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.

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350
  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, ZBASE_SHIFT
  65. li TEMP, 2 * SIZE
  66. blez N, .L999
  67. dsll INCY, INCY, ZBASE_SHIFT
  68. bne INCX, TEMP, .L20
  69. dsra I, N, 1
  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, 1
  152. blez I, .L999
  153. NOP
  154. LD a1, 0 * SIZE(X)
  155. LD a2, 1 * SIZE(X)
  156. LD b1, 0 * SIZE(Y)
  157. LD b2, 1 * SIZE(Y)
  158. MUL t1, S, b1
  159. MUL t2, C, b1
  160. MUL t3, S, b2
  161. MUL t4, C, b2
  162. MADD t1, t1, C, a1
  163. NMSUB t2, t2, S, a1
  164. MADD t3, t3, C, a2
  165. NMSUB t4, t4, S, a2
  166. ST t1, 0 * SIZE(X)
  167. ST t2, 0 * SIZE(Y)
  168. ST t3, 1 * SIZE(X)
  169. j .L999
  170. ST t4, 1 * SIZE(Y)
  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. LD b1, 0 * SIZE(Y)
  179. LD a2, 1 * SIZE(X)
  180. dadd X, X, INCX
  181. LD b2, 1 * SIZE(Y)
  182. dadd Y, Y, INCY
  183. LD a3, 0 * SIZE(X)
  184. LD b3, 0 * SIZE(Y)
  185. LD a4, 1 * SIZE(X)
  186. dadd X, X, INCX
  187. MUL t1, S, b1
  188. LD b4, 1 * SIZE(Y)
  189. MUL t2, C, b1
  190. dadd Y, Y, INCY
  191. MUL t3, S, b2
  192. blez I, .L23
  193. MUL t4, C, b2
  194. .align 3
  195. .L22:
  196. MADD t1, t1, C, a1
  197. LD b1, 0 * SIZE(Y)
  198. NMSUB t2, t2, S, a1
  199. LD a1, 0 * SIZE(X)
  200. MADD t3, t3, C, a2
  201. LD b2, 1 * SIZE(Y)
  202. dadd Y, Y, INCY
  203. NMSUB t4, t4, S, a2
  204. LD a2, 1 * SIZE(X)
  205. dadd X, X, INCX
  206. ST t1, 0 * SIZE(XX)
  207. MUL t1, S, b3
  208. ST t2, 0 * SIZE(YY)
  209. MUL t2, C, b3
  210. ST t3, 1 * SIZE(XX)
  211. dadd XX, XX, INCX
  212. MUL t3, S, b4
  213. ST t4, 1 * SIZE(YY)
  214. dadd YY, YY, INCY
  215. MUL t4, C, b4
  216. MADD t1, t1, C, a3
  217. LD b3, 0 * SIZE(Y)
  218. NMSUB t2, t2, S, a3
  219. LD a3, 0 * SIZE(X)
  220. MADD t3, t3, C, a4
  221. LD b4, 1 * SIZE(Y)
  222. dadd Y, Y, INCY
  223. NMSUB t4, t4, S, a4
  224. LD a4, 1 * SIZE(X)
  225. dadd X, X, INCX
  226. ST t1, 0 * SIZE(XX)
  227. MUL t1, S, b1
  228. ST t2, 0 * SIZE(YY)
  229. MUL t2, C, b1
  230. ST t3, 1 * SIZE(XX)
  231. dadd XX, XX, INCX
  232. MUL t3, S, b2
  233. ST t4, 1 * SIZE(YY)
  234. MUL t4, C, b2
  235. daddiu I, I, -1
  236. bgtz I, .L22
  237. dadd YY, YY, INCY
  238. .align 3
  239. .L23:
  240. MADD t1, t1, C, a1
  241. NMSUB t2, t2, S, a1
  242. MADD t3, t3, C, a2
  243. NMSUB t4, t4, S, a2
  244. ST t1, 0 * SIZE(XX)
  245. MUL t1, S, b3
  246. ST t2, 0 * SIZE(YY)
  247. MUL t2, C, b3
  248. ST t3, 1 * SIZE(XX)
  249. dadd XX, XX, INCX
  250. MUL t3, S, b4
  251. ST t4, 1 * SIZE(YY)
  252. dadd YY, YY, INCY
  253. MUL t4, C, b4
  254. MADD t1, t1, C, a3
  255. NMSUB t2, t2, S, a3
  256. MADD t3, t3, C, a4
  257. NMSUB t4, t4, S, a4
  258. ST t1, 0 * SIZE(XX)
  259. ST t2, 0 * SIZE(YY)
  260. ST t3, 1 * SIZE(XX)
  261. dadd XX, XX, INCX
  262. ST t4, 1 * SIZE(YY)
  263. dadd YY, YY, INCY
  264. .align 3
  265. .L25:
  266. andi I, N, 1
  267. blez I, .L999
  268. NOP
  269. .align 3
  270. .L26:
  271. LD a1, 0 * SIZE(X)
  272. LD a2, 1 * SIZE(X)
  273. LD b1, 0 * SIZE(Y)
  274. LD b2, 1 * SIZE(Y)
  275. MUL t1, S, b1
  276. MUL t2, C, b1
  277. MUL t3, S, b2
  278. MUL t4, C, b2
  279. MADD t1, t1, C, a1
  280. NMSUB t2, t2, S, a1
  281. MADD t3, t3, C, a2
  282. NMSUB t4, t4, S, a2
  283. ST t1, 0 * SIZE(X)
  284. ST t2, 0 * SIZE(Y)
  285. ST t3, 1 * SIZE(X)
  286. ST t4, 1 * SIZE(Y)
  287. .align 3
  288. .L999:
  289. j $31
  290. NOP
  291. EPILOGUE