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 6.9 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336
  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 M ARG1 /* rdi */
  41. #define X ARG4 /* rsi */
  42. #define INCX ARG5 /* rdx */
  43. #define Y ARG6 /* rcx */
  44. #define INCY ARG2 /* r8 */
  45. #ifndef CONJ
  46. #define ADD1 fsubrp
  47. #define ADD2 faddp
  48. #else
  49. #define ADD1 faddp
  50. #define ADD2 fsubrp
  51. #endif
  52. #define ALPHA_R 8(%rsp)
  53. #define ALPHA_I 24(%rsp)
  54. #include "l1param.h"
  55. PROLOGUE
  56. PROFCODE
  57. FLD ALPHA_I
  58. FLD ALPHA_R
  59. movq 40(%rsp), INCY
  60. salq $ZBASE_SHIFT, INCX
  61. salq $ZBASE_SHIFT, INCY
  62. testq M, M
  63. jle .L40
  64. cmpq $2 * SIZE, INCX
  65. jne .L14
  66. cmpq $2 * SIZE, INCY
  67. jne .L14
  68. movq M, %rax
  69. sarq $2, %rax
  70. jle .L15
  71. ALIGN_3
  72. .L16:
  73. #ifdef PREFETCH
  74. PREFETCH (PREFETCHSIZE + 0) - PREOFFSET(X)
  75. #endif
  76. FLD 0 * SIZE(X)
  77. fmul %st(1), %st
  78. FLD 1 * SIZE(X)
  79. fmul %st(3), %st
  80. ADD1 %st, %st(1)
  81. FLD 0 * SIZE(Y)
  82. faddp %st, %st(1)
  83. FST 0 * SIZE(Y)
  84. FLD 0 * SIZE(X)
  85. fmul %st(2), %st
  86. FLD 1 * SIZE(X)
  87. fmul %st(2), %st
  88. ADD2 %st, %st(1)
  89. FLD 1 * SIZE(Y)
  90. faddp %st, %st(1)
  91. FST 1 * SIZE(Y)
  92. FLD 2 * SIZE(X)
  93. fmul %st(1), %st
  94. FLD 3 * SIZE(X)
  95. fmul %st(3), %st
  96. ADD1 %st, %st(1)
  97. FLD 2 * SIZE(Y)
  98. faddp %st, %st(1)
  99. FST 2 * SIZE(Y)
  100. FLD 2 * SIZE(X)
  101. fmul %st(2), %st
  102. FLD 3 * SIZE(X)
  103. fmul %st(2), %st
  104. ADD2 %st, %st(1)
  105. FLD 3 * SIZE(Y)
  106. faddp %st, %st(1)
  107. FST 3 * SIZE(Y)
  108. #ifdef PREFETCHW
  109. PREFETCHW (PREFETCHSIZE + 0) - PREOFFSET(Y)
  110. #endif
  111. FLD 4 * SIZE(X)
  112. fmul %st(1), %st
  113. FLD 5 * SIZE(X)
  114. fmul %st(3), %st
  115. ADD1 %st, %st(1)
  116. FLD 4 * SIZE(Y)
  117. faddp %st, %st(1)
  118. FST 4 * SIZE(Y)
  119. FLD 4 * SIZE(X)
  120. fmul %st(2), %st
  121. FLD 5 * SIZE(X)
  122. fmul %st(2), %st
  123. ADD2 %st, %st(1)
  124. FLD 5 * SIZE(Y)
  125. faddp %st, %st(1)
  126. FST 5 * SIZE(Y)
  127. FLD 6 * SIZE(X)
  128. fmul %st(1), %st
  129. FLD 7 * SIZE(X)
  130. fmul %st(3), %st
  131. ADD1 %st, %st(1)
  132. FLD 6 * SIZE(Y)
  133. faddp %st, %st(1)
  134. FST 6 * SIZE(Y)
  135. FLD 6 * SIZE(X)
  136. fmul %st(2), %st
  137. FLD 7 * SIZE(X)
  138. fmul %st(2), %st
  139. ADD2 %st, %st(1)
  140. FLD 7 * SIZE(Y)
  141. faddp %st, %st(1)
  142. FST 7 * SIZE(Y)
  143. addq $8 * SIZE, X
  144. addq $8 * SIZE, Y
  145. decq %rax
  146. jg .L16
  147. ALIGN_3
  148. .L15:
  149. movq M, %rax
  150. andq $3, %rax
  151. jle .L40
  152. ALIGN_3
  153. .L22:
  154. FLD 0 * SIZE(X)
  155. fmul %st(1), %st
  156. FLD 1 * SIZE(X)
  157. fmul %st(3), %st
  158. ADD1 %st, %st(1)
  159. FLD 0 * SIZE(Y)
  160. faddp %st, %st(1)
  161. FST 0 * SIZE(Y)
  162. FLD 0 * SIZE(X)
  163. fmul %st(2), %st
  164. FLD 1 * SIZE(X)
  165. fmul %st(2), %st
  166. ADD2 %st, %st(1)
  167. FLD 1 * SIZE(Y)
  168. faddp %st, %st(1)
  169. FST 1 * SIZE(Y)
  170. addq $2 * SIZE, X
  171. addq $2 * SIZE, Y
  172. decq %rax
  173. jg .L22
  174. jmp .L40
  175. ALIGN_3
  176. .L14:
  177. movq M, %rax
  178. sarq $2, %rax
  179. jle .L28
  180. ALIGN_3
  181. .L29:
  182. FLD 0 * SIZE(X)
  183. fmul %st(1), %st
  184. FLD 1 * SIZE(X)
  185. fmul %st(3), %st
  186. ADD1 %st, %st(1)
  187. FLD 0 * SIZE(Y)
  188. faddp %st, %st(1)
  189. FST 0 * SIZE(Y)
  190. FLD 0 * SIZE(X)
  191. fmul %st(2), %st
  192. FLD 1 * SIZE(X)
  193. fmul %st(2), %st
  194. ADD2 %st, %st(1)
  195. FLD 1 * SIZE(Y)
  196. faddp %st, %st(1)
  197. FST 1 * SIZE(Y)
  198. addq INCX, X
  199. addq INCY, Y
  200. FLD 0 * SIZE(X)
  201. fmul %st(1), %st
  202. FLD 1 * SIZE(X)
  203. fmul %st(3), %st
  204. ADD1 %st, %st(1)
  205. FLD 0 * SIZE(Y)
  206. faddp %st, %st(1)
  207. FST 0 * SIZE(Y)
  208. FLD 0 * SIZE(X)
  209. fmul %st(2), %st
  210. FLD 1 * SIZE(X)
  211. fmul %st(2), %st
  212. ADD2 %st, %st(1)
  213. FLD 1 * SIZE(Y)
  214. faddp %st, %st(1)
  215. FST 1 * SIZE(Y)
  216. addq INCX, X
  217. addq INCY, Y
  218. FLD 0 * SIZE(X)
  219. fmul %st(1), %st
  220. FLD 1 * SIZE(X)
  221. fmul %st(3), %st
  222. ADD1 %st, %st(1)
  223. FLD 0 * SIZE(Y)
  224. faddp %st, %st(1)
  225. FST 0 * SIZE(Y)
  226. FLD 0 * SIZE(X)
  227. fmul %st(2), %st
  228. FLD 1 * SIZE(X)
  229. fmul %st(2), %st
  230. ADD2 %st, %st(1)
  231. FLD 1 * SIZE(Y)
  232. faddp %st, %st(1)
  233. FST 1 * SIZE(Y)
  234. addq INCX, X
  235. addq INCY, Y
  236. FLD 0 * SIZE(X)
  237. fmul %st(1), %st
  238. FLD 1 * SIZE(X)
  239. fmul %st(3), %st
  240. ADD1 %st, %st(1)
  241. FLD 0 * SIZE(Y)
  242. faddp %st, %st(1)
  243. FST 0 * SIZE(Y)
  244. FLD 0 * SIZE(X)
  245. fmul %st(2), %st
  246. FLD 1 * SIZE(X)
  247. fmul %st(2), %st
  248. ADD2 %st, %st(1)
  249. FLD 1 * SIZE(Y)
  250. faddp %st, %st(1)
  251. FST 1 * SIZE(Y)
  252. addq INCX, X
  253. addq INCY, Y
  254. decq %rax
  255. jg .L29
  256. ALIGN_3
  257. .L28:
  258. movq M, %rax
  259. andq $3, %rax
  260. jle .L40
  261. ALIGN_3
  262. .L35:
  263. FLD 0 * SIZE(X)
  264. fmul %st(1), %st
  265. FLD 1 * SIZE(X)
  266. fmul %st(3), %st
  267. ADD1 %st, %st(1)
  268. FLD 0 * SIZE(Y)
  269. faddp %st, %st(1)
  270. FST 0 * SIZE(Y)
  271. FLD 0 * SIZE(X)
  272. fmul %st(2), %st
  273. FLD 1 * SIZE(X)
  274. fmul %st(2), %st
  275. ADD2 %st, %st(1)
  276. FLD 1 * SIZE(Y)
  277. faddp %st, %st(1)
  278. FST 1 * SIZE(Y)
  279. addq INCX, X
  280. addq INCY, Y
  281. decq %rax
  282. jg .L35
  283. ALIGN_3
  284. .L40:
  285. ffreep %st(0)
  286. ffreep %st(0)
  287. ret
  288. EPILOGUE