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.

swap.S 8.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392
  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 Y $10
  44. #define INCY $11
  45. #define I $2
  46. #define TEMP $3
  47. #define XX $5
  48. #define YY $6
  49. #define a1 $f0
  50. #define a2 $f1
  51. #define a3 $f2
  52. #define a4 $f3
  53. #define a5 $f4
  54. #define a6 $f5
  55. #define a7 $f6
  56. #define a8 $f7
  57. #define b1 $f8
  58. #define b2 $f9
  59. #define b3 $f10
  60. #define b4 $f11
  61. #define b5 $f12
  62. #define b6 $f13
  63. #define b7 $f14
  64. #define b8 $f15
  65. PROLOGUE
  66. li TEMP, SIZE
  67. NOP
  68. blez N, .L999
  69. dsll INCX, INCX, BASE_SHIFT
  70. bne INCX, TEMP, .L20
  71. dsll INCY, INCY, BASE_SHIFT
  72. bne INCY, TEMP, .L20
  73. dsra I, N, 3
  74. blez I, .L15
  75. daddiu I, I, -1
  76. LD a1, 0 * SIZE(X)
  77. LD b1, 0 * SIZE(Y)
  78. LD a2, 1 * SIZE(X)
  79. LD b2, 1 * SIZE(Y)
  80. LD a3, 2 * SIZE(X)
  81. LD b3, 2 * SIZE(Y)
  82. LD a4, 3 * SIZE(X)
  83. LD b4, 3 * SIZE(Y)
  84. LD a5, 4 * SIZE(X)
  85. LD b5, 4 * SIZE(Y)
  86. LD a6, 5 * SIZE(X)
  87. LD b6, 5 * SIZE(Y)
  88. LD a7, 6 * SIZE(X)
  89. LD b7, 6 * SIZE(Y)
  90. LD a8, 7 * SIZE(X)
  91. LD b8, 7 * SIZE(Y)
  92. blez I, .L13
  93. NOP
  94. .align 3
  95. .L12:
  96. ST a1, 0 * SIZE(Y)
  97. LD a1, 8 * SIZE(X)
  98. ST b1, 0 * SIZE(X)
  99. LD b1, 8 * SIZE(Y)
  100. ST a2, 1 * SIZE(Y)
  101. LD a2, 9 * SIZE(X)
  102. ST b2, 1 * SIZE(X)
  103. LD b2, 9 * SIZE(Y)
  104. ST a3, 2 * SIZE(Y)
  105. LD a3, 10 * SIZE(X)
  106. ST b3, 2 * SIZE(X)
  107. LD b3, 10 * SIZE(Y)
  108. ST a4, 3 * SIZE(Y)
  109. LD a4, 11 * SIZE(X)
  110. ST b4, 3 * SIZE(X)
  111. LD b4, 11 * SIZE(Y)
  112. ST a5, 4 * SIZE(Y)
  113. LD a5, 12 * SIZE(X)
  114. ST b5, 4 * SIZE(X)
  115. LD b5, 12 * SIZE(Y)
  116. ST a6, 5 * SIZE(Y)
  117. LD a6, 13 * SIZE(X)
  118. ST b6, 5 * SIZE(X)
  119. LD b6, 13 * SIZE(Y)
  120. ST a7, 6 * SIZE(Y)
  121. LD a7, 14 * SIZE(X)
  122. ST b7, 6 * SIZE(X)
  123. LD b7, 14 * SIZE(Y)
  124. ST a8, 7 * SIZE(Y)
  125. LD a8, 15 * SIZE(X)
  126. ST b8, 7 * SIZE(X)
  127. LD b8, 15 * SIZE(Y)
  128. daddiu I, I, -1
  129. daddiu X, X, 8 * SIZE
  130. bgtz I, .L12
  131. daddiu Y, Y, 8 * SIZE
  132. .align 3
  133. .L13:
  134. ST a1, 0 * SIZE(Y)
  135. ST b1, 0 * SIZE(X)
  136. ST a2, 1 * SIZE(Y)
  137. ST b2, 1 * SIZE(X)
  138. ST a3, 2 * SIZE(Y)
  139. ST b3, 2 * SIZE(X)
  140. ST a4, 3 * SIZE(Y)
  141. ST b4, 3 * SIZE(X)
  142. ST a5, 4 * SIZE(Y)
  143. ST b5, 4 * SIZE(X)
  144. ST a6, 5 * SIZE(Y)
  145. ST b6, 5 * SIZE(X)
  146. ST a7, 6 * SIZE(Y)
  147. ST b7, 6 * SIZE(X)
  148. ST a8, 7 * SIZE(Y)
  149. ST b8, 7 * SIZE(X)
  150. daddiu X, X, 8 * SIZE
  151. daddiu Y, Y, 8 * SIZE
  152. .align 3
  153. .L15:
  154. andi I, N, 7
  155. blez I, .L999
  156. NOP
  157. .align 3
  158. .L16:
  159. LD a1, 0 * SIZE(X)
  160. LD b1, 0 * SIZE(Y)
  161. daddiu X, X, SIZE
  162. daddiu I, I, -1
  163. daddiu Y, Y, SIZE
  164. ST b1, -1 * SIZE(X)
  165. bgtz I, .L16
  166. ST a1, -1 * SIZE(Y)
  167. j .L999
  168. NOP
  169. .align 3
  170. .L20:
  171. dsra I, N, 3
  172. move XX, X
  173. move YY, Y
  174. blez I, .L25
  175. daddiu I, I, -1
  176. LD a1, 0 * SIZE(X)
  177. daddu X, X, INCX
  178. LD b1, 0 * SIZE(Y)
  179. daddu Y, Y, INCY
  180. LD a2, 0 * SIZE(X)
  181. daddu X, X, INCX
  182. LD b2, 0 * SIZE(Y)
  183. daddu Y, Y, INCY
  184. LD a3, 0 * SIZE(X)
  185. daddu X, X, INCX
  186. LD b3, 0 * SIZE(Y)
  187. daddu Y, Y, INCY
  188. LD a4, 0 * SIZE(X)
  189. daddu X, X, INCX
  190. LD b4, 0 * SIZE(Y)
  191. daddu Y, Y, INCY
  192. LD a5, 0 * SIZE(X)
  193. daddu X, X, INCX
  194. LD b5, 0 * SIZE(Y)
  195. daddu Y, Y, INCY
  196. LD a6, 0 * SIZE(X)
  197. daddu X, X, INCX
  198. LD b6, 0 * SIZE(Y)
  199. daddu Y, Y, INCY
  200. LD a7, 0 * SIZE(X)
  201. daddu X, X, INCX
  202. LD b7, 0 * SIZE(Y)
  203. daddu Y, Y, INCY
  204. LD a8, 0 * SIZE(X)
  205. daddu X, X, INCX
  206. LD b8, 0 * SIZE(Y)
  207. daddu Y, Y, INCY
  208. blez I, .L23
  209. NOP
  210. .align 3
  211. .L22:
  212. ST a1, 0 * SIZE(YY)
  213. daddu YY, YY, INCY
  214. LD a1, 0 * SIZE(X)
  215. daddu X, X, INCX
  216. ST b1, 0 * SIZE(XX)
  217. daddu XX, XX, INCX
  218. LD b1, 0 * SIZE(Y)
  219. daddu Y, Y, INCY
  220. ST a2, 0 * SIZE(YY)
  221. daddu YY, YY, INCY
  222. LD a2, 0 * SIZE(X)
  223. daddu X, X, INCX
  224. ST b2, 0 * SIZE(XX)
  225. daddu XX, XX, INCX
  226. LD b2, 0 * SIZE(Y)
  227. daddu Y, Y, INCY
  228. ST a3, 0 * SIZE(YY)
  229. daddu YY, YY, INCY
  230. LD a3, 0 * SIZE(X)
  231. daddu X, X, INCX
  232. ST b3, 0 * SIZE(XX)
  233. daddu XX, XX, INCX
  234. LD b3, 0 * SIZE(Y)
  235. daddu Y, Y, INCY
  236. ST a4, 0 * SIZE(YY)
  237. daddu YY, YY, INCY
  238. LD a4, 0 * SIZE(X)
  239. daddu X, X, INCX
  240. ST b4, 0 * SIZE(XX)
  241. daddu XX, XX, INCX
  242. LD b4, 0 * SIZE(Y)
  243. daddu Y, Y, INCY
  244. ST a5, 0 * SIZE(YY)
  245. daddu YY, YY, INCY
  246. LD a5, 0 * SIZE(X)
  247. daddu X, X, INCX
  248. ST b5, 0 * SIZE(XX)
  249. daddu XX, XX, INCX
  250. LD b5, 0 * SIZE(Y)
  251. daddu Y, Y, INCY
  252. ST a6, 0 * SIZE(YY)
  253. daddu YY, YY, INCY
  254. LD a6, 0 * SIZE(X)
  255. daddu X, X, INCX
  256. ST b6, 0 * SIZE(XX)
  257. daddu XX, XX, INCX
  258. LD b6, 0 * SIZE(Y)
  259. daddu Y, Y, INCY
  260. ST a7, 0 * SIZE(YY)
  261. daddu YY, YY, INCY
  262. LD a7, 0 * SIZE(X)
  263. daddu X, X, INCX
  264. ST b7, 0 * SIZE(XX)
  265. daddu XX, XX, INCX
  266. LD b7, 0 * SIZE(Y)
  267. daddu Y, Y, INCY
  268. ST a8, 0 * SIZE(YY)
  269. daddu YY, YY, INCY
  270. LD a8, 0 * SIZE(X)
  271. daddu X, X, INCX
  272. ST b8, 0 * SIZE(XX)
  273. daddu XX, XX, INCX
  274. LD b8, 0 * SIZE(Y)
  275. daddiu I, I, -1
  276. bgtz I, .L22
  277. daddu Y, Y, INCY
  278. .align 3
  279. .L23:
  280. ST a1, 0 * SIZE(YY)
  281. daddu YY, YY, INCY
  282. ST b1, 0 * SIZE(XX)
  283. daddu XX, XX, INCX
  284. ST a2, 0 * SIZE(YY)
  285. daddu YY, YY, INCY
  286. ST b2, 0 * SIZE(XX)
  287. daddu XX, XX, INCX
  288. ST a3, 0 * SIZE(YY)
  289. daddu YY, YY, INCY
  290. ST b3, 0 * SIZE(XX)
  291. daddu XX, XX, INCX
  292. ST a4, 0 * SIZE(YY)
  293. daddu YY, YY, INCY
  294. ST b4, 0 * SIZE(XX)
  295. daddu XX, XX, INCX
  296. ST a5, 0 * SIZE(YY)
  297. daddu YY, YY, INCY
  298. ST b5, 0 * SIZE(XX)
  299. daddu XX, XX, INCX
  300. ST a6, 0 * SIZE(YY)
  301. daddu YY, YY, INCY
  302. ST b6, 0 * SIZE(XX)
  303. daddu XX, XX, INCX
  304. ST a7, 0 * SIZE(YY)
  305. daddu YY, YY, INCY
  306. ST b7, 0 * SIZE(XX)
  307. daddu XX, XX, INCX
  308. ST a8, 0 * SIZE(YY)
  309. daddu YY, YY, INCY
  310. ST b8, 0 * SIZE(XX)
  311. daddu XX, XX, INCX
  312. .align 3
  313. .L25:
  314. andi I, N, 7
  315. blez I, .L999
  316. NOP
  317. .align 3
  318. .L26:
  319. LD a1, 0 * SIZE(X)
  320. LD b1, 0 * SIZE(Y)
  321. daddiu I, I, -1
  322. ST a1, 0 * SIZE(Y)
  323. ST b1, 0 * SIZE(X)
  324. daddu X, X, INCX
  325. bgtz I, .L26
  326. daddu Y, Y, INCY
  327. .align 3
  328. .L999:
  329. j $31
  330. NOP
  331. EPILOGUE