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
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_RGB32_1 \
108 || (x)==PIX_FMT_BGR24 \
109 || (x)==PIX_FMT_BGR565 \
110 || (x)==PIX_FMT_BGR555 \
111 || (x)==PIX_FMT_BGR32 \
112 || (x)==PIX_FMT_BGR32_1 \
113 || (x)==PIX_FMT_RGB24 \
114 || (x)==PIX_FMT_RGB565 \
115 || (x)==PIX_FMT_RGB555 \
116 || (x)==PIX_FMT_GRAY8 \
117 || (x)==PIX_FMT_YUV410P \
118 || (x)==PIX_FMT_GRAY16BE \
119 || (x)==PIX_FMT_GRAY16LE \
120 || (x)==PIX_FMT_YUV444P \
121 || (x)==PIX_FMT_YUV422P \
122 || (x)==PIX_FMT_YUV411P \
123 || (x)==PIX_FMT_PAL8 \
124 || (x)==PIX_FMT_BGR8 \
125 || (x)==PIX_FMT_RGB8 \
126 || (x)==PIX_FMT_BGR4_BYTE \
127 || (x)==PIX_FMT_RGB4_BYTE \
128 || (x)==PIX_FMT_YUV440P \
130 #define isSupportedOut(x) ( \
131 (x)==PIX_FMT_YUV420P \
132 || (x)==PIX_FMT_YUYV422 \
133 || (x)==PIX_FMT_UYVY422 \
134 || (x)==PIX_FMT_YUV444P \
135 || (x)==PIX_FMT_YUV422P \
136 || (x)==PIX_FMT_YUV411P \
139 || (x)==PIX_FMT_NV12 \
140 || (x)==PIX_FMT_NV21 \
141 || (x)==PIX_FMT_GRAY16BE \
142 || (x)==PIX_FMT_GRAY16LE \
143 || (x)==PIX_FMT_GRAY8 \
144 || (x)==PIX_FMT_YUV410P \
146 #define isPacked(x) ( \
148 || (x)==PIX_FMT_YUYV422 \
149 || (x)==PIX_FMT_UYVY422 \
154 #define RGB2YUV_SHIFT 16
155 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
156 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
157 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
158 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
159 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
160 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
161 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
162 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
163 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
165 extern const int32_t Inverse_Table_6_9[8][4];
169 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
172 more intelligent misalignment avoidance for the horizontal scaler
173 write special vertical cubic upscale version
174 Optimize C code (yv12 / minmax)
175 add support for packed pixel yuv input & output
176 add support for Y8 output
177 optimize bgr24 & bgr32
178 add BGR4 output support
179 write special BGR->BGR scaler
182 #if defined(ARCH_X86) && defined (CONFIG_GPL)
183 DECLARE_ASM_CONST(8, uint64_t, bF8)= 0xF8F8F8F8F8F8F8F8LL;
184 DECLARE_ASM_CONST(8, uint64_t, bFC)= 0xFCFCFCFCFCFCFCFCLL;
185 DECLARE_ASM_CONST(8, uint64_t, w10)= 0x0010001000100010LL;
186 DECLARE_ASM_CONST(8, uint64_t, w02)= 0x0002000200020002LL;
187 DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
188 DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
189 DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
190 DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
192 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
193 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
194 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
195 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
197 const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
198 0x0103010301030103LL,
199 0x0200020002000200LL,};
201 const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
202 0x0602060206020602LL,
203 0x0004000400040004LL,};
205 DECLARE_ASM_CONST(8, uint64_t, b16Mask)= 0x001F001F001F001FLL;
206 DECLARE_ASM_CONST(8, uint64_t, g16Mask)= 0x07E007E007E007E0LL;
207 DECLARE_ASM_CONST(8, uint64_t, r16Mask)= 0xF800F800F800F800LL;
208 DECLARE_ASM_CONST(8, uint64_t, b15Mask)= 0x001F001F001F001FLL;
209 DECLARE_ASM_CONST(8, uint64_t, g15Mask)= 0x03E003E003E003E0LL;
210 DECLARE_ASM_CONST(8, uint64_t, r15Mask)= 0x7C007C007C007C00LL;
212 DECLARE_ALIGNED(8, const uint64_t, ff_M24A) = 0x00FF0000FF0000FFLL;
213 DECLARE_ALIGNED(8, const uint64_t, ff_M24B) = 0xFF0000FF0000FF00LL;
214 DECLARE_ALIGNED(8, const uint64_t, ff_M24C) = 0x0000FF0000FF0000LL;
217 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000000210041000DULL;
218 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000FFEEFFDC0038ULL;
219 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00000038FFD2FFF8ULL;
221 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff) = 0x000020E540830C8BULL;
222 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff) = 0x0000ED0FDAC23831ULL;
223 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff) = 0x00003831D0E6F6EAULL;
224 #endif /* FAST_BGR2YV12 */
225 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset) = 0x1010101010101010ULL;
226 DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
227 DECLARE_ALIGNED(8, const uint64_t, ff_w1111) = 0x0001000100010001ULL;
228 #endif /* defined(ARCH_X86) */
230 // clipping helper table for C implementations:
231 static unsigned char clip_table[768];
233 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
235 extern const uint8_t dither_2x2_4[2][8];
236 extern const uint8_t dither_2x2_8[2][8];
237 extern const uint8_t dither_8x8_32[8][8];
238 extern const uint8_t dither_8x8_73[8][8];
239 extern const uint8_t dither_8x8_220[8][8];
241 const char *sws_format_name(enum PixelFormat format)
244 case PIX_FMT_YUV420P:
246 case PIX_FMT_YUVA420P:
248 case PIX_FMT_YUYV422:
254 case PIX_FMT_YUV422P:
256 case PIX_FMT_YUV444P:
260 case PIX_FMT_YUV410P:
262 case PIX_FMT_YUV411P:
268 case PIX_FMT_GRAY16BE:
270 case PIX_FMT_GRAY16LE:
274 case PIX_FMT_MONOWHITE:
276 case PIX_FMT_MONOBLACK:
280 case PIX_FMT_YUVJ420P:
282 case PIX_FMT_YUVJ422P:
284 case PIX_FMT_YUVJ444P:
286 case PIX_FMT_XVMC_MPEG2_MC:
287 return "xvmc_mpeg2_mc";
288 case PIX_FMT_XVMC_MPEG2_IDCT:
289 return "xvmc_mpeg2_idct";
290 case PIX_FMT_UYVY422:
292 case PIX_FMT_UYYVYY411:
294 case PIX_FMT_RGB32_1:
296 case PIX_FMT_BGR32_1:
308 case PIX_FMT_BGR4_BYTE:
314 case PIX_FMT_RGB4_BYTE:
320 case PIX_FMT_YUV440P:
323 return "Unknown format";
327 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
328 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
329 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
331 //FIXME Optimize (just quickly writen not opti..)
333 for (i=0; i<dstW; i++)
337 for (j=0; j<lumFilterSize; j++)
338 val += lumSrc[j][i] * lumFilter[j];
340 dest[i]= av_clip_uint8(val>>19);
344 for (i=0; i<chrDstW; i++)
349 for (j=0; j<chrFilterSize; j++)
351 u += chrSrc[j][i] * chrFilter[j];
352 v += chrSrc[j][i + VOFW] * chrFilter[j];
355 uDest[i]= av_clip_uint8(u>>19);
356 vDest[i]= av_clip_uint8(v>>19);
360 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
361 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
362 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
364 //FIXME Optimize (just quickly writen not opti..)
366 for (i=0; i<dstW; i++)
370 for (j=0; j<lumFilterSize; j++)
371 val += lumSrc[j][i] * lumFilter[j];
373 dest[i]= av_clip_uint8(val>>19);
379 if (dstFormat == PIX_FMT_NV12)
380 for (i=0; i<chrDstW; i++)
385 for (j=0; j<chrFilterSize; j++)
387 u += chrSrc[j][i] * chrFilter[j];
388 v += chrSrc[j][i + VOFW] * chrFilter[j];
391 uDest[2*i]= av_clip_uint8(u>>19);
392 uDest[2*i+1]= av_clip_uint8(v>>19);
395 for (i=0; i<chrDstW; i++)
400 for (j=0; j<chrFilterSize; j++)
402 u += chrSrc[j][i] * chrFilter[j];
403 v += chrSrc[j][i + VOFW] * chrFilter[j];
406 uDest[2*i]= av_clip_uint8(v>>19);
407 uDest[2*i+1]= av_clip_uint8(u>>19);
411 #define YSCALE_YUV_2_PACKEDX_C(type) \
412 for (i=0; i<(dstW>>1); i++){\
418 type av_unused *r, *b, *g;\
421 for (j=0; j<lumFilterSize; j++)\
423 Y1 += lumSrc[j][i2] * lumFilter[j];\
424 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
426 for (j=0; j<chrFilterSize; j++)\
428 U += chrSrc[j][i] * chrFilter[j];\
429 V += chrSrc[j][i+VOFW] * chrFilter[j];\
435 if ((Y1|Y2|U|V)&256)\
437 if (Y1>255) Y1=255; \
438 else if (Y1<0)Y1=0; \
439 if (Y2>255) Y2=255; \
440 else if (Y2<0)Y2=0; \
447 #define YSCALE_YUV_2_GRAY16_C(type) \
448 for (i=0; i<(dstW>>1); i++){\
454 type av_unused *r, *b, *g;\
457 for (j=0; j<lumFilterSize; j++)\
459 Y1 += lumSrc[j][i2] * lumFilter[j];\
460 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
464 if ((Y1|Y2|U|V)&65536)\
466 if (Y1>65535) Y1=65535; \
467 else if (Y1<0)Y1=0; \
468 if (Y2>65535) Y2=65535; \
469 else if (Y2<0)Y2=0; \
472 #define YSCALE_YUV_2_RGBX_C(type) \
473 YSCALE_YUV_2_PACKEDX_C(type) \
474 r = (type *)c->table_rV[V]; \
475 g = (type *)(c->table_gU[U] + c->table_gV[V]); \
476 b = (type *)c->table_bU[U]; \
478 #define YSCALE_YUV_2_PACKED2_C \
479 for (i=0; i<(dstW>>1); i++){ \
481 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19; \
482 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19; \
483 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19; \
484 int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19; \
486 #define YSCALE_YUV_2_GRAY16_2_C \
487 for (i=0; i<(dstW>>1); i++){ \
489 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>11; \
490 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>11; \
492 #define YSCALE_YUV_2_RGB2_C(type) \
493 YSCALE_YUV_2_PACKED2_C\
495 r = (type *)c->table_rV[V];\
496 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
497 b = (type *)c->table_bU[U];\
499 #define YSCALE_YUV_2_PACKED1_C \
500 for (i=0; i<(dstW>>1); i++){\
502 int Y1= buf0[i2 ]>>7;\
503 int Y2= buf0[i2+1]>>7;\
504 int U= (uvbuf1[i ])>>7;\
505 int V= (uvbuf1[i+VOFW])>>7;\
507 #define YSCALE_YUV_2_GRAY16_1_C \
508 for (i=0; i<(dstW>>1); i++){\
510 int Y1= buf0[i2 ]<<1;\
511 int Y2= buf0[i2+1]<<1;\
513 #define YSCALE_YUV_2_RGB1_C(type) \
514 YSCALE_YUV_2_PACKED1_C\
516 r = (type *)c->table_rV[V];\
517 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
518 b = (type *)c->table_bU[U];\
520 #define YSCALE_YUV_2_PACKED1B_C \
521 for (i=0; i<(dstW>>1); i++){\
523 int Y1= buf0[i2 ]>>7;\
524 int Y2= buf0[i2+1]>>7;\
525 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
526 int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
528 #define YSCALE_YUV_2_RGB1B_C(type) \
529 YSCALE_YUV_2_PACKED1B_C\
531 r = (type *)c->table_rV[V];\
532 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
533 b = (type *)c->table_bU[U];\
535 #define YSCALE_YUV_2_ANYRGB_C(func, func2, func_g16)\
536 switch(c->dstFormat)\
540 case PIX_FMT_RGB32_1:\
541 case PIX_FMT_BGR32_1:\
543 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
544 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
549 ((uint8_t*)dest)[0]= r[Y1];\
550 ((uint8_t*)dest)[1]= g[Y1];\
551 ((uint8_t*)dest)[2]= b[Y1];\
552 ((uint8_t*)dest)[3]= r[Y2];\
553 ((uint8_t*)dest)[4]= g[Y2];\
554 ((uint8_t*)dest)[5]= b[Y2];\
560 ((uint8_t*)dest)[0]= b[Y1];\
561 ((uint8_t*)dest)[1]= g[Y1];\
562 ((uint8_t*)dest)[2]= r[Y1];\
563 ((uint8_t*)dest)[3]= b[Y2];\
564 ((uint8_t*)dest)[4]= g[Y2];\
565 ((uint8_t*)dest)[5]= r[Y2];\
569 case PIX_FMT_RGB565:\
570 case PIX_FMT_BGR565:\
572 const int dr1= dither_2x2_8[y&1 ][0];\
573 const int dg1= dither_2x2_4[y&1 ][0];\
574 const int db1= dither_2x2_8[(y&1)^1][0];\
575 const int dr2= dither_2x2_8[y&1 ][1];\
576 const int dg2= dither_2x2_4[y&1 ][1];\
577 const int db2= dither_2x2_8[(y&1)^1][1];\
579 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
580 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
584 case PIX_FMT_RGB555:\
585 case PIX_FMT_BGR555:\
587 const int dr1= dither_2x2_8[y&1 ][0];\
588 const int dg1= dither_2x2_8[y&1 ][1];\
589 const int db1= dither_2x2_8[(y&1)^1][0];\
590 const int dr2= dither_2x2_8[y&1 ][1];\
591 const int dg2= dither_2x2_8[y&1 ][0];\
592 const int db2= dither_2x2_8[(y&1)^1][1];\
594 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
595 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
602 const uint8_t * const d64= dither_8x8_73[y&7];\
603 const uint8_t * const d32= dither_8x8_32[y&7];\
605 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
606 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
613 const uint8_t * const d64= dither_8x8_73 [y&7];\
614 const uint8_t * const d128=dither_8x8_220[y&7];\
616 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
617 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
621 case PIX_FMT_RGB4_BYTE:\
622 case PIX_FMT_BGR4_BYTE:\
624 const uint8_t * const d64= dither_8x8_73 [y&7];\
625 const uint8_t * const d128=dither_8x8_220[y&7];\
627 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
628 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
632 case PIX_FMT_MONOBLACK:\
634 const uint8_t * const d128=dither_8x8_220[y&7];\
635 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
636 for (i=0; i<dstW-7; i+=8){\
638 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
639 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
640 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
641 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
642 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
643 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
644 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
645 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
646 ((uint8_t*)dest)[0]= acc;\
651 ((uint8_t*)dest)-= dstW>>4;\
655 static int top[1024];\
656 static int last_new[1024][1024];\
657 static int last_in3[1024][1024];\
658 static int drift[1024][1024];\
662 const uint8_t * const d128=dither_8x8_220[y&7];\
667 for (i=dstW>>1; i<dstW; i++){\
668 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
669 int in2 = (76309 * (in - 16) + 32768) >> 16;\
670 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
671 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
672 + (last_new[y][i] - in3)*f/256;\
673 int new= old> 128 ? 255 : 0;\
675 error_new+= FFABS(last_new[y][i] - new);\
676 error_in3+= FFABS(last_in3[y][i] - in3);\
677 f= error_new - error_in3*4;\
682 left= top[i]= old - new;\
683 last_new[y][i]= new;\
684 last_in3[y][i]= in3;\
686 acc+= acc + (new&1);\
688 ((uint8_t*)dest)[0]= acc;\
696 case PIX_FMT_YUYV422:\
698 ((uint8_t*)dest)[2*i2+0]= Y1;\
699 ((uint8_t*)dest)[2*i2+1]= U;\
700 ((uint8_t*)dest)[2*i2+2]= Y2;\
701 ((uint8_t*)dest)[2*i2+3]= V;\
704 case PIX_FMT_UYVY422:\
706 ((uint8_t*)dest)[2*i2+0]= U;\
707 ((uint8_t*)dest)[2*i2+1]= Y1;\
708 ((uint8_t*)dest)[2*i2+2]= V;\
709 ((uint8_t*)dest)[2*i2+3]= Y2;\
712 case PIX_FMT_GRAY16BE:\
714 ((uint8_t*)dest)[2*i2+0]= Y1>>8;\
715 ((uint8_t*)dest)[2*i2+1]= Y1;\
716 ((uint8_t*)dest)[2*i2+2]= Y2>>8;\
717 ((uint8_t*)dest)[2*i2+3]= Y2;\
720 case PIX_FMT_GRAY16LE:\
722 ((uint8_t*)dest)[2*i2+0]= Y1;\
723 ((uint8_t*)dest)[2*i2+1]= Y1>>8;\
724 ((uint8_t*)dest)[2*i2+2]= Y2;\
725 ((uint8_t*)dest)[2*i2+3]= Y2>>8;\
731 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
732 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
733 uint8_t *dest, int dstW, int y)
740 case PIX_FMT_BGR32_1:
741 case PIX_FMT_RGB32_1:
742 YSCALE_YUV_2_RGBX_C(uint32_t)
743 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
744 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
748 YSCALE_YUV_2_RGBX_C(uint8_t)
749 ((uint8_t*)dest)[0]= r[Y1];
750 ((uint8_t*)dest)[1]= g[Y1];
751 ((uint8_t*)dest)[2]= b[Y1];
752 ((uint8_t*)dest)[3]= r[Y2];
753 ((uint8_t*)dest)[4]= g[Y2];
754 ((uint8_t*)dest)[5]= b[Y2];
759 YSCALE_YUV_2_RGBX_C(uint8_t)
760 ((uint8_t*)dest)[0]= b[Y1];
761 ((uint8_t*)dest)[1]= g[Y1];
762 ((uint8_t*)dest)[2]= r[Y1];
763 ((uint8_t*)dest)[3]= b[Y2];
764 ((uint8_t*)dest)[4]= g[Y2];
765 ((uint8_t*)dest)[5]= r[Y2];
772 const int dr1= dither_2x2_8[y&1 ][0];
773 const int dg1= dither_2x2_4[y&1 ][0];
774 const int db1= dither_2x2_8[(y&1)^1][0];
775 const int dr2= dither_2x2_8[y&1 ][1];
776 const int dg2= dither_2x2_4[y&1 ][1];
777 const int db2= dither_2x2_8[(y&1)^1][1];
778 YSCALE_YUV_2_RGBX_C(uint16_t)
779 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
780 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
787 const int dr1= dither_2x2_8[y&1 ][0];
788 const int dg1= dither_2x2_8[y&1 ][1];
789 const int db1= dither_2x2_8[(y&1)^1][0];
790 const int dr2= dither_2x2_8[y&1 ][1];
791 const int dg2= dither_2x2_8[y&1 ][0];
792 const int db2= dither_2x2_8[(y&1)^1][1];
793 YSCALE_YUV_2_RGBX_C(uint16_t)
794 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
795 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
802 const uint8_t * const d64= dither_8x8_73[y&7];
803 const uint8_t * const d32= dither_8x8_32[y&7];
804 YSCALE_YUV_2_RGBX_C(uint8_t)
805 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
806 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
813 const uint8_t * const d64= dither_8x8_73 [y&7];
814 const uint8_t * const d128=dither_8x8_220[y&7];
815 YSCALE_YUV_2_RGBX_C(uint8_t)
816 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
817 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
821 case PIX_FMT_RGB4_BYTE:
822 case PIX_FMT_BGR4_BYTE:
824 const uint8_t * const d64= dither_8x8_73 [y&7];
825 const uint8_t * const d128=dither_8x8_220[y&7];
826 YSCALE_YUV_2_RGBX_C(uint8_t)
827 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
828 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
832 case PIX_FMT_MONOBLACK:
834 const uint8_t * const d128=dither_8x8_220[y&7];
835 uint8_t *g= c->table_gU[128] + c->table_gV[128];
837 for (i=0; i<dstW-1; i+=2){
842 for (j=0; j<lumFilterSize; j++)
844 Y1 += lumSrc[j][i] * lumFilter[j];
845 Y2 += lumSrc[j][i+1] * lumFilter[j];
856 acc+= acc + g[Y1+d128[(i+0)&7]];
857 acc+= acc + g[Y2+d128[(i+1)&7]];
859 ((uint8_t*)dest)[0]= acc;
865 case PIX_FMT_YUYV422:
866 YSCALE_YUV_2_PACKEDX_C(void)
867 ((uint8_t*)dest)[2*i2+0]= Y1;
868 ((uint8_t*)dest)[2*i2+1]= U;
869 ((uint8_t*)dest)[2*i2+2]= Y2;
870 ((uint8_t*)dest)[2*i2+3]= V;
873 case PIX_FMT_UYVY422:
874 YSCALE_YUV_2_PACKEDX_C(void)
875 ((uint8_t*)dest)[2*i2+0]= U;
876 ((uint8_t*)dest)[2*i2+1]= Y1;
877 ((uint8_t*)dest)[2*i2+2]= V;
878 ((uint8_t*)dest)[2*i2+3]= Y2;
881 case PIX_FMT_GRAY16BE:
882 YSCALE_YUV_2_GRAY16_C(void)
883 ((uint8_t*)dest)[2*i2+0]= Y1>>8;
884 ((uint8_t*)dest)[2*i2+1]= Y1;
885 ((uint8_t*)dest)[2*i2+2]= Y2>>8;
886 ((uint8_t*)dest)[2*i2+3]= Y2;
889 case PIX_FMT_GRAY16LE:
890 YSCALE_YUV_2_GRAY16_C(void)
891 ((uint8_t*)dest)[2*i2+0]= Y1;
892 ((uint8_t*)dest)[2*i2+1]= Y1>>8;
893 ((uint8_t*)dest)[2*i2+2]= Y2;
894 ((uint8_t*)dest)[2*i2+3]= Y2>>8;
901 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
903 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
908 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
909 #define COMPILE_ALTIVEC
910 #endif //HAVE_ALTIVEC
911 #endif //ARCH_POWERPC
913 #if defined(ARCH_X86)
915 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
919 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
923 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
924 #define COMPILE_3DNOW
926 #endif //ARCH_X86 || ARCH_X86_64
937 #define RENAME(a) a ## _C
938 #include "swscale_template.c"
941 #ifdef COMPILE_ALTIVEC
944 #define RENAME(a) a ## _altivec
945 #include "swscale_template.c"
948 #if defined(ARCH_X86)
957 #define RENAME(a) a ## _X86
958 #include "swscale_template.c"
966 #define RENAME(a) a ## _MMX
967 #include "swscale_template.c"
976 #define RENAME(a) a ## _MMX2
977 #include "swscale_template.c"
986 #define RENAME(a) a ## _3DNow
987 #include "swscale_template.c"
990 #endif //ARCH_X86 || ARCH_X86_64
992 // minor note: the HAVE_xyz is messed up after that line so don't use it
994 static double getSplineCoeff(double a, double b, double c, double d, double dist)
996 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
997 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
998 else return getSplineCoeff( 0.0,
1005 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
1006 int srcW, int dstW, int filterAlign, int one, int flags,
1007 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
1013 double *filter=NULL;
1014 double *filter2=NULL;
1016 #if defined(ARCH_X86)
1017 if (flags & SWS_CPU_CAPS_MMX)
1018 asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
1021 // Note the +1 is for the MMXscaler which reads over the end
1022 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
1024 if (FFABS(xInc - 0x10000) <10) // unscaled
1028 filter= av_malloc(dstW*sizeof(double)*filterSize);
1029 for (i=0; i<dstW*filterSize; i++) filter[i]=0;
1031 for (i=0; i<dstW; i++)
1033 filter[i*filterSize]=1;
1038 else if (flags&SWS_POINT) // lame looking point sampling mode
1043 filter= av_malloc(dstW*sizeof(double)*filterSize);
1045 xDstInSrc= xInc/2 - 0x8000;
1046 for (i=0; i<dstW; i++)
1048 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1050 (*filterPos)[i]= xx;
1055 else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1059 if (flags&SWS_BICUBIC) filterSize= 4;
1060 else if (flags&SWS_X ) filterSize= 4;
1061 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
1062 filter= av_malloc(dstW*sizeof(double)*filterSize);
1064 xDstInSrc= xInc/2 - 0x8000;
1065 for (i=0; i<dstW; i++)
1067 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1070 (*filterPos)[i]= xx;
1071 //Bilinear upscale / linear interpolate / Area averaging
1072 for (j=0; j<filterSize; j++)
1074 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1075 double coeff= 1.0 - d;
1076 if (coeff<0) coeff=0;
1077 filter[i*filterSize + j]= coeff;
1086 double sizeFactor, filterSizeInSrc;
1087 const double xInc1= (double)xInc / (double)(1<<16);
1089 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
1090 else if (flags&SWS_X) sizeFactor= 8.0;
1091 else if (flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1092 else if (flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
1093 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1094 else if (flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
1095 else if (flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1096 else if (flags&SWS_BILINEAR) sizeFactor= 2.0;
1098 sizeFactor= 0.0; //GCC warning killer
1102 if (xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1103 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1105 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1106 if (filterSize > srcW-2) filterSize=srcW-2;
1108 filter= av_malloc(dstW*sizeof(double)*filterSize);
1110 xDstInSrc= xInc1 / 2.0 - 0.5;
1111 for (i=0; i<dstW; i++)
1113 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1115 (*filterPos)[i]= xx;
1116 for (j=0; j<filterSize; j++)
1118 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1120 if (flags & SWS_BICUBIC)
1122 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1123 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1126 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1128 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1132 /* else if (flags & SWS_X)
1134 double p= param ? param*0.01 : 0.3;
1135 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1136 coeff*= pow(2.0, - p*d*d);
1138 else if (flags & SWS_X)
1140 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1146 if (coeff<0.0) coeff= -pow(-coeff, A);
1147 else coeff= pow( coeff, A);
1148 coeff= coeff*0.5 + 0.5;
1150 else if (flags & SWS_AREA)
1152 double srcPixelSize= 1.0/xInc1;
1153 if (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1154 else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1157 else if (flags & SWS_GAUSS)
1159 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1160 coeff = pow(2.0, - p*d*d);
1162 else if (flags & SWS_SINC)
1164 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1166 else if (flags & SWS_LANCZOS)
1168 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1169 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1172 else if (flags & SWS_BILINEAR)
1175 if (coeff<0) coeff=0;
1177 else if (flags & SWS_SPLINE)
1179 double p=-2.196152422706632;
1180 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1183 coeff= 0.0; //GCC warning killer
1187 filter[i*filterSize + j]= coeff;
1194 /* apply src & dst Filter to filter -> filter2
1197 assert(filterSize>0);
1198 filter2Size= filterSize;
1199 if (srcFilter) filter2Size+= srcFilter->length - 1;
1200 if (dstFilter) filter2Size+= dstFilter->length - 1;
1201 assert(filter2Size>0);
1202 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1204 for (i=0; i<dstW; i++)
1207 SwsVector scaleFilter;
1210 scaleFilter.coeff= filter + i*filterSize;
1211 scaleFilter.length= filterSize;
1213 if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1214 else outVec= &scaleFilter;
1216 assert(outVec->length == filter2Size);
1219 for (j=0; j<outVec->length; j++)
1221 filter2[i*filter2Size + j]= outVec->coeff[j];
1224 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1226 if (outVec != &scaleFilter) sws_freeVec(outVec);
1230 /* try to reduce the filter-size (step1 find size and shift left) */
1231 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1233 for (i=dstW-1; i>=0; i--)
1235 int min= filter2Size;
1239 /* get rid off near zero elements on the left by shifting left */
1240 for (j=0; j<filter2Size; j++)
1243 cutOff += FFABS(filter2[i*filter2Size]);
1245 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1247 /* preserve monotonicity because the core can't handle the filter otherwise */
1248 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1250 // Move filter coeffs left
1251 for (k=1; k<filter2Size; k++)
1252 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1253 filter2[i*filter2Size + k - 1]= 0.0;
1258 /* count near zeros on the right */
1259 for (j=filter2Size-1; j>0; j--)
1261 cutOff += FFABS(filter2[i*filter2Size + j]);
1263 if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1267 if (min>minFilterSize) minFilterSize= min;
1270 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1271 // we can handle the special case 4,
1272 // so we don't want to go to the full 8
1273 if (minFilterSize < 5)
1276 // we really don't want to waste our time
1277 // doing useless computation, so fall-back on
1278 // the scalar C code for very small filter.
1279 // vectorizing is worth it only if you have
1280 // decent-sized vector.
1281 if (minFilterSize < 3)
1285 if (flags & SWS_CPU_CAPS_MMX) {
1286 // special case for unscaled vertical filtering
1287 if (minFilterSize == 1 && filterAlign == 2)
1291 assert(minFilterSize > 0);
1292 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1293 assert(filterSize > 0);
1294 filter= av_malloc(filterSize*dstW*sizeof(double));
1295 if (filterSize >= MAX_FILTER_SIZE || !filter)
1297 *outFilterSize= filterSize;
1299 if (flags&SWS_PRINT_INFO)
1300 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1301 /* try to reduce the filter-size (step2 reduce it) */
1302 for (i=0; i<dstW; i++)
1306 for (j=0; j<filterSize; j++)
1308 if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1309 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1314 //FIXME try to align filterpos if possible
1317 for (i=0; i<dstW; i++)
1320 if ((*filterPos)[i] < 0)
1322 // Move filter coeffs left to compensate for filterPos
1323 for (j=1; j<filterSize; j++)
1325 int left= FFMAX(j + (*filterPos)[i], 0);
1326 filter[i*filterSize + left] += filter[i*filterSize + j];
1327 filter[i*filterSize + j]=0;
1332 if ((*filterPos)[i] + filterSize > srcW)
1334 int shift= (*filterPos)[i] + filterSize - srcW;
1335 // Move filter coeffs right to compensate for filterPos
1336 for (j=filterSize-2; j>=0; j--)
1338 int right= FFMIN(j + shift, filterSize-1);
1339 filter[i*filterSize +right] += filter[i*filterSize +j];
1340 filter[i*filterSize +j]=0;
1342 (*filterPos)[i]= srcW - filterSize;
1346 // Note the +1 is for the MMXscaler which reads over the end
1347 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1348 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1350 /* Normalize & Store in outFilter */
1351 for (i=0; i<dstW; i++)
1358 for (j=0; j<filterSize; j++)
1360 sum+= filter[i*filterSize + j];
1363 for (j=0; j<*outFilterSize; j++)
1365 double v= filter[i*filterSize + j]*scale + error;
1366 int intV= floor(v + 0.5);
1367 (*outFilter)[i*(*outFilterSize) + j]= intV;
1372 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1373 for (i=0; i<*outFilterSize; i++)
1375 int j= dstW*(*outFilterSize);
1376 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1387 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1390 long imm8OfPShufW1A;
1391 long imm8OfPShufW2A;
1392 long fragmentLengthA;
1394 long imm8OfPShufW1B;
1395 long imm8OfPShufW2B;
1396 long fragmentLengthB;
1401 // create an optimized horizontal scaling routine
1409 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1410 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1411 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
1412 "punpcklbw %%mm7, %%mm1 \n\t"
1413 "punpcklbw %%mm7, %%mm0 \n\t"
1414 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1416 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1418 "psubw %%mm1, %%mm0 \n\t"
1419 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1420 "pmullw %%mm3, %%mm0 \n\t"
1421 "psllw $7, %%mm1 \n\t"
1422 "paddw %%mm1, %%mm0 \n\t"
1424 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1426 "add $8, %%"REG_a" \n\t"
1430 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1431 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1432 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1437 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1441 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1442 "=r" (fragmentLengthA)
1449 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
1450 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
1451 "punpcklbw %%mm7, %%mm0 \n\t"
1452 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1454 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1456 "psubw %%mm1, %%mm0 \n\t"
1457 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
1458 "pmullw %%mm3, %%mm0 \n\t"
1459 "psllw $7, %%mm1 \n\t"
1460 "paddw %%mm1, %%mm0 \n\t"
1462 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1464 "add $8, %%"REG_a" \n\t"
1468 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
1469 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
1470 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
1475 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
1479 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1480 "=r" (fragmentLengthB)
1483 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1486 for (i=0; i<dstW/numSplits; i++)
1493 int b=((xpos+xInc)>>16) - xx;
1494 int c=((xpos+xInc*2)>>16) - xx;
1495 int d=((xpos+xInc*3)>>16) - xx;
1497 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1498 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1499 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1500 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1505 int maxShift= 3-(d+1);
1508 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1510 funnyCode[fragmentPos + imm8OfPShufW1B]=
1511 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1512 funnyCode[fragmentPos + imm8OfPShufW2B]=
1513 a | (b<<2) | (c<<4) | (d<<6);
1515 if (i+3>=dstW) shift=maxShift; //avoid overread
1516 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1518 if (shift && i>=shift)
1520 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1521 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1522 filterPos[i/2]-=shift;
1525 fragmentPos+= fragmentLengthB;
1532 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1534 funnyCode[fragmentPos + imm8OfPShufW1A]=
1535 funnyCode[fragmentPos + imm8OfPShufW2A]=
1536 a | (b<<2) | (c<<4) | (d<<6);
1538 if (i+4>=dstW) shift=maxShift; //avoid overread
1539 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1541 if (shift && i>=shift)
1543 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1544 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1545 filterPos[i/2]-=shift;
1548 fragmentPos+= fragmentLengthA;
1551 funnyCode[fragmentPos]= RET;
1555 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1557 #endif /* COMPILE_MMX2 */
1559 static void globalInit(void){
1560 // generating tables:
1562 for (i=0; i<768; i++){
1563 int c= av_clip_uint8(i-256);
1568 static SwsFunc getSwsFunc(int flags){
1570 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1571 #if defined(ARCH_X86)
1572 // ordered per speed fastest first
1573 if (flags & SWS_CPU_CAPS_MMX2)
1574 return swScale_MMX2;
1575 else if (flags & SWS_CPU_CAPS_3DNOW)
1576 return swScale_3DNow;
1577 else if (flags & SWS_CPU_CAPS_MMX)
1584 if (flags & SWS_CPU_CAPS_ALTIVEC)
1585 return swScale_altivec;
1590 #endif /* defined(ARCH_X86) */
1591 #else //RUNTIME_CPUDETECT
1593 return swScale_MMX2;
1594 #elif defined (HAVE_3DNOW)
1595 return swScale_3DNow;
1596 #elif defined (HAVE_MMX)
1598 #elif defined (HAVE_ALTIVEC)
1599 return swScale_altivec;
1603 #endif //!RUNTIME_CPUDETECT
1606 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1607 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1608 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1610 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1611 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1615 uint8_t *srcPtr= src[0];
1616 uint8_t *dstPtr= dst;
1617 for (i=0; i<srcSliceH; i++)
1619 memcpy(dstPtr, srcPtr, c->srcW);
1620 srcPtr+= srcStride[0];
1621 dstPtr+= dstStride[0];
1624 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1625 if (c->dstFormat == PIX_FMT_NV12)
1626 interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1628 interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1633 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1634 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1635 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1637 yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1642 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1643 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1644 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1646 yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1651 static int YUV422PToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1652 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1653 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1655 yuv422ptoyuy2(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1660 static int YUV422PToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1661 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1662 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1664 yuv422ptouyvy(src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0]);
1669 /* {RGB,BGR}{15,16,24,32,32_1} -> {RGB,BGR}{15,16,24,32} */
1670 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1671 int srcSliceH, uint8_t* dst[], int dstStride[]){
1672 const int srcFormat= c->srcFormat;
1673 const int dstFormat= c->dstFormat;
1674 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1675 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1676 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1677 const int dstId= fmt_depth(dstFormat) >> 2;
1678 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1681 if ( (isBGR(srcFormat) && isBGR(dstFormat))
1682 || (isRGB(srcFormat) && isRGB(dstFormat))){
1683 switch(srcId | (dstId<<4)){
1684 case 0x34: conv= rgb16to15; break;
1685 case 0x36: conv= rgb24to15; break;
1686 case 0x38: conv= rgb32to15; break;
1687 case 0x43: conv= rgb15to16; break;
1688 case 0x46: conv= rgb24to16; break;
1689 case 0x48: conv= rgb32to16; break;
1690 case 0x63: conv= rgb15to24; break;
1691 case 0x64: conv= rgb16to24; break;
1692 case 0x68: conv= rgb32to24; break;
1693 case 0x83: conv= rgb15to32; break;
1694 case 0x84: conv= rgb16to32; break;
1695 case 0x86: conv= rgb24to32; break;
1696 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1697 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1699 }else if ( (isBGR(srcFormat) && isRGB(dstFormat))
1700 || (isRGB(srcFormat) && isBGR(dstFormat))){
1701 switch(srcId | (dstId<<4)){
1702 case 0x33: conv= rgb15tobgr15; break;
1703 case 0x34: conv= rgb16tobgr15; break;
1704 case 0x36: conv= rgb24tobgr15; break;
1705 case 0x38: conv= rgb32tobgr15; break;
1706 case 0x43: conv= rgb15tobgr16; break;
1707 case 0x44: conv= rgb16tobgr16; break;
1708 case 0x46: conv= rgb24tobgr16; break;
1709 case 0x48: conv= rgb32tobgr16; break;
1710 case 0x63: conv= rgb15tobgr24; break;
1711 case 0x64: conv= rgb16tobgr24; break;
1712 case 0x66: conv= rgb24tobgr24; break;
1713 case 0x68: conv= rgb32tobgr24; break;
1714 case 0x83: conv= rgb15tobgr32; break;
1715 case 0x84: conv= rgb16tobgr32; break;
1716 case 0x86: conv= rgb24tobgr32; break;
1717 case 0x88: conv= rgb32tobgr32; break;
1718 default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1719 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1722 av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1723 sws_format_name(srcFormat), sws_format_name(dstFormat));
1728 uint8_t *srcPtr= src[0];
1729 if(srcFormat == PIX_FMT_RGB32_1 || srcFormat == PIX_FMT_BGR32_1)
1730 srcPtr += ALT32_CORR;
1732 if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1733 conv(srcPtr, dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1739 for (i=0; i<srcSliceH; i++)
1741 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1742 srcPtr+= srcStride[0];
1743 dstPtr+= dstStride[0];
1750 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1751 int srcSliceH, uint8_t* dst[], int dstStride[]){
1755 dst[0]+ srcSliceY *dstStride[0],
1756 dst[1]+(srcSliceY>>1)*dstStride[1],
1757 dst[2]+(srcSliceY>>1)*dstStride[2],
1759 dstStride[0], dstStride[1], srcStride[0]);
1763 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1764 int srcSliceH, uint8_t* dst[], int dstStride[]){
1768 if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1769 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1771 uint8_t *srcPtr= src[0];
1772 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1774 for (i=0; i<srcSliceH; i++)
1776 memcpy(dstPtr, srcPtr, c->srcW);
1777 srcPtr+= srcStride[0];
1778 dstPtr+= dstStride[0];
1782 if (c->dstFormat==PIX_FMT_YUV420P){
1783 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1784 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1786 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1787 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1792 /* unscaled copy like stuff (assumes nearly identical formats) */
1793 static int packedCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1794 int srcSliceH, uint8_t* dst[], int dstStride[])
1796 if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1797 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1801 uint8_t *srcPtr= src[0];
1802 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1805 /* universal length finder */
1806 while(length+c->srcW <= FFABS(dstStride[0])
1807 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1810 for (i=0; i<srcSliceH; i++)
1812 memcpy(dstPtr, srcPtr, length);
1813 srcPtr+= srcStride[0];
1814 dstPtr+= dstStride[0];
1820 static int planarCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1821 int srcSliceH, uint8_t* dst[], int dstStride[])
1824 for (plane=0; plane<3; plane++)
1826 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1827 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1828 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1830 if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1832 if (!isGray(c->dstFormat))
1833 memset(dst[plane], 128, dstStride[plane]*height);
1837 if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1838 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1842 uint8_t *srcPtr= src[plane];
1843 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1844 for (i=0; i<height; i++)
1846 memcpy(dstPtr, srcPtr, length);
1847 srcPtr+= srcStride[plane];
1848 dstPtr+= dstStride[plane];
1856 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1857 int srcSliceH, uint8_t* dst[], int dstStride[]){
1859 int length= c->srcW;
1861 int height= srcSliceH;
1863 uint8_t *srcPtr= src[0];
1864 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1866 if (!isGray(c->dstFormat)){
1867 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1868 memset(dst[1], 128, dstStride[1]*height);
1869 memset(dst[2], 128, dstStride[2]*height);
1871 if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1872 for (i=0; i<height; i++)
1874 for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1875 srcPtr+= srcStride[0];
1876 dstPtr+= dstStride[0];
1881 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1882 int srcSliceH, uint8_t* dst[], int dstStride[]){
1884 int length= c->srcW;
1886 int height= srcSliceH;
1888 uint8_t *srcPtr= src[0];
1889 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1890 for (i=0; i<height; i++)
1892 for (j=0; j<length; j++)
1894 dstPtr[j<<1] = srcPtr[j];
1895 dstPtr[(j<<1)+1] = srcPtr[j];
1897 srcPtr+= srcStride[0];
1898 dstPtr+= dstStride[0];
1903 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1904 int srcSliceH, uint8_t* dst[], int dstStride[]){
1906 int length= c->srcW;
1908 int height= srcSliceH;
1910 uint16_t *srcPtr= (uint16_t*)src[0];
1911 uint16_t *dstPtr= (uint16_t*)(dst[0] + dstStride[0]*y/2);
1912 for (i=0; i<height; i++)
1914 for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1915 srcPtr+= srcStride[0]/2;
1916 dstPtr+= dstStride[0]/2;
1922 static void getSubSampleFactors(int *h, int *v, int format){
1924 case PIX_FMT_UYVY422:
1925 case PIX_FMT_YUYV422:
1929 case PIX_FMT_YUV420P:
1930 case PIX_FMT_YUVA420P:
1931 case PIX_FMT_GRAY16BE:
1932 case PIX_FMT_GRAY16LE:
1933 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1939 case PIX_FMT_YUV440P:
1943 case PIX_FMT_YUV410P:
1947 case PIX_FMT_YUV444P:
1951 case PIX_FMT_YUV422P:
1955 case PIX_FMT_YUV411P:
1966 static uint16_t roundToInt16(int64_t f){
1967 int r= (f + (1<<15))>>16;
1968 if (r<-0x7FFF) return 0x8000;
1969 else if (r> 0x7FFF) return 0x7FFF;
1974 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1975 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
1976 * @return -1 if not supported
1978 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1979 int64_t crv = inv_table[0];
1980 int64_t cbu = inv_table[1];
1981 int64_t cgu = -inv_table[2];
1982 int64_t cgv = -inv_table[3];
1986 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1987 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1988 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1990 c->brightness= brightness;
1991 c->contrast = contrast;
1992 c->saturation= saturation;
1993 c->srcRange = srcRange;
1994 c->dstRange = dstRange;
1996 c->uOffset= 0x0400040004000400LL;
1997 c->vOffset= 0x0400040004000400LL;
2003 crv= (crv*224) / 255;
2004 cbu= (cbu*224) / 255;
2005 cgu= (cgu*224) / 255;
2006 cgv= (cgv*224) / 255;
2009 cy = (cy *contrast )>>16;
2010 crv= (crv*contrast * saturation)>>32;
2011 cbu= (cbu*contrast * saturation)>>32;
2012 cgu= (cgu*contrast * saturation)>>32;
2013 cgv= (cgv*contrast * saturation)>>32;
2015 oy -= 256*brightness;
2017 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
2018 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
2019 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
2020 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
2021 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
2022 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
2024 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
2027 #ifdef COMPILE_ALTIVEC
2028 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
2029 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
2035 * @return -1 if not supported
2037 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
2038 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
2040 *inv_table = c->srcColorspaceTable;
2041 *table = c->dstColorspaceTable;
2042 *srcRange = c->srcRange;
2043 *dstRange = c->dstRange;
2044 *brightness= c->brightness;
2045 *contrast = c->contrast;
2046 *saturation= c->saturation;
2051 static int handle_jpeg(int *format)
2054 case PIX_FMT_YUVJ420P:
2055 *format = PIX_FMT_YUV420P;
2057 case PIX_FMT_YUVJ422P:
2058 *format = PIX_FMT_YUV422P;
2060 case PIX_FMT_YUVJ444P:
2061 *format = PIX_FMT_YUV444P;
2063 case PIX_FMT_YUVJ440P:
2064 *format = PIX_FMT_YUV440P;
2071 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2072 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2076 int usesVFilter, usesHFilter;
2077 int unscaled, needsDither;
2078 int srcRange, dstRange;
2079 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2080 #if defined(ARCH_X86)
2081 if (flags & SWS_CPU_CAPS_MMX)
2082 asm volatile("emms\n\t"::: "memory");
2085 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2086 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2088 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2089 #elif defined (HAVE_3DNOW)
2090 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2091 #elif defined (HAVE_MMX)
2092 flags |= SWS_CPU_CAPS_MMX;
2093 #elif defined (HAVE_ALTIVEC)
2094 flags |= SWS_CPU_CAPS_ALTIVEC;
2095 #elif defined (ARCH_BFIN)
2096 flags |= SWS_CPU_CAPS_BFIN;
2098 #endif /* RUNTIME_CPUDETECT */
2099 if (clip_table[512] != 255) globalInit();
2100 if (!rgb15to16) sws_rgb2rgb_init(flags);
2102 unscaled = (srcW == dstW && srcH == dstH);
2103 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2104 && (fmt_depth(dstFormat))<24
2105 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2107 srcRange = handle_jpeg(&srcFormat);
2108 dstRange = handle_jpeg(&dstFormat);
2110 if (!isSupportedIn(srcFormat))
2112 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2115 if (!isSupportedOut(dstFormat))
2117 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2121 i= flags & ( SWS_POINT
2132 if(!i || (i & (i-1)))
2134 av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n");
2140 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
2142 av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2143 srcW, srcH, dstW, dstH);
2146 if(srcW > VOFW || dstW > VOFW){
2147 av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2151 if (!dstFilter) dstFilter= &dummyFilter;
2152 if (!srcFilter) srcFilter= &dummyFilter;
2154 c= av_mallocz(sizeof(SwsContext));
2156 c->av_class = &sws_context_class;
2161 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2162 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2164 c->dstFormat= dstFormat;
2165 c->srcFormat= srcFormat;
2166 c->vRounder= 4* 0x0001000100010001ULL;
2168 usesHFilter= usesVFilter= 0;
2169 if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2170 if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2171 if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2172 if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2173 if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2174 if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2175 if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2176 if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2178 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2179 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2181 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2182 if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2184 // drop some chroma lines if the user wants it
2185 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2186 c->chrSrcVSubSample+= c->vChrDrop;
2188 // drop every 2. pixel for chroma calculation unless user wants full chroma
2189 if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2190 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
2191 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
2192 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2193 c->chrSrcHSubSample=1;
2196 c->param[0] = param[0];
2197 c->param[1] = param[1];
2200 c->param[1] = SWS_PARAM_DEFAULT;
2203 c->chrIntHSubSample= c->chrDstHSubSample;
2204 c->chrIntVSubSample= c->chrSrcVSubSample;
2206 // Note the -((-x)>>y) is so that we always round toward +inf.
2207 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2208 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2209 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2210 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2212 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);
2214 /* unscaled special Cases */
2215 if (unscaled && !usesHFilter && !usesVFilter)
2218 if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2220 c->swScale= PlanarToNV12Wrapper;
2224 if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2226 c->swScale= yuv2rgb_get_func_ptr(c);
2230 if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2232 c->swScale= yvu9toyv12Wrapper;
2236 if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2237 c->swScale= bgr24toyv12Wrapper;
2239 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2240 if ( (isBGR(srcFormat) || isRGB(srcFormat))
2241 && (isBGR(dstFormat) || isRGB(dstFormat))
2242 && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2243 && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2244 && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2245 && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2246 && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2247 && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2248 && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2249 && dstFormat != PIX_FMT_RGB32_1
2250 && dstFormat != PIX_FMT_BGR32_1
2251 && (!needsDither || (c->flags&(SWS_FAST_BILINEAR|SWS_POINT))))
2252 c->swScale= rgb2rgbWrapper;
2254 if (srcFormat == PIX_FMT_YUV422P)
2256 if (dstFormat == PIX_FMT_YUYV422)
2257 c->swScale= YUV422PToYuy2Wrapper;
2258 else if (dstFormat == PIX_FMT_UYVY422)
2259 c->swScale= YUV422PToUyvyWrapper;
2262 /* LQ converters if -sws 0 or -sws 4*/
2263 if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2265 if (srcFormat == PIX_FMT_YUV420P &&
2266 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2268 if (dstFormat == PIX_FMT_YUYV422)
2269 c->swScale= PlanarToYuy2Wrapper;
2271 c->swScale= PlanarToUyvyWrapper;
2275 #ifdef COMPILE_ALTIVEC
2276 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2277 ((srcFormat == PIX_FMT_YUV420P &&
2278 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2279 // unscaled YV12 -> packed YUV, we want speed
2280 if (dstFormat == PIX_FMT_YUYV422)
2281 c->swScale= yv12toyuy2_unscaled_altivec;
2283 c->swScale= yv12touyvy_unscaled_altivec;
2288 if ( srcFormat == dstFormat
2289 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2290 || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2292 if (isPacked(c->srcFormat))
2293 c->swScale= packedCopy;
2294 else /* Planar YUV or gray */
2295 c->swScale= planarCopy;
2298 /* gray16{le,be} conversions */
2299 if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2301 c->swScale= gray16togray;
2303 if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2305 c->swScale= graytogray16;
2307 if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2309 c->swScale= gray16swap;
2313 if (flags & SWS_CPU_CAPS_BFIN)
2314 ff_bfin_get_unscaled_swscale (c);
2318 if (flags&SWS_PRINT_INFO)
2319 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2320 sws_format_name(srcFormat), sws_format_name(dstFormat));
2325 if (flags & SWS_CPU_CAPS_MMX2)
2327 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2328 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2330 if (flags&SWS_PRINT_INFO)
2331 av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2333 if (usesHFilter) c->canMMX2BeUsed=0;
2338 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2339 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2341 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2342 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2343 // n-2 is the last chrominance sample available
2344 // this is not perfect, but no one should notice the difference, the more correct variant
2345 // would be like the vertical one, but that would require some special code for the
2346 // first and last pixel
2347 if (flags&SWS_FAST_BILINEAR)
2349 if (c->canMMX2BeUsed)
2354 //we don't use the x86asm scaler if mmx is available
2355 else if (flags & SWS_CPU_CAPS_MMX)
2357 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2358 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2362 /* precalculate horizontal scaler filter coefficients */
2364 const int filterAlign=
2365 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2366 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2369 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2370 srcW , dstW, filterAlign, 1<<14,
2371 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2372 srcFilter->lumH, dstFilter->lumH, c->param);
2373 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2374 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2375 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2376 srcFilter->chrH, dstFilter->chrH, c->param);
2378 #define MAX_FUNNY_CODE_SIZE 10000
2379 #if defined(COMPILE_MMX2)
2380 // can't downscale !!!
2381 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2383 #ifdef MAP_ANONYMOUS
2384 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2385 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2387 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2388 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2391 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2392 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2393 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2394 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2396 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2397 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2399 #endif /* defined(COMPILE_MMX2) */
2400 } // Init Horizontal stuff
2404 /* precalculate vertical scaler filter coefficients */
2406 const int filterAlign=
2407 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2408 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2411 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2412 srcH , dstH, filterAlign, (1<<12)-4,
2413 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2414 srcFilter->lumV, dstFilter->lumV, c->param);
2415 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2416 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2417 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2418 srcFilter->chrV, dstFilter->chrV, c->param);
2421 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2422 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2424 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2426 short *p = (short *)&c->vYCoeffsBank[i];
2428 p[j] = c->vLumFilter[i];
2431 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2433 short *p = (short *)&c->vCCoeffsBank[i];
2435 p[j] = c->vChrFilter[i];
2440 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2441 c->vLumBufSize= c->vLumFilterSize;
2442 c->vChrBufSize= c->vChrFilterSize;
2443 for (i=0; i<dstH; i++)
2445 int chrI= i*c->chrDstH / dstH;
2446 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2447 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2449 nextSlice>>= c->chrSrcVSubSample;
2450 nextSlice<<= c->chrSrcVSubSample;
2451 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2452 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2453 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2454 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2457 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2458 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2459 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2460 //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)
2461 /* align at 16 bytes for AltiVec */
2462 for (i=0; i<c->vLumBufSize; i++)
2463 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2464 for (i=0; i<c->vChrBufSize; i++)
2465 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2467 //try to avoid drawing green stuff between the right end and the stride end
2468 for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2470 assert(2*VOFW == VOF);
2472 assert(c->chrDstH <= dstH);
2474 if (flags&SWS_PRINT_INFO)
2477 const char *dither= " dithered";
2479 const char *dither= "";
2481 if (flags&SWS_FAST_BILINEAR)
2482 av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2483 else if (flags&SWS_BILINEAR)
2484 av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2485 else if (flags&SWS_BICUBIC)
2486 av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2487 else if (flags&SWS_X)
2488 av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2489 else if (flags&SWS_POINT)
2490 av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2491 else if (flags&SWS_AREA)
2492 av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2493 else if (flags&SWS_BICUBLIN)
2494 av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2495 else if (flags&SWS_GAUSS)
2496 av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2497 else if (flags&SWS_SINC)
2498 av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2499 else if (flags&SWS_LANCZOS)
2500 av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2501 else if (flags&SWS_SPLINE)
2502 av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2504 av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2506 if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2507 av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2508 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2510 av_log(c, AV_LOG_INFO, "from %s to %s ",
2511 sws_format_name(srcFormat), sws_format_name(dstFormat));
2513 if (flags & SWS_CPU_CAPS_MMX2)
2514 av_log(c, AV_LOG_INFO, "using MMX2\n");
2515 else if (flags & SWS_CPU_CAPS_3DNOW)
2516 av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2517 else if (flags & SWS_CPU_CAPS_MMX)
2518 av_log(c, AV_LOG_INFO, "using MMX\n");
2519 else if (flags & SWS_CPU_CAPS_ALTIVEC)
2520 av_log(c, AV_LOG_INFO, "using AltiVec\n");
2522 av_log(c, AV_LOG_INFO, "using C\n");
2525 if (flags & SWS_PRINT_INFO)
2527 if (flags & SWS_CPU_CAPS_MMX)
2529 if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2530 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2533 if (c->hLumFilterSize==4)
2534 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2535 else if (c->hLumFilterSize==8)
2536 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2538 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2540 if (c->hChrFilterSize==4)
2541 av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2542 else if (c->hChrFilterSize==8)
2543 av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2545 av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2550 #if defined(ARCH_X86)
2551 av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2553 if (flags & SWS_FAST_BILINEAR)
2554 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2556 av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2559 if (isPlanarYUV(dstFormat))
2561 if (c->vLumFilterSize==1)
2562 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2564 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2568 if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2569 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2570 " 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2571 else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2572 av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2574 av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2577 if (dstFormat==PIX_FMT_BGR24)
2578 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2579 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2580 else if (dstFormat==PIX_FMT_RGB32)
2581 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2582 else if (dstFormat==PIX_FMT_BGR565)
2583 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2584 else if (dstFormat==PIX_FMT_BGR555)
2585 av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2587 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2589 if (flags & SWS_PRINT_INFO)
2591 av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2592 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2593 av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2594 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2597 c->swScale= getSwsFunc(flags);
2602 * swscale wrapper, so we don't need to export the SwsContext.
2603 * assumes planar YUV to be in YUV order instead of YVU
2605 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2606 int srcSliceH, uint8_t* dst[], int dstStride[]){
2608 uint8_t* src2[4]= {src[0], src[1], src[2]};
2610 int use_pal= c->srcFormat == PIX_FMT_PAL8
2611 || c->srcFormat == PIX_FMT_BGR4_BYTE
2612 || c->srcFormat == PIX_FMT_RGB4_BYTE
2613 || c->srcFormat == PIX_FMT_BGR8
2614 || c->srcFormat == PIX_FMT_RGB8;
2616 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2617 av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2620 if (c->sliceDir == 0) {
2621 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2625 for (i=0; i<256; i++){
2626 int p, r, g, b,y,u,v;
2627 if(c->srcFormat == PIX_FMT_PAL8){
2628 p=((uint32_t*)(src[1]))[i];
2632 }else if(c->srcFormat == PIX_FMT_RGB8){
2636 }else if(c->srcFormat == PIX_FMT_BGR8){
2640 }else if(c->srcFormat == PIX_FMT_RGB4_BYTE){
2644 }else if(c->srcFormat == PIX_FMT_BGR4_BYTE){
2649 y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2650 u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2651 v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2652 pal[i]= y + (u<<8) + (v<<16);
2654 src2[1]= (uint8_t*)pal;
2657 // copy strides, so they can safely be modified
2658 if (c->sliceDir == 1) {
2659 // slices go from top to bottom
2660 int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2661 int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2662 return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2664 // slices go from bottom to top => we flip the image internally
2665 uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2666 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2667 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2668 int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2669 int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2671 src2[0] += (srcSliceH-1)*srcStride[0];
2673 src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2674 src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2676 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2681 * swscale wrapper, so we don't need to export the SwsContext
2683 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2684 int srcSliceH, uint8_t* dst[], int dstStride[]){
2685 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2688 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2689 float lumaSharpen, float chromaSharpen,
2690 float chromaHShift, float chromaVShift,
2693 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2695 if (lumaGBlur!=0.0){
2696 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2697 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2699 filter->lumH= sws_getIdentityVec();
2700 filter->lumV= sws_getIdentityVec();
2703 if (chromaGBlur!=0.0){
2704 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2705 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2707 filter->chrH= sws_getIdentityVec();
2708 filter->chrV= sws_getIdentityVec();
2711 if (chromaSharpen!=0.0){
2712 SwsVector *id= sws_getIdentityVec();
2713 sws_scaleVec(filter->chrH, -chromaSharpen);
2714 sws_scaleVec(filter->chrV, -chromaSharpen);
2715 sws_addVec(filter->chrH, id);
2716 sws_addVec(filter->chrV, id);
2720 if (lumaSharpen!=0.0){
2721 SwsVector *id= sws_getIdentityVec();
2722 sws_scaleVec(filter->lumH, -lumaSharpen);
2723 sws_scaleVec(filter->lumV, -lumaSharpen);
2724 sws_addVec(filter->lumH, id);
2725 sws_addVec(filter->lumV, id);
2729 if (chromaHShift != 0.0)
2730 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2732 if (chromaVShift != 0.0)
2733 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2735 sws_normalizeVec(filter->chrH, 1.0);
2736 sws_normalizeVec(filter->chrV, 1.0);
2737 sws_normalizeVec(filter->lumH, 1.0);
2738 sws_normalizeVec(filter->lumV, 1.0);
2740 if (verbose) sws_printVec(filter->chrH);
2741 if (verbose) sws_printVec(filter->lumH);
2747 * returns a normalized gaussian curve used to filter stuff
2748 * quality=3 is high quality, lowwer is lowwer quality
2750 SwsVector *sws_getGaussianVec(double variance, double quality){
2751 const int length= (int)(variance*quality + 0.5) | 1;
2753 double *coeff= av_malloc(length*sizeof(double));
2754 double middle= (length-1)*0.5;
2755 SwsVector *vec= av_malloc(sizeof(SwsVector));
2758 vec->length= length;
2760 for (i=0; i<length; i++)
2762 double dist= i-middle;
2763 coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2766 sws_normalizeVec(vec, 1.0);
2771 SwsVector *sws_getConstVec(double c, int length){
2773 double *coeff= av_malloc(length*sizeof(double));
2774 SwsVector *vec= av_malloc(sizeof(SwsVector));
2777 vec->length= length;
2779 for (i=0; i<length; i++)
2786 SwsVector *sws_getIdentityVec(void){
2787 return sws_getConstVec(1.0, 1);
2790 double sws_dcVec(SwsVector *a){
2794 for (i=0; i<a->length; i++)
2800 void sws_scaleVec(SwsVector *a, double scalar){
2803 for (i=0; i<a->length; i++)
2804 a->coeff[i]*= scalar;
2807 void sws_normalizeVec(SwsVector *a, double height){
2808 sws_scaleVec(a, height/sws_dcVec(a));
2811 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2812 int length= a->length + b->length - 1;
2813 double *coeff= av_malloc(length*sizeof(double));
2815 SwsVector *vec= av_malloc(sizeof(SwsVector));
2818 vec->length= length;
2820 for (i=0; i<length; i++) coeff[i]= 0.0;
2822 for (i=0; i<a->length; i++)
2824 for (j=0; j<b->length; j++)
2826 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2833 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2834 int length= FFMAX(a->length, b->length);
2835 double *coeff= av_malloc(length*sizeof(double));
2837 SwsVector *vec= av_malloc(sizeof(SwsVector));
2840 vec->length= length;
2842 for (i=0; i<length; i++) coeff[i]= 0.0;
2844 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2845 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2850 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2851 int length= FFMAX(a->length, b->length);
2852 double *coeff= av_malloc(length*sizeof(double));
2854 SwsVector *vec= av_malloc(sizeof(SwsVector));
2857 vec->length= length;
2859 for (i=0; i<length; i++) coeff[i]= 0.0;
2861 for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2862 for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2867 /* shift left / or right if "shift" is negative */
2868 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2869 int length= a->length + FFABS(shift)*2;
2870 double *coeff= av_malloc(length*sizeof(double));
2872 SwsVector *vec= av_malloc(sizeof(SwsVector));
2875 vec->length= length;
2877 for (i=0; i<length; i++) coeff[i]= 0.0;
2879 for (i=0; i<a->length; i++)
2881 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2887 void sws_shiftVec(SwsVector *a, int shift){
2888 SwsVector *shifted= sws_getShiftedVec(a, shift);
2890 a->coeff= shifted->coeff;
2891 a->length= shifted->length;
2895 void sws_addVec(SwsVector *a, SwsVector *b){
2896 SwsVector *sum= sws_sumVec(a, b);
2898 a->coeff= sum->coeff;
2899 a->length= sum->length;
2903 void sws_subVec(SwsVector *a, SwsVector *b){
2904 SwsVector *diff= sws_diffVec(a, b);
2906 a->coeff= diff->coeff;
2907 a->length= diff->length;
2911 void sws_convVec(SwsVector *a, SwsVector *b){
2912 SwsVector *conv= sws_getConvVec(a, b);
2914 a->coeff= conv->coeff;
2915 a->length= conv->length;
2919 SwsVector *sws_cloneVec(SwsVector *a){
2920 double *coeff= av_malloc(a->length*sizeof(double));
2922 SwsVector *vec= av_malloc(sizeof(SwsVector));
2925 vec->length= a->length;
2927 for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2932 void sws_printVec(SwsVector *a){
2938 for (i=0; i<a->length; i++)
2939 if (a->coeff[i]>max) max= a->coeff[i];
2941 for (i=0; i<a->length; i++)
2942 if (a->coeff[i]<min) min= a->coeff[i];
2946 for (i=0; i<a->length; i++)
2948 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2949 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2950 for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2951 av_log(NULL, AV_LOG_DEBUG, "|\n");
2955 void sws_freeVec(SwsVector *a){
2957 av_freep(&a->coeff);
2962 void sws_freeFilter(SwsFilter *filter){
2963 if (!filter) return;
2965 if (filter->lumH) sws_freeVec(filter->lumH);
2966 if (filter->lumV) sws_freeVec(filter->lumV);
2967 if (filter->chrH) sws_freeVec(filter->chrH);
2968 if (filter->chrV) sws_freeVec(filter->chrV);
2973 void sws_freeContext(SwsContext *c){
2979 for (i=0; i<c->vLumBufSize; i++)
2980 av_freep(&c->lumPixBuf[i]);
2981 av_freep(&c->lumPixBuf);
2986 for (i=0; i<c->vChrBufSize; i++)
2987 av_freep(&c->chrPixBuf[i]);
2988 av_freep(&c->chrPixBuf);
2991 av_freep(&c->vLumFilter);
2992 av_freep(&c->vChrFilter);
2993 av_freep(&c->hLumFilter);
2994 av_freep(&c->hChrFilter);
2996 av_freep(&c->vYCoeffsBank);
2997 av_freep(&c->vCCoeffsBank);
3000 av_freep(&c->vLumFilterPos);
3001 av_freep(&c->vChrFilterPos);
3002 av_freep(&c->hLumFilterPos);
3003 av_freep(&c->hChrFilterPos);
3005 #if defined(ARCH_X86) && defined(CONFIG_GPL)
3006 #ifdef MAP_ANONYMOUS
3007 if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
3008 if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
3010 av_free(c->funnyYCode);
3011 av_free(c->funnyUVCode);
3014 c->funnyUVCode=NULL;
3015 #endif /* defined(ARCH_X86) */
3017 av_freep(&c->lumMmx2Filter);
3018 av_freep(&c->chrMmx2Filter);
3019 av_freep(&c->lumMmx2FilterPos);
3020 av_freep(&c->chrMmx2FilterPos);
3021 av_freep(&c->yuvTable);
3027 * Checks if context is valid or reallocs a new one instead.
3028 * If context is NULL, just calls sws_getContext() to get a new one.
3029 * Otherwise, checks if the parameters are the same already saved in context.
3030 * If that is the case, returns the current context.
3031 * Otherwise, frees context and gets a new one.
3033 * Be warned that srcFilter, dstFilter are not checked, they are
3034 * asumed to remain valid.
3036 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
3037 int srcW, int srcH, int srcFormat,
3038 int dstW, int dstH, int dstFormat, int flags,
3039 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
3041 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
3044 param = default_param;
3047 if (context->srcW != srcW || context->srcH != srcH ||
3048 context->srcFormat != srcFormat ||
3049 context->dstW != dstW || context->dstH != dstH ||
3050 context->dstFormat != dstFormat || context->flags != flags ||
3051 context->param[0] != param[0] || context->param[1] != param[1])
3053 sws_freeContext(context);
3058 return sws_getContext(srcW, srcH, srcFormat,
3059 dstW, dstH, dstFormat, flags,
3060 srcFilter, dstFilter, param);