| 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 | 6 |
| 7 namespace media { | 7 namespace media { |
| 8 | 8 |
| 9 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) | 9 #define packuswb(x) ((x) < 0 ? 0 : ((x) > 255 ? 255 : (x))) |
| 10 #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ | 10 #define paddsw(x, y) (((x) + (y)) < -32768 ? -32768 : \ |
| (...skipping 16 matching lines...) Expand all Loading... |
| 27 #define SK_B32_SHIFT 0 | 27 #define SK_B32_SHIFT 0 |
| 28 #define SK_G32_SHIFT 8 | 28 #define SK_G32_SHIFT 8 |
| 29 #define SK_R32_SHIFT 16 | 29 #define SK_R32_SHIFT 16 |
| 30 #define SK_A32_SHIFT 24 | 30 #define SK_A32_SHIFT 24 |
| 31 #define B_INDEX 0 | 31 #define B_INDEX 0 |
| 32 #define G_INDEX 1 | 32 #define G_INDEX 1 |
| 33 #define R_INDEX 2 | 33 #define R_INDEX 2 |
| 34 #define A_INDEX 3 | 34 #define A_INDEX 3 |
| 35 #endif | 35 #endif |
| 36 | 36 |
| 37 static inline void ConvertYUVToRGB32_C(uint8 y, | 37 static inline void ConvertYUVToRGB32_C(uint8_t y, |
| 38 uint8 u, | 38 uint8_t u, |
| 39 uint8 v, | 39 uint8_t v, |
| 40 uint8* rgb_buf, | 40 uint8_t* rgb_buf, |
| 41 const int16* convert_table) { | 41 const int16_t* convert_table) { |
| 42 int b = convert_table[4 * (256 + u) + B_INDEX]; | 42 int b = convert_table[4 * (256 + u) + B_INDEX]; |
| 43 int g = convert_table[4 * (256 + u) + G_INDEX]; | 43 int g = convert_table[4 * (256 + u) + G_INDEX]; |
| 44 int r = convert_table[4 * (256 + u) + R_INDEX]; | 44 int r = convert_table[4 * (256 + u) + R_INDEX]; |
| 45 int a = convert_table[4 * (256 + u) + A_INDEX]; | 45 int a = convert_table[4 * (256 + u) + A_INDEX]; |
| 46 | 46 |
| 47 b = paddsw(b, convert_table[4 * (512 + v) + B_INDEX]); | 47 b = paddsw(b, convert_table[4 * (512 + v) + B_INDEX]); |
| 48 g = paddsw(g, convert_table[4 * (512 + v) + G_INDEX]); | 48 g = paddsw(g, convert_table[4 * (512 + v) + G_INDEX]); |
| 49 r = paddsw(r, convert_table[4 * (512 + v) + R_INDEX]); | 49 r = paddsw(r, convert_table[4 * (512 + v) + R_INDEX]); |
| 50 a = paddsw(a, convert_table[4 * (512 + v) + A_INDEX]); | 50 a = paddsw(a, convert_table[4 * (512 + v) + A_INDEX]); |
| 51 | 51 |
| 52 b = paddsw(b, convert_table[4 * y + B_INDEX]); | 52 b = paddsw(b, convert_table[4 * y + B_INDEX]); |
| 53 g = paddsw(g, convert_table[4 * y + G_INDEX]); | 53 g = paddsw(g, convert_table[4 * y + G_INDEX]); |
| 54 r = paddsw(r, convert_table[4 * y + R_INDEX]); | 54 r = paddsw(r, convert_table[4 * y + R_INDEX]); |
| 55 a = paddsw(a, convert_table[4 * y + A_INDEX]); | 55 a = paddsw(a, convert_table[4 * y + A_INDEX]); |
| 56 | 56 |
| 57 b >>= 6; | 57 b >>= 6; |
| 58 g >>= 6; | 58 g >>= 6; |
| 59 r >>= 6; | 59 r >>= 6; |
| 60 a >>= 6; | 60 a >>= 6; |
| 61 | 61 |
| 62 *reinterpret_cast<uint32*>(rgb_buf) = (packuswb(b) << SK_B32_SHIFT) | | 62 *reinterpret_cast<uint32_t*>(rgb_buf) = |
| 63 (packuswb(g) << SK_G32_SHIFT) | | 63 (packuswb(b) << SK_B32_SHIFT) | (packuswb(g) << SK_G32_SHIFT) | |
| 64 (packuswb(r) << SK_R32_SHIFT) | | 64 (packuswb(r) << SK_R32_SHIFT) | (packuswb(a) << SK_A32_SHIFT); |
| 65 (packuswb(a) << SK_A32_SHIFT); | |
| 66 } | 65 } |
| 67 | 66 |
| 68 static inline void ConvertYUVAToARGB_C(uint8 y, | 67 static inline void ConvertYUVAToARGB_C(uint8_t y, |
| 69 uint8 u, | 68 uint8_t u, |
| 70 uint8 v, | 69 uint8_t v, |
| 71 uint8 a, | 70 uint8_t a, |
| 72 uint8* rgb_buf, | 71 uint8_t* rgb_buf, |
| 73 const int16* convert_table) { | 72 const int16_t* convert_table) { |
| 74 int b = convert_table[4 * (256 + u) + 0]; | 73 int b = convert_table[4 * (256 + u) + 0]; |
| 75 int g = convert_table[4 * (256 + u) + 1]; | 74 int g = convert_table[4 * (256 + u) + 1]; |
| 76 int r = convert_table[4 * (256 + u) + 2]; | 75 int r = convert_table[4 * (256 + u) + 2]; |
| 77 | 76 |
| 78 b = paddsw(b, convert_table[4 * (512 + v) + 0]); | 77 b = paddsw(b, convert_table[4 * (512 + v) + 0]); |
| 79 g = paddsw(g, convert_table[4 * (512 + v) + 1]); | 78 g = paddsw(g, convert_table[4 * (512 + v) + 1]); |
| 80 r = paddsw(r, convert_table[4 * (512 + v) + 2]); | 79 r = paddsw(r, convert_table[4 * (512 + v) + 2]); |
| 81 | 80 |
| 82 b = paddsw(b, convert_table[4 * y + 0]); | 81 b = paddsw(b, convert_table[4 * y + 0]); |
| 83 g = paddsw(g, convert_table[4 * y + 1]); | 82 g = paddsw(g, convert_table[4 * y + 1]); |
| 84 r = paddsw(r, convert_table[4 * y + 2]); | 83 r = paddsw(r, convert_table[4 * y + 2]); |
| 85 | 84 |
| 86 b >>= 6; | 85 b >>= 6; |
| 87 g >>= 6; | 86 g >>= 6; |
| 88 r >>= 6; | 87 r >>= 6; |
| 89 | 88 |
| 90 b = packuswb(b) * a >> 8; | 89 b = packuswb(b) * a >> 8; |
| 91 g = packuswb(g) * a >> 8; | 90 g = packuswb(g) * a >> 8; |
| 92 r = packuswb(r) * a >> 8; | 91 r = packuswb(r) * a >> 8; |
| 93 | 92 |
| 94 *reinterpret_cast<uint32*>(rgb_buf) = (b << SK_B32_SHIFT) | | 93 *reinterpret_cast<uint32_t*>(rgb_buf) = |
| 95 (g << SK_G32_SHIFT) | | 94 (b << SK_B32_SHIFT) | (g << SK_G32_SHIFT) | (r << SK_R32_SHIFT) | |
| 96 (r << SK_R32_SHIFT) | | 95 (a << SK_A32_SHIFT); |
| 97 (a << SK_A32_SHIFT); | |
| 98 } | 96 } |
| 99 | 97 |
| 100 void ConvertYUVToRGB32Row_C(const uint8* y_buf, | 98 void ConvertYUVToRGB32Row_C(const uint8_t* y_buf, |
| 101 const uint8* u_buf, | 99 const uint8_t* u_buf, |
| 102 const uint8* v_buf, | 100 const uint8_t* v_buf, |
| 103 uint8* rgb_buf, | 101 uint8_t* rgb_buf, |
| 104 ptrdiff_t width, | 102 ptrdiff_t width, |
| 105 const int16* convert_table) { | 103 const int16_t* convert_table) { |
| 106 for (int x = 0; x < width; x += 2) { | 104 for (int x = 0; x < width; x += 2) { |
| 107 uint8 u = u_buf[x >> 1]; | 105 uint8_t u = u_buf[x >> 1]; |
| 108 uint8 v = v_buf[x >> 1]; | 106 uint8_t v = v_buf[x >> 1]; |
| 109 uint8 y0 = y_buf[x]; | 107 uint8_t y0 = y_buf[x]; |
| 110 ConvertYUVToRGB32_C(y0, u, v, rgb_buf, convert_table); | 108 ConvertYUVToRGB32_C(y0, u, v, rgb_buf, convert_table); |
| 111 if ((x + 1) < width) { | 109 if ((x + 1) < width) { |
| 112 uint8 y1 = y_buf[x + 1]; | 110 uint8_t y1 = y_buf[x + 1]; |
| 113 ConvertYUVToRGB32_C(y1, u, v, rgb_buf + 4, convert_table); | 111 ConvertYUVToRGB32_C(y1, u, v, rgb_buf + 4, convert_table); |
| 114 } | 112 } |
| 115 rgb_buf += 8; // Advance 2 pixels. | 113 rgb_buf += 8; // Advance 2 pixels. |
| 116 } | 114 } |
| 117 } | 115 } |
| 118 | 116 |
| 119 void ConvertYUVAToARGBRow_C(const uint8* y_buf, | 117 void ConvertYUVAToARGBRow_C(const uint8_t* y_buf, |
| 120 const uint8* u_buf, | 118 const uint8_t* u_buf, |
| 121 const uint8* v_buf, | 119 const uint8_t* v_buf, |
| 122 const uint8* a_buf, | 120 const uint8_t* a_buf, |
| 123 uint8* rgba_buf, | 121 uint8_t* rgba_buf, |
| 124 ptrdiff_t width, | 122 ptrdiff_t width, |
| 125 const int16* convert_table) { | 123 const int16_t* convert_table) { |
| 126 for (int x = 0; x < width; x += 2) { | 124 for (int x = 0; x < width; x += 2) { |
| 127 uint8 u = u_buf[x >> 1]; | 125 uint8_t u = u_buf[x >> 1]; |
| 128 uint8 v = v_buf[x >> 1]; | 126 uint8_t v = v_buf[x >> 1]; |
| 129 uint8 y0 = y_buf[x]; | 127 uint8_t y0 = y_buf[x]; |
| 130 uint8 a0 = a_buf[x]; | 128 uint8_t a0 = a_buf[x]; |
| 131 ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf, convert_table); | 129 ConvertYUVAToARGB_C(y0, u, v, a0, rgba_buf, convert_table); |
| 132 if ((x + 1) < width) { | 130 if ((x + 1) < width) { |
| 133 uint8 y1 = y_buf[x + 1]; | 131 uint8_t y1 = y_buf[x + 1]; |
| 134 uint8 a1 = a_buf[x + 1]; | 132 uint8_t a1 = a_buf[x + 1]; |
| 135 ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4, convert_table); | 133 ConvertYUVAToARGB_C(y1, u, v, a1, rgba_buf + 4, convert_table); |
| 136 } | 134 } |
| 137 rgba_buf += 8; // Advance 2 pixels. | 135 rgba_buf += 8; // Advance 2 pixels. |
| 138 } | 136 } |
| 139 } | 137 } |
| 140 | 138 |
| 141 // 16.16 fixed point is used. A shift by 16 isolates the integer. | 139 // 16.16 fixed point is used. A shift by 16 isolates the integer. |
| 142 // A shift by 17 is used to further subsample the chrominence channels. | 140 // A shift by 17 is used to further subsample the chrominence channels. |
| 143 // & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, | 141 // & 0xffff isolates the fixed point fraction. >> 2 to get the upper 2 bits, |
| 144 // for 1/65536 pixel accurate interpolation. | 142 // for 1/65536 pixel accurate interpolation. |
| 145 void ScaleYUVToRGB32Row_C(const uint8* y_buf, | 143 void ScaleYUVToRGB32Row_C(const uint8_t* y_buf, |
| 146 const uint8* u_buf, | 144 const uint8_t* u_buf, |
| 147 const uint8* v_buf, | 145 const uint8_t* v_buf, |
| 148 uint8* rgb_buf, | 146 uint8_t* rgb_buf, |
| 149 ptrdiff_t width, | 147 ptrdiff_t width, |
| 150 ptrdiff_t source_dx, | 148 ptrdiff_t source_dx, |
| 151 const int16* convert_table) { | 149 const int16_t* convert_table) { |
| 152 int x = 0; | 150 int x = 0; |
| 153 for (int i = 0; i < width; i += 2) { | 151 for (int i = 0; i < width; i += 2) { |
| 154 int y = y_buf[x >> 16]; | 152 int y = y_buf[x >> 16]; |
| 155 int u = u_buf[(x >> 17)]; | 153 int u = u_buf[(x >> 17)]; |
| 156 int v = v_buf[(x >> 17)]; | 154 int v = v_buf[(x >> 17)]; |
| 157 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); | 155 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); |
| 158 x += source_dx; | 156 x += source_dx; |
| 159 if ((i + 1) < width) { | 157 if ((i + 1) < width) { |
| 160 y = y_buf[x >> 16]; | 158 y = y_buf[x >> 16]; |
| 161 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); | 159 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); |
| 162 x += source_dx; | 160 x += source_dx; |
| 163 } | 161 } |
| 164 rgb_buf += 8; | 162 rgb_buf += 8; |
| 165 } | 163 } |
| 166 } | 164 } |
| 167 | 165 |
| 168 void LinearScaleYUVToRGB32Row_C(const uint8* y_buf, | 166 void LinearScaleYUVToRGB32Row_C(const uint8_t* y_buf, |
| 169 const uint8* u_buf, | 167 const uint8_t* u_buf, |
| 170 const uint8* v_buf, | 168 const uint8_t* v_buf, |
| 171 uint8* rgb_buf, | 169 uint8_t* rgb_buf, |
| 172 ptrdiff_t width, | 170 ptrdiff_t width, |
| 173 ptrdiff_t source_dx, | 171 ptrdiff_t source_dx, |
| 174 const int16* convert_table) { | 172 const int16_t* convert_table) { |
| 175 // Avoid point-sampling for down-scaling by > 2:1. | 173 // Avoid point-sampling for down-scaling by > 2:1. |
| 176 int source_x = 0; | 174 int source_x = 0; |
| 177 if (source_dx >= 0x20000) | 175 if (source_dx >= 0x20000) |
| 178 source_x += 0x8000; | 176 source_x += 0x8000; |
| 179 LinearScaleYUVToRGB32RowWithRange_C(y_buf, u_buf, v_buf, rgb_buf, width, | 177 LinearScaleYUVToRGB32RowWithRange_C(y_buf, u_buf, v_buf, rgb_buf, width, |
| 180 source_x, source_dx, convert_table); | 178 source_x, source_dx, convert_table); |
| 181 } | 179 } |
| 182 | 180 |
| 183 void LinearScaleYUVToRGB32RowWithRange_C(const uint8* y_buf, | 181 void LinearScaleYUVToRGB32RowWithRange_C(const uint8_t* y_buf, |
| 184 const uint8* u_buf, | 182 const uint8_t* u_buf, |
| 185 const uint8* v_buf, | 183 const uint8_t* v_buf, |
| 186 uint8* rgb_buf, | 184 uint8_t* rgb_buf, |
| 187 int dest_width, | 185 int dest_width, |
| 188 int x, | 186 int x, |
| 189 int source_dx, | 187 int source_dx, |
| 190 const int16* convert_table) { | 188 const int16_t* convert_table) { |
| 191 for (int i = 0; i < dest_width; i += 2) { | 189 for (int i = 0; i < dest_width; i += 2) { |
| 192 int y0 = y_buf[x >> 16]; | 190 int y0 = y_buf[x >> 16]; |
| 193 int y1 = y_buf[(x >> 16) + 1]; | 191 int y1 = y_buf[(x >> 16) + 1]; |
| 194 int u0 = u_buf[(x >> 17)]; | 192 int u0 = u_buf[(x >> 17)]; |
| 195 int u1 = u_buf[(x >> 17) + 1]; | 193 int u1 = u_buf[(x >> 17) + 1]; |
| 196 int v0 = v_buf[(x >> 17)]; | 194 int v0 = v_buf[(x >> 17)]; |
| 197 int v1 = v_buf[(x >> 17) + 1]; | 195 int v1 = v_buf[(x >> 17) + 1]; |
| 198 int y_frac = (x & 65535); | 196 int y_frac = (x & 65535); |
| 199 int uv_frac = ((x >> 1) & 65535); | 197 int uv_frac = ((x >> 1) & 65535); |
| 200 int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; | 198 int y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; |
| 201 int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; | 199 int u = (uv_frac * u1 + (uv_frac ^ 65535) * u0) >> 16; |
| 202 int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; | 200 int v = (uv_frac * v1 + (uv_frac ^ 65535) * v0) >> 16; |
| 203 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); | 201 ConvertYUVToRGB32_C(y, u, v, rgb_buf, convert_table); |
| 204 x += source_dx; | 202 x += source_dx; |
| 205 if ((i + 1) < dest_width) { | 203 if ((i + 1) < dest_width) { |
| 206 y0 = y_buf[x >> 16]; | 204 y0 = y_buf[x >> 16]; |
| 207 y1 = y_buf[(x >> 16) + 1]; | 205 y1 = y_buf[(x >> 16) + 1]; |
| 208 y_frac = (x & 65535); | 206 y_frac = (x & 65535); |
| 209 y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; | 207 y = (y_frac * y1 + (y_frac ^ 65535) * y0) >> 16; |
| 210 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); | 208 ConvertYUVToRGB32_C(y, u, v, rgb_buf+4, convert_table); |
| 211 x += source_dx; | 209 x += source_dx; |
| 212 } | 210 } |
| 213 rgb_buf += 8; | 211 rgb_buf += 8; |
| 214 } | 212 } |
| 215 } | 213 } |
| 216 | 214 |
| 217 void ConvertYUVToRGB32_C(const uint8* yplane, | 215 void ConvertYUVToRGB32_C(const uint8_t* yplane, |
| 218 const uint8* uplane, | 216 const uint8_t* uplane, |
| 219 const uint8* vplane, | 217 const uint8_t* vplane, |
| 220 uint8* rgbframe, | 218 uint8_t* rgbframe, |
| 221 int width, | 219 int width, |
| 222 int height, | 220 int height, |
| 223 int ystride, | 221 int ystride, |
| 224 int uvstride, | 222 int uvstride, |
| 225 int rgbstride, | 223 int rgbstride, |
| 226 YUVType yuv_type) { | 224 YUVType yuv_type) { |
| 227 unsigned int y_shift = GetVerticalShift(yuv_type); | 225 unsigned int y_shift = GetVerticalShift(yuv_type); |
| 228 const int16* lookup_table = GetLookupTable(yuv_type); | 226 const int16_t* lookup_table = GetLookupTable(yuv_type); |
| 229 for (int y = 0; y < height; ++y) { | 227 for (int y = 0; y < height; ++y) { |
| 230 uint8* rgb_row = rgbframe + y * rgbstride; | 228 uint8_t* rgb_row = rgbframe + y * rgbstride; |
| 231 const uint8* y_ptr = yplane + y * ystride; | 229 const uint8_t* y_ptr = yplane + y * ystride; |
| 232 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; | 230 const uint8_t* u_ptr = uplane + (y >> y_shift) * uvstride; |
| 233 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; | 231 const uint8_t* v_ptr = vplane + (y >> y_shift) * uvstride; |
| 234 | 232 |
| 235 ConvertYUVToRGB32Row_C(y_ptr, | 233 ConvertYUVToRGB32Row_C(y_ptr, |
| 236 u_ptr, | 234 u_ptr, |
| 237 v_ptr, | 235 v_ptr, |
| 238 rgb_row, | 236 rgb_row, |
| 239 width, | 237 width, |
| 240 lookup_table); | 238 lookup_table); |
| 241 } | 239 } |
| 242 } | 240 } |
| 243 | 241 |
| 244 void ConvertYUVAToARGB_C(const uint8* yplane, | 242 void ConvertYUVAToARGB_C(const uint8_t* yplane, |
| 245 const uint8* uplane, | 243 const uint8_t* uplane, |
| 246 const uint8* vplane, | 244 const uint8_t* vplane, |
| 247 const uint8* aplane, | 245 const uint8_t* aplane, |
| 248 uint8* rgbaframe, | 246 uint8_t* rgbaframe, |
| 249 int width, | 247 int width, |
| 250 int height, | 248 int height, |
| 251 int ystride, | 249 int ystride, |
| 252 int uvstride, | 250 int uvstride, |
| 253 int astride, | 251 int astride, |
| 254 int rgbastride, | 252 int rgbastride, |
| 255 YUVType yuv_type) { | 253 YUVType yuv_type) { |
| 256 unsigned int y_shift = GetVerticalShift(yuv_type); | 254 unsigned int y_shift = GetVerticalShift(yuv_type); |
| 257 const int16* lookup_table = GetLookupTable(yuv_type); | 255 const int16_t* lookup_table = GetLookupTable(yuv_type); |
| 258 for (int y = 0; y < height; y++) { | 256 for (int y = 0; y < height; y++) { |
| 259 uint8* rgba_row = rgbaframe + y * rgbastride; | 257 uint8_t* rgba_row = rgbaframe + y * rgbastride; |
| 260 const uint8* y_ptr = yplane + y * ystride; | 258 const uint8_t* y_ptr = yplane + y * ystride; |
| 261 const uint8* u_ptr = uplane + (y >> y_shift) * uvstride; | 259 const uint8_t* u_ptr = uplane + (y >> y_shift) * uvstride; |
| 262 const uint8* v_ptr = vplane + (y >> y_shift) * uvstride; | 260 const uint8_t* v_ptr = vplane + (y >> y_shift) * uvstride; |
| 263 const uint8* a_ptr = aplane + y * astride; | 261 const uint8_t* a_ptr = aplane + y * astride; |
| 264 | 262 |
| 265 ConvertYUVAToARGBRow_C(y_ptr, | 263 ConvertYUVAToARGBRow_C(y_ptr, |
| 266 u_ptr, | 264 u_ptr, |
| 267 v_ptr, | 265 v_ptr, |
| 268 a_ptr, | 266 a_ptr, |
| 269 rgba_row, | 267 rgba_row, |
| 270 width, | 268 width, |
| 271 lookup_table); | 269 lookup_table); |
| 272 } | 270 } |
| 273 } | 271 } |
| 274 | 272 |
| 275 } // namespace media | 273 } // namespace media |
| OLD | NEW |