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 \
161 int sws_isSupportedInput(enum PixelFormat pix_fmt)
163 return isSupportedIn(pix_fmt);
166 #define isSupportedOut(x) ( \
167 (x)==PIX_FMT_YUV420P \
168 || (x)==PIX_FMT_YUVA420P \
169 || (x)==PIX_FMT_YUYV422 \
170 || (x)==PIX_FMT_UYVY422 \
171 || (x)==PIX_FMT_YUV444P \
172 || (x)==PIX_FMT_YUV422P \
173 || (x)==PIX_FMT_YUV411P \
176 || (x)==PIX_FMT_NV12 \
177 || (x)==PIX_FMT_NV21 \
178 || (x)==PIX_FMT_GRAY16BE \
179 || (x)==PIX_FMT_GRAY16LE \
180 || (x)==PIX_FMT_GRAY8 \
181 || (x)==PIX_FMT_YUV410P \
182 || (x)==PIX_FMT_YUV440P \
183 || (x)==PIX_FMT_YUV420P16LE \
184 || (x)==PIX_FMT_YUV422P16LE \
185 || (x)==PIX_FMT_YUV444P16LE \
186 || (x)==PIX_FMT_YUV420P16BE \
187 || (x)==PIX_FMT_YUV422P16BE \
188 || (x)==PIX_FMT_YUV444P16BE \
191 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
193 return isSupportedOut(pix_fmt);
196 #define isPacked(x) ( \
198 || (x)==PIX_FMT_YUYV422 \
199 || (x)==PIX_FMT_UYVY422 \
203 #define usePal(x) (av_pix_fmt_descriptors[x].flags & PIX_FMT_PAL)
205 #define RGB2YUV_SHIFT 15
206 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
207 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
208 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
209 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
210 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
211 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
212 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
213 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
214 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
216 extern const int32_t ff_yuv2rgb_coeffs[8][4];
218 static const double rgb2yuv_table[8][9]={
219 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
220 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
221 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
222 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
223 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
224 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
225 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
226 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
231 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
234 more intelligent misalignment avoidance for the horizontal scaler
235 write special vertical cubic upscale version
236 optimize C code (YV12 / minmax)
237 add support for packed pixel YUV input & output
238 add support for Y8 output
239 optimize BGR24 & BGR32
240 add BGR4 output support
241 write special BGR->BGR scaler
244 #if ARCH_X86 && CONFIG_GPL
245 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
246 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
247 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
248 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
249 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
250 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
251 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
252 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
254 const DECLARE_ALIGNED(8, uint64_t, ff_dither4)[2] = {
255 0x0103010301030103LL,
256 0x0200020002000200LL,};
258 const DECLARE_ALIGNED(8, uint64_t, ff_dither8)[2] = {
259 0x0602060206020602LL,
260 0x0004000400040004LL,};
262 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
263 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
264 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
265 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
266 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
267 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
269 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
270 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
271 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
274 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
275 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
276 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
278 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
279 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
280 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
281 #endif /* FAST_BGR2YV12 */
282 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
283 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
284 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
286 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
287 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
288 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
289 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
290 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
292 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV)[2][4] = {
293 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
294 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
297 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
299 #endif /* ARCH_X86 && CONFIG_GPL */
301 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_4)[2][8]={
302 { 1, 3, 1, 3, 1, 3, 1, 3, },
303 { 2, 0, 2, 0, 2, 0, 2, 0, },
306 DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8)[2][8]={
307 { 6, 2, 6, 2, 6, 2, 6, 2, },
308 { 0, 4, 0, 4, 0, 4, 0, 4, },
311 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32)[8][8]={
312 { 17, 9, 23, 15, 16, 8, 22, 14, },
313 { 5, 29, 3, 27, 4, 28, 2, 26, },
314 { 21, 13, 19, 11, 20, 12, 18, 10, },
315 { 0, 24, 6, 30, 1, 25, 7, 31, },
316 { 16, 8, 22, 14, 17, 9, 23, 15, },
317 { 4, 28, 2, 26, 5, 29, 3, 27, },
318 { 20, 12, 18, 10, 21, 13, 19, 11, },
319 { 1, 25, 7, 31, 0, 24, 6, 30, },
322 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73)[8][8]={
323 { 0, 55, 14, 68, 3, 58, 17, 72, },
324 { 37, 18, 50, 32, 40, 22, 54, 35, },
325 { 9, 64, 5, 59, 13, 67, 8, 63, },
326 { 46, 27, 41, 23, 49, 31, 44, 26, },
327 { 2, 57, 16, 71, 1, 56, 15, 70, },
328 { 39, 21, 52, 34, 38, 19, 51, 33, },
329 { 11, 66, 7, 62, 10, 65, 6, 60, },
330 { 48, 30, 43, 25, 47, 29, 42, 24, },
334 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
335 {117, 62, 158, 103, 113, 58, 155, 100, },
336 { 34, 199, 21, 186, 31, 196, 17, 182, },
337 {144, 89, 131, 76, 141, 86, 127, 72, },
338 { 0, 165, 41, 206, 10, 175, 52, 217, },
339 {110, 55, 151, 96, 120, 65, 162, 107, },
340 { 28, 193, 14, 179, 38, 203, 24, 189, },
341 {138, 83, 124, 69, 148, 93, 134, 79, },
342 { 7, 172, 48, 213, 3, 168, 45, 210, },
345 // tries to correct a gamma of 1.5
346 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
347 { 0, 143, 18, 200, 2, 156, 25, 215, },
348 { 78, 28, 125, 64, 89, 36, 138, 74, },
349 { 10, 180, 3, 161, 16, 195, 8, 175, },
350 {109, 51, 93, 38, 121, 60, 105, 47, },
351 { 1, 152, 23, 210, 0, 147, 20, 205, },
352 { 85, 33, 134, 71, 81, 30, 130, 67, },
353 { 14, 190, 6, 171, 12, 185, 5, 166, },
354 {117, 57, 101, 44, 113, 54, 97, 41, },
357 // tries to correct a gamma of 2.0
358 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
359 { 0, 124, 8, 193, 0, 140, 12, 213, },
360 { 55, 14, 104, 42, 66, 19, 119, 52, },
361 { 3, 168, 1, 145, 6, 187, 3, 162, },
362 { 86, 31, 70, 21, 99, 39, 82, 28, },
363 { 0, 134, 11, 206, 0, 129, 9, 200, },
364 { 62, 17, 114, 48, 58, 16, 109, 45, },
365 { 5, 181, 2, 157, 4, 175, 1, 151, },
366 { 95, 36, 78, 26, 90, 34, 74, 24, },
369 // tries to correct a gamma of 2.5
370 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
371 { 0, 107, 3, 187, 0, 125, 6, 212, },
372 { 39, 7, 86, 28, 49, 11, 102, 36, },
373 { 1, 158, 0, 131, 3, 180, 1, 151, },
374 { 68, 19, 52, 12, 81, 25, 64, 17, },
375 { 0, 119, 5, 203, 0, 113, 4, 195, },
376 { 45, 9, 96, 33, 42, 8, 91, 30, },
377 { 2, 172, 1, 144, 2, 165, 0, 137, },
378 { 77, 23, 60, 15, 72, 21, 56, 14, },
382 const char *sws_format_name(enum PixelFormat format)
384 if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
385 return av_pix_fmt_descriptors[format].name;
387 return "Unknown format";
390 static av_always_inline void yuv2yuvX16inC_template(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
391 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
392 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest,
393 int dstW, int chrDstW, int big_endian)
395 //FIXME Optimize (just quickly written not optimized..)
398 for (i = 0; i < dstW; i++) {
402 for (j = 0; j < lumFilterSize; j++)
403 val += lumSrc[j][i] * lumFilter[j];
406 AV_WB16(&dest[i], av_clip_uint16(val >> 11));
408 AV_WL16(&dest[i], av_clip_uint16(val >> 11));
413 for (i = 0; i < chrDstW; i++) {
418 for (j = 0; j < chrFilterSize; j++) {
419 u += chrSrc[j][i ] * chrFilter[j];
420 v += chrSrc[j][i + VOFW] * chrFilter[j];
424 AV_WB16(&uDest[i], av_clip_uint16(u >> 11));
425 AV_WB16(&vDest[i], av_clip_uint16(v >> 11));
427 AV_WL16(&uDest[i], av_clip_uint16(u >> 11));
428 AV_WL16(&vDest[i], av_clip_uint16(v >> 11));
433 if (CONFIG_SWSCALE_ALPHA && aDest) {
434 for (i = 0; i < dstW; i++) {
438 for (j = 0; j < lumFilterSize; j++)
439 val += alpSrc[j][i] * lumFilter[j];
442 AV_WB16(&aDest[i], av_clip_uint16(val >> 11));
444 AV_WL16(&aDest[i], av_clip_uint16(val >> 11));
450 static inline void yuv2yuvX16inC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
451 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
452 const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest, int dstW, int chrDstW,
453 enum PixelFormat dstFormat)
455 if (isBE(dstFormat)) {
456 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
457 chrFilter, chrSrc, chrFilterSize,
459 dest, uDest, vDest, aDest,
462 yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
463 chrFilter, chrSrc, chrFilterSize,
465 dest, uDest, vDest, aDest,
470 static inline void yuv2yuvXinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
471 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
472 const int16_t **alpSrc, uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, int dstW, int chrDstW)
474 //FIXME Optimize (just quickly written not optimized..)
476 for (i=0; i<dstW; i++) {
479 for (j=0; j<lumFilterSize; j++)
480 val += lumSrc[j][i] * lumFilter[j];
482 dest[i]= av_clip_uint8(val>>19);
486 for (i=0; i<chrDstW; i++) {
490 for (j=0; j<chrFilterSize; j++) {
491 u += chrSrc[j][i] * chrFilter[j];
492 v += chrSrc[j][i + VOFW] * chrFilter[j];
495 uDest[i]= av_clip_uint8(u>>19);
496 vDest[i]= av_clip_uint8(v>>19);
499 if (CONFIG_SWSCALE_ALPHA && aDest)
500 for (i=0; i<dstW; i++) {
503 for (j=0; j<lumFilterSize; j++)
504 val += alpSrc[j][i] * lumFilter[j];
506 aDest[i]= av_clip_uint8(val>>19);
511 static inline void yuv2nv12XinC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
512 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
513 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
515 //FIXME Optimize (just quickly written not optimized..)
517 for (i=0; i<dstW; i++) {
520 for (j=0; j<lumFilterSize; j++)
521 val += lumSrc[j][i] * lumFilter[j];
523 dest[i]= av_clip_uint8(val>>19);
529 if (dstFormat == PIX_FMT_NV12)
530 for (i=0; i<chrDstW; i++) {
534 for (j=0; j<chrFilterSize; j++) {
535 u += chrSrc[j][i] * chrFilter[j];
536 v += chrSrc[j][i + VOFW] * chrFilter[j];
539 uDest[2*i]= av_clip_uint8(u>>19);
540 uDest[2*i+1]= av_clip_uint8(v>>19);
543 for (i=0; i<chrDstW; i++) {
547 for (j=0; j<chrFilterSize; j++) {
548 u += chrSrc[j][i] * chrFilter[j];
549 v += chrSrc[j][i + VOFW] * chrFilter[j];
552 uDest[2*i]= av_clip_uint8(v>>19);
553 uDest[2*i+1]= av_clip_uint8(u>>19);
557 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha) \
558 for (i=0; i<(dstW>>1); i++) {\
564 int av_unused A1, A2;\
565 type av_unused *r, *b, *g;\
568 for (j=0; j<lumFilterSize; j++) {\
569 Y1 += lumSrc[j][i2] * lumFilter[j];\
570 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
572 for (j=0; j<chrFilterSize; j++) {\
573 U += chrSrc[j][i] * chrFilter[j];\
574 V += chrSrc[j][i+VOFW] * chrFilter[j];\
583 for (j=0; j<lumFilterSize; j++) {\
584 A1 += alpSrc[j][i2 ] * lumFilter[j];\
585 A2 += alpSrc[j][i2+1] * lumFilter[j];\
591 #define YSCALE_YUV_2_PACKEDX_C(type,alpha) \
592 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type,alpha)\
593 if ((Y1|Y2|U|V)&256) {\
594 if (Y1>255) Y1=255; \
595 else if (Y1<0)Y1=0; \
596 if (Y2>255) Y2=255; \
597 else if (Y2<0)Y2=0; \
603 if (alpha && ((A1|A2)&256)) {\
604 A1=av_clip_uint8(A1);\
605 A2=av_clip_uint8(A2);\
608 #define YSCALE_YUV_2_PACKEDX_FULL_C(rnd,alpha) \
609 for (i=0; i<dstW; i++) {\
617 for (j=0; j<lumFilterSize; j++) {\
618 Y += lumSrc[j][i ] * lumFilter[j];\
620 for (j=0; j<chrFilterSize; j++) {\
621 U += chrSrc[j][i ] * chrFilter[j];\
622 V += chrSrc[j][i+VOFW] * chrFilter[j];\
629 for (j=0; j<lumFilterSize; j++)\
630 A += alpSrc[j][i ] * lumFilter[j];\
633 A = av_clip_uint8(A);\
636 #define YSCALE_YUV_2_RGBX_FULL_C(rnd,alpha) \
637 YSCALE_YUV_2_PACKEDX_FULL_C(rnd>>3,alpha)\
638 Y-= c->yuv2rgb_y_offset;\
639 Y*= c->yuv2rgb_y_coeff;\
641 R= Y + V*c->yuv2rgb_v2r_coeff;\
642 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
643 B= Y + U*c->yuv2rgb_u2b_coeff;\
644 if ((R|G|B)&(0xC0000000)) {\
645 if (R>=(256<<22)) R=(256<<22)-1; \
647 if (G>=(256<<22)) G=(256<<22)-1; \
649 if (B>=(256<<22)) B=(256<<22)-1; \
653 #define YSCALE_YUV_2_GRAY16_C \
654 for (i=0; i<(dstW>>1); i++) {\
663 for (j=0; j<lumFilterSize; j++) {\
664 Y1 += lumSrc[j][i2] * lumFilter[j];\
665 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
669 if ((Y1|Y2|U|V)&65536) {\
670 if (Y1>65535) Y1=65535; \
671 else if (Y1<0)Y1=0; \
672 if (Y2>65535) Y2=65535; \
673 else if (Y2<0)Y2=0; \
676 #define YSCALE_YUV_2_RGBX_C(type,alpha) \
677 YSCALE_YUV_2_PACKEDX_C(type,alpha) /* FIXME fix tables so that clipping is not needed and then use _NOCLIP*/\
678 r = (type *)c->table_rV[V]; \
679 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
680 b = (type *)c->table_bU[U];
682 #define YSCALE_YUV_2_PACKED2_C(type,alpha) \
683 for (i=0; i<(dstW>>1); i++) { \
685 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
686 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
687 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
688 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
689 type av_unused *r, *b, *g; \
690 int av_unused A1, A2; \
692 A1= (abuf0[i2 ]*yalpha1+abuf1[i2 ]*yalpha)>>19; \
693 A2= (abuf0[i2+1]*yalpha1+abuf1[i2+1]*yalpha)>>19; \
696 #define YSCALE_YUV_2_GRAY16_2_C \
697 for (i=0; i<(dstW>>1); i++) { \
699 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
700 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11;
702 #define YSCALE_YUV_2_RGB2_C(type,alpha) \
703 YSCALE_YUV_2_PACKED2_C(type,alpha)\
704 r = (type *)c->table_rV[V];\
705 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
706 b = (type *)c->table_bU[U];
708 #define YSCALE_YUV_2_PACKED1_C(type,alpha) \
709 for (i=0; i<(dstW>>1); i++) {\
711 int Y1= buf0[i2 ]>>7;\
712 int Y2= buf0[i2+1]>>7;\
713 int U= (uvbuf1[i ])>>7;\
714 int V= (uvbuf1[i+VOFW])>>7;\
715 type av_unused *r, *b, *g;\
716 int av_unused A1, A2;\
722 #define YSCALE_YUV_2_GRAY16_1_C \
723 for (i=0; i<(dstW>>1); i++) {\
725 int Y1= buf0[i2 ]<<1;\
726 int Y2= buf0[i2+1]<<1;
728 #define YSCALE_YUV_2_RGB1_C(type,alpha) \
729 YSCALE_YUV_2_PACKED1_C(type,alpha)\
730 r = (type *)c->table_rV[V];\
731 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
732 b = (type *)c->table_bU[U];
734 #define YSCALE_YUV_2_PACKED1B_C(type,alpha) \
735 for (i=0; i<(dstW>>1); i++) {\
737 int Y1= buf0[i2 ]>>7;\
738 int Y2= buf0[i2+1]>>7;\
739 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
740 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
741 type av_unused *r, *b, *g;\
742 int av_unused A1, A2;\
748 #define YSCALE_YUV_2_RGB1B_C(type,alpha) \
749 YSCALE_YUV_2_PACKED1B_C(type,alpha)\
750 r = (type *)c->table_rV[V];\
751 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
752 b = (type *)c->table_bU[U];
754 #define YSCALE_YUV_2_MONO2_C \
755 const uint8_t * const d128=dither_8x8_220[y&7];\
756 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
757 for (i=0; i<dstW-7; i+=8) {\
759 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
760 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
761 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
762 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
763 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
764 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
765 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
766 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
767 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
771 #define YSCALE_YUV_2_MONOX_C \
772 const uint8_t * const d128=dither_8x8_220[y&7];\
773 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
775 for (i=0; i<dstW-1; i+=2) {\
780 for (j=0; j<lumFilterSize; j++) {\
781 Y1 += lumSrc[j][i] * lumFilter[j];\
782 Y2 += lumSrc[j][i+1] * lumFilter[j];\
792 acc+= acc + g[Y1+d128[(i+0)&7]];\
793 acc+= acc + g[Y2+d128[(i+1)&7]];\
795 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
800 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
801 switch(c->dstFormat) {\
802 case PIX_FMT_RGB48BE:\
803 case PIX_FMT_RGB48LE:\
805 ((uint8_t*)dest)[ 0]= r[Y1];\
806 ((uint8_t*)dest)[ 1]= r[Y1];\
807 ((uint8_t*)dest)[ 2]= g[Y1];\
808 ((uint8_t*)dest)[ 3]= g[Y1];\
809 ((uint8_t*)dest)[ 4]= b[Y1];\
810 ((uint8_t*)dest)[ 5]= b[Y1];\
811 ((uint8_t*)dest)[ 6]= r[Y2];\
812 ((uint8_t*)dest)[ 7]= r[Y2];\
813 ((uint8_t*)dest)[ 8]= g[Y2];\
814 ((uint8_t*)dest)[ 9]= g[Y2];\
815 ((uint8_t*)dest)[10]= b[Y2];\
816 ((uint8_t*)dest)[11]= b[Y2];\
823 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
824 func(uint32_t,needAlpha)\
825 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? (A1<<24) : 0);\
826 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? (A2<<24) : 0);\
829 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
831 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (A1<<24);\
832 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (A2<<24);\
836 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
837 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
845 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;\
846 func(uint32_t,needAlpha)\
847 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + (needAlpha ? A1 : 0);\
848 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + (needAlpha ? A2 : 0);\
851 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {\
853 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1] + A1;\
854 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2] + A2;\
858 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
859 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
866 ((uint8_t*)dest)[0]= r[Y1];\
867 ((uint8_t*)dest)[1]= g[Y1];\
868 ((uint8_t*)dest)[2]= b[Y1];\
869 ((uint8_t*)dest)[3]= r[Y2];\
870 ((uint8_t*)dest)[4]= g[Y2];\
871 ((uint8_t*)dest)[5]= b[Y2];\
877 ((uint8_t*)dest)[0]= b[Y1];\
878 ((uint8_t*)dest)[1]= g[Y1];\
879 ((uint8_t*)dest)[2]= r[Y1];\
880 ((uint8_t*)dest)[3]= b[Y2];\
881 ((uint8_t*)dest)[4]= g[Y2];\
882 ((uint8_t*)dest)[5]= r[Y2];\
886 case PIX_FMT_RGB565:\
887 case PIX_FMT_BGR565:\
889 const int dr1= dither_2x2_8[y&1 ][0];\
890 const int dg1= dither_2x2_4[y&1 ][0];\
891 const int db1= dither_2x2_8[(y&1)^1][0];\
892 const int dr2= dither_2x2_8[y&1 ][1];\
893 const int dg2= dither_2x2_4[y&1 ][1];\
894 const int db2= dither_2x2_8[(y&1)^1][1];\
896 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
897 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
901 case PIX_FMT_RGB555:\
902 case PIX_FMT_BGR555:\
904 const int dr1= dither_2x2_8[y&1 ][0];\
905 const int dg1= dither_2x2_8[y&1 ][1];\
906 const int db1= dither_2x2_8[(y&1)^1][0];\
907 const int dr2= dither_2x2_8[y&1 ][1];\
908 const int dg2= dither_2x2_8[y&1 ][0];\
909 const int db2= dither_2x2_8[(y&1)^1][1];\
911 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
912 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
919 const uint8_t * const d64= dither_8x8_73[y&7];\
920 const uint8_t * const d32= dither_8x8_32[y&7];\
922 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
923 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
930 const uint8_t * const d64= dither_8x8_73 [y&7];\
931 const uint8_t * const d128=dither_8x8_220[y&7];\
933 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
934 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
938 case PIX_FMT_RGB4_BYTE:\
939 case PIX_FMT_BGR4_BYTE:\
941 const uint8_t * const d64= dither_8x8_73 [y&7];\
942 const uint8_t * const d128=dither_8x8_220[y&7];\
944 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
945 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
949 case PIX_FMT_MONOBLACK:\
950 case PIX_FMT_MONOWHITE:\
955 case PIX_FMT_YUYV422:\
957 ((uint8_t*)dest)[2*i2+0]= Y1;\
958 ((uint8_t*)dest)[2*i2+1]= U;\
959 ((uint8_t*)dest)[2*i2+2]= Y2;\
960 ((uint8_t*)dest)[2*i2+3]= V;\
963 case PIX_FMT_UYVY422:\
965 ((uint8_t*)dest)[2*i2+0]= U;\
966 ((uint8_t*)dest)[2*i2+1]= Y1;\
967 ((uint8_t*)dest)[2*i2+2]= V;\
968 ((uint8_t*)dest)[2*i2+3]= Y2;\
971 case PIX_FMT_GRAY16BE:\
973 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
974 ((uint8_t*)dest)[2*i2+1]= Y1;\
975 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
976 ((uint8_t*)dest)[2*i2+3]= Y2;\
979 case PIX_FMT_GRAY16LE:\
981 ((uint8_t*)dest)[2*i2+0]= Y1;\
982 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
983 ((uint8_t*)dest)[2*i2+2]= Y2;\
984 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
989 static inline void yuv2packedXinC(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
990 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
991 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
994 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)
997 static inline void yuv2rgbXinC_full(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
998 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
999 const int16_t **alpSrc, uint8_t *dest, int dstW, int y)
1002 int step= fmt_depth(c->dstFormat)/8;
1005 switch(c->dstFormat) {
1013 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1014 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1015 dest[aidx]= needAlpha ? A : 255;
1022 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1023 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1031 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1048 int needAlpha = CONFIG_SWSCALE_ALPHA && c->alpPixBuf;
1049 YSCALE_YUV_2_RGBX_FULL_C(1<<21, needAlpha)
1050 dest[aidx]= needAlpha ? A : 255;
1057 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1058 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 1)
1066 YSCALE_YUV_2_RGBX_FULL_C(1<<21, 0)
1081 static void fillPlane(uint8_t* plane, int stride, int width, int height, int y, uint8_t val)
1084 uint8_t *ptr = plane + stride*y;
1085 for (i=0; i<height; i++) {
1086 memset(ptr, val, width);
1091 static inline void rgb48ToY(uint8_t *dst, const uint8_t *src, int width,
1095 for (i = 0; i < width; i++) {
1100 dst[i] = (RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1104 static inline void rgb48ToUV(uint8_t *dstU, uint8_t *dstV,
1105 const uint8_t *src1, const uint8_t *src2,
1106 int width, uint32_t *unused)
1110 for (i = 0; i < width; i++) {
1111 int r = src1[6*i + 0];
1112 int g = src1[6*i + 2];
1113 int b = src1[6*i + 4];
1115 dstU[i] = (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1116 dstV[i] = (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1))) >> RGB2YUV_SHIFT;
1120 static inline void rgb48ToUV_half(uint8_t *dstU, uint8_t *dstV,
1121 const uint8_t *src1, const uint8_t *src2,
1122 int width, uint32_t *unused)
1126 for (i = 0; i < width; i++) {
1127 int r= src1[12*i + 0] + src1[12*i + 6];
1128 int g= src1[12*i + 2] + src1[12*i + 8];
1129 int b= src1[12*i + 4] + src1[12*i + 10];
1131 dstU[i]= (RU*r + GU*g + BU*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1132 dstV[i]= (RV*r + GV*g + BV*b + (257<<RGB2YUV_SHIFT)) >> (RGB2YUV_SHIFT+1);
1136 #define BGR2Y(type, name, shr, shg, shb, maskr, maskg, maskb, RY, GY, BY, S)\
1137 static inline void name(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)\
1140 for (i=0; i<width; i++) {\
1141 int b= (((const type*)src)[i]>>shb)&maskb;\
1142 int g= (((const type*)src)[i]>>shg)&maskg;\
1143 int r= (((const type*)src)[i]>>shr)&maskr;\
1145 dst[i]= (((RY)*r + (GY)*g + (BY)*b + (33<<((S)-1)))>>(S));\
1149 BGR2Y(uint32_t, bgr32ToY,16, 0, 0, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1150 BGR2Y(uint32_t, rgb32ToY, 0, 0,16, 0x00FF, 0xFF00, 0x00FF, RY<< 8, GY , BY<< 8, RGB2YUV_SHIFT+8)
1151 BGR2Y(uint16_t, bgr16ToY, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RY<<11, GY<<5, BY , RGB2YUV_SHIFT+8)
1152 BGR2Y(uint16_t, bgr15ToY, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RY<<10, GY<<5, BY , RGB2YUV_SHIFT+7)
1153 BGR2Y(uint16_t, rgb16ToY, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RY , GY<<5, BY<<11, RGB2YUV_SHIFT+8)
1154 BGR2Y(uint16_t, rgb15ToY, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RY , GY<<5, BY<<10, RGB2YUV_SHIFT+7)
1156 static inline void abgrToA(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1159 for (i=0; i<width; i++) {
1164 #define BGR2UV(type, name, shr, shg, shb, maska, maskr, maskg, maskb, RU, GU, BU, RV, GV, BV, S)\
1165 static inline void name(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1168 for (i=0; i<width; i++) {\
1169 int b= (((const type*)src)[i]&maskb)>>shb;\
1170 int g= (((const type*)src)[i]&maskg)>>shg;\
1171 int r= (((const type*)src)[i]&maskr)>>shr;\
1173 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<((S)-1)))>>(S);\
1174 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<((S)-1)))>>(S);\
1177 static inline void name ## _half(uint8_t *dstU, uint8_t *dstV, const uint8_t *src, const uint8_t *dummy, long width, uint32_t *unused)\
1180 for (i=0; i<width; i++) {\
1181 int pix0= ((const type*)src)[2*i+0];\
1182 int pix1= ((const type*)src)[2*i+1];\
1183 int g= (pix0&~(maskr|maskb))+(pix1&~(maskr|maskb));\
1184 int b= ((pix0+pix1-g)&(maskb|(2*maskb)))>>shb;\
1185 int r= ((pix0+pix1-g)&(maskr|(2*maskr)))>>shr;\
1186 g&= maskg|(2*maskg);\
1190 dstU[i]= ((RU)*r + (GU)*g + (BU)*b + (257<<(S)))>>((S)+1);\
1191 dstV[i]= ((RV)*r + (GV)*g + (BV)*b + (257<<(S)))>>((S)+1);\
1195 BGR2UV(uint32_t, bgr32ToUV,16, 0, 0, 0xFF000000, 0xFF0000, 0xFF00, 0x00FF, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1196 BGR2UV(uint32_t, rgb32ToUV, 0, 0,16, 0xFF000000, 0x00FF, 0xFF00, 0xFF0000, RU<< 8, GU , BU<< 8, RV<< 8, GV , BV<< 8, RGB2YUV_SHIFT+8)
1197 BGR2UV(uint16_t, bgr16ToUV, 0, 0, 0, 0, 0x001F, 0x07E0, 0xF800, RU<<11, GU<<5, BU , RV<<11, GV<<5, BV , RGB2YUV_SHIFT+8)
1198 BGR2UV(uint16_t, bgr15ToUV, 0, 0, 0, 0, 0x001F, 0x03E0, 0x7C00, RU<<10, GU<<5, BU , RV<<10, GV<<5, BV , RGB2YUV_SHIFT+7)
1199 BGR2UV(uint16_t, rgb16ToUV, 0, 0, 0, 0, 0xF800, 0x07E0, 0x001F, RU , GU<<5, BU<<11, RV , GV<<5, BV<<11, RGB2YUV_SHIFT+8)
1200 BGR2UV(uint16_t, rgb15ToUV, 0, 0, 0, 0, 0x7C00, 0x03E0, 0x001F, RU , GU<<5, BU<<10, RV , GV<<5, BV<<10, RGB2YUV_SHIFT+7)
1202 static inline void palToY(uint8_t *dst, const uint8_t *src, long width, uint32_t *pal)
1205 for (i=0; i<width; i++) {
1208 dst[i]= pal[d] & 0xFF;
1212 static inline void palToUV(uint8_t *dstU, uint8_t *dstV,
1213 const uint8_t *src1, const uint8_t *src2,
1214 long width, uint32_t *pal)
1217 assert(src1 == src2);
1218 for (i=0; i<width; i++) {
1219 int p= pal[src1[i]];
1226 static inline void monowhite2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1229 for (i=0; i<width/8; i++) {
1232 dst[8*i+j]= ((d>>(7-j))&1)*255;
1236 static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
1239 for (i=0; i<width/8; i++) {
1242 dst[8*i+j]= ((d>>(7-j))&1)*255;
1246 //Note: we have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
1248 #if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
1253 #if HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT
1254 #define COMPILE_ALTIVEC
1260 #if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1264 #if (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1265 #define COMPILE_MMX2
1268 #if ((HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
1269 #define COMPILE_3DNOW
1273 #define COMPILE_TEMPLATE_MMX 0
1274 #define COMPILE_TEMPLATE_MMX2 0
1275 #define COMPILE_TEMPLATE_AMD3DNOW 0
1276 #define COMPILE_TEMPLATE_ALTIVEC 0
1279 #define RENAME(a) a ## _C
1280 #include "swscale_template.c"
1283 #ifdef COMPILE_ALTIVEC
1285 #undef COMPILE_TEMPLATE_ALTIVEC
1286 #define COMPILE_TEMPLATE_ALTIVEC 1
1287 #define RENAME(a) a ## _altivec
1288 #include "swscale_template.c"
1296 #undef COMPILE_TEMPLATE_MMX
1297 #undef COMPILE_TEMPLATE_MMX2
1298 #undef COMPILE_TEMPLATE_AMD3DNOW
1299 #define COMPILE_TEMPLATE_MMX 1
1300 #define COMPILE_TEMPLATE_MMX2 0
1301 #define COMPILE_TEMPLATE_AMD3DNOW 0
1302 #define RENAME(a) a ## _MMX
1303 #include "swscale_template.c"
1309 #undef COMPILE_TEMPLATE_MMX
1310 #undef COMPILE_TEMPLATE_MMX2
1311 #undef COMPILE_TEMPLATE_AMD3DNOW
1312 #define COMPILE_TEMPLATE_MMX 1
1313 #define COMPILE_TEMPLATE_MMX2 1
1314 #define COMPILE_TEMPLATE_AMD3DNOW 0
1315 #define RENAME(a) a ## _MMX2
1316 #include "swscale_template.c"
1320 #ifdef COMPILE_3DNOW
1322 #undef COMPILE_TEMPLATE_MMX
1323 #undef COMPILE_TEMPLATE_MMX2
1324 #undef COMPILE_TEMPLATE_AMD3DNOW
1325 #define COMPILE_TEMPLATE_MMX 1
1326 #define COMPILE_TEMPLATE_MMX2 0
1327 #define COMPILE_TEMPLATE_AMD3DNOW 1
1328 #define RENAME(a) a ## _3DNow
1329 #include "swscale_template.c"
1334 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1336 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1337 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1338 else return getSplineCoeff( 0.0,
1345 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1346 int srcW, int dstW, int filterAlign, int one, int flags,
1347 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1353 int64_t *filter=NULL;
1354 int64_t *filter2=NULL;
1355 const int64_t fone= 1LL<<54;
1358 if (flags & SWS_CPU_CAPS_MMX)
1359 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1362 // NOTE: the +1 is for the MMX scaler which reads over the end
1363 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
1365 if (FFABS(xInc - 0x10000) <10) { // unscaled
1368 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1370 for (i=0; i<dstW; i++) {
1371 filter[i*filterSize]= fone;
1375 } else if (flags&SWS_POINT) { // lame looking point sampling mode
1379 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1381 xDstInSrc= xInc/2 - 0x8000;
1382 for (i=0; i<dstW; i++) {
1383 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1385 (*filterPos)[i]= xx;
1389 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
1393 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1395 xDstInSrc= xInc/2 - 0x8000;
1396 for (i=0; i<dstW; i++) {
1397 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1400 (*filterPos)[i]= xx;
1401 //bilinear upscale / linear interpolate / area averaging
1402 for (j=0; j<filterSize; j++) {
1403 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1404 if (coeff<0) coeff=0;
1405 filter[i*filterSize + j]= coeff;
1414 if (flags&SWS_BICUBIC) sizeFactor= 4;
1415 else if (flags&SWS_X) sizeFactor= 8;
1416 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1417 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1418 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1419 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1420 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1421 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1423 sizeFactor= 0; //GCC warning killer
1427 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1428 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1430 if (filterSize > srcW-2) filterSize=srcW-2;
1432 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
1434 xDstInSrc= xInc - 0x10000;
1435 for (i=0; i<dstW; i++) {
1436 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1438 (*filterPos)[i]= xx;
1439 for (j=0; j<filterSize; j++) {
1440 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1446 floatd= d * (1.0/(1<<30));
1448 if (flags & SWS_BICUBIC) {
1449 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1450 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1451 int64_t dd = ( d*d)>>30;
1452 int64_t ddd= (dd*d)>>30;
1455 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1456 else if (d < 1LL<<31)
1457 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1460 coeff *= fone>>(30+24);
1462 /* else if (flags & SWS_X) {
1463 double p= param ? param*0.01 : 0.3;
1464 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1465 coeff*= pow(2.0, - p*d*d);
1467 else if (flags & SWS_X) {
1468 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1475 if (c<0.0) c= -pow(-c, A);
1477 coeff= (c*0.5 + 0.5)*fone;
1478 } else if (flags & SWS_AREA) {
1479 int64_t d2= d - (1<<29);
1480 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1481 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1483 coeff *= fone>>(30+16);
1484 } else if (flags & SWS_GAUSS) {
1485 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1486 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1487 } else if (flags & SWS_SINC) {
1488 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1489 } else if (flags & SWS_LANCZOS) {
1490 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1491 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1492 if (floatd>p) coeff=0;
1493 } else if (flags & SWS_BILINEAR) {
1495 if (coeff<0) coeff=0;
1496 coeff *= fone >> 30;
1497 } else if (flags & SWS_SPLINE) {
1498 double p=-2.196152422706632;
1499 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1501 coeff= 0.0; //GCC warning killer
1505 filter[i*filterSize + j]= coeff;
1512 /* apply src & dst Filter to filter -> filter2
1515 assert(filterSize>0);
1516 filter2Size= filterSize;
1517 if (srcFilter) filter2Size+= srcFilter->length - 1;
1518 if (dstFilter) filter2Size+= dstFilter->length - 1;
1519 assert(filter2Size>0);
1520 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
1522 for (i=0; i<dstW; i++) {
1526 for (k=0; k<srcFilter->length; k++) {
1527 for (j=0; j<filterSize; j++)
1528 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1531 for (j=0; j<filterSize; j++)
1532 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1536 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1540 /* try to reduce the filter-size (step1 find size and shift left) */
1541 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1543 for (i=dstW-1; i>=0; i--) {
1544 int min= filter2Size;
1548 /* get rid of near zero elements on the left by shifting left */
1549 for (j=0; j<filter2Size; j++) {
1551 cutOff += FFABS(filter2[i*filter2Size]);
1553 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1555 /* preserve monotonicity because the core can't handle the filter otherwise */
1556 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1558 // move filter coefficients left
1559 for (k=1; k<filter2Size; k++)
1560 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1561 filter2[i*filter2Size + k - 1]= 0;
1566 /* count near zeros on the right */
1567 for (j=filter2Size-1; j>0; j--) {
1568 cutOff += FFABS(filter2[i*filter2Size + j]);
1570 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1574 if (min>minFilterSize) minFilterSize= min;
1577 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1578 // we can handle the special case 4,
1579 // so we don't want to go to the full 8
1580 if (minFilterSize < 5)
1583 // We really don't want to waste our time
1584 // doing useless computation, so fall back on
1585 // the scalar C code for very small filters.
1586 // Vectorizing is worth it only if you have a
1587 // decent-sized vector.
1588 if (minFilterSize < 3)
1592 if (flags & SWS_CPU_CAPS_MMX) {
1593 // special case for unscaled vertical filtering
1594 if (minFilterSize == 1 && filterAlign == 2)
1598 assert(minFilterSize > 0);
1599 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1600 assert(filterSize > 0);
1601 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1602 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1604 *outFilterSize= filterSize;
1606 if (flags&SWS_PRINT_INFO)
1607 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1608 /* try to reduce the filter-size (step2 reduce it) */
1609 for (i=0; i<dstW; i++) {
1612 for (j=0; j<filterSize; j++) {
1613 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1614 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1615 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1616 filter[i*filterSize + j]= 0;
1620 //FIXME try to align filterPos if possible
1623 for (i=0; i<dstW; i++) {
1625 if ((*filterPos)[i] < 0) {
1626 // move filter coefficients left to compensate for filterPos
1627 for (j=1; j<filterSize; j++) {
1628 int left= FFMAX(j + (*filterPos)[i], 0);
1629 filter[i*filterSize + left] += filter[i*filterSize + j];
1630 filter[i*filterSize + j]=0;
1635 if ((*filterPos)[i] + filterSize > srcW) {
1636 int shift= (*filterPos)[i] + filterSize - srcW;
1637 // move filter coefficients right to compensate for filterPos
1638 for (j=filterSize-2; j>=0; j--) {
1639 int right= FFMIN(j + shift, filterSize-1);
1640 filter[i*filterSize +right] += filter[i*filterSize +j];
1641 filter[i*filterSize +j]=0;
1643 (*filterPos)[i]= srcW - filterSize;
1647 // Note the +1 is for the MMX scaler which reads over the end
1648 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1649 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
1651 /* normalize & store in outFilter */
1652 for (i=0; i<dstW; i++) {
1657 for (j=0; j<filterSize; j++) {
1658 sum+= filter[i*filterSize + j];
1660 sum= (sum + one/2)/ one;
1661 for (j=0; j<*outFilterSize; j++) {
1662 int64_t v= filter[i*filterSize + j] + error;
1663 int intV= ROUNDED_DIV(v, sum);
1664 (*outFilter)[i*(*outFilterSize) + j]= intV;
1665 error= v - intV*sum;
1669 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1670 for (i=0; i<*outFilterSize; i++) {
1671 int j= dstW*(*outFilterSize);
1672 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1683 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
1686 x86_reg imm8OfPShufW1A;
1687 x86_reg imm8OfPShufW2A;
1688 x86_reg fragmentLengthA;
1690 x86_reg imm8OfPShufW1B;
1691 x86_reg imm8OfPShufW2B;
1692 x86_reg fragmentLengthB;
1697 // create an optimized horizontal scaling routine
1698 /* This scaler is made of runtime-generated MMX2 code using specially
1699 * tuned pshufw instructions. For every four output pixels, if four
1700 * input pixels are enough for the fast bilinear scaling, then a chunk
1701 * of fragmentB is used. If five input pixels are needed, then a chunk
1702 * of fragmentA is used.
1711 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1712 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1713 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1714 "punpcklbw %%mm7, %%mm1 \n\t"
1715 "punpcklbw %%mm7, %%mm0 \n\t"
1716 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1718 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1720 "psubw %%mm1, %%mm0 \n\t"
1721 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1722 "pmullw %%mm3, %%mm0 \n\t"
1723 "psllw $7, %%mm1 \n\t"
1724 "paddw %%mm1, %%mm0 \n\t"
1726 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1728 "add $8, %%"REG_a" \n\t"
1732 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1733 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1734 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1739 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1743 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1744 "=r" (fragmentLengthA)
1751 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1752 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1753 "punpcklbw %%mm7, %%mm0 \n\t"
1754 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1756 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1758 "psubw %%mm1, %%mm0 \n\t"
1759 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1760 "pmullw %%mm3, %%mm0 \n\t"
1761 "psllw $7, %%mm1 \n\t"
1762 "paddw %%mm1, %%mm0 \n\t"
1764 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1766 "add $8, %%"REG_a" \n\t"
1770 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1771 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1772 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1777 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1781 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1782 "=r" (fragmentLengthB)
1785 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1788 for (i=0; i<dstW/numSplits; i++) {
1793 int b=((xpos+xInc)>>16) - xx;
1794 int c=((xpos+xInc*2)>>16) - xx;
1795 int d=((xpos+xInc*3)>>16) - xx;
1797 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
1798 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
1799 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
1800 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
1801 int maxShift= 3-(d+inc);
1805 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1806 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1807 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1808 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1811 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
1813 filterCode[fragmentPos + imm8OfPShufW1]=
1814 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
1815 filterCode[fragmentPos + imm8OfPShufW2]=
1816 a | (b<<2) | (c<<4) | (d<<6);
1818 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
1819 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1821 if (shift && i>=shift) {
1822 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
1823 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
1824 filterPos[i/2]-=shift;
1828 fragmentPos+= fragmentLength;
1831 filterCode[fragmentPos]= RET;
1836 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
1838 return fragmentPos + 1;
1840 #endif /* COMPILE_MMX2 */
1842 static SwsFunc getSwsFunc(SwsContext *c)
1844 #if CONFIG_RUNTIME_CPUDETECT
1845 int flags = c->flags;
1847 #if ARCH_X86 && CONFIG_GPL
1848 // ordered per speed fastest first
1849 if (flags & SWS_CPU_CAPS_MMX2) {
1850 sws_init_swScale_MMX2(c);
1851 return swScale_MMX2;
1852 } else if (flags & SWS_CPU_CAPS_3DNOW) {
1853 sws_init_swScale_3DNow(c);
1854 return swScale_3DNow;
1855 } else if (flags & SWS_CPU_CAPS_MMX) {
1856 sws_init_swScale_MMX(c);
1859 sws_init_swScale_C(c);
1865 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1866 sws_init_swScale_altivec(c);
1867 return swScale_altivec;
1869 sws_init_swScale_C(c);
1873 sws_init_swScale_C(c);
1875 #endif /* ARCH_X86 && CONFIG_GPL */
1876 #else //CONFIG_RUNTIME_CPUDETECT
1877 #if COMPILE_TEMPLATE_MMX2
1878 sws_init_swScale_MMX2(c);
1879 return swScale_MMX2;
1880 #elif COMPILE_TEMPLATE_AMD3DNOW
1881 sws_init_swScale_3DNow(c);
1882 return swScale_3DNow;
1883 #elif COMPILE_TEMPLATE_MMX
1884 sws_init_swScale_MMX(c);
1886 #elif COMPILE_TEMPLATE_ALTIVEC
1887 sws_init_swScale_altivec(c);
1888 return swScale_altivec;
1890 sws_init_swScale_C(c);
1893 #endif //!CONFIG_RUNTIME_CPUDETECT
1896 static int PlanarToNV12Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1897 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1899 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1901 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1902 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1905 const uint8_t *srcPtr= src[0];
1906 uint8_t *dstPtr= dst;
1907 for (i=0; i<srcSliceH; i++) {
1908 memcpy(dstPtr, srcPtr, c->srcW);
1909 srcPtr+= srcStride[0];
1910 dstPtr+= dstStride[0];
1913 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1914 if (c->dstFormat == PIX_FMT_NV12)
1915 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1917 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1922 static int PlanarToYuy2Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1923 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1925 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1927 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1932 static int PlanarToUyvyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1933 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1935 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1937 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1942 static int YUV422PToYuy2Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1943 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1945 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1947 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1952 static int YUV422PToUyvyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1953 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1955 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1957 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1962 static int YUYV2YUV420Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1963 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1965 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1966 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
1967 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
1969 yuyvtoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1972 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
1977 static int YUYV2YUV422Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1978 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1980 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1981 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
1982 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
1984 yuyvtoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1989 static int UYVY2YUV420Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
1990 int srcSliceH, uint8_t* dstParam[], int dstStride[])
1992 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
1993 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY/2;
1994 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY/2;
1996 uyvytoyuv420(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
1999 fillPlane(dstParam[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2004 static int UYVY2YUV422Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2005 int srcSliceH, uint8_t* dstParam[], int dstStride[])
2007 uint8_t *ydst=dstParam[0] + dstStride[0]*srcSliceY;
2008 uint8_t *udst=dstParam[1] + dstStride[1]*srcSliceY;
2009 uint8_t *vdst=dstParam[2] + dstStride[2]*srcSliceY;
2011 uyvytoyuv422(ydst, udst, vdst, src[0], c->srcW, srcSliceH, dstStride[0], dstStride[1], srcStride[0]);
2016 static int pal2rgbWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2017 int srcSliceH, uint8_t* dst[], int dstStride[])
2019 const enum PixelFormat srcFormat= c->srcFormat;
2020 const enum PixelFormat dstFormat= c->dstFormat;
2021 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
2022 const uint8_t *palette)=NULL;
2024 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2025 const uint8_t *srcPtr= src[0];
2027 if (usePal(srcFormat)) {
2028 switch (dstFormat) {
2029 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
2030 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
2031 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
2032 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
2033 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
2034 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
2039 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2040 sws_format_name(srcFormat), sws_format_name(dstFormat));
2042 for (i=0; i<srcSliceH; i++) {
2043 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
2044 srcPtr+= srcStride[0];
2045 dstPtr+= dstStride[0];
2052 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
2053 static int rgb2rgbWrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2054 int srcSliceH, uint8_t* dst[], int dstStride[])
2056 const enum PixelFormat srcFormat= c->srcFormat;
2057 const enum PixelFormat dstFormat= c->dstFormat;
2058 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
2059 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
2060 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
2061 const int dstId= fmt_depth(dstFormat) >> 2;
2062 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
2065 if ( (isBGR(srcFormat) && isBGR(dstFormat))
2066 || (isRGB(srcFormat) && isRGB(dstFormat))) {
2067 switch(srcId | (dstId<<4)) {
2068 case 0x34: conv= rgb16to15; break;
2069 case 0x36: conv= rgb24to15; break;
2070 case 0x38: conv= rgb32to15; break;
2071 case 0x43: conv= rgb15to16; break;
2072 case 0x46: conv= rgb24to16; break;
2073 case 0x48: conv= rgb32to16; break;
2074 case 0x63: conv= rgb15to24; break;
2075 case 0x64: conv= rgb16to24; break;
2076 case 0x68: conv= rgb32to24; break;
2077 case 0x83: conv= rgb15to32; break;
2078 case 0x84: conv= rgb16to32; break;
2079 case 0x86: conv= rgb24to32; break;
2081 } else if ( (isBGR(srcFormat) && isRGB(dstFormat))
2082 || (isRGB(srcFormat) && isBGR(dstFormat))) {
2083 switch(srcId | (dstId<<4)) {
2084 case 0x33: conv= rgb15tobgr15; break;
2085 case 0x34: conv= rgb16tobgr15; break;
2086 case 0x36: conv= rgb24tobgr15; break;
2087 case 0x38: conv= rgb32tobgr15; break;
2088 case 0x43: conv= rgb15tobgr16; break;
2089 case 0x44: conv= rgb16tobgr16; break;
2090 case 0x46: conv= rgb24tobgr16; break;
2091 case 0x48: conv= rgb32tobgr16; break;
2092 case 0x63: conv= rgb15tobgr24; break;
2093 case 0x64: conv= rgb16tobgr24; break;
2094 case 0x66: conv= rgb24tobgr24; break;
2095 case 0x68: conv= rgb32tobgr24; break;
2096 case 0x83: conv= rgb15tobgr32; break;
2097 case 0x84: conv= rgb16tobgr32; break;
2098 case 0x86: conv= rgb24tobgr32; break;
2099 case 0x88: conv= rgb32tobgr32; break;
2104 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
2105 sws_format_name(srcFormat), sws_format_name(dstFormat));
2107 const uint8_t *srcPtr= src[0];
2108 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
2109 srcPtr += ALT32_CORR;
2111 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
2112 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
2115 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2117 for (i=0; i<srcSliceH; i++) {
2118 conv(srcPtr, dstPtr, c->srcW*srcBpp);
2119 srcPtr+= srcStride[0];
2120 dstPtr+= dstStride[0];
2127 static int bgr24toyv12Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2128 int srcSliceH, uint8_t* dst[], int dstStride[])
2132 dst[0]+ srcSliceY *dstStride[0],
2133 dst[1]+(srcSliceY>>1)*dstStride[1],
2134 dst[2]+(srcSliceY>>1)*dstStride[2],
2136 dstStride[0], dstStride[1], srcStride[0]);
2138 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2142 static int yvu9toyv12Wrapper(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2143 int srcSliceH, uint8_t* dst[], int dstStride[])
2148 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
2149 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
2151 const uint8_t *srcPtr= src[0];
2152 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2154 for (i=0; i<srcSliceH; i++) {
2155 memcpy(dstPtr, srcPtr, c->srcW);
2156 srcPtr+= srcStride[0];
2157 dstPtr+= dstStride[0];
2161 if (c->dstFormat==PIX_FMT_YUV420P || c->dstFormat==PIX_FMT_YUVA420P) {
2162 planar2x(src[1], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2163 srcSliceH >> 2, srcStride[1], dstStride[1]);
2164 planar2x(src[2], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2165 srcSliceH >> 2, srcStride[2], dstStride[2]);
2167 planar2x(src[1], dst[2] + dstStride[2]*(srcSliceY >> 1), c->chrSrcW,
2168 srcSliceH >> 2, srcStride[1], dstStride[2]);
2169 planar2x(src[2], dst[1] + dstStride[1]*(srcSliceY >> 1), c->chrSrcW,
2170 srcSliceH >> 2, srcStride[2], dstStride[1]);
2173 fillPlane(dst[3], dstStride[3], c->srcW, srcSliceH, srcSliceY, 255);
2177 /* unscaled copy like stuff (assumes nearly identical formats) */
2178 static int packedCopy(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2179 int srcSliceH, uint8_t* dst[], int dstStride[])
2181 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
2182 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
2185 const uint8_t *srcPtr= src[0];
2186 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
2189 /* universal length finder */
2190 while(length+c->srcW <= FFABS(dstStride[0])
2191 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
2194 for (i=0; i<srcSliceH; i++) {
2195 memcpy(dstPtr, srcPtr, length);
2196 srcPtr+= srcStride[0];
2197 dstPtr+= dstStride[0];
2203 static int planarCopy(SwsContext *c, const uint8_t* src[], int srcStride[], int srcSliceY,
2204 int srcSliceH, uint8_t* dst[], int dstStride[])
2207 for (plane=0; plane<4; plane++) {
2208 int length= (plane==0 || plane==3) ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
2209 int y= (plane==0 || plane==3) ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
2210 int height= (plane==0 || plane==3) ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
2211 const uint8_t *srcPtr= src[plane];
2212 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
2214 if (!dst[plane]) continue;
2215 // ignore palette for GRAY8
2216 if (plane == 1 && !dst[2]) continue;
2217 if (!src[plane] || (plane == 1 && !src[2])) {
2218 if(is16BPS(c->dstFormat))
2220 fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
2222 if(is16BPS(c->srcFormat) && !is16BPS(c->dstFormat)) {
2223 if (!isBE(c->srcFormat)) srcPtr++;
2224 for (i=0; i<height; i++) {
2225 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
2226 srcPtr+= srcStride[plane];
2227 dstPtr+= dstStride[plane];
2229 } else if(!is16BPS(c->srcFormat) && is16BPS(c->dstFormat)) {
2230 for (i=0; i<height; i++) {
2231 for (j=0; j<length; j++) {
2232 dstPtr[ j<<1 ] = srcPtr[j];
2233 dstPtr[(j<<1)+1] = srcPtr[j];
2235 srcPtr+= srcStride[plane];
2236 dstPtr+= dstStride[plane];
2238 } else if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat)
2239 && isBE(c->srcFormat) != isBE(c->dstFormat)) {
2241 for (i=0; i<height; i++) {
2242 for (j=0; j<length; j++)
2243 ((uint16_t*)dstPtr)[j] = bswap_16(((const uint16_t*)srcPtr)[j]);
2244 srcPtr+= srcStride[plane];
2245 dstPtr+= dstStride[plane];
2247 } else if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
2248 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
2250 if(is16BPS(c->srcFormat) && is16BPS(c->dstFormat))
2252 for (i=0; i<height; i++) {
2253 memcpy(dstPtr, srcPtr, length);
2254 srcPtr+= srcStride[plane];
2255 dstPtr+= dstStride[plane];
2263 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
2265 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
2266 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
2269 static uint16_t roundToInt16(int64_t f)
2271 int r= (f + (1<<15))>>16;
2272 if (r<-0x7FFF) return 0x8000;
2273 else if (r> 0x7FFF) return 0x7FFF;
2277 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
2279 int64_t crv = inv_table[0];
2280 int64_t cbu = inv_table[1];
2281 int64_t cgu = -inv_table[2];
2282 int64_t cgv = -inv_table[3];
2286 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2287 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2289 c->brightness= brightness;
2290 c->contrast = contrast;
2291 c->saturation= saturation;
2292 c->srcRange = srcRange;
2293 c->dstRange = dstRange;
2294 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2296 c->uOffset= 0x0400040004000400LL;
2297 c->vOffset= 0x0400040004000400LL;
2303 crv= (crv*224) / 255;
2304 cbu= (cbu*224) / 255;
2305 cgu= (cgu*224) / 255;
2306 cgv= (cgv*224) / 255;
2309 cy = (cy *contrast )>>16;
2310 crv= (crv*contrast * saturation)>>32;
2311 cbu= (cbu*contrast * saturation)>>32;
2312 cgu= (cgu*contrast * saturation)>>32;
2313 cgv= (cgv*contrast * saturation)>>32;
2315 oy -= 256*brightness;
2317 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2318 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2319 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2320 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2321 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2322 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2324 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2325 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2326 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2327 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2328 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2329 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2331 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2334 #ifdef COMPILE_ALTIVEC
2335 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2336 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
2341 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
2343 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2345 *inv_table = c->srcColorspaceTable;
2346 *table = c->dstColorspaceTable;
2347 *srcRange = c->srcRange;
2348 *dstRange = c->dstRange;
2349 *brightness= c->brightness;
2350 *contrast = c->contrast;
2351 *saturation= c->saturation;
2356 static int handle_jpeg(enum PixelFormat *format)
2359 case PIX_FMT_YUVJ420P:
2360 *format = PIX_FMT_YUV420P;
2362 case PIX_FMT_YUVJ422P:
2363 *format = PIX_FMT_YUV422P;
2365 case PIX_FMT_YUVJ444P:
2366 *format = PIX_FMT_YUV444P;
2368 case PIX_FMT_YUVJ440P:
2369 *format = PIX_FMT_YUV440P;
2376 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
2377 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2378 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
2383 int usesVFilter, usesHFilter;
2384 int unscaled, needsDither;
2385 int srcRange, dstRange;
2386 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2388 if (flags & SWS_CPU_CAPS_MMX)
2389 __asm__ volatile("emms\n\t"::: "memory");
2392 #if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
2393 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2394 #if COMPILE_TEMPLATE_MMX2
2395 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2396 #elif COMPILE_TEMPLATE_AMD3DNOW
2397 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2398 #elif COMPILE_TEMPLATE_MMX
2399 flags |= SWS_CPU_CAPS_MMX;
2400 #elif COMPILE_TEMPLATE_ALTIVEC
2401 flags |= SWS_CPU_CAPS_ALTIVEC;
2403 flags |= SWS_CPU_CAPS_BFIN;
2405 #endif /* CONFIG_RUNTIME_CPUDETECT */
2406 if (!rgb15to16) sws_rgb2rgb_init(flags);
2408 unscaled = (srcW == dstW && srcH == dstH);
2409 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2410 && (fmt_depth(dstFormat))<24
2411 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2413 srcRange = handle_jpeg(&srcFormat);
2414 dstRange = handle_jpeg(&dstFormat);
2416 if (!isSupportedIn(srcFormat)) {
2417 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2420 if (!isSupportedOut(dstFormat)) {
2421 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2425 i= flags & ( SWS_POINT
2436 if(!i || (i & (i-1))) {
2437 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
2442 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
2443 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2444 srcW, srcH, dstW, dstH);
2447 if(srcW > VOFW || dstW > VOFW) {
2448 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2452 if (!dstFilter) dstFilter= &dummyFilter;
2453 if (!srcFilter) srcFilter= &dummyFilter;
2455 FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
2457 c->av_class = &sws_context_class;
2462 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2463 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2465 c->dstFormat= dstFormat;
2466 c->srcFormat= srcFormat;
2467 c->vRounder= 4* 0x0001000100010001ULL;
2469 usesHFilter= usesVFilter= 0;
2470 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2471 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2472 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2473 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2474 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2475 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2476 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2477 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2479 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2480 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2482 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
2483 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2485 // drop some chroma lines if the user wants it
2486 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2487 c->chrSrcVSubSample+= c->vChrDrop;
2489 // drop every other pixel for chroma calculation unless user wants full chroma
2490 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2491 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2492 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2493 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2494 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2495 c->chrSrcHSubSample=1;
2498 c->param[0] = param[0];
2499 c->param[1] = param[1];
2502 c->param[1] = SWS_PARAM_DEFAULT;
2505 // Note the -((-x)>>y) is so that we always round toward +inf.
2506 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2507 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2508 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2509 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2511 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2513 /* unscaled special cases */
2514 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat))) {
2516 if ((srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21)) {
2517 c->swScale= PlanarToNV12Wrapper;
2520 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P || srcFormat==PIX_FMT_YUVA420P) && (isBGR(dstFormat) || isRGB(dstFormat))
2521 && !(flags & SWS_ACCURATE_RND) && !(dstH&1)) {
2522 c->swScale= ff_yuv2rgb_get_func_ptr(c);
2525 if (srcFormat==PIX_FMT_YUV410P && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_BITEXACT)) {
2526 c->swScale= yvu9toyv12Wrapper;
2530 if (srcFormat==PIX_FMT_BGR24 && (dstFormat==PIX_FMT_YUV420P || dstFormat==PIX_FMT_YUVA420P) && !(flags & SWS_ACCURATE_RND))
2531 c->swScale= bgr24toyv12Wrapper;
2533 /* RGB/BGR -> RGB/BGR (no dither needed forms) */
2534 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2535 && (isBGR(dstFormat) || isRGB(dstFormat))
2536 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2537 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2538 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2539 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2540 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2541 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2542 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2543 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2544 && dstFormat != PIX_FMT_RGB32_1
2545 && dstFormat != PIX_FMT_BGR32_1
2546 && srcFormat != PIX_FMT_RGB48LE && dstFormat != PIX_FMT_RGB48LE
2547 && srcFormat != PIX_FMT_RGB48BE && dstFormat != PIX_FMT_RGB48BE
2548 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2549 c->swScale= rgb2rgbWrapper;
2551 if ((usePal(srcFormat) && (
2552 dstFormat == PIX_FMT_RGB32 ||
2553 dstFormat == PIX_FMT_RGB32_1 ||
2554 dstFormat == PIX_FMT_RGB24 ||
2555 dstFormat == PIX_FMT_BGR32 ||
2556 dstFormat == PIX_FMT_BGR32_1 ||
2557 dstFormat == PIX_FMT_BGR24)))
2558 c->swScale= pal2rgbWrapper;
2560 if (srcFormat == PIX_FMT_YUV422P) {
2561 if (dstFormat == PIX_FMT_YUYV422)
2562 c->swScale= YUV422PToYuy2Wrapper;
2563 else if (dstFormat == PIX_FMT_UYVY422)
2564 c->swScale= YUV422PToUyvyWrapper;
2567 /* LQ converters if -sws 0 or -sws 4*/
2568 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)) {
2570 if (srcFormat == PIX_FMT_YUV420P || srcFormat == PIX_FMT_YUVA420P) {
2571 if (dstFormat == PIX_FMT_YUYV422)
2572 c->swScale= PlanarToYuy2Wrapper;
2573 else if (dstFormat == PIX_FMT_UYVY422)
2574 c->swScale= PlanarToUyvyWrapper;
2577 if(srcFormat == PIX_FMT_YUYV422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2578 c->swScale= YUYV2YUV420Wrapper;
2579 if(srcFormat == PIX_FMT_UYVY422 && (dstFormat == PIX_FMT_YUV420P || dstFormat == PIX_FMT_YUVA420P))
2580 c->swScale= UYVY2YUV420Wrapper;
2581 if(srcFormat == PIX_FMT_YUYV422 && dstFormat == PIX_FMT_YUV422P)
2582 c->swScale= YUYV2YUV422Wrapper;
2583 if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
2584 c->swScale= UYVY2YUV422Wrapper;
2586 #ifdef COMPILE_ALTIVEC
2587 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2588 !(c->flags & SWS_BITEXACT) &&
2589 srcFormat == PIX_FMT_YUV420P) {
2590 // unscaled YV12 -> packed YUV, we want speed
2591 if (dstFormat == PIX_FMT_YUYV422)
2592 c->swScale= yv12toyuy2_unscaled_altivec;
2593 else if (dstFormat == PIX_FMT_UYVY422)
2594 c->swScale= yv12touyvy_unscaled_altivec;
2599 if ( srcFormat == dstFormat
2600 || (srcFormat == PIX_FMT_YUVA420P && dstFormat == PIX_FMT_YUV420P)
2601 || (srcFormat == PIX_FMT_YUV420P && dstFormat == PIX_FMT_YUVA420P)
2602 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2603 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2604 || (isGray(dstFormat) && isGray(srcFormat))
2605 || (isPlanarYUV(srcFormat) && isPlanarYUV(dstFormat)
2606 && c->chrDstHSubSample == c->chrSrcHSubSample
2607 && c->chrDstVSubSample == c->chrSrcVSubSample
2608 && dstFormat != PIX_FMT_NV12 && dstFormat != PIX_FMT_NV21
2609 && srcFormat != PIX_FMT_NV12 && srcFormat != PIX_FMT_NV21))
2611 if (isPacked(c->srcFormat))
2612 c->swScale= packedCopy;
2613 else /* Planar YUV or gray */
2614 c->swScale= planarCopy;
2617 if (flags & SWS_CPU_CAPS_BFIN)
2618 ff_bfin_get_unscaled_swscale (c);
2622 if (flags&SWS_PRINT_INFO)
2623 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2624 sws_format_name(srcFormat), sws_format_name(dstFormat));
2629 if (flags & SWS_CPU_CAPS_MMX2) {
2630 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2631 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
2632 if (flags&SWS_PRINT_INFO)
2633 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
2635 if (usesHFilter) c->canMMX2BeUsed=0;
2640 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2641 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2643 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2644 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2645 // n-2 is the last chrominance sample available
2646 // this is not perfect, but no one should notice the difference, the more correct variant
2647 // would be like the vertical one, but that would require some special code for the
2648 // first and last pixel
2649 if (flags&SWS_FAST_BILINEAR) {
2650 if (c->canMMX2BeUsed) {
2654 //we don't use the x86 asm scaler if MMX is available
2655 else if (flags & SWS_CPU_CAPS_MMX) {
2656 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2657 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2661 /* precalculate horizontal scaler filter coefficients */
2663 #if defined(COMPILE_MMX2)
2664 // can't downscale !!!
2665 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
2666 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
2667 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
2669 #ifdef MAP_ANONYMOUS
2670 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2671 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2672 #elif HAVE_VIRTUALALLOC
2673 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2674 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
2676 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
2677 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
2680 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
2681 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
2682 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
2683 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
2685 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
2686 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
2688 #ifdef MAP_ANONYMOUS
2689 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2690 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
2693 #endif /* defined(COMPILE_MMX2) */
2695 const int filterAlign=
2696 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2697 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2700 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2701 srcW , dstW, filterAlign, 1<<14,
2702 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2703 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
2705 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2706 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2707 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2708 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
2711 } // initialize horizontal stuff
2713 /* precalculate vertical scaler filter coefficients */
2715 const int filterAlign=
2716 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2717 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2720 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2721 srcH , dstH, filterAlign, (1<<12),
2722 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2723 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
2725 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2726 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2727 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2728 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
2731 #ifdef COMPILE_ALTIVEC
2732 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
2733 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
2735 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2737 short *p = (short *)&c->vYCoeffsBank[i];
2739 p[j] = c->vLumFilter[i];
2742 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2744 short *p = (short *)&c->vCCoeffsBank[i];
2746 p[j] = c->vChrFilter[i];
2751 // calculate buffer sizes so that they won't run out while handling these damn slices
2752 c->vLumBufSize= c->vLumFilterSize;
2753 c->vChrBufSize= c->vChrFilterSize;
2754 for (i=0; i<dstH; i++) {
2755 int chrI= i*c->chrDstH / dstH;
2756 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2757 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2759 nextSlice>>= c->chrSrcVSubSample;
2760 nextSlice<<= c->chrSrcVSubSample;
2761 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2762 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2763 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2764 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2767 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2768 // allocate several megabytes to handle all possible cases)
2769 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
2770 FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
2771 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
2772 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
2773 //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)
2774 /* align at 16 bytes for AltiVec */
2775 for (i=0; i<c->vLumBufSize; i++) {
2776 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
2777 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
2779 for (i=0; i<c->vChrBufSize; i++) {
2780 FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
2781 c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
2783 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
2784 for (i=0; i<c->vLumBufSize; i++) {
2785 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
2786 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
2789 //try to avoid drawing green stuff between the right end and the stride end
2790 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2792 assert(2*VOFW == VOF);
2794 assert(c->chrDstH <= dstH);
2796 if (flags&SWS_PRINT_INFO) {
2797 if (flags&SWS_FAST_BILINEAR)
2798 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2799 else if (flags&SWS_BILINEAR)
2800 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2801 else if (flags&SWS_BICUBIC)
2802 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2803 else if (flags&SWS_X)
2804 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2805 else if (flags&SWS_POINT)
2806 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2807 else if (flags&SWS_AREA)
2808 av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
2809 else if (flags&SWS_BICUBLIN)
2810 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2811 else if (flags&SWS_GAUSS)
2812 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2813 else if (flags&SWS_SINC)
2814 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2815 else if (flags&SWS_LANCZOS)
2816 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2817 else if (flags&SWS_SPLINE)
2818 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2820 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2822 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
2823 sws_format_name(srcFormat),
2825 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ? "dithered " : "",
2829 sws_format_name(dstFormat));
2831 if (flags & SWS_CPU_CAPS_MMX2)
2832 av_log(c, AV_LOG_INFO, "using MMX2\n");
2833 else if (flags & SWS_CPU_CAPS_3DNOW)
2834 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2835 else if (flags & SWS_CPU_CAPS_MMX)
2836 av_log(c, AV_LOG_INFO, "using MMX\n");
2837 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2838 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2840 av_log(c, AV_LOG_INFO, "using C\n");
2843 if (flags & SWS_PRINT_INFO) {
2844 if (flags & SWS_CPU_CAPS_MMX) {
2845 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2846 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2848 if (c->hLumFilterSize==4)
2849 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2850 else if (c->hLumFilterSize==8)
2851 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2853 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2855 if (c->hChrFilterSize==4)
2856 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2857 else if (c->hChrFilterSize==8)
2858 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2860 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2864 av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
2866 if (flags & SWS_FAST_BILINEAR)
2867 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2869 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2872 if (isPlanarYUV(dstFormat)) {
2873 if (c->vLumFilterSize==1)
2874 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2876 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2878 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2879 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2880 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2881 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2882 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2884 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2887 if (dstFormat==PIX_FMT_BGR24)
2888 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
2889 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2890 else if (dstFormat==PIX_FMT_RGB32)
2891 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2892 else if (dstFormat==PIX_FMT_BGR565)
2893 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2894 else if (dstFormat==PIX_FMT_BGR555)
2895 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2897 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2899 if (flags & SWS_PRINT_INFO) {
2900 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2901 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2902 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2903 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2906 c->swScale= getSwsFunc(c);
2914 static void reset_ptr(const uint8_t* src[], int format)
2916 if(!isALPHA(format))
2918 if(!isPlanarYUV(format)) {
2921 if (!usePal(format))
2927 * swscale wrapper, so we don't need to export the SwsContext.
2928 * Assumes planar YUV to be in YUV order instead of YVU.
2930 int sws_scale(SwsContext *c, const uint8_t* const src[], const int srcStride[], int srcSliceY,
2931 int srcSliceH, uint8_t* const dst[], const int dstStride[])
2934 const uint8_t* src2[4]= {src[0], src[1], src[2], src[3]};
2935 uint8_t* dst2[4]= {dst[0], dst[1], dst[2], dst[3]};
2937 // do not mess up sliceDir if we have a "trailing" 0-size slice
2941 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2942 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2945 if (c->sliceDir == 0) {
2946 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2949 if (usePal(c->srcFormat)) {
2950 for (i=0; i<256; i++) {
2951 int p, r, g, b,y,u,v;
2952 if(c->srcFormat == PIX_FMT_PAL8) {
2953 p=((const uint32_t*)(src[1]))[i];
2957 } else if(c->srcFormat == PIX_FMT_RGB8) {
2961 } else if(c->srcFormat == PIX_FMT_BGR8) {
2965 } else if(c->srcFormat == PIX_FMT_RGB4_BYTE) {
2970 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2975 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2976 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2977 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2978 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2980 switch(c->dstFormat) {
2985 c->pal_rgb[i]= r + (g<<8) + (b<<16);
2987 case PIX_FMT_BGR32_1:
2991 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
2993 case PIX_FMT_RGB32_1:
2997 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
3004 c->pal_rgb[i]= b + (g<<8) + (r<<16);
3009 // copy strides, so they can safely be modified
3010 if (c->sliceDir == 1) {
3011 // slices go from top to bottom
3012 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2], srcStride[3]};
3013 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2], dstStride[3]};
3015 reset_ptr(src2, c->srcFormat);
3016 reset_ptr((const uint8_t**)dst2, c->dstFormat);
3018 /* reset slice direction at end of frame */
3019 if (srcSliceY + srcSliceH == c->srcH)
3022 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2, dstStride2);
3024 // slices go from bottom to top => we flip the image internally
3025 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2], -srcStride[3]};
3026 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2], -dstStride[3]};
3028 src2[0] += (srcSliceH-1)*srcStride[0];
3029 if (!usePal(c->srcFormat))
3030 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
3031 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
3032 src2[3] += (srcSliceH-1)*srcStride[3];
3033 dst2[0] += ( c->dstH -1)*dstStride[0];
3034 dst2[1] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1];
3035 dst2[2] += ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2];
3036 dst2[3] += ( c->dstH -1)*dstStride[3];
3038 reset_ptr(src2, c->srcFormat);
3039 reset_ptr((const uint8_t**)dst2, c->dstFormat);
3041 /* reset slice direction at end of frame */
3045 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
3049 #if LIBSWSCALE_VERSION_MAJOR < 1
3050 int sws_scale_ordered(SwsContext *c, const uint8_t* const src[], int srcStride[], int srcSliceY,
3051 int srcSliceH, uint8_t* dst[], int dstStride[])
3053 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
3057 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
3058 float lumaSharpen, float chromaSharpen,
3059 float chromaHShift, float chromaVShift,
3062 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
3066 if (lumaGBlur!=0.0) {
3067 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
3068 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
3070 filter->lumH= sws_getIdentityVec();
3071 filter->lumV= sws_getIdentityVec();
3074 if (chromaGBlur!=0.0) {
3075 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
3076 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
3078 filter->chrH= sws_getIdentityVec();
3079 filter->chrV= sws_getIdentityVec();
3082 if (chromaSharpen!=0.0) {
3083 SwsVector *id= sws_getIdentityVec();
3084 sws_scaleVec(filter->chrH, -chromaSharpen);
3085 sws_scaleVec(filter->chrV, -chromaSharpen);
3086 sws_addVec(filter->chrH, id);
3087 sws_addVec(filter->chrV, id);
3091 if (lumaSharpen!=0.0) {
3092 SwsVector *id= sws_getIdentityVec();
3093 sws_scaleVec(filter->lumH, -lumaSharpen);
3094 sws_scaleVec(filter->lumV, -lumaSharpen);
3095 sws_addVec(filter->lumH, id);
3096 sws_addVec(filter->lumV, id);
3100 if (chromaHShift != 0.0)
3101 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
3103 if (chromaVShift != 0.0)
3104 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
3106 sws_normalizeVec(filter->chrH, 1.0);
3107 sws_normalizeVec(filter->chrV, 1.0);
3108 sws_normalizeVec(filter->lumH, 1.0);
3109 sws_normalizeVec(filter->lumV, 1.0);
3111 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
3112 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
3117 SwsVector *sws_allocVec(int length)
3119 SwsVector *vec = av_malloc(sizeof(SwsVector));
3122 vec->length = length;
3123 vec->coeff = av_malloc(sizeof(double) * length);
3129 SwsVector *sws_getGaussianVec(double variance, double quality)
3131 const int length= (int)(variance*quality + 0.5) | 1;
3133 double middle= (length-1)*0.5;
3134 SwsVector *vec= sws_allocVec(length);
3139 for (i=0; i<length; i++) {
3140 double dist= i-middle;
3141 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
3144 sws_normalizeVec(vec, 1.0);
3149 SwsVector *sws_getConstVec(double c, int length)
3152 SwsVector *vec= sws_allocVec(length);
3157 for (i=0; i<length; i++)
3163 SwsVector *sws_getIdentityVec(void)
3165 return sws_getConstVec(1.0, 1);
3168 double sws_dcVec(SwsVector *a)
3173 for (i=0; i<a->length; i++)
3179 void sws_scaleVec(SwsVector *a, double scalar)
3183 for (i=0; i<a->length; i++)
3184 a->coeff[i]*= scalar;
3187 void sws_normalizeVec(SwsVector *a, double height)
3189 sws_scaleVec(a, height/sws_dcVec(a));
3192 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
3194 int length= a->length + b->length - 1;
3196 SwsVector *vec= sws_getConstVec(0.0, length);
3201 for (i=0; i<a->length; i++) {
3202 for (j=0; j<b->length; j++) {
3203 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
3210 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
3212 int length= FFMAX(a->length, b->length);
3214 SwsVector *vec= sws_getConstVec(0.0, length);
3219 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3220 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
3225 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
3227 int length= FFMAX(a->length, b->length);
3229 SwsVector *vec= sws_getConstVec(0.0, length);
3234 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3235 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
3240 /* shift left / or right if "shift" is negative */
3241 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
3243 int length= a->length + FFABS(shift)*2;
3245 SwsVector *vec= sws_getConstVec(0.0, length);
3250 for (i=0; i<a->length; i++) {
3251 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
3257 void sws_shiftVec(SwsVector *a, int shift)
3259 SwsVector *shifted= sws_getShiftedVec(a, shift);
3261 a->coeff= shifted->coeff;
3262 a->length= shifted->length;
3266 void sws_addVec(SwsVector *a, SwsVector *b)
3268 SwsVector *sum= sws_sumVec(a, b);
3270 a->coeff= sum->coeff;
3271 a->length= sum->length;
3275 void sws_subVec(SwsVector *a, SwsVector *b)
3277 SwsVector *diff= sws_diffVec(a, b);
3279 a->coeff= diff->coeff;
3280 a->length= diff->length;
3284 void sws_convVec(SwsVector *a, SwsVector *b)
3286 SwsVector *conv= sws_getConvVec(a, b);
3288 a->coeff= conv->coeff;
3289 a->length= conv->length;
3293 SwsVector *sws_cloneVec(SwsVector *a)
3296 SwsVector *vec= sws_allocVec(a->length);
3301 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
3306 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
3313 for (i=0; i<a->length; i++)
3314 if (a->coeff[i]>max) max= a->coeff[i];
3316 for (i=0; i<a->length; i++)
3317 if (a->coeff[i]<min) min= a->coeff[i];
3321 for (i=0; i<a->length; i++) {
3322 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
3323 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
3324 for (;x>0; x--) av_log(log_ctx, log_level, " ");
3325 av_log(log_ctx, log_level, "|\n");
3329 #if LIBSWSCALE_VERSION_MAJOR < 1
3330 void sws_printVec(SwsVector *a)
3332 sws_printVec2(a, NULL, AV_LOG_DEBUG);
3336 void sws_freeVec(SwsVector *a)
3339 av_freep(&a->coeff);
3344 void sws_freeFilter(SwsFilter *filter)
3346 if (!filter) return;
3348 if (filter->lumH) sws_freeVec(filter->lumH);
3349 if (filter->lumV) sws_freeVec(filter->lumV);
3350 if (filter->chrH) sws_freeVec(filter->chrH);
3351 if (filter->chrV) sws_freeVec(filter->chrV);
3355 void sws_freeContext(SwsContext *c)
3361 for (i=0; i<c->vLumBufSize; i++)
3362 av_freep(&c->lumPixBuf[i]);
3363 av_freep(&c->lumPixBuf);
3367 for (i=0; i<c->vChrBufSize; i++)
3368 av_freep(&c->chrPixBuf[i]);
3369 av_freep(&c->chrPixBuf);
3372 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
3373 for (i=0; i<c->vLumBufSize; i++)
3374 av_freep(&c->alpPixBuf[i]);
3375 av_freep(&c->alpPixBuf);
3378 av_freep(&c->vLumFilter);
3379 av_freep(&c->vChrFilter);
3380 av_freep(&c->hLumFilter);
3381 av_freep(&c->hChrFilter);
3382 #ifdef COMPILE_ALTIVEC
3383 av_freep(&c->vYCoeffsBank);
3384 av_freep(&c->vCCoeffsBank);
3387 av_freep(&c->vLumFilterPos);
3388 av_freep(&c->vChrFilterPos);
3389 av_freep(&c->hLumFilterPos);
3390 av_freep(&c->hChrFilterPos);
3392 #if ARCH_X86 && CONFIG_GPL
3393 #ifdef MAP_ANONYMOUS
3394 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
3395 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
3396 #elif HAVE_VIRTUALALLOC
3397 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, MEM_RELEASE);
3398 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, MEM_RELEASE);
3400 av_free(c->lumMmx2FilterCode);
3401 av_free(c->chrMmx2FilterCode);
3403 c->lumMmx2FilterCode=NULL;
3404 c->chrMmx2FilterCode=NULL;
3405 #endif /* ARCH_X86 && CONFIG_GPL */
3407 av_freep(&c->yuvTable);
3412 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3413 int srcW, int srcH, enum PixelFormat srcFormat,
3414 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
3415 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
3417 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3420 param = default_param;
3423 if (context->srcW != srcW || context->srcH != srcH ||
3424 context->srcFormat != srcFormat ||
3425 context->dstW != dstW || context->dstH != dstH ||
3426 context->dstFormat != dstFormat || context->flags != flags ||
3427 context->param[0] != param[0] || context->param[1] != param[1])
3429 sws_freeContext(context);
3434 return sws_getContext(srcW, srcH, srcFormat,
3435 dstW, dstH, dstFormat, flags,
3436 srcFilter, dstFilter, param);