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 9.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438
  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 $9
  42. #define INCX $10
  43. #define Y $11
  44. #define INCY $8
  45. #define I $2
  46. #define TEMP $3
  47. #define YY $5
  48. #define ALPHA_R $f15
  49. #define ALPHA_I $f16
  50. #define a1 $f0
  51. #define a2 $f1
  52. #define a3 $f2
  53. #define a4 $f3
  54. #define a5 $f4
  55. #define a6 $f5
  56. #define a7 $f6
  57. #define a8 $f7
  58. #define b1 $f8
  59. #define b2 $f9
  60. #define b3 $f10
  61. #define b4 $f11
  62. #define b5 $f12
  63. #define b6 $f13
  64. #define b7 $f14
  65. #define b8 $f17
  66. #define t1 $f18
  67. #define t2 $f19
  68. #define t3 $f20
  69. #define t4 $f21
  70. #ifndef CONJ
  71. #define MADD1 NMSUB
  72. #define MADD2 MADD
  73. #else
  74. #define MADD1 MADD
  75. #define MADD2 NMSUB
  76. #endif
  77. PROLOGUE
  78. LDARG INCY, 0($sp)
  79. li TEMP, 2 * SIZE
  80. #ifndef __64BIT__
  81. daddiu $sp, $sp, -16
  82. sdc1 $f20, 0($sp)
  83. sdc1 $f21, 8($sp)
  84. #endif
  85. blez N, .L999
  86. dsll INCX, INCX, ZBASE_SHIFT
  87. bne INCX, TEMP, .L20
  88. dsll INCY, INCY, ZBASE_SHIFT
  89. bne INCY, TEMP, .L20
  90. dsra I, N, 2
  91. blez I, .L15
  92. daddiu I, I, -1
  93. LD a1, 0 * SIZE(X)
  94. LD b1, 0 * SIZE(Y)
  95. LD a2, 1 * SIZE(X)
  96. LD b2, 1 * SIZE(Y)
  97. LD a3, 2 * SIZE(X)
  98. LD b3, 2 * SIZE(Y)
  99. LD a4, 3 * SIZE(X)
  100. LD b4, 3 * SIZE(Y)
  101. LD a5, 4 * SIZE(X)
  102. LD b5, 4 * SIZE(Y)
  103. LD a6, 5 * SIZE(X)
  104. LD b6, 5 * SIZE(Y)
  105. LD a7, 6 * SIZE(X)
  106. LD b7, 6 * SIZE(Y)
  107. LD a8, 7 * SIZE(X)
  108. LD b8, 7 * SIZE(Y)
  109. blez I, .L13
  110. NOP
  111. .align 3
  112. .L12:
  113. MADD t1, b1, ALPHA_R, a1
  114. LD b1, 8 * SIZE(Y)
  115. MADD t2, b2, ALPHA_I, a1
  116. LD a1, 8 * SIZE(X)
  117. MADD t3, b3, ALPHA_R, a3
  118. LD b3, 10 * SIZE(Y)
  119. MADD t4, b4, ALPHA_I, a3
  120. LD a3, 10 * SIZE(X)
  121. MADD1 t1, t1, ALPHA_I, a2
  122. LD b2, 9 * SIZE(Y)
  123. MADD2 t2, t2, ALPHA_R, a2
  124. LD a2, 9 * SIZE(X)
  125. MADD1 t3, t3, ALPHA_I, a4
  126. LD b4, 11 * SIZE(Y)
  127. MADD2 t4, t4, ALPHA_R, a4
  128. LD a4, 11 * SIZE(X)
  129. ST t1, 0 * SIZE(Y)
  130. ST t2, 1 * SIZE(Y)
  131. ST t3, 2 * SIZE(Y)
  132. ST t4, 3 * SIZE(Y)
  133. MADD t1, b5, ALPHA_R, a5
  134. LD b5, 12 * SIZE(Y)
  135. MADD t2, b6, ALPHA_I, a5
  136. LD a5, 12 * SIZE(X)
  137. MADD t3, b7, ALPHA_R, a7
  138. LD b7, 14 * SIZE(Y)
  139. MADD t4, b8, ALPHA_I, a7
  140. LD a7, 14 * SIZE(X)
  141. MADD1 t1, t1, ALPHA_I, a6
  142. LD b6, 13 * SIZE(Y)
  143. MADD2 t2, t2, ALPHA_R, a6
  144. LD a6, 13 * SIZE(X)
  145. MADD1 t3, t3, ALPHA_I, a8
  146. LD b8, 15 * SIZE(Y)
  147. MADD2 t4, t4, ALPHA_R, a8
  148. LD a8, 15 * SIZE(X)
  149. ST t1, 4 * SIZE(Y)
  150. ST t2, 5 * SIZE(Y)
  151. ST t3, 6 * SIZE(Y)
  152. ST t4, 7 * SIZE(Y)
  153. daddiu I, I, -1
  154. daddiu Y, Y, 8 * SIZE
  155. bgtz I, .L12
  156. daddiu X, X, 8 * SIZE
  157. .align 3
  158. .L13:
  159. MADD t1, b1, ALPHA_R, a1
  160. MADD t2, b2, ALPHA_I, a1
  161. MADD t3, b3, ALPHA_R, a3
  162. MADD t4, b4, ALPHA_I, a3
  163. MADD1 t1, t1, ALPHA_I, a2
  164. MADD2 t2, t2, ALPHA_R, a2
  165. MADD1 t3, t3, ALPHA_I, a4
  166. MADD2 t4, t4, ALPHA_R, a4
  167. ST t1, 0 * SIZE(Y)
  168. MADD t1, b5, ALPHA_R, a5
  169. ST t2, 1 * SIZE(Y)
  170. MADD t2, b6, ALPHA_I, a5
  171. ST t3, 2 * SIZE(Y)
  172. MADD t3, b7, ALPHA_R, a7
  173. ST t4, 3 * SIZE(Y)
  174. MADD t4, b8, ALPHA_I, a7
  175. MADD1 t1, t1, ALPHA_I, a6
  176. MADD2 t2, t2, ALPHA_R, a6
  177. MADD1 t3, t3, ALPHA_I, a8
  178. MADD2 t4, t4, ALPHA_R, a8
  179. ST t1, 4 * SIZE(Y)
  180. ST t2, 5 * SIZE(Y)
  181. ST t3, 6 * SIZE(Y)
  182. ST t4, 7 * SIZE(Y)
  183. daddiu X, X, 8 * SIZE
  184. daddiu Y, Y, 8 * SIZE
  185. .align 3
  186. .L15:
  187. andi I, N, 3
  188. blez I, .L999
  189. NOP
  190. .align 3
  191. .L16:
  192. LD a1, 0 * SIZE(X)
  193. LD a2, 1 * SIZE(X)
  194. LD b1, 0 * SIZE(Y)
  195. LD b2, 1 * SIZE(Y)
  196. MADD t1, b1, ALPHA_R, a1
  197. daddiu X, X, 2 * SIZE
  198. MADD t2, b2, ALPHA_I, a1
  199. MADD1 t1, t1, ALPHA_I, a2
  200. daddiu I, I, -1
  201. MADD2 t2, t2, ALPHA_R, a2
  202. daddiu Y, Y, 2 * SIZE
  203. ST t1, -2 * SIZE(Y)
  204. bgtz I, .L16
  205. ST t2, -1 * SIZE(Y)
  206. #ifndef __64BIT__
  207. ldc1 $f20, 0($sp)
  208. ldc1 $f21, 8($sp)
  209. daddiu $sp, $sp, 16
  210. #endif
  211. j $31
  212. NOP
  213. .align 3
  214. .L20:
  215. dsra I, N, 2
  216. move YY, Y
  217. blez I, .L25
  218. daddiu I, I, -1
  219. LD a1, 0 * SIZE(X)
  220. LD b1, 0 * SIZE(Y)
  221. LD a2, 1 * SIZE(X)
  222. LD b2, 1 * SIZE(Y)
  223. daddu X, X, INCX
  224. daddu Y, Y, INCY
  225. LD a3, 0 * SIZE(X)
  226. LD b3, 0 * SIZE(Y)
  227. LD a4, 1 * SIZE(X)
  228. LD b4, 1 * SIZE(Y)
  229. daddu X, X, INCX
  230. daddu Y, Y, INCY
  231. LD a5, 0 * SIZE(X)
  232. LD b5, 0 * SIZE(Y)
  233. LD a6, 1 * SIZE(X)
  234. LD b6, 1 * SIZE(Y)
  235. daddu X, X, INCX
  236. daddu Y, Y, INCY
  237. LD a7, 0 * SIZE(X)
  238. blez I, .L23
  239. LD b7, 0 * SIZE(Y)
  240. .align 3
  241. .L22:
  242. MADD t1, b1, ALPHA_R, a1
  243. LD b8, 1 * SIZE(Y)
  244. daddu Y, Y, INCY
  245. MADD t2, b2, ALPHA_I, a1
  246. LD a8, 1 * SIZE(X)
  247. daddu X, X, INCX
  248. MADD t3, b3, ALPHA_R, a3
  249. LD b1, 0 * SIZE(Y)
  250. MADD t4, b4, ALPHA_I, a3
  251. LD a1, 0 * SIZE(X)
  252. MADD1 t1, t1, ALPHA_I, a2
  253. LD b2, 1 * SIZE(Y)
  254. daddu Y, Y, INCY
  255. MADD2 t2, t2, ALPHA_R, a2
  256. LD a2, 1 * SIZE(X)
  257. daddu X, X, INCX
  258. MADD1 t3, t3, ALPHA_I, a4
  259. LD a3, 0 * SIZE(X)
  260. MADD2 t4, t4, ALPHA_R, a4
  261. LD b3, 0 * SIZE(Y)
  262. ST t1, 0 * SIZE(YY)
  263. ST t2, 1 * SIZE(YY)
  264. daddu YY, YY, INCY
  265. ST t3, 0 * SIZE(YY)
  266. ST t4, 1 * SIZE(YY)
  267. daddu YY, YY, INCY
  268. MADD t1, b5, ALPHA_R, a5
  269. LD a4, 1 * SIZE(X)
  270. daddu X, X, INCX
  271. MADD t2, b6, ALPHA_I, a5
  272. LD b4, 1 * SIZE(Y)
  273. daddu Y, Y, INCY
  274. MADD t3, b7, ALPHA_R, a7
  275. LD b5, 0 * SIZE(Y)
  276. MADD t4, b8, ALPHA_I, a7
  277. LD a5, 0 * SIZE(X)
  278. MADD1 t1, t1, ALPHA_I, a6
  279. LD b6, 1 * SIZE(Y)
  280. daddu Y, Y, INCY
  281. MADD2 t2, t2, ALPHA_R, a6
  282. LD a6, 1 * SIZE(X)
  283. daddu X, X, INCX
  284. MADD1 t3, t3, ALPHA_I, a8
  285. LD b7, 0 * SIZE(Y)
  286. MADD2 t4, t4, ALPHA_R, a8
  287. LD a7, 0 * SIZE(X)
  288. ST t1, 0 * SIZE(YY)
  289. ST t2, 1 * SIZE(YY)
  290. daddu YY, YY, INCY
  291. ST t3, 0 * SIZE(YY)
  292. ST t4, 1 * SIZE(YY)
  293. daddu YY, YY, INCY
  294. daddiu I, I, -1
  295. bgtz I, .L22
  296. NOP
  297. .align 3
  298. .L23:
  299. MADD t1, b1, ALPHA_R, a1
  300. LD a8, 1 * SIZE(X)
  301. MADD t2, b2, ALPHA_I, a1
  302. LD b8, 1 * SIZE(Y)
  303. MADD t3, b3, ALPHA_R, a3
  304. daddu X, X, INCX
  305. MADD t4, b4, ALPHA_I, a3
  306. daddu Y, Y, INCY
  307. MADD1 t1, t1, ALPHA_I, a2
  308. MADD2 t2, t2, ALPHA_R, a2
  309. MADD1 t3, t3, ALPHA_I, a4
  310. MADD2 t4, t4, ALPHA_R, a4
  311. ST t1, 0 * SIZE(YY)
  312. MADD t1, b5, ALPHA_R, a5
  313. ST t2, 1 * SIZE(YY)
  314. MADD t2, b6, ALPHA_I, a5
  315. daddu YY, YY, INCY
  316. ST t3, 0 * SIZE(YY)
  317. MADD t3, b7, ALPHA_R, a7
  318. ST t4, 1 * SIZE(YY)
  319. MADD t4, b8, ALPHA_I, a7
  320. daddu YY, YY, INCY
  321. MADD1 t1, t1, ALPHA_I, a6
  322. MADD2 t2, t2, ALPHA_R, a6
  323. MADD1 t3, t3, ALPHA_I, a8
  324. MADD2 t4, t4, ALPHA_R, a8
  325. ST t1, 0 * SIZE(YY)
  326. ST t2, 1 * SIZE(YY)
  327. daddu YY, YY, INCY
  328. ST t3, 0 * SIZE(YY)
  329. ST t4, 1 * SIZE(YY)
  330. daddu YY, YY, INCY
  331. .align 3
  332. .L25:
  333. andi I, N, 3
  334. blez I, .L999
  335. NOP
  336. .align 3
  337. .L26:
  338. LD a1, 0 * SIZE(X)
  339. LD a2, 1 * SIZE(X)
  340. LD b1, 0 * SIZE(Y)
  341. LD b2, 1 * SIZE(Y)
  342. MADD t1, b1, ALPHA_R, a1
  343. MADD t2, b2, ALPHA_I, a1
  344. daddu X, X, INCX
  345. MADD1 t1, t1, ALPHA_I, a2
  346. MADD2 t2, t2, ALPHA_R, a2
  347. daddiu I, I, -1
  348. ST t1, 0 * SIZE(Y)
  349. ST t2, 1 * SIZE(Y)
  350. bgtz I, .L26
  351. daddu Y, Y, INCY
  352. .align 3
  353. .L999:
  354. #ifndef __64BIT__
  355. ldc1 $f20, 0($sp)
  356. ldc1 $f21, 8($sp)
  357. daddiu $sp, $sp, 16
  358. #endif
  359. j $31
  360. NOP
  361. EPILOGUE