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
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"
76 #include "libvo/fastmemcpy.h"
86 //#define WORDS_BIGENDIAN
89 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
91 #define RET 0xC3 //near return opcode for X86
94 #define ASSERT(x) assert(x);
102 #define PI 3.14159265358979323846
105 #define isSupportedIn(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
106 || (x)==PIX_FMT_RGB32|| (x)==PIX_FMT_BGR24|| (x)==PIX_FMT_BGR565|| (x)==PIX_FMT_BGR555\
107 || (x)==PIX_FMT_BGR32|| (x)==PIX_FMT_RGB24|| (x)==PIX_FMT_RGB565|| (x)==PIX_FMT_RGB555\
108 || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P\
109 || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
110 || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P)
111 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
112 || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
113 || isRGB(x) || isBGR(x)\
114 || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
115 || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
116 || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
117 #define isPacked(x) ((x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422 ||isRGB(x) || isBGR(x))
119 #define RGB2YUV_SHIFT 16
120 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
121 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
122 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
123 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
124 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
125 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
126 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
127 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
130 extern const int32_t Inverse_Table_6_9[8][4];
134 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
137 more intelligent missalignment avoidance for the horizontal scaler
138 write special vertical cubic upscale version
139 Optimize C code (yv12 / minmax)
140 add support for packed pixel yuv input & output
141 add support for Y8 output
142 optimize bgr24 & bgr32
143 add BGR4 output support
144 write special BGR->BGR scaler
147 #if defined(ARCH_X86) && defined (CONFIG_GPL)
148 static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
149 static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
150 static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
151 static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL;
152 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
153 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
154 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
155 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
157 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
158 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
159 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
160 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
162 static uint64_t __attribute__((aligned(8))) dither4[2]={
163 0x0103010301030103LL,
164 0x0200020002000200LL,};
166 static uint64_t __attribute__((aligned(8))) dither8[2]={
167 0x0602060206020602LL,
168 0x0004000400040004LL,};
170 static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
171 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
172 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
173 static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
174 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
175 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
177 static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
178 static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
179 static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
182 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
183 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
184 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
186 static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
187 static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
188 static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
189 #endif /* FAST_BGR2YV12 */
190 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
191 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
192 static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
193 #endif /* defined(ARCH_X86) */
195 // clipping helper table for C implementations:
196 static unsigned char clip_table[768];
198 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
200 extern const uint8_t dither_2x2_4[2][8];
201 extern const uint8_t dither_2x2_8[2][8];
202 extern const uint8_t dither_8x8_32[8][8];
203 extern const uint8_t dither_8x8_73[8][8];
204 extern const uint8_t dither_8x8_220[8][8];
206 char *sws_format_name(enum PixelFormat format)
209 case PIX_FMT_YUV420P:
211 case PIX_FMT_YUYV422:
217 case PIX_FMT_YUV422P:
219 case PIX_FMT_YUV444P:
223 case PIX_FMT_YUV410P:
225 case PIX_FMT_YUV411P:
231 case PIX_FMT_GRAY16BE:
233 case PIX_FMT_GRAY16LE:
237 case PIX_FMT_MONOWHITE:
239 case PIX_FMT_MONOBLACK:
243 case PIX_FMT_YUVJ420P:
245 case PIX_FMT_YUVJ422P:
247 case PIX_FMT_YUVJ444P:
249 case PIX_FMT_XVMC_MPEG2_MC:
250 return "xvmc_mpeg2_mc";
251 case PIX_FMT_XVMC_MPEG2_IDCT:
252 return "xvmc_mpeg2_idct";
253 case PIX_FMT_UYVY422:
255 case PIX_FMT_UYYVYY411:
257 case PIX_FMT_RGB32_1:
259 case PIX_FMT_BGR32_1:
271 case PIX_FMT_BGR4_BYTE:
277 case PIX_FMT_RGB4_BYTE:
284 return "Unknown format";
288 #if defined(ARCH_X86) && defined (CONFIG_GPL)
289 void in_asm_used_var_warning_killer()
291 volatile int i= bF8+bFC+w10+
292 bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
293 M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
298 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
299 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
300 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
302 //FIXME Optimize (just quickly writen not opti..)
304 for(i=0; i<dstW; i++)
308 for(j=0; j<lumFilterSize; j++)
309 val += lumSrc[j][i] * lumFilter[j];
311 dest[i]= clip_uint8(val>>19);
315 for(i=0; i<chrDstW; i++)
320 for(j=0; j<chrFilterSize; j++)
322 u += chrSrc[j][i] * chrFilter[j];
323 v += chrSrc[j][i + 2048] * chrFilter[j];
326 uDest[i]= clip_uint8(u>>19);
327 vDest[i]= clip_uint8(v>>19);
331 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
332 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
333 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
335 //FIXME Optimize (just quickly writen not opti..)
337 for(i=0; i<dstW; i++)
341 for(j=0; j<lumFilterSize; j++)
342 val += lumSrc[j][i] * lumFilter[j];
344 dest[i]= clip_uint8(val>>19);
350 if(dstFormat == PIX_FMT_NV12)
351 for(i=0; i<chrDstW; i++)
356 for(j=0; j<chrFilterSize; j++)
358 u += chrSrc[j][i] * chrFilter[j];
359 v += chrSrc[j][i + 2048] * chrFilter[j];
362 uDest[2*i]= clip_uint8(u>>19);
363 uDest[2*i+1]= clip_uint8(v>>19);
366 for(i=0; i<chrDstW; i++)
371 for(j=0; j<chrFilterSize; j++)
373 u += chrSrc[j][i] * chrFilter[j];
374 v += chrSrc[j][i + 2048] * chrFilter[j];
377 uDest[2*i]= clip_uint8(v>>19);
378 uDest[2*i+1]= clip_uint8(u>>19);
382 #define YSCALE_YUV_2_PACKEDX_C(type) \
383 for(i=0; i<(dstW>>1); i++){\
389 type attribute_unused *r, *b, *g;\
392 for(j=0; j<lumFilterSize; j++)\
394 Y1 += lumSrc[j][i2] * lumFilter[j];\
395 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
397 for(j=0; j<chrFilterSize; j++)\
399 U += chrSrc[j][i] * chrFilter[j];\
400 V += chrSrc[j][i+2048] * chrFilter[j];\
418 #define YSCALE_YUV_2_RGBX_C(type) \
419 YSCALE_YUV_2_PACKEDX_C(type)\
420 r = (type *)c->table_rV[V];\
421 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
422 b = (type *)c->table_bU[U];\
424 #define YSCALE_YUV_2_PACKED2_C \
425 for(i=0; i<(dstW>>1); i++){\
427 int Y1= (buf0[i2 ]*yalpha1+buf1[i2 ]*yalpha)>>19;\
428 int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
429 int U= (uvbuf0[i ]*uvalpha1+uvbuf1[i ]*uvalpha)>>19;\
430 int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
432 #define YSCALE_YUV_2_RGB2_C(type) \
433 YSCALE_YUV_2_PACKED2_C\
435 r = (type *)c->table_rV[V];\
436 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
437 b = (type *)c->table_bU[U];\
439 #define YSCALE_YUV_2_PACKED1_C \
440 for(i=0; i<(dstW>>1); i++){\
442 int Y1= buf0[i2 ]>>7;\
443 int Y2= buf0[i2+1]>>7;\
444 int U= (uvbuf1[i ])>>7;\
445 int V= (uvbuf1[i+2048])>>7;\
447 #define YSCALE_YUV_2_RGB1_C(type) \
448 YSCALE_YUV_2_PACKED1_C\
450 r = (type *)c->table_rV[V];\
451 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
452 b = (type *)c->table_bU[U];\
454 #define YSCALE_YUV_2_PACKED1B_C \
455 for(i=0; i<(dstW>>1); i++){\
457 int Y1= buf0[i2 ]>>7;\
458 int Y2= buf0[i2+1]>>7;\
459 int U= (uvbuf0[i ] + uvbuf1[i ])>>8;\
460 int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
462 #define YSCALE_YUV_2_RGB1B_C(type) \
463 YSCALE_YUV_2_PACKED1B_C\
465 r = (type *)c->table_rV[V];\
466 g = (type *)(c->table_gU[U] + c->table_gV[V]);\
467 b = (type *)c->table_bU[U];\
469 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
470 switch(c->dstFormat)\
475 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
476 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
481 ((uint8_t*)dest)[0]= r[Y1];\
482 ((uint8_t*)dest)[1]= g[Y1];\
483 ((uint8_t*)dest)[2]= b[Y1];\
484 ((uint8_t*)dest)[3]= r[Y2];\
485 ((uint8_t*)dest)[4]= g[Y2];\
486 ((uint8_t*)dest)[5]= b[Y2];\
492 ((uint8_t*)dest)[0]= b[Y1];\
493 ((uint8_t*)dest)[1]= g[Y1];\
494 ((uint8_t*)dest)[2]= r[Y1];\
495 ((uint8_t*)dest)[3]= b[Y2];\
496 ((uint8_t*)dest)[4]= g[Y2];\
497 ((uint8_t*)dest)[5]= r[Y2];\
501 case PIX_FMT_RGB565:\
502 case PIX_FMT_BGR565:\
504 const int dr1= dither_2x2_8[y&1 ][0];\
505 const int dg1= dither_2x2_4[y&1 ][0];\
506 const int db1= dither_2x2_8[(y&1)^1][0];\
507 const int dr2= dither_2x2_8[y&1 ][1];\
508 const int dg2= dither_2x2_4[y&1 ][1];\
509 const int db2= dither_2x2_8[(y&1)^1][1];\
511 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
512 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
516 case PIX_FMT_RGB555:\
517 case PIX_FMT_BGR555:\
519 const int dr1= dither_2x2_8[y&1 ][0];\
520 const int dg1= dither_2x2_8[y&1 ][1];\
521 const int db1= dither_2x2_8[(y&1)^1][0];\
522 const int dr2= dither_2x2_8[y&1 ][1];\
523 const int dg2= dither_2x2_8[y&1 ][0];\
524 const int db2= dither_2x2_8[(y&1)^1][1];\
526 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
527 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
534 const uint8_t * const d64= dither_8x8_73[y&7];\
535 const uint8_t * const d32= dither_8x8_32[y&7];\
537 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
538 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
545 const uint8_t * const d64= dither_8x8_73 [y&7];\
546 const uint8_t * const d128=dither_8x8_220[y&7];\
548 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
549 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
553 case PIX_FMT_RGB4_BYTE:\
554 case PIX_FMT_BGR4_BYTE:\
556 const uint8_t * const d64= dither_8x8_73 [y&7];\
557 const uint8_t * const d128=dither_8x8_220[y&7];\
559 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
560 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
564 case PIX_FMT_MONOBLACK:\
566 const uint8_t * const d128=dither_8x8_220[y&7];\
567 uint8_t *g= c->table_gU[128] + c->table_gV[128];\
568 for(i=0; i<dstW-7; i+=8){\
570 acc = g[((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19) + d128[0]];\
571 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
572 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
573 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
574 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
575 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
576 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
577 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
578 ((uint8_t*)dest)[0]= acc;\
583 ((uint8_t*)dest)-= dstW>>4;\
587 static int top[1024];\
588 static int last_new[1024][1024];\
589 static int last_in3[1024][1024];\
590 static int drift[1024][1024];\
594 const uint8_t * const d128=dither_8x8_220[y&7];\
599 for(i=dstW>>1; i<dstW; i++){\
600 int in= ((buf0[i ]*yalpha1+buf1[i ]*yalpha)>>19);\
601 int in2 = (76309 * (in - 16) + 32768) >> 16;\
602 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
603 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
604 + (last_new[y][i] - in3)*f/256;\
605 int new= old> 128 ? 255 : 0;\
607 error_new+= FFABS(last_new[y][i] - new);\
608 error_in3+= FFABS(last_in3[y][i] - in3);\
609 f= error_new - error_in3*4;\
614 left= top[i]= old - new;\
615 last_new[y][i]= new;\
616 last_in3[y][i]= in3;\
618 acc+= acc + (new&1);\
620 ((uint8_t*)dest)[0]= acc;\
628 case PIX_FMT_YUYV422:\
630 ((uint8_t*)dest)[2*i2+0]= Y1;\
631 ((uint8_t*)dest)[2*i2+1]= U;\
632 ((uint8_t*)dest)[2*i2+2]= Y2;\
633 ((uint8_t*)dest)[2*i2+3]= V;\
636 case PIX_FMT_UYVY422:\
638 ((uint8_t*)dest)[2*i2+0]= U;\
639 ((uint8_t*)dest)[2*i2+1]= Y1;\
640 ((uint8_t*)dest)[2*i2+2]= V;\
641 ((uint8_t*)dest)[2*i2+3]= Y2;\
647 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
648 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
649 uint8_t *dest, int dstW, int y)
656 YSCALE_YUV_2_RGBX_C(uint32_t)
657 ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
658 ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
662 YSCALE_YUV_2_RGBX_C(uint8_t)
663 ((uint8_t*)dest)[0]= r[Y1];
664 ((uint8_t*)dest)[1]= g[Y1];
665 ((uint8_t*)dest)[2]= b[Y1];
666 ((uint8_t*)dest)[3]= r[Y2];
667 ((uint8_t*)dest)[4]= g[Y2];
668 ((uint8_t*)dest)[5]= b[Y2];
673 YSCALE_YUV_2_RGBX_C(uint8_t)
674 ((uint8_t*)dest)[0]= b[Y1];
675 ((uint8_t*)dest)[1]= g[Y1];
676 ((uint8_t*)dest)[2]= r[Y1];
677 ((uint8_t*)dest)[3]= b[Y2];
678 ((uint8_t*)dest)[4]= g[Y2];
679 ((uint8_t*)dest)[5]= r[Y2];
686 const int dr1= dither_2x2_8[y&1 ][0];
687 const int dg1= dither_2x2_4[y&1 ][0];
688 const int db1= dither_2x2_8[(y&1)^1][0];
689 const int dr2= dither_2x2_8[y&1 ][1];
690 const int dg2= dither_2x2_4[y&1 ][1];
691 const int db2= dither_2x2_8[(y&1)^1][1];
692 YSCALE_YUV_2_RGBX_C(uint16_t)
693 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
694 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
701 const int dr1= dither_2x2_8[y&1 ][0];
702 const int dg1= dither_2x2_8[y&1 ][1];
703 const int db1= dither_2x2_8[(y&1)^1][0];
704 const int dr2= dither_2x2_8[y&1 ][1];
705 const int dg2= dither_2x2_8[y&1 ][0];
706 const int db2= dither_2x2_8[(y&1)^1][1];
707 YSCALE_YUV_2_RGBX_C(uint16_t)
708 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
709 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
716 const uint8_t * const d64= dither_8x8_73[y&7];
717 const uint8_t * const d32= dither_8x8_32[y&7];
718 YSCALE_YUV_2_RGBX_C(uint8_t)
719 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
720 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
727 const uint8_t * const d64= dither_8x8_73 [y&7];
728 const uint8_t * const d128=dither_8x8_220[y&7];
729 YSCALE_YUV_2_RGBX_C(uint8_t)
730 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
731 +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
735 case PIX_FMT_RGB4_BYTE:
736 case PIX_FMT_BGR4_BYTE:
738 const uint8_t * const d64= dither_8x8_73 [y&7];
739 const uint8_t * const d128=dither_8x8_220[y&7];
740 YSCALE_YUV_2_RGBX_C(uint8_t)
741 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
742 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
746 case PIX_FMT_MONOBLACK:
748 const uint8_t * const d128=dither_8x8_220[y&7];
749 uint8_t *g= c->table_gU[128] + c->table_gV[128];
751 for(i=0; i<dstW-1; i+=2){
756 for(j=0; j<lumFilterSize; j++)
758 Y1 += lumSrc[j][i] * lumFilter[j];
759 Y2 += lumSrc[j][i+1] * lumFilter[j];
770 acc+= acc + g[Y1+d128[(i+0)&7]];
771 acc+= acc + g[Y2+d128[(i+1)&7]];
773 ((uint8_t*)dest)[0]= acc;
779 case PIX_FMT_YUYV422:
780 YSCALE_YUV_2_PACKEDX_C(void)
781 ((uint8_t*)dest)[2*i2+0]= Y1;
782 ((uint8_t*)dest)[2*i2+1]= U;
783 ((uint8_t*)dest)[2*i2+2]= Y2;
784 ((uint8_t*)dest)[2*i2+3]= V;
787 case PIX_FMT_UYVY422:
788 YSCALE_YUV_2_PACKEDX_C(void)
789 ((uint8_t*)dest)[2*i2+0]= U;
790 ((uint8_t*)dest)[2*i2+1]= Y1;
791 ((uint8_t*)dest)[2*i2+2]= V;
792 ((uint8_t*)dest)[2*i2+3]= Y2;
799 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
801 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
806 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
807 #define COMPILE_ALTIVEC
808 #endif //HAVE_ALTIVEC
809 #endif //ARCH_POWERPC
811 #if defined(ARCH_X86)
813 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
817 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
821 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
822 #define COMPILE_3DNOW
824 #endif //ARCH_X86 || ARCH_X86_64
835 #define RENAME(a) a ## _C
836 #include "swscale_template.c"
840 #ifdef COMPILE_ALTIVEC
843 #define RENAME(a) a ## _altivec
844 #include "swscale_template.c"
846 #endif //ARCH_POWERPC
848 #if defined(ARCH_X86)
857 #define RENAME(a) a ## _X86
858 #include "swscale_template.c"
866 #define RENAME(a) a ## _MMX
867 #include "swscale_template.c"
876 #define RENAME(a) a ## _MMX2
877 #include "swscale_template.c"
886 #define RENAME(a) a ## _3DNow
887 #include "swscale_template.c"
890 #endif //ARCH_X86 || ARCH_X86_64
892 // minor note: the HAVE_xyz is messed up after that line so don't use it
894 static double getSplineCoeff(double a, double b, double c, double d, double dist)
896 // printf("%f %f %f %f %f\n", a,b,c,d,dist);
897 if(dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
898 else return getSplineCoeff( 0.0,
905 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
906 int srcW, int dstW, int filterAlign, int one, int flags,
907 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
914 double *filter2=NULL;
915 #if defined(ARCH_X86)
916 if(flags & SWS_CPU_CAPS_MMX)
917 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
920 // Note the +1 is for the MMXscaler which reads over the end
921 *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
923 if(FFABS(xInc - 0x10000) <10) // unscaled
927 filter= av_malloc(dstW*sizeof(double)*filterSize);
928 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
930 for(i=0; i<dstW; i++)
932 filter[i*filterSize]=1;
937 else if(flags&SWS_POINT) // lame looking point sampling mode
942 filter= av_malloc(dstW*sizeof(double)*filterSize);
944 xDstInSrc= xInc/2 - 0x8000;
945 for(i=0; i<dstW; i++)
947 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
954 else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
958 if (flags&SWS_BICUBIC) filterSize= 4;
959 else if(flags&SWS_X ) filterSize= 4;
960 else filterSize= 2; // SWS_BILINEAR / SWS_AREA
961 filter= av_malloc(dstW*sizeof(double)*filterSize);
963 xDstInSrc= xInc/2 - 0x8000;
964 for(i=0; i<dstW; i++)
966 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
970 //Bilinear upscale / linear interpolate / Area averaging
971 for(j=0; j<filterSize; j++)
973 double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
974 double coeff= 1.0 - d;
976 filter[i*filterSize + j]= coeff;
985 double sizeFactor, filterSizeInSrc;
986 const double xInc1= (double)xInc / (double)(1<<16);
988 if (flags&SWS_BICUBIC) sizeFactor= 4.0;
989 else if(flags&SWS_X) sizeFactor= 8.0;
990 else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
991 else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
992 else if(flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
993 else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
994 else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
995 else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
997 sizeFactor= 0.0; //GCC warning killer
1001 if(xInc1 <= 1.0) filterSizeInSrc= sizeFactor; // upscale
1002 else filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1004 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1005 if(filterSize > srcW-2) filterSize=srcW-2;
1007 filter= av_malloc(dstW*sizeof(double)*filterSize);
1009 xDstInSrc= xInc1 / 2.0 - 0.5;
1010 for(i=0; i<dstW; i++)
1012 int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1014 (*filterPos)[i]= xx;
1015 for(j=0; j<filterSize; j++)
1017 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1019 if(flags & SWS_BICUBIC)
1021 double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1022 double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1025 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1027 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1031 /* else if(flags & SWS_X)
1033 double p= param ? param*0.01 : 0.3;
1034 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1035 coeff*= pow(2.0, - p*d*d);
1037 else if(flags & SWS_X)
1039 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1045 if(coeff<0.0) coeff= -pow(-coeff, A);
1046 else coeff= pow( coeff, A);
1047 coeff= coeff*0.5 + 0.5;
1049 else if(flags & SWS_AREA)
1051 double srcPixelSize= 1.0/xInc1;
1052 if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1053 else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1056 else if(flags & SWS_GAUSS)
1058 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1059 coeff = pow(2.0, - p*d*d);
1061 else if(flags & SWS_SINC)
1063 coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1065 else if(flags & SWS_LANCZOS)
1067 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1068 coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1071 else if(flags & SWS_BILINEAR)
1074 if(coeff<0) coeff=0;
1076 else if(flags & SWS_SPLINE)
1078 double p=-2.196152422706632;
1079 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1082 coeff= 0.0; //GCC warning killer
1086 filter[i*filterSize + j]= coeff;
1093 /* apply src & dst Filter to filter -> filter2
1096 ASSERT(filterSize>0)
1097 filter2Size= filterSize;
1098 if(srcFilter) filter2Size+= srcFilter->length - 1;
1099 if(dstFilter) filter2Size+= dstFilter->length - 1;
1100 ASSERT(filter2Size>0)
1101 filter2= av_malloc(filter2Size*dstW*sizeof(double));
1103 for(i=0; i<dstW; i++)
1106 SwsVector scaleFilter;
1109 scaleFilter.coeff= filter + i*filterSize;
1110 scaleFilter.length= filterSize;
1112 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1113 else outVec= &scaleFilter;
1115 ASSERT(outVec->length == filter2Size)
1118 for(j=0; j<outVec->length; j++)
1120 filter2[i*filter2Size + j]= outVec->coeff[j];
1123 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1125 if(outVec != &scaleFilter) sws_freeVec(outVec);
1127 av_free(filter); filter=NULL;
1129 /* try to reduce the filter-size (step1 find size and shift left) */
1130 // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1132 for(i=dstW-1; i>=0; i--)
1134 int min= filter2Size;
1138 /* get rid off near zero elements on the left by shifting left */
1139 for(j=0; j<filter2Size; j++)
1142 cutOff += FFABS(filter2[i*filter2Size]);
1144 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1146 /* preserve Monotonicity because the core can't handle the filter otherwise */
1147 if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1149 // Move filter coeffs left
1150 for(k=1; k<filter2Size; k++)
1151 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1152 filter2[i*filter2Size + k - 1]= 0.0;
1157 /* count near zeros on the right */
1158 for(j=filter2Size-1; j>0; j--)
1160 cutOff += FFABS(filter2[i*filter2Size + j]);
1162 if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1166 if(min>minFilterSize) minFilterSize= min;
1169 if (flags & SWS_CPU_CAPS_ALTIVEC) {
1170 // we can handle the special case 4,
1171 // so we don't want to go to the full 8
1172 if (minFilterSize < 5)
1175 // we really don't want to waste our time
1176 // doing useless computation, so fall-back on
1177 // the scalar C code for very small filter.
1178 // vectorizing is worth it only if you have
1179 // decent-sized vector.
1180 if (minFilterSize < 3)
1184 if (flags & SWS_CPU_CAPS_MMX) {
1185 // special case for unscaled vertical filtering
1186 if(minFilterSize == 1 && filterAlign == 2)
1190 ASSERT(minFilterSize > 0)
1191 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1192 ASSERT(filterSize > 0)
1193 filter= av_malloc(filterSize*dstW*sizeof(double));
1194 if(filterSize >= MAX_FILTER_SIZE)
1196 *outFilterSize= filterSize;
1198 if(flags&SWS_PRINT_INFO)
1199 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1200 /* try to reduce the filter-size (step2 reduce it) */
1201 for(i=0; i<dstW; i++)
1205 for(j=0; j<filterSize; j++)
1207 if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1208 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
1211 av_free(filter2); filter2=NULL;
1214 //FIXME try to align filterpos if possible
1217 for(i=0; i<dstW; i++)
1220 if((*filterPos)[i] < 0)
1222 // Move filter coeffs left to compensate for filterPos
1223 for(j=1; j<filterSize; j++)
1225 int left= FFMAX(j + (*filterPos)[i], 0);
1226 filter[i*filterSize + left] += filter[i*filterSize + j];
1227 filter[i*filterSize + j]=0;
1232 if((*filterPos)[i] + filterSize > srcW)
1234 int shift= (*filterPos)[i] + filterSize - srcW;
1235 // Move filter coeffs right to compensate for filterPos
1236 for(j=filterSize-2; j>=0; j--)
1238 int right= FFMIN(j + shift, filterSize-1);
1239 filter[i*filterSize +right] += filter[i*filterSize +j];
1240 filter[i*filterSize +j]=0;
1242 (*filterPos)[i]= srcW - filterSize;
1246 // Note the +1 is for the MMXscaler which reads over the end
1247 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1248 *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1250 /* Normalize & Store in outFilter */
1251 for(i=0; i<dstW; i++)
1258 for(j=0; j<filterSize; j++)
1260 sum+= filter[i*filterSize + j];
1263 for(j=0; j<*outFilterSize; j++)
1265 double v= filter[i*filterSize + j]*scale + error;
1266 int intV= floor(v + 0.5);
1267 (*outFilter)[i*(*outFilterSize) + j]= intV;
1272 (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1273 for(i=0; i<*outFilterSize; i++)
1275 int j= dstW*(*outFilterSize);
1276 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1284 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1287 long imm8OfPShufW1A;
1288 long imm8OfPShufW2A;
1289 long fragmentLengthA;
1291 long imm8OfPShufW1B;
1292 long imm8OfPShufW2B;
1293 long fragmentLengthB;
1298 // create an optimized horizontal scaling routine
1306 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1307 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1308 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1309 "punpcklbw %%mm7, %%mm1 \n\t"
1310 "punpcklbw %%mm7, %%mm0 \n\t"
1311 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
1313 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1315 "psubw %%mm1, %%mm0 \n\t"
1316 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1317 "pmullw %%mm3, %%mm0 \n\t"
1318 "psllw $7, %%mm1 \n\t"
1319 "paddw %%mm1, %%mm0 \n\t"
1321 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1323 "add $8, %%"REG_a" \n\t"
1338 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1339 "=r" (fragmentLengthA)
1346 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
1347 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
1348 "punpcklbw %%mm7, %%mm0 \n\t"
1349 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
1351 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
1353 "psubw %%mm1, %%mm0 \n\t"
1354 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1355 "pmullw %%mm3, %%mm0 \n\t"
1356 "psllw $7, %%mm1 \n\t"
1357 "paddw %%mm1, %%mm0 \n\t"
1359 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1361 "add $8, %%"REG_a" \n\t"
1376 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1377 "=r" (fragmentLengthB)
1380 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1383 for(i=0; i<dstW/numSplits; i++)
1390 int b=((xpos+xInc)>>16) - xx;
1391 int c=((xpos+xInc*2)>>16) - xx;
1392 int d=((xpos+xInc*3)>>16) - xx;
1394 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
1395 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
1396 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1397 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1402 int maxShift= 3-(d+1);
1405 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1407 funnyCode[fragmentPos + imm8OfPShufW1B]=
1408 (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1409 funnyCode[fragmentPos + imm8OfPShufW2B]=
1410 a | (b<<2) | (c<<4) | (d<<6);
1412 if(i+3>=dstW) shift=maxShift; //avoid overread
1413 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1415 if(shift && i>=shift)
1417 funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1418 funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1419 filterPos[i/2]-=shift;
1422 fragmentPos+= fragmentLengthB;
1429 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1431 funnyCode[fragmentPos + imm8OfPShufW1A]=
1432 funnyCode[fragmentPos + imm8OfPShufW2A]=
1433 a | (b<<2) | (c<<4) | (d<<6);
1435 if(i+4>=dstW) shift=maxShift; //avoid overread
1436 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1438 if(shift && i>=shift)
1440 funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1441 funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1442 filterPos[i/2]-=shift;
1445 fragmentPos+= fragmentLengthA;
1448 funnyCode[fragmentPos]= RET;
1452 filterPos[i/2]= xpos>>16; // needed to jump to the next part
1454 #endif /* COMPILE_MMX2 */
1456 static void globalInit(void){
1457 // generating tables:
1459 for(i=0; i<768; i++){
1460 int c= clip_uint8(i-256);
1465 static SwsFunc getSwsFunc(int flags){
1467 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1468 #if defined(ARCH_X86)
1469 // ordered per speed fasterst first
1470 if(flags & SWS_CPU_CAPS_MMX2)
1471 return swScale_MMX2;
1472 else if(flags & SWS_CPU_CAPS_3DNOW)
1473 return swScale_3DNow;
1474 else if(flags & SWS_CPU_CAPS_MMX)
1481 if(flags & SWS_CPU_CAPS_ALTIVEC)
1482 return swScale_altivec;
1487 #endif /* defined(ARCH_X86) */
1488 #else //RUNTIME_CPUDETECT
1490 return swScale_MMX2;
1491 #elif defined (HAVE_3DNOW)
1492 return swScale_3DNow;
1493 #elif defined (HAVE_MMX)
1495 #elif defined (HAVE_ALTIVEC)
1496 return swScale_altivec;
1500 #endif //!RUNTIME_CPUDETECT
1503 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1504 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1505 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1507 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1508 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1512 uint8_t *srcPtr= src[0];
1513 uint8_t *dstPtr= dst;
1514 for(i=0; i<srcSliceH; i++)
1516 memcpy(dstPtr, srcPtr, c->srcW);
1517 srcPtr+= srcStride[0];
1518 dstPtr+= dstStride[0];
1521 dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1522 if (c->dstFormat == PIX_FMT_NV12)
1523 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1525 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1530 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1531 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1532 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1534 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1539 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1540 int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1541 uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1543 yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1548 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1549 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1550 int srcSliceH, uint8_t* dst[], int dstStride[]){
1551 const int srcFormat= c->srcFormat;
1552 const int dstFormat= c->dstFormat;
1553 const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1554 const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1555 const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1556 const int dstId= fmt_depth(dstFormat) >> 2;
1557 void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1560 if( (isBGR(srcFormat) && isBGR(dstFormat))
1561 || (isRGB(srcFormat) && isRGB(dstFormat))){
1562 switch(srcId | (dstId<<4)){
1563 case 0x34: conv= rgb16to15; break;
1564 case 0x36: conv= rgb24to15; break;
1565 case 0x38: conv= rgb32to15; break;
1566 case 0x43: conv= rgb15to16; break;
1567 case 0x46: conv= rgb24to16; break;
1568 case 0x48: conv= rgb32to16; break;
1569 case 0x63: conv= rgb15to24; break;
1570 case 0x64: conv= rgb16to24; break;
1571 case 0x68: conv= rgb32to24; break;
1572 case 0x83: conv= rgb15to32; break;
1573 case 0x84: conv= rgb16to32; break;
1574 case 0x86: conv= rgb24to32; break;
1575 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1576 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1578 }else if( (isBGR(srcFormat) && isRGB(dstFormat))
1579 || (isRGB(srcFormat) && isBGR(dstFormat))){
1580 switch(srcId | (dstId<<4)){
1581 case 0x33: conv= rgb15tobgr15; break;
1582 case 0x34: conv= rgb16tobgr15; break;
1583 case 0x36: conv= rgb24tobgr15; break;
1584 case 0x38: conv= rgb32tobgr15; break;
1585 case 0x43: conv= rgb15tobgr16; break;
1586 case 0x44: conv= rgb16tobgr16; break;
1587 case 0x46: conv= rgb24tobgr16; break;
1588 case 0x48: conv= rgb32tobgr16; break;
1589 case 0x63: conv= rgb15tobgr24; break;
1590 case 0x64: conv= rgb16tobgr24; break;
1591 case 0x66: conv= rgb24tobgr24; break;
1592 case 0x68: conv= rgb32tobgr24; break;
1593 case 0x83: conv= rgb15tobgr32; break;
1594 case 0x84: conv= rgb16tobgr32; break;
1595 case 0x86: conv= rgb24tobgr32; break;
1596 case 0x88: conv= rgb32tobgr32; break;
1597 default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
1598 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1601 MSG_ERR("swScaler: internal error %s -> %s converter\n",
1602 sws_format_name(srcFormat), sws_format_name(dstFormat));
1605 if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1606 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1610 uint8_t *srcPtr= src[0];
1611 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1613 for(i=0; i<srcSliceH; i++)
1615 conv(srcPtr, dstPtr, c->srcW*srcBpp);
1616 srcPtr+= srcStride[0];
1617 dstPtr+= dstStride[0];
1623 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1624 int srcSliceH, uint8_t* dst[], int dstStride[]){
1628 dst[0]+ srcSliceY *dstStride[0],
1629 dst[1]+(srcSliceY>>1)*dstStride[1],
1630 dst[2]+(srcSliceY>>1)*dstStride[2],
1632 dstStride[0], dstStride[1], srcStride[0]);
1636 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1637 int srcSliceH, uint8_t* dst[], int dstStride[]){
1641 if(srcStride[0]==dstStride[0] && srcStride[0] > 0)
1642 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1644 uint8_t *srcPtr= src[0];
1645 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1647 for(i=0; i<srcSliceH; i++)
1649 memcpy(dstPtr, srcPtr, c->srcW);
1650 srcPtr+= srcStride[0];
1651 dstPtr+= dstStride[0];
1655 if(c->dstFormat==PIX_FMT_YUV420P){
1656 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1657 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1659 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1660 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1665 /* unscaled copy like stuff (assumes nearly identical formats) */
1666 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1667 int srcSliceH, uint8_t* dst[], int dstStride[]){
1669 if(isPacked(c->srcFormat))
1671 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1672 memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1676 uint8_t *srcPtr= src[0];
1677 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1680 /* universal length finder */
1681 while(length+c->srcW <= FFABS(dstStride[0])
1682 && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1685 for(i=0; i<srcSliceH; i++)
1687 memcpy(dstPtr, srcPtr, length);
1688 srcPtr+= srcStride[0];
1689 dstPtr+= dstStride[0];
1694 { /* Planar YUV or gray */
1696 for(plane=0; plane<3; plane++)
1698 int length= plane==0 ? c->srcW : -((-c->srcW )>>c->chrDstHSubSample);
1699 int y= plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1700 int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1702 if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1704 if(!isGray(c->dstFormat))
1705 memset(dst[plane], 128, dstStride[plane]*height);
1709 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1710 memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1714 uint8_t *srcPtr= src[plane];
1715 uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1716 for(i=0; i<height; i++)
1718 memcpy(dstPtr, srcPtr, length);
1719 srcPtr+= srcStride[plane];
1720 dstPtr+= dstStride[plane];
1729 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1730 int srcSliceH, uint8_t* dst[], int dstStride[]){
1732 int length= c->srcW;
1734 int height= srcSliceH;
1736 uint8_t *srcPtr= src[0];
1737 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1739 if(!isGray(c->dstFormat)){
1740 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1741 memset(dst[1], 128, dstStride[1]*height);
1742 memset(dst[2], 128, dstStride[2]*height);
1744 if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1745 for(i=0; i<height; i++)
1747 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1748 srcPtr+= srcStride[0];
1749 dstPtr+= dstStride[0];
1754 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1755 int srcSliceH, uint8_t* dst[], int dstStride[]){
1757 int length= c->srcW;
1759 int height= srcSliceH;
1761 uint8_t *srcPtr= src[0];
1762 uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1763 for(i=0; i<height; i++)
1765 for(j=0; j<length; j++)
1767 dstPtr[j<<1] = srcPtr[j];
1768 dstPtr[(j<<1)+1] = srcPtr[j];
1770 srcPtr+= srcStride[0];
1771 dstPtr+= dstStride[0];
1776 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1777 int srcSliceH, uint8_t* dst[], int dstStride[]){
1779 int length= c->srcW;
1781 int height= srcSliceH;
1783 uint16_t *srcPtr= src[0];
1784 uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1785 for(i=0; i<height; i++)
1787 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1788 srcPtr+= srcStride[0]/2;
1789 dstPtr+= dstStride[0]/2;
1795 static void getSubSampleFactors(int *h, int *v, int format){
1797 case PIX_FMT_UYVY422:
1798 case PIX_FMT_YUYV422:
1802 case PIX_FMT_YUV420P:
1803 case PIX_FMT_GRAY16BE:
1804 case PIX_FMT_GRAY16LE:
1805 case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1811 case PIX_FMT_YUV410P:
1815 case PIX_FMT_YUV444P:
1819 case PIX_FMT_YUV422P:
1823 case PIX_FMT_YUV411P:
1834 static uint16_t roundToInt16(int64_t f){
1835 int r= (f + (1<<15))>>16;
1836 if(r<-0x7FFF) return 0x8000;
1837 else if(r> 0x7FFF) return 0x7FFF;
1842 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1843 * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1844 * @return -1 if not supported
1846 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1847 int64_t crv = inv_table[0];
1848 int64_t cbu = inv_table[1];
1849 int64_t cgu = -inv_table[2];
1850 int64_t cgv = -inv_table[3];
1854 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1855 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1856 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
1858 c->brightness= brightness;
1859 c->contrast = contrast;
1860 c->saturation= saturation;
1861 c->srcRange = srcRange;
1862 c->dstRange = dstRange;
1864 c->uOffset= 0x0400040004000400LL;
1865 c->vOffset= 0x0400040004000400LL;
1872 cy = (cy *contrast )>>16;
1873 crv= (crv*contrast * saturation)>>32;
1874 cbu= (cbu*contrast * saturation)>>32;
1875 cgu= (cgu*contrast * saturation)>>32;
1876 cgv= (cgv*contrast * saturation)>>32;
1878 oy -= 256*brightness;
1880 c->yCoeff= roundToInt16(cy *8192) * 0x0001000100010001ULL;
1881 c->vrCoeff= roundToInt16(crv*8192) * 0x0001000100010001ULL;
1882 c->ubCoeff= roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1883 c->vgCoeff= roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1884 c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1885 c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL;
1887 yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1890 #ifdef COMPILE_ALTIVEC
1891 if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1892 yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1898 * @return -1 if not supported
1900 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1901 if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1903 *inv_table = c->srcColorspaceTable;
1904 *table = c->dstColorspaceTable;
1905 *srcRange = c->srcRange;
1906 *dstRange = c->dstRange;
1907 *brightness= c->brightness;
1908 *contrast = c->contrast;
1909 *saturation= c->saturation;
1914 static int handle_jpeg(int *format)
1917 case PIX_FMT_YUVJ420P:
1918 *format = PIX_FMT_YUV420P;
1920 case PIX_FMT_YUVJ422P:
1921 *format = PIX_FMT_YUV422P;
1923 case PIX_FMT_YUVJ444P:
1924 *format = PIX_FMT_YUV444P;
1931 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1932 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1936 int usesVFilter, usesHFilter;
1937 int unscaled, needsDither;
1938 int srcRange, dstRange;
1939 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1940 #if defined(ARCH_X86)
1941 if(flags & SWS_CPU_CAPS_MMX)
1942 asm volatile("emms\n\t"::: "memory");
1945 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1946 flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1948 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1949 #elif defined (HAVE_3DNOW)
1950 flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1951 #elif defined (HAVE_MMX)
1952 flags |= SWS_CPU_CAPS_MMX;
1953 #elif defined (HAVE_ALTIVEC)
1954 flags |= SWS_CPU_CAPS_ALTIVEC;
1956 #endif /* RUNTIME_CPUDETECT */
1957 if(clip_table[512] != 255) globalInit();
1958 if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1960 unscaled = (srcW == dstW && srcH == dstH);
1961 needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
1962 && (fmt_depth(dstFormat))<24
1963 && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1965 srcRange = handle_jpeg(&srcFormat);
1966 dstRange = handle_jpeg(&dstFormat);
1968 if(!isSupportedIn(srcFormat))
1970 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1973 if(!isSupportedOut(dstFormat))
1975 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1980 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
1982 MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
1983 srcW, srcH, dstW, dstH);
1987 if(!dstFilter) dstFilter= &dummyFilter;
1988 if(!srcFilter) srcFilter= &dummyFilter;
1990 c= av_mallocz(sizeof(SwsContext));
1996 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1997 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1999 c->dstFormat= dstFormat;
2000 c->srcFormat= srcFormat;
2001 c->vRounder= 4* 0x0001000100010001ULL;
2003 usesHFilter= usesVFilter= 0;
2004 if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2005 if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2006 if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2007 if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2008 if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2009 if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2010 if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2011 if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2013 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2014 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2016 // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2017 if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2019 // drop some chroma lines if the user wants it
2020 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2021 c->chrSrcVSubSample+= c->vChrDrop;
2023 // drop every 2. pixel for chroma calculation unless user wants full chroma
2024 if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
2025 c->chrSrcHSubSample=1;
2028 c->param[0] = param[0];
2029 c->param[1] = param[1];
2032 c->param[1] = SWS_PARAM_DEFAULT;
2035 c->chrIntHSubSample= c->chrDstHSubSample;
2036 c->chrIntVSubSample= c->chrSrcVSubSample;
2038 // note the -((-x)>>y) is so that we allways round toward +inf
2039 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2040 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2041 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2042 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2044 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);
2046 /* unscaled special Cases */
2047 if(unscaled && !usesHFilter && !usesVFilter)
2050 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2052 c->swScale= PlanarToNV12Wrapper;
2056 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2058 c->swScale= yuv2rgb_get_func_ptr(c);
2062 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2064 c->swScale= yvu9toyv12Wrapper;
2068 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2069 c->swScale= bgr24toyv12Wrapper;
2071 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2072 if( (isBGR(srcFormat) || isRGB(srcFormat))
2073 && (isBGR(dstFormat) || isRGB(dstFormat))
2075 c->swScale= rgb2rgbWrapper;
2077 /* LQ converters if -sws 0 or -sws 4*/
2078 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2079 /* rgb/bgr -> rgb/bgr (dither needed forms) */
2080 if( (isBGR(srcFormat) || isRGB(srcFormat))
2081 && (isBGR(dstFormat) || isRGB(dstFormat))
2083 c->swScale= rgb2rgbWrapper;
2086 if(srcFormat == PIX_FMT_YUV420P &&
2087 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2089 if (dstFormat == PIX_FMT_YUYV422)
2090 c->swScale= PlanarToYuy2Wrapper;
2092 c->swScale= PlanarToUyvyWrapper;
2096 #ifdef COMPILE_ALTIVEC
2097 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2098 ((srcFormat == PIX_FMT_YUV420P &&
2099 (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2100 // unscaled YV12 -> packed YUV, we want speed
2101 if (dstFormat == PIX_FMT_YUYV422)
2102 c->swScale= yv12toyuy2_unscaled_altivec;
2104 c->swScale= yv12touyvy_unscaled_altivec;
2109 if( srcFormat == dstFormat
2110 || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2111 || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2114 c->swScale= simpleCopy;
2117 /* gray16{le,be} conversions */
2118 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2120 c->swScale= gray16togray;
2122 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2124 c->swScale= graytogray16;
2126 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2128 c->swScale= gray16swap;
2132 if(flags&SWS_PRINT_INFO)
2133 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n",
2134 sws_format_name(srcFormat), sws_format_name(dstFormat));
2139 if(flags & SWS_CPU_CAPS_MMX2)
2141 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2142 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2144 if(flags&SWS_PRINT_INFO)
2145 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2147 if(usesHFilter) c->canMMX2BeUsed=0;
2152 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2153 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2155 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2156 // but only for the FAST_BILINEAR mode otherwise do correct scaling
2157 // n-2 is the last chrominance sample available
2158 // this is not perfect, but noone shuld notice the difference, the more correct variant
2159 // would be like the vertical one, but that would require some special code for the
2160 // first and last pixel
2161 if(flags&SWS_FAST_BILINEAR)
2163 if(c->canMMX2BeUsed)
2168 //we don't use the x86asm scaler if mmx is available
2169 else if(flags & SWS_CPU_CAPS_MMX)
2171 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2172 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2176 /* precalculate horizontal scaler filter coefficients */
2178 const int filterAlign=
2179 (flags & SWS_CPU_CAPS_MMX) ? 4 :
2180 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2183 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2184 srcW , dstW, filterAlign, 1<<14,
2185 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2186 srcFilter->lumH, dstFilter->lumH, c->param);
2187 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2188 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2189 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2190 srcFilter->chrH, dstFilter->chrH, c->param);
2192 #define MAX_FUNNY_CODE_SIZE 10000
2193 #if defined(COMPILE_MMX2)
2194 // can't downscale !!!
2195 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2197 #ifdef MAP_ANONYMOUS
2198 c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2199 c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2201 c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2202 c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2205 c->lumMmx2Filter = av_malloc((dstW /8+8)*sizeof(int16_t));
2206 c->chrMmx2Filter = av_malloc((c->chrDstW /4+8)*sizeof(int16_t));
2207 c->lumMmx2FilterPos= av_malloc((dstW /2/8+8)*sizeof(int32_t));
2208 c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2210 initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2211 initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2213 #endif /* defined(COMPILE_MMX2) */
2214 } // Init Horizontal stuff
2218 /* precalculate vertical scaler filter coefficients */
2220 const int filterAlign=
2221 (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2222 (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2225 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2226 srcH , dstH, filterAlign, (1<<12)-4,
2227 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
2228 srcFilter->lumV, dstFilter->lumV, c->param);
2229 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2230 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2231 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2232 srcFilter->chrV, dstFilter->chrV, c->param);
2235 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2236 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2238 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2240 short *p = (short *)&c->vYCoeffsBank[i];
2242 p[j] = c->vLumFilter[i];
2245 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2247 short *p = (short *)&c->vCCoeffsBank[i];
2249 p[j] = c->vChrFilter[i];
2254 // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2255 c->vLumBufSize= c->vLumFilterSize;
2256 c->vChrBufSize= c->vChrFilterSize;
2257 for(i=0; i<dstH; i++)
2259 int chrI= i*c->chrDstH / dstH;
2260 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
2261 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2263 nextSlice>>= c->chrSrcVSubSample;
2264 nextSlice<<= c->chrSrcVSubSample;
2265 if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
2266 c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
2267 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2268 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2271 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2272 c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2273 c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2274 //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)
2275 /* align at 16 bytes for AltiVec */
2276 for(i=0; i<c->vLumBufSize; i++)
2277 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(4000);
2278 for(i=0; i<c->vChrBufSize; i++)
2279 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2281 //try to avoid drawing green stuff between the right end and the stride end
2282 for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2284 ASSERT(c->chrDstH <= dstH)
2286 if(flags&SWS_PRINT_INFO)
2289 char *dither= " dithered";
2293 if(flags&SWS_FAST_BILINEAR)
2294 MSG_INFO("SwScaler: FAST_BILINEAR scaler, ");
2295 else if(flags&SWS_BILINEAR)
2296 MSG_INFO("SwScaler: BILINEAR scaler, ");
2297 else if(flags&SWS_BICUBIC)
2298 MSG_INFO("SwScaler: BICUBIC scaler, ");
2299 else if(flags&SWS_X)
2300 MSG_INFO("SwScaler: Experimental scaler, ");
2301 else if(flags&SWS_POINT)
2302 MSG_INFO("SwScaler: Nearest Neighbor / POINT scaler, ");
2303 else if(flags&SWS_AREA)
2304 MSG_INFO("SwScaler: Area Averageing scaler, ");
2305 else if(flags&SWS_BICUBLIN)
2306 MSG_INFO("SwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2307 else if(flags&SWS_GAUSS)
2308 MSG_INFO("SwScaler: Gaussian scaler, ");
2309 else if(flags&SWS_SINC)
2310 MSG_INFO("SwScaler: Sinc scaler, ");
2311 else if(flags&SWS_LANCZOS)
2312 MSG_INFO("SwScaler: Lanczos scaler, ");
2313 else if(flags&SWS_SPLINE)
2314 MSG_INFO("SwScaler: Bicubic spline scaler, ");
2316 MSG_INFO("SwScaler: ehh flags invalid?! ");
2318 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2319 MSG_INFO("from %s to%s %s ",
2320 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2322 MSG_INFO("from %s to %s ",
2323 sws_format_name(srcFormat), sws_format_name(dstFormat));
2325 if(flags & SWS_CPU_CAPS_MMX2)
2326 MSG_INFO("using MMX2\n");
2327 else if(flags & SWS_CPU_CAPS_3DNOW)
2328 MSG_INFO("using 3DNOW\n");
2329 else if(flags & SWS_CPU_CAPS_MMX)
2330 MSG_INFO("using MMX\n");
2331 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2332 MSG_INFO("using AltiVec\n");
2334 MSG_INFO("using C\n");
2337 if(flags & SWS_PRINT_INFO)
2339 if(flags & SWS_CPU_CAPS_MMX)
2341 if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2342 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2345 if(c->hLumFilterSize==4)
2346 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2347 else if(c->hLumFilterSize==8)
2348 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2350 MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2352 if(c->hChrFilterSize==4)
2353 MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2354 else if(c->hChrFilterSize==8)
2355 MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2357 MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2362 #if defined(ARCH_X86)
2363 MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2365 if(flags & SWS_FAST_BILINEAR)
2366 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2368 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2371 if(isPlanarYUV(dstFormat))
2373 if(c->vLumFilterSize==1)
2374 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2376 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2380 if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2381 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2382 "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2383 else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2384 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2386 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2389 if(dstFormat==PIX_FMT_BGR24)
2390 MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2391 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2392 else if(dstFormat==PIX_FMT_RGB32)
2393 MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2394 else if(dstFormat==PIX_FMT_BGR565)
2395 MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2396 else if(dstFormat==PIX_FMT_BGR555)
2397 MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2399 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2401 if(flags & SWS_PRINT_INFO)
2403 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2404 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2405 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2406 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2409 c->swScale= getSwsFunc(flags);
2414 * swscale warper, so we don't need to export the SwsContext.
2415 * assumes planar YUV to be in YUV order instead of YVU
2417 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2418 int srcSliceH, uint8_t* dst[], int dstStride[]){
2419 if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2420 MSG_ERR("swScaler: slices start in the middle!\n");
2423 if (c->sliceDir == 0) {
2424 if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2427 // copy strides, so they can safely be modified
2428 if (c->sliceDir == 1) {
2429 // slices go from top to bottom
2430 int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2431 int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2432 return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2434 // slices go from bottom to top => we flip the image internally
2435 uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2436 src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2437 src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2439 uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2440 dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2441 dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2442 int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2443 int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2445 return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2450 * swscale warper, so we don't need to export the SwsContext
2452 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2453 int srcSliceH, uint8_t* dst[], int dstStride[]){
2454 return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2457 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2458 float lumaSharpen, float chromaSharpen,
2459 float chromaHShift, float chromaVShift,
2462 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2465 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2466 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2468 filter->lumH= sws_getIdentityVec();
2469 filter->lumV= sws_getIdentityVec();
2472 if(chromaGBlur!=0.0){
2473 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2474 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2476 filter->chrH= sws_getIdentityVec();
2477 filter->chrV= sws_getIdentityVec();
2480 if(chromaSharpen!=0.0){
2481 SwsVector *id= sws_getIdentityVec();
2482 sws_scaleVec(filter->chrH, -chromaSharpen);
2483 sws_scaleVec(filter->chrV, -chromaSharpen);
2484 sws_addVec(filter->chrH, id);
2485 sws_addVec(filter->chrV, id);
2489 if(lumaSharpen!=0.0){
2490 SwsVector *id= sws_getIdentityVec();
2491 sws_scaleVec(filter->lumH, -lumaSharpen);
2492 sws_scaleVec(filter->lumV, -lumaSharpen);
2493 sws_addVec(filter->lumH, id);
2494 sws_addVec(filter->lumV, id);
2498 if(chromaHShift != 0.0)
2499 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2501 if(chromaVShift != 0.0)
2502 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2504 sws_normalizeVec(filter->chrH, 1.0);
2505 sws_normalizeVec(filter->chrV, 1.0);
2506 sws_normalizeVec(filter->lumH, 1.0);
2507 sws_normalizeVec(filter->lumV, 1.0);
2509 if(verbose) sws_printVec(filter->chrH);
2510 if(verbose) sws_printVec(filter->lumH);
2516 * returns a normalized gaussian curve used to filter stuff
2517 * quality=3 is high quality, lowwer is lowwer quality
2519 SwsVector *sws_getGaussianVec(double variance, double quality){
2520 const int length= (int)(variance*quality + 0.5) | 1;
2522 double *coeff= av_malloc(length*sizeof(double));
2523 double middle= (length-1)*0.5;
2524 SwsVector *vec= av_malloc(sizeof(SwsVector));
2527 vec->length= length;
2529 for(i=0; i<length; i++)
2531 double dist= i-middle;
2532 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2535 sws_normalizeVec(vec, 1.0);
2540 SwsVector *sws_getConstVec(double c, int length){
2542 double *coeff= av_malloc(length*sizeof(double));
2543 SwsVector *vec= av_malloc(sizeof(SwsVector));
2546 vec->length= length;
2548 for(i=0; i<length; i++)
2555 SwsVector *sws_getIdentityVec(void){
2556 return sws_getConstVec(1.0, 1);
2559 double sws_dcVec(SwsVector *a){
2563 for(i=0; i<a->length; i++)
2569 void sws_scaleVec(SwsVector *a, double scalar){
2572 for(i=0; i<a->length; i++)
2573 a->coeff[i]*= scalar;
2576 void sws_normalizeVec(SwsVector *a, double height){
2577 sws_scaleVec(a, height/sws_dcVec(a));
2580 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2581 int length= a->length + b->length - 1;
2582 double *coeff= av_malloc(length*sizeof(double));
2584 SwsVector *vec= av_malloc(sizeof(SwsVector));
2587 vec->length= length;
2589 for(i=0; i<length; i++) coeff[i]= 0.0;
2591 for(i=0; i<a->length; i++)
2593 for(j=0; j<b->length; j++)
2595 coeff[i+j]+= a->coeff[i]*b->coeff[j];
2602 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2603 int length= FFMAX(a->length, b->length);
2604 double *coeff= av_malloc(length*sizeof(double));
2606 SwsVector *vec= av_malloc(sizeof(SwsVector));
2609 vec->length= length;
2611 for(i=0; i<length; i++) coeff[i]= 0.0;
2613 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2614 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2619 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2620 int length= FFMAX(a->length, b->length);
2621 double *coeff= av_malloc(length*sizeof(double));
2623 SwsVector *vec= av_malloc(sizeof(SwsVector));
2626 vec->length= length;
2628 for(i=0; i<length; i++) coeff[i]= 0.0;
2630 for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2631 for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2636 /* shift left / or right if "shift" is negative */
2637 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2638 int length= a->length + FFABS(shift)*2;
2639 double *coeff= av_malloc(length*sizeof(double));
2641 SwsVector *vec= av_malloc(sizeof(SwsVector));
2644 vec->length= length;
2646 for(i=0; i<length; i++) coeff[i]= 0.0;
2648 for(i=0; i<a->length; i++)
2650 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2656 void sws_shiftVec(SwsVector *a, int shift){
2657 SwsVector *shifted= sws_getShiftedVec(a, shift);
2659 a->coeff= shifted->coeff;
2660 a->length= shifted->length;
2664 void sws_addVec(SwsVector *a, SwsVector *b){
2665 SwsVector *sum= sws_sumVec(a, b);
2667 a->coeff= sum->coeff;
2668 a->length= sum->length;
2672 void sws_subVec(SwsVector *a, SwsVector *b){
2673 SwsVector *diff= sws_diffVec(a, b);
2675 a->coeff= diff->coeff;
2676 a->length= diff->length;
2680 void sws_convVec(SwsVector *a, SwsVector *b){
2681 SwsVector *conv= sws_getConvVec(a, b);
2683 a->coeff= conv->coeff;
2684 a->length= conv->length;
2688 SwsVector *sws_cloneVec(SwsVector *a){
2689 double *coeff= av_malloc(a->length*sizeof(double));
2691 SwsVector *vec= av_malloc(sizeof(SwsVector));
2694 vec->length= a->length;
2696 for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2701 void sws_printVec(SwsVector *a){
2707 for(i=0; i<a->length; i++)
2708 if(a->coeff[i]>max) max= a->coeff[i];
2710 for(i=0; i<a->length; i++)
2711 if(a->coeff[i]<min) min= a->coeff[i];
2715 for(i=0; i<a->length; i++)
2717 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2718 MSG_DBG2("%1.3f ", a->coeff[i]);
2719 for(;x>0; x--) MSG_DBG2(" ");
2724 void sws_freeVec(SwsVector *a){
2732 void sws_freeFilter(SwsFilter *filter){
2735 if(filter->lumH) sws_freeVec(filter->lumH);
2736 if(filter->lumV) sws_freeVec(filter->lumV);
2737 if(filter->chrH) sws_freeVec(filter->chrH);
2738 if(filter->chrV) sws_freeVec(filter->chrV);
2743 void sws_freeContext(SwsContext *c){
2749 for(i=0; i<c->vLumBufSize; i++)
2751 av_free(c->lumPixBuf[i]);
2752 c->lumPixBuf[i]=NULL;
2754 av_free(c->lumPixBuf);
2760 for(i=0; i<c->vChrBufSize; i++)
2762 av_free(c->chrPixBuf[i]);
2763 c->chrPixBuf[i]=NULL;
2765 av_free(c->chrPixBuf);
2769 av_free(c->vLumFilter);
2770 c->vLumFilter = NULL;
2771 av_free(c->vChrFilter);
2772 c->vChrFilter = NULL;
2773 av_free(c->hLumFilter);
2774 c->hLumFilter = NULL;
2775 av_free(c->hChrFilter);
2776 c->hChrFilter = NULL;
2778 av_free(c->vYCoeffsBank);
2779 c->vYCoeffsBank = NULL;
2780 av_free(c->vCCoeffsBank);
2781 c->vCCoeffsBank = NULL;
2784 av_free(c->vLumFilterPos);
2785 c->vLumFilterPos = NULL;
2786 av_free(c->vChrFilterPos);
2787 c->vChrFilterPos = NULL;
2788 av_free(c->hLumFilterPos);
2789 c->hLumFilterPos = NULL;
2790 av_free(c->hChrFilterPos);
2791 c->hChrFilterPos = NULL;
2793 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2794 #ifdef MAP_ANONYMOUS
2795 if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2796 if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2798 av_free(c->funnyYCode);
2799 av_free(c->funnyUVCode);
2802 c->funnyUVCode=NULL;
2803 #endif /* defined(ARCH_X86) */
2805 av_free(c->lumMmx2Filter);
2806 c->lumMmx2Filter=NULL;
2807 av_free(c->chrMmx2Filter);
2808 c->chrMmx2Filter=NULL;
2809 av_free(c->lumMmx2FilterPos);
2810 c->lumMmx2FilterPos=NULL;
2811 av_free(c->chrMmx2FilterPos);
2812 c->chrMmx2FilterPos=NULL;
2813 av_free(c->yuvTable);
2820 * Checks if context is valid or reallocs a new one instead.
2821 * If context is NULL, just calls sws_getContext() to get a new one.
2822 * Otherwise, checks if the parameters are the same already saved in context.
2823 * If that is the case, returns the current context.
2824 * Otherwise, frees context and gets a new one.
2826 * Be warned that srcFilter, dstFilter are not checked, they are
2827 * asumed to remain valid.
2829 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2830 int srcW, int srcH, int srcFormat,
2831 int dstW, int dstH, int dstFormat, int flags,
2832 SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2834 if (context != NULL) {
2835 if ((context->srcW != srcW) || (context->srcH != srcH) ||
2836 (context->srcFormat != srcFormat) ||
2837 (context->dstW != dstW) || (context->dstH != dstH) ||
2838 (context->dstFormat != dstFormat) || (context->flags != flags) ||
2839 (context->param != param))
2841 sws_freeContext(context);
2845 if (context == NULL) {
2846 return sws_getContext(srcW, srcH, srcFormat,
2847 dstW, dstH, dstFormat, flags,
2848 srcFilter, dstFilter, param);