1 // Copyright 2011 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 // YUV to RGB upsampling functions.
11 //
12 // Author: somnath@google.com (Somnath Banerjee)
13
14 #include "./dsp.h"
15 #include "./yuv.h"
16
17 #if defined(__cplusplus) || defined(c_plusplus)
18 extern "C" {
19 #endif
20
21 //------------------------------------------------------------------------------
22 // Fancy upsampler
23
24 #ifdef FANCY_UPSAMPLING
25
26 // Fancy upsampling functions to convert YUV to RGB
27 WebPUpsampleLinePairFunc WebPUpsamplers[MODE_LAST];
28
29 // Given samples laid out in a square as:
30 // [a b]
31 // [c d]
32 // we interpolate u/v as:
33 // ([9*a + 3*b + 3*c + d 3*a + 9*b + 3*c + d] + [8 8]) / 16
34 // ([3*a + b + 9*c + 3*d a + 3*b + 3*c + 9*d] [8 8]) / 16
35
36 // We process u and v together stashed into 32bit (16bit each).
37 #define LOAD_UV(u, v) ((u) | ((v) << 16))
38
39 #define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
40 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
41 const uint8_t* top_u, const uint8_t* top_v, \
42 const uint8_t* cur_u, const uint8_t* cur_v, \
43 uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
44 int x; \
45 const int last_pixel_pair = (len - 1) >> 1; \
46 uint32_t tl_uv = LOAD_UV(top_u[0], top_v[0]); /* top-left sample */ \
47 uint32_t l_uv = LOAD_UV(cur_u[0], cur_v[0]); /* left-sample */ \
48 if (top_y) { \
49 const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
50 FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst); \
51 } \
52 if (bottom_y) { \
53 const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
54 FUNC(bottom_y[0], uv0 & 0xff, (uv0 >> 16), bottom_dst); \
55 } \
56 for (x = 1; x <= last_pixel_pair; ++x) { \
57 const uint32_t t_uv = LOAD_UV(top_u[x], top_v[x]); /* top sample */ \
58 const uint32_t uv = LOAD_UV(cur_u[x], cur_v[x]); /* sample */ \
59 /* precompute invariant values associated with first and second diagonals*/\
60 const uint32_t avg = tl_uv + t_uv + l_uv + uv + 0x00080008u; \
61 const uint32_t diag_12 = (avg + 2 * (t_uv + l_uv)) >> 3; \
62 const uint32_t diag_03 = (avg + 2 * (tl_uv + uv)) >> 3; \
63 if (top_y) { \
64 const uint32_t uv0 = (diag_12 + tl_uv) >> 1; \
65 const uint32_t uv1 = (diag_03 + t_uv) >> 1; \
66 FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
67 top_dst + (2 * x - 1) * XSTEP); \
68 FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16), \
69 top_dst + (2 * x - 0) * XSTEP); \
70 } \
71 if (bottom_y) { \
72 const uint32_t uv0 = (diag_03 + l_uv) >> 1; \
73 const uint32_t uv1 = (diag_12 + uv) >> 1; \
74 FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
75 bottom_dst + (2 * x - 1) * XSTEP); \
76 FUNC(bottom_y[2 * x + 0], uv1 & 0xff, (uv1 >> 16), \
77 bottom_dst + (2 * x + 0) * XSTEP); \
78 } \
79 tl_uv = t_uv; \
80 l_uv = uv; \
81 } \
82 if (!(len & 1)) { \
83 if (top_y) { \
84 const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
85 FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
86 top_dst + (len - 1) * XSTEP); \
87 } \
88 if (bottom_y) { \
89 const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
90 FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
91 bottom_dst + (len - 1) * XSTEP); \
92 } \
93 } \
94 }
95
96 // All variants implemented.
97 UPSAMPLE_FUNC(UpsampleRgbLinePair, VP8YuvToRgb, 3)
98 UPSAMPLE_FUNC(UpsampleBgrLinePair, VP8YuvToBgr, 3)
99 UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
100 UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
101 UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
102 UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
103 UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2)
104
105 #undef LOAD_UV
106 #undef UPSAMPLE_FUNC
107
108 #endif // FANCY_UPSAMPLING
109
110 //------------------------------------------------------------------------------
111 // simple point-sampling
112
113 #define SAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
114 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
115 const uint8_t* u, const uint8_t* v, \
116 uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
117 int i; \
118 for (i = 0; i < len - 1; i += 2) { \
119 FUNC(top_y[0], u[0], v[0], top_dst); \
120 FUNC(top_y[1], u[0], v[0], top_dst + XSTEP); \
121 FUNC(bottom_y[0], u[0], v[0], bottom_dst); \
122 FUNC(bottom_y[1], u[0], v[0], bottom_dst + XSTEP); \
123 top_y += 2; \
124 bottom_y += 2; \
125 u++; \
126 v++; \
127 top_dst += 2 * XSTEP; \
128 bottom_dst += 2 * XSTEP; \
129 } \
130 if (i == len - 1) { /* last one */ \
131 FUNC(top_y[0], u[0], v[0], top_dst); \
132 FUNC(bottom_y[0], u[0], v[0], bottom_dst); \
133 } \
134 }
135
136 // All variants implemented.
137 SAMPLE_FUNC(SampleRgbLinePair, VP8YuvToRgb, 3)
138 SAMPLE_FUNC(SampleBgrLinePair, VP8YuvToBgr, 3)
139 SAMPLE_FUNC(SampleRgbaLinePair, VP8YuvToRgba, 4)
140 SAMPLE_FUNC(SampleBgraLinePair, VP8YuvToBgra, 4)
141 SAMPLE_FUNC(SampleArgbLinePair, VP8YuvToArgb, 4)
142 SAMPLE_FUNC(SampleRgba4444LinePair, VP8YuvToRgba4444, 2)
143 SAMPLE_FUNC(SampleRgb565LinePair, VP8YuvToRgb565, 2)
144
145 #undef SAMPLE_FUNC
146
147 const WebPSampleLinePairFunc WebPSamplers[MODE_LAST] = {
148 SampleRgbLinePair, // MODE_RGB
149 SampleRgbaLinePair, // MODE_RGBA
150 SampleBgrLinePair, // MODE_BGR
151 SampleBgraLinePair, // MODE_BGRA
152 SampleArgbLinePair, // MODE_ARGB
153 SampleRgba4444LinePair, // MODE_RGBA_4444
154 SampleRgb565LinePair, // MODE_RGB_565
155 SampleRgbaLinePair, // MODE_rgbA
156 SampleBgraLinePair, // MODE_bgrA
157 SampleArgbLinePair, // MODE_Argb
158 SampleRgba4444LinePair // MODE_rgbA_4444
159 };
160
161 //------------------------------------------------------------------------------
162
163 #if !defined(FANCY_UPSAMPLING)
164 #define DUAL_SAMPLE_FUNC(FUNC_NAME, FUNC) \
165 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y, \
166 const uint8_t* top_u, const uint8_t* top_v, \
167 const uint8_t* bot_u, const uint8_t* bot_v, \
168 uint8_t* top_dst, uint8_t* bot_dst, int len) { \
169 const int half_len = len >> 1; \
170 int x; \
171 if (top_dst != NULL) { \
172 for (x = 0; x < half_len; ++x) { \
173 FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x + 0); \
174 FUNC(top_y[2 * x + 1], top_u[x], top_v[x], top_dst + 8 * x + 4); \
175 } \
176 if (len & 1) FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x); \
177 } \
178 if (bot_dst != NULL) { \
179 for (x = 0; x < half_len; ++x) { \
180 FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x + 0); \
181 FUNC(bot_y[2 * x + 1], bot_u[x], bot_v[x], bot_dst + 8 * x + 4); \
182 } \
183 if (len & 1) FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x); \
184 } \
185 }
186
DUAL_SAMPLE_FUNC(DualLineSamplerBGRA,VP8YuvToBgra)187 DUAL_SAMPLE_FUNC(DualLineSamplerBGRA, VP8YuvToBgra)
188 DUAL_SAMPLE_FUNC(DualLineSamplerARGB, VP8YuvToArgb)
189 #undef DUAL_SAMPLE_FUNC
190
191 #endif // !FANCY_UPSAMPLING
192
193 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last) {
194 WebPInitUpsamplers();
195 VP8YUVInit();
196 #ifdef FANCY_UPSAMPLING
197 return WebPUpsamplers[alpha_is_last ? MODE_BGRA : MODE_ARGB];
198 #else
199 return (alpha_is_last ? DualLineSamplerBGRA : DualLineSamplerARGB);
200 #endif
201 }
202
203 //------------------------------------------------------------------------------
204 // YUV444 converter
205
206 #define YUV444_FUNC(FUNC_NAME, FUNC, XSTEP) \
207 static void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
208 uint8_t* dst, int len) { \
209 int i; \
210 for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]); \
211 }
212
213 YUV444_FUNC(Yuv444ToRgb, VP8YuvToRgb, 3)
214 YUV444_FUNC(Yuv444ToBgr, VP8YuvToBgr, 3)
215 YUV444_FUNC(Yuv444ToRgba, VP8YuvToRgba, 4)
216 YUV444_FUNC(Yuv444ToBgra, VP8YuvToBgra, 4)
217 YUV444_FUNC(Yuv444ToArgb, VP8YuvToArgb, 4)
218 YUV444_FUNC(Yuv444ToRgba4444, VP8YuvToRgba4444, 2)
219 YUV444_FUNC(Yuv444ToRgb565, VP8YuvToRgb565, 2)
220
221 #undef YUV444_FUNC
222
223 const WebPYUV444Converter WebPYUV444Converters[MODE_LAST] = {
224 Yuv444ToRgb, // MODE_RGB
225 Yuv444ToRgba, // MODE_RGBA
226 Yuv444ToBgr, // MODE_BGR
227 Yuv444ToBgra, // MODE_BGRA
228 Yuv444ToArgb, // MODE_ARGB
229 Yuv444ToRgba4444, // MODE_RGBA_4444
230 Yuv444ToRgb565, // MODE_RGB_565
231 Yuv444ToRgba, // MODE_rgbA
232 Yuv444ToBgra, // MODE_bgrA
233 Yuv444ToArgb, // MODE_Argb
234 Yuv444ToRgba4444 // MODE_rgbA_4444
235 };
236
237 //------------------------------------------------------------------------------
238 // Premultiplied modes
239
240 // non dithered-modes
241
242 // (x * a * 32897) >> 23 is bit-wise equivalent to (int)(x * a / 255.)
243 // for all 8bit x or a. For bit-wise equivalence to (int)(x * a / 255. + .5),
244 // one can use instead: (x * a * 65793 + (1 << 23)) >> 24
245 #if 1 // (int)(x * a / 255.)
246 #define MULTIPLIER(a) ((a) * 32897UL)
247 #define PREMULTIPLY(x, m) (((x) * (m)) >> 23)
248 #else // (int)(x * a / 255. + .5)
249 #define MULTIPLIER(a) ((a) * 65793UL)
250 #define PREMULTIPLY(x, m) (((x) * (m) + (1UL << 23)) >> 24)
251 #endif
252
ApplyAlphaMultiply(uint8_t * rgba,int alpha_first,int w,int h,int stride)253 static void ApplyAlphaMultiply(uint8_t* rgba, int alpha_first,
254 int w, int h, int stride) {
255 while (h-- > 0) {
256 uint8_t* const rgb = rgba + (alpha_first ? 1 : 0);
257 const uint8_t* const alpha = rgba + (alpha_first ? 0 : 3);
258 int i;
259 for (i = 0; i < w; ++i) {
260 const uint32_t a = alpha[4 * i];
261 if (a != 0xff) {
262 const uint32_t mult = MULTIPLIER(a);
263 rgb[4 * i + 0] = PREMULTIPLY(rgb[4 * i + 0], mult);
264 rgb[4 * i + 1] = PREMULTIPLY(rgb[4 * i + 1], mult);
265 rgb[4 * i + 2] = PREMULTIPLY(rgb[4 * i + 2], mult);
266 }
267 }
268 rgba += stride;
269 }
270 }
271 #undef MULTIPLIER
272 #undef PREMULTIPLY
273
274 // rgbA4444
275
276 #define MULTIPLIER(a) ((a) * 0x1111) // 0x1111 ~= (1 << 16) / 15
277
dither_hi(uint8_t x)278 static WEBP_INLINE uint8_t dither_hi(uint8_t x) {
279 return (x & 0xf0) | (x >> 4);
280 }
281
dither_lo(uint8_t x)282 static WEBP_INLINE uint8_t dither_lo(uint8_t x) {
283 return (x & 0x0f) | (x << 4);
284 }
285
multiply(uint8_t x,uint32_t m)286 static WEBP_INLINE uint8_t multiply(uint8_t x, uint32_t m) {
287 return (x * m) >> 16;
288 }
289
ApplyAlphaMultiply4444(uint8_t * rgba4444,int w,int h,int stride)290 static void ApplyAlphaMultiply4444(uint8_t* rgba4444,
291 int w, int h, int stride) {
292 while (h-- > 0) {
293 int i;
294 for (i = 0; i < w; ++i) {
295 const uint8_t a = (rgba4444[2 * i + 1] & 0x0f);
296 const uint32_t mult = MULTIPLIER(a);
297 const uint8_t r = multiply(dither_hi(rgba4444[2 * i + 0]), mult);
298 const uint8_t g = multiply(dither_lo(rgba4444[2 * i + 0]), mult);
299 const uint8_t b = multiply(dither_hi(rgba4444[2 * i + 1]), mult);
300 rgba4444[2 * i + 0] = (r & 0xf0) | ((g >> 4) & 0x0f);
301 rgba4444[2 * i + 1] = (b & 0xf0) | a;
302 }
303 rgba4444 += stride;
304 }
305 }
306 #undef MULTIPLIER
307
308 void (*WebPApplyAlphaMultiply)(uint8_t*, int, int, int, int)
309 = ApplyAlphaMultiply;
310 void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int)
311 = ApplyAlphaMultiply4444;
312
313 //------------------------------------------------------------------------------
314 // Main call
315
WebPInitUpsamplers(void)316 void WebPInitUpsamplers(void) {
317 #ifdef FANCY_UPSAMPLING
318 WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair;
319 WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
320 WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair;
321 WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
322 WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
323 WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
324 WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
325
326 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
327 if (VP8GetCPUInfo != NULL) {
328 #if defined(WEBP_USE_SSE2)
329 if (VP8GetCPUInfo(kSSE2)) {
330 WebPInitUpsamplersSSE2();
331 }
332 #endif
333 #if defined(WEBP_USE_NEON)
334 if (VP8GetCPUInfo(kNEON)) {
335 WebPInitUpsamplersNEON();
336 }
337 #endif
338 }
339 #endif // FANCY_UPSAMPLING
340 }
341
WebPInitPremultiply(void)342 void WebPInitPremultiply(void) {
343 WebPApplyAlphaMultiply = ApplyAlphaMultiply;
344 WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply4444;
345
346 #ifdef FANCY_UPSAMPLING
347 WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
348 WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
349 WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair;
350 WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
351
352 if (VP8GetCPUInfo != NULL) {
353 #if defined(WEBP_USE_SSE2)
354 if (VP8GetCPUInfo(kSSE2)) {
355 WebPInitPremultiplySSE2();
356 }
357 #endif
358 #if defined(WEBP_USE_NEON)
359 if (VP8GetCPUInfo(kNEON)) {
360 WebPInitPremultiplyNEON();
361 }
362 #endif
363 }
364 #endif // FANCY_UPSAMPLING
365 }
366
367 #if defined(__cplusplus) || defined(c_plusplus)
368 } // extern "C"
369 #endif
370