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.

scal.S 7.6 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412
  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 I $2
  44. #define TEMP $3
  45. #define XX $5
  46. #define ALPHA $f15
  47. #define a1 $f0
  48. #define a2 $f1
  49. #define a3 $f2
  50. #define a4 $f3
  51. #define a5 $f4
  52. #define a6 $f5
  53. #define a7 $f6
  54. #define a8 $f7
  55. #define t1 $f8
  56. #define t2 $f9
  57. #define t3 $f10
  58. #define t4 $f11
  59. PROLOGUE
  60. li TEMP, SIZE
  61. MTC $0, a1
  62. blez N, .L999
  63. dsll INCX, INCX, BASE_SHIFT
  64. CMPEQ $fcc0, ALPHA, a1
  65. NOP
  66. bc1f $fcc0, .L50
  67. NOP
  68. bne INCX, TEMP, .L20
  69. dsra I, N, 3
  70. blez I, .L15
  71. NOP
  72. .align 3
  73. .L12:
  74. ST a1, 0 * SIZE(X)
  75. ST a1, 1 * SIZE(X)
  76. ST a1, 2 * SIZE(X)
  77. ST a1, 3 * SIZE(X)
  78. ST a1, 4 * SIZE(X)
  79. ST a1, 5 * SIZE(X)
  80. ST a1, 6 * SIZE(X)
  81. ST a1, 7 * SIZE(X)
  82. addiu I, I, -1
  83. bgtz I, .L12
  84. daddiu X, X, 8 * SIZE
  85. .align 3
  86. .L15:
  87. andi I, N, 7
  88. blez I, .L999
  89. NOP
  90. .align 3
  91. .L16:
  92. ST a1, 0 * SIZE(X)
  93. daddiu I, I, -1
  94. bgtz I, .L16
  95. daddiu X, X, SIZE
  96. j $31
  97. NOP
  98. .align 3
  99. .L20:
  100. dsra I, N, 3
  101. blez I, .L25
  102. NOP
  103. .align 3
  104. .L22:
  105. ST a1, 0 * SIZE(X)
  106. daddu X, X, INCX
  107. ST a1, 0 * SIZE(X)
  108. daddu X, X, INCX
  109. ST a1, 0 * SIZE(X)
  110. daddu X, X, INCX
  111. ST a1, 0 * SIZE(X)
  112. daddu X, X, INCX
  113. ST a1, 0 * SIZE(X)
  114. daddu X, X, INCX
  115. ST a1, 0 * SIZE(X)
  116. daddu X, X, INCX
  117. ST a1, 0 * SIZE(X)
  118. daddu X, X, INCX
  119. ST a1, 0 * 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, 7
  126. blez I, .L999
  127. NOP
  128. .align 3
  129. .L26:
  130. daddiu I, I, -1
  131. ST a1, 0 * SIZE(X)
  132. bgtz I, .L26
  133. daddu X, X, INCX
  134. j $31
  135. NOP
  136. .align 3
  137. .L50:
  138. bne INCX, TEMP, .L60
  139. dsra I, N, 3
  140. blez I, .L55
  141. daddiu I, I, -1
  142. LD a1, 0 * SIZE(X)
  143. LD a2, 1 * SIZE(X)
  144. LD a3, 2 * SIZE(X)
  145. LD a4, 3 * SIZE(X)
  146. LD a5, 4 * SIZE(X)
  147. LD a6, 5 * SIZE(X)
  148. LD a7, 6 * SIZE(X)
  149. LD a8, 7 * SIZE(X)
  150. blez I, .L53
  151. NOP
  152. .align 3
  153. .L52:
  154. MUL t1, ALPHA, a1
  155. LD a1, 8 * SIZE(X)
  156. MUL t2, ALPHA, a2
  157. LD a2, 9 * SIZE(X)
  158. MUL t3, ALPHA, a3
  159. LD a3, 10 * SIZE(X)
  160. MUL t4, ALPHA, a4
  161. LD a4, 11 * SIZE(X)
  162. ST t1, 0 * SIZE(X)
  163. MUL t1, ALPHA, a5
  164. LD a5, 12 * SIZE(X)
  165. ST t2, 1 * SIZE(X)
  166. MUL t2, ALPHA, a6
  167. LD a6, 13 * SIZE(X)
  168. ST t3, 2 * SIZE(X)
  169. MUL t3, ALPHA, a7
  170. LD a7, 14 * SIZE(X)
  171. ST t4, 3 * SIZE(X)
  172. MUL t4, ALPHA, a8
  173. LD a8, 15 * SIZE(X)
  174. daddiu I, I, -1
  175. ST t1, 4 * SIZE(X)
  176. ST t2, 5 * SIZE(X)
  177. ST t3, 6 * SIZE(X)
  178. ST t4, 7 * SIZE(X)
  179. bgtz I, .L52
  180. daddiu X, X, 8 * SIZE
  181. .align 3
  182. .L53:
  183. MUL t1, ALPHA, a1
  184. MUL t2, ALPHA, a2
  185. MUL t3, ALPHA, a3
  186. MUL t4, ALPHA, a4
  187. ST t1, 0 * SIZE(X)
  188. MUL t1, ALPHA, a5
  189. ST t2, 1 * SIZE(X)
  190. MUL t2, ALPHA, a6
  191. ST t3, 2 * SIZE(X)
  192. MUL t3, ALPHA, a7
  193. ST t4, 3 * SIZE(X)
  194. MUL t4, ALPHA, a8
  195. ST t1, 4 * SIZE(X)
  196. ST t2, 5 * SIZE(X)
  197. ST t3, 6 * SIZE(X)
  198. ST t4, 7 * SIZE(X)
  199. daddiu X, X, 8 * SIZE
  200. .align 3
  201. .L55:
  202. andi I, N, 7
  203. blez I, .L999
  204. NOP
  205. .align 3
  206. .L56:
  207. LD a1, 0 * SIZE(X)
  208. MUL t1, ALPHA, a1
  209. daddiu X, X, SIZE
  210. daddiu I, I, -1
  211. bgtz I, .L56
  212. ST t1, -1 * SIZE(X)
  213. j $31
  214. NOP
  215. .align 3
  216. .L60:
  217. dsra I, N, 3
  218. move XX, X
  219. blez I, .L65
  220. daddiu I, I, -1
  221. LD a1, 0 * SIZE(X)
  222. daddu X, X, INCX
  223. LD a2, 0 * SIZE(X)
  224. daddu X, X, INCX
  225. LD a3, 0 * SIZE(X)
  226. daddu X, X, INCX
  227. LD a4, 0 * SIZE(X)
  228. daddu X, X, INCX
  229. LD a5, 0 * SIZE(X)
  230. daddu X, X, INCX
  231. LD a6, 0 * SIZE(X)
  232. daddu X, X, INCX
  233. LD a7, 0 * SIZE(X)
  234. daddu X, X, INCX
  235. LD a8, 0 * SIZE(X)
  236. daddu X, X, INCX
  237. blez I, .L63
  238. NOP
  239. .align 3
  240. .L62:
  241. MUL t1, ALPHA, a1
  242. LD a1, 0 * SIZE(X)
  243. daddu X, X, INCX
  244. MUL t2, ALPHA, a2
  245. LD a2, 0 * SIZE(X)
  246. daddu X, X, INCX
  247. MUL t3, ALPHA, a3
  248. LD a3, 0 * SIZE(X)
  249. daddu X, X, INCX
  250. MUL t4, ALPHA, a4
  251. LD a4, 0 * SIZE(X)
  252. daddu X, X, INCX
  253. ST t1, 0 * SIZE(XX)
  254. daddu XX, XX, INCX
  255. ST t2, 0 * SIZE(XX)
  256. daddu XX, XX, INCX
  257. ST t3, 0 * SIZE(XX)
  258. daddu XX, XX, INCX
  259. ST t4, 0 * SIZE(XX)
  260. daddu XX, XX, INCX
  261. MUL t1, ALPHA, a5
  262. LD a5, 0 * SIZE(X)
  263. daddu X, X, INCX
  264. MUL t2, ALPHA, a6
  265. LD a6, 0 * SIZE(X)
  266. daddu X, X, INCX
  267. MUL t3, ALPHA, a7
  268. LD a7, 0 * SIZE(X)
  269. daddu X, X, INCX
  270. MUL t4, ALPHA, a8
  271. LD a8, 0 * SIZE(X)
  272. daddu X, X, INCX
  273. ST t1, 0 * SIZE(XX)
  274. daddu XX, XX, INCX
  275. ST t2, 0 * SIZE(XX)
  276. daddu XX, XX, INCX
  277. ST t3, 0 * SIZE(XX)
  278. daddu XX, XX, INCX
  279. ST t4, 0 * SIZE(XX)
  280. daddiu I, I, -1
  281. bgtz I, .L62
  282. daddu XX, XX, INCX
  283. .align 3
  284. .L63:
  285. MUL t1, ALPHA, a1
  286. MUL t2, ALPHA, a2
  287. MUL t3, ALPHA, a3
  288. MUL t4, ALPHA, a4
  289. ST t1, 0 * SIZE(XX)
  290. daddu XX, XX, INCX
  291. ST t2, 0 * SIZE(XX)
  292. daddu XX, XX, INCX
  293. ST t3, 0 * SIZE(XX)
  294. daddu XX, XX, INCX
  295. ST t4, 0 * SIZE(XX)
  296. daddu XX, XX, INCX
  297. MUL t1, ALPHA, a5
  298. MUL t2, ALPHA, a6
  299. MUL t3, ALPHA, a7
  300. MUL t4, ALPHA, a8
  301. ST t1, 0 * SIZE(XX)
  302. daddu XX, XX, INCX
  303. ST t2, 0 * SIZE(XX)
  304. daddu XX, XX, INCX
  305. ST t3, 0 * SIZE(XX)
  306. daddu XX, XX, INCX
  307. ST t4, 0 * SIZE(XX)
  308. daddu XX, XX, INCX
  309. .align 3
  310. .L65:
  311. andi I, N, 7
  312. blez I, .L999
  313. NOP
  314. .align 3
  315. .L66:
  316. LD a1, 0 * SIZE(X)
  317. MUL t1, ALPHA, a1
  318. daddiu I, I, -1
  319. ST t1, 0 * SIZE(X)
  320. bgtz I, .L66
  321. daddu X, X, INCX
  322. .align 3
  323. .L999:
  324. j $31
  325. NOP
  326. EPILOGUE