outputs bit spent on various encoding functions (motion vectors, overhead, etc)
[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 the swscaler which has been written
21  * by Michael Niedermayer can be used under the LGPL license too
22  */
23
24 /*
25   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
26   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27   {BGR,RGB}{1,4,8,15,16} support dithering
28   
29   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
31   x -> x
32   YUV9 -> YV12
33   YUV9/YV12 -> Y800
34   Y800 -> YUV9/YV12
35   BGR24 -> BGR32 & RGB24 -> RGB32
36   BGR32 -> BGR24 & RGB32 -> RGB24
37   BGR15 -> BGR16
38 */
39
40 /* 
41 tested special converters (most are tested actually but i didnt write it down ...)
42  YV12 -> BGR16
43  YV12 -> YV12
44  BGR15 -> BGR16
45  BGR16 -> BGR16
46  YVU9 -> YV12
47
48 untested special converters
49   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
50   YV12/I420 -> YV12/I420
51   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52   BGR24 -> BGR32 & RGB24 -> RGB32
53   BGR32 -> BGR24 & RGB32 -> RGB24
54   BGR24 -> YV12
55 */
56
57 #include <inttypes.h>
58 #include <string.h>
59 #include <math.h>
60 #include <stdio.h>
61 #include <unistd.h>
62 #include "config.h"
63 #include <assert.h>
64 #ifdef HAVE_MALLOC_H
65 #include <malloc.h>
66 #else
67 #include <stdlib.h>
68 #endif
69 #ifdef HAVE_SYS_MMAN_H
70 #include <sys/mman.h>
71 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
72 #define MAP_ANONYMOUS MAP_ANON
73 #endif
74 #endif
75 #include "swscale.h"
76 #include "swscale_internal.h"
77 #include "x86_cpu.h"
78 #include "bswap.h"
79 #include "rgb2rgb.h"
80 #ifdef USE_FASTMEMCPY
81 #include "libvo/fastmemcpy.h"
82 #endif
83
84 #undef MOVNTQ
85 #undef PAVGB
86
87 //#undef HAVE_MMX2
88 //#define HAVE_3DNOW
89 //#undef HAVE_MMX
90 //#undef ARCH_X86
91 //#define WORDS_BIGENDIAN
92 #define DITHER1XBPP
93
94 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
95
96 #define RET 0xC3 //near return opcode for X86
97
98 #ifdef MP_DEBUG
99 #define ASSERT(x) assert(x);
100 #else
101 #define ASSERT(x) ;
102 #endif
103
104 #ifdef M_PI
105 #define PI M_PI
106 #else
107 #define PI 3.14159265358979323846
108 #endif
109
110 #define isSupportedIn(x)  ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
111                         || (x)==PIX_FMT_RGB32|| (x)==PIX_FMT_BGR24|| (x)==PIX_FMT_BGR565|| (x)==PIX_FMT_BGR555\
112                         || (x)==PIX_FMT_BGR32|| (x)==PIX_FMT_RGB24|| (x)==PIX_FMT_RGB565|| (x)==PIX_FMT_RGB555\
113                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P\
114                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
115                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P)
116 #define isSupportedOut(x) ((x)==PIX_FMT_YUV420P || (x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422\
117                         || (x)==PIX_FMT_YUV444P || (x)==PIX_FMT_YUV422P || (x)==PIX_FMT_YUV411P\
118                         || isRGB(x) || isBGR(x)\
119                         || (x)==PIX_FMT_NV12 || (x)==PIX_FMT_NV21\
120                         || (x)==PIX_FMT_GRAY16BE || (x)==PIX_FMT_GRAY16LE\
121                         || (x)==PIX_FMT_GRAY8 || (x)==PIX_FMT_YUV410P)
122 #define isPacked(x)    ((x)==PIX_FMT_YUYV422 || (x)==PIX_FMT_UYVY422 ||isRGB(x) || isBGR(x))
123
124 #define RGB2YUV_SHIFT 16
125 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
126 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
127 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
129 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
130 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
131 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
132 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
133 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
134
135 extern const int32_t Inverse_Table_6_9[8][4];
136
137 /*
138 NOTES
139 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
140
141 TODO
142 more intelligent missalignment avoidance for the horizontal scaler
143 write special vertical cubic upscale version
144 Optimize C code (yv12 / minmax)
145 add support for packed pixel yuv input & output
146 add support for Y8 output
147 optimize bgr24 & bgr32
148 add BGR4 output support
149 write special BGR->BGR scaler
150 */
151
152 #if defined(ARCH_X86)
153 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
154 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
155 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
156 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
157 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
158 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
159 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
160 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
161
162 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
163 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
164 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
165 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
166
167 static uint64_t __attribute__((aligned(8))) dither4[2]={
168         0x0103010301030103LL,
169         0x0200020002000200LL,};
170
171 static uint64_t __attribute__((aligned(8))) dither8[2]={
172         0x0602060206020602LL,
173         0x0004000400040004LL,};
174
175 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
176 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
177 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
178 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
179 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
180 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
181
182 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
183 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
184 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
185
186 #ifdef FAST_BGR2YV12
187 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
188 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
189 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
190 #else
191 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
192 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
193 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
194 #endif /* FAST_BGR2YV12 */
195 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
196 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
197 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
198 #endif /* defined(ARCH_X86) */
199
200 // clipping helper table for C implementations:
201 static unsigned char clip_table[768];
202
203 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
204                   
205 extern const uint8_t dither_2x2_4[2][8];
206 extern const uint8_t dither_2x2_8[2][8];
207 extern const uint8_t dither_8x8_32[8][8];
208 extern const uint8_t dither_8x8_73[8][8];
209 extern const uint8_t dither_8x8_220[8][8];
210
211 char *sws_format_name(enum PixelFormat format)
212 {
213     switch (format) {
214         case PIX_FMT_YUV420P:
215             return "yuv420p";
216         case PIX_FMT_YUYV422:
217             return "yuyv422";
218         case PIX_FMT_RGB24:
219             return "rgb24";
220         case PIX_FMT_BGR24:
221             return "bgr24";
222         case PIX_FMT_YUV422P:
223             return "yuv422p";
224         case PIX_FMT_YUV444P:
225             return "yuv444p";
226         case PIX_FMT_RGB32:
227             return "rgb32";
228         case PIX_FMT_YUV410P:
229             return "yuv410p";
230         case PIX_FMT_YUV411P:
231             return "yuv411p";
232         case PIX_FMT_RGB565:
233             return "rgb565";
234         case PIX_FMT_RGB555:
235             return "rgb555";
236         case PIX_FMT_GRAY16BE:
237             return "gray16be";
238         case PIX_FMT_GRAY16LE:
239             return "gray16le";
240         case PIX_FMT_GRAY8:
241             return "gray8";
242         case PIX_FMT_MONOWHITE:
243             return "mono white";
244         case PIX_FMT_MONOBLACK:
245             return "mono black";
246         case PIX_FMT_PAL8:
247             return "Palette";
248         case PIX_FMT_YUVJ420P:
249             return "yuvj420p";
250         case PIX_FMT_YUVJ422P:
251             return "yuvj422p";
252         case PIX_FMT_YUVJ444P:
253             return "yuvj444p";
254         case PIX_FMT_XVMC_MPEG2_MC:
255             return "xvmc_mpeg2_mc";
256         case PIX_FMT_XVMC_MPEG2_IDCT:
257             return "xvmc_mpeg2_idct";
258         case PIX_FMT_UYVY422:
259             return "uyvy422";
260         case PIX_FMT_UYYVYY411:
261             return "uyyvyy411";
262         case PIX_FMT_RGB32_1:
263             return "rgb32x";
264         case PIX_FMT_BGR32_1:
265             return "bgr32x";
266         case PIX_FMT_BGR32:
267             return "bgr32";
268         case PIX_FMT_BGR565:
269             return "bgr565";
270         case PIX_FMT_BGR555:
271             return "bgr555";
272         case PIX_FMT_BGR8:
273             return "bgr8";
274         case PIX_FMT_BGR4:
275             return "bgr4";
276         case PIX_FMT_BGR4_BYTE:
277             return "bgr4 byte";
278         case PIX_FMT_RGB8:
279             return "rgb8";
280         case PIX_FMT_RGB4:
281             return "rgb4";
282         case PIX_FMT_RGB4_BYTE:
283             return "rgb4 byte";
284         case PIX_FMT_NV12:
285             return "nv12";
286         case PIX_FMT_NV21:
287             return "nv21";
288         default:
289             return "Unknown format";
290     }
291 }
292
293 #if defined(ARCH_X86)
294 void in_asm_used_var_warning_killer()
295 {
296  volatile int i= bF8+bFC+w10+
297  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
298  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
299  if(i) i=0;
300 }
301 #endif
302
303 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
304                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
305                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
306 {
307         //FIXME Optimize (just quickly writen not opti..)
308         int i;
309         for(i=0; i<dstW; i++)
310         {
311                 int val=1<<18;
312                 int j;
313                 for(j=0; j<lumFilterSize; j++)
314                         val += lumSrc[j][i] * lumFilter[j];
315
316                 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
317         }
318
319         if(uDest != NULL)
320                 for(i=0; i<chrDstW; i++)
321                 {
322                         int u=1<<18;
323                         int v=1<<18;
324                         int j;
325                         for(j=0; j<chrFilterSize; j++)
326                         {
327                                 u += chrSrc[j][i] * chrFilter[j];
328                                 v += chrSrc[j][i + 2048] * chrFilter[j];
329                         }
330
331                         uDest[i]= FFMIN(FFMAX(u>>19, 0), 255);
332                         vDest[i]= FFMIN(FFMAX(v>>19, 0), 255);
333                 }
334 }
335
336 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
337                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
338                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
339 {
340         //FIXME Optimize (just quickly writen not opti..)
341         int i;
342         for(i=0; i<dstW; i++)
343         {
344                 int val=1<<18;
345                 int j;
346                 for(j=0; j<lumFilterSize; j++)
347                         val += lumSrc[j][i] * lumFilter[j];
348
349                 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
350         }
351
352         if(uDest == NULL)
353                 return;
354
355         if(dstFormat == PIX_FMT_NV12)
356                 for(i=0; i<chrDstW; i++)
357                 {
358                         int u=1<<18;
359                         int v=1<<18;
360                         int j;
361                         for(j=0; j<chrFilterSize; j++)
362                         {
363                                 u += chrSrc[j][i] * chrFilter[j];
364                                 v += chrSrc[j][i + 2048] * chrFilter[j];
365                         }
366
367                         uDest[2*i]= FFMIN(FFMAX(u>>19, 0), 255);
368                         uDest[2*i+1]= FFMIN(FFMAX(v>>19, 0), 255);
369                 }
370         else
371                 for(i=0; i<chrDstW; i++)
372                 {
373                         int u=1<<18;
374                         int v=1<<18;
375                         int j;
376                         for(j=0; j<chrFilterSize; j++)
377                         {
378                                 u += chrSrc[j][i] * chrFilter[j];
379                                 v += chrSrc[j][i + 2048] * chrFilter[j];
380                         }
381
382                         uDest[2*i]= FFMIN(FFMAX(v>>19, 0), 255);
383                         uDest[2*i+1]= FFMIN(FFMAX(u>>19, 0), 255);
384                 }
385 }
386
387 #define YSCALE_YUV_2_PACKEDX_C(type) \
388                 for(i=0; i<(dstW>>1); i++){\
389                         int j;\
390                         int Y1=1<<18;\
391                         int Y2=1<<18;\
392                         int U=1<<18;\
393                         int V=1<<18;\
394                         type *r, *b, *g;\
395                         const int i2= 2*i;\
396                         \
397                         for(j=0; j<lumFilterSize; j++)\
398                         {\
399                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
400                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
401                         }\
402                         for(j=0; j<chrFilterSize; j++)\
403                         {\
404                                 U += chrSrc[j][i] * chrFilter[j];\
405                                 V += chrSrc[j][i+2048] * chrFilter[j];\
406                         }\
407                         Y1>>=19;\
408                         Y2>>=19;\
409                         U >>=19;\
410                         V >>=19;\
411                         if((Y1|Y2|U|V)&256)\
412                         {\
413                                 if(Y1>255)   Y1=255;\
414                                 else if(Y1<0)Y1=0;\
415                                 if(Y2>255)   Y2=255;\
416                                 else if(Y2<0)Y2=0;\
417                                 if(U>255)    U=255;\
418                                 else if(U<0) U=0;\
419                                 if(V>255)    V=255;\
420                                 else if(V<0) V=0;\
421                         }
422                         
423 #define YSCALE_YUV_2_RGBX_C(type) \
424                         YSCALE_YUV_2_PACKEDX_C(type)\
425                         r = c->table_rV[V];\
426                         g = c->table_gU[U] + c->table_gV[V];\
427                         b = c->table_bU[U];\
428
429 #define YSCALE_YUV_2_PACKED2_C \
430                 for(i=0; i<(dstW>>1); i++){\
431                         const int i2= 2*i;\
432                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
433                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
434                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
435                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
436
437 #define YSCALE_YUV_2_RGB2_C(type) \
438                         YSCALE_YUV_2_PACKED2_C\
439                         type *r, *b, *g;\
440                         r = c->table_rV[V];\
441                         g = c->table_gU[U] + c->table_gV[V];\
442                         b = c->table_bU[U];\
443
444 #define YSCALE_YUV_2_PACKED1_C \
445                 for(i=0; i<(dstW>>1); i++){\
446                         const int i2= 2*i;\
447                         int Y1= buf0[i2  ]>>7;\
448                         int Y2= buf0[i2+1]>>7;\
449                         int U= (uvbuf1[i     ])>>7;\
450                         int V= (uvbuf1[i+2048])>>7;\
451
452 #define YSCALE_YUV_2_RGB1_C(type) \
453                         YSCALE_YUV_2_PACKED1_C\
454                         type *r, *b, *g;\
455                         r = c->table_rV[V];\
456                         g = c->table_gU[U] + c->table_gV[V];\
457                         b = c->table_bU[U];\
458
459 #define YSCALE_YUV_2_PACKED1B_C \
460                 for(i=0; i<(dstW>>1); i++){\
461                         const int i2= 2*i;\
462                         int Y1= buf0[i2  ]>>7;\
463                         int Y2= buf0[i2+1]>>7;\
464                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
465                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
466
467 #define YSCALE_YUV_2_RGB1B_C(type) \
468                         YSCALE_YUV_2_PACKED1B_C\
469                         type *r, *b, *g;\
470                         r = c->table_rV[V];\
471                         g = c->table_gU[U] + c->table_gV[V];\
472                         b = c->table_bU[U];\
473
474 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
475         switch(c->dstFormat)\
476         {\
477         case PIX_FMT_RGB32:\
478         case PIX_FMT_BGR32:\
479                 func(uint32_t)\
480                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
481                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
482                 }               \
483                 break;\
484         case PIX_FMT_RGB24:\
485                 func(uint8_t)\
486                         ((uint8_t*)dest)[0]= r[Y1];\
487                         ((uint8_t*)dest)[1]= g[Y1];\
488                         ((uint8_t*)dest)[2]= b[Y1];\
489                         ((uint8_t*)dest)[3]= r[Y2];\
490                         ((uint8_t*)dest)[4]= g[Y2];\
491                         ((uint8_t*)dest)[5]= b[Y2];\
492                         dest+=6;\
493                 }\
494                 break;\
495         case PIX_FMT_BGR24:\
496                 func(uint8_t)\
497                         ((uint8_t*)dest)[0]= b[Y1];\
498                         ((uint8_t*)dest)[1]= g[Y1];\
499                         ((uint8_t*)dest)[2]= r[Y1];\
500                         ((uint8_t*)dest)[3]= b[Y2];\
501                         ((uint8_t*)dest)[4]= g[Y2];\
502                         ((uint8_t*)dest)[5]= r[Y2];\
503                         dest+=6;\
504                 }\
505                 break;\
506         case PIX_FMT_RGB565:\
507         case PIX_FMT_BGR565:\
508                 {\
509                         const int dr1= dither_2x2_8[y&1    ][0];\
510                         const int dg1= dither_2x2_4[y&1    ][0];\
511                         const int db1= dither_2x2_8[(y&1)^1][0];\
512                         const int dr2= dither_2x2_8[y&1    ][1];\
513                         const int dg2= dither_2x2_4[y&1    ][1];\
514                         const int db2= dither_2x2_8[(y&1)^1][1];\
515                         func(uint16_t)\
516                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
517                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
518                         }\
519                 }\
520                 break;\
521         case PIX_FMT_RGB555:\
522         case PIX_FMT_BGR555:\
523                 {\
524                         const int dr1= dither_2x2_8[y&1    ][0];\
525                         const int dg1= dither_2x2_8[y&1    ][1];\
526                         const int db1= dither_2x2_8[(y&1)^1][0];\
527                         const int dr2= dither_2x2_8[y&1    ][1];\
528                         const int dg2= dither_2x2_8[y&1    ][0];\
529                         const int db2= dither_2x2_8[(y&1)^1][1];\
530                         func(uint16_t)\
531                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
532                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
533                         }\
534                 }\
535                 break;\
536         case PIX_FMT_RGB8:\
537         case PIX_FMT_BGR8:\
538                 {\
539                         const uint8_t * const d64= dither_8x8_73[y&7];\
540                         const uint8_t * const d32= dither_8x8_32[y&7];\
541                         func(uint8_t)\
542                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
543                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
544                         }\
545                 }\
546                 break;\
547         case PIX_FMT_RGB4:\
548         case PIX_FMT_BGR4:\
549                 {\
550                         const uint8_t * const d64= dither_8x8_73 [y&7];\
551                         const uint8_t * const d128=dither_8x8_220[y&7];\
552                         func(uint8_t)\
553                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
554                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
555                         }\
556                 }\
557                 break;\
558         case PIX_FMT_RGB4_BYTE:\
559         case PIX_FMT_BGR4_BYTE:\
560                 {\
561                         const uint8_t * const d64= dither_8x8_73 [y&7];\
562                         const uint8_t * const d128=dither_8x8_220[y&7];\
563                         func(uint8_t)\
564                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
565                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
566                         }\
567                 }\
568                 break;\
569         case PIX_FMT_MONOBLACK:\
570                 {\
571                         const uint8_t * const d128=dither_8x8_220[y&7];\
572                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
573                         for(i=0; i<dstW-7; i+=8){\
574                                 int acc;\
575                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
576                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
577                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
578                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
579                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
580                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
581                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
582                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
583                                 ((uint8_t*)dest)[0]= acc;\
584                                 dest++;\
585                         }\
586 \
587 /*\
588 ((uint8_t*)dest)-= dstW>>4;\
589 {\
590                         int acc=0;\
591                         int left=0;\
592                         static int top[1024];\
593                         static int last_new[1024][1024];\
594                         static int last_in3[1024][1024];\
595                         static int drift[1024][1024];\
596                         int topLeft=0;\
597                         int shift=0;\
598                         int count=0;\
599                         const uint8_t * const d128=dither_8x8_220[y&7];\
600                         int error_new=0;\
601                         int error_in3=0;\
602                         int f=0;\
603                         \
604                         for(i=dstW>>1; i<dstW; i++){\
605                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
606                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
607                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
608                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
609                                         + (last_new[y][i] - in3)*f/256;\
610                                 int new= old> 128 ? 255 : 0;\
611 \
612                                 error_new+= FFABS(last_new[y][i] - new);\
613                                 error_in3+= FFABS(last_in3[y][i] - in3);\
614                                 f= error_new - error_in3*4;\
615                                 if(f<0) f=0;\
616                                 if(f>256) f=256;\
617 \
618                                 topLeft= top[i];\
619                                 left= top[i]= old - new;\
620                                 last_new[y][i]= new;\
621                                 last_in3[y][i]= in3;\
622 \
623                                 acc+= acc + (new&1);\
624                                 if((i&7)==6){\
625                                         ((uint8_t*)dest)[0]= acc;\
626                                         ((uint8_t*)dest)++;\
627                                 }\
628                         }\
629 }\
630 */\
631                 }\
632                 break;\
633         case PIX_FMT_YUYV422:\
634                 func2\
635                         ((uint8_t*)dest)[2*i2+0]= Y1;\
636                         ((uint8_t*)dest)[2*i2+1]= U;\
637                         ((uint8_t*)dest)[2*i2+2]= Y2;\
638                         ((uint8_t*)dest)[2*i2+3]= V;\
639                 }               \
640                 break;\
641         case PIX_FMT_UYVY422:\
642                 func2\
643                         ((uint8_t*)dest)[2*i2+0]= U;\
644                         ((uint8_t*)dest)[2*i2+1]= Y1;\
645                         ((uint8_t*)dest)[2*i2+2]= V;\
646                         ((uint8_t*)dest)[2*i2+3]= Y2;\
647                 }               \
648                 break;\
649         }\
650
651
652 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
653                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
654                                     uint8_t *dest, int dstW, int y)
655 {
656         int i;
657         switch(c->dstFormat)
658         {
659         case PIX_FMT_BGR32:
660         case PIX_FMT_RGB32:
661                 YSCALE_YUV_2_RGBX_C(uint32_t)
662                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
663                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
664                 }
665                 break;
666         case PIX_FMT_RGB24:
667                 YSCALE_YUV_2_RGBX_C(uint8_t)
668                         ((uint8_t*)dest)[0]= r[Y1];
669                         ((uint8_t*)dest)[1]= g[Y1];
670                         ((uint8_t*)dest)[2]= b[Y1];
671                         ((uint8_t*)dest)[3]= r[Y2];
672                         ((uint8_t*)dest)[4]= g[Y2];
673                         ((uint8_t*)dest)[5]= b[Y2];
674                         dest+=6;
675                 }
676                 break;
677         case PIX_FMT_BGR24:
678                 YSCALE_YUV_2_RGBX_C(uint8_t)
679                         ((uint8_t*)dest)[0]= b[Y1];
680                         ((uint8_t*)dest)[1]= g[Y1];
681                         ((uint8_t*)dest)[2]= r[Y1];
682                         ((uint8_t*)dest)[3]= b[Y2];
683                         ((uint8_t*)dest)[4]= g[Y2];
684                         ((uint8_t*)dest)[5]= r[Y2];
685                         dest+=6;
686                 }
687                 break;
688         case PIX_FMT_RGB565:
689         case PIX_FMT_BGR565:
690                 {
691                         const int dr1= dither_2x2_8[y&1    ][0];
692                         const int dg1= dither_2x2_4[y&1    ][0];
693                         const int db1= dither_2x2_8[(y&1)^1][0];
694                         const int dr2= dither_2x2_8[y&1    ][1];
695                         const int dg2= dither_2x2_4[y&1    ][1];
696                         const int db2= dither_2x2_8[(y&1)^1][1];
697                         YSCALE_YUV_2_RGBX_C(uint16_t)
698                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
699                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
700                         }
701                 }
702                 break;
703         case PIX_FMT_RGB555:
704         case PIX_FMT_BGR555:
705                 {
706                         const int dr1= dither_2x2_8[y&1    ][0];
707                         const int dg1= dither_2x2_8[y&1    ][1];
708                         const int db1= dither_2x2_8[(y&1)^1][0];
709                         const int dr2= dither_2x2_8[y&1    ][1];
710                         const int dg2= dither_2x2_8[y&1    ][0];
711                         const int db2= dither_2x2_8[(y&1)^1][1];
712                         YSCALE_YUV_2_RGBX_C(uint16_t)
713                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
714                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
715                         }
716                 }
717                 break;
718         case PIX_FMT_RGB8:
719         case PIX_FMT_BGR8:
720                 {
721                         const uint8_t * const d64= dither_8x8_73[y&7];
722                         const uint8_t * const d32= dither_8x8_32[y&7];
723                         YSCALE_YUV_2_RGBX_C(uint8_t)
724                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
725                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
726                         }
727                 }
728                 break;
729         case PIX_FMT_RGB4:
730         case PIX_FMT_BGR4:
731                 {
732                         const uint8_t * const d64= dither_8x8_73 [y&7];
733                         const uint8_t * const d128=dither_8x8_220[y&7];
734                         YSCALE_YUV_2_RGBX_C(uint8_t)
735                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
736                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
737                         }
738                 }
739                 break;
740         case PIX_FMT_RGB4_BYTE:
741         case PIX_FMT_BGR4_BYTE:
742                 {
743                         const uint8_t * const d64= dither_8x8_73 [y&7];
744                         const uint8_t * const d128=dither_8x8_220[y&7];
745                         YSCALE_YUV_2_RGBX_C(uint8_t)
746                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
747                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
748                         }
749                 }
750                 break;
751         case PIX_FMT_MONOBLACK:
752                 {
753                         const uint8_t * const d128=dither_8x8_220[y&7];
754                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
755                         int acc=0;
756                         for(i=0; i<dstW-1; i+=2){
757                                 int j;
758                                 int Y1=1<<18;
759                                 int Y2=1<<18;
760
761                                 for(j=0; j<lumFilterSize; j++)
762                                 {
763                                         Y1 += lumSrc[j][i] * lumFilter[j];
764                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
765                                 }
766                                 Y1>>=19;
767                                 Y2>>=19;
768                                 if((Y1|Y2)&256)
769                                 {
770                                         if(Y1>255)   Y1=255;
771                                         else if(Y1<0)Y1=0;
772                                         if(Y2>255)   Y2=255;
773                                         else if(Y2<0)Y2=0;
774                                 }
775                                 acc+= acc + g[Y1+d128[(i+0)&7]];
776                                 acc+= acc + g[Y2+d128[(i+1)&7]];
777                                 if((i&7)==6){
778                                         ((uint8_t*)dest)[0]= acc;
779                                         dest++;
780                                 }
781                         }
782                 }
783                 break;
784         case PIX_FMT_YUYV422:
785                 YSCALE_YUV_2_PACKEDX_C(void)
786                         ((uint8_t*)dest)[2*i2+0]= Y1;
787                         ((uint8_t*)dest)[2*i2+1]= U;
788                         ((uint8_t*)dest)[2*i2+2]= Y2;
789                         ((uint8_t*)dest)[2*i2+3]= V;
790                 }
791                 break;
792         case PIX_FMT_UYVY422:
793                 YSCALE_YUV_2_PACKEDX_C(void)
794                         ((uint8_t*)dest)[2*i2+0]= U;
795                         ((uint8_t*)dest)[2*i2+1]= Y1;
796                         ((uint8_t*)dest)[2*i2+2]= V;
797                         ((uint8_t*)dest)[2*i2+3]= Y2;
798                 }
799                 break;
800         }
801 }
802
803
804 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
805 //Plain C versions
806 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
807 #define COMPILE_C
808 #endif
809
810 #ifdef ARCH_POWERPC
811 #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
812 #define COMPILE_ALTIVEC
813 #endif //HAVE_ALTIVEC
814 #endif //ARCH_POWERPC
815
816 #if defined(ARCH_X86)
817
818 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
819 #define COMPILE_MMX
820 #endif
821
822 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
823 #define COMPILE_MMX2
824 #endif
825
826 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
827 #define COMPILE_3DNOW
828 #endif
829 #endif //ARCH_X86 || ARCH_X86_64
830
831 #undef HAVE_MMX
832 #undef HAVE_MMX2
833 #undef HAVE_3DNOW
834
835 #ifdef COMPILE_C
836 #undef HAVE_MMX
837 #undef HAVE_MMX2
838 #undef HAVE_3DNOW
839 #undef HAVE_ALTIVEC
840 #define RENAME(a) a ## _C
841 #include "swscale_template.c"
842 #endif
843
844 #ifdef ARCH_POWERPC
845 #ifdef COMPILE_ALTIVEC
846 #undef RENAME
847 #define HAVE_ALTIVEC
848 #define RENAME(a) a ## _altivec
849 #include "swscale_template.c"
850 #endif
851 #endif //ARCH_POWERPC
852
853 #if defined(ARCH_X86)
854
855 //X86 versions
856 /*
857 #undef RENAME
858 #undef HAVE_MMX
859 #undef HAVE_MMX2
860 #undef HAVE_3DNOW
861 #define ARCH_X86
862 #define RENAME(a) a ## _X86
863 #include "swscale_template.c"
864 */
865 //MMX versions
866 #ifdef COMPILE_MMX
867 #undef RENAME
868 #define HAVE_MMX
869 #undef HAVE_MMX2
870 #undef HAVE_3DNOW
871 #define RENAME(a) a ## _MMX
872 #include "swscale_template.c"
873 #endif
874
875 //MMX2 versions
876 #ifdef COMPILE_MMX2
877 #undef RENAME
878 #define HAVE_MMX
879 #define HAVE_MMX2
880 #undef HAVE_3DNOW
881 #define RENAME(a) a ## _MMX2
882 #include "swscale_template.c"
883 #endif
884
885 //3DNOW versions
886 #ifdef COMPILE_3DNOW
887 #undef RENAME
888 #define HAVE_MMX
889 #undef HAVE_MMX2
890 #define HAVE_3DNOW
891 #define RENAME(a) a ## _3DNow
892 #include "swscale_template.c"
893 #endif
894
895 #endif //ARCH_X86 || ARCH_X86_64
896
897 // minor note: the HAVE_xyz is messed up after that line so don't use it
898
899 static double getSplineCoeff(double a, double b, double c, double d, double dist)
900 {
901 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
902         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
903         else            return getSplineCoeff(  0.0, 
904                                                  b+ 2.0*c + 3.0*d,
905                                                         c + 3.0*d,
906                                                 -b- 3.0*c - 6.0*d,
907                                                 dist-1.0);
908 }
909
910 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
911                               int srcW, int dstW, int filterAlign, int one, int flags,
912                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
913 {
914         int i;
915         int filterSize;
916         int filter2Size;
917         int minFilterSize;
918         double *filter=NULL;
919         double *filter2=NULL;
920 #if defined(ARCH_X86)
921         if(flags & SWS_CPU_CAPS_MMX)
922                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
923 #endif
924
925         // Note the +1 is for the MMXscaler which reads over the end
926         *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
927
928         if(FFABS(xInc - 0x10000) <10) // unscaled
929         {
930                 int i;
931                 filterSize= 1;
932                 filter= av_malloc(dstW*sizeof(double)*filterSize);
933                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
934
935                 for(i=0; i<dstW; i++)
936                 {
937                         filter[i*filterSize]=1;
938                         (*filterPos)[i]=i;
939                 }
940
941         }
942         else if(flags&SWS_POINT) // lame looking point sampling mode
943         {
944                 int i;
945                 int xDstInSrc;
946                 filterSize= 1;
947                 filter= av_malloc(dstW*sizeof(double)*filterSize);
948                 
949                 xDstInSrc= xInc/2 - 0x8000;
950                 for(i=0; i<dstW; i++)
951                 {
952                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
953
954                         (*filterPos)[i]= xx;
955                         filter[i]= 1.0;
956                         xDstInSrc+= xInc;
957                 }
958         }
959         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
960         {
961                 int i;
962                 int xDstInSrc;
963                 if     (flags&SWS_BICUBIC) filterSize= 4;
964                 else if(flags&SWS_X      ) filterSize= 4;
965                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
966                 filter= av_malloc(dstW*sizeof(double)*filterSize);
967
968                 xDstInSrc= xInc/2 - 0x8000;
969                 for(i=0; i<dstW; i++)
970                 {
971                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
972                         int j;
973
974                         (*filterPos)[i]= xx;
975                                 //Bilinear upscale / linear interpolate / Area averaging
976                                 for(j=0; j<filterSize; j++)
977                                 {
978                                         double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
979                                         double coeff= 1.0 - d;
980                                         if(coeff<0) coeff=0;
981                                         filter[i*filterSize + j]= coeff;
982                                         xx++;
983                                 }
984                         xDstInSrc+= xInc;
985                 }
986         }
987         else
988         {
989                 double xDstInSrc;
990                 double sizeFactor, filterSizeInSrc;
991                 const double xInc1= (double)xInc / (double)(1<<16);
992
993                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
994                 else if(flags&SWS_X)            sizeFactor= 8.0;
995                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
996                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
997                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
998                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
999                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
1000                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
1001                 else {
1002                         sizeFactor= 0.0; //GCC warning killer
1003                         ASSERT(0)
1004                 }
1005                 
1006                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
1007                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1008
1009                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1010                 if(filterSize > srcW-2) filterSize=srcW-2;
1011
1012                 filter= av_malloc(dstW*sizeof(double)*filterSize);
1013
1014                 xDstInSrc= xInc1 / 2.0 - 0.5;
1015                 for(i=0; i<dstW; i++)
1016                 {
1017                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1018                         int j;
1019                         (*filterPos)[i]= xx;
1020                         for(j=0; j<filterSize; j++)
1021                         {
1022                                 double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1023                                 double coeff;
1024                                 if(flags & SWS_BICUBIC)
1025                                 {
1026                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1027                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1028
1029                                         if(d<1.0) 
1030                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1031                                         else if(d<2.0)
1032                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1033                                         else
1034                                                 coeff=0.0;
1035                                 }
1036 /*                              else if(flags & SWS_X)
1037                                 {
1038                                         double p= param ? param*0.01 : 0.3;
1039                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1040                                         coeff*= pow(2.0, - p*d*d);
1041                                 }*/
1042                                 else if(flags & SWS_X)
1043                                 {
1044                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1045                                         
1046                                         if(d<1.0)
1047                                                 coeff = cos(d*PI);
1048                                         else
1049                                                 coeff=-1.0;
1050                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
1051                                         else            coeff=  pow( coeff, A);
1052                                         coeff= coeff*0.5 + 0.5;
1053                                 }
1054                                 else if(flags & SWS_AREA)
1055                                 {
1056                                         double srcPixelSize= 1.0/xInc1;
1057                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
1058                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1059                                         else coeff=0.0;
1060                                 }
1061                                 else if(flags & SWS_GAUSS)
1062                                 {
1063                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1064                                         coeff = pow(2.0, - p*d*d);
1065                                 }
1066                                 else if(flags & SWS_SINC)
1067                                 {
1068                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1069                                 }
1070                                 else if(flags & SWS_LANCZOS)
1071                                 {
1072                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
1073                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1074                                         if(d>p) coeff=0;
1075                                 }
1076                                 else if(flags & SWS_BILINEAR)
1077                                 {
1078                                         coeff= 1.0 - d;
1079                                         if(coeff<0) coeff=0;
1080                                 }
1081                                 else if(flags & SWS_SPLINE)
1082                                 {
1083                                         double p=-2.196152422706632;
1084                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1085                                 }
1086                                 else {
1087                                         coeff= 0.0; //GCC warning killer
1088                                         ASSERT(0)
1089                                 }
1090
1091                                 filter[i*filterSize + j]= coeff;
1092                                 xx++;
1093                         }
1094                         xDstInSrc+= xInc1;
1095                 }
1096         }
1097
1098         /* apply src & dst Filter to filter -> filter2
1099            av_free(filter);
1100         */
1101         ASSERT(filterSize>0)
1102         filter2Size= filterSize;
1103         if(srcFilter) filter2Size+= srcFilter->length - 1;
1104         if(dstFilter) filter2Size+= dstFilter->length - 1;
1105         ASSERT(filter2Size>0)
1106         filter2= av_malloc(filter2Size*dstW*sizeof(double));
1107
1108         for(i=0; i<dstW; i++)
1109         {
1110                 int j;
1111                 SwsVector scaleFilter;
1112                 SwsVector *outVec;
1113
1114                 scaleFilter.coeff= filter + i*filterSize;
1115                 scaleFilter.length= filterSize;
1116
1117                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1118                 else          outVec= &scaleFilter;
1119
1120                 ASSERT(outVec->length == filter2Size)
1121                 //FIXME dstFilter
1122
1123                 for(j=0; j<outVec->length; j++)
1124                 {
1125                         filter2[i*filter2Size + j]= outVec->coeff[j];
1126                 }
1127
1128                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1129
1130                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1131         }
1132         av_free(filter); filter=NULL;
1133
1134         /* try to reduce the filter-size (step1 find size and shift left) */
1135         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1136         minFilterSize= 0;
1137         for(i=dstW-1; i>=0; i--)
1138         {
1139                 int min= filter2Size;
1140                 int j;
1141                 double cutOff=0.0;
1142
1143                 /* get rid off near zero elements on the left by shifting left */
1144                 for(j=0; j<filter2Size; j++)
1145                 {
1146                         int k;
1147                         cutOff += FFABS(filter2[i*filter2Size]);
1148
1149                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1150
1151                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1152                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1153
1154                         // Move filter coeffs left
1155                         for(k=1; k<filter2Size; k++)
1156                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1157                         filter2[i*filter2Size + k - 1]= 0.0;
1158                         (*filterPos)[i]++;
1159                 }
1160
1161                 cutOff=0.0;
1162                 /* count near zeros on the right */
1163                 for(j=filter2Size-1; j>0; j--)
1164                 {
1165                         cutOff += FFABS(filter2[i*filter2Size + j]);
1166
1167                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1168                         min--;
1169                 }
1170
1171                 if(min>minFilterSize) minFilterSize= min;
1172         }
1173
1174         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1175           // we can handle the special case 4,
1176           // so we don't want to go to the full 8
1177           if (minFilterSize < 5)
1178             filterAlign = 4;
1179
1180           // we really don't want to waste our time
1181           // doing useless computation, so fall-back on
1182           // the scalar C code for very small filter.
1183           // vectorizing is worth it only if you have
1184           // decent-sized vector.
1185           if (minFilterSize < 3)
1186             filterAlign = 1;
1187         }
1188
1189         if (flags & SWS_CPU_CAPS_MMX) {
1190                 // special case for unscaled vertical filtering
1191                 if(minFilterSize == 1 && filterAlign == 2)
1192                         filterAlign= 1;
1193         }
1194
1195         ASSERT(minFilterSize > 0)
1196         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1197         ASSERT(filterSize > 0)
1198         filter= av_malloc(filterSize*dstW*sizeof(double));
1199         if(filterSize >= MAX_FILTER_SIZE)
1200                 return -1;
1201         *outFilterSize= filterSize;
1202
1203         if(flags&SWS_PRINT_INFO)
1204                 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1205         /* try to reduce the filter-size (step2 reduce it) */
1206         for(i=0; i<dstW; i++)
1207         {
1208                 int j;
1209
1210                 for(j=0; j<filterSize; j++)
1211                 {
1212                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1213                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1214                 }
1215         }
1216         av_free(filter2); filter2=NULL;
1217         
1218
1219         //FIXME try to align filterpos if possible
1220
1221         //fix borders
1222         for(i=0; i<dstW; i++)
1223         {
1224                 int j;
1225                 if((*filterPos)[i] < 0)
1226                 {
1227                         // Move filter coeffs left to compensate for filterPos
1228                         for(j=1; j<filterSize; j++)
1229                         {
1230                                 int left= FFMAX(j + (*filterPos)[i], 0);
1231                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1232                                 filter[i*filterSize + j]=0;
1233                         }
1234                         (*filterPos)[i]= 0;
1235                 }
1236
1237                 if((*filterPos)[i] + filterSize > srcW)
1238                 {
1239                         int shift= (*filterPos)[i] + filterSize - srcW;
1240                         // Move filter coeffs right to compensate for filterPos
1241                         for(j=filterSize-2; j>=0; j--)
1242                         {
1243                                 int right= FFMIN(j + shift, filterSize-1);
1244                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1245                                 filter[i*filterSize +j]=0;
1246                         }
1247                         (*filterPos)[i]= srcW - filterSize;
1248                 }
1249         }
1250
1251         // Note the +1 is for the MMXscaler which reads over the end
1252         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1253         *outFilter= av_malloc(*outFilterSize*(dstW+1)*sizeof(int16_t));
1254         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1255
1256         /* Normalize & Store in outFilter */
1257         for(i=0; i<dstW; i++)
1258         {
1259                 int j;
1260                 double error=0;
1261                 double sum=0;
1262                 double scale= one;
1263
1264                 for(j=0; j<filterSize; j++)
1265                 {
1266                         sum+= filter[i*filterSize + j];
1267                 }
1268                 scale/= sum;
1269                 for(j=0; j<*outFilterSize; j++)
1270                 {
1271                         double v= filter[i*filterSize + j]*scale + error;
1272                         int intV= floor(v + 0.5);
1273                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1274                         error = v - intV;
1275                 }
1276         }
1277         
1278         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1279         for(i=0; i<*outFilterSize; i++)
1280         {
1281                 int j= dstW*(*outFilterSize);
1282                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1283         }
1284
1285         av_free(filter);
1286         return 0;
1287 }
1288
1289 #ifdef COMPILE_MMX2
1290 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1291 {
1292         uint8_t *fragmentA;
1293         long imm8OfPShufW1A;
1294         long imm8OfPShufW2A;
1295         long fragmentLengthA;
1296         uint8_t *fragmentB;
1297         long imm8OfPShufW1B;
1298         long imm8OfPShufW2B;
1299         long fragmentLengthB;
1300         int fragmentPos;
1301
1302         int xpos, i;
1303
1304         // create an optimized horizontal scaling routine
1305
1306         //code fragment
1307
1308         asm volatile(
1309                 "jmp 9f                         \n\t"
1310         // Begin
1311                 "0:                             \n\t"
1312                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1313                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1314                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1315                 "punpcklbw %%mm7, %%mm1         \n\t"
1316                 "punpcklbw %%mm7, %%mm0         \n\t"
1317                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1318                 "1:                             \n\t"
1319                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1320                 "2:                             \n\t"
1321                 "psubw %%mm1, %%mm0             \n\t"
1322                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1323                 "pmullw %%mm3, %%mm0            \n\t"
1324                 "psllw $7, %%mm1                \n\t"
1325                 "paddw %%mm1, %%mm0             \n\t"
1326
1327                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1328
1329                 "add $8, %%"REG_a"              \n\t"
1330         // End
1331                 "9:                             \n\t"
1332 //              "int $3\n\t"
1333                 "lea 0b, %0                     \n\t"
1334                 "lea 1b, %1                     \n\t"
1335                 "lea 2b, %2                     \n\t"
1336                 "dec %1                         \n\t"
1337                 "dec %2                         \n\t"
1338                 "sub %0, %1                     \n\t"
1339                 "sub %0, %2                     \n\t"
1340                 "lea 9b, %3                     \n\t"
1341                 "sub %0, %3                     \n\t"
1342
1343
1344                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1345                 "=r" (fragmentLengthA)
1346         );
1347
1348         asm volatile(
1349                 "jmp 9f                         \n\t"
1350         // Begin
1351                 "0:                             \n\t"
1352                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1353                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1354                 "punpcklbw %%mm7, %%mm0         \n\t"
1355                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1356                 "1:                             \n\t"
1357                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1358                 "2:                             \n\t"
1359                 "psubw %%mm1, %%mm0             \n\t"
1360                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1361                 "pmullw %%mm3, %%mm0            \n\t"
1362                 "psllw $7, %%mm1                \n\t"
1363                 "paddw %%mm1, %%mm0             \n\t"
1364
1365                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1366
1367                 "add $8, %%"REG_a"              \n\t"
1368         // End
1369                 "9:                             \n\t"
1370 //              "int $3\n\t"
1371                 "lea 0b, %0                     \n\t"
1372                 "lea 1b, %1                     \n\t"
1373                 "lea 2b, %2                     \n\t"
1374                 "dec %1                         \n\t"
1375                 "dec %2                         \n\t"
1376                 "sub %0, %1                     \n\t"
1377                 "sub %0, %2                     \n\t"
1378                 "lea 9b, %3                     \n\t"
1379                 "sub %0, %3                     \n\t"
1380
1381
1382                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1383                 "=r" (fragmentLengthB)
1384         );
1385
1386         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1387         fragmentPos=0;
1388         
1389         for(i=0; i<dstW/numSplits; i++)
1390         {
1391                 int xx=xpos>>16;
1392
1393                 if((i&3) == 0)
1394                 {
1395                         int a=0;
1396                         int b=((xpos+xInc)>>16) - xx;
1397                         int c=((xpos+xInc*2)>>16) - xx;
1398                         int d=((xpos+xInc*3)>>16) - xx;
1399
1400                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1401                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1402                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1403                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1404                         filterPos[i/2]= xx;
1405
1406                         if(d+1<4)
1407                         {
1408                                 int maxShift= 3-(d+1);
1409                                 int shift=0;
1410
1411                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1412
1413                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1414                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1415                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1416                                         a | (b<<2) | (c<<4) | (d<<6);
1417
1418                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1419                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1420
1421                                 if(shift && i>=shift)
1422                                 {
1423                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1424                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1425                                         filterPos[i/2]-=shift;
1426                                 }
1427
1428                                 fragmentPos+= fragmentLengthB;
1429                         }
1430                         else
1431                         {
1432                                 int maxShift= 3-d;
1433                                 int shift=0;
1434
1435                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1436
1437                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1438                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1439                                         a | (b<<2) | (c<<4) | (d<<6);
1440
1441                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1442                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1443
1444                                 if(shift && i>=shift)
1445                                 {
1446                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1447                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1448                                         filterPos[i/2]-=shift;
1449                                 }
1450
1451                                 fragmentPos+= fragmentLengthA;
1452                         }
1453
1454                         funnyCode[fragmentPos]= RET;
1455                 }
1456                 xpos+=xInc;
1457         }
1458         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1459 }
1460 #endif /* COMPILE_MMX2 */
1461
1462 static void globalInit(void){
1463     // generating tables:
1464     int i;
1465     for(i=0; i<768; i++){
1466         int c= FFMIN(FFMAX(i-256, 0), 255);
1467         clip_table[i]=c;
1468     }
1469 }
1470
1471 static SwsFunc getSwsFunc(int flags){
1472     
1473 #ifdef RUNTIME_CPUDETECT
1474 #if defined(ARCH_X86)
1475         // ordered per speed fasterst first
1476         if(flags & SWS_CPU_CAPS_MMX2)
1477                 return swScale_MMX2;
1478         else if(flags & SWS_CPU_CAPS_3DNOW)
1479                 return swScale_3DNow;
1480         else if(flags & SWS_CPU_CAPS_MMX)
1481                 return swScale_MMX;
1482         else
1483                 return swScale_C;
1484
1485 #else
1486 #ifdef ARCH_POWERPC
1487         if(flags & SWS_CPU_CAPS_ALTIVEC)
1488           return swScale_altivec;
1489         else
1490           return swScale_C;
1491 #endif
1492         return swScale_C;
1493 #endif /* defined(ARCH_X86) */
1494 #else //RUNTIME_CPUDETECT
1495 #ifdef HAVE_MMX2
1496         return swScale_MMX2;
1497 #elif defined (HAVE_3DNOW)
1498         return swScale_3DNow;
1499 #elif defined (HAVE_MMX)
1500         return swScale_MMX;
1501 #elif defined (HAVE_ALTIVEC)
1502         return swScale_altivec;
1503 #else
1504         return swScale_C;
1505 #endif
1506 #endif //!RUNTIME_CPUDETECT
1507 }
1508
1509 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1510              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1511         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1512         /* Copy Y plane */
1513         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1514                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1515         else
1516         {
1517                 int i;
1518                 uint8_t *srcPtr= src[0];
1519                 uint8_t *dstPtr= dst;
1520                 for(i=0; i<srcSliceH; i++)
1521                 {
1522                         memcpy(dstPtr, srcPtr, c->srcW);
1523                         srcPtr+= srcStride[0];
1524                         dstPtr+= dstStride[0];
1525                 }
1526         }
1527         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1528         if (c->dstFormat == PIX_FMT_NV12)
1529                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1530         else
1531                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1532
1533         return srcSliceH;
1534 }
1535
1536 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1537              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1538         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1539
1540         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1541
1542         return srcSliceH;
1543 }
1544
1545 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1546              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1547         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1548
1549         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1550
1551         return srcSliceH;
1552 }
1553
1554 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1555 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1556                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1557         const int srcFormat= c->srcFormat;
1558         const int dstFormat= c->dstFormat;
1559         const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1560         const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1561         const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1562         const int dstId= fmt_depth(dstFormat) >> 2;
1563         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1564
1565         /* BGR -> BGR */
1566         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1567            || (isRGB(srcFormat) && isRGB(dstFormat))){
1568                 switch(srcId | (dstId<<4)){
1569                 case 0x34: conv= rgb16to15; break;
1570                 case 0x36: conv= rgb24to15; break;
1571                 case 0x38: conv= rgb32to15; break;
1572                 case 0x43: conv= rgb15to16; break;
1573                 case 0x46: conv= rgb24to16; break;
1574                 case 0x48: conv= rgb32to16; break;
1575                 case 0x63: conv= rgb15to24; break;
1576                 case 0x64: conv= rgb16to24; break;
1577                 case 0x68: conv= rgb32to24; break;
1578                 case 0x83: conv= rgb15to32; break;
1579                 case 0x84: conv= rgb16to32; break;
1580                 case 0x86: conv= rgb24to32; break;
1581                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1582                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1583                 }
1584         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1585                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1586                 switch(srcId | (dstId<<4)){
1587                 case 0x33: conv= rgb15tobgr15; break;
1588                 case 0x34: conv= rgb16tobgr15; break;
1589                 case 0x36: conv= rgb24tobgr15; break;
1590                 case 0x38: conv= rgb32tobgr15; break;
1591                 case 0x43: conv= rgb15tobgr16; break;
1592                 case 0x44: conv= rgb16tobgr16; break;
1593                 case 0x46: conv= rgb24tobgr16; break;
1594                 case 0x48: conv= rgb32tobgr16; break;
1595                 case 0x63: conv= rgb15tobgr24; break;
1596                 case 0x64: conv= rgb16tobgr24; break;
1597                 case 0x66: conv= rgb24tobgr24; break;
1598                 case 0x68: conv= rgb32tobgr24; break;
1599                 case 0x83: conv= rgb15tobgr32; break;
1600                 case 0x84: conv= rgb16tobgr32; break;
1601                 case 0x86: conv= rgb24tobgr32; break;
1602                 case 0x88: conv= rgb32tobgr32; break;
1603                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1604                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1605                 }
1606         }else{
1607                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1608                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1609         }
1610
1611         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1612                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1613         else
1614         {
1615                 int i;
1616                 uint8_t *srcPtr= src[0];
1617                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1618
1619                 for(i=0; i<srcSliceH; i++)
1620                 {
1621                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1622                         srcPtr+= srcStride[0];
1623                         dstPtr+= dstStride[0];
1624                 }
1625         }     
1626         return srcSliceH;
1627 }
1628
1629 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1630              int srcSliceH, uint8_t* dst[], int dstStride[]){
1631
1632         rgb24toyv12(
1633                 src[0], 
1634                 dst[0]+ srcSliceY    *dstStride[0], 
1635                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1636                 dst[2]+(srcSliceY>>1)*dstStride[2],
1637                 c->srcW, srcSliceH, 
1638                 dstStride[0], dstStride[1], srcStride[0]);
1639         return srcSliceH;
1640 }
1641
1642 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1643              int srcSliceH, uint8_t* dst[], int dstStride[]){
1644         int i;
1645
1646         /* copy Y */
1647         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1648                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1649         else{
1650                 uint8_t *srcPtr= src[0];
1651                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1652
1653                 for(i=0; i<srcSliceH; i++)
1654                 {
1655                         memcpy(dstPtr, srcPtr, c->srcW);
1656                         srcPtr+= srcStride[0];
1657                         dstPtr+= dstStride[0];
1658                 }
1659         }
1660
1661         if(c->dstFormat==PIX_FMT_YUV420P){
1662                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1663                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1664         }else{
1665                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1666                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1667         }
1668         return srcSliceH;
1669 }
1670
1671 /* unscaled copy like stuff (assumes nearly identical formats) */
1672 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1673              int srcSliceH, uint8_t* dst[], int dstStride[]){
1674
1675         if(isPacked(c->srcFormat))
1676         {
1677                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1678                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1679                 else
1680                 {
1681                         int i;
1682                         uint8_t *srcPtr= src[0];
1683                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1684                         int length=0;
1685
1686                         /* universal length finder */
1687                         while(length+c->srcW <= FFABS(dstStride[0]) 
1688                            && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1689                         ASSERT(length!=0);
1690
1691                         for(i=0; i<srcSliceH; i++)
1692                         {
1693                                 memcpy(dstPtr, srcPtr, length);
1694                                 srcPtr+= srcStride[0];
1695                                 dstPtr+= dstStride[0];
1696                         }
1697                 }
1698         }
1699         else 
1700         { /* Planar YUV or gray */
1701                 int plane;
1702                 for(plane=0; plane<3; plane++)
1703                 {
1704                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1705                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1706                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1707
1708                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1709                         {
1710                                 if(!isGray(c->dstFormat))
1711                                         memset(dst[plane], 128, dstStride[plane]*height);
1712                         }
1713                         else
1714                         {
1715                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1716                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1717                                 else
1718                                 {
1719                                         int i;
1720                                         uint8_t *srcPtr= src[plane];
1721                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1722                                         for(i=0; i<height; i++)
1723                                         {
1724                                                 memcpy(dstPtr, srcPtr, length);
1725                                                 srcPtr+= srcStride[plane];
1726                                                 dstPtr+= dstStride[plane];
1727                                         }
1728                                 }
1729                         }
1730                 }
1731         }
1732         return srcSliceH;
1733 }
1734
1735 static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1736              int srcSliceH, uint8_t* dst[], int dstStride[]){
1737
1738         int length= c->srcW;
1739         int y=      srcSliceY;
1740         int height= srcSliceH;
1741         int i, j;
1742         uint8_t *srcPtr= src[0];
1743         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1744
1745         if(!isGray(c->dstFormat)){
1746                 int height= -((-srcSliceH)>>c->chrDstVSubSample);
1747                 memset(dst[1], 128, dstStride[1]*height);
1748                 memset(dst[2], 128, dstStride[2]*height);
1749         }
1750         if(c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1751         for(i=0; i<height; i++)
1752         {
1753                 for(j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1754                 srcPtr+= srcStride[0];
1755                 dstPtr+= dstStride[0];
1756         }
1757         return srcSliceH;
1758 }
1759
1760 static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1761              int srcSliceH, uint8_t* dst[], int dstStride[]){
1762
1763         int length= c->srcW;
1764         int y=      srcSliceY;
1765         int height= srcSliceH;
1766         int i, j;
1767         uint8_t *srcPtr= src[0];
1768         uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1769         for(i=0; i<height; i++)
1770         {
1771                 for(j=0; j<length; j++)
1772                 {
1773                         dstPtr[j<<1] = srcPtr[j];
1774                         dstPtr[(j<<1)+1] = srcPtr[j];
1775                 }
1776                 srcPtr+= srcStride[0];
1777                 dstPtr+= dstStride[0];
1778         }
1779         return srcSliceH;
1780 }
1781
1782 static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1783              int srcSliceH, uint8_t* dst[], int dstStride[]){
1784
1785         int length= c->srcW;
1786         int y=      srcSliceY;
1787         int height= srcSliceH;
1788         int i, j;
1789         uint16_t *srcPtr= src[0];
1790         uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1791         for(i=0; i<height; i++)
1792         {
1793                 for(j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1794                 srcPtr+= srcStride[0]/2;
1795                 dstPtr+= dstStride[0]/2;
1796         }
1797         return srcSliceH;
1798 }
1799
1800
1801 static void getSubSampleFactors(int *h, int *v, int format){
1802         switch(format){
1803         case PIX_FMT_UYVY422:
1804         case PIX_FMT_YUYV422:
1805                 *h=1;
1806                 *v=0;
1807                 break;
1808         case PIX_FMT_YUV420P:
1809         case PIX_FMT_GRAY16BE:
1810         case PIX_FMT_GRAY16LE:
1811         case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1812         case PIX_FMT_NV12:
1813         case PIX_FMT_NV21:
1814                 *h=1;
1815                 *v=1;
1816                 break;
1817         case PIX_FMT_YUV410P:
1818                 *h=2;
1819                 *v=2;
1820                 break;
1821         case PIX_FMT_YUV444P:
1822                 *h=0;
1823                 *v=0;
1824                 break;
1825         case PIX_FMT_YUV422P:
1826                 *h=1;
1827                 *v=0;
1828                 break;
1829         case PIX_FMT_YUV411P:
1830                 *h=2;
1831                 *v=0;
1832                 break;
1833         default:
1834                 *h=0;
1835                 *v=0;
1836                 break;
1837         }
1838 }
1839
1840 static uint16_t roundToInt16(int64_t f){
1841         int r= (f + (1<<15))>>16;
1842              if(r<-0x7FFF) return 0x8000;
1843         else if(r> 0x7FFF) return 0x7FFF;
1844         else               return r;
1845 }
1846
1847 /**
1848  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1849  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1850  * @return -1 if not supported
1851  */
1852 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1853         int64_t crv =  inv_table[0];
1854         int64_t cbu =  inv_table[1];
1855         int64_t cgu = -inv_table[2];
1856         int64_t cgv = -inv_table[3];
1857         int64_t cy  = 1<<16;
1858         int64_t oy  = 0;
1859
1860         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1861         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1862         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1863
1864         c->brightness= brightness;
1865         c->contrast  = contrast;
1866         c->saturation= saturation;
1867         c->srcRange  = srcRange;
1868         c->dstRange  = dstRange;
1869
1870         c->uOffset=   0x0400040004000400LL;
1871         c->vOffset=   0x0400040004000400LL;
1872
1873         if(!srcRange){
1874                 cy= (cy*255) / 219;
1875                 oy= 16<<16;
1876         }
1877
1878         cy = (cy *contrast             )>>16;
1879         crv= (crv*contrast * saturation)>>32;
1880         cbu= (cbu*contrast * saturation)>>32;
1881         cgu= (cgu*contrast * saturation)>>32;
1882         cgv= (cgv*contrast * saturation)>>32;
1883
1884         oy -= 256*brightness;
1885
1886         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1887         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1888         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1889         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1890         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1891         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1892
1893         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1894         //FIXME factorize
1895
1896 #ifdef COMPILE_ALTIVEC
1897         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1898             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1899 #endif  
1900         return 0;
1901 }
1902
1903 /**
1904  * @return -1 if not supported
1905  */
1906 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1907         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1908
1909         *inv_table = c->srcColorspaceTable;
1910         *table     = c->dstColorspaceTable;
1911         *srcRange  = c->srcRange;
1912         *dstRange  = c->dstRange;
1913         *brightness= c->brightness;
1914         *contrast  = c->contrast;
1915         *saturation= c->saturation;
1916         
1917         return 0;       
1918 }
1919
1920 static int handle_jpeg(int *format)
1921 {
1922         switch (*format) {
1923                 case PIX_FMT_YUVJ420P:
1924                         *format = PIX_FMT_YUV420P;
1925                         return 1;
1926                 case PIX_FMT_YUVJ422P:
1927                         *format = PIX_FMT_YUV422P;
1928                         return 1;
1929                 case PIX_FMT_YUVJ444P:
1930                         *format = PIX_FMT_YUV444P;
1931                         return 1;
1932                 default:
1933                         return 0;
1934         }
1935 }
1936
1937 SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
1938                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1939
1940         SwsContext *c;
1941         int i;
1942         int usesVFilter, usesHFilter;
1943         int unscaled, needsDither;
1944         int srcRange, dstRange;
1945         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1946 #if defined(ARCH_X86)
1947         if(flags & SWS_CPU_CAPS_MMX)
1948                 asm volatile("emms\n\t"::: "memory");
1949 #endif
1950
1951 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1952         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1953 #ifdef HAVE_MMX2
1954         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1955 #elif defined (HAVE_3DNOW)
1956         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1957 #elif defined (HAVE_MMX)
1958         flags |= SWS_CPU_CAPS_MMX;
1959 #elif defined (HAVE_ALTIVEC)
1960         flags |= SWS_CPU_CAPS_ALTIVEC;
1961 #endif
1962 #endif /* RUNTIME_CPUDETECT */
1963         if(clip_table[512] != 255) globalInit();
1964         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1965
1966         unscaled = (srcW == dstW && srcH == dstH);
1967         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1968                      && (fmt_depth(dstFormat))<24
1969                      && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1970
1971         srcRange = handle_jpeg(&srcFormat);
1972         dstRange = handle_jpeg(&dstFormat);
1973
1974         if(!isSupportedIn(srcFormat)) 
1975         {
1976                 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1977                 return NULL;
1978         }
1979         if(!isSupportedOut(dstFormat))
1980         {
1981                 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1982                 return NULL;
1983         }
1984
1985         /* sanity check */
1986         if(srcW<4 || srcH<1 || dstW<8 || dstH<1) //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
1987         {
1988                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1989                         srcW, srcH, dstW, dstH);
1990                 return NULL;
1991         }
1992
1993         if(!dstFilter) dstFilter= &dummyFilter;
1994         if(!srcFilter) srcFilter= &dummyFilter;
1995
1996         c= av_malloc(sizeof(SwsContext));
1997         memset(c, 0, sizeof(SwsContext));
1998
1999         c->srcW= srcW;
2000         c->srcH= srcH;
2001         c->dstW= dstW;
2002         c->dstH= dstH;
2003         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2004         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2005         c->flags= flags;
2006         c->dstFormat= dstFormat;
2007         c->srcFormat= srcFormat;
2008         c->vRounder= 4* 0x0001000100010001ULL;
2009
2010         usesHFilter= usesVFilter= 0;
2011         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
2012         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
2013         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
2014         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
2015         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
2016         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
2017         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
2018         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
2019
2020         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2021         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2022
2023         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2024         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2025
2026         // drop some chroma lines if the user wants it
2027         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2028         c->chrSrcVSubSample+= c->vChrDrop;
2029
2030         // drop every 2. pixel for chroma calculation unless user wants full chroma
2031         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
2032                 c->chrSrcHSubSample=1;
2033
2034         if(param){
2035                 c->param[0] = param[0];
2036                 c->param[1] = param[1];
2037         }else{
2038                 c->param[0] =
2039                 c->param[1] = SWS_PARAM_DEFAULT;
2040         }
2041
2042         c->chrIntHSubSample= c->chrDstHSubSample;
2043         c->chrIntVSubSample= c->chrSrcVSubSample;
2044
2045         // note the -((-x)>>y) is so that we allways round toward +inf
2046         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2047         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2048         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2049         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2050
2051         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16); 
2052
2053         /* unscaled special Cases */
2054         if(unscaled && !usesHFilter && !usesVFilter)
2055         {
2056                 /* yv12_to_nv12 */
2057                 if(srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2058                 {
2059                         c->swScale= PlanarToNV12Wrapper;
2060                 }
2061                 /* yuv2bgr */
2062                 if((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2063                 {
2064                         c->swScale= yuv2rgb_get_func_ptr(c);
2065                 }
2066                 
2067                 if( srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P )
2068                 {
2069                         c->swScale= yvu9toyv12Wrapper;
2070                 }
2071
2072                 /* bgr24toYV12 */
2073                 if(srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2074                         c->swScale= bgr24toyv12Wrapper;
2075                 
2076                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2077                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
2078                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
2079                    && !needsDither)
2080                         c->swScale= rgb2rgbWrapper;
2081
2082                 /* LQ converters if -sws 0 or -sws 4*/
2083                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2084                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
2085                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
2086                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
2087                           && needsDither)
2088                                 c->swScale= rgb2rgbWrapper;
2089
2090                         /* yv12_to_yuy2 */
2091                         if(srcFormat == PIX_FMT_YUV420P && 
2092                             (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2093                         {
2094                                 if (dstFormat == PIX_FMT_YUYV422)
2095                                     c->swScale= PlanarToYuy2Wrapper;
2096                                 else
2097                                     c->swScale= PlanarToUyvyWrapper;
2098                         }
2099                 }
2100
2101 #ifdef COMPILE_ALTIVEC
2102                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2103                     ((srcFormat == PIX_FMT_YUV420P && 
2104                       (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2105                   // unscaled YV12 -> packed YUV, we want speed
2106                   if (dstFormat == PIX_FMT_YUYV422)
2107                     c->swScale= yv12toyuy2_unscaled_altivec;
2108                   else
2109                     c->swScale= yv12touyvy_unscaled_altivec;
2110                 }
2111 #endif
2112
2113                 /* simple copy */
2114                 if(   srcFormat == dstFormat
2115                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2116                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2117                   )
2118                 {
2119                         c->swScale= simpleCopy;
2120                 }
2121
2122                 /* gray16{le,be} conversions */
2123                 if(isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2124                 {
2125                         c->swScale= gray16togray;
2126                 }
2127                 if((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2128                 {
2129                         c->swScale= graytogray16;
2130                 }
2131                 if(srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2132                 {
2133                         c->swScale= gray16swap;
2134                 }               
2135
2136                 if(c->swScale){
2137                         if(flags&SWS_PRINT_INFO)
2138                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2139                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2140                         return c;
2141                 }
2142         }
2143
2144         if(flags & SWS_CPU_CAPS_MMX2)
2145         {
2146                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2147                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2148                 {
2149                         if(flags&SWS_PRINT_INFO)
2150                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2151                 }
2152                 if(usesHFilter) c->canMMX2BeUsed=0;
2153         }
2154         else
2155                 c->canMMX2BeUsed=0;
2156
2157         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2158         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2159
2160         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2161         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2162         // n-2 is the last chrominance sample available
2163         // this is not perfect, but noone shuld notice the difference, the more correct variant
2164         // would be like the vertical one, but that would require some special code for the
2165         // first and last pixel
2166         if(flags&SWS_FAST_BILINEAR)
2167         {
2168                 if(c->canMMX2BeUsed)
2169                 {
2170                         c->lumXInc+= 20;
2171                         c->chrXInc+= 20;
2172                 }
2173                 //we don't use the x86asm scaler if mmx is available
2174                 else if(flags & SWS_CPU_CAPS_MMX)
2175                 {
2176                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2177                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2178                 }
2179         }
2180
2181         /* precalculate horizontal scaler filter coefficients */
2182         {
2183                 const int filterAlign=
2184                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2185                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2186                   1;
2187
2188                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2189                                  srcW      ,       dstW, filterAlign, 1<<14,
2190                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2191                                  srcFilter->lumH, dstFilter->lumH, c->param);
2192                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2193                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2194                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2195                                  srcFilter->chrH, dstFilter->chrH, c->param);
2196
2197 #define MAX_FUNNY_CODE_SIZE 10000
2198 #if defined(COMPILE_MMX2)
2199 // can't downscale !!!
2200                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2201                 {
2202 #ifdef MAP_ANONYMOUS
2203                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2204                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2205 #else
2206                         c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2207                         c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2208 #endif
2209
2210                         c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2211                         c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2212                         c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2213                         c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2214
2215                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2216                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2217                 }
2218 #endif /* defined(COMPILE_MMX2) */
2219         } // Init Horizontal stuff
2220
2221
2222
2223         /* precalculate vertical scaler filter coefficients */
2224         {
2225                 const int filterAlign=
2226                   (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2227                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2228                   1;
2229
2230                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2231                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2232                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2233                                 srcFilter->lumV, dstFilter->lumV, c->param);
2234                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2235                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2236                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2237                                 srcFilter->chrV, dstFilter->chrV, c->param);
2238
2239 #ifdef HAVE_ALTIVEC
2240                 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2241                 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2242
2243                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2244                   int j;
2245                   short *p = (short *)&c->vYCoeffsBank[i];
2246                   for (j=0;j<8;j++)
2247                     p[j] = c->vLumFilter[i];
2248                 }
2249
2250                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2251                   int j;
2252                   short *p = (short *)&c->vCCoeffsBank[i];
2253                   for (j=0;j<8;j++)
2254                     p[j] = c->vChrFilter[i];
2255                 }
2256 #endif
2257         }
2258
2259         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2260         c->vLumBufSize= c->vLumFilterSize;
2261         c->vChrBufSize= c->vChrFilterSize;
2262         for(i=0; i<dstH; i++)
2263         {
2264                 int chrI= i*c->chrDstH / dstH;
2265                 int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2266                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2267
2268                 nextSlice>>= c->chrSrcVSubSample;
2269                 nextSlice<<= c->chrSrcVSubSample;
2270                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2271                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2272                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2273                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2274         }
2275
2276         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2277         c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2278         c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2279         //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)
2280         /* align at 16 bytes for AltiVec */
2281         for(i=0; i<c->vLumBufSize; i++)
2282                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_malloc(4000);
2283         for(i=0; i<c->vChrBufSize; i++)
2284                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2285
2286         //try to avoid drawing green stuff between the right end and the stride end
2287         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2288         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2289
2290         ASSERT(c->chrDstH <= dstH)
2291
2292         if(flags&SWS_PRINT_INFO)
2293         {
2294 #ifdef DITHER1XBPP
2295                 char *dither= " dithered";
2296 #else
2297                 char *dither= "";
2298 #endif
2299                 if(flags&SWS_FAST_BILINEAR)
2300                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2301                 else if(flags&SWS_BILINEAR)
2302                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2303                 else if(flags&SWS_BICUBIC)
2304                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2305                 else if(flags&SWS_X)
2306                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2307                 else if(flags&SWS_POINT)
2308                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2309                 else if(flags&SWS_AREA)
2310                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2311                 else if(flags&SWS_BICUBLIN)
2312                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2313                 else if(flags&SWS_GAUSS)
2314                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2315                 else if(flags&SWS_SINC)
2316                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2317                 else if(flags&SWS_LANCZOS)
2318                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2319                 else if(flags&SWS_SPLINE)
2320                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2321                 else
2322                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2323
2324                 if(dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2325                         MSG_INFO("from %s to%s %s ", 
2326                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2327                 else
2328                         MSG_INFO("from %s to %s ", 
2329                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2330
2331                 if(flags & SWS_CPU_CAPS_MMX2)
2332                         MSG_INFO("using MMX2\n");
2333                 else if(flags & SWS_CPU_CAPS_3DNOW)
2334                         MSG_INFO("using 3DNOW\n");
2335                 else if(flags & SWS_CPU_CAPS_MMX)
2336                         MSG_INFO("using MMX\n");
2337                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2338                         MSG_INFO("using AltiVec\n");
2339                 else 
2340                         MSG_INFO("using C\n");
2341         }
2342
2343         if(flags & SWS_PRINT_INFO)
2344         {
2345                 if(flags & SWS_CPU_CAPS_MMX)
2346                 {
2347                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2348                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2349                         else
2350                         {
2351                                 if(c->hLumFilterSize==4)
2352                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2353                                 else if(c->hLumFilterSize==8)
2354                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2355                                 else
2356                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2357
2358                                 if(c->hChrFilterSize==4)
2359                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2360                                 else if(c->hChrFilterSize==8)
2361                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2362                                 else
2363                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2364                         }
2365                 }
2366                 else
2367                 {
2368 #if defined(ARCH_X86)
2369                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2370 #else
2371                         if(flags & SWS_FAST_BILINEAR)
2372                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2373                         else
2374                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2375 #endif
2376                 }
2377                 if(isPlanarYUV(dstFormat))
2378                 {
2379                         if(c->vLumFilterSize==1)
2380                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2381                         else
2382                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2383                 }
2384                 else
2385                 {
2386                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2387                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2388                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2389                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2390                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2391                         else
2392                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2393                 }
2394
2395                 if(dstFormat==PIX_FMT_BGR24)
2396                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2397                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2398                 else if(dstFormat==PIX_FMT_RGB32)
2399                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2400                 else if(dstFormat==PIX_FMT_BGR565)
2401                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2402                 else if(dstFormat==PIX_FMT_BGR555)
2403                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2404
2405                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2406         }
2407         if(flags & SWS_PRINT_INFO)
2408         {
2409                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2410                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2411                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2412                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2413         }
2414
2415         c->swScale= getSwsFunc(flags);
2416         return c;
2417 }
2418
2419 /**
2420  * swscale warper, so we don't need to export the SwsContext.
2421  * assumes planar YUV to be in YUV order instead of YVU
2422  */
2423 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2424                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2425         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2426             MSG_ERR("swScaler: slices start in the middle!\n");
2427             return 0;
2428         }
2429         if (c->sliceDir == 0) {
2430             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2431         }
2432
2433         // copy strides, so they can safely be modified
2434         if (c->sliceDir == 1) {
2435             // slices go from top to bottom
2436             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2437             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2438             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2439         } else {
2440             // slices go from bottom to top => we flip the image internally
2441             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2442                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2443                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2444             };
2445             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2446                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2447                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2448             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2449             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2450             
2451             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2452         }
2453 }
2454
2455 /**
2456  * swscale warper, so we don't need to export the SwsContext
2457  */
2458 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStride[], int srcSliceY,
2459                            int srcSliceH, uint8_t* dstParam[], int dstStride[]){
2460         uint8_t *src[3];
2461         uint8_t *dst[3];
2462         src[0] = srcParam[0]; src[1] = srcParam[1]; src[2] = srcParam[2];
2463         dst[0] = dstParam[0]; dst[1] = dstParam[1]; dst[2] = dstParam[2];
2464 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2465
2466         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2467 }
2468
2469 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2470                                 float lumaSharpen, float chromaSharpen,
2471                                 float chromaHShift, float chromaVShift,
2472                                 int verbose)
2473 {
2474         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2475
2476         if(lumaGBlur!=0.0){
2477                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2478                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2479         }else{
2480                 filter->lumH= sws_getIdentityVec();
2481                 filter->lumV= sws_getIdentityVec();
2482         }
2483
2484         if(chromaGBlur!=0.0){
2485                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2486                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2487         }else{
2488                 filter->chrH= sws_getIdentityVec();
2489                 filter->chrV= sws_getIdentityVec();
2490         }
2491
2492         if(chromaSharpen!=0.0){
2493                 SwsVector *id= sws_getIdentityVec();
2494                 sws_scaleVec(filter->chrH, -chromaSharpen);
2495                 sws_scaleVec(filter->chrV, -chromaSharpen);
2496                 sws_addVec(filter->chrH, id);
2497                 sws_addVec(filter->chrV, id);
2498                 sws_freeVec(id);
2499         }
2500
2501         if(lumaSharpen!=0.0){
2502                 SwsVector *id= sws_getIdentityVec();
2503                 sws_scaleVec(filter->lumH, -lumaSharpen);
2504                 sws_scaleVec(filter->lumV, -lumaSharpen);
2505                 sws_addVec(filter->lumH, id);
2506                 sws_addVec(filter->lumV, id);
2507                 sws_freeVec(id);
2508         }
2509
2510         if(chromaHShift != 0.0)
2511                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2512
2513         if(chromaVShift != 0.0)
2514                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2515
2516         sws_normalizeVec(filter->chrH, 1.0);
2517         sws_normalizeVec(filter->chrV, 1.0);
2518         sws_normalizeVec(filter->lumH, 1.0);
2519         sws_normalizeVec(filter->lumV, 1.0);
2520
2521         if(verbose) sws_printVec(filter->chrH);
2522         if(verbose) sws_printVec(filter->lumH);
2523
2524         return filter;
2525 }
2526
2527 /**
2528  * returns a normalized gaussian curve used to filter stuff
2529  * quality=3 is high quality, lowwer is lowwer quality
2530  */
2531 SwsVector *sws_getGaussianVec(double variance, double quality){
2532         const int length= (int)(variance*quality + 0.5) | 1;
2533         int i;
2534         double *coeff= av_malloc(length*sizeof(double));
2535         double middle= (length-1)*0.5;
2536         SwsVector *vec= av_malloc(sizeof(SwsVector));
2537
2538         vec->coeff= coeff;
2539         vec->length= length;
2540
2541         for(i=0; i<length; i++)
2542         {
2543                 double dist= i-middle;
2544                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2545         }
2546
2547         sws_normalizeVec(vec, 1.0);
2548
2549         return vec;
2550 }
2551
2552 SwsVector *sws_getConstVec(double c, int length){
2553         int i;
2554         double *coeff= av_malloc(length*sizeof(double));
2555         SwsVector *vec= av_malloc(sizeof(SwsVector));
2556
2557         vec->coeff= coeff;
2558         vec->length= length;
2559
2560         for(i=0; i<length; i++)
2561                 coeff[i]= c;
2562
2563         return vec;
2564 }
2565
2566
2567 SwsVector *sws_getIdentityVec(void){
2568         return sws_getConstVec(1.0, 1);
2569 }
2570
2571 double sws_dcVec(SwsVector *a){
2572         int i;
2573         double sum=0;
2574
2575         for(i=0; i<a->length; i++)
2576                 sum+= a->coeff[i];
2577
2578         return sum;
2579 }
2580
2581 void sws_scaleVec(SwsVector *a, double scalar){
2582         int i;
2583
2584         for(i=0; i<a->length; i++)
2585                 a->coeff[i]*= scalar;
2586 }
2587
2588 void sws_normalizeVec(SwsVector *a, double height){
2589         sws_scaleVec(a, height/sws_dcVec(a));
2590 }
2591
2592 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2593         int length= a->length + b->length - 1;
2594         double *coeff= av_malloc(length*sizeof(double));
2595         int i, j;
2596         SwsVector *vec= av_malloc(sizeof(SwsVector));
2597
2598         vec->coeff= coeff;
2599         vec->length= length;
2600
2601         for(i=0; i<length; i++) coeff[i]= 0.0;
2602
2603         for(i=0; i<a->length; i++)
2604         {
2605                 for(j=0; j<b->length; j++)
2606                 {
2607                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2608                 }
2609         }
2610
2611         return vec;
2612 }
2613
2614 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2615         int length= FFMAX(a->length, b->length);
2616         double *coeff= av_malloc(length*sizeof(double));
2617         int i;
2618         SwsVector *vec= av_malloc(sizeof(SwsVector));
2619
2620         vec->coeff= coeff;
2621         vec->length= length;
2622
2623         for(i=0; i<length; i++) coeff[i]= 0.0;
2624
2625         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2626         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2627
2628         return vec;
2629 }
2630
2631 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2632         int length= FFMAX(a->length, b->length);
2633         double *coeff= av_malloc(length*sizeof(double));
2634         int i;
2635         SwsVector *vec= av_malloc(sizeof(SwsVector));
2636
2637         vec->coeff= coeff;
2638         vec->length= length;
2639
2640         for(i=0; i<length; i++) coeff[i]= 0.0;
2641
2642         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2643         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2644
2645         return vec;
2646 }
2647
2648 /* shift left / or right if "shift" is negative */
2649 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2650         int length= a->length + FFABS(shift)*2;
2651         double *coeff= av_malloc(length*sizeof(double));
2652         int i;
2653         SwsVector *vec= av_malloc(sizeof(SwsVector));
2654
2655         vec->coeff= coeff;
2656         vec->length= length;
2657
2658         for(i=0; i<length; i++) coeff[i]= 0.0;
2659
2660         for(i=0; i<a->length; i++)
2661         {
2662                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2663         }
2664
2665         return vec;
2666 }
2667
2668 void sws_shiftVec(SwsVector *a, int shift){
2669         SwsVector *shifted= sws_getShiftedVec(a, shift);
2670         av_free(a->coeff);
2671         a->coeff= shifted->coeff;
2672         a->length= shifted->length;
2673         av_free(shifted);
2674 }
2675
2676 void sws_addVec(SwsVector *a, SwsVector *b){
2677         SwsVector *sum= sws_sumVec(a, b);
2678         av_free(a->coeff);
2679         a->coeff= sum->coeff;
2680         a->length= sum->length;
2681         av_free(sum);
2682 }
2683
2684 void sws_subVec(SwsVector *a, SwsVector *b){
2685         SwsVector *diff= sws_diffVec(a, b);
2686         av_free(a->coeff);
2687         a->coeff= diff->coeff;
2688         a->length= diff->length;
2689         av_free(diff);
2690 }
2691
2692 void sws_convVec(SwsVector *a, SwsVector *b){
2693         SwsVector *conv= sws_getConvVec(a, b);
2694         av_free(a->coeff);  
2695         a->coeff= conv->coeff;
2696         a->length= conv->length;
2697         av_free(conv);
2698 }
2699
2700 SwsVector *sws_cloneVec(SwsVector *a){
2701         double *coeff= av_malloc(a->length*sizeof(double));
2702         int i;
2703         SwsVector *vec= av_malloc(sizeof(SwsVector));
2704
2705         vec->coeff= coeff;
2706         vec->length= a->length;
2707
2708         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2709
2710         return vec;
2711 }
2712
2713 void sws_printVec(SwsVector *a){
2714         int i;
2715         double max=0;
2716         double min=0;
2717         double range;
2718
2719         for(i=0; i<a->length; i++)
2720                 if(a->coeff[i]>max) max= a->coeff[i];
2721
2722         for(i=0; i<a->length; i++)
2723                 if(a->coeff[i]<min) min= a->coeff[i];
2724
2725         range= max - min;
2726
2727         for(i=0; i<a->length; i++)
2728         {
2729                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2730                 MSG_DBG2("%1.3f ", a->coeff[i]);
2731                 for(;x>0; x--) MSG_DBG2(" ");
2732                 MSG_DBG2("|\n");
2733         }
2734 }
2735
2736 void sws_freeVec(SwsVector *a){
2737         if(!a) return;
2738         av_free(a->coeff);
2739         a->coeff=NULL;
2740         a->length=0;
2741         av_free(a);
2742 }
2743
2744 void sws_freeFilter(SwsFilter *filter){
2745         if(!filter) return;
2746
2747         if(filter->lumH) sws_freeVec(filter->lumH);
2748         if(filter->lumV) sws_freeVec(filter->lumV);
2749         if(filter->chrH) sws_freeVec(filter->chrH);
2750         if(filter->chrV) sws_freeVec(filter->chrV);
2751         av_free(filter);
2752 }
2753
2754
2755 void sws_freeContext(SwsContext *c){
2756         int i;
2757         if(!c) return;
2758
2759         if(c->lumPixBuf)
2760         {
2761                 for(i=0; i<c->vLumBufSize; i++)
2762                 {
2763                         av_free(c->lumPixBuf[i]);
2764                         c->lumPixBuf[i]=NULL;
2765                 }
2766                 av_free(c->lumPixBuf);
2767                 c->lumPixBuf=NULL;
2768         }
2769
2770         if(c->chrPixBuf)
2771         {
2772                 for(i=0; i<c->vChrBufSize; i++)
2773                 {
2774                         av_free(c->chrPixBuf[i]);
2775                         c->chrPixBuf[i]=NULL;
2776                 }
2777                 av_free(c->chrPixBuf);
2778                 c->chrPixBuf=NULL;
2779         }
2780
2781         av_free(c->vLumFilter);
2782         c->vLumFilter = NULL;
2783         av_free(c->vChrFilter);
2784         c->vChrFilter = NULL;
2785         av_free(c->hLumFilter);
2786         c->hLumFilter = NULL;
2787         av_free(c->hChrFilter);
2788         c->hChrFilter = NULL;
2789 #ifdef HAVE_ALTIVEC
2790         av_free(c->vYCoeffsBank);
2791         c->vYCoeffsBank = NULL;
2792         av_free(c->vCCoeffsBank);
2793         c->vCCoeffsBank = NULL;
2794 #endif
2795
2796         av_free(c->vLumFilterPos);
2797         c->vLumFilterPos = NULL;
2798         av_free(c->vChrFilterPos);
2799         c->vChrFilterPos = NULL;
2800         av_free(c->hLumFilterPos);
2801         c->hLumFilterPos = NULL;
2802         av_free(c->hChrFilterPos);
2803         c->hChrFilterPos = NULL;
2804
2805 #if defined(ARCH_X86)
2806 #ifdef MAP_ANONYMOUS
2807         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2808         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2809 #else
2810         av_free(c->funnyYCode);
2811         av_free(c->funnyUVCode);
2812 #endif
2813         c->funnyYCode=NULL;
2814         c->funnyUVCode=NULL;
2815 #endif /* defined(ARCH_X86) */
2816
2817         av_free(c->lumMmx2Filter);
2818         c->lumMmx2Filter=NULL;
2819         av_free(c->chrMmx2Filter);
2820         c->chrMmx2Filter=NULL;
2821         av_free(c->lumMmx2FilterPos);
2822         c->lumMmx2FilterPos=NULL;
2823         av_free(c->chrMmx2FilterPos);
2824         c->chrMmx2FilterPos=NULL;
2825         av_free(c->yuvTable);
2826         c->yuvTable=NULL;
2827
2828         av_free(c);
2829 }
2830
2831 /**
2832  * Checks if context is valid or reallocs a new one instead.
2833  * If context is NULL, just calls sws_getContext() to get a new one.
2834  * Otherwise, checks if the parameters are the same already saved in context.
2835  * If that is the case, returns the current context.
2836  * Otherwise, frees context and gets a new one.
2837  *
2838  * Be warned that srcFilter, dstFilter are not checked, they are
2839  * asumed to remain valid.
2840  */
2841 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2842                         int srcW, int srcH, int srcFormat,
2843                         int dstW, int dstH, int dstFormat, int flags,
2844                         SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2845 {
2846     if (context != NULL) {
2847         if ((context->srcW != srcW) || (context->srcH != srcH) ||
2848             (context->srcFormat != srcFormat) ||
2849             (context->dstW != dstW) || (context->dstH != dstH) ||
2850             (context->dstFormat != dstFormat) || (context->flags != flags) ||
2851             (context->param != param))
2852         {
2853             sws_freeContext(context);
2854             context = NULL;
2855         }
2856     }
2857     if (context == NULL) {
2858         return sws_getContext(srcW, srcH, srcFormat,
2859                         dstW, dstH, dstFormat, flags,
2860                         srcFilter, dstFilter, param);
2861     }
2862     return context;
2863 }
2864