Pass a context to av_log(), when possible
[ffmpeg.git] / libswscale / swscale.c
1 /*
2  * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation; either version 2 of the License, or
9  * (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14  * GNU General Public License for more details.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19  *
20  * the C code (not assembly, mmx, ...) of this file can be used
21  * under the LGPL license too
22  */
23
24 /*
25   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
26   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27   {BGR,RGB}{1,4,8,15,16} support dithering
28   
29   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
31   x -> x
32   YUV9 -> YV12
33   YUV9/YV12 -> Y800
34   Y800 -> YUV9/YV12
35   BGR24 -> BGR32 & RGB24 -> RGB32
36   BGR32 -> BGR24 & RGB32 -> RGB24
37   BGR15 -> BGR16
38 */
39
40 /* 
41 tested special converters (most are tested actually but i didnt write it down ...)
42  YV12 -> BGR16
43  YV12 -> YV12
44  BGR15 -> BGR16
45  BGR16 -> BGR16
46  YVU9 -> YV12
47
48 untested special converters
49   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
50   YV12/I420 -> YV12/I420
51   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52   BGR24 -> BGR32 & RGB24 -> RGB32
53   BGR32 -> BGR24 & RGB32 -> RGB24
54   BGR24 -> YV12
55 */
56
57 #include <inttypes.h>
58 #include <string.h>
59 #include <math.h>
60 #include <stdio.h>
61 #include <unistd.h>
62 #include "config.h"
63 #include <assert.h>
64 #ifdef HAVE_SYS_MMAN_H
65 #include <sys/mman.h>
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
68 #endif
69 #endif
70 #include "swscale.h"
71 #include "swscale_internal.h"
72 #include "x86_cpu.h"
73 #include "bswap.h"
74 #include "rgb2rgb.h"
75 #ifdef USE_FASTMEMCPY
76 #include "libvo/fastmemcpy.h"
77 #endif
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 #define isSupportedIn(x)  ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
106                         || (x)==PIX_FMT_RGB32|| (x)==PIX_FMT_BGR24|| (x)==PIX_FMT_BGR565|| (x)==PIX_FMT_BGR555\
107                         || (x)==PIX_FMT_BGR32|| (x)==PIX_FMT_RGB24|| (x)==PIX_FMT_RGB565|| (x)==PIX_FMT_RGB555\
108                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P\
109                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
110                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P)
111 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
112                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
113                         || isRGB(x) || isBGR(x)\
114                         || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
115                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
116                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
117 #define isPacked(x)    ((x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422 ||isRGB(x) || isBGR(x))
118
119 #define RGB2YUV_SHIFT 16
120 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
121 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
122 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
123 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
124 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
125 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
126 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
127 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
129
130 extern const int32_t Inverse_Table_6_9[8][4];
131
132 /*
133 NOTES
134 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
135
136 TODO
137 more intelligent missalignment avoidance for the horizontal scaler
138 write special vertical cubic upscale version
139 Optimize C code (yv12 / minmax)
140 add support for packed pixel yuv input & output
141 add support for Y8 output
142 optimize bgr24 & bgr32
143 add BGR4 output support
144 write special BGR->BGR scaler
145 */
146
147 #if defined(ARCH_X86) && defined (CONFIG_GPL)
148 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
149 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
150 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
151 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
152 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
153 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
154 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
155 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
156
157 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
158 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
159 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
160 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
161
162 static uint64_t __attribute__((aligned(8))) dither4[2]={
163         0x0103010301030103LL,
164         0x0200020002000200LL,};
165
166 static uint64_t __attribute__((aligned(8))) dither8[2]={
167         0x0602060206020602LL,
168         0x0004000400040004LL,};
169
170 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
171 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
172 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
173 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
174 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
175 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
176
177 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
178 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
179 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
180
181 #ifdef FAST_BGR2YV12
182 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
183 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
184 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
185 #else
186 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
187 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
188 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
189 #endif /* FAST_BGR2YV12 */
190 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
191 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
192 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
193 #endif /* defined(ARCH_X86) */
194
195 // clipping helper table for C implementations:
196 static unsigned char clip_table[768];
197
198 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
199                   
200 extern const uint8_t dither_2x2_4[2][8];
201 extern const uint8_t dither_2x2_8[2][8];
202 extern const uint8_t dither_8x8_32[8][8];
203 extern const uint8_t dither_8x8_73[8][8];
204 extern const uint8_t dither_8x8_220[8][8];
205
206 static const char * sws_context_to_name(void * ptr) {
207     return "swscaler";
208 }
209
210 static AVClass sws_context_class = { "SWScaler", sws_context_to_name, NULL };
211
212 char *sws_format_name(enum PixelFormat format)
213 {
214     switch (format) {
215         case PIX_FMT_YUV420P:
216             return "yuv420p";
217         case PIX_FMT_YUYV422:
218             return "yuyv422";
219         case PIX_FMT_RGB24:
220             return "rgb24";
221         case PIX_FMT_BGR24:
222             return "bgr24";
223         case PIX_FMT_YUV422P:
224             return "yuv422p";
225         case PIX_FMT_YUV444P:
226             return "yuv444p";
227         case PIX_FMT_RGB32:
228             return "rgb32";
229         case PIX_FMT_YUV410P:
230             return "yuv410p";
231         case PIX_FMT_YUV411P:
232             return "yuv411p";
233         case PIX_FMT_RGB565:
234             return "rgb565";
235         case PIX_FMT_RGB555:
236             return "rgb555";
237         case PIX_FMT_GRAY16BE:
238             return "gray16be";
239         case PIX_FMT_GRAY16LE:
240             return "gray16le";
241         case PIX_FMT_GRAY8:
242             return "gray8";
243         case PIX_FMT_MONOWHITE:
244             return "mono white";
245         case PIX_FMT_MONOBLACK:
246             return "mono black";
247         case PIX_FMT_PAL8:
248             return "Palette";
249         case PIX_FMT_YUVJ420P:
250             return "yuvj420p";
251         case PIX_FMT_YUVJ422P:
252             return "yuvj422p";
253         case PIX_FMT_YUVJ444P:
254             return "yuvj444p";
255         case PIX_FMT_XVMC_MPEG2_MC:
256             return "xvmc_mpeg2_mc";
257         case PIX_FMT_XVMC_MPEG2_IDCT:
258             return "xvmc_mpeg2_idct";
259         case PIX_FMT_UYVY422:
260             return "uyvy422";
261         case PIX_FMT_UYYVYY411:
262             return "uyyvyy411";
263         case PIX_FMT_RGB32_1:
264             return "rgb32x";
265         case PIX_FMT_BGR32_1:
266             return "bgr32x";
267         case PIX_FMT_BGR32:
268             return "bgr32";
269         case PIX_FMT_BGR565:
270             return "bgr565";
271         case PIX_FMT_BGR555:
272             return "bgr555";
273         case PIX_FMT_BGR8:
274             return "bgr8";
275         case PIX_FMT_BGR4:
276             return "bgr4";
277         case PIX_FMT_BGR4_BYTE:
278             return "bgr4 byte";
279         case PIX_FMT_RGB8:
280             return "rgb8";
281         case PIX_FMT_RGB4:
282             return "rgb4";
283         case PIX_FMT_RGB4_BYTE:
284             return "rgb4 byte";
285         case PIX_FMT_NV12:
286             return "nv12";
287         case PIX_FMT_NV21:
288             return "nv21";
289         default:
290             return "Unknown format";
291     }
292 }
293
294 #if defined(ARCH_X86) && defined (CONFIG_GPL)
295 void in_asm_used_var_warning_killer()
296 {
297  volatile int i= bF8+bFC+w10+
298  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
299  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
300  if(i) i=0;
301 }
302 #endif
303
304 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
305                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
306                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
307 {
308         //FIXME Optimize (just quickly writen not opti..)
309         int i;
310         for(i=0; i<dstW; i++)
311         {
312                 int val=1<<18;
313                 int j;
314                 for(j=0; j<lumFilterSize; j++)
315                         val += lumSrc[j][i] * lumFilter[j];
316
317                 dest[i]= clip_uint8(val>>19);
318         }
319
320         if(uDest != NULL)
321                 for(i=0; i<chrDstW; i++)
322                 {
323                         int u=1<<18;
324                         int v=1<<18;
325                         int j;
326                         for(j=0; j<chrFilterSize; j++)
327                         {
328                                 u += chrSrc[j][i] * chrFilter[j];
329                                 v += chrSrc[j][i + 2048] * chrFilter[j];
330                         }
331
332                         uDest[i]= clip_uint8(u>>19);
333                         vDest[i]= clip_uint8(v>>19);
334                 }
335 }
336
337 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
338                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
339                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
340 {
341         //FIXME Optimize (just quickly writen not opti..)
342         int i;
343         for(i=0; i<dstW; i++)
344         {
345                 int val=1<<18;
346                 int j;
347                 for(j=0; j<lumFilterSize; j++)
348                         val += lumSrc[j][i] * lumFilter[j];
349
350                 dest[i]= clip_uint8(val>>19);
351         }
352
353         if(uDest == NULL)
354                 return;
355
356         if(dstFormat == PIX_FMT_NV12)
357                 for(i=0; i<chrDstW; i++)
358                 {
359                         int u=1<<18;
360                         int v=1<<18;
361                         int j;
362                         for(j=0; j<chrFilterSize; j++)
363                         {
364                                 u += chrSrc[j][i] * chrFilter[j];
365                                 v += chrSrc[j][i + 2048] * chrFilter[j];
366                         }
367
368                         uDest[2*i]= clip_uint8(u>>19);
369                         uDest[2*i+1]= clip_uint8(v>>19);
370                 }
371         else
372                 for(i=0; i<chrDstW; i++)
373                 {
374                         int u=1<<18;
375                         int v=1<<18;
376                         int j;
377                         for(j=0; j<chrFilterSize; j++)
378                         {
379                                 u += chrSrc[j][i] * chrFilter[j];
380                                 v += chrSrc[j][i + 2048] * chrFilter[j];
381                         }
382
383                         uDest[2*i]= clip_uint8(v>>19);
384                         uDest[2*i+1]= clip_uint8(u>>19);
385                 }
386 }
387
388 #define YSCALE_YUV_2_PACKEDX_C(type) \
389                 for(i=0; i<(dstW>>1); i++){\
390                         int j;\
391                         int Y1=1<<18;\
392                         int Y2=1<<18;\
393                         int U=1<<18;\
394                         int V=1<<18;\
395                         type attribute_unused *r, *b, *g;\
396                         const int i2= 2*i;\
397                         \
398                         for(j=0; j<lumFilterSize; j++)\
399                         {\
400                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
401                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
402                         }\
403                         for(j=0; j<chrFilterSize; j++)\
404                         {\
405                                 U += chrSrc[j][i] * chrFilter[j];\
406                                 V += chrSrc[j][i+2048] * chrFilter[j];\
407                         }\
408                         Y1>>=19;\
409                         Y2>>=19;\
410                         U >>=19;\
411                         V >>=19;\
412                         if((Y1|Y2|U|V)&256)\
413                         {\
414                                 if(Y1>255)   Y1=255;\
415                                 else if(Y1<0)Y1=0;\
416                                 if(Y2>255)   Y2=255;\
417                                 else if(Y2<0)Y2=0;\
418                                 if(U>255)    U=255;\
419                                 else if(U<0) U=0;\
420                                 if(V>255)    V=255;\
421                                 else if(V<0) V=0;\
422                         }
423                         
424 #define YSCALE_YUV_2_RGBX_C(type) \
425                         YSCALE_YUV_2_PACKEDX_C(type)\
426                         r = (type *)c->table_rV[V];\
427                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
428                         b = (type *)c->table_bU[U];\
429
430 #define YSCALE_YUV_2_PACKED2_C \
431                 for(i=0; i<(dstW>>1); i++){\
432                         const int i2= 2*i;\
433                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
434                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
435                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
436                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
437
438 #define YSCALE_YUV_2_RGB2_C(type) \
439                         YSCALE_YUV_2_PACKED2_C\
440                         type *r, *b, *g;\
441                         r = (type *)c->table_rV[V];\
442                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
443                         b = (type *)c->table_bU[U];\
444
445 #define YSCALE_YUV_2_PACKED1_C \
446                 for(i=0; i<(dstW>>1); i++){\
447                         const int i2= 2*i;\
448                         int Y1= buf0[i2  ]>>7;\
449                         int Y2= buf0[i2+1]>>7;\
450                         int U= (uvbuf1[i     ])>>7;\
451                         int V= (uvbuf1[i+2048])>>7;\
452
453 #define YSCALE_YUV_2_RGB1_C(type) \
454                         YSCALE_YUV_2_PACKED1_C\
455                         type *r, *b, *g;\
456                         r = (type *)c->table_rV[V];\
457                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
458                         b = (type *)c->table_bU[U];\
459
460 #define YSCALE_YUV_2_PACKED1B_C \
461                 for(i=0; i<(dstW>>1); i++){\
462                         const int i2= 2*i;\
463                         int Y1= buf0[i2  ]>>7;\
464                         int Y2= buf0[i2+1]>>7;\
465                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
466                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
467
468 #define YSCALE_YUV_2_RGB1B_C(type) \
469                         YSCALE_YUV_2_PACKED1B_C\
470                         type *r, *b, *g;\
471                         r = (type *)c->table_rV[V];\
472                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
473                         b = (type *)c->table_bU[U];\
474
475 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
476         switch(c->dstFormat)\
477         {\
478         case PIX_FMT_RGB32:\
479         case PIX_FMT_BGR32:\
480                 func(uint32_t)\
481                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
482                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
483                 }               \
484                 break;\
485         case PIX_FMT_RGB24:\
486                 func(uint8_t)\
487                         ((uint8_t*)dest)[0]= r[Y1];\
488                         ((uint8_t*)dest)[1]= g[Y1];\
489                         ((uint8_t*)dest)[2]= b[Y1];\
490                         ((uint8_t*)dest)[3]= r[Y2];\
491                         ((uint8_t*)dest)[4]= g[Y2];\
492                         ((uint8_t*)dest)[5]= b[Y2];\
493                         dest+=6;\
494                 }\
495                 break;\
496         case PIX_FMT_BGR24:\
497                 func(uint8_t)\
498                         ((uint8_t*)dest)[0]= b[Y1];\
499                         ((uint8_t*)dest)[1]= g[Y1];\
500                         ((uint8_t*)dest)[2]= r[Y1];\
501                         ((uint8_t*)dest)[3]= b[Y2];\
502                         ((uint8_t*)dest)[4]= g[Y2];\
503                         ((uint8_t*)dest)[5]= r[Y2];\
504                         dest+=6;\
505                 }\
506                 break;\
507         case PIX_FMT_RGB565:\
508         case PIX_FMT_BGR565:\
509                 {\
510                         const int dr1= dither_2x2_8[y&1    ][0];\
511                         const int dg1= dither_2x2_4[y&1    ][0];\
512                         const int db1= dither_2x2_8[(y&1)^1][0];\
513                         const int dr2= dither_2x2_8[y&1    ][1];\
514                         const int dg2= dither_2x2_4[y&1    ][1];\
515                         const int db2= dither_2x2_8[(y&1)^1][1];\
516                         func(uint16_t)\
517                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
518                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
519                         }\
520                 }\
521                 break;\
522         case PIX_FMT_RGB555:\
523         case PIX_FMT_BGR555:\
524                 {\
525                         const int dr1= dither_2x2_8[y&1    ][0];\
526                         const int dg1= dither_2x2_8[y&1    ][1];\
527                         const int db1= dither_2x2_8[(y&1)^1][0];\
528                         const int dr2= dither_2x2_8[y&1    ][1];\
529                         const int dg2= dither_2x2_8[y&1    ][0];\
530                         const int db2= dither_2x2_8[(y&1)^1][1];\
531                         func(uint16_t)\
532                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
533                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
534                         }\
535                 }\
536                 break;\
537         case PIX_FMT_RGB8:\
538         case PIX_FMT_BGR8:\
539                 {\
540                         const uint8_t * const d64= dither_8x8_73[y&7];\
541                         const uint8_t * const d32= dither_8x8_32[y&7];\
542                         func(uint8_t)\
543                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
544                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
545                         }\
546                 }\
547                 break;\
548         case PIX_FMT_RGB4:\
549         case PIX_FMT_BGR4:\
550                 {\
551                         const uint8_t * const d64= dither_8x8_73 [y&7];\
552                         const uint8_t * const d128=dither_8x8_220[y&7];\
553                         func(uint8_t)\
554                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
555                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
556                         }\
557                 }\
558                 break;\
559         case PIX_FMT_RGB4_BYTE:\
560         case PIX_FMT_BGR4_BYTE:\
561                 {\
562                         const uint8_t * const d64= dither_8x8_73 [y&7];\
563                         const uint8_t * const d128=dither_8x8_220[y&7];\
564                         func(uint8_t)\
565                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
566                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
567                         }\
568                 }\
569                 break;\
570         case PIX_FMT_MONOBLACK:\
571                 {\
572                         const uint8_t * const d128=dither_8x8_220[y&7];\
573                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
574                         for(i=0; i<dstW-7; i+=8){\
575                                 int acc;\
576                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
577                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
578                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
579                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
580                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
581                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
582                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
583                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
584                                 ((uint8_t*)dest)[0]= acc;\
585                                 dest++;\
586                         }\
587 \
588 /*\
589 ((uint8_t*)dest)-= dstW>>4;\
590 {\
591                         int acc=0;\
592                         int left=0;\
593                         static int top[1024];\
594                         static int last_new[1024][1024];\
595                         static int last_in3[1024][1024];\
596                         static int drift[1024][1024];\
597                         int topLeft=0;\
598                         int shift=0;\
599                         int count=0;\
600                         const uint8_t * const d128=dither_8x8_220[y&7];\
601                         int error_new=0;\
602                         int error_in3=0;\
603                         int f=0;\
604                         \
605                         for(i=dstW>>1; i<dstW; i++){\
606                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
607                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
608                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
609                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
610                                         + (last_new[y][i] - in3)*f/256;\
611                                 int new= old> 128 ? 255 : 0;\
612 \
613                                 error_new+= FFABS(last_new[y][i] - new);\
614                                 error_in3+= FFABS(last_in3[y][i] - in3);\
615                                 f= error_new - error_in3*4;\
616                                 if(f<0) f=0;\
617                                 if(f>256) f=256;\
618 \
619                                 topLeft= top[i];\
620                                 left= top[i]= old - new;\
621                                 last_new[y][i]= new;\
622                                 last_in3[y][i]= in3;\
623 \
624                                 acc+= acc + (new&1);\
625                                 if((i&7)==6){\
626                                         ((uint8_t*)dest)[0]= acc;\
627                                         ((uint8_t*)dest)++;\
628                                 }\
629                         }\
630 }\
631 */\
632                 }\
633                 break;\
634         case PIX_FMT_YUYV422:\
635                 func2\
636                         ((uint8_t*)dest)[2*i2+0]= Y1;\
637                         ((uint8_t*)dest)[2*i2+1]= U;\
638                         ((uint8_t*)dest)[2*i2+2]= Y2;\
639                         ((uint8_t*)dest)[2*i2+3]= V;\
640                 }               \
641                 break;\
642         case PIX_FMT_UYVY422:\
643                 func2\
644                         ((uint8_t*)dest)[2*i2+0]= U;\
645                         ((uint8_t*)dest)[2*i2+1]= Y1;\
646                         ((uint8_t*)dest)[2*i2+2]= V;\
647                         ((uint8_t*)dest)[2*i2+3]= Y2;\
648                 }               \
649                 break;\
650         }\
651
652
653 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
654                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
655                                     uint8_t *dest, int dstW, int y)
656 {
657         int i;
658         switch(c->dstFormat)
659         {
660         case PIX_FMT_BGR32:
661         case PIX_FMT_RGB32:
662                 YSCALE_YUV_2_RGBX_C(uint32_t)
663                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
664                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
665                 }
666                 break;
667         case PIX_FMT_RGB24:
668                 YSCALE_YUV_2_RGBX_C(uint8_t)
669                         ((uint8_t*)dest)[0]= r[Y1];
670                         ((uint8_t*)dest)[1]= g[Y1];
671                         ((uint8_t*)dest)[2]= b[Y1];
672                         ((uint8_t*)dest)[3]= r[Y2];
673                         ((uint8_t*)dest)[4]= g[Y2];
674                         ((uint8_t*)dest)[5]= b[Y2];
675                         dest+=6;
676                 }
677                 break;
678         case PIX_FMT_BGR24:
679                 YSCALE_YUV_2_RGBX_C(uint8_t)
680                         ((uint8_t*)dest)[0]= b[Y1];
681                         ((uint8_t*)dest)[1]= g[Y1];
682                         ((uint8_t*)dest)[2]= r[Y1];
683                         ((uint8_t*)dest)[3]= b[Y2];
684                         ((uint8_t*)dest)[4]= g[Y2];
685                         ((uint8_t*)dest)[5]= r[Y2];
686                         dest+=6;
687                 }
688                 break;
689         case PIX_FMT_RGB565:
690         case PIX_FMT_BGR565:
691                 {
692                         const int dr1= dither_2x2_8[y&1    ][0];
693                         const int dg1= dither_2x2_4[y&1    ][0];
694                         const int db1= dither_2x2_8[(y&1)^1][0];
695                         const int dr2= dither_2x2_8[y&1    ][1];
696                         const int dg2= dither_2x2_4[y&1    ][1];
697                         const int db2= dither_2x2_8[(y&1)^1][1];
698                         YSCALE_YUV_2_RGBX_C(uint16_t)
699                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
700                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
701                         }
702                 }
703                 break;
704         case PIX_FMT_RGB555:
705         case PIX_FMT_BGR555:
706                 {
707                         const int dr1= dither_2x2_8[y&1    ][0];
708                         const int dg1= dither_2x2_8[y&1    ][1];
709                         const int db1= dither_2x2_8[(y&1)^1][0];
710                         const int dr2= dither_2x2_8[y&1    ][1];
711                         const int dg2= dither_2x2_8[y&1    ][0];
712                         const int db2= dither_2x2_8[(y&1)^1][1];
713                         YSCALE_YUV_2_RGBX_C(uint16_t)
714                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
715                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
716                         }
717                 }
718                 break;
719         case PIX_FMT_RGB8:
720         case PIX_FMT_BGR8:
721                 {
722                         const uint8_t * const d64= dither_8x8_73[y&7];
723                         const uint8_t * const d32= dither_8x8_32[y&7];
724                         YSCALE_YUV_2_RGBX_C(uint8_t)
725                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
726                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
727                         }
728                 }
729                 break;
730         case PIX_FMT_RGB4:
731         case PIX_FMT_BGR4:
732                 {
733                         const uint8_t * const d64= dither_8x8_73 [y&7];
734                         const uint8_t * const d128=dither_8x8_220[y&7];
735                         YSCALE_YUV_2_RGBX_C(uint8_t)
736                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
737                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
738                         }
739                 }
740                 break;
741         case PIX_FMT_RGB4_BYTE:
742         case PIX_FMT_BGR4_BYTE:
743                 {
744                         const uint8_t * const d64= dither_8x8_73 [y&7];
745                         const uint8_t * const d128=dither_8x8_220[y&7];
746                         YSCALE_YUV_2_RGBX_C(uint8_t)
747                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
748                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
749                         }
750                 }
751                 break;
752         case PIX_FMT_MONOBLACK:
753                 {
754                         const uint8_t * const d128=dither_8x8_220[y&7];
755                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
756                         int acc=0;
757                         for(i=0; i<dstW-1; i+=2){
758                                 int j;
759                                 int Y1=1<<18;
760                                 int Y2=1<<18;
761
762                                 for(j=0; j<lumFilterSize; j++)
763                                 {
764                                         Y1 += lumSrc[j][i] * lumFilter[j];
765                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
766                                 }
767                                 Y1>>=19;
768                                 Y2>>=19;
769                                 if((Y1|Y2)&256)
770                                 {
771                                         if(Y1>255)   Y1=255;
772                                         else if(Y1<0)Y1=0;
773                                         if(Y2>255)   Y2=255;
774                                         else if(Y2<0)Y2=0;
775                                 }
776                                 acc+= acc + g[Y1+d128[(i+0)&7]];
777                                 acc+= acc + g[Y2+d128[(i+1)&7]];
778                                 if((i&7)==6){
779                                         ((uint8_t*)dest)[0]= acc;
780                                         dest++;
781                                 }
782                         }
783                 }
784                 break;
785         case PIX_FMT_YUYV422:
786                 YSCALE_YUV_2_PACKEDX_C(void)
787                         ((uint8_t*)dest)[2*i2+0]= Y1;
788                         ((uint8_t*)dest)[2*i2+1]= U;
789                         ((uint8_t*)dest)[2*i2+2]= Y2;
790                         ((uint8_t*)dest)[2*i2+3]= V;
791                 }
792                 break;
793         case PIX_FMT_UYVY422:
794                 YSCALE_YUV_2_PACKEDX_C(void)
795                         ((uint8_t*)dest)[2*i2+0]= U;
796                         ((uint8_t*)dest)[2*i2+1]= Y1;
797                         ((uint8_t*)dest)[2*i2+2]= V;
798                         ((uint8_t*)dest)[2*i2+3]= Y2;
799                 }
800                 break;
801         }
802 }
803
804
805 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
806 //Plain C versions
807 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
808 #define COMPILE_C
809 #endif
810
811 #ifdef ARCH_POWERPC
812 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
813 #define COMPILE_ALTIVEC
814 #endif //HAVE_ALTIVEC
815 #endif //ARCH_POWERPC
816
817 #if defined(ARCH_X86)
818
819 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
820 #define COMPILE_MMX
821 #endif
822
823 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
824 #define COMPILE_MMX2
825 #endif
826
827 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
828 #define COMPILE_3DNOW
829 #endif
830 #endif //ARCH_X86 || ARCH_X86_64
831
832 #undef HAVE_MMX
833 #undef HAVE_MMX2
834 #undef HAVE_3DNOW
835
836 #ifdef COMPILE_C
837 #undef HAVE_MMX
838 #undef HAVE_MMX2
839 #undef HAVE_3DNOW
840 #undef HAVE_ALTIVEC
841 #define RENAME(a) a ## _C
842 #include "swscale_template.c"
843 #endif
844
845 #ifdef ARCH_POWERPC
846 #ifdef COMPILE_ALTIVEC
847 #undef RENAME
848 #define HAVE_ALTIVEC
849 #define RENAME(a) a ## _altivec
850 #include "swscale_template.c"
851 #endif
852 #endif //ARCH_POWERPC
853
854 #if defined(ARCH_X86)
855
856 //X86 versions
857 /*
858 #undef RENAME
859 #undef HAVE_MMX
860 #undef HAVE_MMX2
861 #undef HAVE_3DNOW
862 #define ARCH_X86
863 #define RENAME(a) a ## _X86
864 #include "swscale_template.c"
865 */
866 //MMX versions
867 #ifdef COMPILE_MMX
868 #undef RENAME
869 #define HAVE_MMX
870 #undef HAVE_MMX2
871 #undef HAVE_3DNOW
872 #define RENAME(a) a ## _MMX
873 #include "swscale_template.c"
874 #endif
875
876 //MMX2 versions
877 #ifdef COMPILE_MMX2
878 #undef RENAME
879 #define HAVE_MMX
880 #define HAVE_MMX2
881 #undef HAVE_3DNOW
882 #define RENAME(a) a ## _MMX2
883 #include "swscale_template.c"
884 #endif
885
886 //3DNOW versions
887 #ifdef COMPILE_3DNOW
888 #undef RENAME
889 #define HAVE_MMX
890 #undef HAVE_MMX2
891 #define HAVE_3DNOW
892 #define RENAME(a) a ## _3DNow
893 #include "swscale_template.c"
894 #endif
895
896 #endif //ARCH_X86 || ARCH_X86_64
897
898 // minor note: the HAVE_xyz is messed up after that line so don't use it
899
900 static double getSplineCoeff(double a, double b, double c, double d, double dist)
901 {
902 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
903         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
904         else            return getSplineCoeff(  0.0, 
905                                                  b+ 2.0*c + 3.0*d,
906                                                         c + 3.0*d,
907                                                 -b- 3.0*c - 6.0*d,
908                                                 dist-1.0);
909 }
910
911 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
912                               int srcW, int dstW, int filterAlign, int one, int flags,
913                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
914 {
915         int i;
916         int filterSize;
917         int filter2Size;
918         int minFilterSize;
919         double *filter=NULL;
920         double *filter2=NULL;
921 #if defined(ARCH_X86)
922         if(flags & SWS_CPU_CAPS_MMX)
923                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
924 #endif
925
926         // Note the +1 is for the MMXscaler which reads over the end
927         *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
928
929         if(FFABS(xInc - 0x10000) <10) // unscaled
930         {
931                 int i;
932                 filterSize= 1;
933                 filter= av_malloc(dstW*sizeof(double)*filterSize);
934                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
935
936                 for(i=0; i<dstW; i++)
937                 {
938                         filter[i*filterSize]=1;
939                         (*filterPos)[i]=i;
940                 }
941
942         }
943         else if(flags&SWS_POINT) // lame looking point sampling mode
944         {
945                 int i;
946                 int xDstInSrc;
947                 filterSize= 1;
948                 filter= av_malloc(dstW*sizeof(double)*filterSize);
949                 
950                 xDstInSrc= xInc/2 - 0x8000;
951                 for(i=0; i<dstW; i++)
952                 {
953                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
954
955                         (*filterPos)[i]= xx;
956                         filter[i]= 1.0;
957                         xDstInSrc+= xInc;
958                 }
959         }
960         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
961         {
962                 int i;
963                 int xDstInSrc;
964                 if     (flags&SWS_BICUBIC) filterSize= 4;
965                 else if(flags&SWS_X      ) filterSize= 4;
966                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
967                 filter= av_malloc(dstW*sizeof(double)*filterSize);
968
969                 xDstInSrc= xInc/2 - 0x8000;
970                 for(i=0; i<dstW; i++)
971                 {
972                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
973                         int j;
974
975                         (*filterPos)[i]= xx;
976                                 //Bilinear upscale / linear interpolate / Area averaging
977                                 for(j=0; j<filterSize; j++)
978                                 {
979                                         double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
980                                         double coeff= 1.0 - d;
981                                         if(coeff<0) coeff=0;
982                                         filter[i*filterSize + j]= coeff;
983                                         xx++;
984                                 }
985                         xDstInSrc+= xInc;
986                 }
987         }
988         else
989         {
990                 double xDstInSrc;
991                 double sizeFactor, filterSizeInSrc;
992                 const double xInc1= (double)xInc / (double)(1<<16);
993
994                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
995                 else if(flags&SWS_X)            sizeFactor= 8.0;
996                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
997                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
998                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
999                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
1000                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
1001                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
1002                 else {
1003                         sizeFactor= 0.0; //GCC warning killer
1004                         ASSERT(0)
1005                 }
1006                 
1007                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
1008                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1009
1010                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1011                 if(filterSize > srcW-2) filterSize=srcW-2;
1012
1013                 filter= av_malloc(dstW*sizeof(double)*filterSize);
1014
1015                 xDstInSrc= xInc1 / 2.0 - 0.5;
1016                 for(i=0; i<dstW; i++)
1017                 {
1018                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1019                         int j;
1020                         (*filterPos)[i]= xx;
1021                         for(j=0; j<filterSize; j++)
1022                         {
1023                                 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1024                                 double coeff;
1025                                 if(flags & SWS_BICUBIC)
1026                                 {
1027                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1028                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1029
1030                                         if(d<1.0) 
1031                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1032                                         else if(d<2.0)
1033                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1034                                         else
1035                                                 coeff=0.0;
1036                                 }
1037 /*                              else if(flags & SWS_X)
1038                                 {
1039                                         double p= param ? param*0.01 : 0.3;
1040                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1041                                         coeff*= pow(2.0, - p*d*d);
1042                                 }*/
1043                                 else if(flags & SWS_X)
1044                                 {
1045                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1046                                         
1047                                         if(d<1.0)
1048                                                 coeff = cos(d*PI);
1049                                         else
1050                                                 coeff=-1.0;
1051                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
1052                                         else            coeff=  pow( coeff, A);
1053                                         coeff= coeff*0.5 + 0.5;
1054                                 }
1055                                 else if(flags & SWS_AREA)
1056                                 {
1057                                         double srcPixelSize= 1.0/xInc1;
1058                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1059                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1060                                         else coeff=0.0;
1061                                 }
1062                                 else if(flags & SWS_GAUSS)
1063                                 {
1064                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1065                                         coeff = pow(2.0, - p*d*d);
1066                                 }
1067                                 else if(flags & SWS_SINC)
1068                                 {
1069                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1070                                 }
1071                                 else if(flags & SWS_LANCZOS)
1072                                 {
1073                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
1074                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1075                                         if(d>p) coeff=0;
1076                                 }
1077                                 else if(flags & SWS_BILINEAR)
1078                                 {
1079                                         coeff= 1.0 - d;
1080                                         if(coeff<0) coeff=0;
1081                                 }
1082                                 else if(flags & SWS_SPLINE)
1083                                 {
1084                                         double p=-2.196152422706632;
1085                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1086                                 }
1087                                 else {
1088                                         coeff= 0.0; //GCC warning killer
1089                                         ASSERT(0)
1090                                 }
1091
1092                                 filter[i*filterSize + j]= coeff;
1093                                 xx++;
1094                         }
1095                         xDstInSrc+= xInc1;
1096                 }
1097         }
1098
1099         /* apply src & dst Filter to filter -> filter2
1100            av_free(filter);
1101         */
1102         ASSERT(filterSize>0)
1103         filter2Size= filterSize;
1104         if(srcFilter) filter2Size+= srcFilter->length - 1;
1105         if(dstFilter) filter2Size+= dstFilter->length - 1;
1106         ASSERT(filter2Size>0)
1107         filter2= av_malloc(filter2Size*dstW*sizeof(double));
1108
1109         for(i=0; i<dstW; i++)
1110         {
1111                 int j;
1112                 SwsVector scaleFilter;
1113                 SwsVector *outVec;
1114
1115                 scaleFilter.coeff= filter + i*filterSize;
1116                 scaleFilter.length= filterSize;
1117
1118                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1119                 else          outVec= &scaleFilter;
1120
1121                 ASSERT(outVec->length == filter2Size)
1122                 //FIXME dstFilter
1123
1124                 for(j=0; j<outVec->length; j++)
1125                 {
1126                         filter2[i*filter2Size + j]= outVec->coeff[j];
1127                 }
1128
1129                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1130
1131                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1132         }
1133         av_free(filter); filter=NULL;
1134
1135         /* try to reduce the filter-size (step1 find size and shift left) */
1136         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1137         minFilterSize= 0;
1138         for(i=dstW-1; i>=0; i--)
1139         {
1140                 int min= filter2Size;
1141                 int j;
1142                 double cutOff=0.0;
1143
1144                 /* get rid off near zero elements on the left by shifting left */
1145                 for(j=0; j<filter2Size; j++)
1146                 {
1147                         int k;
1148                         cutOff += FFABS(filter2[i*filter2Size]);
1149
1150                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1151
1152                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1153                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1154
1155                         // Move filter coeffs left
1156                         for(k=1; k<filter2Size; k++)
1157                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1158                         filter2[i*filter2Size + k - 1]= 0.0;
1159                         (*filterPos)[i]++;
1160                 }
1161
1162                 cutOff=0.0;
1163                 /* count near zeros on the right */
1164                 for(j=filter2Size-1; j>0; j--)
1165                 {
1166                         cutOff += FFABS(filter2[i*filter2Size + j]);
1167
1168                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1169                         min--;
1170                 }
1171
1172                 if(min>minFilterSize) minFilterSize= min;
1173         }
1174
1175         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1176           // we can handle the special case 4,
1177           // so we don't want to go to the full 8
1178           if (minFilterSize < 5)
1179             filterAlign = 4;
1180
1181           // we really don't want to waste our time
1182           // doing useless computation, so fall-back on
1183           // the scalar C code for very small filter.
1184           // vectorizing is worth it only if you have
1185           // decent-sized vector.
1186           if (minFilterSize < 3)
1187             filterAlign = 1;
1188         }
1189
1190         if (flags & SWS_CPU_CAPS_MMX) {
1191                 // special case for unscaled vertical filtering
1192                 if(minFilterSize == 1 && filterAlign == 2)
1193                         filterAlign= 1;
1194         }
1195
1196         ASSERT(minFilterSize > 0)
1197         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1198         ASSERT(filterSize > 0)
1199         filter= av_malloc(filterSize*dstW*sizeof(double));
1200         if(filterSize >= MAX_FILTER_SIZE)
1201                 return -1;
1202         *outFilterSize= filterSize;
1203
1204         if(flags&SWS_PRINT_INFO)
1205                 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1206         /* try to reduce the filter-size (step2 reduce it) */
1207         for(i=0; i<dstW; i++)
1208         {
1209                 int j;
1210
1211                 for(j=0; j<filterSize; j++)
1212                 {
1213                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1214                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1215                 }
1216         }
1217         av_free(filter2); filter2=NULL;
1218         
1219
1220         //FIXME try to align filterpos if possible
1221
1222         //fix borders
1223         for(i=0; i<dstW; i++)
1224         {
1225                 int j;
1226                 if((*filterPos)[i] < 0)
1227                 {
1228                         // Move filter coeffs left to compensate for filterPos
1229                         for(j=1; j<filterSize; j++)
1230                         {
1231                                 int left= FFMAX(j + (*filterPos)[i], 0);
1232                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1233                                 filter[i*filterSize + j]=0;
1234                         }
1235                         (*filterPos)[i]= 0;
1236                 }
1237
1238                 if((*filterPos)[i] + filterSize > srcW)
1239                 {
1240                         int shift= (*filterPos)[i] + filterSize - srcW;
1241                         // Move filter coeffs right to compensate for filterPos
1242                         for(j=filterSize-2; j>=0; j--)
1243                         {
1244                                 int right= FFMIN(j + shift, filterSize-1);
1245                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1246                                 filter[i*filterSize +j]=0;
1247                         }
1248                         (*filterPos)[i]= srcW - filterSize;
1249                 }
1250         }
1251
1252         // Note the +1 is for the MMXscaler which reads over the end
1253         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1254         *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1255
1256         /* Normalize & Store in outFilter */
1257         for(i=0; i<dstW; i++)
1258         {
1259                 int j;
1260                 double error=0;
1261                 double sum=0;
1262                 double scale= one;
1263
1264                 for(j=0; j<filterSize; j++)
1265                 {
1266                         sum+= filter[i*filterSize + j];
1267                 }
1268                 scale/= sum;
1269                 for(j=0; j<*outFilterSize; j++)
1270                 {
1271                         double v= filter[i*filterSize + j]*scale + error;
1272                         int intV= floor(v + 0.5);
1273                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1274                         error = v - intV;
1275                 }
1276         }
1277         
1278         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1279         for(i=0; i<*outFilterSize; i++)
1280         {
1281                 int j= dstW*(*outFilterSize);
1282                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1283         }
1284
1285         av_free(filter);
1286         return 0;
1287 }
1288
1289 #ifdef COMPILE_MMX2
1290 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1291 {
1292         uint8_t *fragmentA;
1293         long imm8OfPShufW1A;
1294         long imm8OfPShufW2A;
1295         long fragmentLengthA;
1296         uint8_t *fragmentB;
1297         long imm8OfPShufW1B;
1298         long imm8OfPShufW2B;
1299         long fragmentLengthB;
1300         int fragmentPos;
1301
1302         int xpos, i;
1303
1304         // create an optimized horizontal scaling routine
1305
1306         //code fragment
1307
1308         asm volatile(
1309                 "jmp 9f                         \n\t"
1310         // Begin
1311                 "0:                             \n\t"
1312                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1313                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1314                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1315                 "punpcklbw %%mm7, %%mm1         \n\t"
1316                 "punpcklbw %%mm7, %%mm0         \n\t"
1317                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1318                 "1:                             \n\t"
1319                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1320                 "2:                             \n\t"
1321                 "psubw %%mm1, %%mm0             \n\t"
1322                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1323                 "pmullw %%mm3, %%mm0            \n\t"
1324                 "psllw $7, %%mm1                \n\t"
1325                 "paddw %%mm1, %%mm0             \n\t"
1326
1327                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1328
1329                 "add $8, %%"REG_a"              \n\t"
1330         // End
1331                 "9:                             \n\t"
1332 //              "int $3\n\t"
1333                 "lea 0b, %0                     \n\t"
1334                 "lea 1b, %1                     \n\t"
1335                 "lea 2b, %2                     \n\t"
1336                 "dec %1                         \n\t"
1337                 "dec %2                         \n\t"
1338                 "sub %0, %1                     \n\t"
1339                 "sub %0, %2                     \n\t"
1340                 "lea 9b, %3                     \n\t"
1341                 "sub %0, %3                     \n\t"
1342
1343
1344                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1345                 "=r" (fragmentLengthA)
1346         );
1347
1348         asm volatile(
1349                 "jmp 9f                         \n\t"
1350         // Begin
1351                 "0:                             \n\t"
1352                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1353                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1354                 "punpcklbw %%mm7, %%mm0         \n\t"
1355                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1356                 "1:                             \n\t"
1357                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1358                 "2:                             \n\t"
1359                 "psubw %%mm1, %%mm0             \n\t"
1360                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1361                 "pmullw %%mm3, %%mm0            \n\t"
1362                 "psllw $7, %%mm1                \n\t"
1363                 "paddw %%mm1, %%mm0             \n\t"
1364
1365                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1366
1367                 "add $8, %%"REG_a"              \n\t"
1368         // End
1369                 "9:                             \n\t"
1370 //              "int $3\n\t"
1371                 "lea 0b, %0                     \n\t"
1372                 "lea 1b, %1                     \n\t"
1373                 "lea 2b, %2                     \n\t"
1374                 "dec %1                         \n\t"
1375                 "dec %2                         \n\t"
1376                 "sub %0, %1                     \n\t"
1377                 "sub %0, %2                     \n\t"
1378                 "lea 9b, %3                     \n\t"
1379                 "sub %0, %3                     \n\t"
1380
1381
1382                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1383                 "=r" (fragmentLengthB)
1384         );
1385
1386         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1387         fragmentPos=0;
1388         
1389         for(i=0; i<dstW/numSplits; i++)
1390         {
1391                 int xx=xpos>>16;
1392
1393                 if((i&3) == 0)
1394                 {
1395                         int a=0;
1396                         int b=((xpos+xInc)>>16) - xx;
1397                         int c=((xpos+xInc*2)>>16) - xx;
1398                         int d=((xpos+xInc*3)>>16) - xx;
1399
1400                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1401                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1402                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1403                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1404                         filterPos[i/2]= xx;
1405
1406                         if(d+1<4)
1407                         {
1408                                 int maxShift= 3-(d+1);
1409                                 int shift=0;
1410
1411                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1412
1413                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1414                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1415                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1416                                         a | (b<<2) | (c<<4) | (d<<6);
1417
1418                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1419                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1420
1421                                 if(shift && i>=shift)
1422                                 {
1423                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1424                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1425                                         filterPos[i/2]-=shift;
1426                                 }
1427
1428                                 fragmentPos+= fragmentLengthB;
1429                         }
1430                         else
1431                         {
1432                                 int maxShift= 3-d;
1433                                 int shift=0;
1434
1435                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1436
1437                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1438                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1439                                         a | (b<<2) | (c<<4) | (d<<6);
1440
1441                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1442                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1443
1444                                 if(shift && i>=shift)
1445                                 {
1446                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1447                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1448                                         filterPos[i/2]-=shift;
1449                                 }
1450
1451                                 fragmentPos+= fragmentLengthA;
1452                         }
1453
1454                         funnyCode[fragmentPos]= RET;
1455                 }
1456                 xpos+=xInc;
1457         }
1458         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1459 }
1460 #endif /* COMPILE_MMX2 */
1461
1462 static void globalInit(void){
1463     // generating tables:
1464     int i;
1465     for(i=0; i<768; i++){
1466         int c= clip_uint8(i-256);
1467         clip_table[i]=c;
1468     }
1469 }
1470
1471 static SwsFunc getSwsFunc(int flags){
1472     
1473 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1474 #if defined(ARCH_X86)
1475         // ordered per speed fasterst first
1476         if(flags & SWS_CPU_CAPS_MMX2)
1477                 return swScale_MMX2;
1478         else if(flags & SWS_CPU_CAPS_3DNOW)
1479                 return swScale_3DNow;
1480         else if(flags & SWS_CPU_CAPS_MMX)
1481                 return swScale_MMX;
1482         else
1483                 return swScale_C;
1484
1485 #else
1486 #ifdef ARCH_POWERPC
1487         if(flags & SWS_CPU_CAPS_ALTIVEC)
1488           return swScale_altivec;
1489         else
1490           return swScale_C;
1491 #endif
1492         return swScale_C;
1493 #endif /* defined(ARCH_X86) */
1494 #else //RUNTIME_CPUDETECT
1495 #ifdef HAVE_MMX2
1496         return swScale_MMX2;
1497 #elif defined (HAVE_3DNOW)
1498         return swScale_3DNow;
1499 #elif defined (HAVE_MMX)
1500         return swScale_MMX;
1501 #elif defined (HAVE_ALTIVEC)
1502         return swScale_altivec;
1503 #else
1504         return swScale_C;
1505 #endif
1506 #endif //!RUNTIME_CPUDETECT
1507 }
1508
1509 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1510              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1511         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1512         /* Copy Y plane */
1513         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1514                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1515         else
1516         {
1517                 int i;
1518                 uint8_t *srcPtr= src[0];
1519                 uint8_t *dstPtr= dst;
1520                 for(i=0; i<srcSliceH; i++)
1521                 {
1522                         memcpy(dstPtr, srcPtr, c->srcW);
1523                         srcPtr+= srcStride[0];
1524                         dstPtr+= dstStride[0];
1525                 }
1526         }
1527         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1528         if (c->dstFormat == PIX_FMT_NV12)
1529                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1530         else
1531                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1532
1533         return srcSliceH;
1534 }
1535
1536 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1537              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1538         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1539
1540         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1541
1542         return srcSliceH;
1543 }
1544
1545 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1546              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1547         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1548
1549         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1550
1551         return srcSliceH;
1552 }
1553
1554 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1555 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1556                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1557         const int srcFormat= c->srcFormat;
1558         const int dstFormat= c->dstFormat;
1559         const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1560         const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1561         const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1562         const int dstId= fmt_depth(dstFormat) >> 2;
1563         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1564
1565         /* BGR -> BGR */
1566         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1567            || (isRGB(srcFormat) && isRGB(dstFormat))){
1568                 switch(srcId | (dstId<<4)){
1569                 case 0x34: conv= rgb16to15; break;
1570                 case 0x36: conv= rgb24to15; break;
1571                 case 0x38: conv= rgb32to15; break;
1572                 case 0x43: conv= rgb15to16; break;
1573                 case 0x46: conv= rgb24to16; break;
1574                 case 0x48: conv= rgb32to16; break;
1575                 case 0x63: conv= rgb15to24; break;
1576                 case 0x64: conv= rgb16to24; break;
1577                 case 0x68: conv= rgb32to24; break;
1578                 case 0x83: conv= rgb15to32; break;
1579                 case 0x84: conv= rgb16to32; break;
1580                 case 0x86: conv= rgb24to32; break;
1581                 default: av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1582                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1583                 }
1584         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1585                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1586                 switch(srcId | (dstId<<4)){
1587                 case 0x33: conv= rgb15tobgr15; break;
1588                 case 0x34: conv= rgb16tobgr15; break;
1589                 case 0x36: conv= rgb24tobgr15; break;
1590                 case 0x38: conv= rgb32tobgr15; break;
1591                 case 0x43: conv= rgb15tobgr16; break;
1592                 case 0x44: conv= rgb16tobgr16; break;
1593                 case 0x46: conv= rgb24tobgr16; break;
1594                 case 0x48: conv= rgb32tobgr16; break;
1595                 case 0x63: conv= rgb15tobgr24; break;
1596                 case 0x64: conv= rgb16tobgr24; break;
1597                 case 0x66: conv= rgb24tobgr24; break;
1598                 case 0x68: conv= rgb32tobgr24; break;
1599                 case 0x83: conv= rgb15tobgr32; break;
1600                 case 0x84: conv= rgb16tobgr32; break;
1601                 case 0x86: conv= rgb24tobgr32; break;
1602                 case 0x88: conv= rgb32tobgr32; break;
1603                 default: av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1604                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1605                 }
1606         }else{
1607                 av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1608                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1609         }
1610
1611         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1612                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1613         else
1614         {
1615                 int i;
1616                 uint8_t *srcPtr= src[0];
1617                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1618
1619                 for(i=0; i<srcSliceH; i++)
1620                 {
1621                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1622                         srcPtr+= srcStride[0];
1623                         dstPtr+= dstStride[0];
1624                 }
1625         }     
1626         return srcSliceH;
1627 }
1628
1629 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1630              int srcSliceH, uint8_t* dst[], int dstStride[]){
1631
1632         rgb24toyv12(
1633                 src[0], 
1634                 dst[0]+ srcSliceY    *dstStride[0], 
1635                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1636                 dst[2]+(srcSliceY>>1)*dstStride[2],
1637                 c->srcW, srcSliceH, 
1638                 dstStride[0], dstStride[1], srcStride[0]);
1639         return srcSliceH;
1640 }
1641
1642 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1643              int srcSliceH, uint8_t* dst[], int dstStride[]){
1644         int i;
1645
1646         /* copy Y */
1647         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1648                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1649         else{
1650                 uint8_t *srcPtr= src[0];
1651                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1652
1653                 for(i=0; i<srcSliceH; i++)
1654                 {
1655                         memcpy(dstPtr, srcPtr, c->srcW);
1656                         srcPtr+= srcStride[0];
1657                         dstPtr+= dstStride[0];
1658                 }
1659         }
1660
1661         if(c->dstFormat==PIX_FMT_YUV420P){
1662                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1663                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1664         }else{
1665                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1666                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1667         }
1668         return srcSliceH;
1669 }
1670
1671 /* unscaled copy like stuff (assumes nearly identical formats) */
1672 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1673              int srcSliceH, uint8_t* dst[], int dstStride[]){
1674
1675         if(isPacked(c->srcFormat))
1676         {
1677                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1678                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1679                 else
1680                 {
1681                         int i;
1682                         uint8_t *srcPtr= src[0];
1683                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1684                         int length=0;
1685
1686                         /* universal length finder */
1687                         while(length+c->srcW <= FFABS(dstStride[0]) 
1688                            && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1689                         ASSERT(length!=0);
1690
1691                         for(i=0; i<srcSliceH; i++)
1692                         {
1693                                 memcpy(dstPtr, srcPtr, length);
1694                                 srcPtr+= srcStride[0];
1695                                 dstPtr+= dstStride[0];
1696                         }
1697                 }
1698         }
1699         else 
1700         { /* Planar YUV or gray */
1701                 int plane;
1702                 for(plane=0; plane<3; plane++)
1703                 {
1704                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1705                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1706                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1707
1708                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1709                         {
1710                                 if(!isGray(c->dstFormat))
1711                                         memset(dst[plane], 128, dstStride[plane]*height);
1712                         }
1713                         else
1714                         {
1715                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1716                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1717                                 else
1718                                 {
1719                                         int i;
1720                                         uint8_t *srcPtr= src[plane];
1721                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1722                                         for(i=0; i<height; i++)
1723                                         {
1724                                                 memcpy(dstPtr, srcPtr, length);
1725                                                 srcPtr+= srcStride[plane];
1726                                                 dstPtr+= dstStride[plane];
1727                                         }
1728                                 }
1729                         }
1730                 }
1731         }
1732         return srcSliceH;
1733 }
1734
1735 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1736              int srcSliceH, uint8_t* dst[], int dstStride[]){
1737
1738         int length= c->srcW;
1739         int y=      srcSliceY;
1740         int height= srcSliceH;
1741         int i, j;
1742         uint8_t *srcPtr= src[0];
1743         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1744
1745         if(!isGray(c->dstFormat)){
1746                 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1747                 memset(dst[1], 128, dstStride[1]*height);
1748                 memset(dst[2], 128, dstStride[2]*height);
1749         }
1750         if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1751         for(i=0; i<height; i++)
1752         {
1753                 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1754                 srcPtr+= srcStride[0];
1755                 dstPtr+= dstStride[0];
1756         }
1757         return srcSliceH;
1758 }
1759
1760 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1761              int srcSliceH, uint8_t* dst[], int dstStride[]){
1762
1763         int length= c->srcW;
1764         int y=      srcSliceY;
1765         int height= srcSliceH;
1766         int i, j;
1767         uint8_t *srcPtr= src[0];
1768         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1769         for(i=0; i<height; i++)
1770         {
1771                 for(j=0; j<length; j++)
1772                 {
1773                         dstPtr[j<<1] = srcPtr[j];
1774                         dstPtr[(j<<1)+1] = srcPtr[j];
1775                 }
1776                 srcPtr+= srcStride[0];
1777                 dstPtr+= dstStride[0];
1778         }
1779         return srcSliceH;
1780 }
1781
1782 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1783              int srcSliceH, uint8_t* dst[], int dstStride[]){
1784
1785         int length= c->srcW;
1786         int y=      srcSliceY;
1787         int height= srcSliceH;
1788         int i, j;
1789         uint16_t *srcPtr= src[0];
1790         uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1791         for(i=0; i<height; i++)
1792         {
1793                 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1794                 srcPtr+= srcStride[0]/2;
1795                 dstPtr+= dstStride[0]/2;
1796         }
1797         return srcSliceH;
1798 }
1799
1800
1801 static void getSubSampleFactors(int *h, int *v, int format){
1802         switch(format){
1803         case PIX_FMT_UYVY422:
1804         case PIX_FMT_YUYV422:
1805                 *h=1;
1806                 *v=0;
1807                 break;
1808         case PIX_FMT_YUV420P:
1809         case PIX_FMT_GRAY16BE:
1810         case PIX_FMT_GRAY16LE:
1811         case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1812         case PIX_FMT_NV12:
1813         case PIX_FMT_NV21:
1814                 *h=1;
1815                 *v=1;
1816                 break;
1817         case PIX_FMT_YUV410P:
1818                 *h=2;
1819                 *v=2;
1820                 break;
1821         case PIX_FMT_YUV444P:
1822                 *h=0;
1823                 *v=0;
1824                 break;
1825         case PIX_FMT_YUV422P:
1826                 *h=1;
1827                 *v=0;
1828                 break;
1829         case PIX_FMT_YUV411P:
1830                 *h=2;
1831                 *v=0;
1832                 break;
1833         default:
1834                 *h=0;
1835                 *v=0;
1836                 break;
1837         }
1838 }
1839
1840 static uint16_t roundToInt16(int64_t f){
1841         int r= (f + (1<<15))>>16;
1842              if(r<-0x7FFF) return 0x8000;
1843         else if(r> 0x7FFF) return 0x7FFF;
1844         else               return r;
1845 }
1846
1847 /**
1848  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1849  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1850  * @return -1 if not supported
1851  */
1852 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1853         int64_t crv =  inv_table[0];
1854         int64_t cbu =  inv_table[1];
1855         int64_t cgu = -inv_table[2];
1856         int64_t cgv = -inv_table[3];
1857         int64_t cy  = 1<<16;
1858         int64_t oy  = 0;
1859
1860         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1861         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1862         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1863
1864         c->brightness= brightness;
1865         c->contrast  = contrast;
1866         c->saturation= saturation;
1867         c->srcRange  = srcRange;
1868         c->dstRange  = dstRange;
1869
1870         c->uOffset=   0x0400040004000400LL;
1871         c->vOffset=   0x0400040004000400LL;
1872
1873         if(!srcRange){
1874                 cy= (cy*255) / 219;
1875                 oy= 16<<16;
1876         }
1877
1878         cy = (cy *contrast             )>>16;
1879         crv= (crv*contrast * saturation)>>32;
1880         cbu= (cbu*contrast * saturation)>>32;
1881         cgu= (cgu*contrast * saturation)>>32;
1882         cgv= (cgv*contrast * saturation)>>32;
1883
1884         oy -= 256*brightness;
1885
1886         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1887         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1888         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1889         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1890         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1891         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1892
1893         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1894         //FIXME factorize
1895
1896 #ifdef COMPILE_ALTIVEC
1897         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1898             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1899 #endif  
1900         return 0;
1901 }
1902
1903 /**
1904  * @return -1 if not supported
1905  */
1906 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1907         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1908
1909         *inv_table = c->srcColorspaceTable;
1910         *table     = c->dstColorspaceTable;
1911         *srcRange  = c->srcRange;
1912         *dstRange  = c->dstRange;
1913         *brightness= c->brightness;
1914         *contrast  = c->contrast;
1915         *saturation= c->saturation;
1916         
1917         return 0;       
1918 }
1919
1920 static int handle_jpeg(int *format)
1921 {
1922         switch (*format) {
1923                 case PIX_FMT_YUVJ420P:
1924                         *format = PIX_FMT_YUV420P;
1925                         return 1;
1926                 case PIX_FMT_YUVJ422P:
1927                         *format = PIX_FMT_YUV422P;
1928                         return 1;
1929                 case PIX_FMT_YUVJ444P:
1930                         *format = PIX_FMT_YUV444P;
1931                         return 1;
1932                 default:
1933                         return 0;
1934         }
1935 }
1936
1937 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1938                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1939
1940         SwsContext *c;
1941         int i;
1942         int usesVFilter, usesHFilter;
1943         int unscaled, needsDither;
1944         int srcRange, dstRange;
1945         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1946 #if defined(ARCH_X86)
1947         if(flags & SWS_CPU_CAPS_MMX)
1948                 asm volatile("emms\n\t"::: "memory");
1949 #endif
1950
1951 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1952         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1953 #ifdef HAVE_MMX2
1954         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1955 #elif defined (HAVE_3DNOW)
1956         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1957 #elif defined (HAVE_MMX)
1958         flags |= SWS_CPU_CAPS_MMX;
1959 #elif defined (HAVE_ALTIVEC)
1960         flags |= SWS_CPU_CAPS_ALTIVEC;
1961 #endif
1962 #endif /* RUNTIME_CPUDETECT */
1963         if(clip_table[512] != 255) globalInit();
1964         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1965
1966         unscaled = (srcW == dstW && srcH == dstH);
1967         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1968                      && (fmt_depth(dstFormat))<24
1969                      && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1970
1971         srcRange = handle_jpeg(&srcFormat);
1972         dstRange = handle_jpeg(&dstFormat);
1973
1974         if(!isSupportedIn(srcFormat)) 
1975         {
1976                 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1977                 return NULL;
1978         }
1979         if(!isSupportedOut(dstFormat))
1980         {
1981                 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1982                 return NULL;
1983         }
1984
1985         /* sanity check */
1986         if(srcW<4 || srcH<1 || dstW<8 || dstH<1) //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
1987         {
1988                  av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1989                         srcW, srcH, dstW, dstH);
1990                 return NULL;
1991         }
1992
1993         if(!dstFilter) dstFilter= &dummyFilter;
1994         if(!srcFilter) srcFilter= &dummyFilter;
1995
1996         c= av_mallocz(sizeof(SwsContext));
1997
1998         c->av_class = &sws_context_class;
1999         c->srcW= srcW;
2000         c->srcH= srcH;
2001         c->dstW= dstW;
2002         c->dstH= dstH;
2003         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2004         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2005         c->flags= flags;
2006         c->dstFormat= dstFormat;
2007         c->srcFormat= srcFormat;
2008         c->vRounder= 4* 0x0001000100010001ULL;
2009
2010         usesHFilter= usesVFilter= 0;
2011         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2012         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2013         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2014         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2015         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2016         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2017         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2018         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2019
2020         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2021         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2022
2023         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2024         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2025
2026         // drop some chroma lines if the user wants it
2027         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2028         c->chrSrcVSubSample+= c->vChrDrop;
2029
2030         // drop every 2. pixel for chroma calculation unless user wants full chroma
2031         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
2032                 c->chrSrcHSubSample=1;
2033
2034         if(param){
2035                 c->param[0] = param[0];
2036                 c->param[1] = param[1];
2037         }else{
2038                 c->param[0] =
2039                 c->param[1] = SWS_PARAM_DEFAULT;
2040         }
2041
2042         c->chrIntHSubSample= c->chrDstHSubSample;
2043         c->chrIntVSubSample= c->chrSrcVSubSample;
2044
2045         // note the -((-x)>>y) is so that we allways round toward +inf
2046         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2047         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2048         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2049         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2050
2051         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16); 
2052
2053         /* unscaled special Cases */
2054         if(unscaled && !usesHFilter && !usesVFilter)
2055         {
2056                 /* yv12_to_nv12 */
2057                 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2058                 {
2059                         c->swScale= PlanarToNV12Wrapper;
2060                 }
2061 #ifdef CONFIG_GPL
2062                 /* yuv2bgr */
2063                 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2064                 {
2065                         c->swScale= yuv2rgb_get_func_ptr(c);
2066                 }
2067 #endif
2068                 
2069                 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2070                 {
2071                         c->swScale= yvu9toyv12Wrapper;
2072                 }
2073
2074                 /* bgr24toYV12 */
2075                 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2076                         c->swScale= bgr24toyv12Wrapper;
2077                 
2078                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2079                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
2080                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
2081                    && !needsDither)
2082                         c->swScale= rgb2rgbWrapper;
2083
2084                 /* LQ converters if -sws 0 or -sws 4*/
2085                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2086                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
2087                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
2088                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
2089                           && needsDither)
2090                                 c->swScale= rgb2rgbWrapper;
2091
2092                         /* yv12_to_yuy2 */
2093                         if(srcFormat == PIX_FMT_YUV420P && 
2094                             (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2095                         {
2096                                 if (dstFormat == PIX_FMT_YUYV422)
2097                                     c->swScale= PlanarToYuy2Wrapper;
2098                                 else
2099                                     c->swScale= PlanarToUyvyWrapper;
2100                         }
2101                 }
2102
2103 #ifdef COMPILE_ALTIVEC
2104                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2105                     ((srcFormat == PIX_FMT_YUV420P && 
2106                       (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2107                   // unscaled YV12 -> packed YUV, we want speed
2108                   if (dstFormat == PIX_FMT_YUYV422)
2109                     c->swScale= yv12toyuy2_unscaled_altivec;
2110                   else
2111                     c->swScale= yv12touyvy_unscaled_altivec;
2112                 }
2113 #endif
2114
2115                 /* simple copy */
2116                 if(   srcFormat == dstFormat
2117                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2118                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2119                   )
2120                 {
2121                         c->swScale= simpleCopy;
2122                 }
2123
2124                 /* gray16{le,be} conversions */
2125                 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2126                 {
2127                         c->swScale= gray16togray;
2128                 }
2129                 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2130                 {
2131                         c->swScale= graytogray16;
2132                 }
2133                 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2134                 {
2135                         c->swScale= gray16swap;
2136                 }               
2137
2138                 if(c->swScale){
2139                         if(flags&SWS_PRINT_INFO)
2140                                 av_log(c, AV_LOG_INFO, "SwScaler: using unscaled %s -> %s special converter\n", 
2141                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2142                         return c;
2143                 }
2144         }
2145
2146         if(flags & SWS_CPU_CAPS_MMX2)
2147         {
2148                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2149                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2150                 {
2151                         if(flags&SWS_PRINT_INFO)
2152                                 av_log(c, AV_LOG_INFO, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2153                 }
2154                 if(usesHFilter) c->canMMX2BeUsed=0;
2155         }
2156         else
2157                 c->canMMX2BeUsed=0;
2158
2159         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2160         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2161
2162         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2163         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2164         // n-2 is the last chrominance sample available
2165         // this is not perfect, but noone shuld notice the difference, the more correct variant
2166         // would be like the vertical one, but that would require some special code for the
2167         // first and last pixel
2168         if(flags&SWS_FAST_BILINEAR)
2169         {
2170                 if(c->canMMX2BeUsed)
2171                 {
2172                         c->lumXInc+= 20;
2173                         c->chrXInc+= 20;
2174                 }
2175                 //we don't use the x86asm scaler if mmx is available
2176                 else if(flags & SWS_CPU_CAPS_MMX)
2177                 {
2178                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2179                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2180                 }
2181         }
2182
2183         /* precalculate horizontal scaler filter coefficients */
2184         {
2185                 const int filterAlign=
2186                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2187                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2188                   1;
2189
2190                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2191                                  srcW      ,       dstW, filterAlign, 1<<14,
2192                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2193                                  srcFilter->lumH, dstFilter->lumH, c->param);
2194                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2195                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2196                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2197                                  srcFilter->chrH, dstFilter->chrH, c->param);
2198
2199 #define MAX_FUNNY_CODE_SIZE 10000
2200 #if defined(COMPILE_MMX2)
2201 // can't downscale !!!
2202                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2203                 {
2204 #ifdef MAP_ANONYMOUS
2205                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2206                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2207 #else
2208                         c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2209                         c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2210 #endif
2211
2212                         c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2213                         c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2214                         c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2215                         c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2216
2217                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2218                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2219                 }
2220 #endif /* defined(COMPILE_MMX2) */
2221         } // Init Horizontal stuff
2222
2223
2224
2225         /* precalculate vertical scaler filter coefficients */
2226         {
2227                 const int filterAlign=
2228                   (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2229                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2230                   1;
2231
2232                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2233                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2234                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2235                                 srcFilter->lumV, dstFilter->lumV, c->param);
2236                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2237                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2238                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2239                                 srcFilter->chrV, dstFilter->chrV, c->param);
2240
2241 #ifdef HAVE_ALTIVEC
2242                 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2243                 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2244
2245                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2246                   int j;
2247                   short *p = (short *)&c->vYCoeffsBank[i];
2248                   for (j=0;j<8;j++)
2249                     p[j] = c->vLumFilter[i];
2250                 }
2251
2252                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2253                   int j;
2254                   short *p = (short *)&c->vCCoeffsBank[i];
2255                   for (j=0;j<8;j++)
2256                     p[j] = c->vChrFilter[i];
2257                 }
2258 #endif
2259         }
2260
2261         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2262         c->vLumBufSize= c->vLumFilterSize;
2263         c->vChrBufSize= c->vChrFilterSize;
2264         for(i=0; i<dstH; i++)
2265         {
2266                 int chrI= i*c->chrDstH / dstH;
2267                 int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2268                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2269
2270                 nextSlice>>= c->chrSrcVSubSample;
2271                 nextSlice<<= c->chrSrcVSubSample;
2272                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2273                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2274                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2275                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2276         }
2277
2278         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2279         c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2280         c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2281         //Note we need at least one pixel more at the end because of the mmx code (just in case someone wanna replace the 4000/8000)
2282         /* align at 16 bytes for AltiVec */
2283         for(i=0; i<c->vLumBufSize; i++)
2284                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(4000);
2285         for(i=0; i<c->vChrBufSize; i++)
2286                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2287
2288         //try to avoid drawing green stuff between the right end and the stride end
2289         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2290
2291         ASSERT(c->chrDstH <= dstH)
2292
2293         if(flags&SWS_PRINT_INFO)
2294         {
2295 #ifdef DITHER1XBPP
2296                 char *dither= " dithered";
2297 #else
2298                 char *dither= "";
2299 #endif
2300                 if(flags&SWS_FAST_BILINEAR)
2301                         av_log(c, AV_LOG_INFO, "SwScaler: FAST_BILINEAR scaler, ");
2302                 else if(flags&SWS_BILINEAR)
2303                         av_log(c, AV_LOG_INFO, "SwScaler: BILINEAR scaler, ");
2304                 else if(flags&SWS_BICUBIC)
2305                         av_log(c, AV_LOG_INFO, "SwScaler: BICUBIC scaler, ");
2306                 else if(flags&SWS_X)
2307                         av_log(c, AV_LOG_INFO, "SwScaler: Experimental scaler, ");
2308                 else if(flags&SWS_POINT)
2309                         av_log(c, AV_LOG_INFO, "SwScaler: Nearest Neighbor / POINT scaler, ");
2310                 else if(flags&SWS_AREA)
2311                         av_log(c, AV_LOG_INFO, "SwScaler: Area Averageing scaler, ");
2312                 else if(flags&SWS_BICUBLIN)
2313                         av_log(c, AV_LOG_INFO, "SwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2314                 else if(flags&SWS_GAUSS)
2315                         av_log(c, AV_LOG_INFO, "SwScaler: Gaussian scaler, ");
2316                 else if(flags&SWS_SINC)
2317                         av_log(c, AV_LOG_INFO, "SwScaler: Sinc scaler, ");
2318                 else if(flags&SWS_LANCZOS)
2319                         av_log(c, AV_LOG_INFO, "SwScaler: Lanczos scaler, ");
2320                 else if(flags&SWS_SPLINE)
2321                         av_log(c, AV_LOG_INFO, "SwScaler: Bicubic spline scaler, ");
2322                 else
2323                         av_log(c, AV_LOG_INFO, "SwScaler: ehh flags invalid?! ");
2324
2325                 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2326                         av_log(c, AV_LOG_INFO, "from %s to%s %s ", 
2327                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2328                 else
2329                         av_log(c, AV_LOG_INFO, "from %s to %s ", 
2330                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2331
2332                 if(flags & SWS_CPU_CAPS_MMX2)
2333                         av_log(c, AV_LOG_INFO, "using MMX2\n");
2334                 else if(flags & SWS_CPU_CAPS_3DNOW)
2335                         av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2336                 else if(flags & SWS_CPU_CAPS_MMX)
2337                         av_log(c, AV_LOG_INFO, "using MMX\n");
2338                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2339                         av_log(c, AV_LOG_INFO, "using AltiVec\n");
2340                 else 
2341                         av_log(c, AV_LOG_INFO, "using C\n");
2342         }
2343
2344         if(flags & SWS_PRINT_INFO)
2345         {
2346                 if(flags & SWS_CPU_CAPS_MMX)
2347                 {
2348                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2349                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2350                         else
2351                         {
2352                                 if(c->hLumFilterSize==4)
2353                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2354                                 else if(c->hLumFilterSize==8)
2355                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2356                                 else
2357                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2358
2359                                 if(c->hChrFilterSize==4)
2360                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2361                                 else if(c->hChrFilterSize==8)
2362                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2363                                 else
2364                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2365                         }
2366                 }
2367                 else
2368                 {
2369 #if defined(ARCH_X86)
2370                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using X86-Asm scaler for horizontal scaling\n");
2371 #else
2372                         if(flags & SWS_FAST_BILINEAR)
2373                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2374                         else
2375                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using C scaler for horizontal scaling\n");
2376 #endif
2377                 }
2378                 if(isPlanarYUV(dstFormat))
2379                 {
2380                         if(c->vLumFilterSize==1)
2381                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2382                         else
2383                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2384                 }
2385                 else
2386                 {
2387                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2388                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2389                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2390                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2391                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2392                         else
2393                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2394                 }
2395
2396                 if(dstFormat==PIX_FMT_BGR24)
2397                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR24 Converter\n",
2398                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2399                 else if(dstFormat==PIX_FMT_RGB32)
2400                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2401                 else if(dstFormat==PIX_FMT_BGR565)
2402                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2403                 else if(dstFormat==PIX_FMT_BGR555)
2404                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2405
2406                 av_log(c, AV_LOG_VERBOSE, "SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2407         }
2408         if(flags & SWS_PRINT_INFO)
2409         {
2410                 av_log(c, AV_LOG_DEBUG, "SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2411                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2412                 av_log(c, AV_LOG_DEBUG, "SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2413                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2414         }
2415
2416         c->swScale= getSwsFunc(flags);
2417         return c;
2418 }
2419
2420 /**
2421  * swscale warper, so we don't need to export the SwsContext.
2422  * assumes planar YUV to be in YUV order instead of YVU
2423  */
2424 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2425                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2426         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2427             av_log(c, AV_LOG_ERROR, "swScaler: slices start in the middle!\n");
2428             return 0;
2429         }
2430         if (c->sliceDir == 0) {
2431             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2432         }
2433
2434         // copy strides, so they can safely be modified
2435         if (c->sliceDir == 1) {
2436             // slices go from top to bottom
2437             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2438             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2439             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2440         } else {
2441             // slices go from bottom to top => we flip the image internally
2442             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2443                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2444                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2445             };
2446             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2447                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2448                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2449             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2450             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2451             
2452             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2453         }
2454 }
2455
2456 /**
2457  * swscale warper, so we don't need to export the SwsContext
2458  */
2459 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2460                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2461         return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2462 }
2463
2464 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2465                                 float lumaSharpen, float chromaSharpen,
2466                                 float chromaHShift, float chromaVShift,
2467                                 int verbose)
2468 {
2469         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2470
2471         if(lumaGBlur!=0.0){
2472                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2473                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2474         }else{
2475                 filter->lumH= sws_getIdentityVec();
2476                 filter->lumV= sws_getIdentityVec();
2477         }
2478
2479         if(chromaGBlur!=0.0){
2480                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2481                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2482         }else{
2483                 filter->chrH= sws_getIdentityVec();
2484                 filter->chrV= sws_getIdentityVec();
2485         }
2486
2487         if(chromaSharpen!=0.0){
2488                 SwsVector *id= sws_getIdentityVec();
2489                 sws_scaleVec(filter->chrH, -chromaSharpen);
2490                 sws_scaleVec(filter->chrV, -chromaSharpen);
2491                 sws_addVec(filter->chrH, id);
2492                 sws_addVec(filter->chrV, id);
2493                 sws_freeVec(id);
2494         }
2495
2496         if(lumaSharpen!=0.0){
2497                 SwsVector *id= sws_getIdentityVec();
2498                 sws_scaleVec(filter->lumH, -lumaSharpen);
2499                 sws_scaleVec(filter->lumV, -lumaSharpen);
2500                 sws_addVec(filter->lumH, id);
2501                 sws_addVec(filter->lumV, id);
2502                 sws_freeVec(id);
2503         }
2504
2505         if(chromaHShift != 0.0)
2506                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2507
2508         if(chromaVShift != 0.0)
2509                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2510
2511         sws_normalizeVec(filter->chrH, 1.0);
2512         sws_normalizeVec(filter->chrV, 1.0);
2513         sws_normalizeVec(filter->lumH, 1.0);
2514         sws_normalizeVec(filter->lumV, 1.0);
2515
2516         if(verbose) sws_printVec(filter->chrH);
2517         if(verbose) sws_printVec(filter->lumH);
2518
2519         return filter;
2520 }
2521
2522 /**
2523  * returns a normalized gaussian curve used to filter stuff
2524  * quality=3 is high quality, lowwer is lowwer quality
2525  */
2526 SwsVector *sws_getGaussianVec(double variance, double quality){
2527         const int length= (int)(variance*quality + 0.5) | 1;
2528         int i;
2529         double *coeff= av_malloc(length*sizeof(double));
2530         double middle= (length-1)*0.5;
2531         SwsVector *vec= av_malloc(sizeof(SwsVector));
2532
2533         vec->coeff= coeff;
2534         vec->length= length;
2535
2536         for(i=0; i<length; i++)
2537         {
2538                 double dist= i-middle;
2539                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2540         }
2541
2542         sws_normalizeVec(vec, 1.0);
2543
2544         return vec;
2545 }
2546
2547 SwsVector *sws_getConstVec(double c, int length){
2548         int i;
2549         double *coeff= av_malloc(length*sizeof(double));
2550         SwsVector *vec= av_malloc(sizeof(SwsVector));
2551
2552         vec->coeff= coeff;
2553         vec->length= length;
2554
2555         for(i=0; i<length; i++)
2556                 coeff[i]= c;
2557
2558         return vec;
2559 }
2560
2561
2562 SwsVector *sws_getIdentityVec(void){
2563         return sws_getConstVec(1.0, 1);
2564 }
2565
2566 double sws_dcVec(SwsVector *a){
2567         int i;
2568         double sum=0;
2569
2570         for(i=0; i<a->length; i++)
2571                 sum+= a->coeff[i];
2572
2573         return sum;
2574 }
2575
2576 void sws_scaleVec(SwsVector *a, double scalar){
2577         int i;
2578
2579         for(i=0; i<a->length; i++)
2580                 a->coeff[i]*= scalar;
2581 }
2582
2583 void sws_normalizeVec(SwsVector *a, double height){
2584         sws_scaleVec(a, height/sws_dcVec(a));
2585 }
2586
2587 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2588         int length= a->length + b->length - 1;
2589         double *coeff= av_malloc(length*sizeof(double));
2590         int i, j;
2591         SwsVector *vec= av_malloc(sizeof(SwsVector));
2592
2593         vec->coeff= coeff;
2594         vec->length= length;
2595
2596         for(i=0; i<length; i++) coeff[i]= 0.0;
2597
2598         for(i=0; i<a->length; i++)
2599         {
2600                 for(j=0; j<b->length; j++)
2601                 {
2602                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2603                 }
2604         }
2605
2606         return vec;
2607 }
2608
2609 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2610         int length= FFMAX(a->length, b->length);
2611         double *coeff= av_malloc(length*sizeof(double));
2612         int i;
2613         SwsVector *vec= av_malloc(sizeof(SwsVector));
2614
2615         vec->coeff= coeff;
2616         vec->length= length;
2617
2618         for(i=0; i<length; i++) coeff[i]= 0.0;
2619
2620         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2621         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2622
2623         return vec;
2624 }
2625
2626 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2627         int length= FFMAX(a->length, b->length);
2628         double *coeff= av_malloc(length*sizeof(double));
2629         int i;
2630         SwsVector *vec= av_malloc(sizeof(SwsVector));
2631
2632         vec->coeff= coeff;
2633         vec->length= length;
2634
2635         for(i=0; i<length; i++) coeff[i]= 0.0;
2636
2637         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2638         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2639
2640         return vec;
2641 }
2642
2643 /* shift left / or right if "shift" is negative */
2644 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2645         int length= a->length + FFABS(shift)*2;
2646         double *coeff= av_malloc(length*sizeof(double));
2647         int i;
2648         SwsVector *vec= av_malloc(sizeof(SwsVector));
2649
2650         vec->coeff= coeff;
2651         vec->length= length;
2652
2653         for(i=0; i<length; i++) coeff[i]= 0.0;
2654
2655         for(i=0; i<a->length; i++)
2656         {
2657                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2658         }
2659
2660         return vec;
2661 }
2662
2663 void sws_shiftVec(SwsVector *a, int shift){
2664         SwsVector *shifted= sws_getShiftedVec(a, shift);
2665         av_free(a->coeff);
2666         a->coeff= shifted->coeff;
2667         a->length= shifted->length;
2668         av_free(shifted);
2669 }
2670
2671 void sws_addVec(SwsVector *a, SwsVector *b){
2672         SwsVector *sum= sws_sumVec(a, b);
2673         av_free(a->coeff);
2674         a->coeff= sum->coeff;
2675         a->length= sum->length;
2676         av_free(sum);
2677 }
2678
2679 void sws_subVec(SwsVector *a, SwsVector *b){
2680         SwsVector *diff= sws_diffVec(a, b);
2681         av_free(a->coeff);
2682         a->coeff= diff->coeff;
2683         a->length= diff->length;
2684         av_free(diff);
2685 }
2686
2687 void sws_convVec(SwsVector *a, SwsVector *b){
2688         SwsVector *conv= sws_getConvVec(a, b);
2689         av_free(a->coeff);  
2690         a->coeff= conv->coeff;
2691         a->length= conv->length;
2692         av_free(conv);
2693 }
2694
2695 SwsVector *sws_cloneVec(SwsVector *a){
2696         double *coeff= av_malloc(a->length*sizeof(double));
2697         int i;
2698         SwsVector *vec= av_malloc(sizeof(SwsVector));
2699
2700         vec->coeff= coeff;
2701         vec->length= a->length;
2702
2703         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2704
2705         return vec;
2706 }
2707
2708 void sws_printVec(SwsVector *a){
2709         int i;
2710         double max=0;
2711         double min=0;
2712         double range;
2713
2714         for(i=0; i<a->length; i++)
2715                 if(a->coeff[i]>max) max= a->coeff[i];
2716
2717         for(i=0; i<a->length; i++)
2718                 if(a->coeff[i]<min) min= a->coeff[i];
2719
2720         range= max - min;
2721
2722         for(i=0; i<a->length; i++)
2723         {
2724                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2725                 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2726                 for(;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2727                 av_log(NULL, AV_LOG_DEBUG, "|\n");
2728         }
2729 }
2730
2731 void sws_freeVec(SwsVector *a){
2732         if(!a) return;
2733         av_free(a->coeff);
2734         a->coeff=NULL;
2735         a->length=0;
2736         av_free(a);
2737 }
2738
2739 void sws_freeFilter(SwsFilter *filter){
2740         if(!filter) return;
2741
2742         if(filter->lumH) sws_freeVec(filter->lumH);
2743         if(filter->lumV) sws_freeVec(filter->lumV);
2744         if(filter->chrH) sws_freeVec(filter->chrH);
2745         if(filter->chrV) sws_freeVec(filter->chrV);
2746         av_free(filter);
2747 }
2748
2749
2750 void sws_freeContext(SwsContext *c){
2751         int i;
2752         if(!c) return;
2753
2754         if(c->lumPixBuf)
2755         {
2756                 for(i=0; i<c->vLumBufSize; i++)
2757                 {
2758                         av_free(c->lumPixBuf[i]);
2759                         c->lumPixBuf[i]=NULL;
2760                 }
2761                 av_free(c->lumPixBuf);
2762                 c->lumPixBuf=NULL;
2763         }
2764
2765         if(c->chrPixBuf)
2766         {
2767                 for(i=0; i<c->vChrBufSize; i++)
2768                 {
2769                         av_free(c->chrPixBuf[i]);
2770                         c->chrPixBuf[i]=NULL;
2771                 }
2772                 av_free(c->chrPixBuf);
2773                 c->chrPixBuf=NULL;
2774         }
2775
2776         av_free(c->vLumFilter);
2777         c->vLumFilter = NULL;
2778         av_free(c->vChrFilter);
2779         c->vChrFilter = NULL;
2780         av_free(c->hLumFilter);
2781         c->hLumFilter = NULL;
2782         av_free(c->hChrFilter);
2783         c->hChrFilter = NULL;
2784 #ifdef HAVE_ALTIVEC
2785         av_free(c->vYCoeffsBank);
2786         c->vYCoeffsBank = NULL;
2787         av_free(c->vCCoeffsBank);
2788         c->vCCoeffsBank = NULL;
2789 #endif
2790
2791         av_free(c->vLumFilterPos);
2792         c->vLumFilterPos = NULL;
2793         av_free(c->vChrFilterPos);
2794         c->vChrFilterPos = NULL;
2795         av_free(c->hLumFilterPos);
2796         c->hLumFilterPos = NULL;
2797         av_free(c->hChrFilterPos);
2798         c->hChrFilterPos = NULL;
2799
2800 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2801 #ifdef MAP_ANONYMOUS
2802         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2803         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2804 #else
2805         av_free(c->funnyYCode);
2806         av_free(c->funnyUVCode);
2807 #endif
2808         c->funnyYCode=NULL;
2809         c->funnyUVCode=NULL;
2810 #endif /* defined(ARCH_X86) */
2811
2812         av_free(c->lumMmx2Filter);
2813         c->lumMmx2Filter=NULL;
2814         av_free(c->chrMmx2Filter);
2815         c->chrMmx2Filter=NULL;
2816         av_free(c->lumMmx2FilterPos);
2817         c->lumMmx2FilterPos=NULL;
2818         av_free(c->chrMmx2FilterPos);
2819         c->chrMmx2FilterPos=NULL;
2820         av_free(c->yuvTable);
2821         c->yuvTable=NULL;
2822
2823         av_free(c);
2824 }
2825
2826 /**
2827  * Checks if context is valid or reallocs a new one instead.
2828  * If context is NULL, just calls sws_getContext() to get a new one.
2829  * Otherwise, checks if the parameters are the same already saved in context.
2830  * If that is the case, returns the current context.
2831  * Otherwise, frees context and gets a new one.
2832  *
2833  * Be warned that srcFilter, dstFilter are not checked, they are
2834  * asumed to remain valid.
2835  */
2836 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2837                         int srcW, int srcH, int srcFormat,
2838                         int dstW, int dstH, int dstFormat, int flags,
2839                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2840 {
2841     if (context != NULL) {
2842         if ((context->srcW != srcW) || (context->srcH != srcH) ||
2843             (context->srcFormat != srcFormat) ||
2844             (context->dstW != dstW) || (context->dstH != dstH) ||
2845             (context->dstFormat != dstFormat) || (context->flags != flags) ||
2846             (context->param != param))
2847         {
2848             sws_freeContext(context);
2849             context = NULL;
2850         }
2851     }
2852     if (context == NULL) {
2853         return sws_getContext(srcW, srcH, srcFormat,
2854                         dstW, dstH, dstFormat, flags,
2855                         srcFilter, dstFilter, param);
2856     }
2857     return context;
2858 }
2859