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
67 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
68 #define MAP_ANONYMOUS MAP_ANON
72 #include "swscale_internal.h"
74 #include "libavutil/x86_cpu.h"
75 #include "libavutil/bswap.h"
77 unsigned swscale_version(void)
79 return LIBSWSCALE_VERSION_INT;
89 //#define WORDS_BIGENDIAN
92 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
94 #define RET 0xC3 //near return opcode for X86
99 #define PI 3.14159265358979323846
102 #define isSupportedIn(x) ( \
103 (x)==PIX_FMT_YUV420P \
104 || (x)==PIX_FMT_YUVA420P \
105 || (x)==PIX_FMT_YUYV422 \
106 || (x)==PIX_FMT_UYVY422 \
107 || (x)==PIX_FMT_RGB32 \
108 || (x)==PIX_FMT_RGB32_1 \
109 || (x)==PIX_FMT_BGR24 \
110 || (x)==PIX_FMT_BGR565 \
111 || (x)==PIX_FMT_BGR555 \
112 || (x)==PIX_FMT_BGR32 \
113 || (x)==PIX_FMT_BGR32_1 \
114 || (x)==PIX_FMT_RGB24 \
115 || (x)==PIX_FMT_RGB565 \
116 || (x)==PIX_FMT_RGB555 \
117 || (x)==PIX_FMT_GRAY8 \
118 || (x)==PIX_FMT_YUV410P \
119 || (x)==PIX_FMT_YUV440P \
120 || (x)==PIX_FMT_GRAY16BE \
121 || (x)==PIX_FMT_GRAY16LE \
122 || (x)==PIX_FMT_YUV444P \
123 || (x)==PIX_FMT_YUV422P \
124 || (x)==PIX_FMT_YUV411P \
125 || (x)==PIX_FMT_PAL8 \
126 || (x)==PIX_FMT_BGR8 \
127 || (x)==PIX_FMT_RGB8 \
128 || (x)==PIX_FMT_BGR4_BYTE \
129 || (x)==PIX_FMT_RGB4_BYTE \
130 || (x)==PIX_FMT_YUV440P \
131 || (x)==PIX_FMT_MONOWHITE \
132 || (x)==PIX_FMT_MONOBLACK \
134 #define isSupportedOut(x) ( \
135 (x)==PIX_FMT_YUV420P \
136 || (x)==PIX_FMT_YUYV422 \
137 || (x)==PIX_FMT_UYVY422 \
138 || (x)==PIX_FMT_YUV444P \
139 || (x)==PIX_FMT_YUV422P \
140 || (x)==PIX_FMT_YUV411P \
143 || (x)==PIX_FMT_NV12 \
144 || (x)==PIX_FMT_NV21 \
145 || (x)==PIX_FMT_GRAY16BE \
146 || (x)==PIX_FMT_GRAY16LE \
147 || (x)==PIX_FMT_GRAY8 \
148 || (x)==PIX_FMT_YUV410P \
149 || (x)==PIX_FMT_YUV440P \
151 #define isPacked(x) ( \
153 || (x)==PIX_FMT_YUYV422 \
154 || (x)==PIX_FMT_UYVY422 \
158 #define usePal(x) ( \
160 || (x)==PIX_FMT_BGR4_BYTE \
161 || (x)==PIX_FMT_RGB4_BYTE \
162 || (x)==PIX_FMT_BGR8 \
163 || (x)==PIX_FMT_RGB8 \
166 #define RGB2YUV_SHIFT 15
167 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
168 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
169 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
170 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
171 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
172 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
173 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
174 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
175 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
177 extern const int32_t Inverse_Table_6_9[8][4];
179 static const double rgb2yuv_table[8][9]={
180 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
181 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
182 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
183 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
184 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
185 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
186 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
187 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
192 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
195 more intelligent misalignment avoidance for the horizontal scaler
196 write special vertical cubic upscale version
197 Optimize C code (yv12 / minmax)
198 add support for packed pixel yuv input & output
199 add support for Y8 output
200 optimize bgr24 & bgr32
201 add BGR4 output support
202 write special BGR->BGR scaler
205 #if ARCH_X86 && CONFIG_GPL
206 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
207 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
208 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
209 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
210 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
211 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
212 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
213 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
215 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
216 0x0103010301030103LL,
217 0x0200020002000200LL,};
219 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
220 0x0602060206020602LL,
221 0x0004000400040004LL,};
223 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
224 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
225 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
226 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
227 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
228 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
230 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
231 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
232 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
235 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
236 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
237 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
239 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
240 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
241 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
242 #endif /* FAST_BGR2YV12 */
243 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
244 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
245 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
247 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
248 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
249 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
250 DECLARE_ASM_CONST(8, uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
251 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
253 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUV[2][4]) = {
254 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
255 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
258 DECLARE_ASM_CONST(8, uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
260 #endif /* ARCH_X86 */
262 // clipping helper table for C implementations:
263 static unsigned char clip_table[768];
265 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
267 static const uint8_t __attribute__((aligned(8))) dither_2x2_4[2][8]={
268 { 1, 3, 1, 3, 1, 3, 1, 3, },
269 { 2, 0, 2, 0, 2, 0, 2, 0, },
272 static const uint8_t __attribute__((aligned(8))) dither_2x2_8[2][8]={
273 { 6, 2, 6, 2, 6, 2, 6, 2, },
274 { 0, 4, 0, 4, 0, 4, 0, 4, },
277 const uint8_t __attribute__((aligned(8))) dither_8x8_32[8][8]={
278 { 17, 9, 23, 15, 16, 8, 22, 14, },
279 { 5, 29, 3, 27, 4, 28, 2, 26, },
280 { 21, 13, 19, 11, 20, 12, 18, 10, },
281 { 0, 24, 6, 30, 1, 25, 7, 31, },
282 { 16, 8, 22, 14, 17, 9, 23, 15, },
283 { 4, 28, 2, 26, 5, 29, 3, 27, },
284 { 20, 12, 18, 10, 21, 13, 19, 11, },
285 { 1, 25, 7, 31, 0, 24, 6, 30, },
289 const uint8_t __attribute__((aligned(8))) dither_8x8_64[8][8]={
290 { 0, 48, 12, 60, 3, 51, 15, 63, },
291 { 32, 16, 44, 28, 35, 19, 47, 31, },
292 { 8, 56, 4, 52, 11, 59, 7, 55, },
293 { 40, 24, 36, 20, 43, 27, 39, 23, },
294 { 2, 50, 14, 62, 1, 49, 13, 61, },
295 { 34, 18, 46, 30, 33, 17, 45, 29, },
296 { 10, 58, 6, 54, 9, 57, 5, 53, },
297 { 42, 26, 38, 22, 41, 25, 37, 21, },
301 const uint8_t __attribute__((aligned(8))) dither_8x8_73[8][8]={
302 { 0, 55, 14, 68, 3, 58, 17, 72, },
303 { 37, 18, 50, 32, 40, 22, 54, 35, },
304 { 9, 64, 5, 59, 13, 67, 8, 63, },
305 { 46, 27, 41, 23, 49, 31, 44, 26, },
306 { 2, 57, 16, 71, 1, 56, 15, 70, },
307 { 39, 21, 52, 34, 38, 19, 51, 33, },
308 { 11, 66, 7, 62, 10, 65, 6, 60, },
309 { 48, 30, 43, 25, 47, 29, 42, 24, },
313 const uint8_t __attribute__((aligned(8))) dither_8x8_128[8][8]={
314 { 68, 36, 92, 60, 66, 34, 90, 58, },
315 { 20, 116, 12, 108, 18, 114, 10, 106, },
316 { 84, 52, 76, 44, 82, 50, 74, 42, },
317 { 0, 96, 24, 120, 6, 102, 30, 126, },
318 { 64, 32, 88, 56, 70, 38, 94, 62, },
319 { 16, 112, 8, 104, 22, 118, 14, 110, },
320 { 80, 48, 72, 40, 86, 54, 78, 46, },
321 { 4, 100, 28, 124, 2, 98, 26, 122, },
326 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
327 {117, 62, 158, 103, 113, 58, 155, 100, },
328 { 34, 199, 21, 186, 31, 196, 17, 182, },
329 {144, 89, 131, 76, 141, 86, 127, 72, },
330 { 0, 165, 41, 206, 10, 175, 52, 217, },
331 {110, 55, 151, 96, 120, 65, 162, 107, },
332 { 28, 193, 14, 179, 38, 203, 24, 189, },
333 {138, 83, 124, 69, 148, 93, 134, 79, },
334 { 7, 172, 48, 213, 3, 168, 45, 210, },
337 // tries to correct a gamma of 1.5
338 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
339 { 0, 143, 18, 200, 2, 156, 25, 215, },
340 { 78, 28, 125, 64, 89, 36, 138, 74, },
341 { 10, 180, 3, 161, 16, 195, 8, 175, },
342 {109, 51, 93, 38, 121, 60, 105, 47, },
343 { 1, 152, 23, 210, 0, 147, 20, 205, },
344 { 85, 33, 134, 71, 81, 30, 130, 67, },
345 { 14, 190, 6, 171, 12, 185, 5, 166, },
346 {117, 57, 101, 44, 113, 54, 97, 41, },
349 // tries to correct a gamma of 2.0
350 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
351 { 0, 124, 8, 193, 0, 140, 12, 213, },
352 { 55, 14, 104, 42, 66, 19, 119, 52, },
353 { 3, 168, 1, 145, 6, 187, 3, 162, },
354 { 86, 31, 70, 21, 99, 39, 82, 28, },
355 { 0, 134, 11, 206, 0, 129, 9, 200, },
356 { 62, 17, 114, 48, 58, 16, 109, 45, },
357 { 5, 181, 2, 157, 4, 175, 1, 151, },
358 { 95, 36, 78, 26, 90, 34, 74, 24, },
361 // tries to correct a gamma of 2.5
362 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
363 { 0, 107, 3, 187, 0, 125, 6, 212, },
364 { 39, 7, 86, 28, 49, 11, 102, 36, },
365 { 1, 158, 0, 131, 3, 180, 1, 151, },
366 { 68, 19, 52, 12, 81, 25, 64, 17, },
367 { 0, 119, 5, 203, 0, 113, 4, 195, },
368 { 45, 9, 96, 33, 42, 8, 91, 30, },
369 { 2, 172, 1, 144, 2, 165, 0, 137, },
370 { 77, 23, 60, 15, 72, 21, 56, 14, },
374 const char *sws_format_name(enum PixelFormat format)
377 case PIX_FMT_YUV420P:
379 case PIX_FMT_YUVA420P:
381 case PIX_FMT_YUYV422:
387 case PIX_FMT_YUV422P:
389 case PIX_FMT_YUV444P:
393 case PIX_FMT_YUV410P:
395 case PIX_FMT_YUV411P:
401 case PIX_FMT_GRAY16BE:
403 case PIX_FMT_GRAY16LE:
407 case PIX_FMT_MONOWHITE:
409 case PIX_FMT_MONOBLACK:
413 case PIX_FMT_YUVJ420P:
415 case PIX_FMT_YUVJ422P:
417 case PIX_FMT_YUVJ444P:
419 case PIX_FMT_XVMC_MPEG2_MC:
420 return "xvmc_mpeg2_mc";
421 case PIX_FMT_XVMC_MPEG2_IDCT:
422 return "xvmc_mpeg2_idct";
423 case PIX_FMT_UYVY422:
425 case PIX_FMT_UYYVYY411:
427 case PIX_FMT_RGB32_1:
429 case PIX_FMT_BGR32_1:
441 case PIX_FMT_BGR4_BYTE:
447 case PIX_FMT_RGB4_BYTE:
453 case PIX_FMT_YUV440P:
455 case PIX_FMT_VDPAU_H264:
457 case PIX_FMT_VDPAU_MPEG1:
458 return "vdpau_mpeg1";
459 case PIX_FMT_VDPAU_MPEG2:
460 return "vdpau_mpeg2";
462 return "Unknown format";
466 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
467 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
468 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
470 //FIXME Optimize (just quickly writen not opti..)
472 for (i=0; i<dstW; i++)
476 for (j=0; j<lumFilterSize; j++)
477 val += lumSrc[j][i] * lumFilter[j];
479 dest[i]= av_clip_uint8(val>>19);
483 for (i=0; i<chrDstW; i++)
488 for (j=0; j<chrFilterSize; j++)
490 u += chrSrc[j][i] * chrFilter[j];
491 v += chrSrc[j][i + VOFW] * chrFilter[j];
494 uDest[i]= av_clip_uint8(u>>19);
495 vDest[i]= av_clip_uint8(v>>19);
499 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
500 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
501 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
503 //FIXME Optimize (just quickly writen not opti..)
505 for (i=0; i<dstW; i++)
509 for (j=0; j<lumFilterSize; j++)
510 val += lumSrc[j][i] * lumFilter[j];
512 dest[i]= av_clip_uint8(val>>19);
518 if (dstFormat == PIX_FMT_NV12)
519 for (i=0; i<chrDstW; i++)
524 for (j=0; j<chrFilterSize; j++)
526 u += chrSrc[j][i] * chrFilter[j];
527 v += chrSrc[j][i + VOFW] * chrFilter[j];
530 uDest[2*i]= av_clip_uint8(u>>19);
531 uDest[2*i+1]= av_clip_uint8(v>>19);
534 for (i=0; i<chrDstW; i++)
539 for (j=0; j<chrFilterSize; j++)
541 u += chrSrc[j][i] * chrFilter[j];
542 v += chrSrc[j][i + VOFW] * chrFilter[j];
545 uDest[2*i]= av_clip_uint8(v>>19);
546 uDest[2*i+1]= av_clip_uint8(u>>19);
550 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type) \
551 for (i=0; i<(dstW>>1); i++){\
557 type av_unused *r, *b, *g;\
560 for (j=0; j<lumFilterSize; j++)\
562 Y1 += lumSrc[j][i2] * lumFilter[j];\
563 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
565 for (j=0; j<chrFilterSize; j++)\
567 U += chrSrc[j][i] * chrFilter[j];\
568 V += chrSrc[j][i+VOFW] * chrFilter[j];\
575 #define YSCALE_YUV_2_PACKEDX_C(type) \
576 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)\
577 if ((Y1|Y2|U|V)&256)\
579 if (Y1>255) Y1=255; \
580 else if (Y1<0)Y1=0; \
581 if (Y2>255) Y2=255; \
582 else if (Y2<0)Y2=0; \
589 #define YSCALE_YUV_2_PACKEDX_FULL_C \
590 for (i=0; i<dstW; i++){\
597 for (j=0; j<lumFilterSize; j++){\
598 Y += lumSrc[j][i ] * lumFilter[j];\
600 for (j=0; j<chrFilterSize; j++){\
601 U += chrSrc[j][i ] * chrFilter[j];\
602 V += chrSrc[j][i+VOFW] * chrFilter[j];\
608 #define YSCALE_YUV_2_RGBX_FULL_C(rnd) \
609 YSCALE_YUV_2_PACKEDX_FULL_C\
610 Y-= c->yuv2rgb_y_offset;\
611 Y*= c->yuv2rgb_y_coeff;\
613 R= Y + V*c->yuv2rgb_v2r_coeff;\
614 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
615 B= Y + U*c->yuv2rgb_u2b_coeff;\
616 if ((R|G|B)&(0xC0000000)){\
617 if (R>=(256<<22)) R=(256<<22)-1; \
619 if (G>=(256<<22)) G=(256<<22)-1; \
621 if (B>=(256<<22)) B=(256<<22)-1; \
626 #define YSCALE_YUV_2_GRAY16_C \
627 for (i=0; i<(dstW>>1); i++){\
636 for (j=0; j<lumFilterSize; j++)\
638 Y1 += lumSrc[j][i2] * lumFilter[j];\
639 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
643 if ((Y1|Y2|U|V)&65536)\
645 if (Y1>65535) Y1=65535; \
646 else if (Y1<0)Y1=0; \
647 if (Y2>65535) Y2=65535; \
648 else if (Y2<0)Y2=0; \
651 #define YSCALE_YUV_2_RGBX_C(type) \
652 YSCALE_YUV_2_PACKEDX_C(type) /* FIXME fix tables so that cliping is not needed and then use _NOCLIP*/\
653 r = (type *)c->table_rV[V]; \
654 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
655 b = (type *)c->table_bU[U]; \
657 #define YSCALE_YUV_2_PACKED2_C \
658 for (i=0; i<(dstW>>1); i++){ \
660 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
661 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
662 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
663 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
665 #define YSCALE_YUV_2_GRAY16_2_C \
666 for (i=0; i<(dstW>>1); i++){ \
668 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
669 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
671 #define YSCALE_YUV_2_RGB2_C(type) \
672 YSCALE_YUV_2_PACKED2_C\
674 r = (type *)c->table_rV[V];\
675 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
676 b = (type *)c->table_bU[U];\
678 #define YSCALE_YUV_2_PACKED1_C \
679 for (i=0; i<(dstW>>1); i++){\
681 int Y1= buf0[i2 ]>>7;\
682 int Y2= buf0[i2+1]>>7;\
683 int U= (uvbuf1[i ])>>7;\
684 int V= (uvbuf1[i+VOFW])>>7;\
686 #define YSCALE_YUV_2_GRAY16_1_C \
687 for (i=0; i<(dstW>>1); i++){\
689 int Y1= buf0[i2 ]<<1;\
690 int Y2= buf0[i2+1]<<1;\
692 #define YSCALE_YUV_2_RGB1_C(type) \
693 YSCALE_YUV_2_PACKED1_C\
695 r = (type *)c->table_rV[V];\
696 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
697 b = (type *)c->table_bU[U];\
699 #define YSCALE_YUV_2_PACKED1B_C \
700 for (i=0; i<(dstW>>1); i++){\
702 int Y1= buf0[i2 ]>>7;\
703 int Y2= buf0[i2+1]>>7;\
704 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
705 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
707 #define YSCALE_YUV_2_RGB1B_C(type) \
708 YSCALE_YUV_2_PACKED1B_C\
710 r = (type *)c->table_rV[V];\
711 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
712 b = (type *)c->table_bU[U];\
714 #define YSCALE_YUV_2_MONO2_C \
715 const uint8_t * const d128=dither_8x8_220[y&7];\
716 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
717 for (i=0; i<dstW-7; i+=8){\
719 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
720 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
721 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
722 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
723 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
724 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
725 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
726 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
727 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
732 #define YSCALE_YUV_2_MONOX_C \
733 const uint8_t * const d128=dither_8x8_220[y&7];\
734 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
736 for (i=0; i<dstW-1; i+=2){\
741 for (j=0; j<lumFilterSize; j++)\
743 Y1 += lumSrc[j][i] * lumFilter[j];\
744 Y2 += lumSrc[j][i+1] * lumFilter[j];\
755 acc+= acc + g[Y1+d128[(i+0)&7]];\
756 acc+= acc + g[Y2+d128[(i+1)&7]];\
758 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
764 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
765 switch(c->dstFormat)\
769 case PIX_FMT_RGB32_1:\
770 case PIX_FMT_BGR32_1:\
772 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
773 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
778 ((uint8_t*)dest)[0]= r[Y1];\
779 ((uint8_t*)dest)[1]= g[Y1];\
780 ((uint8_t*)dest)[2]= b[Y1];\
781 ((uint8_t*)dest)[3]= r[Y2];\
782 ((uint8_t*)dest)[4]= g[Y2];\
783 ((uint8_t*)dest)[5]= b[Y2];\
789 ((uint8_t*)dest)[0]= b[Y1];\
790 ((uint8_t*)dest)[1]= g[Y1];\
791 ((uint8_t*)dest)[2]= r[Y1];\
792 ((uint8_t*)dest)[3]= b[Y2];\
793 ((uint8_t*)dest)[4]= g[Y2];\
794 ((uint8_t*)dest)[5]= r[Y2];\
798 case PIX_FMT_RGB565:\
799 case PIX_FMT_BGR565:\
801 const int dr1= dither_2x2_8[y&1 ][0];\
802 const int dg1= dither_2x2_4[y&1 ][0];\
803 const int db1= dither_2x2_8[(y&1)^1][0];\
804 const int dr2= dither_2x2_8[y&1 ][1];\
805 const int dg2= dither_2x2_4[y&1 ][1];\
806 const int db2= dither_2x2_8[(y&1)^1][1];\
808 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
809 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
813 case PIX_FMT_RGB555:\
814 case PIX_FMT_BGR555:\
816 const int dr1= dither_2x2_8[y&1 ][0];\
817 const int dg1= dither_2x2_8[y&1 ][1];\
818 const int db1= dither_2x2_8[(y&1)^1][0];\
819 const int dr2= dither_2x2_8[y&1 ][1];\
820 const int dg2= dither_2x2_8[y&1 ][0];\
821 const int db2= dither_2x2_8[(y&1)^1][1];\
823 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
824 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
831 const uint8_t * const d64= dither_8x8_73[y&7];\
832 const uint8_t * const d32= dither_8x8_32[y&7];\
834 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
835 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
842 const uint8_t * const d64= dither_8x8_73 [y&7];\
843 const uint8_t * const d128=dither_8x8_220[y&7];\
845 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
846 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
850 case PIX_FMT_RGB4_BYTE:\
851 case PIX_FMT_BGR4_BYTE:\
853 const uint8_t * const d64= dither_8x8_73 [y&7];\
854 const uint8_t * const d128=dither_8x8_220[y&7];\
856 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
857 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
861 case PIX_FMT_MONOBLACK:\
862 case PIX_FMT_MONOWHITE:\
867 case PIX_FMT_YUYV422:\
869 ((uint8_t*)dest)[2*i2+0]= Y1;\
870 ((uint8_t*)dest)[2*i2+1]= U;\
871 ((uint8_t*)dest)[2*i2+2]= Y2;\
872 ((uint8_t*)dest)[2*i2+3]= V;\
875 case PIX_FMT_UYVY422:\
877 ((uint8_t*)dest)[2*i2+0]= U;\
878 ((uint8_t*)dest)[2*i2+1]= Y1;\
879 ((uint8_t*)dest)[2*i2+2]= V;\
880 ((uint8_t*)dest)[2*i2+3]= Y2;\
883 case PIX_FMT_GRAY16BE:\
885 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
886 ((uint8_t*)dest)[2*i2+1]= Y1;\
887 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
888 ((uint8_t*)dest)[2*i2+3]= Y2;\
891 case PIX_FMT_GRAY16LE:\
893 ((uint8_t*)dest)[2*i2+0]= Y1;\
894 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
895 ((uint8_t*)dest)[2*i2+2]= Y2;\
896 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
902 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
903 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
904 uint8_t *dest, int dstW, int y)
907 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOX_C)
910 static inline void yuv2rgbXinC_full(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
911 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
912 uint8_t *dest, int dstW, int y)
915 int step= fmt_depth(c->dstFormat)/8;
918 switch(c->dstFormat){
925 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
939 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
952 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
954 #if !HAVE_MMX || defined (RUNTIME_CPUDETECT) || !CONFIG_GPL
959 #if (HAVE_ALTIVEC || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
960 #define COMPILE_ALTIVEC
961 #endif //HAVE_ALTIVEC
966 #if ((HAVE_MMX && !HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
970 #if (HAVE_MMX2 || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
974 #if ((HAVE_3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
975 #define COMPILE_3DNOW
977 #endif //ARCH_X86 || ARCH_X86_64
986 #define HAVE_ALTIVEC 0
989 #define RENAME(a) a ## _C
990 #include "swscale_template.c"
993 #ifdef COMPILE_ALTIVEC
996 #define HAVE_ALTIVEC 1
997 #define RENAME(a) a ## _altivec
998 #include "swscale_template.c"
1010 #define RENAME(a) a ## _X86
1011 #include "swscale_template.c"
1021 #define HAVE_3DNOW 0
1022 #define RENAME(a) a ## _MMX
1023 #include "swscale_template.c"
1034 #define HAVE_3DNOW 0
1035 #define RENAME(a) a ## _MMX2
1036 #include "swscale_template.c"
1040 #ifdef COMPILE_3DNOW
1047 #define HAVE_3DNOW 1
1048 #define RENAME(a) a ## _3DNow
1049 #include "swscale_template.c"
1052 #endif //ARCH_X86 || ARCH_X86_64
1054 // minor note: the HAVE_xyz is messed up after that line so don't use it
1056 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1058 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1059 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1060 else return getSplineCoeff( 0.0,
1067 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1068 int srcW, int dstW, int filterAlign, int one, int flags,
1069 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1075 int64_t *filter=NULL;
1076 int64_t *filter2=NULL;
1077 const int64_t fone= 1LL<<54;
1080 if (flags & SWS_CPU_CAPS_MMX)
1081 __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1084 // Note the +1 is for the MMXscaler which reads over the end
1085 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1087 if (FFABS(xInc - 0x10000) <10) // unscaled
1091 filter= av_mallocz(dstW*sizeof(*filter)*filterSize);
1093 for (i=0; i<dstW; i++)
1095 filter[i*filterSize]= fone;
1100 else if (flags&SWS_POINT) // lame looking point sampling mode
1105 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1107 xDstInSrc= xInc/2 - 0x8000;
1108 for (i=0; i<dstW; i++)
1110 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1112 (*filterPos)[i]= xx;
1117 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1121 if (flags&SWS_BICUBIC) filterSize= 4;
1122 else if (flags&SWS_X ) filterSize= 4;
1123 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1124 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1126 xDstInSrc= xInc/2 - 0x8000;
1127 for (i=0; i<dstW; i++)
1129 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1132 (*filterPos)[i]= xx;
1133 //Bilinear upscale / linear interpolate / Area averaging
1134 for (j=0; j<filterSize; j++)
1136 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1137 if (coeff<0) coeff=0;
1138 filter[i*filterSize + j]= coeff;
1149 if (flags&SWS_BICUBIC) sizeFactor= 4;
1150 else if (flags&SWS_X) sizeFactor= 8;
1151 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1152 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1153 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1154 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1155 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1156 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1158 sizeFactor= 0; //GCC warning killer
1162 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1163 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1165 if (filterSize > srcW-2) filterSize=srcW-2;
1167 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1169 xDstInSrc= xInc - 0x10000;
1170 for (i=0; i<dstW; i++)
1172 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1174 (*filterPos)[i]= xx;
1175 for (j=0; j<filterSize; j++)
1177 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1183 floatd= d * (1.0/(1<<30));
1185 if (flags & SWS_BICUBIC)
1187 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1188 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1189 int64_t dd = ( d*d)>>30;
1190 int64_t ddd= (dd*d)>>30;
1193 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1194 else if (d < 1LL<<31)
1195 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1198 coeff *= fone>>(30+24);
1200 /* else if (flags & SWS_X)
1202 double p= param ? param*0.01 : 0.3;
1203 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1204 coeff*= pow(2.0, - p*d*d);
1206 else if (flags & SWS_X)
1208 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1215 if (c<0.0) c= -pow(-c, A);
1217 coeff= (c*0.5 + 0.5)*fone;
1219 else if (flags & SWS_AREA)
1221 int64_t d2= d - (1<<29);
1222 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1223 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1225 coeff *= fone>>(30+16);
1227 else if (flags & SWS_GAUSS)
1229 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1230 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1232 else if (flags & SWS_SINC)
1234 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1236 else if (flags & SWS_LANCZOS)
1238 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1239 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1240 if (floatd>p) coeff=0;
1242 else if (flags & SWS_BILINEAR)
1245 if (coeff<0) coeff=0;
1246 coeff *= fone >> 30;
1248 else if (flags & SWS_SPLINE)
1250 double p=-2.196152422706632;
1251 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1254 coeff= 0.0; //GCC warning killer
1258 filter[i*filterSize + j]= coeff;
1265 /* apply src & dst Filter to filter -> filter2
1268 assert(filterSize>0);
1269 filter2Size= filterSize;
1270 if (srcFilter) filter2Size+= srcFilter->length - 1;
1271 if (dstFilter) filter2Size+= dstFilter->length - 1;
1272 assert(filter2Size>0);
1273 filter2= av_mallocz(filter2Size*dstW*sizeof(*filter2));
1275 for (i=0; i<dstW; i++)
1280 for (k=0; k<srcFilter->length; k++){
1281 for (j=0; j<filterSize; j++)
1282 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1285 for (j=0; j<filterSize; j++)
1286 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1290 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1294 /* try to reduce the filter-size (step1 find size and shift left) */
1295 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1297 for (i=dstW-1; i>=0; i--)
1299 int min= filter2Size;
1303 /* get rid off near zero elements on the left by shifting left */
1304 for (j=0; j<filter2Size; j++)
1307 cutOff += FFABS(filter2[i*filter2Size]);
1309 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1311 /* preserve monotonicity because the core can't handle the filter otherwise */
1312 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1314 // Move filter coeffs left
1315 for (k=1; k<filter2Size; k++)
1316 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1317 filter2[i*filter2Size + k - 1]= 0;
1322 /* count near zeros on the right */
1323 for (j=filter2Size-1; j>0; j--)
1325 cutOff += FFABS(filter2[i*filter2Size + j]);
1327 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1331 if (min>minFilterSize) minFilterSize= min;
1334 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1335 // we can handle the special case 4,
1336 // so we don't want to go to the full 8
1337 if (minFilterSize < 5)
1340 // we really don't want to waste our time
1341 // doing useless computation, so fall-back on
1342 // the scalar C code for very small filter.
1343 // vectorizing is worth it only if you have
1344 // decent-sized vector.
1345 if (minFilterSize < 3)
1349 if (flags & SWS_CPU_CAPS_MMX) {
1350 // special case for unscaled vertical filtering
1351 if (minFilterSize == 1 && filterAlign == 2)
1355 assert(minFilterSize > 0);
1356 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1357 assert(filterSize > 0);
1358 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1359 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1361 *outFilterSize= filterSize;
1363 if (flags&SWS_PRINT_INFO)
1364 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1365 /* try to reduce the filter-size (step2 reduce it) */
1366 for (i=0; i<dstW; i++)
1370 for (j=0; j<filterSize; j++)
1372 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1373 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1374 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1375 filter[i*filterSize + j]= 0;
1380 //FIXME try to align filterpos if possible
1383 for (i=0; i<dstW; i++)
1386 if ((*filterPos)[i] < 0)
1388 // Move filter coeffs left to compensate for filterPos
1389 for (j=1; j<filterSize; j++)
1391 int left= FFMAX(j + (*filterPos)[i], 0);
1392 filter[i*filterSize + left] += filter[i*filterSize + j];
1393 filter[i*filterSize + j]=0;
1398 if ((*filterPos)[i] + filterSize > srcW)
1400 int shift= (*filterPos)[i] + filterSize - srcW;
1401 // Move filter coeffs right to compensate for filterPos
1402 for (j=filterSize-2; j>=0; j--)
1404 int right= FFMIN(j + shift, filterSize-1);
1405 filter[i*filterSize +right] += filter[i*filterSize +j];
1406 filter[i*filterSize +j]=0;
1408 (*filterPos)[i]= srcW - filterSize;
1412 // Note the +1 is for the MMXscaler which reads over the end
1413 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1414 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1416 /* Normalize & Store in outFilter */
1417 for (i=0; i<dstW; i++)
1423 for (j=0; j<filterSize; j++)
1425 sum+= filter[i*filterSize + j];
1427 sum= (sum + one/2)/ one;
1428 for (j=0; j<*outFilterSize; j++)
1430 int64_t v= filter[i*filterSize + j] + error;
1431 int intV= ROUNDED_DIV(v, sum);
1432 (*outFilter)[i*(*outFilterSize) + j]= intV;
1433 error= v - intV*sum;
1437 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1438 for (i=0; i<*outFilterSize; i++)
1440 int j= dstW*(*outFilterSize);
1441 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1452 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1455 long imm8OfPShufW1A;
1456 long imm8OfPShufW2A;
1457 long fragmentLengthA;
1459 long imm8OfPShufW1B;
1460 long imm8OfPShufW2B;
1461 long fragmentLengthB;
1466 // create an optimized horizontal scaling routine
1474 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1475 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1476 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1477 "punpcklbw %%mm7, %%mm1 \n\t"
1478 "punpcklbw %%mm7, %%mm0 \n\t"
1479 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1481 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1483 "psubw %%mm1, %%mm0 \n\t"
1484 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1485 "pmullw %%mm3, %%mm0 \n\t"
1486 "psllw $7, %%mm1 \n\t"
1487 "paddw %%mm1, %%mm0 \n\t"
1489 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1491 "add $8, %%"REG_a" \n\t"
1495 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1496 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1497 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1502 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1506 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1507 "=r" (fragmentLengthA)
1514 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1515 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1516 "punpcklbw %%mm7, %%mm0 \n\t"
1517 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1519 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1521 "psubw %%mm1, %%mm0 \n\t"
1522 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1523 "pmullw %%mm3, %%mm0 \n\t"
1524 "psllw $7, %%mm1 \n\t"
1525 "paddw %%mm1, %%mm0 \n\t"
1527 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1529 "add $8, %%"REG_a" \n\t"
1533 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1534 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1535 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1540 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1544 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1545 "=r" (fragmentLengthB)
1548 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1551 for (i=0; i<dstW/numSplits; i++)
1558 int b=((xpos+xInc)>>16) - xx;
1559 int c=((xpos+xInc*2)>>16) - xx;
1560 int d=((xpos+xInc*3)>>16) - xx;
1562 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1563 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1564 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1565 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1570 int maxShift= 3-(d+1);
1573 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1575 funnyCode[fragmentPos + imm8OfPShufW1B]=
1576 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1577 funnyCode[fragmentPos + imm8OfPShufW2B]=
1578 a | (b<<2) | (c<<4) | (d<<6);
1580 if (i+3>=dstW) shift=maxShift; //avoid overread
1581 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1583 if (shift && i>=shift)
1585 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1586 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1587 filterPos[i/2]-=shift;
1590 fragmentPos+= fragmentLengthB;
1597 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1599 funnyCode[fragmentPos + imm8OfPShufW1A]=
1600 funnyCode[fragmentPos + imm8OfPShufW2A]=
1601 a | (b<<2) | (c<<4) | (d<<6);
1603 if (i+4>=dstW) shift=maxShift; //avoid overread
1604 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1606 if (shift && i>=shift)
1608 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1609 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1610 filterPos[i/2]-=shift;
1613 fragmentPos+= fragmentLengthA;
1616 funnyCode[fragmentPos]= RET;
1620 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1622 #endif /* COMPILE_MMX2 */
1624 static void globalInit(void){
1625 // generating tables:
1627 for (i=0; i<768; i++){
1628 int c= av_clip_uint8(i-256);
1633 static SwsFunc getSwsFunc(int flags){
1635 #if defined(RUNTIME_CPUDETECT) && CONFIG_GPL
1637 // ordered per speed fastest first
1638 if (flags & SWS_CPU_CAPS_MMX2)
1639 return swScale_MMX2;
1640 else if (flags & SWS_CPU_CAPS_3DNOW)
1641 return swScale_3DNow;
1642 else if (flags & SWS_CPU_CAPS_MMX)
1649 if (flags & SWS_CPU_CAPS_ALTIVEC)
1650 return swScale_altivec;
1655 #endif /* ARCH_X86 */
1656 #else //RUNTIME_CPUDETECT
1658 return swScale_MMX2;
1660 return swScale_3DNow;
1664 return swScale_altivec;
1668 #endif //!RUNTIME_CPUDETECT
1671 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1672 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1673 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1675 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1676 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1680 uint8_t *srcPtr= src[0];
1681 uint8_t *dstPtr= dst;
1682 for (i=0; i<srcSliceH; i++)
1684 memcpy(dstPtr, srcPtr, c->srcW);
1685 srcPtr+= srcStride[0];
1686 dstPtr+= dstStride[0];
1689 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1690 if (c->dstFormat == PIX_FMT_NV12)
1691 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1693 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1698 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1699 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1700 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1702 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1707 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1708 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1709 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1711 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1716 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1717 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1718 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1720 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1725 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1726 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1727 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1729 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1734 static int pal2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1735 int srcSliceH, uint8_t* dst[], int dstStride[]){
1736 const enum PixelFormat srcFormat= c->srcFormat;
1737 const enum PixelFormat dstFormat= c->dstFormat;
1738 void (*conv)(const uint8_t *src, uint8_t *dst, long num_pixels,
1739 const uint8_t *palette)=NULL;
1741 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1742 uint8_t *srcPtr= src[0];
1744 if (!usePal(srcFormat))
1745 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1746 sws_format_name(srcFormat), sws_format_name(dstFormat));
1749 case PIX_FMT_RGB32 : conv = palette8topacked32; break;
1750 case PIX_FMT_BGR32 : conv = palette8topacked32; break;
1751 case PIX_FMT_BGR32_1: conv = palette8topacked32; break;
1752 case PIX_FMT_RGB32_1: conv = palette8topacked32; break;
1753 case PIX_FMT_RGB24 : conv = palette8topacked24; break;
1754 case PIX_FMT_BGR24 : conv = palette8topacked24; break;
1755 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1756 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1760 for (i=0; i<srcSliceH; i++) {
1761 conv(srcPtr, dstPtr, c->srcW, (uint8_t *) c->pal_rgb);
1762 srcPtr+= srcStride[0];
1763 dstPtr+= dstStride[0];
1769 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1770 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1771 int srcSliceH, uint8_t* dst[], int dstStride[]){
1772 const enum PixelFormat srcFormat= c->srcFormat;
1773 const enum PixelFormat dstFormat= c->dstFormat;
1774 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1775 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1776 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1777 const int dstId= fmt_depth(dstFormat) >> 2;
1778 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1781 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1782 || (isRGB(srcFormat) && isRGB(dstFormat))){
1783 switch(srcId | (dstId<<4)){
1784 case 0x34: conv= rgb16to15; break;
1785 case 0x36: conv= rgb24to15; break;
1786 case 0x38: conv= rgb32to15; break;
1787 case 0x43: conv= rgb15to16; break;
1788 case 0x46: conv= rgb24to16; break;
1789 case 0x48: conv= rgb32to16; break;
1790 case 0x63: conv= rgb15to24; break;
1791 case 0x64: conv= rgb16to24; break;
1792 case 0x68: conv= rgb32to24; break;
1793 case 0x83: conv= rgb15to32; break;
1794 case 0x84: conv= rgb16to32; break;
1795 case 0x86: conv= rgb24to32; break;
1796 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1797 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1799 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1800 || (isRGB(srcFormat) && isBGR(dstFormat))){
1801 switch(srcId | (dstId<<4)){
1802 case 0x33: conv= rgb15tobgr15; break;
1803 case 0x34: conv= rgb16tobgr15; break;
1804 case 0x36: conv= rgb24tobgr15; break;
1805 case 0x38: conv= rgb32tobgr15; break;
1806 case 0x43: conv= rgb15tobgr16; break;
1807 case 0x44: conv= rgb16tobgr16; break;
1808 case 0x46: conv= rgb24tobgr16; break;
1809 case 0x48: conv= rgb32tobgr16; break;
1810 case 0x63: conv= rgb15tobgr24; break;
1811 case 0x64: conv= rgb16tobgr24; break;
1812 case 0x66: conv= rgb24tobgr24; break;
1813 case 0x68: conv= rgb32tobgr24; break;
1814 case 0x83: conv= rgb15tobgr32; break;
1815 case 0x84: conv= rgb16tobgr32; break;
1816 case 0x86: conv= rgb24tobgr32; break;
1817 case 0x88: conv= rgb32tobgr32; break;
1818 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1819 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1822 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1823 sws_format_name(srcFormat), sws_format_name(dstFormat));
1828 uint8_t *srcPtr= src[0];
1829 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1830 srcPtr += ALT32_CORR;
1832 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1833 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1837 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1839 for (i=0; i<srcSliceH; i++)
1841 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1842 srcPtr+= srcStride[0];
1843 dstPtr+= dstStride[0];
1850 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1851 int srcSliceH, uint8_t* dst[], int dstStride[]){
1855 dst[0]+ srcSliceY *dstStride[0],
1856 dst[1]+(srcSliceY>>1)*dstStride[1],
1857 dst[2]+(srcSliceY>>1)*dstStride[2],
1859 dstStride[0], dstStride[1], srcStride[0]);
1863 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1864 int srcSliceH, uint8_t* dst[], int dstStride[]){
1868 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1869 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1871 uint8_t *srcPtr= src[0];
1872 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1874 for (i=0; i<srcSliceH; i++)
1876 memcpy(dstPtr, srcPtr, c->srcW);
1877 srcPtr+= srcStride[0];
1878 dstPtr+= dstStride[0];
1882 if (c->dstFormat==PIX_FMT_YUV420P){
1883 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1884 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1886 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1887 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1892 /* unscaled copy like stuff (assumes nearly identical formats) */
1893 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1894 int srcSliceH, uint8_t* dst[], int dstStride[])
1896 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1897 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1901 uint8_t *srcPtr= src[0];
1902 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1905 /* universal length finder */
1906 while(length+c->srcW <= FFABS(dstStride[0])
1907 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1910 for (i=0; i<srcSliceH; i++)
1912 memcpy(dstPtr, srcPtr, length);
1913 srcPtr+= srcStride[0];
1914 dstPtr+= dstStride[0];
1920 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1921 int srcSliceH, uint8_t* dst[], int dstStride[])
1924 for (plane=0; plane<3; plane++)
1926 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1927 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1928 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1930 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1932 if (!isGray(c->dstFormat))
1933 memset(dst[plane], 128, dstStride[plane]*height);
1937 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1938 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1942 uint8_t *srcPtr= src[plane];
1943 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1944 for (i=0; i<height; i++)
1946 memcpy(dstPtr, srcPtr, length);
1947 srcPtr+= srcStride[plane];
1948 dstPtr+= dstStride[plane];
1956 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1957 int srcSliceH, uint8_t* dst[], int dstStride[]){
1959 int length= c->srcW;
1961 int height= srcSliceH;
1963 uint8_t *srcPtr= src[0];
1964 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1966 if (!isGray(c->dstFormat)){
1967 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1968 memset(dst[1], 128, dstStride[1]*height);
1969 memset(dst[2], 128, dstStride[2]*height);
1971 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1972 for (i=0; i<height; i++)
1974 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1975 srcPtr+= srcStride[0];
1976 dstPtr+= dstStride[0];
1981 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1982 int srcSliceH, uint8_t* dst[], int dstStride[]){
1984 int length= c->srcW;
1986 int height= srcSliceH;
1988 uint8_t *srcPtr= src[0];
1989 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1990 for (i=0; i<height; i++)
1992 for (j=0; j<length; j++)
1994 dstPtr[j<<1] = srcPtr[j];
1995 dstPtr[(j<<1)+1] = srcPtr[j];
1997 srcPtr+= srcStride[0];
1998 dstPtr+= dstStride[0];
2003 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2004 int srcSliceH, uint8_t* dst[], int dstStride[]){
2006 int length= c->srcW;
2008 int height= srcSliceH;
2010 uint16_t *srcPtr= (uint16_t*)src[0];
2011 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
2012 for (i=0; i<height; i++)
2014 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
2015 srcPtr+= srcStride[0]/2;
2016 dstPtr+= dstStride[0]/2;
2022 static void getSubSampleFactors(int *h, int *v, int format){
2024 case PIX_FMT_UYVY422:
2025 case PIX_FMT_YUYV422:
2029 case PIX_FMT_YUV420P:
2030 case PIX_FMT_YUVA420P:
2031 case PIX_FMT_GRAY16BE:
2032 case PIX_FMT_GRAY16LE:
2033 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
2039 case PIX_FMT_YUV440P:
2043 case PIX_FMT_YUV410P:
2047 case PIX_FMT_YUV444P:
2051 case PIX_FMT_YUV422P:
2055 case PIX_FMT_YUV411P:
2066 static uint16_t roundToInt16(int64_t f){
2067 int r= (f + (1<<15))>>16;
2068 if (r<-0x7FFF) return 0x8000;
2069 else if (r> 0x7FFF) return 0x7FFF;
2074 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
2075 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2076 * @return -1 if not supported
2078 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2079 int64_t crv = inv_table[0];
2080 int64_t cbu = inv_table[1];
2081 int64_t cgu = -inv_table[2];
2082 int64_t cgv = -inv_table[3];
2086 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2087 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2089 c->brightness= brightness;
2090 c->contrast = contrast;
2091 c->saturation= saturation;
2092 c->srcRange = srcRange;
2093 c->dstRange = dstRange;
2094 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2096 c->uOffset= 0x0400040004000400LL;
2097 c->vOffset= 0x0400040004000400LL;
2103 crv= (crv*224) / 255;
2104 cbu= (cbu*224) / 255;
2105 cgu= (cgu*224) / 255;
2106 cgv= (cgv*224) / 255;
2109 cy = (cy *contrast )>>16;
2110 crv= (crv*contrast * saturation)>>32;
2111 cbu= (cbu*contrast * saturation)>>32;
2112 cgu= (cgu*contrast * saturation)>>32;
2113 cgv= (cgv*contrast * saturation)>>32;
2115 oy -= 256*brightness;
2117 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2118 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2119 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2120 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2121 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2122 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2124 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2125 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2126 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2127 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2128 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2129 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2131 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2134 #ifdef COMPILE_ALTIVEC
2135 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2136 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2142 * @return -1 if not supported
2144 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2145 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2147 *inv_table = c->srcColorspaceTable;
2148 *table = c->dstColorspaceTable;
2149 *srcRange = c->srcRange;
2150 *dstRange = c->dstRange;
2151 *brightness= c->brightness;
2152 *contrast = c->contrast;
2153 *saturation= c->saturation;
2158 static int handle_jpeg(enum PixelFormat *format)
2161 case PIX_FMT_YUVJ420P:
2162 *format = PIX_FMT_YUV420P;
2164 case PIX_FMT_YUVJ422P:
2165 *format = PIX_FMT_YUV422P;
2167 case PIX_FMT_YUVJ444P:
2168 *format = PIX_FMT_YUV444P;
2170 case PIX_FMT_YUVJ440P:
2171 *format = PIX_FMT_YUV440P;
2178 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, int dstW, int dstH, enum PixelFormat dstFormat, int flags,
2179 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2183 int usesVFilter, usesHFilter;
2184 int unscaled, needsDither;
2185 int srcRange, dstRange;
2186 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2188 if (flags & SWS_CPU_CAPS_MMX)
2189 __asm__ volatile("emms\n\t"::: "memory");
2192 #if !defined(RUNTIME_CPUDETECT) || !CONFIG_GPL //ensure that the flags match the compiled variant if cpudetect is off
2193 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2195 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2197 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2199 flags |= SWS_CPU_CAPS_MMX;
2201 flags |= SWS_CPU_CAPS_ALTIVEC;
2203 flags |= SWS_CPU_CAPS_BFIN;
2205 #endif /* RUNTIME_CPUDETECT */
2206 if (clip_table[512] != 255) globalInit();
2207 if (!rgb15to16) sws_rgb2rgb_init(flags);
2209 unscaled = (srcW == dstW && srcH == dstH);
2210 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2211 && (fmt_depth(dstFormat))<24
2212 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2214 srcRange = handle_jpeg(&srcFormat);
2215 dstRange = handle_jpeg(&dstFormat);
2217 if (!isSupportedIn(srcFormat))
2219 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2222 if (!isSupportedOut(dstFormat))
2224 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2228 i= flags & ( SWS_POINT
2239 if(!i || (i & (i-1)))
2241 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2246 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
2248 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2249 srcW, srcH, dstW, dstH);
2252 if(srcW > VOFW || dstW > VOFW){
2253 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2257 if (!dstFilter) dstFilter= &dummyFilter;
2258 if (!srcFilter) srcFilter= &dummyFilter;
2260 c= av_mallocz(sizeof(SwsContext));
2262 c->av_class = &sws_context_class;
2267 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2268 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2270 c->dstFormat= dstFormat;
2271 c->srcFormat= srcFormat;
2272 c->vRounder= 4* 0x0001000100010001ULL;
2274 usesHFilter= usesVFilter= 0;
2275 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2276 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2277 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2278 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2279 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2280 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2281 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2282 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2284 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2285 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2287 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2288 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2290 // drop some chroma lines if the user wants it
2291 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2292 c->chrSrcVSubSample+= c->vChrDrop;
2294 // drop every 2. pixel for chroma calculation unless user wants full chroma
2295 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2296 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2297 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2298 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2299 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2300 c->chrSrcHSubSample=1;
2303 c->param[0] = param[0];
2304 c->param[1] = param[1];
2307 c->param[1] = SWS_PARAM_DEFAULT;
2310 c->chrIntHSubSample= c->chrDstHSubSample;
2311 c->chrIntVSubSample= c->chrSrcVSubSample;
2313 // Note the -((-x)>>y) is so that we always round toward +inf.
2314 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2315 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2316 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2317 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2319 sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2321 /* unscaled special Cases */
2322 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2325 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2327 c->swScale= PlanarToNV12Wrapper;
2331 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat))
2332 && !(flags & SWS_ACCURATE_RND) && !(dstH&1))
2334 c->swScale= yuv2rgb_get_func_ptr(c);
2338 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_BITEXACT))
2340 c->swScale= yvu9toyv12Wrapper;
2344 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_ACCURATE_RND))
2345 c->swScale= bgr24toyv12Wrapper;
2347 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2348 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2349 && (isBGR(dstFormat) || isRGB(dstFormat))
2350 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2351 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2352 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2353 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2354 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2355 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2356 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2357 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2358 && dstFormat != PIX_FMT_RGB32_1
2359 && dstFormat != PIX_FMT_BGR32_1
2360 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2361 c->swScale= rgb2rgbWrapper;
2363 if ((usePal(srcFormat) && (
2364 dstFormat == PIX_FMT_RGB32 ||
2365 dstFormat == PIX_FMT_RGB32_1 ||
2366 dstFormat == PIX_FMT_RGB24 ||
2367 dstFormat == PIX_FMT_BGR32 ||
2368 dstFormat == PIX_FMT_BGR32_1 ||
2369 dstFormat == PIX_FMT_BGR24)))
2370 c->swScale= pal2rgbWrapper;
2372 if (srcFormat == PIX_FMT_YUV422P)
2374 if (dstFormat == PIX_FMT_YUYV422)
2375 c->swScale= YUV422PToYuy2Wrapper;
2376 else if (dstFormat == PIX_FMT_UYVY422)
2377 c->swScale= YUV422PToUyvyWrapper;
2380 /* LQ converters if -sws 0 or -sws 4*/
2381 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2383 if (srcFormat == PIX_FMT_YUV420P)
2385 if (dstFormat == PIX_FMT_YUYV422)
2386 c->swScale= PlanarToYuy2Wrapper;
2387 else if (dstFormat == PIX_FMT_UYVY422)
2388 c->swScale= PlanarToUyvyWrapper;
2392 #ifdef COMPILE_ALTIVEC
2393 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2394 srcFormat == PIX_FMT_YUV420P) {
2395 // unscaled YV12 -> packed YUV, we want speed
2396 if (dstFormat == PIX_FMT_YUYV422)
2397 c->swScale= yv12toyuy2_unscaled_altivec;
2398 else if (dstFormat == PIX_FMT_UYVY422)
2399 c->swScale= yv12touyvy_unscaled_altivec;
2404 if ( srcFormat == dstFormat
2405 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2406 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2408 if (isPacked(c->srcFormat))
2409 c->swScale= packedCopy;
2410 else /* Planar YUV or gray */
2411 c->swScale= planarCopy;
2414 /* gray16{le,be} conversions */
2415 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2417 c->swScale= gray16togray;
2419 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2421 c->swScale= graytogray16;
2423 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2425 c->swScale= gray16swap;
2429 if (flags & SWS_CPU_CAPS_BFIN)
2430 ff_bfin_get_unscaled_swscale (c);
2434 if (flags&SWS_PRINT_INFO)
2435 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2436 sws_format_name(srcFormat), sws_format_name(dstFormat));
2441 if (flags & SWS_CPU_CAPS_MMX2)
2443 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2444 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2446 if (flags&SWS_PRINT_INFO)
2447 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2449 if (usesHFilter) c->canMMX2BeUsed=0;
2454 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2455 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2457 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2458 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2459 // n-2 is the last chrominance sample available
2460 // this is not perfect, but no one should notice the difference, the more correct variant
2461 // would be like the vertical one, but that would require some special code for the
2462 // first and last pixel
2463 if (flags&SWS_FAST_BILINEAR)
2465 if (c->canMMX2BeUsed)
2470 //we don't use the x86asm scaler if mmx is available
2471 else if (flags & SWS_CPU_CAPS_MMX)
2473 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2474 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2478 /* precalculate horizontal scaler filter coefficients */
2480 const int filterAlign=
2481 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2482 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2485 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2486 srcW , dstW, filterAlign, 1<<14,
2487 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2488 srcFilter->lumH, dstFilter->lumH, c->param);
2489 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2490 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2491 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2492 srcFilter->chrH, dstFilter->chrH, c->param);
2494 #define MAX_FUNNY_CODE_SIZE 10000
2495 #if defined(COMPILE_MMX2)
2496 // can't downscale !!!
2497 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2499 #ifdef MAP_ANONYMOUS
2500 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2501 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2503 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2504 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2507 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2508 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2509 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2510 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2512 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2513 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2515 #endif /* defined(COMPILE_MMX2) */
2516 } // Init Horizontal stuff
2520 /* precalculate vertical scaler filter coefficients */
2522 const int filterAlign=
2523 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2524 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2527 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2528 srcH , dstH, filterAlign, (1<<12),
2529 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2530 srcFilter->lumV, dstFilter->lumV, c->param);
2531 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2532 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2533 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2534 srcFilter->chrV, dstFilter->chrV, c->param);
2537 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2538 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2540 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2542 short *p = (short *)&c->vYCoeffsBank[i];
2544 p[j] = c->vLumFilter[i];
2547 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2549 short *p = (short *)&c->vCCoeffsBank[i];
2551 p[j] = c->vChrFilter[i];
2556 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2557 c->vLumBufSize= c->vLumFilterSize;
2558 c->vChrBufSize= c->vChrFilterSize;
2559 for (i=0; i<dstH; i++)
2561 int chrI= i*c->chrDstH / dstH;
2562 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2563 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2565 nextSlice>>= c->chrSrcVSubSample;
2566 nextSlice<<= c->chrSrcVSubSample;
2567 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2568 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2569 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2570 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2573 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2574 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2575 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2576 //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)
2577 /* align at 16 bytes for AltiVec */
2578 for (i=0; i<c->vLumBufSize; i++)
2579 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2580 for (i=0; i<c->vChrBufSize; i++)
2581 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2583 //try to avoid drawing green stuff between the right end and the stride end
2584 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2586 assert(2*VOFW == VOF);
2588 assert(c->chrDstH <= dstH);
2590 if (flags&SWS_PRINT_INFO)
2593 const char *dither= " dithered";
2595 const char *dither= "";
2597 if (flags&SWS_FAST_BILINEAR)
2598 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2599 else if (flags&SWS_BILINEAR)
2600 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2601 else if (flags&SWS_BICUBIC)
2602 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2603 else if (flags&SWS_X)
2604 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2605 else if (flags&SWS_POINT)
2606 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2607 else if (flags&SWS_AREA)
2608 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2609 else if (flags&SWS_BICUBLIN)
2610 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2611 else if (flags&SWS_GAUSS)
2612 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2613 else if (flags&SWS_SINC)
2614 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2615 else if (flags&SWS_LANCZOS)
2616 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2617 else if (flags&SWS_SPLINE)
2618 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2620 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2622 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2623 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2624 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2626 av_log(c, AV_LOG_INFO, "from %s to %s ",
2627 sws_format_name(srcFormat), sws_format_name(dstFormat));
2629 if (flags & SWS_CPU_CAPS_MMX2)
2630 av_log(c, AV_LOG_INFO, "using MMX2\n");
2631 else if (flags & SWS_CPU_CAPS_3DNOW)
2632 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2633 else if (flags & SWS_CPU_CAPS_MMX)
2634 av_log(c, AV_LOG_INFO, "using MMX\n");
2635 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2636 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2638 av_log(c, AV_LOG_INFO, "using C\n");
2641 if (flags & SWS_PRINT_INFO)
2643 if (flags & SWS_CPU_CAPS_MMX)
2645 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2646 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2649 if (c->hLumFilterSize==4)
2650 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2651 else if (c->hLumFilterSize==8)
2652 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2654 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2656 if (c->hChrFilterSize==4)
2657 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2658 else if (c->hChrFilterSize==8)
2659 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2661 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2667 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2669 if (flags & SWS_FAST_BILINEAR)
2670 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2672 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2675 if (isPlanarYUV(dstFormat))
2677 if (c->vLumFilterSize==1)
2678 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2680 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2684 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2685 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2686 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2687 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2688 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2690 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2693 if (dstFormat==PIX_FMT_BGR24)
2694 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2695 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2696 else if (dstFormat==PIX_FMT_RGB32)
2697 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2698 else if (dstFormat==PIX_FMT_BGR565)
2699 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2700 else if (dstFormat==PIX_FMT_BGR555)
2701 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2703 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2705 if (flags & SWS_PRINT_INFO)
2707 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2708 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2709 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2710 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2713 c->swScale= getSwsFunc(flags);
2718 * swscale wrapper, so we don't need to export the SwsContext.
2719 * assumes planar YUV to be in YUV order instead of YVU
2721 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2722 int srcSliceH, uint8_t* dst[], int dstStride[]){
2724 uint8_t* src2[4]= {src[0], src[1], src[2]};
2726 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2727 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2730 if (c->sliceDir == 0) {
2731 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2734 if (usePal(c->srcFormat)){
2735 for (i=0; i<256; i++){
2736 int p, r, g, b,y,u,v;
2737 if(c->srcFormat == PIX_FMT_PAL8){
2738 p=((uint32_t*)(src[1]))[i];
2742 }else if(c->srcFormat == PIX_FMT_RGB8){
2746 }else if(c->srcFormat == PIX_FMT_BGR8){
2750 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2755 assert(c->srcFormat == PIX_FMT_BGR4_BYTE);
2760 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2761 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2762 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2763 c->pal_yuv[i]= y + (u<<8) + (v<<16);
2766 switch(c->dstFormat) {
2768 #ifndef WORDS_BIGENDIAN
2771 c->pal_rgb[i]= r + (g<<8) + (b<<16);
2773 case PIX_FMT_BGR32_1:
2774 #ifdef WORDS_BIGENDIAN
2777 c->pal_rgb[i]= (r + (g<<8) + (b<<16)) << 8;
2779 case PIX_FMT_RGB32_1:
2780 #ifdef WORDS_BIGENDIAN
2783 c->pal_rgb[i]= (b + (g<<8) + (r<<16)) << 8;
2786 #ifndef WORDS_BIGENDIAN
2790 c->pal_rgb[i]= b + (g<<8) + (r<<16);
2795 // copy strides, so they can safely be modified
2796 if (c->sliceDir == 1) {
2797 // slices go from top to bottom
2798 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2799 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2800 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2802 // slices go from bottom to top => we flip the image internally
2803 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2804 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2805 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2806 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2807 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2809 src2[0] += (srcSliceH-1)*srcStride[0];
2810 if (!usePal(c->srcFormat))
2811 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2812 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2814 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2819 * swscale wrapper, so we don't need to export the SwsContext
2821 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2822 int srcSliceH, uint8_t* dst[], int dstStride[]){
2823 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2826 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2827 float lumaSharpen, float chromaSharpen,
2828 float chromaHShift, float chromaVShift,
2831 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2833 if (lumaGBlur!=0.0){
2834 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2835 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2837 filter->lumH= sws_getIdentityVec();
2838 filter->lumV= sws_getIdentityVec();
2841 if (chromaGBlur!=0.0){
2842 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2843 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2845 filter->chrH= sws_getIdentityVec();
2846 filter->chrV= sws_getIdentityVec();
2849 if (chromaSharpen!=0.0){
2850 SwsVector *id= sws_getIdentityVec();
2851 sws_scaleVec(filter->chrH, -chromaSharpen);
2852 sws_scaleVec(filter->chrV, -chromaSharpen);
2853 sws_addVec(filter->chrH, id);
2854 sws_addVec(filter->chrV, id);
2858 if (lumaSharpen!=0.0){
2859 SwsVector *id= sws_getIdentityVec();
2860 sws_scaleVec(filter->lumH, -lumaSharpen);
2861 sws_scaleVec(filter->lumV, -lumaSharpen);
2862 sws_addVec(filter->lumH, id);
2863 sws_addVec(filter->lumV, id);
2867 if (chromaHShift != 0.0)
2868 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2870 if (chromaVShift != 0.0)
2871 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2873 sws_normalizeVec(filter->chrH, 1.0);
2874 sws_normalizeVec(filter->chrV, 1.0);
2875 sws_normalizeVec(filter->lumH, 1.0);
2876 sws_normalizeVec(filter->lumV, 1.0);
2878 if (verbose) sws_printVec(filter->chrH);
2879 if (verbose) sws_printVec(filter->lumH);
2885 * returns a normalized gaussian curve used to filter stuff
2886 * quality=3 is high quality, lowwer is lowwer quality
2888 SwsVector *sws_getGaussianVec(double variance, double quality){
2889 const int length= (int)(variance*quality + 0.5) | 1;
2891 double *coeff= av_malloc(length*sizeof(double));
2892 double middle= (length-1)*0.5;
2893 SwsVector *vec= av_malloc(sizeof(SwsVector));
2896 vec->length= length;
2898 for (i=0; i<length; i++)
2900 double dist= i-middle;
2901 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2904 sws_normalizeVec(vec, 1.0);
2909 SwsVector *sws_getConstVec(double c, int length){
2911 double *coeff= av_malloc(length*sizeof(double));
2912 SwsVector *vec= av_malloc(sizeof(SwsVector));
2915 vec->length= length;
2917 for (i=0; i<length; i++)
2924 SwsVector *sws_getIdentityVec(void){
2925 return sws_getConstVec(1.0, 1);
2928 double sws_dcVec(SwsVector *a){
2932 for (i=0; i<a->length; i++)
2938 void sws_scaleVec(SwsVector *a, double scalar){
2941 for (i=0; i<a->length; i++)
2942 a->coeff[i]*= scalar;
2945 void sws_normalizeVec(SwsVector *a, double height){
2946 sws_scaleVec(a, height/sws_dcVec(a));
2949 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2950 int length= a->length + b->length - 1;
2951 double *coeff= av_malloc(length*sizeof(double));
2953 SwsVector *vec= av_malloc(sizeof(SwsVector));
2956 vec->length= length;
2958 for (i=0; i<length; i++) coeff[i]= 0.0;
2960 for (i=0; i<a->length; i++)
2962 for (j=0; j<b->length; j++)
2964 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2971 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2972 int length= FFMAX(a->length, b->length);
2973 double *coeff= av_malloc(length*sizeof(double));
2975 SwsVector *vec= av_malloc(sizeof(SwsVector));
2978 vec->length= length;
2980 for (i=0; i<length; i++) coeff[i]= 0.0;
2982 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2983 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2988 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2989 int length= FFMAX(a->length, b->length);
2990 double *coeff= av_malloc(length*sizeof(double));
2992 SwsVector *vec= av_malloc(sizeof(SwsVector));
2995 vec->length= length;
2997 for (i=0; i<length; i++) coeff[i]= 0.0;
2999 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
3000 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
3005 /* shift left / or right if "shift" is negative */
3006 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
3007 int length= a->length + FFABS(shift)*2;
3008 double *coeff= av_malloc(length*sizeof(double));
3010 SwsVector *vec= av_malloc(sizeof(SwsVector));
3013 vec->length= length;
3015 for (i=0; i<length; i++) coeff[i]= 0.0;
3017 for (i=0; i<a->length; i++)
3019 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
3025 void sws_shiftVec(SwsVector *a, int shift){
3026 SwsVector *shifted= sws_getShiftedVec(a, shift);
3028 a->coeff= shifted->coeff;
3029 a->length= shifted->length;
3033 void sws_addVec(SwsVector *a, SwsVector *b){
3034 SwsVector *sum= sws_sumVec(a, b);
3036 a->coeff= sum->coeff;
3037 a->length= sum->length;
3041 void sws_subVec(SwsVector *a, SwsVector *b){
3042 SwsVector *diff= sws_diffVec(a, b);
3044 a->coeff= diff->coeff;
3045 a->length= diff->length;
3049 void sws_convVec(SwsVector *a, SwsVector *b){
3050 SwsVector *conv= sws_getConvVec(a, b);
3052 a->coeff= conv->coeff;
3053 a->length= conv->length;
3057 SwsVector *sws_cloneVec(SwsVector *a){
3058 double *coeff= av_malloc(a->length*sizeof(double));
3060 SwsVector *vec= av_malloc(sizeof(SwsVector));
3063 vec->length= a->length;
3065 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
3070 void sws_printVec(SwsVector *a){
3076 for (i=0; i<a->length; i++)
3077 if (a->coeff[i]>max) max= a->coeff[i];
3079 for (i=0; i<a->length; i++)
3080 if (a->coeff[i]<min) min= a->coeff[i];
3084 for (i=0; i<a->length; i++)
3086 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
3087 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
3088 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
3089 av_log(NULL, AV_LOG_DEBUG, "|\n");
3093 void sws_freeVec(SwsVector *a){
3095 av_freep(&a->coeff);
3100 void sws_freeFilter(SwsFilter *filter){
3101 if (!filter) return;
3103 if (filter->lumH) sws_freeVec(filter->lumH);
3104 if (filter->lumV) sws_freeVec(filter->lumV);
3105 if (filter->chrH) sws_freeVec(filter->chrH);
3106 if (filter->chrV) sws_freeVec(filter->chrV);
3111 void sws_freeContext(SwsContext *c){
3117 for (i=0; i<c->vLumBufSize; i++)
3118 av_freep(&c->lumPixBuf[i]);
3119 av_freep(&c->lumPixBuf);
3124 for (i=0; i<c->vChrBufSize; i++)
3125 av_freep(&c->chrPixBuf[i]);
3126 av_freep(&c->chrPixBuf);
3129 av_freep(&c->vLumFilter);
3130 av_freep(&c->vChrFilter);
3131 av_freep(&c->hLumFilter);
3132 av_freep(&c->hChrFilter);
3134 av_freep(&c->vYCoeffsBank);
3135 av_freep(&c->vCCoeffsBank);
3138 av_freep(&c->vLumFilterPos);
3139 av_freep(&c->vChrFilterPos);
3140 av_freep(&c->hLumFilterPos);
3141 av_freep(&c->hChrFilterPos);
3143 #if ARCH_X86 && CONFIG_GPL
3144 #ifdef MAP_ANONYMOUS
3145 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
3146 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
3148 av_free(c->funnyYCode);
3149 av_free(c->funnyUVCode);
3152 c->funnyUVCode=NULL;
3153 #endif /* ARCH_X86 */
3155 av_freep(&c->lumMmx2Filter);
3156 av_freep(&c->chrMmx2Filter);
3157 av_freep(&c->lumMmx2FilterPos);
3158 av_freep(&c->chrMmx2FilterPos);
3159 av_freep(&c->yuvTable);
3165 * Checks if context is valid or reallocs a new one instead.
3166 * If context is NULL, just calls sws_getContext() to get a new one.
3167 * Otherwise, checks if the parameters are the same already saved in context.
3168 * If that is the case, returns the current context.
3169 * Otherwise, frees context and gets a new one.
3171 * Be warned that srcFilter, dstFilter are not checked, they are
3172 * asumed to remain valid.
3174 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3175 int srcW, int srcH, enum PixelFormat srcFormat,
3176 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
3177 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
3179 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3182 param = default_param;
3185 if (context->srcW != srcW || context->srcH != srcH ||
3186 context->srcFormat != srcFormat ||
3187 context->dstW != dstW || context->dstH != dstH ||
3188 context->dstFormat != dstFormat || context->flags != flags ||
3189 context->param[0] != param[0] || context->param[1] != param[1])
3191 sws_freeContext(context);
3196 return sws_getContext(srcW, srcH, srcFormat,
3197 dstW, dstH, dstFormat, flags,
3198 srcFilter, dstFilter, param);