AltiVec hScale, all size patch by (Romain Dolbeau <dolbeaur at club-internet dot...
[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 #ifdef HAVE_ALTIVEC_H
65 #include <altivec.h>
66 #endif
67 #include "swscale.h"
68 #include "swscale_internal.h"
69 #include "../cpudetect.h"
70 #include "../bswap.h"
71 #include "../libvo/img_format.h"
72 #include "rgb2rgb.h"
73 #include "../libvo/fastmemcpy.h"
74
75 #undef MOVNTQ
76 #undef PAVGB
77
78 //#undef HAVE_MMX2
79 //#define HAVE_3DNOW
80 //#undef HAVE_MMX
81 //#undef ARCH_X86
82 //#define WORDS_BIGENDIAN
83 #define DITHER1XBPP
84
85 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
86
87 #define RET 0xC3 //near return opcode for X86
88
89 #ifdef MP_DEBUG
90 #define ASSERT(x) assert(x);
91 #else
92 #define ASSERT(x) ;
93 #endif
94
95 #ifdef M_PI
96 #define PI M_PI
97 #else
98 #define PI 3.14159265358979323846
99 #endif
100
101 //FIXME replace this with something faster
102 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
103                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
104 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
105 #define isGray(x)      ((x)==IMGFMT_Y800)
106 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
107 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
108 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
109                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
110                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
111                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
112                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
113 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
114                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
115                         || isRGB(x) || isBGR(x)\
116                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
117 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
118
119 #define RGB2YUV_SHIFT 16
120 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
121 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
122 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
123 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
124 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
125 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
126 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
127 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
128 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
129
130 extern const int32_t Inverse_Table_6_9[8][4];
131
132 /*
133 NOTES
134 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
135
136 TODO
137 more intelligent missalignment avoidance for the horizontal scaler
138 write special vertical cubic upscale version
139 Optimize C code (yv12 / minmax)
140 add support for packed pixel yuv input & output
141 add support for Y8 output
142 optimize bgr24 & bgr32
143 add BGR4 output support
144 write special BGR->BGR scaler
145 */
146
147 #define ABS(a) ((a) > 0 ? (a) : (-(a)))
148 #define MIN(a,b) ((a) > (b) ? (b) : (a))
149 #define MAX(a,b) ((a) < (b) ? (b) : (a))
150
151 #ifdef ARCH_X86
152 static uint64_t __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
153 static uint64_t __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
154 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
155 static uint64_t __attribute__((aligned(8))) w02=       0x0002000200020002LL;
156 static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
157 static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
158 static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
159 static uint64_t __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
160
161 static volatile uint64_t __attribute__((aligned(8))) b5Dither;
162 static volatile uint64_t __attribute__((aligned(8))) g5Dither;
163 static volatile uint64_t __attribute__((aligned(8))) g6Dither;
164 static volatile uint64_t __attribute__((aligned(8))) r5Dither;
165
166 static uint64_t __attribute__((aligned(8))) dither4[2]={
167         0x0103010301030103LL,
168         0x0200020002000200LL,};
169
170 static uint64_t __attribute__((aligned(8))) dither8[2]={
171         0x0602060206020602LL,
172         0x0004000400040004LL,};
173
174 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
175 static uint64_t __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
176 static uint64_t __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
177 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
178 static uint64_t __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
179 static uint64_t __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
180
181 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
182 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
183 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
184
185 #ifdef FAST_BGR2YV12
186 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
187 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
188 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
189 #else
190 static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
191 static const uint64_t bgr2UCoeff  __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
192 static const uint64_t bgr2VCoeff  __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
193 #endif
194 static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
195 static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
196 static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
197 #endif
198
199 // clipping helper table for C implementations:
200 static unsigned char clip_table[768];
201
202 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
203                   
204 extern const uint8_t dither_2x2_4[2][8];
205 extern const uint8_t dither_2x2_8[2][8];
206 extern const uint8_t dither_8x8_32[8][8];
207 extern const uint8_t dither_8x8_73[8][8];
208 extern const uint8_t dither_8x8_220[8][8];
209
210 #ifdef ARCH_X86
211 void in_asm_used_var_warning_killer()
212 {
213  volatile int i= bF8+bFC+w10+
214  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
215  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
216  if(i) i=0;
217 }
218 #endif
219
220 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
221                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
222                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
223 {
224         //FIXME Optimize (just quickly writen not opti..)
225         int i;
226         for(i=0; i<dstW; i++)
227         {
228                 int val=1<<18;
229                 int j;
230                 for(j=0; j<lumFilterSize; j++)
231                         val += lumSrc[j][i] * lumFilter[j];
232
233                 dest[i]= MIN(MAX(val>>19, 0), 255);
234         }
235
236         if(uDest != NULL)
237                 for(i=0; i<chrDstW; i++)
238                 {
239                         int u=1<<18;
240                         int v=1<<18;
241                         int j;
242                         for(j=0; j<chrFilterSize; j++)
243                         {
244                                 u += chrSrc[j][i] * chrFilter[j];
245                                 v += chrSrc[j][i + 2048] * chrFilter[j];
246                         }
247
248                         uDest[i]= MIN(MAX(u>>19, 0), 255);
249                         vDest[i]= MIN(MAX(v>>19, 0), 255);
250                 }
251 }
252
253
254 #define YSCALE_YUV_2_PACKEDX_C(type) \
255                 for(i=0; i<(dstW>>1); i++){\
256                         int j;\
257                         int Y1=1<<18;\
258                         int Y2=1<<18;\
259                         int U=1<<18;\
260                         int V=1<<18;\
261                         type *r, *b, *g;\
262                         const int i2= 2*i;\
263                         \
264                         for(j=0; j<lumFilterSize; j++)\
265                         {\
266                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
267                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
268                         }\
269                         for(j=0; j<chrFilterSize; j++)\
270                         {\
271                                 U += chrSrc[j][i] * chrFilter[j];\
272                                 V += chrSrc[j][i+2048] * chrFilter[j];\
273                         }\
274                         Y1>>=19;\
275                         Y2>>=19;\
276                         U >>=19;\
277                         V >>=19;\
278                         if((Y1|Y2|U|V)&256)\
279                         {\
280                                 if(Y1>255)   Y1=255;\
281                                 else if(Y1<0)Y1=0;\
282                                 if(Y2>255)   Y2=255;\
283                                 else if(Y2<0)Y2=0;\
284                                 if(U>255)    U=255;\
285                                 else if(U<0) U=0;\
286                                 if(V>255)    V=255;\
287                                 else if(V<0) V=0;\
288                         }
289                         
290 #define YSCALE_YUV_2_RGBX_C(type) \
291                         YSCALE_YUV_2_PACKEDX_C(type)\
292                         r = c->table_rV[V];\
293                         g = c->table_gU[U] + c->table_gV[V];\
294                         b = c->table_bU[U];\
295
296 #define YSCALE_YUV_2_PACKED2_C \
297                 for(i=0; i<(dstW>>1); i++){\
298                         const int i2= 2*i;\
299                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
300                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
301                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
302                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
303
304 #define YSCALE_YUV_2_RGB2_C(type) \
305                         YSCALE_YUV_2_PACKED2_C\
306                         type *r, *b, *g;\
307                         r = c->table_rV[V];\
308                         g = c->table_gU[U] + c->table_gV[V];\
309                         b = c->table_bU[U];\
310
311 #define YSCALE_YUV_2_PACKED1_C \
312                 for(i=0; i<(dstW>>1); i++){\
313                         const int i2= 2*i;\
314                         int Y1= buf0[i2  ]>>7;\
315                         int Y2= buf0[i2+1]>>7;\
316                         int U= (uvbuf1[i     ])>>7;\
317                         int V= (uvbuf1[i+2048])>>7;\
318
319 #define YSCALE_YUV_2_RGB1_C(type) \
320                         YSCALE_YUV_2_PACKED1_C\
321                         type *r, *b, *g;\
322                         r = c->table_rV[V];\
323                         g = c->table_gU[U] + c->table_gV[V];\
324                         b = c->table_bU[U];\
325
326 #define YSCALE_YUV_2_PACKED1B_C \
327                 for(i=0; i<(dstW>>1); i++){\
328                         const int i2= 2*i;\
329                         int Y1= buf0[i2  ]>>7;\
330                         int Y2= buf0[i2+1]>>7;\
331                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
332                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
333
334 #define YSCALE_YUV_2_RGB1B_C(type) \
335                         YSCALE_YUV_2_PACKED1B_C\
336                         type *r, *b, *g;\
337                         r = c->table_rV[V];\
338                         g = c->table_gU[U] + c->table_gV[V];\
339                         b = c->table_bU[U];\
340
341 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
342         switch(c->dstFormat)\
343         {\
344         case IMGFMT_BGR32:\
345         case IMGFMT_RGB32:\
346                 func(uint32_t)\
347                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
348                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
349                 }               \
350                 break;\
351         case IMGFMT_RGB24:\
352                 func(uint8_t)\
353                         ((uint8_t*)dest)[0]= r[Y1];\
354                         ((uint8_t*)dest)[1]= g[Y1];\
355                         ((uint8_t*)dest)[2]= b[Y1];\
356                         ((uint8_t*)dest)[3]= r[Y2];\
357                         ((uint8_t*)dest)[4]= g[Y2];\
358                         ((uint8_t*)dest)[5]= b[Y2];\
359                         ((uint8_t*)dest)+=6;\
360                 }\
361                 break;\
362         case IMGFMT_BGR24:\
363                 func(uint8_t)\
364                         ((uint8_t*)dest)[0]= b[Y1];\
365                         ((uint8_t*)dest)[1]= g[Y1];\
366                         ((uint8_t*)dest)[2]= r[Y1];\
367                         ((uint8_t*)dest)[3]= b[Y2];\
368                         ((uint8_t*)dest)[4]= g[Y2];\
369                         ((uint8_t*)dest)[5]= r[Y2];\
370                         ((uint8_t*)dest)+=6;\
371                 }\
372                 break;\
373         case IMGFMT_RGB16:\
374         case IMGFMT_BGR16:\
375                 {\
376                         const int dr1= dither_2x2_8[y&1    ][0];\
377                         const int dg1= dither_2x2_4[y&1    ][0];\
378                         const int db1= dither_2x2_8[(y&1)^1][0];\
379                         const int dr2= dither_2x2_8[y&1    ][1];\
380                         const int dg2= dither_2x2_4[y&1    ][1];\
381                         const int db2= dither_2x2_8[(y&1)^1][1];\
382                         func(uint16_t)\
383                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
384                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
385                         }\
386                 }\
387                 break;\
388         case IMGFMT_RGB15:\
389         case IMGFMT_BGR15:\
390                 {\
391                         const int dr1= dither_2x2_8[y&1    ][0];\
392                         const int dg1= dither_2x2_8[y&1    ][1];\
393                         const int db1= dither_2x2_8[(y&1)^1][0];\
394                         const int dr2= dither_2x2_8[y&1    ][1];\
395                         const int dg2= dither_2x2_8[y&1    ][0];\
396                         const int db2= dither_2x2_8[(y&1)^1][1];\
397                         func(uint16_t)\
398                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
399                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
400                         }\
401                 }\
402                 break;\
403         case IMGFMT_RGB8:\
404         case IMGFMT_BGR8:\
405                 {\
406                         const uint8_t * const d64= dither_8x8_73[y&7];\
407                         const uint8_t * const d32= dither_8x8_32[y&7];\
408                         func(uint8_t)\
409                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
410                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
411                         }\
412                 }\
413                 break;\
414         case IMGFMT_RGB4:\
415         case IMGFMT_BGR4:\
416                 {\
417                         const uint8_t * const d64= dither_8x8_73 [y&7];\
418                         const uint8_t * const d128=dither_8x8_220[y&7];\
419                         func(uint8_t)\
420                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
421                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
422                         }\
423                 }\
424                 break;\
425         case IMGFMT_RG4B:\
426         case IMGFMT_BG4B:\
427                 {\
428                         const uint8_t * const d64= dither_8x8_73 [y&7];\
429                         const uint8_t * const d128=dither_8x8_220[y&7];\
430                         func(uint8_t)\
431                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
432                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
433                         }\
434                 }\
435                 break;\
436         case IMGFMT_RGB1:\
437         case IMGFMT_BGR1:\
438                 {\
439                         const uint8_t * const d128=dither_8x8_220[y&7];\
440                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
441                         for(i=0; i<dstW-7; i+=8){\
442                                 int acc;\
443                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
444                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
445                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
446                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
447                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
448                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
449                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
450                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
451                                 ((uint8_t*)dest)[0]= acc;\
452                                 ((uint8_t*)dest)++;\
453                         }\
454 \
455 /*\
456 ((uint8_t*)dest)-= dstW>>4;\
457 {\
458                         int acc=0;\
459                         int left=0;\
460                         static int top[1024];\
461                         static int last_new[1024][1024];\
462                         static int last_in3[1024][1024];\
463                         static int drift[1024][1024];\
464                         int topLeft=0;\
465                         int shift=0;\
466                         int count=0;\
467                         const uint8_t * const d128=dither_8x8_220[y&7];\
468                         int error_new=0;\
469                         int error_in3=0;\
470                         int f=0;\
471                         \
472                         for(i=dstW>>1; i<dstW; i++){\
473                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
474                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
475                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
476                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
477                                         + (last_new[y][i] - in3)*f/256;\
478                                 int new= old> 128 ? 255 : 0;\
479 \
480                                 error_new+= ABS(last_new[y][i] - new);\
481                                 error_in3+= ABS(last_in3[y][i] - in3);\
482                                 f= error_new - error_in3*4;\
483                                 if(f<0) f=0;\
484                                 if(f>256) f=256;\
485 \
486                                 topLeft= top[i];\
487                                 left= top[i]= old - new;\
488                                 last_new[y][i]= new;\
489                                 last_in3[y][i]= in3;\
490 \
491                                 acc+= acc + (new&1);\
492                                 if((i&7)==6){\
493                                         ((uint8_t*)dest)[0]= acc;\
494                                         ((uint8_t*)dest)++;\
495                                 }\
496                         }\
497 }\
498 */\
499                 }\
500                 break;\
501         case IMGFMT_YUY2:\
502                 func2\
503                         ((uint8_t*)dest)[2*i2+0]= Y1;\
504                         ((uint8_t*)dest)[2*i2+1]= U;\
505                         ((uint8_t*)dest)[2*i2+2]= Y2;\
506                         ((uint8_t*)dest)[2*i2+3]= V;\
507                 }               \
508                 break;\
509         case IMGFMT_UYVY:\
510                 func2\
511                         ((uint8_t*)dest)[2*i2+0]= U;\
512                         ((uint8_t*)dest)[2*i2+1]= Y1;\
513                         ((uint8_t*)dest)[2*i2+2]= V;\
514                         ((uint8_t*)dest)[2*i2+3]= Y2;\
515                 }               \
516                 break;\
517         }\
518
519
520 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
521                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
522                                     uint8_t *dest, int dstW, int y)
523 {
524         int i;
525         switch(c->dstFormat)
526         {
527         case IMGFMT_RGB32:
528         case IMGFMT_BGR32:
529                 YSCALE_YUV_2_RGBX_C(uint32_t)
530                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
531                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
532                 }
533                 break;
534         case IMGFMT_RGB24:
535                 YSCALE_YUV_2_RGBX_C(uint8_t)
536                         ((uint8_t*)dest)[0]= r[Y1];
537                         ((uint8_t*)dest)[1]= g[Y1];
538                         ((uint8_t*)dest)[2]= b[Y1];
539                         ((uint8_t*)dest)[3]= r[Y2];
540                         ((uint8_t*)dest)[4]= g[Y2];
541                         ((uint8_t*)dest)[5]= b[Y2];
542                         ((uint8_t*)dest)+=6;
543                 }
544                 break;
545         case IMGFMT_BGR24:
546                 YSCALE_YUV_2_RGBX_C(uint8_t)
547                         ((uint8_t*)dest)[0]= b[Y1];
548                         ((uint8_t*)dest)[1]= g[Y1];
549                         ((uint8_t*)dest)[2]= r[Y1];
550                         ((uint8_t*)dest)[3]= b[Y2];
551                         ((uint8_t*)dest)[4]= g[Y2];
552                         ((uint8_t*)dest)[5]= r[Y2];
553                         ((uint8_t*)dest)+=6;
554                 }
555                 break;
556         case IMGFMT_RGB16:
557         case IMGFMT_BGR16:
558                 {
559                         const int dr1= dither_2x2_8[y&1    ][0];
560                         const int dg1= dither_2x2_4[y&1    ][0];
561                         const int db1= dither_2x2_8[(y&1)^1][0];
562                         const int dr2= dither_2x2_8[y&1    ][1];
563                         const int dg2= dither_2x2_4[y&1    ][1];
564                         const int db2= dither_2x2_8[(y&1)^1][1];
565                         YSCALE_YUV_2_RGBX_C(uint16_t)
566                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
567                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
568                         }
569                 }
570                 break;
571         case IMGFMT_RGB15:
572         case IMGFMT_BGR15:
573                 {
574                         const int dr1= dither_2x2_8[y&1    ][0];
575                         const int dg1= dither_2x2_8[y&1    ][1];
576                         const int db1= dither_2x2_8[(y&1)^1][0];
577                         const int dr2= dither_2x2_8[y&1    ][1];
578                         const int dg2= dither_2x2_8[y&1    ][0];
579                         const int db2= dither_2x2_8[(y&1)^1][1];
580                         YSCALE_YUV_2_RGBX_C(uint16_t)
581                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
582                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
583                         }
584                 }
585                 break;
586         case IMGFMT_RGB8:
587         case IMGFMT_BGR8:
588                 {
589                         const uint8_t * const d64= dither_8x8_73[y&7];
590                         const uint8_t * const d32= dither_8x8_32[y&7];
591                         YSCALE_YUV_2_RGBX_C(uint8_t)
592                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
593                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
594                         }
595                 }
596                 break;
597         case IMGFMT_RGB4:
598         case IMGFMT_BGR4:
599                 {
600                         const uint8_t * const d64= dither_8x8_73 [y&7];
601                         const uint8_t * const d128=dither_8x8_220[y&7];
602                         YSCALE_YUV_2_RGBX_C(uint8_t)
603                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
604                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
605                         }
606                 }
607                 break;
608         case IMGFMT_RG4B:
609         case IMGFMT_BG4B:
610                 {
611                         const uint8_t * const d64= dither_8x8_73 [y&7];
612                         const uint8_t * const d128=dither_8x8_220[y&7];
613                         YSCALE_YUV_2_RGBX_C(uint8_t)
614                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
615                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
616                         }
617                 }
618                 break;
619         case IMGFMT_RGB1:
620         case IMGFMT_BGR1:
621                 {
622                         const uint8_t * const d128=dither_8x8_220[y&7];
623                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
624                         int acc=0;
625                         for(i=0; i<dstW-1; i+=2){
626                                 int j;
627                                 int Y1=1<<18;
628                                 int Y2=1<<18;
629
630                                 for(j=0; j<lumFilterSize; j++)
631                                 {
632                                         Y1 += lumSrc[j][i] * lumFilter[j];
633                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
634                                 }
635                                 Y1>>=19;
636                                 Y2>>=19;
637                                 if((Y1|Y2)&256)
638                                 {
639                                         if(Y1>255)   Y1=255;
640                                         else if(Y1<0)Y1=0;
641                                         if(Y2>255)   Y2=255;
642                                         else if(Y2<0)Y2=0;
643                                 }
644                                 acc+= acc + g[Y1+d128[(i+0)&7]];
645                                 acc+= acc + g[Y2+d128[(i+1)&7]];
646                                 if((i&7)==6){
647                                         ((uint8_t*)dest)[0]= acc;
648                                         ((uint8_t*)dest)++;
649                                 }
650                         }
651                 }
652                 break;
653         case IMGFMT_YUY2:
654                 YSCALE_YUV_2_PACKEDX_C(void)
655                         ((uint8_t*)dest)[2*i2+0]= Y1;
656                         ((uint8_t*)dest)[2*i2+1]= U;
657                         ((uint8_t*)dest)[2*i2+2]= Y2;
658                         ((uint8_t*)dest)[2*i2+3]= V;
659                 }
660                 break;
661         case IMGFMT_UYVY:
662                 YSCALE_YUV_2_PACKEDX_C(void)
663                         ((uint8_t*)dest)[2*i2+0]= U;
664                         ((uint8_t*)dest)[2*i2+1]= Y1;
665                         ((uint8_t*)dest)[2*i2+2]= V;
666                         ((uint8_t*)dest)[2*i2+3]= Y2;
667                 }
668                 break;
669         }
670 }
671
672
673 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
674 //Plain C versions
675 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
676 #define COMPILE_C
677 #endif
678
679 #ifdef ARCH_POWERPC
680 #ifdef HAVE_ALTIVEC
681 #define COMPILE_ALTIVEC
682 #endif //HAVE_ALTIVEC
683 #endif //ARCH_POWERPC
684
685 #ifdef ARCH_X86
686
687 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
688 #define COMPILE_MMX
689 #endif
690
691 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
692 #define COMPILE_MMX2
693 #endif
694
695 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
696 #define COMPILE_3DNOW
697 #endif
698 #endif //ARCH_X86
699
700 #undef HAVE_MMX
701 #undef HAVE_MMX2
702 #undef HAVE_3DNOW
703
704 #ifdef COMPILE_C
705 #undef HAVE_MMX
706 #undef HAVE_MMX2
707 #undef HAVE_3DNOW
708 #undef HAVE_ALTIVEC
709 #define RENAME(a) a ## _C
710 #include "swscale_template.c"
711 #endif
712
713 #ifdef ARCH_POWERPC
714 #ifdef COMPILE_ALTIVEC
715 #undef RENAME
716 #define HAVE_ALTIVEC
717 #define RENAME(a) a ## _altivec
718 #include "swscale_template.c"
719 #endif
720 #endif //ARCH_POWERPC
721
722 #ifdef ARCH_X86
723
724 //X86 versions
725 /*
726 #undef RENAME
727 #undef HAVE_MMX
728 #undef HAVE_MMX2
729 #undef HAVE_3DNOW
730 #define ARCH_X86
731 #define RENAME(a) a ## _X86
732 #include "swscale_template.c"
733 */
734 //MMX versions
735 #ifdef COMPILE_MMX
736 #undef RENAME
737 #define HAVE_MMX
738 #undef HAVE_MMX2
739 #undef HAVE_3DNOW
740 #define RENAME(a) a ## _MMX
741 #include "swscale_template.c"
742 #endif
743
744 //MMX2 versions
745 #ifdef COMPILE_MMX2
746 #undef RENAME
747 #define HAVE_MMX
748 #define HAVE_MMX2
749 #undef HAVE_3DNOW
750 #define RENAME(a) a ## _MMX2
751 #include "swscale_template.c"
752 #endif
753
754 //3DNOW versions
755 #ifdef COMPILE_3DNOW
756 #undef RENAME
757 #define HAVE_MMX
758 #undef HAVE_MMX2
759 #define HAVE_3DNOW
760 #define RENAME(a) a ## _3DNow
761 #include "swscale_template.c"
762 #endif
763
764 #endif //ARCH_X86
765
766 // minor note: the HAVE_xyz is messed up after that line so don't use it
767
768 static double getSplineCoeff(double a, double b, double c, double d, double dist)
769 {
770 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
771         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
772         else            return getSplineCoeff(  0.0, 
773                                                  b+ 2.0*c + 3.0*d,
774                                                         c + 3.0*d,
775                                                 -b- 3.0*c - 6.0*d,
776                                                 dist-1.0);
777 }
778
779 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
780                               int srcW, int dstW, int filterAlign, int one, int flags,
781                               SwsVector *srcFilter, SwsVector *dstFilter)
782 {
783         int i;
784         int filterSize;
785         int filter2Size;
786         int minFilterSize;
787         double *filter=NULL;
788         double *filter2=NULL;
789 #ifdef ARCH_X86
790         if(flags & SWS_CPU_CAPS_MMX)
791                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
792 #endif
793
794         // Note the +1 is for the MMXscaler which reads over the end
795         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
796
797         if(ABS(xInc - 0x10000) <10) // unscaled
798         {
799                 int i;
800                 filterSize= 1;
801                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
802                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
803
804                 for(i=0; i<dstW; i++)
805                 {
806                         filter[i*filterSize]=1;
807                         (*filterPos)[i]=i;
808                 }
809
810         }
811         else if(flags&SWS_POINT) // lame looking point sampling mode
812         {
813                 int i;
814                 int xDstInSrc;
815                 filterSize= 1;
816                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
817                 
818                 xDstInSrc= xInc/2 - 0x8000;
819                 for(i=0; i<dstW; i++)
820                 {
821                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
822
823                         (*filterPos)[i]= xx;
824                         filter[i]= 1.0;
825                         xDstInSrc+= xInc;
826                 }
827         }
828         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
829         {
830                 int i;
831                 int xDstInSrc;
832                 if     (flags&SWS_BICUBIC) filterSize= 4;
833                 else if(flags&SWS_X      ) filterSize= 4;
834                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
835                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
836
837                 xDstInSrc= xInc/2 - 0x8000;
838                 for(i=0; i<dstW; i++)
839                 {
840                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
841                         int j;
842
843                         (*filterPos)[i]= xx;
844                                 //Bilinear upscale / linear interpolate / Area averaging
845                                 for(j=0; j<filterSize; j++)
846                                 {
847                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
848                                         double coeff= 1.0 - d;
849                                         if(coeff<0) coeff=0;
850                                         filter[i*filterSize + j]= coeff;
851                                         xx++;
852                                 }
853                         xDstInSrc+= xInc;
854                 }
855         }
856         else
857         {
858                 double xDstInSrc;
859                 double sizeFactor, filterSizeInSrc;
860                 const double xInc1= (double)xInc / (double)(1<<16);
861                 int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
862
863                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
864                 else if(flags&SWS_X)            sizeFactor= 8.0;
865                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
866                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
867                 else if(flags&SWS_LANCZOS)      sizeFactor= param ? 2.0*param : 6.0;
868                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
869                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
870                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
871                 else {
872                         sizeFactor= 0.0; //GCC warning killer
873                         ASSERT(0)
874                 }
875                 
876                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
877                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
878
879                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
880                 if(filterSize > srcW-2) filterSize=srcW-2;
881
882                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
883
884                 xDstInSrc= xInc1 / 2.0 - 0.5;
885                 for(i=0; i<dstW; i++)
886                 {
887                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
888                         int j;
889                         (*filterPos)[i]= xx;
890                         for(j=0; j<filterSize; j++)
891                         {
892                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
893                                 double coeff;
894                                 if(flags & SWS_BICUBIC)
895                                 {
896                                         double A= param ? -param*0.01 : -0.60;
897                                         
898                                         // Equation is from VirtualDub
899                                         if(d<1.0)
900                                                 coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
901                                         else if(d<2.0)
902                                                 coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
903                                         else
904                                                 coeff=0.0;
905                                 }
906 /*                              else if(flags & SWS_X)
907                                 {
908                                         double p= param ? param*0.01 : 0.3;
909                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
910                                         coeff*= pow(2.0, - p*d*d);
911                                 }*/
912                                 else if(flags & SWS_X)
913                                 {
914                                         double A= param ? param*0.1 : 1.0;
915                                         
916                                         if(d<1.0)
917                                                 coeff = cos(d*PI);
918                                         else
919                                                 coeff=-1.0;
920                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
921                                         else            coeff=  pow( coeff, A);
922                                         coeff= coeff*0.5 + 0.5;
923                                 }
924                                 else if(flags & SWS_AREA)
925                                 {
926                                         double srcPixelSize= 1.0/xInc1;
927                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
928                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
929                                         else coeff=0.0;
930                                 }
931                                 else if(flags & SWS_GAUSS)
932                                 {
933                                         double p= param ? param*0.1 : 3.0;
934                                         coeff = pow(2.0, - p*d*d);
935                                 }
936                                 else if(flags & SWS_SINC)
937                                 {
938                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
939                                 }
940                                 else if(flags & SWS_LANCZOS)
941                                 {
942                                         double p= param ? param : 3.0; 
943                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
944                                         if(d>p) coeff=0;
945                                 }
946                                 else if(flags & SWS_BILINEAR)
947                                 {
948                                         coeff= 1.0 - d;
949                                         if(coeff<0) coeff=0;
950                                 }
951                                 else if(flags & SWS_SPLINE)
952                                 {
953                                         double p=-2.196152422706632;
954                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
955                                 }
956                                 else {
957                                         coeff= 0.0; //GCC warning killer
958                                         ASSERT(0)
959                                 }
960
961                                 filter[i*filterSize + j]= coeff;
962                                 xx++;
963                         }
964                         xDstInSrc+= xInc1;
965                 }
966         }
967
968         /* apply src & dst Filter to filter -> filter2
969            free(filter);
970         */
971         ASSERT(filterSize>0)
972         filter2Size= filterSize;
973         if(srcFilter) filter2Size+= srcFilter->length - 1;
974         if(dstFilter) filter2Size+= dstFilter->length - 1;
975         ASSERT(filter2Size>0)
976         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
977
978         for(i=0; i<dstW; i++)
979         {
980                 int j;
981                 SwsVector scaleFilter;
982                 SwsVector *outVec;
983
984                 scaleFilter.coeff= filter + i*filterSize;
985                 scaleFilter.length= filterSize;
986
987                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
988                 else          outVec= &scaleFilter;
989
990                 ASSERT(outVec->length == filter2Size)
991                 //FIXME dstFilter
992
993                 for(j=0; j<outVec->length; j++)
994                 {
995                         filter2[i*filter2Size + j]= outVec->coeff[j];
996                 }
997
998                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
999
1000                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1001         }
1002         free(filter); filter=NULL;
1003
1004         /* try to reduce the filter-size (step1 find size and shift left) */
1005         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1006         minFilterSize= 0;
1007         for(i=dstW-1; i>=0; i--)
1008         {
1009                 int min= filter2Size;
1010                 int j;
1011                 double cutOff=0.0;
1012
1013                 /* get rid off near zero elements on the left by shifting left */
1014                 for(j=0; j<filter2Size; j++)
1015                 {
1016                         int k;
1017                         cutOff += ABS(filter2[i*filter2Size]);
1018
1019                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1020
1021                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1022                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1023
1024                         // Move filter coeffs left
1025                         for(k=1; k<filter2Size; k++)
1026                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1027                         filter2[i*filter2Size + k - 1]= 0.0;
1028                         (*filterPos)[i]++;
1029                 }
1030
1031                 cutOff=0.0;
1032                 /* count near zeros on the right */
1033                 for(j=filter2Size-1; j>0; j--)
1034                 {
1035                         cutOff += ABS(filter2[i*filter2Size + j]);
1036
1037                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1038                         min--;
1039                 }
1040
1041                 if(min>minFilterSize) minFilterSize= min;
1042         }
1043
1044         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1045           // we can handle the special case 4,
1046           // so we don't want to go to the full 8
1047           if (minFilterSize < 5)
1048             filterAlign = 4;
1049
1050           // we really don't want to waste our time
1051           // doing useless computation, so fall-back on
1052           // the scalar C code for very small filter.
1053           // vectorizing is worth it only if you have
1054           // decent-sized vector.
1055           if (minFilterSize < 3)
1056             filterAlign = 1;
1057         }
1058
1059         ASSERT(minFilterSize > 0)
1060         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1061         ASSERT(filterSize > 0)
1062         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1063         *outFilterSize= filterSize;
1064
1065         if(flags&SWS_PRINT_INFO)
1066                 MSG_INFO("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1067         /* try to reduce the filter-size (step2 reduce it) */
1068         for(i=0; i<dstW; i++)
1069         {
1070                 int j;
1071
1072                 for(j=0; j<filterSize; j++)
1073                 {
1074                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1075                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1076                 }
1077         }
1078         free(filter2); filter2=NULL;
1079         
1080
1081         //FIXME try to align filterpos if possible
1082
1083         //fix borders
1084         for(i=0; i<dstW; i++)
1085         {
1086                 int j;
1087                 if((*filterPos)[i] < 0)
1088                 {
1089                         // Move filter coeffs left to compensate for filterPos
1090                         for(j=1; j<filterSize; j++)
1091                         {
1092                                 int left= MAX(j + (*filterPos)[i], 0);
1093                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1094                                 filter[i*filterSize + j]=0;
1095                         }
1096                         (*filterPos)[i]= 0;
1097                 }
1098
1099                 if((*filterPos)[i] + filterSize > srcW)
1100                 {
1101                         int shift= (*filterPos)[i] + filterSize - srcW;
1102                         // Move filter coeffs right to compensate for filterPos
1103                         for(j=filterSize-2; j>=0; j--)
1104                         {
1105                                 int right= MIN(j + shift, filterSize-1);
1106                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1107                                 filter[i*filterSize +j]=0;
1108                         }
1109                         (*filterPos)[i]= srcW - filterSize;
1110                 }
1111         }
1112
1113         // Note the +1 is for the MMXscaler which reads over the end
1114         *outFilter= (int16_t*)memalign(8, *outFilterSize*(dstW+1)*sizeof(int16_t));
1115         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1116
1117         /* Normalize & Store in outFilter */
1118         for(i=0; i<dstW; i++)
1119         {
1120                 int j;
1121                 double error=0;
1122                 double sum=0;
1123                 double scale= one;
1124
1125                 for(j=0; j<filterSize; j++)
1126                 {
1127                         sum+= filter[i*filterSize + j];
1128                 }
1129                 scale/= sum;
1130                 for(j=0; j<*outFilterSize; j++)
1131                 {
1132                         double v= filter[i*filterSize + j]*scale + error;
1133                         int intV= floor(v + 0.5);
1134                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1135                         error = v - intV;
1136                 }
1137         }
1138         
1139         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1140         for(i=0; i<*outFilterSize; i++)
1141         {
1142                 int j= dstW*(*outFilterSize);
1143                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1144         }
1145
1146         free(filter);
1147 }
1148
1149 #ifdef ARCH_X86
1150 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1151 {
1152         uint8_t *fragmentA;
1153         int imm8OfPShufW1A;
1154         int imm8OfPShufW2A;
1155         int fragmentLengthA;
1156         uint8_t *fragmentB;
1157         int imm8OfPShufW1B;
1158         int imm8OfPShufW2B;
1159         int fragmentLengthB;
1160         int fragmentPos;
1161
1162         int xpos, i;
1163
1164         // create an optimized horizontal scaling routine
1165
1166         //code fragment
1167
1168         asm volatile(
1169                 "jmp 9f                         \n\t"
1170         // Begin
1171                 "0:                             \n\t"
1172                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1173                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1174                 "movd 1(%%ecx, %%esi), %%mm1    \n\t"
1175                 "punpcklbw %%mm7, %%mm1         \n\t"
1176                 "punpcklbw %%mm7, %%mm0         \n\t"
1177                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1178                 "1:                             \n\t"
1179                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1180                 "2:                             \n\t"
1181                 "psubw %%mm1, %%mm0             \n\t"
1182                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1183                 "pmullw %%mm3, %%mm0            \n\t"
1184                 "psllw $7, %%mm1                \n\t"
1185                 "paddw %%mm1, %%mm0             \n\t"
1186
1187                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1188
1189                 "addl $8, %%eax                 \n\t"
1190         // End
1191                 "9:                             \n\t"
1192 //              "int $3\n\t"
1193                 "leal 0b, %0                    \n\t"
1194                 "leal 1b, %1                    \n\t"
1195                 "leal 2b, %2                    \n\t"
1196                 "decl %1                        \n\t"
1197                 "decl %2                        \n\t"
1198                 "subl %0, %1                    \n\t"
1199                 "subl %0, %2                    \n\t"
1200                 "leal 9b, %3                    \n\t"
1201                 "subl %0, %3                    \n\t"
1202
1203
1204                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1205                 "=r" (fragmentLengthA)
1206         );
1207
1208         asm volatile(
1209                 "jmp 9f                         \n\t"
1210         // Begin
1211                 "0:                             \n\t"
1212                 "movq (%%edx, %%eax), %%mm3     \n\t" 
1213                 "movd (%%ecx, %%esi), %%mm0     \n\t" 
1214                 "punpcklbw %%mm7, %%mm0         \n\t"
1215                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1216                 "1:                             \n\t"
1217                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1218                 "2:                             \n\t"
1219                 "psubw %%mm1, %%mm0             \n\t"
1220                 "movl 8(%%ebx, %%eax), %%esi    \n\t"
1221                 "pmullw %%mm3, %%mm0            \n\t"
1222                 "psllw $7, %%mm1                \n\t"
1223                 "paddw %%mm1, %%mm0             \n\t"
1224
1225                 "movq %%mm0, (%%edi, %%eax)     \n\t"
1226
1227                 "addl $8, %%eax                 \n\t"
1228         // End
1229                 "9:                             \n\t"
1230 //              "int $3\n\t"
1231                 "leal 0b, %0                    \n\t"
1232                 "leal 1b, %1                    \n\t"
1233                 "leal 2b, %2                    \n\t"
1234                 "decl %1                        \n\t"
1235                 "decl %2                        \n\t"
1236                 "subl %0, %1                    \n\t"
1237                 "subl %0, %2                    \n\t"
1238                 "leal 9b, %3                    \n\t"
1239                 "subl %0, %3                    \n\t"
1240
1241
1242                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1243                 "=r" (fragmentLengthB)
1244         );
1245
1246         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1247         fragmentPos=0;
1248         
1249         for(i=0; i<dstW/numSplits; i++)
1250         {
1251                 int xx=xpos>>16;
1252
1253                 if((i&3) == 0)
1254                 {
1255                         int a=0;
1256                         int b=((xpos+xInc)>>16) - xx;
1257                         int c=((xpos+xInc*2)>>16) - xx;
1258                         int d=((xpos+xInc*3)>>16) - xx;
1259
1260                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1261                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1262                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1263                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1264                         filterPos[i/2]= xx;
1265
1266                         if(d+1<4)
1267                         {
1268                                 int maxShift= 3-(d+1);
1269                                 int shift=0;
1270
1271                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1272
1273                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1274                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1275                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1276                                         a | (b<<2) | (c<<4) | (d<<6);
1277
1278                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1279                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1280
1281                                 if(shift && i>=shift)
1282                                 {
1283                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1284                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1285                                         filterPos[i/2]-=shift;
1286                                 }
1287
1288                                 fragmentPos+= fragmentLengthB;
1289                         }
1290                         else
1291                         {
1292                                 int maxShift= 3-d;
1293                                 int shift=0;
1294
1295                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1296
1297                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1298                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1299                                         a | (b<<2) | (c<<4) | (d<<6);
1300
1301                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1302                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1303
1304                                 if(shift && i>=shift)
1305                                 {
1306                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1307                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1308                                         filterPos[i/2]-=shift;
1309                                 }
1310
1311                                 fragmentPos+= fragmentLengthA;
1312                         }
1313
1314                         funnyCode[fragmentPos]= RET;
1315                 }
1316                 xpos+=xInc;
1317         }
1318         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1319 }
1320 #endif // ARCH_X86
1321
1322 static void globalInit(){
1323     // generating tables:
1324     int i;
1325     for(i=0; i<768; i++){
1326         int c= MIN(MAX(i-256, 0), 255);
1327         clip_table[i]=c;
1328     }
1329 }
1330
1331 static SwsFunc getSwsFunc(int flags){
1332     
1333 #ifdef RUNTIME_CPUDETECT
1334 #ifdef ARCH_X86
1335         // ordered per speed fasterst first
1336         if(flags & SWS_CPU_CAPS_MMX2)
1337                 return swScale_MMX2;
1338         else if(flags & SWS_CPU_CAPS_3DNOW)
1339                 return swScale_3DNow;
1340         else if(flags & SWS_CPU_CAPS_MMX)
1341                 return swScale_MMX;
1342         else
1343                 return swScale_C;
1344
1345 #else
1346 #ifdef ARCH_POWERPC
1347         if(flags & SWS_CPU_CAPS_ALTIVEC)
1348           return swScale_altivec;
1349         else
1350           return swScale_C;
1351 #endif
1352         return swScale_C;
1353 #endif
1354 #else //RUNTIME_CPUDETECT
1355 #ifdef HAVE_MMX2
1356         return swScale_MMX2;
1357 #elif defined (HAVE_3DNOW)
1358         return swScale_3DNow;
1359 #elif defined (HAVE_MMX)
1360         return swScale_MMX;
1361 #elif defined (HAVE_ALTIVEC)
1362         return swScale_altivec;
1363 #else
1364         return swScale_C;
1365 #endif
1366 #endif //!RUNTIME_CPUDETECT
1367 }
1368
1369 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1370              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1371         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1372         /* Copy Y plane */
1373         if(dstStride[0]==srcStride[0])
1374                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1375         else
1376         {
1377                 int i;
1378                 uint8_t *srcPtr= src[0];
1379                 uint8_t *dstPtr= dst;
1380                 for(i=0; i<srcSliceH; i++)
1381                 {
1382                         memcpy(dstPtr, srcPtr, srcStride[0]);
1383                         srcPtr+= srcStride[0];
1384                         dstPtr+= dstStride[0];
1385                 }
1386         }
1387         dst = dstParam[1] + dstStride[1]*srcSliceY;
1388         interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
1389
1390         return srcSliceH;
1391 }
1392
1393 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1394              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1395         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1396
1397         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1398
1399         return srcSliceH;
1400 }
1401
1402 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1403              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1404         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1405
1406         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1407
1408         return srcSliceH;
1409 }
1410
1411 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1412 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1413                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1414         const int srcFormat= c->srcFormat;
1415         const int dstFormat= c->dstFormat;
1416         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1417         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1418         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1419         const int dstId= (dstFormat&0xFF)>>2;
1420         void (*conv)(const uint8_t *src, uint8_t *dst, unsigned src_size)=NULL;
1421
1422         /* BGR -> BGR */
1423         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1424            || (isRGB(srcFormat) && isRGB(dstFormat))){
1425                 switch(srcId | (dstId<<4)){
1426                 case 0x34: conv= rgb16to15; break;
1427                 case 0x36: conv= rgb24to15; break;
1428                 case 0x38: conv= rgb32to15; break;
1429                 case 0x43: conv= rgb15to16; break;
1430                 case 0x46: conv= rgb24to16; break;
1431                 case 0x48: conv= rgb32to16; break;
1432                 case 0x63: conv= rgb15to24; break;
1433                 case 0x64: conv= rgb16to24; break;
1434                 case 0x68: conv= rgb32to24; break;
1435                 case 0x83: conv= rgb15to32; break;
1436                 case 0x84: conv= rgb16to32; break;
1437                 case 0x86: conv= rgb24to32; break;
1438                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1439                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1440                 }
1441         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1442                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1443                 switch(srcId | (dstId<<4)){
1444                 case 0x33: conv= rgb15tobgr15; break;
1445                 case 0x34: conv= rgb16tobgr15; break;
1446                 case 0x36: conv= rgb24tobgr15; break;
1447                 case 0x38: conv= rgb32tobgr15; break;
1448                 case 0x43: conv= rgb15tobgr16; break;
1449                 case 0x44: conv= rgb16tobgr16; break;
1450                 case 0x46: conv= rgb24tobgr16; break;
1451                 case 0x48: conv= rgb32tobgr16; break;
1452                 case 0x63: conv= rgb15tobgr24; break;
1453                 case 0x64: conv= rgb16tobgr24; break;
1454                 case 0x66: conv= rgb24tobgr24; break;
1455                 case 0x68: conv= rgb32tobgr24; break;
1456                 case 0x83: conv= rgb15tobgr32; break;
1457                 case 0x84: conv= rgb16tobgr32; break;
1458                 case 0x86: conv= rgb24tobgr32; break;
1459                 case 0x88: conv= rgb32tobgr32; break;
1460                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1461                                  vo_format_name(srcFormat), vo_format_name(dstFormat)); break;
1462                 }
1463         }else{
1464                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1465                          vo_format_name(srcFormat), vo_format_name(dstFormat));
1466         }
1467
1468         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1469                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1470         else
1471         {
1472                 int i;
1473                 uint8_t *srcPtr= src[0];
1474                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1475
1476                 for(i=0; i<srcSliceH; i++)
1477                 {
1478                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1479                         srcPtr+= srcStride[0];
1480                         dstPtr+= dstStride[0];
1481                 }
1482         }     
1483         return srcSliceH;
1484 }
1485
1486 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1487              int srcSliceH, uint8_t* dst[], int dstStride[]){
1488
1489         rgb24toyv12(
1490                 src[0], 
1491                 dst[0]+ srcSliceY    *dstStride[0], 
1492                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1493                 dst[2]+(srcSliceY>>1)*dstStride[2],
1494                 c->srcW, srcSliceH, 
1495                 dstStride[0], dstStride[1], srcStride[0]);
1496         return srcSliceH;
1497 }
1498
1499 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1500              int srcSliceH, uint8_t* dst[], int dstStride[]){
1501         int i;
1502
1503         /* copy Y */
1504         if(srcStride[0]==dstStride[0]) 
1505                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1506         else{
1507                 uint8_t *srcPtr= src[0];
1508                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1509
1510                 for(i=0; i<srcSliceH; i++)
1511                 {
1512                         memcpy(dstPtr, srcPtr, c->srcW);
1513                         srcPtr+= srcStride[0];
1514                         dstPtr+= dstStride[0];
1515                 }
1516         }
1517
1518         if(c->dstFormat==IMGFMT_YV12){
1519                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1520                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1521         }else{
1522                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1523                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1524         }
1525         return srcSliceH;
1526 }
1527
1528 /**
1529  * bring pointers in YUV order instead of YVU
1530  */
1531 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1532         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1533            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1534                 sortedP[0]= p[0];
1535                 sortedP[1]= p[2];
1536                 sortedP[2]= p[1];
1537                 sortedStride[0]= stride[0];
1538                 sortedStride[1]= stride[2];
1539                 sortedStride[2]= stride[1];
1540         }
1541         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1542         {
1543                 sortedP[0]= p[0];
1544                 sortedP[1]= 
1545                 sortedP[2]= NULL;
1546                 sortedStride[0]= stride[0];
1547                 sortedStride[1]= 
1548                 sortedStride[2]= 0;
1549         }
1550         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1551         {
1552                 sortedP[0]= p[0];
1553                 sortedP[1]= p[1];
1554                 sortedP[2]= p[2];
1555                 sortedStride[0]= stride[0];
1556                 sortedStride[1]= stride[1];
1557                 sortedStride[2]= stride[2];
1558         }else{
1559                 MSG_ERR("internal error in orderYUV\n");
1560         }
1561 }
1562
1563 /* unscaled copy like stuff (assumes nearly identical formats) */
1564 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1565              int srcSliceH, uint8_t* dst[], int dstStride[]){
1566
1567         if(isPacked(c->srcFormat))
1568         {
1569                 if(dstStride[0]==srcStride[0])
1570                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1571                 else
1572                 {
1573                         int i;
1574                         uint8_t *srcPtr= src[0];
1575                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1576                         int length=0;
1577
1578                         /* universal length finder */
1579                         while(length+c->srcW <= ABS(dstStride[0]) 
1580                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1581                         ASSERT(length!=0);
1582
1583                         for(i=0; i<srcSliceH; i++)
1584                         {
1585                                 memcpy(dstPtr, srcPtr, length);
1586                                 srcPtr+= srcStride[0];
1587                                 dstPtr+= dstStride[0];
1588                         }
1589                 }
1590         }
1591         else 
1592         { /* Planar YUV or gray */
1593                 int plane;
1594                 for(plane=0; plane<3; plane++)
1595                 {
1596                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1597                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1598                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1599
1600                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1601                         {
1602                                 if(!isGray(c->dstFormat))
1603                                         memset(dst[plane], 128, dstStride[plane]*height);
1604                         }
1605                         else
1606                         {
1607                                 if(dstStride[plane]==srcStride[plane])
1608                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1609                                 else
1610                                 {
1611                                         int i;
1612                                         uint8_t *srcPtr= src[plane];
1613                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1614                                         for(i=0; i<height; i++)
1615                                         {
1616                                                 memcpy(dstPtr, srcPtr, length);
1617                                                 srcPtr+= srcStride[plane];
1618                                                 dstPtr+= dstStride[plane];
1619                                         }
1620                                 }
1621                         }
1622                 }
1623         }
1624         return srcSliceH;
1625 }
1626
1627 static int remove_dup_fourcc(int fourcc)
1628 {
1629         switch(fourcc)
1630         {
1631             case IMGFMT_I420:
1632             case IMGFMT_IYUV: return IMGFMT_YV12;
1633             case IMGFMT_Y8  : return IMGFMT_Y800;
1634             case IMGFMT_IF09: return IMGFMT_YVU9;
1635             default: return fourcc;
1636         }
1637 }
1638
1639 static void getSubSampleFactors(int *h, int *v, int format){
1640         switch(format){
1641         case IMGFMT_UYVY:
1642         case IMGFMT_YUY2:
1643                 *h=1;
1644                 *v=0;
1645                 break;
1646         case IMGFMT_YV12:
1647         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1648                 *h=1;
1649                 *v=1;
1650                 break;
1651         case IMGFMT_YVU9:
1652                 *h=2;
1653                 *v=2;
1654                 break;
1655         case IMGFMT_444P:
1656                 *h=0;
1657                 *v=0;
1658                 break;
1659         case IMGFMT_422P:
1660                 *h=1;
1661                 *v=0;
1662                 break;
1663         case IMGFMT_411P:
1664                 *h=2;
1665                 *v=0;
1666                 break;
1667         default:
1668                 *h=0;
1669                 *v=0;
1670                 break;
1671         }
1672 }
1673
1674 static uint16_t roundToInt16(int64_t f){
1675         int r= (f + (1<<15))>>16;
1676              if(r<-0x7FFF) return 0x8000;
1677         else if(r> 0x7FFF) return 0x7FFF;
1678         else               return r;
1679 }
1680
1681 /**
1682  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1683  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1684  * @return -1 if not supported
1685  */
1686 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1687         int64_t crv =  inv_table[0];
1688         int64_t cbu =  inv_table[1];
1689         int64_t cgu = -inv_table[2];
1690         int64_t cgv = -inv_table[3];
1691         int64_t cy  = 1<<16;
1692         int64_t oy  = 0;
1693
1694         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1695         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1696         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1697
1698         c->brightness= brightness;
1699         c->contrast  = contrast;
1700         c->saturation= saturation;
1701         c->srcRange  = srcRange;
1702         c->dstRange  = dstRange;
1703
1704         c->uOffset=   0x0400040004000400LL;
1705         c->vOffset=   0x0400040004000400LL;
1706
1707         if(!srcRange){
1708                 cy= (cy*255) / 219;
1709                 oy= 16<<16;
1710         }
1711
1712         cy = (cy *contrast             )>>16;
1713         crv= (crv*contrast * saturation)>>32;
1714         cbu= (cbu*contrast * saturation)>>32;
1715         cgu= (cgu*contrast * saturation)>>32;
1716         cgv= (cgv*contrast * saturation)>>32;
1717
1718         oy -= 256*brightness;
1719
1720         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1721         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1722         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1723         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1724         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1725         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1726
1727         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1728         //FIXME factorize
1729         
1730         return 0;
1731 }
1732
1733 /**
1734  * @return -1 if not supported
1735  */
1736 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1737         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1738
1739         *inv_table = c->srcColorspaceTable;
1740         *table     = c->dstColorspaceTable;
1741         *srcRange  = c->srcRange;
1742         *dstRange  = c->dstRange;
1743         *brightness= c->brightness;
1744         *contrast  = c->contrast;
1745         *saturation= c->saturation;
1746         
1747         return 0;       
1748 }
1749
1750 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1751                          SwsFilter *srcFilter, SwsFilter *dstFilter){
1752
1753         SwsContext *c;
1754         int i;
1755         int usesVFilter, usesHFilter;
1756         int unscaled, needsDither;
1757         int srcFormat, dstFormat;
1758         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1759 #ifdef ARCH_X86
1760         if(flags & SWS_CPU_CAPS_MMX)
1761                 asm volatile("emms\n\t"::: "memory");
1762 #endif
1763
1764 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1765         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1766 #ifdef HAVE_MMX2
1767         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1768 #elif defined (HAVE_3DNOW)
1769         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1770 #elif defined (HAVE_MMX)
1771         flags |= SWS_CPU_CAPS_MMX;
1772 #elif defined (HAVE_ALTIVEC)
1773         flags |= SWS_CPU_CAPS_ALTIVEC;
1774 #endif
1775 #endif
1776         if(clip_table[512] != 255) globalInit();
1777         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1778
1779         /* avoid duplicate Formats, so we don't need to check to much */
1780         srcFormat = remove_dup_fourcc(origSrcFormat);
1781         dstFormat = remove_dup_fourcc(origDstFormat);
1782
1783         unscaled = (srcW == dstW && srcH == dstH);
1784         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1785                      && (dstFormat&0xFF)<24
1786                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1787
1788         if(!isSupportedIn(srcFormat)) 
1789         {
1790                 MSG_ERR("swScaler: %s is not supported as input format\n", vo_format_name(srcFormat));
1791                 return NULL;
1792         }
1793         if(!isSupportedOut(dstFormat))
1794         {
1795                 MSG_ERR("swScaler: %s is not supported as output format\n", vo_format_name(dstFormat));
1796                 return NULL;
1797         }
1798
1799         /* sanity check */
1800         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
1801         {
1802                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1803                         srcW, srcH, dstW, dstH);
1804                 return NULL;
1805         }
1806
1807         if(!dstFilter) dstFilter= &dummyFilter;
1808         if(!srcFilter) srcFilter= &dummyFilter;
1809
1810         c= memalign(64, sizeof(SwsContext));
1811         memset(c, 0, sizeof(SwsContext));
1812
1813         c->srcW= srcW;
1814         c->srcH= srcH;
1815         c->dstW= dstW;
1816         c->dstH= dstH;
1817         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1818         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1819         c->flags= flags;
1820         c->dstFormat= dstFormat;
1821         c->srcFormat= srcFormat;
1822         c->origDstFormat= origDstFormat;
1823         c->origSrcFormat= origSrcFormat;
1824         c->vRounder= 4* 0x0001000100010001ULL;
1825
1826         usesHFilter= usesVFilter= 0;
1827         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1828         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1829         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1830         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1831         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1832         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1833         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1834         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1835
1836         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1837         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1838
1839         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1840         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1841
1842         // drop some chroma lines if the user wants it
1843         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1844         c->chrSrcVSubSample+= c->vChrDrop;
1845
1846         // drop every 2. pixel for chroma calculation unless user wants full chroma
1847         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1848                 c->chrSrcHSubSample=1;
1849
1850         c->chrIntHSubSample= c->chrDstHSubSample;
1851         c->chrIntVSubSample= c->chrSrcVSubSample;
1852
1853         // note the -((-x)>>y) is so that we allways round toward +inf
1854         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1855         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1856         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1857         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1858
1859         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); 
1860
1861         /* unscaled special Cases */
1862         if(unscaled && !usesHFilter && !usesVFilter)
1863         {
1864                 /* yv12_to_nv12 */
1865                 if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
1866                 {
1867                         c->swScale= PlanarToNV12Wrapper;
1868                 }
1869                 /* yuv2bgr */
1870                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1871                 {
1872                         c->swScale= yuv2rgb_get_func_ptr(c);
1873                 }
1874                 
1875                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1876                 {
1877                         c->swScale= yvu9toyv12Wrapper;
1878                 }
1879
1880                 /* bgr24toYV12 */
1881                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1882                         c->swScale= bgr24toyv12Wrapper;
1883                 
1884                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1885                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1886                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
1887                    && !needsDither)
1888                         c->swScale= rgb2rgbWrapper;
1889
1890                 /* LQ converters if -sws 0 or -sws 4*/
1891                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1892                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1893                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1894                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
1895                           && needsDither)
1896                                 c->swScale= rgb2rgbWrapper;
1897
1898                         /* yv12_to_yuy2 */
1899                         if(srcFormat == IMGFMT_YV12 && 
1900                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1901                         {
1902                                 if (dstFormat == IMGFMT_YUY2)
1903                                     c->swScale= PlanarToYuy2Wrapper;
1904                                 else
1905                                     c->swScale= PlanarToUyvyWrapper;
1906                         }
1907                 }
1908
1909                 /* simple copy */
1910                 if(   srcFormat == dstFormat
1911                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
1912                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
1913                   )
1914                 {
1915                         c->swScale= simpleCopy;
1916                 }
1917
1918                 if(c->swScale){
1919                         if(flags&SWS_PRINT_INFO)
1920                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
1921                                         vo_format_name(srcFormat), vo_format_name(dstFormat));
1922                         return c;
1923                 }
1924         }
1925
1926         if(flags & SWS_CPU_CAPS_MMX2)
1927         {
1928                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
1929                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
1930                 {
1931                         if(flags&SWS_PRINT_INFO)
1932                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
1933                 }
1934                 if(usesHFilter) c->canMMX2BeUsed=0;
1935         }
1936         else
1937                 c->canMMX2BeUsed=0;
1938
1939         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
1940         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
1941
1942         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
1943         // but only for the FAST_BILINEAR mode otherwise do correct scaling
1944         // n-2 is the last chrominance sample available
1945         // this is not perfect, but noone shuld notice the difference, the more correct variant
1946         // would be like the vertical one, but that would require some special code for the
1947         // first and last pixel
1948         if(flags&SWS_FAST_BILINEAR)
1949         {
1950                 if(c->canMMX2BeUsed)
1951                 {
1952                         c->lumXInc+= 20;
1953                         c->chrXInc+= 20;
1954                 }
1955                 //we don't use the x86asm scaler if mmx is available
1956                 else if(flags & SWS_CPU_CAPS_MMX)
1957                 {
1958                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
1959                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
1960                 }
1961         }
1962
1963         /* precalculate horizontal scaler filter coefficients */
1964         {
1965                 const int filterAlign=
1966                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
1967                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1968                   1;
1969
1970                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
1971                                  srcW      ,       dstW, filterAlign, 1<<14,
1972                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1973                                  srcFilter->lumH, dstFilter->lumH);
1974                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
1975                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1976                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1977                                  srcFilter->chrH, dstFilter->chrH);
1978
1979 #ifdef ARCH_X86
1980 // can't downscale !!!
1981                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
1982                 {
1983                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
1984                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
1985                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
1986                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
1987
1988                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
1989                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
1990                 }
1991 #endif
1992         } // Init Horizontal stuff
1993
1994
1995
1996         /* precalculate vertical scaler filter coefficients */
1997         {
1998                 const int filterAlign=
1999                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2000                   1;
2001
2002                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2003                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2004                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2005                                 srcFilter->lumV, dstFilter->lumV);
2006                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2007                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2008                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2009                                 srcFilter->chrV, dstFilter->chrV);
2010         }
2011
2012         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2013         c->vLumBufSize= c->vLumFilterSize;
2014         c->vChrBufSize= c->vChrFilterSize;
2015         for(i=0; i<dstH; i++)
2016         {
2017                 int chrI= i*c->chrDstH / dstH;
2018                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2019                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2020
2021                 nextSlice>>= c->chrSrcVSubSample;
2022                 nextSlice<<= c->chrSrcVSubSample;
2023                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2024                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2025                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2026                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2027         }
2028
2029         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2030         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2031         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2032         //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)
2033         for(i=0; i<c->vLumBufSize; i++)
2034                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
2035         for(i=0; i<c->vChrBufSize; i++)
2036                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
2037
2038         //try to avoid drawing green stuff between the right end and the stride end
2039         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2040         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2041
2042         ASSERT(c->chrDstH <= dstH)
2043
2044         if(flags&SWS_PRINT_INFO)
2045         {
2046 #ifdef DITHER1XBPP
2047                 char *dither= " dithered";
2048 #else
2049                 char *dither= "";
2050 #endif
2051                 if(flags&SWS_FAST_BILINEAR)
2052                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2053                 else if(flags&SWS_BILINEAR)
2054                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2055                 else if(flags&SWS_BICUBIC)
2056                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2057                 else if(flags&SWS_X)
2058                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2059                 else if(flags&SWS_POINT)
2060                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2061                 else if(flags&SWS_AREA)
2062                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2063                 else if(flags&SWS_BICUBLIN)
2064                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2065                 else if(flags&SWS_GAUSS)
2066                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2067                 else if(flags&SWS_SINC)
2068                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2069                 else if(flags&SWS_LANCZOS)
2070                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2071                 else if(flags&SWS_SPLINE)
2072                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2073                 else
2074                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2075
2076                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2077                         MSG_INFO("from %s to%s %s ", 
2078                                 vo_format_name(srcFormat), dither, vo_format_name(dstFormat));
2079                 else
2080                         MSG_INFO("from %s to %s ", 
2081                                 vo_format_name(srcFormat), vo_format_name(dstFormat));
2082
2083                 if(flags & SWS_CPU_CAPS_MMX2)
2084                         MSG_INFO("using MMX2\n");
2085                 else if(flags & SWS_CPU_CAPS_3DNOW)
2086                         MSG_INFO("using 3DNOW\n");
2087                 else if(flags & SWS_CPU_CAPS_MMX)
2088                         MSG_INFO("using MMX\n");
2089                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2090                         MSG_INFO("using AltiVec\n");
2091                 else 
2092                         MSG_INFO("using C\n");
2093         }
2094
2095         if(flags & SWS_PRINT_INFO)
2096         {
2097                 if(flags & SWS_CPU_CAPS_MMX)
2098                 {
2099                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2100                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2101                         else
2102                         {
2103                                 if(c->hLumFilterSize==4)
2104                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2105                                 else if(c->hLumFilterSize==8)
2106                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2107                                 else
2108                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2109
2110                                 if(c->hChrFilterSize==4)
2111                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2112                                 else if(c->hChrFilterSize==8)
2113                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2114                                 else
2115                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2116                         }
2117                 }
2118                 else
2119                 {
2120 #ifdef ARCH_X86
2121                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2122 #else
2123                         if(flags & SWS_FAST_BILINEAR)
2124                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2125                         else
2126                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2127 #endif
2128                 }
2129                 if(isPlanarYUV(dstFormat))
2130                 {
2131                         if(c->vLumFilterSize==1)
2132                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2133                         else
2134                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2135                 }
2136                 else
2137                 {
2138                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2139                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2140                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2141                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2142                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2143                         else
2144                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2145                 }
2146
2147                 if(dstFormat==IMGFMT_BGR24)
2148                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2149                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2150                 else if(dstFormat==IMGFMT_BGR32)
2151                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2152                 else if(dstFormat==IMGFMT_BGR16)
2153                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2154                 else if(dstFormat==IMGFMT_BGR15)
2155                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2156
2157                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2158         }
2159         if(flags & SWS_PRINT_INFO)
2160         {
2161                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2162                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2163                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2164                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2165         }
2166
2167         c->swScale= getSwsFunc(flags);
2168         return c;
2169 }
2170
2171 /**
2172  * swscale warper, so we don't need to export the SwsContext.
2173  * assumes planar YUV to be in YUV order instead of YVU
2174  */
2175 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2176                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2177         //copy strides, so they can safely be modified
2178         int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2179         int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2180         return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2181 }
2182
2183 /**
2184  * swscale warper, so we don't need to export the SwsContext
2185  */
2186 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2187                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2188         int srcStride[3];
2189         int dstStride[3];
2190         uint8_t *src[3];
2191         uint8_t *dst[3];
2192         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2193         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2194 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2195
2196         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2197 }
2198
2199 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2200                                 float lumaSharpen, float chromaSharpen,
2201                                 float chromaHShift, float chromaVShift,
2202                                 int verbose)
2203 {
2204         SwsFilter *filter= malloc(sizeof(SwsFilter));
2205
2206         if(lumaGBlur!=0.0){
2207                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2208                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2209         }else{
2210                 filter->lumH= sws_getIdentityVec();
2211                 filter->lumV= sws_getIdentityVec();
2212         }
2213
2214         if(chromaGBlur!=0.0){
2215                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2216                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2217         }else{
2218                 filter->chrH= sws_getIdentityVec();
2219                 filter->chrV= sws_getIdentityVec();
2220         }
2221
2222         if(chromaSharpen!=0.0){
2223                 SwsVector *g= sws_getConstVec(-1.0, 3);
2224                 SwsVector *id= sws_getConstVec(10.0/chromaSharpen, 1);
2225                 g->coeff[1]=2.0;
2226                 sws_addVec(id, g);
2227                 sws_convVec(filter->chrH, id);
2228                 sws_convVec(filter->chrV, id);
2229                 sws_freeVec(g);
2230                 sws_freeVec(id);
2231         }
2232
2233         if(lumaSharpen!=0.0){
2234                 SwsVector *g= sws_getConstVec(-1.0, 3);
2235                 SwsVector *id= sws_getConstVec(10.0/lumaSharpen, 1);
2236                 g->coeff[1]=2.0;
2237                 sws_addVec(id, g);
2238                 sws_convVec(filter->lumH, id);
2239                 sws_convVec(filter->lumV, id);
2240                 sws_freeVec(g);
2241                 sws_freeVec(id);
2242         }
2243
2244         if(chromaHShift != 0.0)
2245                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2246
2247         if(chromaVShift != 0.0)
2248                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2249
2250         sws_normalizeVec(filter->chrH, 1.0);
2251         sws_normalizeVec(filter->chrV, 1.0);
2252         sws_normalizeVec(filter->lumH, 1.0);
2253         sws_normalizeVec(filter->lumV, 1.0);
2254
2255         if(verbose) sws_printVec(filter->chrH);
2256         if(verbose) sws_printVec(filter->lumH);
2257
2258         return filter;
2259 }
2260
2261 /**
2262  * returns a normalized gaussian curve used to filter stuff
2263  * quality=3 is high quality, lowwer is lowwer quality
2264  */
2265 SwsVector *sws_getGaussianVec(double variance, double quality){
2266         const int length= (int)(variance*quality + 0.5) | 1;
2267         int i;
2268         double *coeff= memalign(sizeof(double), length*sizeof(double));
2269         double middle= (length-1)*0.5;
2270         SwsVector *vec= malloc(sizeof(SwsVector));
2271
2272         vec->coeff= coeff;
2273         vec->length= length;
2274
2275         for(i=0; i<length; i++)
2276         {
2277                 double dist= i-middle;
2278                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2279         }
2280
2281         sws_normalizeVec(vec, 1.0);
2282
2283         return vec;
2284 }
2285
2286 SwsVector *sws_getConstVec(double c, int length){
2287         int i;
2288         double *coeff= memalign(sizeof(double), length*sizeof(double));
2289         SwsVector *vec= malloc(sizeof(SwsVector));
2290
2291         vec->coeff= coeff;
2292         vec->length= length;
2293
2294         for(i=0; i<length; i++)
2295                 coeff[i]= c;
2296
2297         return vec;
2298 }
2299
2300
2301 SwsVector *sws_getIdentityVec(void){
2302         double *coeff= memalign(sizeof(double), sizeof(double));
2303         SwsVector *vec= malloc(sizeof(SwsVector));
2304         coeff[0]= 1.0;
2305
2306         vec->coeff= coeff;
2307         vec->length= 1;
2308
2309         return vec;
2310 }
2311
2312 void sws_normalizeVec(SwsVector *a, double height){
2313         int i;
2314         double sum=0;
2315         double inv;
2316
2317         for(i=0; i<a->length; i++)
2318                 sum+= a->coeff[i];
2319
2320         inv= height/sum;
2321
2322         for(i=0; i<a->length; i++)
2323                 a->coeff[i]*= inv;
2324 }
2325
2326 void sws_scaleVec(SwsVector *a, double scalar){
2327         int i;
2328
2329         for(i=0; i<a->length; i++)
2330                 a->coeff[i]*= scalar;
2331 }
2332
2333 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2334         int length= a->length + b->length - 1;
2335         double *coeff= memalign(sizeof(double), length*sizeof(double));
2336         int i, j;
2337         SwsVector *vec= malloc(sizeof(SwsVector));
2338
2339         vec->coeff= coeff;
2340         vec->length= length;
2341
2342         for(i=0; i<length; i++) coeff[i]= 0.0;
2343
2344         for(i=0; i<a->length; i++)
2345         {
2346                 for(j=0; j<b->length; j++)
2347                 {
2348                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2349                 }
2350         }
2351
2352         return vec;
2353 }
2354
2355 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2356         int length= MAX(a->length, b->length);
2357         double *coeff= memalign(sizeof(double), length*sizeof(double));
2358         int i;
2359         SwsVector *vec= malloc(sizeof(SwsVector));
2360
2361         vec->coeff= coeff;
2362         vec->length= length;
2363
2364         for(i=0; i<length; i++) coeff[i]= 0.0;
2365
2366         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2367         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2368
2369         return vec;
2370 }
2371
2372 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2373         int length= MAX(a->length, b->length);
2374         double *coeff= memalign(sizeof(double), length*sizeof(double));
2375         int i;
2376         SwsVector *vec= malloc(sizeof(SwsVector));
2377
2378         vec->coeff= coeff;
2379         vec->length= length;
2380
2381         for(i=0; i<length; i++) coeff[i]= 0.0;
2382
2383         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2384         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2385
2386         return vec;
2387 }
2388
2389 /* shift left / or right if "shift" is negative */
2390 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2391         int length= a->length + ABS(shift)*2;
2392         double *coeff= memalign(sizeof(double), length*sizeof(double));
2393         int i;
2394         SwsVector *vec= malloc(sizeof(SwsVector));
2395
2396         vec->coeff= coeff;
2397         vec->length= length;
2398
2399         for(i=0; i<length; i++) coeff[i]= 0.0;
2400
2401         for(i=0; i<a->length; i++)
2402         {
2403                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2404         }
2405
2406         return vec;
2407 }
2408
2409 void sws_shiftVec(SwsVector *a, int shift){
2410         SwsVector *shifted= sws_getShiftedVec(a, shift);
2411         free(a->coeff);
2412         a->coeff= shifted->coeff;
2413         a->length= shifted->length;
2414         free(shifted);
2415 }
2416
2417 void sws_addVec(SwsVector *a, SwsVector *b){
2418         SwsVector *sum= sws_sumVec(a, b);
2419         free(a->coeff);
2420         a->coeff= sum->coeff;
2421         a->length= sum->length;
2422         free(sum);
2423 }
2424
2425 void sws_subVec(SwsVector *a, SwsVector *b){
2426         SwsVector *diff= sws_diffVec(a, b);
2427         free(a->coeff);
2428         a->coeff= diff->coeff;
2429         a->length= diff->length;
2430         free(diff);
2431 }
2432
2433 void sws_convVec(SwsVector *a, SwsVector *b){
2434         SwsVector *conv= sws_getConvVec(a, b);
2435         free(a->coeff);  
2436         a->coeff= conv->coeff;
2437         a->length= conv->length;
2438         free(conv);
2439 }
2440
2441 SwsVector *sws_cloneVec(SwsVector *a){
2442         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2443         int i;
2444         SwsVector *vec= malloc(sizeof(SwsVector));
2445
2446         vec->coeff= coeff;
2447         vec->length= a->length;
2448
2449         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2450
2451         return vec;
2452 }
2453
2454 void sws_printVec(SwsVector *a){
2455         int i;
2456         double max=0;
2457         double min=0;
2458         double range;
2459
2460         for(i=0; i<a->length; i++)
2461                 if(a->coeff[i]>max) max= a->coeff[i];
2462
2463         for(i=0; i<a->length; i++)
2464                 if(a->coeff[i]<min) min= a->coeff[i];
2465
2466         range= max - min;
2467
2468         for(i=0; i<a->length; i++)
2469         {
2470                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2471                 MSG_DBG2("%1.3f ", a->coeff[i]);
2472                 for(;x>0; x--) MSG_DBG2(" ");
2473                 MSG_DBG2("|\n");
2474         }
2475 }
2476
2477 void sws_freeVec(SwsVector *a){
2478         if(!a) return;
2479         if(a->coeff) free(a->coeff);
2480         a->coeff=NULL;
2481         a->length=0;
2482         free(a);
2483 }
2484
2485 void sws_freeFilter(SwsFilter *filter){
2486         if(!filter) return;
2487
2488         if(filter->lumH) sws_freeVec(filter->lumH);
2489         if(filter->lumV) sws_freeVec(filter->lumV);
2490         if(filter->chrH) sws_freeVec(filter->chrH);
2491         if(filter->chrV) sws_freeVec(filter->chrV);
2492         free(filter);
2493 }
2494
2495
2496 void sws_freeContext(SwsContext *c){
2497         int i;
2498         if(!c) return;
2499
2500         if(c->lumPixBuf)
2501         {
2502                 for(i=0; i<c->vLumBufSize; i++)
2503                 {
2504                         if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
2505                         c->lumPixBuf[i]=NULL;
2506                 }
2507                 free(c->lumPixBuf);
2508                 c->lumPixBuf=NULL;
2509         }
2510
2511         if(c->chrPixBuf)
2512         {
2513                 for(i=0; i<c->vChrBufSize; i++)
2514                 {
2515                         if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
2516                         c->chrPixBuf[i]=NULL;
2517                 }
2518                 free(c->chrPixBuf);
2519                 c->chrPixBuf=NULL;
2520         }
2521
2522         if(c->vLumFilter) free(c->vLumFilter);
2523         c->vLumFilter = NULL;
2524         if(c->vChrFilter) free(c->vChrFilter);
2525         c->vChrFilter = NULL;
2526         if(c->hLumFilter) free(c->hLumFilter);
2527         c->hLumFilter = NULL;
2528         if(c->hChrFilter) free(c->hChrFilter);
2529         c->hChrFilter = NULL;
2530
2531         if(c->vLumFilterPos) free(c->vLumFilterPos);
2532         c->vLumFilterPos = NULL;
2533         if(c->vChrFilterPos) free(c->vChrFilterPos);
2534         c->vChrFilterPos = NULL;
2535         if(c->hLumFilterPos) free(c->hLumFilterPos);
2536         c->hLumFilterPos = NULL;
2537         if(c->hChrFilterPos) free(c->hChrFilterPos);
2538         c->hChrFilterPos = NULL;
2539
2540         if(c->lumMmx2Filter) free(c->lumMmx2Filter);
2541         c->lumMmx2Filter=NULL;
2542         if(c->chrMmx2Filter) free(c->chrMmx2Filter);
2543         c->chrMmx2Filter=NULL;
2544         if(c->lumMmx2FilterPos) free(c->lumMmx2FilterPos);
2545         c->lumMmx2FilterPos=NULL;
2546         if(c->chrMmx2FilterPos) free(c->chrMmx2FilterPos);
2547         c->chrMmx2FilterPos=NULL;
2548         if(c->yuvTable) free(c->yuvTable);
2549         c->yuvTable=NULL;
2550
2551         free(c);
2552 }
2553