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.

dot.S 6.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310
  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 $f2
  48. #define a2 $f3
  49. #define a3 $f4
  50. #define a4 $f5
  51. #define b1 $f6
  52. #define b2 $f7
  53. #define b3 $f8
  54. #define b4 $f9
  55. #define s1 $f0
  56. #define s2 $f1
  57. PROLOGUE
  58. #ifdef F_INTERFACE
  59. LDINT N, 0(N)
  60. LDINT INCX, 0(INCX)
  61. LDINT INCY, 0(INCY)
  62. #endif
  63. MTC $0, s1
  64. MTC $0, s2
  65. dsll INCX, INCX, BASE_SHIFT
  66. li TEMP, SIZE
  67. blez N, .L999
  68. dsll INCY, INCY, BASE_SHIFT
  69. bne INCX, TEMP, .L20
  70. dsra I, N, 3
  71. bne INCY, TEMP, .L20
  72. NOP
  73. blez I, .L15
  74. NOP
  75. LD a1, 0 * SIZE(X)
  76. LD b1, 0 * SIZE(Y)
  77. LD a2, 1 * SIZE(X)
  78. LD b2, 1 * SIZE(Y)
  79. LD a3, 2 * SIZE(X)
  80. LD b3, 2 * SIZE(Y)
  81. LD a4, 3 * SIZE(X)
  82. daddiu I, I, -1
  83. blez I, .L13
  84. LD b4, 3 * SIZE(Y)
  85. .align 3
  86. .L12:
  87. MADD s1, s1, a1, b1
  88. LD a1, 4 * SIZE(X)
  89. LD b1, 4 * SIZE(Y)
  90. MADD s2, s2, a2, b2
  91. LD a2, 5 * SIZE(X)
  92. LD b2, 5 * SIZE(Y)
  93. MADD s1, s1, a3, b3
  94. LD a3, 6 * SIZE(X)
  95. LD b3, 6 * SIZE(Y)
  96. MADD s2, s2, a4, b4
  97. LD a4, 7 * SIZE(X)
  98. LD b4, 7 * SIZE(Y)
  99. MADD s1, s1, a1, b1
  100. LD a1, 8 * SIZE(X)
  101. LD b1, 8 * SIZE(Y)
  102. MADD s2, s2, a2, b2
  103. LD a2, 9 * SIZE(X)
  104. LD b2, 9 * SIZE(Y)
  105. MADD s1, s1, a3, b3
  106. LD a3, 10 * SIZE(X)
  107. LD b3, 10 * SIZE(Y)
  108. MADD s2, s2, a4, b4
  109. LD a4, 11 * SIZE(X)
  110. LD b4, 11 * SIZE(Y)
  111. daddiu I, I, -1
  112. daddiu X, X, 8 * SIZE
  113. bgtz I, .L12
  114. daddiu Y, Y, 8 * SIZE
  115. .align 3
  116. .L13:
  117. MADD s1, s1, a1, b1
  118. LD a1, 4 * SIZE(X)
  119. LD b1, 4 * SIZE(Y)
  120. MADD s2, s2, a2, b2
  121. LD a2, 5 * SIZE(X)
  122. LD b2, 5 * SIZE(Y)
  123. MADD s1, s1, a3, b3
  124. LD a3, 6 * SIZE(X)
  125. LD b3, 6 * SIZE(Y)
  126. MADD s2, s2, a4, b4
  127. LD a4, 7 * SIZE(X)
  128. LD b4, 7 * SIZE(Y)
  129. MADD s1, s1, a1, b1
  130. daddiu X, X, 8 * SIZE
  131. MADD s2, s2, a2, b2
  132. daddiu Y, Y, 8 * SIZE
  133. MADD s1, s1, a3, b3
  134. MADD s2, s2, a4, b4
  135. .align 3
  136. .L15:
  137. andi I, N, 7
  138. blez I, .L999
  139. NOP
  140. .align 3
  141. .L16:
  142. LD a1, 0 * SIZE(X)
  143. LD b1, 0 * SIZE(Y)
  144. MADD s1, s1, a1, b1
  145. daddiu I, I, -1
  146. daddiu X, X, SIZE
  147. daddiu Y, Y, SIZE
  148. bgtz I, .L16
  149. NOP
  150. j .L999
  151. NOP
  152. .align 3
  153. .L20:
  154. #ifdef F_INTERFACE
  155. bgez INCX, .L21
  156. daddiu TEMP, N, -1
  157. mult TEMP, INCX
  158. mflo TEMP
  159. dsub X, X, TEMP
  160. .align 3
  161. .L21:
  162. bgez INCY, .L22
  163. daddiu TEMP, N, -1
  164. mult TEMP, INCY
  165. mflo TEMP
  166. dsub Y, Y, TEMP
  167. .align 3
  168. .L22:
  169. #endif
  170. blez I, .L25
  171. NOP
  172. .align 3
  173. .L23:
  174. LD a1, 0 * SIZE(X)
  175. dadd X, X, INCX
  176. LD b1, 0 * SIZE(Y)
  177. dadd Y, Y, INCY
  178. MADD s1, s1, a1, b1
  179. LD a1, 0 * SIZE(X)
  180. dadd X, X, INCX
  181. LD b1, 0 * SIZE(Y)
  182. dadd Y, Y, INCY
  183. MADD s2, s2, a1, b1
  184. LD a1, 0 * SIZE(X)
  185. dadd X, X, INCX
  186. LD b1, 0 * SIZE(Y)
  187. dadd Y, Y, INCY
  188. MADD s1, s1, a1, b1
  189. LD a1, 0 * SIZE(X)
  190. dadd X, X, INCX
  191. LD b1, 0 * SIZE(Y)
  192. dadd Y, Y, INCY
  193. MADD s2, s2, a1, b1
  194. LD a1, 0 * SIZE(X)
  195. dadd X, X, INCX
  196. LD b1, 0 * SIZE(Y)
  197. dadd Y, Y, INCY
  198. MADD s1, s1, a1, b1
  199. LD a1, 0 * SIZE(X)
  200. dadd X, X, INCX
  201. LD b1, 0 * SIZE(Y)
  202. dadd Y, Y, INCY
  203. MADD s2, s2, a1, b1
  204. LD a1, 0 * SIZE(X)
  205. dadd X, X, INCX
  206. LD b1, 0 * SIZE(Y)
  207. dadd Y, Y, INCY
  208. MADD s1, s1, a1, b1
  209. LD a1, 0 * SIZE(X)
  210. dadd X, X, INCX
  211. LD b1, 0 * SIZE(Y)
  212. dadd Y, Y, INCY
  213. daddiu I, I, -1
  214. bgtz I, .L23
  215. MADD s2, s2, a1, b1
  216. .align 3
  217. .L25:
  218. andi I, N, 7
  219. blez I, .L999
  220. NOP
  221. .align 3
  222. .L26:
  223. LD a1, 0 * SIZE(X)
  224. dadd X, X, INCX
  225. LD b1, 0 * SIZE(Y)
  226. dadd Y, Y, INCY
  227. daddiu I, I, -1
  228. bgtz I, .L26
  229. MADD s1, s1, a1, b1
  230. .align 3
  231. .L999:
  232. ADD s1, s1, s2
  233. #ifdef DSDOT
  234. cvt.d.s s1, s1
  235. #endif
  236. j $31
  237. NOP
  238. EPILOGUE