| OLD | NEW |
| 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. | 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 | 2 // Use of this source code is governed by a BSD-style license that can be |
| 3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
| 4 | 4 |
| 5 #include "media/base/simd/convert_yuv_to_rgb.h" | 5 #include "media/base/simd/convert_yuv_to_rgb.h" |
| 6 #include "media/base/simd/yuv_to_rgb_table.h" | 6 #include "media/base/simd/yuv_to_rgb_table.h" |
| 7 | 7 |
| 8 namespace media { | 8 namespace media { |
| 9 | 9 |
| 10 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) | 10 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) |
| (...skipping 20 matching lines...) Expand all Loading... |
| 31 #define SK_A32_SHIFT 24 | 31 #define SK_A32_SHIFT 24 |
| 32 #define B_INDEX 0 | 32 #define B_INDEX 0 |
| 33 #define G_INDEX 1 | 33 #define G_INDEX 1 |
| 34 #define R_INDEX 2 | 34 #define R_INDEX 2 |
| 35 #define A_INDEX 3 | 35 #define A_INDEX 3 |
| 36 #endif | 36 #endif |
| 37 | 37 |
| 38 static inline void ConvertYUVToRGB32_C(uint8 y, | 38 static inline void ConvertYUVToRGB32_C(uint8 y, |
| 39 uint8 u, | 39 uint8 u, |
| 40 uint8 v, | 40 uint8 v, |
| 41 uint8* rgb_buf, | 41 uint8* rgb_buf) { |
| 42 const int16 convert_table[1024][4]) { | 42 int b = kCoefficientsRgbY[256+u][B_INDEX]; |
| 43 int b = convert_table[256+u][B_INDEX]; | 43 int g = kCoefficientsRgbY[256+u][G_INDEX]; |
| 44 int g = convert_table[256+u][G_INDEX]; | 44 int r = kCoefficientsRgbY[256+u][R_INDEX]; |
| 45 int r = convert_table[256+u][R_INDEX]; | 45 int a = kCoefficientsRgbY[256+u][A_INDEX]; |
| 46 int a = convert_table[256+u][A_INDEX]; | |
| 47 | 46 |
| 48 b = paddsw(b, convert_table[512+v][B_INDEX]); | 47 b = paddsw(b, kCoefficientsRgbY[512+v][B_INDEX]); |
| 49 g = paddsw(g, convert_table[512+v][G_INDEX]); | 48 g = paddsw(g, kCoefficientsRgbY[512+v][G_INDEX]); |
| 50 r = paddsw(r, convert_table[512+v][R_INDEX]); | 49 r = paddsw(r, kCoefficientsRgbY[512+v][R_INDEX]); |
| 51 a = paddsw(a, convert_table[512+v][A_INDEX]); | 50 a = paddsw(a, kCoefficientsRgbY[512+v][A_INDEX]); |
| 52 | 51 |
| 53 b = paddsw(b, convert_table[y][B_INDEX]); | 52 b = paddsw(b, kCoefficientsRgbY[y][B_INDEX]); |
| 54 g = paddsw(g, convert_table[y][G_INDEX]); | 53 g = paddsw(g, kCoefficientsRgbY[y][G_INDEX]); |
| 55 r = paddsw(r, convert_table[y][R_INDEX]); | 54 r = paddsw(r, kCoefficientsRgbY[y][R_INDEX]); |
| 56 a = paddsw(a, convert_table[y][A_INDEX]); | 55 a = paddsw(a, kCoefficientsRgbY[y][A_INDEX]); |
| 57 | 56 |
| 58 b >>= 6; | 57 b >>= 6; |
| 59 g >>= 6; | 58 g >>= 6; |
| 60 r >>= 6; | 59 r >>= 6; |
| 61 a >>= 6; | 60 a >>= 6; |
| 62 | 61 |
| 63 *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b) << SK_B32_SHIFT) | | 62 *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b) << SK_B32_SHIFT) | |
| 64 (packuswb(g) << SK_G32_SHIFT) | | 63 (packuswb(g) << SK_G32_SHIFT) | |
| 65 (packuswb(r) << SK_R32_SHIFT) | | 64 (packuswb(r) << SK_R32_SHIFT) | |
| 66 (packuswb(a) << SK_A32_SHIFT); | 65 (packuswb(a) << SK_A32_SHIFT); |
| 67 } | 66 } |
| 68 | 67 |
| 69 static inline void ConvertYUVAToARGB_C(uint8 y, | 68 static inline void ConvertYUVAToARGB_C(uint8 y, |
| 70 uint8 u, | 69 uint8 u, |
| 71 uint8 v, | 70 uint8 v, |
| 72 uint8 a, | 71 uint8 a, |
| 73 uint8* rgb_buf, | 72 uint8* rgb_buf) { |
| 74 const int16 convert_table[1024][4]) { | 73 int b = kCoefficientsRgbY[256+u][0]; |
| 75 int b = convert_table[256+u][0]; | 74 int g = kCoefficientsRgbY[256+u][1]; |
| 76 int g = convert_table[256+u][1]; | 75 int r = kCoefficientsRgbY[256+u][2]; |
| 77 int r = convert_table[256+u][2]; | |
| 78 | 76 |
| 79 b = paddsw(b, convert_table[512+v][0]); | 77 b = paddsw(b, kCoefficientsRgbY[512+v][0]); |
| 80 g = paddsw(g, convert_table[512+v][1]); | 78 g = paddsw(g, kCoefficientsRgbY[512+v][1]); |
| 81 r = paddsw(r, convert_table[512+v][2]); | 79 r = paddsw(r, kCoefficientsRgbY[512+v][2]); |
| 82 | 80 |
| 83 b = paddsw(b, convert_table[y][0]); | 81 b = paddsw(b, kCoefficientsRgbY[y][0]); |
| 84 g = paddsw(g, convert_table[y][1]); | 82 g = paddsw(g, kCoefficientsRgbY[y][1]); |
| 85 r = paddsw(r, convert_table[y][2]); | 83 r = paddsw(r, kCoefficientsRgbY[y][2]); |
| 86 | 84 |
| 87 b >>= 6; | 85 b >>= 6; |
| 88 g >>= 6; | 86 g >>= 6; |
| 89 r >>= 6; | 87 r >>= 6; |
| 90 | 88 |
| 91 b = packuswb(b) * a >> 8; | 89 b = packuswb(b) * a >> 8; |
| 92 g = packuswb(g) * a >> 8; | 90 g = packuswb(g) * a >> 8; |
| 93 r = packuswb(r) * a >> 8; | 91 r = packuswb(r) * a >> 8; |
| 94 | 92 |
| 95 *reinterpret_cast<uint32*>(rgb_buf) = (b << SK_B32_SHIFT) | | 93 *reinterpret_cast<uint32*>(rgb_buf) = (b << SK_B32_SHIFT) | |
| 96 (g << SK_G32_SHIFT) | | 94 (g << SK_G32_SHIFT) | |
| 97 (r << SK_R32_SHIFT) | | 95 (r << SK_R32_SHIFT) | |
| 98 (a << SK_A32_SHIFT); | 96 (a << SK_A32_SHIFT); |
| 99 } | 97 } |
| 100 | 98 |
| 101 void ConvertYUVToRGB32Row_C(const uint8* y_buf, | 99 void ConvertYUVToRGB32Row_C(const uint8* y_buf, |
| 102 const uint8* u_buf, | 100 const uint8* u_buf, |
| 103 const uint8* v_buf, | 101 const uint8* v_buf, |
| 104 uint8* rgb_buf, | 102 uint8* rgb_buf, |
| 105 ptrdiff_t width, | 103 ptrdiff_t width) { |
| 106 const int16 convert_table[1024][4]) { | |
| 107 for (int x = 0; x < width; x += 2) { | 104 for (int x = 0; x < width; x += 2) { |
| 108 uint8 u = u_buf[x >> 1]; | 105 uint8 u = u_buf[x >> 1]; |
| 109 uint8 v = v_buf[x >> 1]; | 106 uint8 v = v_buf[x >> 1]; |
| 110 uint8 y0 = y_buf[x]; | 107 uint8 y0 = y_buf[x]; |
| 111 ConvertYUVToRGB32_C(y0, u, v, rgb_buf, convert_table); | 108 ConvertYUVToRGB32_C(y0, u, v, rgb_buf); |
| 112 if ((x + 1) < width) { | 109 if ((x + 1) < width) { |
| 113 uint8 y1 = y_buf[x + 1]; | 110 uint8 y1 = y_buf[x + 1]; |
| 114 ConvertYUVToRGB32_C(y1, u, v, rgb_buf + 4, convert_table); | 111 ConvertYUVToRGB32_C(y1, u, v, rgb_buf + 4); |
| 115 } | 112 } |
| 116 rgb_buf += 8; // Advance 2 pixels. | 113 rgb_buf += 8; // Advance 2 pixels. |
| 117 } | 114 } |
| 118 } | 115 } |
| 119 | 116 |
| 120 void ConvertYUVAToARGBRow_C(const uint8* y_buf, | 117 void ConvertYUVAToARGBRow_C(const uint8* y_buf, |
| 121 const uint8* u_buf, | 118 const uint8* u_buf, |
| 122 const uint8* v_buf, | 119 const uint8* v_buf, |
| 123 const uint8* a_buf, | 120 const uint8* a_buf, |
| 124 uint8* rgba_buf, | 121 uint8* rgba_buf, |
| 125 ptrdiff_t width, | 122 ptrdiff_t width) { |
| 126 const int16 convert_table[1024][4]) { | |
| 127 for (int x = 0; x < width; x += 2) { | 123 for (int x = 0; x < width; x += 2) { |
| 128 uint8 u = u_buf[x >> 1]; | 124 uint8 u = u_buf[x >> 1]; |
| 129 uint8 v = v_buf[x >> 1]; | 125 uint8 v = v_buf[x >> 1]; |
| 130 uint8 y0 = y_buf[x]; | 126 uint8 y0 = y_buf[x]; |
| 131 uint8 a0 = a_buf[x]; | 127 uint8 a0 = a_buf[x]; |
| 132 ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf, convert_table); | 128 ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf); |
| 133 if ((x + 1) < width) { | 129 if ((x + 1) < width) { |
| 134 uint8 y1 = y_buf[x + 1]; | 130 uint8 y1 = y_buf[x + 1]; |
| 135 uint8 a1 = a_buf[x + 1]; | 131 uint8 a1 = a_buf[x + 1]; |
| 136 ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4, convert_table); | 132 ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4); |
| 137 } | 133 } |
| 138 rgba_buf += 8; // Advance 2 pixels. | 134 rgba_buf += 8; // Advance 2 pixels. |
| 139 } | 135 } |
| 140 } | 136 } |
| 141 | 137 |
| 142 // 16.16 fixed point is used. A shift by 16 isolates the integer. | 138 // 16.16 fixed point is used. A shift by 16 isolates the integer. |
| 143 // A shift by 17 is used to further subsample the chrominence channels. | 139 // A shift by 17 is used to further subsample the chrominence channels. |
| 144 // & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, | 140 // & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, |
| 145 // for 1/65536 pixel accurate interpolation. | 141 // for 1/65536 pixel accurate interpolation. |
| 146 void ScaleYUVToRGB32Row_C(const uint8* y_buf, | 142 void ScaleYUVToRGB32Row_C(const uint8* y_buf, |
| 147 const uint8* u_buf, | 143 const uint8* u_buf, |
| 148 const uint8* v_buf, | 144 const uint8* v_buf, |
| 149 uint8* rgb_buf, | 145 uint8* rgb_buf, |
| 150 ptrdiff_t width, | 146 ptrdiff_t width, |
| 151 ptrdiff_t source_dx, | 147 ptrdiff_t source_dx) { |
| 152 const int16 convert_table[1024][4]) { | |
| 153 int x = 0; | 148 int x = 0; |
| 154 for (int i = 0; i < width; i += 2) { | 149 for (int i = 0; i < width; i += 2) { |
| 155 int y = y_buf[x >> 16]; | 150 int y = y_buf[x >> 16]; |
| 156 int u = u_buf[(x >> 17)]; | 151 int u = u_buf[(x >> 17)]; |
| 157 int v = v_buf[(x >> 17)]; | 152 int v = v_buf[(x >> 17)]; |
| 158 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); | 153 ConvertYUVToRGB32_C(y, u, v, rgb_buf); |
| 159 x += source_dx; | 154 x += source_dx; |
| 160 if ((i + 1) < width) { | 155 if ((i + 1) < width) { |
| 161 y = y_buf[x >> 16]; | 156 y = y_buf[x >> 16]; |
| 162 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); | 157 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4); |
| 163 x += source_dx; | 158 x += source_dx; |
| 164 } | 159 } |
| 165 rgb_buf += 8; | 160 rgb_buf += 8; |
| 166 } | 161 } |
| 167 } | 162 } |
| 168 | 163 |
| 169 void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, | 164 void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, |
| 170 const uint8* u_buf, | 165 const uint8* u_buf, |
| 171 const uint8* v_buf, | 166 const uint8* v_buf, |
| 172 uint8* rgb_buf, | 167 uint8* rgb_buf, |
| 173 ptrdiff_t width, | 168 ptrdiff_t width, |
| 174 ptrdiff_t source_dx, | 169 ptrdiff_t source_dx) { |
| 175 const int16 convert_table[1024][4]) { | |
| 176 // Avoid point-sampling for down-scaling by > 2:1. | 170 // Avoid point-sampling for down-scaling by > 2:1. |
| 177 int source_x = 0; | 171 int source_x = 0; |
| 178 if (source_dx >= 0x20000) | 172 if (source_dx >= 0x20000) |
| 179 source_x += 0x8000; | 173 source_x += 0x8000; |
| 180 LinearScaleYUVToRGB32RowWithRange_C(y_buf, u_buf, v_buf, rgb_buf, width, | 174 LinearScaleYUVToRGB32RowWithRange_C(y_buf, u_buf, v_buf, rgb_buf, width, |
| 181 source_x, source_dx, convert_table); | 175 source_x, source_dx); |
| 182 } | 176 } |
| 183 | 177 |
| 184 void LinearScaleYUVToRGB32RowWithRange_C(const uint8* y_buf, | 178 void LinearScaleYUVToRGB32RowWithRange_C(const uint8* y_buf, |
| 185 const uint8* u_buf, | 179 const uint8* u_buf, |
| 186 const uint8* v_buf, | 180 const uint8* v_buf, |
| 187 uint8* rgb_buf, | 181 uint8* rgb_buf, |
| 188 int dest_width, | 182 int dest_width, |
| 189 int x, | 183 int x, |
| 190 int source_dx, | 184 int source_dx) { |
| 191 const int16 convert_table[1024][4]) { | |
| 192 for (int i = 0; i < dest_width; i += 2) { | 185 for (int i = 0; i < dest_width; i += 2) { |
| 193 int y0 = y_buf[x >> 16]; | 186 int y0 = y_buf[x >> 16]; |
| 194 int y1 = y_buf[(x >> 16) + 1]; | 187 int y1 = y_buf[(x >> 16) + 1]; |
| 195 int u0 = u_buf[(x >> 17)]; | 188 int u0 = u_buf[(x >> 17)]; |
| 196 int u1 = u_buf[(x >> 17) + 1]; | 189 int u1 = u_buf[(x >> 17) + 1]; |
| 197 int v0 = v_buf[(x >> 17)]; | 190 int v0 = v_buf[(x >> 17)]; |
| 198 int v1 = v_buf[(x >> 17) + 1]; | 191 int v1 = v_buf[(x >> 17) + 1]; |
| 199 int y_frac = (x & 65535); | 192 int y_frac = (x & 65535); |
| 200 int uv_frac = ((x >> 1) & 65535); | 193 int uv_frac = ((x >> 1) & 65535); |
| 201 int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; | 194 int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; |
| 202 int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; | 195 int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; |
| 203 int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; | 196 int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; |
| 204 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); | 197 ConvertYUVToRGB32_C(y, u, v, rgb_buf); |
| 205 x += source_dx; | 198 x += source_dx; |
| 206 if ((i + 1) < dest_width) { | 199 if ((i + 1) < dest_width) { |
| 207 y0 = y_buf[x >> 16]; | 200 y0 = y_buf[x >> 16]; |
| 208 y1 = y_buf[(x >> 16) + 1]; | 201 y1 = y_buf[(x >> 16) + 1]; |
| 209 y_frac = (x & 65535); | 202 y_frac = (x & 65535); |
| 210 y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; | 203 y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; |
| 211 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); | 204 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4); |
| 212 x += source_dx; | 205 x += source_dx; |
| 213 } | 206 } |
| 214 rgb_buf += 8; | 207 rgb_buf += 8; |
| 215 } | 208 } |
| 216 } | 209 } |
| 217 | 210 |
| 218 void ConvertYUVToRGB32_C(const uint8* yplane, | 211 void ConvertYUVToRGB32_C(const uint8* yplane, |
| 219 const uint8* uplane, | 212 const uint8* uplane, |
| 220 const uint8* vplane, | 213 const uint8* vplane, |
| 221 uint8* rgbframe, | 214 uint8* rgbframe, |
| 222 int width, | 215 int width, |
| 223 int height, | 216 int height, |
| 224 int ystride, | 217 int ystride, |
| 225 int uvstride, | 218 int uvstride, |
| 226 int rgbstride, | 219 int rgbstride, |
| 227 YUVType yuv_type) { | 220 YUVType yuv_type) { |
| 228 unsigned int y_shift = GetVerticalShift(yuv_type); | 221 unsigned int y_shift = yuv_type; |
| 229 for (int y = 0; y < height; ++y) { | 222 for (int y = 0; y < height; ++y) { |
| 230 uint8* rgb_row = rgbframe + y * rgbstride; | 223 uint8* rgb_row = rgbframe + y * rgbstride; |
| 231 const uint8* y_ptr = yplane + y * ystride; | 224 const uint8* y_ptr = yplane + y * ystride; |
| 232 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; | 225 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; |
| 233 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; | 226 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; |
| 234 | 227 |
| 235 ConvertYUVToRGB32Row_C(y_ptr, | 228 ConvertYUVToRGB32Row_C(y_ptr, |
| 236 u_ptr, | 229 u_ptr, |
| 237 v_ptr, | 230 v_ptr, |
| 238 rgb_row, | 231 rgb_row, |
| 239 width, | 232 width); |
| 240 GetLookupTable(yuv_type)); | |
| 241 } | 233 } |
| 242 } | 234 } |
| 243 | 235 |
| 244 void ConvertYUVAToARGB_C(const uint8* yplane, | 236 void ConvertYUVAToARGB_C(const uint8* yplane, |
| 245 const uint8* uplane, | 237 const uint8* uplane, |
| 246 const uint8* vplane, | 238 const uint8* vplane, |
| 247 const uint8* aplane, | 239 const uint8* aplane, |
| 248 uint8* rgbaframe, | 240 uint8* rgbaframe, |
| 249 int width, | 241 int width, |
| 250 int height, | 242 int height, |
| 251 int ystride, | 243 int ystride, |
| 252 int uvstride, | 244 int uvstride, |
| 253 int astride, | 245 int astride, |
| 254 int rgbastride, | 246 int rgbastride, |
| 255 YUVType yuv_type) { | 247 YUVType yuv_type) { |
| 256 unsigned int y_shift = yuv_type; | 248 unsigned int y_shift = yuv_type; |
| 257 for (int y = 0; y < height; y++) { | 249 for (int y = 0; y < height; y++) { |
| 258 uint8* rgba_row = rgbaframe + y * rgbastride; | 250 uint8* rgba_row = rgbaframe + y * rgbastride; |
| 259 const uint8* y_ptr = yplane + y * ystride; | 251 const uint8* y_ptr = yplane + y * ystride; |
| 260 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; | 252 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; |
| 261 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; | 253 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; |
| 262 const uint8* a_ptr = aplane + y * astride; | 254 const uint8* a_ptr = aplane + y * astride; |
| 263 | 255 |
| 264 ConvertYUVAToARGBRow_C(y_ptr, | 256 ConvertYUVAToARGBRow_C(y_ptr, |
| 265 u_ptr, | 257 u_ptr, |
| 266 v_ptr, | 258 v_ptr, |
| 267 a_ptr, | 259 a_ptr, |
| 268 rgba_row, | 260 rgba_row, |
| 269 width, | 261 width); |
| 270 GetLookupTable(yuv_type)); | |
| 271 } | 262 } |
| 272 } | 263 } |
| 273 | 264 |
| 274 } // namespace media | 265 } // namespace media |
| OLD | NEW |