fabs f12, f28
  110. LFDUX f28, X, INCX
  111. fabs f13, f29
  112. LFDUX f29, X, INCX
  113. fabs f14, f30
  114. LFDUX f30, X, INCX
  115. fabs f15, f31
  116. LFDUX f31, X, INCX
  117. bdz LL(120)
  118. .align 4
  119. LL(110):
  120. fsub f16, f0, f8
  121. #ifdef PPCG4
  122. dcbt X, PREX
  123. #endif
  124. fsub f17, f1, f9
  125. fsub f18, f2, f10
  126. fsub f19, f3, f11
  127. fsub f20, f4, f12
  128. fsub f21, f5, f13
  129. fsub f22, f6, f14
  130. fsub f23, f7, f15
  131. fsel f0, f16, f8, f0
  132. fabs f8, f24
  133. LFDUX f24, X, INCX
  134. fsel f1, f17, f9, f1
  135. fabs f9, f25
  136. LFDUX f25, X, INCX
  137. fsel f2, f18, f10, f2
  138. fabs f10, f26
  139. LFDUX f26, X, INCX
  140. fsel f3, f19, f11, f3
  141. fabs f11, f27
  142. LFDUX f27, X, INCX
  143. fsel f4, f20, f12, f4
  144. #if defined(PPCG4) && defined(DOUBLE)
  145. dcbt X, PREX
  146. #endif
  147. fabs f12, f28
  148. LFDUX f28, X, INCX
  149. fsel f5, f21, f13, f5
  150. fabs f13, f29
  151. LFDUX f29, X, INCX
  152. fsel f6, f22, f14, f6
  153. fabs f14, f30
  154. LFDUX f30, X, INCX
  155. fsel f7, f23, f15, f7
  156. fabs f15, f31
  157. LFDUX f31, X, INCX
  158. fsub f16, f0, f8
  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, f8, f0
  167. #ifdef PPCG4
  168. dcbt X, PREX
  169. #endif
  170. fabs f8, f24
  171. LFDUX f24, X, INCX
  172. fsel f1, f17, f9, f1
  173. fabs f9, f25
  174. LFDUX f25, X, INCX
  175. fsel f2, f18, f10, f2
  176. fabs f10, f26
  177. LFDUX f26, X, INCX
  178. fsel f3, f19, f11, f3
  179. fabs f11, f27
  180. LFDUX f27, X, INCX
  181. fsel f4, f20, f12, f4
  182. #if defined(PPCG4) && defined(DOUBLE)
  183. dcbt X, PREX
  184. #endif
  185. fabs f12, f28
  186. LFDUX f28, X, INCX
  187. fsel f5, f21, f13, f5
  188. fabs f13, f29
  189. LFDUX f29, X, INCX
  190. fsel f6, f22, f14, f6
  191. fabs f14, f30
  192. LFDUX f30, X, INCX
  193. fsel f7, f23, f15, f7
  194. fabs f15, f31
  195. LFDUX f31, X, INCX
  196. bdnz LL(110)
  197. .align 4
  198. LL(120):
  199. fsub f16, f0, f8
  200. fsub f17, f1, f9
  201. fsub f18, f2, f10
  202. fsub f19, f3, f11
  203. fsub f20, f4, f12
  204. fsub f21, f5, f13
  205. fsub f22, f6, f14
  206. fsub f23, f7, f15
  207. fsel f0, f16, f8, f0
  208. fabs f8, f24
  209. fsel f1, f17, f9, f1
  210. fabs f9, f25
  211. fsel f2, f18, f10, f2
  212. fabs f10, f26
  213. fsel f3, f19, f11, f3
  214. fabs f11, f27
  215. fsel f4, f20, f12, f4
  216. fabs f12, f28
  217. fsel f5, f21, f13, f5
  218. fabs f13, f29
  219. fsel f6, f22, f14, f6
  220. fabs f14, f30
  221. fsel f7, f23, f15, f7
  222. fabs f15, f31
  223. fsub f16, f0, f8
  224. fsub f17, f1, f9
  225. fsub f18, f2, f10
  226. fsub f19, f3, f11
  227. fsub f20, f4, f12
  228. fsub f21, f5, f13
  229. fsub f22, f6, f14
  230. fsub f23, f7, f15
  231. fsel f0, f16, f8, f0
  232. fsel f1, f17, f9, f1
  233. fsel f2, f18, f10, f2
  234. fsel f3, f19, f11, f3
  235. fsel f4, f20, f12, f4
  236. fsel f5, f21, f13, f5
  237. fsel f6, f22, f14, f6
  238. fsel f7, f23, f15, f7
  239. .align 4
  240. LL(150):
  241. andi. r0, N, 15
  242. mtspr CTR, r0
  243. beq LL(999)
  244. .align 4
  245. LL(160):
  246. LFDUX f8, X, INCX
  247. fabs f8, f8
  248. fsub f16, f1, f8
  249. fsel f1, f16, f8, f1
  250. bdnz LL(160)
  251. .align 4
  252. LL(999):
  253. fsub f8, f0, f1
  254. fsub f9, f2, f3
  255. fsub f10, f4, f5
  256. fsub f11, f6, f7
  257. fsel f0, f8, f1, f0
  258. fsel f2, f9, f3, f2
  259. fsel f4, f10, f5, f4
  260. fsel f6, f11, f7, f6
  261. fsub f8, f0, f2
  262. fsub f9, f4, f6
  263. fsel f0, f8, f2, f0
  264. fsel f4, f9, f6, f4
  265. fsub f8, f0, f4
  266. fsel f1, f8, f4, f0
  267. .align 4
  268. LL(9999):
  269. lfd f14, 0(SP)
  270. lfd f15, 8(SP)
  271. lfd f16, 16(SP)
  272. lfd f17, 24(SP)
  273. lfd f18, 32(SP)
  274. lfd f19, 40(SP)
  275. lfd f20, 48(SP)
  276. lfd f21, 56(SP)
  277. lfd f22, 64(SP)
  278. lfd f23, 72(SP)
  279. lfd f24, 80(SP)
  280. lfd f25, 88(SP)
  281. lfd f26, 96(SP)
  282. lfd f27, 104(SP)
  283. lfd f28, 112(SP)
  284. lfd f29, 120(SP)
  285. lfd f30, 128(SP)
  286. lfd f31, 136(SP)
  287. addi SP, SP, STACKSIZE
  288. blr
  289. EPILOGUE