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.

snrm2_ppc440.S 6.6 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 PRE r8
  44. #define FZERO 144(SP)
  45. #define FONE 148(SP)
  46. #define C1 152(SP)
  47. #define C2 156(SP)
  48. #define STACKSIZE 160
  49. PROLOGUE
  50. PROFCODE
  51. addi SP, SP, -STACKSIZE
  52. li r10, 0
  53. lis r11, 0x3f80
  54. lis r6, 0x3f00
  55. lis r7, 0x4040
  56. stfd f14, 0(SP)
  57. stfd f15, 8(SP)
  58. stfd f16, 16(SP)
  59. stfd f17, 24(SP)
  60. stfd f18, 32(SP)
  61. stfd f19, 40(SP)
  62. stfd f20, 48(SP)
  63. stfd f21, 56(SP)
  64. stfd f22, 64(SP)
  65. stfd f23, 72(SP)
  66. stfd f24, 80(SP)
  67. stfd f25, 88(SP)
  68. stfd f26, 96(SP)
  69. stfd f27, 104(SP)
  70. stfd f28, 112(SP)
  71. stfd f29, 120(SP)
  72. stfd f30, 128(SP)
  73. stfd f31, 136(SP)
  74. stw r10, FZERO
  75. stw r11, FONE
  76. stw r6, C1
  77. stw r7, C2
  78. lfs f1, FZERO
  79. #ifdef F_INTERFACE
  80. LDINT N, 0(N)
  81. LDINT INCX, 0(INCX)
  82. #endif
  83. slwi INCX, INCX, BASE_SHIFT
  84. li PRE, 3 * 16 * SIZE
  85. sub X, X, INCX
  86. cmpwi cr0, N, 0
  87. ble- LL(999)
  88. cmpwi cr0, INCX, 0
  89. ble- LL(999)
  90. fmr f0, f1
  91. fmr f2, f1
  92. fmr f3, f1
  93. fmr f4, f1
  94. fmr f5, f1
  95. fmr f6, f1
  96. fmr f7, f1
  97. fmr f8, f1
  98. fmr f9, f1
  99. fmr f10, f1
  100. fmr f11, f1
  101. fmr f12, f1
  102. fmr f13, f1
  103. fmr f14, f1
  104. fmr f15, f1
  105. srawi. r0, N, 4
  106. mtspr CTR, r0
  107. beq- cr0, LL(50)
  108. LFDUX f16, X, INCX
  109. LFDUX f17, X, INCX
  110. LFDUX f18, X, INCX
  111. LFDUX f19, X, INCX
  112. LFDUX f20, X, INCX
  113. LFDUX f21, X, INCX
  114. LFDUX f22, X, INCX
  115. LFDUX f23, X, INCX
  116. LFDUX f24, X, INCX
  117. LFDUX f25, X, INCX
  118. LFDUX f26, X, INCX
  119. LFDUX f27, X, INCX
  120. LFDUX f28, X, INCX
  121. LFDUX f29, X, INCX
  122. LFDUX f30, X, INCX
  123. LFDUX f31, X, INCX
  124. bdz LL(20)
  125. .align 4
  126. LL(10):
  127. fmadd f0, f16, f16, f0
  128. LFDUX f16, X, INCX
  129. fmadd f1, f17, f17, f1
  130. LFDUX f17, X, INCX
  131. fmadd f2, f18, f18, f2
  132. LFDUX f18, X, INCX
  133. fmadd f3, f19, f19, f3
  134. LFDUX f19, X, INCX
  135. #ifdef PPCG4
  136. dcbt X, PRE
  137. #endif
  138. fmadd f4, f20, f20, f4
  139. LFDUX f20, X, INCX
  140. fmadd f5, f21, f21, f5
  141. LFDUX f21, X, INCX
  142. fmadd f6, f22, f22, f6
  143. LFDUX f22, X, INCX
  144. fmadd f7, f23, f23, f7
  145. LFDUX f23, X, INCX
  146. fmadd f8, f24, f24, f8
  147. LFDUX f24, X, INCX
  148. fmadd f9, f25, f25, f9
  149. LFDUX f25, X, INCX
  150. fmadd f10, f26, f26, f10
  151. LFDUX f26, X, INCX
  152. fmadd f11, f27, f27, f11
  153. LFDUX f27, X, INCX
  154. #ifdef PPCG4
  155. dcbt X, PRE
  156. #endif
  157. fmadd f12, f28, f28, f12
  158. LFDUX f28, X, INCX
  159. fmadd f13, f29, f29, f13
  160. LFDUX f29, X, INCX
  161. fmadd f14, f30, f30, f14
  162. LFDUX f30, X, INCX
  163. fmadd f15, f31, f31, f15
  164. LFDUX f31, X, INCX
  165. bdnz LL(10)
  166. .align 4
  167. LL(20):
  168. fmadd f0, f16, f16, f0
  169. fmadd f1, f17, f17, f1
  170. fmadd f2, f18, f18, f2
  171. fmadd f3, f19, f19, f3
  172. fmadd f4, f20, f20, f4
  173. fmadd f5, f21, f21, f5
  174. fmadd f6, f22, f22, f6
  175. fmadd f7, f23, f23, f7
  176. fmadd f8, f24, f24, f8
  177. fmadd f9, f25, f25, f9
  178. fmadd f10, f26, f26, f10
  179. fmadd f11, f27, f27, f11
  180. fmadd f12, f28, f28, f12
  181. fmadd f13, f29, f29, f13
  182. fmadd f14, f30, f30, f14
  183. fmadd f15, f31, f31, f15
  184. .align 4
  185. LL(50):
  186. andi. r0, N, 15
  187. mtspr CTR, r0
  188. beq- cr0, LL(70)
  189. .align 4
  190. LL(60):
  191. LFDUX f16, X, INCX
  192. fmadd f0, f16, f16, f0
  193. bdnz LL(60)
  194. .align 4
  195. LL(70):
  196. fadd f0, f0, f1
  197. fadd f2, f2, f3
  198. fadd f4, f4, f5
  199. fadd f6, f6, f7
  200. fadd f8, f8, f9
  201. fadd f10, f10, f11
  202. fadd f12, f12, f13
  203. fadd f14, f14, f15
  204. fadd f0, f0, f2
  205. fadd f4, f4, f6
  206. fadd f8, f8, f10
  207. fadd f12, f12, f14
  208. fadd f0, f0, f4
  209. fadd f8, f8, f12
  210. fadd f1, f0, f8
  211. lfs f4, FZERO
  212. fcmpu cr0, f1, f4
  213. beq cr0, LL(999)
  214. frsqrte f0, f1
  215. lfs f8, C1
  216. lfs f9, C2
  217. fmul f2, f1, f0
  218. fadd f7, f8, f8
  219. fmul f3, f0, f8
  220. fnmsub f4, f2, f0, f9
  221. fmul f0, f3, f4
  222. fmul f5, f1, f0
  223. fmul f2, f5, f8
  224. fnmsub f3, f5, f0, f7
  225. fmadd f1, f2, f3, f5
  226. .align 4
  227. LL(999):
  228. lfd f14, 0(SP)
  229. lfd f15, 8(SP)
  230. lfd f16, 16(SP)
  231. lfd f17, 24(SP)
  232. lfd f18, 32(SP)
  233. lfd f19, 40(SP)
  234. lfd f20, 48(SP)
  235. lfd f21, 56(SP)
  236. lfd f22, 64(SP)
  237. lfd f23, 72(SP)
  238. lfd f24, 80(SP)
  239. lfd f25, 88(SP)
  240. lfd f26, 96(SP)
  241. lfd f27, 104(SP)
  242. lfd f28, 112(SP)
  243. lfd f29, 120(SP)
  244. lfd f30, 128(SP)
  245. lfd f31, 136(SP)
  246. addi SP, SP, STACKSIZE
  247. blr
  248. EPILOGUE