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.

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480
  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 PREFETCHSIZE 88
  41. #define N $16
  42. #define X $20
  43. #define INCX $21
  44. #define XX $18
  45. #define I $19
  46. #define ALPHA $f19
  47. #define s0 $f0
  48. #define s1 $f1
  49. #define s2 $f10
  50. #define s3 $f11
  51. #define a0 $f12
  52. #define a1 $f13
  53. #define a2 $f14
  54. #define a3 $f15
  55. #define a4 $f16
  56. #define a5 $f17
  57. #define a6 $f18
  58. #define a7 $f21
  59. #define t0 $f22
  60. #define t1 $f23
  61. #define t2 $f24
  62. #define t3 $f25
  63. PROLOGUE
  64. PROFCODE
  65. mov X, XX
  66. ble N, $L999
  67. cmpeq INCX, 1, $0
  68. beq $0, $L20
  69. #ifndef DOUBLE
  70. sra N, 4, I
  71. ble I, $L15
  72. LD a0, 0 * SIZE(X)
  73. LD a1, 1 * SIZE(X)
  74. LD a2, 2 * SIZE(X)
  75. LD a3, 3 * SIZE(X)
  76. LD a4, 4 * SIZE(X)
  77. MUL a0, ALPHA, t0
  78. LD a5, 5 * SIZE(X)
  79. MUL a1, ALPHA, t1
  80. LD a6, 6 * SIZE(X)
  81. MUL a2, ALPHA, t2
  82. LD a7, 7 * SIZE(X)
  83. MUL a3, ALPHA, t3
  84. ST t0, 0 * SIZE(X)
  85. MUL a4, ALPHA, t0
  86. ST t1, 1 * SIZE(X)
  87. MUL a5, ALPHA, t1
  88. ST t2, 2 * SIZE(X)
  89. MUL a6, ALPHA, t2
  90. ST t3, 3 * SIZE(X)
  91. MUL a7, ALPHA, t3
  92. LD a0, 8 * SIZE(X)
  93. LD a1, 9 * SIZE(X)
  94. LD a2, 10 * SIZE(X)
  95. LD a3, 11 * SIZE(X)
  96. ST t0, 4 * SIZE(X)
  97. MUL a0, ALPHA, t0
  98. ST t1, 5 * SIZE(X)
  99. MUL a1, ALPHA, t1
  100. ST t2, 6 * SIZE(X)
  101. MUL a2, ALPHA, t2
  102. ST t3, 7 * SIZE(X)
  103. MUL a3, ALPHA, t3
  104. LD a4, 12 * SIZE(X)
  105. LD a5, 13 * SIZE(X)
  106. LD a6, 14 * SIZE(X)
  107. LD a7, 15 * SIZE(X)
  108. lda I, -1(I)
  109. ble I, $L13
  110. .align 4
  111. $L12:
  112. ST t0, 8 * SIZE(X)
  113. MUL a4, ALPHA, t0
  114. ST t1, 9 * SIZE(X)
  115. MUL a5, ALPHA, t1
  116. ST t2, 10 * SIZE(X)
  117. MUL a6, ALPHA, t2
  118. ST t3, 11 * SIZE(X)
  119. MUL a7, ALPHA, t3
  120. LD a0, 16 * SIZE(X)
  121. LD a1, 17 * SIZE(X)
  122. LD a2, 18 * SIZE(X)
  123. LD a3, 19 * SIZE(X)
  124. ST t0, 12 * SIZE(X)
  125. MUL a0, ALPHA, t0
  126. ST t1, 13 * SIZE(X)
  127. MUL a1, ALPHA, t1
  128. ST t2, 14 * SIZE(X)
  129. MUL a2, ALPHA, t2
  130. ST t3, 15 * SIZE(X)
  131. MUL a3, ALPHA, t3
  132. LD a4, 20 * SIZE(X)
  133. LD a5, 21 * SIZE(X)
  134. LD a6, 22 * SIZE(X)
  135. LD a7, 23 * SIZE(X)
  136. ST t0, 16 * SIZE(X)
  137. MUL a4, ALPHA, t0
  138. ST t1, 17 * SIZE(X)
  139. MUL a5, ALPHA, t1
  140. ST t2, 18 * SIZE(X)
  141. MUL a6, ALPHA, t2
  142. ST t3, 19 * SIZE(X)
  143. MUL a7, ALPHA, t3
  144. LD a0, 24 * SIZE(X)
  145. LD a1, 25 * SIZE(X)
  146. LD a2, 26 * SIZE(X)
  147. LD a3, 27 * SIZE(X)
  148. ST t0, 20 * SIZE(X)
  149. MUL a0, ALPHA, t0
  150. ST t1, 21 * SIZE(X)
  151. MUL a1, ALPHA, t1
  152. ST t2, 22 * SIZE(X)
  153. MUL a2, ALPHA, t2
  154. ST t3, 23 * SIZE(X)
  155. MUL a3, ALPHA, t3
  156. LD a4, 28 * SIZE(X)
  157. LD a5, 29 * SIZE(X)
  158. LD a6, 30 * SIZE(X)
  159. LD a7, 31 * SIZE(X)
  160. lds $f31, PREFETCHSIZE * SIZE(X)
  161. lda I, -1(I)
  162. addq X, 16 * SIZE, X
  163. bne I, $L12
  164. .align 4
  165. $L13:
  166. ST t0, 8 * SIZE(X)
  167. MUL a4, ALPHA, t0
  168. ST t1, 9 * SIZE(X)
  169. MUL a5, ALPHA, t1
  170. ST t2, 10 * SIZE(X)
  171. MUL a6, ALPHA, t2
  172. ST t3, 11 * SIZE(X)
  173. MUL a7, ALPHA, t3
  174. ST t0, 12 * SIZE(X)
  175. ST t1, 13 * SIZE(X)
  176. ST t2, 14 * SIZE(X)
  177. ST t3, 15 * SIZE(X)
  178. addq X, 16 * SIZE, X
  179. .align 4
  180. $L15:
  181. and N, 15, I
  182. #else
  183. sra N, 3, I
  184. ble I, $L15
  185. LD a0, 0 * SIZE(X)
  186. LD a1, 1 * SIZE(X)
  187. LD a2, 2 * SIZE(X)
  188. LD a3, 3 * SIZE(X)
  189. LD a4, 4 * SIZE(X)
  190. MUL a0, ALPHA, t0
  191. LD a5, 5 * SIZE(X)
  192. MUL a1, ALPHA, t1
  193. LD a6, 6 * SIZE(X)
  194. MUL a2, ALPHA, t2
  195. LD a7, 7 * SIZE(X)
  196. MUL a3, ALPHA, t3
  197. lda I, -1(I)
  198. ble I, $L13
  199. .align 4
  200. $L12:
  201. ST t0, 0 * SIZE(X)
  202. MUL a4, ALPHA, t0
  203. ST t1, 1 * SIZE(X)
  204. MUL a5, ALPHA, t1
  205. ST t2, 2 * SIZE(X)
  206. MUL a6, ALPHA, t2
  207. ST t3, 3 * SIZE(X)
  208. MUL a7, ALPHA, t3
  209. LD a0, 8 * SIZE(X)
  210. lda I, -1(I)
  211. LD a1, 9 * SIZE(X)
  212. addq X, 8 * SIZE, X
  213. LD a2, 2 * SIZE(X)
  214. LD a3, 3 * SIZE(X)
  215. ST t0, -4 * SIZE(X)
  216. MUL a0, ALPHA, t0
  217. ST t1, -3 * SIZE(X)
  218. MUL a1, ALPHA, t1
  219. ST t2, -2 * SIZE(X)
  220. MUL a2, ALPHA, t2
  221. ST t3, -1 * SIZE(X)
  222. MUL a3, ALPHA, t3
  223. LD a4, 4 * SIZE(X)
  224. LD a5, 5 * SIZE(X)
  225. LD a6, 6 * SIZE(X)
  226. LD a7, 7 * SIZE(X)
  227. lds $f31, PREFETCHSIZE * SIZE(X)
  228. bne I, $L12
  229. .align 4
  230. $L13:
  231. ST t0, 0 * SIZE(X)
  232. MUL a4, ALPHA, t0
  233. ST t1, 1 * SIZE(X)
  234. MUL a5, ALPHA, t1
  235. ST t2, 2 * SIZE(X)
  236. MUL a6, ALPHA, t2
  237. ST t3, 3 * SIZE(X)
  238. MUL a7, ALPHA, t3
  239. ST t0, 4 * SIZE(X)
  240. ST t1, 5 * SIZE(X)
  241. ST t2, 6 * SIZE(X)
  242. ST t3, 7 * SIZE(X)
  243. addq X, 8 * SIZE, X
  244. .align 4
  245. $L15:
  246. and N, 7, I
  247. #endif
  248. unop
  249. unop
  250. ble I, $L999
  251. .align 4
  252. $L17:
  253. LD a0, 0 * SIZE(X)
  254. MUL a0, ALPHA, t0
  255. ST t0, 0 * SIZE(X)
  256. addq X, SIZE, X
  257. lda I, -1(I)
  258. bne I, $L17
  259. ret
  260. .align 4
  261. $L20:
  262. sra N, 3, I
  263. ble I, $L25
  264. LD a0, 0 * SIZE(X)
  265. SXADDQ INCX, X, X
  266. LD a1, 0 * SIZE(X)
  267. SXADDQ INCX, X, X
  268. LD a2, 0 * SIZE(X)
  269. SXADDQ INCX, X, X
  270. LD a3, 0 * SIZE(X)
  271. SXADDQ INCX, X, X
  272. LD a4, 0 * SIZE(X)
  273. MUL a0, ALPHA, t0
  274. lda I, -1(I)
  275. SXADDQ INCX, X, X
  276. LD a5, 0 * SIZE(X)
  277. MUL a1, ALPHA, t1
  278. SXADDQ INCX, X, X
  279. unop
  280. LD a6, 0 * SIZE(X)
  281. MUL a2, ALPHA, t2
  282. SXADDQ INCX, X, X
  283. unop
  284. LD a7, 0 * SIZE(X)
  285. MUL a3, ALPHA, t3
  286. SXADDQ INCX, X, X
  287. ble I, $L23
  288. .align 4
  289. $L22:
  290. ST t0, 0 * SIZE(XX)
  291. MUL a4, ALPHA, t0
  292. lds $f31, PREFETCHSIZE * SIZE(X)
  293. SXADDQ INCX, XX, XX
  294. LD a0, 0 * SIZE(X)
  295. SXADDQ INCX, X, X
  296. lda I, -1(I)
  297. unop
  298. ST t1, 0 * SIZE(XX)
  299. MUL a5, ALPHA, t1
  300. SXADDQ INCX, XX, XX
  301. unop
  302. LD a1, 0 * SIZE(X)
  303. SXADDQ INCX, X, X
  304. ST t2, 0 * SIZE(XX)
  305. MUL a6, ALPHA, t2
  306. SXADDQ INCX, XX, XX
  307. unop
  308. LD a2, 0 * SIZE(X)
  309. SXADDQ INCX, X, X
  310. ST t3, 0 * SIZE(XX)
  311. MUL a7, ALPHA, t3
  312. SXADDQ INCX, XX, XX
  313. unop
  314. LD a3, 0 * SIZE(X)
  315. SXADDQ INCX, X, X
  316. ST t0, 0 * SIZE(XX)
  317. MUL a0, ALPHA, t0
  318. SXADDQ INCX, XX, XX
  319. unop
  320. LD a4, 0 * SIZE(X)
  321. SXADDQ INCX, X, X
  322. ST t1, 0 * SIZE(XX)
  323. MUL a1, ALPHA, t1
  324. SXADDQ INCX, XX, XX
  325. unop
  326. LD a5, 0 * SIZE(X)
  327. SXADDQ INCX, X, X
  328. ST t2, 0 * SIZE(XX)
  329. MUL a2, ALPHA, t2
  330. SXADDQ INCX, XX, XX
  331. unop
  332. LD a6, 0 * SIZE(X)
  333. SXADDQ INCX, X, X
  334. ST t3, 0 * SIZE(XX)
  335. MUL a3, ALPHA, t3
  336. SXADDQ INCX, XX, XX
  337. unop
  338. LD a7, 0 * SIZE(X)
  339. SXADDQ INCX, X, X
  340. unop
  341. bne I, $L22
  342. .align 4
  343. $L23:
  344. ST t0, 0 * SIZE(XX)
  345. MUL a4, ALPHA, t0
  346. SXADDQ INCX, XX, XX
  347. ST t1, 0 * SIZE(XX)
  348. MUL a5, ALPHA, t1
  349. SXADDQ INCX, XX, XX
  350. ST t2, 0 * SIZE(XX)
  351. MUL a6, ALPHA, t2
  352. SXADDQ INCX, XX, XX
  353. ST t3, 0 * SIZE(XX)
  354. MUL a7, ALPHA, t3
  355. SXADDQ INCX, XX, XX
  356. ST t0, 0 * SIZE(XX)
  357. SXADDQ INCX, XX, XX
  358. ST t1, 0 * SIZE(XX)
  359. SXADDQ INCX, XX, XX
  360. ST t2, 0 * SIZE(XX)
  361. SXADDQ INCX, XX, XX
  362. ST t3, 0 * SIZE(XX)
  363. SXADDQ INCX, XX, XX
  364. .align 4
  365. $L25:
  366. and N, 7, I
  367. unop
  368. unop
  369. ble I, $L999
  370. .align 4
  371. $L27:
  372. LD a0, 0 * SIZE(X)
  373. MUL a0, ALPHA, t0
  374. ST t0, 0 * SIZE(XX)
  375. SXADDQ INCX, X, X
  376. SXADDQ INCX, XX, XX
  377. lda I, -1(I)
  378. bne I, $L27
  379. .align 4
  380. $L999:
  381. ret
  382. EPILOGUE