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.

dot_ppc440.S 7.0 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301
  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 Y r6
  44. #define INCY r7
  45. #define PRE r8
  46. #define FZERO f0
  47. #define STACKSIZE 96
  48. PROLOGUE
  49. PROFCODE
  50. addi SP, SP, -STACKSIZE
  51. li r0, 0
  52. stfd f14, 0(SP)
  53. stfd f15, 8(SP)
  54. stfd f16, 16(SP)
  55. stfd f17, 24(SP)
  56. stfd f18, 32(SP)
  57. stfd f19, 40(SP)
  58. stfd f20, 48(SP)
  59. stfd f21, 56(SP)
  60. stfd f22, 64(SP)
  61. stfd f23, 72(SP)
  62. stw r0, 80(SP)
  63. lfs FZERO, 80(SP)
  64. #ifdef F_INTERFACE
  65. LDINT N, 0(N)
  66. LDINT INCX, 0(INCX)
  67. LDINT INCY, 0(INCY)
  68. #endif
  69. slwi INCX, INCX, BASE_SHIFT
  70. slwi INCY, INCY, BASE_SHIFT
  71. fmr f1, FZERO
  72. fmr f2, FZERO
  73. fmr f3, FZERO
  74. fmr f4, FZERO
  75. fmr f5, FZERO
  76. fmr f6, FZERO
  77. fmr f7, FZERO
  78. li PRE, 3 * 16 * SIZE
  79. cmpwi cr0, N, 0
  80. ble- LL(999)
  81. #ifdef F_INTERFACE
  82. cmpwi cr0, INCX, 0
  83. bge+ LL(102)
  84. subi r0, N, 1
  85. mullw r0, r0, INCX
  86. sub X, X, r0
  87. .align 4
  88. LL(102):
  89. cmpwi cr0, INCY, 0
  90. bge+ LL(104)
  91. subi r0, N, 1
  92. mullw r0, r0, INCY
  93. sub Y, Y, r0
  94. .align 4
  95. LL(104):
  96. #endif
  97. sub X, X, INCX
  98. sub Y, Y, INCY
  99. srawi. r0, N, 4
  100. mtspr CTR, r0
  101. beq- LL(150)
  102. LFDUX f8, X, INCX
  103. LFDUX f16, Y, INCY
  104. LFDUX f9, X, INCX
  105. LFDUX f17, Y, INCY
  106. LFDUX f10, X, INCX
  107. LFDUX f18, Y, INCY
  108. LFDUX f11, X, INCX
  109. LFDUX f19, Y, INCY
  110. LFDUX f12, X, INCX
  111. LFDUX f20, Y, INCY
  112. LFDUX f13, X, INCX
  113. LFDUX f21, Y, INCY
  114. LFDUX f14, X, INCX
  115. LFDUX f22, Y, INCY
  116. LFDUX f15, X, INCX
  117. LFDUX f23, Y, INCY
  118. bdz LL(120)
  119. .align 4
  120. LL(110):
  121. FMADD f0, f8, f16, f0
  122. LFDUX f8, X, INCX
  123. LFDUX f16, Y, INCY
  124. #ifdef PPCG4
  125. dcbt X, PRE
  126. #endif
  127. FMADD f1, f9, f17, f1
  128. LFDUX f9, X, INCX
  129. LFDUX f17, Y, INCY
  130. FMADD f2, f10, f18, f2
  131. LFDUX f10, X, INCX
  132. LFDUX f18, Y, INCY
  133. #ifdef PPCG4
  134. dcbt Y, PRE
  135. #endif
  136. FMADD f3, f11, f19, f3
  137. LFDUX f11, X, INCX
  138. LFDUX f19, Y, INCY
  139. FMADD f4, f12, f20, f4
  140. LFDUX f12, X, INCX
  141. LFDUX f20, Y, INCY
  142. #if defined(PPCG4) && defined(DOUBLE)
  143. dcbt X, PRE
  144. #endif
  145. FMADD f5, f13, f21, f5
  146. LFDUX f13, X, INCX
  147. LFDUX f21, Y, INCY
  148. FMADD f6, f14, f22, f6
  149. LFDUX f14, X, INCX
  150. LFDUX f22, Y, INCY
  151. #if defined(PPCG4) && defined(DOUBLE)
  152. dcbt Y, PRE
  153. #endif
  154. FMADD f7, f15, f23, f7
  155. LFDUX f15, X, INCX
  156. LFDUX f23, Y, INCY
  157. FMADD f0, f8, f16, f0
  158. LFDUX f8, X, INCX
  159. LFDUX f16, Y, INCY
  160. #ifdef PPCG4
  161. dcbt X, PRE
  162. #endif
  163. FMADD f1, f9, f17, f1
  164. LFDUX f9, X, INCX
  165. LFDUX f17, Y, INCY
  166. FMADD f2, f10, f18, f2
  167. LFDUX f10, X, INCX
  168. LFDUX f18, Y, INCY
  169. #ifdef PPCG4
  170. dcbt Y, PRE
  171. #endif
  172. FMADD f3, f11, f19, f3
  173. LFDUX f11, X, INCX
  174. LFDUX f19, Y, INCY
  175. FMADD f4, f12, f20, f4
  176. LFDUX f12, X, INCX
  177. LFDUX f20, Y, INCY
  178. #if defined(PPCG4) && defined(DOUBLE)
  179. dcbt X, PRE
  180. #endif
  181. FMADD f5, f13, f21, f5
  182. LFDUX f13, X, INCX
  183. LFDUX f21, Y, INCY
  184. FMADD f6, f14, f22, f6
  185. LFDUX f14, X, INCX
  186. LFDUX f22, Y, INCY
  187. #if defined(PPCG4) && defined(DOUBLE)
  188. dcbt Y, PRE
  189. #endif
  190. FMADD f7, f15, f23, f7
  191. LFDUX f15, X, INCX
  192. LFDUX f23, Y, INCY
  193. bdnz LL(110)
  194. .align 4
  195. LL(120):
  196. FMADD f0, f8, f16, f0
  197. LFDUX f8, X, INCX
  198. LFDUX f16, Y, INCY
  199. FMADD f1, f9, f17, f1
  200. LFDUX f9, X, INCX
  201. LFDUX f17, Y, INCY
  202. FMADD f2, f10, f18, f2
  203. LFDUX f10, X, INCX
  204. LFDUX f18, Y, INCY
  205. FMADD f3, f11, f19, f3
  206. LFDUX f11, X, INCX
  207. LFDUX f19, Y, INCY
  208. FMADD f4, f12, f20, f4
  209. LFDUX f12, X, INCX
  210. LFDUX f20, Y, INCY
  211. FMADD f5, f13, f21, f5
  212. LFDUX f13, X, INCX
  213. LFDUX f21, Y, INCY
  214. FMADD f6, f14, f22, f6
  215. LFDUX f14, X, INCX
  216. LFDUX f22, Y, INCY
  217. FMADD f7, f15, f23, f7
  218. LFDUX f15, X, INCX
  219. LFDUX f23, Y, INCY
  220. FMADD f0, f8, f16, f0
  221. FMADD f1, f9, f17, f1
  222. FMADD f2, f10, f18, f2
  223. FMADD f3, f11, f19, f3
  224. FMADD f4, f12, f20, f4
  225. FMADD f5, f13, f21, f5
  226. FMADD f6, f14, f22, f6
  227. FMADD f7, f15, f23, f7
  228. .align 4
  229. LL(150):
  230. andi. r0, N, 15
  231. mtspr CTR, r0
  232. beq LL(999)
  233. .align 4
  234. LL(160):
  235. LFDUX f8, X, INCX
  236. LFDUX f16, Y, INCY
  237. FMADD f0, f8, f16, f0
  238. bdnz LL(160)
  239. .align 4
  240. LL(999):
  241. FADD f0, f0, f1
  242. FADD f2, f2, f3
  243. FADD f4, f4, f5
  244. FADD f6, f6, f7
  245. FADD f0, f0, f2
  246. FADD f4, f4, f6
  247. FADD f1, f0, f4
  248. lfd f14, 0(SP)
  249. lfd f15, 8(SP)
  250. lfd f16, 16(SP)
  251. lfd f17, 24(SP)
  252. lfd f18, 32(SP)
  253. lfd f19, 40(SP)
  254. lfd f20, 48(SP)
  255. lfd f21, 56(SP)
  256. lfd f22, 64(SP)
  257. lfd f23, 72(SP)
  258. addi SP, SP, STACKSIZE
  259. blr
  260. EPILOGUE