test widths all down to one. this tests whether the scalers correctly switch
[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, PAL8
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                         || (x)==PIX_FMT_PAL8 || (x)==PIX_FMT_BGR8 || (x)==PIX_FMT_RGB8\
112                         || (x)==PIX_FMT_BGR4_BYTE  || (x)==PIX_FMT_RGB4_BYTE)
113 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
114                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
115                         || isRGB(x) || isBGR(x)\
116                         || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
117                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
118                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
119 #define isPacked(x)    ((x)==PIX_FMT_PAL8 || (x)==PIX_FMT_YUYV422 ||\
120                         (x)==PIX_FMT_UYVY422 || isRGB(x) || isBGR(x))
121
122 #define RGB2YUV_SHIFT 16
123 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
124 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
125 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
126 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
127 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
128 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
129 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
130 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
131 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
132
133 extern const int32_t Inverse_Table_6_9[8][4];
134
135 /*
136 NOTES
137 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
138
139 TODO
140 more intelligent missalignment avoidance for the horizontal scaler
141 write special vertical cubic upscale version
142 Optimize C code (yv12 / minmax)
143 add support for packed pixel yuv input & output
144 add support for Y8 output
145 optimize bgr24 & bgr32
146 add BGR4 output support
147 write special BGR->BGR scaler
148 */
149
150 #if defined(ARCH_X86) && defined (CONFIG_GPL)
151 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
152 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
153 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
154 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
155 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
156 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
157 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
158 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
159
160 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
161 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
162 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
163 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
164
165 static uint64_t __attribute__((aligned(8))) dither4[2]={
166         0x0103010301030103LL,
167         0x0200020002000200LL,};
168
169 static uint64_t __attribute__((aligned(8))) dither8[2]={
170         0x0602060206020602LL,
171         0x0004000400040004LL,};
172
173 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
174 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
175 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
176 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
177 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
178 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
179
180 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
181 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
182 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
183
184 #ifdef FAST_BGR2YV12
185 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
186 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
187 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
188 #else
189 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
190 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
191 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
192 #endif /* FAST_BGR2YV12 */
193 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
194 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
195 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
196 #endif /* defined(ARCH_X86) */
197
198 // clipping helper table for C implementations:
199 static unsigned char clip_table[768];
200
201 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
202                   
203 extern const uint8_t dither_2x2_4[2][8];
204 extern const uint8_t dither_2x2_8[2][8];
205 extern const uint8_t dither_8x8_32[8][8];
206 extern const uint8_t dither_8x8_73[8][8];
207 extern const uint8_t dither_8x8_220[8][8];
208
209 static const char * sws_context_to_name(void * ptr) {
210     return "swscaler";
211 }
212
213 static AVClass sws_context_class = { "SWScaler", sws_context_to_name, NULL };
214
215 char *sws_format_name(enum PixelFormat format)
216 {
217     switch (format) {
218         case PIX_FMT_YUV420P:
219             return "yuv420p";
220         case PIX_FMT_YUYV422:
221             return "yuyv422";
222         case PIX_FMT_RGB24:
223             return "rgb24";
224         case PIX_FMT_BGR24:
225             return "bgr24";
226         case PIX_FMT_YUV422P:
227             return "yuv422p";
228         case PIX_FMT_YUV444P:
229             return "yuv444p";
230         case PIX_FMT_RGB32:
231             return "rgb32";
232         case PIX_FMT_YUV410P:
233             return "yuv410p";
234         case PIX_FMT_YUV411P:
235             return "yuv411p";
236         case PIX_FMT_RGB565:
237             return "rgb565";
238         case PIX_FMT_RGB555:
239             return "rgb555";
240         case PIX_FMT_GRAY16BE:
241             return "gray16be";
242         case PIX_FMT_GRAY16LE:
243             return "gray16le";
244         case PIX_FMT_GRAY8:
245             return "gray8";
246         case PIX_FMT_MONOWHITE:
247             return "mono white";
248         case PIX_FMT_MONOBLACK:
249             return "mono black";
250         case PIX_FMT_PAL8:
251             return "Palette";
252         case PIX_FMT_YUVJ420P:
253             return "yuvj420p";
254         case PIX_FMT_YUVJ422P:
255             return "yuvj422p";
256         case PIX_FMT_YUVJ444P:
257             return "yuvj444p";
258         case PIX_FMT_XVMC_MPEG2_MC:
259             return "xvmc_mpeg2_mc";
260         case PIX_FMT_XVMC_MPEG2_IDCT:
261             return "xvmc_mpeg2_idct";
262         case PIX_FMT_UYVY422:
263             return "uyvy422";
264         case PIX_FMT_UYYVYY411:
265             return "uyyvyy411";
266         case PIX_FMT_RGB32_1:
267             return "rgb32x";
268         case PIX_FMT_BGR32_1:
269             return "bgr32x";
270         case PIX_FMT_BGR32:
271             return "bgr32";
272         case PIX_FMT_BGR565:
273             return "bgr565";
274         case PIX_FMT_BGR555:
275             return "bgr555";
276         case PIX_FMT_BGR8:
277             return "bgr8";
278         case PIX_FMT_BGR4:
279             return "bgr4";
280         case PIX_FMT_BGR4_BYTE:
281             return "bgr4 byte";
282         case PIX_FMT_RGB8:
283             return "rgb8";
284         case PIX_FMT_RGB4:
285             return "rgb4";
286         case PIX_FMT_RGB4_BYTE:
287             return "rgb4 byte";
288         case PIX_FMT_NV12:
289             return "nv12";
290         case PIX_FMT_NV21:
291             return "nv21";
292         default:
293             return "Unknown format";
294     }
295 }
296
297 #if defined(ARCH_X86) && defined (CONFIG_GPL)
298 void in_asm_used_var_warning_killer()
299 {
300  volatile int i= bF8+bFC+w10+
301  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
302  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
303  if(i) i=0;
304 }
305 #endif
306
307 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
308                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
309                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
310 {
311         //FIXME Optimize (just quickly writen not opti..)
312         int i;
313         for(i=0; i<dstW; i++)
314         {
315                 int val=1<<18;
316                 int j;
317                 for(j=0; j<lumFilterSize; j++)
318                         val += lumSrc[j][i] * lumFilter[j];
319
320                 dest[i]= av_clip_uint8(val>>19);
321         }
322
323         if(uDest != NULL)
324                 for(i=0; i<chrDstW; i++)
325                 {
326                         int u=1<<18;
327                         int v=1<<18;
328                         int j;
329                         for(j=0; j<chrFilterSize; j++)
330                         {
331                                 u += chrSrc[j][i] * chrFilter[j];
332                                 v += chrSrc[j][i + 2048] * chrFilter[j];
333                         }
334
335                         uDest[i]= av_clip_uint8(u>>19);
336                         vDest[i]= av_clip_uint8(v>>19);
337                 }
338 }
339
340 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
341                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
342                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
343 {
344         //FIXME Optimize (just quickly writen not opti..)
345         int i;
346         for(i=0; i<dstW; i++)
347         {
348                 int val=1<<18;
349                 int j;
350                 for(j=0; j<lumFilterSize; j++)
351                         val += lumSrc[j][i] * lumFilter[j];
352
353                 dest[i]= av_clip_uint8(val>>19);
354         }
355
356         if(uDest == NULL)
357                 return;
358
359         if(dstFormat == PIX_FMT_NV12)
360                 for(i=0; i<chrDstW; i++)
361                 {
362                         int u=1<<18;
363                         int v=1<<18;
364                         int j;
365                         for(j=0; j<chrFilterSize; j++)
366                         {
367                                 u += chrSrc[j][i] * chrFilter[j];
368                                 v += chrSrc[j][i + 2048] * chrFilter[j];
369                         }
370
371                         uDest[2*i]= av_clip_uint8(u>>19);
372                         uDest[2*i+1]= av_clip_uint8(v>>19);
373                 }
374         else
375                 for(i=0; i<chrDstW; i++)
376                 {
377                         int u=1<<18;
378                         int v=1<<18;
379                         int j;
380                         for(j=0; j<chrFilterSize; j++)
381                         {
382                                 u += chrSrc[j][i] * chrFilter[j];
383                                 v += chrSrc[j][i + 2048] * chrFilter[j];
384                         }
385
386                         uDest[2*i]= av_clip_uint8(v>>19);
387                         uDest[2*i+1]= av_clip_uint8(u>>19);
388                 }
389 }
390
391 #define YSCALE_YUV_2_PACKEDX_C(type) \
392                 for(i=0; i<(dstW>>1); i++){\
393                         int j;\
394                         int Y1=1<<18;\
395                         int Y2=1<<18;\
396                         int U=1<<18;\
397                         int V=1<<18;\
398                         type attribute_unused *r, *b, *g;\
399                         const int i2= 2*i;\
400                         \
401                         for(j=0; j<lumFilterSize; j++)\
402                         {\
403                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
404                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
405                         }\
406                         for(j=0; j<chrFilterSize; j++)\
407                         {\
408                                 U += chrSrc[j][i] * chrFilter[j];\
409                                 V += chrSrc[j][i+2048] * chrFilter[j];\
410                         }\
411                         Y1>>=19;\
412                         Y2>>=19;\
413                         U >>=19;\
414                         V >>=19;\
415                         if((Y1|Y2|U|V)&256)\
416                         {\
417                                 if(Y1>255)   Y1=255;\
418                                 else if(Y1<0)Y1=0;\
419                                 if(Y2>255)   Y2=255;\
420                                 else if(Y2<0)Y2=0;\
421                                 if(U>255)    U=255;\
422                                 else if(U<0) U=0;\
423                                 if(V>255)    V=255;\
424                                 else if(V<0) V=0;\
425                         }
426                         
427 #define YSCALE_YUV_2_RGBX_C(type) \
428                         YSCALE_YUV_2_PACKEDX_C(type)\
429                         r = (type *)c->table_rV[V];\
430                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
431                         b = (type *)c->table_bU[U];\
432
433 #define YSCALE_YUV_2_PACKED2_C \
434                 for(i=0; i<(dstW>>1); i++){\
435                         const int i2= 2*i;\
436                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
437                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
438                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
439                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
440
441 #define YSCALE_YUV_2_RGB2_C(type) \
442                         YSCALE_YUV_2_PACKED2_C\
443                         type *r, *b, *g;\
444                         r = (type *)c->table_rV[V];\
445                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
446                         b = (type *)c->table_bU[U];\
447
448 #define YSCALE_YUV_2_PACKED1_C \
449                 for(i=0; i<(dstW>>1); i++){\
450                         const int i2= 2*i;\
451                         int Y1= buf0[i2  ]>>7;\
452                         int Y2= buf0[i2+1]>>7;\
453                         int U= (uvbuf1[i     ])>>7;\
454                         int V= (uvbuf1[i+2048])>>7;\
455
456 #define YSCALE_YUV_2_RGB1_C(type) \
457                         YSCALE_YUV_2_PACKED1_C\
458                         type *r, *b, *g;\
459                         r = (type *)c->table_rV[V];\
460                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
461                         b = (type *)c->table_bU[U];\
462
463 #define YSCALE_YUV_2_PACKED1B_C \
464                 for(i=0; i<(dstW>>1); i++){\
465                         const int i2= 2*i;\
466                         int Y1= buf0[i2  ]>>7;\
467                         int Y2= buf0[i2+1]>>7;\
468                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
469                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
470
471 #define YSCALE_YUV_2_RGB1B_C(type) \
472                         YSCALE_YUV_2_PACKED1B_C\
473                         type *r, *b, *g;\
474                         r = (type *)c->table_rV[V];\
475                         g = (type *)(c->table_gU[U] + c->table_gV[V]);\
476                         b = (type *)c->table_bU[U];\
477
478 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
479         switch(c->dstFormat)\
480         {\
481         case PIX_FMT_RGB32:\
482         case PIX_FMT_BGR32:\
483                 func(uint32_t)\
484                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
485                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
486                 }               \
487                 break;\
488         case PIX_FMT_RGB24:\
489                 func(uint8_t)\
490                         ((uint8_t*)dest)[0]= r[Y1];\
491                         ((uint8_t*)dest)[1]= g[Y1];\
492                         ((uint8_t*)dest)[2]= b[Y1];\
493                         ((uint8_t*)dest)[3]= r[Y2];\
494                         ((uint8_t*)dest)[4]= g[Y2];\
495                         ((uint8_t*)dest)[5]= b[Y2];\
496                         dest+=6;\
497                 }\
498                 break;\
499         case PIX_FMT_BGR24:\
500                 func(uint8_t)\
501                         ((uint8_t*)dest)[0]= b[Y1];\
502                         ((uint8_t*)dest)[1]= g[Y1];\
503                         ((uint8_t*)dest)[2]= r[Y1];\
504                         ((uint8_t*)dest)[3]= b[Y2];\
505                         ((uint8_t*)dest)[4]= g[Y2];\
506                         ((uint8_t*)dest)[5]= r[Y2];\
507                         dest+=6;\
508                 }\
509                 break;\
510         case PIX_FMT_RGB565:\
511         case PIX_FMT_BGR565:\
512                 {\
513                         const int dr1= dither_2x2_8[y&1    ][0];\
514                         const int dg1= dither_2x2_4[y&1    ][0];\
515                         const int db1= dither_2x2_8[(y&1)^1][0];\
516                         const int dr2= dither_2x2_8[y&1    ][1];\
517                         const int dg2= dither_2x2_4[y&1    ][1];\
518                         const int db2= dither_2x2_8[(y&1)^1][1];\
519                         func(uint16_t)\
520                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
521                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
522                         }\
523                 }\
524                 break;\
525         case PIX_FMT_RGB555:\
526         case PIX_FMT_BGR555:\
527                 {\
528                         const int dr1= dither_2x2_8[y&1    ][0];\
529                         const int dg1= dither_2x2_8[y&1    ][1];\
530                         const int db1= dither_2x2_8[(y&1)^1][0];\
531                         const int dr2= dither_2x2_8[y&1    ][1];\
532                         const int dg2= dither_2x2_8[y&1    ][0];\
533                         const int db2= dither_2x2_8[(y&1)^1][1];\
534                         func(uint16_t)\
535                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
536                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
537                         }\
538                 }\
539                 break;\
540         case PIX_FMT_RGB8:\
541         case PIX_FMT_BGR8:\
542                 {\
543                         const uint8_t * const d64= dither_8x8_73[y&7];\
544                         const uint8_t * const d32= dither_8x8_32[y&7];\
545                         func(uint8_t)\
546                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
547                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
548                         }\
549                 }\
550                 break;\
551         case PIX_FMT_RGB4:\
552         case PIX_FMT_BGR4:\
553                 {\
554                         const uint8_t * const d64= dither_8x8_73 [y&7];\
555                         const uint8_t * const d128=dither_8x8_220[y&7];\
556                         func(uint8_t)\
557                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
558                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
559                         }\
560                 }\
561                 break;\
562         case PIX_FMT_RGB4_BYTE:\
563         case PIX_FMT_BGR4_BYTE:\
564                 {\
565                         const uint8_t * const d64= dither_8x8_73 [y&7];\
566                         const uint8_t * const d128=dither_8x8_220[y&7];\
567                         func(uint8_t)\
568                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
569                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
570                         }\
571                 }\
572                 break;\
573         case PIX_FMT_MONOBLACK:\
574                 {\
575                         const uint8_t * const d128=dither_8x8_220[y&7];\
576                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
577                         for(i=0; i<dstW-7; i+=8){\
578                                 int acc;\
579                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
580                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
581                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
582                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
583                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
584                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
585                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
586                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
587                                 ((uint8_t*)dest)[0]= acc;\
588                                 dest++;\
589                         }\
590 \
591 /*\
592 ((uint8_t*)dest)-= dstW>>4;\
593 {\
594                         int acc=0;\
595                         int left=0;\
596                         static int top[1024];\
597                         static int last_new[1024][1024];\
598                         static int last_in3[1024][1024];\
599                         static int drift[1024][1024];\
600                         int topLeft=0;\
601                         int shift=0;\
602                         int count=0;\
603                         const uint8_t * const d128=dither_8x8_220[y&7];\
604                         int error_new=0;\
605                         int error_in3=0;\
606                         int f=0;\
607                         \
608                         for(i=dstW>>1; i<dstW; i++){\
609                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
610                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
611                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
612                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
613                                         + (last_new[y][i] - in3)*f/256;\
614                                 int new= old> 128 ? 255 : 0;\
615 \
616                                 error_new+= FFABS(last_new[y][i] - new);\
617                                 error_in3+= FFABS(last_in3[y][i] - in3);\
618                                 f= error_new - error_in3*4;\
619                                 if(f<0) f=0;\
620                                 if(f>256) f=256;\
621 \
622                                 topLeft= top[i];\
623                                 left= top[i]= old - new;\
624                                 last_new[y][i]= new;\
625                                 last_in3[y][i]= in3;\
626 \
627                                 acc+= acc + (new&1);\
628                                 if((i&7)==6){\
629                                         ((uint8_t*)dest)[0]= acc;\
630                                         ((uint8_t*)dest)++;\
631                                 }\
632                         }\
633 }\
634 */\
635                 }\
636                 break;\
637         case PIX_FMT_YUYV422:\
638                 func2\
639                         ((uint8_t*)dest)[2*i2+0]= Y1;\
640                         ((uint8_t*)dest)[2*i2+1]= U;\
641                         ((uint8_t*)dest)[2*i2+2]= Y2;\
642                         ((uint8_t*)dest)[2*i2+3]= V;\
643                 }               \
644                 break;\
645         case PIX_FMT_UYVY422:\
646                 func2\
647                         ((uint8_t*)dest)[2*i2+0]= U;\
648                         ((uint8_t*)dest)[2*i2+1]= Y1;\
649                         ((uint8_t*)dest)[2*i2+2]= V;\
650                         ((uint8_t*)dest)[2*i2+3]= Y2;\
651                 }               \
652                 break;\
653         }\
654
655
656 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
657                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
658                                     uint8_t *dest, int dstW, int y)
659 {
660         int i;
661         switch(c->dstFormat)
662         {
663         case PIX_FMT_BGR32:
664         case PIX_FMT_RGB32:
665                 YSCALE_YUV_2_RGBX_C(uint32_t)
666                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
667                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
668                 }
669                 break;
670         case PIX_FMT_RGB24:
671                 YSCALE_YUV_2_RGBX_C(uint8_t)
672                         ((uint8_t*)dest)[0]= r[Y1];
673                         ((uint8_t*)dest)[1]= g[Y1];
674                         ((uint8_t*)dest)[2]= b[Y1];
675                         ((uint8_t*)dest)[3]= r[Y2];
676                         ((uint8_t*)dest)[4]= g[Y2];
677                         ((uint8_t*)dest)[5]= b[Y2];
678                         dest+=6;
679                 }
680                 break;
681         case PIX_FMT_BGR24:
682                 YSCALE_YUV_2_RGBX_C(uint8_t)
683                         ((uint8_t*)dest)[0]= b[Y1];
684                         ((uint8_t*)dest)[1]= g[Y1];
685                         ((uint8_t*)dest)[2]= r[Y1];
686                         ((uint8_t*)dest)[3]= b[Y2];
687                         ((uint8_t*)dest)[4]= g[Y2];
688                         ((uint8_t*)dest)[5]= r[Y2];
689                         dest+=6;
690                 }
691                 break;
692         case PIX_FMT_RGB565:
693         case PIX_FMT_BGR565:
694                 {
695                         const int dr1= dither_2x2_8[y&1    ][0];
696                         const int dg1= dither_2x2_4[y&1    ][0];
697                         const int db1= dither_2x2_8[(y&1)^1][0];
698                         const int dr2= dither_2x2_8[y&1    ][1];
699                         const int dg2= dither_2x2_4[y&1    ][1];
700                         const int db2= dither_2x2_8[(y&1)^1][1];
701                         YSCALE_YUV_2_RGBX_C(uint16_t)
702                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
703                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
704                         }
705                 }
706                 break;
707         case PIX_FMT_RGB555:
708         case PIX_FMT_BGR555:
709                 {
710                         const int dr1= dither_2x2_8[y&1    ][0];
711                         const int dg1= dither_2x2_8[y&1    ][1];
712                         const int db1= dither_2x2_8[(y&1)^1][0];
713                         const int dr2= dither_2x2_8[y&1    ][1];
714                         const int dg2= dither_2x2_8[y&1    ][0];
715                         const int db2= dither_2x2_8[(y&1)^1][1];
716                         YSCALE_YUV_2_RGBX_C(uint16_t)
717                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
718                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
719                         }
720                 }
721                 break;
722         case PIX_FMT_RGB8:
723         case PIX_FMT_BGR8:
724                 {
725                         const uint8_t * const d64= dither_8x8_73[y&7];
726                         const uint8_t * const d32= dither_8x8_32[y&7];
727                         YSCALE_YUV_2_RGBX_C(uint8_t)
728                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
729                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
730                         }
731                 }
732                 break;
733         case PIX_FMT_RGB4:
734         case PIX_FMT_BGR4:
735                 {
736                         const uint8_t * const d64= dither_8x8_73 [y&7];
737                         const uint8_t * const d128=dither_8x8_220[y&7];
738                         YSCALE_YUV_2_RGBX_C(uint8_t)
739                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
740                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
741                         }
742                 }
743                 break;
744         case PIX_FMT_RGB4_BYTE:
745         case PIX_FMT_BGR4_BYTE:
746                 {
747                         const uint8_t * const d64= dither_8x8_73 [y&7];
748                         const uint8_t * const d128=dither_8x8_220[y&7];
749                         YSCALE_YUV_2_RGBX_C(uint8_t)
750                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
751                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
752                         }
753                 }
754                 break;
755         case PIX_FMT_MONOBLACK:
756                 {
757                         const uint8_t * const d128=dither_8x8_220[y&7];
758                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
759                         int acc=0;
760                         for(i=0; i<dstW-1; i+=2){
761                                 int j;
762                                 int Y1=1<<18;
763                                 int Y2=1<<18;
764
765                                 for(j=0; j<lumFilterSize; j++)
766                                 {
767                                         Y1 += lumSrc[j][i] * lumFilter[j];
768                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
769                                 }
770                                 Y1>>=19;
771                                 Y2>>=19;
772                                 if((Y1|Y2)&256)
773                                 {
774                                         if(Y1>255)   Y1=255;
775                                         else if(Y1<0)Y1=0;
776                                         if(Y2>255)   Y2=255;
777                                         else if(Y2<0)Y2=0;
778                                 }
779                                 acc+= acc + g[Y1+d128[(i+0)&7]];
780                                 acc+= acc + g[Y2+d128[(i+1)&7]];
781                                 if((i&7)==6){
782                                         ((uint8_t*)dest)[0]= acc;
783                                         dest++;
784                                 }
785                         }
786                 }
787                 break;
788         case PIX_FMT_YUYV422:
789                 YSCALE_YUV_2_PACKEDX_C(void)
790                         ((uint8_t*)dest)[2*i2+0]= Y1;
791                         ((uint8_t*)dest)[2*i2+1]= U;
792                         ((uint8_t*)dest)[2*i2+2]= Y2;
793                         ((uint8_t*)dest)[2*i2+3]= V;
794                 }
795                 break;
796         case PIX_FMT_UYVY422:
797                 YSCALE_YUV_2_PACKEDX_C(void)
798                         ((uint8_t*)dest)[2*i2+0]= U;
799                         ((uint8_t*)dest)[2*i2+1]= Y1;
800                         ((uint8_t*)dest)[2*i2+2]= V;
801                         ((uint8_t*)dest)[2*i2+3]= Y2;
802                 }
803                 break;
804         }
805 }
806
807
808 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
809 //Plain C versions
810 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
811 #define COMPILE_C
812 #endif
813
814 #ifdef ARCH_POWERPC
815 #if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
816 #define COMPILE_ALTIVEC
817 #endif //HAVE_ALTIVEC
818 #endif //ARCH_POWERPC
819
820 #if defined(ARCH_X86)
821
822 #if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
823 #define COMPILE_MMX
824 #endif
825
826 #if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
827 #define COMPILE_MMX2
828 #endif
829
830 #if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
831 #define COMPILE_3DNOW
832 #endif
833 #endif //ARCH_X86 || ARCH_X86_64
834
835 #undef HAVE_MMX
836 #undef HAVE_MMX2
837 #undef HAVE_3DNOW
838
839 #ifdef COMPILE_C
840 #undef HAVE_MMX
841 #undef HAVE_MMX2
842 #undef HAVE_3DNOW
843 #undef HAVE_ALTIVEC
844 #define RENAME(a) a ## _C
845 #include "swscale_template.c"
846 #endif
847
848 #ifdef ARCH_POWERPC
849 #ifdef COMPILE_ALTIVEC
850 #undef RENAME
851 #define HAVE_ALTIVEC
852 #define RENAME(a) a ## _altivec
853 #include "swscale_template.c"
854 #endif
855 #endif //ARCH_POWERPC
856
857 #if defined(ARCH_X86)
858
859 //X86 versions
860 /*
861 #undef RENAME
862 #undef HAVE_MMX
863 #undef HAVE_MMX2
864 #undef HAVE_3DNOW
865 #define ARCH_X86
866 #define RENAME(a) a ## _X86
867 #include "swscale_template.c"
868 */
869 //MMX versions
870 #ifdef COMPILE_MMX
871 #undef RENAME
872 #define HAVE_MMX
873 #undef HAVE_MMX2
874 #undef HAVE_3DNOW
875 #define RENAME(a) a ## _MMX
876 #include "swscale_template.c"
877 #endif
878
879 //MMX2 versions
880 #ifdef COMPILE_MMX2
881 #undef RENAME
882 #define HAVE_MMX
883 #define HAVE_MMX2
884 #undef HAVE_3DNOW
885 #define RENAME(a) a ## _MMX2
886 #include "swscale_template.c"
887 #endif
888
889 //3DNOW versions
890 #ifdef COMPILE_3DNOW
891 #undef RENAME
892 #define HAVE_MMX
893 #undef HAVE_MMX2
894 #define HAVE_3DNOW
895 #define RENAME(a) a ## _3DNow
896 #include "swscale_template.c"
897 #endif
898
899 #endif //ARCH_X86 || ARCH_X86_64
900
901 // minor note: the HAVE_xyz is messed up after that line so don't use it
902
903 static double getSplineCoeff(double a, double b, double c, double d, double dist)
904 {
905 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
906         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
907         else            return getSplineCoeff(  0.0, 
908                                                  b+ 2.0*c + 3.0*d,
909                                                         c + 3.0*d,
910                                                 -b- 3.0*c - 6.0*d,
911                                                 dist-1.0);
912 }
913
914 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
915                               int srcW, int dstW, int filterAlign, int one, int flags,
916                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
917 {
918         int i;
919         int filterSize;
920         int filter2Size;
921         int minFilterSize;
922         double *filter=NULL;
923         double *filter2=NULL;
924 #if defined(ARCH_X86)
925         if(flags & SWS_CPU_CAPS_MMX)
926                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
927 #endif
928
929         // Note the +1 is for the MMXscaler which reads over the end
930         *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
931
932         if(FFABS(xInc - 0x10000) <10) // unscaled
933         {
934                 int i;
935                 filterSize= 1;
936                 filter= av_malloc(dstW*sizeof(double)*filterSize);
937                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
938
939                 for(i=0; i<dstW; i++)
940                 {
941                         filter[i*filterSize]=1;
942                         (*filterPos)[i]=i;
943                 }
944
945         }
946         else if(flags&SWS_POINT) // lame looking point sampling mode
947         {
948                 int i;
949                 int xDstInSrc;
950                 filterSize= 1;
951                 filter= av_malloc(dstW*sizeof(double)*filterSize);
952                 
953                 xDstInSrc= xInc/2 - 0x8000;
954                 for(i=0; i<dstW; i++)
955                 {
956                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
957
958                         (*filterPos)[i]= xx;
959                         filter[i]= 1.0;
960                         xDstInSrc+= xInc;
961                 }
962         }
963         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
964         {
965                 int i;
966                 int xDstInSrc;
967                 if     (flags&SWS_BICUBIC) filterSize= 4;
968                 else if(flags&SWS_X      ) filterSize= 4;
969                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
970                 filter= av_malloc(dstW*sizeof(double)*filterSize);
971
972                 xDstInSrc= xInc/2 - 0x8000;
973                 for(i=0; i<dstW; i++)
974                 {
975                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
976                         int j;
977
978                         (*filterPos)[i]= xx;
979                                 //Bilinear upscale / linear interpolate / Area averaging
980                                 for(j=0; j<filterSize; j++)
981                                 {
982                                         double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
983                                         double coeff= 1.0 - d;
984                                         if(coeff<0) coeff=0;
985                                         filter[i*filterSize + j]= coeff;
986                                         xx++;
987                                 }
988                         xDstInSrc+= xInc;
989                 }
990         }
991         else
992         {
993                 double xDstInSrc;
994                 double sizeFactor, filterSizeInSrc;
995                 const double xInc1= (double)xInc / (double)(1<<16);
996
997                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
998                 else if(flags&SWS_X)            sizeFactor= 8.0;
999                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
1000                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
1001                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1002                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
1003                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
1004                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
1005                 else {
1006                         sizeFactor= 0.0; //GCC warning killer
1007                         ASSERT(0)
1008                 }
1009                 
1010                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
1011                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1012
1013                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1014                 if(filterSize > srcW-2) filterSize=srcW-2;
1015
1016                 filter= av_malloc(dstW*sizeof(double)*filterSize);
1017
1018                 xDstInSrc= xInc1 / 2.0 - 0.5;
1019                 for(i=0; i<dstW; i++)
1020                 {
1021                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1022                         int j;
1023                         (*filterPos)[i]= xx;
1024                         for(j=0; j<filterSize; j++)
1025                         {
1026                                 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1027                                 double coeff;
1028                                 if(flags & SWS_BICUBIC)
1029                                 {
1030                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1031                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1032
1033                                         if(d<1.0) 
1034                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1035                                         else if(d<2.0)
1036                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1037                                         else
1038                                                 coeff=0.0;
1039                                 }
1040 /*                              else if(flags & SWS_X)
1041                                 {
1042                                         double p= param ? param*0.01 : 0.3;
1043                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1044                                         coeff*= pow(2.0, - p*d*d);
1045                                 }*/
1046                                 else if(flags & SWS_X)
1047                                 {
1048                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1049                                         
1050                                         if(d<1.0)
1051                                                 coeff = cos(d*PI);
1052                                         else
1053                                                 coeff=-1.0;
1054                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
1055                                         else            coeff=  pow( coeff, A);
1056                                         coeff= coeff*0.5 + 0.5;
1057                                 }
1058                                 else if(flags & SWS_AREA)
1059                                 {
1060                                         double srcPixelSize= 1.0/xInc1;
1061                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1062                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1063                                         else coeff=0.0;
1064                                 }
1065                                 else if(flags & SWS_GAUSS)
1066                                 {
1067                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1068                                         coeff = pow(2.0, - p*d*d);
1069                                 }
1070                                 else if(flags & SWS_SINC)
1071                                 {
1072                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1073                                 }
1074                                 else if(flags & SWS_LANCZOS)
1075                                 {
1076                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
1077                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1078                                         if(d>p) coeff=0;
1079                                 }
1080                                 else if(flags & SWS_BILINEAR)
1081                                 {
1082                                         coeff= 1.0 - d;
1083                                         if(coeff<0) coeff=0;
1084                                 }
1085                                 else if(flags & SWS_SPLINE)
1086                                 {
1087                                         double p=-2.196152422706632;
1088                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1089                                 }
1090                                 else {
1091                                         coeff= 0.0; //GCC warning killer
1092                                         ASSERT(0)
1093                                 }
1094
1095                                 filter[i*filterSize + j]= coeff;
1096                                 xx++;
1097                         }
1098                         xDstInSrc+= xInc1;
1099                 }
1100         }
1101
1102         /* apply src & dst Filter to filter -> filter2
1103            av_free(filter);
1104         */
1105         ASSERT(filterSize>0)
1106         filter2Size= filterSize;
1107         if(srcFilter) filter2Size+= srcFilter->length - 1;
1108         if(dstFilter) filter2Size+= dstFilter->length - 1;
1109         ASSERT(filter2Size>0)
1110         filter2= av_malloc(filter2Size*dstW*sizeof(double));
1111
1112         for(i=0; i<dstW; i++)
1113         {
1114                 int j;
1115                 SwsVector scaleFilter;
1116                 SwsVector *outVec;
1117
1118                 scaleFilter.coeff= filter + i*filterSize;
1119                 scaleFilter.length= filterSize;
1120
1121                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1122                 else          outVec= &scaleFilter;
1123
1124                 ASSERT(outVec->length == filter2Size)
1125                 //FIXME dstFilter
1126
1127                 for(j=0; j<outVec->length; j++)
1128                 {
1129                         filter2[i*filter2Size + j]= outVec->coeff[j];
1130                 }
1131
1132                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1133
1134                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1135         }
1136         av_free(filter); filter=NULL;
1137
1138         /* try to reduce the filter-size (step1 find size and shift left) */
1139         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1140         minFilterSize= 0;
1141         for(i=dstW-1; i>=0; i--)
1142         {
1143                 int min= filter2Size;
1144                 int j;
1145                 double cutOff=0.0;
1146
1147                 /* get rid off near zero elements on the left by shifting left */
1148                 for(j=0; j<filter2Size; j++)
1149                 {
1150                         int k;
1151                         cutOff += FFABS(filter2[i*filter2Size]);
1152
1153                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1154
1155                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1156                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1157
1158                         // Move filter coeffs left
1159                         for(k=1; k<filter2Size; k++)
1160                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1161                         filter2[i*filter2Size + k - 1]= 0.0;
1162                         (*filterPos)[i]++;
1163                 }
1164
1165                 cutOff=0.0;
1166                 /* count near zeros on the right */
1167                 for(j=filter2Size-1; j>0; j--)
1168                 {
1169                         cutOff += FFABS(filter2[i*filter2Size + j]);
1170
1171                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1172                         min--;
1173                 }
1174
1175                 if(min>minFilterSize) minFilterSize= min;
1176         }
1177
1178         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1179           // we can handle the special case 4,
1180           // so we don't want to go to the full 8
1181           if (minFilterSize < 5)
1182             filterAlign = 4;
1183
1184           // we really don't want to waste our time
1185           // doing useless computation, so fall-back on
1186           // the scalar C code for very small filter.
1187           // vectorizing is worth it only if you have
1188           // decent-sized vector.
1189           if (minFilterSize < 3)
1190             filterAlign = 1;
1191         }
1192
1193         if (flags & SWS_CPU_CAPS_MMX) {
1194                 // special case for unscaled vertical filtering
1195                 if(minFilterSize == 1 && filterAlign == 2)
1196                         filterAlign= 1;
1197         }
1198
1199         ASSERT(minFilterSize > 0)
1200         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1201         ASSERT(filterSize > 0)
1202         filter= av_malloc(filterSize*dstW*sizeof(double));
1203         if(filterSize >= MAX_FILTER_SIZE)
1204                 return -1;
1205         *outFilterSize= filterSize;
1206
1207         if(flags&SWS_PRINT_INFO)
1208                 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1209         /* try to reduce the filter-size (step2 reduce it) */
1210         for(i=0; i<dstW; i++)
1211         {
1212                 int j;
1213
1214                 for(j=0; j<filterSize; j++)
1215                 {
1216                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1217                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1218                 }
1219         }
1220         av_free(filter2); filter2=NULL;
1221         
1222
1223         //FIXME try to align filterpos if possible
1224
1225         //fix borders
1226         for(i=0; i<dstW; i++)
1227         {
1228                 int j;
1229                 if((*filterPos)[i] < 0)
1230                 {
1231                         // Move filter coeffs left to compensate for filterPos
1232                         for(j=1; j<filterSize; j++)
1233                         {
1234                                 int left= FFMAX(j + (*filterPos)[i], 0);
1235                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1236                                 filter[i*filterSize + j]=0;
1237                         }
1238                         (*filterPos)[i]= 0;
1239                 }
1240
1241                 if((*filterPos)[i] + filterSize > srcW)
1242                 {
1243                         int shift= (*filterPos)[i] + filterSize - srcW;
1244                         // Move filter coeffs right to compensate for filterPos
1245                         for(j=filterSize-2; j>=0; j--)
1246                         {
1247                                 int right= FFMIN(j + shift, filterSize-1);
1248                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1249                                 filter[i*filterSize +j]=0;
1250                         }
1251                         (*filterPos)[i]= srcW - filterSize;
1252                 }
1253         }
1254
1255         // Note the +1 is for the MMXscaler which reads over the end
1256         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1257         *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1258
1259         /* Normalize & Store in outFilter */
1260         for(i=0; i<dstW; i++)
1261         {
1262                 int j;
1263                 double error=0;
1264                 double sum=0;
1265                 double scale= one;
1266
1267                 for(j=0; j<filterSize; j++)
1268                 {
1269                         sum+= filter[i*filterSize + j];
1270                 }
1271                 scale/= sum;
1272                 for(j=0; j<*outFilterSize; j++)
1273                 {
1274                         double v= filter[i*filterSize + j]*scale + error;
1275                         int intV= floor(v + 0.5);
1276                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1277                         error = v - intV;
1278                 }
1279         }
1280         
1281         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1282         for(i=0; i<*outFilterSize; i++)
1283         {
1284                 int j= dstW*(*outFilterSize);
1285                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1286         }
1287
1288         av_free(filter);
1289         return 0;
1290 }
1291
1292 #ifdef COMPILE_MMX2
1293 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1294 {
1295         uint8_t *fragmentA;
1296         long imm8OfPShufW1A;
1297         long imm8OfPShufW2A;
1298         long fragmentLengthA;
1299         uint8_t *fragmentB;
1300         long imm8OfPShufW1B;
1301         long imm8OfPShufW2B;
1302         long fragmentLengthB;
1303         int fragmentPos;
1304
1305         int xpos, i;
1306
1307         // create an optimized horizontal scaling routine
1308
1309         //code fragment
1310
1311         asm volatile(
1312                 "jmp 9f                         \n\t"
1313         // Begin
1314                 "0:                             \n\t"
1315                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1316                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1317                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1318                 "punpcklbw %%mm7, %%mm1         \n\t"
1319                 "punpcklbw %%mm7, %%mm0         \n\t"
1320                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1321                 "1:                             \n\t"
1322                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1323                 "2:                             \n\t"
1324                 "psubw %%mm1, %%mm0             \n\t"
1325                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1326                 "pmullw %%mm3, %%mm0            \n\t"
1327                 "psllw $7, %%mm1                \n\t"
1328                 "paddw %%mm1, %%mm0             \n\t"
1329
1330                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1331
1332                 "add $8, %%"REG_a"              \n\t"
1333         // End
1334                 "9:                             \n\t"
1335 //              "int $3\n\t"
1336                 "lea 0b, %0                     \n\t"
1337                 "lea 1b, %1                     \n\t"
1338                 "lea 2b, %2                     \n\t"
1339                 "dec %1                         \n\t"
1340                 "dec %2                         \n\t"
1341                 "sub %0, %1                     \n\t"
1342                 "sub %0, %2                     \n\t"
1343                 "lea 9b, %3                     \n\t"
1344                 "sub %0, %3                     \n\t"
1345
1346
1347                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1348                 "=r" (fragmentLengthA)
1349         );
1350
1351         asm volatile(
1352                 "jmp 9f                         \n\t"
1353         // Begin
1354                 "0:                             \n\t"
1355                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1356                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1357                 "punpcklbw %%mm7, %%mm0         \n\t"
1358                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1359                 "1:                             \n\t"
1360                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1361                 "2:                             \n\t"
1362                 "psubw %%mm1, %%mm0             \n\t"
1363                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1364                 "pmullw %%mm3, %%mm0            \n\t"
1365                 "psllw $7, %%mm1                \n\t"
1366                 "paddw %%mm1, %%mm0             \n\t"
1367
1368                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1369
1370                 "add $8, %%"REG_a"              \n\t"
1371         // End
1372                 "9:                             \n\t"
1373 //              "int $3\n\t"
1374                 "lea 0b, %0                     \n\t"
1375                 "lea 1b, %1                     \n\t"
1376                 "lea 2b, %2                     \n\t"
1377                 "dec %1                         \n\t"
1378                 "dec %2                         \n\t"
1379                 "sub %0, %1                     \n\t"
1380                 "sub %0, %2                     \n\t"
1381                 "lea 9b, %3                     \n\t"
1382                 "sub %0, %3                     \n\t"
1383
1384
1385                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1386                 "=r" (fragmentLengthB)
1387         );
1388
1389         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1390         fragmentPos=0;
1391         
1392         for(i=0; i<dstW/numSplits; i++)
1393         {
1394                 int xx=xpos>>16;
1395
1396                 if((i&3) == 0)
1397                 {
1398                         int a=0;
1399                         int b=((xpos+xInc)>>16) - xx;
1400                         int c=((xpos+xInc*2)>>16) - xx;
1401                         int d=((xpos+xInc*3)>>16) - xx;
1402
1403                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1404                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1405                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1406                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1407                         filterPos[i/2]= xx;
1408
1409                         if(d+1<4)
1410                         {
1411                                 int maxShift= 3-(d+1);
1412                                 int shift=0;
1413
1414                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1415
1416                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1417                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1418                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1419                                         a | (b<<2) | (c<<4) | (d<<6);
1420
1421                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1422                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1423
1424                                 if(shift && i>=shift)
1425                                 {
1426                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1427                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1428                                         filterPos[i/2]-=shift;
1429                                 }
1430
1431                                 fragmentPos+= fragmentLengthB;
1432                         }
1433                         else
1434                         {
1435                                 int maxShift= 3-d;
1436                                 int shift=0;
1437
1438                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1439
1440                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1441                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1442                                         a | (b<<2) | (c<<4) | (d<<6);
1443
1444                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1445                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1446
1447                                 if(shift && i>=shift)
1448                                 {
1449                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1450                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1451                                         filterPos[i/2]-=shift;
1452                                 }
1453
1454                                 fragmentPos+= fragmentLengthA;
1455                         }
1456
1457                         funnyCode[fragmentPos]= RET;
1458                 }
1459                 xpos+=xInc;
1460         }
1461         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1462 }
1463 #endif /* COMPILE_MMX2 */
1464
1465 static void globalInit(void){
1466     // generating tables:
1467     int i;
1468     for(i=0; i<768; i++){
1469         int c= av_clip_uint8(i-256);
1470         clip_table[i]=c;
1471     }
1472 }
1473
1474 static SwsFunc getSwsFunc(int flags){
1475     
1476 #if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1477 #if defined(ARCH_X86)
1478         // ordered per speed fasterst first
1479         if(flags & SWS_CPU_CAPS_MMX2)
1480                 return swScale_MMX2;
1481         else if(flags & SWS_CPU_CAPS_3DNOW)
1482                 return swScale_3DNow;
1483         else if(flags & SWS_CPU_CAPS_MMX)
1484                 return swScale_MMX;
1485         else
1486                 return swScale_C;
1487
1488 #else
1489 #ifdef ARCH_POWERPC
1490         if(flags & SWS_CPU_CAPS_ALTIVEC)
1491           return swScale_altivec;
1492         else
1493           return swScale_C;
1494 #endif
1495         return swScale_C;
1496 #endif /* defined(ARCH_X86) */
1497 #else //RUNTIME_CPUDETECT
1498 #ifdef HAVE_MMX2
1499         return swScale_MMX2;
1500 #elif defined (HAVE_3DNOW)
1501         return swScale_3DNow;
1502 #elif defined (HAVE_MMX)
1503         return swScale_MMX;
1504 #elif defined (HAVE_ALTIVEC)
1505         return swScale_altivec;
1506 #else
1507         return swScale_C;
1508 #endif
1509 #endif //!RUNTIME_CPUDETECT
1510 }
1511
1512 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1513              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1514         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1515         /* Copy Y plane */
1516         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1517                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1518         else
1519         {
1520                 int i;
1521                 uint8_t *srcPtr= src[0];
1522                 uint8_t *dstPtr= dst;
1523                 for(i=0; i<srcSliceH; i++)
1524                 {
1525                         memcpy(dstPtr, srcPtr, c->srcW);
1526                         srcPtr+= srcStride[0];
1527                         dstPtr+= dstStride[0];
1528                 }
1529         }
1530         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1531         if (c->dstFormat == PIX_FMT_NV12)
1532                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1533         else
1534                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1535
1536         return srcSliceH;
1537 }
1538
1539 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1540              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1541         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1542
1543         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1544
1545         return srcSliceH;
1546 }
1547
1548 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1549              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1550         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1551
1552         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1553
1554         return srcSliceH;
1555 }
1556
1557 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1558 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1559                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1560         const int srcFormat= c->srcFormat;
1561         const int dstFormat= c->dstFormat;
1562         const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1563         const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1564         const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1565         const int dstId= fmt_depth(dstFormat) >> 2;
1566         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1567
1568         /* BGR -> BGR */
1569         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1570            || (isRGB(srcFormat) && isRGB(dstFormat))){
1571                 switch(srcId | (dstId<<4)){
1572                 case 0x34: conv= rgb16to15; break;
1573                 case 0x36: conv= rgb24to15; break;
1574                 case 0x38: conv= rgb32to15; break;
1575                 case 0x43: conv= rgb15to16; break;
1576                 case 0x46: conv= rgb24to16; break;
1577                 case 0x48: conv= rgb32to16; break;
1578                 case 0x63: conv= rgb15to24; break;
1579                 case 0x64: conv= rgb16to24; break;
1580                 case 0x68: conv= rgb32to24; break;
1581                 case 0x83: conv= rgb15to32; break;
1582                 case 0x84: conv= rgb16to32; break;
1583                 case 0x86: conv= rgb24to32; break;
1584                 default: av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1585                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1586                 }
1587         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1588                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1589                 switch(srcId | (dstId<<4)){
1590                 case 0x33: conv= rgb15tobgr15; break;
1591                 case 0x34: conv= rgb16tobgr15; break;
1592                 case 0x36: conv= rgb24tobgr15; break;
1593                 case 0x38: conv= rgb32tobgr15; break;
1594                 case 0x43: conv= rgb15tobgr16; break;
1595                 case 0x44: conv= rgb16tobgr16; break;
1596                 case 0x46: conv= rgb24tobgr16; break;
1597                 case 0x48: conv= rgb32tobgr16; break;
1598                 case 0x63: conv= rgb15tobgr24; break;
1599                 case 0x64: conv= rgb16tobgr24; break;
1600                 case 0x66: conv= rgb24tobgr24; break;
1601                 case 0x68: conv= rgb32tobgr24; break;
1602                 case 0x83: conv= rgb15tobgr32; break;
1603                 case 0x84: conv= rgb16tobgr32; break;
1604                 case 0x86: conv= rgb24tobgr32; break;
1605                 case 0x88: conv= rgb32tobgr32; break;
1606                 default: av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1607                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1608                 }
1609         }else{
1610                 av_log(c, AV_LOG_ERROR, "swScaler: internal error %s -> %s converter\n", 
1611                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1612         }
1613
1614         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1615                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1616         else
1617         {
1618                 int i;
1619                 uint8_t *srcPtr= src[0];
1620                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1621
1622                 for(i=0; i<srcSliceH; i++)
1623                 {
1624                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1625                         srcPtr+= srcStride[0];
1626                         dstPtr+= dstStride[0];
1627                 }
1628         }     
1629         return srcSliceH;
1630 }
1631
1632 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1633              int srcSliceH, uint8_t* dst[], int dstStride[]){
1634
1635         rgb24toyv12(
1636                 src[0], 
1637                 dst[0]+ srcSliceY    *dstStride[0], 
1638                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1639                 dst[2]+(srcSliceY>>1)*dstStride[2],
1640                 c->srcW, srcSliceH, 
1641                 dstStride[0], dstStride[1], srcStride[0]);
1642         return srcSliceH;
1643 }
1644
1645 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1646              int srcSliceH, uint8_t* dst[], int dstStride[]){
1647         int i;
1648
1649         /* copy Y */
1650         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1651                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1652         else{
1653                 uint8_t *srcPtr= src[0];
1654                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1655
1656                 for(i=0; i<srcSliceH; i++)
1657                 {
1658                         memcpy(dstPtr, srcPtr, c->srcW);
1659                         srcPtr+= srcStride[0];
1660                         dstPtr+= dstStride[0];
1661                 }
1662         }
1663
1664         if(c->dstFormat==PIX_FMT_YUV420P){
1665                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1666                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1667         }else{
1668                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1669                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1670         }
1671         return srcSliceH;
1672 }
1673
1674 /* unscaled copy like stuff (assumes nearly identical formats) */
1675 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1676              int srcSliceH, uint8_t* dst[], int dstStride[]){
1677
1678         if(isPacked(c->srcFormat))
1679         {
1680                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1681                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1682                 else
1683                 {
1684                         int i;
1685                         uint8_t *srcPtr= src[0];
1686                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1687                         int length=0;
1688
1689                         /* universal length finder */
1690                         while(length+c->srcW <= FFABS(dstStride[0]) 
1691                            && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1692                         ASSERT(length!=0);
1693
1694                         for(i=0; i<srcSliceH; i++)
1695                         {
1696                                 memcpy(dstPtr, srcPtr, length);
1697                                 srcPtr+= srcStride[0];
1698                                 dstPtr+= dstStride[0];
1699                         }
1700                 }
1701         }
1702         else 
1703         { /* Planar YUV or gray */
1704                 int plane;
1705                 for(plane=0; plane<3; plane++)
1706                 {
1707                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1708                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1709                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1710
1711                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1712                         {
1713                                 if(!isGray(c->dstFormat))
1714                                         memset(dst[plane], 128, dstStride[plane]*height);
1715                         }
1716                         else
1717                         {
1718                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1719                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1720                                 else
1721                                 {
1722                                         int i;
1723                                         uint8_t *srcPtr= src[plane];
1724                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1725                                         for(i=0; i<height; i++)
1726                                         {
1727                                                 memcpy(dstPtr, srcPtr, length);
1728                                                 srcPtr+= srcStride[plane];
1729                                                 dstPtr+= dstStride[plane];
1730                                         }
1731                                 }
1732                         }
1733                 }
1734         }
1735         return srcSliceH;
1736 }
1737
1738 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1739              int srcSliceH, uint8_t* dst[], int dstStride[]){
1740
1741         int length= c->srcW;
1742         int y=      srcSliceY;
1743         int height= srcSliceH;
1744         int i, j;
1745         uint8_t *srcPtr= src[0];
1746         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1747
1748         if(!isGray(c->dstFormat)){
1749                 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1750                 memset(dst[1], 128, dstStride[1]*height);
1751                 memset(dst[2], 128, dstStride[2]*height);
1752         }
1753         if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1754         for(i=0; i<height; i++)
1755         {
1756                 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1757                 srcPtr+= srcStride[0];
1758                 dstPtr+= dstStride[0];
1759         }
1760         return srcSliceH;
1761 }
1762
1763 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1764              int srcSliceH, uint8_t* dst[], int dstStride[]){
1765
1766         int length= c->srcW;
1767         int y=      srcSliceY;
1768         int height= srcSliceH;
1769         int i, j;
1770         uint8_t *srcPtr= src[0];
1771         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1772         for(i=0; i<height; i++)
1773         {
1774                 for(j=0; j<length; j++)
1775                 {
1776                         dstPtr[j<<1] = srcPtr[j];
1777                         dstPtr[(j<<1)+1] = srcPtr[j];
1778                 }
1779                 srcPtr+= srcStride[0];
1780                 dstPtr+= dstStride[0];
1781         }
1782         return srcSliceH;
1783 }
1784
1785 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1786              int srcSliceH, uint8_t* dst[], int dstStride[]){
1787
1788         int length= c->srcW;
1789         int y=      srcSliceY;
1790         int height= srcSliceH;
1791         int i, j;
1792         uint16_t *srcPtr= src[0];
1793         uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1794         for(i=0; i<height; i++)
1795         {
1796                 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1797                 srcPtr+= srcStride[0]/2;
1798                 dstPtr+= dstStride[0]/2;
1799         }
1800         return srcSliceH;
1801 }
1802
1803
1804 static void getSubSampleFactors(int *h, int *v, int format){
1805         switch(format){
1806         case PIX_FMT_UYVY422:
1807         case PIX_FMT_YUYV422:
1808                 *h=1;
1809                 *v=0;
1810                 break;
1811         case PIX_FMT_YUV420P:
1812         case PIX_FMT_GRAY16BE:
1813         case PIX_FMT_GRAY16LE:
1814         case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1815         case PIX_FMT_NV12:
1816         case PIX_FMT_NV21:
1817                 *h=1;
1818                 *v=1;
1819                 break;
1820         case PIX_FMT_YUV410P:
1821                 *h=2;
1822                 *v=2;
1823                 break;
1824         case PIX_FMT_YUV444P:
1825                 *h=0;
1826                 *v=0;
1827                 break;
1828         case PIX_FMT_YUV422P:
1829                 *h=1;
1830                 *v=0;
1831                 break;
1832         case PIX_FMT_YUV411P:
1833                 *h=2;
1834                 *v=0;
1835                 break;
1836         default:
1837                 *h=0;
1838                 *v=0;
1839                 break;
1840         }
1841 }
1842
1843 static uint16_t roundToInt16(int64_t f){
1844         int r= (f + (1<<15))>>16;
1845              if(r<-0x7FFF) return 0x8000;
1846         else if(r> 0x7FFF) return 0x7FFF;
1847         else               return r;
1848 }
1849
1850 /**
1851  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1852  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1853  * @return -1 if not supported
1854  */
1855 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1856         int64_t crv =  inv_table[0];
1857         int64_t cbu =  inv_table[1];
1858         int64_t cgu = -inv_table[2];
1859         int64_t cgv = -inv_table[3];
1860         int64_t cy  = 1<<16;
1861         int64_t oy  = 0;
1862
1863         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1864         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1865         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1866
1867         c->brightness= brightness;
1868         c->contrast  = contrast;
1869         c->saturation= saturation;
1870         c->srcRange  = srcRange;
1871         c->dstRange  = dstRange;
1872
1873         c->uOffset=   0x0400040004000400LL;
1874         c->vOffset=   0x0400040004000400LL;
1875
1876         if(!srcRange){
1877                 cy= (cy*255) / 219;
1878                 oy= 16<<16;
1879         }else{
1880                 crv= (crv*224) / 255;
1881                 cbu= (cbu*224) / 255;
1882                 cgu= (cgu*224) / 255;
1883                 cgv= (cgv*224) / 255;
1884         }
1885
1886         cy = (cy *contrast             )>>16;
1887         crv= (crv*contrast * saturation)>>32;
1888         cbu= (cbu*contrast * saturation)>>32;
1889         cgu= (cgu*contrast * saturation)>>32;
1890         cgv= (cgv*contrast * saturation)>>32;
1891
1892         oy -= 256*brightness;
1893
1894         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1895         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1896         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1897         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1898         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1899         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1900
1901         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1902         //FIXME factorize
1903
1904 #ifdef COMPILE_ALTIVEC
1905         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1906             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1907 #endif  
1908         return 0;
1909 }
1910
1911 /**
1912  * @return -1 if not supported
1913  */
1914 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1915         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1916
1917         *inv_table = c->srcColorspaceTable;
1918         *table     = c->dstColorspaceTable;
1919         *srcRange  = c->srcRange;
1920         *dstRange  = c->dstRange;
1921         *brightness= c->brightness;
1922         *contrast  = c->contrast;
1923         *saturation= c->saturation;
1924         
1925         return 0;       
1926 }
1927
1928 static int handle_jpeg(int *format)
1929 {
1930         switch (*format) {
1931                 case PIX_FMT_YUVJ420P:
1932                         *format = PIX_FMT_YUV420P;
1933                         return 1;
1934                 case PIX_FMT_YUVJ422P:
1935                         *format = PIX_FMT_YUV422P;
1936                         return 1;
1937                 case PIX_FMT_YUVJ444P:
1938                         *format = PIX_FMT_YUV444P;
1939                         return 1;
1940                 default:
1941                         return 0;
1942         }
1943 }
1944
1945 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1946                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1947
1948         SwsContext *c;
1949         int i;
1950         int usesVFilter, usesHFilter;
1951         int unscaled, needsDither;
1952         int srcRange, dstRange;
1953         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1954 #if defined(ARCH_X86)
1955         if(flags & SWS_CPU_CAPS_MMX)
1956                 asm volatile("emms\n\t"::: "memory");
1957 #endif
1958
1959 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
1960         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1961 #ifdef HAVE_MMX2
1962         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1963 #elif defined (HAVE_3DNOW)
1964         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1965 #elif defined (HAVE_MMX)
1966         flags |= SWS_CPU_CAPS_MMX;
1967 #elif defined (HAVE_ALTIVEC)
1968         flags |= SWS_CPU_CAPS_ALTIVEC;
1969 #endif
1970 #endif /* RUNTIME_CPUDETECT */
1971         if(clip_table[512] != 255) globalInit();
1972         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1973
1974         unscaled = (srcW == dstW && srcH == dstH);
1975         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1976                      && (fmt_depth(dstFormat))<24
1977                      && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1978
1979         srcRange = handle_jpeg(&srcFormat);
1980         dstRange = handle_jpeg(&dstFormat);
1981
1982         if(!isSupportedIn(srcFormat)) 
1983         {
1984                 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1985                 return NULL;
1986         }
1987         if(!isSupportedOut(dstFormat))
1988         {
1989                 av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1990                 return NULL;
1991         }
1992
1993         /* sanity check */
1994         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
1995         {
1996                  av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1997                         srcW, srcH, dstW, dstH);
1998                 return NULL;
1999         }
2000
2001         if(!dstFilter) dstFilter= &dummyFilter;
2002         if(!srcFilter) srcFilter= &dummyFilter;
2003
2004         c= av_mallocz(sizeof(SwsContext));
2005
2006         c->av_class = &sws_context_class;
2007         c->srcW= srcW;
2008         c->srcH= srcH;
2009         c->dstW= dstW;
2010         c->dstH= dstH;
2011         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2012         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2013         c->flags= flags;
2014         c->dstFormat= dstFormat;
2015         c->srcFormat= srcFormat;
2016         c->vRounder= 4* 0x0001000100010001ULL;
2017
2018         usesHFilter= usesVFilter= 0;
2019         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2020         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2021         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2022         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2023         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2024         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2025         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2026         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2027
2028         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2029         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2030
2031         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2032         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2033
2034         // drop some chroma lines if the user wants it
2035         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2036         c->chrSrcVSubSample+= c->vChrDrop;
2037
2038         // drop every 2. pixel for chroma calculation unless user wants full chroma
2039         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2040            && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
2041            && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
2042            && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2043                 c->chrSrcHSubSample=1;
2044
2045         if(param){
2046                 c->param[0] = param[0];
2047                 c->param[1] = param[1];
2048         }else{
2049                 c->param[0] =
2050                 c->param[1] = SWS_PARAM_DEFAULT;
2051         }
2052
2053         c->chrIntHSubSample= c->chrDstHSubSample;
2054         c->chrIntVSubSample= c->chrSrcVSubSample;
2055
2056         // Note the -((-x)>>y) is so that we always round toward +inf.
2057         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2058         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2059         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2060         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2061
2062         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); 
2063
2064         /* unscaled special Cases */
2065         if(unscaled && !usesHFilter && !usesVFilter)
2066         {
2067                 /* yv12_to_nv12 */
2068                 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2069                 {
2070                         c->swScale= PlanarToNV12Wrapper;
2071                 }
2072 #ifdef CONFIG_GPL
2073                 /* yuv2bgr */
2074                 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2075                 {
2076                         c->swScale= yuv2rgb_get_func_ptr(c);
2077                 }
2078 #endif
2079                 
2080                 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2081                 {
2082                         c->swScale= yvu9toyv12Wrapper;
2083                 }
2084
2085                 /* bgr24toYV12 */
2086                 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2087                         c->swScale= bgr24toyv12Wrapper;
2088                 
2089                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2090                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
2091                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
2092                    && srcFormat != PIX_FMT_BGR8 && dstFormat != PIX_FMT_BGR8
2093                    && srcFormat != PIX_FMT_RGB8 && dstFormat != PIX_FMT_RGB8
2094                    && srcFormat != PIX_FMT_BGR4 && dstFormat != PIX_FMT_BGR4
2095                    && srcFormat != PIX_FMT_RGB4 && dstFormat != PIX_FMT_RGB4
2096                    && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2097                    && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2098                    && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2099                    && !needsDither)
2100                         c->swScale= rgb2rgbWrapper;
2101
2102                 /* LQ converters if -sws 0 or -sws 4*/
2103                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2104                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
2105                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
2106                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
2107                           && needsDither)
2108                                 c->swScale= rgb2rgbWrapper;
2109
2110                         /* yv12_to_yuy2 */
2111                         if(srcFormat == PIX_FMT_YUV420P && 
2112                             (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2113                         {
2114                                 if (dstFormat == PIX_FMT_YUYV422)
2115                                     c->swScale= PlanarToYuy2Wrapper;
2116                                 else
2117                                     c->swScale= PlanarToUyvyWrapper;
2118                         }
2119                 }
2120
2121 #ifdef COMPILE_ALTIVEC
2122                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2123                     ((srcFormat == PIX_FMT_YUV420P && 
2124                       (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2125                   // unscaled YV12 -> packed YUV, we want speed
2126                   if (dstFormat == PIX_FMT_YUYV422)
2127                     c->swScale= yv12toyuy2_unscaled_altivec;
2128                   else
2129                     c->swScale= yv12touyvy_unscaled_altivec;
2130                 }
2131 #endif
2132
2133                 /* simple copy */
2134                 if(   srcFormat == dstFormat
2135                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2136                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2137                   )
2138                 {
2139                         c->swScale= simpleCopy;
2140                 }
2141
2142                 /* gray16{le,be} conversions */
2143                 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2144                 {
2145                         c->swScale= gray16togray;
2146                 }
2147                 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2148                 {
2149                         c->swScale= graytogray16;
2150                 }
2151                 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2152                 {
2153                         c->swScale= gray16swap;
2154                 }               
2155
2156                 if(c->swScale){
2157                         if(flags&SWS_PRINT_INFO)
2158                                 av_log(c, AV_LOG_INFO, "SwScaler: using unscaled %s -> %s special converter\n", 
2159                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2160                         return c;
2161                 }
2162         }
2163
2164         if(flags & SWS_CPU_CAPS_MMX2)
2165         {
2166                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2167                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2168                 {
2169                         if(flags&SWS_PRINT_INFO)
2170                                 av_log(c, AV_LOG_INFO, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2171                 }
2172                 if(usesHFilter) c->canMMX2BeUsed=0;
2173         }
2174         else
2175                 c->canMMX2BeUsed=0;
2176
2177         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2178         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2179
2180         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2181         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2182         // n-2 is the last chrominance sample available
2183         // this is not perfect, but noone shuld notice the difference, the more correct variant
2184         // would be like the vertical one, but that would require some special code for the
2185         // first and last pixel
2186         if(flags&SWS_FAST_BILINEAR)
2187         {
2188                 if(c->canMMX2BeUsed)
2189                 {
2190                         c->lumXInc+= 20;
2191                         c->chrXInc+= 20;
2192                 }
2193                 //we don't use the x86asm scaler if mmx is available
2194                 else if(flags & SWS_CPU_CAPS_MMX)
2195                 {
2196                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2197                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2198                 }
2199         }
2200
2201         /* precalculate horizontal scaler filter coefficients */
2202         {
2203                 const int filterAlign=
2204                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2205                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2206                   1;
2207
2208                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2209                                  srcW      ,       dstW, filterAlign, 1<<14,
2210                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2211                                  srcFilter->lumH, dstFilter->lumH, c->param);
2212                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2213                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2214                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2215                                  srcFilter->chrH, dstFilter->chrH, c->param);
2216
2217 #define MAX_FUNNY_CODE_SIZE 10000
2218 #if defined(COMPILE_MMX2)
2219 // can't downscale !!!
2220                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2221                 {
2222 #ifdef MAP_ANONYMOUS
2223                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2224                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2225 #else
2226                         c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2227                         c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2228 #endif
2229
2230                         c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2231                         c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2232                         c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2233                         c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2234
2235                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2236                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2237                 }
2238 #endif /* defined(COMPILE_MMX2) */
2239         } // Init Horizontal stuff
2240
2241
2242
2243         /* precalculate vertical scaler filter coefficients */
2244         {
2245                 const int filterAlign=
2246                   (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2247                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2248                   1;
2249
2250                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2251                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2252                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2253                                 srcFilter->lumV, dstFilter->lumV, c->param);
2254                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2255                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2256                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2257                                 srcFilter->chrV, dstFilter->chrV, c->param);
2258
2259 #ifdef HAVE_ALTIVEC
2260                 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2261                 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2262
2263                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2264                   int j;
2265                   short *p = (short *)&c->vYCoeffsBank[i];
2266                   for (j=0;j<8;j++)
2267                     p[j] = c->vLumFilter[i];
2268                 }
2269
2270                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2271                   int j;
2272                   short *p = (short *)&c->vCCoeffsBank[i];
2273                   for (j=0;j<8;j++)
2274                     p[j] = c->vChrFilter[i];
2275                 }
2276 #endif
2277         }
2278
2279         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2280         c->vLumBufSize= c->vLumFilterSize;
2281         c->vChrBufSize= c->vChrFilterSize;
2282         for(i=0; i<dstH; i++)
2283         {
2284                 int chrI= i*c->chrDstH / dstH;
2285                 int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2286                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2287
2288                 nextSlice>>= c->chrSrcVSubSample;
2289                 nextSlice<<= c->chrSrcVSubSample;
2290                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2291                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2292                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2293                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2294         }
2295
2296         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2297         c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2298         c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2299         //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)
2300         /* align at 16 bytes for AltiVec */
2301         for(i=0; i<c->vLumBufSize; i++)
2302                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(4000);
2303         for(i=0; i<c->vChrBufSize; i++)
2304                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2305
2306         //try to avoid drawing green stuff between the right end and the stride end
2307         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2308
2309         ASSERT(c->chrDstH <= dstH)
2310
2311         if(flags&SWS_PRINT_INFO)
2312         {
2313 #ifdef DITHER1XBPP
2314                 char *dither= " dithered";
2315 #else
2316                 char *dither= "";
2317 #endif
2318                 if(flags&SWS_FAST_BILINEAR)
2319                         av_log(c, AV_LOG_INFO, "SwScaler: FAST_BILINEAR scaler, ");
2320                 else if(flags&SWS_BILINEAR)
2321                         av_log(c, AV_LOG_INFO, "SwScaler: BILINEAR scaler, ");
2322                 else if(flags&SWS_BICUBIC)
2323                         av_log(c, AV_LOG_INFO, "SwScaler: BICUBIC scaler, ");
2324                 else if(flags&SWS_X)
2325                         av_log(c, AV_LOG_INFO, "SwScaler: Experimental scaler, ");
2326                 else if(flags&SWS_POINT)
2327                         av_log(c, AV_LOG_INFO, "SwScaler: Nearest Neighbor / POINT scaler, ");
2328                 else if(flags&SWS_AREA)
2329                         av_log(c, AV_LOG_INFO, "SwScaler: Area Averageing scaler, ");
2330                 else if(flags&SWS_BICUBLIN)
2331                         av_log(c, AV_LOG_INFO, "SwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2332                 else if(flags&SWS_GAUSS)
2333                         av_log(c, AV_LOG_INFO, "SwScaler: Gaussian scaler, ");
2334                 else if(flags&SWS_SINC)
2335                         av_log(c, AV_LOG_INFO, "SwScaler: Sinc scaler, ");
2336                 else if(flags&SWS_LANCZOS)
2337                         av_log(c, AV_LOG_INFO, "SwScaler: Lanczos scaler, ");
2338                 else if(flags&SWS_SPLINE)
2339                         av_log(c, AV_LOG_INFO, "SwScaler: Bicubic spline scaler, ");
2340                 else
2341                         av_log(c, AV_LOG_INFO, "SwScaler: ehh flags invalid?! ");
2342
2343                 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2344                         av_log(c, AV_LOG_INFO, "from %s to%s %s ", 
2345                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2346                 else
2347                         av_log(c, AV_LOG_INFO, "from %s to %s ", 
2348                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2349
2350                 if(flags & SWS_CPU_CAPS_MMX2)
2351                         av_log(c, AV_LOG_INFO, "using MMX2\n");
2352                 else if(flags & SWS_CPU_CAPS_3DNOW)
2353                         av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2354                 else if(flags & SWS_CPU_CAPS_MMX)
2355                         av_log(c, AV_LOG_INFO, "using MMX\n");
2356                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2357                         av_log(c, AV_LOG_INFO, "using AltiVec\n");
2358                 else 
2359                         av_log(c, AV_LOG_INFO, "using C\n");
2360         }
2361
2362         if(flags & SWS_PRINT_INFO)
2363         {
2364                 if(flags & SWS_CPU_CAPS_MMX)
2365                 {
2366                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2367                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2368                         else
2369                         {
2370                                 if(c->hLumFilterSize==4)
2371                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2372                                 else if(c->hLumFilterSize==8)
2373                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2374                                 else
2375                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2376
2377                                 if(c->hChrFilterSize==4)
2378                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2379                                 else if(c->hChrFilterSize==8)
2380                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2381                                 else
2382                                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2383                         }
2384                 }
2385                 else
2386                 {
2387 #if defined(ARCH_X86)
2388                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using X86-Asm scaler for horizontal scaling\n");
2389 #else
2390                         if(flags & SWS_FAST_BILINEAR)
2391                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2392                         else
2393                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using C scaler for horizontal scaling\n");
2394 #endif
2395                 }
2396                 if(isPlanarYUV(dstFormat))
2397                 {
2398                         if(c->vLumFilterSize==1)
2399                                 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");
2400                         else
2401                                 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");
2402                 }
2403                 else
2404                 {
2405                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2406                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2407                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2408                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2409                                 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");
2410                         else
2411                                 av_log(c, AV_LOG_VERBOSE, "SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2412                 }
2413
2414                 if(dstFormat==PIX_FMT_BGR24)
2415                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR24 Converter\n",
2416                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2417                 else if(dstFormat==PIX_FMT_RGB32)
2418                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2419                 else if(dstFormat==PIX_FMT_BGR565)
2420                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2421                 else if(dstFormat==PIX_FMT_BGR555)
2422                         av_log(c, AV_LOG_VERBOSE, "SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2423
2424                 av_log(c, AV_LOG_VERBOSE, "SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2425         }
2426         if(flags & SWS_PRINT_INFO)
2427         {
2428                 av_log(c, AV_LOG_DEBUG, "SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2429                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2430                 av_log(c, AV_LOG_DEBUG, "SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2431                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2432         }
2433
2434         c->swScale= getSwsFunc(flags);
2435         return c;
2436 }
2437
2438 /**
2439  * swscale warper, so we don't need to export the SwsContext.
2440  * assumes planar YUV to be in YUV order instead of YVU
2441  */
2442 int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2443                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2444         int i;
2445         uint8_t* src2[4]= {src[0], src[1], src[2]};
2446         uint32_t pal[256];
2447         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2448             av_log(c, AV_LOG_ERROR, "swScaler: slices start in the middle!\n");
2449             return 0;
2450         }
2451         if (c->sliceDir == 0) {
2452             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2453         }
2454
2455         if(c->srcFormat == PIX_FMT_PAL8){
2456                 for(i=0; i<256; i++){
2457                         int p= ((uint32_t*)(src[1]))[i];
2458                         int r= (p>>16)&0xFF;
2459                         int g= (p>> 8)&0xFF;
2460                         int b=  p     &0xFF;
2461                         int y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2462                         int u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2463                         int v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2464                         pal[i]= y + (u<<8) + (v<<16);
2465                 }
2466                 src2[1]= pal;
2467         }
2468
2469         // copy strides, so they can safely be modified
2470         if (c->sliceDir == 1) {
2471             // slices go from top to bottom
2472             int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2473             int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2474             return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2475         } else {
2476             // slices go from bottom to top => we flip the image internally
2477             uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2478                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2479                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2480             int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2481             int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2482
2483             src2[0] += (srcSliceH-1)*srcStride[0];
2484             if(c->srcFormat != PIX_FMT_PAL8)
2485                     src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2486             src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2487             
2488             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2489         }
2490 }
2491
2492 /**
2493  * swscale warper, so we don't need to export the SwsContext
2494  */
2495 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2496                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2497         return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2498 }
2499
2500 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2501                                 float lumaSharpen, float chromaSharpen,
2502                                 float chromaHShift, float chromaVShift,
2503                                 int verbose)
2504 {
2505         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2506
2507         if(lumaGBlur!=0.0){
2508                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2509                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2510         }else{
2511                 filter->lumH= sws_getIdentityVec();
2512                 filter->lumV= sws_getIdentityVec();
2513         }
2514
2515         if(chromaGBlur!=0.0){
2516                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2517                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2518         }else{
2519                 filter->chrH= sws_getIdentityVec();
2520                 filter->chrV= sws_getIdentityVec();
2521         }
2522
2523         if(chromaSharpen!=0.0){
2524                 SwsVector *id= sws_getIdentityVec();
2525                 sws_scaleVec(filter->chrH, -chromaSharpen);
2526                 sws_scaleVec(filter->chrV, -chromaSharpen);
2527                 sws_addVec(filter->chrH, id);
2528                 sws_addVec(filter->chrV, id);
2529                 sws_freeVec(id);
2530         }
2531
2532         if(lumaSharpen!=0.0){
2533                 SwsVector *id= sws_getIdentityVec();
2534                 sws_scaleVec(filter->lumH, -lumaSharpen);
2535                 sws_scaleVec(filter->lumV, -lumaSharpen);
2536                 sws_addVec(filter->lumH, id);
2537                 sws_addVec(filter->lumV, id);
2538                 sws_freeVec(id);
2539         }
2540
2541         if(chromaHShift != 0.0)
2542                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2543
2544         if(chromaVShift != 0.0)
2545                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2546
2547         sws_normalizeVec(filter->chrH, 1.0);
2548         sws_normalizeVec(filter->chrV, 1.0);
2549         sws_normalizeVec(filter->lumH, 1.0);
2550         sws_normalizeVec(filter->lumV, 1.0);
2551
2552         if(verbose) sws_printVec(filter->chrH);
2553         if(verbose) sws_printVec(filter->lumH);
2554
2555         return filter;
2556 }
2557
2558 /**
2559  * returns a normalized gaussian curve used to filter stuff
2560  * quality=3 is high quality, lowwer is lowwer quality
2561  */
2562 SwsVector *sws_getGaussianVec(double variance, double quality){
2563         const int length= (int)(variance*quality + 0.5) | 1;
2564         int i;
2565         double *coeff= av_malloc(length*sizeof(double));
2566         double middle= (length-1)*0.5;
2567         SwsVector *vec= av_malloc(sizeof(SwsVector));
2568
2569         vec->coeff= coeff;
2570         vec->length= length;
2571
2572         for(i=0; i<length; i++)
2573         {
2574                 double dist= i-middle;
2575                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2576         }
2577
2578         sws_normalizeVec(vec, 1.0);
2579
2580         return vec;
2581 }
2582
2583 SwsVector *sws_getConstVec(double c, int length){
2584         int i;
2585         double *coeff= av_malloc(length*sizeof(double));
2586         SwsVector *vec= av_malloc(sizeof(SwsVector));
2587
2588         vec->coeff= coeff;
2589         vec->length= length;
2590
2591         for(i=0; i<length; i++)
2592                 coeff[i]= c;
2593
2594         return vec;
2595 }
2596
2597
2598 SwsVector *sws_getIdentityVec(void){
2599         return sws_getConstVec(1.0, 1);
2600 }
2601
2602 double sws_dcVec(SwsVector *a){
2603         int i;
2604         double sum=0;
2605
2606         for(i=0; i<a->length; i++)
2607                 sum+= a->coeff[i];
2608
2609         return sum;
2610 }
2611
2612 void sws_scaleVec(SwsVector *a, double scalar){
2613         int i;
2614
2615         for(i=0; i<a->length; i++)
2616                 a->coeff[i]*= scalar;
2617 }
2618
2619 void sws_normalizeVec(SwsVector *a, double height){
2620         sws_scaleVec(a, height/sws_dcVec(a));
2621 }
2622
2623 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2624         int length= a->length + b->length - 1;
2625         double *coeff= av_malloc(length*sizeof(double));
2626         int i, j;
2627         SwsVector *vec= av_malloc(sizeof(SwsVector));
2628
2629         vec->coeff= coeff;
2630         vec->length= length;
2631
2632         for(i=0; i<length; i++) coeff[i]= 0.0;
2633
2634         for(i=0; i<a->length; i++)
2635         {
2636                 for(j=0; j<b->length; j++)
2637                 {
2638                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2639                 }
2640         }
2641
2642         return vec;
2643 }
2644
2645 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2646         int length= FFMAX(a->length, b->length);
2647         double *coeff= av_malloc(length*sizeof(double));
2648         int i;
2649         SwsVector *vec= av_malloc(sizeof(SwsVector));
2650
2651         vec->coeff= coeff;
2652         vec->length= length;
2653
2654         for(i=0; i<length; i++) coeff[i]= 0.0;
2655
2656         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2657         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2658
2659         return vec;
2660 }
2661
2662 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2663         int length= FFMAX(a->length, b->length);
2664         double *coeff= av_malloc(length*sizeof(double));
2665         int i;
2666         SwsVector *vec= av_malloc(sizeof(SwsVector));
2667
2668         vec->coeff= coeff;
2669         vec->length= length;
2670
2671         for(i=0; i<length; i++) coeff[i]= 0.0;
2672
2673         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2674         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2675
2676         return vec;
2677 }
2678
2679 /* shift left / or right if "shift" is negative */
2680 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2681         int length= a->length + FFABS(shift)*2;
2682         double *coeff= av_malloc(length*sizeof(double));
2683         int i;
2684         SwsVector *vec= av_malloc(sizeof(SwsVector));
2685
2686         vec->coeff= coeff;
2687         vec->length= length;
2688
2689         for(i=0; i<length; i++) coeff[i]= 0.0;
2690
2691         for(i=0; i<a->length; i++)
2692         {
2693                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2694         }
2695
2696         return vec;
2697 }
2698
2699 void sws_shiftVec(SwsVector *a, int shift){
2700         SwsVector *shifted= sws_getShiftedVec(a, shift);
2701         av_free(a->coeff);
2702         a->coeff= shifted->coeff;
2703         a->length= shifted->length;
2704         av_free(shifted);
2705 }
2706
2707 void sws_addVec(SwsVector *a, SwsVector *b){
2708         SwsVector *sum= sws_sumVec(a, b);
2709         av_free(a->coeff);
2710         a->coeff= sum->coeff;
2711         a->length= sum->length;
2712         av_free(sum);
2713 }
2714
2715 void sws_subVec(SwsVector *a, SwsVector *b){
2716         SwsVector *diff= sws_diffVec(a, b);
2717         av_free(a->coeff);
2718         a->coeff= diff->coeff;
2719         a->length= diff->length;
2720         av_free(diff);
2721 }
2722
2723 void sws_convVec(SwsVector *a, SwsVector *b){
2724         SwsVector *conv= sws_getConvVec(a, b);
2725         av_free(a->coeff);  
2726         a->coeff= conv->coeff;
2727         a->length= conv->length;
2728         av_free(conv);
2729 }
2730
2731 SwsVector *sws_cloneVec(SwsVector *a){
2732         double *coeff= av_malloc(a->length*sizeof(double));
2733         int i;
2734         SwsVector *vec= av_malloc(sizeof(SwsVector));
2735
2736         vec->coeff= coeff;
2737         vec->length= a->length;
2738
2739         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2740
2741         return vec;
2742 }
2743
2744 void sws_printVec(SwsVector *a){
2745         int i;
2746         double max=0;
2747         double min=0;
2748         double range;
2749
2750         for(i=0; i<a->length; i++)
2751                 if(a->coeff[i]>max) max= a->coeff[i];
2752
2753         for(i=0; i<a->length; i++)
2754                 if(a->coeff[i]<min) min= a->coeff[i];
2755
2756         range= max - min;
2757
2758         for(i=0; i<a->length; i++)
2759         {
2760                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2761                 av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2762                 for(;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2763                 av_log(NULL, AV_LOG_DEBUG, "|\n");
2764         }
2765 }
2766
2767 void sws_freeVec(SwsVector *a){
2768         if(!a) return;
2769         av_free(a->coeff);
2770         a->coeff=NULL;
2771         a->length=0;
2772         av_free(a);
2773 }
2774
2775 void sws_freeFilter(SwsFilter *filter){
2776         if(!filter) return;
2777
2778         if(filter->lumH) sws_freeVec(filter->lumH);
2779         if(filter->lumV) sws_freeVec(filter->lumV);
2780         if(filter->chrH) sws_freeVec(filter->chrH);
2781         if(filter->chrV) sws_freeVec(filter->chrV);
2782         av_free(filter);
2783 }
2784
2785
2786 void sws_freeContext(SwsContext *c){
2787         int i;
2788         if(!c) return;
2789
2790         if(c->lumPixBuf)
2791         {
2792                 for(i=0; i<c->vLumBufSize; i++)
2793                 {
2794                         av_free(c->lumPixBuf[i]);
2795                         c->lumPixBuf[i]=NULL;
2796                 }
2797                 av_free(c->lumPixBuf);
2798                 c->lumPixBuf=NULL;
2799         }
2800
2801         if(c->chrPixBuf)
2802         {
2803                 for(i=0; i<c->vChrBufSize; i++)
2804                 {
2805                         av_free(c->chrPixBuf[i]);
2806                         c->chrPixBuf[i]=NULL;
2807                 }
2808                 av_free(c->chrPixBuf);
2809                 c->chrPixBuf=NULL;
2810         }
2811
2812         av_free(c->vLumFilter);
2813         c->vLumFilter = NULL;
2814         av_free(c->vChrFilter);
2815         c->vChrFilter = NULL;
2816         av_free(c->hLumFilter);
2817         c->hLumFilter = NULL;
2818         av_free(c->hChrFilter);
2819         c->hChrFilter = NULL;
2820 #ifdef HAVE_ALTIVEC
2821         av_free(c->vYCoeffsBank);
2822         c->vYCoeffsBank = NULL;
2823         av_free(c->vCCoeffsBank);
2824         c->vCCoeffsBank = NULL;
2825 #endif
2826
2827         av_free(c->vLumFilterPos);
2828         c->vLumFilterPos = NULL;
2829         av_free(c->vChrFilterPos);
2830         c->vChrFilterPos = NULL;
2831         av_free(c->hLumFilterPos);
2832         c->hLumFilterPos = NULL;
2833         av_free(c->hChrFilterPos);
2834         c->hChrFilterPos = NULL;
2835
2836 #if defined(ARCH_X86) && defined(CONFIG_GPL)
2837 #ifdef MAP_ANONYMOUS
2838         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2839         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2840 #else
2841         av_free(c->funnyYCode);
2842         av_free(c->funnyUVCode);
2843 #endif
2844         c->funnyYCode=NULL;
2845         c->funnyUVCode=NULL;
2846 #endif /* defined(ARCH_X86) */
2847
2848         av_free(c->lumMmx2Filter);
2849         c->lumMmx2Filter=NULL;
2850         av_free(c->chrMmx2Filter);
2851         c->chrMmx2Filter=NULL;
2852         av_free(c->lumMmx2FilterPos);
2853         c->lumMmx2FilterPos=NULL;
2854         av_free(c->chrMmx2FilterPos);
2855         c->chrMmx2FilterPos=NULL;
2856         av_free(c->yuvTable);
2857         c->yuvTable=NULL;
2858
2859         av_free(c);
2860 }
2861
2862 /**
2863  * Checks if context is valid or reallocs a new one instead.
2864  * If context is NULL, just calls sws_getContext() to get a new one.
2865  * Otherwise, checks if the parameters are the same already saved in context.
2866  * If that is the case, returns the current context.
2867  * Otherwise, frees context and gets a new one.
2868  *
2869  * Be warned that srcFilter, dstFilter are not checked, they are
2870  * asumed to remain valid.
2871  */
2872 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2873                         int srcW, int srcH, int srcFormat,
2874                         int dstW, int dstH, int dstFormat, int flags,
2875                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2876 {
2877     if (context != NULL) {
2878         if ((context->srcW != srcW) || (context->srcH != srcH) ||
2879             (context->srcFormat != srcFormat) ||
2880             (context->dstW != dstW) || (context->dstH != dstH) ||
2881             (context->dstFormat != dstFormat) || (context->flags != flags) ||
2882             (context->param != param))
2883         {
2884             sws_freeContext(context);
2885             context = NULL;
2886         }
2887     }
2888     if (context == NULL) {
2889         return sws_getContext(srcW, srcH, srcFormat,
2890                         dstW, dstH, dstFormat, flags,
2891                         srcFilter, dstFilter, param);
2892     }
2893     return context;
2894 }
2895