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
65 #ifdef HAVE_SYS_MMAN_H
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 \
159 #define RGB2YUV_SHIFT 15
160 #define BY ( (int)(0.114*219/255*(1<<RGB2YUV_SHIFT)+0.5))
161 #define BV (-(int)(0.081*224/255*(1<<RGB2YUV_SHIFT)+0.5))
162 #define BU ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
163 #define GY ( (int)(0.587*219/255*(1<<RGB2YUV_SHIFT)+0.5))
164 #define GV (-(int)(0.419*224/255*(1<<RGB2YUV_SHIFT)+0.5))
165 #define GU (-(int)(0.331*224/255*(1<<RGB2YUV_SHIFT)+0.5))
166 #define RY ( (int)(0.299*219/255*(1<<RGB2YUV_SHIFT)+0.5))
167 #define RV ( (int)(0.500*224/255*(1<<RGB2YUV_SHIFT)+0.5))
168 #define RU (-(int)(0.169*224/255*(1<<RGB2YUV_SHIFT)+0.5))
170 extern const int32_t Inverse_Table_6_9[8][4];
172 static const double rgb2yuv_table[8][9]={
173 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
174 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
175 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
176 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
177 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
178 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
179 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
180 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
185 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
188 more intelligent misalignment avoidance for the horizontal scaler
189 write special vertical cubic upscale version
190 Optimize C code (yv12 / minmax)
191 add support for packed pixel yuv input & output
192 add support for Y8 output
193 optimize bgr24 & bgr32
194 add BGR4 output support
195 write special BGR->BGR scaler
198 #if defined(ARCH_X86) && defined (CONFIG_GPL)
199 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
200 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
201 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
202 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
203 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
204 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
205 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
206 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
208 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
209 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
210 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
211 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
213 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
214 0x0103010301030103LL,
215 0x0200020002000200LL,};
217 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
218 0x0602060206020602LL,
219 0x0004000400040004LL,};
221 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
222 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
223 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
224 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
225 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
226 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
228 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
229 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
230 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
233 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
234 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
235 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
237 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
238 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
239 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
240 #endif /* FAST_BGR2YV12 */
241 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
242 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
243 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
245 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
246 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
247 DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
248 DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
249 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
251 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUV[2][4]) = {
252 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
253 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
256 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
258 #endif /* defined(ARCH_X86) */
260 // clipping helper table for C implementations:
261 static unsigned char clip_table[768];
263 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
265 const uint8_t __attribute__((aligned(8))) dither_2x2_4[2][8]={
266 { 1, 3, 1, 3, 1, 3, 1, 3, },
267 { 2, 0, 2, 0, 2, 0, 2, 0, },
270 const uint8_t __attribute__((aligned(8))) dither_2x2_8[2][8]={
271 { 6, 2, 6, 2, 6, 2, 6, 2, },
272 { 0, 4, 0, 4, 0, 4, 0, 4, },
275 const uint8_t __attribute__((aligned(8))) dither_8x8_32[8][8]={
276 { 17, 9, 23, 15, 16, 8, 22, 14, },
277 { 5, 29, 3, 27, 4, 28, 2, 26, },
278 { 21, 13, 19, 11, 20, 12, 18, 10, },
279 { 0, 24, 6, 30, 1, 25, 7, 31, },
280 { 16, 8, 22, 14, 17, 9, 23, 15, },
281 { 4, 28, 2, 26, 5, 29, 3, 27, },
282 { 20, 12, 18, 10, 21, 13, 19, 11, },
283 { 1, 25, 7, 31, 0, 24, 6, 30, },
287 const uint8_t __attribute__((aligned(8))) dither_8x8_64[8][8]={
288 { 0, 48, 12, 60, 3, 51, 15, 63, },
289 { 32, 16, 44, 28, 35, 19, 47, 31, },
290 { 8, 56, 4, 52, 11, 59, 7, 55, },
291 { 40, 24, 36, 20, 43, 27, 39, 23, },
292 { 2, 50, 14, 62, 1, 49, 13, 61, },
293 { 34, 18, 46, 30, 33, 17, 45, 29, },
294 { 10, 58, 6, 54, 9, 57, 5, 53, },
295 { 42, 26, 38, 22, 41, 25, 37, 21, },
299 const uint8_t __attribute__((aligned(8))) dither_8x8_73[8][8]={
300 { 0, 55, 14, 68, 3, 58, 17, 72, },
301 { 37, 18, 50, 32, 40, 22, 54, 35, },
302 { 9, 64, 5, 59, 13, 67, 8, 63, },
303 { 46, 27, 41, 23, 49, 31, 44, 26, },
304 { 2, 57, 16, 71, 1, 56, 15, 70, },
305 { 39, 21, 52, 34, 38, 19, 51, 33, },
306 { 11, 66, 7, 62, 10, 65, 6, 60, },
307 { 48, 30, 43, 25, 47, 29, 42, 24, },
311 const uint8_t __attribute__((aligned(8))) dither_8x8_128[8][8]={
312 { 68, 36, 92, 60, 66, 34, 90, 58, },
313 { 20, 116, 12, 108, 18, 114, 10, 106, },
314 { 84, 52, 76, 44, 82, 50, 74, 42, },
315 { 0, 96, 24, 120, 6, 102, 30, 126, },
316 { 64, 32, 88, 56, 70, 38, 94, 62, },
317 { 16, 112, 8, 104, 22, 118, 14, 110, },
318 { 80, 48, 72, 40, 86, 54, 78, 46, },
319 { 4, 100, 28, 124, 2, 98, 26, 122, },
324 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
325 {117, 62, 158, 103, 113, 58, 155, 100, },
326 { 34, 199, 21, 186, 31, 196, 17, 182, },
327 {144, 89, 131, 76, 141, 86, 127, 72, },
328 { 0, 165, 41, 206, 10, 175, 52, 217, },
329 {110, 55, 151, 96, 120, 65, 162, 107, },
330 { 28, 193, 14, 179, 38, 203, 24, 189, },
331 {138, 83, 124, 69, 148, 93, 134, 79, },
332 { 7, 172, 48, 213, 3, 168, 45, 210, },
335 // tries to correct a gamma of 1.5
336 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
337 { 0, 143, 18, 200, 2, 156, 25, 215, },
338 { 78, 28, 125, 64, 89, 36, 138, 74, },
339 { 10, 180, 3, 161, 16, 195, 8, 175, },
340 {109, 51, 93, 38, 121, 60, 105, 47, },
341 { 1, 152, 23, 210, 0, 147, 20, 205, },
342 { 85, 33, 134, 71, 81, 30, 130, 67, },
343 { 14, 190, 6, 171, 12, 185, 5, 166, },
344 {117, 57, 101, 44, 113, 54, 97, 41, },
347 // tries to correct a gamma of 2.0
348 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
349 { 0, 124, 8, 193, 0, 140, 12, 213, },
350 { 55, 14, 104, 42, 66, 19, 119, 52, },
351 { 3, 168, 1, 145, 6, 187, 3, 162, },
352 { 86, 31, 70, 21, 99, 39, 82, 28, },
353 { 0, 134, 11, 206, 0, 129, 9, 200, },
354 { 62, 17, 114, 48, 58, 16, 109, 45, },
355 { 5, 181, 2, 157, 4, 175, 1, 151, },
356 { 95, 36, 78, 26, 90, 34, 74, 24, },
359 // tries to correct a gamma of 2.5
360 const uint8_t __attribute__((aligned(8))) dither_8x8_220[8][8]={
361 { 0, 107, 3, 187, 0, 125, 6, 212, },
362 { 39, 7, 86, 28, 49, 11, 102, 36, },
363 { 1, 158, 0, 131, 3, 180, 1, 151, },
364 { 68, 19, 52, 12, 81, 25, 64, 17, },
365 { 0, 119, 5, 203, 0, 113, 4, 195, },
366 { 45, 9, 96, 33, 42, 8, 91, 30, },
367 { 2, 172, 1, 144, 2, 165, 0, 137, },
368 { 77, 23, 60, 15, 72, 21, 56, 14, },
372 const char *sws_format_name(enum PixelFormat format)
375 case PIX_FMT_YUV420P:
377 case PIX_FMT_YUVA420P:
379 case PIX_FMT_YUYV422:
385 case PIX_FMT_YUV422P:
387 case PIX_FMT_YUV444P:
391 case PIX_FMT_YUV410P:
393 case PIX_FMT_YUV411P:
399 case PIX_FMT_GRAY16BE:
401 case PIX_FMT_GRAY16LE:
405 case PIX_FMT_MONOWHITE:
407 case PIX_FMT_MONOBLACK:
411 case PIX_FMT_YUVJ420P:
413 case PIX_FMT_YUVJ422P:
415 case PIX_FMT_YUVJ444P:
417 case PIX_FMT_XVMC_MPEG2_MC:
418 return "xvmc_mpeg2_mc";
419 case PIX_FMT_XVMC_MPEG2_IDCT:
420 return "xvmc_mpeg2_idct";
421 case PIX_FMT_UYVY422:
423 case PIX_FMT_UYYVYY411:
425 case PIX_FMT_RGB32_1:
427 case PIX_FMT_BGR32_1:
439 case PIX_FMT_BGR4_BYTE:
445 case PIX_FMT_RGB4_BYTE:
451 case PIX_FMT_YUV440P:
454 return "Unknown format";
458 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
459 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
460 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
462 //FIXME Optimize (just quickly writen not opti..)
464 for (i=0; i<dstW; i++)
468 for (j=0; j<lumFilterSize; j++)
469 val += lumSrc[j][i] * lumFilter[j];
471 dest[i]= av_clip_uint8(val>>19);
475 for (i=0; i<chrDstW; i++)
480 for (j=0; j<chrFilterSize; j++)
482 u += chrSrc[j][i] * chrFilter[j];
483 v += chrSrc[j][i + VOFW] * chrFilter[j];
486 uDest[i]= av_clip_uint8(u>>19);
487 vDest[i]= av_clip_uint8(v>>19);
491 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
492 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
493 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
495 //FIXME Optimize (just quickly writen not opti..)
497 for (i=0; i<dstW; i++)
501 for (j=0; j<lumFilterSize; j++)
502 val += lumSrc[j][i] * lumFilter[j];
504 dest[i]= av_clip_uint8(val>>19);
510 if (dstFormat == PIX_FMT_NV12)
511 for (i=0; i<chrDstW; i++)
516 for (j=0; j<chrFilterSize; j++)
518 u += chrSrc[j][i] * chrFilter[j];
519 v += chrSrc[j][i + VOFW] * chrFilter[j];
522 uDest[2*i]= av_clip_uint8(u>>19);
523 uDest[2*i+1]= av_clip_uint8(v>>19);
526 for (i=0; i<chrDstW; i++)
531 for (j=0; j<chrFilterSize; j++)
533 u += chrSrc[j][i] * chrFilter[j];
534 v += chrSrc[j][i + VOFW] * chrFilter[j];
537 uDest[2*i]= av_clip_uint8(v>>19);
538 uDest[2*i+1]= av_clip_uint8(u>>19);
542 #define YSCALE_YUV_2_PACKEDX_NOCLIP_C(type) \
543 for (i=0; i<(dstW>>1); i++){\
549 type av_unused *r, *b, *g;\
552 for (j=0; j<lumFilterSize; j++)\
554 Y1 += lumSrc[j][i2] * lumFilter[j];\
555 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
557 for (j=0; j<chrFilterSize; j++)\
559 U += chrSrc[j][i] * chrFilter[j];\
560 V += chrSrc[j][i+VOFW] * chrFilter[j];\
567 #define YSCALE_YUV_2_PACKEDX_C(type) \
568 YSCALE_YUV_2_PACKEDX_NOCLIP_C(type)\
569 if ((Y1|Y2|U|V)&256)\
571 if (Y1>255) Y1=255; \
572 else if (Y1<0)Y1=0; \
573 if (Y2>255) Y2=255; \
574 else if (Y2<0)Y2=0; \
581 #define YSCALE_YUV_2_PACKEDX_FULL_C \
582 for (i=0; i<dstW; i++){\
589 for (j=0; j<lumFilterSize; j++){\
590 Y += lumSrc[j][i ] * lumFilter[j];\
592 for (j=0; j<chrFilterSize; j++){\
593 U += chrSrc[j][i ] * chrFilter[j];\
594 V += chrSrc[j][i+VOFW] * chrFilter[j];\
600 #define YSCALE_YUV_2_RGBX_FULL_C(rnd) \
601 YSCALE_YUV_2_PACKEDX_FULL_C\
602 Y-= c->yuv2rgb_y_offset;\
603 Y*= c->yuv2rgb_y_coeff;\
605 R= Y + V*c->yuv2rgb_v2r_coeff;\
606 G= Y + V*c->yuv2rgb_v2g_coeff + U*c->yuv2rgb_u2g_coeff;\
607 B= Y + U*c->yuv2rgb_u2b_coeff;\
608 if ((R|G|B)&(0xC0000000)){\
609 if (R>=(256<<22)) R=(256<<22)-1; \
611 if (G>=(256<<22)) G=(256<<22)-1; \
613 if (B>=(256<<22)) B=(256<<22)-1; \
618 #define YSCALE_YUV_2_GRAY16_C \
619 for (i=0; i<(dstW>>1); i++){\
628 for (j=0; j<lumFilterSize; j++)\
630 Y1 += lumSrc[j][i2] * lumFilter[j];\
631 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
635 if ((Y1|Y2|U|V)&65536)\
637 if (Y1>65535) Y1=65535; \
638 else if (Y1<0)Y1=0; \
639 if (Y2>65535) Y2=65535; \
640 else if (Y2<0)Y2=0; \
643 #define YSCALE_YUV_2_RGBX_C(type) \
644 YSCALE_YUV_2_PACKEDX_C(type) /* FIXME fix tables so that cliping is not needed and then use _NOCLIP*/\
645 r = (type *)c->table_rV[V]; \
646 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
647 b = (type *)c->table_bU[U]; \
649 #define YSCALE_YUV_2_PACKED2_C \
650 for (i=0; i<(dstW>>1); i++){ \
652 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
653 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
654 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
655 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
657 #define YSCALE_YUV_2_GRAY16_2_C \
658 for (i=0; i<(dstW>>1); i++){ \
660 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
661 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
663 #define YSCALE_YUV_2_RGB2_C(type) \
664 YSCALE_YUV_2_PACKED2_C\
666 r = (type *)c->table_rV[V];\
667 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
668 b = (type *)c->table_bU[U];\
670 #define YSCALE_YUV_2_PACKED1_C \
671 for (i=0; i<(dstW>>1); i++){\
673 int Y1= buf0[i2 ]>>7;\
674 int Y2= buf0[i2+1]>>7;\
675 int U= (uvbuf1[i ])>>7;\
676 int V= (uvbuf1[i+VOFW])>>7;\
678 #define YSCALE_YUV_2_GRAY16_1_C \
679 for (i=0; i<(dstW>>1); i++){\
681 int Y1= buf0[i2 ]<<1;\
682 int Y2= buf0[i2+1]<<1;\
684 #define YSCALE_YUV_2_RGB1_C(type) \
685 YSCALE_YUV_2_PACKED1_C\
687 r = (type *)c->table_rV[V];\
688 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
689 b = (type *)c->table_bU[U];\
691 #define YSCALE_YUV_2_PACKED1B_C \
692 for (i=0; i<(dstW>>1); i++){\
694 int Y1= buf0[i2 ]>>7;\
695 int Y2= buf0[i2+1]>>7;\
696 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
697 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
699 #define YSCALE_YUV_2_RGB1B_C(type) \
700 YSCALE_YUV_2_PACKED1B_C\
702 r = (type *)c->table_rV[V];\
703 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
704 b = (type *)c->table_bU[U];\
706 #define YSCALE_YUV_2_MONO2_C \
707 const uint8_t * const d128=dither_8x8_220[y&7];\
708 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
709 for (i=0; i<dstW-7; i+=8){\
711 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
712 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
713 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
714 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
715 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
716 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
717 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
718 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
719 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
724 #define YSCALE_YUV_2_MONOX_C \
725 const uint8_t * const d128=dither_8x8_220[y&7];\
726 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
728 for (i=0; i<dstW-1; i+=2){\
733 for (j=0; j<lumFilterSize; j++)\
735 Y1 += lumSrc[j][i] * lumFilter[j];\
736 Y2 += lumSrc[j][i+1] * lumFilter[j];\
747 acc+= acc + g[Y1+d128[(i+0)&7]];\
748 acc+= acc + g[Y2+d128[(i+1)&7]];\
750 ((uint8_t*)dest)[0]= c->dstFormat == PIX_FMT_MONOBLACK ? acc : ~acc;\
756 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16, func_monoblack)\
757 switch(c->dstFormat)\
761 case PIX_FMT_RGB32_1:\
762 case PIX_FMT_BGR32_1:\
764 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
765 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
770 ((uint8_t*)dest)[0]= r[Y1];\
771 ((uint8_t*)dest)[1]= g[Y1];\
772 ((uint8_t*)dest)[2]= b[Y1];\
773 ((uint8_t*)dest)[3]= r[Y2];\
774 ((uint8_t*)dest)[4]= g[Y2];\
775 ((uint8_t*)dest)[5]= b[Y2];\
781 ((uint8_t*)dest)[0]= b[Y1];\
782 ((uint8_t*)dest)[1]= g[Y1];\
783 ((uint8_t*)dest)[2]= r[Y1];\
784 ((uint8_t*)dest)[3]= b[Y2];\
785 ((uint8_t*)dest)[4]= g[Y2];\
786 ((uint8_t*)dest)[5]= r[Y2];\
790 case PIX_FMT_RGB565:\
791 case PIX_FMT_BGR565:\
793 const int dr1= dither_2x2_8[y&1 ][0];\
794 const int dg1= dither_2x2_4[y&1 ][0];\
795 const int db1= dither_2x2_8[(y&1)^1][0];\
796 const int dr2= dither_2x2_8[y&1 ][1];\
797 const int dg2= dither_2x2_4[y&1 ][1];\
798 const int db2= dither_2x2_8[(y&1)^1][1];\
800 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
801 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
805 case PIX_FMT_RGB555:\
806 case PIX_FMT_BGR555:\
808 const int dr1= dither_2x2_8[y&1 ][0];\
809 const int dg1= dither_2x2_8[y&1 ][1];\
810 const int db1= dither_2x2_8[(y&1)^1][0];\
811 const int dr2= dither_2x2_8[y&1 ][1];\
812 const int dg2= dither_2x2_8[y&1 ][0];\
813 const int db2= dither_2x2_8[(y&1)^1][1];\
815 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
816 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
823 const uint8_t * const d64= dither_8x8_73[y&7];\
824 const uint8_t * const d32= dither_8x8_32[y&7];\
826 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
827 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
834 const uint8_t * const d64= dither_8x8_73 [y&7];\
835 const uint8_t * const d128=dither_8x8_220[y&7];\
837 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
838 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
842 case PIX_FMT_RGB4_BYTE:\
843 case PIX_FMT_BGR4_BYTE:\
845 const uint8_t * const d64= dither_8x8_73 [y&7];\
846 const uint8_t * const d128=dither_8x8_220[y&7];\
848 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
849 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
853 case PIX_FMT_MONOBLACK:\
854 case PIX_FMT_MONOWHITE:\
859 case PIX_FMT_YUYV422:\
861 ((uint8_t*)dest)[2*i2+0]= Y1;\
862 ((uint8_t*)dest)[2*i2+1]= U;\
863 ((uint8_t*)dest)[2*i2+2]= Y2;\
864 ((uint8_t*)dest)[2*i2+3]= V;\
867 case PIX_FMT_UYVY422:\
869 ((uint8_t*)dest)[2*i2+0]= U;\
870 ((uint8_t*)dest)[2*i2+1]= Y1;\
871 ((uint8_t*)dest)[2*i2+2]= V;\
872 ((uint8_t*)dest)[2*i2+3]= Y2;\
875 case PIX_FMT_GRAY16BE:\
877 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
878 ((uint8_t*)dest)[2*i2+1]= Y1;\
879 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
880 ((uint8_t*)dest)[2*i2+3]= Y2;\
883 case PIX_FMT_GRAY16LE:\
885 ((uint8_t*)dest)[2*i2+0]= Y1;\
886 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
887 ((uint8_t*)dest)[2*i2+2]= Y2;\
888 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
894 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
895 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
896 uint8_t *dest, int dstW, int y)
899 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)
902 static inline void yuv2rgbXinC_full(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 int step= fmt_depth(c->dstFormat)/8;
910 switch(c->dstFormat){
917 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
931 YSCALE_YUV_2_RGBX_FULL_C(1<<21)
944 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
946 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
951 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
952 #define COMPILE_ALTIVEC
953 #endif //HAVE_ALTIVEC
954 #endif //ARCH_POWERPC
956 #if defined(ARCH_X86)
958 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
962 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
966 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
967 #define COMPILE_3DNOW
969 #endif //ARCH_X86 || ARCH_X86_64
980 #define RENAME(a) a ## _C
981 #include "swscale_template.c"
984 #ifdef COMPILE_ALTIVEC
987 #define RENAME(a) a ## _altivec
988 #include "swscale_template.c"
991 #if defined(ARCH_X86)
1000 #define RENAME(a) a ## _X86
1001 #include "swscale_template.c"
1009 #define RENAME(a) a ## _MMX
1010 #include "swscale_template.c"
1019 #define RENAME(a) a ## _MMX2
1020 #include "swscale_template.c"
1024 #ifdef COMPILE_3DNOW
1029 #define RENAME(a) a ## _3DNow
1030 #include "swscale_template.c"
1033 #endif //ARCH_X86 || ARCH_X86_64
1035 // minor note: the HAVE_xyz is messed up after that line so don't use it
1037 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1039 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1040 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1041 else return getSplineCoeff( 0.0,
1048 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1049 int srcW, int dstW, int filterAlign, int one, int flags,
1050 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1056 int64_t *filter=NULL;
1057 int64_t *filter2=NULL;
1058 const int64_t fone= 1LL<<54;
1060 #if defined(ARCH_X86)
1061 if (flags & SWS_CPU_CAPS_MMX)
1062 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1065 // Note the +1 is for the MMXscaler which reads over the end
1066 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1068 if (FFABS(xInc - 0x10000) <10) // unscaled
1072 filter= av_mallocz(dstW*sizeof(*filter)*filterSize);
1074 for (i=0; i<dstW; i++)
1076 filter[i*filterSize]= fone;
1081 else if (flags&SWS_POINT) // lame looking point sampling mode
1086 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1088 xDstInSrc= xInc/2 - 0x8000;
1089 for (i=0; i<dstW; i++)
1091 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1093 (*filterPos)[i]= xx;
1098 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1102 if (flags&SWS_BICUBIC) filterSize= 4;
1103 else if (flags&SWS_X ) filterSize= 4;
1104 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
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;
1113 (*filterPos)[i]= xx;
1114 //Bilinear upscale / linear interpolate / Area averaging
1115 for (j=0; j<filterSize; j++)
1117 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
1118 if (coeff<0) coeff=0;
1119 filter[i*filterSize + j]= coeff;
1130 if (flags&SWS_BICUBIC) sizeFactor= 4;
1131 else if (flags&SWS_X) sizeFactor= 8;
1132 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
1133 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
1134 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
1135 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
1136 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
1137 else if (flags&SWS_BILINEAR) sizeFactor= 2;
1139 sizeFactor= 0; //GCC warning killer
1143 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
1144 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
1146 if (filterSize > srcW-2) filterSize=srcW-2;
1148 filter= av_malloc(dstW*sizeof(*filter)*filterSize);
1150 xDstInSrc= xInc - 0x10000;
1151 for (i=0; i<dstW; i++)
1153 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
1155 (*filterPos)[i]= xx;
1156 for (j=0; j<filterSize; j++)
1158 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
1164 floatd= d * (1.0/(1<<30));
1166 if (flags & SWS_BICUBIC)
1168 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
1169 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
1170 int64_t dd = ( d*d)>>30;
1171 int64_t ddd= (dd*d)>>30;
1174 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
1175 else if (d < 1LL<<31)
1176 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
1179 coeff *= fone>>(30+24);
1181 /* else if (flags & SWS_X)
1183 double p= param ? param*0.01 : 0.3;
1184 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1185 coeff*= pow(2.0, - p*d*d);
1187 else if (flags & SWS_X)
1189 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1196 if (c<0.0) c= -pow(-c, A);
1198 coeff= (c*0.5 + 0.5)*fone;
1200 else if (flags & SWS_AREA)
1202 int64_t d2= d - (1<<29);
1203 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
1204 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
1206 coeff *= fone>>(30+16);
1208 else if (flags & SWS_GAUSS)
1210 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1211 coeff = (pow(2.0, - p*floatd*floatd))*fone;
1213 else if (flags & SWS_SINC)
1215 coeff = (d ? sin(floatd*PI)/(floatd*PI) : 1.0)*fone;
1217 else if (flags & SWS_LANCZOS)
1219 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1220 coeff = (d ? sin(floatd*PI)*sin(floatd*PI/p)/(floatd*floatd*PI*PI/p) : 1.0)*fone;
1221 if (floatd>p) coeff=0;
1223 else if (flags & SWS_BILINEAR)
1226 if (coeff<0) coeff=0;
1227 coeff *= fone >> 30;
1229 else if (flags & SWS_SPLINE)
1231 double p=-2.196152422706632;
1232 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
1235 coeff= 0.0; //GCC warning killer
1239 filter[i*filterSize + j]= coeff;
1246 /* apply src & dst Filter to filter -> filter2
1249 assert(filterSize>0);
1250 filter2Size= filterSize;
1251 if (srcFilter) filter2Size+= srcFilter->length - 1;
1252 if (dstFilter) filter2Size+= dstFilter->length - 1;
1253 assert(filter2Size>0);
1254 filter2= av_mallocz(filter2Size*dstW*sizeof(*filter2));
1256 for (i=0; i<dstW; i++)
1261 for (k=0; k<srcFilter->length; k++){
1262 for (j=0; j<filterSize; j++)
1263 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
1266 for (j=0; j<filterSize; j++)
1267 filter2[i*filter2Size + j]= filter[i*filterSize + j];
1271 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1275 /* try to reduce the filter-size (step1 find size and shift left) */
1276 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1278 for (i=dstW-1; i>=0; i--)
1280 int min= filter2Size;
1284 /* get rid off near zero elements on the left by shifting left */
1285 for (j=0; j<filter2Size; j++)
1288 cutOff += FFABS(filter2[i*filter2Size]);
1290 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1292 /* preserve monotonicity because the core can't handle the filter otherwise */
1293 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1295 // Move filter coeffs left
1296 for (k=1; k<filter2Size; k++)
1297 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1298 filter2[i*filter2Size + k - 1]= 0;
1303 /* count near zeros on the right */
1304 for (j=filter2Size-1; j>0; j--)
1306 cutOff += FFABS(filter2[i*filter2Size + j]);
1308 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
1312 if (min>minFilterSize) minFilterSize= min;
1315 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1316 // we can handle the special case 4,
1317 // so we don't want to go to the full 8
1318 if (minFilterSize < 5)
1321 // we really don't want to waste our time
1322 // doing useless computation, so fall-back on
1323 // the scalar C code for very small filter.
1324 // vectorizing is worth it only if you have
1325 // decent-sized vector.
1326 if (minFilterSize < 3)
1330 if (flags & SWS_CPU_CAPS_MMX) {
1331 // special case for unscaled vertical filtering
1332 if (minFilterSize == 1 && filterAlign == 2)
1336 assert(minFilterSize > 0);
1337 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1338 assert(filterSize > 0);
1339 filter= av_malloc(filterSize*dstW*sizeof(*filter));
1340 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1342 *outFilterSize= filterSize;
1344 if (flags&SWS_PRINT_INFO)
1345 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1346 /* try to reduce the filter-size (step2 reduce it) */
1347 for (i=0; i<dstW; i++)
1351 for (j=0; j<filterSize; j++)
1353 if (j>=filter2Size) filter[i*filterSize + j]= 0;
1354 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1355 if((flags & SWS_BITEXACT) && j>=minFilterSize)
1356 filter[i*filterSize + j]= 0;
1361 //FIXME try to align filterpos if possible
1364 for (i=0; i<dstW; i++)
1367 if ((*filterPos)[i] < 0)
1369 // Move filter coeffs left to compensate for filterPos
1370 for (j=1; j<filterSize; j++)
1372 int left= FFMAX(j + (*filterPos)[i], 0);
1373 filter[i*filterSize + left] += filter[i*filterSize + j];
1374 filter[i*filterSize + j]=0;
1379 if ((*filterPos)[i] + filterSize > srcW)
1381 int shift= (*filterPos)[i] + filterSize - srcW;
1382 // Move filter coeffs right to compensate for filterPos
1383 for (j=filterSize-2; j>=0; j--)
1385 int right= FFMIN(j + shift, filterSize-1);
1386 filter[i*filterSize +right] += filter[i*filterSize +j];
1387 filter[i*filterSize +j]=0;
1389 (*filterPos)[i]= srcW - filterSize;
1393 // Note the +1 is for the MMXscaler which reads over the end
1394 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1395 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1397 /* Normalize & Store in outFilter */
1398 for (i=0; i<dstW; i++)
1404 for (j=0; j<filterSize; j++)
1406 sum+= filter[i*filterSize + j];
1408 sum= (sum + one/2)/ one;
1409 for (j=0; j<*outFilterSize; j++)
1411 int64_t v= filter[i*filterSize + j] + error;
1412 int intV= ROUNDED_DIV(v, sum);
1413 (*outFilter)[i*(*outFilterSize) + j]= intV;
1414 error= v - intV*sum;
1418 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1419 for (i=0; i<*outFilterSize; i++)
1421 int j= dstW*(*outFilterSize);
1422 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1433 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1436 long imm8OfPShufW1A;
1437 long imm8OfPShufW2A;
1438 long fragmentLengthA;
1440 long imm8OfPShufW1B;
1441 long imm8OfPShufW2B;
1442 long fragmentLengthB;
1447 // create an optimized horizontal scaling routine
1455 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1456 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1457 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1458 "punpcklbw %%mm7, %%mm1 \n\t"
1459 "punpcklbw %%mm7, %%mm0 \n\t"
1460 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1462 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1464 "psubw %%mm1, %%mm0 \n\t"
1465 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1466 "pmullw %%mm3, %%mm0 \n\t"
1467 "psllw $7, %%mm1 \n\t"
1468 "paddw %%mm1, %%mm0 \n\t"
1470 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1472 "add $8, %%"REG_a" \n\t"
1476 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1477 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1478 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1483 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1487 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1488 "=r" (fragmentLengthA)
1495 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1496 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1497 "punpcklbw %%mm7, %%mm0 \n\t"
1498 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1500 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1502 "psubw %%mm1, %%mm0 \n\t"
1503 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1504 "pmullw %%mm3, %%mm0 \n\t"
1505 "psllw $7, %%mm1 \n\t"
1506 "paddw %%mm1, %%mm0 \n\t"
1508 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1510 "add $8, %%"REG_a" \n\t"
1514 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1515 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1516 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1521 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1525 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1526 "=r" (fragmentLengthB)
1529 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1532 for (i=0; i<dstW/numSplits; i++)
1539 int b=((xpos+xInc)>>16) - xx;
1540 int c=((xpos+xInc*2)>>16) - xx;
1541 int d=((xpos+xInc*3)>>16) - xx;
1543 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1544 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1545 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1546 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1551 int maxShift= 3-(d+1);
1554 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1556 funnyCode[fragmentPos + imm8OfPShufW1B]=
1557 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1558 funnyCode[fragmentPos + imm8OfPShufW2B]=
1559 a | (b<<2) | (c<<4) | (d<<6);
1561 if (i+3>=dstW) shift=maxShift; //avoid overread
1562 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1564 if (shift && i>=shift)
1566 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1567 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1568 filterPos[i/2]-=shift;
1571 fragmentPos+= fragmentLengthB;
1578 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1580 funnyCode[fragmentPos + imm8OfPShufW1A]=
1581 funnyCode[fragmentPos + imm8OfPShufW2A]=
1582 a | (b<<2) | (c<<4) | (d<<6);
1584 if (i+4>=dstW) shift=maxShift; //avoid overread
1585 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1587 if (shift && i>=shift)
1589 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1590 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1591 filterPos[i/2]-=shift;
1594 fragmentPos+= fragmentLengthA;
1597 funnyCode[fragmentPos]= RET;
1601 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1603 #endif /* COMPILE_MMX2 */
1605 static void globalInit(void){
1606 // generating tables:
1608 for (i=0; i<768; i++){
1609 int c= av_clip_uint8(i-256);
1614 static SwsFunc getSwsFunc(int flags){
1616 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1617 #if defined(ARCH_X86)
1618 // ordered per speed fastest first
1619 if (flags & SWS_CPU_CAPS_MMX2)
1620 return swScale_MMX2;
1621 else if (flags & SWS_CPU_CAPS_3DNOW)
1622 return swScale_3DNow;
1623 else if (flags & SWS_CPU_CAPS_MMX)
1630 if (flags & SWS_CPU_CAPS_ALTIVEC)
1631 return swScale_altivec;
1636 #endif /* defined(ARCH_X86) */
1637 #else //RUNTIME_CPUDETECT
1639 return swScale_MMX2;
1640 #elif defined (HAVE_3DNOW)
1641 return swScale_3DNow;
1642 #elif defined (HAVE_MMX)
1644 #elif defined (HAVE_ALTIVEC)
1645 return swScale_altivec;
1649 #endif //!RUNTIME_CPUDETECT
1652 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1653 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1654 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1656 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1657 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1661 uint8_t *srcPtr= src[0];
1662 uint8_t *dstPtr= dst;
1663 for (i=0; i<srcSliceH; i++)
1665 memcpy(dstPtr, srcPtr, c->srcW);
1666 srcPtr+= srcStride[0];
1667 dstPtr+= dstStride[0];
1670 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1671 if (c->dstFormat == PIX_FMT_NV12)
1672 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1674 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1679 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1680 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1681 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1683 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1688 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1689 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1690 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1692 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1697 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1698 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1699 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1701 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1706 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1707 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1708 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1710 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1715 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1716 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1717 int srcSliceH, uint8_t* dst[], int dstStride[]){
1718 const int srcFormat= c->srcFormat;
1719 const int dstFormat= c->dstFormat;
1720 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1721 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1722 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1723 const int dstId= fmt_depth(dstFormat) >> 2;
1724 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1727 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1728 || (isRGB(srcFormat) && isRGB(dstFormat))){
1729 switch(srcId | (dstId<<4)){
1730 case 0x34: conv= rgb16to15; break;
1731 case 0x36: conv= rgb24to15; break;
1732 case 0x38: conv= rgb32to15; break;
1733 case 0x43: conv= rgb15to16; break;
1734 case 0x46: conv= rgb24to16; break;
1735 case 0x48: conv= rgb32to16; break;
1736 case 0x63: conv= rgb15to24; break;
1737 case 0x64: conv= rgb16to24; break;
1738 case 0x68: conv= rgb32to24; break;
1739 case 0x83: conv= rgb15to32; break;
1740 case 0x84: conv= rgb16to32; break;
1741 case 0x86: conv= rgb24to32; break;
1742 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1743 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1745 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1746 || (isRGB(srcFormat) && isBGR(dstFormat))){
1747 switch(srcId | (dstId<<4)){
1748 case 0x33: conv= rgb15tobgr15; break;
1749 case 0x34: conv= rgb16tobgr15; break;
1750 case 0x36: conv= rgb24tobgr15; break;
1751 case 0x38: conv= rgb32tobgr15; break;
1752 case 0x43: conv= rgb15tobgr16; break;
1753 case 0x44: conv= rgb16tobgr16; break;
1754 case 0x46: conv= rgb24tobgr16; break;
1755 case 0x48: conv= rgb32tobgr16; break;
1756 case 0x63: conv= rgb15tobgr24; break;
1757 case 0x64: conv= rgb16tobgr24; break;
1758 case 0x66: conv= rgb24tobgr24; break;
1759 case 0x68: conv= rgb32tobgr24; break;
1760 case 0x83: conv= rgb15tobgr32; break;
1761 case 0x84: conv= rgb16tobgr32; break;
1762 case 0x86: conv= rgb24tobgr32; break;
1763 case 0x88: conv= rgb32tobgr32; break;
1764 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1765 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1768 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1769 sws_format_name(srcFormat), sws_format_name(dstFormat));
1774 uint8_t *srcPtr= src[0];
1775 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1776 srcPtr += ALT32_CORR;
1778 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1779 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1783 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1785 for (i=0; i<srcSliceH; i++)
1787 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1788 srcPtr+= srcStride[0];
1789 dstPtr+= dstStride[0];
1796 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1797 int srcSliceH, uint8_t* dst[], int dstStride[]){
1801 dst[0]+ srcSliceY *dstStride[0],
1802 dst[1]+(srcSliceY>>1)*dstStride[1],
1803 dst[2]+(srcSliceY>>1)*dstStride[2],
1805 dstStride[0], dstStride[1], srcStride[0]);
1809 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1810 int srcSliceH, uint8_t* dst[], int dstStride[]){
1814 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1815 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1817 uint8_t *srcPtr= src[0];
1818 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1820 for (i=0; i<srcSliceH; i++)
1822 memcpy(dstPtr, srcPtr, c->srcW);
1823 srcPtr+= srcStride[0];
1824 dstPtr+= dstStride[0];
1828 if (c->dstFormat==PIX_FMT_YUV420P){
1829 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1830 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1832 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1833 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1838 /* unscaled copy like stuff (assumes nearly identical formats) */
1839 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1840 int srcSliceH, uint8_t* dst[], int dstStride[])
1842 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1843 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1847 uint8_t *srcPtr= src[0];
1848 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1851 /* universal length finder */
1852 while(length+c->srcW <= FFABS(dstStride[0])
1853 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1856 for (i=0; i<srcSliceH; i++)
1858 memcpy(dstPtr, srcPtr, length);
1859 srcPtr+= srcStride[0];
1860 dstPtr+= dstStride[0];
1866 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1867 int srcSliceH, uint8_t* dst[], int dstStride[])
1870 for (plane=0; plane<3; plane++)
1872 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1873 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1874 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1876 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1878 if (!isGray(c->dstFormat))
1879 memset(dst[plane], 128, dstStride[plane]*height);
1883 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1884 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1888 uint8_t *srcPtr= src[plane];
1889 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1890 for (i=0; i<height; i++)
1892 memcpy(dstPtr, srcPtr, length);
1893 srcPtr+= srcStride[plane];
1894 dstPtr+= dstStride[plane];
1902 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1903 int srcSliceH, uint8_t* dst[], int dstStride[]){
1905 int length= c->srcW;
1907 int height= srcSliceH;
1909 uint8_t *srcPtr= src[0];
1910 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1912 if (!isGray(c->dstFormat)){
1913 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1914 memset(dst[1], 128, dstStride[1]*height);
1915 memset(dst[2], 128, dstStride[2]*height);
1917 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1918 for (i=0; i<height; i++)
1920 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1921 srcPtr+= srcStride[0];
1922 dstPtr+= dstStride[0];
1927 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1928 int srcSliceH, uint8_t* dst[], int dstStride[]){
1930 int length= c->srcW;
1932 int height= srcSliceH;
1934 uint8_t *srcPtr= src[0];
1935 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1936 for (i=0; i<height; i++)
1938 for (j=0; j<length; j++)
1940 dstPtr[j<<1] = srcPtr[j];
1941 dstPtr[(j<<1)+1] = srcPtr[j];
1943 srcPtr+= srcStride[0];
1944 dstPtr+= dstStride[0];
1949 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1950 int srcSliceH, uint8_t* dst[], int dstStride[]){
1952 int length= c->srcW;
1954 int height= srcSliceH;
1956 uint16_t *srcPtr= (uint16_t*)src[0];
1957 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1958 for (i=0; i<height; i++)
1960 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1961 srcPtr+= srcStride[0]/2;
1962 dstPtr+= dstStride[0]/2;
1968 static void getSubSampleFactors(int *h, int *v, int format){
1970 case PIX_FMT_UYVY422:
1971 case PIX_FMT_YUYV422:
1975 case PIX_FMT_YUV420P:
1976 case PIX_FMT_YUVA420P:
1977 case PIX_FMT_GRAY16BE:
1978 case PIX_FMT_GRAY16LE:
1979 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1985 case PIX_FMT_YUV440P:
1989 case PIX_FMT_YUV410P:
1993 case PIX_FMT_YUV444P:
1997 case PIX_FMT_YUV422P:
2001 case PIX_FMT_YUV411P:
2012 static uint16_t roundToInt16(int64_t f){
2013 int r= (f + (1<<15))>>16;
2014 if (r<-0x7FFF) return 0x8000;
2015 else if (r> 0x7FFF) return 0x7FFF;
2020 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
2021 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2022 * @return -1 if not supported
2024 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2025 int64_t crv = inv_table[0];
2026 int64_t cbu = inv_table[1];
2027 int64_t cgu = -inv_table[2];
2028 int64_t cgv = -inv_table[3];
2032 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2033 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2035 c->brightness= brightness;
2036 c->contrast = contrast;
2037 c->saturation= saturation;
2038 c->srcRange = srcRange;
2039 c->dstRange = dstRange;
2040 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2042 c->uOffset= 0x0400040004000400LL;
2043 c->vOffset= 0x0400040004000400LL;
2049 crv= (crv*224) / 255;
2050 cbu= (cbu*224) / 255;
2051 cgu= (cgu*224) / 255;
2052 cgv= (cgv*224) / 255;
2055 cy = (cy *contrast )>>16;
2056 crv= (crv*contrast * saturation)>>32;
2057 cbu= (cbu*contrast * saturation)>>32;
2058 cgu= (cgu*contrast * saturation)>>32;
2059 cgv= (cgv*contrast * saturation)>>32;
2061 oy -= 256*brightness;
2063 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2064 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2065 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2066 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2067 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2068 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2070 c->yuv2rgb_y_coeff = (int16_t)roundToInt16(cy <<13);
2071 c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
2072 c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
2073 c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
2074 c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
2075 c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
2077 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2080 #ifdef COMPILE_ALTIVEC
2081 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2082 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2088 * @return -1 if not supported
2090 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2091 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2093 *inv_table = c->srcColorspaceTable;
2094 *table = c->dstColorspaceTable;
2095 *srcRange = c->srcRange;
2096 *dstRange = c->dstRange;
2097 *brightness= c->brightness;
2098 *contrast = c->contrast;
2099 *saturation= c->saturation;
2104 static int handle_jpeg(int *format)
2107 case PIX_FMT_YUVJ420P:
2108 *format = PIX_FMT_YUV420P;
2110 case PIX_FMT_YUVJ422P:
2111 *format = PIX_FMT_YUV422P;
2113 case PIX_FMT_YUVJ444P:
2114 *format = PIX_FMT_YUV444P;
2116 case PIX_FMT_YUVJ440P:
2117 *format = PIX_FMT_YUV440P;
2124 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2125 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2129 int usesVFilter, usesHFilter;
2130 int unscaled, needsDither;
2131 int srcRange, dstRange;
2132 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2133 #if defined(ARCH_X86)
2134 if (flags & SWS_CPU_CAPS_MMX)
2135 asm volatile("emms\n\t"::: "memory");
2138 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2139 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2141 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2142 #elif defined (HAVE_3DNOW)
2143 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2144 #elif defined (HAVE_MMX)
2145 flags |= SWS_CPU_CAPS_MMX;
2146 #elif defined (HAVE_ALTIVEC)
2147 flags |= SWS_CPU_CAPS_ALTIVEC;
2148 #elif defined (ARCH_BFIN)
2149 flags |= SWS_CPU_CAPS_BFIN;
2151 #endif /* RUNTIME_CPUDETECT */
2152 if (clip_table[512] != 255) globalInit();
2153 if (!rgb15to16) sws_rgb2rgb_init(flags);
2155 unscaled = (srcW == dstW && srcH == dstH);
2156 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2157 && (fmt_depth(dstFormat))<24
2158 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2160 srcRange = handle_jpeg(&srcFormat);
2161 dstRange = handle_jpeg(&dstFormat);
2163 if (!isSupportedIn(srcFormat))
2165 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2168 if (!isSupportedOut(dstFormat))
2170 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2174 i= flags & ( SWS_POINT
2185 if(!i || (i & (i-1)))
2187 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2192 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
2194 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2195 srcW, srcH, dstW, dstH);
2198 if(srcW > VOFW || dstW > VOFW){
2199 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2203 if (!dstFilter) dstFilter= &dummyFilter;
2204 if (!srcFilter) srcFilter= &dummyFilter;
2206 c= av_mallocz(sizeof(SwsContext));
2208 c->av_class = &sws_context_class;
2213 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2214 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2216 c->dstFormat= dstFormat;
2217 c->srcFormat= srcFormat;
2218 c->vRounder= 4* 0x0001000100010001ULL;
2220 usesHFilter= usesVFilter= 0;
2221 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2222 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2223 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2224 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2225 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2226 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2227 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2228 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2230 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2231 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2233 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2234 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2236 // drop some chroma lines if the user wants it
2237 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2238 c->chrSrcVSubSample+= c->vChrDrop;
2240 // drop every 2. pixel for chroma calculation unless user wants full chroma
2241 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2242 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2243 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2244 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2245 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2246 c->chrSrcHSubSample=1;
2249 c->param[0] = param[0];
2250 c->param[1] = param[1];
2253 c->param[1] = SWS_PARAM_DEFAULT;
2256 c->chrIntHSubSample= c->chrDstHSubSample;
2257 c->chrIntVSubSample= c->chrSrcVSubSample;
2259 // Note the -((-x)>>y) is so that we always round toward +inf.
2260 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2261 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2262 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2263 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2265 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);
2267 /* unscaled special Cases */
2268 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2271 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2273 c->swScale= PlanarToNV12Wrapper;
2277 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat))
2278 && !(flags & SWS_ACCURATE_RND))
2280 c->swScale= yuv2rgb_get_func_ptr(c);
2284 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_BITEXACT))
2286 c->swScale= yvu9toyv12Wrapper;
2290 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P && !(flags & SWS_ACCURATE_RND))
2291 c->swScale= bgr24toyv12Wrapper;
2293 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2294 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2295 && (isBGR(dstFormat) || isRGB(dstFormat))
2296 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2297 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2298 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2299 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2300 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2301 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2302 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2303 && srcFormat != PIX_FMT_MONOWHITE && dstFormat != PIX_FMT_MONOWHITE
2304 && dstFormat != PIX_FMT_RGB32_1
2305 && dstFormat != PIX_FMT_BGR32_1
2306 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2307 c->swScale= rgb2rgbWrapper;
2309 if (srcFormat == PIX_FMT_YUV422P)
2311 if (dstFormat == PIX_FMT_YUYV422)
2312 c->swScale= YUV422PToYuy2Wrapper;
2313 else if (dstFormat == PIX_FMT_UYVY422)
2314 c->swScale= YUV422PToUyvyWrapper;
2317 /* LQ converters if -sws 0 or -sws 4*/
2318 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2320 if (srcFormat == PIX_FMT_YUV420P)
2322 if (dstFormat == PIX_FMT_YUYV422)
2323 c->swScale= PlanarToYuy2Wrapper;
2324 else if (dstFormat == PIX_FMT_UYVY422)
2325 c->swScale= PlanarToUyvyWrapper;
2329 #ifdef COMPILE_ALTIVEC
2330 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2331 srcFormat == PIX_FMT_YUV420P) {
2332 // unscaled YV12 -> packed YUV, we want speed
2333 if (dstFormat == PIX_FMT_YUYV422)
2334 c->swScale= yv12toyuy2_unscaled_altivec;
2335 else if (dstFormat == PIX_FMT_UYVY422)
2336 c->swScale= yv12touyvy_unscaled_altivec;
2341 if ( srcFormat == dstFormat
2342 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2343 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2345 if (isPacked(c->srcFormat))
2346 c->swScale= packedCopy;
2347 else /* Planar YUV or gray */
2348 c->swScale= planarCopy;
2351 /* gray16{le,be} conversions */
2352 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2354 c->swScale= gray16togray;
2356 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2358 c->swScale= graytogray16;
2360 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2362 c->swScale= gray16swap;
2366 if (flags & SWS_CPU_CAPS_BFIN)
2367 ff_bfin_get_unscaled_swscale (c);
2371 if (flags&SWS_PRINT_INFO)
2372 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2373 sws_format_name(srcFormat), sws_format_name(dstFormat));
2378 if (flags & SWS_CPU_CAPS_MMX2)
2380 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2381 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2383 if (flags&SWS_PRINT_INFO)
2384 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2386 if (usesHFilter) c->canMMX2BeUsed=0;
2391 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2392 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2394 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2395 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2396 // n-2 is the last chrominance sample available
2397 // this is not perfect, but no one should notice the difference, the more correct variant
2398 // would be like the vertical one, but that would require some special code for the
2399 // first and last pixel
2400 if (flags&SWS_FAST_BILINEAR)
2402 if (c->canMMX2BeUsed)
2407 //we don't use the x86asm scaler if mmx is available
2408 else if (flags & SWS_CPU_CAPS_MMX)
2410 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2411 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2415 /* precalculate horizontal scaler filter coefficients */
2417 const int filterAlign=
2418 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2419 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2422 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2423 srcW , dstW, filterAlign, 1<<14,
2424 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2425 srcFilter->lumH, dstFilter->lumH, c->param);
2426 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2427 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2428 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2429 srcFilter->chrH, dstFilter->chrH, c->param);
2431 #define MAX_FUNNY_CODE_SIZE 10000
2432 #if defined(COMPILE_MMX2)
2433 // can't downscale !!!
2434 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2436 #ifdef MAP_ANONYMOUS
2437 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2438 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2440 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2441 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2444 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2445 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2446 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2447 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2449 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2450 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2452 #endif /* defined(COMPILE_MMX2) */
2453 } // Init Horizontal stuff
2457 /* precalculate vertical scaler filter coefficients */
2459 const int filterAlign=
2460 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2461 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2464 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2465 srcH , dstH, filterAlign, (1<<12),
2466 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2467 srcFilter->lumV, dstFilter->lumV, c->param);
2468 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2469 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
2470 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2471 srcFilter->chrV, dstFilter->chrV, c->param);
2474 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2475 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2477 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2479 short *p = (short *)&c->vYCoeffsBank[i];
2481 p[j] = c->vLumFilter[i];
2484 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2486 short *p = (short *)&c->vCCoeffsBank[i];
2488 p[j] = c->vChrFilter[i];
2493 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2494 c->vLumBufSize= c->vLumFilterSize;
2495 c->vChrBufSize= c->vChrFilterSize;
2496 for (i=0; i<dstH; i++)
2498 int chrI= i*c->chrDstH / dstH;
2499 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2500 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2502 nextSlice>>= c->chrSrcVSubSample;
2503 nextSlice<<= c->chrSrcVSubSample;
2504 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2505 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2506 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2507 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2510 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2511 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2512 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2513 //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)
2514 /* align at 16 bytes for AltiVec */
2515 for (i=0; i<c->vLumBufSize; i++)
2516 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2517 for (i=0; i<c->vChrBufSize; i++)
2518 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2520 //try to avoid drawing green stuff between the right end and the stride end
2521 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2523 assert(2*VOFW == VOF);
2525 assert(c->chrDstH <= dstH);
2527 if (flags&SWS_PRINT_INFO)
2530 const char *dither= " dithered";
2532 const char *dither= "";
2534 if (flags&SWS_FAST_BILINEAR)
2535 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2536 else if (flags&SWS_BILINEAR)
2537 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2538 else if (flags&SWS_BICUBIC)
2539 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2540 else if (flags&SWS_X)
2541 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2542 else if (flags&SWS_POINT)
2543 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2544 else if (flags&SWS_AREA)
2545 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2546 else if (flags&SWS_BICUBLIN)
2547 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2548 else if (flags&SWS_GAUSS)
2549 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2550 else if (flags&SWS_SINC)
2551 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2552 else if (flags&SWS_LANCZOS)
2553 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2554 else if (flags&SWS_SPLINE)
2555 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2557 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2559 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2560 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2561 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2563 av_log(c, AV_LOG_INFO, "from %s to %s ",
2564 sws_format_name(srcFormat), sws_format_name(dstFormat));
2566 if (flags & SWS_CPU_CAPS_MMX2)
2567 av_log(c, AV_LOG_INFO, "using MMX2\n");
2568 else if (flags & SWS_CPU_CAPS_3DNOW)
2569 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2570 else if (flags & SWS_CPU_CAPS_MMX)
2571 av_log(c, AV_LOG_INFO, "using MMX\n");
2572 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2573 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2575 av_log(c, AV_LOG_INFO, "using C\n");
2578 if (flags & SWS_PRINT_INFO)
2580 if (flags & SWS_CPU_CAPS_MMX)
2582 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2583 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2586 if (c->hLumFilterSize==4)
2587 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2588 else if (c->hLumFilterSize==8)
2589 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2591 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2593 if (c->hChrFilterSize==4)
2594 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2595 else if (c->hChrFilterSize==8)
2596 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2598 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2603 #if defined(ARCH_X86)
2604 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2606 if (flags & SWS_FAST_BILINEAR)
2607 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2609 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2612 if (isPlanarYUV(dstFormat))
2614 if (c->vLumFilterSize==1)
2615 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2617 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2621 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2622 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2623 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2624 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2625 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2627 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2630 if (dstFormat==PIX_FMT_BGR24)
2631 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2632 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2633 else if (dstFormat==PIX_FMT_RGB32)
2634 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2635 else if (dstFormat==PIX_FMT_BGR565)
2636 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2637 else if (dstFormat==PIX_FMT_BGR555)
2638 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2640 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2642 if (flags & SWS_PRINT_INFO)
2644 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2645 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2646 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2647 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2650 c->swScale= getSwsFunc(flags);
2655 * swscale wrapper, so we don't need to export the SwsContext.
2656 * assumes planar YUV to be in YUV order instead of YVU
2658 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2659 int srcSliceH, uint8_t* dst[], int dstStride[]){
2661 uint8_t* src2[4]= {src[0], src[1], src[2]};
2663 int use_pal= c->srcFormat == PIX_FMT_PAL8
2664 || c->srcFormat == PIX_FMT_BGR4_BYTE
2665 || c->srcFormat == PIX_FMT_RGB4_BYTE
2666 || c->srcFormat == PIX_FMT_BGR8
2667 || c->srcFormat == PIX_FMT_RGB8;
2669 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2670 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2673 if (c->sliceDir == 0) {
2674 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2678 for (i=0; i<256; i++){
2679 int p, r, g, b,y,u,v;
2680 if(c->srcFormat == PIX_FMT_PAL8){
2681 p=((uint32_t*)(src[1]))[i];
2685 }else if(c->srcFormat == PIX_FMT_RGB8){
2689 }else if(c->srcFormat == PIX_FMT_BGR8){
2693 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2697 }else if(c->srcFormat == PIX_FMT_BGR4_BYTE){
2702 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2703 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2704 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2705 pal[i]= y + (u<<8) + (v<<16);
2707 src2[1]= (uint8_t*)pal;
2710 // copy strides, so they can safely be modified
2711 if (c->sliceDir == 1) {
2712 // slices go from top to bottom
2713 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2714 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2715 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2717 // slices go from bottom to top => we flip the image internally
2718 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2719 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2720 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2721 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2722 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2724 src2[0] += (srcSliceH-1)*srcStride[0];
2726 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2727 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2729 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2734 * swscale wrapper, so we don't need to export the SwsContext
2736 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2737 int srcSliceH, uint8_t* dst[], int dstStride[]){
2738 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2741 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2742 float lumaSharpen, float chromaSharpen,
2743 float chromaHShift, float chromaVShift,
2746 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2748 if (lumaGBlur!=0.0){
2749 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2750 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2752 filter->lumH= sws_getIdentityVec();
2753 filter->lumV= sws_getIdentityVec();
2756 if (chromaGBlur!=0.0){
2757 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2758 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2760 filter->chrH= sws_getIdentityVec();
2761 filter->chrV= sws_getIdentityVec();
2764 if (chromaSharpen!=0.0){
2765 SwsVector *id= sws_getIdentityVec();
2766 sws_scaleVec(filter->chrH, -chromaSharpen);
2767 sws_scaleVec(filter->chrV, -chromaSharpen);
2768 sws_addVec(filter->chrH, id);
2769 sws_addVec(filter->chrV, id);
2773 if (lumaSharpen!=0.0){
2774 SwsVector *id= sws_getIdentityVec();
2775 sws_scaleVec(filter->lumH, -lumaSharpen);
2776 sws_scaleVec(filter->lumV, -lumaSharpen);
2777 sws_addVec(filter->lumH, id);
2778 sws_addVec(filter->lumV, id);
2782 if (chromaHShift != 0.0)
2783 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2785 if (chromaVShift != 0.0)
2786 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2788 sws_normalizeVec(filter->chrH, 1.0);
2789 sws_normalizeVec(filter->chrV, 1.0);
2790 sws_normalizeVec(filter->lumH, 1.0);
2791 sws_normalizeVec(filter->lumV, 1.0);
2793 if (verbose) sws_printVec(filter->chrH);
2794 if (verbose) sws_printVec(filter->lumH);
2800 * returns a normalized gaussian curve used to filter stuff
2801 * quality=3 is high quality, lowwer is lowwer quality
2803 SwsVector *sws_getGaussianVec(double variance, double quality){
2804 const int length= (int)(variance*quality + 0.5) | 1;
2806 double *coeff= av_malloc(length*sizeof(double));
2807 double middle= (length-1)*0.5;
2808 SwsVector *vec= av_malloc(sizeof(SwsVector));
2811 vec->length= length;
2813 for (i=0; i<length; i++)
2815 double dist= i-middle;
2816 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2819 sws_normalizeVec(vec, 1.0);
2824 SwsVector *sws_getConstVec(double c, int length){
2826 double *coeff= av_malloc(length*sizeof(double));
2827 SwsVector *vec= av_malloc(sizeof(SwsVector));
2830 vec->length= length;
2832 for (i=0; i<length; i++)
2839 SwsVector *sws_getIdentityVec(void){
2840 return sws_getConstVec(1.0, 1);
2843 double sws_dcVec(SwsVector *a){
2847 for (i=0; i<a->length; i++)
2853 void sws_scaleVec(SwsVector *a, double scalar){
2856 for (i=0; i<a->length; i++)
2857 a->coeff[i]*= scalar;
2860 void sws_normalizeVec(SwsVector *a, double height){
2861 sws_scaleVec(a, height/sws_dcVec(a));
2864 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2865 int length= a->length + b->length - 1;
2866 double *coeff= av_malloc(length*sizeof(double));
2868 SwsVector *vec= av_malloc(sizeof(SwsVector));
2871 vec->length= length;
2873 for (i=0; i<length; i++) coeff[i]= 0.0;
2875 for (i=0; i<a->length; i++)
2877 for (j=0; j<b->length; j++)
2879 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2886 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2887 int length= FFMAX(a->length, b->length);
2888 double *coeff= av_malloc(length*sizeof(double));
2890 SwsVector *vec= av_malloc(sizeof(SwsVector));
2893 vec->length= length;
2895 for (i=0; i<length; i++) coeff[i]= 0.0;
2897 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2898 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2903 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2904 int length= FFMAX(a->length, b->length);
2905 double *coeff= av_malloc(length*sizeof(double));
2907 SwsVector *vec= av_malloc(sizeof(SwsVector));
2910 vec->length= length;
2912 for (i=0; i<length; i++) coeff[i]= 0.0;
2914 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2915 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2920 /* shift left / or right if "shift" is negative */
2921 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2922 int length= a->length + FFABS(shift)*2;
2923 double *coeff= av_malloc(length*sizeof(double));
2925 SwsVector *vec= av_malloc(sizeof(SwsVector));
2928 vec->length= length;
2930 for (i=0; i<length; i++) coeff[i]= 0.0;
2932 for (i=0; i<a->length; i++)
2934 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2940 void sws_shiftVec(SwsVector *a, int shift){
2941 SwsVector *shifted= sws_getShiftedVec(a, shift);
2943 a->coeff= shifted->coeff;
2944 a->length= shifted->length;
2948 void sws_addVec(SwsVector *a, SwsVector *b){
2949 SwsVector *sum= sws_sumVec(a, b);
2951 a->coeff= sum->coeff;
2952 a->length= sum->length;
2956 void sws_subVec(SwsVector *a, SwsVector *b){
2957 SwsVector *diff= sws_diffVec(a, b);
2959 a->coeff= diff->coeff;
2960 a->length= diff->length;
2964 void sws_convVec(SwsVector *a, SwsVector *b){
2965 SwsVector *conv= sws_getConvVec(a, b);
2967 a->coeff= conv->coeff;
2968 a->length= conv->length;
2972 SwsVector *sws_cloneVec(SwsVector *a){
2973 double *coeff= av_malloc(a->length*sizeof(double));
2975 SwsVector *vec= av_malloc(sizeof(SwsVector));
2978 vec->length= a->length;
2980 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2985 void sws_printVec(SwsVector *a){
2991 for (i=0; i<a->length; i++)
2992 if (a->coeff[i]>max) max= a->coeff[i];
2994 for (i=0; i<a->length; i++)
2995 if (a->coeff[i]<min) min= a->coeff[i];
2999 for (i=0; i<a->length; i++)
3001 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
3002 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
3003 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
3004 av_log(NULL, AV_LOG_DEBUG, "|\n");
3008 void sws_freeVec(SwsVector *a){
3010 av_freep(&a->coeff);
3015 void sws_freeFilter(SwsFilter *filter){
3016 if (!filter) return;
3018 if (filter->lumH) sws_freeVec(filter->lumH);
3019 if (filter->lumV) sws_freeVec(filter->lumV);
3020 if (filter->chrH) sws_freeVec(filter->chrH);
3021 if (filter->chrV) sws_freeVec(filter->chrV);
3026 void sws_freeContext(SwsContext *c){
3032 for (i=0; i<c->vLumBufSize; i++)
3033 av_freep(&c->lumPixBuf[i]);
3034 av_freep(&c->lumPixBuf);
3039 for (i=0; i<c->vChrBufSize; i++)
3040 av_freep(&c->chrPixBuf[i]);
3041 av_freep(&c->chrPixBuf);
3044 av_freep(&c->vLumFilter);
3045 av_freep(&c->vChrFilter);
3046 av_freep(&c->hLumFilter);
3047 av_freep(&c->hChrFilter);
3049 av_freep(&c->vYCoeffsBank);
3050 av_freep(&c->vCCoeffsBank);
3053 av_freep(&c->vLumFilterPos);
3054 av_freep(&c->vChrFilterPos);
3055 av_freep(&c->hLumFilterPos);
3056 av_freep(&c->hChrFilterPos);
3058 #if defined(ARCH_X86) && defined(CONFIG_GPL)
3059 #ifdef MAP_ANONYMOUS
3060 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
3061 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
3063 av_free(c->funnyYCode);
3064 av_free(c->funnyUVCode);
3067 c->funnyUVCode=NULL;
3068 #endif /* defined(ARCH_X86) */
3070 av_freep(&c->lumMmx2Filter);
3071 av_freep(&c->chrMmx2Filter);
3072 av_freep(&c->lumMmx2FilterPos);
3073 av_freep(&c->chrMmx2FilterPos);
3074 av_freep(&c->yuvTable);
3080 * Checks if context is valid or reallocs a new one instead.
3081 * If context is NULL, just calls sws_getContext() to get a new one.
3082 * Otherwise, checks if the parameters are the same already saved in context.
3083 * If that is the case, returns the current context.
3084 * Otherwise, frees context and gets a new one.
3086 * Be warned that srcFilter, dstFilter are not checked, they are
3087 * asumed to remain valid.
3089 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3090 int srcW, int srcH, int srcFormat,
3091 int dstW, int dstH, int dstFormat, int flags,
3092 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
3094 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3097 param = default_param;
3100 if (context->srcW != srcW || context->srcH != srcH ||
3101 context->srcFormat != srcFormat ||
3102 context->dstW != dstW || context->dstH != dstH ||
3103 context->dstFormat != dstFormat || context->flags != flags ||
3104 context->param[0] != param[0] || context->param[1] != param[1])
3106 sws_freeContext(context);
3111 return sws_getContext(srcW, srcH, srcFormat,
3112 dstW, dstH, dstFormat, flags,
3113 srcFilter, dstFilter, param);