gfx/ycbcr/yuv_row.h

   1 // Copyright (c) 2010 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 // yuv_row internal functions to handle YUV conversion and scaling to RGB.
   6 // These functions are used from both yuv_convert.cc and yuv_scale.cc.
   7
   8 // TODO(fbarchard): Write function that can handle rotation and scaling.
   9
  10 #ifndef MEDIA_BASE_YUV_ROW_H_
  11 #define MEDIA_BASE_YUV_ROW_H_
  12
  13 #include "chromium_types.h"
  14
  15 extern "C" {
  16 // Can only do 1x.
  17 // This is the second fastest of the scalers.
  18 void FastConvertYUVToRGB32Row(const uint8_t* y_buf,
  19                               const uint8_t* u_buf,
  20                               const uint8_t* v_buf,
  21                               uint8_t* rgb_buf,
  22                               int width);
  23
  24 void FastConvertYUVToRGB32Row_C(const uint8_t* y_buf,
  25                                 const uint8_t* u_buf,
  26                                 const uint8_t* v_buf,
  27                                 uint8_t* rgb_buf,
  28                                 int width,
  29                                 unsigned int x_shift);
  30
  31 void FastConvertYUVToRGB32Row(const uint8_t* y_buf,
  32                               const uint8_t* u_buf,
  33                               const uint8_t* v_buf,
  34                               uint8_t* rgb_buf,
  35                               int width);
  36
  37 // Can do 1x, half size or any scale down by an integer amount.
  38 // Step can be negative (mirroring, rotate 180).
  39 // This is the third fastest of the scalers.
  40 // Only defined on Windows x86-32.
  41 void ConvertYUVToRGB32Row_SSE(const uint8_t* y_buf,
  42                               const uint8_t* u_buf,
  43                               const uint8_t* v_buf,
  44                               uint8_t* rgb_buf,
  45                               int width,
  46                               int step);
  47
  48 // Rotate is like Convert, but applies different step to Y versus U and V.
  49 // This allows rotation by 90 or 270, by stepping by stride.
  50 // This is the forth fastest of the scalers.
  51 // Only defined on Windows x86-32.
  52 void RotateConvertYUVToRGB32Row_SSE(const uint8_t* y_buf,
  53                                     const uint8_t* u_buf,
  54                                     const uint8_t* v_buf,
  55                                     uint8_t* rgb_buf,
  56                                     int width,
  57                                     int ystep,
  58                                     int uvstep);
  59
  60 // Doubler does 4 pixels at a time.  Each pixel is replicated.
  61 // This is the fastest of the scalers.
  62 // Only defined on Windows x86-32.
  63 void DoubleYUVToRGB32Row_SSE(const uint8_t* y_buf,
  64                              const uint8_t* u_buf,
  65                              const uint8_t* v_buf,
  66                              uint8_t* rgb_buf,
  67                              int width);
  68
  69 // Handles arbitrary scaling up or down.
  70 // Mirroring is supported, but not 90 or 270 degree rotation.
  71 // Chroma is under sampled every 2 pixels for performance.
  72 void ScaleYUVToRGB32Row(const uint8_t* y_buf,
  73                         const uint8_t* u_buf,
  74                         const uint8_t* v_buf,
  75                         uint8_t* rgb_buf,
  76                         int width,
  77                         int source_dx);
  78
  79 void ScaleYUVToRGB32Row(const uint8_t* y_buf,
  80                         const uint8_t* u_buf,
  81                         const uint8_t* v_buf,
  82                         uint8_t* rgb_buf,
  83                         int width,
  84                         int source_dx);
  85
  86 void ScaleYUVToRGB32Row_C(const uint8_t* y_buf,
  87                           const uint8_t* u_buf,
  88                           const uint8_t* v_buf,
  89                           uint8_t* rgb_buf,
  90                           int width,
  91                           int source_dx);
  92
  93 // Handles arbitrary scaling up or down with bilinear filtering.
  94 // Mirroring is supported, but not 90 or 270 degree rotation.
  95 // Chroma is under sampled every 2 pixels for performance.
  96 // This is the slowest of the scalers.
  97 void LinearScaleYUVToRGB32Row(const uint8_t* y_buf,
  98                               const uint8_t* u_buf,
  99                               const uint8_t* v_buf,
 100                               uint8_t* rgb_buf,
 101                               int width,
 102                               int source_dx);
 103
 104 void LinearScaleYUVToRGB32Row(const uint8_t* y_buf,
 105                               const uint8_t* u_buf,
 106                               const uint8_t* v_buf,
 107                               uint8_t* rgb_buf,
 108                               int width,
 109                               int source_dx);
 110
 111 void LinearScaleYUVToRGB32Row_C(const uint8_t* y_buf,
 112                                 const uint8_t* u_buf,
 113                                 const uint8_t* v_buf,
 114                                 uint8_t* rgb_buf,
 115                                 int width,
 116                                 int source_dx);
 117
 118
 119 #if defined(_MSC_VER) && !defined(__CLR_VER) && !defined(__clang__)
 120 #if defined(VISUALC_HAS_AVX2)
 121 #define SIMD_ALIGNED(var) __declspec(align(32)) var
 122 #else
 123 #define SIMD_ALIGNED(var) __declspec(align(16)) var
 124 #endif
 125 #elif defined(__GNUC__) || defined(__clang__)
 126 // Caveat GCC 4.2 to 4.7 have a known issue using vectors with const.
 127 #if defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2)
 128 #define SIMD_ALIGNED(var) var __attribute__((aligned(32)))
 129 #else
 130 #define SIMD_ALIGNED(var) var __attribute__((aligned(16)))
 131 #endif
 132 #else
 133 #define SIMD_ALIGNED(var) var
 134 #endif
 135
 136 extern SIMD_ALIGNED(const int16_t kCoefficientsRgbY[768][4]);
 137
 138 // x64 uses MMX2 (SSE) so emms is not required.
 139 // Warning C4799: function has no EMMS instruction.
 140 // EMMS() is slow and should be called by the calling function once per image.
 141 #if defined(ARCH_CPU_X86) && !defined(ARCH_CPU_X86_64)
 142 #if defined(_MSC_VER)
 143 #define EMMS() __asm emms
 144 #pragma warning(disable: 4799)
 145 #else
 146 #define EMMS() asm("emms")
 147 #endif
 148 #else
 149 #define EMMS() ((void)0)
 150 #endif
 151
 152 }  // extern "C"
 153
 154 #endif  // MEDIA_BASE_YUV_ROW_H_