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.

zaxpy.S 7.2 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348
  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 STACK 12
  41. #define ARGS 0
  42. #define STACK_M 4 + STACK + ARGS(%esp)
  43. #ifdef DOUBLE
  44. #define STACK_ALPHA_R 16 + STACK + ARGS(%esp)
  45. #define STACK_ALPHA_I 24 + STACK + ARGS(%esp)
  46. #define STACK_X 32 + STACK + ARGS(%esp)
  47. #define STACK_INCX 36 + STACK + ARGS(%esp)
  48. #define STACK_Y 40 + STACK + ARGS(%esp)
  49. #define STACK_INCY 44 + STACK + ARGS(%esp)
  50. #else
  51. #define STACK_ALPHA_R 16 + STACK + ARGS(%esp)
  52. #define STACK_ALPHA_I 20 + STACK + ARGS(%esp)
  53. #define STACK_X 24 + STACK + ARGS(%esp)
  54. #define STACK_INCX 28 + STACK + ARGS(%esp)
  55. #define STACK_Y 32 + STACK + ARGS(%esp)
  56. #define STACK_INCY 36 + STACK + ARGS(%esp)
  57. #endif
  58. #define M %ebx
  59. #define X %esi
  60. #define INCX %ecx
  61. #define Y %edi
  62. #define INCY %edx
  63. #ifndef CONJ
  64. #define ADD1 fsubrp
  65. #define ADD2 faddp
  66. #else
  67. #define ADD1 faddp
  68. #define ADD2 fsubrp
  69. #endif
  70. PROLOGUE
  71. pushl %edi
  72. pushl %esi
  73. pushl %ebx
  74. PROFCODE
  75. #if defined(F_INTERFACE_GFORT) || defined(F_INTERFACE_G95)
  76. EMMS
  77. #endif
  78. FLD STACK_ALPHA_I
  79. FLD STACK_ALPHA_R
  80. movl STACK_M, M
  81. movl STACK_X, X
  82. movl STACK_INCX, INCX
  83. movl STACK_Y, Y
  84. movl STACK_INCY, INCY
  85. addl INCX, INCX
  86. addl INCY, INCY
  87. leal (, INCX, SIZE), INCX
  88. leal (, INCY, SIZE), INCY
  89. testl M, M
  90. jle .L40
  91. cmpl $2 * SIZE, INCX
  92. jne .L14
  93. cmpl $2 * SIZE, INCY
  94. jne .L14
  95. movl M, %eax
  96. sarl $2, %eax
  97. jle .L15
  98. ALIGN_3
  99. .L16:
  100. FLD 0 * SIZE(X)
  101. fmul %st(1), %st
  102. FLD 1 * SIZE(X)
  103. fmul %st(3), %st
  104. ADD1 %st, %st(1)
  105. FADD 0 * SIZE(Y)
  106. FST 0 * SIZE(Y)
  107. FLD 0 * SIZE(X)
  108. fmul %st(2), %st
  109. FLD 1 * SIZE(X)
  110. fmul %st(2), %st
  111. ADD2 %st, %st(1)
  112. FADD 1 * SIZE(Y)
  113. FST 1 * SIZE(Y)
  114. FLD 2 * SIZE(X)
  115. fmul %st(1), %st
  116. FLD 3 * SIZE(X)
  117. fmul %st(3), %st
  118. ADD1 %st, %st(1)
  119. FADD 2 * SIZE(Y)
  120. FST 2 * SIZE(Y)
  121. FLD 2 * SIZE(X)
  122. fmul %st(2), %st
  123. FLD 3 * SIZE(X)
  124. fmul %st(2), %st
  125. ADD2 %st, %st(1)
  126. FADD 3 * SIZE(Y)
  127. FST 3 * SIZE(Y)
  128. FLD 4 * SIZE(X)
  129. fmul %st(1), %st
  130. FLD 5 * SIZE(X)
  131. fmul %st(3), %st
  132. ADD1 %st, %st(1)
  133. FADD 4 * SIZE(Y)
  134. FST 4 * SIZE(Y)
  135. FLD 4 * SIZE(X)
  136. fmul %st(2), %st
  137. FLD 5 * SIZE(X)
  138. fmul %st(2), %st
  139. ADD2 %st, %st(1)
  140. FADD 5 * SIZE(Y)
  141. FST 5 * SIZE(Y)
  142. FLD 6 * SIZE(X)
  143. fmul %st(1), %st
  144. FLD 7 * SIZE(X)
  145. fmul %st(3), %st
  146. ADD1 %st, %st(1)
  147. FADD 6 * SIZE(Y)
  148. FST 6 * SIZE(Y)
  149. FLD 6 * SIZE(X)
  150. fmul %st(2), %st
  151. FLD 7 * SIZE(X)
  152. fmul %st(2), %st
  153. ADD2 %st, %st(1)
  154. FADD 7 * SIZE(Y)
  155. FST 7 * SIZE(Y)
  156. #ifdef HAVE_3DNOW
  157. prefetch 20 * SIZE(X)
  158. prefetchw 20 * SIZE(Y)
  159. #endif
  160. addl $8 * SIZE, X
  161. addl $8 * SIZE, Y
  162. decl %eax
  163. jg .L16
  164. ALIGN_3
  165. .L15:
  166. movl M, %eax
  167. andl $3, %eax
  168. jle .L40
  169. ALIGN_3
  170. .L22:
  171. FLD 0 * SIZE(X)
  172. fmul %st(1), %st
  173. FLD 1 * SIZE(X)
  174. fmul %st(3), %st
  175. ADD1 %st, %st(1)
  176. FADD 0 * SIZE(Y)
  177. FST 0 * SIZE(Y)
  178. FLD 0 * SIZE(X)
  179. fmul %st(2), %st
  180. FLD 1 * SIZE(X)
  181. fmul %st(2), %st
  182. ADD2 %st, %st(1)
  183. FADD 1 * SIZE(Y)
  184. FST 1 * SIZE(Y)
  185. addl $2 * SIZE, X
  186. addl $2 * SIZE, Y
  187. decl %eax
  188. jg .L22
  189. jmp .L40
  190. ALIGN_3
  191. .L14:
  192. movl M, %eax
  193. sarl $2, %eax
  194. jle .L28
  195. ALIGN_3
  196. .L29:
  197. FLD 0 * SIZE(X)
  198. fmul %st(1), %st
  199. FLD 1 * SIZE(X)
  200. fmul %st(3), %st
  201. ADD1 %st, %st(1)
  202. FADD 0 * SIZE(Y)
  203. FST 0 * SIZE(Y)
  204. FLD 0 * SIZE(X)
  205. fmul %st(2), %st
  206. FLD 1 * SIZE(X)
  207. fmul %st(2), %st
  208. ADD2 %st, %st(1)
  209. FADD 1 * SIZE(Y)
  210. FST 1 * SIZE(Y)
  211. addl INCX, X
  212. addl INCY, Y
  213. FLD 0 * SIZE(X)
  214. fmul %st(1), %st
  215. FLD 1 * SIZE(X)
  216. fmul %st(3), %st
  217. ADD1 %st, %st(1)
  218. FADD 0 * SIZE(Y)
  219. FST 0 * SIZE(Y)
  220. FLD 0 * SIZE(X)
  221. fmul %st(2), %st
  222. FLD 1 * SIZE(X)
  223. fmul %st(2), %st
  224. ADD2 %st, %st(1)
  225. FADD 1 * SIZE(Y)
  226. FST 1 * SIZE(Y)
  227. addl INCX, X
  228. addl INCY, Y
  229. FLD 0 * SIZE(X)
  230. fmul %st(1), %st
  231. FLD 1 * SIZE(X)
  232. fmul %st(3), %st
  233. ADD1 %st, %st(1)
  234. FADD 0 * SIZE(Y)
  235. FST 0 * SIZE(Y)
  236. FLD 0 * SIZE(X)
  237. fmul %st(2), %st
  238. FLD 1 * SIZE(X)
  239. fmul %st(2), %st
  240. ADD2 %st, %st(1)
  241. FADD 1 * SIZE(Y)
  242. FST 1 * SIZE(Y)
  243. addl INCX, X
  244. addl INCY, Y
  245. FLD 0 * SIZE(X)
  246. fmul %st(1), %st
  247. FLD 1 * SIZE(X)
  248. fmul %st(3), %st
  249. ADD1 %st, %st(1)
  250. FADD 0 * SIZE(Y)
  251. FST 0 * SIZE(Y)
  252. FLD 0 * SIZE(X)
  253. fmul %st(2), %st
  254. FLD 1 * SIZE(X)
  255. fmul %st(2), %st
  256. ADD2 %st, %st(1)
  257. FADD 1 * SIZE(Y)
  258. FST 1 * SIZE(Y)
  259. addl INCX, X
  260. addl INCY, Y
  261. decl %eax
  262. jg .L29
  263. ALIGN_3
  264. .L28:
  265. movl M, %eax
  266. andl $3, %eax
  267. jle .L40
  268. ALIGN_3
  269. .L35:
  270. FLD 0 * SIZE(X)
  271. fmul %st(1), %st
  272. FLD 1 * SIZE(X)
  273. fmul %st(3), %st
  274. ADD1 %st, %st(1)
  275. FADD 0 * SIZE(Y)
  276. FST 0 * SIZE(Y)
  277. FLD 0 * SIZE(X)
  278. fmul %st(2), %st
  279. FLD 1 * SIZE(X)
  280. fmul %st(2), %st
  281. ADD2 %st, %st(1)
  282. FADD 1 * SIZE(Y)
  283. FST 1 * SIZE(Y)
  284. addl INCX, X
  285. addl INCY, Y
  286. decl %eax
  287. jg .L35
  288. ALIGN_3
  289. .L40:
  290. ffreep %st(0)
  291. ffreep %st(0)
  292. xorl %eax,%eax
  293. popl %ebx
  294. popl %esi
  295. popl %edi
  296. ret
  297. EPILOGUE