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_GRAY16BE \
120 || (x)==PIX_FMT_GRAY16LE \
121 || (x)==PIX_FMT_YUV444P \
122 || (x)==PIX_FMT_YUV422P \
123 || (x)==PIX_FMT_YUV411P \
124 || (x)==PIX_FMT_PAL8 \
125 || (x)==PIX_FMT_BGR8 \
126 || (x)==PIX_FMT_RGB8 \
127 || (x)==PIX_FMT_BGR4_BYTE \
128 || (x)==PIX_FMT_RGB4_BYTE \
129 || (x)==PIX_FMT_YUV440P \
131 #define isSupportedOut(x) ( \
132 (x)==PIX_FMT_YUV420P \
133 || (x)==PIX_FMT_YUYV422 \
134 || (x)==PIX_FMT_UYVY422 \
135 || (x)==PIX_FMT_YUV444P \
136 || (x)==PIX_FMT_YUV422P \
137 || (x)==PIX_FMT_YUV411P \
140 || (x)==PIX_FMT_NV12 \
141 || (x)==PIX_FMT_NV21 \
142 || (x)==PIX_FMT_GRAY16BE \
143 || (x)==PIX_FMT_GRAY16LE \
144 || (x)==PIX_FMT_GRAY8 \
145 || (x)==PIX_FMT_YUV410P \
147 #define isPacked(x) ( \
149 || (x)==PIX_FMT_YUYV422 \
150 || (x)==PIX_FMT_UYVY422 \
155 #define RGB2YUV_SHIFT 15
156 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
157 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
158 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
159 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
160 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
161 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
162 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
163 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
164 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
166 extern const int32_t Inverse_Table_6_9[8][4];
168 static const double rgb2yuv_table[8][9]={
169 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
170 {0.7152, 0.0722, 0.2126, -0.386, 0.5, -0.115, -0.454, -0.046, 0.5},
171 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
172 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
173 {0.59 , 0.11 , 0.30 , -0.331, 0.5, -0.169, -0.421, -0.079, 0.5}, //FCC
174 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5},
175 {0.587 , 0.114 , 0.299 , -0.331, 0.5, -0.169, -0.419, -0.081, 0.5}, //SMPTE 170M
176 {0.701 , 0.087 , 0.212 , -0.384, 0.5 -0.116, -0.445, -0.055, 0.5}, //SMPTE 240M
181 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
184 more intelligent misalignment avoidance for the horizontal scaler
185 write special vertical cubic upscale version
186 Optimize C code (yv12 / minmax)
187 add support for packed pixel yuv input & output
188 add support for Y8 output
189 optimize bgr24 & bgr32
190 add BGR4 output support
191 write special BGR->BGR scaler
194 #if defined(ARCH_X86) && defined (CONFIG_GPL)
195 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
196 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
197 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
198 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
199 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
200 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
201 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
202 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
204 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
205 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
206 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
207 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
209 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
210 0x0103010301030103LL,
211 0x0200020002000200LL,};
213 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
214 0x0602060206020602LL,
215 0x0004000400040004LL,};
217 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
218 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
219 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
220 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
221 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
222 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
224 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
225 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
226 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
229 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
230 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
231 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
233 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
234 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
235 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
236 #endif /* FAST_BGR2YV12 */
237 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
238 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
239 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
241 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY1Coeff) = 0x0C88000040870C88ULL;
242 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toY2Coeff) = 0x20DE4087000020DEULL;
243 DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY1Coeff) = 0x20DE0000408720DEULL;
244 DECLARE_ALIGNED(8, const uint64_t, ff_rgb24toY2Coeff) = 0x0C88408700000C88ULL;
245 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toYOffset) = 0x0008400000084000ULL;
247 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUV[2][4]) = {
248 {0x38380000DAC83838ULL, 0xECFFDAC80000ECFFULL, 0xF6E40000D0E3F6E4ULL, 0x3838D0E300003838ULL},
249 {0xECFF0000DAC8ECFFULL, 0x3838DAC800003838ULL, 0x38380000D0E33838ULL, 0xF6E4D0E30000F6E4ULL},
252 DECLARE_ALIGNED(8, const uint64_t, ff_bgr24toUVOffset)= 0x0040400000404000ULL;
254 #endif /* defined(ARCH_X86) */
256 // clipping helper table for C implementations:
257 static unsigned char clip_table[768];
259 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
261 extern const uint8_t dither_2x2_4[2][8];
262 extern const uint8_t dither_2x2_8[2][8];
263 extern const uint8_t dither_8x8_32[8][8];
264 extern const uint8_t dither_8x8_73[8][8];
265 extern const uint8_t dither_8x8_220[8][8];
267 const char *sws_format_name(enum PixelFormat format)
270 case PIX_FMT_YUV420P:
272 case PIX_FMT_YUVA420P:
274 case PIX_FMT_YUYV422:
280 case PIX_FMT_YUV422P:
282 case PIX_FMT_YUV444P:
286 case PIX_FMT_YUV410P:
288 case PIX_FMT_YUV411P:
294 case PIX_FMT_GRAY16BE:
296 case PIX_FMT_GRAY16LE:
300 case PIX_FMT_MONOWHITE:
302 case PIX_FMT_MONOBLACK:
306 case PIX_FMT_YUVJ420P:
308 case PIX_FMT_YUVJ422P:
310 case PIX_FMT_YUVJ444P:
312 case PIX_FMT_XVMC_MPEG2_MC:
313 return "xvmc_mpeg2_mc";
314 case PIX_FMT_XVMC_MPEG2_IDCT:
315 return "xvmc_mpeg2_idct";
316 case PIX_FMT_UYVY422:
318 case PIX_FMT_UYYVYY411:
320 case PIX_FMT_RGB32_1:
322 case PIX_FMT_BGR32_1:
334 case PIX_FMT_BGR4_BYTE:
340 case PIX_FMT_RGB4_BYTE:
346 case PIX_FMT_YUV440P:
349 return "Unknown format";
353 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
354 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
355 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
357 //FIXME Optimize (just quickly writen not opti..)
359 for (i=0; i<dstW; i++)
363 for (j=0; j<lumFilterSize; j++)
364 val += lumSrc[j][i] * lumFilter[j];
366 dest[i]= av_clip_uint8(val>>19);
370 for (i=0; i<chrDstW; i++)
375 for (j=0; j<chrFilterSize; j++)
377 u += chrSrc[j][i] * chrFilter[j];
378 v += chrSrc[j][i + VOFW] * chrFilter[j];
381 uDest[i]= av_clip_uint8(u>>19);
382 vDest[i]= av_clip_uint8(v>>19);
386 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
387 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
388 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
390 //FIXME Optimize (just quickly writen not opti..)
392 for (i=0; i<dstW; i++)
396 for (j=0; j<lumFilterSize; j++)
397 val += lumSrc[j][i] * lumFilter[j];
399 dest[i]= av_clip_uint8(val>>19);
405 if (dstFormat == PIX_FMT_NV12)
406 for (i=0; i<chrDstW; i++)
411 for (j=0; j<chrFilterSize; j++)
413 u += chrSrc[j][i] * chrFilter[j];
414 v += chrSrc[j][i + VOFW] * chrFilter[j];
417 uDest[2*i]= av_clip_uint8(u>>19);
418 uDest[2*i+1]= av_clip_uint8(v>>19);
421 for (i=0; i<chrDstW; i++)
426 for (j=0; j<chrFilterSize; j++)
428 u += chrSrc[j][i] * chrFilter[j];
429 v += chrSrc[j][i + VOFW] * chrFilter[j];
432 uDest[2*i]= av_clip_uint8(v>>19);
433 uDest[2*i+1]= av_clip_uint8(u>>19);
437 #define YSCALE_YUV_2_PACKEDX_C(type) \
438 for (i=0; i<(dstW>>1); i++){\
444 type av_unused *r, *b, *g;\
447 for (j=0; j<lumFilterSize; j++)\
449 Y1 += lumSrc[j][i2] * lumFilter[j];\
450 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
452 for (j=0; j<chrFilterSize; j++)\
454 U += chrSrc[j][i] * chrFilter[j];\
455 V += chrSrc[j][i+VOFW] * chrFilter[j];\
461 if ((Y1|Y2|U|V)&256)\
463 if (Y1>255) Y1=255; \
464 else if (Y1<0)Y1=0; \
465 if (Y2>255) Y2=255; \
466 else if (Y2<0)Y2=0; \
473 #define YSCALE_YUV_2_GRAY16_C(type) \
474 for (i=0; i<(dstW>>1); i++){\
480 type av_unused *r, *b, *g;\
483 for (j=0; j<lumFilterSize; j++)\
485 Y1 += lumSrc[j][i2] * lumFilter[j];\
486 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
490 if ((Y1|Y2|U|V)&65536)\
492 if (Y1>65535) Y1=65535; \
493 else if (Y1<0)Y1=0; \
494 if (Y2>65535) Y2=65535; \
495 else if (Y2<0)Y2=0; \
498 #define YSCALE_YUV_2_RGBX_C(type) \
499 YSCALE_YUV_2_PACKEDX_C(type) \
500 r = (type *)c->table_rV[V]; \
501 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
502 b = (type *)c->table_bU[U]; \
504 #define YSCALE_YUV_2_PACKED2_C \
505 for (i=0; i<(dstW>>1); i++){ \
507 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
508 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
509 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
510 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
512 #define YSCALE_YUV_2_GRAY16_2_C \
513 for (i=0; i<(dstW>>1); i++){ \
515 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
516 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
518 #define YSCALE_YUV_2_RGB2_C(type) \
519 YSCALE_YUV_2_PACKED2_C\
521 r = (type *)c->table_rV[V];\
522 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
523 b = (type *)c->table_bU[U];\
525 #define YSCALE_YUV_2_PACKED1_C \
526 for (i=0; i<(dstW>>1); i++){\
528 int Y1= buf0[i2 ]>>7;\
529 int Y2= buf0[i2+1]>>7;\
530 int U= (uvbuf1[i ])>>7;\
531 int V= (uvbuf1[i+VOFW])>>7;\
533 #define YSCALE_YUV_2_GRAY16_1_C \
534 for (i=0; i<(dstW>>1); i++){\
536 int Y1= buf0[i2 ]<<1;\
537 int Y2= buf0[i2+1]<<1;\
539 #define YSCALE_YUV_2_RGB1_C(type) \
540 YSCALE_YUV_2_PACKED1_C\
542 r = (type *)c->table_rV[V];\
543 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
544 b = (type *)c->table_bU[U];\
546 #define YSCALE_YUV_2_PACKED1B_C \
547 for (i=0; i<(dstW>>1); i++){\
549 int Y1= buf0[i2 ]>>7;\
550 int Y2= buf0[i2+1]>>7;\
551 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
552 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
554 #define YSCALE_YUV_2_RGB1B_C(type) \
555 YSCALE_YUV_2_PACKED1B_C\
557 r = (type *)c->table_rV[V];\
558 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
559 b = (type *)c->table_bU[U];\
561 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16)\
562 switch(c->dstFormat)\
566 case PIX_FMT_RGB32_1:\
567 case PIX_FMT_BGR32_1:\
569 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
570 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
575 ((uint8_t*)dest)[0]= r[Y1];\
576 ((uint8_t*)dest)[1]= g[Y1];\
577 ((uint8_t*)dest)[2]= b[Y1];\
578 ((uint8_t*)dest)[3]= r[Y2];\
579 ((uint8_t*)dest)[4]= g[Y2];\
580 ((uint8_t*)dest)[5]= b[Y2];\
586 ((uint8_t*)dest)[0]= b[Y1];\
587 ((uint8_t*)dest)[1]= g[Y1];\
588 ((uint8_t*)dest)[2]= r[Y1];\
589 ((uint8_t*)dest)[3]= b[Y2];\
590 ((uint8_t*)dest)[4]= g[Y2];\
591 ((uint8_t*)dest)[5]= r[Y2];\
595 case PIX_FMT_RGB565:\
596 case PIX_FMT_BGR565:\
598 const int dr1= dither_2x2_8[y&1 ][0];\
599 const int dg1= dither_2x2_4[y&1 ][0];\
600 const int db1= dither_2x2_8[(y&1)^1][0];\
601 const int dr2= dither_2x2_8[y&1 ][1];\
602 const int dg2= dither_2x2_4[y&1 ][1];\
603 const int db2= dither_2x2_8[(y&1)^1][1];\
605 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
606 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
610 case PIX_FMT_RGB555:\
611 case PIX_FMT_BGR555:\
613 const int dr1= dither_2x2_8[y&1 ][0];\
614 const int dg1= dither_2x2_8[y&1 ][1];\
615 const int db1= dither_2x2_8[(y&1)^1][0];\
616 const int dr2= dither_2x2_8[y&1 ][1];\
617 const int dg2= dither_2x2_8[y&1 ][0];\
618 const int db2= dither_2x2_8[(y&1)^1][1];\
620 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
621 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
628 const uint8_t * const d64= dither_8x8_73[y&7];\
629 const uint8_t * const d32= dither_8x8_32[y&7];\
631 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
632 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
639 const uint8_t * const d64= dither_8x8_73 [y&7];\
640 const uint8_t * const d128=dither_8x8_220[y&7];\
642 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
643 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
647 case PIX_FMT_RGB4_BYTE:\
648 case PIX_FMT_BGR4_BYTE:\
650 const uint8_t * const d64= dither_8x8_73 [y&7];\
651 const uint8_t * const d128=dither_8x8_220[y&7];\
653 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
654 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
658 case PIX_FMT_MONOBLACK:\
660 const uint8_t * const d128=dither_8x8_220[y&7];\
661 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
662 for (i=0; i<dstW-7; i+=8){\
664 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
665 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
666 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
667 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
668 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
669 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
670 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
671 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
672 ((uint8_t*)dest)[0]= acc;\
677 ((uint8_t*)dest)-= dstW>>4;\
681 static int top[1024];\
682 static int last_new[1024][1024];\
683 static int last_in3[1024][1024];\
684 static int drift[1024][1024];\
688 const uint8_t * const d128=dither_8x8_220[y&7];\
693 for (i=dstW>>1; i<dstW; i++){\
694 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
695 int in2 = (76309 * (in - 16) + 32768) >> 16;\
696 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
697 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
698 + (last_new[y][i] - in3)*f/256;\
699 int new= old> 128 ? 255 : 0;\
701 error_new+= FFABS(last_new[y][i] - new);\
702 error_in3+= FFABS(last_in3[y][i] - in3);\
703 f= error_new - error_in3*4;\
708 left= top[i]= old - new;\
709 last_new[y][i]= new;\
710 last_in3[y][i]= in3;\
712 acc+= acc + (new&1);\
714 ((uint8_t*)dest)[0]= acc;\
722 case PIX_FMT_YUYV422:\
724 ((uint8_t*)dest)[2*i2+0]= Y1;\
725 ((uint8_t*)dest)[2*i2+1]= U;\
726 ((uint8_t*)dest)[2*i2+2]= Y2;\
727 ((uint8_t*)dest)[2*i2+3]= V;\
730 case PIX_FMT_UYVY422:\
732 ((uint8_t*)dest)[2*i2+0]= U;\
733 ((uint8_t*)dest)[2*i2+1]= Y1;\
734 ((uint8_t*)dest)[2*i2+2]= V;\
735 ((uint8_t*)dest)[2*i2+3]= Y2;\
738 case PIX_FMT_GRAY16BE:\
740 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
741 ((uint8_t*)dest)[2*i2+1]= Y1;\
742 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
743 ((uint8_t*)dest)[2*i2+3]= Y2;\
746 case PIX_FMT_GRAY16LE:\
748 ((uint8_t*)dest)[2*i2+0]= Y1;\
749 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
750 ((uint8_t*)dest)[2*i2+2]= Y2;\
751 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
757 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
758 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
759 uint8_t *dest, int dstW, int y)
766 case PIX_FMT_BGR32_1:
767 case PIX_FMT_RGB32_1:
768 YSCALE_YUV_2_RGBX_C(uint32_t)
769 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
770 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
774 YSCALE_YUV_2_RGBX_C(uint8_t)
775 ((uint8_t*)dest)[0]= r[Y1];
776 ((uint8_t*)dest)[1]= g[Y1];
777 ((uint8_t*)dest)[2]= b[Y1];
778 ((uint8_t*)dest)[3]= r[Y2];
779 ((uint8_t*)dest)[4]= g[Y2];
780 ((uint8_t*)dest)[5]= b[Y2];
785 YSCALE_YUV_2_RGBX_C(uint8_t)
786 ((uint8_t*)dest)[0]= b[Y1];
787 ((uint8_t*)dest)[1]= g[Y1];
788 ((uint8_t*)dest)[2]= r[Y1];
789 ((uint8_t*)dest)[3]= b[Y2];
790 ((uint8_t*)dest)[4]= g[Y2];
791 ((uint8_t*)dest)[5]= r[Y2];
798 const int dr1= dither_2x2_8[y&1 ][0];
799 const int dg1= dither_2x2_4[y&1 ][0];
800 const int db1= dither_2x2_8[(y&1)^1][0];
801 const int dr2= dither_2x2_8[y&1 ][1];
802 const int dg2= dither_2x2_4[y&1 ][1];
803 const int db2= dither_2x2_8[(y&1)^1][1];
804 YSCALE_YUV_2_RGBX_C(uint16_t)
805 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
806 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
813 const int dr1= dither_2x2_8[y&1 ][0];
814 const int dg1= dither_2x2_8[y&1 ][1];
815 const int db1= dither_2x2_8[(y&1)^1][0];
816 const int dr2= dither_2x2_8[y&1 ][1];
817 const int dg2= dither_2x2_8[y&1 ][0];
818 const int db2= dither_2x2_8[(y&1)^1][1];
819 YSCALE_YUV_2_RGBX_C(uint16_t)
820 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
821 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
828 const uint8_t * const d64= dither_8x8_73[y&7];
829 const uint8_t * const d32= dither_8x8_32[y&7];
830 YSCALE_YUV_2_RGBX_C(uint8_t)
831 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
832 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
839 const uint8_t * const d64= dither_8x8_73 [y&7];
840 const uint8_t * const d128=dither_8x8_220[y&7];
841 YSCALE_YUV_2_RGBX_C(uint8_t)
842 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
843 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
847 case PIX_FMT_RGB4_BYTE:
848 case PIX_FMT_BGR4_BYTE:
850 const uint8_t * const d64= dither_8x8_73 [y&7];
851 const uint8_t * const d128=dither_8x8_220[y&7];
852 YSCALE_YUV_2_RGBX_C(uint8_t)
853 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
854 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
858 case PIX_FMT_MONOBLACK:
860 const uint8_t * const d128=dither_8x8_220[y&7];
861 uint8_t *g= c->table_gU[128] + c->table_gV[128];
863 for (i=0; i<dstW-1; i+=2){
868 for (j=0; j<lumFilterSize; j++)
870 Y1 += lumSrc[j][i] * lumFilter[j];
871 Y2 += lumSrc[j][i+1] * lumFilter[j];
882 acc+= acc + g[Y1+d128[(i+0)&7]];
883 acc+= acc + g[Y2+d128[(i+1)&7]];
885 ((uint8_t*)dest)[0]= acc;
891 case PIX_FMT_YUYV422:
892 YSCALE_YUV_2_PACKEDX_C(void)
893 ((uint8_t*)dest)[2*i2+0]= Y1;
894 ((uint8_t*)dest)[2*i2+1]= U;
895 ((uint8_t*)dest)[2*i2+2]= Y2;
896 ((uint8_t*)dest)[2*i2+3]= V;
899 case PIX_FMT_UYVY422:
900 YSCALE_YUV_2_PACKEDX_C(void)
901 ((uint8_t*)dest)[2*i2+0]= U;
902 ((uint8_t*)dest)[2*i2+1]= Y1;
903 ((uint8_t*)dest)[2*i2+2]= V;
904 ((uint8_t*)dest)[2*i2+3]= Y2;
907 case PIX_FMT_GRAY16BE:
908 YSCALE_YUV_2_GRAY16_C(void)
909 ((uint8_t*)dest)[2*i2+0]= Y1>>8;
910 ((uint8_t*)dest)[2*i2+1]= Y1;
911 ((uint8_t*)dest)[2*i2+2]= Y2>>8;
912 ((uint8_t*)dest)[2*i2+3]= Y2;
915 case PIX_FMT_GRAY16LE:
916 YSCALE_YUV_2_GRAY16_C(void)
917 ((uint8_t*)dest)[2*i2+0]= Y1;
918 ((uint8_t*)dest)[2*i2+1]= Y1>>8;
919 ((uint8_t*)dest)[2*i2+2]= Y2;
920 ((uint8_t*)dest)[2*i2+3]= Y2>>8;
927 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
929 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
934 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
935 #define COMPILE_ALTIVEC
936 #endif //HAVE_ALTIVEC
937 #endif //ARCH_POWERPC
939 #if defined(ARCH_X86)
941 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
945 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
949 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
950 #define COMPILE_3DNOW
952 #endif //ARCH_X86 || ARCH_X86_64
963 #define RENAME(a) a ## _C
964 #include "swscale_template.c"
967 #ifdef COMPILE_ALTIVEC
970 #define RENAME(a) a ## _altivec
971 #include "swscale_template.c"
974 #if defined(ARCH_X86)
983 #define RENAME(a) a ## _X86
984 #include "swscale_template.c"
992 #define RENAME(a) a ## _MMX
993 #include "swscale_template.c"
1002 #define RENAME(a) a ## _MMX2
1003 #include "swscale_template.c"
1007 #ifdef COMPILE_3DNOW
1012 #define RENAME(a) a ## _3DNow
1013 #include "swscale_template.c"
1016 #endif //ARCH_X86 || ARCH_X86_64
1018 // minor note: the HAVE_xyz is messed up after that line so don't use it
1020 static double getSplineCoeff(double a, double b, double c, double d, double dist)
1022 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
1023 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
1024 else return getSplineCoeff( 0.0,
1031 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1032 int srcW, int dstW, int filterAlign, int one, int flags,
1033 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1039 double *filter=NULL;
1040 double *filter2=NULL;
1042 #if defined(ARCH_X86)
1043 if (flags & SWS_CPU_CAPS_MMX)
1044 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1047 // Note the +1 is for the MMXscaler which reads over the end
1048 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1050 if (FFABS(xInc - 0x10000) <10) // unscaled
1054 filter= av_malloc(dstW*sizeof(double)*filterSize);
1055 for (i=0; i<dstW*filterSize; i++) filter[i]=0;
1057 for (i=0; i<dstW; i++)
1059 filter[i*filterSize]=1;
1064 else if (flags&SWS_POINT) // lame looking point sampling mode
1069 filter= av_malloc(dstW*sizeof(double)*filterSize);
1071 xDstInSrc= xInc/2 - 0x8000;
1072 for (i=0; i<dstW; i++)
1074 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1076 (*filterPos)[i]= xx;
1081 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1085 if (flags&SWS_BICUBIC) filterSize= 4;
1086 else if (flags&SWS_X ) filterSize= 4;
1087 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1088 filter= av_malloc(dstW*sizeof(double)*filterSize);
1090 xDstInSrc= xInc/2 - 0x8000;
1091 for (i=0; i<dstW; i++)
1093 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1096 (*filterPos)[i]= xx;
1097 //Bilinear upscale / linear interpolate / Area averaging
1098 for (j=0; j<filterSize; j++)
1100 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1101 double coeff= 1.0 - d;
1102 if (coeff<0) coeff=0;
1103 filter[i*filterSize + j]= coeff;
1112 double sizeFactor, filterSizeInSrc;
1113 const double xInc1= (double)xInc / (double)(1<<16);
1115 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1116 else if (flags&SWS_X) sizeFactor= 8.0;
1117 else if (flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1118 else if (flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1119 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1120 else if (flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1121 else if (flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1122 else if (flags&SWS_BILINEAR) sizeFactor= 2.0;
1124 sizeFactor= 0.0; //GCC warning killer
1128 if (xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1129 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1131 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1132 if (filterSize > srcW-2) filterSize=srcW-2;
1134 filter= av_malloc(dstW*sizeof(double)*filterSize);
1136 xDstInSrc= xInc1 / 2.0 - 0.5;
1137 for (i=0; i<dstW; i++)
1139 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1141 (*filterPos)[i]= xx;
1142 for (j=0; j<filterSize; j++)
1144 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1146 if (flags & SWS_BICUBIC)
1148 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1149 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1152 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1154 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1158 /* else if (flags & SWS_X)
1160 double p= param ? param*0.01 : 0.3;
1161 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1162 coeff*= pow(2.0, - p*d*d);
1164 else if (flags & SWS_X)
1166 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1172 if (coeff<0.0) coeff= -pow(-coeff, A);
1173 else coeff= pow( coeff, A);
1174 coeff= coeff*0.5 + 0.5;
1176 else if (flags & SWS_AREA)
1178 double srcPixelSize= 1.0/xInc1;
1179 if (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1180 else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1183 else if (flags & SWS_GAUSS)
1185 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1186 coeff = pow(2.0, - p*d*d);
1188 else if (flags & SWS_SINC)
1190 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1192 else if (flags & SWS_LANCZOS)
1194 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1195 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1198 else if (flags & SWS_BILINEAR)
1201 if (coeff<0) coeff=0;
1203 else if (flags & SWS_SPLINE)
1205 double p=-2.196152422706632;
1206 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1209 coeff= 0.0; //GCC warning killer
1213 filter[i*filterSize + j]= coeff;
1220 /* apply src & dst Filter to filter -> filter2
1223 assert(filterSize>0);
1224 filter2Size= filterSize;
1225 if (srcFilter) filter2Size+= srcFilter->length - 1;
1226 if (dstFilter) filter2Size+= dstFilter->length - 1;
1227 assert(filter2Size>0);
1228 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1230 for (i=0; i<dstW; i++)
1233 SwsVector scaleFilter;
1236 scaleFilter.coeff= filter + i*filterSize;
1237 scaleFilter.length= filterSize;
1239 if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1240 else outVec= &scaleFilter;
1242 assert(outVec->length == filter2Size);
1245 for (j=0; j<outVec->length; j++)
1247 filter2[i*filter2Size + j]= outVec->coeff[j];
1250 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1252 if (outVec != &scaleFilter) sws_freeVec(outVec);
1256 /* try to reduce the filter-size (step1 find size and shift left) */
1257 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1259 for (i=dstW-1; i>=0; i--)
1261 int min= filter2Size;
1265 /* get rid off near zero elements on the left by shifting left */
1266 for (j=0; j<filter2Size; j++)
1269 cutOff += FFABS(filter2[i*filter2Size]);
1271 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1273 /* preserve monotonicity because the core can't handle the filter otherwise */
1274 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1276 // Move filter coeffs left
1277 for (k=1; k<filter2Size; k++)
1278 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1279 filter2[i*filter2Size + k - 1]= 0.0;
1284 /* count near zeros on the right */
1285 for (j=filter2Size-1; j>0; j--)
1287 cutOff += FFABS(filter2[i*filter2Size + j]);
1289 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1293 if (min>minFilterSize) minFilterSize= min;
1296 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1297 // we can handle the special case 4,
1298 // so we don't want to go to the full 8
1299 if (minFilterSize < 5)
1302 // we really don't want to waste our time
1303 // doing useless computation, so fall-back on
1304 // the scalar C code for very small filter.
1305 // vectorizing is worth it only if you have
1306 // decent-sized vector.
1307 if (minFilterSize < 3)
1311 if (flags & SWS_CPU_CAPS_MMX) {
1312 // special case for unscaled vertical filtering
1313 if (minFilterSize == 1 && filterAlign == 2)
1317 assert(minFilterSize > 0);
1318 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1319 assert(filterSize > 0);
1320 filter= av_malloc(filterSize*dstW*sizeof(double));
1321 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
1323 *outFilterSize= filterSize;
1325 if (flags&SWS_PRINT_INFO)
1326 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1327 /* try to reduce the filter-size (step2 reduce it) */
1328 for (i=0; i<dstW; i++)
1332 for (j=0; j<filterSize; j++)
1334 if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1335 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1340 //FIXME try to align filterpos if possible
1343 for (i=0; i<dstW; i++)
1346 if ((*filterPos)[i] < 0)
1348 // Move filter coeffs left to compensate for filterPos
1349 for (j=1; j<filterSize; j++)
1351 int left= FFMAX(j + (*filterPos)[i], 0);
1352 filter[i*filterSize + left] += filter[i*filterSize + j];
1353 filter[i*filterSize + j]=0;
1358 if ((*filterPos)[i] + filterSize > srcW)
1360 int shift= (*filterPos)[i] + filterSize - srcW;
1361 // Move filter coeffs right to compensate for filterPos
1362 for (j=filterSize-2; j>=0; j--)
1364 int right= FFMIN(j + shift, filterSize-1);
1365 filter[i*filterSize +right] += filter[i*filterSize +j];
1366 filter[i*filterSize +j]=0;
1368 (*filterPos)[i]= srcW - filterSize;
1372 // Note the +1 is for the MMXscaler which reads over the end
1373 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1374 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1376 /* Normalize & Store in outFilter */
1377 for (i=0; i<dstW; i++)
1384 for (j=0; j<filterSize; j++)
1386 sum+= filter[i*filterSize + j];
1389 for (j=0; j<*outFilterSize; j++)
1391 double v= filter[i*filterSize + j]*scale + error;
1392 int intV= floor(v + 0.5);
1393 (*outFilter)[i*(*outFilterSize) + j]= intV;
1398 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1399 for (i=0; i<*outFilterSize; i++)
1401 int j= dstW*(*outFilterSize);
1402 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1413 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1416 long imm8OfPShufW1A;
1417 long imm8OfPShufW2A;
1418 long fragmentLengthA;
1420 long imm8OfPShufW1B;
1421 long imm8OfPShufW2B;
1422 long fragmentLengthB;
1427 // create an optimized horizontal scaling routine
1435 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1436 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1437 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1438 "punpcklbw %%mm7, %%mm1 \n\t"
1439 "punpcklbw %%mm7, %%mm0 \n\t"
1440 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1442 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1444 "psubw %%mm1, %%mm0 \n\t"
1445 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1446 "pmullw %%mm3, %%mm0 \n\t"
1447 "psllw $7, %%mm1 \n\t"
1448 "paddw %%mm1, %%mm0 \n\t"
1450 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1452 "add $8, %%"REG_a" \n\t"
1456 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1457 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1458 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1463 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1467 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1468 "=r" (fragmentLengthA)
1475 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1476 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1477 "punpcklbw %%mm7, %%mm0 \n\t"
1478 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1480 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1482 "psubw %%mm1, %%mm0 \n\t"
1483 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1484 "pmullw %%mm3, %%mm0 \n\t"
1485 "psllw $7, %%mm1 \n\t"
1486 "paddw %%mm1, %%mm0 \n\t"
1488 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1490 "add $8, %%"REG_a" \n\t"
1494 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1495 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1496 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1501 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1505 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1506 "=r" (fragmentLengthB)
1509 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1512 for (i=0; i<dstW/numSplits; i++)
1519 int b=((xpos+xInc)>>16) - xx;
1520 int c=((xpos+xInc*2)>>16) - xx;
1521 int d=((xpos+xInc*3)>>16) - xx;
1523 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1524 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1525 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1526 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1531 int maxShift= 3-(d+1);
1534 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1536 funnyCode[fragmentPos + imm8OfPShufW1B]=
1537 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1538 funnyCode[fragmentPos + imm8OfPShufW2B]=
1539 a | (b<<2) | (c<<4) | (d<<6);
1541 if (i+3>=dstW) shift=maxShift; //avoid overread
1542 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1544 if (shift && i>=shift)
1546 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1547 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1548 filterPos[i/2]-=shift;
1551 fragmentPos+= fragmentLengthB;
1558 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1560 funnyCode[fragmentPos + imm8OfPShufW1A]=
1561 funnyCode[fragmentPos + imm8OfPShufW2A]=
1562 a | (b<<2) | (c<<4) | (d<<6);
1564 if (i+4>=dstW) shift=maxShift; //avoid overread
1565 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1567 if (shift && i>=shift)
1569 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1570 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1571 filterPos[i/2]-=shift;
1574 fragmentPos+= fragmentLengthA;
1577 funnyCode[fragmentPos]= RET;
1581 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1583 #endif /* COMPILE_MMX2 */
1585 static void globalInit(void){
1586 // generating tables:
1588 for (i=0; i<768; i++){
1589 int c= av_clip_uint8(i-256);
1594 static SwsFunc getSwsFunc(int flags){
1596 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1597 #if defined(ARCH_X86)
1598 // ordered per speed fastest first
1599 if (flags & SWS_CPU_CAPS_MMX2)
1600 return swScale_MMX2;
1601 else if (flags & SWS_CPU_CAPS_3DNOW)
1602 return swScale_3DNow;
1603 else if (flags & SWS_CPU_CAPS_MMX)
1610 if (flags & SWS_CPU_CAPS_ALTIVEC)
1611 return swScale_altivec;
1616 #endif /* defined(ARCH_X86) */
1617 #else //RUNTIME_CPUDETECT
1619 return swScale_MMX2;
1620 #elif defined (HAVE_3DNOW)
1621 return swScale_3DNow;
1622 #elif defined (HAVE_MMX)
1624 #elif defined (HAVE_ALTIVEC)
1625 return swScale_altivec;
1629 #endif //!RUNTIME_CPUDETECT
1632 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1633 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1634 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1636 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1637 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1641 uint8_t *srcPtr= src[0];
1642 uint8_t *dstPtr= dst;
1643 for (i=0; i<srcSliceH; i++)
1645 memcpy(dstPtr, srcPtr, c->srcW);
1646 srcPtr+= srcStride[0];
1647 dstPtr+= dstStride[0];
1650 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1651 if (c->dstFormat == PIX_FMT_NV12)
1652 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1654 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1659 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1660 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1661 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1663 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1668 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1669 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1670 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1672 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1677 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1678 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1679 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1681 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1686 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1687 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1688 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1690 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1695 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1696 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1697 int srcSliceH, uint8_t* dst[], int dstStride[]){
1698 const int srcFormat= c->srcFormat;
1699 const int dstFormat= c->dstFormat;
1700 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1701 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1702 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1703 const int dstId= fmt_depth(dstFormat) >> 2;
1704 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1707 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1708 || (isRGB(srcFormat) && isRGB(dstFormat))){
1709 switch(srcId | (dstId<<4)){
1710 case 0x34: conv= rgb16to15; break;
1711 case 0x36: conv= rgb24to15; break;
1712 case 0x38: conv= rgb32to15; break;
1713 case 0x43: conv= rgb15to16; break;
1714 case 0x46: conv= rgb24to16; break;
1715 case 0x48: conv= rgb32to16; break;
1716 case 0x63: conv= rgb15to24; break;
1717 case 0x64: conv= rgb16to24; break;
1718 case 0x68: conv= rgb32to24; break;
1719 case 0x83: conv= rgb15to32; break;
1720 case 0x84: conv= rgb16to32; break;
1721 case 0x86: conv= rgb24to32; break;
1722 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1723 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1725 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1726 || (isRGB(srcFormat) && isBGR(dstFormat))){
1727 switch(srcId | (dstId<<4)){
1728 case 0x33: conv= rgb15tobgr15; break;
1729 case 0x34: conv= rgb16tobgr15; break;
1730 case 0x36: conv= rgb24tobgr15; break;
1731 case 0x38: conv= rgb32tobgr15; break;
1732 case 0x43: conv= rgb15tobgr16; break;
1733 case 0x44: conv= rgb16tobgr16; break;
1734 case 0x46: conv= rgb24tobgr16; break;
1735 case 0x48: conv= rgb32tobgr16; break;
1736 case 0x63: conv= rgb15tobgr24; break;
1737 case 0x64: conv= rgb16tobgr24; break;
1738 case 0x66: conv= rgb24tobgr24; break;
1739 case 0x68: conv= rgb32tobgr24; break;
1740 case 0x83: conv= rgb15tobgr32; break;
1741 case 0x84: conv= rgb16tobgr32; break;
1742 case 0x86: conv= rgb24tobgr32; break;
1743 case 0x88: conv= rgb32tobgr32; break;
1744 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1745 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1748 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1749 sws_format_name(srcFormat), sws_format_name(dstFormat));
1754 uint8_t *srcPtr= src[0];
1755 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1756 srcPtr += ALT32_CORR;
1758 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1759 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1763 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1765 for (i=0; i<srcSliceH; i++)
1767 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1768 srcPtr+= srcStride[0];
1769 dstPtr+= dstStride[0];
1776 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1777 int srcSliceH, uint8_t* dst[], int dstStride[]){
1781 dst[0]+ srcSliceY *dstStride[0],
1782 dst[1]+(srcSliceY>>1)*dstStride[1],
1783 dst[2]+(srcSliceY>>1)*dstStride[2],
1785 dstStride[0], dstStride[1], srcStride[0]);
1789 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1790 int srcSliceH, uint8_t* dst[], int dstStride[]){
1794 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1795 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1797 uint8_t *srcPtr= src[0];
1798 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1800 for (i=0; i<srcSliceH; i++)
1802 memcpy(dstPtr, srcPtr, c->srcW);
1803 srcPtr+= srcStride[0];
1804 dstPtr+= dstStride[0];
1808 if (c->dstFormat==PIX_FMT_YUV420P){
1809 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1810 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1812 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1813 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1818 /* unscaled copy like stuff (assumes nearly identical formats) */
1819 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1820 int srcSliceH, uint8_t* dst[], int dstStride[])
1822 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1823 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1827 uint8_t *srcPtr= src[0];
1828 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1831 /* universal length finder */
1832 while(length+c->srcW <= FFABS(dstStride[0])
1833 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1836 for (i=0; i<srcSliceH; i++)
1838 memcpy(dstPtr, srcPtr, length);
1839 srcPtr+= srcStride[0];
1840 dstPtr+= dstStride[0];
1846 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1847 int srcSliceH, uint8_t* dst[], int dstStride[])
1850 for (plane=0; plane<3; plane++)
1852 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1853 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1854 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1856 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1858 if (!isGray(c->dstFormat))
1859 memset(dst[plane], 128, dstStride[plane]*height);
1863 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1864 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1868 uint8_t *srcPtr= src[plane];
1869 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1870 for (i=0; i<height; i++)
1872 memcpy(dstPtr, srcPtr, length);
1873 srcPtr+= srcStride[plane];
1874 dstPtr+= dstStride[plane];
1882 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1883 int srcSliceH, uint8_t* dst[], int dstStride[]){
1885 int length= c->srcW;
1887 int height= srcSliceH;
1889 uint8_t *srcPtr= src[0];
1890 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1892 if (!isGray(c->dstFormat)){
1893 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1894 memset(dst[1], 128, dstStride[1]*height);
1895 memset(dst[2], 128, dstStride[2]*height);
1897 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1898 for (i=0; i<height; i++)
1900 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1901 srcPtr+= srcStride[0];
1902 dstPtr+= dstStride[0];
1907 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1908 int srcSliceH, uint8_t* dst[], int dstStride[]){
1910 int length= c->srcW;
1912 int height= srcSliceH;
1914 uint8_t *srcPtr= src[0];
1915 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1916 for (i=0; i<height; i++)
1918 for (j=0; j<length; j++)
1920 dstPtr[j<<1] = srcPtr[j];
1921 dstPtr[(j<<1)+1] = srcPtr[j];
1923 srcPtr+= srcStride[0];
1924 dstPtr+= dstStride[0];
1929 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1930 int srcSliceH, uint8_t* dst[], int dstStride[]){
1932 int length= c->srcW;
1934 int height= srcSliceH;
1936 uint16_t *srcPtr= (uint16_t*)src[0];
1937 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1938 for (i=0; i<height; i++)
1940 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1941 srcPtr+= srcStride[0]/2;
1942 dstPtr+= dstStride[0]/2;
1948 static void getSubSampleFactors(int *h, int *v, int format){
1950 case PIX_FMT_UYVY422:
1951 case PIX_FMT_YUYV422:
1955 case PIX_FMT_YUV420P:
1956 case PIX_FMT_YUVA420P:
1957 case PIX_FMT_GRAY16BE:
1958 case PIX_FMT_GRAY16LE:
1959 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1965 case PIX_FMT_YUV440P:
1969 case PIX_FMT_YUV410P:
1973 case PIX_FMT_YUV444P:
1977 case PIX_FMT_YUV422P:
1981 case PIX_FMT_YUV411P:
1992 static uint16_t roundToInt16(int64_t f){
1993 int r= (f + (1<<15))>>16;
1994 if (r<-0x7FFF) return 0x8000;
1995 else if (r> 0x7FFF) return 0x7FFF;
2000 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
2001 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
2002 * @return -1 if not supported
2004 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
2005 int64_t crv = inv_table[0];
2006 int64_t cbu = inv_table[1];
2007 int64_t cgu = -inv_table[2];
2008 int64_t cgv = -inv_table[3];
2012 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
2013 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
2015 c->brightness= brightness;
2016 c->contrast = contrast;
2017 c->saturation= saturation;
2018 c->srcRange = srcRange;
2019 c->dstRange = dstRange;
2020 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return 0;
2022 c->uOffset= 0x0400040004000400LL;
2023 c->vOffset= 0x0400040004000400LL;
2029 crv= (crv*224) / 255;
2030 cbu= (cbu*224) / 255;
2031 cgu= (cgu*224) / 255;
2032 cgv= (cgv*224) / 255;
2035 cy = (cy *contrast )>>16;
2036 crv= (crv*contrast * saturation)>>32;
2037 cbu= (cbu*contrast * saturation)>>32;
2038 cgu= (cgu*contrast * saturation)>>32;
2039 cgv= (cgv*contrast * saturation)>>32;
2041 oy -= 256*brightness;
2043 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2044 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2045 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2046 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2047 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2048 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2050 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2053 #ifdef COMPILE_ALTIVEC
2054 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2055 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2061 * @return -1 if not supported
2063 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2064 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2066 *inv_table = c->srcColorspaceTable;
2067 *table = c->dstColorspaceTable;
2068 *srcRange = c->srcRange;
2069 *dstRange = c->dstRange;
2070 *brightness= c->brightness;
2071 *contrast = c->contrast;
2072 *saturation= c->saturation;
2077 static int handle_jpeg(int *format)
2080 case PIX_FMT_YUVJ420P:
2081 *format = PIX_FMT_YUV420P;
2083 case PIX_FMT_YUVJ422P:
2084 *format = PIX_FMT_YUV422P;
2086 case PIX_FMT_YUVJ444P:
2087 *format = PIX_FMT_YUV444P;
2089 case PIX_FMT_YUVJ440P:
2090 *format = PIX_FMT_YUV440P;
2097 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2098 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2102 int usesVFilter, usesHFilter;
2103 int unscaled, needsDither;
2104 int srcRange, dstRange;
2105 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2106 #if defined(ARCH_X86)
2107 if (flags & SWS_CPU_CAPS_MMX)
2108 asm volatile("emms\n\t"::: "memory");
2111 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2112 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2114 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2115 #elif defined (HAVE_3DNOW)
2116 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2117 #elif defined (HAVE_MMX)
2118 flags |= SWS_CPU_CAPS_MMX;
2119 #elif defined (HAVE_ALTIVEC)
2120 flags |= SWS_CPU_CAPS_ALTIVEC;
2121 #elif defined (ARCH_BFIN)
2122 flags |= SWS_CPU_CAPS_BFIN;
2124 #endif /* RUNTIME_CPUDETECT */
2125 if (clip_table[512] != 255) globalInit();
2126 if (!rgb15to16) sws_rgb2rgb_init(flags);
2128 unscaled = (srcW == dstW && srcH == dstH);
2129 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2130 && (fmt_depth(dstFormat))<24
2131 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2133 srcRange = handle_jpeg(&srcFormat);
2134 dstRange = handle_jpeg(&dstFormat);
2136 if (!isSupportedIn(srcFormat))
2138 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2141 if (!isSupportedOut(dstFormat))
2143 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2147 i= flags & ( SWS_POINT
2158 if(!i || (i & (i-1)))
2160 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2166 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
2168 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2169 srcW, srcH, dstW, dstH);
2172 if(srcW > VOFW || dstW > VOFW){
2173 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2177 if (!dstFilter) dstFilter= &dummyFilter;
2178 if (!srcFilter) srcFilter= &dummyFilter;
2180 c= av_mallocz(sizeof(SwsContext));
2182 c->av_class = &sws_context_class;
2187 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2188 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2190 c->dstFormat= dstFormat;
2191 c->srcFormat= srcFormat;
2192 c->vRounder= 4* 0x0001000100010001ULL;
2194 usesHFilter= usesVFilter= 0;
2195 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2196 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2197 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2198 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2199 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2200 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2201 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2202 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2204 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2205 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2207 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2208 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2210 // drop some chroma lines if the user wants it
2211 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2212 c->chrSrcVSubSample+= c->vChrDrop;
2214 // drop every 2. pixel for chroma calculation unless user wants full chroma
2215 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2216 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2217 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2218 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
2219 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2220 c->chrSrcHSubSample=1;
2223 c->param[0] = param[0];
2224 c->param[1] = param[1];
2227 c->param[1] = SWS_PARAM_DEFAULT;
2230 c->chrIntHSubSample= c->chrDstHSubSample;
2231 c->chrIntVSubSample= c->chrSrcVSubSample;
2233 // Note the -((-x)>>y) is so that we always round toward +inf.
2234 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2235 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2236 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2237 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2239 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);
2241 /* unscaled special Cases */
2242 if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isBGR(dstFormat) || isRGB(dstFormat)))
2245 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2247 c->swScale= PlanarToNV12Wrapper;
2251 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2253 c->swScale= yuv2rgb_get_func_ptr(c);
2257 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2259 c->swScale= yvu9toyv12Wrapper;
2263 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2264 c->swScale= bgr24toyv12Wrapper;
2266 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2267 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2268 && (isBGR(dstFormat) || isRGB(dstFormat))
2269 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2270 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2271 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2272 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2273 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2274 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2275 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2276 && dstFormat != PIX_FMT_RGB32_1
2277 && dstFormat != PIX_FMT_BGR32_1
2278 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2279 c->swScale= rgb2rgbWrapper;
2281 if (srcFormat == PIX_FMT_YUV422P)
2283 if (dstFormat == PIX_FMT_YUYV422)
2284 c->swScale= YUV422PToYuy2Wrapper;
2285 else if (dstFormat == PIX_FMT_UYVY422)
2286 c->swScale= YUV422PToUyvyWrapper;
2289 /* LQ converters if -sws 0 or -sws 4*/
2290 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2292 if (srcFormat == PIX_FMT_YUV420P)
2294 if (dstFormat == PIX_FMT_YUYV422)
2295 c->swScale= PlanarToYuy2Wrapper;
2296 else if (dstFormat == PIX_FMT_UYVY422)
2297 c->swScale= PlanarToUyvyWrapper;
2301 #ifdef COMPILE_ALTIVEC
2302 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2303 srcFormat == PIX_FMT_YUV420P) {
2304 // unscaled YV12 -> packed YUV, we want speed
2305 if (dstFormat == PIX_FMT_YUYV422)
2306 c->swScale= yv12toyuy2_unscaled_altivec;
2307 else if (dstFormat == PIX_FMT_UYVY422)
2308 c->swScale= yv12touyvy_unscaled_altivec;
2313 if ( srcFormat == dstFormat
2314 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2315 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2317 if (isPacked(c->srcFormat))
2318 c->swScale= packedCopy;
2319 else /* Planar YUV or gray */
2320 c->swScale= planarCopy;
2323 /* gray16{le,be} conversions */
2324 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2326 c->swScale= gray16togray;
2328 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2330 c->swScale= graytogray16;
2332 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2334 c->swScale= gray16swap;
2338 if (flags & SWS_CPU_CAPS_BFIN)
2339 ff_bfin_get_unscaled_swscale (c);
2343 if (flags&SWS_PRINT_INFO)
2344 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2345 sws_format_name(srcFormat), sws_format_name(dstFormat));
2350 if (flags & SWS_CPU_CAPS_MMX2)
2352 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2353 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2355 if (flags&SWS_PRINT_INFO)
2356 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2358 if (usesHFilter) c->canMMX2BeUsed=0;
2363 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2364 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2366 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2367 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2368 // n-2 is the last chrominance sample available
2369 // this is not perfect, but no one should notice the difference, the more correct variant
2370 // would be like the vertical one, but that would require some special code for the
2371 // first and last pixel
2372 if (flags&SWS_FAST_BILINEAR)
2374 if (c->canMMX2BeUsed)
2379 //we don't use the x86asm scaler if mmx is available
2380 else if (flags & SWS_CPU_CAPS_MMX)
2382 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2383 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2387 /* precalculate horizontal scaler filter coefficients */
2389 const int filterAlign=
2390 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2391 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2394 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2395 srcW , dstW, filterAlign, 1<<14,
2396 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2397 srcFilter->lumH, dstFilter->lumH, c->param);
2398 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2399 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2400 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2401 srcFilter->chrH, dstFilter->chrH, c->param);
2403 #define MAX_FUNNY_CODE_SIZE 10000
2404 #if defined(COMPILE_MMX2)
2405 // can't downscale !!!
2406 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2408 #ifdef MAP_ANONYMOUS
2409 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2410 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2412 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2413 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2416 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2417 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2418 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2419 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2421 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2422 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2424 #endif /* defined(COMPILE_MMX2) */
2425 } // Init Horizontal stuff
2429 /* precalculate vertical scaler filter coefficients */
2431 const int filterAlign=
2432 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2433 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2436 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2437 srcH , dstH, filterAlign, (1<<12)-4,
2438 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2439 srcFilter->lumV, dstFilter->lumV, c->param);
2440 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2441 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2442 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2443 srcFilter->chrV, dstFilter->chrV, c->param);
2446 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2447 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2449 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2451 short *p = (short *)&c->vYCoeffsBank[i];
2453 p[j] = c->vLumFilter[i];
2456 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2458 short *p = (short *)&c->vCCoeffsBank[i];
2460 p[j] = c->vChrFilter[i];
2465 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2466 c->vLumBufSize= c->vLumFilterSize;
2467 c->vChrBufSize= c->vChrFilterSize;
2468 for (i=0; i<dstH; i++)
2470 int chrI= i*c->chrDstH / dstH;
2471 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2472 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2474 nextSlice>>= c->chrSrcVSubSample;
2475 nextSlice<<= c->chrSrcVSubSample;
2476 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2477 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2478 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2479 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2482 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2483 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2484 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2485 //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)
2486 /* align at 16 bytes for AltiVec */
2487 for (i=0; i<c->vLumBufSize; i++)
2488 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2489 for (i=0; i<c->vChrBufSize; i++)
2490 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2492 //try to avoid drawing green stuff between the right end and the stride end
2493 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2495 assert(2*VOFW == VOF);
2497 assert(c->chrDstH <= dstH);
2499 if (flags&SWS_PRINT_INFO)
2502 const char *dither= " dithered";
2504 const char *dither= "";
2506 if (flags&SWS_FAST_BILINEAR)
2507 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2508 else if (flags&SWS_BILINEAR)
2509 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2510 else if (flags&SWS_BICUBIC)
2511 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2512 else if (flags&SWS_X)
2513 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2514 else if (flags&SWS_POINT)
2515 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2516 else if (flags&SWS_AREA)
2517 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2518 else if (flags&SWS_BICUBLIN)
2519 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2520 else if (flags&SWS_GAUSS)
2521 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2522 else if (flags&SWS_SINC)
2523 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2524 else if (flags&SWS_LANCZOS)
2525 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2526 else if (flags&SWS_SPLINE)
2527 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2529 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2531 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2532 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2533 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2535 av_log(c, AV_LOG_INFO, "from %s to %s ",
2536 sws_format_name(srcFormat), sws_format_name(dstFormat));
2538 if (flags & SWS_CPU_CAPS_MMX2)
2539 av_log(c, AV_LOG_INFO, "using MMX2\n");
2540 else if (flags & SWS_CPU_CAPS_3DNOW)
2541 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2542 else if (flags & SWS_CPU_CAPS_MMX)
2543 av_log(c, AV_LOG_INFO, "using MMX\n");
2544 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2545 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2547 av_log(c, AV_LOG_INFO, "using C\n");
2550 if (flags & SWS_PRINT_INFO)
2552 if (flags & SWS_CPU_CAPS_MMX)
2554 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2555 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2558 if (c->hLumFilterSize==4)
2559 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2560 else if (c->hLumFilterSize==8)
2561 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2563 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2565 if (c->hChrFilterSize==4)
2566 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2567 else if (c->hChrFilterSize==8)
2568 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2570 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2575 #if defined(ARCH_X86)
2576 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2578 if (flags & SWS_FAST_BILINEAR)
2579 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2581 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2584 if (isPlanarYUV(dstFormat))
2586 if (c->vLumFilterSize==1)
2587 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2589 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2593 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2594 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2595 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2596 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2597 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2599 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2602 if (dstFormat==PIX_FMT_BGR24)
2603 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2604 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2605 else if (dstFormat==PIX_FMT_RGB32)
2606 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2607 else if (dstFormat==PIX_FMT_BGR565)
2608 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2609 else if (dstFormat==PIX_FMT_BGR555)
2610 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2612 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2614 if (flags & SWS_PRINT_INFO)
2616 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2617 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2618 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2619 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2622 c->swScale= getSwsFunc(flags);
2627 * swscale wrapper, so we don't need to export the SwsContext.
2628 * assumes planar YUV to be in YUV order instead of YVU
2630 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2631 int srcSliceH, uint8_t* dst[], int dstStride[]){
2633 uint8_t* src2[4]= {src[0], src[1], src[2]};
2635 int use_pal= c->srcFormat == PIX_FMT_PAL8
2636 || c->srcFormat == PIX_FMT_BGR4_BYTE
2637 || c->srcFormat == PIX_FMT_RGB4_BYTE
2638 || c->srcFormat == PIX_FMT_BGR8
2639 || c->srcFormat == PIX_FMT_RGB8;
2641 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2642 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2645 if (c->sliceDir == 0) {
2646 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2650 for (i=0; i<256; i++){
2651 int p, r, g, b,y,u,v;
2652 if(c->srcFormat == PIX_FMT_PAL8){
2653 p=((uint32_t*)(src[1]))[i];
2657 }else if(c->srcFormat == PIX_FMT_RGB8){
2661 }else if(c->srcFormat == PIX_FMT_BGR8){
2665 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2669 }else if(c->srcFormat == PIX_FMT_BGR4_BYTE){
2674 y= av_clip_uint8((RY*r + GY*g + BY*b + ( 33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2675 u= av_clip_uint8((RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2676 v= av_clip_uint8((RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
2677 pal[i]= y + (u<<8) + (v<<16);
2679 src2[1]= (uint8_t*)pal;
2682 // copy strides, so they can safely be modified
2683 if (c->sliceDir == 1) {
2684 // slices go from top to bottom
2685 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2686 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2687 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2689 // slices go from bottom to top => we flip the image internally
2690 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2691 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2692 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2693 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2694 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2696 src2[0] += (srcSliceH-1)*srcStride[0];
2698 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2699 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2701 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2706 * swscale wrapper, so we don't need to export the SwsContext
2708 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2709 int srcSliceH, uint8_t* dst[], int dstStride[]){
2710 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2713 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2714 float lumaSharpen, float chromaSharpen,
2715 float chromaHShift, float chromaVShift,
2718 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2720 if (lumaGBlur!=0.0){
2721 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2722 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2724 filter->lumH= sws_getIdentityVec();
2725 filter->lumV= sws_getIdentityVec();
2728 if (chromaGBlur!=0.0){
2729 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2730 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2732 filter->chrH= sws_getIdentityVec();
2733 filter->chrV= sws_getIdentityVec();
2736 if (chromaSharpen!=0.0){
2737 SwsVector *id= sws_getIdentityVec();
2738 sws_scaleVec(filter->chrH, -chromaSharpen);
2739 sws_scaleVec(filter->chrV, -chromaSharpen);
2740 sws_addVec(filter->chrH, id);
2741 sws_addVec(filter->chrV, id);
2745 if (lumaSharpen!=0.0){
2746 SwsVector *id= sws_getIdentityVec();
2747 sws_scaleVec(filter->lumH, -lumaSharpen);
2748 sws_scaleVec(filter->lumV, -lumaSharpen);
2749 sws_addVec(filter->lumH, id);
2750 sws_addVec(filter->lumV, id);
2754 if (chromaHShift != 0.0)
2755 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2757 if (chromaVShift != 0.0)
2758 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2760 sws_normalizeVec(filter->chrH, 1.0);
2761 sws_normalizeVec(filter->chrV, 1.0);
2762 sws_normalizeVec(filter->lumH, 1.0);
2763 sws_normalizeVec(filter->lumV, 1.0);
2765 if (verbose) sws_printVec(filter->chrH);
2766 if (verbose) sws_printVec(filter->lumH);
2772 * returns a normalized gaussian curve used to filter stuff
2773 * quality=3 is high quality, lowwer is lowwer quality
2775 SwsVector *sws_getGaussianVec(double variance, double quality){
2776 const int length= (int)(variance*quality + 0.5) | 1;
2778 double *coeff= av_malloc(length*sizeof(double));
2779 double middle= (length-1)*0.5;
2780 SwsVector *vec= av_malloc(sizeof(SwsVector));
2783 vec->length= length;
2785 for (i=0; i<length; i++)
2787 double dist= i-middle;
2788 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2791 sws_normalizeVec(vec, 1.0);
2796 SwsVector *sws_getConstVec(double c, int length){
2798 double *coeff= av_malloc(length*sizeof(double));
2799 SwsVector *vec= av_malloc(sizeof(SwsVector));
2802 vec->length= length;
2804 for (i=0; i<length; i++)
2811 SwsVector *sws_getIdentityVec(void){
2812 return sws_getConstVec(1.0, 1);
2815 double sws_dcVec(SwsVector *a){
2819 for (i=0; i<a->length; i++)
2825 void sws_scaleVec(SwsVector *a, double scalar){
2828 for (i=0; i<a->length; i++)
2829 a->coeff[i]*= scalar;
2832 void sws_normalizeVec(SwsVector *a, double height){
2833 sws_scaleVec(a, height/sws_dcVec(a));
2836 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2837 int length= a->length + b->length - 1;
2838 double *coeff= av_malloc(length*sizeof(double));
2840 SwsVector *vec= av_malloc(sizeof(SwsVector));
2843 vec->length= length;
2845 for (i=0; i<length; i++) coeff[i]= 0.0;
2847 for (i=0; i<a->length; i++)
2849 for (j=0; j<b->length; j++)
2851 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2858 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2859 int length= FFMAX(a->length, b->length);
2860 double *coeff= av_malloc(length*sizeof(double));
2862 SwsVector *vec= av_malloc(sizeof(SwsVector));
2865 vec->length= length;
2867 for (i=0; i<length; i++) coeff[i]= 0.0;
2869 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2870 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2875 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2876 int length= FFMAX(a->length, b->length);
2877 double *coeff= av_malloc(length*sizeof(double));
2879 SwsVector *vec= av_malloc(sizeof(SwsVector));
2882 vec->length= length;
2884 for (i=0; i<length; i++) coeff[i]= 0.0;
2886 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2887 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2892 /* shift left / or right if "shift" is negative */
2893 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2894 int length= a->length + FFABS(shift)*2;
2895 double *coeff= av_malloc(length*sizeof(double));
2897 SwsVector *vec= av_malloc(sizeof(SwsVector));
2900 vec->length= length;
2902 for (i=0; i<length; i++) coeff[i]= 0.0;
2904 for (i=0; i<a->length; i++)
2906 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2912 void sws_shiftVec(SwsVector *a, int shift){
2913 SwsVector *shifted= sws_getShiftedVec(a, shift);
2915 a->coeff= shifted->coeff;
2916 a->length= shifted->length;
2920 void sws_addVec(SwsVector *a, SwsVector *b){
2921 SwsVector *sum= sws_sumVec(a, b);
2923 a->coeff= sum->coeff;
2924 a->length= sum->length;
2928 void sws_subVec(SwsVector *a, SwsVector *b){
2929 SwsVector *diff= sws_diffVec(a, b);
2931 a->coeff= diff->coeff;
2932 a->length= diff->length;
2936 void sws_convVec(SwsVector *a, SwsVector *b){
2937 SwsVector *conv= sws_getConvVec(a, b);
2939 a->coeff= conv->coeff;
2940 a->length= conv->length;
2944 SwsVector *sws_cloneVec(SwsVector *a){
2945 double *coeff= av_malloc(a->length*sizeof(double));
2947 SwsVector *vec= av_malloc(sizeof(SwsVector));
2950 vec->length= a->length;
2952 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2957 void sws_printVec(SwsVector *a){
2963 for (i=0; i<a->length; i++)
2964 if (a->coeff[i]>max) max= a->coeff[i];
2966 for (i=0; i<a->length; i++)
2967 if (a->coeff[i]<min) min= a->coeff[i];
2971 for (i=0; i<a->length; i++)
2973 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2974 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2975 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2976 av_log(NULL, AV_LOG_DEBUG, "|\n");
2980 void sws_freeVec(SwsVector *a){
2982 av_freep(&a->coeff);
2987 void sws_freeFilter(SwsFilter *filter){
2988 if (!filter) return;
2990 if (filter->lumH) sws_freeVec(filter->lumH);
2991 if (filter->lumV) sws_freeVec(filter->lumV);
2992 if (filter->chrH) sws_freeVec(filter->chrH);
2993 if (filter->chrV) sws_freeVec(filter->chrV);
2998 void sws_freeContext(SwsContext *c){
3004 for (i=0; i<c->vLumBufSize; i++)
3005 av_freep(&c->lumPixBuf[i]);
3006 av_freep(&c->lumPixBuf);
3011 for (i=0; i<c->vChrBufSize; i++)
3012 av_freep(&c->chrPixBuf[i]);
3013 av_freep(&c->chrPixBuf);
3016 av_freep(&c->vLumFilter);
3017 av_freep(&c->vChrFilter);
3018 av_freep(&c->hLumFilter);
3019 av_freep(&c->hChrFilter);
3021 av_freep(&c->vYCoeffsBank);
3022 av_freep(&c->vCCoeffsBank);
3025 av_freep(&c->vLumFilterPos);
3026 av_freep(&c->vChrFilterPos);
3027 av_freep(&c->hLumFilterPos);
3028 av_freep(&c->hChrFilterPos);
3030 #if defined(ARCH_X86) && defined(CONFIG_GPL)
3031 #ifdef MAP_ANONYMOUS
3032 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
3033 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
3035 av_free(c->funnyYCode);
3036 av_free(c->funnyUVCode);
3039 c->funnyUVCode=NULL;
3040 #endif /* defined(ARCH_X86) */
3042 av_freep(&c->lumMmx2Filter);
3043 av_freep(&c->chrMmx2Filter);
3044 av_freep(&c->lumMmx2FilterPos);
3045 av_freep(&c->chrMmx2FilterPos);
3046 av_freep(&c->yuvTable);
3052 * Checks if context is valid or reallocs a new one instead.
3053 * If context is NULL, just calls sws_getContext() to get a new one.
3054 * Otherwise, checks if the parameters are the same already saved in context.
3055 * If that is the case, returns the current context.
3056 * Otherwise, frees context and gets a new one.
3058 * Be warned that srcFilter, dstFilter are not checked, they are
3059 * asumed to remain valid.
3061 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3062 int srcW, int srcH, int srcFormat,
3063 int dstW, int dstH, int dstFormat, int flags,
3064 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
3066 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3069 param = default_param;
3072 if (context->srcW != srcW || context->srcH != srcH ||
3073 context->srcFormat != srcFormat ||
3074 context->dstW != dstW || context->dstH != dstH ||
3075 context->dstFormat != dstFormat || context->flags != flags ||
3076 context->param[0] != param[0] || context->param[1] != param[1])
3078 sws_freeContext(context);
3083 return sws_getContext(srcW, srcH, srcFormat,
3084 dstW, dstH, dstFormat, flags,
3085 srcFilter, dstFilter, param);