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.

axpy.S 8.2 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409
  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 $4
  41. #define X $8
  42. #define INCX $9
  43. #define Y $10
  44. #define INCY $11
  45. #define I $2
  46. #define TEMP $3
  47. #define YY $5
  48. #define ALPHA $f15
  49. #define a1 $f0
  50. #define a2 $f1
  51. #define a3 $f2
  52. #define a4 $f3
  53. #define a5 $f4
  54. #define a6 $f5
  55. #define a7 $f6
  56. #define a8 $f7
  57. #define b1 $f8
  58. #define b2 $f9
  59. #define b3 $f10
  60. #define b4 $f11
  61. #define b5 $f12
  62. #define b6 $f13
  63. #define b7 $f14
  64. #define b8 $f17
  65. #define t1 $f18
  66. #define t2 $f19
  67. #define t3 $f20
  68. #define t4 $f21
  69. PROLOGUE
  70. #ifndef __64BIT__
  71. daddiu $sp, $sp, -16
  72. sdc1 $f20, 0($sp)
  73. sdc1 $f21, 8($sp)
  74. #endif
  75. li TEMP, SIZE
  76. blez N, .L999
  77. dsll INCX, INCX, BASE_SHIFT
  78. bne INCX, TEMP, .L20
  79. dsll INCY, INCY, BASE_SHIFT
  80. bne INCY, TEMP, .L20
  81. dsra I, N, 3
  82. blez I, .L15
  83. daddiu I, I, -1
  84. LD a1, 0 * SIZE(X)
  85. LD b1, 0 * SIZE(Y)
  86. LD a2, 1 * SIZE(X)
  87. LD b2, 1 * SIZE(Y)
  88. LD a3, 2 * SIZE(X)
  89. LD b3, 2 * SIZE(Y)
  90. LD a4, 3 * SIZE(X)
  91. LD b4, 3 * SIZE(Y)
  92. LD a5, 4 * SIZE(X)
  93. LD b5, 4 * SIZE(Y)
  94. LD a6, 5 * SIZE(X)
  95. LD b6, 5 * SIZE(Y)
  96. LD a7, 6 * SIZE(X)
  97. LD b7, 6 * SIZE(Y)
  98. LD a8, 7 * SIZE(X)
  99. LD b8, 7 * SIZE(Y)
  100. blez I, .L13
  101. NOP
  102. .align 3
  103. .L12:
  104. MADD t1, b1, ALPHA, a1
  105. LD a1, 8 * SIZE(X)
  106. LD b1, 8 * SIZE(Y)
  107. MADD t2, b2, ALPHA, a2
  108. LD a2, 9 * SIZE(X)
  109. LD b2, 9 * SIZE(Y)
  110. MADD t3, b3, ALPHA, a3
  111. LD a3, 10 * SIZE(X)
  112. LD b3, 10 * SIZE(Y)
  113. MADD t4, b4, ALPHA, a4
  114. LD a4, 11 * SIZE(X)
  115. LD b4, 11 * SIZE(Y)
  116. ST t1, 0 * SIZE(Y)
  117. ST t2, 1 * SIZE(Y)
  118. ST t3, 2 * SIZE(Y)
  119. ST t4, 3 * SIZE(Y)
  120. MADD t1, b5, ALPHA, a5
  121. LD a5, 12 * SIZE(X)
  122. LD b5, 12 * SIZE(Y)
  123. MADD t2, b6, ALPHA, a6
  124. LD a6, 13 * SIZE(X)
  125. LD b6, 13 * SIZE(Y)
  126. MADD t3, b7, ALPHA, a7
  127. LD a7, 14 * SIZE(X)
  128. LD b7, 14 * SIZE(Y)
  129. MADD t4, b8, ALPHA, a8
  130. LD a8, 15 * SIZE(X)
  131. LD b8, 15 * SIZE(Y)
  132. ST t1, 4 * SIZE(Y)
  133. ST t2, 5 * SIZE(Y)
  134. ST t3, 6 * SIZE(Y)
  135. ST t4, 7 * SIZE(Y)
  136. daddiu I, I, -1
  137. daddiu Y, Y, 8 * SIZE
  138. bgtz I, .L12
  139. daddiu X, X, 8 * SIZE
  140. .align 3
  141. .L13:
  142. MADD t1, b1, ALPHA, a1
  143. MADD t2, b2, ALPHA, a2
  144. MADD t3, b3, ALPHA, a3
  145. MADD t4, b4, ALPHA, a4
  146. ST t1, 0 * SIZE(Y)
  147. MADD t1, b5, ALPHA, a5
  148. ST t2, 1 * SIZE(Y)
  149. MADD t2, b6, ALPHA, a6
  150. ST t3, 2 * SIZE(Y)
  151. MADD t3, b7, ALPHA, a7
  152. ST t4, 3 * SIZE(Y)
  153. MADD t4, b8, ALPHA, a8
  154. ST t1, 4 * SIZE(Y)
  155. ST t2, 5 * SIZE(Y)
  156. ST t3, 6 * SIZE(Y)
  157. ST t4, 7 * SIZE(Y)
  158. daddiu X, X, 8 * SIZE
  159. daddiu Y, Y, 8 * SIZE
  160. .align 3
  161. .L15:
  162. andi I, N, 7
  163. blez I, .L999
  164. NOP
  165. .align 3
  166. .L16:
  167. LD a1, 0 * SIZE(X)
  168. LD b1, 0 * SIZE(Y)
  169. daddiu X, X, SIZE
  170. daddiu Y, Y, SIZE
  171. MADD t1, b1, ALPHA, a1
  172. daddiu I, I, -1
  173. bgtz I, .L16
  174. ST t1, -1 * SIZE(Y)
  175. #ifndef __64BIT__
  176. ldc1 $f20, 0($sp)
  177. ldc1 $f21, 8($sp)
  178. daddiu $sp, $sp, 16
  179. #endif
  180. j $31
  181. NOP
  182. .align 3
  183. .L20:
  184. dsra I, N, 3
  185. move YY, Y
  186. blez I, .L25
  187. daddiu I, I, -1
  188. LD a1, 0 * SIZE(X)
  189. daddu X, X, INCX
  190. LD b1, 0 * SIZE(Y)
  191. daddu Y, Y, INCY
  192. LD a2, 0 * SIZE(X)
  193. daddu X, X, INCX
  194. LD b2, 0 * SIZE(Y)
  195. daddu Y, Y, INCY
  196. LD a3, 0 * SIZE(X)
  197. daddu X, X, INCX
  198. LD b3, 0 * SIZE(Y)
  199. daddu Y, Y, INCY
  200. LD a4, 0 * SIZE(X)
  201. daddu X, X, INCX
  202. LD b4, 0 * SIZE(Y)
  203. daddu Y, Y, INCY
  204. LD a5, 0 * SIZE(X)
  205. daddu X, X, INCX
  206. LD b5, 0 * SIZE(Y)
  207. daddu Y, Y, INCY
  208. LD a6, 0 * SIZE(X)
  209. daddu X, X, INCX
  210. LD b6, 0 * SIZE(Y)
  211. daddu Y, Y, INCY
  212. LD a7, 0 * SIZE(X)
  213. daddu X, X, INCX
  214. LD b7, 0 * SIZE(Y)
  215. daddu Y, Y, INCY
  216. LD a8, 0 * SIZE(X)
  217. daddu X, X, INCX
  218. LD b8, 0 * SIZE(Y)
  219. daddu Y, Y, INCY
  220. blez I, .L23
  221. NOP
  222. .align 3
  223. .L22:
  224. MADD t1, b1, ALPHA, a1
  225. LD a1, 0 * SIZE(X)
  226. LD b1, 0 * SIZE(Y)
  227. daddu X, X, INCX
  228. daddu Y, Y, INCY
  229. MADD t2, b2, ALPHA, a2
  230. LD a2, 0 * SIZE(X)
  231. LD b2, 0 * SIZE(Y)
  232. daddu X, X, INCX
  233. daddu Y, Y, INCY
  234. MADD t3, b3, ALPHA, a3
  235. LD a3, 0 * SIZE(X)
  236. LD b3, 0 * SIZE(Y)
  237. daddu X, X, INCX
  238. daddu Y, Y, INCY
  239. MADD t4, b4, ALPHA, a4
  240. LD a4, 0 * SIZE(X)
  241. LD b4, 0 * SIZE(Y)
  242. daddu X, X, INCX
  243. daddu Y, Y, INCY
  244. ST t1, 0 * SIZE(YY)
  245. daddu YY, YY, INCY
  246. MADD t1, b5, ALPHA, a5
  247. LD a5, 0 * SIZE(X)
  248. LD b5, 0 * SIZE(Y)
  249. daddu X, X, INCX
  250. daddu Y, Y, INCY
  251. ST t2, 0 * SIZE(YY)
  252. daddu YY, YY, INCY
  253. MADD t2, b6, ALPHA, a6
  254. LD a6, 0 * SIZE(X)
  255. LD b6, 0 * SIZE(Y)
  256. daddu X, X, INCX
  257. daddu Y, Y, INCY
  258. ST t3, 0 * SIZE(YY)
  259. daddu YY, YY, INCY
  260. MADD t3, b7, ALPHA, a7
  261. LD a7, 0 * SIZE(X)
  262. LD b7, 0 * SIZE(Y)
  263. daddu X, X, INCX
  264. daddu Y, Y, INCY
  265. ST t4, 0 * SIZE(YY)
  266. daddu YY, YY, INCY
  267. MADD t4, b8, ALPHA, a8
  268. LD a8, 0 * SIZE(X)
  269. daddu X, X, INCX
  270. LD b8, 0 * SIZE(Y)
  271. daddu Y, Y, INCY
  272. ST t1, 0 * SIZE(YY)
  273. daddu YY, YY, INCY
  274. ST t2, 0 * SIZE(YY)
  275. daddu YY, YY, INCY
  276. ST t3, 0 * SIZE(YY)
  277. daddu YY, YY, INCY
  278. ST t4, 0 * SIZE(YY)
  279. daddiu I, I, -1
  280. bgtz I, .L22
  281. daddu YY, YY, INCY
  282. .align 3
  283. .L23:
  284. MADD t1, b1, ALPHA, a1
  285. MADD t2, b2, ALPHA, a2
  286. MADD t3, b3, ALPHA, a3
  287. MADD t4, b4, ALPHA, a4
  288. ST t1, 0 * SIZE(YY)
  289. daddu YY, YY, INCY
  290. MADD t1, b5, ALPHA, a5
  291. ST t2, 0 * SIZE(YY)
  292. daddu YY, YY, INCY
  293. MADD t2, b6, ALPHA, a6
  294. ST t3, 0 * SIZE(YY)
  295. daddu YY, YY, INCY
  296. MADD t3, b7, ALPHA, a7
  297. ST t4, 0 * SIZE(YY)
  298. daddu YY, YY, INCY
  299. MADD t4, b8, ALPHA, a8
  300. ST t1, 0 * SIZE(YY)
  301. daddu YY, YY, INCY
  302. ST t2, 0 * SIZE(YY)
  303. daddu YY, YY, INCY
  304. ST t3, 0 * SIZE(YY)
  305. daddu YY, YY, INCY
  306. ST t4, 0 * SIZE(YY)
  307. daddu YY, YY, INCY
  308. .align 3
  309. .L25:
  310. andi I, N, 7
  311. blez I, .L999
  312. NOP
  313. .align 3
  314. .L26:
  315. LD a1, 0 * SIZE(X)
  316. LD b1, 0 * SIZE(Y)
  317. MADD t1, b1, ALPHA, a1
  318. daddu X, X, INCX
  319. ST t1, 0 * SIZE(Y)
  320. daddiu I, I, -1
  321. bgtz I, .L26
  322. daddu Y, Y, INCY
  323. .align 3
  324. .L999:
  325. #ifndef __64BIT__
  326. ldc1 $f20, 0($sp)
  327. ldc1 $f21, 8($sp)
  328. daddiu $sp, $sp, 16
  329. #endif
  330. j $31
  331. NOP
  332. EPILOGUE