yuv.h (8467B)
1 // Copyright 2010 Google Inc. All Rights Reserved. 2 // 3 // Use of this source code is governed by a BSD-style license 4 // that can be found in the COPYING file in the root of the source 5 // tree. An additional intellectual property rights grant can be found 6 // in the file PATENTS. All contributing project authors may 7 // be found in the AUTHORS file in the root of the source tree. 8 // ----------------------------------------------------------------------------- 9 // 10 // inline YUV<->RGB conversion function 11 // 12 // The exact naming is Y'CbCr, following the ITU-R BT.601 standard. 13 // More information at: https://en.wikipedia.org/wiki/YCbCr 14 // Y = 0.2568 * R + 0.5041 * G + 0.0979 * B + 16 15 // U = -0.1482 * R - 0.2910 * G + 0.4392 * B + 128 16 // V = 0.4392 * R - 0.3678 * G - 0.0714 * B + 128 17 // We use 16bit fixed point operations for RGB->YUV conversion (YUV_FIX). 18 // 19 // For the Y'CbCr to RGB conversion, the BT.601 specification reads: 20 // R = 1.164 * (Y-16) + 1.596 * (V-128) 21 // G = 1.164 * (Y-16) - 0.813 * (V-128) - 0.392 * (U-128) 22 // B = 1.164 * (Y-16) + 2.017 * (U-128) 23 // where Y is in the [16,235] range, and U/V in the [16,240] range. 24 // 25 // The fixed-point implementation used here is: 26 // R = (19077 . y + 26149 . v - 14234) >> 6 27 // G = (19077 . y - 6419 . u - 13320 . v + 8708) >> 6 28 // B = (19077 . y + 33050 . u - 17685) >> 6 29 // where the '.' operator is the mulhi_epu16 variant: 30 // a . b = ((a << 8) * b) >> 16 31 // that preserves 8 bits of fractional precision before final descaling. 32 33 // Author: Skal (pascal.massimino@gmail.com) 34 35 #ifndef WEBP_DSP_YUV_H_ 36 #define WEBP_DSP_YUV_H_ 37 38 #include "src/dec/vp8_dec.h" 39 #include "src/dsp/cpu.h" 40 #include "src/dsp/dsp.h" 41 #include "src/webp/types.h" 42 43 //------------------------------------------------------------------------------ 44 // YUV -> RGB conversion 45 46 #ifdef __cplusplus 47 extern "C" { 48 #endif 49 50 enum { 51 YUV_FIX = 16, // fixed-point precision for RGB->YUV 52 YUV_HALF = 1 << (YUV_FIX - 1), 53 54 YUV_FIX2 = 6, // fixed-point precision for YUV->RGB 55 YUV_MASK2 = (256 << YUV_FIX2) - 1 56 }; 57 58 //------------------------------------------------------------------------------ 59 // slower on x86 by ~7-8%, but bit-exact with the SSE2/NEON version 60 61 static WEBP_INLINE int MultHi(int v, int coeff) { // _mm_mulhi_epu16 emulation 62 return (v * coeff) >> 8; 63 } 64 65 static WEBP_INLINE int VP8Clip8(int v) { 66 return ((v & ~YUV_MASK2) == 0) ? (v >> YUV_FIX2) : (v < 0) ? 0 : 255; 67 } 68 69 static WEBP_INLINE int VP8YUVToR(int y, int v) { 70 return VP8Clip8(MultHi(y, 19077) + MultHi(v, 26149) - 14234); 71 } 72 73 static WEBP_INLINE int VP8YUVToG(int y, int u, int v) { 74 return VP8Clip8(MultHi(y, 19077) - MultHi(u, 6419) - MultHi(v, 13320) + 8708); 75 } 76 77 static WEBP_INLINE int VP8YUVToB(int y, int u) { 78 return VP8Clip8(MultHi(y, 19077) + MultHi(u, 33050) - 17685); 79 } 80 81 static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v, 82 uint8_t* const rgb) { 83 rgb[0] = VP8YUVToR(y, v); 84 rgb[1] = VP8YUVToG(y, u, v); 85 rgb[2] = VP8YUVToB(y, u); 86 } 87 88 static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v, 89 uint8_t* const bgr) { 90 bgr[0] = VP8YUVToB(y, u); 91 bgr[1] = VP8YUVToG(y, u, v); 92 bgr[2] = VP8YUVToR(y, v); 93 } 94 95 static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v, 96 uint8_t* const rgb) { 97 const int r = VP8YUVToR(y, v); // 5 usable bits 98 const int g = VP8YUVToG(y, u, v); // 6 usable bits 99 const int b = VP8YUVToB(y, u); // 5 usable bits 100 const int rg = (r & 0xf8) | (g >> 5); 101 const int gb = ((g << 3) & 0xe0) | (b >> 3); 102 #if (WEBP_SWAP_16BIT_CSP == 1) 103 rgb[0] = gb; 104 rgb[1] = rg; 105 #else 106 rgb[0] = rg; 107 rgb[1] = gb; 108 #endif 109 } 110 111 static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v, 112 uint8_t* const argb) { 113 const int r = VP8YUVToR(y, v); // 4 usable bits 114 const int g = VP8YUVToG(y, u, v); // 4 usable bits 115 const int b = VP8YUVToB(y, u); // 4 usable bits 116 const int rg = (r & 0xf0) | (g >> 4); 117 const int ba = (b & 0xf0) | 0x0f; // overwrite the lower 4 bits 118 #if (WEBP_SWAP_16BIT_CSP == 1) 119 argb[0] = ba; 120 argb[1] = rg; 121 #else 122 argb[0] = rg; 123 argb[1] = ba; 124 #endif 125 } 126 127 //----------------------------------------------------------------------------- 128 // Alpha handling variants 129 130 static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v, 131 uint8_t* const argb) { 132 argb[0] = 0xff; 133 VP8YuvToRgb(y, u, v, argb + 1); 134 } 135 136 static WEBP_INLINE void VP8YuvToBgra(uint8_t y, uint8_t u, uint8_t v, 137 uint8_t* const bgra) { 138 VP8YuvToBgr(y, u, v, bgra); 139 bgra[3] = 0xff; 140 } 141 142 static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v, 143 uint8_t* const rgba) { 144 VP8YuvToRgb(y, u, v, rgba); 145 rgba[3] = 0xff; 146 } 147 148 //----------------------------------------------------------------------------- 149 // SSE2 extra functions (mostly for upsampling_sse2.c) 150 151 #if defined(WEBP_USE_SSE2) 152 153 // Process 32 pixels and store the result (16b, 24b or 32b per pixel) in *dst. 154 void VP8YuvToRgba32_SSE2(const uint8_t* WEBP_RESTRICT y, 155 const uint8_t* WEBP_RESTRICT u, 156 const uint8_t* WEBP_RESTRICT v, 157 uint8_t* WEBP_RESTRICT dst); 158 void VP8YuvToRgb32_SSE2(const uint8_t* WEBP_RESTRICT y, 159 const uint8_t* WEBP_RESTRICT u, 160 const uint8_t* WEBP_RESTRICT v, 161 uint8_t* WEBP_RESTRICT dst); 162 void VP8YuvToBgra32_SSE2(const uint8_t* WEBP_RESTRICT y, 163 const uint8_t* WEBP_RESTRICT u, 164 const uint8_t* WEBP_RESTRICT v, 165 uint8_t* WEBP_RESTRICT dst); 166 void VP8YuvToBgr32_SSE2(const uint8_t* WEBP_RESTRICT y, 167 const uint8_t* WEBP_RESTRICT u, 168 const uint8_t* WEBP_RESTRICT v, 169 uint8_t* WEBP_RESTRICT dst); 170 void VP8YuvToArgb32_SSE2(const uint8_t* WEBP_RESTRICT y, 171 const uint8_t* WEBP_RESTRICT u, 172 const uint8_t* WEBP_RESTRICT v, 173 uint8_t* WEBP_RESTRICT dst); 174 void VP8YuvToRgba444432_SSE2(const uint8_t* WEBP_RESTRICT y, 175 const uint8_t* WEBP_RESTRICT u, 176 const uint8_t* WEBP_RESTRICT v, 177 uint8_t* WEBP_RESTRICT dst); 178 void VP8YuvToRgb56532_SSE2(const uint8_t* WEBP_RESTRICT y, 179 const uint8_t* WEBP_RESTRICT u, 180 const uint8_t* WEBP_RESTRICT v, 181 uint8_t* WEBP_RESTRICT dst); 182 183 #endif // WEBP_USE_SSE2 184 185 //----------------------------------------------------------------------------- 186 // SSE41 extra functions (mostly for upsampling_sse41.c) 187 188 #if defined(WEBP_USE_SSE41) 189 190 // Process 32 pixels and store the result (16b, 24b or 32b per pixel) in *dst. 191 void VP8YuvToRgb32_SSE41(const uint8_t* WEBP_RESTRICT y, 192 const uint8_t* WEBP_RESTRICT u, 193 const uint8_t* WEBP_RESTRICT v, 194 uint8_t* WEBP_RESTRICT dst); 195 void VP8YuvToBgr32_SSE41(const uint8_t* WEBP_RESTRICT y, 196 const uint8_t* WEBP_RESTRICT u, 197 const uint8_t* WEBP_RESTRICT v, 198 uint8_t* WEBP_RESTRICT dst); 199 200 #endif // WEBP_USE_SSE41 201 202 //------------------------------------------------------------------------------ 203 // RGB -> YUV conversion 204 205 // Stub functions that can be called with various rounding values: 206 static WEBP_INLINE int VP8ClipUV(int uv, int rounding) { 207 uv = (uv + rounding + (128 << (YUV_FIX + 2))) >> (YUV_FIX + 2); 208 return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255; 209 } 210 211 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) { 212 const int luma = 16839 * r + 33059 * g + 6420 * b; 213 return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX; // no need to clip 214 } 215 216 static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) { 217 const int u = -9719 * r - 19081 * g + 28800 * b; 218 return VP8ClipUV(u, rounding); 219 } 220 221 static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) { 222 const int v = +28800 * r - 24116 * g - 4684 * b; 223 return VP8ClipUV(v, rounding); 224 } 225 226 #ifdef __cplusplus 227 } // extern "C" 228 #endif 229 230 #endif // WEBP_DSP_YUV_H_