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 St, 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
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 didnt write it down ...)
48 untested special converters
49 YV12/I420 -> BGR15/BGR24/BGR32 (its 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
69 #ifdef HAVE_SYS_MMAN_H
71 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
72 #define MAP_ANONYMOUS MAP_ANON
76 #include "swscale_internal.h"
81 #include "libvo/fastmemcpy.h"
91 //#define WORDS_BIGENDIAN
94 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
96 #define RET 0xC3 //near return opcode for X86
99 #define ASSERT(x) assert(x);
107 #define PI 3.14159265358979323846
110 #define isSupportedIn(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
111 || (x)==PIX_FMT_RGB32|| (x)==PIX_FMT_BGR24|| (x)==PIX_FMT_BGR565|| (x)==PIX_FMT_BGR555\
112 || (x)==PIX_FMT_BGR32|| (x)==PIX_FMT_RGB24|| (x)==PIX_FMT_RGB565|| (x)==PIX_FMT_RGB555\
113 || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P\
114 || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
115 || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P)
116 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
117 || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
118 || isRGB(x) || isBGR(x)\
119 || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
120 || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
121 || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
122 #define isPacked(x) ((x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422 ||isRGB(x) || isBGR(x))
124 #define RGB2YUV_SHIFT 16
125 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
126 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
127 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
129 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
130 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
131 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
132 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
133 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
135 extern const int32_t Inverse_Table_6_9[8][4];
139 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
142 more intelligent missalignment avoidance for the horizontal scaler
143 write special vertical cubic upscale version
144 Optimize C code (yv12 / minmax)
145 add support for packed pixel yuv input & output
146 add support for Y8 output
147 optimize bgr24 & bgr32
148 add BGR4 output support
149 write special BGR->BGR scaler
152 #if defined(ARCH_X86) && defined (CONFIG_GPL)
153 static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
154 static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
155 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
156 static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL;
157 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
158 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
159 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
160 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
162 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
163 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
164 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
165 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
167 static uint64_t __attribute__((aligned(8))) dither4[2]={
168 0x0103010301030103LL,
169 0x0200020002000200LL,};
171 static uint64_t __attribute__((aligned(8))) dither8[2]={
172 0x0602060206020602LL,
173 0x0004000400040004LL,};
175 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
176 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
177 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
178 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
179 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
180 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
182 static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
183 static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
184 static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
187 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
188 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
189 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
191 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
192 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
193 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
194 #endif /* FAST_BGR2YV12 */
195 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
196 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
197 static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
198 #endif /* defined(ARCH_X86) */
200 // clipping helper table for C implementations:
201 static unsigned char clip_table[768];
203 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
205 extern const uint8_t dither_2x2_4[2][8];
206 extern const uint8_t dither_2x2_8[2][8];
207 extern const uint8_t dither_8x8_32[8][8];
208 extern const uint8_t dither_8x8_73[8][8];
209 extern const uint8_t dither_8x8_220[8][8];
211 char *sws_format_name(enum PixelFormat format)
214 case PIX_FMT_YUV420P:
216 case PIX_FMT_YUYV422:
222 case PIX_FMT_YUV422P:
224 case PIX_FMT_YUV444P:
228 case PIX_FMT_YUV410P:
230 case PIX_FMT_YUV411P:
236 case PIX_FMT_GRAY16BE:
238 case PIX_FMT_GRAY16LE:
242 case PIX_FMT_MONOWHITE:
244 case PIX_FMT_MONOBLACK:
248 case PIX_FMT_YUVJ420P:
250 case PIX_FMT_YUVJ422P:
252 case PIX_FMT_YUVJ444P:
254 case PIX_FMT_XVMC_MPEG2_MC:
255 return "xvmc_mpeg2_mc";
256 case PIX_FMT_XVMC_MPEG2_IDCT:
257 return "xvmc_mpeg2_idct";
258 case PIX_FMT_UYVY422:
260 case PIX_FMT_UYYVYY411:
262 case PIX_FMT_RGB32_1:
264 case PIX_FMT_BGR32_1:
276 case PIX_FMT_BGR4_BYTE:
282 case PIX_FMT_RGB4_BYTE:
289 return "Unknown format";
293 #if defined(ARCH_X86) && defined (CONFIG_GPL)
294 void in_asm_used_var_warning_killer()
296 volatile int i= bF8+bFC+w10+
297 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
298 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
303 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
304 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
305 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
307 //FIXME Optimize (just quickly writen not opti..)
309 for(i=0; i<dstW; i++)
313 for(j=0; j<lumFilterSize; j++)
314 val += lumSrc[j][i] * lumFilter[j];
316 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
320 for(i=0; i<chrDstW; i++)
325 for(j=0; j<chrFilterSize; j++)
327 u += chrSrc[j][i] * chrFilter[j];
328 v += chrSrc[j][i + 2048] * chrFilter[j];
331 uDest[i]= FFMIN(FFMAX(u>>19, 0), 255);
332 vDest[i]= FFMIN(FFMAX(v>>19, 0), 255);
336 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
337 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
338 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
340 //FIXME Optimize (just quickly writen not opti..)
342 for(i=0; i<dstW; i++)
346 for(j=0; j<lumFilterSize; j++)
347 val += lumSrc[j][i] * lumFilter[j];
349 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
355 if(dstFormat == PIX_FMT_NV12)
356 for(i=0; i<chrDstW; i++)
361 for(j=0; j<chrFilterSize; j++)
363 u += chrSrc[j][i] * chrFilter[j];
364 v += chrSrc[j][i + 2048] * chrFilter[j];
367 uDest[2*i]= FFMIN(FFMAX(u>>19, 0), 255);
368 uDest[2*i+1]= FFMIN(FFMAX(v>>19, 0), 255);
371 for(i=0; i<chrDstW; i++)
376 for(j=0; j<chrFilterSize; j++)
378 u += chrSrc[j][i] * chrFilter[j];
379 v += chrSrc[j][i + 2048] * chrFilter[j];
382 uDest[2*i]= FFMIN(FFMAX(v>>19, 0), 255);
383 uDest[2*i+1]= FFMIN(FFMAX(u>>19, 0), 255);
387 #define YSCALE_YUV_2_PACKEDX_C(type) \
388 for(i=0; i<(dstW>>1); i++){\
397 for(j=0; j<lumFilterSize; j++)\
399 Y1 += lumSrc[j][i2] * lumFilter[j];\
400 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
402 for(j=0; j<chrFilterSize; j++)\
404 U += chrSrc[j][i] * chrFilter[j];\
405 V += chrSrc[j][i+2048] * chrFilter[j];\
423 #define YSCALE_YUV_2_RGBX_C(type) \
424 YSCALE_YUV_2_PACKEDX_C(type)\
426 g = c->table_gU[U] + c->table_gV[V];\
429 #define YSCALE_YUV_2_PACKED2_C \
430 for(i=0; i<(dstW>>1); i++){\
432 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
433 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
434 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
435 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
437 #define YSCALE_YUV_2_RGB2_C(type) \
438 YSCALE_YUV_2_PACKED2_C\
441 g = c->table_gU[U] + c->table_gV[V];\
444 #define YSCALE_YUV_2_PACKED1_C \
445 for(i=0; i<(dstW>>1); i++){\
447 int Y1= buf0[i2 ]>>7;\
448 int Y2= buf0[i2+1]>>7;\
449 int U= (uvbuf1[i ])>>7;\
450 int V= (uvbuf1[i+2048])>>7;\
452 #define YSCALE_YUV_2_RGB1_C(type) \
453 YSCALE_YUV_2_PACKED1_C\
456 g = c->table_gU[U] + c->table_gV[V];\
459 #define YSCALE_YUV_2_PACKED1B_C \
460 for(i=0; i<(dstW>>1); i++){\
462 int Y1= buf0[i2 ]>>7;\
463 int Y2= buf0[i2+1]>>7;\
464 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
465 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
467 #define YSCALE_YUV_2_RGB1B_C(type) \
468 YSCALE_YUV_2_PACKED1B_C\
471 g = c->table_gU[U] + c->table_gV[V];\
474 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
475 switch(c->dstFormat)\
480 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
481 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
486 ((uint8_t*)dest)[0]= r[Y1];\
487 ((uint8_t*)dest)[1]= g[Y1];\
488 ((uint8_t*)dest)[2]= b[Y1];\
489 ((uint8_t*)dest)[3]= r[Y2];\
490 ((uint8_t*)dest)[4]= g[Y2];\
491 ((uint8_t*)dest)[5]= b[Y2];\
497 ((uint8_t*)dest)[0]= b[Y1];\
498 ((uint8_t*)dest)[1]= g[Y1];\
499 ((uint8_t*)dest)[2]= r[Y1];\
500 ((uint8_t*)dest)[3]= b[Y2];\
501 ((uint8_t*)dest)[4]= g[Y2];\
502 ((uint8_t*)dest)[5]= r[Y2];\
506 case PIX_FMT_RGB565:\
507 case PIX_FMT_BGR565:\
509 const int dr1= dither_2x2_8[y&1 ][0];\
510 const int dg1= dither_2x2_4[y&1 ][0];\
511 const int db1= dither_2x2_8[(y&1)^1][0];\
512 const int dr2= dither_2x2_8[y&1 ][1];\
513 const int dg2= dither_2x2_4[y&1 ][1];\
514 const int db2= dither_2x2_8[(y&1)^1][1];\
516 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
517 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
521 case PIX_FMT_RGB555:\
522 case PIX_FMT_BGR555:\
524 const int dr1= dither_2x2_8[y&1 ][0];\
525 const int dg1= dither_2x2_8[y&1 ][1];\
526 const int db1= dither_2x2_8[(y&1)^1][0];\
527 const int dr2= dither_2x2_8[y&1 ][1];\
528 const int dg2= dither_2x2_8[y&1 ][0];\
529 const int db2= dither_2x2_8[(y&1)^1][1];\
531 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
532 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
539 const uint8_t * const d64= dither_8x8_73[y&7];\
540 const uint8_t * const d32= dither_8x8_32[y&7];\
542 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
543 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
550 const uint8_t * const d64= dither_8x8_73 [y&7];\
551 const uint8_t * const d128=dither_8x8_220[y&7];\
553 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
554 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
558 case PIX_FMT_RGB4_BYTE:\
559 case PIX_FMT_BGR4_BYTE:\
561 const uint8_t * const d64= dither_8x8_73 [y&7];\
562 const uint8_t * const d128=dither_8x8_220[y&7];\
564 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
565 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
569 case PIX_FMT_MONOBLACK:\
571 const uint8_t * const d128=dither_8x8_220[y&7];\
572 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
573 for(i=0; i<dstW-7; i+=8){\
575 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
576 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
577 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
578 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
579 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
580 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
581 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
582 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
583 ((uint8_t*)dest)[0]= acc;\
588 ((uint8_t*)dest)-= dstW>>4;\
592 static int top[1024];\
593 static int last_new[1024][1024];\
594 static int last_in3[1024][1024];\
595 static int drift[1024][1024];\
599 const uint8_t * const d128=dither_8x8_220[y&7];\
604 for(i=dstW>>1; i<dstW; i++){\
605 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
606 int in2 = (76309 * (in - 16) + 32768) >> 16;\
607 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
608 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
609 + (last_new[y][i] - in3)*f/256;\
610 int new= old> 128 ? 255 : 0;\
612 error_new+= FFABS(last_new[y][i] - new);\
613 error_in3+= FFABS(last_in3[y][i] - in3);\
614 f= error_new - error_in3*4;\
619 left= top[i]= old - new;\
620 last_new[y][i]= new;\
621 last_in3[y][i]= in3;\
623 acc+= acc + (new&1);\
625 ((uint8_t*)dest)[0]= acc;\
633 case PIX_FMT_YUYV422:\
635 ((uint8_t*)dest)[2*i2+0]= Y1;\
636 ((uint8_t*)dest)[2*i2+1]= U;\
637 ((uint8_t*)dest)[2*i2+2]= Y2;\
638 ((uint8_t*)dest)[2*i2+3]= V;\
641 case PIX_FMT_UYVY422:\
643 ((uint8_t*)dest)[2*i2+0]= U;\
644 ((uint8_t*)dest)[2*i2+1]= Y1;\
645 ((uint8_t*)dest)[2*i2+2]= V;\
646 ((uint8_t*)dest)[2*i2+3]= Y2;\
652 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
653 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
654 uint8_t *dest, int dstW, int y)
661 YSCALE_YUV_2_RGBX_C(uint32_t)
662 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
663 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
667 YSCALE_YUV_2_RGBX_C(uint8_t)
668 ((uint8_t*)dest)[0]= r[Y1];
669 ((uint8_t*)dest)[1]= g[Y1];
670 ((uint8_t*)dest)[2]= b[Y1];
671 ((uint8_t*)dest)[3]= r[Y2];
672 ((uint8_t*)dest)[4]= g[Y2];
673 ((uint8_t*)dest)[5]= b[Y2];
678 YSCALE_YUV_2_RGBX_C(uint8_t)
679 ((uint8_t*)dest)[0]= b[Y1];
680 ((uint8_t*)dest)[1]= g[Y1];
681 ((uint8_t*)dest)[2]= r[Y1];
682 ((uint8_t*)dest)[3]= b[Y2];
683 ((uint8_t*)dest)[4]= g[Y2];
684 ((uint8_t*)dest)[5]= r[Y2];
691 const int dr1= dither_2x2_8[y&1 ][0];
692 const int dg1= dither_2x2_4[y&1 ][0];
693 const int db1= dither_2x2_8[(y&1)^1][0];
694 const int dr2= dither_2x2_8[y&1 ][1];
695 const int dg2= dither_2x2_4[y&1 ][1];
696 const int db2= dither_2x2_8[(y&1)^1][1];
697 YSCALE_YUV_2_RGBX_C(uint16_t)
698 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
699 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
706 const int dr1= dither_2x2_8[y&1 ][0];
707 const int dg1= dither_2x2_8[y&1 ][1];
708 const int db1= dither_2x2_8[(y&1)^1][0];
709 const int dr2= dither_2x2_8[y&1 ][1];
710 const int dg2= dither_2x2_8[y&1 ][0];
711 const int db2= dither_2x2_8[(y&1)^1][1];
712 YSCALE_YUV_2_RGBX_C(uint16_t)
713 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
714 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
721 const uint8_t * const d64= dither_8x8_73[y&7];
722 const uint8_t * const d32= dither_8x8_32[y&7];
723 YSCALE_YUV_2_RGBX_C(uint8_t)
724 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
725 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
732 const uint8_t * const d64= dither_8x8_73 [y&7];
733 const uint8_t * const d128=dither_8x8_220[y&7];
734 YSCALE_YUV_2_RGBX_C(uint8_t)
735 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
736 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
740 case PIX_FMT_RGB4_BYTE:
741 case PIX_FMT_BGR4_BYTE:
743 const uint8_t * const d64= dither_8x8_73 [y&7];
744 const uint8_t * const d128=dither_8x8_220[y&7];
745 YSCALE_YUV_2_RGBX_C(uint8_t)
746 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
747 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
751 case PIX_FMT_MONOBLACK:
753 const uint8_t * const d128=dither_8x8_220[y&7];
754 uint8_t *g= c->table_gU[128] + c->table_gV[128];
756 for(i=0; i<dstW-1; i+=2){
761 for(j=0; j<lumFilterSize; j++)
763 Y1 += lumSrc[j][i] * lumFilter[j];
764 Y2 += lumSrc[j][i+1] * lumFilter[j];
775 acc+= acc + g[Y1+d128[(i+0)&7]];
776 acc+= acc + g[Y2+d128[(i+1)&7]];
778 ((uint8_t*)dest)[0]= acc;
784 case PIX_FMT_YUYV422:
785 YSCALE_YUV_2_PACKEDX_C(void)
786 ((uint8_t*)dest)[2*i2+0]= Y1;
787 ((uint8_t*)dest)[2*i2+1]= U;
788 ((uint8_t*)dest)[2*i2+2]= Y2;
789 ((uint8_t*)dest)[2*i2+3]= V;
792 case PIX_FMT_UYVY422:
793 YSCALE_YUV_2_PACKEDX_C(void)
794 ((uint8_t*)dest)[2*i2+0]= U;
795 ((uint8_t*)dest)[2*i2+1]= Y1;
796 ((uint8_t*)dest)[2*i2+2]= V;
797 ((uint8_t*)dest)[2*i2+3]= Y2;
804 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
806 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
811 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
812 #define COMPILE_ALTIVEC
813 #endif //HAVE_ALTIVEC
814 #endif //ARCH_POWERPC
816 #if defined(ARCH_X86)
818 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
822 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
826 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
827 #define COMPILE_3DNOW
829 #endif //ARCH_X86 || ARCH_X86_64
840 #define RENAME(a) a ## _C
841 #include "swscale_template.c"
845 #ifdef COMPILE_ALTIVEC
848 #define RENAME(a) a ## _altivec
849 #include "swscale_template.c"
851 #endif //ARCH_POWERPC
853 #if defined(ARCH_X86)
862 #define RENAME(a) a ## _X86
863 #include "swscale_template.c"
871 #define RENAME(a) a ## _MMX
872 #include "swscale_template.c"
881 #define RENAME(a) a ## _MMX2
882 #include "swscale_template.c"
891 #define RENAME(a) a ## _3DNow
892 #include "swscale_template.c"
895 #endif //ARCH_X86 || ARCH_X86_64
897 // minor note: the HAVE_xyz is messed up after that line so don't use it
899 static double getSplineCoeff(double a, double b, double c, double d, double dist)
901 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
902 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
903 else return getSplineCoeff( 0.0,
910 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
911 int srcW, int dstW, int filterAlign, int one, int flags,
912 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
919 double *filter2=NULL;
920 #if defined(ARCH_X86)
921 if(flags & SWS_CPU_CAPS_MMX)
922 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
925 // Note the +1 is for the MMXscaler which reads over the end
926 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
928 if(FFABS(xInc - 0x10000) <10) // unscaled
932 filter= av_malloc(dstW*sizeof(double)*filterSize);
933 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
935 for(i=0; i<dstW; i++)
937 filter[i*filterSize]=1;
942 else if(flags&SWS_POINT) // lame looking point sampling mode
947 filter= av_malloc(dstW*sizeof(double)*filterSize);
949 xDstInSrc= xInc/2 - 0x8000;
950 for(i=0; i<dstW; i++)
952 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
959 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
963 if (flags&SWS_BICUBIC) filterSize= 4;
964 else if(flags&SWS_X ) filterSize= 4;
965 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
966 filter= av_malloc(dstW*sizeof(double)*filterSize);
968 xDstInSrc= xInc/2 - 0x8000;
969 for(i=0; i<dstW; i++)
971 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
975 //Bilinear upscale / linear interpolate / Area averaging
976 for(j=0; j<filterSize; j++)
978 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
979 double coeff= 1.0 - d;
981 filter[i*filterSize + j]= coeff;
990 double sizeFactor, filterSizeInSrc;
991 const double xInc1= (double)xInc / (double)(1<<16);
993 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
994 else if(flags&SWS_X) sizeFactor= 8.0;
995 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
996 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
997 else if(flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
998 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
999 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
1000 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
1002 sizeFactor= 0.0; //GCC warning killer
1006 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1007 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1009 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1010 if(filterSize > srcW-2) filterSize=srcW-2;
1012 filter= av_malloc(dstW*sizeof(double)*filterSize);
1014 xDstInSrc= xInc1 / 2.0 - 0.5;
1015 for(i=0; i<dstW; i++)
1017 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1019 (*filterPos)[i]= xx;
1020 for(j=0; j<filterSize; j++)
1022 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1024 if(flags & SWS_BICUBIC)
1026 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1027 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1030 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1032 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1036 /* else if(flags & SWS_X)
1038 double p= param ? param*0.01 : 0.3;
1039 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1040 coeff*= pow(2.0, - p*d*d);
1042 else if(flags & SWS_X)
1044 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1050 if(coeff<0.0) coeff= -pow(-coeff, A);
1051 else coeff= pow( coeff, A);
1052 coeff= coeff*0.5 + 0.5;
1054 else if(flags & SWS_AREA)
1056 double srcPixelSize= 1.0/xInc1;
1057 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1058 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1061 else if(flags & SWS_GAUSS)
1063 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1064 coeff = pow(2.0, - p*d*d);
1066 else if(flags & SWS_SINC)
1068 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1070 else if(flags & SWS_LANCZOS)
1072 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1073 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1076 else if(flags & SWS_BILINEAR)
1079 if(coeff<0) coeff=0;
1081 else if(flags & SWS_SPLINE)
1083 double p=-2.196152422706632;
1084 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1087 coeff= 0.0; //GCC warning killer
1091 filter[i*filterSize + j]= coeff;
1098 /* apply src & dst Filter to filter -> filter2
1101 ASSERT(filterSize>0)
1102 filter2Size= filterSize;
1103 if(srcFilter) filter2Size+= srcFilter->length - 1;
1104 if(dstFilter) filter2Size+= dstFilter->length - 1;
1105 ASSERT(filter2Size>0)
1106 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1108 for(i=0; i<dstW; i++)
1111 SwsVector scaleFilter;
1114 scaleFilter.coeff= filter + i*filterSize;
1115 scaleFilter.length= filterSize;
1117 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1118 else outVec= &scaleFilter;
1120 ASSERT(outVec->length == filter2Size)
1123 for(j=0; j<outVec->length; j++)
1125 filter2[i*filter2Size + j]= outVec->coeff[j];
1128 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1130 if(outVec != &scaleFilter) sws_freeVec(outVec);
1132 av_free(filter); filter=NULL;
1134 /* try to reduce the filter-size (step1 find size and shift left) */
1135 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1137 for(i=dstW-1; i>=0; i--)
1139 int min= filter2Size;
1143 /* get rid off near zero elements on the left by shifting left */
1144 for(j=0; j<filter2Size; j++)
1147 cutOff += FFABS(filter2[i*filter2Size]);
1149 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1151 /* preserve Monotonicity because the core can't handle the filter otherwise */
1152 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1154 // Move filter coeffs left
1155 for(k=1; k<filter2Size; k++)
1156 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1157 filter2[i*filter2Size + k - 1]= 0.0;
1162 /* count near zeros on the right */
1163 for(j=filter2Size-1; j>0; j--)
1165 cutOff += FFABS(filter2[i*filter2Size + j]);
1167 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1171 if(min>minFilterSize) minFilterSize= min;
1174 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1175 // we can handle the special case 4,
1176 // so we don't want to go to the full 8
1177 if (minFilterSize < 5)
1180 // we really don't want to waste our time
1181 // doing useless computation, so fall-back on
1182 // the scalar C code for very small filter.
1183 // vectorizing is worth it only if you have
1184 // decent-sized vector.
1185 if (minFilterSize < 3)
1189 if (flags & SWS_CPU_CAPS_MMX) {
1190 // special case for unscaled vertical filtering
1191 if(minFilterSize == 1 && filterAlign == 2)
1195 ASSERT(minFilterSize > 0)
1196 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1197 ASSERT(filterSize > 0)
1198 filter= av_malloc(filterSize*dstW*sizeof(double));
1199 if(filterSize >= MAX_FILTER_SIZE)
1201 *outFilterSize= filterSize;
1203 if(flags&SWS_PRINT_INFO)
1204 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1205 /* try to reduce the filter-size (step2 reduce it) */
1206 for(i=0; i<dstW; i++)
1210 for(j=0; j<filterSize; j++)
1212 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1213 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1216 av_free(filter2); filter2=NULL;
1219 //FIXME try to align filterpos if possible
1222 for(i=0; i<dstW; i++)
1225 if((*filterPos)[i] < 0)
1227 // Move filter coeffs left to compensate for filterPos
1228 for(j=1; j<filterSize; j++)
1230 int left= FFMAX(j + (*filterPos)[i], 0);
1231 filter[i*filterSize + left] += filter[i*filterSize + j];
1232 filter[i*filterSize + j]=0;
1237 if((*filterPos)[i] + filterSize > srcW)
1239 int shift= (*filterPos)[i] + filterSize - srcW;
1240 // Move filter coeffs right to compensate for filterPos
1241 for(j=filterSize-2; j>=0; j--)
1243 int right= FFMIN(j + shift, filterSize-1);
1244 filter[i*filterSize +right] += filter[i*filterSize +j];
1245 filter[i*filterSize +j]=0;
1247 (*filterPos)[i]= srcW - filterSize;
1251 // Note the +1 is for the MMXscaler which reads over the end
1252 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1253 *outFilter= av_malloc(*outFilterSize*(dstW+1)*sizeof(int16_t));
1254 memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1256 /* Normalize & Store in outFilter */
1257 for(i=0; i<dstW; i++)
1264 for(j=0; j<filterSize; j++)
1266 sum+= filter[i*filterSize + j];
1269 for(j=0; j<*outFilterSize; j++)
1271 double v= filter[i*filterSize + j]*scale + error;
1272 int intV= floor(v + 0.5);
1273 (*outFilter)[i*(*outFilterSize) + j]= intV;
1278 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1279 for(i=0; i<*outFilterSize; i++)
1281 int j= dstW*(*outFilterSize);
1282 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1290 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1293 long imm8OfPShufW1A;
1294 long imm8OfPShufW2A;
1295 long fragmentLengthA;
1297 long imm8OfPShufW1B;
1298 long imm8OfPShufW2B;
1299 long fragmentLengthB;
1304 // create an optimized horizontal scaling routine
1312 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1313 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1314 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1315 "punpcklbw %%mm7, %%mm1 \n\t"
1316 "punpcklbw %%mm7, %%mm0 \n\t"
1317 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1319 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1321 "psubw %%mm1, %%mm0 \n\t"
1322 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1323 "pmullw %%mm3, %%mm0 \n\t"
1324 "psllw $7, %%mm1 \n\t"
1325 "paddw %%mm1, %%mm0 \n\t"
1327 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1329 "add $8, %%"REG_a" \n\t"
1344 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1345 "=r" (fragmentLengthA)
1352 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1353 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1354 "punpcklbw %%mm7, %%mm0 \n\t"
1355 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1357 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1359 "psubw %%mm1, %%mm0 \n\t"
1360 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1361 "pmullw %%mm3, %%mm0 \n\t"
1362 "psllw $7, %%mm1 \n\t"
1363 "paddw %%mm1, %%mm0 \n\t"
1365 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1367 "add $8, %%"REG_a" \n\t"
1382 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1383 "=r" (fragmentLengthB)
1386 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1389 for(i=0; i<dstW/numSplits; i++)
1396 int b=((xpos+xInc)>>16) - xx;
1397 int c=((xpos+xInc*2)>>16) - xx;
1398 int d=((xpos+xInc*3)>>16) - xx;
1400 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1401 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1402 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1403 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1408 int maxShift= 3-(d+1);
1411 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1413 funnyCode[fragmentPos + imm8OfPShufW1B]=
1414 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1415 funnyCode[fragmentPos + imm8OfPShufW2B]=
1416 a | (b<<2) | (c<<4) | (d<<6);
1418 if(i+3>=dstW) shift=maxShift; //avoid overread
1419 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1421 if(shift && i>=shift)
1423 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1424 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1425 filterPos[i/2]-=shift;
1428 fragmentPos+= fragmentLengthB;
1435 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1437 funnyCode[fragmentPos + imm8OfPShufW1A]=
1438 funnyCode[fragmentPos + imm8OfPShufW2A]=
1439 a | (b<<2) | (c<<4) | (d<<6);
1441 if(i+4>=dstW) shift=maxShift; //avoid overread
1442 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1444 if(shift && i>=shift)
1446 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1447 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1448 filterPos[i/2]-=shift;
1451 fragmentPos+= fragmentLengthA;
1454 funnyCode[fragmentPos]= RET;
1458 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1460 #endif /* COMPILE_MMX2 */
1462 static void globalInit(void){
1463 // generating tables:
1465 for(i=0; i<768; i++){
1466 int c= FFMIN(FFMAX(i-256, 0), 255);
1471 static SwsFunc getSwsFunc(int flags){
1473 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1474 #if defined(ARCH_X86)
1475 // ordered per speed fasterst first
1476 if(flags & SWS_CPU_CAPS_MMX2)
1477 return swScale_MMX2;
1478 else if(flags & SWS_CPU_CAPS_3DNOW)
1479 return swScale_3DNow;
1480 else if(flags & SWS_CPU_CAPS_MMX)
1487 if(flags & SWS_CPU_CAPS_ALTIVEC)
1488 return swScale_altivec;
1493 #endif /* defined(ARCH_X86) */
1494 #else //RUNTIME_CPUDETECT
1496 return swScale_MMX2;
1497 #elif defined (HAVE_3DNOW)
1498 return swScale_3DNow;
1499 #elif defined (HAVE_MMX)
1501 #elif defined (HAVE_ALTIVEC)
1502 return swScale_altivec;
1506 #endif //!RUNTIME_CPUDETECT
1509 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1510 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1511 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1513 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1514 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1518 uint8_t *srcPtr= src[0];
1519 uint8_t *dstPtr= dst;
1520 for(i=0; i<srcSliceH; i++)
1522 memcpy(dstPtr, srcPtr, c->srcW);
1523 srcPtr+= srcStride[0];
1524 dstPtr+= dstStride[0];
1527 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1528 if (c->dstFormat == PIX_FMT_NV12)
1529 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1531 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1536 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1537 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1538 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1540 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1545 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1546 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1547 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1549 yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1554 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1555 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1556 int srcSliceH, uint8_t* dst[], int dstStride[]){
1557 const int srcFormat= c->srcFormat;
1558 const int dstFormat= c->dstFormat;
1559 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1560 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1561 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1562 const int dstId= fmt_depth(dstFormat) >> 2;
1563 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1566 if( (isBGR(srcFormat) && isBGR(dstFormat))
1567 || (isRGB(srcFormat) && isRGB(dstFormat))){
1568 switch(srcId | (dstId<<4)){
1569 case 0x34: conv= rgb16to15; break;
1570 case 0x36: conv= rgb24to15; break;
1571 case 0x38: conv= rgb32to15; break;
1572 case 0x43: conv= rgb15to16; break;
1573 case 0x46: conv= rgb24to16; break;
1574 case 0x48: conv= rgb32to16; break;
1575 case 0x63: conv= rgb15to24; break;
1576 case 0x64: conv= rgb16to24; break;
1577 case 0x68: conv= rgb32to24; break;
1578 case 0x83: conv= rgb15to32; break;
1579 case 0x84: conv= rgb16to32; break;
1580 case 0x86: conv= rgb24to32; break;
1581 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1582 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1584 }else if( (isBGR(srcFormat) && isRGB(dstFormat))
1585 || (isRGB(srcFormat) && isBGR(dstFormat))){
1586 switch(srcId | (dstId<<4)){
1587 case 0x33: conv= rgb15tobgr15; break;
1588 case 0x34: conv= rgb16tobgr15; break;
1589 case 0x36: conv= rgb24tobgr15; break;
1590 case 0x38: conv= rgb32tobgr15; break;
1591 case 0x43: conv= rgb15tobgr16; break;
1592 case 0x44: conv= rgb16tobgr16; break;
1593 case 0x46: conv= rgb24tobgr16; break;
1594 case 0x48: conv= rgb32tobgr16; break;
1595 case 0x63: conv= rgb15tobgr24; break;
1596 case 0x64: conv= rgb16tobgr24; break;
1597 case 0x66: conv= rgb24tobgr24; break;
1598 case 0x68: conv= rgb32tobgr24; break;
1599 case 0x83: conv= rgb15tobgr32; break;
1600 case 0x84: conv= rgb16tobgr32; break;
1601 case 0x86: conv= rgb24tobgr32; break;
1602 case 0x88: conv= rgb32tobgr32; break;
1603 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1604 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1607 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1608 sws_format_name(srcFormat), sws_format_name(dstFormat));
1611 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1612 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1616 uint8_t *srcPtr= src[0];
1617 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1619 for(i=0; i<srcSliceH; i++)
1621 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1622 srcPtr+= srcStride[0];
1623 dstPtr+= dstStride[0];
1629 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1630 int srcSliceH, uint8_t* dst[], int dstStride[]){
1634 dst[0]+ srcSliceY *dstStride[0],
1635 dst[1]+(srcSliceY>>1)*dstStride[1],
1636 dst[2]+(srcSliceY>>1)*dstStride[2],
1638 dstStride[0], dstStride[1], srcStride[0]);
1642 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1643 int srcSliceH, uint8_t* dst[], int dstStride[]){
1647 if(srcStride[0]==dstStride[0] && srcStride[0] > 0)
1648 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1650 uint8_t *srcPtr= src[0];
1651 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1653 for(i=0; i<srcSliceH; i++)
1655 memcpy(dstPtr, srcPtr, c->srcW);
1656 srcPtr+= srcStride[0];
1657 dstPtr+= dstStride[0];
1661 if(c->dstFormat==PIX_FMT_YUV420P){
1662 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1663 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1665 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1666 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1671 /* unscaled copy like stuff (assumes nearly identical formats) */
1672 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1673 int srcSliceH, uint8_t* dst[], int dstStride[]){
1675 if(isPacked(c->srcFormat))
1677 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1678 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1682 uint8_t *srcPtr= src[0];
1683 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1686 /* universal length finder */
1687 while(length+c->srcW <= FFABS(dstStride[0])
1688 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1691 for(i=0; i<srcSliceH; i++)
1693 memcpy(dstPtr, srcPtr, length);
1694 srcPtr+= srcStride[0];
1695 dstPtr+= dstStride[0];
1700 { /* Planar YUV or gray */
1702 for(plane=0; plane<3; plane++)
1704 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1705 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1706 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1708 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1710 if(!isGray(c->dstFormat))
1711 memset(dst[plane], 128, dstStride[plane]*height);
1715 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1716 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1720 uint8_t *srcPtr= src[plane];
1721 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1722 for(i=0; i<height; i++)
1724 memcpy(dstPtr, srcPtr, length);
1725 srcPtr+= srcStride[plane];
1726 dstPtr+= dstStride[plane];
1735 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1736 int srcSliceH, uint8_t* dst[], int dstStride[]){
1738 int length= c->srcW;
1740 int height= srcSliceH;
1742 uint8_t *srcPtr= src[0];
1743 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1745 if(!isGray(c->dstFormat)){
1746 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1747 memset(dst[1], 128, dstStride[1]*height);
1748 memset(dst[2], 128, dstStride[2]*height);
1750 if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1751 for(i=0; i<height; i++)
1753 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1754 srcPtr+= srcStride[0];
1755 dstPtr+= dstStride[0];
1760 static int graytogray16(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;
1769 for(i=0; i<height; i++)
1771 for(j=0; j<length; j++)
1773 dstPtr[j<<1] = srcPtr[j];
1774 dstPtr[(j<<1)+1] = srcPtr[j];
1776 srcPtr+= srcStride[0];
1777 dstPtr+= dstStride[0];
1782 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1783 int srcSliceH, uint8_t* dst[], int dstStride[]){
1785 int length= c->srcW;
1787 int height= srcSliceH;
1789 uint16_t *srcPtr= src[0];
1790 uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1791 for(i=0; i<height; i++)
1793 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1794 srcPtr+= srcStride[0]/2;
1795 dstPtr+= dstStride[0]/2;
1801 static void getSubSampleFactors(int *h, int *v, int format){
1803 case PIX_FMT_UYVY422:
1804 case PIX_FMT_YUYV422:
1808 case PIX_FMT_YUV420P:
1809 case PIX_FMT_GRAY16BE:
1810 case PIX_FMT_GRAY16LE:
1811 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1817 case PIX_FMT_YUV410P:
1821 case PIX_FMT_YUV444P:
1825 case PIX_FMT_YUV422P:
1829 case PIX_FMT_YUV411P:
1840 static uint16_t roundToInt16(int64_t f){
1841 int r= (f + (1<<15))>>16;
1842 if(r<-0x7FFF) return 0x8000;
1843 else if(r> 0x7FFF) return 0x7FFF;
1848 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1849 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1850 * @return -1 if not supported
1852 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1853 int64_t crv = inv_table[0];
1854 int64_t cbu = inv_table[1];
1855 int64_t cgu = -inv_table[2];
1856 int64_t cgv = -inv_table[3];
1860 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1861 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1862 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1864 c->brightness= brightness;
1865 c->contrast = contrast;
1866 c->saturation= saturation;
1867 c->srcRange = srcRange;
1868 c->dstRange = dstRange;
1870 c->uOffset= 0x0400040004000400LL;
1871 c->vOffset= 0x0400040004000400LL;
1878 cy = (cy *contrast )>>16;
1879 crv= (crv*contrast * saturation)>>32;
1880 cbu= (cbu*contrast * saturation)>>32;
1881 cgu= (cgu*contrast * saturation)>>32;
1882 cgv= (cgv*contrast * saturation)>>32;
1884 oy -= 256*brightness;
1886 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1887 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1888 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1889 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1890 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1891 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1893 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1896 #ifdef COMPILE_ALTIVEC
1897 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1898 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1904 * @return -1 if not supported
1906 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1907 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1909 *inv_table = c->srcColorspaceTable;
1910 *table = c->dstColorspaceTable;
1911 *srcRange = c->srcRange;
1912 *dstRange = c->dstRange;
1913 *brightness= c->brightness;
1914 *contrast = c->contrast;
1915 *saturation= c->saturation;
1920 static int handle_jpeg(int *format)
1923 case PIX_FMT_YUVJ420P:
1924 *format = PIX_FMT_YUV420P;
1926 case PIX_FMT_YUVJ422P:
1927 *format = PIX_FMT_YUV422P;
1929 case PIX_FMT_YUVJ444P:
1930 *format = PIX_FMT_YUV444P;
1937 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1938 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1942 int usesVFilter, usesHFilter;
1943 int unscaled, needsDither;
1944 int srcRange, dstRange;
1945 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1946 #if defined(ARCH_X86)
1947 if(flags & SWS_CPU_CAPS_MMX)
1948 asm volatile("emms\n\t"::: "memory");
1951 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1952 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1954 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1955 #elif defined (HAVE_3DNOW)
1956 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1957 #elif defined (HAVE_MMX)
1958 flags |= SWS_CPU_CAPS_MMX;
1959 #elif defined (HAVE_ALTIVEC)
1960 flags |= SWS_CPU_CAPS_ALTIVEC;
1962 #endif /* RUNTIME_CPUDETECT */
1963 if(clip_table[512] != 255) globalInit();
1964 if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1966 unscaled = (srcW == dstW && srcH == dstH);
1967 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1968 && (fmt_depth(dstFormat))<24
1969 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1971 srcRange = handle_jpeg(&srcFormat);
1972 dstRange = handle_jpeg(&dstFormat);
1974 if(!isSupportedIn(srcFormat))
1976 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1979 if(!isSupportedOut(dstFormat))
1981 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1986 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
1988 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1989 srcW, srcH, dstW, dstH);
1993 if(!dstFilter) dstFilter= &dummyFilter;
1994 if(!srcFilter) srcFilter= &dummyFilter;
1996 c= av_malloc(sizeof(SwsContext));
1997 memset(c, 0, sizeof(SwsContext));
2003 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2004 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2006 c->dstFormat= dstFormat;
2007 c->srcFormat= srcFormat;
2008 c->vRounder= 4* 0x0001000100010001ULL;
2010 usesHFilter= usesVFilter= 0;
2011 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2012 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2013 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2014 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2015 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2016 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2017 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2018 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2020 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2021 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2023 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2024 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2026 // drop some chroma lines if the user wants it
2027 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2028 c->chrSrcVSubSample+= c->vChrDrop;
2030 // drop every 2. pixel for chroma calculation unless user wants full chroma
2031 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
2032 c->chrSrcHSubSample=1;
2035 c->param[0] = param[0];
2036 c->param[1] = param[1];
2039 c->param[1] = SWS_PARAM_DEFAULT;
2042 c->chrIntHSubSample= c->chrDstHSubSample;
2043 c->chrIntVSubSample= c->chrSrcVSubSample;
2045 // note the -((-x)>>y) is so that we allways round toward +inf
2046 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2047 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2048 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2049 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2051 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);
2053 /* unscaled special Cases */
2054 if(unscaled && !usesHFilter && !usesVFilter)
2057 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2059 c->swScale= PlanarToNV12Wrapper;
2063 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2065 c->swScale= yuv2rgb_get_func_ptr(c);
2069 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2071 c->swScale= yvu9toyv12Wrapper;
2075 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2076 c->swScale= bgr24toyv12Wrapper;
2078 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2079 if( (isBGR(srcFormat) || isRGB(srcFormat))
2080 && (isBGR(dstFormat) || isRGB(dstFormat))
2082 c->swScale= rgb2rgbWrapper;
2084 /* LQ converters if -sws 0 or -sws 4*/
2085 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2086 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2087 if( (isBGR(srcFormat) || isRGB(srcFormat))
2088 && (isBGR(dstFormat) || isRGB(dstFormat))
2090 c->swScale= rgb2rgbWrapper;
2093 if(srcFormat == PIX_FMT_YUV420P &&
2094 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2096 if (dstFormat == PIX_FMT_YUYV422)
2097 c->swScale= PlanarToYuy2Wrapper;
2099 c->swScale= PlanarToUyvyWrapper;
2103 #ifdef COMPILE_ALTIVEC
2104 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2105 ((srcFormat == PIX_FMT_YUV420P &&
2106 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2107 // unscaled YV12 -> packed YUV, we want speed
2108 if (dstFormat == PIX_FMT_YUYV422)
2109 c->swScale= yv12toyuy2_unscaled_altivec;
2111 c->swScale= yv12touyvy_unscaled_altivec;
2116 if( srcFormat == dstFormat
2117 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2118 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2121 c->swScale= simpleCopy;
2124 /* gray16{le,be} conversions */
2125 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2127 c->swScale= gray16togray;
2129 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2131 c->swScale= graytogray16;
2133 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2135 c->swScale= gray16swap;
2139 if(flags&SWS_PRINT_INFO)
2140 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2141 sws_format_name(srcFormat), sws_format_name(dstFormat));
2146 if(flags & SWS_CPU_CAPS_MMX2)
2148 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2149 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2151 if(flags&SWS_PRINT_INFO)
2152 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2154 if(usesHFilter) c->canMMX2BeUsed=0;
2159 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2160 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2162 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2163 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2164 // n-2 is the last chrominance sample available
2165 // this is not perfect, but noone shuld notice the difference, the more correct variant
2166 // would be like the vertical one, but that would require some special code for the
2167 // first and last pixel
2168 if(flags&SWS_FAST_BILINEAR)
2170 if(c->canMMX2BeUsed)
2175 //we don't use the x86asm scaler if mmx is available
2176 else if(flags & SWS_CPU_CAPS_MMX)
2178 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2179 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2183 /* precalculate horizontal scaler filter coefficients */
2185 const int filterAlign=
2186 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2187 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2190 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2191 srcW , dstW, filterAlign, 1<<14,
2192 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2193 srcFilter->lumH, dstFilter->lumH, c->param);
2194 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2195 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2196 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2197 srcFilter->chrH, dstFilter->chrH, c->param);
2199 #define MAX_FUNNY_CODE_SIZE 10000
2200 #if defined(COMPILE_MMX2)
2201 // can't downscale !!!
2202 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2204 #ifdef MAP_ANONYMOUS
2205 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2206 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2208 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2209 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2212 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2213 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2214 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2215 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2217 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2218 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2220 #endif /* defined(COMPILE_MMX2) */
2221 } // Init Horizontal stuff
2225 /* precalculate vertical scaler filter coefficients */
2227 const int filterAlign=
2228 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2229 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2232 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2233 srcH , dstH, filterAlign, (1<<12)-4,
2234 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2235 srcFilter->lumV, dstFilter->lumV, c->param);
2236 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2237 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2238 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2239 srcFilter->chrV, dstFilter->chrV, c->param);
2242 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2243 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2245 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2247 short *p = (short *)&c->vYCoeffsBank[i];
2249 p[j] = c->vLumFilter[i];
2252 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2254 short *p = (short *)&c->vCCoeffsBank[i];
2256 p[j] = c->vChrFilter[i];
2261 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2262 c->vLumBufSize= c->vLumFilterSize;
2263 c->vChrBufSize= c->vChrFilterSize;
2264 for(i=0; i<dstH; i++)
2266 int chrI= i*c->chrDstH / dstH;
2267 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2268 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2270 nextSlice>>= c->chrSrcVSubSample;
2271 nextSlice<<= c->chrSrcVSubSample;
2272 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2273 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
2274 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2275 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2278 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2279 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2280 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2281 //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)
2282 /* align at 16 bytes for AltiVec */
2283 for(i=0; i<c->vLumBufSize; i++)
2284 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_malloc(4000);
2285 for(i=0; i<c->vChrBufSize; i++)
2286 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2288 //try to avoid drawing green stuff between the right end and the stride end
2289 for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2290 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2292 ASSERT(c->chrDstH <= dstH)
2294 if(flags&SWS_PRINT_INFO)
2297 char *dither= " dithered";
2301 if(flags&SWS_FAST_BILINEAR)
2302 MSG_INFO("SwScaler: FAST_BILINEAR scaler, ");
2303 else if(flags&SWS_BILINEAR)
2304 MSG_INFO("SwScaler: BILINEAR scaler, ");
2305 else if(flags&SWS_BICUBIC)
2306 MSG_INFO("SwScaler: BICUBIC scaler, ");
2307 else if(flags&SWS_X)
2308 MSG_INFO("SwScaler: Experimental scaler, ");
2309 else if(flags&SWS_POINT)
2310 MSG_INFO("SwScaler: Nearest Neighbor / POINT scaler, ");
2311 else if(flags&SWS_AREA)
2312 MSG_INFO("SwScaler: Area Averageing scaler, ");
2313 else if(flags&SWS_BICUBLIN)
2314 MSG_INFO("SwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2315 else if(flags&SWS_GAUSS)
2316 MSG_INFO("SwScaler: Gaussian scaler, ");
2317 else if(flags&SWS_SINC)
2318 MSG_INFO("SwScaler: Sinc scaler, ");
2319 else if(flags&SWS_LANCZOS)
2320 MSG_INFO("SwScaler: Lanczos scaler, ");
2321 else if(flags&SWS_SPLINE)
2322 MSG_INFO("SwScaler: Bicubic spline scaler, ");
2324 MSG_INFO("SwScaler: ehh flags invalid?! ");
2326 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2327 MSG_INFO("from %s to%s %s ",
2328 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2330 MSG_INFO("from %s to %s ",
2331 sws_format_name(srcFormat), sws_format_name(dstFormat));
2333 if(flags & SWS_CPU_CAPS_MMX2)
2334 MSG_INFO("using MMX2\n");
2335 else if(flags & SWS_CPU_CAPS_3DNOW)
2336 MSG_INFO("using 3DNOW\n");
2337 else if(flags & SWS_CPU_CAPS_MMX)
2338 MSG_INFO("using MMX\n");
2339 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2340 MSG_INFO("using AltiVec\n");
2342 MSG_INFO("using C\n");
2345 if(flags & SWS_PRINT_INFO)
2347 if(flags & SWS_CPU_CAPS_MMX)
2349 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2350 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2353 if(c->hLumFilterSize==4)
2354 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2355 else if(c->hLumFilterSize==8)
2356 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2358 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2360 if(c->hChrFilterSize==4)
2361 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2362 else if(c->hChrFilterSize==8)
2363 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2365 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2370 #if defined(ARCH_X86)
2371 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2373 if(flags & SWS_FAST_BILINEAR)
2374 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2376 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2379 if(isPlanarYUV(dstFormat))
2381 if(c->vLumFilterSize==1)
2382 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2384 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2388 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2389 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2390 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2391 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2392 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2394 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2397 if(dstFormat==PIX_FMT_BGR24)
2398 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2399 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2400 else if(dstFormat==PIX_FMT_RGB32)
2401 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2402 else if(dstFormat==PIX_FMT_BGR565)
2403 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2404 else if(dstFormat==PIX_FMT_BGR555)
2405 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2407 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2409 if(flags & SWS_PRINT_INFO)
2411 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2412 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2413 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2414 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2417 c->swScale= getSwsFunc(flags);
2422 * swscale warper, so we don't need to export the SwsContext.
2423 * assumes planar YUV to be in YUV order instead of YVU
2425 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2426 int srcSliceH, uint8_t* dst[], int dstStride[]){
2427 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2428 MSG_ERR("swScaler: slices start in the middle!\n");
2431 if (c->sliceDir == 0) {
2432 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2435 // copy strides, so they can safely be modified
2436 if (c->sliceDir == 1) {
2437 // slices go from top to bottom
2438 int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2439 int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2440 return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2442 // slices go from bottom to top => we flip the image internally
2443 uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2444 src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2445 src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2447 uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2448 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2449 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2450 int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2451 int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2453 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2458 * swscale warper, so we don't need to export the SwsContext
2460 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStride[], int srcSliceY,
2461 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
2464 src[0] = srcParam[0]; src[1] = srcParam[1]; src[2] = srcParam[2];
2465 dst[0] = dstParam[0]; dst[1] = dstParam[1]; dst[2] = dstParam[2];
2466 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2468 return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2471 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2472 float lumaSharpen, float chromaSharpen,
2473 float chromaHShift, float chromaVShift,
2476 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2479 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2480 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2482 filter->lumH= sws_getIdentityVec();
2483 filter->lumV= sws_getIdentityVec();
2486 if(chromaGBlur!=0.0){
2487 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2488 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2490 filter->chrH= sws_getIdentityVec();
2491 filter->chrV= sws_getIdentityVec();
2494 if(chromaSharpen!=0.0){
2495 SwsVector *id= sws_getIdentityVec();
2496 sws_scaleVec(filter->chrH, -chromaSharpen);
2497 sws_scaleVec(filter->chrV, -chromaSharpen);
2498 sws_addVec(filter->chrH, id);
2499 sws_addVec(filter->chrV, id);
2503 if(lumaSharpen!=0.0){
2504 SwsVector *id= sws_getIdentityVec();
2505 sws_scaleVec(filter->lumH, -lumaSharpen);
2506 sws_scaleVec(filter->lumV, -lumaSharpen);
2507 sws_addVec(filter->lumH, id);
2508 sws_addVec(filter->lumV, id);
2512 if(chromaHShift != 0.0)
2513 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2515 if(chromaVShift != 0.0)
2516 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2518 sws_normalizeVec(filter->chrH, 1.0);
2519 sws_normalizeVec(filter->chrV, 1.0);
2520 sws_normalizeVec(filter->lumH, 1.0);
2521 sws_normalizeVec(filter->lumV, 1.0);
2523 if(verbose) sws_printVec(filter->chrH);
2524 if(verbose) sws_printVec(filter->lumH);
2530 * returns a normalized gaussian curve used to filter stuff
2531 * quality=3 is high quality, lowwer is lowwer quality
2533 SwsVector *sws_getGaussianVec(double variance, double quality){
2534 const int length= (int)(variance*quality + 0.5) | 1;
2536 double *coeff= av_malloc(length*sizeof(double));
2537 double middle= (length-1)*0.5;
2538 SwsVector *vec= av_malloc(sizeof(SwsVector));
2541 vec->length= length;
2543 for(i=0; i<length; i++)
2545 double dist= i-middle;
2546 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2549 sws_normalizeVec(vec, 1.0);
2554 SwsVector *sws_getConstVec(double c, int length){
2556 double *coeff= av_malloc(length*sizeof(double));
2557 SwsVector *vec= av_malloc(sizeof(SwsVector));
2560 vec->length= length;
2562 for(i=0; i<length; i++)
2569 SwsVector *sws_getIdentityVec(void){
2570 return sws_getConstVec(1.0, 1);
2573 double sws_dcVec(SwsVector *a){
2577 for(i=0; i<a->length; i++)
2583 void sws_scaleVec(SwsVector *a, double scalar){
2586 for(i=0; i<a->length; i++)
2587 a->coeff[i]*= scalar;
2590 void sws_normalizeVec(SwsVector *a, double height){
2591 sws_scaleVec(a, height/sws_dcVec(a));
2594 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2595 int length= a->length + b->length - 1;
2596 double *coeff= av_malloc(length*sizeof(double));
2598 SwsVector *vec= av_malloc(sizeof(SwsVector));
2601 vec->length= length;
2603 for(i=0; i<length; i++) coeff[i]= 0.0;
2605 for(i=0; i<a->length; i++)
2607 for(j=0; j<b->length; j++)
2609 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2616 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2617 int length= FFMAX(a->length, b->length);
2618 double *coeff= av_malloc(length*sizeof(double));
2620 SwsVector *vec= av_malloc(sizeof(SwsVector));
2623 vec->length= length;
2625 for(i=0; i<length; i++) coeff[i]= 0.0;
2627 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2628 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2633 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2634 int length= FFMAX(a->length, b->length);
2635 double *coeff= av_malloc(length*sizeof(double));
2637 SwsVector *vec= av_malloc(sizeof(SwsVector));
2640 vec->length= length;
2642 for(i=0; i<length; i++) coeff[i]= 0.0;
2644 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2645 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2650 /* shift left / or right if "shift" is negative */
2651 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2652 int length= a->length + FFABS(shift)*2;
2653 double *coeff= av_malloc(length*sizeof(double));
2655 SwsVector *vec= av_malloc(sizeof(SwsVector));
2658 vec->length= length;
2660 for(i=0; i<length; i++) coeff[i]= 0.0;
2662 for(i=0; i<a->length; i++)
2664 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2670 void sws_shiftVec(SwsVector *a, int shift){
2671 SwsVector *shifted= sws_getShiftedVec(a, shift);
2673 a->coeff= shifted->coeff;
2674 a->length= shifted->length;
2678 void sws_addVec(SwsVector *a, SwsVector *b){
2679 SwsVector *sum= sws_sumVec(a, b);
2681 a->coeff= sum->coeff;
2682 a->length= sum->length;
2686 void sws_subVec(SwsVector *a, SwsVector *b){
2687 SwsVector *diff= sws_diffVec(a, b);
2689 a->coeff= diff->coeff;
2690 a->length= diff->length;
2694 void sws_convVec(SwsVector *a, SwsVector *b){
2695 SwsVector *conv= sws_getConvVec(a, b);
2697 a->coeff= conv->coeff;
2698 a->length= conv->length;
2702 SwsVector *sws_cloneVec(SwsVector *a){
2703 double *coeff= av_malloc(a->length*sizeof(double));
2705 SwsVector *vec= av_malloc(sizeof(SwsVector));
2708 vec->length= a->length;
2710 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2715 void sws_printVec(SwsVector *a){
2721 for(i=0; i<a->length; i++)
2722 if(a->coeff[i]>max) max= a->coeff[i];
2724 for(i=0; i<a->length; i++)
2725 if(a->coeff[i]<min) min= a->coeff[i];
2729 for(i=0; i<a->length; i++)
2731 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2732 MSG_DBG2("%1.3f ", a->coeff[i]);
2733 for(;x>0; x--) MSG_DBG2(" ");
2738 void sws_freeVec(SwsVector *a){
2746 void sws_freeFilter(SwsFilter *filter){
2749 if(filter->lumH) sws_freeVec(filter->lumH);
2750 if(filter->lumV) sws_freeVec(filter->lumV);
2751 if(filter->chrH) sws_freeVec(filter->chrH);
2752 if(filter->chrV) sws_freeVec(filter->chrV);
2757 void sws_freeContext(SwsContext *c){
2763 for(i=0; i<c->vLumBufSize; i++)
2765 av_free(c->lumPixBuf[i]);
2766 c->lumPixBuf[i]=NULL;
2768 av_free(c->lumPixBuf);
2774 for(i=0; i<c->vChrBufSize; i++)
2776 av_free(c->chrPixBuf[i]);
2777 c->chrPixBuf[i]=NULL;
2779 av_free(c->chrPixBuf);
2783 av_free(c->vLumFilter);
2784 c->vLumFilter = NULL;
2785 av_free(c->vChrFilter);
2786 c->vChrFilter = NULL;
2787 av_free(c->hLumFilter);
2788 c->hLumFilter = NULL;
2789 av_free(c->hChrFilter);
2790 c->hChrFilter = NULL;
2792 av_free(c->vYCoeffsBank);
2793 c->vYCoeffsBank = NULL;
2794 av_free(c->vCCoeffsBank);
2795 c->vCCoeffsBank = NULL;
2798 av_free(c->vLumFilterPos);
2799 c->vLumFilterPos = NULL;
2800 av_free(c->vChrFilterPos);
2801 c->vChrFilterPos = NULL;
2802 av_free(c->hLumFilterPos);
2803 c->hLumFilterPos = NULL;
2804 av_free(c->hChrFilterPos);
2805 c->hChrFilterPos = NULL;
2807 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2808 #ifdef MAP_ANONYMOUS
2809 if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2810 if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2812 av_free(c->funnyYCode);
2813 av_free(c->funnyUVCode);
2816 c->funnyUVCode=NULL;
2817 #endif /* defined(ARCH_X86) */
2819 av_free(c->lumMmx2Filter);
2820 c->lumMmx2Filter=NULL;
2821 av_free(c->chrMmx2Filter);
2822 c->chrMmx2Filter=NULL;
2823 av_free(c->lumMmx2FilterPos);
2824 c->lumMmx2FilterPos=NULL;
2825 av_free(c->chrMmx2FilterPos);
2826 c->chrMmx2FilterPos=NULL;
2827 av_free(c->yuvTable);
2834 * Checks if context is valid or reallocs a new one instead.
2835 * If context is NULL, just calls sws_getContext() to get a new one.
2836 * Otherwise, checks if the parameters are the same already saved in context.
2837 * If that is the case, returns the current context.
2838 * Otherwise, frees context and gets a new one.
2840 * Be warned that srcFilter, dstFilter are not checked, they are
2841 * asumed to remain valid.
2843 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2844 int srcW, int srcH, int srcFormat,
2845 int dstW, int dstH, int dstFormat, int flags,
2846 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2848 if (context != NULL) {
2849 if ((context->srcW != srcW) || (context->srcH != srcH) ||
2850 (context->srcFormat != srcFormat) ||
2851 (context->dstW != dstW) || (context->dstH != dstH) ||
2852 (context->dstFormat != dstFormat) || (context->flags != flags) ||
2853 (context->param != param))
2855 sws_freeContext(context);
2859 if (context == NULL) {
2860 return sws_getContext(srcW, srcH, srcFormat,
2861 dstW, dstH, dstFormat, flags,
2862 srcFilter, dstFilter, param);