check for negative strides before memcpy
[ffmpeg.git] / postproc / swscale.c
1 /*
2     Copyright (C) 2001-2003 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, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {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 <unistd.h>
57 #include "../config.h"
58 #include "../mangle.h"
59 #include <assert.h>
60 #ifdef HAVE_MALLOC_H
61 #include <malloc.h>
62 #else
63 #include <stdlib.h>
64 #endif
65 #ifdef HAVE_SYS_MMAN_H
66 #include <sys/mman.h>
67 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
68 #define MAP_ANONYMOUS MAP_ANON
69 #endif
70 #endif
71 #include "swscale.h"
72 #include "swscale_internal.h"
73 #include "../cpudetect.h"
74 #include "../bswap.h"
75 #include "../libvo/img_format.h"
76 #include "rgb2rgb.h"
77 #include "../libvo/fastmemcpy.h"
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_YVU9 \
107                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \
108                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
109 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
110 #define isGray(x)      ((x)==IMGFMT_Y800)
111 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
112 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
113 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
114                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
115                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
116                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
117                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
118 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
119                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
120                         || isRGB(x) || isBGR(x)\
121                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\
122                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
123 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
124
125 #define RGB2YUV_SHIFT 16
126 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
127 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
128 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
129 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
130 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
131 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
132 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
133 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
134 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
135
136 extern const int32_t Inverse_Table_6_9[8][4];
137
138 /*
139 NOTES
140 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
141
142 TODO
143 more intelligent missalignment avoidance for the horizontal scaler
144 write special vertical cubic upscale version
145 Optimize C code (yv12 / minmax)
146 add support for packed pixel yuv input & output
147 add support for Y8 output
148 optimize bgr24 & bgr32
149 add BGR4 output support
150 write special BGR->BGR scaler
151 */
152
153 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
154 #define MIN(a,b) ((a) > (b) ? (b) : (a))
155 #define MAX(a,b) ((a) < (b) ? (b) : (a))
156
157 #if defined(ARCH_X86) || defined(ARCH_X86_64)
158 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
159 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
160 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
161 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
162 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
163 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
164 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
165 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
166
167 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
168 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
169 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
170 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
171
172 static uint64_t __attribute__((aligned(8))) dither4[2]={
173         0x0103010301030103LL,
174         0x0200020002000200LL,};
175
176 static uint64_t __attribute__((aligned(8))) dither8[2]={
177         0x0602060206020602LL,
178         0x0004000400040004LL,};
179
180 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
181 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
182 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
183 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
184 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
185 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
186
187 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
188 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
189 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
190
191 #ifdef FAST_BGR2YV12
192 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
193 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
194 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
195 #else
196 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
197 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
198 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
199 #endif
200 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
201 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
202 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
203 #endif
204
205 // clipping helper table for C implementations:
206 static unsigned char clip_table[768];
207
208 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
209                   
210 extern const uint8_t dither_2x2_4[2][8];
211 extern const uint8_t dither_2x2_8[2][8];
212 extern const uint8_t dither_8x8_32[8][8];
213 extern const uint8_t dither_8x8_73[8][8];
214 extern const uint8_t dither_8x8_220[8][8];
215
216 #if defined(ARCH_X86) || defined(ARCH_X86_64)
217 void in_asm_used_var_warning_killer()
218 {
219  volatile int i= bF8+bFC+w10+
220  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
221  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
222  if(i) i=0;
223 }
224 #endif
225
226 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
227                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
228                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
229 {
230         //FIXME Optimize (just quickly writen not opti..)
231         int i;
232         for(i=0; i<dstW; i++)
233         {
234                 int val=1<<18;
235                 int j;
236                 for(j=0; j<lumFilterSize; j++)
237                         val += lumSrc[j][i] * lumFilter[j];
238
239                 dest[i]= MIN(MAX(val>>19, 0), 255);
240         }
241
242         if(uDest != NULL)
243                 for(i=0; i<chrDstW; i++)
244                 {
245                         int u=1<<18;
246                         int v=1<<18;
247                         int j;
248                         for(j=0; j<chrFilterSize; j++)
249                         {
250                                 u += chrSrc[j][i] * chrFilter[j];
251                                 v += chrSrc[j][i + 2048] * chrFilter[j];
252                         }
253
254                         uDest[i]= MIN(MAX(u>>19, 0), 255);
255                         vDest[i]= MIN(MAX(v>>19, 0), 255);
256                 }
257 }
258
259 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
260                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
261                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
262 {
263         //FIXME Optimize (just quickly writen not opti..)
264         int i;
265         for(i=0; i<dstW; i++)
266         {
267                 int val=1<<18;
268                 int j;
269                 for(j=0; j<lumFilterSize; j++)
270                         val += lumSrc[j][i] * lumFilter[j];
271
272                 dest[i]= MIN(MAX(val>>19, 0), 255);
273         }
274
275         if(uDest == NULL)
276                 return;
277
278         if(dstFormat == IMGFMT_NV12)
279                 for(i=0; i<chrDstW; i++)
280                 {
281                         int u=1<<18;
282                         int v=1<<18;
283                         int j;
284                         for(j=0; j<chrFilterSize; j++)
285                         {
286                                 u += chrSrc[j][i] * chrFilter[j];
287                                 v += chrSrc[j][i + 2048] * chrFilter[j];
288                         }
289
290                         uDest[2*i]= MIN(MAX(u>>19, 0), 255);
291                         uDest[2*i+1]= MIN(MAX(v>>19, 0), 255);
292                 }
293         else
294                 for(i=0; i<chrDstW; i++)
295                 {
296                         int u=1<<18;
297                         int v=1<<18;
298                         int j;
299                         for(j=0; j<chrFilterSize; j++)
300                         {
301                                 u += chrSrc[j][i] * chrFilter[j];
302                                 v += chrSrc[j][i + 2048] * chrFilter[j];
303                         }
304
305                         uDest[2*i]= MIN(MAX(v>>19, 0), 255);
306                         uDest[2*i+1]= MIN(MAX(u>>19, 0), 255);
307                 }
308 }
309
310 #define YSCALE_YUV_2_PACKEDX_C(type) \
311                 for(i=0; i<(dstW>>1); i++){\
312                         int j;\
313                         int Y1=1<<18;\
314                         int Y2=1<<18;\
315                         int U=1<<18;\
316                         int V=1<<18;\
317                         type *r, *b, *g;\
318                         const int i2= 2*i;\
319                         \
320                         for(j=0; j<lumFilterSize; j++)\
321                         {\
322                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
323                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
324                         }\
325                         for(j=0; j<chrFilterSize; j++)\
326                         {\
327                                 U += chrSrc[j][i] * chrFilter[j];\
328                                 V += chrSrc[j][i+2048] * chrFilter[j];\
329                         }\
330                         Y1>>=19;\
331                         Y2>>=19;\
332                         U >>=19;\
333                         V >>=19;\
334                         if((Y1|Y2|U|V)&256)\
335                         {\
336                                 if(Y1>255)   Y1=255;\
337                                 else if(Y1<0)Y1=0;\
338                                 if(Y2>255)   Y2=255;\
339                                 else if(Y2<0)Y2=0;\
340                                 if(U>255)    U=255;\
341                                 else if(U<0) U=0;\
342                                 if(V>255)    V=255;\
343                                 else if(V<0) V=0;\
344                         }
345                         
346 #define YSCALE_YUV_2_RGBX_C(type) \
347                         YSCALE_YUV_2_PACKEDX_C(type)\
348                         r = c->table_rV[V];\
349                         g = c->table_gU[U] + c->table_gV[V];\
350                         b = c->table_bU[U];\
351
352 #define YSCALE_YUV_2_PACKED2_C \
353                 for(i=0; i<(dstW>>1); i++){\
354                         const int i2= 2*i;\
355                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
356                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
357                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
358                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
359
360 #define YSCALE_YUV_2_RGB2_C(type) \
361                         YSCALE_YUV_2_PACKED2_C\
362                         type *r, *b, *g;\
363                         r = c->table_rV[V];\
364                         g = c->table_gU[U] + c->table_gV[V];\
365                         b = c->table_bU[U];\
366
367 #define YSCALE_YUV_2_PACKED1_C \
368                 for(i=0; i<(dstW>>1); i++){\
369                         const int i2= 2*i;\
370                         int Y1= buf0[i2  ]>>7;\
371                         int Y2= buf0[i2+1]>>7;\
372                         int U= (uvbuf1[i     ])>>7;\
373                         int V= (uvbuf1[i+2048])>>7;\
374
375 #define YSCALE_YUV_2_RGB1_C(type) \
376                         YSCALE_YUV_2_PACKED1_C\
377                         type *r, *b, *g;\
378                         r = c->table_rV[V];\
379                         g = c->table_gU[U] + c->table_gV[V];\
380                         b = c->table_bU[U];\
381
382 #define YSCALE_YUV_2_PACKED1B_C \
383                 for(i=0; i<(dstW>>1); i++){\
384                         const int i2= 2*i;\
385                         int Y1= buf0[i2  ]>>7;\
386                         int Y2= buf0[i2+1]>>7;\
387                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
388                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
389
390 #define YSCALE_YUV_2_RGB1B_C(type) \
391                         YSCALE_YUV_2_PACKED1B_C\
392                         type *r, *b, *g;\
393                         r = c->table_rV[V];\
394                         g = c->table_gU[U] + c->table_gV[V];\
395                         b = c->table_bU[U];\
396
397 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
398         switch(c->dstFormat)\
399         {\
400         case IMGFMT_BGR32:\
401         case IMGFMT_RGB32:\
402                 func(uint32_t)\
403                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
404                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
405                 }               \
406                 break;\
407         case IMGFMT_RGB24:\
408                 func(uint8_t)\
409                         ((uint8_t*)dest)[0]= r[Y1];\
410                         ((uint8_t*)dest)[1]= g[Y1];\
411                         ((uint8_t*)dest)[2]= b[Y1];\
412                         ((uint8_t*)dest)[3]= r[Y2];\
413                         ((uint8_t*)dest)[4]= g[Y2];\
414                         ((uint8_t*)dest)[5]= b[Y2];\
415                         dest+=6;\
416                 }\
417                 break;\
418         case IMGFMT_BGR24:\
419                 func(uint8_t)\
420                         ((uint8_t*)dest)[0]= b[Y1];\
421                         ((uint8_t*)dest)[1]= g[Y1];\
422                         ((uint8_t*)dest)[2]= r[Y1];\
423                         ((uint8_t*)dest)[3]= b[Y2];\
424                         ((uint8_t*)dest)[4]= g[Y2];\
425                         ((uint8_t*)dest)[5]= r[Y2];\
426                         dest+=6;\
427                 }\
428                 break;\
429         case IMGFMT_RGB16:\
430         case IMGFMT_BGR16:\
431                 {\
432                         const int dr1= dither_2x2_8[y&1    ][0];\
433                         const int dg1= dither_2x2_4[y&1    ][0];\
434                         const int db1= dither_2x2_8[(y&1)^1][0];\
435                         const int dr2= dither_2x2_8[y&1    ][1];\
436                         const int dg2= dither_2x2_4[y&1    ][1];\
437                         const int db2= dither_2x2_8[(y&1)^1][1];\
438                         func(uint16_t)\
439                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
440                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
441                         }\
442                 }\
443                 break;\
444         case IMGFMT_RGB15:\
445         case IMGFMT_BGR15:\
446                 {\
447                         const int dr1= dither_2x2_8[y&1    ][0];\
448                         const int dg1= dither_2x2_8[y&1    ][1];\
449                         const int db1= dither_2x2_8[(y&1)^1][0];\
450                         const int dr2= dither_2x2_8[y&1    ][1];\
451                         const int dg2= dither_2x2_8[y&1    ][0];\
452                         const int db2= dither_2x2_8[(y&1)^1][1];\
453                         func(uint16_t)\
454                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
455                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
456                         }\
457                 }\
458                 break;\
459         case IMGFMT_RGB8:\
460         case IMGFMT_BGR8:\
461                 {\
462                         const uint8_t * const d64= dither_8x8_73[y&7];\
463                         const uint8_t * const d32= dither_8x8_32[y&7];\
464                         func(uint8_t)\
465                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
466                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
467                         }\
468                 }\
469                 break;\
470         case IMGFMT_RGB4:\
471         case IMGFMT_BGR4:\
472                 {\
473                         const uint8_t * const d64= dither_8x8_73 [y&7];\
474                         const uint8_t * const d128=dither_8x8_220[y&7];\
475                         func(uint8_t)\
476                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
477                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
478                         }\
479                 }\
480                 break;\
481         case IMGFMT_RG4B:\
482         case IMGFMT_BG4B:\
483                 {\
484                         const uint8_t * const d64= dither_8x8_73 [y&7];\
485                         const uint8_t * const d128=dither_8x8_220[y&7];\
486                         func(uint8_t)\
487                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
488                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
489                         }\
490                 }\
491                 break;\
492         case IMGFMT_RGB1:\
493         case IMGFMT_BGR1:\
494                 {\
495                         const uint8_t * const d128=dither_8x8_220[y&7];\
496                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
497                         for(i=0; i<dstW-7; i+=8){\
498                                 int acc;\
499                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
500                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
501                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
502                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
503                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
504                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
505                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
506                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
507                                 ((uint8_t*)dest)[0]= acc;\
508                                 dest++;\
509                         }\
510 \
511 /*\
512 ((uint8_t*)dest)-= dstW>>4;\
513 {\
514                         int acc=0;\
515                         int left=0;\
516                         static int top[1024];\
517                         static int last_new[1024][1024];\
518                         static int last_in3[1024][1024];\
519                         static int drift[1024][1024];\
520                         int topLeft=0;\
521                         int shift=0;\
522                         int count=0;\
523                         const uint8_t * const d128=dither_8x8_220[y&7];\
524                         int error_new=0;\
525                         int error_in3=0;\
526                         int f=0;\
527                         \
528                         for(i=dstW>>1; i<dstW; i++){\
529                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
530                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
531                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
532                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
533                                         + (last_new[y][i] - in3)*f/256;\
534                                 int new= old> 128 ? 255 : 0;\
535 \
536                                 error_new+= ABS(last_new[y][i] - new);\
537                                 error_in3+= ABS(last_in3[y][i] - in3);\
538                                 f= error_new - error_in3*4;\
539                                 if(f<0) f=0;\
540                                 if(f>256) f=256;\
541 \
542                                 topLeft= top[i];\
543                                 left= top[i]= old - new;\
544                                 last_new[y][i]= new;\
545                                 last_in3[y][i]= in3;\
546 \
547                                 acc+= acc + (new&1);\
548                                 if((i&7)==6){\
549                                         ((uint8_t*)dest)[0]= acc;\
550                                         ((uint8_t*)dest)++;\
551                                 }\
552                         }\
553 }\
554 */\
555                 }\
556                 break;\
557         case IMGFMT_YUY2:\
558                 func2\
559                         ((uint8_t*)dest)[2*i2+0]= Y1;\
560                         ((uint8_t*)dest)[2*i2+1]= U;\
561                         ((uint8_t*)dest)[2*i2+2]= Y2;\
562                         ((uint8_t*)dest)[2*i2+3]= V;\
563                 }               \
564                 break;\
565         case IMGFMT_UYVY:\
566                 func2\
567                         ((uint8_t*)dest)[2*i2+0]= U;\
568                         ((uint8_t*)dest)[2*i2+1]= Y1;\
569                         ((uint8_t*)dest)[2*i2+2]= V;\
570                         ((uint8_t*)dest)[2*i2+3]= Y2;\
571                 }               \
572                 break;\
573         }\
574
575
576 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
577                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
578                                     uint8_t *dest, int dstW, int y)
579 {
580         int i;
581         switch(c->dstFormat)
582         {
583         case IMGFMT_RGB32:
584         case IMGFMT_BGR32:
585                 YSCALE_YUV_2_RGBX_C(uint32_t)
586                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
587                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
588                 }
589                 break;
590         case IMGFMT_RGB24:
591                 YSCALE_YUV_2_RGBX_C(uint8_t)
592                         ((uint8_t*)dest)[0]= r[Y1];
593                         ((uint8_t*)dest)[1]= g[Y1];
594                         ((uint8_t*)dest)[2]= b[Y1];
595                         ((uint8_t*)dest)[3]= r[Y2];
596                         ((uint8_t*)dest)[4]= g[Y2];
597                         ((uint8_t*)dest)[5]= b[Y2];
598                         dest+=6;
599                 }
600                 break;
601         case IMGFMT_BGR24:
602                 YSCALE_YUV_2_RGBX_C(uint8_t)
603                         ((uint8_t*)dest)[0]= b[Y1];
604                         ((uint8_t*)dest)[1]= g[Y1];
605                         ((uint8_t*)dest)[2]= r[Y1];
606                         ((uint8_t*)dest)[3]= b[Y2];
607                         ((uint8_t*)dest)[4]= g[Y2];
608                         ((uint8_t*)dest)[5]= r[Y2];
609                         dest+=6;
610                 }
611                 break;
612         case IMGFMT_RGB16:
613         case IMGFMT_BGR16:
614                 {
615                         const int dr1= dither_2x2_8[y&1    ][0];
616                         const int dg1= dither_2x2_4[y&1    ][0];
617                         const int db1= dither_2x2_8[(y&1)^1][0];
618                         const int dr2= dither_2x2_8[y&1    ][1];
619                         const int dg2= dither_2x2_4[y&1    ][1];
620                         const int db2= dither_2x2_8[(y&1)^1][1];
621                         YSCALE_YUV_2_RGBX_C(uint16_t)
622                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
623                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
624                         }
625                 }
626                 break;
627         case IMGFMT_RGB15:
628         case IMGFMT_BGR15:
629                 {
630                         const int dr1= dither_2x2_8[y&1    ][0];
631                         const int dg1= dither_2x2_8[y&1    ][1];
632                         const int db1= dither_2x2_8[(y&1)^1][0];
633                         const int dr2= dither_2x2_8[y&1    ][1];
634                         const int dg2= dither_2x2_8[y&1    ][0];
635                         const int db2= dither_2x2_8[(y&1)^1][1];
636                         YSCALE_YUV_2_RGBX_C(uint16_t)
637                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
638                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
639                         }
640                 }
641                 break;
642         case IMGFMT_RGB8:
643         case IMGFMT_BGR8:
644                 {
645                         const uint8_t * const d64= dither_8x8_73[y&7];
646                         const uint8_t * const d32= dither_8x8_32[y&7];
647                         YSCALE_YUV_2_RGBX_C(uint8_t)
648                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
649                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
650                         }
651                 }
652                 break;
653         case IMGFMT_RGB4:
654         case IMGFMT_BGR4:
655                 {
656                         const uint8_t * const d64= dither_8x8_73 [y&7];
657                         const uint8_t * const d128=dither_8x8_220[y&7];
658                         YSCALE_YUV_2_RGBX_C(uint8_t)
659                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
660                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
661                         }
662                 }
663                 break;
664         case IMGFMT_RG4B:
665         case IMGFMT_BG4B:
666                 {
667                         const uint8_t * const d64= dither_8x8_73 [y&7];
668                         const uint8_t * const d128=dither_8x8_220[y&7];
669                         YSCALE_YUV_2_RGBX_C(uint8_t)
670                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
671                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
672                         }
673                 }
674                 break;
675         case IMGFMT_RGB1:
676         case IMGFMT_BGR1:
677                 {
678                         const uint8_t * const d128=dither_8x8_220[y&7];
679                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
680                         int acc=0;
681                         for(i=0; i<dstW-1; i+=2){
682                                 int j;
683                                 int Y1=1<<18;
684                                 int Y2=1<<18;
685
686                                 for(j=0; j<lumFilterSize; j++)
687                                 {
688                                         Y1 += lumSrc[j][i] * lumFilter[j];
689                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
690                                 }
691                                 Y1>>=19;
692                                 Y2>>=19;
693                                 if((Y1|Y2)&256)
694                                 {
695                                         if(Y1>255)   Y1=255;
696                                         else if(Y1<0)Y1=0;
697                                         if(Y2>255)   Y2=255;
698                                         else if(Y2<0)Y2=0;
699                                 }
700                                 acc+= acc + g[Y1+d128[(i+0)&7]];
701                                 acc+= acc + g[Y2+d128[(i+1)&7]];
702                                 if((i&7)==6){
703                                         ((uint8_t*)dest)[0]= acc;
704                                         dest++;
705                                 }
706                         }
707                 }
708                 break;
709         case IMGFMT_YUY2:
710                 YSCALE_YUV_2_PACKEDX_C(void)
711                         ((uint8_t*)dest)[2*i2+0]= Y1;
712                         ((uint8_t*)dest)[2*i2+1]= U;
713                         ((uint8_t*)dest)[2*i2+2]= Y2;
714                         ((uint8_t*)dest)[2*i2+3]= V;
715                 }
716                 break;
717         case IMGFMT_UYVY:
718                 YSCALE_YUV_2_PACKEDX_C(void)
719                         ((uint8_t*)dest)[2*i2+0]= U;
720                         ((uint8_t*)dest)[2*i2+1]= Y1;
721                         ((uint8_t*)dest)[2*i2+2]= V;
722                         ((uint8_t*)dest)[2*i2+3]= Y2;
723                 }
724                 break;
725         }
726 }
727
728
729 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
730 //Plain C versions
731 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
732 #define COMPILE_C
733 #endif
734
735 #ifdef ARCH_POWERPC
736 #ifdef HAVE_ALTIVEC
737 #define COMPILE_ALTIVEC
738 #endif //HAVE_ALTIVEC
739 #endif //ARCH_POWERPC
740
741 #if defined(ARCH_X86) || defined(ARCH_X86_64)
742
743 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
744 #define COMPILE_MMX
745 #endif
746
747 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
748 #define COMPILE_MMX2
749 #endif
750
751 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
752 #define COMPILE_3DNOW
753 #endif
754 #endif //ARCH_X86 || ARCH_X86_64
755
756 #undef HAVE_MMX
757 #undef HAVE_MMX2
758 #undef HAVE_3DNOW
759
760 #ifdef COMPILE_C
761 #undef HAVE_MMX
762 #undef HAVE_MMX2
763 #undef HAVE_3DNOW
764 #undef HAVE_ALTIVEC
765 #define RENAME(a) a ## _C
766 #include "swscale_template.c"
767 #endif
768
769 #ifdef ARCH_POWERPC
770 #ifdef COMPILE_ALTIVEC
771 #undef RENAME
772 #define HAVE_ALTIVEC
773 #define RENAME(a) a ## _altivec
774 #include "swscale_template.c"
775 #endif
776 #endif //ARCH_POWERPC
777
778 #if defined(ARCH_X86) || defined(ARCH_X86_64)
779
780 //X86 versions
781 /*
782 #undef RENAME
783 #undef HAVE_MMX
784 #undef HAVE_MMX2
785 #undef HAVE_3DNOW
786 #define ARCH_X86
787 #define RENAME(a) a ## _X86
788 #include "swscale_template.c"
789 */
790 //MMX versions
791 #ifdef COMPILE_MMX
792 #undef RENAME
793 #define HAVE_MMX
794 #undef HAVE_MMX2
795 #undef HAVE_3DNOW
796 #define RENAME(a) a ## _MMX
797 #include "swscale_template.c"
798 #endif
799
800 //MMX2 versions
801 #ifdef COMPILE_MMX2
802 #undef RENAME
803 #define HAVE_MMX
804 #define HAVE_MMX2
805 #undef HAVE_3DNOW
806 #define RENAME(a) a ## _MMX2
807 #include "swscale_template.c"
808 #endif
809
810 //3DNOW versions
811 #ifdef COMPILE_3DNOW
812 #undef RENAME
813 #define HAVE_MMX
814 #undef HAVE_MMX2
815 #define HAVE_3DNOW
816 #define RENAME(a) a ## _3DNow
817 #include "swscale_template.c"
818 #endif
819
820 #endif //ARCH_X86 || ARCH_X86_64
821
822 // minor note: the HAVE_xyz is messed up after that line so don't use it
823
824 static double getSplineCoeff(double a, double b, double c, double d, double dist)
825 {
826 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
827         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
828         else            return getSplineCoeff(  0.0, 
829                                                  b+ 2.0*c + 3.0*d,
830                                                         c + 3.0*d,
831                                                 -b- 3.0*c - 6.0*d,
832                                                 dist-1.0);
833 }
834
835 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
836                               int srcW, int dstW, int filterAlign, int one, int flags,
837                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
838 {
839         int i;
840         int filterSize;
841         int filter2Size;
842         int minFilterSize;
843         double *filter=NULL;
844         double *filter2=NULL;
845 #if defined(ARCH_X86) || defined(ARCH_X86_64)
846         if(flags & SWS_CPU_CAPS_MMX)
847                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
848 #endif
849
850         // Note the +1 is for the MMXscaler which reads over the end
851         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
852
853         if(ABS(xInc - 0x10000) <10) // unscaled
854         {
855                 int i;
856                 filterSize= 1;
857                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
858                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
859
860                 for(i=0; i<dstW; i++)
861                 {
862                         filter[i*filterSize]=1;
863                         (*filterPos)[i]=i;
864                 }
865
866         }
867         else if(flags&SWS_POINT) // lame looking point sampling mode
868         {
869                 int i;
870                 int xDstInSrc;
871                 filterSize= 1;
872                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
873                 
874                 xDstInSrc= xInc/2 - 0x8000;
875                 for(i=0; i<dstW; i++)
876                 {
877                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
878
879                         (*filterPos)[i]= xx;
880                         filter[i]= 1.0;
881                         xDstInSrc+= xInc;
882                 }
883         }
884         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
885         {
886                 int i;
887                 int xDstInSrc;
888                 if     (flags&SWS_BICUBIC) filterSize= 4;
889                 else if(flags&SWS_X      ) filterSize= 4;
890                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
891                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
892
893                 xDstInSrc= xInc/2 - 0x8000;
894                 for(i=0; i<dstW; i++)
895                 {
896                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
897                         int j;
898
899                         (*filterPos)[i]= xx;
900                                 //Bilinear upscale / linear interpolate / Area averaging
901                                 for(j=0; j<filterSize; j++)
902                                 {
903                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
904                                         double coeff= 1.0 - d;
905                                         if(coeff<0) coeff=0;
906                                         filter[i*filterSize + j]= coeff;
907                                         xx++;
908                                 }
909                         xDstInSrc+= xInc;
910                 }
911         }
912         else
913         {
914                 double xDstInSrc;
915                 double sizeFactor, filterSizeInSrc;
916                 const double xInc1= (double)xInc / (double)(1<<16);
917
918                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
919                 else if(flags&SWS_X)            sizeFactor= 8.0;
920                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
921                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
922                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
923                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
924                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
925                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
926                 else {
927                         sizeFactor= 0.0; //GCC warning killer
928                         ASSERT(0)
929                 }
930                 
931                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
932                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
933
934                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
935                 if(filterSize > srcW-2) filterSize=srcW-2;
936
937                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
938
939                 xDstInSrc= xInc1 / 2.0 - 0.5;
940                 for(i=0; i<dstW; i++)
941                 {
942                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
943                         int j;
944                         (*filterPos)[i]= xx;
945                         for(j=0; j<filterSize; j++)
946                         {
947                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
948                                 double coeff;
949                                 if(flags & SWS_BICUBIC)
950                                 {
951                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
952                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
953
954                                         if(d<1.0) 
955                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
956                                         else if(d<2.0)
957                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
958                                         else
959                                                 coeff=0.0;
960                                 }
961 /*                              else if(flags & SWS_X)
962                                 {
963                                         double p= param ? param*0.01 : 0.3;
964                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
965                                         coeff*= pow(2.0, - p*d*d);
966                                 }*/
967                                 else if(flags & SWS_X)
968                                 {
969                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
970                                         
971                                         if(d<1.0)
972                                                 coeff = cos(d*PI);
973                                         else
974                                                 coeff=-1.0;
975                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
976                                         else            coeff=  pow( coeff, A);
977                                         coeff= coeff*0.5 + 0.5;
978                                 }
979                                 else if(flags & SWS_AREA)
980                                 {
981                                         double srcPixelSize= 1.0/xInc1;
982                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
983                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
984                                         else coeff=0.0;
985                                 }
986                                 else if(flags & SWS_GAUSS)
987                                 {
988                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
989                                         coeff = pow(2.0, - p*d*d);
990                                 }
991                                 else if(flags & SWS_SINC)
992                                 {
993                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
994                                 }
995                                 else if(flags & SWS_LANCZOS)
996                                 {
997                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
998                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
999                                         if(d>p) coeff=0;
1000                                 }
1001                                 else if(flags & SWS_BILINEAR)
1002                                 {
1003                                         coeff= 1.0 - d;
1004                                         if(coeff<0) coeff=0;
1005                                 }
1006                                 else if(flags & SWS_SPLINE)
1007                                 {
1008                                         double p=-2.196152422706632;
1009                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1010                                 }
1011                                 else {
1012                                         coeff= 0.0; //GCC warning killer
1013                                         ASSERT(0)
1014                                 }
1015
1016                                 filter[i*filterSize + j]= coeff;
1017                                 xx++;
1018                         }
1019                         xDstInSrc+= xInc1;
1020                 }
1021         }
1022
1023         /* apply src & dst Filter to filter -> filter2
1024            free(filter);
1025         */
1026         ASSERT(filterSize>0)
1027         filter2Size= filterSize;
1028         if(srcFilter) filter2Size+= srcFilter->length - 1;
1029         if(dstFilter) filter2Size+= dstFilter->length - 1;
1030         ASSERT(filter2Size>0)
1031         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1032
1033         for(i=0; i<dstW; i++)
1034         {
1035                 int j;
1036                 SwsVector scaleFilter;
1037                 SwsVector *outVec;
1038
1039                 scaleFilter.coeff= filter + i*filterSize;
1040                 scaleFilter.length= filterSize;
1041
1042                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1043                 else          outVec= &scaleFilter;
1044
1045                 ASSERT(outVec->length == filter2Size)
1046                 //FIXME dstFilter
1047
1048                 for(j=0; j<outVec->length; j++)
1049                 {
1050                         filter2[i*filter2Size + j]= outVec->coeff[j];
1051                 }
1052
1053                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1054
1055                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1056         }
1057         free(filter); filter=NULL;
1058
1059         /* try to reduce the filter-size (step1 find size and shift left) */
1060         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1061         minFilterSize= 0;
1062         for(i=dstW-1; i>=0; i--)
1063         {
1064                 int min= filter2Size;
1065                 int j;
1066                 double cutOff=0.0;
1067
1068                 /* get rid off near zero elements on the left by shifting left */
1069                 for(j=0; j<filter2Size; j++)
1070                 {
1071                         int k;
1072                         cutOff += ABS(filter2[i*filter2Size]);
1073
1074                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1075
1076                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1077                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1078
1079                         // Move filter coeffs left
1080                         for(k=1; k<filter2Size; k++)
1081                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1082                         filter2[i*filter2Size + k - 1]= 0.0;
1083                         (*filterPos)[i]++;
1084                 }
1085
1086                 cutOff=0.0;
1087                 /* count near zeros on the right */
1088                 for(j=filter2Size-1; j>0; j--)
1089                 {
1090                         cutOff += ABS(filter2[i*filter2Size + j]);
1091
1092                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1093                         min--;
1094                 }
1095
1096                 if(min>minFilterSize) minFilterSize= min;
1097         }
1098
1099         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1100           // we can handle the special case 4,
1101           // so we don't want to go to the full 8
1102           if (minFilterSize < 5)
1103             filterAlign = 4;
1104
1105           // we really don't want to waste our time
1106           // doing useless computation, so fall-back on
1107           // the scalar C code for very small filter.
1108           // vectorizing is worth it only if you have
1109           // decent-sized vector.
1110           if (minFilterSize < 3)
1111             filterAlign = 1;
1112         }
1113
1114         ASSERT(minFilterSize > 0)
1115         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1116         ASSERT(filterSize > 0)
1117         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1118         *outFilterSize= filterSize;
1119
1120         if(flags&SWS_PRINT_INFO)
1121                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1122         /* try to reduce the filter-size (step2 reduce it) */
1123         for(i=0; i<dstW; i++)
1124         {
1125                 int j;
1126
1127                 for(j=0; j<filterSize; j++)
1128                 {
1129                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1130                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1131                 }
1132         }
1133         free(filter2); filter2=NULL;
1134         
1135
1136         //FIXME try to align filterpos if possible
1137
1138         //fix borders
1139         for(i=0; i<dstW; i++)
1140         {
1141                 int j;
1142                 if((*filterPos)[i] < 0)
1143                 {
1144                         // Move filter coeffs left to compensate for filterPos
1145                         for(j=1; j<filterSize; j++)
1146                         {
1147                                 int left= MAX(j + (*filterPos)[i], 0);
1148                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1149                                 filter[i*filterSize + j]=0;
1150                         }
1151                         (*filterPos)[i]= 0;
1152                 }
1153
1154                 if((*filterPos)[i] + filterSize > srcW)
1155                 {
1156                         int shift= (*filterPos)[i] + filterSize - srcW;
1157                         // Move filter coeffs right to compensate for filterPos
1158                         for(j=filterSize-2; j>=0; j--)
1159                         {
1160                                 int right= MIN(j + shift, filterSize-1);
1161                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1162                                 filter[i*filterSize +j]=0;
1163                         }
1164                         (*filterPos)[i]= srcW - filterSize;
1165                 }
1166         }
1167
1168         // Note the +1 is for the MMXscaler which reads over the end
1169         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1170         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1171
1172         /* Normalize & Store in outFilter */
1173         for(i=0; i<dstW; i++)
1174         {
1175                 int j;
1176                 double error=0;
1177                 double sum=0;
1178                 double scale= one;
1179
1180                 for(j=0; j<filterSize; j++)
1181                 {
1182                         sum+= filter[i*filterSize + j];
1183                 }
1184                 scale/= sum;
1185                 for(j=0; j<*outFilterSize; j++)
1186                 {
1187                         double v= filter[i*filterSize + j]*scale + error;
1188                         int intV= floor(v + 0.5);
1189                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1190                         error = v - intV;
1191                 }
1192         }
1193         
1194         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1195         for(i=0; i<*outFilterSize; i++)
1196         {
1197                 int j= dstW*(*outFilterSize);
1198                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1199         }
1200
1201         free(filter);
1202 }
1203
1204 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1205 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1206 {
1207         uint8_t *fragmentA;
1208         long imm8OfPShufW1A;
1209         long imm8OfPShufW2A;
1210         long fragmentLengthA;
1211         uint8_t *fragmentB;
1212         long imm8OfPShufW1B;
1213         long imm8OfPShufW2B;
1214         long fragmentLengthB;
1215         int fragmentPos;
1216
1217         int xpos, i;
1218
1219         // create an optimized horizontal scaling routine
1220
1221         //code fragment
1222
1223         asm volatile(
1224                 "jmp 9f                         \n\t"
1225         // Begin
1226                 "0:                             \n\t"
1227                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1228                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1229                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1230                 "punpcklbw %%mm7, %%mm1         \n\t"
1231                 "punpcklbw %%mm7, %%mm0         \n\t"
1232                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1233                 "1:                             \n\t"
1234                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1235                 "2:                             \n\t"
1236                 "psubw %%mm1, %%mm0             \n\t"
1237                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1238                 "pmullw %%mm3, %%mm0            \n\t"
1239                 "psllw $7, %%mm1                \n\t"
1240                 "paddw %%mm1, %%mm0             \n\t"
1241
1242                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1243
1244                 "add $8, %%"REG_a"              \n\t"
1245         // End
1246                 "9:                             \n\t"
1247 //              "int $3\n\t"
1248                 "lea 0b, %0                     \n\t"
1249                 "lea 1b, %1                     \n\t"
1250                 "lea 2b, %2                     \n\t"
1251                 "dec %1                         \n\t"
1252                 "dec %2                         \n\t"
1253                 "sub %0, %1                     \n\t"
1254                 "sub %0, %2                     \n\t"
1255                 "lea 9b, %3                     \n\t"
1256                 "sub %0, %3                     \n\t"
1257
1258
1259                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1260                 "=r" (fragmentLengthA)
1261         );
1262
1263         asm volatile(
1264                 "jmp 9f                         \n\t"
1265         // Begin
1266                 "0:                             \n\t"
1267                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1268                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1269                 "punpcklbw %%mm7, %%mm0         \n\t"
1270                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1271                 "1:                             \n\t"
1272                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1273                 "2:                             \n\t"
1274                 "psubw %%mm1, %%mm0             \n\t"
1275                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1276                 "pmullw %%mm3, %%mm0            \n\t"
1277                 "psllw $7, %%mm1                \n\t"
1278                 "paddw %%mm1, %%mm0             \n\t"
1279
1280                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1281
1282                 "add $8, %%"REG_a"              \n\t"
1283         // End
1284                 "9:                             \n\t"
1285 //              "int $3\n\t"
1286                 "lea 0b, %0                     \n\t"
1287                 "lea 1b, %1                     \n\t"
1288                 "lea 2b, %2                     \n\t"
1289                 "dec %1                         \n\t"
1290                 "dec %2                         \n\t"
1291                 "sub %0, %1                     \n\t"
1292                 "sub %0, %2                     \n\t"
1293                 "lea 9b, %3                     \n\t"
1294                 "sub %0, %3                     \n\t"
1295
1296
1297                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1298                 "=r" (fragmentLengthB)
1299         );
1300
1301         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1302         fragmentPos=0;
1303         
1304         for(i=0; i<dstW/numSplits; i++)
1305         {
1306                 int xx=xpos>>16;
1307
1308                 if((i&3) == 0)
1309                 {
1310                         int a=0;
1311                         int b=((xpos+xInc)>>16) - xx;
1312                         int c=((xpos+xInc*2)>>16) - xx;
1313                         int d=((xpos+xInc*3)>>16) - xx;
1314
1315                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1316                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1317                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1318                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1319                         filterPos[i/2]= xx;
1320
1321                         if(d+1<4)
1322                         {
1323                                 int maxShift= 3-(d+1);
1324                                 int shift=0;
1325
1326                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1327
1328                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1329                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1330                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1331                                         a | (b<<2) | (c<<4) | (d<<6);
1332
1333                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1334                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1335
1336                                 if(shift && i>=shift)
1337                                 {
1338                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1339                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1340                                         filterPos[i/2]-=shift;
1341                                 }
1342
1343                                 fragmentPos+= fragmentLengthB;
1344                         }
1345                         else
1346                         {
1347                                 int maxShift= 3-d;
1348                                 int shift=0;
1349
1350                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1351
1352                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1353                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1354                                         a | (b<<2) | (c<<4) | (d<<6);
1355
1356                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1357                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1358
1359                                 if(shift && i>=shift)
1360                                 {
1361                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1362                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1363                                         filterPos[i/2]-=shift;
1364                                 }
1365
1366                                 fragmentPos+= fragmentLengthA;
1367                         }
1368
1369                         funnyCode[fragmentPos]= RET;
1370                 }
1371                 xpos+=xInc;
1372         }
1373         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1374 }
1375 #endif // ARCH_X86 || ARCH_X86_64
1376
1377 static void globalInit(){
1378     // generating tables:
1379     int i;
1380     for(i=0; i<768; i++){
1381         int c= MIN(MAX(i-256, 0), 255);
1382         clip_table[i]=c;
1383     }
1384 }
1385
1386 static SwsFunc getSwsFunc(int flags){
1387     
1388 #ifdef RUNTIME_CPUDETECT
1389 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1390         // ordered per speed fasterst first
1391         if(flags & SWS_CPU_CAPS_MMX2)
1392                 return swScale_MMX2;
1393         else if(flags & SWS_CPU_CAPS_3DNOW)
1394                 return swScale_3DNow;
1395         else if(flags & SWS_CPU_CAPS_MMX)
1396                 return swScale_MMX;
1397         else
1398                 return swScale_C;
1399
1400 #else
1401 #ifdef ARCH_POWERPC
1402         if(flags & SWS_CPU_CAPS_ALTIVEC)
1403           return swScale_altivec;
1404         else
1405           return swScale_C;
1406 #endif
1407         return swScale_C;
1408 #endif
1409 #else //RUNTIME_CPUDETECT
1410 #ifdef HAVE_MMX2
1411         return swScale_MMX2;
1412 #elif defined (HAVE_3DNOW)
1413         return swScale_3DNow;
1414 #elif defined (HAVE_MMX)
1415         return swScale_MMX;
1416 #elif defined (HAVE_ALTIVEC)
1417         return swScale_altivec;
1418 #else
1419         return swScale_C;
1420 #endif
1421 #endif //!RUNTIME_CPUDETECT
1422 }
1423
1424 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1425              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1426         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1427         /* Copy Y plane */
1428         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1429                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1430         else
1431         {
1432                 int i;
1433                 uint8_t *srcPtr= src[0];
1434                 uint8_t *dstPtr= dst;
1435                 for(i=0; i<srcSliceH; i++)
1436                 {
1437                         memcpy(dstPtr, srcPtr, c->srcW);
1438                         srcPtr+= srcStride[0];
1439                         dstPtr+= dstStride[0];
1440                 }
1441         }
1442         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1443         if (c->dstFormat == IMGFMT_NV12)
1444                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1445         else
1446                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1447
1448         return srcSliceH;
1449 }
1450
1451 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1452              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1453         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1454
1455         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1456
1457         return srcSliceH;
1458 }
1459
1460 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1461              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1462         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1463
1464         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1465
1466         return srcSliceH;
1467 }
1468
1469 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1470 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1471                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1472         const int srcFormat= c->srcFormat;
1473         const int dstFormat= c->dstFormat;
1474         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1475         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1476         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1477         const int dstId= (dstFormat&0xFF)>>2;
1478         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1479
1480         /* BGR -> BGR */
1481         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1482            || (isRGB(srcFormat) && isRGB(dstFormat))){
1483                 switch(srcId | (dstId<<4)){
1484                 case 0x34: conv= rgb16to15; break;
1485                 case 0x36: conv= rgb24to15; break;
1486                 case 0x38: conv= rgb32to15; break;
1487                 case 0x43: conv= rgb15to16; break;
1488                 case 0x46: conv= rgb24to16; break;
1489                 case 0x48: conv= rgb32to16; break;
1490                 case 0x63: conv= rgb15to24; break;
1491                 case 0x64: conv= rgb16to24; break;
1492                 case 0x68: conv= rgb32to24; break;
1493                 case 0x83: conv= rgb15to32; break;
1494                 case 0x84: conv= rgb16to32; break;
1495                 case 0x86: conv= rgb24to32; break;
1496                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1497                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1498                 }
1499         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1500                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1501                 switch(srcId | (dstId<<4)){
1502                 case 0x33: conv= rgb15tobgr15; break;
1503                 case 0x34: conv= rgb16tobgr15; break;
1504                 case 0x36: conv= rgb24tobgr15; break;
1505                 case 0x38: conv= rgb32tobgr15; break;
1506                 case 0x43: conv= rgb15tobgr16; break;
1507                 case 0x44: conv= rgb16tobgr16; break;
1508                 case 0x46: conv= rgb24tobgr16; break;
1509                 case 0x48: conv= rgb32tobgr16; break;
1510                 case 0x63: conv= rgb15tobgr24; break;
1511                 case 0x64: conv= rgb16tobgr24; break;
1512                 case 0x66: conv= rgb24tobgr24; break;
1513                 case 0x68: conv= rgb32tobgr24; break;
1514                 case 0x83: conv= rgb15tobgr32; break;
1515                 case 0x84: conv= rgb16tobgr32; break;
1516                 case 0x86: conv= rgb24tobgr32; break;
1517                 case 0x88: conv= rgb32tobgr32; break;
1518                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1519                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1520                 }
1521         }else{
1522                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1523                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1524         }
1525
1526         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1527                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1528         else
1529         {
1530                 int i;
1531                 uint8_t *srcPtr= src[0];
1532                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1533
1534                 for(i=0; i<srcSliceH; i++)
1535                 {
1536                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1537                         srcPtr+= srcStride[0];
1538                         dstPtr+= dstStride[0];
1539                 }
1540         }     
1541         return srcSliceH;
1542 }
1543
1544 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1545              int srcSliceH, uint8_t* dst[], int dstStride[]){
1546
1547         rgb24toyv12(
1548                 src[0], 
1549                 dst[0]+ srcSliceY    *dstStride[0], 
1550                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1551                 dst[2]+(srcSliceY>>1)*dstStride[2],
1552                 c->srcW, srcSliceH, 
1553                 dstStride[0], dstStride[1], srcStride[0]);
1554         return srcSliceH;
1555 }
1556
1557 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1558              int srcSliceH, uint8_t* dst[], int dstStride[]){
1559         int i;
1560
1561         /* copy Y */
1562         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1563                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1564         else{
1565                 uint8_t *srcPtr= src[0];
1566                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1567
1568                 for(i=0; i<srcSliceH; i++)
1569                 {
1570                         memcpy(dstPtr, srcPtr, c->srcW);
1571                         srcPtr+= srcStride[0];
1572                         dstPtr+= dstStride[0];
1573                 }
1574         }
1575
1576         if(c->dstFormat==IMGFMT_YV12){
1577                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1578                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1579         }else{
1580                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1581                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1582         }
1583         return srcSliceH;
1584 }
1585
1586 /**
1587  * bring pointers in YUV order instead of YVU
1588  */
1589 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1590         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1591            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1592                 sortedP[0]= p[0];
1593                 sortedP[1]= p[2];
1594                 sortedP[2]= p[1];
1595                 sortedStride[0]= stride[0];
1596                 sortedStride[1]= stride[2];
1597                 sortedStride[2]= stride[1];
1598         }
1599         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1600         {
1601                 sortedP[0]= p[0];
1602                 sortedP[1]= 
1603                 sortedP[2]= NULL;
1604                 sortedStride[0]= stride[0];
1605                 sortedStride[1]= 
1606                 sortedStride[2]= 0;
1607         }
1608         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1609         {
1610                 sortedP[0]= p[0];
1611                 sortedP[1]= p[1];
1612                 sortedP[2]= p[2];
1613                 sortedStride[0]= stride[0];
1614                 sortedStride[1]= stride[1];
1615                 sortedStride[2]= stride[2];
1616         }
1617         else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
1618         {
1619                 sortedP[0]= p[0];
1620                 sortedP[1]= p[1];
1621                 sortedP[2]= NULL;
1622                 sortedStride[0]= stride[0];
1623                 sortedStride[1]= stride[1];
1624                 sortedStride[2]= 0;
1625         }else{
1626                 MSG_ERR("internal error in orderYUV\n");
1627         }
1628 }
1629
1630 /* unscaled copy like stuff (assumes nearly identical formats) */
1631 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1632              int srcSliceH, uint8_t* dst[], int dstStride[]){
1633
1634         if(isPacked(c->srcFormat))
1635         {
1636                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1637                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1638                 else
1639                 {
1640                         int i;
1641                         uint8_t *srcPtr= src[0];
1642                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1643                         int length=0;
1644
1645                         /* universal length finder */
1646                         while(length+c->srcW <= ABS(dstStride[0]) 
1647                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1648                         ASSERT(length!=0);
1649
1650                         for(i=0; i<srcSliceH; i++)
1651                         {
1652                                 memcpy(dstPtr, srcPtr, length);
1653                                 srcPtr+= srcStride[0];
1654                                 dstPtr+= dstStride[0];
1655                         }
1656                 }
1657         }
1658         else 
1659         { /* Planar YUV or gray */
1660                 int plane;
1661                 for(plane=0; plane<3; plane++)
1662                 {
1663                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1664                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1665                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1666
1667                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1668                         {
1669                                 if(!isGray(c->dstFormat))
1670                                         memset(dst[plane], 128, dstStride[plane]*height);
1671                         }
1672                         else
1673                         {
1674                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1675                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1676                                 else
1677                                 {
1678                                         int i;
1679                                         uint8_t *srcPtr= src[plane];
1680                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1681                                         for(i=0; i<height; i++)
1682                                         {
1683                                                 memcpy(dstPtr, srcPtr, length);
1684                                                 srcPtr+= srcStride[plane];
1685                                                 dstPtr+= dstStride[plane];
1686                                         }
1687                                 }
1688                         }
1689                 }
1690         }
1691         return srcSliceH;
1692 }
1693
1694 static int remove_dup_fourcc(int fourcc)
1695 {
1696         switch(fourcc)
1697         {
1698             case IMGFMT_I420:
1699             case IMGFMT_IYUV: return IMGFMT_YV12;
1700             case IMGFMT_Y8  : return IMGFMT_Y800;
1701             case IMGFMT_IF09: return IMGFMT_YVU9;
1702             default: return fourcc;
1703         }
1704 }
1705
1706 static void getSubSampleFactors(int *h, int *v, int format){
1707         switch(format){
1708         case IMGFMT_UYVY:
1709         case IMGFMT_YUY2:
1710                 *h=1;
1711                 *v=0;
1712                 break;
1713         case IMGFMT_YV12:
1714         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1715         case IMGFMT_NV12:
1716         case IMGFMT_NV21:
1717                 *h=1;
1718                 *v=1;
1719                 break;
1720         case IMGFMT_YVU9:
1721                 *h=2;
1722                 *v=2;
1723                 break;
1724         case IMGFMT_444P:
1725                 *h=0;
1726                 *v=0;
1727                 break;
1728         case IMGFMT_422P:
1729                 *h=1;
1730                 *v=0;
1731                 break;
1732         case IMGFMT_411P:
1733                 *h=2;
1734                 *v=0;
1735                 break;
1736         default:
1737                 *h=0;
1738                 *v=0;
1739                 break;
1740         }
1741 }
1742
1743 static uint16_t roundToInt16(int64_t f){
1744         int r= (f + (1<<15))>>16;
1745              if(r<-0x7FFF) return 0x8000;
1746         else if(r> 0x7FFF) return 0x7FFF;
1747         else               return r;
1748 }
1749
1750 /**
1751  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1752  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1753  * @return -1 if not supported
1754  */
1755 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1756         int64_t crv =  inv_table[0];
1757         int64_t cbu =  inv_table[1];
1758         int64_t cgu = -inv_table[2];
1759         int64_t cgv = -inv_table[3];
1760         int64_t cy  = 1<<16;
1761         int64_t oy  = 0;
1762
1763         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1764         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1765         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1766
1767         c->brightness= brightness;
1768         c->contrast  = contrast;
1769         c->saturation= saturation;
1770         c->srcRange  = srcRange;
1771         c->dstRange  = dstRange;
1772
1773         c->uOffset=   0x0400040004000400LL;
1774         c->vOffset=   0x0400040004000400LL;
1775
1776         if(!srcRange){
1777                 cy= (cy*255) / 219;
1778                 oy= 16<<16;
1779         }
1780
1781         cy = (cy *contrast             )>>16;
1782         crv= (crv*contrast * saturation)>>32;
1783         cbu= (cbu*contrast * saturation)>>32;
1784         cgu= (cgu*contrast * saturation)>>32;
1785         cgv= (cgv*contrast * saturation)>>32;
1786
1787         oy -= 256*brightness;
1788
1789         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1790         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1791         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1792         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1793         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1794         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1795
1796         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1797         //FIXME factorize
1798
1799 #ifdef HAVE_ALTIVEC
1800         yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1801 #endif  
1802         return 0;
1803 }
1804
1805 /**
1806  * @return -1 if not supported
1807  */
1808 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1809         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1810
1811         *inv_table = c->srcColorspaceTable;
1812         *table     = c->dstColorspaceTable;
1813         *srcRange  = c->srcRange;
1814         *dstRange  = c->dstRange;
1815         *brightness= c->brightness;
1816         *contrast  = c->contrast;
1817         *saturation= c->saturation;
1818         
1819         return 0;       
1820 }
1821
1822 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1823                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1824
1825         SwsContext *c;
1826         int i;
1827         int usesVFilter, usesHFilter;
1828         int unscaled, needsDither;
1829         int srcFormat, dstFormat;
1830         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1831 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1832         if(flags & SWS_CPU_CAPS_MMX)
1833                 asm volatile("emms\n\t"::: "memory");
1834 #endif
1835
1836 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1837         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1838 #ifdef HAVE_MMX2
1839         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1840 #elif defined (HAVE_3DNOW)
1841         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1842 #elif defined (HAVE_MMX)
1843         flags |= SWS_CPU_CAPS_MMX;
1844 #elif defined (HAVE_ALTIVEC)
1845         flags |= SWS_CPU_CAPS_ALTIVEC;
1846 #endif
1847 #endif
1848         if(clip_table[512] != 255) globalInit();
1849         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1850
1851         /* avoid duplicate Formats, so we don't need to check to much */
1852         srcFormat = remove_dup_fourcc(origSrcFormat);
1853         dstFormat = remove_dup_fourcc(origDstFormat);
1854
1855         unscaled = (srcW == dstW && srcH == dstH);
1856         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1857                      && (dstFormat&0xFF)<24
1858                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1859
1860         if(!isSupportedIn(srcFormat)) 
1861         {
1862                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1863                 return NULL;
1864         }
1865         if(!isSupportedOut(dstFormat))
1866         {
1867                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1868                 return NULL;
1869         }
1870
1871         /* sanity check */
1872         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
1873         {
1874                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1875                         srcW, srcH, dstW, dstH);
1876                 return NULL;
1877         }
1878
1879         if(!dstFilter) dstFilter= &dummyFilter;
1880         if(!srcFilter) srcFilter= &dummyFilter;
1881
1882         c= memalign(64, sizeof(SwsContext));
1883         memset(c, 0, sizeof(SwsContext));
1884
1885         c->srcW= srcW;
1886         c->srcH= srcH;
1887         c->dstW= dstW;
1888         c->dstH= dstH;
1889         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1890         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1891         c->flags= flags;
1892         c->dstFormat= dstFormat;
1893         c->srcFormat= srcFormat;
1894         c->origDstFormat= origDstFormat;
1895         c->origSrcFormat= origSrcFormat;
1896         c->vRounder= 4* 0x0001000100010001ULL;
1897
1898         usesHFilter= usesVFilter= 0;
1899         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1900         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1901         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1902         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1903         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1904         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1905         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1906         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1907
1908         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1909         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1910
1911         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1912         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1913
1914         // drop some chroma lines if the user wants it
1915         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1916         c->chrSrcVSubSample+= c->vChrDrop;
1917
1918         // drop every 2. pixel for chroma calculation unless user wants full chroma
1919         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1920                 c->chrSrcHSubSample=1;
1921
1922         if(param){
1923                 c->param[0] = param[0];
1924                 c->param[1] = param[1];
1925         }else{
1926                 c->param[0] =
1927                 c->param[1] = SWS_PARAM_DEFAULT;
1928         }
1929
1930         c->chrIntHSubSample= c->chrDstHSubSample;
1931         c->chrIntVSubSample= c->chrSrcVSubSample;
1932
1933         // note the -((-x)>>y) is so that we allways round toward +inf
1934         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1935         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1936         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1937         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1938
1939         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16); 
1940
1941         /* unscaled special Cases */
1942         if(unscaled && !usesHFilter && !usesVFilter)
1943         {
1944                 /* yv12_to_nv12 */
1945                 if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
1946                 {
1947                         c->swScale= PlanarToNV12Wrapper;
1948                 }
1949                 /* yuv2bgr */
1950                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1951                 {
1952                         c->swScale= yuv2rgb_get_func_ptr(c);
1953                 }
1954                 
1955                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1956                 {
1957                         c->swScale= yvu9toyv12Wrapper;
1958                 }
1959
1960                 /* bgr24toYV12 */
1961                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1962                         c->swScale= bgr24toyv12Wrapper;
1963                 
1964                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1965                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1966                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
1967                    && !needsDither)
1968                         c->swScale= rgb2rgbWrapper;
1969
1970                 /* LQ converters if -sws 0 or -sws 4*/
1971                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1972                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1973                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1974                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
1975                           && needsDither)
1976                                 c->swScale= rgb2rgbWrapper;
1977
1978                         /* yv12_to_yuy2 */
1979                         if(srcFormat == IMGFMT_YV12 && 
1980                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1981                         {
1982                                 if (dstFormat == IMGFMT_YUY2)
1983                                     c->swScale= PlanarToYuy2Wrapper;
1984                                 else
1985                                     c->swScale= PlanarToUyvyWrapper;
1986                         }
1987                 }
1988
1989 #ifdef HAVE_ALTIVEC
1990                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
1991                     ((srcFormat == IMGFMT_YV12 && 
1992                       (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
1993                   // unscaled YV12 -> packed YUV, we want speed
1994                   if (dstFormat == IMGFMT_YUY2)
1995                     c->swScale= yv12toyuy2_unscaled_altivec;
1996                   else
1997                     c->swScale= yv12touyvy_unscaled_altivec;
1998                 }
1999 #endif
2000
2001                 /* simple copy */
2002                 if(   srcFormat == dstFormat
2003                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2004                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2005                   )
2006                 {
2007                         c->swScale= simpleCopy;
2008                 }
2009
2010                 if(c->swScale){
2011                         if(flags&SWS_PRINT_INFO)
2012                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2013                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
2014                         return c;
2015                 }
2016         }
2017
2018         if(flags & SWS_CPU_CAPS_MMX2)
2019         {
2020                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2021                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2022                 {
2023                         if(flags&SWS_PRINT_INFO)
2024                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2025                 }
2026                 if(usesHFilter) c->canMMX2BeUsed=0;
2027         }
2028         else
2029                 c->canMMX2BeUsed=0;
2030
2031         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2032         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2033
2034         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2035         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2036         // n-2 is the last chrominance sample available
2037         // this is not perfect, but noone shuld notice the difference, the more correct variant
2038         // would be like the vertical one, but that would require some special code for the
2039         // first and last pixel
2040         if(flags&SWS_FAST_BILINEAR)
2041         {
2042                 if(c->canMMX2BeUsed)
2043                 {
2044                         c->lumXInc+= 20;
2045                         c->chrXInc+= 20;
2046                 }
2047                 //we don't use the x86asm scaler if mmx is available
2048                 else if(flags & SWS_CPU_CAPS_MMX)
2049                 {
2050                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2051                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2052                 }
2053         }
2054
2055         /* precalculate horizontal scaler filter coefficients */
2056         {
2057                 const int filterAlign=
2058                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2059                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2060                   1;
2061
2062                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2063                                  srcW      ,       dstW, filterAlign, 1<<14,
2064                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2065                                  srcFilter->lumH, dstFilter->lumH, c->param);
2066                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2067                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2068                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2069                                  srcFilter->chrH, dstFilter->chrH, c->param);
2070
2071 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2072 // can't downscale !!!
2073                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2074                 {
2075 #define MAX_FUNNY_CODE_SIZE 10000
2076 #ifdef MAP_ANONYMOUS
2077                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2078                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2079 #else
2080                         c->funnyYCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2081                         c->funnyUVCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2082 #endif
2083
2084                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
2085                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
2086                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
2087                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2088
2089                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2090                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2091                 }
2092 #endif
2093         } // Init Horizontal stuff
2094
2095
2096
2097         /* precalculate vertical scaler filter coefficients */
2098         {
2099                 const int filterAlign=
2100                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2101                   1;
2102
2103                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2104                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2105                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2106                                 srcFilter->lumV, dstFilter->lumV, c->param);
2107                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2108                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2109                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2110                                 srcFilter->chrV, dstFilter->chrV, c->param);
2111         }
2112
2113         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2114         c->vLumBufSize= c->vLumFilterSize;
2115         c->vChrBufSize= c->vChrFilterSize;
2116         for(i=0; i<dstH; i++)
2117         {
2118                 int chrI= i*c->chrDstH / dstH;
2119                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2120                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2121
2122                 nextSlice>>= c->chrSrcVSubSample;
2123                 nextSlice<<= c->chrSrcVSubSample;
2124                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2125                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2126                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2127                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2128         }
2129
2130         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2131         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2132         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2133         //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)
2134         for(i=0; i<c->vLumBufSize; i++)
2135                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2136         for(i=0; i<c->vChrBufSize; i++)
2137                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2138
2139         //try to avoid drawing green stuff between the right end and the stride end
2140         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2141         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2142
2143         ASSERT(c->chrDstH <= dstH)
2144
2145         if(flags&SWS_PRINT_INFO)
2146         {
2147 #ifdef DITHER1XBPP
2148                 char *dither= " dithered";
2149 #else
2150                 char *dither= "";
2151 #endif
2152                 if(flags&SWS_FAST_BILINEAR)
2153                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2154                 else if(flags&SWS_BILINEAR)
2155                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2156                 else if(flags&SWS_BICUBIC)
2157                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2158                 else if(flags&SWS_X)
2159                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2160                 else if(flags&SWS_POINT)
2161                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2162                 else if(flags&SWS_AREA)
2163                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2164                 else if(flags&SWS_BICUBLIN)
2165                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2166                 else if(flags&SWS_GAUSS)
2167                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2168                 else if(flags&SWS_SINC)
2169                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2170                 else if(flags&SWS_LANCZOS)
2171                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2172                 else if(flags&SWS_SPLINE)
2173                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2174                 else
2175                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2176
2177                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2178                         MSG_INFO("from %s to%s %s ", 
2179                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2180                 else
2181                         MSG_INFO("from %s to %s ", 
2182                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2183
2184                 if(flags & SWS_CPU_CAPS_MMX2)
2185                         MSG_INFO("using MMX2\n");
2186                 else if(flags & SWS_CPU_CAPS_3DNOW)
2187                         MSG_INFO("using 3DNOW\n");
2188                 else if(flags & SWS_CPU_CAPS_MMX)
2189                         MSG_INFO("using MMX\n");
2190                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2191                         MSG_INFO("using AltiVec\n");
2192                 else 
2193                         MSG_INFO("using C\n");
2194         }
2195
2196         if(flags & SWS_PRINT_INFO)
2197         {
2198                 if(flags & SWS_CPU_CAPS_MMX)
2199                 {
2200                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2201                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2202                         else
2203                         {
2204                                 if(c->hLumFilterSize==4)
2205                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2206                                 else if(c->hLumFilterSize==8)
2207                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2208                                 else
2209                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2210
2211                                 if(c->hChrFilterSize==4)
2212                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2213                                 else if(c->hChrFilterSize==8)
2214                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2215                                 else
2216                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2217                         }
2218                 }
2219                 else
2220                 {
2221 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2222                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2223 #else
2224                         if(flags & SWS_FAST_BILINEAR)
2225                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2226                         else
2227                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2228 #endif
2229                 }
2230                 if(isPlanarYUV(dstFormat))
2231                 {
2232                         if(c->vLumFilterSize==1)
2233                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2234                         else
2235                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2236                 }
2237                 else
2238                 {
2239                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2240                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2241                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2242                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2243                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2244                         else
2245                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2246                 }
2247
2248                 if(dstFormat==IMGFMT_BGR24)
2249                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2250                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2251                 else if(dstFormat==IMGFMT_BGR32)
2252                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2253                 else if(dstFormat==IMGFMT_BGR16)
2254                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2255                 else if(dstFormat==IMGFMT_BGR15)
2256                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2257
2258                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2259         }
2260         if(flags & SWS_PRINT_INFO)
2261         {
2262                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2263                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2264                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2265                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2266         }
2267
2268         c->swScale= getSwsFunc(flags);
2269         return c;
2270 }
2271
2272 /**
2273  * swscale warper, so we don't need to export the SwsContext.
2274  * assumes planar YUV to be in YUV order instead of YVU
2275  */
2276 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2277                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2278         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2279             MSG_ERR("swScaler: slices start in the middle!\n");
2280             return 0;
2281         }
2282         if (c->sliceDir == 0) {
2283             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2284         }
2285
2286         // copy strides, so they can safely be modified
2287         if (c->sliceDir == 1) {
2288             // slices go from top to bottom
2289             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2290             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2291             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2292         } else {
2293             // slices go from bottom to top => we flip the image internally
2294             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2295                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2296                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2297             };
2298             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2299                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2300                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2301             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2302             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2303             
2304             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2305         }
2306 }
2307
2308 /**
2309  * swscale warper, so we don't need to export the SwsContext
2310  */
2311 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2312                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2313         int srcStride[3];
2314         int dstStride[3];
2315         uint8_t *src[3];
2316         uint8_t *dst[3];
2317         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2318         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2319 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2320
2321         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2322 }
2323
2324 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2325                                 float lumaSharpen, float chromaSharpen,
2326                                 float chromaHShift, float chromaVShift,
2327                                 int verbose)
2328 {
2329         SwsFilter *filter= malloc(sizeof(SwsFilter));
2330
2331         if(lumaGBlur!=0.0){
2332                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2333                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2334         }else{
2335                 filter->lumH= sws_getIdentityVec();
2336                 filter->lumV= sws_getIdentityVec();
2337         }
2338
2339         if(chromaGBlur!=0.0){
2340                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2341                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2342         }else{
2343                 filter->chrH= sws_getIdentityVec();
2344                 filter->chrV= sws_getIdentityVec();
2345         }
2346
2347         if(chromaSharpen!=0.0){
2348                 SwsVector *g= sws_getConstVec(-1.0, 3);
2349                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2350                 g->coeff[1]=2.0;
2351                 sws_addVec(id, g);
2352                 sws_convVec(filter->chrH, id);
2353                 sws_convVec(filter->chrV, id);
2354                 sws_freeVec(g);
2355                 sws_freeVec(id);
2356         }
2357
2358         if(lumaSharpen!=0.0){
2359                 SwsVector *g= sws_getConstVec(-1.0, 3);
2360                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2361                 g->coeff[1]=2.0;
2362                 sws_addVec(id, g);
2363                 sws_convVec(filter->lumH, id);
2364                 sws_convVec(filter->lumV, id);
2365                 sws_freeVec(g);
2366                 sws_freeVec(id);
2367         }
2368
2369         if(chromaHShift != 0.0)
2370                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2371
2372         if(chromaVShift != 0.0)
2373                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2374
2375         sws_normalizeVec(filter->chrH, 1.0);
2376         sws_normalizeVec(filter->chrV, 1.0);
2377         sws_normalizeVec(filter->lumH, 1.0);
2378         sws_normalizeVec(filter->lumV, 1.0);
2379
2380         if(verbose) sws_printVec(filter->chrH);
2381         if(verbose) sws_printVec(filter->lumH);
2382
2383         return filter;
2384 }
2385
2386 /**
2387  * returns a normalized gaussian curve used to filter stuff
2388  * quality=3 is high quality, lowwer is lowwer quality
2389  */
2390 SwsVector *sws_getGaussianVec(double variance, double quality){
2391         const int length= (int)(variance*quality + 0.5) | 1;
2392         int i;
2393         double *coeff= memalign(sizeof(double), length*sizeof(double));
2394         double middle= (length-1)*0.5;
2395         SwsVector *vec= malloc(sizeof(SwsVector));
2396
2397         vec->coeff= coeff;
2398         vec->length= length;
2399
2400         for(i=0; i<length; i++)
2401         {
2402                 double dist= i-middle;
2403                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2404         }
2405
2406         sws_normalizeVec(vec, 1.0);
2407
2408         return vec;
2409 }
2410
2411 SwsVector *sws_getConstVec(double c, int length){
2412         int i;
2413         double *coeff= memalign(sizeof(double), length*sizeof(double));
2414         SwsVector *vec= malloc(sizeof(SwsVector));
2415
2416         vec->coeff= coeff;
2417         vec->length= length;
2418
2419         for(i=0; i<length; i++)
2420                 coeff[i]= c;
2421
2422         return vec;
2423 }
2424
2425
2426 SwsVector *sws_getIdentityVec(void){
2427         double *coeff= memalign(sizeof(double), sizeof(double));
2428         SwsVector *vec= malloc(sizeof(SwsVector));
2429         coeff[0]= 1.0;
2430
2431         vec->coeff= coeff;
2432         vec->length= 1;
2433
2434         return vec;
2435 }
2436
2437 void sws_normalizeVec(SwsVector *a, double height){
2438         int i;
2439         double sum=0;
2440         double inv;
2441
2442         for(i=0; i<a->length; i++)
2443                 sum+= a->coeff[i];
2444
2445         inv= height/sum;
2446
2447         for(i=0; i<a->length; i++)
2448                 a->coeff[i]*= inv;
2449 }
2450
2451 void sws_scaleVec(SwsVector *a, double scalar){
2452         int i;
2453
2454         for(i=0; i<a->length; i++)
2455                 a->coeff[i]*= scalar;
2456 }
2457
2458 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2459         int length= a->length + b->length - 1;
2460         double *coeff= memalign(sizeof(double), length*sizeof(double));
2461         int i, j;
2462         SwsVector *vec= malloc(sizeof(SwsVector));
2463
2464         vec->coeff= coeff;
2465         vec->length= length;
2466
2467         for(i=0; i<length; i++) coeff[i]= 0.0;
2468
2469         for(i=0; i<a->length; i++)
2470         {
2471                 for(j=0; j<b->length; j++)
2472                 {
2473                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2474                 }
2475         }
2476
2477         return vec;
2478 }
2479
2480 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2481         int length= MAX(a->length, b->length);
2482         double *coeff= memalign(sizeof(double), length*sizeof(double));
2483         int i;
2484         SwsVector *vec= malloc(sizeof(SwsVector));
2485
2486         vec->coeff= coeff;
2487         vec->length= length;
2488
2489         for(i=0; i<length; i++) coeff[i]= 0.0;
2490
2491         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2492         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2493
2494         return vec;
2495 }
2496
2497 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2498         int length= MAX(a->length, b->length);
2499         double *coeff= memalign(sizeof(double), length*sizeof(double));
2500         int i;
2501         SwsVector *vec= malloc(sizeof(SwsVector));
2502
2503         vec->coeff= coeff;
2504         vec->length= length;
2505
2506         for(i=0; i<length; i++) coeff[i]= 0.0;
2507
2508         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2509         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2510
2511         return vec;
2512 }
2513
2514 /* shift left / or right if "shift" is negative */
2515 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2516         int length= a->length + ABS(shift)*2;
2517         double *coeff= memalign(sizeof(double), length*sizeof(double));
2518         int i;
2519         SwsVector *vec= malloc(sizeof(SwsVector));
2520
2521         vec->coeff= coeff;
2522         vec->length= length;
2523
2524         for(i=0; i<length; i++) coeff[i]= 0.0;
2525
2526         for(i=0; i<a->length; i++)
2527         {
2528                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2529         }
2530
2531         return vec;
2532 }
2533
2534 void sws_shiftVec(SwsVector *a, int shift){
2535         SwsVector *shifted= sws_getShiftedVec(a, shift);
2536         free(a->coeff);
2537         a->coeff= shifted->coeff;
2538         a->length= shifted->length;
2539         free(shifted);
2540 }
2541
2542 void sws_addVec(SwsVector *a, SwsVector *b){
2543         SwsVector *sum= sws_sumVec(a, b);
2544         free(a->coeff);
2545         a->coeff= sum->coeff;
2546         a->length= sum->length;
2547         free(sum);
2548 }
2549
2550 void sws_subVec(SwsVector *a, SwsVector *b){
2551         SwsVector *diff= sws_diffVec(a, b);
2552         free(a->coeff);
2553         a->coeff= diff->coeff;
2554         a->length= diff->length;
2555         free(diff);
2556 }
2557
2558 void sws_convVec(SwsVector *a, SwsVector *b){
2559         SwsVector *conv= sws_getConvVec(a, b);
2560         free(a->coeff);  
2561         a->coeff= conv->coeff;
2562         a->length= conv->length;
2563         free(conv);
2564 }
2565
2566 SwsVector *sws_cloneVec(SwsVector *a){
2567         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2568         int i;
2569         SwsVector *vec= malloc(sizeof(SwsVector));
2570
2571         vec->coeff= coeff;
2572         vec->length= a->length;
2573
2574         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2575
2576         return vec;
2577 }
2578
2579 void sws_printVec(SwsVector *a){
2580         int i;
2581         double max=0;
2582         double min=0;
2583         double range;
2584
2585         for(i=0; i<a->length; i++)
2586                 if(a->coeff[i]>max) max= a->coeff[i];
2587
2588         for(i=0; i<a->length; i++)
2589                 if(a->coeff[i]<min) min= a->coeff[i];
2590
2591         range= max - min;
2592
2593         for(i=0; i<a->length; i++)
2594         {
2595                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2596                 MSG_DBG2("%1.3f ", a->coeff[i]);
2597                 for(;x>0; x--) MSG_DBG2(" ");
2598                 MSG_DBG2("|\n");
2599         }
2600 }
2601
2602 void sws_freeVec(SwsVector *a){
2603         if(!a) return;
2604         if(a->coeff) free(a->coeff);
2605         a->coeff=NULL;
2606         a->length=0;
2607         free(a);
2608 }
2609
2610 void sws_freeFilter(SwsFilter *filter){
2611         if(!filter) return;
2612
2613         if(filter->lumH) sws_freeVec(filter->lumH);
2614         if(filter->lumV) sws_freeVec(filter->lumV);
2615         if(filter->chrH) sws_freeVec(filter->chrH);
2616         if(filter->chrV) sws_freeVec(filter->chrV);
2617         free(filter);
2618 }
2619
2620
2621 void sws_freeContext(SwsContext *c){
2622         int i;
2623         if(!c) return;
2624
2625         if(c->lumPixBuf)
2626         {
2627                 for(i=0; i<c->vLumBufSize; i++)
2628                 {
2629                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2630                         c->lumPixBuf[i]=NULL;
2631                 }
2632                 free(c->lumPixBuf);
2633                 c->lumPixBuf=NULL;
2634         }
2635
2636         if(c->chrPixBuf)
2637         {
2638                 for(i=0; i<c->vChrBufSize; i++)
2639                 {
2640                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2641                         c->chrPixBuf[i]=NULL;
2642                 }
2643                 free(c->chrPixBuf);
2644                 c->chrPixBuf=NULL;
2645         }
2646
2647         if(c->vLumFilter) free(c->vLumFilter);
2648         c->vLumFilter = NULL;
2649         if(c->vChrFilter) free(c->vChrFilter);
2650         c->vChrFilter = NULL;
2651         if(c->hLumFilter) free(c->hLumFilter);
2652         c->hLumFilter = NULL;
2653         if(c->hChrFilter) free(c->hChrFilter);
2654         c->hChrFilter = NULL;
2655
2656         if(c->vLumFilterPos) free(c->vLumFilterPos);
2657         c->vLumFilterPos = NULL;
2658         if(c->vChrFilterPos) free(c->vChrFilterPos);
2659         c->vChrFilterPos = NULL;
2660         if(c->hLumFilterPos) free(c->hLumFilterPos);
2661         c->hLumFilterPos = NULL;
2662         if(c->hChrFilterPos) free(c->hChrFilterPos);
2663         c->hChrFilterPos = NULL;
2664
2665 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2666 #ifdef MAP_ANONYMOUS
2667         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2668         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2669 #else
2670         if(c->funnyYCode) free(c->funnyYCode);
2671         if(c->funnyUVCode) free(c->funnyUVCode);
2672 #endif
2673         c->funnyYCode=NULL;
2674         c->funnyUVCode=NULL;
2675 #endif
2676
2677         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2678         c->lumMmx2Filter=NULL;
2679         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2680         c->chrMmx2Filter=NULL;
2681         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2682         c->lumMmx2FilterPos=NULL;
2683         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2684         c->chrMmx2FilterPos=NULL;
2685         if(c->yuvTable) free(c->yuvTable);
2686         c->yuvTable=NULL;
2687
2688         free(c);
2689 }
2690