* resync without endless loop
[ffmpeg.git] / postproc / swscale.c
1 /*
2     Copyright (C) 2001-2002 Michael Niedermayer <michaelni@gmx.at>
3
4     This program is free software; you can redistribute it and/or modify
5     it under the terms of the GNU General Public License as published by
6     the Free Software Foundation; either version 2 of the License, or
7     (at your option) any later version.
8
9     This program is distributed in the hope that it will be useful,
10     but WITHOUT ANY WARRANTY; without even the implied warranty of
11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12     GNU General Public License for more details.
13
14     You should have received a copy of the GNU General Public License
15     along with this program; if not, write to the Free Software
16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17 */
18
19 /*
20   supported Input formats: YV12, I420/IYUV, YUY2, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21   supported output formats: YV12, I420/IYUV, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
22   {BGR,RGB}{1,4,8,15,16} support dithering
23   
24   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
25   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
26   x -> x
27   YUV9 -> YV12
28   YUV9/YV12 -> Y800
29   Y800 -> YUV9/YV12
30   BGR24 -> BGR32 & RGB24 -> RGB32
31   BGR32 -> BGR24 & RGB32 -> RGB24
32   BGR15 -> BGR16
33 */
34
35 /* 
36 tested special converters (most are tested actually but i didnt write it down ...)
37  YV12 -> BGR16
38  YV12 -> YV12
39  BGR15 -> BGR16
40  BGR16 -> BGR16
41  YVU9 -> YV12
42
43 untested special converters
44   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
45   YV12/I420 -> YV12/I420
46   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
47   BGR24 -> BGR32 & RGB24 -> RGB32
48   BGR32 -> BGR24 & RGB32 -> RGB24
49   BGR24 -> YV12
50 */
51
52 #include <inttypes.h>
53 #include <string.h>
54 #include <math.h>
55 #include <stdio.h>
56 #include "../config.h"
57 #include "../mangle.h"
58 #include <assert.h>
59 #ifdef HAVE_MALLOC_H
60 #include <malloc.h>
61 #else
62 #include <stdlib.h>
63 #endif
64 #include "swscale.h"
65 #include "../cpudetect.h"
66 #include "../bswap.h"
67 #include "../libvo/img_format.h"
68 #include "rgb2rgb.h"
69 #include "../libvo/fastmemcpy.h"
70 #include "../mp_msg.h"
71
72 #define MSG_WARN(args...) mp_msg(MSGT_SWS,MSGL_WARN, ##args )
73 #define MSG_FATAL(args...) mp_msg(MSGT_SWS,MSGL_FATAL, ##args )
74 #define MSG_ERR(args...) mp_msg(MSGT_SWS,MSGL_ERR, ##args )
75 #define MSG_V(args...) mp_msg(MSGT_SWS,MSGL_V, ##args )
76 #define MSG_DBG2(args...) mp_msg(MSGT_SWS,MSGL_DBG2, ##args )
77 #define MSG_INFO(args...) mp_msg(MSGT_SWS,MSGL_INFO, ##args )
78
79 #undef MOVNTQ
80 #undef PAVGB
81
82 //#undef HAVE_MMX2
83 //#define HAVE_3DNOW
84 //#undef HAVE_MMX
85 //#undef ARCH_X86
86 //#define WORDS_BIGENDIAN
87 #define DITHER1XBPP
88
89 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
90
91 #define RET 0xC3 //near return opcode for X86
92
93 #ifdef MP_DEBUG
94 #define ASSERT(x) assert(x);
95 #else
96 #define ASSERT(x) ;
97 #endif
98
99 #ifdef M_PI
100 #define PI M_PI
101 #else
102 #define PI 3.14159265358979323846
103 #endif
104
105 //FIXME replace this with something faster
106 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YVU9 \
107                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
108 #define isYUV(x)       ((x)==IMGFMT_YUY2 || isPlanarYUV(x))
109 #define isGray(x)      ((x)==IMGFMT_Y800)
110 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
111 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
112 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 || (x)==IMGFMT_YUY2 \
113                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
114                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
115                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
116                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
117 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_I420 \
118                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
119                         || isRGB(x) || isBGR(x)\
120                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
121 #define isPacked(x)    ((x)==IMGFMT_YUY2 || isRGB(x) || isBGR(x))
122
123 #define RGB2YUV_SHIFT 16
124 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
125 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
126 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
127 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
128 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
129 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
130 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
131 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
132 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
133
134 extern int verbose; // defined in mplayer.c
135 /*
136 NOTES
137 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
138
139 TODO
140 more intelligent missalignment avoidance for the horizontal scaler
141 write special vertical cubic upscale version
142 Optimize C code (yv12 / minmax)
143 add support for packed pixel yuv input & output
144 add support for Y8 output
145 optimize bgr24 & bgr32
146 add BGR4 output support
147 write special BGR->BGR scaler
148 deglobalize yuv2rgb*.c
149 */
150
151 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
152 #define MIN(a,b) ((a) > (b) ? (b) : (a))
153 #define MAX(a,b) ((a) < (b) ? (b) : (a))
154
155 #ifdef ARCH_X86
156 #define CAN_COMPILE_X86_ASM
157 #endif
158
159 #ifdef CAN_COMPILE_X86_ASM
160 static uint64_t __attribute__((aligned(8))) yCoeff=    0x2568256825682568LL;
161 static uint64_t __attribute__((aligned(8))) vrCoeff=   0x3343334333433343LL;
162 static uint64_t __attribute__((aligned(8))) ubCoeff=   0x40cf40cf40cf40cfLL;
163 static uint64_t __attribute__((aligned(8))) vgCoeff=   0xE5E2E5E2E5E2E5E2LL;
164 static uint64_t __attribute__((aligned(8))) ugCoeff=   0xF36EF36EF36EF36ELL;
165 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
166 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
167 static uint64_t __attribute__((aligned(8))) w400=      0x0400040004000400LL;
168 static uint64_t __attribute__((aligned(8))) w80=       0x0080008000800080LL;
169 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
170 static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;
171 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
172 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
173 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
174 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
175
176 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
177 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
178 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
179 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
180
181 static uint64_t __attribute__((aligned(8))) dither4[2]={
182         0x0103010301030103LL,
183         0x0200020002000200LL,};
184
185 static uint64_t __attribute__((aligned(8))) dither8[2]={
186         0x0602060206020602LL,
187         0x0004000400040004LL,};
188
189 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
190 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
191 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
192 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
193 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
194 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
195
196 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
197 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
198 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
199
200 #ifdef FAST_BGR2YV12
201 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
202 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
203 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
204 #else
205 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
206 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
207 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
208 #endif
209 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
210 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
211 static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
212 #endif
213
214 // clipping helper table for C implementations:
215 static unsigned char clip_table[768];
216
217 //global sws_flags from the command line
218 int sws_flags=2;
219
220 //global srcFilter
221 SwsFilter src_filter= {NULL, NULL, NULL, NULL};
222
223 float sws_lum_gblur= 0.0;
224 float sws_chr_gblur= 0.0;
225 int sws_chr_vshift= 0;
226 int sws_chr_hshift= 0;
227 float sws_chr_sharpen= 0.0;
228 float sws_lum_sharpen= 0.0;
229
230 /* cpuCaps combined from cpudetect and whats actually compiled in
231    (if there is no support for something compiled in it wont appear here) */
232 static CpuCaps cpuCaps;
233
234 void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
235              int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
236
237 static SwsVector *getConvVec(SwsVector *a, SwsVector *b);
238 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]);
239 void *yuv2rgb_c_init (unsigned bpp, int mode, void *table_rV[256], void *table_gU[256], int table_gV[256], void *table_bU[256]);
240
241 extern const uint8_t dither_2x2_4[2][8];
242 extern const uint8_t dither_2x2_8[2][8];
243 extern const uint8_t dither_8x8_32[8][8];
244 extern const uint8_t dither_8x8_73[8][8];
245 extern const uint8_t dither_8x8_220[8][8];
246
247 #ifdef CAN_COMPILE_X86_ASM
248 void in_asm_used_var_warning_killer()
249 {
250  volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
251  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
252  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
253  if(i) i=0;
254 }
255 #endif
256
257 static int testFormat[]={
258 IMGFMT_YVU9,
259 IMGFMT_YV12,
260 //IMGFMT_IYUV,
261 IMGFMT_I420,
262 IMGFMT_BGR15,
263 IMGFMT_BGR16,
264 IMGFMT_BGR24,
265 IMGFMT_BGR32,
266 IMGFMT_RGB24,
267 IMGFMT_RGB32,
268 //IMGFMT_Y8,
269 IMGFMT_Y800,
270 //IMGFMT_YUY2,
271 0
272 };
273
274 static uint64_t getSSD(uint8_t *src1, uint8_t *src2, int stride1, int stride2, int w, int h){
275         int x,y;
276         uint64_t ssd=0;
277
278         for(y=0; y<h; y++){
279                 for(x=0; x<w; x++){
280                         int d= src1[x + y*stride1] - src2[x + y*stride2];
281                         ssd+= d*d;
282                 }
283         }
284         return ssd;
285 }
286
287 // test by ref -> src -> dst -> out & compare out against ref
288 // ref & out are YV12
289 static void doTest(uint8_t *ref[3], int refStride[3], int w, int h, int srcFormat, int dstFormat, 
290                    int srcW, int srcH, int dstW, int dstH, int flags){
291         uint8_t *src[3];
292         uint8_t *dst[3];
293         uint8_t *out[3];
294         int srcStride[3], dstStride[3];
295         int i;
296         uint64_t ssdY, ssdU, ssdV;
297         SwsContext *srcContext, *dstContext, *outContext;
298         
299         for(i=0; i<3; i++){
300                 // avoid stride % bpp != 0
301                 if(srcFormat==IMGFMT_RGB24 || srcFormat==IMGFMT_BGR24)
302                         srcStride[i]= srcW*3;
303                 else
304                         srcStride[i]= srcW*4;
305                 
306                 if(dstFormat==IMGFMT_RGB24 || dstFormat==IMGFMT_BGR24)
307                         dstStride[i]= dstW*3;
308                 else
309                         dstStride[i]= dstW*4;
310         
311                 src[i]= malloc(srcStride[i]*srcH);
312                 dst[i]= malloc(dstStride[i]*dstH);
313                 out[i]= malloc(refStride[i]*h);
314         }
315
316         srcContext= getSwsContext(w, h, IMGFMT_YV12, srcW, srcH, srcFormat, flags, NULL, NULL);
317         dstContext= getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, NULL, NULL);
318         outContext= getSwsContext(dstW, dstH, dstFormat, w, h, IMGFMT_YV12, flags, NULL, NULL);
319         if(srcContext==NULL ||dstContext==NULL ||outContext==NULL){
320                 printf("Failed allocating swsContext\n");
321                 goto end;
322         }
323 //      printf("test %X %X %X -> %X %X %X\n", (int)ref[0], (int)ref[1], (int)ref[2],
324 //              (int)src[0], (int)src[1], (int)src[2]);
325
326         srcContext->swScale(srcContext, ref, refStride, 0, h   , src, srcStride);
327         dstContext->swScale(dstContext, src, srcStride, 0, srcH, dst, dstStride);
328         outContext->swScale(outContext, dst, dstStride, 0, dstH, out, refStride);
329              
330         ssdY= getSSD(ref[0], out[0], refStride[0], refStride[0], w, h);
331         ssdU= getSSD(ref[1], out[1], refStride[1], refStride[1], (w+1)>>1, (h+1)>>1);
332         ssdV= getSSD(ref[2], out[2], refStride[2], refStride[2], (w+1)>>1, (h+1)>>1);
333         
334         if(isGray(srcFormat) || isGray(dstFormat)) ssdU=ssdV=0; //FIXME check that output is really gray
335         
336         ssdY/= w*h;
337         ssdU/= w*h/4;
338         ssdV/= w*h/4;
339         
340         if(ssdY>100 || ssdU>50 || ssdV>50){
341                 printf(" %s %dx%d -> %s %4dx%4d flags=%2d SSD=%5lld,%5lld,%5lld\n", 
342                         vo_format_name(srcFormat), srcW, srcH, 
343                         vo_format_name(dstFormat), dstW, dstH,
344                         flags,
345                         ssdY, ssdU, ssdV);
346         }
347
348         end:
349         
350         freeSwsContext(srcContext);
351         freeSwsContext(dstContext);
352         freeSwsContext(outContext);
353
354         for(i=0; i<3; i++){
355                 free(src[i]);
356                 free(dst[i]);
357                 free(out[i]);
358         }
359 }
360
361 static void selfTest(uint8_t *src[3], int stride[3], int w, int h){
362         int srcFormat, dstFormat, srcFormatIndex, dstFormatIndex;
363         int srcW, srcH, dstW, dstH;
364         int flags;
365
366         for(srcFormatIndex=0; ;srcFormatIndex++){
367                 srcFormat= testFormat[srcFormatIndex];
368                 if(!srcFormat) break;
369                 for(dstFormatIndex=0; ;dstFormatIndex++){
370                         dstFormat= testFormat[dstFormatIndex];
371                         if(!dstFormat) break;
372                         if(!isSupportedOut(dstFormat)) continue;
373 printf("%s -> %s\n", 
374         vo_format_name(srcFormat),
375         vo_format_name(dstFormat));
376
377                         srcW= w+w/3;
378                         srcH= h+h/3;
379                         for(dstW=w; dstW<w*2; dstW+= dstW/3){
380                                 for(dstH=h; dstH<h*2; dstH+= dstH/3){
381                                         for(flags=1; flags<33; flags*=2)
382                                                 doTest(src, stride, w, h, srcFormat, dstFormat,
383                                                         srcW, srcH, dstW, dstH, flags);
384                                 }
385                         }
386                 }
387         }
388 }
389
390 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
391                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
392                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
393 {
394         //FIXME Optimize (just quickly writen not opti..)
395         int i;
396         for(i=0; i<dstW; i++)
397         {
398                 int val=0;
399                 int j;
400                 for(j=0; j<lumFilterSize; j++)
401                         val += lumSrc[j][i] * lumFilter[j];
402
403                 dest[i]= MIN(MAX(val>>19, 0), 255);
404         }
405
406         if(uDest != NULL)
407                 for(i=0; i<chrDstW; i++)
408                 {
409                         int u=0;
410                         int v=0;
411                         int j;
412                         for(j=0; j<chrFilterSize; j++)
413                         {
414                                 u += chrSrc[j][i] * chrFilter[j];
415                                 v += chrSrc[j][i + 2048] * chrFilter[j];
416                         }
417
418                         uDest[i]= MIN(MAX(u>>19, 0), 255);
419                         vDest[i]= MIN(MAX(v>>19, 0), 255);
420                 }
421 }
422
423 #define YSCALE_YUV_2_RGBX_C(type) \
424                 for(i=0; i<(dstW>>1); i++){\
425                         int j;\
426                         int Y1=0;\
427                         int Y2=0;\
428                         int U=0;\
429                         int V=0;\
430                         type *r, *b, *g;\
431                         const int i2= 2*i;\
432                         \
433                         for(j=0; j<lumFilterSize; j++)\
434                         {\
435                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
436                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
437                         }\
438                         for(j=0; j<chrFilterSize; j++)\
439                         {\
440                                 U += chrSrc[j][i] * chrFilter[j];\
441                                 V += chrSrc[j][i+2048] * chrFilter[j];\
442                         }\
443                         Y1>>=19;\
444                         Y2>>=19;\
445                         U >>=19;\
446                         V >>=19;\
447                         if((Y1|Y2|U|V)&256)\
448                         {\
449                                 if(Y1>255)   Y1=255;\
450                                 else if(Y1<0)Y1=0;\
451                                 if(Y2>255)   Y2=255;\
452                                 else if(Y2<0)Y2=0;\
453                                 if(U>255)    U=255;\
454                                 else if(U<0) U=0;\
455                                 if(V>255)    V=255;\
456                                 else if(V<0) V=0;\
457                         }\
458                         r = c->table_rV[V];\
459                         g = c->table_gU[U] + c->table_gV[V];\
460                         b = c->table_bU[U];\
461
462 #define YSCALE_YUV_2_RGB2_C(type) \
463                 for(i=0; i<(dstW>>1); i++){\
464                         const int i2= 2*i;\
465                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
466                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
467                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
468                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
469                         type *r, *b, *g;\
470                         r = c->table_rV[V];\
471                         g = c->table_gU[U] + c->table_gV[V];\
472                         b = c->table_bU[U];\
473
474 #define YSCALE_YUV_2_RGB1_C(type) \
475                 for(i=0; i<(dstW>>1); i++){\
476                         const int i2= 2*i;\
477                         int Y1= buf0[i2  ]>>7;\
478                         int Y2= buf0[i2+1]>>7;\
479                         int U= (uvbuf1[i     ])>>7;\
480                         int V= (uvbuf1[i+2048])>>7;\
481                         type *r, *b, *g;\
482                         r = c->table_rV[V];\
483                         g = c->table_gU[U] + c->table_gV[V];\
484                         b = c->table_bU[U];\
485
486 #define YSCALE_YUV_2_RGB1B_C(type) \
487                 for(i=0; i<(dstW>>1); i++){\
488                         const int i2= 2*i;\
489                         int Y1= buf0[i2  ]>>7;\
490                         int Y2= buf0[i2+1]>>7;\
491                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
492                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
493                         type *r, *b, *g;\
494                         r = c->table_rV[V];\
495                         g = c->table_gU[U] + c->table_gV[V];\
496                         b = c->table_bU[U];\
497
498 #define YSCALE_YUV_2_ANYRGB_C(func)\
499         switch(c->dstFormat)\
500         {\
501         case IMGFMT_BGR32:\
502         case IMGFMT_RGB32:\
503                 func(uint32_t)\
504                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
505                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
506                 }               \
507                 break;\
508         case IMGFMT_RGB24:\
509                 func(uint8_t)\
510                         ((uint8_t*)dest)[0]= r[Y1];\
511                         ((uint8_t*)dest)[1]= g[Y1];\
512                         ((uint8_t*)dest)[2]= b[Y1];\
513                         ((uint8_t*)dest)[3]= r[Y2];\
514                         ((uint8_t*)dest)[4]= g[Y2];\
515                         ((uint8_t*)dest)[5]= b[Y2];\
516                         ((uint8_t*)dest)+=6;\
517                 }\
518                 break;\
519         case IMGFMT_BGR24:\
520                 func(uint8_t)\
521                         ((uint8_t*)dest)[0]= b[Y1];\
522                         ((uint8_t*)dest)[1]= g[Y1];\
523                         ((uint8_t*)dest)[2]= r[Y1];\
524                         ((uint8_t*)dest)[3]= b[Y2];\
525                         ((uint8_t*)dest)[4]= g[Y2];\
526                         ((uint8_t*)dest)[5]= r[Y2];\
527                         ((uint8_t*)dest)+=6;\
528                 }\
529                 break;\
530         case IMGFMT_RGB16:\
531         case IMGFMT_BGR16:\
532                 {\
533                         const int dr1= dither_2x2_8[y&1    ][0];\
534                         const int dg1= dither_2x2_4[y&1    ][0];\
535                         const int db1= dither_2x2_8[(y&1)^1][0];\
536                         const int dr2= dither_2x2_8[y&1    ][1];\
537                         const int dg2= dither_2x2_4[y&1    ][1];\
538                         const int db2= dither_2x2_8[(y&1)^1][1];\
539                         func(uint16_t)\
540                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
541                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
542                         }\
543                 }\
544                 break;\
545         case IMGFMT_RGB15:\
546         case IMGFMT_BGR15:\
547                 {\
548                         const int dr1= dither_2x2_8[y&1    ][0];\
549                         const int dg1= dither_2x2_8[y&1    ][1];\
550                         const int db1= dither_2x2_8[(y&1)^1][0];\
551                         const int dr2= dither_2x2_8[y&1    ][1];\
552                         const int dg2= dither_2x2_8[y&1    ][0];\
553                         const int db2= dither_2x2_8[(y&1)^1][1];\
554                         func(uint16_t)\
555                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
556                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
557                         }\
558                 }\
559                 break;\
560         case IMGFMT_RGB8:\
561         case IMGFMT_BGR8:\
562                 {\
563                         const uint8_t * const d64= dither_8x8_73[y&7];\
564                         const uint8_t * const d32= dither_8x8_32[y&7];\
565                         func(uint8_t)\
566                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
567                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
568                         }\
569                 }\
570                 break;\
571         case IMGFMT_RGB4:\
572         case IMGFMT_BGR4:\
573                 {\
574                         const uint8_t * const d64= dither_8x8_73 [y&7];\
575                         const uint8_t * const d128=dither_8x8_220[y&7];\
576                         func(uint8_t)\
577                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
578                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
579                         }\
580                 }\
581                 break;\
582         case IMGFMT_RGB1:\
583         case IMGFMT_BGR1:\
584                 {\
585                         const uint8_t * const d128=dither_8x8_220[y&7];\
586                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
587                         for(i=0; i<dstW-7; i+=8){\
588                                 int acc;\
589                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
590                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
591                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
592                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
593                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
594                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
595                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
596                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
597                                 ((uint8_t*)dest)[0]= acc;\
598                                 ((uint8_t*)dest)++;\
599                         }\
600 \
601 /*\
602 ((uint8_t*)dest)-= dstW>>4;\
603 {\
604                         int acc=0;\
605                         int left=0;\
606                         static int top[1024];\
607                         static int last_new[1024][1024];\
608                         static int last_in3[1024][1024];\
609                         static int drift[1024][1024];\
610                         int topLeft=0;\
611                         int shift=0;\
612                         int count=0;\
613                         const uint8_t * const d128=dither_8x8_220[y&7];\
614                         int error_new=0;\
615                         int error_in3=0;\
616                         int f=0;\
617                         \
618                         for(i=dstW>>1; i<dstW; i++){\
619                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
620                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
621                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
622                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
623                                         + (last_new[y][i] - in3)*f/256;\
624                                 int new= old> 128 ? 255 : 0;\
625 \
626                                 error_new+= ABS(last_new[y][i] - new);\
627                                 error_in3+= ABS(last_in3[y][i] - in3);\
628                                 f= error_new - error_in3*4;\
629                                 if(f<0) f=0;\
630                                 if(f>256) f=256;\
631 \
632                                 topLeft= top[i];\
633                                 left= top[i]= old - new;\
634                                 last_new[y][i]= new;\
635                                 last_in3[y][i]= in3;\
636 \
637                                 acc+= acc + (new&1);\
638                                 if((i&7)==6){\
639                                         ((uint8_t*)dest)[0]= acc;\
640                                         ((uint8_t*)dest)++;\
641                                 }\
642                         }\
643 }\
644 */\
645                 }\
646                 break;\
647         }\
648
649
650 static inline void yuv2rgbXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
651                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
652                                     uint8_t *dest, int dstW, int y)
653 {
654         int i;
655         switch(c->dstFormat)
656         {
657         case IMGFMT_RGB32:
658         case IMGFMT_BGR32:
659                 YSCALE_YUV_2_RGBX_C(uint32_t)
660                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
661                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
662                 }
663                 break;
664         case IMGFMT_RGB24:
665                 YSCALE_YUV_2_RGBX_C(uint8_t)
666                         ((uint8_t*)dest)[0]= r[Y1];
667                         ((uint8_t*)dest)[1]= g[Y1];
668                         ((uint8_t*)dest)[2]= b[Y1];
669                         ((uint8_t*)dest)[3]= r[Y2];
670                         ((uint8_t*)dest)[4]= g[Y2];
671                         ((uint8_t*)dest)[5]= b[Y2];
672                         ((uint8_t*)dest)+=6;
673                 }
674                 break;
675         case IMGFMT_BGR24:
676                 YSCALE_YUV_2_RGBX_C(uint8_t)
677                         ((uint8_t*)dest)[0]= b[Y1];
678                         ((uint8_t*)dest)[1]= g[Y1];
679                         ((uint8_t*)dest)[2]= r[Y1];
680                         ((uint8_t*)dest)[3]= b[Y2];
681                         ((uint8_t*)dest)[4]= g[Y2];
682                         ((uint8_t*)dest)[5]= r[Y2];
683                         ((uint8_t*)dest)+=6;
684                 }
685                 break;
686         case IMGFMT_RGB16:
687         case IMGFMT_BGR16:
688                 {
689                         const int dr1= dither_2x2_8[y&1    ][0];
690                         const int dg1= dither_2x2_4[y&1    ][0];
691                         const int db1= dither_2x2_8[(y&1)^1][0];
692                         const int dr2= dither_2x2_8[y&1    ][1];
693                         const int dg2= dither_2x2_4[y&1    ][1];
694                         const int db2= dither_2x2_8[(y&1)^1][1];
695                         YSCALE_YUV_2_RGBX_C(uint16_t)
696                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
697                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
698                         }
699                 }
700                 break;
701         case IMGFMT_RGB15:
702         case IMGFMT_BGR15:
703                 {
704                         const int dr1= dither_2x2_8[y&1    ][0];
705                         const int dg1= dither_2x2_8[y&1    ][1];
706                         const int db1= dither_2x2_8[(y&1)^1][0];
707                         const int dr2= dither_2x2_8[y&1    ][1];
708                         const int dg2= dither_2x2_8[y&1    ][0];
709                         const int db2= dither_2x2_8[(y&1)^1][1];
710                         YSCALE_YUV_2_RGBX_C(uint16_t)
711                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
712                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
713                         }
714                 }
715                 break;
716         case IMGFMT_RGB8:
717         case IMGFMT_BGR8:
718                 {
719                         const uint8_t * const d64= dither_8x8_73[y&7];
720                         const uint8_t * const d32= dither_8x8_32[y&7];
721                         YSCALE_YUV_2_RGBX_C(uint8_t)
722                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
723                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
724                         }
725                 }
726                 break;
727         case IMGFMT_RGB4:
728         case IMGFMT_BGR4:
729                 {
730                         const uint8_t * const d64= dither_8x8_73 [y&7];
731                         const uint8_t * const d128=dither_8x8_220[y&7];
732                         YSCALE_YUV_2_RGBX_C(uint8_t)
733                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
734                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
735                         }
736                 }
737                 break;
738         case IMGFMT_RGB1:
739         case IMGFMT_BGR1:
740                 {
741                         const uint8_t * const d128=dither_8x8_220[y&7];
742                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
743                         int acc=0;
744                         for(i=0; i<dstW-1; i+=2){
745                                 int j;
746                                 int Y1=0;
747                                 int Y2=0;
748
749                                 for(j=0; j<lumFilterSize; j++)
750                                 {
751                                         Y1 += lumSrc[j][i] * lumFilter[j];
752                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
753                                 }
754                                 Y1>>=19;
755                                 Y2>>=19;
756                                 if((Y1|Y2)&256)
757                                 {
758                                         if(Y1>255)   Y1=255;
759                                         else if(Y1<0)Y1=0;
760                                         if(Y2>255)   Y2=255;
761                                         else if(Y2<0)Y2=0;
762                                 }
763                                 acc+= acc + g[Y1+d128[(i+0)&7]];
764                                 acc+= acc + g[Y2+d128[(i+1)&7]];
765                                 if((i&7)==6){
766                                         ((uint8_t*)dest)[0]= acc;
767                                         ((uint8_t*)dest)++;
768                                 }
769                         }
770                 }
771                 break;
772         }
773 }
774
775
776 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
777 //Plain C versions
778 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
779 #define COMPILE_C
780 #endif
781
782 #ifdef CAN_COMPILE_X86_ASM
783
784 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
785 #define COMPILE_MMX
786 #endif
787
788 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
789 #define COMPILE_MMX2
790 #endif
791
792 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
793 #define COMPILE_3DNOW
794 #endif
795 #endif //CAN_COMPILE_X86_ASM
796
797 #undef HAVE_MMX
798 #undef HAVE_MMX2
799 #undef HAVE_3DNOW
800
801 #ifdef COMPILE_C
802 #undef HAVE_MMX
803 #undef HAVE_MMX2
804 #undef HAVE_3DNOW
805 #define RENAME(a) a ## _C
806 #include "swscale_template.c"
807 #endif
808
809 #ifdef CAN_COMPILE_X86_ASM
810
811 //X86 versions
812 /*
813 #undef RENAME
814 #undef HAVE_MMX
815 #undef HAVE_MMX2
816 #undef HAVE_3DNOW
817 #define ARCH_X86
818 #define RENAME(a) a ## _X86
819 #include "swscale_template.c"
820 */
821 //MMX versions
822 #ifdef COMPILE_MMX
823 #undef RENAME
824 #define HAVE_MMX
825 #undef HAVE_MMX2
826 #undef HAVE_3DNOW
827 #define RENAME(a) a ## _MMX
828 #include "swscale_template.c"
829 #endif
830
831 //MMX2 versions
832 #ifdef COMPILE_MMX2
833 #undef RENAME
834 #define HAVE_MMX
835 #define HAVE_MMX2
836 #undef HAVE_3DNOW
837 #define RENAME(a) a ## _MMX2
838 #include "swscale_template.c"
839 #endif
840
841 //3DNOW versions
842 #ifdef COMPILE_3DNOW
843 #undef RENAME
844 #define HAVE_MMX
845 #undef HAVE_MMX2
846 #define HAVE_3DNOW
847 #define RENAME(a) a ## _3DNow
848 #include "swscale_template.c"
849 #endif
850
851 #endif //CAN_COMPILE_X86_ASM
852
853 // minor note: the HAVE_xyz is messed up after that line so dont use it
854
855
856 // old global scaler, dont use for new code
857 // will use sws_flags from the command line
858 void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
859                              int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
860                              int srcW, int srcH, int dstW, int dstH){
861
862         static SwsContext *context=NULL;
863         int dstFormat;
864         int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
865
866         switch(dstbpp)
867         {
868                 case 8 : dstFormat= IMGFMT_Y8;          break;
869                 case 12: dstFormat= IMGFMT_YV12;        break;
870                 case 15: dstFormat= IMGFMT_BGR15;       break;
871                 case 16: dstFormat= IMGFMT_BGR16;       break;
872                 case 24: dstFormat= IMGFMT_BGR24;       break;
873                 case 32: dstFormat= IMGFMT_BGR32;       break;
874                 default: return;
875         }
876
877         if(!context) context=getSwsContextFromCmdLine(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat);
878
879         context->swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
880 }
881
882 void swsGetFlagsAndFilterFromCmdLine(int *flags, SwsFilter **srcFilterParam, SwsFilter **dstFilterParam)
883 {
884         static int firstTime=1;
885         *flags=0;
886
887 #ifdef ARCH_X86
888         if(gCpuCaps.hasMMX)
889                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
890 #endif
891         if(firstTime)
892         {
893                 firstTime=0;
894                 *flags= SWS_PRINT_INFO;
895         }
896         else if(verbose>1) *flags= SWS_PRINT_INFO;
897
898         if(src_filter.lumH) freeVec(src_filter.lumH);
899         if(src_filter.lumV) freeVec(src_filter.lumV);
900         if(src_filter.chrH) freeVec(src_filter.chrH);
901         if(src_filter.chrV) freeVec(src_filter.chrV);
902
903         if(sws_lum_gblur!=0.0){
904                 src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);
905                 src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);
906         }else{
907                 src_filter.lumH= getIdentityVec();
908                 src_filter.lumV= getIdentityVec();
909         }
910
911         if(sws_chr_gblur!=0.0){
912                 src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);
913                 src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);
914         }else{
915                 src_filter.chrH= getIdentityVec();
916                 src_filter.chrV= getIdentityVec();
917         }
918
919         if(sws_chr_sharpen!=0.0){
920                 SwsVector *g= getConstVec(-1.0, 3);
921                 SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
922                 g->coeff[1]=2.0;
923                 addVec(id, g);
924                 convVec(src_filter.chrH, id);
925                 convVec(src_filter.chrV, id);
926                 freeVec(g);
927                 freeVec(id);
928         }
929
930         if(sws_lum_sharpen!=0.0){
931                 SwsVector *g= getConstVec(-1.0, 3);
932                 SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
933                 g->coeff[1]=2.0;
934                 addVec(id, g);
935                 convVec(src_filter.lumH, id);
936                 convVec(src_filter.lumV, id);
937                 freeVec(g);
938                 freeVec(id);
939         }
940
941         if(sws_chr_hshift)
942                 shiftVec(src_filter.chrH, sws_chr_hshift);
943
944         if(sws_chr_vshift)
945                 shiftVec(src_filter.chrV, sws_chr_vshift);
946
947         normalizeVec(src_filter.chrH, 1.0);
948         normalizeVec(src_filter.chrV, 1.0);
949         normalizeVec(src_filter.lumH, 1.0);
950         normalizeVec(src_filter.lumV, 1.0);
951
952         if(verbose > 1) printVec(src_filter.chrH);
953         if(verbose > 1) printVec(src_filter.lumH);
954
955         switch(sws_flags)
956         {
957                 case 0: *flags|= SWS_FAST_BILINEAR; break;
958                 case 1: *flags|= SWS_BILINEAR; break;
959                 case 2: *flags|= SWS_BICUBIC; break;
960                 case 3: *flags|= SWS_X; break;
961                 case 4: *flags|= SWS_POINT; break;
962                 case 5: *flags|= SWS_AREA; break;
963                 case 6: *flags|= SWS_BICUBLIN; break;
964                 case 7: *flags|= SWS_GAUSS; break;
965                 case 8: *flags|= SWS_SINC; break;
966                 case 9: *flags|= SWS_LANCZOS; break;
967                 case 10:*flags|= SWS_SPLINE; break;
968                 default:*flags|= SWS_BILINEAR; break;
969         }
970         
971         *srcFilterParam= &src_filter;
972         *dstFilterParam= NULL;
973 }
974
975 // will use sws_flags & src_filter (from cmd line)
976 SwsContext *getSwsContextFromCmdLine(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat)
977 {
978         int flags;
979         SwsFilter *dstFilterParam, *srcFilterParam;
980         swsGetFlagsAndFilterFromCmdLine(&flags, &srcFilterParam, &dstFilterParam);
981
982         return getSwsContext(srcW, srcH, srcFormat, dstW, dstH, dstFormat, flags, srcFilterParam, dstFilterParam);
983 }
984
985 static double getSplineCoeff(double a, double b, double c, double d, double dist)
986 {
987 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
988         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
989         else            return getSplineCoeff(  0.0, 
990                                                  b+ 2.0*c + 3.0*d,
991                                                         c + 3.0*d,
992                                                 -b- 3.0*c - 6.0*d,
993                                                 dist-1.0);
994 }
995
996 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
997                               int srcW, int dstW, int filterAlign, int one, int flags,
998                               SwsVector *srcFilter, SwsVector *dstFilter)
999 {
1000         int i;
1001         int filterSize;
1002         int filter2Size;
1003         int minFilterSize;
1004         double *filter=NULL;
1005         double *filter2=NULL;
1006 #ifdef ARCH_X86
1007         if(gCpuCaps.hasMMX)
1008                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
1009 #endif
1010
1011         // Note the +1 is for the MMXscaler which reads over the end
1012         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
1013
1014         if(ABS(xInc - 0x10000) <10) // unscaled
1015         {
1016                 int i;
1017                 filterSize= 1;
1018                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1019                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
1020
1021                 for(i=0; i<dstW; i++)
1022                 {
1023                         filter[i*filterSize]=1;
1024                         (*filterPos)[i]=i;
1025                 }
1026
1027         }
1028         else if(flags&SWS_POINT) // lame looking point sampling mode
1029         {
1030                 int i;
1031                 int xDstInSrc;
1032                 filterSize= 1;
1033                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1034                 
1035                 xDstInSrc= xInc/2 - 0x8000;
1036                 for(i=0; i<dstW; i++)
1037                 {
1038                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1039
1040                         (*filterPos)[i]= xx;
1041                         filter[i]= 1.0;
1042                         xDstInSrc+= xInc;
1043                 }
1044         }
1045         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1046         {
1047                 int i;
1048                 int xDstInSrc;
1049                 if     (flags&SWS_BICUBIC) filterSize= 4;
1050                 else if(flags&SWS_X      ) filterSize= 4;
1051                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
1052                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
1053
1054                 xDstInSrc= xInc/2 - 0x8000;
1055                 for(i=0; i<dstW; i++)
1056                 {
1057                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1058                         int j;
1059
1060                         (*filterPos)[i]= xx;
1061                                 //Bilinear upscale / linear interpolate / Area averaging
1062                                 for(j=0; j<filterSize; j++)
1063                                 {
1064                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1065                                         double coeff= 1.0 - d;
1066                                         if(coeff<0) coeff=0;
1067                                         filter[i*filterSize + j]= coeff;
1068                                         xx++;
1069                                 }
1070                         xDstInSrc+= xInc;
1071                 }
1072         }
1073         else
1074         {
1075                 double xDstInSrc;
1076                 double sizeFactor, filterSizeInSrc;
1077                 const double xInc1= (double)xInc / (double)(1<<16);
1078                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
1079
1080                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
1081                 else if(flags&SWS_X)            sizeFactor= 8.0;
1082                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1083                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
1084                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
1085                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
1086                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
1087                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
1088                 else {
1089                         sizeFactor= 0.0; //GCC warning killer
1090                         ASSERT(0)
1091                 }
1092                 
1093                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
1094                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1095
1096                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1097                 if(filterSize > srcW-2) filterSize=srcW-2;
1098
1099                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
1100
1101                 xDstInSrc= xInc1 / 2.0 - 0.5;
1102                 for(i=0; i<dstW; i++)
1103                 {
1104                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1105                         int j;
1106                         (*filterPos)[i]= xx;
1107                         for(j=0; j<filterSize; j++)
1108                         {
1109                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1110                                 double coeff;
1111                                 if(flags & SWS_BICUBIC)
1112                                 {
1113                                         double A= param ? -param*0.01 : -0.60;
1114                                         
1115                                         // Equation is from VirtualDub
1116                                         if(d<1.0)
1117                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
1118                                         else if(d<2.0)
1119                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
1120                                         else
1121                                                 coeff=0.0;
1122                                 }
1123 /*                              else if(flags & SWS_X)
1124                                 {
1125                                         double p= param ? param*0.01 : 0.3;
1126                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1127                                         coeff*= pow(2.0, - p*d*d);
1128                                 }*/
1129                                 else if(flags & SWS_X)
1130                                 {
1131                                         double A= param ? param*0.1 : 1.0;
1132                                         
1133                                         if(d<1.0)
1134                                                 coeff = cos(d*PI);
1135                                         else
1136                                                 coeff=-1.0;
1137                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
1138                                         else            coeff=  pow( coeff, A);
1139                                         coeff= coeff*0.5 + 0.5;
1140                                 }
1141                                 else if(flags & SWS_AREA)
1142                                 {
1143                                         double srcPixelSize= 1.0/xInc1;
1144                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1145                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1146                                         else coeff=0.0;
1147                                 }
1148                                 else if(flags & SWS_GAUSS)
1149                                 {
1150                                         double p= param ? param*0.1 : 3.0;
1151                                         coeff = pow(2.0, - p*d*d);
1152                                 }
1153                                 else if(flags & SWS_SINC)
1154                                 {
1155                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1156                                 }
1157                                 else if(flags & SWS_LANCZOS)
1158                                 {
1159                                         double p= param ? param : 3.0; 
1160                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1161                                         if(d>p) coeff=0;
1162                                 }
1163                                 else if(flags & SWS_BILINEAR)
1164                                 {
1165                                         coeff= 1.0 - d;
1166                                         if(coeff<0) coeff=0;
1167                                 }
1168                                 else if(flags & SWS_SPLINE)
1169                                 {
1170                                         double p=-2.196152422706632;
1171                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1172                                 }
1173                                 else {
1174                                         coeff= 0.0; //GCC warning killer
1175                                         ASSERT(0)
1176                                 }
1177
1178                                 filter[i*filterSize + j]= coeff;
1179                                 xx++;
1180                         }
1181                         xDstInSrc+= xInc1;
1182                 }
1183         }
1184
1185         /* apply src & dst Filter to filter -> filter2
1186            free(filter);
1187         */
1188         ASSERT(filterSize>0)
1189         filter2Size= filterSize;
1190         if(srcFilter) filter2Size+= srcFilter->length - 1;
1191         if(dstFilter) filter2Size+= dstFilter->length - 1;
1192         ASSERT(filter2Size>0)
1193         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1194
1195         for(i=0; i<dstW; i++)
1196         {
1197                 int j;
1198                 SwsVector scaleFilter;
1199                 SwsVector *outVec;
1200
1201                 scaleFilter.coeff= filter + i*filterSize;
1202                 scaleFilter.length= filterSize;
1203
1204                 if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
1205                 else          outVec= &scaleFilter;
1206
1207                 ASSERT(outVec->length == filter2Size)
1208                 //FIXME dstFilter
1209
1210                 for(j=0; j<outVec->length; j++)
1211                 {
1212                         filter2[i*filter2Size + j]= outVec->coeff[j];
1213                 }
1214
1215                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1216
1217                 if(outVec != &scaleFilter) freeVec(outVec);
1218         }
1219         free(filter); filter=NULL;
1220
1221         /* try to reduce the filter-size (step1 find size and shift left) */
1222         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1223         minFilterSize= 0;
1224         for(i=dstW-1; i>=0; i--)
1225         {
1226                 int min= filter2Size;
1227                 int j;
1228                 double cutOff=0.0;
1229
1230                 /* get rid off near zero elements on the left by shifting left */
1231                 for(j=0; j<filter2Size; j++)
1232                 {
1233                         int k;
1234                         cutOff += ABS(filter2[i*filter2Size]);
1235
1236                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1237
1238                         /* preserve Monotonicity because the core cant handle the filter otherwise */
1239                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1240
1241                         // Move filter coeffs left
1242                         for(k=1; k<filter2Size; k++)
1243                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1244                         filter2[i*filter2Size + k - 1]= 0.0;
1245                         (*filterPos)[i]++;
1246                 }
1247
1248                 cutOff=0.0;
1249                 /* count near zeros on the right */
1250                 for(j=filter2Size-1; j>0; j--)
1251                 {
1252                         cutOff += ABS(filter2[i*filter2Size + j]);
1253
1254                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1255                         min--;
1256                 }
1257
1258                 if(min>minFilterSize) minFilterSize= min;
1259         }
1260
1261         ASSERT(minFilterSize > 0)
1262         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1263         ASSERT(filterSize > 0)
1264         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1265         *outFilterSize= filterSize;
1266
1267         if(flags&SWS_PRINT_INFO)
1268                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1269         /* try to reduce the filter-size (step2 reduce it) */
1270         for(i=0; i<dstW; i++)
1271         {
1272                 int j;
1273
1274                 for(j=0; j<filterSize; j++)
1275                 {
1276                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1277                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1278                 }
1279         }
1280         free(filter2); filter2=NULL;
1281         
1282
1283         //FIXME try to align filterpos if possible
1284
1285         //fix borders
1286         for(i=0; i<dstW; i++)
1287         {
1288                 int j;
1289                 if((*filterPos)[i] < 0)
1290                 {
1291                         // Move filter coeffs left to compensate for filterPos
1292                         for(j=1; j<filterSize; j++)
1293                         {
1294                                 int left= MAX(j + (*filterPos)[i], 0);
1295                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1296                                 filter[i*filterSize + j]=0;
1297                         }
1298                         (*filterPos)[i]= 0;
1299                 }
1300
1301                 if((*filterPos)[i] + filterSize > srcW)
1302                 {
1303                         int shift= (*filterPos)[i] + filterSize - srcW;
1304                         // Move filter coeffs right to compensate for filterPos
1305                         for(j=filterSize-2; j>=0; j--)
1306                         {
1307                                 int right= MIN(j + shift, filterSize-1);
1308                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1309                                 filter[i*filterSize +j]=0;
1310                         }
1311                         (*filterPos)[i]= srcW - filterSize;
1312                 }
1313         }
1314
1315         // Note the +1 is for the MMXscaler which reads over the end
1316         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1317         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1318
1319         /* Normalize & Store in outFilter */
1320         for(i=0; i<dstW; i++)
1321         {
1322                 int j;
1323                 double sum=0;
1324                 double scale= one;
1325                 for(j=0; j<filterSize; j++)
1326                 {
1327                         sum+= filter[i*filterSize + j];
1328                 }
1329                 scale/= sum;
1330                 for(j=0; j<*outFilterSize; j++)
1331                 {
1332                         (*outFilter)[i*(*outFilterSize) + j]= (int)(filter[i*filterSize + j]*scale);
1333                 }
1334         }
1335         
1336         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1337         for(i=0; i<*outFilterSize; i++)
1338         {
1339                 int j= dstW*(*outFilterSize);
1340                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1341         }
1342
1343         free(filter);
1344 }
1345
1346 #ifdef ARCH_X86
1347 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1348 {
1349         uint8_t *fragmentA;
1350         int imm8OfPShufW1A;
1351         int imm8OfPShufW2A;
1352         int fragmentLengthA;
1353         uint8_t *fragmentB;
1354         int imm8OfPShufW1B;
1355         int imm8OfPShufW2B;
1356         int fragmentLengthB;
1357         int fragmentPos;
1358
1359         int xpos, i;
1360
1361         // create an optimized horizontal scaling routine
1362
1363         //code fragment
1364
1365         asm volatile(
1366                 "jmp 9f                         \n\t"
1367         // Begin
1368                 "0:                             \n\t"
1369                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1370                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1371                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1372                 "punpcklbw %%mm7, %%mm1         \n\t"
1373                 "punpcklbw %%mm7, %%mm0         \n\t"
1374                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1375                 "1:                             \n\t"
1376                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1377                 "2:                             \n\t"
1378                 "psubw %%mm1, %%mm0             \n\t"
1379                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1380                 "pmullw %%mm3, %%mm0            \n\t"
1381                 "psllw $7, %%mm1                \n\t"
1382                 "paddw %%mm1, %%mm0             \n\t"
1383
1384                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1385
1386                 "addl $8, %%eax                 \n\t"
1387         // End
1388                 "9:                             \n\t"
1389 //              "int $3\n\t"
1390                 "leal 0b, %0                    \n\t"
1391                 "leal 1b, %1                    \n\t"
1392                 "leal 2b, %2                    \n\t"
1393                 "decl %1                        \n\t"
1394                 "decl %2                        \n\t"
1395                 "subl %0, %1                    \n\t"
1396                 "subl %0, %2                    \n\t"
1397                 "leal 9b, %3                    \n\t"
1398                 "subl %0, %3                    \n\t"
1399
1400
1401                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1402                 "=r" (fragmentLengthA)
1403         );
1404
1405         asm volatile(
1406                 "jmp 9f                         \n\t"
1407         // Begin
1408                 "0:                             \n\t"
1409                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1410                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1411                 "punpcklbw %%mm7, %%mm0         \n\t"
1412                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1413                 "1:                             \n\t"
1414                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1415                 "2:                             \n\t"
1416                 "psubw %%mm1, %%mm0             \n\t"
1417                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1418                 "pmullw %%mm3, %%mm0            \n\t"
1419                 "psllw $7, %%mm1                \n\t"
1420                 "paddw %%mm1, %%mm0             \n\t"
1421
1422                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1423
1424                 "addl $8, %%eax                 \n\t"
1425         // End
1426                 "9:                             \n\t"
1427 //              "int $3\n\t"
1428                 "leal 0b, %0                    \n\t"
1429                 "leal 1b, %1                    \n\t"
1430                 "leal 2b, %2                    \n\t"
1431                 "decl %1                        \n\t"
1432                 "decl %2                        \n\t"
1433                 "subl %0, %1                    \n\t"
1434                 "subl %0, %2                    \n\t"
1435                 "leal 9b, %3                    \n\t"
1436                 "subl %0, %3                    \n\t"
1437
1438
1439                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1440                 "=r" (fragmentLengthB)
1441         );
1442
1443         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1444         fragmentPos=0;
1445         
1446         for(i=0; i<dstW/numSplits; i++)
1447         {
1448                 int xx=xpos>>16;
1449
1450                 if((i&3) == 0)
1451                 {
1452                         int a=0;
1453                         int b=((xpos+xInc)>>16) - xx;
1454                         int c=((xpos+xInc*2)>>16) - xx;
1455                         int d=((xpos+xInc*3)>>16) - xx;
1456
1457                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1458                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1459                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1460                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1461                         filterPos[i/2]= xx;
1462
1463                         if(d+1<4)
1464                         {
1465                                 int maxShift= 3-(d+1);
1466                                 int shift=0;
1467
1468                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1469
1470                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1471                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1472                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1473                                         a | (b<<2) | (c<<4) | (d<<6);
1474
1475                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1476                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1477
1478                                 if(shift && i>=shift)
1479                                 {
1480                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1481                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1482                                         filterPos[i/2]-=shift;
1483                                 }
1484
1485                                 fragmentPos+= fragmentLengthB;
1486                         }
1487                         else
1488                         {
1489                                 int maxShift= 3-d;
1490                                 int shift=0;
1491
1492                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1493
1494                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1495                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1496                                         a | (b<<2) | (c<<4) | (d<<6);
1497
1498                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1499                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1500
1501                                 if(shift && i>=shift)
1502                                 {
1503                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1504                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1505                                         filterPos[i/2]-=shift;
1506                                 }
1507
1508                                 fragmentPos+= fragmentLengthA;
1509                         }
1510
1511                         funnyCode[fragmentPos]= RET;
1512                 }
1513                 xpos+=xInc;
1514         }
1515         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1516 }
1517 #endif // ARCH_X86
1518
1519 //FIXME remove
1520 void SwScale_Init(){
1521 }
1522
1523 static void globalInit(){
1524     // generating tables:
1525     int i;
1526     for(i=0; i<768; i++){
1527         int c= MIN(MAX(i-256, 0), 255);
1528         clip_table[i]=c;
1529     }
1530
1531 cpuCaps= gCpuCaps;
1532
1533 #ifdef RUNTIME_CPUDETECT
1534 #ifdef CAN_COMPILE_X86_ASM
1535         // ordered per speed fasterst first
1536         if(gCpuCaps.hasMMX2)
1537                 swScale= swScale_MMX2;
1538         else if(gCpuCaps.has3DNow)
1539                 swScale= swScale_3DNow;
1540         else if(gCpuCaps.hasMMX)
1541                 swScale= swScale_MMX;
1542         else
1543                 swScale= swScale_C;
1544
1545 #else
1546         swScale= swScale_C;
1547         cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1548 #endif
1549 #else //RUNTIME_CPUDETECT
1550 #ifdef HAVE_MMX2
1551         swScale= swScale_MMX2;
1552         cpuCaps.has3DNow = 0;
1553 #elif defined (HAVE_3DNOW)
1554         swScale= swScale_3DNow;
1555         cpuCaps.hasMMX2 = 0;
1556 #elif defined (HAVE_MMX)
1557         swScale= swScale_MMX;
1558         cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
1559 #else
1560         swScale= swScale_C;
1561         cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
1562 #endif
1563 #endif //!RUNTIME_CPUDETECT
1564 }
1565
1566 static void PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1567              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1568         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1569         /* Copy Y plane */
1570         if(dstStride[0]==srcStride[0])
1571                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1572         else
1573         {
1574                 int i;
1575                 uint8_t *srcPtr= src[0];
1576                 uint8_t *dstPtr= dst;
1577                 for(i=0; i<srcSliceH; i++)
1578                 {
1579                         memcpy(dstPtr, srcPtr, srcStride[0]);
1580                         srcPtr+= srcStride[0];
1581                         dstPtr+= dstStride[0];
1582                 }
1583         }
1584         dst = dstParam[1] + dstStride[1]*srcSliceY;
1585         if(c->srcFormat==IMGFMT_YV12)
1586                 interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1587         else /* I420 & IYUV */
1588                 interleaveBytes( src[2],src[1],dst,c->srcW,srcSliceH,srcStride[2],srcStride[1],dstStride[0] );
1589 }
1590
1591
1592 /* Warper functions for yuv2bgr */
1593 static void planarYuvToBgr(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1594              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1595         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1596
1597         if(c->srcFormat==IMGFMT_YV12)
1598                 yuv2rgb( dst,src[0],src[1],src[2],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1599         else /* I420 & IYUV */
1600                 yuv2rgb( dst,src[0],src[2],src[1],c->srcW,srcSliceH,dstStride[0],srcStride[0],srcStride[1] );
1601 }
1602
1603 static void PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1604              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1605         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1606
1607         if(c->srcFormat==IMGFMT_YV12)
1608                 yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1609         else /* I420 & IYUV */
1610                 yv12toyuy2( src[0],src[2],src[1],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1611 }
1612
1613 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1614 static void rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1615                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1616         const int srcFormat= c->srcFormat;
1617         const int dstFormat= c->dstFormat;
1618         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1619         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1620         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1621         const int dstId= (dstFormat&0xFF)>>2;
1622         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1623
1624         /* BGR -> BGR */
1625         if(isBGR(srcFormat) && isBGR(dstFormat)){
1626                 switch(srcId | (dstId<<4)){
1627                 case 0x34: conv= rgb16to15; break;
1628                 case 0x36: conv= rgb24to15; break;
1629                 case 0x38: conv= rgb32to15; break;
1630                 case 0x43: conv= rgb15to16; break;
1631                 case 0x46: conv= rgb24to16; break;
1632                 case 0x48: conv= rgb32to16; break;
1633                 case 0x63: conv= rgb15to24; break;
1634                 case 0x64: conv= rgb16to24; break;
1635                 case 0x68: conv= rgb32to24; break;
1636                 case 0x83: conv= rgb15to32; break;
1637                 case 0x84: conv= rgb16to32; break;
1638                 case 0x86: conv= rgb24to32; break;
1639                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1640                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1641                 }
1642         }else if(isBGR(srcFormat) && isRGB(dstFormat)){
1643                 switch(srcId | (dstId<<4)){
1644                 case 0x33: conv= rgb15tobgr15; break;
1645                 case 0x34: conv= rgb16tobgr15; break;
1646                 case 0x36: conv= rgb24tobgr15; break;
1647                 case 0x38: conv= rgb32tobgr15; break;
1648                 case 0x43: conv= rgb15tobgr16; break;
1649                 case 0x44: conv= rgb16tobgr16; break;
1650                 case 0x46: conv= rgb24tobgr16; break;
1651                 case 0x48: conv= rgb32tobgr16; break;
1652                 case 0x63: conv= rgb15tobgr24; break;
1653                 case 0x64: conv= rgb16tobgr24; break;
1654                 case 0x66: conv= rgb24tobgr24; break;
1655                 case 0x68: conv= rgb32tobgr24; break;
1656                 case 0x83: conv= rgb15tobgr32; break;
1657                 case 0x84: conv= rgb16tobgr32; break;
1658                 case 0x86: conv= rgb24tobgr32; break;
1659                 case 0x88: conv= rgb32tobgr32; break;
1660                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1661                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1662                 }
1663         }else if(isRGB(srcFormat) && isRGB(dstFormat)){
1664                 switch(srcId | (dstId<<4)){
1665                 case 0x34: conv= rgb16to15; break;
1666                 case 0x36: conv= rgb24to15; break;
1667                 case 0x38: conv= rgb32to15; break;
1668                 case 0x43: conv= rgb15to16; break;
1669                 case 0x46: conv= rgb24to16; break;
1670                 case 0x48: conv= rgb32to16; break;
1671                 case 0x63: conv= rgb15to24; break;
1672                 case 0x64: conv= rgb16to24; break;
1673                 case 0x68: conv= rgb32to24; break;
1674                 case 0x83: conv= rgb15to32; break;
1675                 case 0x84: conv= rgb16to32; break;
1676                 case 0x86: conv= rgb24to32; break;
1677                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1678                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1679                 }
1680         }else if(isRGB(srcFormat) && isBGR(dstFormat)){
1681                 switch(srcId | (dstId<<4)){
1682                 case 0x33: conv= rgb15tobgr15; break;
1683                 case 0x34: conv= rgb16tobgr15; break;
1684                 case 0x36: conv= rgb24tobgr15; break;
1685                 case 0x38: conv= rgb32tobgr15; break;
1686                 case 0x43: conv= rgb15tobgr16; break;
1687                 case 0x44: conv= rgb16tobgr16; break;
1688                 case 0x46: conv= rgb24tobgr16; break;
1689                 case 0x48: conv= rgb32tobgr16; break;
1690                 case 0x63: conv= rgb15tobgr24; break;
1691                 case 0x64: conv= rgb16tobgr24; break;
1692                 case 0x66: conv= rgb24tobgr24; break;
1693                 case 0x68: conv= rgb32tobgr24; break;
1694                 case 0x83: conv= rgb15tobgr32; break;
1695                 case 0x84: conv= rgb16tobgr32; break;
1696                 case 0x86: conv= rgb24tobgr32; break;
1697                 case 0x88: conv= rgb32tobgr32; break;
1698                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1699                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1700                 }
1701         }
1702         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1703                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1704         else
1705         {
1706                 int i;
1707                 uint8_t *srcPtr= src[0];
1708                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1709
1710                 for(i=0; i<srcSliceH; i++)
1711                 {
1712                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1713                         srcPtr+= srcStride[0];
1714                         dstPtr+= dstStride[0];
1715                 }
1716         }     
1717 }
1718
1719 static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1720              int srcSliceH, uint8_t* dst[], int dstStride[]){
1721
1722         rgb24toyv12(
1723                 src[0], 
1724                 dst[0]+ srcSliceY    *dstStride[0], 
1725                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1726                 dst[2]+(srcSliceY>>1)*dstStride[2],
1727                 c->srcW, srcSliceH, 
1728                 dstStride[0], dstStride[1], srcStride[0]);
1729 }
1730
1731 static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1732              int srcSliceH, uint8_t* dst[], int dstStride[]){
1733         int i;
1734
1735         /* copy Y */
1736         if(srcStride[0]==dstStride[0]) 
1737                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1738         else{
1739                 uint8_t *srcPtr= src[0];
1740                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1741
1742                 for(i=0; i<srcSliceH; i++)
1743                 {
1744                         memcpy(dstPtr, srcPtr, c->srcW);
1745                         srcPtr+= srcStride[0];
1746                         dstPtr+= dstStride[0];
1747                 }
1748         }
1749
1750         if(c->dstFormat==IMGFMT_YV12){
1751                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1752                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1753         }else{
1754                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1755                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1756         }
1757 }
1758
1759 /**
1760  * bring pointers in YUV order instead of YVU
1761  */
1762 static inline void orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1763         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9 
1764            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1765                 sortedP[0]= p[0];
1766                 sortedP[1]= p[1];
1767                 sortedP[2]= p[2];
1768                 sortedStride[0]= stride[0];
1769                 sortedStride[1]= stride[1];
1770                 sortedStride[2]= stride[2];
1771         }
1772         else if(isPacked(format) || isGray(format))
1773         {
1774                 sortedP[0]= p[0];
1775                 sortedP[1]= 
1776                 sortedP[2]= NULL;
1777                 sortedStride[0]= stride[0];
1778                 sortedStride[1]= 
1779                 sortedStride[2]= 0;
1780         }
1781         else if(format == IMGFMT_I420)
1782         {
1783                 sortedP[0]= p[0];
1784                 sortedP[1]= p[2];
1785                 sortedP[2]= p[1];
1786                 sortedStride[0]= stride[0];
1787                 sortedStride[1]= stride[2];
1788                 sortedStride[2]= stride[1];
1789         }else{
1790                 MSG_ERR("internal error in orderYUV\n");
1791         }
1792 }
1793
1794 /* unscaled copy like stuff (assumes nearly identical formats) */
1795 static void simpleCopy(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
1796              int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
1797
1798         int srcStride[3];
1799         int dstStride[3];
1800         uint8_t *src[3];
1801         uint8_t *dst[3];
1802
1803         orderYUV(c->srcFormat, src, srcStride, srcParam, srcStrideParam);
1804         orderYUV(c->dstFormat, dst, dstStride, dstParam, dstStrideParam);
1805
1806         if(isPacked(c->srcFormat))
1807         {
1808                 if(dstStride[0]==srcStride[0])
1809                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1810                 else
1811                 {
1812                         int i;
1813                         uint8_t *srcPtr= src[0];
1814                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1815                         int length=0;
1816
1817                         /* universal length finder */
1818                         while(length+c->srcW <= ABS(dstStride[0]) 
1819                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1820                         ASSERT(length!=0);
1821
1822                         for(i=0; i<srcSliceH; i++)
1823                         {
1824                                 memcpy(dstPtr, srcPtr, length);
1825                                 srcPtr+= srcStride[0];
1826                                 dstPtr+= dstStride[0];
1827                         }
1828                 }
1829         }
1830         else 
1831         { /* Planar YUV or gray */
1832                 int plane;
1833                 for(plane=0; plane<3; plane++)
1834                 {
1835                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1836                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1837                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1838
1839                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1840                         {
1841                                 if(!isGray(c->dstFormat))
1842                                         memset(dst[plane], 128, dstStride[plane]*height);
1843                         }
1844                         else
1845                         {
1846                                 if(dstStride[plane]==srcStride[plane])
1847                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1848                                 else
1849                                 {
1850                                         int i;
1851                                         uint8_t *srcPtr= src[plane];
1852                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1853                                         for(i=0; i<height; i++)
1854                                         {
1855                                                 memcpy(dstPtr, srcPtr, length);
1856                                                 srcPtr+= srcStride[plane];
1857                                                 dstPtr+= dstStride[plane];
1858                                         }
1859                                 }
1860                         }
1861                 }
1862         }
1863 }
1864
1865 static int remove_dup_fourcc(int fourcc)
1866 {
1867         switch(fourcc)
1868         {
1869             case IMGFMT_IYUV: return IMGFMT_I420;
1870             case IMGFMT_Y8  : return IMGFMT_Y800;
1871             case IMGFMT_IF09: return IMGFMT_YVU9;
1872             default: return fourcc;
1873         }
1874 }
1875
1876 static void getSubSampleFactors(int *h, int *v, int format){
1877         switch(format){
1878         case IMGFMT_YUY2:
1879                 *h=1;
1880                 *v=0;
1881                 break;
1882         case IMGFMT_YV12:
1883         case IMGFMT_I420:
1884         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1885                 *h=1;
1886                 *v=1;
1887                 break;
1888         case IMGFMT_YVU9:
1889                 *h=2;
1890                 *v=2;
1891                 break;
1892         case IMGFMT_444P:
1893                 *h=0;
1894                 *v=0;
1895                 break;
1896         case IMGFMT_422P:
1897                 *h=1;
1898                 *v=0;
1899                 break;
1900         case IMGFMT_411P:
1901                 *h=2;
1902                 *v=0;
1903                 break;
1904         default:
1905                 *h=0;
1906                 *v=0;
1907                 break;
1908         }
1909 }
1910
1911 SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1912                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1913
1914         SwsContext *c;
1915         int i;
1916         int usesFilter;
1917         int unscaled, needsDither;
1918         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1919 #ifdef ARCH_X86
1920         if(gCpuCaps.hasMMX)
1921                 asm volatile("emms\n\t"::: "memory");
1922 #endif
1923         if(swScale==NULL) globalInit();
1924 //srcFormat= IMGFMT_Y800;
1925 //dstFormat= IMGFMT_Y800;
1926         /* avoid dupplicate Formats, so we dont need to check to much */
1927         srcFormat = remove_dup_fourcc(srcFormat);
1928         dstFormat = remove_dup_fourcc(dstFormat);
1929
1930         unscaled = (srcW == dstW && srcH == dstH);
1931         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1932                      && (dstFormat&0xFF)<24
1933                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1934
1935         if(!isSupportedIn(srcFormat)) 
1936         {
1937                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1938                 return NULL;
1939         }
1940         if(!isSupportedOut(dstFormat))
1941         {
1942                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1943                 return NULL;
1944         }
1945
1946         /* sanity check */
1947         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
1948         {
1949                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1950                         srcW, srcH, dstW, dstH);
1951                 return NULL;
1952         }
1953
1954         if(!dstFilter) dstFilter= &dummyFilter;
1955         if(!srcFilter) srcFilter= &dummyFilter;
1956
1957         c= memalign(64, sizeof(SwsContext));
1958         memset(c, 0, sizeof(SwsContext));
1959
1960         c->srcW= srcW;
1961         c->srcH= srcH;
1962         c->dstW= dstW;
1963         c->dstH= dstH;
1964         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1965         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1966         c->flags= flags;
1967         c->dstFormat= dstFormat;
1968         c->srcFormat= srcFormat;
1969
1970         usesFilter=0;
1971         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesFilter=1;
1972         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesFilter=1;
1973         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesFilter=1;
1974         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesFilter=1;
1975         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesFilter=1;
1976         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesFilter=1;
1977         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesFilter=1;
1978         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesFilter=1;
1979
1980         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1981         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1982
1983         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1984         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1985
1986         // drop some chroma lines if the user wants it
1987         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1988         c->chrSrcVSubSample+= c->vChrDrop;
1989
1990         // drop every 2. pixel for chroma calculation unless user wants full chroma
1991         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1992                 c->chrSrcHSubSample=1;
1993
1994         c->chrIntHSubSample= c->chrDstHSubSample;
1995         c->chrIntVSubSample= c->chrSrcVSubSample;
1996         
1997         // note the -((-x)>>y) is so that we allways round toward +inf
1998         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1999         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2000         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2001         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2002         
2003         if(isBGR(dstFormat))
2004                 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_RGB, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2005         if(isRGB(dstFormat))
2006                 c->yuvTable= yuv2rgb_c_init(dstFormat & 0xFF, MODE_BGR, c->table_rV, c->table_gU, c->table_gV, c->table_bU);
2007
2008         /* unscaled special Cases */
2009         if(unscaled && !usesFilter)
2010         {
2011                 /* yv12_to_nv12 */
2012                 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_NV12)
2013                 {
2014                         c->swScale= PlanarToNV12Wrapper;
2015
2016                         if(flags&SWS_PRINT_INFO)
2017                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2018                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2019                         return c;
2020                 }
2021                 /* yv12_to_yuy2 */
2022                 if((srcFormat == IMGFMT_YV12||srcFormat==IMGFMT_I420)&&dstFormat == IMGFMT_YUY2)
2023                 {
2024                         c->swScale= PlanarToYuy2Wrapper;
2025
2026                         if(flags&SWS_PRINT_INFO)
2027                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2028                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2029                         return c;
2030                 }
2031                 /* yuv2bgr */
2032                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_I420) && isBGR(dstFormat))
2033                 {
2034                         // FIXME multiple yuv2rgb converters wont work that way cuz that thing is full of globals&statics
2035                         //FIXME rgb vs. bgr ? 
2036 #ifdef WORDS_BIGENDIAN
2037                         if(dstFormat==IMGFMT_BGR32)
2038                                 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_BGR);
2039                         else
2040                                 yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2041 #else
2042                         yuv2rgb_init( dstFormat&0xFF /* =bpp */, MODE_RGB);
2043 #endif
2044                         c->swScale= planarYuvToBgr;
2045
2046                         if(flags&SWS_PRINT_INFO)
2047                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2048                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2049                         return c;
2050                 }
2051                 
2052                 /* simple copy */
2053                 if(   srcFormat == dstFormat
2054                    || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
2055                    || (srcFormat==IMGFMT_I420 && dstFormat==IMGFMT_YV12)
2056                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2057                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2058                   )
2059                 {
2060                         c->swScale= simpleCopy;
2061
2062                         if(flags&SWS_PRINT_INFO)
2063                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2064                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2065                         return c;
2066                 }
2067                 
2068                 if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
2069                 {
2070                         c->swScale= yvu9toyv12Wrapper;
2071
2072                         if(flags&SWS_PRINT_INFO)
2073                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2074                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2075                         return c;
2076                 }
2077
2078                 /* bgr24toYV12 */
2079                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
2080                         c->swScale= bgr24toyv12Wrapper;
2081                 
2082                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2083                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
2084                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
2085                    && !needsDither)
2086                         c->swScale= rgb2rgbWrapper;
2087
2088                 /* LQ converters if -sws 0 or -sws 4*/
2089                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2090                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
2091                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
2092                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
2093                           && needsDither)
2094                                 c->swScale= rgb2rgbWrapper;
2095                 }
2096
2097                 if(c->swScale){
2098                         if(flags&SWS_PRINT_INFO)
2099                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2100                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2101                         return c;
2102                 }
2103         }
2104
2105         if(cpuCaps.hasMMX2)
2106         {
2107                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2108                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2109                 {
2110                         if(flags&SWS_PRINT_INFO)
2111                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2112                 }
2113         }
2114         else
2115                 c->canMMX2BeUsed=0;
2116
2117         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2118         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2119
2120         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2121         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2122         // n-2 is the last chrominance sample available
2123         // this is not perfect, but noone shuld notice the difference, the more correct variant
2124         // would be like the vertical one, but that would require some special code for the
2125         // first and last pixel
2126         if(flags&SWS_FAST_BILINEAR)
2127         {
2128                 if(c->canMMX2BeUsed)
2129                 {
2130                         c->lumXInc+= 20;
2131                         c->chrXInc+= 20;
2132                 }
2133                 //we dont use the x86asm scaler if mmx is available
2134                 else if(cpuCaps.hasMMX)
2135                 {
2136                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2137                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2138                 }
2139         }
2140
2141         /* precalculate horizontal scaler filter coefficients */
2142         {
2143                 const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
2144
2145                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2146                                  srcW      ,       dstW, filterAlign, 1<<14,
2147                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2148                                  srcFilter->lumH, dstFilter->lumH);
2149                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2150                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2151                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2152                                  srcFilter->chrH, dstFilter->chrH);
2153
2154 #ifdef ARCH_X86
2155 // cant downscale !!!
2156                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2157                 {
2158                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
2159                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
2160                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
2161                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2162
2163                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2164                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2165                 }
2166 #endif
2167         } // Init Horizontal stuff
2168
2169
2170
2171         /* precalculate vertical scaler filter coefficients */
2172         initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2173                         srcH      ,        dstH, 1, (1<<12)-4,
2174                         (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2175                         srcFilter->lumV, dstFilter->lumV);
2176         initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2177                         c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
2178                         (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2179                         srcFilter->chrV, dstFilter->chrV);
2180
2181         // Calculate Buffer Sizes so that they wont run out while handling these damn slices
2182         c->vLumBufSize= c->vLumFilterSize;
2183         c->vChrBufSize= c->vChrFilterSize;
2184         for(i=0; i<dstH; i++)
2185         {
2186                 int chrI= i*c->chrDstH / dstH;
2187                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2188                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2189                 nextSlice&= ~3; // Slices start at boundaries which are divisable through 4
2190                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2191                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2192                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2193                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2194         }
2195
2196         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2197         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2198         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2199         //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)
2200         for(i=0; i<c->vLumBufSize; i++)
2201                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2202         for(i=0; i<c->vChrBufSize; i++)
2203                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2204
2205         //try to avoid drawing green stuff between the right end and the stride end
2206         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2207         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2208
2209         ASSERT(c->chrDstH <= dstH)
2210
2211         // pack filter data for mmx code
2212         if(cpuCaps.hasMMX)
2213         {
2214                 c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize*      dstH*4*sizeof(int16_t));
2215                 c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
2216                 for(i=0; i<c->vLumFilterSize*dstH; i++)
2217                         c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
2218                                 c->vLumFilter[i];
2219                 for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
2220                         c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
2221                                 c->vChrFilter[i];
2222         }
2223
2224         if(flags&SWS_PRINT_INFO)
2225         {
2226 #ifdef DITHER1XBPP
2227                 char *dither= " dithered";
2228 #else
2229                 char *dither= "";
2230 #endif
2231                 if(flags&SWS_FAST_BILINEAR)
2232                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2233                 else if(flags&SWS_BILINEAR)
2234                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2235                 else if(flags&SWS_BICUBIC)
2236                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2237                 else if(flags&SWS_X)
2238                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2239                 else if(flags&SWS_POINT)
2240                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2241                 else if(flags&SWS_AREA)
2242                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2243                 else if(flags&SWS_BICUBLIN)
2244                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2245                 else if(flags&SWS_GAUSS)
2246                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2247                 else if(flags&SWS_SINC)
2248                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2249                 else if(flags&SWS_LANCZOS)
2250                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2251                 else if(flags&SWS_SPLINE)
2252                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2253                 else
2254                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2255
2256                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2257                         MSG_INFO("from %s to%s %s ", 
2258                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2259                 else
2260                         MSG_INFO("from %s to %s ", 
2261                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2262
2263                 if(cpuCaps.hasMMX2)
2264                         MSG_INFO("using MMX2\n");
2265                 else if(cpuCaps.has3DNow)
2266                         MSG_INFO("using 3DNOW\n");
2267                 else if(cpuCaps.hasMMX)
2268                         MSG_INFO("using MMX\n");
2269                 else
2270                         MSG_INFO("using C\n");
2271         }
2272
2273         if((flags & SWS_PRINT_INFO) && verbose)
2274         {
2275                 if(cpuCaps.hasMMX)
2276                 {
2277                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2278                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2279                         else
2280                         {
2281                                 if(c->hLumFilterSize==4)
2282                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2283                                 else if(c->hLumFilterSize==8)
2284                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2285                                 else
2286                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2287
2288                                 if(c->hChrFilterSize==4)
2289                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2290                                 else if(c->hChrFilterSize==8)
2291                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2292                                 else
2293                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2294                         }
2295                 }
2296                 else
2297                 {
2298 #ifdef ARCH_X86
2299                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2300 #else
2301                         if(flags & SWS_FAST_BILINEAR)
2302                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2303                         else
2304                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2305 #endif
2306                 }
2307                 if(isPlanarYUV(dstFormat))
2308                 {
2309                         if(c->vLumFilterSize==1)
2310                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2311                         else
2312                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", cpuCaps.hasMMX ? "MMX" : "C");
2313                 }
2314                 else
2315                 {
2316                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2317                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2318                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
2319                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2320                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2321                         else
2322                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
2323                 }
2324
2325                 if(dstFormat==IMGFMT_BGR24)
2326                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2327                                 cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
2328                 else if(dstFormat==IMGFMT_BGR32)
2329                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2330                 else if(dstFormat==IMGFMT_BGR16)
2331                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2332                 else if(dstFormat==IMGFMT_BGR15)
2333                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
2334
2335                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2336         }
2337         if((flags & SWS_PRINT_INFO) && verbose>1)
2338         {
2339                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2340                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2341                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2342                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2343         }
2344
2345         c->swScale= swScale;
2346         return c;
2347 }
2348
2349 /**
2350  * returns a normalized gaussian curve used to filter stuff
2351  * quality=3 is high quality, lowwer is lowwer quality
2352  */
2353
2354 SwsVector *getGaussianVec(double variance, double quality){
2355         const int length= (int)(variance*quality + 0.5) | 1;
2356         int i;
2357         double *coeff= memalign(sizeof(double), length*sizeof(double));
2358         double middle= (length-1)*0.5;
2359         SwsVector *vec= malloc(sizeof(SwsVector));
2360
2361         vec->coeff= coeff;
2362         vec->length= length;
2363
2364         for(i=0; i<length; i++)
2365         {
2366                 double dist= i-middle;
2367                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2368         }
2369
2370         normalizeVec(vec, 1.0);
2371
2372         return vec;
2373 }
2374
2375 SwsVector *getConstVec(double c, int length){
2376         int i;
2377         double *coeff= memalign(sizeof(double), length*sizeof(double));
2378         SwsVector *vec= malloc(sizeof(SwsVector));
2379
2380         vec->coeff= coeff;
2381         vec->length= length;
2382
2383         for(i=0; i<length; i++)
2384                 coeff[i]= c;
2385
2386         return vec;
2387 }
2388
2389
2390 SwsVector *getIdentityVec(void){
2391         double *coeff= memalign(sizeof(double), sizeof(double));
2392         SwsVector *vec= malloc(sizeof(SwsVector));
2393         coeff[0]= 1.0;
2394
2395         vec->coeff= coeff;
2396         vec->length= 1;
2397
2398         return vec;
2399 }
2400
2401 void normalizeVec(SwsVector *a, double height){
2402         int i;
2403         double sum=0;
2404         double inv;
2405
2406         for(i=0; i<a->length; i++)
2407                 sum+= a->coeff[i];
2408
2409         inv= height/sum;
2410
2411         for(i=0; i<a->length; i++)
2412                 a->coeff[i]*= height;
2413 }
2414
2415 void scaleVec(SwsVector *a, double scalar){
2416         int i;
2417
2418         for(i=0; i<a->length; i++)
2419                 a->coeff[i]*= scalar;
2420 }
2421
2422 static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
2423         int length= a->length + b->length - 1;
2424         double *coeff= memalign(sizeof(double), length*sizeof(double));
2425         int i, j;
2426         SwsVector *vec= malloc(sizeof(SwsVector));
2427
2428         vec->coeff= coeff;
2429         vec->length= length;
2430
2431         for(i=0; i<length; i++) coeff[i]= 0.0;
2432
2433         for(i=0; i<a->length; i++)
2434         {
2435                 for(j=0; j<b->length; j++)
2436                 {
2437                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2438                 }
2439         }
2440
2441         return vec;
2442 }
2443
2444 static SwsVector *sumVec(SwsVector *a, SwsVector *b){
2445         int length= MAX(a->length, b->length);
2446         double *coeff= memalign(sizeof(double), length*sizeof(double));
2447         int i;
2448         SwsVector *vec= malloc(sizeof(SwsVector));
2449
2450         vec->coeff= coeff;
2451         vec->length= length;
2452
2453         for(i=0; i<length; i++) coeff[i]= 0.0;
2454
2455         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2456         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2457
2458         return vec;
2459 }
2460
2461 static SwsVector *diffVec(SwsVector *a, SwsVector *b){
2462         int length= MAX(a->length, b->length);
2463         double *coeff= memalign(sizeof(double), length*sizeof(double));
2464         int i;
2465         SwsVector *vec= malloc(sizeof(SwsVector));
2466
2467         vec->coeff= coeff;
2468         vec->length= length;
2469
2470         for(i=0; i<length; i++) coeff[i]= 0.0;
2471
2472         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2473         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2474
2475         return vec;
2476 }
2477
2478 /* shift left / or right if "shift" is negative */
2479 static SwsVector *getShiftedVec(SwsVector *a, int shift){
2480         int length= a->length + ABS(shift)*2;
2481         double *coeff= memalign(sizeof(double), length*sizeof(double));
2482         int i;
2483         SwsVector *vec= malloc(sizeof(SwsVector));
2484
2485         vec->coeff= coeff;
2486         vec->length= length;
2487
2488         for(i=0; i<length; i++) coeff[i]= 0.0;
2489
2490         for(i=0; i<a->length; i++)
2491         {
2492                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2493         }
2494
2495         return vec;
2496 }
2497
2498 void shiftVec(SwsVector *a, int shift){
2499         SwsVector *shifted= getShiftedVec(a, shift);
2500         free(a->coeff);
2501         a->coeff= shifted->coeff;
2502         a->length= shifted->length;
2503         free(shifted);
2504 }
2505
2506 void addVec(SwsVector *a, SwsVector *b){
2507         SwsVector *sum= sumVec(a, b);
2508         free(a->coeff);
2509         a->coeff= sum->coeff;
2510         a->length= sum->length;
2511         free(sum);
2512 }
2513
2514 void subVec(SwsVector *a, SwsVector *b){
2515         SwsVector *diff= diffVec(a, b);
2516         free(a->coeff);
2517         a->coeff= diff->coeff;
2518         a->length= diff->length;
2519         free(diff);
2520 }
2521
2522 void convVec(SwsVector *a, SwsVector *b){
2523         SwsVector *conv= getConvVec(a, b);
2524         free(a->coeff);
2525         a->coeff= conv->coeff;
2526         a->length= conv->length;
2527         free(conv);
2528 }
2529
2530 SwsVector *cloneVec(SwsVector *a){
2531         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2532         int i;
2533         SwsVector *vec= malloc(sizeof(SwsVector));
2534
2535         vec->coeff= coeff;
2536         vec->length= a->length;
2537
2538         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2539
2540         return vec;
2541 }
2542
2543 void printVec(SwsVector *a){
2544         int i;
2545         double max=0;
2546         double min=0;
2547         double range;
2548
2549         for(i=0; i<a->length; i++)
2550                 if(a->coeff[i]>max) max= a->coeff[i];
2551
2552         for(i=0; i<a->length; i++)
2553                 if(a->coeff[i]<min) min= a->coeff[i];
2554
2555         range= max - min;
2556
2557         for(i=0; i<a->length; i++)
2558         {
2559                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2560                 MSG_DBG2("%1.3f ", a->coeff[i]);
2561                 for(;x>0; x--) MSG_DBG2(" ");
2562                 MSG_DBG2("|\n");
2563         }
2564 }
2565
2566 void freeVec(SwsVector *a){
2567         if(!a) return;
2568         if(a->coeff) free(a->coeff);
2569         a->coeff=NULL;
2570         a->length=0;
2571         free(a);
2572 }
2573
2574 void freeSwsContext(SwsContext *c){
2575         int i;
2576         if(!c) return;
2577
2578         if(c->lumPixBuf)
2579         {
2580                 for(i=0; i<c->vLumBufSize; i++)
2581                 {
2582                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2583                         c->lumPixBuf[i]=NULL;
2584                 }
2585                 free(c->lumPixBuf);
2586                 c->lumPixBuf=NULL;
2587         }
2588
2589         if(c->chrPixBuf)
2590         {
2591                 for(i=0; i<c->vChrBufSize; i++)
2592                 {
2593                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2594                         c->chrPixBuf[i]=NULL;
2595                 }
2596                 free(c->chrPixBuf);
2597                 c->chrPixBuf=NULL;
2598         }
2599
2600         if(c->vLumFilter) free(c->vLumFilter);
2601         c->vLumFilter = NULL;
2602         if(c->vChrFilter) free(c->vChrFilter);
2603         c->vChrFilter = NULL;
2604         if(c->hLumFilter) free(c->hLumFilter);
2605         c->hLumFilter = NULL;
2606         if(c->hChrFilter) free(c->hChrFilter);
2607         c->hChrFilter = NULL;
2608
2609         if(c->vLumFilterPos) free(c->vLumFilterPos);
2610         c->vLumFilterPos = NULL;
2611         if(c->vChrFilterPos) free(c->vChrFilterPos);
2612         c->vChrFilterPos = NULL;
2613         if(c->hLumFilterPos) free(c->hLumFilterPos);
2614         c->hLumFilterPos = NULL;
2615         if(c->hChrFilterPos) free(c->hChrFilterPos);
2616         c->hChrFilterPos = NULL;
2617
2618         if(c->lumMmxFilter) free(c->lumMmxFilter);
2619         c->lumMmxFilter = NULL;
2620         if(c->chrMmxFilter) free(c->chrMmxFilter);
2621         c->chrMmxFilter = NULL;
2622
2623         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2624         c->lumMmx2Filter=NULL;
2625         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2626         c->chrMmx2Filter=NULL;
2627         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2628         c->lumMmx2FilterPos=NULL;
2629         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2630         c->chrMmx2FilterPos=NULL;
2631         if(c->yuvTable) free(c->yuvTable);
2632         c->yuvTable=NULL;
2633
2634         free(c);
2635 }
2636
2637