media/base/simd/linear_scale_yuv_to_rgb_mmx_x64.asm

   1 ; Copyright (c) 2011 The Chromium Authors. All rights reserved.
   2 ; Use of this source code is governed by a BSD-style license that can be
   3 ; found in the LICENSE file.
   4
   5 %include "media/base/simd/media_export.asm"
   6 %include "third_party/x86inc/x86inc.asm"
   7
   8 ;
   9 ; This file uses MMX instructions.
  10 ;
  11   SECTION_TEXT
  12   CPU       MMX
  13
  14 ;void LinearScaleYUVToRGB32Row_MMX_X64(const uint8* y_buf,
  15 ;                                      const uint8* u_buf,
  16 ;                                      const uint8* v_buf,
  17 ;                                      uint8* rgb_buf,
  18 ;                                      ptrdiff_t width,
  19 ;                                      ptrdiff_t source_dx);
  20 %define SYMBOL LinearScaleYUVToRGB32Row_MMX_X64
  21   EXPORT    SYMBOL
  22   align     function_align
  23
  24 mangle(SYMBOL):
  25   %assign   stack_offset 0
  26   extern    mangle(kCoefficientsRgbY)
  27
  28 ; Parameters are in the following order:
  29 ; 1. Y plane
  30 ; 2. U plane
  31 ; 3. V plane
  32 ; 4. ARGB frame
  33 ; 5. Width
  34 ; 6. Source dx
  35 ; 7. Conversion lookup table
  36
  37 PROLOGUE  7, 7, 3, Y, U, V, ARGB, WIDTH, SOURCE_DX, R1
  38
  39 %define     TABLEq     r10
  40 %define     Xq         r11
  41 %define     INDEXq     r12
  42 %define     COMPRd     r13d
  43 %define     COMPRq     r13
  44 %define     FRACTIONq  r14
  45 %define     COMPL      R1
  46 %define     COMPLq     R1q
  47 %define     COMPLd     R1d
  48
  49   PUSH      TABLEq
  50   PUSH      Xq
  51   PUSH      INDEXq
  52   PUSH      COMPRq
  53   PUSH      FRACTIONq
  54
  55 %macro EPILOGUE 0
  56   POP       FRACTIONq
  57   POP       COMPRq
  58   POP       INDEXq
  59   POP       Xq
  60   POP       TABLEq
  61 %endmacro
  62
  63   mov       TABLEq, R1q
  64
  65   imul      WIDTHq, SOURCE_DXq           ; source_width = width * source_dx
  66   xor       Xq, Xq                       ; x = 0
  67   cmp       SOURCE_DXq, 0x20000
  68   jl        .lscaleend
  69   mov       Xq, 0x8000                   ; x = 0.5 for 1/2 or less
  70   jmp       .lscaleend
  71
  72 .lscaleloop:
  73   ; Interpolate U
  74   mov       INDEXq, Xq
  75   sar       INDEXq, 0x11
  76   movzx     COMPLd, BYTE [Uq + INDEXq]
  77   movzx     COMPRd, BYTE [Uq + INDEXq + 1]
  78   mov       FRACTIONq, Xq
  79   and       FRACTIONq, 0x1fffe
  80   imul      COMPRq, FRACTIONq
  81   xor       FRACTIONq, 0x1fffe
  82   imul      COMPLq, FRACTIONq
  83   add       COMPLq, COMPRq
  84   shr       COMPLq, 17
  85   movq      mm0, [TABLEq + 2048 + 8 * COMPLq]
  86
  87   ; Interpolate V
  88   movzx     COMPLd, BYTE [Vq + INDEXq]
  89   movzx     COMPRd, BYTE [Vq + INDEXq + 1]
  90   ; Trick here to imul COMPL first then COMPR.
  91   ; Saves two instruction. :)
  92   imul      COMPLq, FRACTIONq
  93   xor       FRACTIONq, 0x1fffe
  94   imul      COMPRq, FRACTIONq
  95   add       COMPLq, COMPRq
  96   shr       COMPLq, 17
  97   paddsw    mm0, [TABLEq + 4096 + 8 * COMPLq]
  98
  99   ; Interpolate first Y1.
 100   lea       INDEXq, [Xq + SOURCE_DXq]   ; INDEXq now points to next pixel.
 101                                         ; Xq points to current pixel.
 102   mov       FRACTIONq, Xq
 103   sar       Xq, 0x10
 104   movzx     COMPLd, BYTE [Yq + Xq]
 105   movzx     COMPRd, BYTE [Yq + Xq + 1]
 106   and       FRACTIONq, 0xffff
 107   imul      COMPRq, FRACTIONq
 108   xor       FRACTIONq, 0xffff
 109   imul      COMPLq, FRACTIONq
 110   add       COMPLq, COMPRq
 111   shr       COMPLq, 16
 112   movq      mm1, [TABLEq + 8 * COMPLq]
 113
 114   ; Interpolate Y2 if available.
 115   cmp       INDEXq, WIDTHq
 116   jge       .lscalelastpixel
 117
 118   lea       Xq, [INDEXq + SOURCE_DXq]    ; Xq points to next pixel.
 119                                          ; INDEXq points to current pixel.
 120   mov       FRACTIONq, INDEXq
 121   sar       INDEXq, 0x10
 122   movzx     COMPLd, BYTE [Yq + INDEXq]
 123   movzx     COMPRd, BYTE [Yq + INDEXq + 1]
 124   and       FRACTIONq, 0xffff
 125   imul      COMPRq, FRACTIONq
 126   xor       FRACTIONq, 0xffff
 127   imul      COMPLq, FRACTIONq
 128   add       COMPLq, COMPRq
 129   shr       COMPLq, 16
 130   movq      mm2, [TABLEq + 8 * COMPLq]
 131
 132   paddsw    mm1, mm0
 133   paddsw    mm2, mm0
 134   psraw     mm1, 0x6
 135   psraw     mm2, 0x6
 136   packuswb  mm1, mm2
 137   movntq    [ARGBq], mm1
 138   add       ARGBq, 0x8
 139
 140 .lscaleend:
 141   cmp       Xq, WIDTHq
 142   jl        .lscaleloop
 143   jmp       .epilogue
 144
 145 .lscalelastpixel:
 146   paddsw    mm1, mm0
 147   psraw     mm1, 6
 148   packuswb  mm1, mm1
 149   movd      [ARGBq], mm1
 150
 151 .epilogue
 152   EPILOGUE
 153   RET