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.

rot.S 7.3 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367
  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 $5
  42. #define INCX $6
  43. #define Y $7
  44. #define INCY $8
  45. #define XX $9
  46. #define YY $10
  47. #define C $f17
  48. #define S $f18
  49. #define I $2
  50. #define TEMP $3
  51. #define a1 $f4
  52. #define a2 $f5
  53. #define a3 $f6
  54. #define a4 $f7
  55. #define b1 $f8
  56. #define b2 $f9
  57. #define b3 $f10
  58. #define b4 $f11
  59. #define t1 $f0
  60. #define t2 $f1
  61. #define t3 $f2
  62. #define t4 $f3
  63. PROLOGUE
  64. dsll INCX, INCX, BASE_SHIFT
  65. li TEMP, SIZE
  66. blez N, .L999
  67. dsll INCY, INCY, BASE_SHIFT
  68. bne INCX, TEMP, .L20
  69. dsra I, N, 2
  70. bne INCY, TEMP, .L20
  71. NOP
  72. blez I, .L15
  73. daddiu I, I, -1
  74. LD a1, 0 * SIZE(X)
  75. LD b1, 0 * SIZE(Y)
  76. LD a2, 1 * SIZE(X)
  77. LD b2, 1 * SIZE(Y)
  78. LD a3, 2 * SIZE(X)
  79. LD b3, 2 * SIZE(Y)
  80. MUL t1, S, b1
  81. LD a4, 3 * SIZE(X)
  82. MUL t2, C, b1
  83. LD b4, 3 * SIZE(Y)
  84. MUL t3, S, b2
  85. blez I, .L13
  86. MUL t4, C, b2
  87. .align 3
  88. .L12:
  89. MADD t1, t1, C, a1
  90. LD b1, 4 * SIZE(Y)
  91. NMSUB t2, t2, S, a1
  92. LD a1, 4 * SIZE(X)
  93. MADD t3, t3, C, a2
  94. LD b2, 5 * SIZE(Y)
  95. NMSUB t4, t4, S, a2
  96. LD a2, 5 * SIZE(X)
  97. ST t1, 0 * SIZE(X)
  98. MUL t1, S, b3
  99. ST t2, 0 * SIZE(Y)
  100. MUL t2, C, b3
  101. ST t3, 1 * SIZE(X)
  102. MUL t3, S, b4
  103. ST t4, 1 * SIZE(Y)
  104. MUL t4, C, b4
  105. MADD t1, t1, C, a3
  106. LD b3, 6 * SIZE(Y)
  107. NMSUB t2, t2, S, a3
  108. LD a3, 6 * SIZE(X)
  109. MADD t3, t3, C, a4
  110. LD b4, 7 * SIZE(Y)
  111. NMSUB t4, t4, S, a4
  112. LD a4, 7 * SIZE(X)
  113. ST t1, 2 * SIZE(X)
  114. MUL t1, S, b1
  115. ST t2, 2 * SIZE(Y)
  116. MUL t2, C, b1
  117. ST t3, 3 * SIZE(X)
  118. MUL t3, S, b2
  119. ST t4, 3 * SIZE(Y)
  120. MUL t4, C, b2
  121. daddiu I, I, -1
  122. daddiu X, X, 4 * SIZE
  123. bgtz I, .L12
  124. daddiu Y, Y, 4 * SIZE
  125. .align 3
  126. .L13:
  127. MADD t1, t1, C, a1
  128. NMSUB t2, t2, S, a1
  129. MADD t3, t3, C, a2
  130. NMSUB t4, t4, S, a2
  131. ST t1, 0 * SIZE(X)
  132. MUL t1, S, b3
  133. ST t2, 0 * SIZE(Y)
  134. MUL t2, C, b3
  135. ST t3, 1 * SIZE(X)
  136. MUL t3, S, b4
  137. ST t4, 1 * SIZE(Y)
  138. MUL t4, C, b4
  139. MADD t1, t1, C, a3
  140. NMSUB t2, t2, S, a3
  141. MADD t3, t3, C, a4
  142. daddiu X, X, 4 * SIZE
  143. NMSUB t4, t4, S, a4
  144. daddiu Y, Y, 4 * SIZE
  145. ST t1, -2 * SIZE(X)
  146. ST t2, -2 * SIZE(Y)
  147. ST t3, -1 * SIZE(X)
  148. ST t4, -1 * SIZE(Y)
  149. .align 3
  150. .L15:
  151. andi I, N, 3
  152. blez I, .L999
  153. NOP
  154. .align 3
  155. .L16:
  156. LD a1, 0 * SIZE(X)
  157. LD b1, 0 * SIZE(Y)
  158. MUL t1, S, b1
  159. MUL t2, C, b1
  160. MADD t1, t1, C, a1
  161. NMSUB t2, t2, S, a1
  162. ST t1, 0 * SIZE(X)
  163. ST t2, 0 * SIZE(Y)
  164. daddiu I, I, -1
  165. daddiu X, X, SIZE
  166. daddiu Y, Y, SIZE
  167. bgtz I, .L16
  168. NOP
  169. j .L999
  170. NOP
  171. .align 3
  172. .L20:
  173. move XX, X
  174. move YY, Y
  175. blez I, .L25
  176. daddiu I, I, -1
  177. LD a1, 0 * SIZE(X)
  178. dadd X, X, INCX
  179. LD b1, 0 * SIZE(Y)
  180. dadd Y, Y, INCY
  181. LD a2, 0 * SIZE(X)
  182. dadd X, X, INCX
  183. LD b2, 0 * SIZE(Y)
  184. dadd Y, Y, INCY
  185. LD a3, 0 * SIZE(X)
  186. dadd X, X, INCX
  187. LD b3, 0 * SIZE(Y)
  188. dadd Y, Y, INCY
  189. MUL t1, S, b1
  190. LD a4, 0 * SIZE(X)
  191. dadd X, X, INCX
  192. MUL t2, C, b1
  193. LD b4, 0 * SIZE(Y)
  194. dadd Y, Y, INCY
  195. MUL t3, S, b2
  196. blez I, .L23
  197. MUL t4, C, b2
  198. .align 3
  199. .L22:
  200. MADD t1, t1, C, a1
  201. LD b1, 0 * SIZE(Y)
  202. dadd Y, Y, INCY
  203. NMSUB t2, t2, S, a1
  204. LD a1, 0 * SIZE(X)
  205. dadd X, X, INCX
  206. MADD t3, t3, C, a2
  207. LD b2, 0 * SIZE(Y)
  208. dadd Y, Y, INCY
  209. NMSUB t4, t4, S, a2
  210. LD a2, 0 * SIZE(X)
  211. dadd X, X, INCX
  212. ST t1, 0 * SIZE(XX)
  213. dadd XX, XX, INCX
  214. MUL t1, S, b3
  215. ST t2, 0 * SIZE(YY)
  216. dadd YY, YY, INCY
  217. MUL t2, C, b3
  218. ST t3, 0 * SIZE(XX)
  219. dadd XX, XX, INCX
  220. MUL t3, S, b4
  221. ST t4, 0 * SIZE(YY)
  222. dadd YY, YY, INCY
  223. MUL t4, C, b4
  224. MADD t1, t1, C, a3
  225. LD b3, 0 * SIZE(Y)
  226. dadd Y, Y, INCY
  227. NMSUB t2, t2, S, a3
  228. LD a3, 0 * SIZE(X)
  229. dadd X, X, INCX
  230. MADD t3, t3, C, a4
  231. LD b4, 0 * SIZE(Y)
  232. dadd Y, Y, INCY
  233. NMSUB t4, t4, S, a4
  234. LD a4, 0 * SIZE(X)
  235. dadd X, X, INCX
  236. ST t1, 0 * SIZE(XX)
  237. dadd XX, XX, INCX
  238. MUL t1, S, b1
  239. ST t2, 0 * SIZE(YY)
  240. dadd YY, YY, INCY
  241. MUL t2, C, b1
  242. ST t3, 0 * SIZE(XX)
  243. dadd XX, XX, INCX
  244. MUL t3, S, b2
  245. ST t4, 0 * SIZE(YY)
  246. MUL t4, C, b2
  247. daddiu I, I, -1
  248. bgtz I, .L22
  249. dadd YY, YY, INCY
  250. .align 3
  251. .L23:
  252. MADD t1, t1, C, a1
  253. NMSUB t2, t2, S, a1
  254. MADD t3, t3, C, a2
  255. NMSUB t4, t4, S, a2
  256. ST t1, 0 * SIZE(XX)
  257. dadd XX, XX, INCX
  258. MUL t1, S, b3
  259. ST t2, 0 * SIZE(YY)
  260. dadd YY, YY, INCY
  261. MUL t2, C, b3
  262. ST t3, 0 * SIZE(XX)
  263. dadd XX, XX, INCX
  264. MUL t3, S, b4
  265. ST t4, 0 * SIZE(YY)
  266. dadd YY, YY, INCY
  267. MUL t4, C, b4
  268. MADD t1, t1, C, a3
  269. NMSUB t2, t2, S, a3
  270. MADD t3, t3, C, a4
  271. NMSUB t4, t4, S, a4
  272. ST t1, 0 * SIZE(XX)
  273. dadd XX, XX, INCX
  274. ST t2, 0 * SIZE(YY)
  275. dadd YY, YY, INCY
  276. ST t3, 0 * SIZE(XX)
  277. dadd XX, XX, INCX
  278. ST t4, 0 * SIZE(YY)
  279. dadd YY, YY, INCY
  280. .align 3
  281. .L25:
  282. andi I, N, 3
  283. blez I, .L999
  284. NOP
  285. .align 3
  286. .L26:
  287. LD a1, 0 * SIZE(X)
  288. LD b1, 0 * SIZE(Y)
  289. MUL t1, S, b1
  290. MUL t2, C, b1
  291. MADD t1, t1, C, a1
  292. daddiu I, I, -1
  293. NMSUB t2, t2, S, a1
  294. ST t1, 0 * SIZE(X)
  295. ST t2, 0 * SIZE(Y)
  296. dadd X, X, INCX
  297. bgtz I, .L26
  298. dadd Y, Y, INCY
  299. .align 3
  300. .L999:
  301. j $31
  302. NOP
  303. EPILOGUE