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.

min_ppc440.S 6.7 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284
  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 PREA r8
  44. #define FZERO f1
  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. stw r0, 144(SP)
  69. lfs FZERO,144(SP)
  70. #ifdef F_INTERFACE
  71. LDINT N, 0(N)
  72. LDINT INCX, 0(INCX)
  73. #endif
  74. slwi INCX, INCX, BASE_SHIFT
  75. sub X, X, INCX
  76. cmpwi cr0, N, 0
  77. ble- LL(9999)
  78. cmpwi cr0, INCX, 0
  79. ble- LL(9999)
  80. LFDUX f1, X, INCX
  81. fmr f0, f1
  82. subi N, N, 1
  83. fmr f2, f1
  84. fmr f3, f1
  85. fmr f4, f1
  86. fmr f5, f1
  87. srawi. r0, N, 4
  88. fmr f6, f1
  89. mtspr CTR, r0
  90. fmr f7, f1
  91. beq- LL(150)
  92. LFDUX f16, X, INCX
  93. LFDUX f17, X, INCX
  94. LFDUX f18, X, INCX
  95. LFDUX f19, X, INCX
  96. LFDUX f20, X, INCX
  97. LFDUX f21, X, INCX
  98. LFDUX f22, X, INCX
  99. LFDUX f23, X, INCX
  100. LFDUX f24, X, INCX
  101. fsub f8, f0, f16
  102. LFDUX f25, X, INCX
  103. fsub f9, f1, f17
  104. LFDUX f26, X, INCX
  105. fsub f10, f2, f18
  106. LFDUX f27, X, INCX
  107. fsub f11, f3, f19
  108. LFDUX f28, X, INCX
  109. fsub f12, f4, f20
  110. LFDUX f29, X, INCX
  111. fsub f13, f5, f21
  112. LFDUX f30, X, INCX
  113. fsub f14, f6, f22
  114. LFDUX f31, X, INCX
  115. fsub f15, f7, f23
  116. bdz LL(120)
  117. .align 4
  118. LL(110):
  119. fsel f0, f8, f16, f0
  120. LFDUX f16, X, INCX
  121. fsub f8, f0, f24
  122. fsel f1, f9, f17, f1
  123. LFDUX f17, X, INCX
  124. fsub f9, f1, f25
  125. fsel f2, f10, f18, f2
  126. LFDUX f18, X, INCX
  127. fsub f10, f2, f26
  128. fsel f3, f11, f19, f3
  129. LFDUX f19, X, INCX
  130. fsub f11, f3, f27
  131. fsel f4, f12, f20, f4
  132. LFDUX f20, X, INCX
  133. fsub f12, f4, f28
  134. fsel f5, f13, f21, f5
  135. LFDUX f21, X, INCX
  136. fsub f13, f5, f29
  137. fsel f6, f14, f22, f6
  138. LFDUX f22, X, INCX
  139. fsub f14, f6, f30
  140. fsel f7, f15, f23, f7
  141. LFDUX f23, X, INCX
  142. fsub f15, f7, f31
  143. fsel f0, f8, f24, f0
  144. LFDUX f24, X, INCX
  145. fsub f8, f0, f16
  146. fsel f1, f9, f25, f1
  147. LFDUX f25, X, INCX
  148. fsub f9, f1, f17
  149. fsel f2, f10, f26, f2
  150. LFDUX f26, X, INCX
  151. fsub f10, f2, f18
  152. fsel f3, f11, f27, f3
  153. LFDUX f27, X, INCX
  154. fsub f11, f3, f19
  155. fsel f4, f12, f28, f4
  156. LFDUX f28, X, INCX
  157. fsub f12, f4, f20
  158. fsel f5, f13, f29, f5
  159. LFDUX f29, X, INCX
  160. fsub f13, f5, f21
  161. fsel f6, f14, f30, f6
  162. LFDUX f30, X, INCX
  163. fsub f14, f6, f22
  164. fsel f7, f15, f31, f7
  165. LFDUX f31, X, INCX
  166. fsub f15, f7, f23
  167. bdnz LL(110)
  168. .align 4
  169. LL(120):
  170. fsel f0, f8, f16, f0
  171. fsub f8, f0, f24
  172. fsel f1, f9, f17, f1
  173. fsub f9, f1, f25
  174. fsel f2, f10, f18, f2
  175. fsub f10, f2, f26
  176. fsel f3, f11, f19, f3
  177. fsub f11, f3, f27
  178. fsel f4, f12, f20, f4
  179. fsub f12, f4, f28
  180. fsel f5, f13, f21, f5
  181. fsub f13, f5, f29
  182. fsel f6, f14, f22, f6
  183. fsub f14, f6, f30
  184. fsel f7, f15, f23, f7
  185. fsub f15, f7, f31
  186. fsel f0, f8, f24, f0
  187. fsel f1, f9, f25, f1
  188. fsel f2, f10, f26, f2
  189. fsel f3, f11, f27, f3
  190. fsel f4, f12, f28, f4
  191. fsel f5, f13, f29, f5
  192. fsel f6, f14, f30, f6
  193. fsel f7, f15, f31, f7
  194. .align 4
  195. LL(150):
  196. andi. r0, N, 15
  197. mtspr CTR, r0
  198. beq LL(999)
  199. .align 4
  200. LL(160):
  201. LFDUX f8, X, INCX
  202. fsub f16, f1, f8
  203. fsel f1, f16, f8, f1
  204. bdnz LL(160)
  205. .align 4
  206. LL(999):
  207. fsub f8, f0, f1
  208. fsub f9, f2, f3
  209. fsub f10, f4, f5
  210. fsub f11, f6, f7
  211. fsel f0, f8, f1, f0
  212. fsel f2, f9, f3, f2
  213. fsel f4, f10, f5, f4
  214. fsel f6, f11, f7, f6
  215. fsub f8, f0, f2
  216. fsub f9, f4, f6
  217. fsel f0, f8, f2, f0
  218. fsel f4, f9, f6, f4
  219. fsub f8, f0, f4
  220. fsel f1, f8, f4, f0
  221. .align 4
  222. LL(9999):
  223. lfd f14, 0(SP)
  224. lfd f15, 8(SP)
  225. lfd f16, 16(SP)
  226. lfd f17, 24(SP)
  227. lfd f18, 32(SP)
  228. lfd f19, 40(SP)
  229. lfd f20, 48(SP)
  230. lfd f21, 56(SP)
  231. lfd f22, 64(SP)
  232. lfd f23, 72(SP)
  233. lfd f24, 80(SP)
  234. lfd f25, 88(SP)
  235. lfd f26, 96(SP)
  236. lfd f27, 104(SP)
  237. lfd f28, 112(SP)
  238. lfd f29, 120(SP)
  239. lfd f30, 128(SP)
  240. lfd f31, 136(SP)
  241. addi SP, SP, STACKSIZE
  242. blr
  243. EPILOGUE