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_vfp.S 6.1 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286
  1. /***************************************************************************
  2. Copyright (c) 2013, The OpenBLAS Project
  3. All rights reserved.
  4. Redistribution and use in source and binary forms, with or without
  5. modification, are permitted provided that the following conditions are
  6. met:
  7. 1. Redistributions of source code must retain the above copyright
  8. notice, this list of conditions and the following disclaimer.
  9. 2. Redistributions in binary form must reproduce the above copyright
  10. notice, this list of conditions and the following disclaimer in
  11. the documentation and/or other materials provided with the
  12. distribution.
  13. 3. Neither the name of the OpenBLAS project nor the names of
  14. its contributors may be used to endorse or promote products
  15. derived from this software without specific prior written permission.
  16. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  17. AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18. IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19. ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
  20. LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21. DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
  22. SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
  23. CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  24. OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
  25. USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  26. *****************************************************************************/
  27. /**************************************************************************************
  28. * 2013/11/11 Saar
  29. * BLASTEST : OK
  30. * CTEST : OK
  31. * TEST : OK
  32. *
  33. **************************************************************************************/
  34. #define ASSEMBLER
  35. #include "common.h"
  36. #define STACKSIZE 256
  37. #define N r0
  38. #define X r1
  39. #define INC_X r2
  40. #define OLD_Y r3
  41. /******************************************************
  42. * [fp, #-128] - [fp, #-64] is reserved
  43. * for store and restore of floating point
  44. * registers
  45. *******************************************************/
  46. #define OLD_INC_Y [fp, #4 ]
  47. #define I r5
  48. #define Y r6
  49. #define INC_Y r7
  50. #define X_PRE 512
  51. /**************************************************************************************
  52. * Macro definitions
  53. **************************************************************************************/
  54. .macro KERNEL_F4
  55. pld [ X, #X_PRE ]
  56. pld [ Y, #X_PRE ]
  57. fldmiad X!, { d4 - d5 }
  58. fldmiad Y!, { d8 - d9 }
  59. fmacd d0 , d4, d8
  60. fmacd d1 , d4, d9
  61. fldmiad X!, { d6 - d7 }
  62. fmacd d2 , d5, d9
  63. fmacd d3 , d5, d8
  64. fldmiad Y!, { d10 - d11 }
  65. fmacd d0 , d6, d10
  66. fmacd d1 , d6, d11
  67. pld [ X, #X_PRE ]
  68. fmacd d2 , d7, d11
  69. fmacd d3 , d7, d10
  70. pld [ Y, #X_PRE ]
  71. fldmiad X!, { d4 - d5 }
  72. fldmiad Y!, { d8 - d9 }
  73. fmacd d0 , d4, d8
  74. fmacd d1 , d4, d9
  75. fldmiad X!, { d6 - d7 }
  76. fmacd d2 , d5, d9
  77. fmacd d3 , d5, d8
  78. fldmiad Y!, { d10 - d11 }
  79. fmacd d0 , d6, d10
  80. fmacd d1 , d6, d11
  81. fmacd d2 , d7, d11
  82. fmacd d3 , d7, d10
  83. .endm
  84. .macro KERNEL_F1
  85. fldmiad X!, { d4 - d5 }
  86. fldmiad Y!, { d8 - d9 }
  87. fmacd d0 , d4, d8
  88. fmacd d1 , d4, d9
  89. fmacd d2 , d5, d9
  90. fmacd d3 , d5, d8
  91. .endm
  92. /*************************************************************************************************************************/
  93. .macro KERNEL_S4
  94. nop
  95. fldmiad X, { d4 - d5 }
  96. fldmiad Y, { d8 - d9 }
  97. fmacd d0 , d4, d8
  98. fmacd d1 , d4, d9
  99. fmacd d2 , d5, d9
  100. fmacd d3 , d5, d8
  101. add X, X, INC_X
  102. add Y, Y, INC_Y
  103. fldmiad X, { d4 - d5 }
  104. fldmiad Y, { d8 - d9 }
  105. fmacd d0 , d4, d8
  106. fmacd d1 , d4, d9
  107. fmacd d2 , d5, d9
  108. fmacd d3 , d5, d8
  109. add X, X, INC_X
  110. add Y, Y, INC_Y
  111. fldmiad X, { d4 - d5 }
  112. fldmiad Y, { d8 - d9 }
  113. fmacd d0 , d4, d8
  114. fmacd d1 , d4, d9
  115. fmacd d2 , d5, d9
  116. fmacd d3 , d5, d8
  117. add X, X, INC_X
  118. add Y, Y, INC_Y
  119. fldmiad X, { d4 - d5 }
  120. fldmiad Y, { d8 - d9 }
  121. fmacd d0 , d4, d8
  122. fmacd d1 , d4, d9
  123. fmacd d2 , d5, d9
  124. fmacd d3 , d5, d8
  125. add X, X, INC_X
  126. add Y, Y, INC_Y
  127. .endm
  128. .macro KERNEL_S1
  129. fldmiad X, { d4 - d5 }
  130. fldmiad Y, { d8 - d9 }
  131. fmacd d0 , d4, d8
  132. fmacd d1 , d4, d9
  133. fmacd d2 , d5, d9
  134. fmacd d3 , d5, d8
  135. add X, X, INC_X
  136. add Y, Y, INC_Y
  137. .endm
  138. /**************************************************************************************
  139. * End of macro definitions
  140. **************************************************************************************/
  141. PROLOGUE
  142. .align 5
  143. push {r4 - r9, fp}
  144. add fp, sp, #24
  145. sub sp, sp, #STACKSIZE // reserve stack
  146. sub r4, fp, #128
  147. vstm r4, { d8 - d15} // store floating point registers
  148. mov Y, OLD_Y
  149. ldr INC_Y, OLD_INC_Y
  150. vsub.f64 d0 , d0 , d0
  151. vsub.f64 d1 , d1 , d1
  152. vsub.f64 d2 , d2 , d2
  153. vsub.f64 d3 , d3 , d3
  154. cmp N, #0
  155. ble zdot_kernel_L999
  156. cmp INC_X, #0
  157. beq zdot_kernel_L999
  158. cmp INC_Y, #0
  159. beq zdot_kernel_L999
  160. cmp INC_X, #1
  161. bne zdot_kernel_S_BEGIN
  162. cmp INC_Y, #1
  163. bne zdot_kernel_S_BEGIN
  164. zdot_kernel_F_BEGIN:
  165. asrs I, N, #2 // I = N / 4
  166. ble zdot_kernel_F1
  167. zdot_kernel_F4:
  168. KERNEL_F4
  169. subs I, I, #1
  170. bne zdot_kernel_F4
  171. zdot_kernel_F1:
  172. ands I, N, #3
  173. ble zdot_kernel_L999
  174. zdot_kernel_F10:
  175. KERNEL_F1
  176. subs I, I, #1
  177. bne zdot_kernel_F10
  178. b zdot_kernel_L999
  179. zdot_kernel_S_BEGIN:
  180. lsl INC_X, INC_X, #4 // INC_X * SIZE * 2
  181. lsl INC_Y, INC_Y, #4 // INC_Y * SIZE * 2
  182. asrs I, N, #2 // I = N / 4
  183. ble zdot_kernel_S1
  184. zdot_kernel_S4:
  185. KERNEL_S4
  186. subs I, I, #1
  187. bne zdot_kernel_S4
  188. zdot_kernel_S1:
  189. ands I, N, #3
  190. ble zdot_kernel_L999
  191. zdot_kernel_S10:
  192. KERNEL_S1
  193. subs I, I, #1
  194. bne zdot_kernel_S10
  195. zdot_kernel_L999:
  196. sub r3, fp, #128
  197. vldm r3, { d8 - d15} // restore floating point registers
  198. #if !defined(CONJ)
  199. vsub.f64 d0 , d0, d2
  200. vadd.f64 d1 , d1, d3
  201. #else
  202. vadd.f64 d0 , d0, d2
  203. vsub.f64 d1 , d1, d3
  204. #endif
  205. sub sp, fp, #24
  206. pop {r4 - r9, fp}
  207. bx lr
  208. EPILOGUE