media/base/simd/convert_yuv_to_rgb_c.cc

   1 // Copyright (c) 2012 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/convert_yuv_to_rgb.h"
   6 #include "media/base/simd/yuv_to_rgb_table.h"
   7
   8 namespace media {
   9
  10 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x)))
  11 #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \
  12     (((x) + (y)) > 32767 ? 32767 : ((x) + (y))))
  13
  14 // On Android, pixel layout is RGBA (see skia/include/core/SkColorPriv.h);
  15 // however, other Chrome platforms use BGRA (see skia/config/SkUserConfig.h).
  16 // Ideally, android should not use the functions here due to performance issue
  17 // (http://crbug.com/249980).
  18 #if defined(OS_ANDROID)
  19 #define SK_R32_SHIFT    0
  20 #define SK_G32_SHIFT    8
  21 #define SK_B32_SHIFT    16
  22 #define SK_A32_SHIFT    24
  23 #define R_INDEX         0
  24 #define G_INDEX         1
  25 #define B_INDEX         2
  26 #define A_INDEX         3
  27 #else
  28 #define SK_B32_SHIFT    0
  29 #define SK_G32_SHIFT    8
  30 #define SK_R32_SHIFT    16
  31 #define SK_A32_SHIFT    24
  32 #define B_INDEX         0
  33 #define G_INDEX         1
  34 #define R_INDEX         2
  35 #define A_INDEX         3
  36 #endif
  37
  38 static inline void ConvertYUVToRGB32_C(uint8 y,
  39                                        uint8 u,
  40                                        uint8 v,
  41                                        uint8* rgb_buf) {
  42   int b = kCoefficientsRgbY[256+u][B_INDEX];
  43   int g = kCoefficientsRgbY[256+u][G_INDEX];
  44   int r = kCoefficientsRgbY[256+u][R_INDEX];
  45   int a = kCoefficientsRgbY[256+u][A_INDEX];
  46
  47   b = paddsw(b, kCoefficientsRgbY[512+v][B_INDEX]);
  48   g = paddsw(g, kCoefficientsRgbY[512+v][G_INDEX]);
  49   r = paddsw(r, kCoefficientsRgbY[512+v][R_INDEX]);
  50   a = paddsw(a, kCoefficientsRgbY[512+v][A_INDEX]);
  51
  52   b = paddsw(b, kCoefficientsRgbY[y][B_INDEX]);
  53   g = paddsw(g, kCoefficientsRgbY[y][G_INDEX]);
  54   r = paddsw(r, kCoefficientsRgbY[y][R_INDEX]);
  55   a = paddsw(a, kCoefficientsRgbY[y][A_INDEX]);
  56
  57   b >>= 6;
  58   g >>= 6;
  59   r >>= 6;
  60   a >>= 6;
  61
  62   *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b) << SK_B32_SHIFT) |
  63                                         (packuswb(g) << SK_G32_SHIFT) |
  64                                         (packuswb(r) << SK_R32_SHIFT) |
  65                                         (packuswb(a) << SK_A32_SHIFT);
  66 }
  67
  68 static inline void ConvertYUVAToARGB_C(uint8 y,
  69                                        uint8 u,
  70                                        uint8 v,
  71                                        uint8 a,
  72                                        uint8* rgb_buf) {
  73   int b = kCoefficientsRgbY[256+u][0];
  74   int g = kCoefficientsRgbY[256+u][1];
  75   int r = kCoefficientsRgbY[256+u][2];
  76
  77   b = paddsw(b, kCoefficientsRgbY[512+v][0]);
  78   g = paddsw(g, kCoefficientsRgbY[512+v][1]);
  79   r = paddsw(r, kCoefficientsRgbY[512+v][2]);
  80
  81   b = paddsw(b, kCoefficientsRgbY[y][0]);
  82   g = paddsw(g, kCoefficientsRgbY[y][1]);
  83   r = paddsw(r, kCoefficientsRgbY[y][2]);
  84
  85   b >>= 6;
  86   g >>= 6;
  87   r >>= 6;
  88
  89   b = packuswb(b) * a >> 8;
  90   g = packuswb(g) * a >> 8;
  91   r = packuswb(r) * a >> 8;
  92
  93   *reinterpret_cast<uint32*>(rgb_buf) = (b << SK_B32_SHIFT) |
  94                                         (g << SK_G32_SHIFT) |
  95                                         (r << SK_R32_SHIFT) |
  96                                         (a << SK_A32_SHIFT);
  97 }
  98
  99 void ConvertYUVToRGB32Row_C(const uint8* y_buf,
 100                             const uint8* u_buf,
 101                             const uint8* v_buf,
 102                             uint8* rgb_buf,
 103                             ptrdiff_t width) {
 104   for (int x = 0; x < width; x += 2) {
 105     uint8 u = u_buf[x >> 1];
 106     uint8 v = v_buf[x >> 1];
 107     uint8 y0 = y_buf[x];
 108     ConvertYUVToRGB32_C(y0, u, v, rgb_buf);
 109     if ((x + 1) < width) {
 110       uint8 y1 = y_buf[x + 1];
 111       ConvertYUVToRGB32_C(y1, u, v, rgb_buf + 4);
 112     }
 113     rgb_buf += 8;  // Advance 2 pixels.
 114   }
 115 }
 116
 117 void ConvertYUVAToARGBRow_C(const uint8* y_buf,
 118                             const uint8* u_buf,
 119                             const uint8* v_buf,
 120                             const uint8* a_buf,
 121                             uint8* rgba_buf,
 122                             ptrdiff_t width) {
 123   for (int x = 0; x < width; x += 2) {
 124     uint8 u = u_buf[x >> 1];
 125     uint8 v = v_buf[x >> 1];
 126     uint8 y0 = y_buf[x];
 127     uint8 a0 = a_buf[x];
 128     ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf);
 129     if ((x + 1) < width) {
 130       uint8 y1 = y_buf[x + 1];
 131       uint8 a1 = a_buf[x + 1];
 132       ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4);
 133     }
 134     rgba_buf += 8;  // Advance 2 pixels.
 135   }
 136 }
 137
 138 // 16.16 fixed point is used.  A shift by 16 isolates the integer.
 139 // A shift by 17 is used to further subsample the chrominence channels.
 140 // & 0xffff isolates the fixed point fraction.  >> 2 to get the upper 2 bits,
 141 // for 1/65536 pixel accurate interpolation.
 142 void ScaleYUVToRGB32Row_C(const uint8* y_buf,
 143                           const uint8* u_buf,
 144                           const uint8* v_buf,
 145                           uint8* rgb_buf,
 146                           ptrdiff_t width,
 147                           ptrdiff_t source_dx) {
 148   int x = 0;
 149   for (int i = 0; i < width; i += 2) {
 150     int y = y_buf[x >> 16];
 151     int u = u_buf[(x >> 17)];
 152     int v = v_buf[(x >> 17)];
 153     ConvertYUVToRGB32_C(y, u, v, rgb_buf);
 154     x += source_dx;
 155     if ((i + 1) < width) {
 156       y = y_buf[x >> 16];
 157       ConvertYUVToRGB32_C(y, u, v, rgb_buf+4);
 158       x += source_dx;
 159     }
 160     rgb_buf += 8;
 161   }
 162 }
 163
 164 void LinearScaleYUVToRGB32Row_C(const uint8* y_buf,
 165                                 const uint8* u_buf,
 166                                 const uint8* v_buf,
 167                                 uint8* rgb_buf,
 168                                 ptrdiff_t width,
 169                                 ptrdiff_t source_dx) {
 170   // Avoid point-sampling for down-scaling by > 2:1.
 171   int source_x = 0;
 172   if (source_dx >= 0x20000)
 173     source_x += 0x8000;
 174   LinearScaleYUVToRGB32RowWithRange_C(y_buf, u_buf, v_buf, rgb_buf, width,
 175                                       source_x, source_dx);
 176 }
 177
 178 void LinearScaleYUVToRGB32RowWithRange_C(const uint8* y_buf,
 179                                          const uint8* u_buf,
 180                                          const uint8* v_buf,
 181                                          uint8* rgb_buf,
 182                                          int dest_width,
 183                                          int x,
 184                                          int source_dx) {
 185   for (int i = 0; i < dest_width; i += 2) {
 186     int y0 = y_buf[x >> 16];
 187     int y1 = y_buf[(x >> 16) + 1];
 188     int u0 = u_buf[(x >> 17)];
 189     int u1 = u_buf[(x >> 17) + 1];
 190     int v0 = v_buf[(x >> 17)];
 191     int v1 = v_buf[(x >> 17) + 1];
 192     int y_frac = (x & 65535);
 193     int uv_frac = ((x >> 1) & 65535);
 194     int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
 195     int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16;
 196     int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16;
 197     ConvertYUVToRGB32_C(y, u, v, rgb_buf);
 198     x += source_dx;
 199     if ((i + 1) < dest_width) {
 200       y0 = y_buf[x >> 16];
 201       y1 = y_buf[(x >> 16) + 1];
 202       y_frac = (x & 65535);
 203       y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16;
 204       ConvertYUVToRGB32_C(y, u, v, rgb_buf+4);
 205       x += source_dx;
 206     }
 207     rgb_buf += 8;
 208   }
 209 }
 210
 211 void ConvertYUVToRGB32_C(const uint8* yplane,
 212                          const uint8* uplane,
 213                          const uint8* vplane,
 214                          uint8* rgbframe,
 215                          int width,
 216                          int height,
 217                          int ystride,
 218                          int uvstride,
 219                          int rgbstride,
 220                          YUVType yuv_type) {
 221   unsigned int y_shift = yuv_type;
 222   for (int y = 0; y < height; ++y) {
 223     uint8* rgb_row = rgbframe + y * rgbstride;
 224     const uint8* y_ptr = yplane + y * ystride;
 225     const uint8* u_ptr = uplane + (y >> y_shift) * uvstride;
 226     const uint8* v_ptr = vplane + (y >> y_shift) * uvstride;
 227
 228     ConvertYUVToRGB32Row_C(y_ptr,
 229                            u_ptr,
 230                            v_ptr,
 231                            rgb_row,
 232                            width);
 233   }
 234 }
 235
 236 void ConvertYUVAToARGB_C(const uint8* yplane,
 237                          const uint8* uplane,
 238                          const uint8* vplane,
 239                          const uint8* aplane,
 240                          uint8* rgbaframe,
 241                          int width,
 242                          int height,
 243                          int ystride,
 244                          int uvstride,
 245                          int astride,
 246                          int rgbastride,
 247                          YUVType yuv_type) {
 248   unsigned int y_shift = yuv_type;
 249   for (int y = 0; y < height; y++) {
 250     uint8* rgba_row = rgbaframe + y * rgbastride;
 251     const uint8* y_ptr = yplane + y * ystride;
 252     const uint8* u_ptr = uplane + (y >> y_shift) * uvstride;
 253     const uint8* v_ptr = vplane + (y >> y_shift) * uvstride;
 254     const uint8* a_ptr = aplane + y * astride;
 255
 256     ConvertYUVAToARGBRow_C(y_ptr,
 257                            u_ptr,
 258                            v_ptr,
 259                            a_ptr,
 260                            rgba_row,
 261                            width);
 262   }
 263 }
 264
 265 }  // namespace media