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.

zamin_ppc440.S 6.8 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317
  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 INC1 r9
  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, ZBASE_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. LFDX f2, X, INC1
  81. fabs f1, f1
  82. li PREX, 4 * 8 * SIZE
  83. fabs f2, f2
  84. subi N, N, 1
  85. fadd f1, f1, f2
  86. fmr f0, f1
  87. srawi. r0, N, 3
  88. fmr f2, f1
  89. mtspr CTR, r0
  90. fmr f3, f1
  91. beq- LL(150)
  92. LFDUX f24, X, INCX
  93. LFDX f25, X, INC1
  94. LFDUX f26, X, INCX
  95. LFDX f27, X, INC1
  96. LFDUX f28, X, INCX
  97. LFDX f29, X, INC1
  98. LFDUX f30, X, INCX
  99. LFDX f31, X, INC1
  100. fabs f8, f24
  101. LFDUX f24, X, INCX
  102. fabs f9, f25
  103. LFDX f25, X, INC1
  104. fabs f10, f26
  105. LFDUX f26, X, INCX
  106. fabs f11, f27
  107. LFDX f27, X, INC1
  108. fabs f12, f28
  109. LFDUX f28, X, INCX
  110. fabs f13, f29
  111. LFDX f29, X, INC1
  112. fabs f14, f30
  113. LFDUX f30, X, INCX
  114. fabs f15, f31
  115. LFDX f31, X, INC1
  116. bdz LL(120)
  117. .align 4
  118. LL(110):
  119. fadd f4, f8, f9
  120. #ifdef PPCG4
  121. dcbt X, PREX
  122. #endif
  123. fadd f5, f10, f11
  124. fadd f6, f12, f13
  125. fadd f7, f14, f15
  126. fabs f8, f24
  127. LFDUX f24, X, INCX
  128. fabs f9, f25
  129. LFDX f25, X, INC1
  130. fabs f10, f26
  131. LFDUX f26, X, INCX
  132. fabs f11, f27
  133. LFDX f27, X, INC1
  134. fabs f12, f28
  135. #ifdef PPCG4
  136. dcbt X, PREX
  137. #endif
  138. fabs f13, f29
  139. LFDUX f28, X, INCX
  140. fabs f14, f30
  141. LFDX f29, X, INC1
  142. fabs f15, f31
  143. LFDUX f30, X, INCX
  144. fsub f16, f0, f4
  145. LFDX f31, X, INC1
  146. fsub f17, f1, f5
  147. fsub f18, f2, f6
  148. fsub f19, f3, f7
  149. fadd f20, f8, f9
  150. #ifdef PPCG4
  151. dcbt X, PREX
  152. #endif
  153. fadd f21, f10, f11
  154. fadd f22, f12, f13
  155. fadd f23, f14, f15
  156. fabs f8, f24
  157. LFDUX f24, X, INCX
  158. fabs f9, f25
  159. LFDX f25, X, INC1
  160. fabs f10, f26
  161. LFDUX f26, X, INCX
  162. fabs f11, f27
  163. LFDX f27, X, INC1
  164. fsel f0, f16, f4, f0
  165. #ifdef PPCG4
  166. dcbt X, PREX
  167. #endif
  168. fsel f1, f17, f5, f1
  169. fsel f2, f18, f6, f2
  170. fsel f3, f19, f7, f3
  171. fabs f12, f28
  172. LFDUX f28, X, INCX
  173. fabs f13, f29
  174. LFDX f29, X, INC1
  175. fabs f14, f30
  176. LFDUX f30, X, INCX
  177. fabs f15, f31
  178. LFDX f31, X, INC1
  179. fsub f16, f0, f20
  180. fsub f17, f1, f21
  181. fsub f18, f2, f22
  182. fsub f19, f3, f23
  183. fsel f0, f16, f20, f0
  184. fsel f1, f17, f21, f1
  185. fsel f2, f18, f22, f2
  186. fsel f3, f19, f23, f3
  187. bdnz LL(110)
  188. .align 4
  189. LL(120):
  190. fadd f4, f8, f9
  191. fadd f5, f10, f11
  192. fadd f6, f12, f13
  193. fadd f7, f14, f15
  194. fabs f8, f24
  195. fabs f9, f25
  196. fabs f10, f26
  197. fabs f11, f27
  198. fabs f12, f28
  199. fabs f13, f29
  200. fabs f14, f30
  201. fabs f15, f31
  202. fsub f16, f0, f4
  203. fsub f17, f1, f5
  204. fsub f18, f2, f6
  205. fsub f19, f3, f7
  206. fadd f20, f8, f9
  207. fadd f21, f10, f11
  208. fadd f22, f12, f13
  209. fadd f23, f14, f15
  210. fsel f0, f16, f4, f0
  211. fsel f1, f17, f5, f1
  212. fsel f2, f18, f6, f2
  213. fsel f3, f19, f7, f3
  214. fsub f16, f0, f20
  215. fsub f17, f1, f21
  216. fsub f18, f2, f22
  217. fsub f19, f3, f23
  218. fsel f0, f16, f20, f0
  219. fsel f1, f17, f21, f1
  220. fsel f2, f18, f22, f2
  221. fsel f3, f19, f23, f3
  222. .align 4
  223. LL(150):
  224. andi. r0, N, 7
  225. mtspr CTR, r0
  226. beq LL(999)
  227. .align 4
  228. LL(160):
  229. LFDUX f8, X, INCX
  230. LFDX f9, X, INC1
  231. fabs f8, f8
  232. fabs f9, f9
  233. fadd f8, f8, f9
  234. fsub f16, f1, f8
  235. fsel f1, f16, f8, f1
  236. bdnz LL(160)
  237. .align 4
  238. LL(999):
  239. fsub f8, f0, f1
  240. fsub f9, f2, f3
  241. fsel f0, f8, f1, f0
  242. fsel f2, f9, f3, f2
  243. fsub f8, f0, f2
  244. fsel f1, f8, f2, f0
  245. .align 4
  246. LL(9999):
  247. lfd f14, 0(SP)
  248. lfd f15, 8(SP)
  249. lfd f16, 16(SP)
  250. lfd f17, 24(SP)
  251. lfd f18, 32(SP)
  252. lfd f19, 40(SP)
  253. lfd f20, 48(SP)
  254. lfd f21, 56(SP)
  255. lfd f22, 64(SP)
  256. lfd f23, 72(SP)
  257. lfd f24, 80(SP)
  258. lfd f25, 88(SP)
  259. lfd f26, 96(SP)
  260. lfd f27, 104(SP)
  261. lfd f28, 112(SP)
  262. lfd f29, 120(SP)
  263. lfd f30, 128(SP)
  264. lfd f31, 136(SP)
  265. addi SP, SP, STACKSIZE
  266. blr
  267. EPILOGUE