AltiVec support in postproc/ + altivec optimizations for yuv2yuvX patch by (Romain...
[ffmpeg.git] / postproc / swscale.c
1 /*
2     Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3
4     This program is free software; you can redistribute it and/or modify
5     it under the terms of the GNU General Public License as published by
6     the Free Software Foundation; either version 2 of the License, or
7     (at your option) any later version.
8
9     This program is distributed in the hope that it will be useful,
10     but WITHOUT ANY WARRANTY; without even the implied warranty of
11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12     GNU General Public License for more details.
13
14     You should have received a copy of the GNU General Public License
15     along with this program; if not, write to the Free Software
16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17 */
18
19 /*
20   supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09
21   supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
22   {BGR,RGB}{1,4,8,15,16} support dithering
23   
24   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
25   YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
26   x -> x
27   YUV9 -> YV12
28   YUV9/YV12 -> Y800
29   Y800 -> YUV9/YV12
30   BGR24 -> BGR32 & RGB24 -> RGB32
31   BGR32 -> BGR24 & RGB32 -> RGB24
32   BGR15 -> BGR16
33 */
34
35 /* 
36 tested special converters (most are tested actually but i didnt write it down ...)
37  YV12 -> BGR16
38  YV12 -> YV12
39  BGR15 -> BGR16
40  BGR16 -> BGR16
41  YVU9 -> YV12
42
43 untested special converters
44   YV12/I420 -> BGR15/BGR24/BGR32 (its the yuv2rgb stuff, so it should be ok)
45   YV12/I420 -> YV12/I420
46   YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
47   BGR24 -> BGR32 & RGB24 -> RGB32
48   BGR32 -> BGR24 & RGB32 -> RGB24
49   BGR24 -> YV12
50 */
51
52 #include <inttypes.h>
53 #include <string.h>
54 #include <math.h>
55 #include <stdio.h>
56 #include "../config.h"
57 #include "../mangle.h"
58 #include <assert.h>
59 #ifdef HAVE_MALLOC_H
60 #include <malloc.h>
61 #else
62 #include <stdlib.h>
63 #endif
64 #include "swscale.h"
65 #include "swscale_internal.h"
66 #include "../cpudetect.h"
67 #include "../bswap.h"
68 #include "../libvo/img_format.h"
69 #include "rgb2rgb.h"
70 #include "../libvo/fastmemcpy.h"
71
72 #undef MOVNTQ
73 #undef PAVGB
74
75 //#undef HAVE_MMX2
76 //#define HAVE_3DNOW
77 //#undef HAVE_MMX
78 //#undef ARCH_X86
79 //#define WORDS_BIGENDIAN
80 #define DITHER1XBPP
81
82 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
83
84 #define RET 0xC3 //near return opcode for X86
85
86 #ifdef MP_DEBUG
87 #define ASSERT(x) assert(x);
88 #else
89 #define ASSERT(x) ;
90 #endif
91
92 #ifdef M_PI
93 #define PI M_PI
94 #else
95 #define PI 3.14159265358979323846
96 #endif
97
98 //FIXME replace this with something faster
99 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
100                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
101 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
102 #define isGray(x)      ((x)==IMGFMT_Y800)
103 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
104 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
105 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
106                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
107                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
108                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
109                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
110 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
111                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
112                         || isRGB(x) || isBGR(x)\
113                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
114 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
115
116 #define RGB2YUV_SHIFT 16
117 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
118 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
119 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
120 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
121 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
122 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
123 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
124 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
125 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
126
127 extern const int32_t Inverse_Table_6_9[8][4];
128
129 /*
130 NOTES
131 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
132
133 TODO
134 more intelligent missalignment avoidance for the horizontal scaler
135 write special vertical cubic upscale version
136 Optimize C code (yv12 / minmax)
137 add support for packed pixel yuv input & output
138 add support for Y8 output
139 optimize bgr24 & bgr32
140 add BGR4 output support
141 write special BGR->BGR scaler
142 */
143
144 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
145 #define MIN(a,b) ((a) > (b) ? (b) : (a))
146 #define MAX(a,b) ((a) < (b) ? (b) : (a))
147
148 #ifdef ARCH_X86
149 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
150 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
151 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
152 static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;
153 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
154 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
155 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
156 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
157
158 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
159 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
160 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
161 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
162
163 static uint64_t __attribute__((aligned(8))) dither4[2]={
164         0x0103010301030103LL,
165         0x0200020002000200LL,};
166
167 static uint64_t __attribute__((aligned(8))) dither8[2]={
168         0x0602060206020602LL,
169         0x0004000400040004LL,};
170
171 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
172 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
173 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
174 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
175 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
176 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
177
178 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
179 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
180 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
181
182 #ifdef FAST_BGR2YV12
183 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
184 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
185 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
186 #else
187 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
188 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
189 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
190 #endif
191 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
192 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
193 static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
194 #endif
195
196 // clipping helper table for C implementations:
197 static unsigned char clip_table[768];
198
199 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
200                   
201 extern const uint8_t dither_2x2_4[2][8];
202 extern const uint8_t dither_2x2_8[2][8];
203 extern const uint8_t dither_8x8_32[8][8];
204 extern const uint8_t dither_8x8_73[8][8];
205 extern const uint8_t dither_8x8_220[8][8];
206
207 #ifdef ARCH_X86
208 void in_asm_used_var_warning_killer()
209 {
210  volatile int i= bF8+bFC+w10+
211  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
212  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
213  if(i) i=0;
214 }
215 #endif
216
217 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
218                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
219                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
220 {
221         //FIXME Optimize (just quickly writen not opti..)
222         int i;
223         for(i=0; i<dstW; i++)
224         {
225                 int val=1<<18;
226                 int j;
227                 for(j=0; j<lumFilterSize; j++)
228                         val += lumSrc[j][i] * lumFilter[j];
229
230                 dest[i]= MIN(MAX(val>>19, 0), 255);
231         }
232
233         if(uDest != NULL)
234                 for(i=0; i<chrDstW; i++)
235                 {
236                         int u=1<<18;
237                         int v=1<<18;
238                         int j;
239                         for(j=0; j<chrFilterSize; j++)
240                         {
241                                 u += chrSrc[j][i] * chrFilter[j];
242                                 v += chrSrc[j][i + 2048] * chrFilter[j];
243                         }
244
245                         uDest[i]= MIN(MAX(u>>19, 0), 255);
246                         vDest[i]= MIN(MAX(v>>19, 0), 255);
247                 }
248 }
249
250
251 #define YSCALE_YUV_2_PACKEDX_C(type) \
252                 for(i=0; i<(dstW>>1); i++){\
253                         int j;\
254                         int Y1=1<<18;\
255                         int Y2=1<<18;\
256                         int U=1<<18;\
257                         int V=1<<18;\
258                         type *r, *b, *g;\
259                         const int i2= 2*i;\
260                         \
261                         for(j=0; j<lumFilterSize; j++)\
262                         {\
263                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
264                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
265                         }\
266                         for(j=0; j<chrFilterSize; j++)\
267                         {\
268                                 U += chrSrc[j][i] * chrFilter[j];\
269                                 V += chrSrc[j][i+2048] * chrFilter[j];\
270                         }\
271                         Y1>>=19;\
272                         Y2>>=19;\
273                         U >>=19;\
274                         V >>=19;\
275                         if((Y1|Y2|U|V)&256)\
276                         {\
277                                 if(Y1>255)   Y1=255;\
278                                 else if(Y1<0)Y1=0;\
279                                 if(Y2>255)   Y2=255;\
280                                 else if(Y2<0)Y2=0;\
281                                 if(U>255)    U=255;\
282                                 else if(U<0) U=0;\
283                                 if(V>255)    V=255;\
284                                 else if(V<0) V=0;\
285                         }
286                         
287 #define YSCALE_YUV_2_RGBX_C(type) \
288                         YSCALE_YUV_2_PACKEDX_C(type)\
289                         r = c->table_rV[V];\
290                         g = c->table_gU[U] + c->table_gV[V];\
291                         b = c->table_bU[U];\
292
293 #define YSCALE_YUV_2_PACKED2_C \
294                 for(i=0; i<(dstW>>1); i++){\
295                         const int i2= 2*i;\
296                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
297                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
298                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
299                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
300
301 #define YSCALE_YUV_2_RGB2_C(type) \
302                         YSCALE_YUV_2_PACKED2_C\
303                         type *r, *b, *g;\
304                         r = c->table_rV[V];\
305                         g = c->table_gU[U] + c->table_gV[V];\
306                         b = c->table_bU[U];\
307
308 #define YSCALE_YUV_2_PACKED1_C \
309                 for(i=0; i<(dstW>>1); i++){\
310                         const int i2= 2*i;\
311                         int Y1= buf0[i2  ]>>7;\
312                         int Y2= buf0[i2+1]>>7;\
313                         int U= (uvbuf1[i     ])>>7;\
314                         int V= (uvbuf1[i+2048])>>7;\
315
316 #define YSCALE_YUV_2_RGB1_C(type) \
317                         YSCALE_YUV_2_PACKED1_C\
318                         type *r, *b, *g;\
319                         r = c->table_rV[V];\
320                         g = c->table_gU[U] + c->table_gV[V];\
321                         b = c->table_bU[U];\
322
323 #define YSCALE_YUV_2_PACKED1B_C \
324                 for(i=0; i<(dstW>>1); i++){\
325                         const int i2= 2*i;\
326                         int Y1= buf0[i2  ]>>7;\
327                         int Y2= buf0[i2+1]>>7;\
328                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
329                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
330
331 #define YSCALE_YUV_2_RGB1B_C(type) \
332                         YSCALE_YUV_2_PACKED1B_C\
333                         type *r, *b, *g;\
334                         r = c->table_rV[V];\
335                         g = c->table_gU[U] + c->table_gV[V];\
336                         b = c->table_bU[U];\
337
338 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
339         switch(c->dstFormat)\
340         {\
341         case IMGFMT_BGR32:\
342         case IMGFMT_RGB32:\
343                 func(uint32_t)\
344                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
345                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
346                 }               \
347                 break;\
348         case IMGFMT_RGB24:\
349                 func(uint8_t)\
350                         ((uint8_t*)dest)[0]= r[Y1];\
351                         ((uint8_t*)dest)[1]= g[Y1];\
352                         ((uint8_t*)dest)[2]= b[Y1];\
353                         ((uint8_t*)dest)[3]= r[Y2];\
354                         ((uint8_t*)dest)[4]= g[Y2];\
355                         ((uint8_t*)dest)[5]= b[Y2];\
356                         ((uint8_t*)dest)+=6;\
357                 }\
358                 break;\
359         case IMGFMT_BGR24:\
360                 func(uint8_t)\
361                         ((uint8_t*)dest)[0]= b[Y1];\
362                         ((uint8_t*)dest)[1]= g[Y1];\
363                         ((uint8_t*)dest)[2]= r[Y1];\
364                         ((uint8_t*)dest)[3]= b[Y2];\
365                         ((uint8_t*)dest)[4]= g[Y2];\
366                         ((uint8_t*)dest)[5]= r[Y2];\
367                         ((uint8_t*)dest)+=6;\
368                 }\
369                 break;\
370         case IMGFMT_RGB16:\
371         case IMGFMT_BGR16:\
372                 {\
373                         const int dr1= dither_2x2_8[y&1    ][0];\
374                         const int dg1= dither_2x2_4[y&1    ][0];\
375                         const int db1= dither_2x2_8[(y&1)^1][0];\
376                         const int dr2= dither_2x2_8[y&1    ][1];\
377                         const int dg2= dither_2x2_4[y&1    ][1];\
378                         const int db2= dither_2x2_8[(y&1)^1][1];\
379                         func(uint16_t)\
380                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
381                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
382                         }\
383                 }\
384                 break;\
385         case IMGFMT_RGB15:\
386         case IMGFMT_BGR15:\
387                 {\
388                         const int dr1= dither_2x2_8[y&1    ][0];\
389                         const int dg1= dither_2x2_8[y&1    ][1];\
390                         const int db1= dither_2x2_8[(y&1)^1][0];\
391                         const int dr2= dither_2x2_8[y&1    ][1];\
392                         const int dg2= dither_2x2_8[y&1    ][0];\
393                         const int db2= dither_2x2_8[(y&1)^1][1];\
394                         func(uint16_t)\
395                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
396                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
397                         }\
398                 }\
399                 break;\
400         case IMGFMT_RGB8:\
401         case IMGFMT_BGR8:\
402                 {\
403                         const uint8_t * const d64= dither_8x8_73[y&7];\
404                         const uint8_t * const d32= dither_8x8_32[y&7];\
405                         func(uint8_t)\
406                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
407                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
408                         }\
409                 }\
410                 break;\
411         case IMGFMT_RGB4:\
412         case IMGFMT_BGR4:\
413                 {\
414                         const uint8_t * const d64= dither_8x8_73 [y&7];\
415                         const uint8_t * const d128=dither_8x8_220[y&7];\
416                         func(uint8_t)\
417                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
418                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
419                         }\
420                 }\
421                 break;\
422         case IMGFMT_RG4B:\
423         case IMGFMT_BG4B:\
424                 {\
425                         const uint8_t * const d64= dither_8x8_73 [y&7];\
426                         const uint8_t * const d128=dither_8x8_220[y&7];\
427                         func(uint8_t)\
428                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
429                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
430                         }\
431                 }\
432                 break;\
433         case IMGFMT_RGB1:\
434         case IMGFMT_BGR1:\
435                 {\
436                         const uint8_t * const d128=dither_8x8_220[y&7];\
437                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
438                         for(i=0; i<dstW-7; i+=8){\
439                                 int acc;\
440                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
441                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
442                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
443                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
444                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
445                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
446                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
447                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
448                                 ((uint8_t*)dest)[0]= acc;\
449                                 ((uint8_t*)dest)++;\
450                         }\
451 \
452 /*\
453 ((uint8_t*)dest)-= dstW>>4;\
454 {\
455                         int acc=0;\
456                         int left=0;\
457                         static int top[1024];\
458                         static int last_new[1024][1024];\
459                         static int last_in3[1024][1024];\
460                         static int drift[1024][1024];\
461                         int topLeft=0;\
462                         int shift=0;\
463                         int count=0;\
464                         const uint8_t * const d128=dither_8x8_220[y&7];\
465                         int error_new=0;\
466                         int error_in3=0;\
467                         int f=0;\
468                         \
469                         for(i=dstW>>1; i<dstW; i++){\
470                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
471                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
472                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
473                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
474                                         + (last_new[y][i] - in3)*f/256;\
475                                 int new= old> 128 ? 255 : 0;\
476 \
477                                 error_new+= ABS(last_new[y][i] - new);\
478                                 error_in3+= ABS(last_in3[y][i] - in3);\
479                                 f= error_new - error_in3*4;\
480                                 if(f<0) f=0;\
481                                 if(f>256) f=256;\
482 \
483                                 topLeft= top[i];\
484                                 left= top[i]= old - new;\
485                                 last_new[y][i]= new;\
486                                 last_in3[y][i]= in3;\
487 \
488                                 acc+= acc + (new&1);\
489                                 if((i&7)==6){\
490                                         ((uint8_t*)dest)[0]= acc;\
491                                         ((uint8_t*)dest)++;\
492                                 }\
493                         }\
494 }\
495 */\
496                 }\
497                 break;\
498         case IMGFMT_YUY2:\
499                 func2\
500                         ((uint8_t*)dest)[2*i2+0]= Y1;\
501                         ((uint8_t*)dest)[2*i2+1]= U;\
502                         ((uint8_t*)dest)[2*i2+2]= Y2;\
503                         ((uint8_t*)dest)[2*i2+3]= V;\
504                 }               \
505                 break;\
506         case IMGFMT_UYVY:\
507                 func2\
508                         ((uint8_t*)dest)[2*i2+0]= U;\
509                         ((uint8_t*)dest)[2*i2+1]= Y1;\
510                         ((uint8_t*)dest)[2*i2+2]= V;\
511                         ((uint8_t*)dest)[2*i2+3]= Y2;\
512                 }               \
513                 break;\
514         }\
515
516
517 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
518                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
519                                     uint8_t *dest, int dstW, int y)
520 {
521         int i;
522         switch(c->dstFormat)
523         {
524         case IMGFMT_RGB32:
525         case IMGFMT_BGR32:
526                 YSCALE_YUV_2_RGBX_C(uint32_t)
527                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
528                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
529                 }
530                 break;
531         case IMGFMT_RGB24:
532                 YSCALE_YUV_2_RGBX_C(uint8_t)
533                         ((uint8_t*)dest)[0]= r[Y1];
534                         ((uint8_t*)dest)[1]= g[Y1];
535                         ((uint8_t*)dest)[2]= b[Y1];
536                         ((uint8_t*)dest)[3]= r[Y2];
537                         ((uint8_t*)dest)[4]= g[Y2];
538                         ((uint8_t*)dest)[5]= b[Y2];
539                         ((uint8_t*)dest)+=6;
540                 }
541                 break;
542         case IMGFMT_BGR24:
543                 YSCALE_YUV_2_RGBX_C(uint8_t)
544                         ((uint8_t*)dest)[0]= b[Y1];
545                         ((uint8_t*)dest)[1]= g[Y1];
546                         ((uint8_t*)dest)[2]= r[Y1];
547                         ((uint8_t*)dest)[3]= b[Y2];
548                         ((uint8_t*)dest)[4]= g[Y2];
549                         ((uint8_t*)dest)[5]= r[Y2];
550                         ((uint8_t*)dest)+=6;
551                 }
552                 break;
553         case IMGFMT_RGB16:
554         case IMGFMT_BGR16:
555                 {
556                         const int dr1= dither_2x2_8[y&1    ][0];
557                         const int dg1= dither_2x2_4[y&1    ][0];
558                         const int db1= dither_2x2_8[(y&1)^1][0];
559                         const int dr2= dither_2x2_8[y&1    ][1];
560                         const int dg2= dither_2x2_4[y&1    ][1];
561                         const int db2= dither_2x2_8[(y&1)^1][1];
562                         YSCALE_YUV_2_RGBX_C(uint16_t)
563                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
564                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
565                         }
566                 }
567                 break;
568         case IMGFMT_RGB15:
569         case IMGFMT_BGR15:
570                 {
571                         const int dr1= dither_2x2_8[y&1    ][0];
572                         const int dg1= dither_2x2_8[y&1    ][1];
573                         const int db1= dither_2x2_8[(y&1)^1][0];
574                         const int dr2= dither_2x2_8[y&1    ][1];
575                         const int dg2= dither_2x2_8[y&1    ][0];
576                         const int db2= dither_2x2_8[(y&1)^1][1];
577                         YSCALE_YUV_2_RGBX_C(uint16_t)
578                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
579                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
580                         }
581                 }
582                 break;
583         case IMGFMT_RGB8:
584         case IMGFMT_BGR8:
585                 {
586                         const uint8_t * const d64= dither_8x8_73[y&7];
587                         const uint8_t * const d32= dither_8x8_32[y&7];
588                         YSCALE_YUV_2_RGBX_C(uint8_t)
589                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
590                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
591                         }
592                 }
593                 break;
594         case IMGFMT_RGB4:
595         case IMGFMT_BGR4:
596                 {
597                         const uint8_t * const d64= dither_8x8_73 [y&7];
598                         const uint8_t * const d128=dither_8x8_220[y&7];
599                         YSCALE_YUV_2_RGBX_C(uint8_t)
600                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
601                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
602                         }
603                 }
604                 break;
605         case IMGFMT_RG4B:
606         case IMGFMT_BG4B:
607                 {
608                         const uint8_t * const d64= dither_8x8_73 [y&7];
609                         const uint8_t * const d128=dither_8x8_220[y&7];
610                         YSCALE_YUV_2_RGBX_C(uint8_t)
611                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
612                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
613                         }
614                 }
615                 break;
616         case IMGFMT_RGB1:
617         case IMGFMT_BGR1:
618                 {
619                         const uint8_t * const d128=dither_8x8_220[y&7];
620                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
621                         int acc=0;
622                         for(i=0; i<dstW-1; i+=2){
623                                 int j;
624                                 int Y1=1<<18;
625                                 int Y2=1<<18;
626
627                                 for(j=0; j<lumFilterSize; j++)
628                                 {
629                                         Y1 += lumSrc[j][i] * lumFilter[j];
630                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
631                                 }
632                                 Y1>>=19;
633                                 Y2>>=19;
634                                 if((Y1|Y2)&256)
635                                 {
636                                         if(Y1>255)   Y1=255;
637                                         else if(Y1<0)Y1=0;
638                                         if(Y2>255)   Y2=255;
639                                         else if(Y2<0)Y2=0;
640                                 }
641                                 acc+= acc + g[Y1+d128[(i+0)&7]];
642                                 acc+= acc + g[Y2+d128[(i+1)&7]];
643                                 if((i&7)==6){
644                                         ((uint8_t*)dest)[0]= acc;
645                                         ((uint8_t*)dest)++;
646                                 }
647                         }
648                 }
649                 break;
650         case IMGFMT_YUY2:
651                 YSCALE_YUV_2_PACKEDX_C(void)
652                         ((uint8_t*)dest)[2*i2+0]= Y1;
653                         ((uint8_t*)dest)[2*i2+1]= U;
654                         ((uint8_t*)dest)[2*i2+2]= Y2;
655                         ((uint8_t*)dest)[2*i2+3]= V;
656                 }
657                 break;
658         case IMGFMT_UYVY:
659                 YSCALE_YUV_2_PACKEDX_C(void)
660                         ((uint8_t*)dest)[2*i2+0]= U;
661                         ((uint8_t*)dest)[2*i2+1]= Y1;
662                         ((uint8_t*)dest)[2*i2+2]= V;
663                         ((uint8_t*)dest)[2*i2+3]= Y2;
664                 }
665                 break;
666         }
667 }
668
669
670 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
671 //Plain C versions
672 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
673 #define COMPILE_C
674 #endif
675
676 #ifdef ARCH_POWERPC
677 #ifdef HAVE_ALTIVEC
678 #define COMPILE_ALTIVEC
679 #endif //HAVE_ALTIVEC
680 #endif //ARCH_POWERPC
681
682 #ifdef ARCH_X86
683
684 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
685 #define COMPILE_MMX
686 #endif
687
688 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
689 #define COMPILE_MMX2
690 #endif
691
692 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
693 #define COMPILE_3DNOW
694 #endif
695 #endif //ARCH_X86
696
697 #undef HAVE_MMX
698 #undef HAVE_MMX2
699 #undef HAVE_3DNOW
700
701 #ifdef COMPILE_C
702 #undef HAVE_MMX
703 #undef HAVE_MMX2
704 #undef HAVE_3DNOW
705 #undef HAVE_ALTIVEC
706 #define RENAME(a) a ## _C
707 #include "swscale_template.c"
708 #endif
709
710 #ifdef ARCH_POWERPC
711 #ifdef COMPILE_ALTIVEC
712 #undef RENAME
713 #define HAVE_ALTIVEC
714 #define RENAME(a) a ## _altivec
715 #include "swscale_template.c"
716 #endif
717 #endif //ARCH_POWERPC
718
719 #ifdef ARCH_X86
720
721 //X86 versions
722 /*
723 #undef RENAME
724 #undef HAVE_MMX
725 #undef HAVE_MMX2
726 #undef HAVE_3DNOW
727 #define ARCH_X86
728 #define RENAME(a) a ## _X86
729 #include "swscale_template.c"
730 */
731 //MMX versions
732 #ifdef COMPILE_MMX
733 #undef RENAME
734 #define HAVE_MMX
735 #undef HAVE_MMX2
736 #undef HAVE_3DNOW
737 #define RENAME(a) a ## _MMX
738 #include "swscale_template.c"
739 #endif
740
741 //MMX2 versions
742 #ifdef COMPILE_MMX2
743 #undef RENAME
744 #define HAVE_MMX
745 #define HAVE_MMX2
746 #undef HAVE_3DNOW
747 #define RENAME(a) a ## _MMX2
748 #include "swscale_template.c"
749 #endif
750
751 //3DNOW versions
752 #ifdef COMPILE_3DNOW
753 #undef RENAME
754 #define HAVE_MMX
755 #undef HAVE_MMX2
756 #define HAVE_3DNOW
757 #define RENAME(a) a ## _3DNow
758 #include "swscale_template.c"
759 #endif
760
761 #endif //ARCH_X86
762
763 // minor note: the HAVE_xyz is messed up after that line so don't use it
764
765 static double getSplineCoeff(double a, double b, double c, double d, double dist)
766 {
767 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
768         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
769         else            return getSplineCoeff(  0.0, 
770                                                  b+ 2.0*c + 3.0*d,
771                                                         c + 3.0*d,
772                                                 -b- 3.0*c - 6.0*d,
773                                                 dist-1.0);
774 }
775
776 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
777                               int srcW, int dstW, int filterAlign, int one, int flags,
778                               SwsVector *srcFilter, SwsVector *dstFilter)
779 {
780         int i;
781         int filterSize;
782         int filter2Size;
783         int minFilterSize;
784         double *filter=NULL;
785         double *filter2=NULL;
786 #ifdef ARCH_X86
787         if(flags & SWS_CPU_CAPS_MMX)
788                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
789 #endif
790
791         // Note the +1 is for the MMXscaler which reads over the end
792         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
793
794         if(ABS(xInc - 0x10000) <10) // unscaled
795         {
796                 int i;
797                 filterSize= 1;
798                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
799                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
800
801                 for(i=0; i<dstW; i++)
802                 {
803                         filter[i*filterSize]=1;
804                         (*filterPos)[i]=i;
805                 }
806
807         }
808         else if(flags&SWS_POINT) // lame looking point sampling mode
809         {
810                 int i;
811                 int xDstInSrc;
812                 filterSize= 1;
813                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
814                 
815                 xDstInSrc= xInc/2 - 0x8000;
816                 for(i=0; i<dstW; i++)
817                 {
818                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
819
820                         (*filterPos)[i]= xx;
821                         filter[i]= 1.0;
822                         xDstInSrc+= xInc;
823                 }
824         }
825         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
826         {
827                 int i;
828                 int xDstInSrc;
829                 if     (flags&SWS_BICUBIC) filterSize= 4;
830                 else if(flags&SWS_X      ) filterSize= 4;
831                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
832                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
833
834                 xDstInSrc= xInc/2 - 0x8000;
835                 for(i=0; i<dstW; i++)
836                 {
837                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
838                         int j;
839
840                         (*filterPos)[i]= xx;
841                                 //Bilinear upscale / linear interpolate / Area averaging
842                                 for(j=0; j<filterSize; j++)
843                                 {
844                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
845                                         double coeff= 1.0 - d;
846                                         if(coeff<0) coeff=0;
847                                         filter[i*filterSize + j]= coeff;
848                                         xx++;
849                                 }
850                         xDstInSrc+= xInc;
851                 }
852         }
853         else
854         {
855                 double xDstInSrc;
856                 double sizeFactor, filterSizeInSrc;
857                 const double xInc1= (double)xInc / (double)(1<<16);
858                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
859
860                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
861                 else if(flags&SWS_X)            sizeFactor= 8.0;
862                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
863                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
864                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
865                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
866                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
867                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
868                 else {
869                         sizeFactor= 0.0; //GCC warning killer
870                         ASSERT(0)
871                 }
872                 
873                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
874                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
875
876                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
877                 if(filterSize > srcW-2) filterSize=srcW-2;
878
879                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
880
881                 xDstInSrc= xInc1 / 2.0 - 0.5;
882                 for(i=0; i<dstW; i++)
883                 {
884                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
885                         int j;
886                         (*filterPos)[i]= xx;
887                         for(j=0; j<filterSize; j++)
888                         {
889                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
890                                 double coeff;
891                                 if(flags & SWS_BICUBIC)
892                                 {
893                                         double A= param ? -param*0.01 : -0.60;
894                                         
895                                         // Equation is from VirtualDub
896                                         if(d<1.0)
897                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
898                                         else if(d<2.0)
899                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
900                                         else
901                                                 coeff=0.0;
902                                 }
903 /*                              else if(flags & SWS_X)
904                                 {
905                                         double p= param ? param*0.01 : 0.3;
906                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
907                                         coeff*= pow(2.0, - p*d*d);
908                                 }*/
909                                 else if(flags & SWS_X)
910                                 {
911                                         double A= param ? param*0.1 : 1.0;
912                                         
913                                         if(d<1.0)
914                                                 coeff = cos(d*PI);
915                                         else
916                                                 coeff=-1.0;
917                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
918                                         else            coeff=  pow( coeff, A);
919                                         coeff= coeff*0.5 + 0.5;
920                                 }
921                                 else if(flags & SWS_AREA)
922                                 {
923                                         double srcPixelSize= 1.0/xInc1;
924                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
925                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
926                                         else coeff=0.0;
927                                 }
928                                 else if(flags & SWS_GAUSS)
929                                 {
930                                         double p= param ? param*0.1 : 3.0;
931                                         coeff = pow(2.0, - p*d*d);
932                                 }
933                                 else if(flags & SWS_SINC)
934                                 {
935                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
936                                 }
937                                 else if(flags & SWS_LANCZOS)
938                                 {
939                                         double p= param ? param : 3.0; 
940                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
941                                         if(d>p) coeff=0;
942                                 }
943                                 else if(flags & SWS_BILINEAR)
944                                 {
945                                         coeff= 1.0 - d;
946                                         if(coeff<0) coeff=0;
947                                 }
948                                 else if(flags & SWS_SPLINE)
949                                 {
950                                         double p=-2.196152422706632;
951                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
952                                 }
953                                 else {
954                                         coeff= 0.0; //GCC warning killer
955                                         ASSERT(0)
956                                 }
957
958                                 filter[i*filterSize + j]= coeff;
959                                 xx++;
960                         }
961                         xDstInSrc+= xInc1;
962                 }
963         }
964
965         /* apply src & dst Filter to filter -> filter2
966            free(filter);
967         */
968         ASSERT(filterSize>0)
969         filter2Size= filterSize;
970         if(srcFilter) filter2Size+= srcFilter->length - 1;
971         if(dstFilter) filter2Size+= dstFilter->length - 1;
972         ASSERT(filter2Size>0)
973         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
974
975         for(i=0; i<dstW; i++)
976         {
977                 int j;
978                 SwsVector scaleFilter;
979                 SwsVector *outVec;
980
981                 scaleFilter.coeff= filter + i*filterSize;
982                 scaleFilter.length= filterSize;
983
984                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
985                 else          outVec= &scaleFilter;
986
987                 ASSERT(outVec->length == filter2Size)
988                 //FIXME dstFilter
989
990                 for(j=0; j<outVec->length; j++)
991                 {
992                         filter2[i*filter2Size + j]= outVec->coeff[j];
993                 }
994
995                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
996
997                 if(outVec != &scaleFilter) sws_freeVec(outVec);
998         }
999         free(filter); filter=NULL;
1000
1001         /* try to reduce the filter-size (step1 find size and shift left) */
1002         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1003         minFilterSize= 0;
1004         for(i=dstW-1; i>=0; i--)
1005         {
1006                 int min= filter2Size;
1007                 int j;
1008                 double cutOff=0.0;
1009
1010                 /* get rid off near zero elements on the left by shifting left */
1011                 for(j=0; j<filter2Size; j++)
1012                 {
1013                         int k;
1014                         cutOff += ABS(filter2[i*filter2Size]);
1015
1016                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1017
1018                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1019                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1020
1021                         // Move filter coeffs left
1022                         for(k=1; k<filter2Size; k++)
1023                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1024                         filter2[i*filter2Size + k - 1]= 0.0;
1025                         (*filterPos)[i]++;
1026                 }
1027
1028                 cutOff=0.0;
1029                 /* count near zeros on the right */
1030                 for(j=filter2Size-1; j>0; j--)
1031                 {
1032                         cutOff += ABS(filter2[i*filter2Size + j]);
1033
1034                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1035                         min--;
1036                 }
1037
1038                 if(min>minFilterSize) minFilterSize= min;
1039         }
1040
1041         ASSERT(minFilterSize > 0)
1042         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1043         ASSERT(filterSize > 0)
1044         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1045         *outFilterSize= filterSize;
1046
1047         if(flags&SWS_PRINT_INFO)
1048                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1049         /* try to reduce the filter-size (step2 reduce it) */
1050         for(i=0; i<dstW; i++)
1051         {
1052                 int j;
1053
1054                 for(j=0; j<filterSize; j++)
1055                 {
1056                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1057                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1058                 }
1059         }
1060         free(filter2); filter2=NULL;
1061         
1062
1063         //FIXME try to align filterpos if possible
1064
1065         //fix borders
1066         for(i=0; i<dstW; i++)
1067         {
1068                 int j;
1069                 if((*filterPos)[i] < 0)
1070                 {
1071                         // Move filter coeffs left to compensate for filterPos
1072                         for(j=1; j<filterSize; j++)
1073                         {
1074                                 int left= MAX(j + (*filterPos)[i], 0);
1075                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1076                                 filter[i*filterSize + j]=0;
1077                         }
1078                         (*filterPos)[i]= 0;
1079                 }
1080
1081                 if((*filterPos)[i] + filterSize > srcW)
1082                 {
1083                         int shift= (*filterPos)[i] + filterSize - srcW;
1084                         // Move filter coeffs right to compensate for filterPos
1085                         for(j=filterSize-2; j>=0; j--)
1086                         {
1087                                 int right= MIN(j + shift, filterSize-1);
1088                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1089                                 filter[i*filterSize +j]=0;
1090                         }
1091                         (*filterPos)[i]= srcW - filterSize;
1092                 }
1093         }
1094
1095         // Note the +1 is for the MMXscaler which reads over the end
1096         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1097         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1098
1099         /* Normalize & Store in outFilter */
1100         for(i=0; i<dstW; i++)
1101         {
1102                 int j;
1103                 double error=0;
1104                 double sum=0;
1105                 double scale= one;
1106
1107                 for(j=0; j<filterSize; j++)
1108                 {
1109                         sum+= filter[i*filterSize + j];
1110                 }
1111                 scale/= sum;
1112                 for(j=0; j<*outFilterSize; j++)
1113                 {
1114                         double v= filter[i*filterSize + j]*scale + error;
1115                         int intV= floor(v + 0.5);
1116                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1117                         error = v - intV;
1118                 }
1119         }
1120         
1121         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1122         for(i=0; i<*outFilterSize; i++)
1123         {
1124                 int j= dstW*(*outFilterSize);
1125                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1126         }
1127
1128         free(filter);
1129 }
1130
1131 #ifdef ARCH_X86
1132 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1133 {
1134         uint8_t *fragmentA;
1135         int imm8OfPShufW1A;
1136         int imm8OfPShufW2A;
1137         int fragmentLengthA;
1138         uint8_t *fragmentB;
1139         int imm8OfPShufW1B;
1140         int imm8OfPShufW2B;
1141         int fragmentLengthB;
1142         int fragmentPos;
1143
1144         int xpos, i;
1145
1146         // create an optimized horizontal scaling routine
1147
1148         //code fragment
1149
1150         asm volatile(
1151                 "jmp 9f                         \n\t"
1152         // Begin
1153                 "0:                             \n\t"
1154                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1155                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1156                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1157                 "punpcklbw %%mm7, %%mm1         \n\t"
1158                 "punpcklbw %%mm7, %%mm0         \n\t"
1159                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1160                 "1:                             \n\t"
1161                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1162                 "2:                             \n\t"
1163                 "psubw %%mm1, %%mm0             \n\t"
1164                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1165                 "pmullw %%mm3, %%mm0            \n\t"
1166                 "psllw $7, %%mm1                \n\t"
1167                 "paddw %%mm1, %%mm0             \n\t"
1168
1169                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1170
1171                 "addl $8, %%eax                 \n\t"
1172         // End
1173                 "9:                             \n\t"
1174 //              "int $3\n\t"
1175                 "leal 0b, %0                    \n\t"
1176                 "leal 1b, %1                    \n\t"
1177                 "leal 2b, %2                    \n\t"
1178                 "decl %1                        \n\t"
1179                 "decl %2                        \n\t"
1180                 "subl %0, %1                    \n\t"
1181                 "subl %0, %2                    \n\t"
1182                 "leal 9b, %3                    \n\t"
1183                 "subl %0, %3                    \n\t"
1184
1185
1186                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1187                 "=r" (fragmentLengthA)
1188         );
1189
1190         asm volatile(
1191                 "jmp 9f                         \n\t"
1192         // Begin
1193                 "0:                             \n\t"
1194                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1195                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1196                 "punpcklbw %%mm7, %%mm0         \n\t"
1197                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1198                 "1:                             \n\t"
1199                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1200                 "2:                             \n\t"
1201                 "psubw %%mm1, %%mm0             \n\t"
1202                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1203                 "pmullw %%mm3, %%mm0            \n\t"
1204                 "psllw $7, %%mm1                \n\t"
1205                 "paddw %%mm1, %%mm0             \n\t"
1206
1207                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1208
1209                 "addl $8, %%eax                 \n\t"
1210         // End
1211                 "9:                             \n\t"
1212 //              "int $3\n\t"
1213                 "leal 0b, %0                    \n\t"
1214                 "leal 1b, %1                    \n\t"
1215                 "leal 2b, %2                    \n\t"
1216                 "decl %1                        \n\t"
1217                 "decl %2                        \n\t"
1218                 "subl %0, %1                    \n\t"
1219                 "subl %0, %2                    \n\t"
1220                 "leal 9b, %3                    \n\t"
1221                 "subl %0, %3                    \n\t"
1222
1223
1224                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1225                 "=r" (fragmentLengthB)
1226         );
1227
1228         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1229         fragmentPos=0;
1230         
1231         for(i=0; i<dstW/numSplits; i++)
1232         {
1233                 int xx=xpos>>16;
1234
1235                 if((i&3) == 0)
1236                 {
1237                         int a=0;
1238                         int b=((xpos+xInc)>>16) - xx;
1239                         int c=((xpos+xInc*2)>>16) - xx;
1240                         int d=((xpos+xInc*3)>>16) - xx;
1241
1242                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1243                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1244                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1245                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1246                         filterPos[i/2]= xx;
1247
1248                         if(d+1<4)
1249                         {
1250                                 int maxShift= 3-(d+1);
1251                                 int shift=0;
1252
1253                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1254
1255                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1256                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1257                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1258                                         a | (b<<2) | (c<<4) | (d<<6);
1259
1260                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1261                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1262
1263                                 if(shift && i>=shift)
1264                                 {
1265                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1266                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1267                                         filterPos[i/2]-=shift;
1268                                 }
1269
1270                                 fragmentPos+= fragmentLengthB;
1271                         }
1272                         else
1273                         {
1274                                 int maxShift= 3-d;
1275                                 int shift=0;
1276
1277                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1278
1279                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1280                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1281                                         a | (b<<2) | (c<<4) | (d<<6);
1282
1283                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1284                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1285
1286                                 if(shift && i>=shift)
1287                                 {
1288                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1289                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1290                                         filterPos[i/2]-=shift;
1291                                 }
1292
1293                                 fragmentPos+= fragmentLengthA;
1294                         }
1295
1296                         funnyCode[fragmentPos]= RET;
1297                 }
1298                 xpos+=xInc;
1299         }
1300         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1301 }
1302 #endif // ARCH_X86
1303
1304 static void globalInit(){
1305     // generating tables:
1306     int i;
1307     for(i=0; i<768; i++){
1308         int c= MIN(MAX(i-256, 0), 255);
1309         clip_table[i]=c;
1310     }
1311 }
1312
1313 static SwsFunc getSwsFunc(int flags){
1314     
1315 #ifdef RUNTIME_CPUDETECT
1316 #ifdef ARCH_X86
1317         // ordered per speed fasterst first
1318         if(flags & SWS_CPU_CAPS_MMX2)
1319                 return swScale_MMX2;
1320         else if(flags & SWS_CPU_CAPS_3DNOW)
1321                 return swScale_3DNow;
1322         else if(flags & SWS_CPU_CAPS_MMX)
1323                 return swScale_MMX;
1324         else
1325                 return swScale_C;
1326
1327 #else
1328 #ifdef ARCH_POWERPC
1329         if(flags & SWS_CPU_CAPS_ALTIVEC)
1330           return swScale_altivec;
1331         else
1332           return swScale_C;
1333 #endif
1334         return swScale_C;
1335 #endif
1336 #else //RUNTIME_CPUDETECT
1337 #ifdef HAVE_MMX2
1338         return swScale_MMX2;
1339 #elif defined (HAVE_3DNOW)
1340         return swScale_3DNow;
1341 #elif defined (HAVE_MMX)
1342         return swScale_MMX;
1343 #elif defined (HAVE_ALTIVEC)
1344         return swScale_altivec;
1345 #else
1346         return swScale_C;
1347 #endif
1348 #endif //!RUNTIME_CPUDETECT
1349 }
1350
1351 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1352              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1353         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1354         /* Copy Y plane */
1355         if(dstStride[0]==srcStride[0])
1356                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1357         else
1358         {
1359                 int i;
1360                 uint8_t *srcPtr= src[0];
1361                 uint8_t *dstPtr= dst;
1362                 for(i=0; i<srcSliceH; i++)
1363                 {
1364                         memcpy(dstPtr, srcPtr, srcStride[0]);
1365                         srcPtr+= srcStride[0];
1366                         dstPtr+= dstStride[0];
1367                 }
1368         }
1369         dst = dstParam[1] + dstStride[1]*srcSliceY;
1370         interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1371
1372         return srcSliceH;
1373 }
1374
1375 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1376              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1377         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1378
1379         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1380
1381         return srcSliceH;
1382 }
1383
1384 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1385              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1386         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1387
1388         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1389
1390         return srcSliceH;
1391 }
1392
1393 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1394 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1395                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1396         const int srcFormat= c->srcFormat;
1397         const int dstFormat= c->dstFormat;
1398         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1399         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1400         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1401         const int dstId= (dstFormat&0xFF)>>2;
1402         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1403
1404         /* BGR -> BGR */
1405         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1406            || (isRGB(srcFormat) && isRGB(dstFormat))){
1407                 switch(srcId | (dstId<<4)){
1408                 case 0x34: conv= rgb16to15; break;
1409                 case 0x36: conv= rgb24to15; break;
1410                 case 0x38: conv= rgb32to15; break;
1411                 case 0x43: conv= rgb15to16; break;
1412                 case 0x46: conv= rgb24to16; break;
1413                 case 0x48: conv= rgb32to16; break;
1414                 case 0x63: conv= rgb15to24; break;
1415                 case 0x64: conv= rgb16to24; break;
1416                 case 0x68: conv= rgb32to24; break;
1417                 case 0x83: conv= rgb15to32; break;
1418                 case 0x84: conv= rgb16to32; break;
1419                 case 0x86: conv= rgb24to32; break;
1420                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1421                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1422                 }
1423         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1424                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1425                 switch(srcId | (dstId<<4)){
1426                 case 0x33: conv= rgb15tobgr15; break;
1427                 case 0x34: conv= rgb16tobgr15; break;
1428                 case 0x36: conv= rgb24tobgr15; break;
1429                 case 0x38: conv= rgb32tobgr15; break;
1430                 case 0x43: conv= rgb15tobgr16; break;
1431                 case 0x44: conv= rgb16tobgr16; break;
1432                 case 0x46: conv= rgb24tobgr16; break;
1433                 case 0x48: conv= rgb32tobgr16; break;
1434                 case 0x63: conv= rgb15tobgr24; break;
1435                 case 0x64: conv= rgb16tobgr24; break;
1436                 case 0x66: conv= rgb24tobgr24; break;
1437                 case 0x68: conv= rgb32tobgr24; break;
1438                 case 0x83: conv= rgb15tobgr32; break;
1439                 case 0x84: conv= rgb16tobgr32; break;
1440                 case 0x86: conv= rgb24tobgr32; break;
1441                 case 0x88: conv= rgb32tobgr32; break;
1442                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1443                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1444                 }
1445         }else{
1446                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1447                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1448         }
1449
1450         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1451                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1452         else
1453         {
1454                 int i;
1455                 uint8_t *srcPtr= src[0];
1456                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1457
1458                 for(i=0; i<srcSliceH; i++)
1459                 {
1460                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1461                         srcPtr+= srcStride[0];
1462                         dstPtr+= dstStride[0];
1463                 }
1464         }     
1465         return srcSliceH;
1466 }
1467
1468 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1469              int srcSliceH, uint8_t* dst[], int dstStride[]){
1470
1471         rgb24toyv12(
1472                 src[0], 
1473                 dst[0]+ srcSliceY    *dstStride[0], 
1474                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1475                 dst[2]+(srcSliceY>>1)*dstStride[2],
1476                 c->srcW, srcSliceH, 
1477                 dstStride[0], dstStride[1], srcStride[0]);
1478         return srcSliceH;
1479 }
1480
1481 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1482              int srcSliceH, uint8_t* dst[], int dstStride[]){
1483         int i;
1484
1485         /* copy Y */
1486         if(srcStride[0]==dstStride[0]) 
1487                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1488         else{
1489                 uint8_t *srcPtr= src[0];
1490                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1491
1492                 for(i=0; i<srcSliceH; i++)
1493                 {
1494                         memcpy(dstPtr, srcPtr, c->srcW);
1495                         srcPtr+= srcStride[0];
1496                         dstPtr+= dstStride[0];
1497                 }
1498         }
1499
1500         if(c->dstFormat==IMGFMT_YV12){
1501                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1502                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1503         }else{
1504                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1505                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1506         }
1507         return srcSliceH;
1508 }
1509
1510 /**
1511  * bring pointers in YUV order instead of YVU
1512  */
1513 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1514         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1515            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1516                 sortedP[0]= p[0];
1517                 sortedP[1]= p[2];
1518                 sortedP[2]= p[1];
1519                 sortedStride[0]= stride[0];
1520                 sortedStride[1]= stride[2];
1521                 sortedStride[2]= stride[1];
1522         }
1523         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1524         {
1525                 sortedP[0]= p[0];
1526                 sortedP[1]= 
1527                 sortedP[2]= NULL;
1528                 sortedStride[0]= stride[0];
1529                 sortedStride[1]= 
1530                 sortedStride[2]= 0;
1531         }
1532         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1533         {
1534                 sortedP[0]= p[0];
1535                 sortedP[1]= p[1];
1536                 sortedP[2]= p[2];
1537                 sortedStride[0]= stride[0];
1538                 sortedStride[1]= stride[1];
1539                 sortedStride[2]= stride[2];
1540         }else{
1541                 MSG_ERR("internal error in orderYUV\n");
1542         }
1543 }
1544
1545 /* unscaled copy like stuff (assumes nearly identical formats) */
1546 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1547              int srcSliceH, uint8_t* dst[], int dstStride[]){
1548
1549         if(isPacked(c->srcFormat))
1550         {
1551                 if(dstStride[0]==srcStride[0])
1552                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1553                 else
1554                 {
1555                         int i;
1556                         uint8_t *srcPtr= src[0];
1557                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1558                         int length=0;
1559
1560                         /* universal length finder */
1561                         while(length+c->srcW <= ABS(dstStride[0]) 
1562                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1563                         ASSERT(length!=0);
1564
1565                         for(i=0; i<srcSliceH; i++)
1566                         {
1567                                 memcpy(dstPtr, srcPtr, length);
1568                                 srcPtr+= srcStride[0];
1569                                 dstPtr+= dstStride[0];
1570                         }
1571                 }
1572         }
1573         else 
1574         { /* Planar YUV or gray */
1575                 int plane;
1576                 for(plane=0; plane<3; plane++)
1577                 {
1578                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1579                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1580                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1581
1582                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1583                         {
1584                                 if(!isGray(c->dstFormat))
1585                                         memset(dst[plane], 128, dstStride[plane]*height);
1586                         }
1587                         else
1588                         {
1589                                 if(dstStride[plane]==srcStride[plane])
1590                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1591                                 else
1592                                 {
1593                                         int i;
1594                                         uint8_t *srcPtr= src[plane];
1595                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1596                                         for(i=0; i<height; i++)
1597                                         {
1598                                                 memcpy(dstPtr, srcPtr, length);
1599                                                 srcPtr+= srcStride[plane];
1600                                                 dstPtr+= dstStride[plane];
1601                                         }
1602                                 }
1603                         }
1604                 }
1605         }
1606         return srcSliceH;
1607 }
1608
1609 static int remove_dup_fourcc(int fourcc)
1610 {
1611         switch(fourcc)
1612         {
1613             case IMGFMT_I420:
1614             case IMGFMT_IYUV: return IMGFMT_YV12;
1615             case IMGFMT_Y8  : return IMGFMT_Y800;
1616             case IMGFMT_IF09: return IMGFMT_YVU9;
1617             default: return fourcc;
1618         }
1619 }
1620
1621 static void getSubSampleFactors(int *h, int *v, int format){
1622         switch(format){
1623         case IMGFMT_UYVY:
1624         case IMGFMT_YUY2:
1625                 *h=1;
1626                 *v=0;
1627                 break;
1628         case IMGFMT_YV12:
1629         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1630                 *h=1;
1631                 *v=1;
1632                 break;
1633         case IMGFMT_YVU9:
1634                 *h=2;
1635                 *v=2;
1636                 break;
1637         case IMGFMT_444P:
1638                 *h=0;
1639                 *v=0;
1640                 break;
1641         case IMGFMT_422P:
1642                 *h=1;
1643                 *v=0;
1644                 break;
1645         case IMGFMT_411P:
1646                 *h=2;
1647                 *v=0;
1648                 break;
1649         default:
1650                 *h=0;
1651                 *v=0;
1652                 break;
1653         }
1654 }
1655
1656 static uint16_t roundToInt16(int64_t f){
1657         int r= (f + (1<<15))>>16;
1658              if(r<-0x7FFF) return 0x8000;
1659         else if(r> 0x7FFF) return 0x7FFF;
1660         else               return r;
1661 }
1662
1663 /**
1664  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1665  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1666  * @return -1 if not supported
1667  */
1668 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1669         int64_t crv =  inv_table[0];
1670         int64_t cbu =  inv_table[1];
1671         int64_t cgu = -inv_table[2];
1672         int64_t cgv = -inv_table[3];
1673         int64_t cy  = 1<<16;
1674         int64_t oy  = 0;
1675
1676         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1677         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1678         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1679
1680         c->brightness= brightness;
1681         c->contrast  = contrast;
1682         c->saturation= saturation;
1683         c->srcRange  = srcRange;
1684         c->dstRange  = dstRange;
1685
1686         c->uOffset=   0x0400040004000400LL;
1687         c->vOffset=   0x0400040004000400LL;
1688
1689         if(!srcRange){
1690                 cy= (cy*255) / 219;
1691                 oy= 16<<16;
1692         }
1693
1694         cy = (cy *contrast             )>>16;
1695         crv= (crv*contrast * saturation)>>32;
1696         cbu= (cbu*contrast * saturation)>>32;
1697         cgu= (cgu*contrast * saturation)>>32;
1698         cgv= (cgv*contrast * saturation)>>32;
1699
1700         oy -= 256*brightness;
1701
1702         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1703         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1704         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1705         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1706         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1707         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1708
1709         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1710         //FIXME factorize
1711         
1712         return 0;
1713 }
1714
1715 /**
1716  * @return -1 if not supported
1717  */
1718 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1719         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1720
1721         *inv_table = c->srcColorspaceTable;
1722         *table     = c->dstColorspaceTable;
1723         *srcRange  = c->srcRange;
1724         *dstRange  = c->dstRange;
1725         *brightness= c->brightness;
1726         *contrast  = c->contrast;
1727         *saturation= c->saturation;
1728         
1729         return 0;       
1730 }
1731
1732 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1733                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1734
1735         SwsContext *c;
1736         int i;
1737         int usesVFilter, usesHFilter;
1738         int unscaled, needsDither;
1739         int srcFormat, dstFormat;
1740         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1741 #ifdef ARCH_X86
1742         if(flags & SWS_CPU_CAPS_MMX)
1743                 asm volatile("emms\n\t"::: "memory");
1744 #endif
1745
1746 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1747         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1748 #ifdef HAVE_MMX2
1749         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1750 #elif defined (HAVE_3DNOW)
1751         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1752 #elif defined (HAVE_MMX)
1753         flags |= SWS_CPU_CAPS_MMX;
1754 #elif defined (HAVE_ALTIVEC)
1755         flags |= SWS_CPU_CAPS_ALTIVEC;
1756 #endif
1757 #endif
1758         if(clip_table[512] != 255) globalInit();
1759         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1760
1761         /* avoid duplicate Formats, so we don't need to check to much */
1762         srcFormat = remove_dup_fourcc(origSrcFormat);
1763         dstFormat = remove_dup_fourcc(origDstFormat);
1764
1765         unscaled = (srcW == dstW && srcH == dstH);
1766         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1767                      && (dstFormat&0xFF)<24
1768                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1769
1770         if(!isSupportedIn(srcFormat)) 
1771         {
1772                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1773                 return NULL;
1774         }
1775         if(!isSupportedOut(dstFormat))
1776         {
1777                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1778                 return NULL;
1779         }
1780
1781         /* sanity check */
1782         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
1783         {
1784                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1785                         srcW, srcH, dstW, dstH);
1786                 return NULL;
1787         }
1788
1789         if(!dstFilter) dstFilter= &dummyFilter;
1790         if(!srcFilter) srcFilter= &dummyFilter;
1791
1792         c= memalign(64, sizeof(SwsContext));
1793         memset(c, 0, sizeof(SwsContext));
1794
1795         c->srcW= srcW;
1796         c->srcH= srcH;
1797         c->dstW= dstW;
1798         c->dstH= dstH;
1799         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1800         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1801         c->flags= flags;
1802         c->dstFormat= dstFormat;
1803         c->srcFormat= srcFormat;
1804         c->origDstFormat= origDstFormat;
1805         c->origSrcFormat= origSrcFormat;
1806         c->vRounder= 4* 0x0001000100010001ULL;
1807
1808         usesHFilter= usesVFilter= 0;
1809         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1810         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1811         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1812         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1813         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1814         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1815         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1816         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1817
1818         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1819         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1820
1821         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1822         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1823
1824         // drop some chroma lines if the user wants it
1825         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1826         c->chrSrcVSubSample+= c->vChrDrop;
1827
1828         // drop every 2. pixel for chroma calculation unless user wants full chroma
1829         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1830                 c->chrSrcHSubSample=1;
1831
1832         c->chrIntHSubSample= c->chrDstHSubSample;
1833         c->chrIntVSubSample= c->chrSrcVSubSample;
1834
1835         // note the -((-x)>>y) is so that we allways round toward +inf
1836         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1837         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1838         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1839         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1840
1841         sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], 0, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, 0, 0, 1<<16, 1<<16); 
1842
1843         /* unscaled special Cases */
1844         if(unscaled && !usesHFilter && !usesVFilter)
1845         {
1846                 /* yv12_to_nv12 */
1847                 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1848                 {
1849                         c->swScale= PlanarToNV12Wrapper;
1850                 }
1851                 /* yuv2bgr */
1852                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1853                 {
1854                         c->swScale= yuv2rgb_get_func_ptr(c);
1855                 }
1856                 
1857                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1858                 {
1859                         c->swScale= yvu9toyv12Wrapper;
1860                 }
1861
1862                 /* bgr24toYV12 */
1863                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1864                         c->swScale= bgr24toyv12Wrapper;
1865                 
1866                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1867                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1868                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
1869                    && !needsDither)
1870                         c->swScale= rgb2rgbWrapper;
1871
1872                 /* LQ converters if -sws 0 or -sws 4*/
1873                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1874                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1875                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1876                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
1877                           && needsDither)
1878                                 c->swScale= rgb2rgbWrapper;
1879
1880                         /* yv12_to_yuy2 */
1881                         if(srcFormat == IMGFMT_YV12 && 
1882                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1883                         {
1884                                 if (dstFormat == IMGFMT_YUY2)
1885                                     c->swScale= PlanarToYuy2Wrapper;
1886                                 else
1887                                     c->swScale= PlanarToUyvyWrapper;
1888                         }
1889                 }
1890
1891                 /* simple copy */
1892                 if(   srcFormat == dstFormat
1893                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1894                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1895                   )
1896                 {
1897                         c->swScale= simpleCopy;
1898                 }
1899
1900                 if(c->swScale){
1901                         if(flags&SWS_PRINT_INFO)
1902                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
1903                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
1904                         return c;
1905                 }
1906         }
1907
1908         if(flags & SWS_CPU_CAPS_MMX2)
1909         {
1910                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1911                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1912                 {
1913                         if(flags&SWS_PRINT_INFO)
1914                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1915                 }
1916                 if(usesHFilter) c->canMMX2BeUsed=0;
1917         }
1918         else
1919                 c->canMMX2BeUsed=0;
1920
1921         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1922         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1923
1924         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1925         // but only for the FAST_BILINEAR mode otherwise do correct scaling
1926         // n-2 is the last chrominance sample available
1927         // this is not perfect, but noone shuld notice the difference, the more correct variant
1928         // would be like the vertical one, but that would require some special code for the
1929         // first and last pixel
1930         if(flags&SWS_FAST_BILINEAR)
1931         {
1932                 if(c->canMMX2BeUsed)
1933                 {
1934                         c->lumXInc+= 20;
1935                         c->chrXInc+= 20;
1936                 }
1937                 //we don't use the x86asm scaler if mmx is available
1938                 else if(flags & SWS_CPU_CAPS_MMX)
1939                 {
1940                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1941                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1942                 }
1943         }
1944
1945         /* precalculate horizontal scaler filter coefficients */
1946         {
1947                 const int filterAlign= (flags & SWS_CPU_CAPS_MMX) ? 4 : 1;
1948
1949                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1950                                  srcW      ,       dstW, filterAlign, 1<<14,
1951                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1952                                  srcFilter->lumH, dstFilter->lumH);
1953                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1954                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1955                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1956                                  srcFilter->chrH, dstFilter->chrH);
1957
1958 #ifdef ARCH_X86
1959 // can't downscale !!!
1960                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1961                 {
1962                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
1963                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
1964                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
1965                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1966
1967                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1968                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
1969                 }
1970 #endif
1971         } // Init Horizontal stuff
1972
1973
1974
1975         /* precalculate vertical scaler filter coefficients */
1976         initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1977                         srcH      ,        dstH, 1, (1<<12)-4,
1978                         (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1979                         srcFilter->lumV, dstFilter->lumV);
1980         initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1981                         c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
1982                         (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1983                         srcFilter->chrV, dstFilter->chrV);
1984
1985         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
1986         c->vLumBufSize= c->vLumFilterSize;
1987         c->vChrBufSize= c->vChrFilterSize;
1988         for(i=0; i<dstH; i++)
1989         {
1990                 int chrI= i*c->chrDstH / dstH;
1991                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
1992                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1993
1994                 nextSlice>>= c->chrSrcVSubSample;
1995                 nextSlice<<= c->chrSrcVSubSample;
1996                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
1997                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
1998                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1999                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2000         }
2001
2002         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2003         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2004         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2005         //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)
2006         for(i=0; i<c->vLumBufSize; i++)
2007                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2008         for(i=0; i<c->vChrBufSize; i++)
2009                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2010
2011         //try to avoid drawing green stuff between the right end and the stride end
2012         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2013         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2014
2015         ASSERT(c->chrDstH <= dstH)
2016
2017         if(flags&SWS_PRINT_INFO)
2018         {
2019 #ifdef DITHER1XBPP
2020                 char *dither= " dithered";
2021 #else
2022                 char *dither= "";
2023 #endif
2024                 if(flags&SWS_FAST_BILINEAR)
2025                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2026                 else if(flags&SWS_BILINEAR)
2027                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2028                 else if(flags&SWS_BICUBIC)
2029                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2030                 else if(flags&SWS_X)
2031                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2032                 else if(flags&SWS_POINT)
2033                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2034                 else if(flags&SWS_AREA)
2035                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2036                 else if(flags&SWS_BICUBLIN)
2037                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2038                 else if(flags&SWS_GAUSS)
2039                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2040                 else if(flags&SWS_SINC)
2041                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2042                 else if(flags&SWS_LANCZOS)
2043                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2044                 else if(flags&SWS_SPLINE)
2045                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2046                 else
2047                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2048
2049                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2050                         MSG_INFO("from %s to%s %s ", 
2051                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2052                 else
2053                         MSG_INFO("from %s to %s ", 
2054                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2055
2056                 if(flags & SWS_CPU_CAPS_MMX2)
2057                         MSG_INFO("using MMX2\n");
2058                 else if(flags & SWS_CPU_CAPS_3DNOW)
2059                         MSG_INFO("using 3DNOW\n");
2060                 else if(flags & SWS_CPU_CAPS_MMX)
2061                         MSG_INFO("using MMX\n");
2062                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2063                         MSG_INFO("using AltiVec\n");
2064                 else 
2065                         MSG_INFO("using C\n");
2066         }
2067
2068         if(flags & SWS_PRINT_INFO)
2069         {
2070                 if(flags & SWS_CPU_CAPS_MMX)
2071                 {
2072                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2073                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2074                         else
2075                         {
2076                                 if(c->hLumFilterSize==4)
2077                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2078                                 else if(c->hLumFilterSize==8)
2079                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2080                                 else
2081                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2082
2083                                 if(c->hChrFilterSize==4)
2084                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2085                                 else if(c->hChrFilterSize==8)
2086                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2087                                 else
2088                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2089                         }
2090                 }
2091                 else
2092                 {
2093 #ifdef ARCH_X86
2094                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2095 #else
2096                         if(flags & SWS_FAST_BILINEAR)
2097                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2098                         else
2099                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2100 #endif
2101                 }
2102                 if(isPlanarYUV(dstFormat))
2103                 {
2104                         if(c->vLumFilterSize==1)
2105                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2106                         else
2107                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2108                 }
2109                 else
2110                 {
2111                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2112                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2113                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2114                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2115                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2116                         else
2117                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2118                 }
2119
2120                 if(dstFormat==IMGFMT_BGR24)
2121                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2122                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2123                 else if(dstFormat==IMGFMT_BGR32)
2124                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2125                 else if(dstFormat==IMGFMT_BGR16)
2126                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2127                 else if(dstFormat==IMGFMT_BGR15)
2128                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2129
2130                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2131         }
2132         if(flags & SWS_PRINT_INFO)
2133         {
2134                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2135                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2136                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2137                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2138         }
2139
2140         c->swScale= getSwsFunc(flags);
2141         return c;
2142 }
2143
2144 /**
2145  * swscale warper, so we don't need to export the SwsContext.
2146  * assumes planar YUV to be in YUV order instead of YVU
2147  */
2148 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2149                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2150         //copy strides, so they can safely be modified
2151         int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2152         int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2153         return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2154 }
2155
2156 /**
2157  * swscale warper, so we don't need to export the SwsContext
2158  */
2159 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2160                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2161         int srcStride[3];
2162         int dstStride[3];
2163         uint8_t *src[3];
2164         uint8_t *dst[3];
2165         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2166         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2167 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2168
2169         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2170 }
2171
2172 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2173                                 float lumaSharpen, float chromaSharpen,
2174                                 float chromaHShift, float chromaVShift,
2175                                 int verbose)
2176 {
2177         SwsFilter *filter= malloc(sizeof(SwsFilter));
2178
2179         if(lumaGBlur!=0.0){
2180                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2181                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2182         }else{
2183                 filter->lumH= sws_getIdentityVec();
2184                 filter->lumV= sws_getIdentityVec();
2185         }
2186
2187         if(chromaGBlur!=0.0){
2188                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2189                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2190         }else{
2191                 filter->chrH= sws_getIdentityVec();
2192                 filter->chrV= sws_getIdentityVec();
2193         }
2194
2195         if(chromaSharpen!=0.0){
2196                 SwsVector *g= sws_getConstVec(-1.0, 3);
2197                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2198                 g->coeff[1]=2.0;
2199                 sws_addVec(id, g);
2200                 sws_convVec(filter->chrH, id);
2201                 sws_convVec(filter->chrV, id);
2202                 sws_freeVec(g);
2203                 sws_freeVec(id);
2204         }
2205
2206         if(lumaSharpen!=0.0){
2207                 SwsVector *g= sws_getConstVec(-1.0, 3);
2208                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2209                 g->coeff[1]=2.0;
2210                 sws_addVec(id, g);
2211                 sws_convVec(filter->lumH, id);
2212                 sws_convVec(filter->lumV, id);
2213                 sws_freeVec(g);
2214                 sws_freeVec(id);
2215         }
2216
2217         if(chromaHShift != 0.0)
2218                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2219
2220         if(chromaVShift != 0.0)
2221                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2222
2223         sws_normalizeVec(filter->chrH, 1.0);
2224         sws_normalizeVec(filter->chrV, 1.0);
2225         sws_normalizeVec(filter->lumH, 1.0);
2226         sws_normalizeVec(filter->lumV, 1.0);
2227
2228         if(verbose) sws_printVec(filter->chrH);
2229         if(verbose) sws_printVec(filter->lumH);
2230
2231         return filter;
2232 }
2233
2234 /**
2235  * returns a normalized gaussian curve used to filter stuff
2236  * quality=3 is high quality, lowwer is lowwer quality
2237  */
2238 SwsVector *sws_getGaussianVec(double variance, double quality){
2239         const int length= (int)(variance*quality + 0.5) | 1;
2240         int i;
2241         double *coeff= memalign(sizeof(double), length*sizeof(double));
2242         double middle= (length-1)*0.5;
2243         SwsVector *vec= malloc(sizeof(SwsVector));
2244
2245         vec->coeff= coeff;
2246         vec->length= length;
2247
2248         for(i=0; i<length; i++)
2249         {
2250                 double dist= i-middle;
2251                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2252         }
2253
2254         sws_normalizeVec(vec, 1.0);
2255
2256         return vec;
2257 }
2258
2259 SwsVector *sws_getConstVec(double c, int length){
2260         int i;
2261         double *coeff= memalign(sizeof(double), length*sizeof(double));
2262         SwsVector *vec= malloc(sizeof(SwsVector));
2263
2264         vec->coeff= coeff;
2265         vec->length= length;
2266
2267         for(i=0; i<length; i++)
2268                 coeff[i]= c;
2269
2270         return vec;
2271 }
2272
2273
2274 SwsVector *sws_getIdentityVec(void){
2275         double *coeff= memalign(sizeof(double), sizeof(double));
2276         SwsVector *vec= malloc(sizeof(SwsVector));
2277         coeff[0]= 1.0;
2278
2279         vec->coeff= coeff;
2280         vec->length= 1;
2281
2282         return vec;
2283 }
2284
2285 void sws_normalizeVec(SwsVector *a, double height){
2286         int i;
2287         double sum=0;
2288         double inv;
2289
2290         for(i=0; i<a->length; i++)
2291                 sum+= a->coeff[i];
2292
2293         inv= height/sum;
2294
2295         for(i=0; i<a->length; i++)
2296                 a->coeff[i]*= inv;
2297 }
2298
2299 void sws_scaleVec(SwsVector *a, double scalar){
2300         int i;
2301
2302         for(i=0; i<a->length; i++)
2303                 a->coeff[i]*= scalar;
2304 }
2305
2306 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2307         int length= a->length + b->length - 1;
2308         double *coeff= memalign(sizeof(double), length*sizeof(double));
2309         int i, j;
2310         SwsVector *vec= malloc(sizeof(SwsVector));
2311
2312         vec->coeff= coeff;
2313         vec->length= length;
2314
2315         for(i=0; i<length; i++) coeff[i]= 0.0;
2316
2317         for(i=0; i<a->length; i++)
2318         {
2319                 for(j=0; j<b->length; j++)
2320                 {
2321                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2322                 }
2323         }
2324
2325         return vec;
2326 }
2327
2328 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2329         int length= MAX(a->length, b->length);
2330         double *coeff= memalign(sizeof(double), length*sizeof(double));
2331         int i;
2332         SwsVector *vec= malloc(sizeof(SwsVector));
2333
2334         vec->coeff= coeff;
2335         vec->length= length;
2336
2337         for(i=0; i<length; i++) coeff[i]= 0.0;
2338
2339         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2340         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2341
2342         return vec;
2343 }
2344
2345 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2346         int length= MAX(a->length, b->length);
2347         double *coeff= memalign(sizeof(double), length*sizeof(double));
2348         int i;
2349         SwsVector *vec= malloc(sizeof(SwsVector));
2350
2351         vec->coeff= coeff;
2352         vec->length= length;
2353
2354         for(i=0; i<length; i++) coeff[i]= 0.0;
2355
2356         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2357         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2358
2359         return vec;
2360 }
2361
2362 /* shift left / or right if "shift" is negative */
2363 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2364         int length= a->length + ABS(shift)*2;
2365         double *coeff= memalign(sizeof(double), length*sizeof(double));
2366         int i;
2367         SwsVector *vec= malloc(sizeof(SwsVector));
2368
2369         vec->coeff= coeff;
2370         vec->length= length;
2371
2372         for(i=0; i<length; i++) coeff[i]= 0.0;
2373
2374         for(i=0; i<a->length; i++)
2375         {
2376                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2377         }
2378
2379         return vec;
2380 }
2381
2382 void sws_shiftVec(SwsVector *a, int shift){
2383         SwsVector *shifted= sws_getShiftedVec(a, shift);
2384         free(a->coeff);
2385         a->coeff= shifted->coeff;
2386         a->length= shifted->length;
2387         free(shifted);
2388 }
2389
2390 void sws_addVec(SwsVector *a, SwsVector *b){
2391         SwsVector *sum= sws_sumVec(a, b);
2392         free(a->coeff);
2393         a->coeff= sum->coeff;
2394         a->length= sum->length;
2395         free(sum);
2396 }
2397
2398 void sws_subVec(SwsVector *a, SwsVector *b){
2399         SwsVector *diff= sws_diffVec(a, b);
2400         free(a->coeff);
2401         a->coeff= diff->coeff;
2402         a->length= diff->length;
2403         free(diff);
2404 }
2405
2406 void sws_convVec(SwsVector *a, SwsVector *b){
2407         SwsVector *conv= sws_getConvVec(a, b);
2408         free(a->coeff);  
2409         a->coeff= conv->coeff;
2410         a->length= conv->length;
2411         free(conv);
2412 }
2413
2414 SwsVector *sws_cloneVec(SwsVector *a){
2415         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2416         int i;
2417         SwsVector *vec= malloc(sizeof(SwsVector));
2418
2419         vec->coeff= coeff;
2420         vec->length= a->length;
2421
2422         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2423
2424         return vec;
2425 }
2426
2427 void sws_printVec(SwsVector *a){
2428         int i;
2429         double max=0;
2430         double min=0;
2431         double range;
2432
2433         for(i=0; i<a->length; i++)
2434                 if(a->coeff[i]>max) max= a->coeff[i];
2435
2436         for(i=0; i<a->length; i++)
2437                 if(a->coeff[i]<min) min= a->coeff[i];
2438
2439         range= max - min;
2440
2441         for(i=0; i<a->length; i++)
2442         {
2443                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2444                 MSG_DBG2("%1.3f ", a->coeff[i]);
2445                 for(;x>0; x--) MSG_DBG2(" ");
2446                 MSG_DBG2("|\n");
2447         }
2448 }
2449
2450 void sws_freeVec(SwsVector *a){
2451         if(!a) return;
2452         if(a->coeff) free(a->coeff);
2453         a->coeff=NULL;
2454         a->length=0;
2455         free(a);
2456 }
2457
2458 void sws_freeFilter(SwsFilter *filter){
2459         if(!filter) return;
2460
2461         if(filter->lumH) sws_freeVec(filter->lumH);
2462         if(filter->lumV) sws_freeVec(filter->lumV);
2463         if(filter->chrH) sws_freeVec(filter->chrH);
2464         if(filter->chrV) sws_freeVec(filter->chrV);
2465         free(filter);
2466 }
2467
2468
2469 void sws_freeContext(SwsContext *c){
2470         int i;
2471         if(!c) return;
2472
2473         if(c->lumPixBuf)
2474         {
2475                 for(i=0; i<c->vLumBufSize; i++)
2476                 {
2477                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2478                         c->lumPixBuf[i]=NULL;
2479                 }
2480                 free(c->lumPixBuf);
2481                 c->lumPixBuf=NULL;
2482         }
2483
2484         if(c->chrPixBuf)
2485         {
2486                 for(i=0; i<c->vChrBufSize; i++)
2487                 {
2488                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2489                         c->chrPixBuf[i]=NULL;
2490                 }
2491                 free(c->chrPixBuf);
2492                 c->chrPixBuf=NULL;
2493         }
2494
2495         if(c->vLumFilter) free(c->vLumFilter);
2496         c->vLumFilter = NULL;
2497         if(c->vChrFilter) free(c->vChrFilter);
2498         c->vChrFilter = NULL;
2499         if(c->hLumFilter) free(c->hLumFilter);
2500         c->hLumFilter = NULL;
2501         if(c->hChrFilter) free(c->hChrFilter);
2502         c->hChrFilter = NULL;
2503
2504         if(c->vLumFilterPos) free(c->vLumFilterPos);
2505         c->vLumFilterPos = NULL;
2506         if(c->vChrFilterPos) free(c->vChrFilterPos);
2507         c->vChrFilterPos = NULL;
2508         if(c->hLumFilterPos) free(c->hLumFilterPos);
2509         c->hLumFilterPos = NULL;
2510         if(c->hChrFilterPos) free(c->hChrFilterPos);
2511         c->hChrFilterPos = NULL;
2512
2513         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2514         c->lumMmx2Filter=NULL;
2515         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2516         c->chrMmx2Filter=NULL;
2517         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2518         c->lumMmx2FilterPos=NULL;
2519         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2520         c->chrMmx2FilterPos=NULL;
2521         if(c->yuvTable) free(c->yuvTable);
2522         c->yuvTable=NULL;
2523
2524         free(c);
2525 }
2526