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 "src/dsp/dsp.h"
15 #include "src/dsp/yuv.h"
16
17 #include <assert.h>
18
19 //------------------------------------------------------------------------------
20 // Fancy upsampler
21
22 #ifdef FANCY_UPSAMPLING
23
24 // Fancy upsampling functions to convert YUV to RGB
25 WebPUpsampleLinePairFunc WebPUpsamplers[MODE_LAST];
26
27 // Given samples laid out in a square as:
28 // [a b]
29 // [c d]
30 // we interpolate u/v as:
31 // ([9*a + 3*b + 3*c + d 3*a + 9*b + 3*c + d] + [8 8]) / 16
32 // ([3*a + b + 9*c + 3*d a + 3*b + 3*c + 9*d] [8 8]) / 16
33
34 // We process u and v together stashed into 32bit (16bit each).
35 #define LOAD_UV(u, v) ((u) | ((v) << 16))
36
37 #define UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
38 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bottom_y, \
39 const uint8_t* top_u, const uint8_t* top_v, \
40 const uint8_t* cur_u, const uint8_t* cur_v, \
41 uint8_t* top_dst, uint8_t* bottom_dst, int len) { \
42 int x; \
43 const int last_pixel_pair = (len - 1) >> 1; \
44 uint32_t tl_uv = LOAD_UV(top_u[0], top_v[0]); /* top-left sample */ \
45 uint32_t l_uv = LOAD_UV(cur_u[0], cur_v[0]); /* left-sample */ \
46 assert(top_y != NULL); \
47 { \
48 const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
49 FUNC(top_y[0], uv0 & 0xff, (uv0 >> 16), top_dst); \
50 } \
51 if (bottom_y != NULL) { \
52 const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
53 FUNC(bottom_y[0], uv0 & 0xff, (uv0 >> 16), bottom_dst); \
54 } \
55 for (x = 1; x <= last_pixel_pair; ++x) { \
56 const uint32_t t_uv = LOAD_UV(top_u[x], top_v[x]); /* top sample */ \
57 const uint32_t uv = LOAD_UV(cur_u[x], cur_v[x]); /* sample */ \
58 /* precompute invariant values associated with first and second diagonals*/\
59 const uint32_t avg = tl_uv + t_uv + l_uv + uv + 0x00080008u; \
60 const uint32_t diag_12 = (avg + 2 * (t_uv + l_uv)) >> 3; \
61 const uint32_t diag_03 = (avg + 2 * (tl_uv + uv)) >> 3; \
62 { \
63 const uint32_t uv0 = (diag_12 + tl_uv) >> 1; \
64 const uint32_t uv1 = (diag_03 + t_uv) >> 1; \
65 FUNC(top_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
66 top_dst + (2 * x - 1) * (XSTEP)); \
67 FUNC(top_y[2 * x - 0], uv1 & 0xff, (uv1 >> 16), \
68 top_dst + (2 * x - 0) * (XSTEP)); \
69 } \
70 if (bottom_y != NULL) { \
71 const uint32_t uv0 = (diag_03 + l_uv) >> 1; \
72 const uint32_t uv1 = (diag_12 + uv) >> 1; \
73 FUNC(bottom_y[2 * x - 1], uv0 & 0xff, (uv0 >> 16), \
74 bottom_dst + (2 * x - 1) * (XSTEP)); \
75 FUNC(bottom_y[2 * x + 0], uv1 & 0xff, (uv1 >> 16), \
76 bottom_dst + (2 * x + 0) * (XSTEP)); \
77 } \
78 tl_uv = t_uv; \
79 l_uv = uv; \
80 } \
81 if (!(len & 1)) { \
82 { \
83 const uint32_t uv0 = (3 * tl_uv + l_uv + 0x00020002u) >> 2; \
84 FUNC(top_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
85 top_dst + (len - 1) * (XSTEP)); \
86 } \
87 if (bottom_y != NULL) { \
88 const uint32_t uv0 = (3 * l_uv + tl_uv + 0x00020002u) >> 2; \
89 FUNC(bottom_y[len - 1], uv0 & 0xff, (uv0 >> 16), \
90 bottom_dst + (len - 1) * (XSTEP)); \
91 } \
92 } \
93 }
94
95 // All variants implemented.
96 #if !WEBP_NEON_OMIT_C_CODE
97 UPSAMPLE_FUNC(UpsampleRgbaLinePair_C, VP8YuvToRgba, 4)
98 UPSAMPLE_FUNC(UpsampleBgraLinePair_C, VP8YuvToBgra, 4)
99 #if !defined(WEBP_REDUCE_CSP)
100 UPSAMPLE_FUNC(UpsampleArgbLinePair_C, VP8YuvToArgb, 4)
101 UPSAMPLE_FUNC(UpsampleRgbLinePair_C, VP8YuvToRgb, 3)
102 UPSAMPLE_FUNC(UpsampleBgrLinePair_C, VP8YuvToBgr, 3)
103 UPSAMPLE_FUNC(UpsampleRgba4444LinePair_C, VP8YuvToRgba4444, 2)
104 UPSAMPLE_FUNC(UpsampleRgb565LinePair_C, VP8YuvToRgb565, 2)
105 #else
106 static void EmptyUpsampleFunc(const uint8_t* top_y, const uint8_t* bottom_y,
107 const uint8_t* top_u, const uint8_t* top_v,
108 const uint8_t* cur_u, const uint8_t* cur_v,
109 uint8_t* top_dst, uint8_t* bottom_dst, int len) {
110 (void)top_y;
111 (void)bottom_y;
112 (void)top_u;
113 (void)top_v;
114 (void)cur_u;
115 (void)cur_v;
116 (void)top_dst;
117 (void)bottom_dst;
118 (void)len;
119 assert(0); // COLORSPACE SUPPORT NOT COMPILED
120 }
121 #define UpsampleArgbLinePair_C EmptyUpsampleFunc
122 #define UpsampleRgbLinePair_C EmptyUpsampleFunc
123 #define UpsampleBgrLinePair_C EmptyUpsampleFunc
124 #define UpsampleRgba4444LinePair_C EmptyUpsampleFunc
125 #define UpsampleRgb565LinePair_C EmptyUpsampleFunc
126 #endif // WEBP_REDUCE_CSP
127
128 #endif
129
130 #undef LOAD_UV
131 #undef UPSAMPLE_FUNC
132
133 #endif // FANCY_UPSAMPLING
134
135 //------------------------------------------------------------------------------
136
137 #if !defined(FANCY_UPSAMPLING)
138 #define DUAL_SAMPLE_FUNC(FUNC_NAME, FUNC) \
139 static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y, \
140 const uint8_t* top_u, const uint8_t* top_v, \
141 const uint8_t* bot_u, const uint8_t* bot_v, \
142 uint8_t* top_dst, uint8_t* bot_dst, int len) { \
143 const int half_len = len >> 1; \
144 int x; \
145 assert(top_dst != NULL); \
146 { \
147 for (x = 0; x < half_len; ++x) { \
148 FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x + 0); \
149 FUNC(top_y[2 * x + 1], top_u[x], top_v[x], top_dst + 8 * x + 4); \
150 } \
151 if (len & 1) FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x); \
152 } \
153 if (bot_dst != NULL) { \
154 for (x = 0; x < half_len; ++x) { \
155 FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x + 0); \
156 FUNC(bot_y[2 * x + 1], bot_u[x], bot_v[x], bot_dst + 8 * x + 4); \
157 } \
158 if (len & 1) FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x); \
159 } \
160 }
161
DUAL_SAMPLE_FUNC(DualLineSamplerBGRA,VP8YuvToBgra)162 DUAL_SAMPLE_FUNC(DualLineSamplerBGRA, VP8YuvToBgra)
163 DUAL_SAMPLE_FUNC(DualLineSamplerARGB, VP8YuvToArgb)
164 #undef DUAL_SAMPLE_FUNC
165
166 #endif // !FANCY_UPSAMPLING
167
168 WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last) {
169 WebPInitUpsamplers();
170 #ifdef FANCY_UPSAMPLING
171 return WebPUpsamplers[alpha_is_last ? MODE_BGRA : MODE_ARGB];
172 #else
173 return (alpha_is_last ? DualLineSamplerBGRA : DualLineSamplerARGB);
174 #endif
175 }
176
177 //------------------------------------------------------------------------------
178 // YUV444 converter
179
180 #define YUV444_FUNC(FUNC_NAME, FUNC, XSTEP) \
181 extern void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
182 uint8_t* dst, int len); \
183 void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
184 uint8_t* dst, int len) { \
185 int i; \
186 for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * (XSTEP)]); \
187 }
188
189 YUV444_FUNC(WebPYuv444ToRgba_C, VP8YuvToRgba, 4)
190 YUV444_FUNC(WebPYuv444ToBgra_C, VP8YuvToBgra, 4)
191 #if !defined(WEBP_REDUCE_CSP)
192 YUV444_FUNC(WebPYuv444ToRgb_C, VP8YuvToRgb, 3)
193 YUV444_FUNC(WebPYuv444ToBgr_C, VP8YuvToBgr, 3)
194 YUV444_FUNC(WebPYuv444ToArgb_C, VP8YuvToArgb, 4)
195 YUV444_FUNC(WebPYuv444ToRgba4444_C, VP8YuvToRgba4444, 2)
196 YUV444_FUNC(WebPYuv444ToRgb565_C, VP8YuvToRgb565, 2)
197 #else
198 static void EmptyYuv444Func(const uint8_t* y,
199 const uint8_t* u, const uint8_t* v,
200 uint8_t* dst, int len) {
201 (void)y;
202 (void)u;
203 (void)v;
204 (void)dst;
205 (void)len;
206 }
207 #define WebPYuv444ToRgb_C EmptyYuv444Func
208 #define WebPYuv444ToBgr_C EmptyYuv444Func
209 #define WebPYuv444ToArgb_C EmptyYuv444Func
210 #define WebPYuv444ToRgba4444_C EmptyYuv444Func
211 #define WebPYuv444ToRgb565_C EmptyYuv444Func
212 #endif // WEBP_REDUCE_CSP
213
214 #undef YUV444_FUNC
215
216 WebPYUV444Converter WebPYUV444Converters[MODE_LAST];
217
218 extern void WebPInitYUV444ConvertersMIPSdspR2(void);
219 extern void WebPInitYUV444ConvertersSSE2(void);
220
221 static volatile VP8CPUInfo upsampling_last_cpuinfo_used1 =
222 (VP8CPUInfo)&upsampling_last_cpuinfo_used1;
223
WebPInitYUV444Converters(void)224 WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444Converters(void) {
225 if (upsampling_last_cpuinfo_used1 == VP8GetCPUInfo) return;
226
227 WebPYUV444Converters[MODE_RGBA] = WebPYuv444ToRgba_C;
228 WebPYUV444Converters[MODE_BGRA] = WebPYuv444ToBgra_C;
229 WebPYUV444Converters[MODE_RGB] = WebPYuv444ToRgb_C;
230 WebPYUV444Converters[MODE_BGR] = WebPYuv444ToBgr_C;
231 WebPYUV444Converters[MODE_ARGB] = WebPYuv444ToArgb_C;
232 WebPYUV444Converters[MODE_RGBA_4444] = WebPYuv444ToRgba4444_C;
233 WebPYUV444Converters[MODE_RGB_565] = WebPYuv444ToRgb565_C;
234 WebPYUV444Converters[MODE_rgbA] = WebPYuv444ToRgba_C;
235 WebPYUV444Converters[MODE_bgrA] = WebPYuv444ToBgra_C;
236 WebPYUV444Converters[MODE_Argb] = WebPYuv444ToArgb_C;
237 WebPYUV444Converters[MODE_rgbA_4444] = WebPYuv444ToRgba4444_C;
238
239 if (VP8GetCPUInfo != NULL) {
240 #if defined(WEBP_USE_SSE2)
241 if (VP8GetCPUInfo(kSSE2)) {
242 WebPInitYUV444ConvertersSSE2();
243 }
244 #endif
245 #if defined(WEBP_USE_MIPS_DSP_R2)
246 if (VP8GetCPUInfo(kMIPSdspR2)) {
247 WebPInitYUV444ConvertersMIPSdspR2();
248 }
249 #endif
250 }
251 upsampling_last_cpuinfo_used1 = VP8GetCPUInfo;
252 }
253
254 //------------------------------------------------------------------------------
255 // Main calls
256
257 extern void WebPInitUpsamplersSSE2(void);
258 extern void WebPInitUpsamplersNEON(void);
259 extern void WebPInitUpsamplersMIPSdspR2(void);
260 extern void WebPInitUpsamplersMSA(void);
261
262 static volatile VP8CPUInfo upsampling_last_cpuinfo_used2 =
263 (VP8CPUInfo)&upsampling_last_cpuinfo_used2;
264
WebPInitUpsamplers(void)265 WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplers(void) {
266 if (upsampling_last_cpuinfo_used2 == VP8GetCPUInfo) return;
267
268 #ifdef FANCY_UPSAMPLING
269 #if !WEBP_NEON_OMIT_C_CODE
270 WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair_C;
271 WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair_C;
272 WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair_C;
273 WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair_C;
274 WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair_C;
275 WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair_C;
276 WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair_C;
277 WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair_C;
278 WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair_C;
279 WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair_C;
280 WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair_C;
281 #endif
282
283 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
284 if (VP8GetCPUInfo != NULL) {
285 #if defined(WEBP_USE_SSE2)
286 if (VP8GetCPUInfo(kSSE2)) {
287 WebPInitUpsamplersSSE2();
288 }
289 #endif
290 #if defined(WEBP_USE_MIPS_DSP_R2)
291 if (VP8GetCPUInfo(kMIPSdspR2)) {
292 WebPInitUpsamplersMIPSdspR2();
293 }
294 #endif
295 #if defined(WEBP_USE_MSA)
296 if (VP8GetCPUInfo(kMSA)) {
297 WebPInitUpsamplersMSA();
298 }
299 #endif
300 }
301
302 #if defined(WEBP_USE_NEON)
303 if (WEBP_NEON_OMIT_C_CODE ||
304 (VP8GetCPUInfo != NULL && VP8GetCPUInfo(kNEON))) {
305 WebPInitUpsamplersNEON();
306 }
307 #endif
308
309 assert(WebPUpsamplers[MODE_RGBA] != NULL);
310 assert(WebPUpsamplers[MODE_BGRA] != NULL);
311 assert(WebPUpsamplers[MODE_rgbA] != NULL);
312 assert(WebPUpsamplers[MODE_bgrA] != NULL);
313 assert(WebPUpsamplers[MODE_RGB] != NULL);
314 assert(WebPUpsamplers[MODE_BGR] != NULL);
315 assert(WebPUpsamplers[MODE_ARGB] != NULL);
316 assert(WebPUpsamplers[MODE_RGBA_4444] != NULL);
317 assert(WebPUpsamplers[MODE_RGB_565] != NULL);
318 assert(WebPUpsamplers[MODE_Argb] != NULL);
319 assert(WebPUpsamplers[MODE_rgbA_4444] != NULL);
320
321 #endif // FANCY_UPSAMPLING
322 upsampling_last_cpuinfo_used2 = VP8GetCPUInfo;
323 }
324
325 //------------------------------------------------------------------------------
326