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.

zscal.S 5.8 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255
  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 PREFETCHSIZE 88
  41. #define N $16
  42. #define X $21
  43. #define INCX $17
  44. #define XX $18
  45. #define I $19
  46. #define ALPHA_R $f19
  47. #define ALPHA_I $f20
  48. #define s0 $f0
  49. #define s1 $f1
  50. #define s2 $f10
  51. #define s3 $f11
  52. #define a0 $f12
  53. #define a1 $f13
  54. #define a2 $f14
  55. #define a3 $f15
  56. #define a4 $f16
  57. #define a5 $f17
  58. #define a6 $f18
  59. #define a7 $f21
  60. #define t0 $f22
  61. #define t1 $f23
  62. #define t2 $f24
  63. #define t3 $f25
  64. #define t4 $f26
  65. #define t5 $f27
  66. #define t6 $f28
  67. #define t7 $f29
  68. PROLOGUE
  69. PROFCODE
  70. ldq INCX, 0($sp)
  71. mov X, XX
  72. ble N, $L999
  73. addq INCX, INCX, INCX
  74. sra N, 2, I
  75. ble I, $L15
  76. LD a0, 0 * SIZE(X)
  77. LD a1, 1 * SIZE(X)
  78. SXADDQ INCX, X, X
  79. LD a2, 0 * SIZE(X)
  80. LD a3, 1 * SIZE(X)
  81. SXADDQ INCX, X, X
  82. LD a4, 0 * SIZE(X)
  83. LD a5, 1 * SIZE(X)
  84. SXADDQ INCX, X, X
  85. LD a6, 0 * SIZE(X)
  86. LD a7, 1 * SIZE(X)
  87. SXADDQ INCX, X, X
  88. MUL a0, ALPHA_R, t0
  89. MUL a1, ALPHA_I, t1
  90. MUL a0, ALPHA_I, t2
  91. MUL a1, ALPHA_R, t3
  92. SUB t0, t1, t4
  93. ADD t2, t3, t5
  94. lda I, -1(I)
  95. ble I, $L13
  96. .align 4
  97. $L12:
  98. ST t4, 0 * SIZE(XX)
  99. MUL a2, ALPHA_R, t0
  100. ST t5, 1 * SIZE(XX)
  101. MUL a3, ALPHA_I, t1
  102. MUL a2, ALPHA_I, t2
  103. LD a0, 0 * SIZE(X)
  104. MUL a3, ALPHA_R, t3
  105. LD a1, 1 * SIZE(X)
  106. SUB t0, t1, t6
  107. SXADDQ INCX, XX, XX
  108. ADD t2, t3, t7
  109. SXADDQ INCX, X, X
  110. MUL a4, ALPHA_R, t0
  111. ST t6, 0 * SIZE(XX)
  112. MUL a5, ALPHA_I, t1
  113. ST t7, 1 * SIZE(XX)
  114. MUL a4, ALPHA_I, t2
  115. LD a2, 0 * SIZE(X)
  116. MUL a5, ALPHA_R, t3
  117. LD a3, 1 * SIZE(X)
  118. SUB t0, t1, t4
  119. SXADDQ INCX, XX, XX
  120. ADD t2, t3, t5
  121. SXADDQ INCX, X, X
  122. MUL a6, ALPHA_R, t0
  123. ST t4, 0 * SIZE(XX)
  124. MUL a7, ALPHA_I, t1
  125. ST t5, 1 * SIZE(XX)
  126. MUL a6, ALPHA_I, t2
  127. LD a4, 0 * SIZE(X)
  128. MUL a7, ALPHA_R, t3
  129. LD a5, 1 * SIZE(X)
  130. SUB t0, t1, t6
  131. SXADDQ INCX, XX, XX
  132. ADD t2, t3, t7
  133. SXADDQ INCX, X, X
  134. MUL a0, ALPHA_R, t0
  135. ST t6, 0 * SIZE(XX)
  136. MUL a1, ALPHA_I, t1
  137. ST t7, 1 * SIZE(XX)
  138. MUL a0, ALPHA_I, t2
  139. LD a6, 0 * SIZE(X)
  140. MUL a1, ALPHA_R, t3
  141. LD a7, 1 * SIZE(X)
  142. SUB t0, t1, t4
  143. lda I, -1(I)
  144. ADD t2, t3, t5
  145. SXADDQ INCX, XX, XX
  146. lds $f31, PREFETCHSIZE * SIZE(X)
  147. unop
  148. SXADDQ INCX, X, X
  149. bne I, $L12
  150. .align 4
  151. $L13:
  152. MUL a2, ALPHA_R, t0
  153. MUL a3, ALPHA_I, t1
  154. ST t4, 0 * SIZE(XX)
  155. MUL a2, ALPHA_I, t2
  156. ST t5, 1 * SIZE(XX)
  157. MUL a3, ALPHA_R, t3
  158. SUB t0, t1, t6
  159. SXADDQ INCX, XX, XX
  160. ADD t2, t3, t7
  161. unop
  162. ST t6, 0 * SIZE(XX)
  163. MUL a4, ALPHA_R, t0
  164. ST t7, 1 * SIZE(XX)
  165. MUL a5, ALPHA_I, t1
  166. MUL a4, ALPHA_I, t2
  167. MUL a5, ALPHA_R, t3
  168. SUB t0, t1, t4
  169. SXADDQ INCX, XX, XX
  170. ADD t2, t3, t5
  171. unop
  172. MUL a6, ALPHA_R, t0
  173. ST t4, 0 * SIZE(XX)
  174. MUL a7, ALPHA_I, t1
  175. ST t5, 1 * SIZE(XX)
  176. MUL a6, ALPHA_I, t2
  177. MUL a7, ALPHA_R, t3
  178. SUB t0, t1, t6
  179. SXADDQ INCX, XX, XX
  180. ADD t2, t3, t7
  181. ST t6, 0 * SIZE(XX)
  182. ST t7, 1 * SIZE(XX)
  183. SXADDQ INCX, XX, XX
  184. .align 4
  185. $L15:
  186. and N, 3, I
  187. unop
  188. unop
  189. ble I, $L999
  190. .align 4
  191. $L17:
  192. LD a0, 0 * SIZE(X)
  193. LD a1, 1 * SIZE(X)
  194. SXADDQ INCX, X, X
  195. MUL a0, ALPHA_R, t0
  196. MUL a1, ALPHA_I, t1
  197. MUL a0, ALPHA_I, t2
  198. MUL a1, ALPHA_R, t3
  199. SUB t0, t1, t4
  200. ADD t2, t3, t5
  201. ST t4, 0 * SIZE(XX)
  202. ST t5, 1 * SIZE(XX)
  203. SXADDQ INCX, XX, XX
  204. lda I, -1(I)
  205. bne I, $L17
  206. .align 4
  207. $L999:
  208. ret
  209. EPILOGUE