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.

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379
  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 $16
  41. #define X $17
  42. #define INCX $18
  43. #define Y $19
  44. #define INCY $20
  45. PROLOGUE
  46. PROFCODE
  47. .frame $sp, 0, $26, 0
  48. #ifndef PROFILE
  49. .prologue 0
  50. #else
  51. .prologue 1
  52. #endif
  53. cmpeq INCX, 1, $0
  54. ble N, $End
  55. #ifndef COMPLEX
  56. sra N, 4, $4
  57. #else
  58. sra N, 3, $4
  59. #endif
  60. cmpeq INCY, 1, $1
  61. and $0, $1, $0
  62. beq $0, $Sub
  63. #ifndef COMPLEX
  64. and N, 15, $5
  65. #else
  66. and N, 7, $5
  67. #endif
  68. ble $4, $Remain
  69. LD $f10, 0*SIZE(X)
  70. LD $f11, 1*SIZE(X)
  71. LD $f12, 2*SIZE(X)
  72. LD $f13, 3*SIZE(X)
  73. LD $f14, 4*SIZE(X)
  74. LD $f15, 5*SIZE(X)
  75. LD $f16, 6*SIZE(X)
  76. LD $f17, 7*SIZE(X)
  77. LD $f18, 8*SIZE(X)
  78. LD $f19, 9*SIZE(X)
  79. LD $f20, 10*SIZE(X)
  80. LD $f21, 11*SIZE(X)
  81. LD $f22, 12*SIZE(X)
  82. LD $f23, 13*SIZE(X)
  83. LD $f24, 14*SIZE(X)
  84. LD $f25, 15*SIZE(X)
  85. subq $4, 1, $4
  86. lda X, 16*SIZE(X)
  87. ble $4, $MainLoopEnd
  88. .align 4
  89. $MainLoop:
  90. ST $f10, 0*SIZE(Y)
  91. ST $f11, 1*SIZE(Y)
  92. ST $f12, 2*SIZE(Y)
  93. ST $f13, 3*SIZE(Y)
  94. LD $f10, 0*SIZE(X)
  95. LD $f11, 1*SIZE(X)
  96. LD $f12, 2*SIZE(X)
  97. LD $f13, 3*SIZE(X)
  98. ST $f14, 4*SIZE(Y)
  99. ST $f15, 5*SIZE(Y)
  100. ST $f16, 6*SIZE(Y)
  101. ST $f17, 7*SIZE(Y)
  102. LD $f14, 4*SIZE(X)
  103. LD $f15, 5*SIZE(X)
  104. LD $f16, 6*SIZE(X)
  105. LD $f17, 7*SIZE(X)
  106. ST $f18, 8*SIZE(Y)
  107. ST $f19, 9*SIZE(Y)
  108. ST $f20, 10*SIZE(Y)
  109. ST $f21, 11*SIZE(Y)
  110. LD $f18, 8*SIZE(X)
  111. LD $f19, 9*SIZE(X)
  112. LD $f20, 10*SIZE(X)
  113. LD $f21, 11*SIZE(X)
  114. ST $f22, 12*SIZE(Y)
  115. ST $f23, 13*SIZE(Y)
  116. ST $f24, 14*SIZE(Y)
  117. ST $f25, 15*SIZE(Y)
  118. LD $f22, 12*SIZE(X)
  119. LD $f23, 13*SIZE(X)
  120. LD $f24, 14*SIZE(X)
  121. LD $f25, 15*SIZE(X)
  122. subq $4, 1, $4
  123. lda Y, 16*SIZE(Y)
  124. lda X, 16*SIZE(X)
  125. bgt $4, $MainLoop
  126. .align 4
  127. $MainLoopEnd:
  128. ST $f10, 0*SIZE(Y)
  129. ST $f11, 1*SIZE(Y)
  130. ST $f12, 2*SIZE(Y)
  131. ST $f13, 3*SIZE(Y)
  132. ST $f14, 4*SIZE(Y)
  133. ST $f15, 5*SIZE(Y)
  134. ST $f16, 6*SIZE(Y)
  135. ST $f17, 7*SIZE(Y)
  136. ST $f18, 8*SIZE(Y)
  137. ST $f19, 9*SIZE(Y)
  138. ST $f20, 10*SIZE(Y)
  139. ST $f21, 11*SIZE(Y)
  140. ST $f22, 12*SIZE(Y)
  141. ST $f23, 13*SIZE(Y)
  142. ST $f24, 14*SIZE(Y)
  143. ST $f25, 15*SIZE(Y)
  144. lda Y, 16*SIZE(Y)
  145. .align 4
  146. $Remain:
  147. ble $5, $End
  148. .align 4
  149. $RemainLoop:
  150. #ifndef COMPLEX
  151. LD $f10, 0*SIZE(X)
  152. lda X, 1*SIZE(X)
  153. ST $f10, 0*SIZE(Y)
  154. lda Y, 1*SIZE(Y)
  155. #else
  156. LD $f10, 0*SIZE(X)
  157. LD $f11, 1*SIZE(X)
  158. lda X, 2*SIZE(X)
  159. ST $f10, 0*SIZE(Y)
  160. ST $f11, 1*SIZE(Y)
  161. lda Y, 2*SIZE(Y)
  162. #endif
  163. subq $5, 1, $5
  164. bgt $5, $RemainLoop
  165. .align 4
  166. $End:
  167. ret
  168. .align 4
  169. $Sub:
  170. #ifdef COMPLEX
  171. addq INCX, INCX, INCX
  172. addq INCY, INCY, INCY
  173. and N, 7, $5
  174. #else
  175. and N, 15, $5
  176. #endif
  177. ble $4, $SubRemain
  178. .align 4
  179. $SubMainLoop:
  180. #ifndef COMPLEX
  181. LD $f10, 0(X)
  182. SXADDQ INCX, X, X
  183. LD $f11, 0(X)
  184. SXADDQ INCX, X, X
  185. LD $f12, 0(X)
  186. SXADDQ INCX, X, X
  187. LD $f13, 0(X)
  188. SXADDQ INCX, X, X
  189. LD $f14, 0(X)
  190. SXADDQ INCX, X, X
  191. LD $f15, 0(X)
  192. SXADDQ INCX, X, X
  193. LD $f16, 0(X)
  194. SXADDQ INCX, X, X
  195. LD $f17, 0(X)
  196. SXADDQ INCX, X, X
  197. LD $f18, 0(X)
  198. SXADDQ INCX, X, X
  199. LD $f19, 0(X)
  200. SXADDQ INCX, X, X
  201. LD $f20, 0(X)
  202. SXADDQ INCX, X, X
  203. LD $f21, 0(X)
  204. SXADDQ INCX, X, X
  205. LD $f22, 0(X)
  206. SXADDQ INCX, X, X
  207. LD $f23, 0(X)
  208. SXADDQ INCX, X, X
  209. LD $f24, 0(X)
  210. SXADDQ INCX, X, X
  211. LD $f25, 0(X)
  212. SXADDQ INCX, X, X
  213. ST $f10, 0(Y)
  214. SXADDQ INCY, Y, Y
  215. ST $f11, 0(Y)
  216. SXADDQ INCY, Y, Y
  217. ST $f12, 0(Y)
  218. SXADDQ INCY, Y, Y
  219. ST $f13, 0(Y)
  220. SXADDQ INCY, Y, Y
  221. ST $f14, 0(Y)
  222. SXADDQ INCY, Y, Y
  223. ST $f15, 0(Y)
  224. SXADDQ INCY, Y, Y
  225. ST $f16, 0(Y)
  226. SXADDQ INCY, Y, Y
  227. ST $f17, 0(Y)
  228. SXADDQ INCY, Y, Y
  229. ST $f18, 0(Y)
  230. SXADDQ INCY, Y, Y
  231. ST $f19, 0(Y)
  232. SXADDQ INCY, Y, Y
  233. ST $f20, 0(Y)
  234. SXADDQ INCY, Y, Y
  235. ST $f21, 0(Y)
  236. SXADDQ INCY, Y, Y
  237. ST $f22, 0(Y)
  238. SXADDQ INCY, Y, Y
  239. ST $f23, 0(Y)
  240. SXADDQ INCY, Y, Y
  241. ST $f24, 0(Y)
  242. SXADDQ INCY, Y, Y
  243. ST $f25, 0(Y)
  244. SXADDQ INCY, Y, Y
  245. #else
  246. LD $f10, 0(X)
  247. LD $f11, SIZE(X)
  248. SXADDQ INCX, X, X
  249. LD $f12, 0(X)
  250. LD $f13, SIZE(X)
  251. SXADDQ INCX, X, X
  252. LD $f14, 0(X)
  253. LD $f15, SIZE(X)
  254. SXADDQ INCX, X, X
  255. LD $f16, 0(X)
  256. LD $f17, SIZE(X)
  257. SXADDQ INCX, X, X
  258. LD $f18, 0(X)
  259. LD $f19, SIZE(X)
  260. SXADDQ INCX, X, X
  261. LD $f20, 0(X)
  262. LD $f21, SIZE(X)
  263. SXADDQ INCX, X, X
  264. LD $f22, 0(X)
  265. LD $f23, SIZE(X)
  266. SXADDQ INCX, X, X
  267. LD $f24, 0(X)
  268. LD $f25, SIZE(X)
  269. SXADDQ INCX, X, X
  270. ST $f10, 0(Y)
  271. ST $f11, SIZE(Y)
  272. SXADDQ INCY, Y, Y
  273. ST $f12, 0(Y)
  274. ST $f13, SIZE(Y)
  275. SXADDQ INCY, Y, Y
  276. ST $f14, 0(Y)
  277. ST $f15, SIZE(Y)
  278. SXADDQ INCY, Y, Y
  279. ST $f16, 0(Y)
  280. ST $f17, SIZE(Y)
  281. SXADDQ INCY, Y, Y
  282. ST $f18, 0(Y)
  283. ST $f19, SIZE(Y)
  284. SXADDQ INCY, Y, Y
  285. ST $f20, 0(Y)
  286. ST $f21, SIZE(Y)
  287. SXADDQ INCY, Y, Y
  288. ST $f22, 0(Y)
  289. ST $f23, SIZE(Y)
  290. SXADDQ INCY, Y, Y
  291. ST $f24, 0(Y)
  292. ST $f25, SIZE(Y)
  293. SXADDQ INCY, Y, Y
  294. #endif
  295. subq $4, 1, $4
  296. bgt $4, $SubMainLoop
  297. .align 4
  298. $SubRemain:
  299. ble $5, $SubEnd
  300. .align 4
  301. $SubRemainLoop:
  302. #ifndef COMPLEX
  303. LD $f10, 0(X)
  304. SXADDQ INCX, X, X
  305. ST $f10, 0(Y)
  306. SXADDQ INCY, Y, Y
  307. #else
  308. LD $f10, 0(X)
  309. LD $f11, SIZE(X)
  310. SXADDQ INCX, X, X
  311. ST $f10, 0(Y)
  312. ST $f11, SIZE(Y)
  313. SXADDQ INCY, Y, Y
  314. #endif
  315. subq $5, 1, $5
  316. bgt $5, $SubRemainLoop
  317. .align 4
  318. $SubEnd:
  319. ret
  320. EPILOGUE