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.

zdot.S 7.9 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402
  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 I $2
  46. #define TEMP $3
  47. #define a1 $f4
  48. #define a2 $f5
  49. #define a3 $f6
  50. #define a4 $f7
  51. #define b1 $f8
  52. #define b2 $f9
  53. #define b3 $f10
  54. #define b4 $f11
  55. #define s1 $f0
  56. #define s2 $f1
  57. #define s3 $f2
  58. #define s4 $f3
  59. PROLOGUE
  60. #ifdef F_INTERFACE
  61. LDINT N, 0(N)
  62. LDINT INCX, 0(INCX)
  63. LDINT INCY, 0(INCY)
  64. #endif
  65. MTC $0, s1
  66. MOV s2, s1
  67. MOV s3, s2
  68. MOV s4, s3
  69. dsll INCX, INCX, ZBASE_SHIFT
  70. li TEMP, 2 * SIZE
  71. blez N, .L999
  72. dsll INCY, INCY, ZBASE_SHIFT
  73. bne INCX, TEMP, .L20
  74. dsra I, N, 2
  75. bne INCY, TEMP, .L20
  76. NOP
  77. blez I, .L15
  78. NOP
  79. LD a1, 0 * SIZE(X)
  80. LD a2, 1 * SIZE(X)
  81. LD b1, 0 * SIZE(Y)
  82. daddiu I, I, -1
  83. blez I, .L14
  84. LD b2, 1 * SIZE(Y)
  85. .align 3
  86. .L13:
  87. MADD s1, s1, a1, b1
  88. LD a3, 2 * SIZE(X)
  89. MADD s2, s2, a2, b1
  90. LD a4, 3 * SIZE(X)
  91. MADD s3, s3, a1, b2
  92. LD b3, 2 * SIZE(Y)
  93. MADD s4, s4, a2, b2
  94. LD b4, 3 * SIZE(Y)
  95. MADD s1, s1, a3, b3
  96. LD a1, 4 * SIZE(X)
  97. MADD s2, s2, a4, b3
  98. LD a2, 5 * SIZE(X)
  99. MADD s3, s3, a3, b4
  100. LD b1, 4 * SIZE(Y)
  101. MADD s4, s4, a4, b4
  102. LD b2, 5 * SIZE(Y)
  103. MADD s1, s1, a1, b1
  104. LD a3, 6 * SIZE(X)
  105. MADD s2, s2, a2, b1
  106. LD a4, 7 * SIZE(X)
  107. MADD s3, s3, a1, b2
  108. LD b3, 6 * SIZE(Y)
  109. MADD s4, s4, a2, b2
  110. LD b4, 7 * SIZE(Y)
  111. MADD s1, s1, a3, b3
  112. LD a1, 8 * SIZE(X)
  113. MADD s2, s2, a4, b3
  114. LD a2, 9 * SIZE(X)
  115. MADD s3, s3, a3, b4
  116. LD b1, 8 * SIZE(Y)
  117. MADD s4, s4, a4, b4
  118. LD b2, 9 * SIZE(Y)
  119. daddiu I, I, -1
  120. daddiu X, X, 8 * SIZE
  121. bgtz I, .L13
  122. daddiu Y, Y, 8 * SIZE
  123. .align 3
  124. .L14:
  125. MADD s1, s1, a1, b1
  126. LD a3, 2 * SIZE(X)
  127. MADD s2, s2, a2, b1
  128. LD a4, 3 * SIZE(X)
  129. MADD s3, s3, a1, b2
  130. LD b3, 2 * SIZE(Y)
  131. MADD s4, s4, a2, b2
  132. LD b4, 3 * SIZE(Y)
  133. MADD s1, s1, a3, b3
  134. LD a1, 4 * SIZE(X)
  135. MADD s2, s2, a4, b3
  136. LD a2, 5 * SIZE(X)
  137. MADD s3, s3, a3, b4
  138. LD b1, 4 * SIZE(Y)
  139. MADD s4, s4, a4, b4
  140. LD b2, 5 * SIZE(Y)
  141. MADD s1, s1, a1, b1
  142. LD a3, 6 * SIZE(X)
  143. MADD s2, s2, a2, b1
  144. LD a4, 7 * SIZE(X)
  145. MADD s3, s3, a1, b2
  146. LD b3, 6 * SIZE(Y)
  147. MADD s4, s4, a2, b2
  148. LD b4, 7 * SIZE(Y)
  149. MADD s1, s1, a3, b3
  150. daddiu X, X, 8 * SIZE
  151. MADD s2, s2, a4, b3
  152. daddiu Y, Y, 8 * SIZE
  153. MADD s3, s3, a3, b4
  154. MADD s4, s4, a4, b4
  155. .align 3
  156. .L15:
  157. andi I, N, 3
  158. blez I, .L999
  159. NOP
  160. LD a1, 0 * SIZE(X)
  161. LD a2, 1 * SIZE(X)
  162. LD b1, 0 * SIZE(Y)
  163. daddiu I, I, -1
  164. blez I, .L17
  165. LD b2, 1 * SIZE(Y)
  166. .align 3
  167. .L16:
  168. MADD s1, s1, a1, b1
  169. daddiu I, I, -1
  170. MADD s2, s2, a2, b1
  171. LD b1, 2 * SIZE(Y)
  172. MADD s3, s3, a1, b2
  173. LD a1, 2 * SIZE(X)
  174. MADD s4, s4, a2, b2
  175. LD a2, 3 * SIZE(X)
  176. LD b2, 3 * SIZE(Y)
  177. daddiu X, X, 2 * SIZE
  178. bgtz I, .L16
  179. daddiu Y, Y, 2 * SIZE
  180. .align 3
  181. .L17:
  182. MADD s1, s1, a1, b1
  183. MADD s2, s2, a2, b1
  184. NOP
  185. MADD s3, s3, a1, b2
  186. j .L999
  187. MADD s4, s4, a2, b2
  188. .align 3
  189. .L20:
  190. #ifdef F_INTERFACE
  191. bgez INCX, .L21
  192. daddiu TEMP, N, -1
  193. mult TEMP, INCX
  194. mflo TEMP
  195. dsub X, X, TEMP
  196. .align 3
  197. .L21:
  198. bgez INCY, .L22
  199. daddiu TEMP, N, -1
  200. mult TEMP, INCY
  201. mflo TEMP
  202. dsub Y, Y, TEMP
  203. .align 3
  204. .L22:
  205. #endif
  206. blez I, .L25
  207. NOP
  208. LD a1, 0 * SIZE(X)
  209. LD a2, 1 * SIZE(X)
  210. LD b1, 0 * SIZE(Y)
  211. LD b2, 1 * SIZE(Y)
  212. dadd X, X, INCX
  213. daddiu I, I, -1
  214. blez I, .L24
  215. dadd Y, Y, INCY
  216. .align 3
  217. .L23:
  218. MADD s1, s1, a1, b1
  219. LD a3, 0 * SIZE(X)
  220. MADD s2, s2, a2, b1
  221. LD a4, 1 * SIZE(X)
  222. MADD s3, s3, a1, b2
  223. LD b3, 0 * SIZE(Y)
  224. MADD s4, s4, a2, b2
  225. LD b4, 1 * SIZE(Y)
  226. dadd X, X, INCX
  227. dadd Y, Y, INCY
  228. MADD s1, s1, a3, b3
  229. LD a1, 0 * SIZE(X)
  230. MADD s2, s2, a4, b3
  231. LD a2, 1 * SIZE(X)
  232. MADD s3, s3, a3, b4
  233. LD b1, 0 * SIZE(Y)
  234. MADD s4, s4, a4, b4
  235. LD b2, 1 * SIZE(Y)
  236. dadd X, X, INCX
  237. dadd Y, Y, INCY
  238. MADD s1, s1, a1, b1
  239. LD a3, 0 * SIZE(X)
  240. MADD s2, s2, a2, b1
  241. LD a4, 1 * SIZE(X)
  242. MADD s3, s3, a1, b2
  243. LD b3, 0 * SIZE(Y)
  244. MADD s4, s4, a2, b2
  245. LD b4, 1 * SIZE(Y)
  246. dadd X, X, INCX
  247. dadd Y, Y, INCY
  248. MADD s1, s1, a3, b3
  249. LD a1, 0 * SIZE(X)
  250. MADD s2, s2, a4, b3
  251. LD a2, 1 * SIZE(X)
  252. MADD s3, s3, a3, b4
  253. LD b1, 0 * SIZE(Y)
  254. MADD s4, s4, a4, b4
  255. LD b2, 1 * SIZE(Y)
  256. dadd X, X, INCX
  257. daddiu I, I, -1
  258. bgtz I, .L23
  259. dadd Y, Y, INCY
  260. .align 3
  261. .L24:
  262. MADD s1, s1, a1, b1
  263. LD a3, 0 * SIZE(X)
  264. MADD s2, s2, a2, b1
  265. LD a4, 1 * SIZE(X)
  266. MADD s3, s3, a1, b2
  267. LD b3, 0 * SIZE(Y)
  268. MADD s4, s4, a2, b2
  269. LD b4, 1 * SIZE(Y)
  270. dadd X, X, INCX
  271. dadd Y, Y, INCY
  272. MADD s1, s1, a3, b3
  273. LD a1, 0 * SIZE(X)
  274. MADD s2, s2, a4, b3
  275. LD a2, 1 * SIZE(X)
  276. MADD s3, s3, a3, b4
  277. LD b1, 0 * SIZE(Y)
  278. MADD s4, s4, a4, b4
  279. LD b2, 1 * SIZE(Y)
  280. dadd X, X, INCX
  281. dadd Y, Y, INCY
  282. MADD s1, s1, a1, b1
  283. LD a3, 0 * SIZE(X)
  284. MADD s2, s2, a2, b1
  285. LD a4, 1 * SIZE(X)
  286. MADD s3, s3, a1, b2
  287. LD b3, 0 * SIZE(Y)
  288. MADD s4, s4, a2, b2
  289. LD b4, 1 * SIZE(Y)
  290. MADD s1, s1, a3, b3
  291. dadd X, X, INCX
  292. MADD s2, s2, a4, b3
  293. dadd Y, Y, INCY
  294. MADD s3, s3, a3, b4
  295. MADD s4, s4, a4, b4
  296. .align 3
  297. .L25:
  298. andi I, N, 3
  299. blez I, .L999
  300. NOP
  301. .align 3
  302. .L26:
  303. LD a1, 0 * SIZE(X)
  304. LD a2, 1 * SIZE(X)
  305. LD b1, 0 * SIZE(Y)
  306. LD b2, 1 * SIZE(Y)
  307. MADD s1, s1, a1, b1
  308. MADD s2, s2, a2, b1
  309. MADD s3, s3, a1, b2
  310. MADD s4, s4, a2, b2
  311. dadd X, X, INCX
  312. dadd Y, Y, INCY
  313. daddiu I, I, -1
  314. bgtz I, .L26
  315. NOP
  316. .align 3
  317. .L999:
  318. NOP
  319. #ifndef CONJ
  320. SUB s1, s1, s4
  321. #else
  322. ADD s1, s1, s4
  323. #endif
  324. j $31
  325. #ifndef CONJ
  326. ADD s3, s3, s2
  327. #else
  328. SUB s3, s3, s2
  329. #endif
  330. EPILOGUE