2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
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.
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.
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
20 * the C code (not assembly, mmx, ...) of this file can be used
21 * under the LGPL license too
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
29 unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30 YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
35 BGR24 -> BGR32 & RGB24 -> RGB32
36 BGR32 -> BGR24 & RGB32 -> RGB24
41 tested special converters (most are tested actually, but I did not write it down ...)
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
57 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
71 #define WIN32_LEAN_AND_MEAN
75 #include "swscale_internal.h"
77 #include "libavutil/intreadwrite.h"
78 #include "libavutil/x86_cpu.h"
79 #include "libavutil/avutil.h"
80 #include "libavutil/bswap.h"
81 #include "libavutil/pixdesc.h"
83 unsigned swscale_version(void)
85 return LIBSWSCALE_VERSION_INT;
88 const char *swscale_configuration(void)
90 return FFMPEG_CONFIGURATION;
93 const char *swscale_license(void)
95 #define LICENSE_PREFIX "libswscale license: "
96 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
103 //#define HAVE_AMD3DNOW
108 #define FAST_BGR2YV12 // use 7 bit coefficients instead of 15 bit
110 #define RET 0xC3 //near return opcode for x86
115 #define PI 3.14159265358979323846
118 #define isSupportedIn(x) ( \
119 (x)==PIX_FMT_YUV420P \
120 || (x)==PIX_FMT_YUVA420P \
121 || (x)==PIX_FMT_YUYV422 \
122 || (x)==PIX_FMT_UYVY422 \
123 || (x)==PIX_FMT_RGB48BE \
124 || (x)==PIX_FMT_RGB48LE \
125 || (x)==PIX_FMT_RGB32 \
126 || (x)==PIX_FMT_RGB32_1 \
127 || (x)==PIX_FMT_BGR24 \
128 || (x)==PIX_FMT_BGR565 \
129 || (x)==PIX_FMT_BGR555 \
130 || (x)==PIX_FMT_BGR32 \
131 || (x)==PIX_FMT_BGR32_1 \
132 || (x)==PIX_FMT_RGB24 \
133 || (x)==PIX_FMT_RGB565 \
134 || (x)==PIX_FMT_RGB555 \
135 || (x)==PIX_FMT_GRAY8 \
136 || (x)==PIX_FMT_YUV410P \
137 || (x)==PIX_FMT_YUV440P \
138 || (x)==PIX_FMT_NV12 \
139 || (x)==PIX_FMT_NV21 \
140 || (x)==PIX_FMT_GRAY16BE \
141 || (x)==PIX_FMT_GRAY16LE \
142 || (x)==PIX_FMT_YUV444P \
143 || (x)==PIX_FMT_YUV422P \
144 || (x)==PIX_FMT_YUV411P \
145 || (x)==PIX_FMT_PAL8 \
146 || (x)==PIX_FMT_BGR8 \
147 || (x)==PIX_FMT_RGB8 \
148 || (x)==PIX_FMT_BGR4_BYTE \
149 || (x)==PIX_FMT_RGB4_BYTE \
150 || (x)==PIX_FMT_YUV440P \
151 || (x)==PIX_FMT_MONOWHITE \
152 || (x)==PIX_FMT_MONOBLACK \
153 || (x)==PIX_FMT_YUV420P16LE \
154 || (x)==PIX_FMT_YUV422P16LE \
155 || (x)==PIX_FMT_YUV444P16LE \
156 || (x)==PIX_FMT_YUV420P16BE \
157 || (x)==PIX_FMT_YUV422P16BE \
158 || (x)==PIX_FMT_YUV444P16BE \
160 #define isSupportedOut(x) ( \
161 (x)==PIX_FMT_YUV420P \
162 || (x)==PIX_FMT_YUVA420P \
163 || (x)==PIX_FMT_YUYV422 \
164 || (x)==PIX_FMT_UYVY422 \
165 || (x)==PIX_FMT_YUV444P \
166 || (x)==PIX_FMT_YUV422P \
167 || (x)==PIX_FMT_YUV411P \
170 || (x)==PIX_FMT_NV12 \
171 || (x)==PIX_FMT_NV21 \
172 || (x)==PIX_FMT_GRAY16BE \
173 || (x)==PIX_FMT_GRAY16LE \
174 || (x)==PIX_FMT_GRAY8 \
175 || (x)==PIX_FMT_YUV410P \
176 || (x)==PIX_FMT_YUV440P \
177 || (x)==PIX_FMT_YUV420P16LE \
178 || (x)==PIX_FMT_YUV422P16LE \
179 || (x)==PIX_FMT_YUV444P16LE \
180 || (x)==PIX_FMT_YUV420P16BE \
181 || (x)==PIX_FMT_YUV422P16BE \
182 || (x)==PIX_FMT_YUV444P16BE \
184 #define isPacked(x) ( \
186 || (x)==PIX_FMT_YUYV422 \
187 || (x)==PIX_FMT_UYVY422 \
191 #define usePal(x) (av_pix_fmt_descriptors[x].flags & PIX_FMT_PAL)
193 #define RGB2YUV_SHIFT 15
194 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
195 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
196 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
197 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
198 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
199 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
200 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
201 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
202 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
204 extern const int32_t ff_yuv2rgb_coeffs[8][4];
206 static const double rgb2yuv_table[8][9]={
207 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
208 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
209 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
210 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
211 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
212 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
213 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
214 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
219 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
222 more intelligent misalignment avoidance for the horizontal scaler
223 write special vertical cubic upscale version
224 optimize C code (YV12 / minmax)
225 add support for packed pixel YUV input & output
226 add support for Y8 output
227 optimize BGR24 & BGR32
228 add BGR4 output support
229 write special BGR->BGR scaler
232 #if ARCH_X86 && CONFIG_GPL
233 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
234 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
235 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
236 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
237 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
238 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
239 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
240 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
242 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
243 0x0103010301030103LL,
244 0x0200020002000200LL,};
246 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
247 0x0602060206020602LL,
248 0x0004000400040004LL,};
250 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
251 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
252 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
253 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
254 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
255 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
257 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
258 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
259 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
262 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
263 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
264 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
266 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
267 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
268 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
269 #endif /* FAST_BGR2YV12 */
270 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
271 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
272 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
274 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
275 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
276 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
277 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
278 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
280 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
281 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
282 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
285 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
287 #endif /* ARCH_X86 && CONFIG_GPL */
289 // clipping helper table for C implementations:
290 static unsigned char clip_table[768];
292 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
294 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_4[2][8])={
295 { 1, 3, 1, 3, 1, 3, 1, 3, },
296 { 2, 0, 2, 0, 2, 0, 2, 0, },
299 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8[2][8])={
300 { 6, 2, 6, 2, 6, 2, 6, 2, },
301 { 0, 4, 0, 4, 0, 4, 0, 4, },
304 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32[8][8])={
305 { 17, 9, 23, 15, 16, 8, 22, 14, },
306 { 5, 29, 3, 27, 4, 28, 2, 26, },
307 { 21, 13, 19, 11, 20, 12, 18, 10, },
308 { 0, 24, 6, 30, 1, 25, 7, 31, },
309 { 16, 8, 22, 14, 17, 9, 23, 15, },
310 { 4, 28, 2, 26, 5, 29, 3, 27, },
311 { 20, 12, 18, 10, 21, 13, 19, 11, },
312 { 1, 25, 7, 31, 0, 24, 6, 30, },
315 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73[8][8])={
316 { 0, 55, 14, 68, 3, 58, 17, 72, },
317 { 37, 18, 50, 32, 40, 22, 54, 35, },
318 { 9, 64, 5, 59, 13, 67, 8, 63, },
319 { 46, 27, 41, 23, 49, 31, 44, 26, },
320 { 2, 57, 16, 71, 1, 56, 15, 70, },
321 { 39, 21, 52, 34, 38, 19, 51, 33, },
322 { 11, 66, 7, 62, 10, 65, 6, 60, },
323 { 48, 30, 43, 25, 47, 29, 42, 24, },
327 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
328 {117, 62, 158, 103, 113, 58, 155, 100, },
329 { 34, 199, 21, 186, 31, 196, 17, 182, },
330 {144, 89, 131, 76, 141, 86, 127, 72, },
331 { 0, 165, 41, 206, 10, 175, 52, 217, },
332 {110, 55, 151, 96, 120, 65, 162, 107, },
333 { 28, 193, 14, 179, 38, 203, 24, 189, },
334 {138, 83, 124, 69, 148, 93, 134, 79, },
335 { 7, 172, 48, 213, 3, 168, 45, 210, },
338 // tries to correct a gamma of 1.5
339 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
340 { 0, 143, 18, 200, 2, 156, 25, 215, },
341 { 78, 28, 125, 64, 89, 36, 138, 74, },
342 { 10, 180, 3, 161, 16, 195, 8, 175, },
343 {109, 51, 93, 38, 121, 60, 105, 47, },
344 { 1, 152, 23, 210, 0, 147, 20, 205, },
345 { 85, 33, 134, 71, 81, 30, 130, 67, },
346 { 14, 190, 6, 171, 12, 185, 5, 166, },
347 {117, 57, 101, 44, 113, 54, 97, 41, },
350 // tries to correct a gamma of 2.0
351 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
352 { 0, 124, 8, 193, 0, 140, 12, 213, },
353 { 55, 14, 104, 42, 66, 19, 119, 52, },
354 { 3, 168, 1, 145, 6, 187, 3, 162, },
355 { 86, 31, 70, 21, 99, 39, 82, 28, },
356 { 0, 134, 11, 206, 0, 129, 9, 200, },
357 { 62, 17, 114, 48, 58, 16, 109, 45, },
358 { 5, 181, 2, 157, 4, 175, 1, 151, },
359 { 95, 36, 78, 26, 90, 34, 74, 24, },
362 // tries to correct a gamma of 2.5
363 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220[8][8])={
364 { 0, 107, 3, 187, 0, 125, 6, 212, },
365 { 39, 7, 86, 28, 49, 11, 102, 36, },
366 { 1, 158, 0, 131, 3, 180, 1, 151, },
367 { 68, 19, 52, 12, 81, 25, 64, 17, },
368 { 0, 119, 5, 203, 0, 113, 4, 195, },
369 { 45, 9, 96, 33, 42, 8, 91, 30, },
370 { 2, 172, 1, 144, 2, 165, 0, 137, },
371 { 77, 23, 60, 15, 72, 21, 56, 14, },
375 const char *sws_format_name(enum PixelFormat format)
377 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
378 return av_pix_fmt_descriptors[format].name;
380 return "Unknown format";
383 static av_always_inline void yuv2yuvX16inC_template(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
384 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
385 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest,
386 int dstW, int chrDstW, int big_endian)
388 //FIXME Optimize (just quickly written not optimized..)
391 for (i = 0; i < dstW; i++) {
395 for (j = 0; j < lumFilterSize; j++)
396 val += lumSrc[j][i] * lumFilter[j];
399 AV_WB16(&dest[i], av_clip_uint16(val >> 11));
401 AV_WL16(&dest[i], av_clip_uint16(val >> 11));
406 for (i = 0; i < chrDstW; i++) {
411 for (j = 0; j < chrFilterSize; j++) {
412 u += chrSrc[j][i ] * chrFilter[j];
413 v += chrSrc[j][i + VOFW] * chrFilter[j];
417 AV_WB16(&uDest[i], av_clip_uint16(u >> 11));
418 AV_WB16(&vDest[i], av_clip_uint16(v >> 11));
420 AV_WL16(&uDest[i], av_clip_uint16(u >> 11));
421 AV_WL16(&vDest[i], av_clip_uint16(v >> 11));
426 if (CONFIG_SWSCALE_ALPHA && aDest) {
427 for (i = 0; i < dstW; i++) {
431 for (j = 0; j < lumFilterSize; j++)
432 val += alpSrc[j][i] * lumFilter[j];
435 AV_WB16(&aDest[i], av_clip_uint16(val >> 11));
437 AV_WL16(&aDest[i], av_clip_uint16(val >> 11));
443 static inline void yuv2yuvX16inC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
444 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
445 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest, int dstW, int chrDstW,
446 enum PixelFormat dstFormat)
448 if (isBE(dstFormat)) {
449 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
450 chrFilter, chrSrc, chrFilterSize,
452 dest, uDest, vDest, aDest,
455 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
456 chrFilter, chrSrc, chrFilterSize,
458 dest, uDest, vDest, aDest,
463 static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
464 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
465 const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW)
467 //FIXME Optimize (just quickly written not optimized..)
469 for (i=0; i<dstW; i++) {
472 for (j=0; j<lumFilterSize; j++)
473 val += lumSrc[j][i] * lumFilter[j];
475 dest[i]= av_clip_uint8(val>>19);
479 for (i=0; i<chrDstW; i++) {
483 for (j=0; j<chrFilterSize; j++) {
484 u += chrSrc[j][i] * chrFilter[j];
485 v += chrSrc[j][i + VOFW] * chrFilter[j];
488 uDest[i]= av_clip_uint8(u>>19);
489 vDest[i]= av_clip_uint8(v>>19);
492 if (CONFIG_SWSCALE_ALPHA && aDest)
493 for (i=0; i<dstW; i++) {
496 for (j=0; j<lumFilterSize; j++)
497 val += alpSrc[j][i] * lumFilter[j];
499 aDest[i]= av_clip_uint8(val>>19);
504 static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
505 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
506 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
508 //FIXME Optimize (just quickly written not optimized..)
510 for (i=0; i<dstW; i++) {
513 for (j=0; j<lumFilterSize; j++)
514 val += lumSrc[j][i] * lumFilter[j];
516 dest[i]= av_clip_uint8(val>>19);
522 if (dstFormat == PIX_FMT_NV12)
523 for (i=0; i<chrDstW; i++) {
527 for (j=0; j<chrFilterSize; j++) {
528 u += chrSrc[j][i] * chrFilter[j];
529 v += chrSrc[j][i + VOFW] * chrFilter[j];
532 uDest[2*i]= av_clip_uint8(u>>19);
533 uDest[2*i+1]= av_clip_uint8(v>>19);
536 for (i=0; i<chrDstW; i++) {
540 for (j=0; j<chrFilterSize; j++) {
541 u += chrSrc[j][i] * chrFilter[j];
542 v += chrSrc[j][i + VOFW] * chrFilter[j];
545 uDest[2*i]= av_clip_uint8(v>>19);
546 uDest[2*i+1]= av_clip_uint8(u>>19);
550 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha) \
551 for (i=0; i<(dstW>>1); i++) {\
557 int av_unused A1, A2;\
558 type av_unused *r, *b, *g;\
561 for (j=0; j<lumFilterSize; j++) {\
562 Y1 += lumSrc[j][i2] * lumFilter[j];\
563 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
565 for (j=0; j<chrFilterSize; j++) {\
566 U += chrSrc[j][i] * chrFilter[j];\
567 V += chrSrc[j][i+VOFW] * chrFilter[j];\
576 for (j=0; j<lumFilterSize; j++) {\
577 A1 += alpSrc[j][i2 ] * lumFilter[j];\
578 A2 += alpSrc[j][i2+1] * lumFilter[j];\
584 #define YSCALE_YUV_2_PACKEDX_C(type,alpha) \
585 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha)\
586 if ((Y1|Y2|U|V)&256) {\
587 if (Y1>255) Y1=255; \
588 else if (Y1<0)Y1=0; \
589 if (Y2>255) Y2=255; \
590 else if (Y2<0)Y2=0; \
596 if (alpha && ((A1|A2)&256)) {\
597 A1=av_clip_uint8(A1);\
598 A2=av_clip_uint8(A2);\
601 #define YSCALE_YUV_2_PACKEDX_FULL_C(rnd,alpha) \
602 for (i=0; i<dstW; i++) {\
610 for (j=0; j<lumFilterSize; j++) {\
611 Y += lumSrc[j][i ] * lumFilter[j];\
613 for (j=0; j<chrFilterSize; j++) {\
614 U += chrSrc[j][i ] * chrFilter[j];\
615 V += chrSrc[j][i+VOFW] * chrFilter[j];\
622 for (j=0; j<lumFilterSize; j++)\
623 A += alpSrc[j][i ] * lumFilter[j];\
626 A = av_clip_uint8(A);\
629 #define YSCALE_YUV_2_RGBX_FULL_C(rnd,alpha) \
630 YSCALE_YUV_2_PACKEDX_FULL_C(rnd>>3,alpha)\
631 Y-= c->yuv2rgb_y_offset;\
632 Y*= c->yuv2rgb_y_coeff;\
634 R= Y + V*c->yuv2rgb_v2r_coeff;\
635 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
636 B= Y + U*c->yuv2rgb_u2b_coeff;\
637 if ((R|G|B)&(0xC0000000)) {\
638 if (R>=(256<<22)) R=(256<<22)-1; \
640 if (G>=(256<<22)) G=(256<<22)-1; \
642 if (B>=(256<<22)) B=(256<<22)-1; \
647 #define YSCALE_YUV_2_GRAY16_C \
648 for (i=0; i<(dstW>>1); i++) {\
657 for (j=0; j<lumFilterSize; j++) {\
658 Y1 += lumSrc[j][i2] * lumFilter[j];\
659 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
663 if ((Y1|Y2|U|V)&65536) {\
664 if (Y1>65535) Y1=65535; \
665 else if (Y1<0)Y1=0; \
666 if (Y2>65535) Y2=65535; \
667 else if (Y2<0)Y2=0; \
670 #define YSCALE_YUV_2_RGBX_C(type,alpha) \
671 YSCALE_YUV_2_PACKEDX_C(type,alpha) /* FIXME fix tables so that clipping is not needed and then use _NOCLIP*/\
672 r = (type *)c->table_rV[V]; \
673 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
674 b = (type *)c->table_bU[U]; \
676 #define YSCALE_YUV_2_PACKED2_C(type,alpha) \
677 for (i=0; i<(dstW>>1); i++) { \
679 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
680 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
681 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
682 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
683 type av_unused *r, *b, *g; \
684 int av_unused A1, A2; \
686 A1= (abuf0[i2 ]*yalpha1+abuf1[i2 ]*yalpha)>>19; \
687 A2= (abuf0[i2+1]*yalpha1+abuf1[i2+1]*yalpha)>>19; \
690 #define YSCALE_YUV_2_GRAY16_2_C \
691 for (i=0; i<(dstW>>1); i++) { \
693 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
694 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
696 #define YSCALE_YUV_2_RGB2_C(type,alpha) \
697 YSCALE_YUV_2_PACKED2_C(type,alpha)\
698 r = (type *)c->table_rV[V];\
699 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
700 b = (type *)c->table_bU[U];\
702 #define YSCALE_YUV_2_PACKED1_C(type,alpha) \
703 for (i=0; i<(dstW>>1); i++) {\
705 int Y1= buf0[i2 ]>>7;\
706 int Y2= buf0[i2+1]>>7;\
707 int U= (uvbuf1[i ])>>7;\
708 int V= (uvbuf1[i+VOFW])>>7;\
709 type av_unused *r, *b, *g;\
710 int av_unused A1, A2;\
716 #define YSCALE_YUV_2_GRAY16_1_C \
717 for (i=0; i<(dstW>>1); i++) {\
719 int Y1= buf0[i2 ]<<1;\
720 int Y2= buf0[i2+1]<<1;\
722 #define YSCALE_YUV_2_RGB1_C(type,alpha) \
723 YSCALE_YUV_2_PACKED1_C(type,alpha)\
724 r = (type *)c->table_rV[V];\
725 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
726 b = (type *)c->table_bU[U];\
728 #define YSCALE_YUV_2_PACKED1B_C(type,alpha) \
729 for (i=0; i<(dstW>>1); i++) {\
731 int Y1= buf0[i2 ]>>7;\
732 int Y2= buf0[i2+1]>>7;\
733 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
734 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
735 type av_unused *r, *b, *g;\
736 int av_unused A1, A2;\
742 #define YSCALE_YUV_2_RGB1B_C(type,alpha) \
743 YSCALE_YUV_2_PACKED1B_C(type,alpha)\
744 r = (type *)c->table_rV[V];\
745 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
746 b = (type *)c->table_bU[U];\
748 #define YSCALE_YUV_2_MONO2_C \
749 const uint8_t * const d128=dither_8x8_220[y&7];\
750 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
751 for (i=0; i<dstW-7; i+=8) {\
753 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
754 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
755 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
756 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
757 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
758 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
759 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
760 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
761 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
766 #define YSCALE_YUV_2_MONOX_C \
767 const uint8_t * const d128=dither_8x8_220[y&7];\
768 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
770 for (i=0; i<dstW-1; i+=2) {\
775 for (j=0; j<lumFilterSize; j++) {\
776 Y1 += lumSrc[j][i] * lumFilter[j];\
777 Y2 += lumSrc[j][i+1] * lumFilter[j];\
787 acc+= acc + g[Y1+d128[(i+0)&7]];\
788 acc+= acc + g[Y2+d128[(i+1)&7]];\
790 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
796 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
797 switch(c->dstFormat) {\
798 case PIX_FMT_RGB48BE:\
799 case PIX_FMT_RGB48LE:\
801 ((uint8_t*)dest)[ 0]= r[Y1];\
802 ((uint8_t*)dest)[ 1]= r[Y1];\
803 ((uint8_t*)dest)[ 2]= g[Y1];\
804 ((uint8_t*)dest)[ 3]= g[Y1];\
805 ((uint8_t*)dest)[ 4]= b[Y1];\
806 ((uint8_t*)dest)[ 5]= b[Y1];\
807 ((uint8_t*)dest)[ 6]= r[Y2];\
808 ((uint8_t*)dest)[ 7]= r[Y2];\
809 ((uint8_t*)dest)[ 8]= g[Y2];\
810 ((uint8_t*)dest)[ 9]= g[Y2];\
811 ((uint8_t*)dest)[10]= b[Y2];\
812 ((uint8_t*)dest)[11]= b[Y2];\
819 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
820 func(uint32_t,needAlpha)\
821 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? (A1<<24) : 0);\
822 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? (A2<<24) : 0);\
825 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
827 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (A1<<24);\
828 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (A2<<24);\
832 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
833 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
841 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
842 func(uint32_t,needAlpha)\
843 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? A1 : 0);\
844 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? A2 : 0);\
847 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
849 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + A1;\
850 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + A2;\
854 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
855 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
862 ((uint8_t*)dest)[0]= r[Y1];\
863 ((uint8_t*)dest)[1]= g[Y1];\
864 ((uint8_t*)dest)[2]= b[Y1];\
865 ((uint8_t*)dest)[3]= r[Y2];\
866 ((uint8_t*)dest)[4]= g[Y2];\
867 ((uint8_t*)dest)[5]= b[Y2];\
873 ((uint8_t*)dest)[0]= b[Y1];\
874 ((uint8_t*)dest)[1]= g[Y1];\
875 ((uint8_t*)dest)[2]= r[Y1];\
876 ((uint8_t*)dest)[3]= b[Y2];\
877 ((uint8_t*)dest)[4]= g[Y2];\
878 ((uint8_t*)dest)[5]= r[Y2];\
882 case PIX_FMT_RGB565:\
883 case PIX_FMT_BGR565:\
885 const int dr1= dither_2x2_8[y&1 ][0];\
886 const int dg1= dither_2x2_4[y&1 ][0];\
887 const int db1= dither_2x2_8[(y&1)^1][0];\
888 const int dr2= dither_2x2_8[y&1 ][1];\
889 const int dg2= dither_2x2_4[y&1 ][1];\
890 const int db2= dither_2x2_8[(y&1)^1][1];\
892 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
893 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
897 case PIX_FMT_RGB555:\
898 case PIX_FMT_BGR555:\
900 const int dr1= dither_2x2_8[y&1 ][0];\
901 const int dg1= dither_2x2_8[y&1 ][1];\
902 const int db1= dither_2x2_8[(y&1)^1][0];\
903 const int dr2= dither_2x2_8[y&1 ][1];\
904 const int dg2= dither_2x2_8[y&1 ][0];\
905 const int db2= dither_2x2_8[(y&1)^1][1];\
907 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
908 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
915 const uint8_t * const d64= dither_8x8_73[y&7];\
916 const uint8_t * const d32= dither_8x8_32[y&7];\
918 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
919 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
926 const uint8_t * const d64= dither_8x8_73 [y&7];\
927 const uint8_t * const d128=dither_8x8_220[y&7];\
929 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
930 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
934 case PIX_FMT_RGB4_BYTE:\
935 case PIX_FMT_BGR4_BYTE:\
937 const uint8_t * const d64= dither_8x8_73 [y&7];\
938 const uint8_t * const d128=dither_8x8_220[y&7];\
940 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
941 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
945 case PIX_FMT_MONOBLACK:\
946 case PIX_FMT_MONOWHITE:\
951 case PIX_FMT_YUYV422:\
953 ((uint8_t*)dest)[2*i2+0]= Y1;\
954 ((uint8_t*)dest)[2*i2+1]= U;\
955 ((uint8_t*)dest)[2*i2+2]= Y2;\
956 ((uint8_t*)dest)[2*i2+3]= V;\
959 case PIX_FMT_UYVY422:\
961 ((uint8_t*)dest)[2*i2+0]= U;\
962 ((uint8_t*)dest)[2*i2+1]= Y1;\
963 ((uint8_t*)dest)[2*i2+2]= V;\
964 ((uint8_t*)dest)[2*i2+3]= Y2;\
967 case PIX_FMT_GRAY16BE:\
969 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
970 ((uint8_t*)dest)[2*i2+1]= Y1;\
971 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
972 ((uint8_t*)dest)[2*i2+3]= Y2;\
975 case PIX_FMT_GRAY16LE:\
977 ((uint8_t*)dest)[2*i2+0]= Y1;\
978 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
979 ((uint8_t*)dest)[2*i2+2]= Y2;\
980 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
986 static inline void yuv2packedXinC(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
987 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
988 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
991 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)
994 static inline void yuv2rgbXinC_full(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
995 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
996 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
999 int step= fmt_depth(c->dstFormat)/8;
1002 switch(c->dstFormat) {
1010 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1011 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1012 dest[aidx]= needAlpha ? A : 255;
1019 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1020 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1028 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1045 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1046 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1047 dest[aidx]= needAlpha ? A : 255;
1054 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1055 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1063 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1078 static void fillPlane(uint8_t* plane, int stride, int width, int height, int y, uint8_t val)
1081 uint8_t *ptr = plane + stride*y;
1082 for (i=0; i<height; i++) {
1083 memset(ptr, val, width);
1088 static inline void rgb48ToY(uint8_t *dst, const uint8_t *src, int width)
1091 for (i = 0; i < width; i++) {
1096 dst[i] = (RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1100 static inline void rgb48ToUV(uint8_t *dstU, uint8_t *dstV,
1101 uint8_t *src1, uint8_t *src2, int width)
1105 for (i = 0; i < width; i++) {
1106 int r = src1[6*i + 0];
1107 int g = src1[6*i + 2];
1108 int b = src1[6*i + 4];
1110 dstU[i] = (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1111 dstV[i] = (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1115 static inline void rgb48ToUV_half(uint8_t *dstU, uint8_t *dstV,
1116 uint8_t *src1, uint8_t *src2, int width)
1120 for (i = 0; i < width; i++) {
1121 int r= src1[12*i + 0] + src1[12*i + 6];
1122 int g= src1[12*i + 2] + src1[12*i + 8];
1123 int b= src1[12*i + 4] + src1[12*i + 10];
1125 dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1126 dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1130 #define BGR2Y(type, name, shr, shg, shb, maskr, maskg, maskb, RY, GY, BY, S)\
1131 static inline void name(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)\
1134 for (i=0; i<width; i++) {\
1135 int b= (((const type*)src)[i]>>shb)&maskb;\
1136 int g= (((const type*)src)[i]>>shg)&maskg;\
1137 int r= (((const type*)src)[i]>>shr)&maskr;\
1139 dst[i]= (((RY)*r + (GY)*g + (BY)*b + (33<<((S)-1)))>>(S));\
1143 BGR2Y(uint32_t, bgr32ToY,16, 0, 0, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1144 BGR2Y(uint32_t, rgb32ToY, 0, 0,16, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1145 BGR2Y(uint16_t, bgr16ToY, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RY<<11, GY<<5, BY , RGB2YUV_SHIFT+8)
1146 BGR2Y(uint16_t, bgr15ToY, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RY<<10, GY<<5, BY , RGB2YUV_SHIFT+7)
1147 BGR2Y(uint16_t, rgb16ToY, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RY , GY<<5, BY<<11, RGB2YUV_SHIFT+8)
1148 BGR2Y(uint16_t, rgb15ToY, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RY , GY<<5, BY<<10, RGB2YUV_SHIFT+7)
1150 static inline void abgrToA(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1153 for (i=0; i<width; i++) {
1158 #define BGR2UV(type, name, shr, shg, shb, maska, maskr, maskg, maskb, RU, GU, BU, RV, GV, BV, S)\
1159 static inline void name(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1162 for (i=0; i<width; i++) {\
1163 int b= (((const type*)src)[i]&maskb)>>shb;\
1164 int g= (((const type*)src)[i]&maskg)>>shg;\
1165 int r= (((const type*)src)[i]&maskr)>>shr;\
1167 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<((S)-1)))>>(S);\
1168 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<((S)-1)))>>(S);\
1171 static inline void name ## _half(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1174 for (i=0; i<width; i++) {\
1175 int pix0= ((const type*)src)[2*i+0];\
1176 int pix1= ((const type*)src)[2*i+1];\
1177 int g= (pix0&~(maskr|maskb))+(pix1&~(maskr|maskb));\
1178 int b= ((pix0+pix1-g)&(maskb|(2*maskb)))>>shb;\
1179 int r= ((pix0+pix1-g)&(maskr|(2*maskr)))>>shr;\
1180 g&= maskg|(2*maskg);\
1184 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<(S)))>>((S)+1);\
1185 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<(S)))>>((S)+1);\
1189 BGR2UV(uint32_t, bgr32ToUV,16, 0, 0, 0xFF000000, 0xFF0000, 0xFF00, 0x00FF, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1190 BGR2UV(uint32_t, rgb32ToUV, 0, 0,16, 0xFF000000, 0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1191 BGR2UV(uint16_t, bgr16ToUV, 0, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RU<<11, GU<<5, BU , RV<<11, GV<<5, BV , RGB2YUV_SHIFT+8)
1192 BGR2UV(uint16_t, bgr15ToUV, 0, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RU<<10, GU<<5, BU , RV<<10, GV<<5, BV , RGB2YUV_SHIFT+7)
1193 BGR2UV(uint16_t, rgb16ToUV, 0, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RU , GU<<5, BU<<11, RV , GV<<5, BV<<11, RGB2YUV_SHIFT+8)
1194 BGR2UV(uint16_t, rgb15ToUV, 0, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RU , GU<<5, BU<<10, RV , GV<<5, BV<<10, RGB2YUV_SHIFT+7)
1196 static inline void palToY(uint8_t *dst, const uint8_t *src, long width, uint32_t *pal)
1199 for (i=0; i<width; i++) {
1202 dst[i]= pal[d] & 0xFF;
1206 static inline void palToUV(uint8_t *dstU, uint8_t *dstV,
1207 const uint8_t *src1, const uint8_t *src2,
1208 long width, uint32_t *pal)
1211 assert(src1 == src2);
1212 for (i=0; i<width; i++) {
1213 int p= pal[src1[i]];
1220 static inline void monowhite2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1223 for (i=0; i<width/8; i++) {
1226 dst[8*i+j]= ((d>>(7-j))&1)*255;
1230 static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1233 for (i=0; i<width/8; i++) {
1236 dst[8*i+j]= ((d>>(7-j))&1)*255;
1241 //Note: we have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
1243 #if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
1248 #if HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT
1249 #define COMPILE_ALTIVEC
1255 #if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1259 #if (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1260 #define COMPILE_MMX2
1263 #if ((HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1264 #define COMPILE_3DNOW
1268 #define COMPILE_TEMPLATE_MMX 0
1269 #define COMPILE_TEMPLATE_MMX2 0
1270 #define COMPILE_TEMPLATE_AMD3DNOW 0
1271 #define COMPILE_TEMPLATE_ALTIVEC 0
1274 #define RENAME(a) a ## _C
1275 #include "swscale_template.c"
1278 #ifdef COMPILE_ALTIVEC
1280 #undef COMPILE_TEMPLATE_ALTIVEC
1281 #define COMPILE_TEMPLATE_ALTIVEC 1
1282 #define RENAME(a) a ## _altivec
1283 #include "swscale_template.c"
1291 #undef COMPILE_TEMPLATE_MMX
1292 #undef COMPILE_TEMPLATE_MMX2
1293 #undef COMPILE_TEMPLATE_AMD3DNOW
1294 #define COMPILE_TEMPLATE_MMX 1
1295 #define COMPILE_TEMPLATE_MMX2 0
1296 #define COMPILE_TEMPLATE_AMD3DNOW 0
1297 #define RENAME(a) a ## _MMX
1298 #include "swscale_template.c"
1304 #undef COMPILE_TEMPLATE_MMX
1305 #undef COMPILE_TEMPLATE_MMX2
1306 #undef COMPILE_TEMPLATE_AMD3DNOW
1307 #define COMPILE_TEMPLATE_MMX 1
1308 #define COMPILE_TEMPLATE_MMX2 1
1309 #define COMPILE_TEMPLATE_AMD3DNOW 0
1310 #define RENAME(a) a ## _MMX2
1311 #include "swscale_template.c"
1315 #ifdef COMPILE_3DNOW
1317 #undef COMPILE_TEMPLATE_MMX
1318 #undef COMPILE_TEMPLATE_MMX2
1319 #undef COMPILE_TEMPLATE_AMD3DNOW
1320 #define COMPILE_TEMPLATE_MMX 1
1321 #define COMPILE_TEMPLATE_MMX2 0
1322 #define COMPILE_TEMPLATE_AMD3DNOW 1
1323 #define RENAME(a) a ## _3DNow
1324 #include "swscale_template.c"
1329 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1331 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1332 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1333 else return getSplineCoeff( 0.0,
1340 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1341 int srcW, int dstW, int filterAlign, int one, int flags,
1342 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1348 int64_t *filter=NULL;
1349 int64_t *filter2=NULL;
1350 const int64_t fone= 1LL<<54;
1353 if (flags & SWS_CPU_CAPS_MMX)
1354 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1357 // NOTE: the +1 is for the MMX scaler which reads over the end
1358 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
1360 if (FFABS(xInc - 0x10000) <10) { // unscaled
1363 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1365 for (i=0; i<dstW; i++) {
1366 filter[i*filterSize]= fone;
1370 } else if (flags&SWS_POINT) { // lame looking point sampling mode
1374 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1376 xDstInSrc= xInc/2 - 0x8000;
1377 for (i=0; i<dstW; i++) {
1378 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1380 (*filterPos)[i]= xx;
1384 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
1388 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1390 xDstInSrc= xInc/2 - 0x8000;
1391 for (i=0; i<dstW; i++) {
1392 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1395 (*filterPos)[i]= xx;
1396 //bilinear upscale / linear interpolate / area averaging
1397 for (j=0; j<filterSize; j++) {
1398 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1399 if (coeff<0) coeff=0;
1400 filter[i*filterSize + j]= coeff;
1409 if (flags&SWS_BICUBIC) sizeFactor= 4;
1410 else if (flags&SWS_X) sizeFactor= 8;
1411 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1412 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1413 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1414 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1415 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1416 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1418 sizeFactor= 0; //GCC warning killer
1422 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1423 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1425 if (filterSize > srcW-2) filterSize=srcW-2;
1427 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1429 xDstInSrc= xInc - 0x10000;
1430 for (i=0; i<dstW; i++) {
1431 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1433 (*filterPos)[i]= xx;
1434 for (j=0; j<filterSize; j++) {
1435 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1441 floatd= d * (1.0/(1<<30));
1443 if (flags & SWS_BICUBIC) {
1444 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1445 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1446 int64_t dd = ( d*d)>>30;
1447 int64_t ddd= (dd*d)>>30;
1450 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1451 else if (d < 1LL<<31)
1452 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1455 coeff *= fone>>(30+24);
1457 /* else if (flags & SWS_X) {
1458 double p= param ? param*0.01 : 0.3;
1459 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1460 coeff*= pow(2.0, - p*d*d);
1462 else if (flags & SWS_X) {
1463 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1470 if (c<0.0) c= -pow(-c, A);
1472 coeff= (c*0.5 + 0.5)*fone;
1473 } else if (flags & SWS_AREA) {
1474 int64_t d2= d - (1<<29);
1475 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1476 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1478 coeff *= fone>>(30+16);
1479 } else if (flags & SWS_GAUSS) {
1480 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1481 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1482 } else if (flags & SWS_SINC) {
1483 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1484 } else if (flags & SWS_LANCZOS) {
1485 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1486 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1487 if (floatd>p) coeff=0;
1488 } else if (flags & SWS_BILINEAR) {
1490 if (coeff<0) coeff=0;
1491 coeff *= fone >> 30;
1492 } else if (flags & SWS_SPLINE) {
1493 double p=-2.196152422706632;
1494 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1496 coeff= 0.0; //GCC warning killer
1500 filter[i*filterSize + j]= coeff;
1507 /* apply src & dst Filter to filter -> filter2
1510 assert(filterSize>0);
1511 filter2Size= filterSize;
1512 if (srcFilter) filter2Size+= srcFilter->length - 1;
1513 if (dstFilter) filter2Size+= dstFilter->length - 1;
1514 assert(filter2Size>0);
1515 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
1517 for (i=0; i<dstW; i++) {
1521 for (k=0; k<srcFilter->length; k++) {
1522 for (j=0; j<filterSize; j++)
1523 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1526 for (j=0; j<filterSize; j++)
1527 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1531 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1535 /* try to reduce the filter-size (step1 find size and shift left) */
1536 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1538 for (i=dstW-1; i>=0; i--) {
1539 int min= filter2Size;
1543 /* get rid off near zero elements on the left by shifting left */
1544 for (j=0; j<filter2Size; j++) {
1546 cutOff += FFABS(filter2[i*filter2Size]);
1548 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1550 /* preserve monotonicity because the core can't handle the filter otherwise */
1551 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1553 // move filter coefficients left
1554 for (k=1; k<filter2Size; k++)
1555 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1556 filter2[i*filter2Size + k - 1]= 0;
1561 /* count near zeros on the right */
1562 for (j=filter2Size-1; j>0; j--) {
1563 cutOff += FFABS(filter2[i*filter2Size + j]);
1565 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1569 if (min>minFilterSize) minFilterSize= min;
1572 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1573 // we can handle the special case 4,
1574 // so we don't want to go to the full 8
1575 if (minFilterSize < 5)
1578 // We really don't want to waste our time
1579 // doing useless computation, so fall back on
1580 // the scalar C code for very small filters.
1581 // Vectorizing is worth it only if you have a
1582 // decent-sized vector.
1583 if (minFilterSize < 3)
1587 if (flags & SWS_CPU_CAPS_MMX) {
1588 // special case for unscaled vertical filtering
1589 if (minFilterSize == 1 && filterAlign == 2)
1593 assert(minFilterSize > 0);
1594 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1595 assert(filterSize > 0);
1596 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1597 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1599 *outFilterSize= filterSize;
1601 if (flags&SWS_PRINT_INFO)
1602 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1603 /* try to reduce the filter-size (step2 reduce it) */
1604 for (i=0; i<dstW; i++) {
1607 for (j=0; j<filterSize; j++) {
1608 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1609 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1610 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1611 filter[i*filterSize + j]= 0;
1616 //FIXME try to align filterPos if possible
1619 for (i=0; i<dstW; i++) {
1621 if ((*filterPos)[i] < 0) {
1622 // move filter coefficients left to compensate for filterPos
1623 for (j=1; j<filterSize; j++) {
1624 int left= FFMAX(j + (*filterPos)[i], 0);
1625 filter[i*filterSize + left] += filter[i*filterSize + j];
1626 filter[i*filterSize + j]=0;
1631 if ((*filterPos)[i] + filterSize > srcW) {
1632 int shift= (*filterPos)[i] + filterSize - srcW;
1633 // move filter coefficients right to compensate for filterPos
1634 for (j=filterSize-2; j>=0; j--) {
1635 int right= FFMIN(j + shift, filterSize-1);
1636 filter[i*filterSize +right] += filter[i*filterSize +j];
1637 filter[i*filterSize +j]=0;
1639 (*filterPos)[i]= srcW - filterSize;
1643 // Note the +1 is for the MMX scaler which reads over the end
1644 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1645 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
1647 /* normalize & store in outFilter */
1648 for (i=0; i<dstW; i++) {
1653 for (j=0; j<filterSize; j++) {
1654 sum+= filter[i*filterSize + j];
1656 sum= (sum + one/2)/ one;
1657 for (j=0; j<*outFilterSize; j++) {
1658 int64_t v= filter[i*filterSize + j] + error;
1659 int intV= ROUNDED_DIV(v, sum);
1660 (*outFilter)[i*(*outFilterSize) + j]= intV;
1661 error= v - intV*sum;
1665 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1666 for (i=0; i<*outFilterSize; i++) {
1667 int j= dstW*(*outFilterSize);
1668 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1679 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
1682 x86_reg imm8OfPShufW1A;
1683 x86_reg imm8OfPShufW2A;
1684 x86_reg fragmentLengthA;
1686 x86_reg imm8OfPShufW1B;
1687 x86_reg imm8OfPShufW2B;
1688 x86_reg fragmentLengthB;
1693 // create an optimized horizontal scaling routine
1694 /* This scaler is made of runtime-generated MMX2 code using specially
1695 * tuned pshufw instructions. For every four output pixels, if four
1696 * input pixels are enough for the fast bilinear scaling, then a chunk
1697 * of fragmentB is used. If five input pixels are needed, then a chunk
1698 * of fragmentA is used.
1707 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1708 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1709 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1710 "punpcklbw %%mm7, %%mm1 \n\t"
1711 "punpcklbw %%mm7, %%mm0 \n\t"
1712 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1714 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1716 "psubw %%mm1, %%mm0 \n\t"
1717 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1718 "pmullw %%mm3, %%mm0 \n\t"
1719 "psllw $7, %%mm1 \n\t"
1720 "paddw %%mm1, %%mm0 \n\t"
1722 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1724 "add $8, %%"REG_a" \n\t"
1728 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1729 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1730 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1735 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1739 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1740 "=r" (fragmentLengthA)
1747 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1748 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1749 "punpcklbw %%mm7, %%mm0 \n\t"
1750 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1752 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1754 "psubw %%mm1, %%mm0 \n\t"
1755 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1756 "pmullw %%mm3, %%mm0 \n\t"
1757 "psllw $7, %%mm1 \n\t"
1758 "paddw %%mm1, %%mm0 \n\t"
1760 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1762 "add $8, %%"REG_a" \n\t"
1766 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1767 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1768 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1773 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1777 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1778 "=r" (fragmentLengthB)
1781 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1784 for (i=0; i<dstW/numSplits; i++) {
1789 int b=((xpos+xInc)>>16) - xx;
1790 int c=((xpos+xInc*2)>>16) - xx;
1791 int d=((xpos+xInc*3)>>16) - xx;
1793 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
1794 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
1795 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
1796 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
1797 int maxShift= 3-(d+inc);
1801 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1802 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1803 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1804 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1807 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
1809 filterCode[fragmentPos + imm8OfPShufW1]=
1810 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
1811 filterCode[fragmentPos + imm8OfPShufW2]=
1812 a | (b<<2) | (c<<4) | (d<<6);
1814 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
1815 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1817 if (shift && i>=shift) {
1818 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
1819 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
1820 filterPos[i/2]-=shift;
1824 fragmentPos+= fragmentLength;
1827 filterCode[fragmentPos]= RET;
1832 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
1834 return fragmentPos + 1;
1836 #endif /* COMPILE_MMX2 */
1838 static void globalInit(void)
1840 // generating tables:
1842 for (i=0; i<768; i++) {
1843 int c= av_clip_uint8(i-256);
1848 static SwsFunc getSwsFunc(SwsContext *c)
1850 #if CONFIG_RUNTIME_CPUDETECT
1851 int flags = c->flags;
1853 #if ARCH_X86 && CONFIG_GPL
1854 // ordered per speed fastest first
1855 if (flags & SWS_CPU_CAPS_MMX2) {
1856 sws_init_swScale_MMX2(c);
1857 return swScale_MMX2;
1858 } else if (flags & SWS_CPU_CAPS_3DNOW) {
1859 sws_init_swScale_3DNow(c);
1860 return swScale_3DNow;
1861 } else if (flags & SWS_CPU_CAPS_MMX) {
1862 sws_init_swScale_MMX(c);
1865 sws_init_swScale_C(c);
1871 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1872 sws_init_swScale_altivec(c);
1873 return swScale_altivec;
1875 sws_init_swScale_C(c);
1879 sws_init_swScale_C(c);
1881 #endif /* ARCH_X86 && CONFIG_GPL */
1882 #else //CONFIG_RUNTIME_CPUDETECT
1883 #if COMPILE_TEMPLATE_MMX2
1884 sws_init_swScale_MMX2(c);
1885 return swScale_MMX2;
1886 #elif COMPILE_TEMPLATE_AMD3DNOW
1887 sws_init_swScale_3DNow(c);
1888 return swScale_3DNow;
1889 #elif COMPILE_TEMPLATE_MMX
1890 sws_init_swScale_MMX(c);
1892 #elif COMPILE_TEMPLATE_ALTIVEC
1893 sws_init_swScale_altivec(c);
1894 return swScale_altivec;
1896 sws_init_swScale_C(c);
1899 #endif //!CONFIG_RUNTIME_CPUDETECT
1902 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1903 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1905 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1907 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1908 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1911 const uint8_t *srcPtr= src[0];
1912 uint8_t *dstPtr= dst;
1913 for (i=0; i<srcSliceH; i++) {
1914 memcpy(dstPtr, srcPtr, c->srcW);
1915 srcPtr+= srcStride[0];
1916 dstPtr+= dstStride[0];
1919 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1920 if (c->dstFormat == PIX_FMT_NV12)
1921 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1923 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1928 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1929 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1931 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1933 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1938 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1939 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1941 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1943 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1948 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1949 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1951 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1953 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1958 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1959 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1961 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1963 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1968 static int YUYV2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1969 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1971 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1972 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
1973 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
1975 yuyvtoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1978 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
1983 static int YUYV2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1984 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1986 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1987 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
1988 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
1990 yuyvtoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1995 static int UYVY2YUV420Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1996 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1998 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1999 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
2000 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
2002 uyvytoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2005 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2010 static int UYVY2YUV422Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2011 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2013 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2014 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
2015 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
2017 uyvytoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2022 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2023 int srcSliceH, uint8_t* dst[], int dstStride[])
2025 const enum PixelFormat srcFormat= c->srcFormat;
2026 const enum PixelFormat dstFormat= c->dstFormat;
2027 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
2028 const uint8_t *palette)=NULL;
2030 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2031 uint8_t *srcPtr= src[0];
2033 if (!usePal(srcFormat))
2034 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2035 sws_format_name(srcFormat), sws_format_name(dstFormat));
2038 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
2039 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
2040 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
2041 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
2042 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
2043 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
2044 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2045 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2049 for (i=0; i<srcSliceH; i++) {
2050 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
2051 srcPtr+= srcStride[0];
2052 dstPtr+= dstStride[0];
2058 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
2059 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2060 int srcSliceH, uint8_t* dst[], int dstStride[])
2062 const enum PixelFormat srcFormat= c->srcFormat;
2063 const enum PixelFormat dstFormat= c->dstFormat;
2064 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
2065 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
2066 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
2067 const int dstId= fmt_depth(dstFormat) >> 2;
2068 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
2071 if ( (isBGR(srcFormat) && isBGR(dstFormat))
2072 || (isRGB(srcFormat) && isRGB(dstFormat))) {
2073 switch(srcId | (dstId<<4)) {
2074 case 0x34: conv= rgb16to15; break;
2075 case 0x36: conv= rgb24to15; break;
2076 case 0x38: conv= rgb32to15; break;
2077 case 0x43: conv= rgb15to16; break;
2078 case 0x46: conv= rgb24to16; break;
2079 case 0x48: conv= rgb32to16; break;
2080 case 0x63: conv= rgb15to24; break;
2081 case 0x64: conv= rgb16to24; break;
2082 case 0x68: conv= rgb32to24; break;
2083 case 0x83: conv= rgb15to32; break;
2084 case 0x84: conv= rgb16to32; break;
2085 case 0x86: conv= rgb24to32; break;
2086 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2087 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2089 } else if ( (isBGR(srcFormat) && isRGB(dstFormat))
2090 || (isRGB(srcFormat) && isBGR(dstFormat))) {
2091 switch(srcId | (dstId<<4)) {
2092 case 0x33: conv= rgb15tobgr15; break;
2093 case 0x34: conv= rgb16tobgr15; break;
2094 case 0x36: conv= rgb24tobgr15; break;
2095 case 0x38: conv= rgb32tobgr15; break;
2096 case 0x43: conv= rgb15tobgr16; break;
2097 case 0x44: conv= rgb16tobgr16; break;
2098 case 0x46: conv= rgb24tobgr16; break;
2099 case 0x48: conv= rgb32tobgr16; break;
2100 case 0x63: conv= rgb15tobgr24; break;
2101 case 0x64: conv= rgb16tobgr24; break;
2102 case 0x66: conv= rgb24tobgr24; break;
2103 case 0x68: conv= rgb32tobgr24; break;
2104 case 0x83: conv= rgb15tobgr32; break;
2105 case 0x84: conv= rgb16tobgr32; break;
2106 case 0x86: conv= rgb24tobgr32; break;
2107 case 0x88: conv= rgb32tobgr32; break;
2108 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2109 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
2112 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2113 sws_format_name(srcFormat), sws_format_name(dstFormat));
2117 uint8_t *srcPtr= src[0];
2118 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
2119 srcPtr += ALT32_CORR;
2121 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
2122 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
2125 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2127 for (i=0; i<srcSliceH; i++) {
2128 conv(srcPtr, dstPtr, c->srcW*srcBpp);
2129 srcPtr+= srcStride[0];
2130 dstPtr+= dstStride[0];
2137 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2138 int srcSliceH, uint8_t* dst[], int dstStride[])
2143 dst[0]+ srcSliceY *dstStride[0],
2144 dst[1]+(srcSliceY>>1)*dstStride[1],
2145 dst[2]+(srcSliceY>>1)*dstStride[2],
2147 dstStride[0], dstStride[1], srcStride[0]);
2149 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2153 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2154 int srcSliceH, uint8_t* dst[], int dstStride[])
2159 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
2160 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
2162 uint8_t *srcPtr= src[0];
2163 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2165 for (i=0; i<srcSliceH; i++) {
2166 memcpy(dstPtr, srcPtr, c->srcW);
2167 srcPtr+= srcStride[0];
2168 dstPtr+= dstStride[0];
2172 if (c->dstFormat==PIX_FMT_YUV420P || c->dstFormat==PIX_FMT_YUVA420P) {
2173 planar2x(src[1], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2174 srcSliceH >> 2, srcStride[1], dstStride[1]);
2175 planar2x(src[2], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2176 srcSliceH >> 2, srcStride[2], dstStride[2]);
2178 planar2x(src[1], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2179 srcSliceH >> 2, srcStride[1], dstStride[2]);
2180 planar2x(src[2], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2181 srcSliceH >> 2, srcStride[2], dstStride[1]);
2184 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2188 /* unscaled copy like stuff (assumes nearly identical formats) */
2189 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2190 int srcSliceH, uint8_t* dst[], int dstStride[])
2192 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
2193 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
2196 uint8_t *srcPtr= src[0];
2197 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2200 /* universal length finder */
2201 while(length+c->srcW <= FFABS(dstStride[0])
2202 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
2205 for (i=0; i<srcSliceH; i++) {
2206 memcpy(dstPtr, srcPtr, length);
2207 srcPtr+= srcStride[0];
2208 dstPtr+= dstStride[0];
2214 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2215 int srcSliceH, uint8_t* dst[], int dstStride[])
2218 for (plane=0; plane<4; plane++) {
2219 int length= (plane==0 || plane==3) ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
2220 int y= (plane==0 || plane==3) ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
2221 int height= (plane==0 || plane==3) ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
2222 uint8_t *srcPtr= src[plane];
2223 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
2225 if (!dst[plane]) continue;
2226 // ignore palette for GRAY8
2227 if (plane == 1 && !dst[2]) continue;
2228 if (!src[plane] || (plane == 1 && !src[2])) {
2229 if(is16BPS(c->dstFormat))
2231 fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
2233 if(is16BPS(c->srcFormat) && !is16BPS(c->dstFormat)) {
2234 if (!isBE(c->srcFormat)) srcPtr++;
2235 for (i=0; i<height; i++) {
2236 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
2237 srcPtr+= srcStride[plane];
2238 dstPtr+= dstStride[plane];
2240 } else if(!is16BPS(c->srcFormat) && is16BPS(c->dstFormat)) {
2241 for (i=0; i<height; i++) {
2242 for (j=0; j<length; j++) {
2243 dstPtr[ j<<1 ] = srcPtr[j];
2244 dstPtr[(j<<1)+1] = srcPtr[j];
2246 srcPtr+= srcStride[plane];
2247 dstPtr+= dstStride[plane];
2249 } else if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat)
2250 && isBE(c->srcFormat) != isBE(c->dstFormat)) {
2252 for (i=0; i<height; i++) {
2253 for (j=0; j<length; j++)
2254 ((uint16_t*)dstPtr)[j] = bswap_16(((uint16_t*)srcPtr)[j]);
2255 srcPtr+= srcStride[plane];
2256 dstPtr+= dstStride[plane];
2258 } else if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
2259 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
2261 if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat))
2263 for (i=0; i<height; i++) {
2264 memcpy(dstPtr, srcPtr, length);
2265 srcPtr+= srcStride[plane];
2266 dstPtr+= dstStride[plane];
2275 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
2277 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
2278 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
2281 static uint16_t roundToInt16(int64_t f)
2283 int r= (f + (1<<15))>>16;
2284 if (r<-0x7FFF) return 0x8000;
2285 else if (r> 0x7FFF) return 0x7FFF;
2289 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
2291 int64_t crv = inv_table[0];
2292 int64_t cbu = inv_table[1];
2293 int64_t cgu = -inv_table[2];
2294 int64_t cgv = -inv_table[3];
2298 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2299 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2301 c->brightness= brightness;
2302 c->contrast = contrast;
2303 c->saturation= saturation;
2304 c->srcRange = srcRange;
2305 c->dstRange = dstRange;
2306 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2308 c->uOffset= 0x0400040004000400LL;
2309 c->vOffset= 0x0400040004000400LL;
2315 crv= (crv*224) / 255;
2316 cbu= (cbu*224) / 255;
2317 cgu= (cgu*224) / 255;
2318 cgv= (cgv*224) / 255;
2321 cy = (cy *contrast )>>16;
2322 crv= (crv*contrast * saturation)>>32;
2323 cbu= (cbu*contrast * saturation)>>32;
2324 cgu= (cgu*contrast * saturation)>>32;
2325 cgv= (cgv*contrast * saturation)>>32;
2327 oy -= 256*brightness;
2329 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2330 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2331 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2332 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2333 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2334 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2336 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2337 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2338 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2339 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2340 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2341 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2343 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2346 #ifdef COMPILE_ALTIVEC
2347 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2348 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
2353 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
2355 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2357 *inv_table = c->srcColorspaceTable;
2358 *table = c->dstColorspaceTable;
2359 *srcRange = c->srcRange;
2360 *dstRange = c->dstRange;
2361 *brightness= c->brightness;
2362 *contrast = c->contrast;
2363 *saturation= c->saturation;
2368 static int handle_jpeg(enum PixelFormat *format)
2371 case PIX_FMT_YUVJ420P:
2372 *format = PIX_FMT_YUV420P;
2374 case PIX_FMT_YUVJ422P:
2375 *format = PIX_FMT_YUV422P;
2377 case PIX_FMT_YUVJ444P:
2378 *format = PIX_FMT_YUV444P;
2380 case PIX_FMT_YUVJ440P:
2381 *format = PIX_FMT_YUV440P;
2388 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2389 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
2394 int usesVFilter, usesHFilter;
2395 int unscaled, needsDither;
2396 int srcRange, dstRange;
2397 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2399 if (flags & SWS_CPU_CAPS_MMX)
2400 __asm__ volatile("emms\n\t"::: "memory");
2403 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
2404 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2405 #if COMPILE_TEMPLATE_MMX2
2406 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2407 #elif COMPILE_TEMPLATE_AMD3DNOW
2408 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2409 #elif COMPILE_TEMPLATE_MMX
2410 flags |= SWS_CPU_CAPS_MMX;
2411 #elif COMPILE_TEMPLATE_ALTIVEC
2412 flags |= SWS_CPU_CAPS_ALTIVEC;
2414 flags |= SWS_CPU_CAPS_BFIN;
2416 #endif /* CONFIG_RUNTIME_CPUDETECT */
2417 if (clip_table[512] != 255) globalInit();
2418 if (!rgb15to16) sws_rgb2rgb_init(flags);
2420 unscaled = (srcW == dstW && srcH == dstH);
2421 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2422 && (fmt_depth(dstFormat))<24
2423 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2425 srcRange = handle_jpeg(&srcFormat);
2426 dstRange = handle_jpeg(&dstFormat);
2428 if (!isSupportedIn(srcFormat)) {
2429 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2432 if (!isSupportedOut(dstFormat)) {
2433 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2437 i= flags & ( SWS_POINT
2448 if(!i || (i & (i-1))) {
2449 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
2454 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
2455 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2456 srcW, srcH, dstW, dstH);
2459 if(srcW > VOFW || dstW > VOFW) {
2460 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2464 if (!dstFilter) dstFilter= &dummyFilter;
2465 if (!srcFilter) srcFilter= &dummyFilter;
2467 FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
2469 c->av_class = &sws_context_class;
2474 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2475 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2477 c->dstFormat= dstFormat;
2478 c->srcFormat= srcFormat;
2479 c->vRounder= 4* 0x0001000100010001ULL;
2481 usesHFilter= usesVFilter= 0;
2482 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2483 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2484 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2485 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2486 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2487 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2488 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2489 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2491 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2492 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2494 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
2495 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2497 // drop some chroma lines if the user wants it
2498 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2499 c->chrSrcVSubSample+= c->vChrDrop;
2501 // drop every other pixel for chroma calculation unless user wants full chroma
2502 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2503 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2504 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2505 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2506 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2507 c->chrSrcHSubSample=1;
2510 c->param[0] = param[0];
2511 c->param[1] = param[1];
2514 c->param[1] = SWS_PARAM_DEFAULT;
2517 // Note the -((-x)>>y) is so that we always round toward +inf.
2518 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2519 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2520 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2521 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2523 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2525 /* unscaled special cases */
2526 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat))) {
2528 if ((srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21)) {
2529 c->swScale= PlanarToNV12Wrapper;
2532 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P || srcFormat==PIX_FMT_YUVA420P) && (isBGR(dstFormat) || isRGB(dstFormat))
2533 && !(flags & SWS_ACCURATE_RND) && !(dstH&1)) {
2534 c->swScale= ff_yuv2rgb_get_func_ptr(c);
2537 if (srcFormat==PIX_FMT_YUV410P && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_BITEXACT)) {
2538 c->swScale= yvu9toyv12Wrapper;
2542 if (srcFormat==PIX_FMT_BGR24 && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_ACCURATE_RND))
2543 c->swScale= bgr24toyv12Wrapper;
2545 /* RGB/BGR -> RGB/BGR (no dither needed forms) */
2546 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2547 && (isBGR(dstFormat) || isRGB(dstFormat))
2548 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2549 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2550 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2551 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2552 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2553 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2554 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2555 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2556 && dstFormat != PIX_FMT_RGB32_1
2557 && dstFormat != PIX_FMT_BGR32_1
2558 && srcFormat != PIX_FMT_RGB48LE && dstFormat != PIX_FMT_RGB48LE
2559 && srcFormat != PIX_FMT_RGB48BE && dstFormat != PIX_FMT_RGB48BE
2560 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2561 c->swScale= rgb2rgbWrapper;
2563 if ((usePal(srcFormat) && (
2564 dstFormat == PIX_FMT_RGB32 ||
2565 dstFormat == PIX_FMT_RGB32_1 ||
2566 dstFormat == PIX_FMT_RGB24 ||
2567 dstFormat == PIX_FMT_BGR32 ||
2568 dstFormat == PIX_FMT_BGR32_1 ||
2569 dstFormat == PIX_FMT_BGR24)))
2570 c->swScale= pal2rgbWrapper;
2572 if (srcFormat == PIX_FMT_YUV422P) {
2573 if (dstFormat == PIX_FMT_YUYV422)
2574 c->swScale= YUV422PToYuy2Wrapper;
2575 else if (dstFormat == PIX_FMT_UYVY422)
2576 c->swScale= YUV422PToUyvyWrapper;
2579 /* LQ converters if -sws 0 or -sws 4*/
2580 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)) {
2582 if (srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) {
2583 if (dstFormat == PIX_FMT_YUYV422)
2584 c->swScale= PlanarToYuy2Wrapper;
2585 else if (dstFormat == PIX_FMT_UYVY422)
2586 c->swScale= PlanarToUyvyWrapper;
2589 if(srcFormat == PIX_FMT_YUYV422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2590 c->swScale= YUYV2YUV420Wrapper;
2591 if(srcFormat == PIX_FMT_UYVY422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2592 c->swScale= UYVY2YUV420Wrapper;
2593 if(srcFormat == PIX_FMT_YUYV422 && dstFormat == PIX_FMT_YUV422P)
2594 c->swScale= YUYV2YUV422Wrapper;
2595 if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
2596 c->swScale= UYVY2YUV422Wrapper;
2598 #ifdef COMPILE_ALTIVEC
2599 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2600 !(c->flags & SWS_BITEXACT) &&
2601 srcFormat == PIX_FMT_YUV420P) {
2602 // unscaled YV12 -> packed YUV, we want speed
2603 if (dstFormat == PIX_FMT_YUYV422)
2604 c->swScale= yv12toyuy2_unscaled_altivec;
2605 else if (dstFormat == PIX_FMT_UYVY422)
2606 c->swScale= yv12touyvy_unscaled_altivec;
2611 if ( srcFormat == dstFormat
2612 || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P)
2613 || (srcFormat == PIX_FMT_YUV420P && dstFormat == PIX_FMT_YUVA420P)
2614 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2615 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2616 || (isGray(dstFormat) && isGray(srcFormat))
2617 || (isPlanarYUV(srcFormat) && isPlanarYUV(dstFormat)
2618 && c->chrDstHSubSample == c->chrSrcHSubSample
2619 && c->chrDstVSubSample == c->chrSrcVSubSample
2620 && dstFormat != PIX_FMT_NV12 && dstFormat != PIX_FMT_NV21
2621 && srcFormat != PIX_FMT_NV12 && srcFormat != PIX_FMT_NV21))
2623 if (isPacked(c->srcFormat))
2624 c->swScale= packedCopy;
2625 else /* Planar YUV or gray */
2626 c->swScale= planarCopy;
2629 if (flags & SWS_CPU_CAPS_BFIN)
2630 ff_bfin_get_unscaled_swscale (c);
2634 if (flags&SWS_PRINT_INFO)
2635 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2636 sws_format_name(srcFormat), sws_format_name(dstFormat));
2641 if (flags & SWS_CPU_CAPS_MMX2) {
2642 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2643 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
2644 if (flags&SWS_PRINT_INFO)
2645 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
2647 if (usesHFilter) c->canMMX2BeUsed=0;
2652 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2653 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2655 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2656 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2657 // n-2 is the last chrominance sample available
2658 // this is not perfect, but no one should notice the difference, the more correct variant
2659 // would be like the vertical one, but that would require some special code for the
2660 // first and last pixel
2661 if (flags&SWS_FAST_BILINEAR) {
2662 if (c->canMMX2BeUsed) {
2666 //we don't use the x86 asm scaler if MMX is available
2667 else if (flags & SWS_CPU_CAPS_MMX) {
2668 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2669 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2673 /* precalculate horizontal scaler filter coefficients */
2675 const int filterAlign=
2676 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2677 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2680 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2681 srcW , dstW, filterAlign, 1<<14,
2682 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2683 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
2685 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2686 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2687 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2688 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
2691 #if defined(COMPILE_MMX2)
2692 // can't downscale !!!
2693 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
2694 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
2695 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
2697 #ifdef MAP_ANONYMOUS
2698 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2699 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2700 #elif HAVE_VIRTUALALLOC
2701 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2702 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2704 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
2705 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
2708 FF_ALLOCZ_OR_GOTO(c, c->lumMmx2Filter , (dstW /8+8)*sizeof(int16_t), fail);
2709 FF_ALLOCZ_OR_GOTO(c, c->chrMmx2Filter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
2710 FF_ALLOCZ_OR_GOTO(c, c->lumMmx2FilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
2711 FF_ALLOCZ_OR_GOTO(c, c->chrMmx2FilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
2713 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2714 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2716 #ifdef MAP_ANONYMOUS
2717 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2718 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2721 #endif /* defined(COMPILE_MMX2) */
2722 } // initialize horizontal stuff
2726 /* precalculate vertical scaler filter coefficients */
2728 const int filterAlign=
2729 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2730 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2733 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2734 srcH , dstH, filterAlign, (1<<12),
2735 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2736 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
2738 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2739 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2740 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2741 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
2744 #ifdef COMPILE_ALTIVEC
2745 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
2746 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
2748 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2750 short *p = (short *)&c->vYCoeffsBank[i];
2752 p[j] = c->vLumFilter[i];
2755 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2757 short *p = (short *)&c->vCCoeffsBank[i];
2759 p[j] = c->vChrFilter[i];
2764 // calculate buffer sizes so that they won't run out while handling these damn slices
2765 c->vLumBufSize= c->vLumFilterSize;
2766 c->vChrBufSize= c->vChrFilterSize;
2767 for (i=0; i<dstH; i++) {
2768 int chrI= i*c->chrDstH / dstH;
2769 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2770 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2772 nextSlice>>= c->chrSrcVSubSample;
2773 nextSlice<<= c->chrSrcVSubSample;
2774 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2775 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2776 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2777 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2780 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2781 // allocate several megabytes to handle all possible cases)
2782 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
2783 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
2784 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
2785 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
2786 //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)
2787 /* align at 16 bytes for AltiVec */
2788 for (i=0; i<c->vLumBufSize; i++) {
2789 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
2790 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
2792 for (i=0; i<c->vChrBufSize; i++) {
2793 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
2794 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
2796 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
2797 for (i=0; i<c->vLumBufSize; i++) {
2798 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
2799 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
2802 //try to avoid drawing green stuff between the right end and the stride end
2803 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2805 assert(2*VOFW == VOF);
2807 assert(c->chrDstH <= dstH);
2809 if (flags&SWS_PRINT_INFO) {
2811 const char *dither= " dithered";
2813 const char *dither= "";
2815 if (flags&SWS_FAST_BILINEAR)
2816 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2817 else if (flags&SWS_BILINEAR)
2818 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2819 else if (flags&SWS_BICUBIC)
2820 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2821 else if (flags&SWS_X)
2822 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2823 else if (flags&SWS_POINT)
2824 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2825 else if (flags&SWS_AREA)
2826 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2827 else if (flags&SWS_BICUBLIN)
2828 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2829 else if (flags&SWS_GAUSS)
2830 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2831 else if (flags&SWS_SINC)
2832 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2833 else if (flags&SWS_LANCZOS)
2834 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2835 else if (flags&SWS_SPLINE)
2836 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2838 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2840 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2841 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2842 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2844 av_log(c, AV_LOG_INFO, "from %s to %s ",
2845 sws_format_name(srcFormat), sws_format_name(dstFormat));
2847 if (flags & SWS_CPU_CAPS_MMX2)
2848 av_log(c, AV_LOG_INFO, "using MMX2\n");
2849 else if (flags & SWS_CPU_CAPS_3DNOW)
2850 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2851 else if (flags & SWS_CPU_CAPS_MMX)
2852 av_log(c, AV_LOG_INFO, "using MMX\n");
2853 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2854 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2856 av_log(c, AV_LOG_INFO, "using C\n");
2859 if (flags & SWS_PRINT_INFO) {
2860 if (flags & SWS_CPU_CAPS_MMX) {
2861 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2862 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2864 if (c->hLumFilterSize==4)
2865 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2866 else if (c->hLumFilterSize==8)
2867 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2869 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2871 if (c->hChrFilterSize==4)
2872 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2873 else if (c->hChrFilterSize==8)
2874 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2876 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2880 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
2882 if (flags & SWS_FAST_BILINEAR)
2883 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2885 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2888 if (isPlanarYUV(dstFormat)) {
2889 if (c->vLumFilterSize==1)
2890 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2892 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2894 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2895 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2896 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2897 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2898 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2900 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2903 if (dstFormat==PIX_FMT_BGR24)
2904 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
2905 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2906 else if (dstFormat==PIX_FMT_RGB32)
2907 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2908 else if (dstFormat==PIX_FMT_BGR565)
2909 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2910 else if (dstFormat==PIX_FMT_BGR555)
2911 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2913 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2915 if (flags & SWS_PRINT_INFO) {
2916 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2917 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2918 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2919 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2922 c->swScale= getSwsFunc(c);
2930 static void reset_ptr(uint8_t* src[], int format)
2932 if(!isALPHA(format))
2934 if(!isPlanarYUV(format)) {
2936 if( format != PIX_FMT_PAL8
2937 && format != PIX_FMT_RGB8
2938 && format != PIX_FMT_BGR8
2939 && format != PIX_FMT_RGB4_BYTE
2940 && format != PIX_FMT_BGR4_BYTE
2947 * swscale wrapper, so we don't need to export the SwsContext.
2948 * Assumes planar YUV to be in YUV order instead of YVU.
2950 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2951 int srcSliceH, uint8_t* dst[], int dstStride[])
2954 uint8_t* src2[4]= {src[0], src[1], src[2], src[3]};
2955 uint8_t* dst2[4]= {dst[0], dst[1], dst[2], dst[3]};
2957 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2958 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2961 if (c->sliceDir == 0) {
2962 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2965 if (usePal(c->srcFormat)) {
2966 for (i=0; i<256; i++) {
2967 int p, r, g, b,y,u,v;
2968 if(c->srcFormat == PIX_FMT_PAL8) {
2969 p=((uint32_t*)(src[1]))[i];
2973 } else if(c->srcFormat == PIX_FMT_RGB8) {
2977 } else if(c->srcFormat == PIX_FMT_BGR8) {
2981 } else if(c->srcFormat == PIX_FMT_RGB4_BYTE) {
2986 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2991 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2992 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2993 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2994 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2997 switch(c->dstFormat) {
3002 c->pal_rgb[i]= r + (g<<8) + (b<<16);
3004 case PIX_FMT_BGR32_1:
3008 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
3010 case PIX_FMT_RGB32_1:
3014 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
3021 c->pal_rgb[i]= b + (g<<8) + (r<<16);
3026 // copy strides, so they can safely be modified
3027 if (c->sliceDir == 1) {
3028 // slices go from top to bottom
3029 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2], srcStride[3]};
3030 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2], dstStride[3]};
3032 reset_ptr(src2, c->srcFormat);
3033 reset_ptr(dst2, c->dstFormat);
3035 /* reset slice direction at end of frame */
3036 if (srcSliceY + srcSliceH == c->srcH)
3039 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2, dstStride2);
3041 // slices go from bottom to top => we flip the image internally
3042 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2], -srcStride[3]};
3043 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2], -dstStride[3]};
3045 src2[0] += (srcSliceH-1)*srcStride[0];
3046 if (!usePal(c->srcFormat))
3047 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
3048 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
3049 src2[3] += (srcSliceH-1)*srcStride[3];
3050 dst2[0] += ( c->dstH -1)*dstStride[0];
3051 dst2[1] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1];
3052 dst2[2] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2];
3053 dst2[3] += ( c->dstH -1)*dstStride[3];
3055 reset_ptr(src2, c->srcFormat);
3056 reset_ptr(dst2, c->dstFormat);
3058 /* reset slice direction at end of frame */
3062 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
3066 #if LIBSWSCALE_VERSION_MAJOR < 1
3067 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
3068 int srcSliceH, uint8_t* dst[], int dstStride[])
3070 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
3074 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
3075 float lumaSharpen, float chromaSharpen,
3076 float chromaHShift, float chromaVShift,
3079 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
3083 if (lumaGBlur!=0.0) {
3084 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
3085 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
3087 filter->lumH= sws_getIdentityVec();
3088 filter->lumV= sws_getIdentityVec();
3091 if (chromaGBlur!=0.0) {
3092 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
3093 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
3095 filter->chrH= sws_getIdentityVec();
3096 filter->chrV= sws_getIdentityVec();
3099 if (chromaSharpen!=0.0) {
3100 SwsVector *id= sws_getIdentityVec();
3101 sws_scaleVec(filter->chrH, -chromaSharpen);
3102 sws_scaleVec(filter->chrV, -chromaSharpen);
3103 sws_addVec(filter->chrH, id);
3104 sws_addVec(filter->chrV, id);
3108 if (lumaSharpen!=0.0) {
3109 SwsVector *id= sws_getIdentityVec();
3110 sws_scaleVec(filter->lumH, -lumaSharpen);
3111 sws_scaleVec(filter->lumV, -lumaSharpen);
3112 sws_addVec(filter->lumH, id);
3113 sws_addVec(filter->lumV, id);
3117 if (chromaHShift != 0.0)
3118 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
3120 if (chromaVShift != 0.0)
3121 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
3123 sws_normalizeVec(filter->chrH, 1.0);
3124 sws_normalizeVec(filter->chrV, 1.0);
3125 sws_normalizeVec(filter->lumH, 1.0);
3126 sws_normalizeVec(filter->lumV, 1.0);
3128 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
3129 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
3134 SwsVector *sws_allocVec(int length)
3136 SwsVector *vec = av_malloc(sizeof(SwsVector));
3139 vec->length = length;
3140 vec->coeff = av_malloc(sizeof(double) * length);
3146 SwsVector *sws_getGaussianVec(double variance, double quality)
3148 const int length= (int)(variance*quality + 0.5) | 1;
3150 double middle= (length-1)*0.5;
3151 SwsVector *vec= sws_allocVec(length);
3156 for (i=0; i<length; i++) {
3157 double dist= i-middle;
3158 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
3161 sws_normalizeVec(vec, 1.0);
3166 SwsVector *sws_getConstVec(double c, int length)
3169 SwsVector *vec= sws_allocVec(length);
3174 for (i=0; i<length; i++)
3181 SwsVector *sws_getIdentityVec(void)
3183 return sws_getConstVec(1.0, 1);
3186 double sws_dcVec(SwsVector *a)
3191 for (i=0; i<a->length; i++)
3197 void sws_scaleVec(SwsVector *a, double scalar)
3201 for (i=0; i<a->length; i++)
3202 a->coeff[i]*= scalar;
3205 void sws_normalizeVec(SwsVector *a, double height)
3207 sws_scaleVec(a, height/sws_dcVec(a));
3210 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
3212 int length= a->length + b->length - 1;
3214 SwsVector *vec= sws_getConstVec(0.0, length);
3219 for (i=0; i<a->length; i++) {
3220 for (j=0; j<b->length; j++) {
3221 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
3228 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
3230 int length= FFMAX(a->length, b->length);
3232 SwsVector *vec= sws_getConstVec(0.0, length);
3237 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3238 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
3243 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
3245 int length= FFMAX(a->length, b->length);
3247 SwsVector *vec= sws_getConstVec(0.0, length);
3252 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3253 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
3258 /* shift left / or right if "shift" is negative */
3259 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
3261 int length= a->length + FFABS(shift)*2;
3263 SwsVector *vec= sws_getConstVec(0.0, length);
3268 for (i=0; i<a->length; i++) {
3269 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
3275 void sws_shiftVec(SwsVector *a, int shift)
3277 SwsVector *shifted= sws_getShiftedVec(a, shift);
3279 a->coeff= shifted->coeff;
3280 a->length= shifted->length;
3284 void sws_addVec(SwsVector *a, SwsVector *b)
3286 SwsVector *sum= sws_sumVec(a, b);
3288 a->coeff= sum->coeff;
3289 a->length= sum->length;
3293 void sws_subVec(SwsVector *a, SwsVector *b)
3295 SwsVector *diff= sws_diffVec(a, b);
3297 a->coeff= diff->coeff;
3298 a->length= diff->length;
3302 void sws_convVec(SwsVector *a, SwsVector *b)
3304 SwsVector *conv= sws_getConvVec(a, b);
3306 a->coeff= conv->coeff;
3307 a->length= conv->length;
3311 SwsVector *sws_cloneVec(SwsVector *a)
3314 SwsVector *vec= sws_allocVec(a->length);
3319 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
3324 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
3331 for (i=0; i<a->length; i++)
3332 if (a->coeff[i]>max) max= a->coeff[i];
3334 for (i=0; i<a->length; i++)
3335 if (a->coeff[i]<min) min= a->coeff[i];
3339 for (i=0; i<a->length; i++) {
3340 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
3341 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
3342 for (;x>0; x--) av_log(log_ctx, log_level, " ");
3343 av_log(log_ctx, log_level, "|\n");
3347 #if LIBSWSCALE_VERSION_MAJOR < 1
3348 void sws_printVec(SwsVector *a)
3350 sws_printVec2(a, NULL, AV_LOG_DEBUG);
3354 void sws_freeVec(SwsVector *a)
3357 av_freep(&a->coeff);
3362 void sws_freeFilter(SwsFilter *filter)
3364 if (!filter) return;
3366 if (filter->lumH) sws_freeVec(filter->lumH);
3367 if (filter->lumV) sws_freeVec(filter->lumV);
3368 if (filter->chrH) sws_freeVec(filter->chrH);
3369 if (filter->chrV) sws_freeVec(filter->chrV);
3374 void sws_freeContext(SwsContext *c)
3380 for (i=0; i<c->vLumBufSize; i++)
3381 av_freep(&c->lumPixBuf[i]);
3382 av_freep(&c->lumPixBuf);
3386 for (i=0; i<c->vChrBufSize; i++)
3387 av_freep(&c->chrPixBuf[i]);
3388 av_freep(&c->chrPixBuf);
3391 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
3392 for (i=0; i<c->vLumBufSize; i++)
3393 av_freep(&c->alpPixBuf[i]);
3394 av_freep(&c->alpPixBuf);
3397 av_freep(&c->vLumFilter);
3398 av_freep(&c->vChrFilter);
3399 av_freep(&c->hLumFilter);
3400 av_freep(&c->hChrFilter);
3401 #ifdef COMPILE_ALTIVEC
3402 av_freep(&c->vYCoeffsBank);
3403 av_freep(&c->vCCoeffsBank);
3406 av_freep(&c->vLumFilterPos);
3407 av_freep(&c->vChrFilterPos);
3408 av_freep(&c->hLumFilterPos);
3409 av_freep(&c->hChrFilterPos);
3411 #if ARCH_X86 && CONFIG_GPL
3412 #ifdef MAP_ANONYMOUS
3413 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
3414 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
3415 #elif HAVE_VIRTUALALLOC
3416 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, MEM_RELEASE);
3417 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, MEM_RELEASE);
3419 av_free(c->lumMmx2FilterCode);
3420 av_free(c->chrMmx2FilterCode);
3422 c->lumMmx2FilterCode=NULL;
3423 c->chrMmx2FilterCode=NULL;
3424 #endif /* ARCH_X86 && CONFIG_GPL */
3426 av_freep(&c->lumMmx2Filter);
3427 av_freep(&c->chrMmx2Filter);
3428 av_freep(&c->lumMmx2FilterPos);
3429 av_freep(&c->chrMmx2FilterPos);
3430 av_freep(&c->yuvTable);
3435 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3436 int srcW, int srcH, enum PixelFormat srcFormat,
3437 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
3438 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
3440 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3443 param = default_param;
3446 if (context->srcW != srcW || context->srcH != srcH ||
3447 context->srcFormat != srcFormat ||
3448 context->dstW != dstW || context->dstH != dstH ||
3449 context->dstFormat != dstFormat || context->flags != flags ||
3450 context->param[0] != param[0] || context->param[1] != param[1])
3452 sws_freeContext(context);
3457 return sws_getContext(srcW, srcH, srcFormat,
3458 dstW, dstH, dstFormat, flags,
3459 srcFilter, dstFilter, param);