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.

zscal.S 8.4 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441
  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 I $2
  44. #define TEMP $3
  45. #define XX $5
  46. #define ALPHA_R $f15
  47. #define ALPHA_I $f16
  48. #define a1 $f0
  49. #define a2 $f1
  50. #define a3 $f2
  51. #define a4 $f3
  52. #define a5 $f4
  53. #define a6 $f5
  54. #define a7 $f6
  55. #define a8 $f7
  56. #define t1 $f8
  57. #define t2 $f9
  58. #define t3 $f10
  59. #define t4 $f11
  60. PROLOGUE
  61. li TEMP, 2 * SIZE
  62. MTC $0, a1
  63. blez N, .L999
  64. dsll INCX, INCX, ZBASE_SHIFT
  65. CMPEQ $fcc0, ALPHA_R, a1
  66. CMPEQ $fcc1, ALPHA_I, a1
  67. bc1f $fcc0, .L50
  68. NOP
  69. bc1f $fcc1, .L50
  70. NOP
  71. bne INCX, TEMP, .L20
  72. dsra I, N, 2
  73. blez I, .L15
  74. NOP
  75. .align 3
  76. .L12:
  77. ST a1, 0 * SIZE(X)
  78. ST a1, 1 * SIZE(X)
  79. ST a1, 2 * SIZE(X)
  80. ST a1, 3 * SIZE(X)
  81. ST a1, 4 * SIZE(X)
  82. ST a1, 5 * SIZE(X)
  83. ST a1, 6 * SIZE(X)
  84. ST a1, 7 * SIZE(X)
  85. addiu I, I, -1
  86. bgtz I, .L12
  87. daddiu X, X, 8 * SIZE
  88. .align 3
  89. .L15:
  90. andi I, N, 3
  91. blez I, .L999
  92. NOP
  93. .align 3
  94. .L16:
  95. ST a1, 0 * SIZE(X)
  96. ST a1, 1 * SIZE(X)
  97. daddiu I, I, -1
  98. bgtz I, .L16
  99. daddiu X, X, 2 * SIZE
  100. j $31
  101. NOP
  102. .align 3
  103. .L20:
  104. dsra I, N, 2
  105. blez I, .L25
  106. NOP
  107. .align 3
  108. .L22:
  109. ST a1, 0 * SIZE(X)
  110. ST a1, 1 * SIZE(X)
  111. daddu X, X, INCX
  112. ST a1, 0 * SIZE(X)
  113. ST a1, 1 * SIZE(X)
  114. daddu X, X, INCX
  115. ST a1, 0 * SIZE(X)
  116. ST a1, 1 * SIZE(X)
  117. daddu X, X, INCX
  118. ST a1, 0 * SIZE(X)
  119. ST a1, 1 * SIZE(X)
  120. daddiu I, I, -1
  121. bgtz I, .L22
  122. daddu X, X, INCX
  123. .align 3
  124. .L25:
  125. andi I, N, 3
  126. blez I, .L999
  127. NOP
  128. .align 3
  129. .L26:
  130. ST a1, 0 * SIZE(X)
  131. daddiu I, I, -1
  132. ST a1, 1 * SIZE(X)
  133. bgtz I, .L26
  134. daddu X, X, INCX
  135. j $31
  136. NOP
  137. .align 3
  138. .L50:
  139. bne INCX, TEMP, .L60
  140. dsra I, N, 2
  141. blez I, .L55
  142. daddiu I, I, -1
  143. LD a1, 0 * SIZE(X)
  144. LD a2, 1 * SIZE(X)
  145. LD a3, 2 * SIZE(X)
  146. LD a4, 3 * SIZE(X)
  147. LD a5, 4 * SIZE(X)
  148. LD a6, 5 * SIZE(X)
  149. MUL t1, ALPHA_R, a1
  150. LD a7, 6 * SIZE(X)
  151. MUL t2, ALPHA_I, a1
  152. LD a8, 7 * SIZE(X)
  153. MUL t3, ALPHA_R, a3
  154. blez I, .L53
  155. MUL t4, ALPHA_I, a3
  156. .align 3
  157. .L52:
  158. NMSUB t1, t1, ALPHA_I, a2
  159. LD a1, 8 * SIZE(X)
  160. MADD t2, t2, ALPHA_R, a2
  161. LD a2, 9 * SIZE(X)
  162. NMSUB t3, t3, ALPHA_I, a4
  163. LD a3, 10 * SIZE(X)
  164. MADD t4, t4, ALPHA_R, a4
  165. LD a4, 11 * SIZE(X)
  166. ST t1, 0 * SIZE(X)
  167. MUL t1, ALPHA_R, a5
  168. ST t2, 1 * SIZE(X)
  169. MUL t2, ALPHA_I, a5
  170. ST t3, 2 * SIZE(X)
  171. MUL t3, ALPHA_R, a7
  172. ST t4, 3 * SIZE(X)
  173. MUL t4, ALPHA_I, a7
  174. NMSUB t1, t1, ALPHA_I, a6
  175. LD a5, 12 * SIZE(X)
  176. MADD t2, t2, ALPHA_R, a6
  177. LD a6, 13 * SIZE(X)
  178. NMSUB t3, t3, ALPHA_I, a8
  179. LD a7, 14 * SIZE(X)
  180. MADD t4, t4, ALPHA_R, a8
  181. LD a8, 15 * SIZE(X)
  182. ST t1, 4 * SIZE(X)
  183. MUL t1, ALPHA_R, a1
  184. ST t2, 5 * SIZE(X)
  185. MUL t2, ALPHA_I, a1
  186. ST t3, 6 * SIZE(X)
  187. MUL t3, ALPHA_R, a3
  188. ST t4, 7 * SIZE(X)
  189. MUL t4, ALPHA_I, a3
  190. daddiu I, I, -1
  191. bgtz I, .L52
  192. daddiu X, X, 8 * SIZE
  193. .align 3
  194. .L53:
  195. NMSUB t1, t1, ALPHA_I, a2
  196. MADD t2, t2, ALPHA_R, a2
  197. NMSUB t3, t3, ALPHA_I, a4
  198. MADD t4, t4, ALPHA_R, a4
  199. ST t1, 0 * SIZE(X)
  200. MUL t1, ALPHA_R, a5
  201. ST t2, 1 * SIZE(X)
  202. MUL t2, ALPHA_I, a5
  203. ST t3, 2 * SIZE(X)
  204. MUL t3, ALPHA_R, a7
  205. ST t4, 3 * SIZE(X)
  206. MUL t4, ALPHA_I, a7
  207. NMSUB t1, t1, ALPHA_I, a6
  208. MADD t2, t2, ALPHA_R, a6
  209. NMSUB t3, t3, ALPHA_I, a8
  210. MADD t4, t4, ALPHA_R, a8
  211. ST t1, 4 * SIZE(X)
  212. ST t2, 5 * SIZE(X)
  213. ST t3, 6 * SIZE(X)
  214. ST t4, 7 * SIZE(X)
  215. daddiu X, X, 8 * SIZE
  216. .align 3
  217. .L55:
  218. andi I, N, 3
  219. blez I, .L999
  220. NOP
  221. .align 3
  222. .L56:
  223. LD a1, 0 * SIZE(X)
  224. LD a2, 1 * SIZE(X)
  225. MUL t1, ALPHA_R, a1
  226. MUL t2, ALPHA_I, a1
  227. NMSUB t1, t1, ALPHA_I, a2
  228. MADD t2, t2, ALPHA_R, a2
  229. daddiu X, X, 2 * SIZE
  230. daddiu I, I, -1
  231. ST t1, -2 * SIZE(X)
  232. bgtz I, .L56
  233. ST t2, -1 * SIZE(X)
  234. j $31
  235. NOP
  236. .align 3
  237. .L60:
  238. dsra I, N, 2
  239. move XX, X
  240. blez I, .L65
  241. daddiu I, I, -1
  242. LD a1, 0 * SIZE(X)
  243. LD a2, 1 * SIZE(X)
  244. daddu X, X, INCX
  245. LD a3, 0 * SIZE(X)
  246. LD a4, 1 * SIZE(X)
  247. daddu X, X, INCX
  248. LD a5, 0 * SIZE(X)
  249. LD a6, 1 * SIZE(X)
  250. daddu X, X, INCX
  251. MUL t1, ALPHA_R, a1
  252. LD a7, 0 * SIZE(X)
  253. MUL t2, ALPHA_I, a1
  254. LD a8, 1 * SIZE(X)
  255. MUL t3, ALPHA_R, a3
  256. daddu X, X, INCX
  257. blez I, .L63
  258. MUL t4, ALPHA_I, a3
  259. .align 3
  260. .L62:
  261. NMSUB t1, t1, ALPHA_I, a2
  262. LD a1, 0 * SIZE(X)
  263. MADD t2, t2, ALPHA_R, a2
  264. LD a2, 1 * SIZE(X)
  265. daddu X, X, INCX
  266. NMSUB t3, t3, ALPHA_I, a4
  267. LD a3, 0 * SIZE(X)
  268. MADD t4, t4, ALPHA_R, a4
  269. LD a4, 1 * SIZE(X)
  270. daddu X, X, INCX
  271. ST t1, 0 * SIZE(XX)
  272. MUL t1, ALPHA_R, a5
  273. ST t2, 1 * SIZE(XX)
  274. MUL t2, ALPHA_I, a5
  275. daddu XX, XX, INCX
  276. ST t3, 0 * SIZE(XX)
  277. MUL t3, ALPHA_R, a7
  278. ST t4, 1 * SIZE(XX)
  279. MUL t4, ALPHA_I, a7
  280. daddu XX, XX, INCX
  281. NMSUB t1, t1, ALPHA_I, a6
  282. LD a5, 0 * SIZE(X)
  283. MADD t2, t2, ALPHA_R, a6
  284. LD a6, 1 * SIZE(X)
  285. daddu X, X, INCX
  286. NMSUB t3, t3, ALPHA_I, a8
  287. LD a7, 0 * SIZE(X)
  288. MADD t4, t4, ALPHA_R, a8
  289. LD a8, 1 * SIZE(X)
  290. daddu X, X, INCX
  291. ST t1, 0 * SIZE(XX)
  292. MUL t1, ALPHA_R, a1
  293. ST t2, 1 * SIZE(XX)
  294. MUL t2, ALPHA_I, a1
  295. daddu XX, XX, INCX
  296. ST t3, 0 * SIZE(XX)
  297. MUL t3, ALPHA_R, a3
  298. ST t4, 1 * SIZE(XX)
  299. MUL t4, ALPHA_I, a3
  300. daddiu I, I, -1
  301. bgtz I, .L62
  302. daddu XX, XX, INCX
  303. .align 3
  304. .L63:
  305. NMSUB t1, t1, ALPHA_I, a2
  306. MADD t2, t2, ALPHA_R, a2
  307. NMSUB t3, t3, ALPHA_I, a4
  308. MADD t4, t4, ALPHA_R, a4
  309. ST t1, 0 * SIZE(XX)
  310. MUL t1, ALPHA_R, a5
  311. ST t2, 1 * SIZE(XX)
  312. MUL t2, ALPHA_I, a5
  313. daddu XX, XX, INCX
  314. ST t3, 0 * SIZE(XX)
  315. MUL t3, ALPHA_R, a7
  316. ST t4, 1 * SIZE(XX)
  317. MUL t4, ALPHA_I, a7
  318. daddu XX, XX, INCX
  319. NMSUB t1, t1, ALPHA_I, a6
  320. MADD t2, t2, ALPHA_R, a6
  321. NMSUB t3, t3, ALPHA_I, a8
  322. MADD t4, t4, ALPHA_R, a8
  323. ST t1, 0 * SIZE(XX)
  324. ST t2, 1 * SIZE(XX)
  325. daddu XX, XX, INCX
  326. ST t3, 0 * SIZE(XX)
  327. ST t4, 1 * SIZE(XX)
  328. daddu XX, XX, INCX
  329. .align 3
  330. .L65:
  331. andi I, N, 3
  332. blez I, .L999
  333. NOP
  334. .align 3
  335. .L66:
  336. LD a1, 0 * SIZE(X)
  337. LD a2, 1 * SIZE(X)
  338. MUL t1, ALPHA_R, a1
  339. MUL t2, ALPHA_I, a1
  340. NMSUB t1, t1, ALPHA_I, a2
  341. MADD t2, t2, ALPHA_R, a2
  342. daddiu I, I, -1
  343. ST t1, 0 * SIZE(X)
  344. ST t2, 1 * SIZE(X)
  345. bgtz I, .L66
  346. daddu X, X, INCX
  347. .align 3
  348. .L999:
  349. j $31
  350. NOP
  351. EPILOGUE