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#include <assert.h>
18
19//------------------------------------------------------------------------------
20// Fancy upsampler
21
22#ifdef FANCY_UPSAMPLING
23
24// Fancy upsampling functions to convert YUV to RGB
25WebPUpsampleLinePairFunc 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) \
38static 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.
96UPSAMPLE_FUNC(UpsampleRgbLinePair, VP8YuvToRgb, 3)
97UPSAMPLE_FUNC(UpsampleBgrLinePair, VP8YuvToBgr, 3)
98UPSAMPLE_FUNC(UpsampleRgbaLinePair, VP8YuvToRgba, 4)
99UPSAMPLE_FUNC(UpsampleBgraLinePair, VP8YuvToBgra, 4)
100UPSAMPLE_FUNC(UpsampleArgbLinePair, VP8YuvToArgb, 4)
101UPSAMPLE_FUNC(UpsampleRgba4444LinePair, VP8YuvToRgba4444, 2)
102UPSAMPLE_FUNC(UpsampleRgb565LinePair, VP8YuvToRgb565, 2)
103
104#undef LOAD_UV
105#undef UPSAMPLE_FUNC
106
107#endif // FANCY_UPSAMPLING
108
109//------------------------------------------------------------------------------
110
111#if !defined(FANCY_UPSAMPLING)
112#define DUAL_SAMPLE_FUNC(FUNC_NAME, FUNC) \
113static void FUNC_NAME(const uint8_t* top_y, const uint8_t* bot_y, \
114 const uint8_t* top_u, const uint8_t* top_v, \
115 const uint8_t* bot_u, const uint8_t* bot_v, \
116 uint8_t* top_dst, uint8_t* bot_dst, int len) { \
117 const int half_len = len >> 1; \
118 int x; \
119 assert(top_dst != NULL); \
120 { \
121 for (x = 0; x < half_len; ++x) { \
122 FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x + 0); \
123 FUNC(top_y[2 * x + 1], top_u[x], top_v[x], top_dst + 8 * x + 4); \
124 } \
125 if (len & 1) FUNC(top_y[2 * x + 0], top_u[x], top_v[x], top_dst + 8 * x); \
126 } \
127 if (bot_dst != NULL) { \
128 for (x = 0; x < half_len; ++x) { \
129 FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x + 0); \
130 FUNC(bot_y[2 * x + 1], bot_u[x], bot_v[x], bot_dst + 8 * x + 4); \
131 } \
132 if (len & 1) FUNC(bot_y[2 * x + 0], bot_u[x], bot_v[x], bot_dst + 8 * x); \
133 } \
134}
135
136DUAL_SAMPLE_FUNC(DualLineSamplerBGRA, VP8YuvToBgra)
137DUAL_SAMPLE_FUNC(DualLineSamplerARGB, VP8YuvToArgb)
138#undef DUAL_SAMPLE_FUNC
139
140#endif // !FANCY_UPSAMPLING
141
142WebPUpsampleLinePairFunc WebPGetLinePairConverter(int alpha_is_last) {
143 WebPInitUpsamplers();
144 VP8YUVInit();
145#ifdef FANCY_UPSAMPLING
146 return WebPUpsamplers[alpha_is_last ? MODE_BGRA : MODE_ARGB];
147#else
148 return (alpha_is_last ? DualLineSamplerBGRA : DualLineSamplerARGB);
149#endif
150}
151
152//------------------------------------------------------------------------------
153// YUV444 converter
154
155#define YUV444_FUNC(FUNC_NAME, FUNC, XSTEP) \
156extern void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
157 uint8_t* dst, int len); \
158void FUNC_NAME(const uint8_t* y, const uint8_t* u, const uint8_t* v, \
159 uint8_t* dst, int len) { \
160 int i; \
161 for (i = 0; i < len; ++i) FUNC(y[i], u[i], v[i], &dst[i * XSTEP]); \
162}
163
164YUV444_FUNC(WebPYuv444ToRgbC, VP8YuvToRgb, 3)
165YUV444_FUNC(WebPYuv444ToBgrC, VP8YuvToBgr, 3)
166YUV444_FUNC(WebPYuv444ToRgbaC, VP8YuvToRgba, 4)
167YUV444_FUNC(WebPYuv444ToBgraC, VP8YuvToBgra, 4)
168YUV444_FUNC(WebPYuv444ToArgbC, VP8YuvToArgb, 4)
169YUV444_FUNC(WebPYuv444ToRgba4444C, VP8YuvToRgba4444, 2)
170YUV444_FUNC(WebPYuv444ToRgb565C, VP8YuvToRgb565, 2)
171
172#undef YUV444_FUNC
173
174WebPYUV444Converter WebPYUV444Converters[MODE_LAST];
175
176extern void WebPInitYUV444ConvertersMIPSdspR2(void);
177extern void WebPInitYUV444ConvertersSSE2(void);
178
179static volatile VP8CPUInfo upsampling_last_cpuinfo_used1 =
180 (VP8CPUInfo)&upsampling_last_cpuinfo_used1;
181
182WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444Converters(void) {
183 if (upsampling_last_cpuinfo_used1 == VP8GetCPUInfo) return;
184
185 WebPYUV444Converters[MODE_RGB] = WebPYuv444ToRgbC;
186 WebPYUV444Converters[MODE_RGBA] = WebPYuv444ToRgbaC;
187 WebPYUV444Converters[MODE_BGR] = WebPYuv444ToBgrC;
188 WebPYUV444Converters[MODE_BGRA] = WebPYuv444ToBgraC;
189 WebPYUV444Converters[MODE_ARGB] = WebPYuv444ToArgbC;
190 WebPYUV444Converters[MODE_RGBA_4444] = WebPYuv444ToRgba4444C;
191 WebPYUV444Converters[MODE_RGB_565] = WebPYuv444ToRgb565C;
192 WebPYUV444Converters[MODE_rgbA] = WebPYuv444ToRgbaC;
193 WebPYUV444Converters[MODE_bgrA] = WebPYuv444ToBgraC;
194 WebPYUV444Converters[MODE_Argb] = WebPYuv444ToArgbC;
195 WebPYUV444Converters[MODE_rgbA_4444] = WebPYuv444ToRgba4444C;
196
197 if (VP8GetCPUInfo != NULL) {
198#if defined(WEBP_USE_SSE2)
199 if (VP8GetCPUInfo(kSSE2)) {
200 WebPInitYUV444ConvertersSSE2();
201 }
202#endif
203#if defined(WEBP_USE_MIPS_DSP_R2)
204 if (VP8GetCPUInfo(kMIPSdspR2)) {
205 WebPInitYUV444ConvertersMIPSdspR2();
206 }
207#endif
208 }
209 upsampling_last_cpuinfo_used1 = VP8GetCPUInfo;
210}
211
212//------------------------------------------------------------------------------
213// Main calls
214
215extern void WebPInitUpsamplersSSE2(void);
216extern void WebPInitUpsamplersNEON(void);
217extern void WebPInitUpsamplersMIPSdspR2(void);
218extern void WebPInitUpsamplersMSA(void);
219
220static volatile VP8CPUInfo upsampling_last_cpuinfo_used2 =
221 (VP8CPUInfo)&upsampling_last_cpuinfo_used2;
222
223WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplers(void) {
224 if (upsampling_last_cpuinfo_used2 == VP8GetCPUInfo) return;
225
226#ifdef FANCY_UPSAMPLING
227 WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair;
228 WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair;
229 WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair;
230 WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair;
231 WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair;
232 WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair;
233 WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair;
234 WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair;
235 WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair;
236 WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair;
237 WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair;
238
239 // If defined, use CPUInfo() to overwrite some pointers with faster versions.
240 if (VP8GetCPUInfo != NULL) {
241#if defined(WEBP_USE_SSE2)
242 if (VP8GetCPUInfo(kSSE2)) {
243 WebPInitUpsamplersSSE2();
244 }
245#endif
246#if defined(WEBP_USE_NEON)
247 if (VP8GetCPUInfo(kNEON)) {
248 WebPInitUpsamplersNEON();
249 }
250#endif
251#if defined(WEBP_USE_MIPS_DSP_R2)
252 if (VP8GetCPUInfo(kMIPSdspR2)) {
253 WebPInitUpsamplersMIPSdspR2();
254 }
255#endif
256#if defined(WEBP_USE_MSA)
257 if (VP8GetCPUInfo(kMSA)) {
258 WebPInitUpsamplersMSA();
259 }
260#endif
261 }
262#endif // FANCY_UPSAMPLING
263 upsampling_last_cpuinfo_used2 = VP8GetCPUInfo;
264}
265
266//------------------------------------------------------------------------------
267