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_hummer.S 7.0 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347
  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 INCX2 r6
  44. #define C1 f1
  45. #define C2 f0
  46. #define C3 f2
  47. #define C4 f3
  48. #define A1 f4
  49. #define A2 f5
  50. #define A3 f6
  51. #define A4 f7
  52. #define A5 f8
  53. #define A6 f9
  54. #define A7 f10
  55. #define A8 f11
  56. #define F1 f12
  57. #define F2 f13
  58. #define F3 f14
  59. #define F4 f15
  60. #define T1 f16
  61. #define T2 f17
  62. #define T3 f18
  63. #define T4 f19
  64. #define B1 f20
  65. #define B2 f21
  66. #define B3 f22
  67. #define B4 f23
  68. #define B5 f24
  69. #define B6 f25
  70. #define B7 f26
  71. #define B8 f27
  72. PROLOGUE
  73. PROFCODE
  74. li r10, -16
  75. stfpdux f14, SP, r10
  76. stfpdux f15, SP, r10
  77. stfpdux f16, SP, r10
  78. stfpdux f17, SP, r10
  79. stfpdux f18, SP, r10
  80. stfpdux f19, SP, r10
  81. stfpdux f20, SP, r10
  82. stfpdux f21, SP, r10
  83. stfpdux f22, SP, r10
  84. stfpdux f23, SP, r10
  85. stfpdux f24, SP, r10
  86. stfpdux f25, SP, r10
  87. stfpdux f26, SP, r10
  88. stfpdux f27, SP, r10
  89. li r10, 0
  90. stwu r10, -4(SP)
  91. stwu r10, -4(SP)
  92. stwu r10, -4(SP)
  93. stwu r10, -4(SP)
  94. #ifdef F_INTERFACE
  95. LDINT N, 0(N)
  96. LDINT INCX, 0(INCX)
  97. #endif
  98. lfpdx C1, SP, r10 # Zero clear
  99. slwi INCX, INCX, BASE_SHIFT
  100. add INCX2, INCX, INCX
  101. cmpwi cr0, N, 0
  102. ble LL(999)
  103. cmpwi cr0, INCX, 0
  104. ble LL(999)
  105. LFD A1, 0 * SIZE(X)
  106. LFD A2, 1 * SIZE(X)
  107. add X, X, INCX2
  108. fabs A1, A1
  109. fabs A2, A2
  110. addi N, N, -1
  111. cmpwi cr0, N, 0
  112. fadd C1, A1, A2
  113. ble LL(999)
  114. subi INCX2, INCX2, SIZE
  115. fsmfp C1, C1
  116. li INCX, SIZE
  117. fpmr C2, C1
  118. sub X, X, INCX2
  119. fpmr C3, C1
  120. srawi. r0, N, 3
  121. fpmr C4, C1
  122. mtspr CTR, r0
  123. beq- LL(105)
  124. LFDUX A1, X, INCX2
  125. LFDUX A2, X, INCX
  126. LFDUX A3, X, INCX2
  127. LFDUX A4, X, INCX
  128. LFSDUX A1, X, INCX2
  129. LFSDUX A2, X, INCX
  130. LFSDUX A3, X, INCX2
  131. LFSDUX A4, X, INCX
  132. LFDUX A5, X, INCX2
  133. LFDUX A6, X, INCX
  134. LFDUX A7, X, INCX2
  135. LFDUX A8, X, INCX
  136. LFSDUX A5, X, INCX2
  137. LFSDUX A6, X, INCX
  138. LFSDUX A7, X, INCX2
  139. LFSDUX A8, X, INCX
  140. bdz LL(103)
  141. .align 4
  142. LL(102):
  143. fpabs B1, A1
  144. LFDUX A1, X, INCX2
  145. fpabs B2, A2
  146. LFDUX A2, X, INCX
  147. fpabs B3, A3
  148. LFDUX A3, X, INCX2
  149. fpabs B4, A4
  150. LFDUX A4, X, INCX
  151. fpabs B5, A5
  152. LFSDUX A1, X, INCX2
  153. fpabs B6, A6
  154. LFSDUX A2, X, INCX
  155. fpabs B7, A7
  156. LFSDUX A3, X, INCX2
  157. fpabs B8, A8
  158. LFSDUX A4, X, INCX
  159. fpadd T1, B1, B2
  160. LFDUX A5, X, INCX2
  161. fpadd T2, B3, B4
  162. LFDUX A6, X, INCX
  163. fpadd T3, B5, B6
  164. LFDUX A7, X, INCX2
  165. fpadd T4, B7, B8
  166. LFDUX A8, X, INCX
  167. fpsub F1, T1, C1
  168. LFSDUX A5, X, INCX2
  169. fpsub F2, T2, C2
  170. LFSDUX A6, X, INCX
  171. fpsub F3, T3, C3
  172. LFSDUX A7, X, INCX2
  173. fpsub F4, T4, C4
  174. LFSDUX A8, X, INCX
  175. fpsel C1, F1, C1, T1
  176. fpsel C2, F2, C2, T2
  177. fpsel C3, F3, C3, T3
  178. fpsel C4, F4, C4, T4
  179. bdnz LL(102)
  180. .align 4
  181. LL(103):
  182. fpabs B1, A1
  183. fpabs B2, A2
  184. fpabs B3, A3
  185. fpabs B4, A4
  186. fpabs B5, A5
  187. fpabs B6, A6
  188. fpabs B7, A7
  189. fpabs B8, A8
  190. fpadd T1, B1, B2
  191. fpadd T2, B3, B4
  192. fpadd T3, B5, B6
  193. fpadd T4, B7, B8
  194. fpsub F1, T1, C1
  195. fpsub F2, T2, C2
  196. fpsub F3, T3, C3
  197. fpsub F4, T4, C4
  198. fpsel C1, F1, C1, T1
  199. fpsel C2, F2, C2, T2
  200. fpsel C3, F3, C3, T3
  201. fpsel C4, F4, C4, T4
  202. .align 4
  203. LL(105):
  204. andi. r0, N, 7
  205. beq LL(998)
  206. andi. r0, N, 4
  207. beq LL(106)
  208. LFDUX A1, X, INCX2
  209. LFDUX A2, X, INCX
  210. LFDUX A3, X, INCX2
  211. LFDUX A4, X, INCX
  212. LFSDUX A1, X, INCX2
  213. LFSDUX A2, X, INCX
  214. LFSDUX A3, X, INCX2
  215. LFSDUX A4, X, INCX
  216. fpabs A1, A1
  217. fpabs A2, A2
  218. fpabs A3, A3
  219. fpabs A4, A4
  220. fpadd A1, A1, A2
  221. fpadd A3, A3, A4
  222. fpsub F1, A1, C1
  223. fpsub F2, A3, C2
  224. fpsel C1, F1, C1, A1
  225. fpsel C2, F2, C2, A3
  226. .align 4
  227. LL(106):
  228. andi. r0, N, 2
  229. beq LL(107)
  230. LFDUX A1, X, INCX2
  231. LFDUX A2, X, INCX
  232. LFSDUX A1, X, INCX2
  233. LFSDUX A2, X, INCX
  234. fpabs A1, A1
  235. fpabs A2, A2
  236. fpadd A1, A1, A2
  237. fpsub F1, A1, C1
  238. fpsel C1, F1, C1, A1
  239. .align 4
  240. LL(107):
  241. andi. r0, N, 1
  242. beq LL(998)
  243. LFDUX A1, X, INCX2
  244. LFDUX A2, X, INCX
  245. fabs A1, A1
  246. fabs A2, A2
  247. fadd A1, A1, A2
  248. fsub F1, A1, C1
  249. fsel C1, F1, C1, A1
  250. .align 4
  251. LL(998):
  252. fpsub F1, C2, C1
  253. fpsub F2, C4, C3
  254. fpsel C1, F1, C1, C2
  255. fpsel C3, F2, C3, C4
  256. fpsub F1, C3, C1
  257. fpsel C1, F1, C1, C3
  258. fsmtp C2, C1
  259. fsub F1, C2, C1
  260. fsel C1, F1, C1, C2
  261. .align 4
  262. LL(999):
  263. li r10, 16
  264. lfpdux f27, SP, r10
  265. lfpdux f26, SP, r10
  266. lfpdux f25, SP, r10
  267. lfpdux f24, SP, r10
  268. lfpdux f23, SP, r10
  269. lfpdux f22, SP, r10
  270. lfpdux f21, SP, r10
  271. lfpdux f20, SP, r10
  272. lfpdux f19, SP, r10
  273. lfpdux f18, SP, r10
  274. lfpdux f17, SP, r10
  275. lfpdux f16, SP, r10
  276. lfpdux f15, SP, r10
  277. lfpdux f14, SP, r10
  278. addi SP, SP, 16
  279. blr
  280. EPILOGUE