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, BGR24, BGR16, BGR15, RGB32, 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
64 #ifdef HAVE_SYS_MMAN_H
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
71 #include "swscale_internal.h"
73 #include "libavutil/x86_cpu.h"
74 #include "libavutil/bswap.h"
76 unsigned swscale_version(void)
78 return LIBSWSCALE_VERSION_INT;
88 //#define WORDS_BIGENDIAN
91 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
93 #define RET 0xC3 //near return opcode for X86
98 #define PI 3.14159265358979323846
101 #define isSupportedIn(x) ( \
102 (x)==PIX_FMT_YUV420P \
103 || (x)==PIX_FMT_YUVA420P \
104 || (x)==PIX_FMT_YUYV422 \
105 || (x)==PIX_FMT_UYVY422 \
106 || (x)==PIX_FMT_RGB32 \
107 || (x)==PIX_FMT_BGR24 \
108 || (x)==PIX_FMT_BGR565 \
109 || (x)==PIX_FMT_BGR555 \
110 || (x)==PIX_FMT_BGR32 \
111 || (x)==PIX_FMT_RGB24 \
112 || (x)==PIX_FMT_RGB565 \
113 || (x)==PIX_FMT_RGB555 \
114 || (x)==PIX_FMT_GRAY8 \
115 || (x)==PIX_FMT_YUV410P \
116 || (x)==PIX_FMT_GRAY16BE \
117 || (x)==PIX_FMT_GRAY16LE \
118 || (x)==PIX_FMT_YUV444P \
119 || (x)==PIX_FMT_YUV422P \
120 || (x)==PIX_FMT_YUV411P \
121 || (x)==PIX_FMT_PAL8 \
122 || (x)==PIX_FMT_BGR8 \
123 || (x)==PIX_FMT_RGB8 \
124 || (x)==PIX_FMT_BGR4_BYTE \
125 || (x)==PIX_FMT_RGB4_BYTE \
126 || (x)==PIX_FMT_YUV440P \
128 #define isSupportedOut(x) ( \
129 (x)==PIX_FMT_YUV420P \
130 || (x)==PIX_FMT_YUYV422 \
131 || (x)==PIX_FMT_UYVY422 \
132 || (x)==PIX_FMT_YUV444P \
133 || (x)==PIX_FMT_YUV422P \
134 || (x)==PIX_FMT_YUV411P \
137 || (x)==PIX_FMT_NV12 \
138 || (x)==PIX_FMT_NV21 \
139 || (x)==PIX_FMT_GRAY16BE \
140 || (x)==PIX_FMT_GRAY16LE \
141 || (x)==PIX_FMT_GRAY8 \
142 || (x)==PIX_FMT_YUV410P \
144 #define isPacked(x) ( \
146 || (x)==PIX_FMT_YUYV422 \
147 || (x)==PIX_FMT_UYVY422 \
152 #define RGB2YUV_SHIFT 16
153 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
154 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
155 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
156 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
157 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
158 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
159 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
160 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
161 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
163 extern const int32_t Inverse_Table_6_9[8][4];
167 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
170 more intelligent misalignment avoidance for the horizontal scaler
171 write special vertical cubic upscale version
172 Optimize C code (yv12 / minmax)
173 add support for packed pixel yuv input & output
174 add support for Y8 output
175 optimize bgr24 & bgr32
176 add BGR4 output support
177 write special BGR->BGR scaler
180 #if defined(ARCH_X86) && defined (CONFIG_GPL)
181 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
182 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
183 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
184 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
185 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
186 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
187 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
188 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
190 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
191 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
192 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
193 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
195 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
196 0x0103010301030103LL,
197 0x0200020002000200LL,};
199 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
200 0x0602060206020602LL,
201 0x0004000400040004LL,};
203 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
204 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
205 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
206 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
207 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
208 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
210 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
211 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
212 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
215 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
216 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
217 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
219 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
220 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
221 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
222 #endif /* FAST_BGR2YV12 */
223 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
224 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
225 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
226 #endif /* defined(ARCH_X86) */
228 // clipping helper table for C implementations:
229 static unsigned char clip_table[768];
231 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
233 extern const uint8_t dither_2x2_4[2][8];
234 extern const uint8_t dither_2x2_8[2][8];
235 extern const uint8_t dither_8x8_32[8][8];
236 extern const uint8_t dither_8x8_73[8][8];
237 extern const uint8_t dither_8x8_220[8][8];
239 const char *sws_format_name(enum PixelFormat format)
242 case PIX_FMT_YUV420P:
244 case PIX_FMT_YUVA420P:
246 case PIX_FMT_YUYV422:
252 case PIX_FMT_YUV422P:
254 case PIX_FMT_YUV444P:
258 case PIX_FMT_YUV410P:
260 case PIX_FMT_YUV411P:
266 case PIX_FMT_GRAY16BE:
268 case PIX_FMT_GRAY16LE:
272 case PIX_FMT_MONOWHITE:
274 case PIX_FMT_MONOBLACK:
278 case PIX_FMT_YUVJ420P:
280 case PIX_FMT_YUVJ422P:
282 case PIX_FMT_YUVJ444P:
284 case PIX_FMT_XVMC_MPEG2_MC:
285 return "xvmc_mpeg2_mc";
286 case PIX_FMT_XVMC_MPEG2_IDCT:
287 return "xvmc_mpeg2_idct";
288 case PIX_FMT_UYVY422:
290 case PIX_FMT_UYYVYY411:
292 case PIX_FMT_RGB32_1:
294 case PIX_FMT_BGR32_1:
306 case PIX_FMT_BGR4_BYTE:
312 case PIX_FMT_RGB4_BYTE:
318 case PIX_FMT_YUV440P:
321 return "Unknown format";
325 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
326 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
327 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
329 //FIXME Optimize (just quickly writen not opti..)
331 for (i=0; i<dstW; i++)
335 for (j=0; j<lumFilterSize; j++)
336 val += lumSrc[j][i] * lumFilter[j];
338 dest[i]= av_clip_uint8(val>>19);
342 for (i=0; i<chrDstW; i++)
347 for (j=0; j<chrFilterSize; j++)
349 u += chrSrc[j][i] * chrFilter[j];
350 v += chrSrc[j][i + VOFW] * chrFilter[j];
353 uDest[i]= av_clip_uint8(u>>19);
354 vDest[i]= av_clip_uint8(v>>19);
358 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
359 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
360 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
362 //FIXME Optimize (just quickly writen not opti..)
364 for (i=0; i<dstW; i++)
368 for (j=0; j<lumFilterSize; j++)
369 val += lumSrc[j][i] * lumFilter[j];
371 dest[i]= av_clip_uint8(val>>19);
377 if (dstFormat == PIX_FMT_NV12)
378 for (i=0; i<chrDstW; i++)
383 for (j=0; j<chrFilterSize; j++)
385 u += chrSrc[j][i] * chrFilter[j];
386 v += chrSrc[j][i + VOFW] * chrFilter[j];
389 uDest[2*i]= av_clip_uint8(u>>19);
390 uDest[2*i+1]= av_clip_uint8(v>>19);
393 for (i=0; i<chrDstW; i++)
398 for (j=0; j<chrFilterSize; j++)
400 u += chrSrc[j][i] * chrFilter[j];
401 v += chrSrc[j][i + VOFW] * chrFilter[j];
404 uDest[2*i]= av_clip_uint8(v>>19);
405 uDest[2*i+1]= av_clip_uint8(u>>19);
409 #define YSCALE_YUV_2_PACKEDX_C(type) \
410 for (i=0; i<(dstW>>1); i++){\
416 type av_unused *r, *b, *g;\
419 for (j=0; j<lumFilterSize; j++)\
421 Y1 += lumSrc[j][i2] * lumFilter[j];\
422 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
424 for (j=0; j<chrFilterSize; j++)\
426 U += chrSrc[j][i] * chrFilter[j];\
427 V += chrSrc[j][i+VOFW] * chrFilter[j];\
433 if ((Y1|Y2|U|V)&256)\
435 if (Y1>255) Y1=255; \
436 else if (Y1<0)Y1=0; \
437 if (Y2>255) Y2=255; \
438 else if (Y2<0)Y2=0; \
445 #define YSCALE_YUV_2_RGBX_C(type) \
446 YSCALE_YUV_2_PACKEDX_C(type) \
447 r = (type *)c->table_rV[V]; \
448 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
449 b = (type *)c->table_bU[U]; \
451 #define YSCALE_YUV_2_PACKED2_C \
452 for (i=0; i<(dstW>>1); i++){ \
454 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
455 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
456 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
457 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
459 #define YSCALE_YUV_2_RGB2_C(type) \
460 YSCALE_YUV_2_PACKED2_C\
462 r = (type *)c->table_rV[V];\
463 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
464 b = (type *)c->table_bU[U];\
466 #define YSCALE_YUV_2_PACKED1_C \
467 for (i=0; i<(dstW>>1); i++){\
469 int Y1= buf0[i2 ]>>7;\
470 int Y2= buf0[i2+1]>>7;\
471 int U= (uvbuf1[i ])>>7;\
472 int V= (uvbuf1[i+VOFW])>>7;\
474 #define YSCALE_YUV_2_RGB1_C(type) \
475 YSCALE_YUV_2_PACKED1_C\
477 r = (type *)c->table_rV[V];\
478 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
479 b = (type *)c->table_bU[U];\
481 #define YSCALE_YUV_2_PACKED1B_C \
482 for (i=0; i<(dstW>>1); i++){\
484 int Y1= buf0[i2 ]>>7;\
485 int Y2= buf0[i2+1]>>7;\
486 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
487 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
489 #define YSCALE_YUV_2_RGB1B_C(type) \
490 YSCALE_YUV_2_PACKED1B_C\
492 r = (type *)c->table_rV[V];\
493 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
494 b = (type *)c->table_bU[U];\
496 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
497 switch(c->dstFormat)\
502 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
503 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
508 ((uint8_t*)dest)[0]= r[Y1];\
509 ((uint8_t*)dest)[1]= g[Y1];\
510 ((uint8_t*)dest)[2]= b[Y1];\
511 ((uint8_t*)dest)[3]= r[Y2];\
512 ((uint8_t*)dest)[4]= g[Y2];\
513 ((uint8_t*)dest)[5]= b[Y2];\
519 ((uint8_t*)dest)[0]= b[Y1];\
520 ((uint8_t*)dest)[1]= g[Y1];\
521 ((uint8_t*)dest)[2]= r[Y1];\
522 ((uint8_t*)dest)[3]= b[Y2];\
523 ((uint8_t*)dest)[4]= g[Y2];\
524 ((uint8_t*)dest)[5]= r[Y2];\
528 case PIX_FMT_RGB565:\
529 case PIX_FMT_BGR565:\
531 const int dr1= dither_2x2_8[y&1 ][0];\
532 const int dg1= dither_2x2_4[y&1 ][0];\
533 const int db1= dither_2x2_8[(y&1)^1][0];\
534 const int dr2= dither_2x2_8[y&1 ][1];\
535 const int dg2= dither_2x2_4[y&1 ][1];\
536 const int db2= dither_2x2_8[(y&1)^1][1];\
538 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
539 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
543 case PIX_FMT_RGB555:\
544 case PIX_FMT_BGR555:\
546 const int dr1= dither_2x2_8[y&1 ][0];\
547 const int dg1= dither_2x2_8[y&1 ][1];\
548 const int db1= dither_2x2_8[(y&1)^1][0];\
549 const int dr2= dither_2x2_8[y&1 ][1];\
550 const int dg2= dither_2x2_8[y&1 ][0];\
551 const int db2= dither_2x2_8[(y&1)^1][1];\
553 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
554 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
561 const uint8_t * const d64= dither_8x8_73[y&7];\
562 const uint8_t * const d32= dither_8x8_32[y&7];\
564 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
565 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
572 const uint8_t * const d64= dither_8x8_73 [y&7];\
573 const uint8_t * const d128=dither_8x8_220[y&7];\
575 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
576 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
580 case PIX_FMT_RGB4_BYTE:\
581 case PIX_FMT_BGR4_BYTE:\
583 const uint8_t * const d64= dither_8x8_73 [y&7];\
584 const uint8_t * const d128=dither_8x8_220[y&7];\
586 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
587 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
591 case PIX_FMT_MONOBLACK:\
593 const uint8_t * const d128=dither_8x8_220[y&7];\
594 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
595 for (i=0; i<dstW-7; i+=8){\
597 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
598 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
599 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
600 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
601 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
602 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
603 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
604 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
605 ((uint8_t*)dest)[0]= acc;\
610 ((uint8_t*)dest)-= dstW>>4;\
614 static int top[1024];\
615 static int last_new[1024][1024];\
616 static int last_in3[1024][1024];\
617 static int drift[1024][1024];\
621 const uint8_t * const d128=dither_8x8_220[y&7];\
626 for (i=dstW>>1; i<dstW; i++){\
627 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
628 int in2 = (76309 * (in - 16) + 32768) >> 16;\
629 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
630 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
631 + (last_new[y][i] - in3)*f/256;\
632 int new= old> 128 ? 255 : 0;\
634 error_new+= FFABS(last_new[y][i] - new);\
635 error_in3+= FFABS(last_in3[y][i] - in3);\
636 f= error_new - error_in3*4;\
641 left= top[i]= old - new;\
642 last_new[y][i]= new;\
643 last_in3[y][i]= in3;\
645 acc+= acc + (new&1);\
647 ((uint8_t*)dest)[0]= acc;\
655 case PIX_FMT_YUYV422:\
657 ((uint8_t*)dest)[2*i2+0]= Y1;\
658 ((uint8_t*)dest)[2*i2+1]= U;\
659 ((uint8_t*)dest)[2*i2+2]= Y2;\
660 ((uint8_t*)dest)[2*i2+3]= V;\
663 case PIX_FMT_UYVY422:\
665 ((uint8_t*)dest)[2*i2+0]= U;\
666 ((uint8_t*)dest)[2*i2+1]= Y1;\
667 ((uint8_t*)dest)[2*i2+2]= V;\
668 ((uint8_t*)dest)[2*i2+3]= Y2;\
674 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
675 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
676 uint8_t *dest, int dstW, int y)
683 YSCALE_YUV_2_RGBX_C(uint32_t)
684 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
685 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
689 YSCALE_YUV_2_RGBX_C(uint8_t)
690 ((uint8_t*)dest)[0]= r[Y1];
691 ((uint8_t*)dest)[1]= g[Y1];
692 ((uint8_t*)dest)[2]= b[Y1];
693 ((uint8_t*)dest)[3]= r[Y2];
694 ((uint8_t*)dest)[4]= g[Y2];
695 ((uint8_t*)dest)[5]= b[Y2];
700 YSCALE_YUV_2_RGBX_C(uint8_t)
701 ((uint8_t*)dest)[0]= b[Y1];
702 ((uint8_t*)dest)[1]= g[Y1];
703 ((uint8_t*)dest)[2]= r[Y1];
704 ((uint8_t*)dest)[3]= b[Y2];
705 ((uint8_t*)dest)[4]= g[Y2];
706 ((uint8_t*)dest)[5]= r[Y2];
713 const int dr1= dither_2x2_8[y&1 ][0];
714 const int dg1= dither_2x2_4[y&1 ][0];
715 const int db1= dither_2x2_8[(y&1)^1][0];
716 const int dr2= dither_2x2_8[y&1 ][1];
717 const int dg2= dither_2x2_4[y&1 ][1];
718 const int db2= dither_2x2_8[(y&1)^1][1];
719 YSCALE_YUV_2_RGBX_C(uint16_t)
720 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
721 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
728 const int dr1= dither_2x2_8[y&1 ][0];
729 const int dg1= dither_2x2_8[y&1 ][1];
730 const int db1= dither_2x2_8[(y&1)^1][0];
731 const int dr2= dither_2x2_8[y&1 ][1];
732 const int dg2= dither_2x2_8[y&1 ][0];
733 const int db2= dither_2x2_8[(y&1)^1][1];
734 YSCALE_YUV_2_RGBX_C(uint16_t)
735 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
736 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
743 const uint8_t * const d64= dither_8x8_73[y&7];
744 const uint8_t * const d32= dither_8x8_32[y&7];
745 YSCALE_YUV_2_RGBX_C(uint8_t)
746 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
747 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
754 const uint8_t * const d64= dither_8x8_73 [y&7];
755 const uint8_t * const d128=dither_8x8_220[y&7];
756 YSCALE_YUV_2_RGBX_C(uint8_t)
757 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
758 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
762 case PIX_FMT_RGB4_BYTE:
763 case PIX_FMT_BGR4_BYTE:
765 const uint8_t * const d64= dither_8x8_73 [y&7];
766 const uint8_t * const d128=dither_8x8_220[y&7];
767 YSCALE_YUV_2_RGBX_C(uint8_t)
768 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
769 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
773 case PIX_FMT_MONOBLACK:
775 const uint8_t * const d128=dither_8x8_220[y&7];
776 uint8_t *g= c->table_gU[128] + c->table_gV[128];
778 for (i=0; i<dstW-1; i+=2){
783 for (j=0; j<lumFilterSize; j++)
785 Y1 += lumSrc[j][i] * lumFilter[j];
786 Y2 += lumSrc[j][i+1] * lumFilter[j];
797 acc+= acc + g[Y1+d128[(i+0)&7]];
798 acc+= acc + g[Y2+d128[(i+1)&7]];
800 ((uint8_t*)dest)[0]= acc;
806 case PIX_FMT_YUYV422:
807 YSCALE_YUV_2_PACKEDX_C(void)
808 ((uint8_t*)dest)[2*i2+0]= Y1;
809 ((uint8_t*)dest)[2*i2+1]= U;
810 ((uint8_t*)dest)[2*i2+2]= Y2;
811 ((uint8_t*)dest)[2*i2+3]= V;
814 case PIX_FMT_UYVY422:
815 YSCALE_YUV_2_PACKEDX_C(void)
816 ((uint8_t*)dest)[2*i2+0]= U;
817 ((uint8_t*)dest)[2*i2+1]= Y1;
818 ((uint8_t*)dest)[2*i2+2]= V;
819 ((uint8_t*)dest)[2*i2+3]= Y2;
826 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
828 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
833 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
834 #define COMPILE_ALTIVEC
835 #endif //HAVE_ALTIVEC
836 #endif //ARCH_POWERPC
838 #if defined(ARCH_X86)
840 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
844 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
848 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
849 #define COMPILE_3DNOW
851 #endif //ARCH_X86 || ARCH_X86_64
862 #define RENAME(a) a ## _C
863 #include "swscale_template.c"
866 #ifdef COMPILE_ALTIVEC
869 #define RENAME(a) a ## _altivec
870 #include "swscale_template.c"
873 #if defined(ARCH_X86)
882 #define RENAME(a) a ## _X86
883 #include "swscale_template.c"
891 #define RENAME(a) a ## _MMX
892 #include "swscale_template.c"
901 #define RENAME(a) a ## _MMX2
902 #include "swscale_template.c"
911 #define RENAME(a) a ## _3DNow
912 #include "swscale_template.c"
915 #endif //ARCH_X86 || ARCH_X86_64
917 // minor note: the HAVE_xyz is messed up after that line so don't use it
919 static double getSplineCoeff(double a, double b, double c, double d, double dist)
921 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
922 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
923 else return getSplineCoeff( 0.0,
930 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
931 int srcW, int dstW, int filterAlign, int one, int flags,
932 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
939 double *filter2=NULL;
941 #if defined(ARCH_X86)
942 if (flags & SWS_CPU_CAPS_MMX)
943 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
946 // Note the +1 is for the MMXscaler which reads over the end
947 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
949 if (FFABS(xInc - 0x10000) <10) // unscaled
953 filter= av_malloc(dstW*sizeof(double)*filterSize);
954 for (i=0; i<dstW*filterSize; i++) filter[i]=0;
956 for (i=0; i<dstW; i++)
958 filter[i*filterSize]=1;
963 else if (flags&SWS_POINT) // lame looking point sampling mode
968 filter= av_malloc(dstW*sizeof(double)*filterSize);
970 xDstInSrc= xInc/2 - 0x8000;
971 for (i=0; i<dstW; i++)
973 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
980 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
984 if (flags&SWS_BICUBIC) filterSize= 4;
985 else if (flags&SWS_X ) filterSize= 4;
986 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
987 filter= av_malloc(dstW*sizeof(double)*filterSize);
989 xDstInSrc= xInc/2 - 0x8000;
990 for (i=0; i<dstW; i++)
992 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
996 //Bilinear upscale / linear interpolate / Area averaging
997 for (j=0; j<filterSize; j++)
999 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1000 double coeff= 1.0 - d;
1001 if (coeff<0) coeff=0;
1002 filter[i*filterSize + j]= coeff;
1011 double sizeFactor, filterSizeInSrc;
1012 const double xInc1= (double)xInc / (double)(1<<16);
1014 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1015 else if (flags&SWS_X) sizeFactor= 8.0;
1016 else if (flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1017 else if (flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1018 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1019 else if (flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1020 else if (flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1021 else if (flags&SWS_BILINEAR) sizeFactor= 2.0;
1023 sizeFactor= 0.0; //GCC warning killer
1027 if (xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1028 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1030 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1031 if (filterSize > srcW-2) filterSize=srcW-2;
1033 filter= av_malloc(dstW*sizeof(double)*filterSize);
1035 xDstInSrc= xInc1 / 2.0 - 0.5;
1036 for (i=0; i<dstW; i++)
1038 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1040 (*filterPos)[i]= xx;
1041 for (j=0; j<filterSize; j++)
1043 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1045 if (flags & SWS_BICUBIC)
1047 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1048 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1051 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1053 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1057 /* else if (flags & SWS_X)
1059 double p= param ? param*0.01 : 0.3;
1060 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1061 coeff*= pow(2.0, - p*d*d);
1063 else if (flags & SWS_X)
1065 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1071 if (coeff<0.0) coeff= -pow(-coeff, A);
1072 else coeff= pow( coeff, A);
1073 coeff= coeff*0.5 + 0.5;
1075 else if (flags & SWS_AREA)
1077 double srcPixelSize= 1.0/xInc1;
1078 if (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1079 else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1082 else if (flags & SWS_GAUSS)
1084 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1085 coeff = pow(2.0, - p*d*d);
1087 else if (flags & SWS_SINC)
1089 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1091 else if (flags & SWS_LANCZOS)
1093 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1094 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1097 else if (flags & SWS_BILINEAR)
1100 if (coeff<0) coeff=0;
1102 else if (flags & SWS_SPLINE)
1104 double p=-2.196152422706632;
1105 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1108 coeff= 0.0; //GCC warning killer
1112 filter[i*filterSize + j]= coeff;
1119 /* apply src & dst Filter to filter -> filter2
1122 assert(filterSize>0);
1123 filter2Size= filterSize;
1124 if (srcFilter) filter2Size+= srcFilter->length - 1;
1125 if (dstFilter) filter2Size+= dstFilter->length - 1;
1126 assert(filter2Size>0);
1127 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1129 for (i=0; i<dstW; i++)
1132 SwsVector scaleFilter;
1135 scaleFilter.coeff= filter + i*filterSize;
1136 scaleFilter.length= filterSize;
1138 if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1139 else outVec= &scaleFilter;
1141 assert(outVec->length == filter2Size);
1144 for (j=0; j<outVec->length; j++)
1146 filter2[i*filter2Size + j]= outVec->coeff[j];
1149 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1151 if (outVec != &scaleFilter) sws_freeVec(outVec);
1155 /* try to reduce the filter-size (step1 find size and shift left) */
1156 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1158 for (i=dstW-1; i>=0; i--)
1160 int min= filter2Size;
1164 /* get rid off near zero elements on the left by shifting left */
1165 for (j=0; j<filter2Size; j++)
1168 cutOff += FFABS(filter2[i*filter2Size]);
1170 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1172 /* preserve monotonicity because the core can't handle the filter otherwise */
1173 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1175 // Move filter coeffs left
1176 for (k=1; k<filter2Size; k++)
1177 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1178 filter2[i*filter2Size + k - 1]= 0.0;
1183 /* count near zeros on the right */
1184 for (j=filter2Size-1; j>0; j--)
1186 cutOff += FFABS(filter2[i*filter2Size + j]);
1188 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1192 if (min>minFilterSize) minFilterSize= min;
1195 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1196 // we can handle the special case 4,
1197 // so we don't want to go to the full 8
1198 if (minFilterSize < 5)
1201 // we really don't want to waste our time
1202 // doing useless computation, so fall-back on
1203 // the scalar C code for very small filter.
1204 // vectorizing is worth it only if you have
1205 // decent-sized vector.
1206 if (minFilterSize < 3)
1210 if (flags & SWS_CPU_CAPS_MMX) {
1211 // special case for unscaled vertical filtering
1212 if (minFilterSize == 1 && filterAlign == 2)
1216 assert(minFilterSize > 0);
1217 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1218 assert(filterSize > 0);
1219 filter= av_malloc(filterSize*dstW*sizeof(double));
1220 if (filterSize >= MAX_FILTER_SIZE || !filter)
1222 *outFilterSize= filterSize;
1224 if (flags&SWS_PRINT_INFO)
1225 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1226 /* try to reduce the filter-size (step2 reduce it) */
1227 for (i=0; i<dstW; i++)
1231 for (j=0; j<filterSize; j++)
1233 if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1234 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1239 //FIXME try to align filterpos if possible
1242 for (i=0; i<dstW; i++)
1245 if ((*filterPos)[i] < 0)
1247 // Move filter coeffs left to compensate for filterPos
1248 for (j=1; j<filterSize; j++)
1250 int left= FFMAX(j + (*filterPos)[i], 0);
1251 filter[i*filterSize + left] += filter[i*filterSize + j];
1252 filter[i*filterSize + j]=0;
1257 if ((*filterPos)[i] + filterSize > srcW)
1259 int shift= (*filterPos)[i] + filterSize - srcW;
1260 // Move filter coeffs right to compensate for filterPos
1261 for (j=filterSize-2; j>=0; j--)
1263 int right= FFMIN(j + shift, filterSize-1);
1264 filter[i*filterSize +right] += filter[i*filterSize +j];
1265 filter[i*filterSize +j]=0;
1267 (*filterPos)[i]= srcW - filterSize;
1271 // Note the +1 is for the MMXscaler which reads over the end
1272 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1273 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1275 /* Normalize & Store in outFilter */
1276 for (i=0; i<dstW; i++)
1283 for (j=0; j<filterSize; j++)
1285 sum+= filter[i*filterSize + j];
1288 for (j=0; j<*outFilterSize; j++)
1290 double v= filter[i*filterSize + j]*scale + error;
1291 int intV= floor(v + 0.5);
1292 (*outFilter)[i*(*outFilterSize) + j]= intV;
1297 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1298 for (i=0; i<*outFilterSize; i++)
1300 int j= dstW*(*outFilterSize);
1301 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1312 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1315 long imm8OfPShufW1A;
1316 long imm8OfPShufW2A;
1317 long fragmentLengthA;
1319 long imm8OfPShufW1B;
1320 long imm8OfPShufW2B;
1321 long fragmentLengthB;
1326 // create an optimized horizontal scaling routine
1334 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1335 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1336 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1337 "punpcklbw %%mm7, %%mm1 \n\t"
1338 "punpcklbw %%mm7, %%mm0 \n\t"
1339 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1341 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1343 "psubw %%mm1, %%mm0 \n\t"
1344 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1345 "pmullw %%mm3, %%mm0 \n\t"
1346 "psllw $7, %%mm1 \n\t"
1347 "paddw %%mm1, %%mm0 \n\t"
1349 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1351 "add $8, %%"REG_a" \n\t"
1355 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1356 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1357 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1362 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1366 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1367 "=r" (fragmentLengthA)
1374 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1375 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1376 "punpcklbw %%mm7, %%mm0 \n\t"
1377 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1379 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1381 "psubw %%mm1, %%mm0 \n\t"
1382 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1383 "pmullw %%mm3, %%mm0 \n\t"
1384 "psllw $7, %%mm1 \n\t"
1385 "paddw %%mm1, %%mm0 \n\t"
1387 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1389 "add $8, %%"REG_a" \n\t"
1393 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1394 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1395 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1400 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1404 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1405 "=r" (fragmentLengthB)
1408 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1411 for (i=0; i<dstW/numSplits; i++)
1418 int b=((xpos+xInc)>>16) - xx;
1419 int c=((xpos+xInc*2)>>16) - xx;
1420 int d=((xpos+xInc*3)>>16) - xx;
1422 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1423 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1424 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1425 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1430 int maxShift= 3-(d+1);
1433 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1435 funnyCode[fragmentPos + imm8OfPShufW1B]=
1436 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1437 funnyCode[fragmentPos + imm8OfPShufW2B]=
1438 a | (b<<2) | (c<<4) | (d<<6);
1440 if (i+3>=dstW) shift=maxShift; //avoid overread
1441 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1443 if (shift && i>=shift)
1445 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1446 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1447 filterPos[i/2]-=shift;
1450 fragmentPos+= fragmentLengthB;
1457 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1459 funnyCode[fragmentPos + imm8OfPShufW1A]=
1460 funnyCode[fragmentPos + imm8OfPShufW2A]=
1461 a | (b<<2) | (c<<4) | (d<<6);
1463 if (i+4>=dstW) shift=maxShift; //avoid overread
1464 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1466 if (shift && i>=shift)
1468 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1469 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1470 filterPos[i/2]-=shift;
1473 fragmentPos+= fragmentLengthA;
1476 funnyCode[fragmentPos]= RET;
1480 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1482 #endif /* COMPILE_MMX2 */
1484 static void globalInit(void){
1485 // generating tables:
1487 for (i=0; i<768; i++){
1488 int c= av_clip_uint8(i-256);
1493 static SwsFunc getSwsFunc(int flags){
1495 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1496 #if defined(ARCH_X86)
1497 // ordered per speed fastest first
1498 if (flags & SWS_CPU_CAPS_MMX2)
1499 return swScale_MMX2;
1500 else if (flags & SWS_CPU_CAPS_3DNOW)
1501 return swScale_3DNow;
1502 else if (flags & SWS_CPU_CAPS_MMX)
1509 if (flags & SWS_CPU_CAPS_ALTIVEC)
1510 return swScale_altivec;
1515 #endif /* defined(ARCH_X86) */
1516 #else //RUNTIME_CPUDETECT
1518 return swScale_MMX2;
1519 #elif defined (HAVE_3DNOW)
1520 return swScale_3DNow;
1521 #elif defined (HAVE_MMX)
1523 #elif defined (HAVE_ALTIVEC)
1524 return swScale_altivec;
1528 #endif //!RUNTIME_CPUDETECT
1531 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1532 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1533 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1535 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1536 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1540 uint8_t *srcPtr= src[0];
1541 uint8_t *dstPtr= dst;
1542 for (i=0; i<srcSliceH; i++)
1544 memcpy(dstPtr, srcPtr, c->srcW);
1545 srcPtr+= srcStride[0];
1546 dstPtr+= dstStride[0];
1549 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1550 if (c->dstFormat == PIX_FMT_NV12)
1551 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1553 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1558 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1559 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1560 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1562 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1567 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1568 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1569 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1571 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1576 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1577 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1578 int srcSliceH, uint8_t* dst[], int dstStride[]){
1579 const int srcFormat= c->srcFormat;
1580 const int dstFormat= c->dstFormat;
1581 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1582 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1583 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1584 const int dstId= fmt_depth(dstFormat) >> 2;
1585 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1588 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1589 || (isRGB(srcFormat) && isRGB(dstFormat))){
1590 switch(srcId | (dstId<<4)){
1591 case 0x34: conv= rgb16to15; break;
1592 case 0x36: conv= rgb24to15; break;
1593 case 0x38: conv= rgb32to15; break;
1594 case 0x43: conv= rgb15to16; break;
1595 case 0x46: conv= rgb24to16; break;
1596 case 0x48: conv= rgb32to16; break;
1597 case 0x63: conv= rgb15to24; break;
1598 case 0x64: conv= rgb16to24; break;
1599 case 0x68: conv= rgb32to24; break;
1600 case 0x83: conv= rgb15to32; break;
1601 case 0x84: conv= rgb16to32; break;
1602 case 0x86: conv= rgb24to32; break;
1603 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1604 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1606 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1607 || (isRGB(srcFormat) && isBGR(dstFormat))){
1608 switch(srcId | (dstId<<4)){
1609 case 0x33: conv= rgb15tobgr15; break;
1610 case 0x34: conv= rgb16tobgr15; break;
1611 case 0x36: conv= rgb24tobgr15; break;
1612 case 0x38: conv= rgb32tobgr15; break;
1613 case 0x43: conv= rgb15tobgr16; break;
1614 case 0x44: conv= rgb16tobgr16; break;
1615 case 0x46: conv= rgb24tobgr16; break;
1616 case 0x48: conv= rgb32tobgr16; break;
1617 case 0x63: conv= rgb15tobgr24; break;
1618 case 0x64: conv= rgb16tobgr24; break;
1619 case 0x66: conv= rgb24tobgr24; break;
1620 case 0x68: conv= rgb32tobgr24; break;
1621 case 0x83: conv= rgb15tobgr32; break;
1622 case 0x84: conv= rgb16tobgr32; break;
1623 case 0x86: conv= rgb24tobgr32; break;
1624 case 0x88: conv= rgb32tobgr32; break;
1625 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1626 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1629 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1630 sws_format_name(srcFormat), sws_format_name(dstFormat));
1635 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1636 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1640 uint8_t *srcPtr= src[0];
1641 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1643 for (i=0; i<srcSliceH; i++)
1645 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1646 srcPtr+= srcStride[0];
1647 dstPtr+= dstStride[0];
1654 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1655 int srcSliceH, uint8_t* dst[], int dstStride[]){
1659 dst[0]+ srcSliceY *dstStride[0],
1660 dst[1]+(srcSliceY>>1)*dstStride[1],
1661 dst[2]+(srcSliceY>>1)*dstStride[2],
1663 dstStride[0], dstStride[1], srcStride[0]);
1667 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1668 int srcSliceH, uint8_t* dst[], int dstStride[]){
1672 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1673 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1675 uint8_t *srcPtr= src[0];
1676 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1678 for (i=0; i<srcSliceH; i++)
1680 memcpy(dstPtr, srcPtr, c->srcW);
1681 srcPtr+= srcStride[0];
1682 dstPtr+= dstStride[0];
1686 if (c->dstFormat==PIX_FMT_YUV420P){
1687 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1688 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1690 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1691 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1696 /* unscaled copy like stuff (assumes nearly identical formats) */
1697 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1698 int srcSliceH, uint8_t* dst[], int dstStride[])
1700 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1701 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1705 uint8_t *srcPtr= src[0];
1706 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1709 /* universal length finder */
1710 while(length+c->srcW <= FFABS(dstStride[0])
1711 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1714 for (i=0; i<srcSliceH; i++)
1716 memcpy(dstPtr, srcPtr, length);
1717 srcPtr+= srcStride[0];
1718 dstPtr+= dstStride[0];
1724 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1725 int srcSliceH, uint8_t* dst[], int dstStride[])
1728 for (plane=0; plane<3; plane++)
1730 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1731 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1732 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1734 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1736 if (!isGray(c->dstFormat))
1737 memset(dst[plane], 128, dstStride[plane]*height);
1741 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1742 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1746 uint8_t *srcPtr= src[plane];
1747 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1748 for (i=0; i<height; i++)
1750 memcpy(dstPtr, srcPtr, length);
1751 srcPtr+= srcStride[plane];
1752 dstPtr+= dstStride[plane];
1760 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1761 int srcSliceH, uint8_t* dst[], int dstStride[]){
1763 int length= c->srcW;
1765 int height= srcSliceH;
1767 uint8_t *srcPtr= src[0];
1768 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1770 if (!isGray(c->dstFormat)){
1771 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1772 memset(dst[1], 128, dstStride[1]*height);
1773 memset(dst[2], 128, dstStride[2]*height);
1775 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1776 for (i=0; i<height; i++)
1778 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1779 srcPtr+= srcStride[0];
1780 dstPtr+= dstStride[0];
1785 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1786 int srcSliceH, uint8_t* dst[], int dstStride[]){
1788 int length= c->srcW;
1790 int height= srcSliceH;
1792 uint8_t *srcPtr= src[0];
1793 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1794 for (i=0; i<height; i++)
1796 for (j=0; j<length; j++)
1798 dstPtr[j<<1] = srcPtr[j];
1799 dstPtr[(j<<1)+1] = srcPtr[j];
1801 srcPtr+= srcStride[0];
1802 dstPtr+= dstStride[0];
1807 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1808 int srcSliceH, uint8_t* dst[], int dstStride[]){
1810 int length= c->srcW;
1812 int height= srcSliceH;
1814 uint16_t *srcPtr= (uint16_t*)src[0];
1815 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1816 for (i=0; i<height; i++)
1818 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1819 srcPtr+= srcStride[0]/2;
1820 dstPtr+= dstStride[0]/2;
1826 static void getSubSampleFactors(int *h, int *v, int format){
1828 case PIX_FMT_UYVY422:
1829 case PIX_FMT_YUYV422:
1833 case PIX_FMT_YUV420P:
1834 case PIX_FMT_YUVA420P:
1835 case PIX_FMT_GRAY16BE:
1836 case PIX_FMT_GRAY16LE:
1837 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1843 case PIX_FMT_YUV440P:
1847 case PIX_FMT_YUV410P:
1851 case PIX_FMT_YUV444P:
1855 case PIX_FMT_YUV422P:
1859 case PIX_FMT_YUV411P:
1870 static uint16_t roundToInt16(int64_t f){
1871 int r= (f + (1<<15))>>16;
1872 if (r<-0x7FFF) return 0x8000;
1873 else if (r> 0x7FFF) return 0x7FFF;
1878 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1879 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
1880 * @return -1 if not supported
1882 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1883 int64_t crv = inv_table[0];
1884 int64_t cbu = inv_table[1];
1885 int64_t cgu = -inv_table[2];
1886 int64_t cgv = -inv_table[3];
1890 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1891 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1892 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1894 c->brightness= brightness;
1895 c->contrast = contrast;
1896 c->saturation= saturation;
1897 c->srcRange = srcRange;
1898 c->dstRange = dstRange;
1900 c->uOffset= 0x0400040004000400LL;
1901 c->vOffset= 0x0400040004000400LL;
1907 crv= (crv*224) / 255;
1908 cbu= (cbu*224) / 255;
1909 cgu= (cgu*224) / 255;
1910 cgv= (cgv*224) / 255;
1913 cy = (cy *contrast )>>16;
1914 crv= (crv*contrast * saturation)>>32;
1915 cbu= (cbu*contrast * saturation)>>32;
1916 cgu= (cgu*contrast * saturation)>>32;
1917 cgv= (cgv*contrast * saturation)>>32;
1919 oy -= 256*brightness;
1921 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1922 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1923 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1924 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1925 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1926 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1928 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1931 #ifdef COMPILE_ALTIVEC
1932 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1933 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1939 * @return -1 if not supported
1941 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1942 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1944 *inv_table = c->srcColorspaceTable;
1945 *table = c->dstColorspaceTable;
1946 *srcRange = c->srcRange;
1947 *dstRange = c->dstRange;
1948 *brightness= c->brightness;
1949 *contrast = c->contrast;
1950 *saturation= c->saturation;
1955 static int handle_jpeg(int *format)
1958 case PIX_FMT_YUVJ420P:
1959 *format = PIX_FMT_YUV420P;
1961 case PIX_FMT_YUVJ422P:
1962 *format = PIX_FMT_YUV422P;
1964 case PIX_FMT_YUVJ444P:
1965 *format = PIX_FMT_YUV444P;
1967 case PIX_FMT_YUVJ440P:
1968 *format = PIX_FMT_YUV440P;
1975 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1976 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1980 int usesVFilter, usesHFilter;
1981 int unscaled, needsDither;
1982 int srcRange, dstRange;
1983 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1984 #if defined(ARCH_X86)
1985 if (flags & SWS_CPU_CAPS_MMX)
1986 asm volatile("emms\n\t"::: "memory");
1989 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1990 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
1992 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1993 #elif defined (HAVE_3DNOW)
1994 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1995 #elif defined (HAVE_MMX)
1996 flags |= SWS_CPU_CAPS_MMX;
1997 #elif defined (HAVE_ALTIVEC)
1998 flags |= SWS_CPU_CAPS_ALTIVEC;
1999 #elif defined (ARCH_BFIN)
2000 flags |= SWS_CPU_CAPS_BFIN;
2002 #endif /* RUNTIME_CPUDETECT */
2003 if (clip_table[512] != 255) globalInit();
2004 if (!rgb15to16) sws_rgb2rgb_init(flags);
2006 unscaled = (srcW == dstW && srcH == dstH);
2007 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2008 && (fmt_depth(dstFormat))<24
2009 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2011 srcRange = handle_jpeg(&srcFormat);
2012 dstRange = handle_jpeg(&dstFormat);
2014 if (!isSupportedIn(srcFormat))
2016 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2019 if (!isSupportedOut(dstFormat))
2021 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2025 i= flags & ( SWS_POINT
2036 if(!i || (i & (i-1)))
2038 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2044 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
2046 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2047 srcW, srcH, dstW, dstH);
2050 if(srcW > VOFW || dstW > VOFW){
2051 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2055 if (!dstFilter) dstFilter= &dummyFilter;
2056 if (!srcFilter) srcFilter= &dummyFilter;
2058 c= av_mallocz(sizeof(SwsContext));
2060 c->av_class = &sws_context_class;
2065 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2066 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2068 c->dstFormat= dstFormat;
2069 c->srcFormat= srcFormat;
2070 c->vRounder= 4* 0x0001000100010001ULL;
2072 usesHFilter= usesVFilter= 0;
2073 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2074 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2075 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2076 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2077 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2078 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2079 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2080 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2082 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2083 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2085 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2086 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2088 // drop some chroma lines if the user wants it
2089 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2090 c->chrSrcVSubSample+= c->vChrDrop;
2092 // drop every 2. pixel for chroma calculation unless user wants full chroma
2093 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2094 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2095 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2096 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2097 c->chrSrcHSubSample=1;
2100 c->param[0] = param[0];
2101 c->param[1] = param[1];
2104 c->param[1] = SWS_PARAM_DEFAULT;
2107 c->chrIntHSubSample= c->chrDstHSubSample;
2108 c->chrIntVSubSample= c->chrSrcVSubSample;
2110 // Note the -((-x)>>y) is so that we always round toward +inf.
2111 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2112 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2113 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2114 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2116 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);
2118 /* unscaled special Cases */
2119 if (unscaled && !usesHFilter && !usesVFilter)
2122 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2124 c->swScale= PlanarToNV12Wrapper;
2128 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2130 c->swScale= yuv2rgb_get_func_ptr(c);
2134 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2136 c->swScale= yvu9toyv12Wrapper;
2140 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2141 c->swScale= bgr24toyv12Wrapper;
2143 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2144 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2145 && (isBGR(dstFormat) || isRGB(dstFormat))
2146 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2147 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2148 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2149 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2150 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2151 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2152 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2154 c->swScale= rgb2rgbWrapper;
2156 /* LQ converters if -sws 0 or -sws 4*/
2157 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2158 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2159 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2160 && (isBGR(dstFormat) || isRGB(dstFormat))
2162 c->swScale= rgb2rgbWrapper;
2165 if (srcFormat == PIX_FMT_YUV420P &&
2166 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2168 if (dstFormat == PIX_FMT_YUYV422)
2169 c->swScale= PlanarToYuy2Wrapper;
2171 c->swScale= PlanarToUyvyWrapper;
2175 #ifdef COMPILE_ALTIVEC
2176 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2177 ((srcFormat == PIX_FMT_YUV420P &&
2178 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2179 // unscaled YV12 -> packed YUV, we want speed
2180 if (dstFormat == PIX_FMT_YUYV422)
2181 c->swScale= yv12toyuy2_unscaled_altivec;
2183 c->swScale= yv12touyvy_unscaled_altivec;
2188 if ( srcFormat == dstFormat
2189 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2190 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2192 if (isPacked(c->srcFormat))
2193 c->swScale= packedCopy;
2194 else /* Planar YUV or gray */
2195 c->swScale= planarCopy;
2198 /* gray16{le,be} conversions */
2199 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2201 c->swScale= gray16togray;
2203 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2205 c->swScale= graytogray16;
2207 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2209 c->swScale= gray16swap;
2213 if (flags & SWS_CPU_CAPS_BFIN)
2214 ff_bfin_get_unscaled_swscale (c);
2218 if (flags&SWS_PRINT_INFO)
2219 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2220 sws_format_name(srcFormat), sws_format_name(dstFormat));
2225 if (flags & SWS_CPU_CAPS_MMX2)
2227 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2228 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2230 if (flags&SWS_PRINT_INFO)
2231 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2233 if (usesHFilter) c->canMMX2BeUsed=0;
2238 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2239 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2241 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2242 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2243 // n-2 is the last chrominance sample available
2244 // this is not perfect, but no one should notice the difference, the more correct variant
2245 // would be like the vertical one, but that would require some special code for the
2246 // first and last pixel
2247 if (flags&SWS_FAST_BILINEAR)
2249 if (c->canMMX2BeUsed)
2254 //we don't use the x86asm scaler if mmx is available
2255 else if (flags & SWS_CPU_CAPS_MMX)
2257 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2258 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2262 /* precalculate horizontal scaler filter coefficients */
2264 const int filterAlign=
2265 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2266 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2269 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2270 srcW , dstW, filterAlign, 1<<14,
2271 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2272 srcFilter->lumH, dstFilter->lumH, c->param);
2273 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2274 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2275 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2276 srcFilter->chrH, dstFilter->chrH, c->param);
2278 #define MAX_FUNNY_CODE_SIZE 10000
2279 #if defined(COMPILE_MMX2)
2280 // can't downscale !!!
2281 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2283 #ifdef MAP_ANONYMOUS
2284 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2285 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2287 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2288 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2291 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2292 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2293 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2294 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2296 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2297 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2299 #endif /* defined(COMPILE_MMX2) */
2300 } // Init Horizontal stuff
2304 /* precalculate vertical scaler filter coefficients */
2306 const int filterAlign=
2307 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2308 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2311 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2312 srcH , dstH, filterAlign, (1<<12)-4,
2313 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2314 srcFilter->lumV, dstFilter->lumV, c->param);
2315 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2316 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2317 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2318 srcFilter->chrV, dstFilter->chrV, c->param);
2321 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2322 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2324 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2326 short *p = (short *)&c->vYCoeffsBank[i];
2328 p[j] = c->vLumFilter[i];
2331 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2333 short *p = (short *)&c->vCCoeffsBank[i];
2335 p[j] = c->vChrFilter[i];
2340 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2341 c->vLumBufSize= c->vLumFilterSize;
2342 c->vChrBufSize= c->vChrFilterSize;
2343 for (i=0; i<dstH; i++)
2345 int chrI= i*c->chrDstH / dstH;
2346 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2347 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2349 nextSlice>>= c->chrSrcVSubSample;
2350 nextSlice<<= c->chrSrcVSubSample;
2351 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2352 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2353 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2354 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2357 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2358 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2359 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2360 //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)
2361 /* align at 16 bytes for AltiVec */
2362 for (i=0; i<c->vLumBufSize; i++)
2363 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2364 for (i=0; i<c->vChrBufSize; i++)
2365 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2367 //try to avoid drawing green stuff between the right end and the stride end
2368 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2370 assert(2*VOFW == VOF);
2372 assert(c->chrDstH <= dstH);
2374 if (flags&SWS_PRINT_INFO)
2377 const char *dither= " dithered";
2379 const char *dither= "";
2381 if (flags&SWS_FAST_BILINEAR)
2382 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2383 else if (flags&SWS_BILINEAR)
2384 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2385 else if (flags&SWS_BICUBIC)
2386 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2387 else if (flags&SWS_X)
2388 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2389 else if (flags&SWS_POINT)
2390 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2391 else if (flags&SWS_AREA)
2392 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2393 else if (flags&SWS_BICUBLIN)
2394 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2395 else if (flags&SWS_GAUSS)
2396 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2397 else if (flags&SWS_SINC)
2398 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2399 else if (flags&SWS_LANCZOS)
2400 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2401 else if (flags&SWS_SPLINE)
2402 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2404 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2406 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2407 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2408 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2410 av_log(c, AV_LOG_INFO, "from %s to %s ",
2411 sws_format_name(srcFormat), sws_format_name(dstFormat));
2413 if (flags & SWS_CPU_CAPS_MMX2)
2414 av_log(c, AV_LOG_INFO, "using MMX2\n");
2415 else if (flags & SWS_CPU_CAPS_3DNOW)
2416 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2417 else if (flags & SWS_CPU_CAPS_MMX)
2418 av_log(c, AV_LOG_INFO, "using MMX\n");
2419 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2420 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2422 av_log(c, AV_LOG_INFO, "using C\n");
2425 if (flags & SWS_PRINT_INFO)
2427 if (flags & SWS_CPU_CAPS_MMX)
2429 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2430 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2433 if (c->hLumFilterSize==4)
2434 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2435 else if (c->hLumFilterSize==8)
2436 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2438 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2440 if (c->hChrFilterSize==4)
2441 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2442 else if (c->hChrFilterSize==8)
2443 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2445 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2450 #if defined(ARCH_X86)
2451 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2453 if (flags & SWS_FAST_BILINEAR)
2454 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2456 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2459 if (isPlanarYUV(dstFormat))
2461 if (c->vLumFilterSize==1)
2462 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2464 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2468 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2469 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2470 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2471 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2472 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2474 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2477 if (dstFormat==PIX_FMT_BGR24)
2478 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2479 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2480 else if (dstFormat==PIX_FMT_RGB32)
2481 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2482 else if (dstFormat==PIX_FMT_BGR565)
2483 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2484 else if (dstFormat==PIX_FMT_BGR555)
2485 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2487 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2489 if (flags & SWS_PRINT_INFO)
2491 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2492 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2493 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2494 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2497 c->swScale= getSwsFunc(flags);
2502 * swscale wrapper, so we don't need to export the SwsContext.
2503 * assumes planar YUV to be in YUV order instead of YVU
2505 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2506 int srcSliceH, uint8_t* dst[], int dstStride[]){
2508 uint8_t* src2[4]= {src[0], src[1], src[2]};
2510 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2511 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2514 if (c->sliceDir == 0) {
2515 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2518 if (c->srcFormat == PIX_FMT_PAL8){
2519 for (i=0; i<256; i++){
2520 int p= ((uint32_t*)(src[1]))[i];
2521 int r= (p>>16)&0xFF;
2522 int g= (p>> 8)&0xFF;
2524 int y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2525 int u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2526 int v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2527 pal[i]= y + (u<<8) + (v<<16);
2529 src2[1]= (uint8_t*)pal;
2532 // copy strides, so they can safely be modified
2533 if (c->sliceDir == 1) {
2534 // slices go from top to bottom
2535 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2536 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2537 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2539 // slices go from bottom to top => we flip the image internally
2540 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2541 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2542 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2543 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2544 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2546 src2[0] += (srcSliceH-1)*srcStride[0];
2547 if (c->srcFormat != PIX_FMT_PAL8)
2548 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2549 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2551 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2556 * swscale wrapper, so we don't need to export the SwsContext
2558 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2559 int srcSliceH, uint8_t* dst[], int dstStride[]){
2560 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2563 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2564 float lumaSharpen, float chromaSharpen,
2565 float chromaHShift, float chromaVShift,
2568 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2570 if (lumaGBlur!=0.0){
2571 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2572 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2574 filter->lumH= sws_getIdentityVec();
2575 filter->lumV= sws_getIdentityVec();
2578 if (chromaGBlur!=0.0){
2579 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2580 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2582 filter->chrH= sws_getIdentityVec();
2583 filter->chrV= sws_getIdentityVec();
2586 if (chromaSharpen!=0.0){
2587 SwsVector *id= sws_getIdentityVec();
2588 sws_scaleVec(filter->chrH, -chromaSharpen);
2589 sws_scaleVec(filter->chrV, -chromaSharpen);
2590 sws_addVec(filter->chrH, id);
2591 sws_addVec(filter->chrV, id);
2595 if (lumaSharpen!=0.0){
2596 SwsVector *id= sws_getIdentityVec();
2597 sws_scaleVec(filter->lumH, -lumaSharpen);
2598 sws_scaleVec(filter->lumV, -lumaSharpen);
2599 sws_addVec(filter->lumH, id);
2600 sws_addVec(filter->lumV, id);
2604 if (chromaHShift != 0.0)
2605 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2607 if (chromaVShift != 0.0)
2608 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2610 sws_normalizeVec(filter->chrH, 1.0);
2611 sws_normalizeVec(filter->chrV, 1.0);
2612 sws_normalizeVec(filter->lumH, 1.0);
2613 sws_normalizeVec(filter->lumV, 1.0);
2615 if (verbose) sws_printVec(filter->chrH);
2616 if (verbose) sws_printVec(filter->lumH);
2622 * returns a normalized gaussian curve used to filter stuff
2623 * quality=3 is high quality, lowwer is lowwer quality
2625 SwsVector *sws_getGaussianVec(double variance, double quality){
2626 const int length= (int)(variance*quality + 0.5) | 1;
2628 double *coeff= av_malloc(length*sizeof(double));
2629 double middle= (length-1)*0.5;
2630 SwsVector *vec= av_malloc(sizeof(SwsVector));
2633 vec->length= length;
2635 for (i=0; i<length; i++)
2637 double dist= i-middle;
2638 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2641 sws_normalizeVec(vec, 1.0);
2646 SwsVector *sws_getConstVec(double c, int length){
2648 double *coeff= av_malloc(length*sizeof(double));
2649 SwsVector *vec= av_malloc(sizeof(SwsVector));
2652 vec->length= length;
2654 for (i=0; i<length; i++)
2661 SwsVector *sws_getIdentityVec(void){
2662 return sws_getConstVec(1.0, 1);
2665 double sws_dcVec(SwsVector *a){
2669 for (i=0; i<a->length; i++)
2675 void sws_scaleVec(SwsVector *a, double scalar){
2678 for (i=0; i<a->length; i++)
2679 a->coeff[i]*= scalar;
2682 void sws_normalizeVec(SwsVector *a, double height){
2683 sws_scaleVec(a, height/sws_dcVec(a));
2686 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2687 int length= a->length + b->length - 1;
2688 double *coeff= av_malloc(length*sizeof(double));
2690 SwsVector *vec= av_malloc(sizeof(SwsVector));
2693 vec->length= length;
2695 for (i=0; i<length; i++) coeff[i]= 0.0;
2697 for (i=0; i<a->length; i++)
2699 for (j=0; j<b->length; j++)
2701 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2708 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2709 int length= FFMAX(a->length, b->length);
2710 double *coeff= av_malloc(length*sizeof(double));
2712 SwsVector *vec= av_malloc(sizeof(SwsVector));
2715 vec->length= length;
2717 for (i=0; i<length; i++) coeff[i]= 0.0;
2719 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2720 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2725 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2726 int length= FFMAX(a->length, b->length);
2727 double *coeff= av_malloc(length*sizeof(double));
2729 SwsVector *vec= av_malloc(sizeof(SwsVector));
2732 vec->length= length;
2734 for (i=0; i<length; i++) coeff[i]= 0.0;
2736 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2737 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2742 /* shift left / or right if "shift" is negative */
2743 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2744 int length= a->length + FFABS(shift)*2;
2745 double *coeff= av_malloc(length*sizeof(double));
2747 SwsVector *vec= av_malloc(sizeof(SwsVector));
2750 vec->length= length;
2752 for (i=0; i<length; i++) coeff[i]= 0.0;
2754 for (i=0; i<a->length; i++)
2756 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2762 void sws_shiftVec(SwsVector *a, int shift){
2763 SwsVector *shifted= sws_getShiftedVec(a, shift);
2765 a->coeff= shifted->coeff;
2766 a->length= shifted->length;
2770 void sws_addVec(SwsVector *a, SwsVector *b){
2771 SwsVector *sum= sws_sumVec(a, b);
2773 a->coeff= sum->coeff;
2774 a->length= sum->length;
2778 void sws_subVec(SwsVector *a, SwsVector *b){
2779 SwsVector *diff= sws_diffVec(a, b);
2781 a->coeff= diff->coeff;
2782 a->length= diff->length;
2786 void sws_convVec(SwsVector *a, SwsVector *b){
2787 SwsVector *conv= sws_getConvVec(a, b);
2789 a->coeff= conv->coeff;
2790 a->length= conv->length;
2794 SwsVector *sws_cloneVec(SwsVector *a){
2795 double *coeff= av_malloc(a->length*sizeof(double));
2797 SwsVector *vec= av_malloc(sizeof(SwsVector));
2800 vec->length= a->length;
2802 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2807 void sws_printVec(SwsVector *a){
2813 for (i=0; i<a->length; i++)
2814 if (a->coeff[i]>max) max= a->coeff[i];
2816 for (i=0; i<a->length; i++)
2817 if (a->coeff[i]<min) min= a->coeff[i];
2821 for (i=0; i<a->length; i++)
2823 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2824 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2825 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2826 av_log(NULL, AV_LOG_DEBUG, "|\n");
2830 void sws_freeVec(SwsVector *a){
2832 av_freep(&a->coeff);
2837 void sws_freeFilter(SwsFilter *filter){
2838 if (!filter) return;
2840 if (filter->lumH) sws_freeVec(filter->lumH);
2841 if (filter->lumV) sws_freeVec(filter->lumV);
2842 if (filter->chrH) sws_freeVec(filter->chrH);
2843 if (filter->chrV) sws_freeVec(filter->chrV);
2848 void sws_freeContext(SwsContext *c){
2854 for (i=0; i<c->vLumBufSize; i++)
2855 av_freep(&c->lumPixBuf[i]);
2856 av_freep(&c->lumPixBuf);
2861 for (i=0; i<c->vChrBufSize; i++)
2862 av_freep(&c->chrPixBuf[i]);
2863 av_freep(&c->chrPixBuf);
2866 av_freep(&c->vLumFilter);
2867 av_freep(&c->vChrFilter);
2868 av_freep(&c->hLumFilter);
2869 av_freep(&c->hChrFilter);
2871 av_freep(&c->vYCoeffsBank);
2872 av_freep(&c->vCCoeffsBank);
2875 av_freep(&c->vLumFilterPos);
2876 av_freep(&c->vChrFilterPos);
2877 av_freep(&c->hLumFilterPos);
2878 av_freep(&c->hChrFilterPos);
2880 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2881 #ifdef MAP_ANONYMOUS
2882 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2883 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2885 av_free(c->funnyYCode);
2886 av_free(c->funnyUVCode);
2889 c->funnyUVCode=NULL;
2890 #endif /* defined(ARCH_X86) */
2892 av_freep(&c->lumMmx2Filter);
2893 av_freep(&c->chrMmx2Filter);
2894 av_freep(&c->lumMmx2FilterPos);
2895 av_freep(&c->chrMmx2FilterPos);
2896 av_freep(&c->yuvTable);
2902 * Checks if context is valid or reallocs a new one instead.
2903 * If context is NULL, just calls sws_getContext() to get a new one.
2904 * Otherwise, checks if the parameters are the same already saved in context.
2905 * If that is the case, returns the current context.
2906 * Otherwise, frees context and gets a new one.
2908 * Be warned that srcFilter, dstFilter are not checked, they are
2909 * asumed to remain valid.
2911 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2912 int srcW, int srcH, int srcFormat,
2913 int dstW, int dstH, int dstFormat, int flags,
2914 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2916 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
2919 param = default_param;
2922 if (context->srcW != srcW || context->srcH != srcH ||
2923 context->srcFormat != srcFormat ||
2924 context->dstW != dstW || context->dstH != dstH ||
2925 context->dstFormat != dstFormat || context->flags != flags ||
2926 context->param[0] != param[0] || context->param[1] != param[1])
2928 sws_freeContext(context);
2933 return sws_getContext(srcW, srcH, srcFormat,
2934 dstW, dstH, dstFormat, flags,
2935 srcFilter, dstFilter, param);