projects / libav.git / blob
? search:


60b1d4b66eac62bbb47daae0c3e0bcfecb7a152e
[libav.git] / libswscale / swscale.c
1 /*
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * the C code (not assembly, mmx, ...) of this file can be used
21 * under the LGPL license too
22 */
23
24 /*
25 supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR32_1, BGR24, BGR16, BGR15, RGB32, RGB32_1, RGB24, Y8/Y800, YVU9/IF09, PAL8
26 supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27 {BGR,RGB}{1,4,8,15,16} support dithering
28
29 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
31 x -> x
32 YUV9 -> YV12
33 YUV9/YV12 -> Y800
34 Y800 -> YUV9/YV12
35 BGR24 -> BGR32 & RGB24 -> RGB32
36 BGR32 -> BGR24 & RGB32 -> RGB24
37 BGR15 -> BGR16
38 */
39
40 /*
41 tested special converters (most are tested actually, but I did not write it down ...)
42 YV12 -> BGR16
43 YV12 -> YV12
44 BGR15 -> BGR16
45 BGR16 -> BGR16
46 YVU9 -> YV12
47
48 untested special converters
49 YV12/I420 -> BGR15/BGR24/BGR32 (it is the yuv2rgb stuff, so it should be OK)
50 YV12/I420 -> YV12/I420
51 YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52 BGR24 -> BGR32 & RGB24 -> RGB32
53 BGR32 -> BGR24 & RGB32 -> RGB24
54 BGR24 -> YV12
55 */
56
57 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
58 #include <inttypes.h>
59 #include <string.h>
60 #include <math.h>
61 #include <stdio.h>
62 #include "config.h"
63 #include <assert.h>
64 #if HAVE_SYS_MMAN_H
65 #include <sys/mman.h>
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
68 #endif
69 #endif
70 #if HAVE_VIRTUALALLOC
71 #define WIN32_LEAN_AND_MEAN
72 #include <windows.h>
73 #endif
74 #include "swscale.h"
75 #include "swscale_internal.h"
76 #include "rgb2rgb.h"
77 #include "libavutil/intreadwrite.h"
78 #include "libavutil/x86_cpu.h"
79 #include "libavutil/avutil.h"
80 #include "libavutil/bswap.h"
81
82 unsigned swscale_version(void)
83 {
84 return LIBSWSCALE_VERSION_INT;
85 }
86
87 #undef MOVNTQ
88 #undef PAVGB
89
90 //#undef HAVE_MMX2
91 //#define HAVE_AMD3DNOW
92 //#undef HAVE_MMX
93 //#undef ARCH_X86
94 #define DITHER1XBPP
95
96 #define FAST_BGR2YV12 // use 7 bit coefficients instead of 15 bit
97
98 #define RET 0xC3 //near return opcode for x86
99
100 #ifdef M_PI
101 #define PI M_PI
102 #else
103 #define PI 3.14159265358979323846
104 #endif
105
106 #define isSupportedIn(x) ( \
107 (x)==PIX_FMT_YUV420P \
108 || (x)==PIX_FMT_YUVA420P \
109 || (x)==PIX_FMT_YUYV422 \
110 || (x)==PIX_FMT_UYVY422 \
111 || (x)==PIX_FMT_RGB48BE \
112 || (x)==PIX_FMT_RGB48LE \
113 || (x)==PIX_FMT_RGB32 \
114 || (x)==PIX_FMT_RGB32_1 \
115 || (x)==PIX_FMT_BGR24 \
116 || (x)==PIX_FMT_BGR565 \
117 || (x)==PIX_FMT_BGR555 \
118 || (x)==PIX_FMT_BGR32 \
119 || (x)==PIX_FMT_BGR32_1 \
120 || (x)==PIX_FMT_RGB24 \
121 || (x)==PIX_FMT_RGB565 \
122 || (x)==PIX_FMT_RGB555 \
123 || (x)==PIX_FMT_GRAY8 \
124 || (x)==PIX_FMT_YUV410P \
125 || (x)==PIX_FMT_YUV440P \
126 || (x)==PIX_FMT_GRAY16BE \
127 || (x)==PIX_FMT_GRAY16LE \
128 || (x)==PIX_FMT_YUV444P \
129 || (x)==PIX_FMT_YUV422P \
130 || (x)==PIX_FMT_YUV411P \
131 || (x)==PIX_FMT_PAL8 \
132 || (x)==PIX_FMT_BGR8 \
133 || (x)==PIX_FMT_RGB8 \
134 || (x)==PIX_FMT_BGR4_BYTE \
135 || (x)==PIX_FMT_RGB4_BYTE \
136 || (x)==PIX_FMT_YUV440P \
137 || (x)==PIX_FMT_MONOWHITE \
138 || (x)==PIX_FMT_MONOBLACK \
139 || (x)==PIX_FMT_YUV420PLE \
140 || (x)==PIX_FMT_YUV422PLE \
141 || (x)==PIX_FMT_YUV444PLE \
142 || (x)==PIX_FMT_YUV420PBE \
143 || (x)==PIX_FMT_YUV422PBE \
144 || (x)==PIX_FMT_YUV444PBE \
145 )
146 #define isSupportedOut(x) ( \
147 (x)==PIX_FMT_YUV420P \
148 || (x)==PIX_FMT_YUVA420P \
149 || (x)==PIX_FMT_YUYV422 \
150 || (x)==PIX_FMT_UYVY422 \
151 || (x)==PIX_FMT_YUV444P \
152 || (x)==PIX_FMT_YUV422P \
153 || (x)==PIX_FMT_YUV411P \
154 || isRGB(x) \
155 || isBGR(x) \
156 || (x)==PIX_FMT_NV12 \
157 || (x)==PIX_FMT_NV21 \
158 || (x)==PIX_FMT_GRAY16BE \
159 || (x)==PIX_FMT_GRAY16LE \
160 || (x)==PIX_FMT_GRAY8 \
161 || (x)==PIX_FMT_YUV410P \
162 || (x)==PIX_FMT_YUV440P \
163 || (x)==PIX_FMT_YUV420PLE \
164 || (x)==PIX_FMT_YUV422PLE \
165 || (x)==PIX_FMT_YUV444PLE \
166 || (x)==PIX_FMT_YUV420PBE \
167 || (x)==PIX_FMT_YUV422PBE \
168 || (x)==PIX_FMT_YUV444PBE \
169 )
170 #define isPacked(x) ( \
171 (x)==PIX_FMT_PAL8 \
172 || (x)==PIX_FMT_YUYV422 \
173 || (x)==PIX_FMT_UYVY422 \
174 || isRGB(x) \
175 || isBGR(x) \
176 )
177 #define usePal(x) ( \
178 (x)==PIX_FMT_PAL8 \
179 || (x)==PIX_FMT_BGR4_BYTE \
180 || (x)==PIX_FMT_RGB4_BYTE \
181 || (x)==PIX_FMT_BGR8 \
182 || (x)==PIX_FMT_RGB8 \
183 )
184
185 #define RGB2YUV_SHIFT 15
186 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
187 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
188 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
189 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
190 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
191 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
192 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
193 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
194 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
195
196 extern const int32_t ff_yuv2rgb_coeffs[8][4];
197
198 static const double rgb2yuv_table[8][9]={
199 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
200 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
201 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
202 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
203 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
204 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
205 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
206 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
207 };
208
209 /*
210 NOTES
211 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
212
213 TODO
214 more intelligent misalignment avoidance for the horizontal scaler
215 write special vertical cubic upscale version
216 optimize C code (YV12 / minmax)
217 add support for packed pixel YUV input & output
218 add support for Y8 output
219 optimize BGR24 & BGR32
220 add BGR4 output support
221 write special BGR->BGR scaler
222 */
223
224 #if ARCH_X86 && CONFIG_GPL
225 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
226 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
227 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
228 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
229 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
230 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
231 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
232 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
233
234 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
235 0x0103010301030103LL,
236 0x0200020002000200LL,};
237
238 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
239 0x0602060206020602LL,
240 0x0004000400040004LL,};
241
242 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
243 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
244 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
245 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
246 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
247 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
248
249 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
250 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
251 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
252
253 #ifdef FAST_BGR2YV12
254 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
255 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
256 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
257 #else
258 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
259 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
260 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
261 #endif /* FAST_BGR2YV12 */
262 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
263 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
264 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
265
266 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
267 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
268 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
269 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
270 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
271
272 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
273 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
274 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
275 };
276
277 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
278
279 #endif /* ARCH_X86 && CONFIG_GPL */
280
281 // clipping helper table for C implementations:
282 static unsigned char clip_table[768];
283
284 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
285
286 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_4[2][8])={
287 { 1, 3, 1, 3, 1, 3, 1, 3, },
288 { 2, 0, 2, 0, 2, 0, 2, 0, },
289 };
290
291 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8[2][8])={
292 { 6, 2, 6, 2, 6, 2, 6, 2, },
293 { 0, 4, 0, 4, 0, 4, 0, 4, },
294 };
295
296 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32[8][8])={
297 { 17, 9, 23, 15, 16, 8, 22, 14, },
298 { 5, 29, 3, 27, 4, 28, 2, 26, },
299 { 21, 13, 19, 11, 20, 12, 18, 10, },
300 { 0, 24, 6, 30, 1, 25, 7, 31, },
301 { 16, 8, 22, 14, 17, 9, 23, 15, },
302 { 4, 28, 2, 26, 5, 29, 3, 27, },
303 { 20, 12, 18, 10, 21, 13, 19, 11, },
304 { 1, 25, 7, 31, 0, 24, 6, 30, },
305 };
306
307 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73[8][8])={
308 { 0, 55, 14, 68, 3, 58, 17, 72, },
309 { 37, 18, 50, 32, 40, 22, 54, 35, },
310 { 9, 64, 5, 59, 13, 67, 8, 63, },
311 { 46, 27, 41, 23, 49, 31, 44, 26, },
312 { 2, 57, 16, 71, 1, 56, 15, 70, },
313 { 39, 21, 52, 34, 38, 19, 51, 33, },
314 { 11, 66, 7, 62, 10, 65, 6, 60, },
315 { 48, 30, 43, 25, 47, 29, 42, 24, },
316 };
317
318 #if 1
319 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
320 {117, 62, 158, 103, 113, 58, 155, 100, },
321 { 34, 199, 21, 186, 31, 196, 17, 182, },
322 {144, 89, 131, 76, 141, 86, 127, 72, },
323 { 0, 165, 41, 206, 10, 175, 52, 217, },
324 {110, 55, 151, 96, 120, 65, 162, 107, },
325 { 28, 193, 14, 179, 38, 203, 24, 189, },
326 {138, 83, 124, 69, 148, 93, 134, 79, },
327 { 7, 172, 48, 213, 3, 168, 45, 210, },
328 };
329 #elif 1
330 // tries to correct a gamma of 1.5
331 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
332 { 0, 143, 18, 200, 2, 156, 25, 215, },
333 { 78, 28, 125, 64, 89, 36, 138, 74, },
334 { 10, 180, 3, 161, 16, 195, 8, 175, },
335 {109, 51, 93, 38, 121, 60, 105, 47, },
336 { 1, 152, 23, 210, 0, 147, 20, 205, },
337 { 85, 33, 134, 71, 81, 30, 130, 67, },
338 { 14, 190, 6, 171, 12, 185, 5, 166, },
339 {117, 57, 101, 44, 113, 54, 97, 41, },
340 };
341 #elif 1
342 // tries to correct a gamma of 2.0
343 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
344 { 0, 124, 8, 193, 0, 140, 12, 213, },
345 { 55, 14, 104, 42, 66, 19, 119, 52, },
346 { 3, 168, 1, 145, 6, 187, 3, 162, },
347 { 86, 31, 70, 21, 99, 39, 82, 28, },
348 { 0, 134, 11, 206, 0, 129, 9, 200, },
349 { 62, 17, 114, 48, 58, 16, 109, 45, },
350 { 5, 181, 2, 157, 4, 175, 1, 151, },
351 { 95, 36, 78, 26, 90, 34, 74, 24, },
352 };
353 #else
354 // tries to correct a gamma of 2.5
355 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
356 { 0, 107, 3, 187, 0, 125, 6, 212, },
357 { 39, 7, 86, 28, 49, 11, 102, 36, },
358 { 1, 158, 0, 131, 3, 180, 1, 151, },
359 { 68, 19, 52, 12, 81, 25, 64, 17, },
360 { 0, 119, 5, 203, 0, 113, 4, 195, },
361 { 45, 9, 96, 33, 42, 8, 91, 30, },
362 { 2, 172, 1, 144, 2, 165, 0, 137, },
363 { 77, 23, 60, 15, 72, 21, 56, 14, },
364 };
365 #endif
366
367 const char *sws_format_name(enum PixelFormat format)
368 {
369 switch (format) {
370 case PIX_FMT_YUV420P:
371 return "yuv420p";
372 case PIX_FMT_YUVA420P:
373 return "yuva420p";
374 case PIX_FMT_YUYV422:
375 return "yuyv422";
376 case PIX_FMT_RGB24:
377 return "rgb24";
378 case PIX_FMT_BGR24:
379 return "bgr24";
380 case PIX_FMT_YUV422P:
381 return "yuv422p";
382 case PIX_FMT_YUV444P:
383 return "yuv444p";
384 case PIX_FMT_RGB32:
385 return "rgb32";
386 case PIX_FMT_YUV410P:
387 return "yuv410p";
388 case PIX_FMT_YUV411P:
389 return "yuv411p";
390 case PIX_FMT_RGB565:
391 return "rgb565";
392 case PIX_FMT_RGB555:
393 return "rgb555";
394 case PIX_FMT_GRAY16BE:
395 return "gray16be";
396 case PIX_FMT_GRAY16LE:
397 return "gray16le";
398 case PIX_FMT_GRAY8:
399 return "gray8";
400 case PIX_FMT_MONOWHITE:
401 return "mono white";
402 case PIX_FMT_MONOBLACK:
403 return "mono black";
404 case PIX_FMT_PAL8:
405 return "Palette";
406 case PIX_FMT_YUVJ420P:
407 return "yuvj420p";
408 case PIX_FMT_YUVJ422P:
409 return "yuvj422p";
410 case PIX_FMT_YUVJ444P:
411 return "yuvj444p";
412 case PIX_FMT_XVMC_MPEG2_MC:
413 return "xvmc_mpeg2_mc";
414 case PIX_FMT_XVMC_MPEG2_IDCT:
415 return "xvmc_mpeg2_idct";
416 case PIX_FMT_UYVY422:
417 return "uyvy422";
418 case PIX_FMT_UYYVYY411:
419 return "uyyvyy411";
420 case PIX_FMT_RGB32_1:
421 return "rgb32x";
422 case PIX_FMT_BGR32_1:
423 return "bgr32x";
424 case PIX_FMT_BGR32:
425 return "bgr32";
426 case PIX_FMT_BGR565:
427 return "bgr565";
428 case PIX_FMT_BGR555:
429 return "bgr555";
430 case PIX_FMT_BGR8:
431 return "bgr8";
432 case PIX_FMT_BGR4:
433 return "bgr4";
434 case PIX_FMT_BGR4_BYTE:
435 return "bgr4 byte";
436 case PIX_FMT_RGB8:
437 return "rgb8";
438 case PIX_FMT_RGB4:
439 return "rgb4";
440 case PIX_FMT_RGB4_BYTE:
441 return "rgb4 byte";
442 case PIX_FMT_RGB48BE:
443 return "rgb48be";
444 case PIX_FMT_RGB48LE:
445 return "rgb48le";
446 case PIX_FMT_NV12:
447 return "nv12";
448 case PIX_FMT_NV21:
449 return "nv21";
450 case PIX_FMT_YUV440P:
451 return "yuv440p";
452 case PIX_FMT_VDPAU_H264:
453 return "vdpau_h264";
454 case PIX_FMT_VDPAU_MPEG1:
455 return "vdpau_mpeg1";
456 case PIX_FMT_VDPAU_MPEG2:
457 return "vdpau_mpeg2";
458 case PIX_FMT_VDPAU_WMV3:
459 return "vdpau_wmv3";
460 case PIX_FMT_VDPAU_VC1:
461 return "vdpau_vc1";
462 case PIX_FMT_YUV420PLE:
463 return "yuv420ple";
464 case PIX_FMT_YUV422PLE:
465 return "yuv422ple";
466 case PIX_FMT_YUV444PLE:
467 return "yuv444ple";
468 case PIX_FMT_YUV420PBE:
469 return "yuv420pbe";
470 case PIX_FMT_YUV422PBE:
471 return "yuv422pbe";
472 case PIX_FMT_YUV444PBE:
473 return "yuv444pbe";
474 default:
475 return "Unknown format";
476 }
477 }
478
479 static av_always_inline void yuv2yuvX16inC_template(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
480 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
481 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest,
482 int dstW, int chrDstW, int big_endian)
483 {
484 //FIXME Optimize (just quickly written not optimized..)
485 int i;
486
487 for (i = 0; i < dstW; i++) {
488 int val = 1 << 10;
489 int j;
490
491 for (j = 0; j < lumFilterSize; j++)
492 val += lumSrc[j][i] * lumFilter[j];
493
494 if (big_endian) {
495 AV_WB16(&dest[i], av_clip_uint16(val >> 11));
496 } else {
497 AV_WL16(&dest[i], av_clip_uint16(val >> 11));
498 }
499 }
500
501 if (uDest) {
502 for (i = 0; i < chrDstW; i++) {
503 int u = 1 << 10;
504 int v = 1 << 10;
505 int j;
506
507 for (j = 0; j < chrFilterSize; j++) {
508 u += chrSrc[j][i ] * chrFilter[j];
509 v += chrSrc[j][i + VOFW] * chrFilter[j];
510 }
511
512 if (big_endian) {
513 AV_WB16(&uDest[i], av_clip_uint16(u >> 11));
514 AV_WB16(&vDest[i], av_clip_uint16(v >> 11));
515 } else {
516 AV_WL16(&uDest[i], av_clip_uint16(u >> 11));
517 AV_WL16(&vDest[i], av_clip_uint16(v >> 11));
518 }
519 }
520 }
521
522 if (CONFIG_SWSCALE_ALPHA && aDest) {
523 for (i = 0; i < dstW; i++) {
524 int val = 1 << 10;
525 int j;
526
527 for (j = 0; j < lumFilterSize; j++)
528 val += alpSrc[j][i] * lumFilter[j];
529
530 if (big_endian) {
531 AV_WB16(&aDest[i], av_clip_uint16(val >> 11));
532 } else {
533 AV_WL16(&aDest[i], av_clip_uint16(val >> 11));
534 }
535 }
536 }
537 }
538
539 static inline void yuv2yuvX16inC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
540 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
541 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest, int dstW, int chrDstW,
542 enum PixelFormat dstFormat)
543 {
544 if (isBE(dstFormat)) {
545 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
546 chrFilter, chrSrc, chrFilterSize,
547 alpSrc,
548 dest, uDest, vDest, aDest,
549 dstW, chrDstW, 1);
550 } else {
551 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
552 chrFilter, chrSrc, chrFilterSize,
553 alpSrc,
554 dest, uDest, vDest, aDest,
555 dstW, chrDstW, 0);
556 }
557 }
558
559 static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
560 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
561 const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW)
562 {
563 //FIXME Optimize (just quickly written not optimized..)
564 int i;
565 for (i=0; i<dstW; i++) {
566 int val=1<<18;
567 int j;
568 for (j=0; j<lumFilterSize; j++)
569 val += lumSrc[j][i] * lumFilter[j];
570
571 dest[i]= av_clip_uint8(val>>19);
572 }
573
574 if (uDest)
575 for (i=0; i<chrDstW; i++) {
576 int u=1<<18;
577 int v=1<<18;
578 int j;
579 for (j=0; j<chrFilterSize; j++) {
580 u += chrSrc[j][i] * chrFilter[j];
581 v += chrSrc[j][i + VOFW] * chrFilter[j];
582 }
583
584 uDest[i]= av_clip_uint8(u>>19);
585 vDest[i]= av_clip_uint8(v>>19);
586 }
587
588 if (CONFIG_SWSCALE_ALPHA && aDest)
589 for (i=0; i<dstW; i++) {
590 int val=1<<18;
591 int j;
592 for (j=0; j<lumFilterSize; j++)
593 val += alpSrc[j][i] * lumFilter[j];
594
595 aDest[i]= av_clip_uint8(val>>19);
596 }
597
598 }
599
600 static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
601 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
602 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
603 {
604 //FIXME Optimize (just quickly written not optimized..)
605 int i;
606 for (i=0; i<dstW; i++) {
607 int val=1<<18;
608 int j;
609 for (j=0; j<lumFilterSize; j++)
610 val += lumSrc[j][i] * lumFilter[j];
611
612 dest[i]= av_clip_uint8(val>>19);
613 }
614
615 if (!uDest)
616 return;
617
618 if (dstFormat == PIX_FMT_NV12)
619 for (i=0; i<chrDstW; i++) {
620 int u=1<<18;
621 int v=1<<18;
622 int j;
623 for (j=0; j<chrFilterSize; j++) {
624 u += chrSrc[j][i] * chrFilter[j];
625 v += chrSrc[j][i + VOFW] * chrFilter[j];
626 }
627
628 uDest[2*i]= av_clip_uint8(u>>19);
629 uDest[2*i+1]= av_clip_uint8(v>>19);
630 }
631 else
632 for (i=0; i<chrDstW; i++) {
633 int u=1<<18;
634 int v=1<<18;
635 int j;
636 for (j=0; j<chrFilterSize; j++) {
637 u += chrSrc[j][i] * chrFilter[j];
638 v += chrSrc[j][i + VOFW] * chrFilter[j];
639 }
640
641 uDest[2*i]= av_clip_uint8(v>>19);
642 uDest[2*i+1]= av_clip_uint8(u>>19);
643 }
644 }
645
646 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha) \
647 for (i=0; i<(dstW>>1); i++) {\
648 int j;\
649 int Y1 = 1<<18;\
650 int Y2 = 1<<18;\
651 int U = 1<<18;\
652 int V = 1<<18;\
653 int av_unused A1, A2;\
654 type av_unused *r, *b, *g;\
655 const int i2= 2*i;\
656 \
657 for (j=0; j<lumFilterSize; j++) {\
658 Y1 += lumSrc[j][i2] * lumFilter[j];\
659 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
660 }\
661 for (j=0; j<chrFilterSize; j++) {\
662 U += chrSrc[j][i] * chrFilter[j];\
663 V += chrSrc[j][i+VOFW] * chrFilter[j];\
664 }\
665 Y1>>=19;\
666 Y2>>=19;\
667 U >>=19;\
668 V >>=19;\
669 if (alpha) {\
670 A1 = 1<<18;\
671 A2 = 1<<18;\
672 for (j=0; j<lumFilterSize; j++) {\
673 A1 += alpSrc[j][i2 ] * lumFilter[j];\
674 A2 += alpSrc[j][i2+1] * lumFilter[j];\
675 }\
676 A1>>=19;\
677 A2>>=19;\
678 }\
679
680 #define YSCALE_YUV_2_PACKEDX_C(type,alpha) \
681 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha)\
682 if ((Y1|Y2|U|V)&256) {\
683 if (Y1>255) Y1=255; \
684 else if (Y1<0)Y1=0; \
685 if (Y2>255) Y2=255; \
686 else if (Y2<0)Y2=0; \
687 if (U>255) U=255; \
688 else if (U<0) U=0; \
689 if (V>255) V=255; \
690 else if (V<0) V=0; \
691 }\
692 if (alpha && ((A1|A2)&256)) {\
693 A1=av_clip_uint8(A1);\
694 A2=av_clip_uint8(A2);\
695 }
696
697 #define YSCALE_YUV_2_PACKEDX_FULL_C(rnd,alpha) \
698 for (i=0; i<dstW; i++) {\
699 int j;\
700 int Y = 0;\
701 int U = -128<<19;\
702 int V = -128<<19;\
703 int av_unused A;\
704 int R,G,B;\
705 \
706 for (j=0; j<lumFilterSize; j++) {\
707 Y += lumSrc[j][i ] * lumFilter[j];\
708 }\
709 for (j=0; j<chrFilterSize; j++) {\
710 U += chrSrc[j][i ] * chrFilter[j];\
711 V += chrSrc[j][i+VOFW] * chrFilter[j];\
712 }\
713 Y >>=10;\
714 U >>=10;\
715 V >>=10;\
716 if (alpha) {\
717 A = rnd;\
718 for (j=0; j<lumFilterSize; j++)\
719 A += alpSrc[j][i ] * lumFilter[j];\
720 A >>=19;\
721 if (A&256)\
722 A = av_clip_uint8(A);\
723 }\
724
725 #define YSCALE_YUV_2_RGBX_FULL_C(rnd,alpha) \
726 YSCALE_YUV_2_PACKEDX_FULL_C(rnd>>3,alpha)\
727 Y-= c->yuv2rgb_y_offset;\
728 Y*= c->yuv2rgb_y_coeff;\
729 Y+= rnd;\
730 R= Y + V*c->yuv2rgb_v2r_coeff;\
731 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
732 B= Y + U*c->yuv2rgb_u2b_coeff;\
733 if ((R|G|B)&(0xC0000000)) {\
734 if (R>=(256<<22)) R=(256<<22)-1; \
735 else if (R<0)R=0; \
736 if (G>=(256<<22)) G=(256<<22)-1; \
737 else if (G<0)G=0; \
738 if (B>=(256<<22)) B=(256<<22)-1; \
739 else if (B<0)B=0; \
740 }\
741
742
743 #define YSCALE_YUV_2_GRAY16_C \
744 for (i=0; i<(dstW>>1); i++) {\
745 int j;\
746 int Y1 = 1<<18;\
747 int Y2 = 1<<18;\
748 int U = 1<<18;\
749 int V = 1<<18;\
750 \
751 const int i2= 2*i;\
752 \
753 for (j=0; j<lumFilterSize; j++) {\
754 Y1 += lumSrc[j][i2] * lumFilter[j];\
755 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
756 }\
757 Y1>>=11;\
758 Y2>>=11;\
759 if ((Y1|Y2|U|V)&65536) {\
760 if (Y1>65535) Y1=65535; \
761 else if (Y1<0)Y1=0; \
762 if (Y2>65535) Y2=65535; \
763 else if (Y2<0)Y2=0; \
764 }
765
766 #define YSCALE_YUV_2_RGBX_C(type,alpha) \
767 YSCALE_YUV_2_PACKEDX_C(type,alpha) /* FIXME fix tables so that clipping is not needed and then use _NOCLIP*/\
768 r = (type *)c->table_rV[V]; \
769 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
770 b = (type *)c->table_bU[U]; \
771
772 #define YSCALE_YUV_2_PACKED2_C(type,alpha) \
773 for (i=0; i<(dstW>>1); i++) { \
774 const int i2= 2*i; \
775 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
776 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
777 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
778 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
779 type av_unused *r, *b, *g; \
780 int av_unused A1, A2; \
781 if (alpha) {\
782 A1= (abuf0[i2 ]*yalpha1+abuf1[i2 ]*yalpha)>>19; \
783 A2= (abuf0[i2+1]*yalpha1+abuf1[i2+1]*yalpha)>>19; \
784 }\
785
786 #define YSCALE_YUV_2_GRAY16_2_C \
787 for (i=0; i<(dstW>>1); i++) { \
788 const int i2= 2*i; \
789 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
790 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
791
792 #define YSCALE_YUV_2_RGB2_C(type,alpha) \
793 YSCALE_YUV_2_PACKED2_C(type,alpha)\
794 r = (type *)c->table_rV[V];\
795 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
796 b = (type *)c->table_bU[U];\
797
798 #define YSCALE_YUV_2_PACKED1_C(type,alpha) \
799 for (i=0; i<(dstW>>1); i++) {\
800 const int i2= 2*i;\
801 int Y1= buf0[i2 ]>>7;\
802 int Y2= buf0[i2+1]>>7;\
803 int U= (uvbuf1[i ])>>7;\
804 int V= (uvbuf1[i+VOFW])>>7;\
805 type av_unused *r, *b, *g;\
806 int av_unused A1, A2;\
807 if (alpha) {\
808 A1= abuf0[i2 ]>>7;\
809 A2= abuf0[i2+1]>>7;\
810 }\
811
812 #define YSCALE_YUV_2_GRAY16_1_C \
813 for (i=0; i<(dstW>>1); i++) {\
814 const int i2= 2*i;\
815 int Y1= buf0[i2 ]<<1;\
816 int Y2= buf0[i2+1]<<1;\
817
818 #define YSCALE_YUV_2_RGB1_C(type,alpha) \
819 YSCALE_YUV_2_PACKED1_C(type,alpha)\
820 r = (type *)c->table_rV[V];\
821 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
822 b = (type *)c->table_bU[U];\
823
824 #define YSCALE_YUV_2_PACKED1B_C(type,alpha) \
825 for (i=0; i<(dstW>>1); i++) {\
826 const int i2= 2*i;\
827 int Y1= buf0[i2 ]>>7;\
828 int Y2= buf0[i2+1]>>7;\
829 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
830 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
831 type av_unused *r, *b, *g;\
832 int av_unused A1, A2;\
833 if (alpha) {\
834 A1= abuf0[i2 ]>>7;\
835 A2= abuf0[i2+1]>>7;\
836 }\
837
838 #define YSCALE_YUV_2_RGB1B_C(type,alpha) \
839 YSCALE_YUV_2_PACKED1B_C(type,alpha)\
840 r = (type *)c->table_rV[V];\
841 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
842 b = (type *)c->table_bU[U];\
843
844 #define YSCALE_YUV_2_MONO2_C \
845 const uint8_t * const d128=dither_8x8_220[y&7];\
846 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
847 for (i=0; i<dstW-7; i+=8) {\
848 int acc;\
849 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
850 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
851 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
852 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
853 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
854 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
855 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
856 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
857 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
858 dest++;\
859 }\
860
861
862 #define YSCALE_YUV_2_MONOX_C \
863 const uint8_t * const d128=dither_8x8_220[y&7];\
864 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
865 int acc=0;\
866 for (i=0; i<dstW-1; i+=2) {\
867 int j;\
868 int Y1=1<<18;\
869 int Y2=1<<18;\
870 \
871 for (j=0; j<lumFilterSize; j++) {\
872 Y1 += lumSrc[j][i] * lumFilter[j];\
873 Y2 += lumSrc[j][i+1] * lumFilter[j];\
874 }\
875 Y1>>=19;\
876 Y2>>=19;\
877 if ((Y1|Y2)&256) {\
878 if (Y1>255) Y1=255;\
879 else if (Y1<0)Y1=0;\
880 if (Y2>255) Y2=255;\
881 else if (Y2<0)Y2=0;\
882 }\
883 acc+= acc + g[Y1+d128[(i+0)&7]];\
884 acc+= acc + g[Y2+d128[(i+1)&7]];\
885 if ((i&7)==6) {\
886 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
887 dest++;\
888 }\
889 }
890
891
892 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
893 switch(c->dstFormat) {\
894 case PIX_FMT_RGB48BE:\
895 case PIX_FMT_RGB48LE:\
896 func(uint8_t,0)\
897 ((uint8_t*)dest)[ 0]= r[Y1];\
898 ((uint8_t*)dest)[ 1]= r[Y1];\
899 ((uint8_t*)dest)[ 2]= g[Y1];\
900 ((uint8_t*)dest)[ 3]= g[Y1];\
901 ((uint8_t*)dest)[ 4]= b[Y1];\
902 ((uint8_t*)dest)[ 5]= b[Y1];\
903 ((uint8_t*)dest)[ 6]= r[Y2];\
904 ((uint8_t*)dest)[ 7]= r[Y2];\
905 ((uint8_t*)dest)[ 8]= g[Y2];\
906 ((uint8_t*)dest)[ 9]= g[Y2];\
907 ((uint8_t*)dest)[10]= b[Y2];\
908 ((uint8_t*)dest)[11]= b[Y2];\
909 dest+=12;\
910 }\
911 break;\
912 case PIX_FMT_RGBA:\
913 case PIX_FMT_BGRA:\
914 if (CONFIG_SMALL) {\
915 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
916 func(uint32_t,needAlpha)\
917 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? (A1<<24) : 0);\
918 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? (A2<<24) : 0);\
919 }\
920 } else {\
921 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
922 func(uint32_t,1)\
923 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (A1<<24);\
924 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (A2<<24);\
925 }\
926 } else {\
927 func(uint32_t,0)\
928 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
929 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
930 }\
931 }\
932 }\
933 break;\
934 case PIX_FMT_ARGB:\
935 case PIX_FMT_ABGR:\
936 if (CONFIG_SMALL) {\
937 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
938 func(uint32_t,needAlpha)\
939 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? A1 : 0);\
940 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? A2 : 0);\
941 }\
942 } else {\
943 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
944 func(uint32_t,1)\
945 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + A1;\
946 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + A2;\
947 }\
948 } else {\
949 func(uint32_t,0)\
950 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
951 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
952 }\
953 }\
954 } \
955 break;\
956 case PIX_FMT_RGB24:\
957 func(uint8_t,0)\
958 ((uint8_t*)dest)[0]= r[Y1];\
959 ((uint8_t*)dest)[1]= g[Y1];\
960 ((uint8_t*)dest)[2]= b[Y1];\
961 ((uint8_t*)dest)[3]= r[Y2];\
962 ((uint8_t*)dest)[4]= g[Y2];\
963 ((uint8_t*)dest)[5]= b[Y2];\
964 dest+=6;\
965 }\
966 break;\
967 case PIX_FMT_BGR24:\
968 func(uint8_t,0)\
969 ((uint8_t*)dest)[0]= b[Y1];\
970 ((uint8_t*)dest)[1]= g[Y1];\
971 ((uint8_t*)dest)[2]= r[Y1];\
972 ((uint8_t*)dest)[3]= b[Y2];\
973 ((uint8_t*)dest)[4]= g[Y2];\
974 ((uint8_t*)dest)[5]= r[Y2];\
975 dest+=6;\
976 }\
977 break;\
978 case PIX_FMT_RGB565:\
979 case PIX_FMT_BGR565:\
980 {\
981 const int dr1= dither_2x2_8[y&1 ][0];\
982 const int dg1= dither_2x2_4[y&1 ][0];\
983 const int db1= dither_2x2_8[(y&1)^1][0];\
984 const int dr2= dither_2x2_8[y&1 ][1];\
985 const int dg2= dither_2x2_4[y&1 ][1];\
986 const int db2= dither_2x2_8[(y&1)^1][1];\
987 func(uint16_t,0)\
988 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
989 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
990 }\
991 }\
992 break;\
993 case PIX_FMT_RGB555:\
994 case PIX_FMT_BGR555:\
995 {\
996 const int dr1= dither_2x2_8[y&1 ][0];\
997 const int dg1= dither_2x2_8[y&1 ][1];\
998 const int db1= dither_2x2_8[(y&1)^1][0];\
999 const int dr2= dither_2x2_8[y&1 ][1];\
1000 const int dg2= dither_2x2_8[y&1 ][0];\
1001 const int db2= dither_2x2_8[(y&1)^1][1];\
1002 func(uint16_t,0)\
1003 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
1004 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
1005 }\
1006 }\
1007 break;\
1008 case PIX_FMT_RGB8:\
1009 case PIX_FMT_BGR8:\
1010 {\
1011 const uint8_t * const d64= dither_8x8_73[y&7];\
1012 const uint8_t * const d32= dither_8x8_32[y&7];\
1013 func(uint8_t,0)\
1014 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
1015 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
1016 }\
1017 }\
1018 break;\
1019 case PIX_FMT_RGB4:\
1020 case PIX_FMT_BGR4:\
1021 {\
1022 const uint8_t * const d64= dither_8x8_73 [y&7];\
1023 const uint8_t * const d128=dither_8x8_220[y&7];\
1024 func(uint8_t,0)\
1025 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
1026 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
1027 }\
1028 }\
1029 break;\
1030 case PIX_FMT_RGB4_BYTE:\
1031 case PIX_FMT_BGR4_BYTE:\
1032 {\
1033 const uint8_t * const d64= dither_8x8_73 [y&7];\
1034 const uint8_t * const d128=dither_8x8_220[y&7];\
1035 func(uint8_t,0)\
1036 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
1037 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
1038 }\
1039 }\
1040 break;\
1041 case PIX_FMT_MONOBLACK:\
1042 case PIX_FMT_MONOWHITE:\
1043 {\
1044 func_monoblack\
1045 }\
1046 break;\
1047 case PIX_FMT_YUYV422:\
1048 func2\
1049 ((uint8_t*)dest)[2*i2+0]= Y1;\
1050 ((uint8_t*)dest)[2*i2+1]= U;\
1051 ((uint8_t*)dest)[2*i2+2]= Y2;\
1052 ((uint8_t*)dest)[2*i2+3]= V;\
1053 } \
1054 break;\
1055 case PIX_FMT_UYVY422:\
1056 func2\
1057 ((uint8_t*)dest)[2*i2+0]= U;\
1058 ((uint8_t*)dest)[2*i2+1]= Y1;\
1059 ((uint8_t*)dest)[2*i2+2]= V;\
1060 ((uint8_t*)dest)[2*i2+3]= Y2;\
1061 } \
1062 break;\
1063 case PIX_FMT_GRAY16BE:\
1064 func_g16\
1065 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
1066 ((uint8_t*)dest)[2*i2+1]= Y1;\
1067 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
1068 ((uint8_t*)dest)[2*i2+3]= Y2;\
1069 } \
1070 break;\
1071 case PIX_FMT_GRAY16LE:\
1072 func_g16\
1073 ((uint8_t*)dest)[2*i2+0]= Y1;\
1074 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
1075 ((uint8_t*)dest)[2*i2+2]= Y2;\
1076 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
1077 } \
1078 break;\
1079 }\
1080
1081
1082 static inline void yuv2packedXinC(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
1083 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
1084 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
1085 {
1086 int i;
1087 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void,0), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
1088 }
1089
1090 static inline void yuv2rgbXinC_full(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
1091 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
1092 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
1093 {
1094 int i;
1095 int step= fmt_depth(c->dstFormat)/8;
1096 int aidx= 3;
1097
1098 switch(c->dstFormat) {
1099 case PIX_FMT_ARGB:
1100 dest++;
1101 aidx= 0;
1102 case PIX_FMT_RGB24:
1103 aidx--;
1104 case PIX_FMT_RGBA:
1105 if (CONFIG_SMALL) {
1106 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1107 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1108 dest[aidx]= needAlpha ? A : 255;
1109 dest[0]= R>>22;
1110 dest[1]= G>>22;
1111 dest[2]= B>>22;
1112 dest+= step;
1113 }
1114 } else {
1115 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1116 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1117 dest[aidx]= A;
1118 dest[0]= R>>22;
1119 dest[1]= G>>22;
1120 dest[2]= B>>22;
1121 dest+= step;
1122 }
1123 } else {
1124 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1125 dest[aidx]= 255;
1126 dest[0]= R>>22;
1127 dest[1]= G>>22;
1128 dest[2]= B>>22;
1129 dest+= step;
1130 }
1131 }
1132 }
1133 break;
1134 case PIX_FMT_ABGR:
1135 dest++;
1136 aidx= 0;
1137 case PIX_FMT_BGR24:
1138 aidx--;
1139 case PIX_FMT_BGRA:
1140 if (CONFIG_SMALL) {
1141 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1142 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1143 dest[aidx]= needAlpha ? A : 255;
1144 dest[0]= B>>22;
1145 dest[1]= G>>22;
1146 dest[2]= R>>22;
1147 dest+= step;
1148 }
1149 } else {
1150 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1151 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1152 dest[aidx]= A;
1153 dest[0]= B>>22;
1154 dest[1]= G>>22;
1155 dest[2]= R>>22;
1156 dest+= step;
1157 }
1158 } else {
1159 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1160 dest[aidx]= 255;
1161 dest[0]= B>>22;
1162 dest[1]= G>>22;
1163 dest[2]= R>>22;
1164 dest+= step;
1165 }
1166 }
1167 }
1168 break;
1169 default:
1170 assert(0);
1171 }
1172 }
1173
1174 static void fillPlane(uint8_t* plane, int stride, int width, int height, int y, uint8_t val)
1175 {
1176 int i;
1177 uint8_t *ptr = plane + stride*y;
1178 for (i=0; i<height; i++) {
1179 memset(ptr, val, width);
1180 ptr += stride;
1181 }
1182 }
1183
1184 static inline void rgb48ToY(uint8_t *dst, const uint8_t *src, int width)
1185 {
1186 int i;
1187 for (i = 0; i < width; i++) {
1188 int r = src[i*6+0];
1189 int g = src[i*6+2];
1190 int b = src[i*6+4];
1191
1192 dst[i] = (RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1193 }
1194 }
1195
1196 static inline void rgb48ToUV(uint8_t *dstU, uint8_t *dstV,
1197 uint8_t *src1, uint8_t *src2, int width)
1198 {
1199 int i;
1200 assert(src1==src2);
1201 for (i = 0; i < width; i++) {
1202 int r = src1[6*i + 0];
1203 int g = src1[6*i + 2];
1204 int b = src1[6*i + 4];
1205
1206 dstU[i] = (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1207 dstV[i] = (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1208 }
1209 }
1210
1211 static inline void rgb48ToUV_half(uint8_t *dstU, uint8_t *dstV,
1212 uint8_t *src1, uint8_t *src2, int width)
1213 {
1214 int i;
1215 assert(src1==src2);
1216 for (i = 0; i < width; i++) {
1217 int r= src1[12*i + 0] + src1[12*i + 6];
1218 int g= src1[12*i + 2] + src1[12*i + 8];
1219 int b= src1[12*i + 4] + src1[12*i + 10];
1220
1221 dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1222 dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1223 }
1224 }
1225
1226 #define BGR2Y(type, name, shr, shg, shb, maskr, maskg, maskb, RY, GY, BY, S)\
1227 static inline void name(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)\
1228 {\
1229 int i;\
1230 for (i=0; i<width; i++) {\
1231 int b= (((const type*)src)[i]>>shb)&maskb;\
1232 int g= (((const type*)src)[i]>>shg)&maskg;\
1233 int r= (((const type*)src)[i]>>shr)&maskr;\
1234 \
1235 dst[i]= (((RY)*r + (GY)*g + (BY)*b + (33<<((S)-1)))>>(S));\
1236 }\
1237 }
1238
1239 BGR2Y(uint32_t, bgr32ToY,16, 0, 0, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1240 BGR2Y(uint32_t, rgb32ToY, 0, 0,16, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1241 BGR2Y(uint16_t, bgr16ToY, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RY<<11, GY<<5, BY , RGB2YUV_SHIFT+8)
1242 BGR2Y(uint16_t, bgr15ToY, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RY<<10, GY<<5, BY , RGB2YUV_SHIFT+7)
1243 BGR2Y(uint16_t, rgb16ToY, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RY , GY<<5, BY<<11, RGB2YUV_SHIFT+8)
1244 BGR2Y(uint16_t, rgb15ToY, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RY , GY<<5, BY<<10, RGB2YUV_SHIFT+7)
1245
1246 static inline void abgrToA(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1247 {
1248 int i;
1249 for (i=0; i<width; i++) {
1250 dst[i]= src[4*i];
1251 }
1252 }
1253
1254 #define BGR2UV(type, name, shr, shg, shb, maska, maskr, maskg, maskb, RU, GU, BU, RV, GV, BV, S)\
1255 static inline void name(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1256 {\
1257 int i;\
1258 for (i=0; i<width; i++) {\
1259 int b= (((const type*)src)[i]&maskb)>>shb;\
1260 int g= (((const type*)src)[i]&maskg)>>shg;\
1261 int r= (((const type*)src)[i]&maskr)>>shr;\
1262 \
1263 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<((S)-1)))>>(S);\
1264 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<((S)-1)))>>(S);\
1265 }\
1266 }\
1267 static inline void name ## _half(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1268 {\
1269 int i;\
1270 for (i=0; i<width; i++) {\
1271 int pix0= ((const type*)src)[2*i+0];\
1272 int pix1= ((const type*)src)[2*i+1];\
1273 int g= (pix0&~(maskr|maskb))+(pix1&~(maskr|maskb));\
1274 int b= ((pix0+pix1-g)&(maskb|(2*maskb)))>>shb;\
1275 int r= ((pix0+pix1-g)&(maskr|(2*maskr)))>>shr;\
1276 g&= maskg|(2*maskg);\
1277 \
1278 g>>=shg;\
1279 \
1280 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<(S)))>>((S)+1);\
1281 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<(S)))>>((S)+1);\
1282 }\
1283 }
1284
1285 BGR2UV(uint32_t, bgr32ToUV,16, 0, 0, 0xFF000000, 0xFF0000, 0xFF00, 0x00FF, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1286 BGR2UV(uint32_t, rgb32ToUV, 0, 0,16, 0xFF000000, 0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1287 BGR2UV(uint16_t, bgr16ToUV, 0, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RU<<11, GU<<5, BU , RV<<11, GV<<5, BV , RGB2YUV_SHIFT+8)
1288 BGR2UV(uint16_t, bgr15ToUV, 0, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RU<<10, GU<<5, BU , RV<<10, GV<<5, BV , RGB2YUV_SHIFT+7)
1289 BGR2UV(uint16_t, rgb16ToUV, 0, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RU , GU<<5, BU<<11, RV , GV<<5, BV<<11, RGB2YUV_SHIFT+8)
1290 BGR2UV(uint16_t, rgb15ToUV, 0, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RU , GU<<5, BU<<10, RV , GV<<5, BV<<10, RGB2YUV_SHIFT+7)
1291
1292 static inline void palToY(uint8_t *dst, const uint8_t *src, long width, uint32_t *pal)
1293 {
1294 int i;
1295 for (i=0; i<width; i++) {
1296 int d= src[i];
1297
1298 dst[i]= pal[d] & 0xFF;
1299 }
1300 }
1301
1302 static inline void palToUV(uint8_t *dstU, uint8_t *dstV,
1303 const uint8_t *src1, const uint8_t *src2,
1304 long width, uint32_t *pal)
1305 {
1306 int i;
1307 assert(src1 == src2);
1308 for (i=0; i<width; i++) {
1309 int p= pal[src1[i]];
1310
1311 dstU[i]= p>>8;
1312 dstV[i]= p>>16;
1313 }
1314 }
1315
1316 static inline void monowhite2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1317 {
1318 int i, j;
1319 for (i=0; i<width/8; i++) {
1320 int d= ~src[i];
1321 for(j=0; j<8; j++)
1322 dst[8*i+j]= ((d>>(7-j))&1)*255;
1323 }
1324 }
1325
1326 static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1327 {
1328 int i, j;
1329 for (i=0; i<width/8; i++) {
1330 int d= src[i];
1331 for(j=0; j<8; j++)
1332 dst[8*i+j]= ((d>>(7-j))&1)*255;
1333 }
1334 }
1335
1336
1337 //Note: we have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
1338 //Plain C versions
1339 #if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
1340 #define COMPILE_C
1341 #endif
1342
1343 #if ARCH_PPC
1344 #if HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT
1345 #define COMPILE_ALTIVEC
1346 #endif
1347 #endif //ARCH_PPC
1348
1349 #if ARCH_X86
1350
1351 #if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1352 #define COMPILE_MMX
1353 #endif
1354
1355 #if (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1356 #define COMPILE_MMX2
1357 #endif
1358
1359 #if ((HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1360 #define COMPILE_3DNOW
1361 #endif
1362 #endif //ARCH_X86
1363
1364 #define COMPILE_TEMPLATE_MMX 0
1365 #define COMPILE_TEMPLATE_MMX2 0
1366 #define COMPILE_TEMPLATE_AMD3DNOW 0
1367 #define COMPILE_TEMPLATE_ALTIVEC 0
1368
1369 #ifdef COMPILE_C
1370 #define RENAME(a) a ## _C
1371 #include "swscale_template.c"
1372 #endif
1373
1374 #ifdef COMPILE_ALTIVEC
1375 #undef RENAME
1376 #undef COMPILE_TEMPLATE_ALTIVEC
1377 #define COMPILE_TEMPLATE_ALTIVEC 1
1378 #define RENAME(a) a ## _altivec
1379 #include "swscale_template.c"
1380 #endif
1381
1382 #if ARCH_X86
1383
1384 //MMX versions
1385 #ifdef COMPILE_MMX
1386 #undef RENAME
1387 #undef COMPILE_TEMPLATE_MMX
1388 #undef COMPILE_TEMPLATE_MMX2
1389 #undef COMPILE_TEMPLATE_AMD3DNOW
1390 #define COMPILE_TEMPLATE_MMX 1
1391 #define COMPILE_TEMPLATE_MMX2 0
1392 #define COMPILE_TEMPLATE_AMD3DNOW 0
1393 #define RENAME(a) a ## _MMX
1394 #include "swscale_template.c"
1395 #endif
1396
1397 //MMX2 versions
1398 #ifdef COMPILE_MMX2
1399 #undef RENAME
1400 #undef COMPILE_TEMPLATE_MMX
1401 #undef COMPILE_TEMPLATE_MMX2
1402 #undef COMPILE_TEMPLATE_AMD3DNOW
1403 #define COMPILE_TEMPLATE_MMX 1
1404 #define COMPILE_TEMPLATE_MMX2 1
1405 #define COMPILE_TEMPLATE_AMD3DNOW 0
1406 #define RENAME(a) a ## _MMX2
1407 #include "swscale_template.c"
1408 #endif
1409
1410 //3DNOW versions
1411 #ifdef COMPILE_3DNOW
1412 #undef RENAME
1413 #undef COMPILE_TEMPLATE_MMX
1414 #undef COMPILE_TEMPLATE_MMX2
1415 #undef COMPILE_TEMPLATE_AMD3DNOW
1416 #define COMPILE_TEMPLATE_MMX 1
1417 #define COMPILE_TEMPLATE_MMX2 0
1418 #define COMPILE_TEMPLATE_AMD3DNOW 1
1419 #define RENAME(a) a ## _3DNow
1420 #include "swscale_template.c"
1421 #endif
1422
1423 #endif //ARCH_X86
1424
1425 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1426 {
1427 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1428 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1429 else return getSplineCoeff( 0.0,
1430 b+ 2.0*c + 3.0*d,
1431 c + 3.0*d,
1432 -b- 3.0*c - 6.0*d,
1433 dist-1.0);
1434 }
1435
1436 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1437 int srcW, int dstW, int filterAlign, int one, int flags,
1438 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1439 {
1440 int i;
1441 int filterSize;
1442 int filter2Size;
1443 int minFilterSize;
1444 int64_t *filter=NULL;
1445 int64_t *filter2=NULL;
1446 const int64_t fone= 1LL<<54;
1447 int ret= -1;
1448 #if ARCH_X86
1449 if (flags & SWS_CPU_CAPS_MMX)
1450 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1451 #endif
1452
1453 // NOTE: the +1 is for the MMX scaler which reads over the end
1454 CHECKED_ALLOC(*filterPos, (dstW+1)*sizeof(int16_t));
1455
1456 if (FFABS(xInc - 0x10000) <10) { // unscaled
1457 int i;
1458 filterSize= 1;
1459 CHECKED_ALLOCZ(filter, dstW*sizeof(*filter)*filterSize);
1460
1461 for (i=0; i<dstW; i++) {
1462 filter[i*filterSize]= fone;
1463 (*filterPos)[i]=i;
1464 }
1465
1466 } else if (flags&SWS_POINT) { // lame looking point sampling mode
1467 int i;
1468 int xDstInSrc;
1469 filterSize= 1;
1470 CHECKED_ALLOC(filter, dstW*sizeof(*filter)*filterSize);
1471
1472 xDstInSrc= xInc/2 - 0x8000;
1473 for (i=0; i<dstW; i++) {
1474 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1475
1476 (*filterPos)[i]= xx;
1477 filter[i]= fone;
1478 xDstInSrc+= xInc;
1479 }
1480 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
1481 int i;
1482 int xDstInSrc;
1483 filterSize= 2;
1484 CHECKED_ALLOC(filter, dstW*sizeof(*filter)*filterSize);
1485
1486 xDstInSrc= xInc/2 - 0x8000;
1487 for (i=0; i<dstW; i++) {
1488 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1489 int j;
1490
1491 (*filterPos)[i]= xx;
1492 //bilinear upscale / linear interpolate / area averaging
1493 for (j=0; j<filterSize; j++) {
1494 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1495 if (coeff<0) coeff=0;
1496 filter[i*filterSize + j]= coeff;
1497 xx++;
1498 }
1499 xDstInSrc+= xInc;
1500 }
1501 } else {
1502 int xDstInSrc;
1503 int sizeFactor;
1504
1505 if (flags&SWS_BICUBIC) sizeFactor= 4;
1506 else if (flags&SWS_X) sizeFactor= 8;
1507 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1508 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1509 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1510 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1511 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1512 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1513 else {
1514 sizeFactor= 0; //GCC warning killer
1515 assert(0);
1516 }
1517
1518 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1519 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1520
1521 if (filterSize > srcW-2) filterSize=srcW-2;
1522
1523 CHECKED_ALLOC(filter, dstW*sizeof(*filter)*filterSize);
1524
1525 xDstInSrc= xInc - 0x10000;
1526 for (i=0; i<dstW; i++) {
1527 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1528 int j;
1529 (*filterPos)[i]= xx;
1530 for (j=0; j<filterSize; j++) {
1531 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1532 double floatd;
1533 int64_t coeff;
1534
1535 if (xInc > 1<<16)
1536 d= d*dstW/srcW;
1537 floatd= d * (1.0/(1<<30));
1538
1539 if (flags & SWS_BICUBIC) {
1540 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1541 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1542 int64_t dd = ( d*d)>>30;
1543 int64_t ddd= (dd*d)>>30;
1544
1545 if (d < 1LL<<30)
1546 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1547 else if (d < 1LL<<31)
1548 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1549 else
1550 coeff=0.0;
1551 coeff *= fone>>(30+24);
1552 }
1553 /* else if (flags & SWS_X) {
1554 double p= param ? param*0.01 : 0.3;
1555 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1556 coeff*= pow(2.0, - p*d*d);
1557 }*/
1558 else if (flags & SWS_X) {
1559 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1560 double c;
1561
1562 if (floatd<1.0)
1563 c = cos(floatd*PI);
1564 else
1565 c=-1.0;
1566 if (c<0.0) c= -pow(-c, A);
1567 else c= pow( c, A);
1568 coeff= (c*0.5 + 0.5)*fone;
1569 } else if (flags & SWS_AREA) {
1570 int64_t d2= d - (1<<29);
1571 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1572 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1573 else coeff=0.0;
1574 coeff *= fone>>(30+16);
1575 } else if (flags & SWS_GAUSS) {
1576 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1577 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1578 } else if (flags & SWS_SINC) {
1579 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1580 } else if (flags & SWS_LANCZOS) {
1581 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1582 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1583 if (floatd>p) coeff=0;
1584 } else if (flags & SWS_BILINEAR) {
1585 coeff= (1<<30) - d;
1586 if (coeff<0) coeff=0;
1587 coeff *= fone >> 30;
1588 } else if (flags & SWS_SPLINE) {
1589 double p=-2.196152422706632;
1590 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1591 } else {
1592 coeff= 0.0; //GCC warning killer
1593 assert(0);
1594 }
1595
1596 filter[i*filterSize + j]= coeff;
1597 xx++;
1598 }
1599 xDstInSrc+= 2*xInc;
1600 }
1601 }
1602
1603 /* apply src & dst Filter to filter -> filter2
1604 av_free(filter);
1605 */
1606 assert(filterSize>0);
1607 filter2Size= filterSize;
1608 if (srcFilter) filter2Size+= srcFilter->length - 1;
1609 if (dstFilter) filter2Size+= dstFilter->length - 1;
1610 assert(filter2Size>0);
1611 CHECKED_ALLOCZ(filter2, filter2Size*dstW*sizeof(*filter2));
1612
1613 for (i=0; i<dstW; i++) {
1614 int j, k;
1615
1616 if(srcFilter) {
1617 for (k=0; k<srcFilter->length; k++) {
1618 for (j=0; j<filterSize; j++)
1619 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1620 }
1621 } else {
1622 for (j=0; j<filterSize; j++)
1623 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1624 }
1625 //FIXME dstFilter
1626
1627 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1628 }
1629 av_freep(&filter);
1630
1631 /* try to reduce the filter-size (step1 find size and shift left) */
1632 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1633 minFilterSize= 0;
1634 for (i=dstW-1; i>=0; i--) {
1635 int min= filter2Size;
1636 int j;
1637 int64_t cutOff=0.0;
1638
1639 /* get rid off near zero elements on the left by shifting left */
1640 for (j=0; j<filter2Size; j++) {
1641 int k;
1642 cutOff += FFABS(filter2[i*filter2Size]);
1643
1644 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1645
1646 /* preserve monotonicity because the core can't handle the filter otherwise */
1647 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1648
1649 // move filter coefficients left
1650 for (k=1; k<filter2Size; k++)
1651 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1652 filter2[i*filter2Size + k - 1]= 0;
1653 (*filterPos)[i]++;
1654 }
1655
1656 cutOff=0;
1657 /* count near zeros on the right */
1658 for (j=filter2Size-1; j>0; j--) {
1659 cutOff += FFABS(filter2[i*filter2Size + j]);
1660
1661 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1662 min--;
1663 }
1664
1665 if (min>minFilterSize) minFilterSize= min;
1666 }
1667
1668 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1669 // we can handle the special case 4,
1670 // so we don't want to go to the full 8
1671 if (minFilterSize < 5)
1672 filterAlign = 4;
1673
1674 // We really don't want to waste our time
1675 // doing useless computation, so fall back on
1676 // the scalar C code for very small filters.
1677 // Vectorizing is worth it only if you have a
1678 // decent-sized vector.
1679 if (minFilterSize < 3)
1680 filterAlign = 1;
1681 }
1682
1683 if (flags & SWS_CPU_CAPS_MMX) {
1684 // special case for unscaled vertical filtering
1685 if (minFilterSize == 1 && filterAlign == 2)
1686 filterAlign= 1;
1687 }
1688
1689 assert(minFilterSize > 0);
1690 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1691 assert(filterSize > 0);
1692 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1693 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1694 goto fail;
1695 *outFilterSize= filterSize;
1696
1697 if (flags&SWS_PRINT_INFO)
1698 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1699 /* try to reduce the filter-size (step2 reduce it) */
1700 for (i=0; i<dstW; i++) {
1701 int j;
1702
1703 for (j=0; j<filterSize; j++) {
1704 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1705 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1706 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1707 filter[i*filterSize + j]= 0;
1708 }
1709 }
1710
1711
1712 //FIXME try to align filterPos if possible
1713
1714 //fix borders
1715 for (i=0; i<dstW; i++) {
1716 int j;
1717 if ((*filterPos)[i] < 0) {
1718 // move filter coefficients left to compensate for filterPos
1719 for (j=1; j<filterSize; j++) {
1720 int left= FFMAX(j + (*filterPos)[i], 0);
1721 filter[i*filterSize + left] += filter[i*filterSize + j];
1722 filter[i*filterSize + j]=0;
1723 }
1724 (*filterPos)[i]= 0;
1725 }
1726
1727 if ((*filterPos)[i] + filterSize > srcW) {
1728 int shift= (*filterPos)[i] + filterSize - srcW;
1729 // move filter coefficients right to compensate for filterPos
1730 for (j=filterSize-2; j>=0; j--) {
1731 int right= FFMIN(j + shift, filterSize-1);
1732 filter[i*filterSize +right] += filter[i*filterSize +j];
1733 filter[i*filterSize +j]=0;
1734 }
1735 (*filterPos)[i]= srcW - filterSize;
1736 }
1737 }
1738
1739 // Note the +1 is for the MMX scaler which reads over the end
1740 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1741 CHECKED_ALLOCZ(*outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t));
1742
1743 /* normalize & store in outFilter */
1744 for (i=0; i<dstW; i++) {
1745 int j;
1746 int64_t error=0;
1747 int64_t sum=0;
1748
1749 for (j=0; j<filterSize; j++) {
1750 sum+= filter[i*filterSize + j];
1751 }
1752 sum= (sum + one/2)/ one;
1753 for (j=0; j<*outFilterSize; j++) {
1754 int64_t v= filter[i*filterSize + j] + error;
1755 int intV= ROUNDED_DIV(v, sum);
1756 (*outFilter)[i*(*outFilterSize) + j]= intV;
1757 error= v - intV*sum;
1758 }
1759 }
1760
1761 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1762 for (i=0; i<*outFilterSize; i++) {
1763 int j= dstW*(*outFilterSize);
1764 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1765 }
1766
1767 ret=0;
1768 fail:
1769 av_free(filter);
1770 av_free(filter2);
1771 return ret;
1772 }
1773
1774 #ifdef COMPILE_MMX2
1775 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
1776 {
1777 uint8_t *fragmentA;
1778 x86_reg imm8OfPShufW1A;
1779 x86_reg imm8OfPShufW2A;
1780 x86_reg fragmentLengthA;
1781 uint8_t *fragmentB;
1782 x86_reg imm8OfPShufW1B;
1783 x86_reg imm8OfPShufW2B;
1784 x86_reg fragmentLengthB;
1785 int fragmentPos;
1786
1787 int xpos, i;
1788
1789 // create an optimized horizontal scaling routine
1790
1791 //code fragment
1792
1793 __asm__ volatile(
1794 "jmp 9f \n\t"
1795 // Begin
1796 "0: \n\t"
1797 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1798 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1799 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1800 "punpcklbw %%mm7, %%mm1 \n\t"
1801 "punpcklbw %%mm7, %%mm0 \n\t"
1802 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1803 "1: \n\t"
1804 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1805 "2: \n\t"
1806 "psubw %%mm1, %%mm0 \n\t"
1807 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1808 "pmullw %%mm3, %%mm0 \n\t"
1809 "psllw $7, %%mm1 \n\t"
1810 "paddw %%mm1, %%mm0 \n\t"
1811
1812 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1813
1814 "add $8, %%"REG_a" \n\t"
1815 // End
1816 "9: \n\t"
1817 // "int $3 \n\t"
1818 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1819 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1820 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1821 "dec %1 \n\t"
1822 "dec %2 \n\t"
1823 "sub %0, %1 \n\t"
1824 "sub %0, %2 \n\t"
1825 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1826 "sub %0, %3 \n\t"
1827
1828
1829 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1830 "=r" (fragmentLengthA)
1831 );
1832
1833 __asm__ volatile(
1834 "jmp 9f \n\t"
1835 // Begin
1836 "0: \n\t"
1837 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1838 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1839 "punpcklbw %%mm7, %%mm0 \n\t"
1840 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1841 "1: \n\t"
1842 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1843 "2: \n\t"
1844 "psubw %%mm1, %%mm0 \n\t"
1845 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1846 "pmullw %%mm3, %%mm0 \n\t"
1847 "psllw $7, %%mm1 \n\t"
1848 "paddw %%mm1, %%mm0 \n\t"
1849
1850 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1851
1852 "add $8, %%"REG_a" \n\t"
1853 // End
1854 "9: \n\t"
1855 // "int $3 \n\t"
1856 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1857 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1858 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1859 "dec %1 \n\t"
1860 "dec %2 \n\t"
1861 "sub %0, %1 \n\t"
1862 "sub %0, %2 \n\t"
1863 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1864 "sub %0, %3 \n\t"
1865
1866
1867 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1868 "=r" (fragmentLengthB)
1869 );
1870
1871 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1872 fragmentPos=0;
1873
1874 for (i=0; i<dstW/numSplits; i++) {
1875 int xx=xpos>>16;
1876
1877 if ((i&3) == 0) {
1878 int a=0;
1879 int b=((xpos+xInc)>>16) - xx;
1880 int c=((xpos+xInc*2)>>16) - xx;
1881 int d=((xpos+xInc*3)>>16) - xx;
1882 int inc = (d+1<4);
1883 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
1884 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
1885 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
1886 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
1887 int maxShift= 3-(d+inc);
1888 int shift=0;
1889
1890 if (filterCode) {
1891 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1892 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1893 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1894 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1895 filterPos[i/2]= xx;
1896
1897 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
1898
1899 filterCode[fragmentPos + imm8OfPShufW1]=
1900 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
1901 filterCode[fragmentPos + imm8OfPShufW2]=
1902 a | (b<<2) | (c<<4) | (d<<6);
1903
1904 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
1905 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1906
1907 if (shift && i>=shift) {
1908 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
1909 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
1910 filterPos[i/2]-=shift;
1911 }
1912 }
1913
1914 fragmentPos+= fragmentLength;
1915
1916 if (filterCode)
1917 filterCode[fragmentPos]= RET;
1918 }
1919 xpos+=xInc;
1920 }
1921 if (filterCode)
1922 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
1923
1924 return fragmentPos + 1;
1925 }
1926 #endif /* COMPILE_MMX2 */
1927
1928 static void globalInit(void)
1929 {
1930 // generating tables:
1931 int i;
1932 for (i=0; i<768; i++) {
1933 int c= av_clip_uint8(i-256);
1934 clip_table[i]=c;
1935 }
1936 }
1937
1938 static SwsFunc getSwsFunc(SwsContext *c)
1939 {
1940 #if CONFIG_RUNTIME_CPUDETECT
1941 int flags = c->flags;
1942
1943 #if ARCH_X86 && CONFIG_GPL
1944 // ordered per speed fastest first
1945 if (flags & SWS_CPU_CAPS_MMX2) {
1946 sws_init_swScale_MMX2(c);
1947 return swScale_MMX2;
1948 } else if (flags & SWS_CPU_CAPS_3DNOW) {
1949 sws_init_swScale_3DNow(c);
1950 return swScale_3DNow;
1951 } else if (flags & SWS_CPU_CAPS_MMX) {
1952 sws_init_swScale_MMX(c);
1953 return swScale_MMX;
1954 } else {
1955 sws_init_swScale_C(c);
1956 return swScale_C;
1957 }
1958
1959 #else
1960 #if ARCH_PPC
1961 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1962 sws_init_swScale_altivec(c);
1963 return swScale_altivec;
1964 } else {
1965 sws_init_swScale_C(c);
1966 return swScale_C;
1967 }
1968 #endif
1969 sws_init_swScale_C(c);
1970 return swScale_C;
1971 #endif /* ARCH_X86 && CONFIG_GPL */
1972 #else //CONFIG_RUNTIME_CPUDETECT
1973 #if COMPILE_TEMPLATE_MMX2
1974 sws_init_swScale_MMX2(c);
1975 return swScale_MMX2;
1976 #elif COMPILE_TEMPLATE_AMD3DNOW
1977 sws_init_swScale_3DNow(c);
1978 return swScale_3DNow;
1979 #elif COMPILE_TEMPLATE_MMX
1980 sws_init_swScale_MMX(c);
1981 return swScale_MMX;
1982 #elif COMPILE_TEMPLATE_ALTIVEC
1983 sws_init_swScale_altivec(c);
1984 return swScale_altivec;
1985 #else
1986 sws_init_swScale_C(c);
1987 return swScale_C;
1988 #endif
1989 #endif //!CONFIG_RUNTIME_CPUDETECT
1990 }
1991
1992 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1993 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1994 {
1995 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1996 /* Copy Y plane */
1997 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1998 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1999 else {
2000 int i;
2001 const uint8_t *srcPtr= src[0];
2002 uint8_t *dstPtr= dst;
2003 for (i=0; i<srcSliceH; i++) {
2004 memcpy(dstPtr, srcPtr, c->srcW);
2005 srcPtr+= srcStride[0];
2006 dstPtr+= dstStride[0];
2007 }
2008 }
2009 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
2010 if (c->dstFormat == PIX_FMT_NV12)
2011 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
2012 else
2013 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
2014
2015 return srcSliceH;
2016 }
2017
2018 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2019 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2020 {
2021 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
2022
2023 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
2024
2025 return srcSliceH;
2026 }
2027
2028 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2029 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2030 {
2031 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
2032
2033 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
2034
2035 return srcSliceH;
2036 }
2037
2038 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2039 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2040 {
2041 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
2042
2043 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
2044
2045 return srcSliceH;
2046 }
2047
2048 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2049 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2050 {
2051 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
2052
2053 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
2054
2055 return srcSliceH;
2056 }
2057
2058 static int YUYV2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2059 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2060 {
2061 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2062 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
2063 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
2064
2065 yuyvtoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2066
2067 if (dstParam[3])
2068 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2069
2070 return srcSliceH;
2071 }
2072
2073 static int YUYV2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2074 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2075 {
2076 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2077 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
2078 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
2079
2080 yuyvtoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2081
2082 return srcSliceH;
2083 }
2084
2085 static int UYVY2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2086 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2087 {
2088 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2089 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
2090 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
2091
2092 uyvytoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2093
2094 if (dstParam[3])
2095 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2096
2097 return srcSliceH;
2098 }
2099
2100 static int UYVY2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2101 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2102 {
2103 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2104 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
2105 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
2106
2107 uyvytoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2108
2109 return srcSliceH;
2110 }
2111
2112 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2113 int srcSliceH, uint8_t* dst[], int dstStride[])
2114 {
2115 const enum PixelFormat srcFormat= c->srcFormat;
2116 const enum PixelFormat dstFormat= c->dstFormat;
2117 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
2118 const uint8_t *palette)=NULL;
2119 int i;
2120 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2121 uint8_t *srcPtr= src[0];
2122
2123 if (!usePal(srcFormat))
2124 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2125 sws_format_name(srcFormat), sws_format_name(dstFormat));
2126
2127 switch(dstFormat) {
2128 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
2129 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
2130 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
2131 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
2132 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
2133 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
2134 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2135 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2136 }
2137
2138
2139 for (i=0; i<srcSliceH; i++) {
2140 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
2141 srcPtr+= srcStride[0];
2142 dstPtr+= dstStride[0];
2143 }
2144
2145 return srcSliceH;
2146 }
2147
2148 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
2149 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2150 int srcSliceH, uint8_t* dst[], int dstStride[])
2151 {
2152 const enum PixelFormat srcFormat= c->srcFormat;
2153 const enum PixelFormat dstFormat= c->dstFormat;
2154 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
2155 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
2156 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
2157 const int dstId= fmt_depth(dstFormat) >> 2;
2158 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
2159
2160 /* BGR -> BGR */
2161 if ( (isBGR(srcFormat) && isBGR(dstFormat))
2162 || (isRGB(srcFormat) && isRGB(dstFormat))) {
2163 switch(srcId | (dstId<<4)) {
2164 case 0x34: conv= rgb16to15; break;
2165 case 0x36: conv= rgb24to15; break;
2166 case 0x38: conv= rgb32to15; break;
2167 case 0x43: conv= rgb15to16; break;
2168 case 0x46: conv= rgb24to16; break;
2169 case 0x48: conv= rgb32to16; break;
2170 case 0x63: conv= rgb15to24; break;
2171 case 0x64: conv= rgb16to24; break;
2172 case 0x68: conv= rgb32to24; break;
2173 case 0x83: conv= rgb15to32; break;
2174 case 0x84: conv= rgb16to32; break;
2175 case 0x86: conv= rgb24to32; break;
2176 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2177 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2178 }
2179 } else if ( (isBGR(srcFormat) && isRGB(dstFormat))
2180 || (isRGB(srcFormat) && isBGR(dstFormat))) {
2181 switch(srcId | (dstId<<4)) {
2182 case 0x33: conv= rgb15tobgr15; break;
2183 case 0x34: conv= rgb16tobgr15; break;
2184 case 0x36: conv= rgb24tobgr15; break;
2185 case 0x38: conv= rgb32tobgr15; break;
2186 case 0x43: conv= rgb15tobgr16; break;
2187 case 0x44: conv= rgb16tobgr16; break;
2188 case 0x46: conv= rgb24tobgr16; break;
2189 case 0x48: conv= rgb32tobgr16; break;
2190 case 0x63: conv= rgb15tobgr24; break;
2191 case 0x64: conv= rgb16tobgr24; break;
2192 case 0x66: conv= rgb24tobgr24; break;
2193 case 0x68: conv= rgb32tobgr24; break;
2194 case 0x83: conv= rgb15tobgr32; break;
2195 case 0x84: conv= rgb16tobgr32; break;
2196 case 0x86: conv= rgb24tobgr32; break;
2197 case 0x88: conv= rgb32tobgr32; break;
2198 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2199 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2200 }
2201 } else {
2202 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2203 sws_format_name(srcFormat), sws_format_name(dstFormat));
2204 }
2205
2206 if(conv) {
2207 uint8_t *srcPtr= src[0];
2208 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
2209 srcPtr += ALT32_CORR;
2210
2211 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
2212 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
2213 else {
2214 int i;
2215 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2216
2217 for (i=0; i<srcSliceH; i++) {
2218 conv(srcPtr, dstPtr, c->srcW*srcBpp);
2219 srcPtr+= srcStride[0];
2220 dstPtr+= dstStride[0];
2221 }
2222 }
2223 }
2224 return srcSliceH;
2225 }
2226
2227 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2228 int srcSliceH, uint8_t* dst[], int dstStride[])
2229 {
2230
2231 rgb24toyv12(
2232 src[0],
2233 dst[0]+ srcSliceY *dstStride[0],
2234 dst[1]+(srcSliceY>>1)*dstStride[1],
2235 dst[2]+(srcSliceY>>1)*dstStride[2],
2236 c->srcW, srcSliceH,
2237 dstStride[0], dstStride[1], srcStride[0]);
2238 if (dst[3])
2239 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2240 return srcSliceH;
2241 }
2242
2243 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2244 int srcSliceH, uint8_t* dst[], int dstStride[])
2245 {
2246 int i;
2247
2248 /* copy Y */
2249 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
2250 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
2251 else {
2252 uint8_t *srcPtr= src[0];
2253 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2254
2255 for (i=0; i<srcSliceH; i++) {
2256 memcpy(dstPtr, srcPtr, c->srcW);
2257 srcPtr+= srcStride[0];
2258 dstPtr+= dstStride[0];
2259 }
2260 }
2261
2262 if (c->dstFormat==PIX_FMT_YUV420P || c->dstFormat==PIX_FMT_YUVA420P) {
2263 planar2x(src[1], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2264 srcSliceH >> 2, srcStride[1], dstStride[1]);
2265 planar2x(src[2], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2266 srcSliceH >> 2, srcStride[2], dstStride[2]);
2267 } else {
2268 planar2x(src[1], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2269 srcSliceH >> 2, srcStride[1], dstStride[2]);
2270 planar2x(src[2], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2271 srcSliceH >> 2, srcStride[2], dstStride[1]);
2272 }
2273 if (dst[3])
2274 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2275 return srcSliceH;
2276 }
2277
2278 /* unscaled copy like stuff (assumes nearly identical formats) */
2279 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2280 int srcSliceH, uint8_t* dst[], int dstStride[])
2281 {
2282 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
2283 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
2284 else {
2285 int i;
2286 uint8_t *srcPtr= src[0];
2287 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2288 int length=0;
2289
2290 /* universal length finder */
2291 while(length+c->srcW <= FFABS(dstStride[0])
2292 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
2293 assert(length!=0);
2294
2295 for (i=0; i<srcSliceH; i++) {
2296 memcpy(dstPtr, srcPtr, length);
2297 srcPtr+= srcStride[0];
2298 dstPtr+= dstStride[0];
2299 }
2300 }
2301 return srcSliceH;
2302 }
2303
2304 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2305 int srcSliceH, uint8_t* dst[], int dstStride[])
2306 {
2307 int plane, i, j;
2308 for (plane=0; plane<4; plane++) {
2309 int length= (plane==0 || plane==3) ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
2310 int y= (plane==0 || plane==3) ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
2311 int height= (plane==0 || plane==3) ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
2312 uint8_t *srcPtr= src[plane];
2313 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
2314
2315 if (!dst[plane]) continue;
2316 // ignore palette for GRAY8
2317 if (plane == 1 && !dst[2]) continue;
2318 if (!src[plane] || (plane == 1 && !src[2])) {
2319 if(is16BPS(c->dstFormat))
2320 length*=2;
2321 fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
2322 } else {
2323 if(is16BPS(c->srcFormat) && !is16BPS(c->dstFormat)) {
2324 if (!isBE(c->srcFormat)) srcPtr++;
2325 for (i=0; i<height; i++) {
2326 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
2327 srcPtr+= srcStride[plane];
2328 dstPtr+= dstStride[plane];
2329 }
2330 } else if(!is16BPS(c->srcFormat) && is16BPS(c->dstFormat)) {
2331 for (i=0; i<height; i++) {
2332 for (j=0; j<length; j++) {
2333 dstPtr[ j<<1 ] = srcPtr[j];
2334 dstPtr[(j<<1)+1] = srcPtr[j];
2335 }
2336 srcPtr+= srcStride[plane];
2337 dstPtr+= dstStride[plane];
2338 }
2339 } else if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat)
2340 && isBE(c->srcFormat) != isBE(c->dstFormat)) {
2341
2342 for (i=0; i<height; i++) {
2343 for (j=0; j<length; j++)
2344 ((uint16_t*)dstPtr)[j] = bswap_16(((uint16_t*)srcPtr)[j]);
2345 srcPtr+= srcStride[plane];
2346 dstPtr+= dstStride[plane];
2347 }
2348 } else if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
2349 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
2350 else {
2351 if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat))
2352 length*=2;
2353 for (i=0; i<height; i++) {
2354 memcpy(dstPtr, srcPtr, length);
2355 srcPtr+= srcStride[plane];
2356 dstPtr+= dstStride[plane];
2357 }
2358 }
2359 }
2360 }
2361 return srcSliceH;
2362 }
2363
2364
2365 static void getSubSampleFactors(int *h, int *v, int format)
2366 {
2367 switch(format) {
2368 case PIX_FMT_UYVY422:
2369 case PIX_FMT_YUYV422:
2370 *h=1;
2371 *v=0;
2372 break;
2373 case PIX_FMT_YUV420P:
2374 case PIX_FMT_YUV420PLE:
2375 case PIX_FMT_YUV420PBE:
2376 case PIX_FMT_YUVA420P:
2377 case PIX_FMT_GRAY16BE:
2378 case PIX_FMT_GRAY16LE:
2379 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
2380 case PIX_FMT_NV12:
2381 case PIX_FMT_NV21:
2382 *h=1;
2383 *v=1;
2384 break;
2385 case PIX_FMT_YUV440P:
2386 *h=0;
2387 *v=1;
2388 break;
2389 case PIX_FMT_YUV410P:
2390 *h=2;
2391 *v=2;
2392 break;
2393 case PIX_FMT_YUV444P:
2394 case PIX_FMT_YUV444PLE:
2395 case PIX_FMT_YUV444PBE:
2396 *h=0;
2397 *v=0;
2398 break;
2399 case PIX_FMT_YUV422P:
2400 case PIX_FMT_YUV422PLE:
2401 case PIX_FMT_YUV422PBE:
2402 *h=1;
2403 *v=0;
2404 break;
2405 case PIX_FMT_YUV411P:
2406 *h=2;
2407 *v=0;
2408 break;
2409 default:
2410 *h=0;
2411 *v=0;
2412 break;
2413 }
2414 }
2415
2416 static uint16_t roundToInt16(int64_t f)
2417 {
2418 int r= (f + (1<<15))>>16;
2419 if (r<-0x7FFF) return 0x8000;
2420 else if (r> 0x7FFF) return 0x7FFF;
2421 else return r;
2422 }
2423
2424 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
2425 {
2426 int64_t crv = inv_table[0];
2427 int64_t cbu = inv_table[1];
2428 int64_t cgu = -inv_table[2];
2429 int64_t cgv = -inv_table[3];
2430 int64_t cy = 1<<16;
2431 int64_t oy = 0;
2432
2433 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2434 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2435
2436 c->brightness= brightness;
2437 c->contrast = contrast;
2438 c->saturation= saturation;
2439 c->srcRange = srcRange;
2440 c->dstRange = dstRange;
2441 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2442
2443 c->uOffset= 0x0400040004000400LL;
2444 c->vOffset= 0x0400040004000400LL;
2445
2446 if (!srcRange) {
2447 cy= (cy*255) / 219;
2448 oy= 16<<16;
2449 } else {
2450 crv= (crv*224) / 255;
2451 cbu= (cbu*224) / 255;
2452 cgu= (cgu*224) / 255;
2453 cgv= (cgv*224) / 255;
2454 }
2455
2456 cy = (cy *contrast )>>16;
2457 crv= (crv*contrast * saturation)>>32;
2458 cbu= (cbu*contrast * saturation)>>32;
2459 cgu= (cgu*contrast * saturation)>>32;
2460 cgv= (cgv*contrast * saturation)>>32;
2461
2462 oy -= 256*brightness;
2463
2464 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2465 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2466 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2467 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2468 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2469 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2470
2471 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2472 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2473 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2474 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2475 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2476 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2477
2478 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2479 //FIXME factorize
2480
2481 #ifdef COMPILE_ALTIVEC
2482 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2483 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
2484 #endif
2485 return 0;
2486 }
2487
2488 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
2489 {
2490 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2491
2492 *inv_table = c->srcColorspaceTable;
2493 *table = c->dstColorspaceTable;
2494 *srcRange = c->srcRange;
2495 *dstRange = c->dstRange;
2496 *brightness= c->brightness;
2497 *contrast = c->contrast;
2498 *saturation= c->saturation;
2499
2500 return 0;
2501 }
2502
2503 static int handle_jpeg(enum PixelFormat *format)
2504 {
2505 switch (*format) {
2506 case PIX_FMT_YUVJ420P:
2507 *format = PIX_FMT_YUV420P;
2508 return 1;
2509 case PIX_FMT_YUVJ422P:
2510 *format = PIX_FMT_YUV422P;
2511 return 1;
2512 case PIX_FMT_YUVJ444P:
2513 *format = PIX_FMT_YUV444P;
2514 return 1;
2515 case PIX_FMT_YUVJ440P:
2516 *format = PIX_FMT_YUV440P;
2517 return 1;
2518 default:
2519 return 0;
2520 }
2521 }
2522
2523 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2524 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
2525 {
2526
2527 SwsContext *c;
2528 int i;
2529 int usesVFilter, usesHFilter;
2530 int unscaled, needsDither;
2531 int srcRange, dstRange;
2532 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2533 #if ARCH_X86
2534 if (flags & SWS_CPU_CAPS_MMX)
2535 __asm__ volatile("emms\n\t"::: "memory");
2536 #endif
2537
2538 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
2539 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2540 #if COMPILE_TEMPLATE_MMX2
2541 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2542 #elif COMPILE_TEMPLATE_AMD3DNOW
2543 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2544 #elif COMPILE_TEMPLATE_MMX
2545 flags |= SWS_CPU_CAPS_MMX;
2546 #elif COMPILE_TEMPLATE_ALTIVEC
2547 flags |= SWS_CPU_CAPS_ALTIVEC;
2548 #elif ARCH_BFIN
2549 flags |= SWS_CPU_CAPS_BFIN;
2550 #endif
2551 #endif /* CONFIG_RUNTIME_CPUDETECT */
2552 if (clip_table[512] != 255) globalInit();
2553 if (!rgb15to16) sws_rgb2rgb_init(flags);
2554
2555 unscaled = (srcW == dstW && srcH == dstH);
2556 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2557 && (fmt_depth(dstFormat))<24
2558 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2559
2560 srcRange = handle_jpeg(&srcFormat);
2561 dstRange = handle_jpeg(&dstFormat);
2562
2563 if (!isSupportedIn(srcFormat)) {
2564 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2565 return NULL;
2566 }
2567 if (!isSupportedOut(dstFormat)) {
2568 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2569 return NULL;
2570 }
2571
2572 i= flags & ( SWS_POINT
2573 |SWS_AREA
2574 |SWS_BILINEAR
2575 |SWS_FAST_BILINEAR
2576 |SWS_BICUBIC
2577 |SWS_X
2578 |SWS_GAUSS
2579 |SWS_LANCZOS
2580 |SWS_SINC
2581 |SWS_SPLINE
2582 |SWS_BICUBLIN);
2583 if(!i || (i & (i-1))) {
2584 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
2585 return NULL;
2586 }
2587
2588 /* sanity check */
2589 if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
2590 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2591 srcW, srcH, dstW, dstH);
2592 return NULL;
2593 }
2594 if(srcW > VOFW || dstW > VOFW) {
2595 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2596 return NULL;
2597 }
2598
2599 if (!dstFilter) dstFilter= &dummyFilter;
2600 if (!srcFilter) srcFilter= &dummyFilter;
2601
2602 CHECKED_ALLOCZ(c, sizeof(SwsContext));
2603
2604 c->av_class = &sws_context_class;
2605 c->srcW= srcW;
2606 c->srcH= srcH;
2607 c->dstW= dstW;
2608 c->dstH= dstH;
2609 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2610 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2611 c->flags= flags;
2612 c->dstFormat= dstFormat;
2613 c->srcFormat= srcFormat;
2614 c->vRounder= 4* 0x0001000100010001ULL;
2615
2616 usesHFilter= usesVFilter= 0;
2617 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2618 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2619 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2620 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2621 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2622 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2623 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2624 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2625
2626 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2627 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2628
2629 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
2630 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2631
2632 // drop some chroma lines if the user wants it
2633 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2634 c->chrSrcVSubSample+= c->vChrDrop;
2635
2636 // drop every other pixel for chroma calculation unless user wants full chroma
2637 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2638 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2639 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2640 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2641 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2642 c->chrSrcHSubSample=1;
2643
2644 if (param) {
2645 c->param[0] = param[0];
2646 c->param[1] = param[1];
2647 } else {
2648 c->param[0] =
2649 c->param[1] = SWS_PARAM_DEFAULT;
2650 }
2651
2652 // Note the -((-x)>>y) is so that we always round toward +inf.
2653 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2654 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2655 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2656 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2657
2658 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2659
2660 /* unscaled special cases */
2661 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat))) {
2662 /* yv12_to_nv12 */
2663 if ((srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21)) {
2664 c->swScale= PlanarToNV12Wrapper;
2665 }
2666 /* yuv2bgr */
2667 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P || srcFormat==PIX_FMT_YUVA420P) && (isBGR(dstFormat) || isRGB(dstFormat))
2668 && !(flags & SWS_ACCURATE_RND) && !(dstH&1)) {
2669 c->swScale= ff_yuv2rgb_get_func_ptr(c);
2670 }
2671
2672 if (srcFormat==PIX_FMT_YUV410P && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_BITEXACT)) {
2673 c->swScale= yvu9toyv12Wrapper;
2674 }
2675
2676 /* bgr24toYV12 */
2677 if (srcFormat==PIX_FMT_BGR24 && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_ACCURATE_RND))
2678 c->swScale= bgr24toyv12Wrapper;
2679
2680 /* RGB/BGR -> RGB/BGR (no dither needed forms) */
2681 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2682 && (isBGR(dstFormat) || isRGB(dstFormat))
2683 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2684 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2685 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2686 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2687 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2688 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2689 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2690 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2691 && dstFormat != PIX_FMT_RGB32_1
2692 && dstFormat != PIX_FMT_BGR32_1
2693 && srcFormat != PIX_FMT_RGB48LE && dstFormat != PIX_FMT_RGB48LE
2694 && srcFormat != PIX_FMT_RGB48BE && dstFormat != PIX_FMT_RGB48BE
2695 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2696 c->swScale= rgb2rgbWrapper;
2697
2698 if ((usePal(srcFormat) && (
2699 dstFormat == PIX_FMT_RGB32 ||
2700 dstFormat == PIX_FMT_RGB32_1 ||
2701 dstFormat == PIX_FMT_RGB24 ||
2702 dstFormat == PIX_FMT_BGR32 ||
2703 dstFormat == PIX_FMT_BGR32_1 ||
2704 dstFormat == PIX_FMT_BGR24)))
2705 c->swScale= pal2rgbWrapper;
2706
2707 if (srcFormat == PIX_FMT_YUV422P) {
2708 if (dstFormat == PIX_FMT_YUYV422)
2709 c->swScale= YUV422PToYuy2Wrapper;
2710 else if (dstFormat == PIX_FMT_UYVY422)
2711 c->swScale= YUV422PToUyvyWrapper;
2712 }
2713
2714 /* LQ converters if -sws 0 or -sws 4*/
2715 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)) {
2716 /* yv12_to_yuy2 */
2717 if (srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) {
2718 if (dstFormat == PIX_FMT_YUYV422)
2719 c->swScale= PlanarToYuy2Wrapper;
2720 else if (dstFormat == PIX_FMT_UYVY422)
2721 c->swScale= PlanarToUyvyWrapper;
2722 }
2723 }
2724 if(srcFormat == PIX_FMT_YUYV422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2725 c->swScale= YUYV2YUV420Wrapper;
2726 if(srcFormat == PIX_FMT_UYVY422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2727 c->swScale= UYVY2YUV420Wrapper;
2728 if(srcFormat == PIX_FMT_YUYV422 && dstFormat == PIX_FMT_YUV422P)
2729 c->swScale= YUYV2YUV422Wrapper;
2730 if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
2731 c->swScale= UYVY2YUV422Wrapper;
2732
2733 #ifdef COMPILE_ALTIVEC
2734 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2735 !(c->flags & SWS_BITEXACT) &&
2736 srcFormat == PIX_FMT_YUV420P) {
2737 // unscaled YV12 -> packed YUV, we want speed
2738 if (dstFormat == PIX_FMT_YUYV422)
2739 c->swScale= yv12toyuy2_unscaled_altivec;
2740 else if (dstFormat == PIX_FMT_UYVY422)
2741 c->swScale= yv12touyvy_unscaled_altivec;
2742 }
2743 #endif
2744
2745 /* simple copy */
2746 if ( srcFormat == dstFormat
2747 || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P)
2748 || (srcFormat == PIX_FMT_YUV420P && dstFormat == PIX_FMT_YUVA420P)
2749 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2750 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2751 || (isGray(dstFormat) && isGray(srcFormat))
2752 || (isPlanarYUV(srcFormat) && isPlanarYUV(dstFormat)
2753 && c->chrDstHSubSample == c->chrSrcHSubSample
2754 && c->chrDstVSubSample == c->chrSrcVSubSample
2755 && dstFormat != PIX_FMT_NV12 && dstFormat != PIX_FMT_NV21
2756 && srcFormat != PIX_FMT_NV12 && srcFormat != PIX_FMT_NV21))
2757 {
2758 if (isPacked(c->srcFormat))
2759 c->swScale= packedCopy;
2760 else /* Planar YUV or gray */
2761 c->swScale= planarCopy;
2762 }
2763 #if ARCH_BFIN
2764 if (flags & SWS_CPU_CAPS_BFIN)
2765 ff_bfin_get_unscaled_swscale (c);
2766 #endif
2767
2768 if (c->swScale) {
2769 if (flags&SWS_PRINT_INFO)
2770 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2771 sws_format_name(srcFormat), sws_format_name(dstFormat));
2772 return c;
2773 }
2774 }
2775
2776 if (flags & SWS_CPU_CAPS_MMX2) {
2777 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2778 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
2779 if (flags&SWS_PRINT_INFO)
2780 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
2781 }
2782 if (usesHFilter) c->canMMX2BeUsed=0;
2783 }
2784 else
2785 c->canMMX2BeUsed=0;
2786
2787 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2788 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2789
2790 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2791 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2792 // n-2 is the last chrominance sample available
2793 // this is not perfect, but no one should notice the difference, the more correct variant
2794 // would be like the vertical one, but that would require some special code for the
2795 // first and last pixel
2796 if (flags&SWS_FAST_BILINEAR) {
2797 if (c->canMMX2BeUsed) {
2798 c->lumXInc+= 20;
2799 c->chrXInc+= 20;
2800 }
2801 //we don't use the x86 asm scaler if MMX is available
2802 else if (flags & SWS_CPU_CAPS_MMX) {
2803 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2804 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2805 }
2806 }
2807
2808 /* precalculate horizontal scaler filter coefficients */
2809 {
2810 const int filterAlign=
2811 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2812 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2813 1;
2814
2815 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2816 srcW , dstW, filterAlign, 1<<14,
2817 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2818 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
2819 goto fail;
2820 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2821 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2822 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2823 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
2824 goto fail;
2825
2826 #if defined(COMPILE_MMX2)
2827 // can't downscale !!!
2828 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
2829 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
2830 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
2831
2832 #ifdef MAP_ANONYMOUS
2833 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2834 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2835 #elif HAVE_VIRTUALALLOC
2836 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2837 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2838 #else
2839 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
2840 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
2841 #endif
2842
2843 CHECKED_ALLOCZ(c->lumMmx2Filter , (dstW /8+8)*sizeof(int16_t));
2844 CHECKED_ALLOCZ(c->chrMmx2Filter , (c->chrDstW /4+8)*sizeof(int16_t));
2845 CHECKED_ALLOCZ(c->lumMmx2FilterPos, (dstW /2/8+8)*sizeof(int32_t));
2846 CHECKED_ALLOCZ(c->chrMmx2FilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t));
2847
2848 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2849 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2850
2851 #ifdef MAP_ANONYMOUS
2852 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2853 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2854 #endif
2855 }
2856 #endif /* defined(COMPILE_MMX2) */
2857 } // initialize horizontal stuff
2858
2859
2860
2861 /* precalculate vertical scaler filter coefficients */
2862 {
2863 const int filterAlign=
2864 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2865 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2866 1;
2867
2868 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2869 srcH , dstH, filterAlign, (1<<12),
2870 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2871 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
2872 goto fail;
2873 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2874 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2875 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2876 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
2877 goto fail;
2878
2879 #ifdef COMPILE_ALTIVEC
2880 CHECKED_ALLOC(c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2881 CHECKED_ALLOC(c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2882
2883 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2884 int j;
2885 short *p = (short *)&c->vYCoeffsBank[i];
2886 for (j=0;j<8;j++)
2887 p[j] = c->vLumFilter[i];
2888 }
2889
2890 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2891 int j;
2892 short *p = (short *)&c->vCCoeffsBank[i];
2893 for (j=0;j<8;j++)
2894 p[j] = c->vChrFilter[i];
2895 }
2896 #endif
2897 }
2898
2899 // calculate buffer sizes so that they won't run out while handling these damn slices
2900 c->vLumBufSize= c->vLumFilterSize;
2901 c->vChrBufSize= c->vChrFilterSize;
2902 for (i=0; i<dstH; i++) {
2903 int chrI= i*c->chrDstH / dstH;
2904 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2905 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2906
2907 nextSlice>>= c->chrSrcVSubSample;
2908 nextSlice<<= c->chrSrcVSubSample;
2909 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2910 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2911 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2912 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2913 }
2914
2915 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2916 // allocate several megabytes to handle all possible cases)
2917 CHECKED_ALLOC(c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*));
2918 CHECKED_ALLOC(c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*));
2919 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
2920 CHECKED_ALLOCZ(c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*));
2921 //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
2922 /* align at 16 bytes for AltiVec */
2923 for (i=0; i<c->vLumBufSize; i++)
2924 {
2925 CHECKED_ALLOCZ(c->lumPixBuf[i+c->vLumBufSize], VOF+1);
2926 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
2927 }
2928 for (i=0; i<c->vChrBufSize; i++)
2929 {
2930 CHECKED_ALLOC(c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2);
2931 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
2932 }
2933 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
2934 for (i=0; i<c->vLumBufSize; i++)
2935 {
2936 CHECKED_ALLOCZ(c->alpPixBuf[i+c->vLumBufSize], VOF+1);
2937 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
2938 }
2939
2940 //try to avoid drawing green stuff between the right end and the stride end
2941 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2942
2943 assert(2*VOFW == VOF);
2944
2945 assert(c->chrDstH <= dstH);
2946
2947 if (flags&SWS_PRINT_INFO) {
2948 #ifdef DITHER1XBPP
2949 const char *dither= " dithered";
2950 #else
2951 const char *dither= "";
2952 #endif
2953 if (flags&SWS_FAST_BILINEAR)
2954 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2955 else if (flags&SWS_BILINEAR)
2956 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2957 else if (flags&SWS_BICUBIC)
2958 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2959 else if (flags&SWS_X)
2960 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2961 else if (flags&SWS_POINT)
2962 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2963 else if (flags&SWS_AREA)
2964 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2965 else if (flags&SWS_BICUBLIN)
2966 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2967 else if (flags&SWS_GAUSS)
2968 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2969 else if (flags&SWS_SINC)
2970 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2971 else if (flags&SWS_LANCZOS)
2972 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2973 else if (flags&SWS_SPLINE)
2974 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2975 else
2976 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2977
2978 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2979 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2980 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2981 else
2982 av_log(c, AV_LOG_INFO, "from %s to %s ",
2983 sws_format_name(srcFormat), sws_format_name(dstFormat));
2984
2985 if (flags & SWS_CPU_CAPS_MMX2)
2986 av_log(c, AV_LOG_INFO, "using MMX2\n");
2987 else if (flags & SWS_CPU_CAPS_3DNOW)
2988 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2989 else if (flags & SWS_CPU_CAPS_MMX)
2990 av_log(c, AV_LOG_INFO, "using MMX\n");
2991 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2992 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2993 else
2994 av_log(c, AV_LOG_INFO, "using C\n");
2995 }
2996
2997 if (flags & SWS_PRINT_INFO) {
2998 if (flags & SWS_CPU_CAPS_MMX) {
2999 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
3000 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
3001 else {
3002 if (c->hLumFilterSize==4)
3003 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
3004 else if (c->hLumFilterSize==8)
3005 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
3006 else
3007 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
3008
3009 if (c->hChrFilterSize==4)
3010 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
3011 else if (c->hChrFilterSize==8)
3012 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
3013 else
3014 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
3015 }
3016 } else {
3017 #if ARCH_X86
3018 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
3019 #else
3020 if (flags & SWS_FAST_BILINEAR)
3021 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
3022 else
3023 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
3024 #endif
3025 }
3026 if (isPlanarYUV(dstFormat)) {
3027 if (c->vLumFilterSize==1)
3028 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3029 else
3030 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3031 } else {
3032 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
3033 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
3034 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3035 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
3036 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3037 else
3038 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3039 }
3040
3041 if (dstFormat==PIX_FMT_BGR24)
3042 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
3043 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
3044 else if (dstFormat==PIX_FMT_RGB32)
3045 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3046 else if (dstFormat==PIX_FMT_BGR565)
3047 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3048 else if (dstFormat==PIX_FMT_BGR555)
3049 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
3050
3051 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
3052 }
3053 if (flags & SWS_PRINT_INFO) {
3054 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
3055 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
3056 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
3057 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
3058 }
3059
3060 c->swScale= getSwsFunc(c);
3061 return c;
3062
3063 fail:
3064 sws_freeContext(c);
3065 return NULL;
3066 }
3067
3068 static void reset_ptr(uint8_t* src[], int format)
3069 {
3070 if(!isALPHA(format))
3071 src[3]=NULL;
3072 if(!isPlanarYUV(format)) {
3073 src[3]=src[2]=NULL;
3074 if( format != PIX_FMT_PAL8
3075 && format != PIX_FMT_RGB8
3076 && format != PIX_FMT_BGR8
3077 && format != PIX_FMT_RGB4_BYTE
3078 && format != PIX_FMT_BGR4_BYTE
3079 )
3080 src[1]= NULL;
3081 }
3082 }
3083
3084 /**
3085 * swscale wrapper, so we don't need to export the SwsContext.
3086 * Assumes planar YUV to be in YUV order instead of YVU.
3087 */
3088 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
3089 int srcSliceH, uint8_t* dst[], int dstStride[])
3090 {
3091 int i;
3092 uint8_t* src2[4]= {src[0], src[1], src[2], src[3]};
3093 uint8_t* dst2[4]= {dst[0], dst[1], dst[2], dst[3]};
3094
3095 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
3096 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
3097 return 0;
3098 }
3099 if (c->sliceDir == 0) {
3100 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
3101 }
3102
3103 if (usePal(c->srcFormat)) {
3104 for (i=0; i<256; i++) {
3105 int p, r, g, b,y,u,v;
3106 if(c->srcFormat == PIX_FMT_PAL8) {
3107 p=((uint32_t*)(src[1]))[i];
3108 r= (p>>16)&0xFF;
3109 g= (p>> 8)&0xFF;
3110 b= p &0xFF;
3111 } else if(c->srcFormat == PIX_FMT_RGB8) {
3112 r= (i>>5 )*36;
3113 g= ((i>>2)&7)*36;
3114 b= (i&3 )*85;
3115 } else if(c->srcFormat == PIX_FMT_BGR8) {
3116 b= (i>>6 )*85;
3117 g= ((i>>3)&7)*36;
3118 r= (i&7 )*36;
3119 } else if(c->srcFormat == PIX_FMT_RGB4_BYTE) {
3120 r= (i>>3 )*255;
3121 g= ((i>>1)&3)*85;
3122 b= (i&1 )*255;
3123 } else {
3124 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
3125 b= (i>>3 )*255;
3126 g= ((i>>1)&3)*85;
3127 r= (i&1 )*255;
3128 }
3129 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
3130 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
3131 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
3132 c->pal_yuv[i]= y + (u<<8) + (v<<16);
3133
3134
3135 switch(c->dstFormat) {
3136 case PIX_FMT_BGR32:
3137 #if !HAVE_BIGENDIAN
3138 case PIX_FMT_RGB24:
3139 #endif
3140 c->pal_rgb[i]= r + (g<<8) + (b<<16);
3141 break;
3142 case PIX_FMT_BGR32_1:
3143 #if HAVE_BIGENDIAN
3144 case PIX_FMT_BGR24:
3145 #endif
3146 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
3147 break;
3148 case PIX_FMT_RGB32_1:
3149 #if HAVE_BIGENDIAN
3150 case PIX_FMT_RGB24:
3151 #endif
3152 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
3153 break;
3154 case PIX_FMT_RGB32:
3155 #if !HAVE_BIGENDIAN
3156 case PIX_FMT_BGR24:
3157 #endif
3158 default:
3159 c->pal_rgb[i]= b + (g<<8) + (r<<16);
3160 }
3161 }
3162 }
3163
3164 // copy strides, so they can safely be modified
3165 if (c->sliceDir == 1) {
3166 // slices go from top to bottom
3167 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2], srcStride[3]};
3168 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2], dstStride[3]};
3169
3170 reset_ptr(src2, c->srcFormat);
3171 reset_ptr(dst2, c->dstFormat);
3172
3173 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2, dstStride2);
3174 } else {
3175 // slices go from bottom to top => we flip the image internally
3176 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2], -srcStride[3]};
3177 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2], -dstStride[3]};
3178
3179 src2[0] += (srcSliceH-1)*srcStride[0];
3180 if (!usePal(c->srcFormat))
3181 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
3182 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
3183 src2[3] += (srcSliceH-1)*srcStride[3];
3184 dst2[0] += ( c->dstH -1)*dstStride[0];
3185 dst2[1] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1];
3186 dst2[2] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2];
3187 dst2[3] += ( c->dstH -1)*dstStride[3];
3188
3189 reset_ptr(src2, c->srcFormat);
3190 reset_ptr(dst2, c->dstFormat);
3191
3192 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
3193 }
3194 }
3195
3196 #if LIBSWSCALE_VERSION_MAJOR < 1
3197 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
3198 int srcSliceH, uint8_t* dst[], int dstStride[])
3199 {
3200 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
3201 }
3202 #endif
3203
3204 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
3205 float lumaSharpen, float chromaSharpen,
3206 float chromaHShift, float chromaVShift,
3207 int verbose)
3208 {
3209 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
3210 if (!filter)
3211 return NULL;
3212
3213 if (lumaGBlur!=0.0) {
3214 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
3215 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
3216 } else {
3217 filter->lumH= sws_getIdentityVec();
3218 filter->lumV= sws_getIdentityVec();
3219 }
3220
3221 if (chromaGBlur!=0.0) {
3222 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
3223 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
3224 } else {
3225 filter->chrH= sws_getIdentityVec();
3226 filter->chrV= sws_getIdentityVec();
3227 }
3228
3229 if (chromaSharpen!=0.0) {
3230 SwsVector *id= sws_getIdentityVec();
3231 sws_scaleVec(filter->chrH, -chromaSharpen);
3232 sws_scaleVec(filter->chrV, -chromaSharpen);
3233 sws_addVec(filter->chrH, id);
3234 sws_addVec(filter->chrV, id);
3235 sws_freeVec(id);
3236 }
3237
3238 if (lumaSharpen!=0.0) {
3239 SwsVector *id= sws_getIdentityVec();
3240 sws_scaleVec(filter->lumH, -lumaSharpen);
3241 sws_scaleVec(filter->lumV, -lumaSharpen);
3242 sws_addVec(filter->lumH, id);
3243 sws_addVec(filter->lumV, id);
3244 sws_freeVec(id);
3245 }
3246
3247 if (chromaHShift != 0.0)
3248 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
3249
3250 if (chromaVShift != 0.0)
3251 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
3252
3253 sws_normalizeVec(filter->chrH, 1.0);
3254 sws_normalizeVec(filter->chrV, 1.0);
3255 sws_normalizeVec(filter->lumH, 1.0);
3256 sws_normalizeVec(filter->lumV, 1.0);
3257
3258 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
3259 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
3260
3261 return filter;
3262 }
3263
3264 SwsVector *sws_allocVec(int length)
3265 {
3266 SwsVector *vec = av_malloc(sizeof(SwsVector));
3267 if (!vec)
3268 return NULL;
3269 vec->length = length;
3270 vec->coeff = av_malloc(sizeof(double) * length);
3271 if (!vec->coeff)
3272 av_freep(&vec);
3273 return vec;
3274 }
3275
3276 SwsVector *sws_getGaussianVec(double variance, double quality)
3277 {
3278 const int length= (int)(variance*quality + 0.5) | 1;
3279 int i;
3280 double middle= (length-1)*0.5;
3281 SwsVector *vec= sws_allocVec(length);
3282
3283 if (!vec)
3284 return NULL;
3285
3286 for (i=0; i<length; i++) {
3287 double dist= i-middle;
3288 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
3289 }
3290
3291 sws_normalizeVec(vec, 1.0);
3292
3293 return vec;
3294 }
3295
3296 SwsVector *sws_getConstVec(double c, int length)
3297 {
3298 int i;
3299 SwsVector *vec= sws_allocVec(length);
3300
3301 if (!vec)
3302 return NULL;
3303
3304 for (i=0; i<length; i++)
3305 vec->coeff[i]= c;
3306
3307 return vec;
3308 }
3309
3310
3311 SwsVector *sws_getIdentityVec(void)
3312 {
3313 return sws_getConstVec(1.0, 1);
3314 }
3315
3316 double sws_dcVec(SwsVector *a)
3317 {
3318 int i;
3319 double sum=0;
3320
3321 for (i=0; i<a->length; i++)
3322 sum+= a->coeff[i];
3323
3324 return sum;
3325 }
3326
3327 void sws_scaleVec(SwsVector *a, double scalar)
3328 {
3329 int i;
3330
3331 for (i=0; i<a->length; i++)
3332 a->coeff[i]*= scalar;
3333 }
3334
3335 void sws_normalizeVec(SwsVector *a, double height)
3336 {
3337 sws_scaleVec(a, height/sws_dcVec(a));
3338 }
3339
3340 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
3341 {
3342 int length= a->length + b->length - 1;
3343 int i, j;
3344 SwsVector *vec= sws_getConstVec(0.0, length);
3345
3346 if (!vec)
3347 return NULL;
3348
3349 for (i=0; i<a->length; i++) {
3350 for (j=0; j<b->length; j++) {
3351 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
3352 }
3353 }
3354
3355 return vec;
3356 }
3357
3358 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
3359 {
3360 int length= FFMAX(a->length, b->length);