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: http://en.wikipedia.org/wiki/YCbCr
14 // Y = 0.2569 * R + 0.5044 * G + 0.0979 * B + 16
15 // U = -0.1483 * R - 0.2911 * G + 0.4394 * B + 128
16 // V = 0.4394 * R - 0.3679 * G - 0.0715 * 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) - 0.391 * (U-128)
22 //   B = 1.164 * (Y-16)                   + 2.018 * (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 "./dsp.h"
39 #include "../dec/decode_vp8.h"
40 
41 #if defined(WEBP_EXPERIMENTAL_FEATURES)
42 // Do NOT activate this feature for real compression. This is only experimental!
43 // This flag is for comparison purpose against JPEG's "YUVj" natural colorspace.
44 // This colorspace is close to Rec.601's Y'CbCr model with the notable
45 // difference of allowing larger range for luma/chroma.
46 // See http://en.wikipedia.org/wiki/YCbCr#JPEG_conversion paragraph, and its
47 // difference with http://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.601_conversion
48 // #define USE_YUVj
49 #endif
50 
51 //------------------------------------------------------------------------------
52 // YUV -> RGB conversion
53 
54 #ifdef __cplusplus
55 extern "C" {
56 #endif
57 
58 enum {
59   YUV_FIX = 16,                    // fixed-point precision for RGB->YUV
60   YUV_HALF = 1 << (YUV_FIX - 1),
61   YUV_MASK = (256 << YUV_FIX) - 1,
62   YUV_RANGE_MIN = -227,            // min value of r/g/b output
63   YUV_RANGE_MAX = 256 + 226,       // max value of r/g/b output
64 
65   YUV_FIX2 = 6,                   // fixed-point precision for YUV->RGB
66   YUV_HALF2 = 1 << YUV_FIX2 >> 1,
67   YUV_MASK2 = (256 << YUV_FIX2) - 1
68 };
69 
70 //------------------------------------------------------------------------------
71 // slower on x86 by ~7-8%, but bit-exact with the SSE2/NEON version
72 
MultHi(int v,int coeff)73 static WEBP_INLINE int MultHi(int v, int coeff) {   // _mm_mulhi_epu16 emulation
74   return (v * coeff) >> 8;
75 }
76 
VP8Clip8(int v)77 static WEBP_INLINE int VP8Clip8(int v) {
78   return ((v & ~YUV_MASK2) == 0) ? (v >> YUV_FIX2) : (v < 0) ? 0 : 255;
79 }
80 
VP8YUVToR(int y,int v)81 static WEBP_INLINE int VP8YUVToR(int y, int v) {
82   return VP8Clip8(MultHi(y, 19077) + MultHi(v, 26149) - 14234);
83 }
84 
VP8YUVToG(int y,int u,int v)85 static WEBP_INLINE int VP8YUVToG(int y, int u, int v) {
86   return VP8Clip8(MultHi(y, 19077) - MultHi(u, 6419) - MultHi(v, 13320) + 8708);
87 }
88 
VP8YUVToB(int y,int u)89 static WEBP_INLINE int VP8YUVToB(int y, int u) {
90   return VP8Clip8(MultHi(y, 19077) + MultHi(u, 33050) - 17685);
91 }
92 
VP8YuvToRgb(int y,int u,int v,uint8_t * const rgb)93 static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
94                                     uint8_t* const rgb) {
95   rgb[0] = VP8YUVToR(y, v);
96   rgb[1] = VP8YUVToG(y, u, v);
97   rgb[2] = VP8YUVToB(y, u);
98 }
99 
VP8YuvToBgr(int y,int u,int v,uint8_t * const bgr)100 static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
101                                     uint8_t* const bgr) {
102   bgr[0] = VP8YUVToB(y, u);
103   bgr[1] = VP8YUVToG(y, u, v);
104   bgr[2] = VP8YUVToR(y, v);
105 }
106 
VP8YuvToRgb565(int y,int u,int v,uint8_t * const rgb)107 static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
108                                        uint8_t* const rgb) {
109   const int r = VP8YUVToR(y, v);      // 5 usable bits
110   const int g = VP8YUVToG(y, u, v);   // 6 usable bits
111   const int b = VP8YUVToB(y, u);      // 5 usable bits
112   const int rg = (r & 0xf8) | (g >> 5);
113   const int gb = ((g << 3) & 0xe0) | (b >> 3);
114 #ifdef WEBP_SWAP_16BIT_CSP
115   rgb[0] = gb;
116   rgb[1] = rg;
117 #else
118   rgb[0] = rg;
119   rgb[1] = gb;
120 #endif
121 }
122 
VP8YuvToRgba4444(int y,int u,int v,uint8_t * const argb)123 static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
124                                          uint8_t* const argb) {
125   const int r = VP8YUVToR(y, v);        // 4 usable bits
126   const int g = VP8YUVToG(y, u, v);     // 4 usable bits
127   const int b = VP8YUVToB(y, u);        // 4 usable bits
128   const int rg = (r & 0xf0) | (g >> 4);
129   const int ba = (b & 0xf0) | 0x0f;     // overwrite the lower 4 bits
130 #ifdef WEBP_SWAP_16BIT_CSP
131   argb[0] = ba;
132   argb[1] = rg;
133 #else
134   argb[0] = rg;
135   argb[1] = ba;
136 #endif
137 }
138 
139 //-----------------------------------------------------------------------------
140 // Alpha handling variants
141 
VP8YuvToArgb(uint8_t y,uint8_t u,uint8_t v,uint8_t * const argb)142 static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
143                                      uint8_t* const argb) {
144   argb[0] = 0xff;
145   VP8YuvToRgb(y, u, v, argb + 1);
146 }
147 
VP8YuvToBgra(uint8_t y,uint8_t u,uint8_t v,uint8_t * const bgra)148 static WEBP_INLINE void VP8YuvToBgra(uint8_t y, uint8_t u, uint8_t v,
149                                      uint8_t* const bgra) {
150   VP8YuvToBgr(y, u, v, bgra);
151   bgra[3] = 0xff;
152 }
153 
VP8YuvToRgba(uint8_t y,uint8_t u,uint8_t v,uint8_t * const rgba)154 static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
155                                      uint8_t* const rgba) {
156   VP8YuvToRgb(y, u, v, rgba);
157   rgba[3] = 0xff;
158 }
159 
160 // Must be called before everything, to initialize the tables.
161 void VP8YUVInit(void);
162 
163 //-----------------------------------------------------------------------------
164 // SSE2 extra functions (mostly for upsampling_sse2.c)
165 
166 #if defined(WEBP_USE_SSE2)
167 
168 // Process 32 pixels and store the result (16b, 24b or 32b per pixel) in *dst.
169 void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
170                     uint8_t* dst);
171 void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
172                    uint8_t* dst);
173 void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
174                     uint8_t* dst);
175 void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
176                    uint8_t* dst);
177 void VP8YuvToArgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
178                     uint8_t* dst);
179 void VP8YuvToRgba444432(const uint8_t* y, const uint8_t* u, const uint8_t* v,
180                         uint8_t* dst);
181 void VP8YuvToRgb56532(const uint8_t* y, const uint8_t* u, const uint8_t* v,
182                       uint8_t* dst);
183 
184 #endif    // WEBP_USE_SSE2
185 
186 //------------------------------------------------------------------------------
187 // RGB -> YUV conversion
188 
189 // Stub functions that can be called with various rounding values:
VP8ClipUV(int uv,int rounding)190 static WEBP_INLINE int VP8ClipUV(int uv, int rounding) {
191   uv = (uv + rounding + (128 << (YUV_FIX + 2))) >> (YUV_FIX + 2);
192   return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255;
193 }
194 
195 #ifndef USE_YUVj
196 
VP8RGBToY(int r,int g,int b,int rounding)197 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
198   const int luma = 16839 * r + 33059 * g + 6420 * b;
199   return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX;  // no need to clip
200 }
201 
VP8RGBToU(int r,int g,int b,int rounding)202 static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
203   const int u = -9719 * r - 19081 * g + 28800 * b;
204   return VP8ClipUV(u, rounding);
205 }
206 
VP8RGBToV(int r,int g,int b,int rounding)207 static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
208   const int v = +28800 * r - 24116 * g - 4684 * b;
209   return VP8ClipUV(v, rounding);
210 }
211 
212 #else
213 
214 // This JPEG-YUV colorspace, only for comparison!
215 // These are also 16bit precision coefficients from Rec.601, but with full
216 // [0..255] output range.
VP8RGBToY(int r,int g,int b,int rounding)217 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
218   const int luma = 19595 * r + 38470 * g + 7471 * b;
219   return (luma + rounding) >> YUV_FIX;  // no need to clip
220 }
221 
VP8RGBToU(int r,int g,int b,int rounding)222 static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
223   const int u = -11058 * r - 21710 * g + 32768 * b;
224   return VP8ClipUV(u, rounding);
225 }
226 
VP8RGBToV(int r,int g,int b,int rounding)227 static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
228   const int v = 32768 * r - 27439 * g - 5329 * b;
229   return VP8ClipUV(v, rounding);
230 }
231 
232 #endif    // USE_YUVj
233 
234 #ifdef __cplusplus
235 }    // extern "C"
236 #endif
237 
238 #endif  /* WEBP_DSP_YUV_H_ */
239