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 // In the table-lookup version (WEBP_YUV_USE_TABLE), the common factor
25 // "1.164 * (Y-16)" can be handled as an offset in the VP8kClip[] table.
26 // So in this case the formulae should read:
27 // R = 1.164 * [Y + 1.371 * (V-128) ] - 18.624
28 // G = 1.164 * [Y - 0.698 * (V-128) - 0.336 * (U-128)] - 18.624
29 // B = 1.164 * [Y + 1.733 * (U-128)] - 18.624
30 // once factorized.
31 // For YUV->RGB conversion, only 14bit fixed precision is used (YUV_FIX2).
32 // That's the maximum possible for a convenient ARM implementation.
33 //
34 // Author: Skal (pascal.massimino@gmail.com)
35
36 #ifndef WEBP_DSP_YUV_H_
37 #define WEBP_DSP_YUV_H_
38
39 #include "./dsp.h"
40 #include "../dec/decode_vp8.h"
41
42 // Define the following to use the LUT-based code:
43 // #define WEBP_YUV_USE_TABLE
44
45 #if defined(WEBP_EXPERIMENTAL_FEATURES)
46 // Do NOT activate this feature for real compression. This is only experimental!
47 // This flag is for comparison purpose against JPEG's "YUVj" natural colorspace.
48 // This colorspace is close to Rec.601's Y'CbCr model with the notable
49 // difference of allowing larger range for luma/chroma.
50 // See http://en.wikipedia.org/wiki/YCbCr#JPEG_conversion paragraph, and its
51 // difference with http://en.wikipedia.org/wiki/YCbCr#ITU-R_BT.601_conversion
52 // #define USE_YUVj
53 #endif
54
55 //------------------------------------------------------------------------------
56 // YUV -> RGB conversion
57
58 #ifdef __cplusplus
59 extern "C" {
60 #endif
61
62 enum {
63 YUV_FIX = 16, // fixed-point precision for RGB->YUV
64 YUV_HALF = 1 << (YUV_FIX - 1),
65 YUV_MASK = (256 << YUV_FIX) - 1,
66 YUV_RANGE_MIN = -227, // min value of r/g/b output
67 YUV_RANGE_MAX = 256 + 226, // max value of r/g/b output
68
69 YUV_FIX2 = 14, // fixed-point precision for YUV->RGB
70 YUV_HALF2 = 1 << (YUV_FIX2 - 1),
71 YUV_MASK2 = (256 << YUV_FIX2) - 1
72 };
73
74 // These constants are 14b fixed-point version of ITU-R BT.601 constants.
75 #define kYScale 19077 // 1.164 = 255 / 219
76 #define kVToR 26149 // 1.596 = 255 / 112 * 0.701
77 #define kUToG 6419 // 0.391 = 255 / 112 * 0.886 * 0.114 / 0.587
78 #define kVToG 13320 // 0.813 = 255 / 112 * 0.701 * 0.299 / 0.587
79 #define kUToB 33050 // 2.018 = 255 / 112 * 0.886
80 #define kRCst (-kYScale * 16 - kVToR * 128 + YUV_HALF2)
81 #define kGCst (-kYScale * 16 + kUToG * 128 + kVToG * 128 + YUV_HALF2)
82 #define kBCst (-kYScale * 16 - kUToB * 128 + YUV_HALF2)
83
84 //------------------------------------------------------------------------------
85
86 #if !defined(WEBP_YUV_USE_TABLE)
87
88 // slower on x86 by ~7-8%, but bit-exact with the SSE2 version
89
VP8Clip8(int v)90 static WEBP_INLINE int VP8Clip8(int v) {
91 return ((v & ~YUV_MASK2) == 0) ? (v >> YUV_FIX2) : (v < 0) ? 0 : 255;
92 }
93
VP8YUVToR(int y,int v)94 static WEBP_INLINE int VP8YUVToR(int y, int v) {
95 return VP8Clip8(kYScale * y + kVToR * v + kRCst);
96 }
97
VP8YUVToG(int y,int u,int v)98 static WEBP_INLINE int VP8YUVToG(int y, int u, int v) {
99 return VP8Clip8(kYScale * y - kUToG * u - kVToG * v + kGCst);
100 }
101
VP8YUVToB(int y,int u)102 static WEBP_INLINE int VP8YUVToB(int y, int u) {
103 return VP8Clip8(kYScale * y + kUToB * u + kBCst);
104 }
105
VP8YuvToRgb(int y,int u,int v,uint8_t * const rgb)106 static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
107 uint8_t* const rgb) {
108 rgb[0] = VP8YUVToR(y, v);
109 rgb[1] = VP8YUVToG(y, u, v);
110 rgb[2] = VP8YUVToB(y, u);
111 }
112
VP8YuvToBgr(int y,int u,int v,uint8_t * const bgr)113 static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
114 uint8_t* const bgr) {
115 bgr[0] = VP8YUVToB(y, u);
116 bgr[1] = VP8YUVToG(y, u, v);
117 bgr[2] = VP8YUVToR(y, v);
118 }
119
VP8YuvToRgb565(int y,int u,int v,uint8_t * const rgb)120 static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
121 uint8_t* const rgb) {
122 const int r = VP8YUVToR(y, v); // 5 usable bits
123 const int g = VP8YUVToG(y, u, v); // 6 usable bits
124 const int b = VP8YUVToB(y, u); // 5 usable bits
125 const int rg = (r & 0xf8) | (g >> 5);
126 const int gb = ((g << 3) & 0xe0) | (b >> 3);
127 #ifdef WEBP_SWAP_16BIT_CSP
128 rgb[0] = gb;
129 rgb[1] = rg;
130 #else
131 rgb[0] = rg;
132 rgb[1] = gb;
133 #endif
134 }
135
VP8YuvToRgba4444(int y,int u,int v,uint8_t * const argb)136 static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
137 uint8_t* const argb) {
138 const int r = VP8YUVToR(y, v); // 4 usable bits
139 const int g = VP8YUVToG(y, u, v); // 4 usable bits
140 const int b = VP8YUVToB(y, u); // 4 usable bits
141 const int rg = (r & 0xf0) | (g >> 4);
142 const int ba = (b & 0xf0) | 0x0f; // overwrite the lower 4 bits
143 #ifdef WEBP_SWAP_16BIT_CSP
144 argb[0] = ba;
145 argb[1] = rg;
146 #else
147 argb[0] = rg;
148 argb[1] = ba;
149 #endif
150 }
151
152 #else
153
154 // Table-based version, not totally equivalent to the SSE2 version.
155 // Rounding diff is only +/-1 though.
156
157 extern int16_t VP8kVToR[256], VP8kUToB[256];
158 extern int32_t VP8kVToG[256], VP8kUToG[256];
159 extern uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
160 extern uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
161
VP8YuvToRgb(int y,int u,int v,uint8_t * const rgb)162 static WEBP_INLINE void VP8YuvToRgb(int y, int u, int v,
163 uint8_t* const rgb) {
164 const int r_off = VP8kVToR[v];
165 const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
166 const int b_off = VP8kUToB[u];
167 rgb[0] = VP8kClip[y + r_off - YUV_RANGE_MIN];
168 rgb[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
169 rgb[2] = VP8kClip[y + b_off - YUV_RANGE_MIN];
170 }
171
VP8YuvToBgr(int y,int u,int v,uint8_t * const bgr)172 static WEBP_INLINE void VP8YuvToBgr(int y, int u, int v,
173 uint8_t* const bgr) {
174 const int r_off = VP8kVToR[v];
175 const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
176 const int b_off = VP8kUToB[u];
177 bgr[0] = VP8kClip[y + b_off - YUV_RANGE_MIN];
178 bgr[1] = VP8kClip[y + g_off - YUV_RANGE_MIN];
179 bgr[2] = VP8kClip[y + r_off - YUV_RANGE_MIN];
180 }
181
VP8YuvToRgb565(int y,int u,int v,uint8_t * const rgb)182 static WEBP_INLINE void VP8YuvToRgb565(int y, int u, int v,
183 uint8_t* const rgb) {
184 const int r_off = VP8kVToR[v];
185 const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
186 const int b_off = VP8kUToB[u];
187 const int rg = ((VP8kClip[y + r_off - YUV_RANGE_MIN] & 0xf8) |
188 (VP8kClip[y + g_off - YUV_RANGE_MIN] >> 5));
189 const int gb = (((VP8kClip[y + g_off - YUV_RANGE_MIN] << 3) & 0xe0) |
190 (VP8kClip[y + b_off - YUV_RANGE_MIN] >> 3));
191 #ifdef WEBP_SWAP_16BIT_CSP
192 rgb[0] = gb;
193 rgb[1] = rg;
194 #else
195 rgb[0] = rg;
196 rgb[1] = gb;
197 #endif
198 }
199
VP8YuvToRgba4444(int y,int u,int v,uint8_t * const argb)200 static WEBP_INLINE void VP8YuvToRgba4444(int y, int u, int v,
201 uint8_t* const argb) {
202 const int r_off = VP8kVToR[v];
203 const int g_off = (VP8kVToG[v] + VP8kUToG[u]) >> YUV_FIX;
204 const int b_off = VP8kUToB[u];
205 const int rg = ((VP8kClip4Bits[y + r_off - YUV_RANGE_MIN] << 4) |
206 VP8kClip4Bits[y + g_off - YUV_RANGE_MIN]);
207 const int ba = (VP8kClip4Bits[y + b_off - YUV_RANGE_MIN] << 4) | 0x0f;
208 #ifdef WEBP_SWAP_16BIT_CSP
209 argb[0] = ba;
210 argb[1] = rg;
211 #else
212 argb[0] = rg;
213 argb[1] = ba;
214 #endif
215 }
216
217 #endif // WEBP_YUV_USE_TABLE
218
219 //-----------------------------------------------------------------------------
220 // Alpha handling variants
221
VP8YuvToArgb(uint8_t y,uint8_t u,uint8_t v,uint8_t * const argb)222 static WEBP_INLINE void VP8YuvToArgb(uint8_t y, uint8_t u, uint8_t v,
223 uint8_t* const argb) {
224 argb[0] = 0xff;
225 VP8YuvToRgb(y, u, v, argb + 1);
226 }
227
VP8YuvToBgra(uint8_t y,uint8_t u,uint8_t v,uint8_t * const bgra)228 static WEBP_INLINE void VP8YuvToBgra(uint8_t y, uint8_t u, uint8_t v,
229 uint8_t* const bgra) {
230 VP8YuvToBgr(y, u, v, bgra);
231 bgra[3] = 0xff;
232 }
233
VP8YuvToRgba(uint8_t y,uint8_t u,uint8_t v,uint8_t * const rgba)234 static WEBP_INLINE void VP8YuvToRgba(uint8_t y, uint8_t u, uint8_t v,
235 uint8_t* const rgba) {
236 VP8YuvToRgb(y, u, v, rgba);
237 rgba[3] = 0xff;
238 }
239
240 // Must be called before everything, to initialize the tables.
241 void VP8YUVInit(void);
242
243 //-----------------------------------------------------------------------------
244 // SSE2 extra functions (mostly for upsampling_sse2.c)
245
246 #if defined(WEBP_USE_SSE2)
247
248 // When the following is defined, tables are initialized statically, adding ~12k
249 // to the binary size. Otherwise, they are initialized at run-time (small cost).
250 #define WEBP_YUV_USE_SSE2_TABLES
251
252 #if defined(FANCY_UPSAMPLING)
253 // Process 32 pixels and store the result (24b or 32b per pixel) in *dst.
254 void VP8YuvToRgba32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
255 uint8_t* dst);
256 void VP8YuvToRgb32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
257 uint8_t* dst);
258 void VP8YuvToBgra32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
259 uint8_t* dst);
260 void VP8YuvToBgr32(const uint8_t* y, const uint8_t* u, const uint8_t* v,
261 uint8_t* dst);
262 #endif // FANCY_UPSAMPLING
263
264 // Must be called to initialize tables before using the functions.
265 void VP8YUVInitSSE2(void);
266
267 #endif // WEBP_USE_SSE2
268
269 //------------------------------------------------------------------------------
270 // RGB -> YUV conversion
271
272 // Stub functions that can be called with various rounding values:
VP8ClipUV(int uv,int rounding)273 static WEBP_INLINE int VP8ClipUV(int uv, int rounding) {
274 uv = (uv + rounding + (128 << (YUV_FIX + 2))) >> (YUV_FIX + 2);
275 return ((uv & ~0xff) == 0) ? uv : (uv < 0) ? 0 : 255;
276 }
277
278 #ifndef USE_YUVj
279
VP8RGBToY(int r,int g,int b,int rounding)280 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
281 const int luma = 16839 * r + 33059 * g + 6420 * b;
282 return (luma + rounding + (16 << YUV_FIX)) >> YUV_FIX; // no need to clip
283 }
284
VP8RGBToU(int r,int g,int b,int rounding)285 static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
286 const int u = -9719 * r - 19081 * g + 28800 * b;
287 return VP8ClipUV(u, rounding);
288 }
289
VP8RGBToV(int r,int g,int b,int rounding)290 static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
291 const int v = +28800 * r - 24116 * g - 4684 * b;
292 return VP8ClipUV(v, rounding);
293 }
294
295 #else
296
297 // This JPEG-YUV colorspace, only for comparison!
298 // These are also 16bit precision coefficients from Rec.601, but with full
299 // [0..255] output range.
VP8RGBToY(int r,int g,int b,int rounding)300 static WEBP_INLINE int VP8RGBToY(int r, int g, int b, int rounding) {
301 const int luma = 19595 * r + 38470 * g + 7471 * b;
302 return (luma + rounding) >> YUV_FIX; // no need to clip
303 }
304
VP8RGBToU(int r,int g,int b,int rounding)305 static WEBP_INLINE int VP8RGBToU(int r, int g, int b, int rounding) {
306 const int u = -11058 * r - 21710 * g + 32768 * b;
307 return VP8ClipUV(u, rounding);
308 }
309
VP8RGBToV(int r,int g,int b,int rounding)310 static WEBP_INLINE int VP8RGBToV(int r, int g, int b, int rounding) {
311 const int v = 32768 * r - 27439 * g - 5329 * b;
312 return VP8ClipUV(v, rounding);
313 }
314
315 #endif // USE_YUVj
316
317 #ifdef __cplusplus
318 } // extern "C"
319 #endif
320
321 #endif /* WEBP_DSP_YUV_H_ */
322