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// YUV->RGB conversion functions
11//
12// Author: Skal (pascal.massimino@gmail.com)
13
14#include "./yuv.h"
15
16#include <stdlib.h>
17
18#if defined(WEBP_YUV_USE_TABLE)
19
20static int done = 0;
21
22static WEBP_INLINE uint8_t clip(int v, int max_value) {
23 return v < 0 ? 0 : v > max_value ? max_value : v;
24}
25
26int16_t VP8kVToR[256], VP8kUToB[256];
27int32_t VP8kVToG[256], VP8kUToG[256];
28uint8_t VP8kClip[YUV_RANGE_MAX - YUV_RANGE_MIN];
29uint8_t VP8kClip4Bits[YUV_RANGE_MAX - YUV_RANGE_MIN];
30
31WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {
32 int i;
33 if (done) {
34 return;
35 }
36#ifndef USE_YUVj
37 for (i = 0; i < 256; ++i) {
38 VP8kVToR[i] = (89858 * (i - 128) + YUV_HALF) >> YUV_FIX;
39 VP8kUToG[i] = -22014 * (i - 128) + YUV_HALF;
40 VP8kVToG[i] = -45773 * (i - 128);
41 VP8kUToB[i] = (113618 * (i - 128) + YUV_HALF) >> YUV_FIX;
42 }
43 for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
44 const int k = ((i - 16) * 76283 + YUV_HALF) >> YUV_FIX;
45 VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
46 VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
47 }
48#else
49 for (i = 0; i < 256; ++i) {
50 VP8kVToR[i] = (91881 * (i - 128) + YUV_HALF) >> YUV_FIX;
51 VP8kUToG[i] = -22554 * (i - 128) + YUV_HALF;
52 VP8kVToG[i] = -46802 * (i - 128);
53 VP8kUToB[i] = (116130 * (i - 128) + YUV_HALF) >> YUV_FIX;
54 }
55 for (i = YUV_RANGE_MIN; i < YUV_RANGE_MAX; ++i) {
56 const int k = i;
57 VP8kClip[i - YUV_RANGE_MIN] = clip(k, 255);
58 VP8kClip4Bits[i - YUV_RANGE_MIN] = clip((k + 8) >> 4, 15);
59 }
60#endif
61
62 done = 1;
63}
64
65#else
66
67WEBP_TSAN_IGNORE_FUNCTION void VP8YUVInit(void) {}
68
69#endif // WEBP_YUV_USE_TABLE
70
71//-----------------------------------------------------------------------------
72// Plain-C version
73
74#define ROW_FUNC(FUNC_NAME, FUNC, XSTEP) \
75static void FUNC_NAME(const uint8_t* y, \
76 const uint8_t* u, const uint8_t* v, \
77 uint8_t* dst, int len) { \
78 const uint8_t* const end = dst + (len & ~1) * XSTEP; \
79 while (dst != end) { \
80 FUNC(y[0], u[0], v[0], dst); \
81 FUNC(y[1], u[0], v[0], dst + XSTEP); \
82 y += 2; \
83 ++u; \
84 ++v; \
85 dst += 2 * XSTEP; \
86 } \
87 if (len & 1) { \
88 FUNC(y[0], u[0], v[0], dst); \
89 } \
90} \
91
92// All variants implemented.
93ROW_FUNC(YuvToRgbRow, VP8YuvToRgb, 3)
94ROW_FUNC(YuvToBgrRow, VP8YuvToBgr, 3)
95ROW_FUNC(YuvToRgbaRow, VP8YuvToRgba, 4)
96ROW_FUNC(YuvToBgraRow, VP8YuvToBgra, 4)
97ROW_FUNC(YuvToArgbRow, VP8YuvToArgb, 4)
98ROW_FUNC(YuvToRgba4444Row, VP8YuvToRgba4444, 2)
99ROW_FUNC(YuvToRgb565Row, VP8YuvToRgb565, 2)
100
101#undef ROW_FUNC
102
103// Main call for processing a plane with a WebPSamplerRowFunc function:
104void WebPSamplerProcessPlane(const uint8_t* y, int y_stride,
105 const uint8_t* u, const uint8_t* v, int uv_stride,
106 uint8_t* dst, int dst_stride,
107 int width, int height, WebPSamplerRowFunc func) {
108 int j;
109 for (j = 0; j < height; ++j) {
110 func(y, u, v, dst, width);
111 y += y_stride;
112 if (j & 1) {
113 u += uv_stride;
114 v += uv_stride;
115 }
116 dst += dst_stride;
117 }
118}
119
120//-----------------------------------------------------------------------------
121// Main call
122
123WebPSamplerRowFunc WebPSamplers[MODE_LAST];
124
125extern void WebPInitSamplersSSE2(void);
126extern void WebPInitSamplersMIPS32(void);
127extern void WebPInitSamplersMIPSdspR2(void);
128
129static volatile VP8CPUInfo yuv_last_cpuinfo_used =
130 (VP8CPUInfo)&yuv_last_cpuinfo_used;
131
132WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplers(void) {
133 if (yuv_last_cpuinfo_used == VP8GetCPUInfo) return;
134
135 WebPSamplers[MODE_RGB] = YuvToRgbRow;
136 WebPSamplers[MODE_RGBA] = YuvToRgbaRow;
137 WebPSamplers[MODE_BGR] = YuvToBgrRow;
138 WebPSamplers[MODE_BGRA] = YuvToBgraRow;
139 WebPSamplers[MODE_ARGB] = YuvToArgbRow;
140 WebPSamplers[MODE_RGBA_4444] = YuvToRgba4444Row;
141 WebPSamplers[MODE_RGB_565] = YuvToRgb565Row;
142 WebPSamplers[MODE_rgbA] = YuvToRgbaRow;
143 WebPSamplers[MODE_bgrA] = YuvToBgraRow;
144 WebPSamplers[MODE_Argb] = YuvToArgbRow;
145 WebPSamplers[MODE_rgbA_4444] = YuvToRgba4444Row;
146
147 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
148 if (VP8GetCPUInfo != NULL) {
149#if defined(WEBP_USE_SSE2)
150 if (VP8GetCPUInfo(kSSE2)) {
151 WebPInitSamplersSSE2();
152 }
153#endif // WEBP_USE_SSE2
154#if defined(WEBP_USE_MIPS32)
155 if (VP8GetCPUInfo(kMIPS32)) {
156 WebPInitSamplersMIPS32();
157 }
158#endif // WEBP_USE_MIPS32
159#if defined(WEBP_USE_MIPS_DSP_R2)
160 if (VP8GetCPUInfo(kMIPSdspR2)) {
161 WebPInitSamplersMIPSdspR2();
162 }
163#endif // WEBP_USE_MIPS_DSP_R2
164 }
165 yuv_last_cpuinfo_used = VP8GetCPUInfo;
166}
167
168//-----------------------------------------------------------------------------
169// ARGB -> YUV converters
170
171static void ConvertARGBToY(const uint32_t* argb, uint8_t* y, int width) {
172 int i;
173 for (i = 0; i < width; ++i) {
174 const uint32_t p = argb[i];
175 y[i] = VP8RGBToY((p >> 16) & 0xff, (p >> 8) & 0xff, (p >> 0) & 0xff,
176 YUV_HALF);
177 }
178}
179
180void WebPConvertARGBToUV_C(const uint32_t* argb, uint8_t* u, uint8_t* v,
181 int src_width, int do_store) {
182 // No rounding. Last pixel is dealt with separately.
183 const int uv_width = src_width >> 1;
184 int i;
185 for (i = 0; i < uv_width; ++i) {
186 const uint32_t v0 = argb[2 * i + 0];
187 const uint32_t v1 = argb[2 * i + 1];
188 // VP8RGBToU/V expects four accumulated pixels. Hence we need to
189 // scale r/g/b value by a factor 2. We just shift v0/v1 one bit less.
190 const int r = ((v0 >> 15) & 0x1fe) + ((v1 >> 15) & 0x1fe);
191 const int g = ((v0 >> 7) & 0x1fe) + ((v1 >> 7) & 0x1fe);
192 const int b = ((v0 << 1) & 0x1fe) + ((v1 << 1) & 0x1fe);
193 const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
194 const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
195 if (do_store) {
196 u[i] = tmp_u;
197 v[i] = tmp_v;
198 } else {
199 // Approximated average-of-four. But it's an acceptable diff.
200 u[i] = (u[i] + tmp_u + 1) >> 1;
201 v[i] = (v[i] + tmp_v + 1) >> 1;
202 }
203 }
204 if (src_width & 1) { // last pixel
205 const uint32_t v0 = argb[2 * i + 0];
206 const int r = (v0 >> 14) & 0x3fc;
207 const int g = (v0 >> 6) & 0x3fc;
208 const int b = (v0 << 2) & 0x3fc;
209 const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
210 const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
211 if (do_store) {
212 u[i] = tmp_u;
213 v[i] = tmp_v;
214 } else {
215 u[i] = (u[i] + tmp_u + 1) >> 1;
216 v[i] = (v[i] + tmp_v + 1) >> 1;
217 }
218 }
219}
220
221//-----------------------------------------------------------------------------
222
223static void ConvertRGB24ToY(const uint8_t* rgb, uint8_t* y, int width) {
224 int i;
225 for (i = 0; i < width; ++i, rgb += 3) {
226 y[i] = VP8RGBToY(rgb[0], rgb[1], rgb[2], YUV_HALF);
227 }
228}
229
230static void ConvertBGR24ToY(const uint8_t* bgr, uint8_t* y, int width) {
231 int i;
232 for (i = 0; i < width; ++i, bgr += 3) {
233 y[i] = VP8RGBToY(bgr[2], bgr[1], bgr[0], YUV_HALF);
234 }
235}
236
237void WebPConvertRGBA32ToUV_C(const uint16_t* rgb,
238 uint8_t* u, uint8_t* v, int width) {
239 int i;
240 for (i = 0; i < width; i += 1, rgb += 4) {
241 const int r = rgb[0], g = rgb[1], b = rgb[2];
242 u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
243 v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
244 }
245}
246
247//-----------------------------------------------------------------------------
248
249#define MAX_Y ((1 << 10) - 1) // 10b precision over 16b-arithmetic
250static uint16_t clip_y(int v) {
251 return (v < 0) ? 0 : (v > MAX_Y) ? MAX_Y : (uint16_t)v;
252}
253
254static uint64_t SharpYUVUpdateY_C(const uint16_t* ref, const uint16_t* src,
255 uint16_t* dst, int len) {
256 uint64_t diff = 0;
257 int i;
258 for (i = 0; i < len; ++i) {
259 const int diff_y = ref[i] - src[i];
260 const int new_y = (int)dst[i] + diff_y;
261 dst[i] = clip_y(new_y);
262 diff += (uint64_t)abs(diff_y);
263 }
264 return diff;
265}
266
267static void SharpYUVUpdateRGB_C(const int16_t* ref, const int16_t* src,
268 int16_t* dst, int len) {
269 int i;
270 for (i = 0; i < len; ++i) {
271 const int diff_uv = ref[i] - src[i];
272 dst[i] += diff_uv;
273 }
274}
275
276static void SharpYUVFilterRow_C(const int16_t* A, const int16_t* B, int len,
277 const uint16_t* best_y, uint16_t* out) {
278 int i;
279 for (i = 0; i < len; ++i, ++A, ++B) {
280 const int v0 = (A[0] * 9 + A[1] * 3 + B[0] * 3 + B[1] + 8) >> 4;
281 const int v1 = (A[1] * 9 + A[0] * 3 + B[1] * 3 + B[0] + 8) >> 4;
282 out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
283 out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
284 }
285}
286
287#undef MAX_Y
288
289//-----------------------------------------------------------------------------
290
291void (*WebPConvertRGB24ToY)(const uint8_t* rgb, uint8_t* y, int width);
292void (*WebPConvertBGR24ToY)(const uint8_t* bgr, uint8_t* y, int width);
293void (*WebPConvertRGBA32ToUV)(const uint16_t* rgb,
294 uint8_t* u, uint8_t* v, int width);
295
296void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
297void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
298 int src_width, int do_store);
299
300uint64_t (*WebPSharpYUVUpdateY)(const uint16_t* ref, const uint16_t* src,
301 uint16_t* dst, int len);
302void (*WebPSharpYUVUpdateRGB)(const int16_t* ref, const int16_t* src,
303 int16_t* dst, int len);
304void (*WebPSharpYUVFilterRow)(const int16_t* A, const int16_t* B, int len,
305 const uint16_t* best_y, uint16_t* out);
306
307static volatile VP8CPUInfo rgba_to_yuv_last_cpuinfo_used =
308 (VP8CPUInfo)&rgba_to_yuv_last_cpuinfo_used;
309
310extern void WebPInitConvertARGBToYUVSSE2(void);
311extern void WebPInitSharpYUVSSE2(void);
312
313WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUV(void) {
314 if (rgba_to_yuv_last_cpuinfo_used == VP8GetCPUInfo) return;
315
316 WebPConvertARGBToY = ConvertARGBToY;
317 WebPConvertARGBToUV = WebPConvertARGBToUV_C;
318
319 WebPConvertRGB24ToY = ConvertRGB24ToY;
320 WebPConvertBGR24ToY = ConvertBGR24ToY;
321
322 WebPConvertRGBA32ToUV = WebPConvertRGBA32ToUV_C;
323
324 WebPSharpYUVUpdateY = SharpYUVUpdateY_C;
325 WebPSharpYUVUpdateRGB = SharpYUVUpdateRGB_C;
326 WebPSharpYUVFilterRow = SharpYUVFilterRow_C;
327
328 if (VP8GetCPUInfo != NULL) {
329#if defined(WEBP_USE_SSE2)
330 if (VP8GetCPUInfo(kSSE2)) {
331 WebPInitConvertARGBToYUVSSE2();
332 WebPInitSharpYUVSSE2();
333 }
334#endif // WEBP_USE_SSE2
335 }
336 rgba_to_yuv_last_cpuinfo_used = VP8GetCPUInfo;
337}
338