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.

copy.S 7.6 kB

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