7bd049aa154357ba62a22fdc95e4bfb260deb17c
[ffmpeg.git] / libswscale / 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., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 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 <unistd.h>
57 #include "config.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_SYS_MMAN_H
65 #include <sys/mman.h>
66 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
67 #define MAP_ANONYMOUS MAP_ANON
68 #endif
69 #endif
70 #include "swscale.h"
71 #include "swscale_internal.h"
72 #include "x86_cpu.h"
73 #include "bswap.h"
74 #include "img_format.h"
75 #include "rgb2rgb.h"
76 #ifdef USE_FASTMEMCPY
77 #include "libvo/fastmemcpy.h"
78 #endif
79
80 #undef MOVNTQ
81 #undef PAVGB
82
83 //#undef HAVE_MMX2
84 //#define HAVE_3DNOW
85 //#undef HAVE_MMX
86 //#undef ARCH_X86
87 //#define WORDS_BIGENDIAN
88 #define DITHER1XBPP
89
90 #define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
91
92 #define RET 0xC3 //near return opcode for X86
93
94 #ifdef MP_DEBUG
95 #define ASSERT(x) assert(x);
96 #else
97 #define ASSERT(x) ;
98 #endif
99
100 #ifdef M_PI
101 #define PI M_PI
102 #else
103 #define PI 3.14159265358979323846
104 #endif
105
106 //FIXME replace this with something faster
107 #define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
108                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \
109                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
110 #define isYUV(x)       ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
111 #define isGray(x)      ((x)==IMGFMT_Y800)
112 #define isRGB(x)       (((x)&IMGFMT_RGB_MASK)==IMGFMT_RGB)
113 #define isBGR(x)       (((x)&IMGFMT_BGR_MASK)==IMGFMT_BGR)
114 #define isSupportedIn(x)  ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
115                         || (x)==IMGFMT_BGR32|| (x)==IMGFMT_BGR24|| (x)==IMGFMT_BGR16|| (x)==IMGFMT_BGR15\
116                         || (x)==IMGFMT_RGB32|| (x)==IMGFMT_RGB24\
117                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9\
118                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
119 #define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
120                         || (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
121                         || isRGB(x) || isBGR(x)\
122                         || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\
123                         || (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
124 #define isPacked(x)    ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
125
126 #define RGB2YUV_SHIFT 16
127 #define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
128 #define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
129 #define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
130 #define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
131 #define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
132 #define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
133 #define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
134 #define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
135 #define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
136
137 extern const int32_t Inverse_Table_6_9[8][4];
138
139 /*
140 NOTES
141 Special versions: fast Y 1:1 scaling (no interpolation in y direction)
142
143 TODO
144 more intelligent missalignment avoidance for the horizontal scaler
145 write special vertical cubic upscale version
146 Optimize C code (yv12 / minmax)
147 add support for packed pixel yuv input & output
148 add support for Y8 output
149 optimize bgr24 & bgr32
150 add BGR4 output support
151 write special BGR->BGR scaler
152 */
153
154 #define MIN(a,b) ((a) > (b) ? (b) : (a))
155 #define MAX(a,b) ((a) < (b) ? (b) : (a))
156
157 #if defined(ARCH_X86) || defined(ARCH_X86_64)
158 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
159 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
160 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
161 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
162 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
163 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
164 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
165 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
166
167 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
168 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
169 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
170 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
171
172 static uint64_t __attribute__((aligned(8))) dither4[2]={
173         0x0103010301030103LL,
174         0x0200020002000200LL,};
175
176 static uint64_t __attribute__((aligned(8))) dither8[2]={
177         0x0602060206020602LL,
178         0x0004000400040004LL,};
179
180 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
181 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
182 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
183 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
184 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
185 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
186
187 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
188 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
189 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
190
191 #ifdef FAST_BGR2YV12
192 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
193 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
194 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
195 #else
196 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
197 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
198 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
199 #endif
200 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
201 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
202 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
203 #endif
204
205 // clipping helper table for C implementations:
206 static unsigned char clip_table[768];
207
208 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
209                   
210 extern const uint8_t dither_2x2_4[2][8];
211 extern const uint8_t dither_2x2_8[2][8];
212 extern const uint8_t dither_8x8_32[8][8];
213 extern const uint8_t dither_8x8_73[8][8];
214 extern const uint8_t dither_8x8_220[8][8];
215
216 char *sws_format_name(int format)
217 {
218     static char fmt_name[64];
219     char *res;
220     static int buffer;
221
222     res = fmt_name + buffer * 32;
223     buffer = 1 - buffer;
224     snprintf(res, 32, "0x%x (%c%c%c%c)", format,
225                     format >> 24, (format >> 16) & 0xFF,
226                     (format >> 8) & 0xFF,
227                     format & 0xFF);
228
229     return res;
230 }
231
232 #if defined(ARCH_X86) || defined(ARCH_X86_64)
233 void in_asm_used_var_warning_killer()
234 {
235  volatile int i= bF8+bFC+w10+
236  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
237  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
238  if(i) i=0;
239 }
240 #endif
241
242 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
243                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
244                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
245 {
246         //FIXME Optimize (just quickly writen not opti..)
247         int i;
248         for(i=0; i<dstW; i++)
249         {
250                 int val=1<<18;
251                 int j;
252                 for(j=0; j<lumFilterSize; j++)
253                         val += lumSrc[j][i] * lumFilter[j];
254
255                 dest[i]= MIN(MAX(val>>19, 0), 255);
256         }
257
258         if(uDest != NULL)
259                 for(i=0; i<chrDstW; i++)
260                 {
261                         int u=1<<18;
262                         int v=1<<18;
263                         int j;
264                         for(j=0; j<chrFilterSize; j++)
265                         {
266                                 u += chrSrc[j][i] * chrFilter[j];
267                                 v += chrSrc[j][i + 2048] * chrFilter[j];
268                         }
269
270                         uDest[i]= MIN(MAX(u>>19, 0), 255);
271                         vDest[i]= MIN(MAX(v>>19, 0), 255);
272                 }
273 }
274
275 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
276                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
277                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
278 {
279         //FIXME Optimize (just quickly writen not opti..)
280         int i;
281         for(i=0; i<dstW; i++)
282         {
283                 int val=1<<18;
284                 int j;
285                 for(j=0; j<lumFilterSize; j++)
286                         val += lumSrc[j][i] * lumFilter[j];
287
288                 dest[i]= MIN(MAX(val>>19, 0), 255);
289         }
290
291         if(uDest == NULL)
292                 return;
293
294         if(dstFormat == IMGFMT_NV12)
295                 for(i=0; i<chrDstW; i++)
296                 {
297                         int u=1<<18;
298                         int v=1<<18;
299                         int j;
300                         for(j=0; j<chrFilterSize; j++)
301                         {
302                                 u += chrSrc[j][i] * chrFilter[j];
303                                 v += chrSrc[j][i + 2048] * chrFilter[j];
304                         }
305
306                         uDest[2*i]= MIN(MAX(u>>19, 0), 255);
307                         uDest[2*i+1]= MIN(MAX(v>>19, 0), 255);
308                 }
309         else
310                 for(i=0; i<chrDstW; i++)
311                 {
312                         int u=1<<18;
313                         int v=1<<18;
314                         int j;
315                         for(j=0; j<chrFilterSize; j++)
316                         {
317                                 u += chrSrc[j][i] * chrFilter[j];
318                                 v += chrSrc[j][i + 2048] * chrFilter[j];
319                         }
320
321                         uDest[2*i]= MIN(MAX(v>>19, 0), 255);
322                         uDest[2*i+1]= MIN(MAX(u>>19, 0), 255);
323                 }
324 }
325
326 #define YSCALE_YUV_2_PACKEDX_C(type) \
327                 for(i=0; i<(dstW>>1); i++){\
328                         int j;\
329                         int Y1=1<<18;\
330                         int Y2=1<<18;\
331                         int U=1<<18;\
332                         int V=1<<18;\
333                         type *r, *b, *g;\
334                         const int i2= 2*i;\
335                         \
336                         for(j=0; j<lumFilterSize; j++)\
337                         {\
338                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
339                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
340                         }\
341                         for(j=0; j<chrFilterSize; j++)\
342                         {\
343                                 U += chrSrc[j][i] * chrFilter[j];\
344                                 V += chrSrc[j][i+2048] * chrFilter[j];\
345                         }\
346                         Y1>>=19;\
347                         Y2>>=19;\
348                         U >>=19;\
349                         V >>=19;\
350                         if((Y1|Y2|U|V)&256)\
351                         {\
352                                 if(Y1>255)   Y1=255;\
353                                 else if(Y1<0)Y1=0;\
354                                 if(Y2>255)   Y2=255;\
355                                 else if(Y2<0)Y2=0;\
356                                 if(U>255)    U=255;\
357                                 else if(U<0) U=0;\
358                                 if(V>255)    V=255;\
359                                 else if(V<0) V=0;\
360                         }
361                         
362 #define YSCALE_YUV_2_RGBX_C(type) \
363                         YSCALE_YUV_2_PACKEDX_C(type)\
364                         r = c->table_rV[V];\
365                         g = c->table_gU[U] + c->table_gV[V];\
366                         b = c->table_bU[U];\
367
368 #define YSCALE_YUV_2_PACKED2_C \
369                 for(i=0; i<(dstW>>1); i++){\
370                         const int i2= 2*i;\
371                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
372                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
373                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
374                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
375
376 #define YSCALE_YUV_2_RGB2_C(type) \
377                         YSCALE_YUV_2_PACKED2_C\
378                         type *r, *b, *g;\
379                         r = c->table_rV[V];\
380                         g = c->table_gU[U] + c->table_gV[V];\
381                         b = c->table_bU[U];\
382
383 #define YSCALE_YUV_2_PACKED1_C \
384                 for(i=0; i<(dstW>>1); i++){\
385                         const int i2= 2*i;\
386                         int Y1= buf0[i2  ]>>7;\
387                         int Y2= buf0[i2+1]>>7;\
388                         int U= (uvbuf1[i     ])>>7;\
389                         int V= (uvbuf1[i+2048])>>7;\
390
391 #define YSCALE_YUV_2_RGB1_C(type) \
392                         YSCALE_YUV_2_PACKED1_C\
393                         type *r, *b, *g;\
394                         r = c->table_rV[V];\
395                         g = c->table_gU[U] + c->table_gV[V];\
396                         b = c->table_bU[U];\
397
398 #define YSCALE_YUV_2_PACKED1B_C \
399                 for(i=0; i<(dstW>>1); i++){\
400                         const int i2= 2*i;\
401                         int Y1= buf0[i2  ]>>7;\
402                         int Y2= buf0[i2+1]>>7;\
403                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
404                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
405
406 #define YSCALE_YUV_2_RGB1B_C(type) \
407                         YSCALE_YUV_2_PACKED1B_C\
408                         type *r, *b, *g;\
409                         r = c->table_rV[V];\
410                         g = c->table_gU[U] + c->table_gV[V];\
411                         b = c->table_bU[U];\
412
413 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
414         switch(c->dstFormat)\
415         {\
416         case IMGFMT_BGR32:\
417         case IMGFMT_RGB32:\
418                 func(uint32_t)\
419                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
420                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
421                 }               \
422                 break;\
423         case IMGFMT_RGB24:\
424                 func(uint8_t)\
425                         ((uint8_t*)dest)[0]= r[Y1];\
426                         ((uint8_t*)dest)[1]= g[Y1];\
427                         ((uint8_t*)dest)[2]= b[Y1];\
428                         ((uint8_t*)dest)[3]= r[Y2];\
429                         ((uint8_t*)dest)[4]= g[Y2];\
430                         ((uint8_t*)dest)[5]= b[Y2];\
431                         dest+=6;\
432                 }\
433                 break;\
434         case IMGFMT_BGR24:\
435                 func(uint8_t)\
436                         ((uint8_t*)dest)[0]= b[Y1];\
437                         ((uint8_t*)dest)[1]= g[Y1];\
438                         ((uint8_t*)dest)[2]= r[Y1];\
439                         ((uint8_t*)dest)[3]= b[Y2];\
440                         ((uint8_t*)dest)[4]= g[Y2];\
441                         ((uint8_t*)dest)[5]= r[Y2];\
442                         dest+=6;\
443                 }\
444                 break;\
445         case IMGFMT_RGB16:\
446         case IMGFMT_BGR16:\
447                 {\
448                         const int dr1= dither_2x2_8[y&1    ][0];\
449                         const int dg1= dither_2x2_4[y&1    ][0];\
450                         const int db1= dither_2x2_8[(y&1)^1][0];\
451                         const int dr2= dither_2x2_8[y&1    ][1];\
452                         const int dg2= dither_2x2_4[y&1    ][1];\
453                         const int db2= dither_2x2_8[(y&1)^1][1];\
454                         func(uint16_t)\
455                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
456                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
457                         }\
458                 }\
459                 break;\
460         case IMGFMT_RGB15:\
461         case IMGFMT_BGR15:\
462                 {\
463                         const int dr1= dither_2x2_8[y&1    ][0];\
464                         const int dg1= dither_2x2_8[y&1    ][1];\
465                         const int db1= dither_2x2_8[(y&1)^1][0];\
466                         const int dr2= dither_2x2_8[y&1    ][1];\
467                         const int dg2= dither_2x2_8[y&1    ][0];\
468                         const int db2= dither_2x2_8[(y&1)^1][1];\
469                         func(uint16_t)\
470                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
471                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
472                         }\
473                 }\
474                 break;\
475         case IMGFMT_RGB8:\
476         case IMGFMT_BGR8:\
477                 {\
478                         const uint8_t * const d64= dither_8x8_73[y&7];\
479                         const uint8_t * const d32= dither_8x8_32[y&7];\
480                         func(uint8_t)\
481                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
482                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
483                         }\
484                 }\
485                 break;\
486         case IMGFMT_RGB4:\
487         case IMGFMT_BGR4:\
488                 {\
489                         const uint8_t * const d64= dither_8x8_73 [y&7];\
490                         const uint8_t * const d128=dither_8x8_220[y&7];\
491                         func(uint8_t)\
492                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
493                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
494                         }\
495                 }\
496                 break;\
497         case IMGFMT_RG4B:\
498         case IMGFMT_BG4B:\
499                 {\
500                         const uint8_t * const d64= dither_8x8_73 [y&7];\
501                         const uint8_t * const d128=dither_8x8_220[y&7];\
502                         func(uint8_t)\
503                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
504                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
505                         }\
506                 }\
507                 break;\
508         case IMGFMT_RGB1:\
509         case IMGFMT_BGR1:\
510                 {\
511                         const uint8_t * const d128=dither_8x8_220[y&7];\
512                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
513                         for(i=0; i<dstW-7; i+=8){\
514                                 int acc;\
515                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
516                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
517                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
518                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
519                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
520                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
521                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
522                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
523                                 ((uint8_t*)dest)[0]= acc;\
524                                 dest++;\
525                         }\
526 \
527 /*\
528 ((uint8_t*)dest)-= dstW>>4;\
529 {\
530                         int acc=0;\
531                         int left=0;\
532                         static int top[1024];\
533                         static int last_new[1024][1024];\
534                         static int last_in3[1024][1024];\
535                         static int drift[1024][1024];\
536                         int topLeft=0;\
537                         int shift=0;\
538                         int count=0;\
539                         const uint8_t * const d128=dither_8x8_220[y&7];\
540                         int error_new=0;\
541                         int error_in3=0;\
542                         int f=0;\
543                         \
544                         for(i=dstW>>1; i<dstW; i++){\
545                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
546                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
547                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
548                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
549                                         + (last_new[y][i] - in3)*f/256;\
550                                 int new= old> 128 ? 255 : 0;\
551 \
552                                 error_new+= ABS(last_new[y][i] - new);\
553                                 error_in3+= ABS(last_in3[y][i] - in3);\
554                                 f= error_new - error_in3*4;\
555                                 if(f<0) f=0;\
556                                 if(f>256) f=256;\
557 \
558                                 topLeft= top[i];\
559                                 left= top[i]= old - new;\
560                                 last_new[y][i]= new;\
561                                 last_in3[y][i]= in3;\
562 \
563                                 acc+= acc + (new&1);\
564                                 if((i&7)==6){\
565                                         ((uint8_t*)dest)[0]= acc;\
566                                         ((uint8_t*)dest)++;\
567                                 }\
568                         }\
569 }\
570 */\
571                 }\
572                 break;\
573         case IMGFMT_YUY2:\
574                 func2\
575                         ((uint8_t*)dest)[2*i2+0]= Y1;\
576                         ((uint8_t*)dest)[2*i2+1]= U;\
577                         ((uint8_t*)dest)[2*i2+2]= Y2;\
578                         ((uint8_t*)dest)[2*i2+3]= V;\
579                 }               \
580                 break;\
581         case IMGFMT_UYVY:\
582                 func2\
583                         ((uint8_t*)dest)[2*i2+0]= U;\
584                         ((uint8_t*)dest)[2*i2+1]= Y1;\
585                         ((uint8_t*)dest)[2*i2+2]= V;\
586                         ((uint8_t*)dest)[2*i2+3]= Y2;\
587                 }               \
588                 break;\
589         }\
590
591
592 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
593                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
594                                     uint8_t *dest, int dstW, int y)
595 {
596         int i;
597         switch(c->dstFormat)
598         {
599         case IMGFMT_RGB32:
600         case IMGFMT_BGR32:
601                 YSCALE_YUV_2_RGBX_C(uint32_t)
602                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
603                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
604                 }
605                 break;
606         case IMGFMT_RGB24:
607                 YSCALE_YUV_2_RGBX_C(uint8_t)
608                         ((uint8_t*)dest)[0]= r[Y1];
609                         ((uint8_t*)dest)[1]= g[Y1];
610                         ((uint8_t*)dest)[2]= b[Y1];
611                         ((uint8_t*)dest)[3]= r[Y2];
612                         ((uint8_t*)dest)[4]= g[Y2];
613                         ((uint8_t*)dest)[5]= b[Y2];
614                         dest+=6;
615                 }
616                 break;
617         case IMGFMT_BGR24:
618                 YSCALE_YUV_2_RGBX_C(uint8_t)
619                         ((uint8_t*)dest)[0]= b[Y1];
620                         ((uint8_t*)dest)[1]= g[Y1];
621                         ((uint8_t*)dest)[2]= r[Y1];
622                         ((uint8_t*)dest)[3]= b[Y2];
623                         ((uint8_t*)dest)[4]= g[Y2];
624                         ((uint8_t*)dest)[5]= r[Y2];
625                         dest+=6;
626                 }
627                 break;
628         case IMGFMT_RGB16:
629         case IMGFMT_BGR16:
630                 {
631                         const int dr1= dither_2x2_8[y&1    ][0];
632                         const int dg1= dither_2x2_4[y&1    ][0];
633                         const int db1= dither_2x2_8[(y&1)^1][0];
634                         const int dr2= dither_2x2_8[y&1    ][1];
635                         const int dg2= dither_2x2_4[y&1    ][1];
636                         const int db2= dither_2x2_8[(y&1)^1][1];
637                         YSCALE_YUV_2_RGBX_C(uint16_t)
638                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
639                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
640                         }
641                 }
642                 break;
643         case IMGFMT_RGB15:
644         case IMGFMT_BGR15:
645                 {
646                         const int dr1= dither_2x2_8[y&1    ][0];
647                         const int dg1= dither_2x2_8[y&1    ][1];
648                         const int db1= dither_2x2_8[(y&1)^1][0];
649                         const int dr2= dither_2x2_8[y&1    ][1];
650                         const int dg2= dither_2x2_8[y&1    ][0];
651                         const int db2= dither_2x2_8[(y&1)^1][1];
652                         YSCALE_YUV_2_RGBX_C(uint16_t)
653                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
654                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
655                         }
656                 }
657                 break;
658         case IMGFMT_RGB8:
659         case IMGFMT_BGR8:
660                 {
661                         const uint8_t * const d64= dither_8x8_73[y&7];
662                         const uint8_t * const d32= dither_8x8_32[y&7];
663                         YSCALE_YUV_2_RGBX_C(uint8_t)
664                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
665                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
666                         }
667                 }
668                 break;
669         case IMGFMT_RGB4:
670         case IMGFMT_BGR4:
671                 {
672                         const uint8_t * const d64= dither_8x8_73 [y&7];
673                         const uint8_t * const d128=dither_8x8_220[y&7];
674                         YSCALE_YUV_2_RGBX_C(uint8_t)
675                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
676                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
677                         }
678                 }
679                 break;
680         case IMGFMT_RG4B:
681         case IMGFMT_BG4B:
682                 {
683                         const uint8_t * const d64= dither_8x8_73 [y&7];
684                         const uint8_t * const d128=dither_8x8_220[y&7];
685                         YSCALE_YUV_2_RGBX_C(uint8_t)
686                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
687                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
688                         }
689                 }
690                 break;
691         case IMGFMT_RGB1:
692         case IMGFMT_BGR1:
693                 {
694                         const uint8_t * const d128=dither_8x8_220[y&7];
695                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
696                         int acc=0;
697                         for(i=0; i<dstW-1; i+=2){
698                                 int j;
699                                 int Y1=1<<18;
700                                 int Y2=1<<18;
701
702                                 for(j=0; j<lumFilterSize; j++)
703                                 {
704                                         Y1 += lumSrc[j][i] * lumFilter[j];
705                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
706                                 }
707                                 Y1>>=19;
708                                 Y2>>=19;
709                                 if((Y1|Y2)&256)
710                                 {
711                                         if(Y1>255)   Y1=255;
712                                         else if(Y1<0)Y1=0;
713                                         if(Y2>255)   Y2=255;
714                                         else if(Y2<0)Y2=0;
715                                 }
716                                 acc+= acc + g[Y1+d128[(i+0)&7]];
717                                 acc+= acc + g[Y2+d128[(i+1)&7]];
718                                 if((i&7)==6){
719                                         ((uint8_t*)dest)[0]= acc;
720                                         dest++;
721                                 }
722                         }
723                 }
724                 break;
725         case IMGFMT_YUY2:
726                 YSCALE_YUV_2_PACKEDX_C(void)
727                         ((uint8_t*)dest)[2*i2+0]= Y1;
728                         ((uint8_t*)dest)[2*i2+1]= U;
729                         ((uint8_t*)dest)[2*i2+2]= Y2;
730                         ((uint8_t*)dest)[2*i2+3]= V;
731                 }
732                 break;
733         case IMGFMT_UYVY:
734                 YSCALE_YUV_2_PACKEDX_C(void)
735                         ((uint8_t*)dest)[2*i2+0]= U;
736                         ((uint8_t*)dest)[2*i2+1]= Y1;
737                         ((uint8_t*)dest)[2*i2+2]= V;
738                         ((uint8_t*)dest)[2*i2+3]= Y2;
739                 }
740                 break;
741         }
742 }
743
744
745 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
746 //Plain C versions
747 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
748 #define COMPILE_C
749 #endif
750
751 #ifdef ARCH_POWERPC
752 #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
753 #define COMPILE_ALTIVEC
754 #endif //HAVE_ALTIVEC
755 #endif //ARCH_POWERPC
756
757 #if defined(ARCH_X86) || defined(ARCH_X86_64)
758
759 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
760 #define COMPILE_MMX
761 #endif
762
763 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
764 #define COMPILE_MMX2
765 #endif
766
767 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
768 #define COMPILE_3DNOW
769 #endif
770 #endif //ARCH_X86 || ARCH_X86_64
771
772 #undef HAVE_MMX
773 #undef HAVE_MMX2
774 #undef HAVE_3DNOW
775
776 #ifdef COMPILE_C
777 #undef HAVE_MMX
778 #undef HAVE_MMX2
779 #undef HAVE_3DNOW
780 #undef HAVE_ALTIVEC
781 #define RENAME(a) a ## _C
782 #include "swscale_template.c"
783 #endif
784
785 #ifdef ARCH_POWERPC
786 #ifdef COMPILE_ALTIVEC
787 #undef RENAME
788 #define HAVE_ALTIVEC
789 #define RENAME(a) a ## _altivec
790 #include "swscale_template.c"
791 #endif
792 #endif //ARCH_POWERPC
793
794 #if defined(ARCH_X86) || defined(ARCH_X86_64)
795
796 //X86 versions
797 /*
798 #undef RENAME
799 #undef HAVE_MMX
800 #undef HAVE_MMX2
801 #undef HAVE_3DNOW
802 #define ARCH_X86
803 #define RENAME(a) a ## _X86
804 #include "swscale_template.c"
805 */
806 //MMX versions
807 #ifdef COMPILE_MMX
808 #undef RENAME
809 #define HAVE_MMX
810 #undef HAVE_MMX2
811 #undef HAVE_3DNOW
812 #define RENAME(a) a ## _MMX
813 #include "swscale_template.c"
814 #endif
815
816 //MMX2 versions
817 #ifdef COMPILE_MMX2
818 #undef RENAME
819 #define HAVE_MMX
820 #define HAVE_MMX2
821 #undef HAVE_3DNOW
822 #define RENAME(a) a ## _MMX2
823 #include "swscale_template.c"
824 #endif
825
826 //3DNOW versions
827 #ifdef COMPILE_3DNOW
828 #undef RENAME
829 #define HAVE_MMX
830 #undef HAVE_MMX2
831 #define HAVE_3DNOW
832 #define RENAME(a) a ## _3DNow
833 #include "swscale_template.c"
834 #endif
835
836 #endif //ARCH_X86 || ARCH_X86_64
837
838 // minor note: the HAVE_xyz is messed up after that line so don't use it
839
840 static double getSplineCoeff(double a, double b, double c, double d, double dist)
841 {
842 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
843         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
844         else            return getSplineCoeff(  0.0, 
845                                                  b+ 2.0*c + 3.0*d,
846                                                         c + 3.0*d,
847                                                 -b- 3.0*c - 6.0*d,
848                                                 dist-1.0);
849 }
850
851 static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
852                               int srcW, int dstW, int filterAlign, int one, int flags,
853                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
854 {
855         int i;
856         int filterSize;
857         int filter2Size;
858         int minFilterSize;
859         double *filter=NULL;
860         double *filter2=NULL;
861 #if defined(ARCH_X86) || defined(ARCH_X86_64)
862         if(flags & SWS_CPU_CAPS_MMX)
863                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
864 #endif
865
866         // Note the +1 is for the MMXscaler which reads over the end
867         *filterPos = (int16_t*)memalign(8, (dstW+1)*sizeof(int16_t));
868
869         if(ABS(xInc - 0x10000) <10) // unscaled
870         {
871                 int i;
872                 filterSize= 1;
873                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
874                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
875
876                 for(i=0; i<dstW; i++)
877                 {
878                         filter[i*filterSize]=1;
879                         (*filterPos)[i]=i;
880                 }
881
882         }
883         else if(flags&SWS_POINT) // lame looking point sampling mode
884         {
885                 int i;
886                 int xDstInSrc;
887                 filterSize= 1;
888                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
889                 
890                 xDstInSrc= xInc/2 - 0x8000;
891                 for(i=0; i<dstW; i++)
892                 {
893                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
894
895                         (*filterPos)[i]= xx;
896                         filter[i]= 1.0;
897                         xDstInSrc+= xInc;
898                 }
899         }
900         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
901         {
902                 int i;
903                 int xDstInSrc;
904                 if     (flags&SWS_BICUBIC) filterSize= 4;
905                 else if(flags&SWS_X      ) filterSize= 4;
906                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
907                 filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
908
909                 xDstInSrc= xInc/2 - 0x8000;
910                 for(i=0; i<dstW; i++)
911                 {
912                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
913                         int j;
914
915                         (*filterPos)[i]= xx;
916                                 //Bilinear upscale / linear interpolate / Area averaging
917                                 for(j=0; j<filterSize; j++)
918                                 {
919                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
920                                         double coeff= 1.0 - d;
921                                         if(coeff<0) coeff=0;
922                                         filter[i*filterSize + j]= coeff;
923                                         xx++;
924                                 }
925                         xDstInSrc+= xInc;
926                 }
927         }
928         else
929         {
930                 double xDstInSrc;
931                 double sizeFactor, filterSizeInSrc;
932                 const double xInc1= (double)xInc / (double)(1<<16);
933
934                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
935                 else if(flags&SWS_X)            sizeFactor= 8.0;
936                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
937                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
938                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
939                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
940                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
941                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
942                 else {
943                         sizeFactor= 0.0; //GCC warning killer
944                         ASSERT(0)
945                 }
946                 
947                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
948                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
949
950                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
951                 if(filterSize > srcW-2) filterSize=srcW-2;
952
953                 filter= (double*)memalign(16, dstW*sizeof(double)*filterSize);
954
955                 xDstInSrc= xInc1 / 2.0 - 0.5;
956                 for(i=0; i<dstW; i++)
957                 {
958                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
959                         int j;
960                         (*filterPos)[i]= xx;
961                         for(j=0; j<filterSize; j++)
962                         {
963                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
964                                 double coeff;
965                                 if(flags & SWS_BICUBIC)
966                                 {
967                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
968                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
969
970                                         if(d<1.0) 
971                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
972                                         else if(d<2.0)
973                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
974                                         else
975                                                 coeff=0.0;
976                                 }
977 /*                              else if(flags & SWS_X)
978                                 {
979                                         double p= param ? param*0.01 : 0.3;
980                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
981                                         coeff*= pow(2.0, - p*d*d);
982                                 }*/
983                                 else if(flags & SWS_X)
984                                 {
985                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
986                                         
987                                         if(d<1.0)
988                                                 coeff = cos(d*PI);
989                                         else
990                                                 coeff=-1.0;
991                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
992                                         else            coeff=  pow( coeff, A);
993                                         coeff= coeff*0.5 + 0.5;
994                                 }
995                                 else if(flags & SWS_AREA)
996                                 {
997                                         double srcPixelSize= 1.0/xInc1;
998                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
999                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1000                                         else coeff=0.0;
1001                                 }
1002                                 else if(flags & SWS_GAUSS)
1003                                 {
1004                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1005                                         coeff = pow(2.0, - p*d*d);
1006                                 }
1007                                 else if(flags & SWS_SINC)
1008                                 {
1009                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1010                                 }
1011                                 else if(flags & SWS_LANCZOS)
1012                                 {
1013                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
1014                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1015                                         if(d>p) coeff=0;
1016                                 }
1017                                 else if(flags & SWS_BILINEAR)
1018                                 {
1019                                         coeff= 1.0 - d;
1020                                         if(coeff<0) coeff=0;
1021                                 }
1022                                 else if(flags & SWS_SPLINE)
1023                                 {
1024                                         double p=-2.196152422706632;
1025                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1026                                 }
1027                                 else {
1028                                         coeff= 0.0; //GCC warning killer
1029                                         ASSERT(0)
1030                                 }
1031
1032                                 filter[i*filterSize + j]= coeff;
1033                                 xx++;
1034                         }
1035                         xDstInSrc+= xInc1;
1036                 }
1037         }
1038
1039         /* apply src & dst Filter to filter -> filter2
1040            av_free(filter);
1041         */
1042         ASSERT(filterSize>0)
1043         filter2Size= filterSize;
1044         if(srcFilter) filter2Size+= srcFilter->length - 1;
1045         if(dstFilter) filter2Size+= dstFilter->length - 1;
1046         ASSERT(filter2Size>0)
1047         filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
1048
1049         for(i=0; i<dstW; i++)
1050         {
1051                 int j;
1052                 SwsVector scaleFilter;
1053                 SwsVector *outVec;
1054
1055                 scaleFilter.coeff= filter + i*filterSize;
1056                 scaleFilter.length= filterSize;
1057
1058                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1059                 else          outVec= &scaleFilter;
1060
1061                 ASSERT(outVec->length == filter2Size)
1062                 //FIXME dstFilter
1063
1064                 for(j=0; j<outVec->length; j++)
1065                 {
1066                         filter2[i*filter2Size + j]= outVec->coeff[j];
1067                 }
1068
1069                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1070
1071                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1072         }
1073         av_free(filter); filter=NULL;
1074
1075         /* try to reduce the filter-size (step1 find size and shift left) */
1076         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1077         minFilterSize= 0;
1078         for(i=dstW-1; i>=0; i--)
1079         {
1080                 int min= filter2Size;
1081                 int j;
1082                 double cutOff=0.0;
1083
1084                 /* get rid off near zero elements on the left by shifting left */
1085                 for(j=0; j<filter2Size; j++)
1086                 {
1087                         int k;
1088                         cutOff += ABS(filter2[i*filter2Size]);
1089
1090                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1091
1092                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1093                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1094
1095                         // Move filter coeffs left
1096                         for(k=1; k<filter2Size; k++)
1097                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1098                         filter2[i*filter2Size + k - 1]= 0.0;
1099                         (*filterPos)[i]++;
1100                 }
1101
1102                 cutOff=0.0;
1103                 /* count near zeros on the right */
1104                 for(j=filter2Size-1; j>0; j--)
1105                 {
1106                         cutOff += ABS(filter2[i*filter2Size + j]);
1107
1108                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1109                         min--;
1110                 }
1111
1112                 if(min>minFilterSize) minFilterSize= min;
1113         }
1114
1115         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1116           // we can handle the special case 4,
1117           // so we don't want to go to the full 8
1118           if (minFilterSize < 5)
1119             filterAlign = 4;
1120
1121           // we really don't want to waste our time
1122           // doing useless computation, so fall-back on
1123           // the scalar C code for very small filter.
1124           // vectorizing is worth it only if you have
1125           // decent-sized vector.
1126           if (minFilterSize < 3)
1127             filterAlign = 1;
1128         }
1129
1130         ASSERT(minFilterSize > 0)
1131         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1132         ASSERT(filterSize > 0)
1133         filter= (double*)memalign(8, filterSize*dstW*sizeof(double));
1134         *outFilterSize= filterSize;
1135
1136         if(flags&SWS_PRINT_INFO)
1137                 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1138         /* try to reduce the filter-size (step2 reduce it) */
1139         for(i=0; i<dstW; i++)
1140         {
1141                 int j;
1142
1143                 for(j=0; j<filterSize; j++)
1144                 {
1145                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1146                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1147                 }
1148         }
1149         av_free(filter2); filter2=NULL;
1150         
1151
1152         //FIXME try to align filterpos if possible
1153
1154         //fix borders
1155         for(i=0; i<dstW; i++)
1156         {
1157                 int j;
1158                 if((*filterPos)[i] < 0)
1159                 {
1160                         // Move filter coeffs left to compensate for filterPos
1161                         for(j=1; j<filterSize; j++)
1162                         {
1163                                 int left= MAX(j + (*filterPos)[i], 0);
1164                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1165                                 filter[i*filterSize + j]=0;
1166                         }
1167                         (*filterPos)[i]= 0;
1168                 }
1169
1170                 if((*filterPos)[i] + filterSize > srcW)
1171                 {
1172                         int shift= (*filterPos)[i] + filterSize - srcW;
1173                         // Move filter coeffs right to compensate for filterPos
1174                         for(j=filterSize-2; j>=0; j--)
1175                         {
1176                                 int right= MIN(j + shift, filterSize-1);
1177                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1178                                 filter[i*filterSize +j]=0;
1179                         }
1180                         (*filterPos)[i]= srcW - filterSize;
1181                 }
1182         }
1183
1184         // Note the +1 is for the MMXscaler which reads over the end
1185         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1186         *outFilter= (int16_t*)memalign(16, *outFilterSize*(dstW+1)*sizeof(int16_t));
1187         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1188
1189         /* Normalize & Store in outFilter */
1190         for(i=0; i<dstW; i++)
1191         {
1192                 int j;
1193                 double error=0;
1194                 double sum=0;
1195                 double scale= one;
1196
1197                 for(j=0; j<filterSize; j++)
1198                 {
1199                         sum+= filter[i*filterSize + j];
1200                 }
1201                 scale/= sum;
1202                 for(j=0; j<*outFilterSize; j++)
1203                 {
1204                         double v= filter[i*filterSize + j]*scale + error;
1205                         int intV= floor(v + 0.5);
1206                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1207                         error = v - intV;
1208                 }
1209         }
1210         
1211         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1212         for(i=0; i<*outFilterSize; i++)
1213         {
1214                 int j= dstW*(*outFilterSize);
1215                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1216         }
1217
1218         av_free(filter);
1219 }
1220
1221 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1222 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1223 {
1224         uint8_t *fragmentA;
1225         long imm8OfPShufW1A;
1226         long imm8OfPShufW2A;
1227         long fragmentLengthA;
1228         uint8_t *fragmentB;
1229         long imm8OfPShufW1B;
1230         long imm8OfPShufW2B;
1231         long fragmentLengthB;
1232         int fragmentPos;
1233
1234         int xpos, i;
1235
1236         // create an optimized horizontal scaling routine
1237
1238         //code fragment
1239
1240         asm volatile(
1241                 "jmp 9f                         \n\t"
1242         // Begin
1243                 "0:                             \n\t"
1244                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1245                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1246                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1247                 "punpcklbw %%mm7, %%mm1         \n\t"
1248                 "punpcklbw %%mm7, %%mm0         \n\t"
1249                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1250                 "1:                             \n\t"
1251                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1252                 "2:                             \n\t"
1253                 "psubw %%mm1, %%mm0             \n\t"
1254                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1255                 "pmullw %%mm3, %%mm0            \n\t"
1256                 "psllw $7, %%mm1                \n\t"
1257                 "paddw %%mm1, %%mm0             \n\t"
1258
1259                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1260
1261                 "add $8, %%"REG_a"              \n\t"
1262         // End
1263                 "9:                             \n\t"
1264 //              "int $3\n\t"
1265                 "lea 0b, %0                     \n\t"
1266                 "lea 1b, %1                     \n\t"
1267                 "lea 2b, %2                     \n\t"
1268                 "dec %1                         \n\t"
1269                 "dec %2                         \n\t"
1270                 "sub %0, %1                     \n\t"
1271                 "sub %0, %2                     \n\t"
1272                 "lea 9b, %3                     \n\t"
1273                 "sub %0, %3                     \n\t"
1274
1275
1276                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1277                 "=r" (fragmentLengthA)
1278         );
1279
1280         asm volatile(
1281                 "jmp 9f                         \n\t"
1282         // Begin
1283                 "0:                             \n\t"
1284                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1285                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1286                 "punpcklbw %%mm7, %%mm0         \n\t"
1287                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1288                 "1:                             \n\t"
1289                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1290                 "2:                             \n\t"
1291                 "psubw %%mm1, %%mm0             \n\t"
1292                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1293                 "pmullw %%mm3, %%mm0            \n\t"
1294                 "psllw $7, %%mm1                \n\t"
1295                 "paddw %%mm1, %%mm0             \n\t"
1296
1297                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1298
1299                 "add $8, %%"REG_a"              \n\t"
1300         // End
1301                 "9:                             \n\t"
1302 //              "int $3\n\t"
1303                 "lea 0b, %0                     \n\t"
1304                 "lea 1b, %1                     \n\t"
1305                 "lea 2b, %2                     \n\t"
1306                 "dec %1                         \n\t"
1307                 "dec %2                         \n\t"
1308                 "sub %0, %1                     \n\t"
1309                 "sub %0, %2                     \n\t"
1310                 "lea 9b, %3                     \n\t"
1311                 "sub %0, %3                     \n\t"
1312
1313
1314                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1315                 "=r" (fragmentLengthB)
1316         );
1317
1318         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1319         fragmentPos=0;
1320         
1321         for(i=0; i<dstW/numSplits; i++)
1322         {
1323                 int xx=xpos>>16;
1324
1325                 if((i&3) == 0)
1326                 {
1327                         int a=0;
1328                         int b=((xpos+xInc)>>16) - xx;
1329                         int c=((xpos+xInc*2)>>16) - xx;
1330                         int d=((xpos+xInc*3)>>16) - xx;
1331
1332                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1333                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1334                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1335                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1336                         filterPos[i/2]= xx;
1337
1338                         if(d+1<4)
1339                         {
1340                                 int maxShift= 3-(d+1);
1341                                 int shift=0;
1342
1343                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1344
1345                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1346                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1347                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1348                                         a | (b<<2) | (c<<4) | (d<<6);
1349
1350                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1351                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1352
1353                                 if(shift && i>=shift)
1354                                 {
1355                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1356                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1357                                         filterPos[i/2]-=shift;
1358                                 }
1359
1360                                 fragmentPos+= fragmentLengthB;
1361                         }
1362                         else
1363                         {
1364                                 int maxShift= 3-d;
1365                                 int shift=0;
1366
1367                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1368
1369                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1370                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1371                                         a | (b<<2) | (c<<4) | (d<<6);
1372
1373                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1374                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1375
1376                                 if(shift && i>=shift)
1377                                 {
1378                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1379                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1380                                         filterPos[i/2]-=shift;
1381                                 }
1382
1383                                 fragmentPos+= fragmentLengthA;
1384                         }
1385
1386                         funnyCode[fragmentPos]= RET;
1387                 }
1388                 xpos+=xInc;
1389         }
1390         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1391 }
1392 #endif // ARCH_X86 || ARCH_X86_64
1393
1394 static void globalInit(void){
1395     // generating tables:
1396     int i;
1397     for(i=0; i<768; i++){
1398         int c= MIN(MAX(i-256, 0), 255);
1399         clip_table[i]=c;
1400     }
1401 }
1402
1403 static SwsFunc getSwsFunc(int flags){
1404     
1405 #ifdef RUNTIME_CPUDETECT
1406 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1407         // ordered per speed fasterst first
1408         if(flags & SWS_CPU_CAPS_MMX2)
1409                 return swScale_MMX2;
1410         else if(flags & SWS_CPU_CAPS_3DNOW)
1411                 return swScale_3DNow;
1412         else if(flags & SWS_CPU_CAPS_MMX)
1413                 return swScale_MMX;
1414         else
1415                 return swScale_C;
1416
1417 #else
1418 #ifdef ARCH_POWERPC
1419         if(flags & SWS_CPU_CAPS_ALTIVEC)
1420           return swScale_altivec;
1421         else
1422           return swScale_C;
1423 #endif
1424         return swScale_C;
1425 #endif
1426 #else //RUNTIME_CPUDETECT
1427 #ifdef HAVE_MMX2
1428         return swScale_MMX2;
1429 #elif defined (HAVE_3DNOW)
1430         return swScale_3DNow;
1431 #elif defined (HAVE_MMX)
1432         return swScale_MMX;
1433 #elif defined (HAVE_ALTIVEC)
1434         return swScale_altivec;
1435 #else
1436         return swScale_C;
1437 #endif
1438 #endif //!RUNTIME_CPUDETECT
1439 }
1440
1441 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1442              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1443         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1444         /* Copy Y plane */
1445         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1446                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1447         else
1448         {
1449                 int i;
1450                 uint8_t *srcPtr= src[0];
1451                 uint8_t *dstPtr= dst;
1452                 for(i=0; i<srcSliceH; i++)
1453                 {
1454                         memcpy(dstPtr, srcPtr, c->srcW);
1455                         srcPtr+= srcStride[0];
1456                         dstPtr+= dstStride[0];
1457                 }
1458         }
1459         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1460         if (c->dstFormat == IMGFMT_NV12)
1461                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1462         else
1463                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1464
1465         return srcSliceH;
1466 }
1467
1468 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1469              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1470         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1471
1472         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1473
1474         return srcSliceH;
1475 }
1476
1477 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1478              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1479         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1480
1481         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1482
1483         return srcSliceH;
1484 }
1485
1486 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1487 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1488                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1489         const int srcFormat= c->srcFormat;
1490         const int dstFormat= c->dstFormat;
1491         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1492         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1493         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1494         const int dstId= (dstFormat&0xFF)>>2;
1495         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1496
1497         /* BGR -> BGR */
1498         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1499            || (isRGB(srcFormat) && isRGB(dstFormat))){
1500                 switch(srcId | (dstId<<4)){
1501                 case 0x34: conv= rgb16to15; break;
1502                 case 0x36: conv= rgb24to15; break;
1503                 case 0x38: conv= rgb32to15; break;
1504                 case 0x43: conv= rgb15to16; break;
1505                 case 0x46: conv= rgb24to16; break;
1506                 case 0x48: conv= rgb32to16; break;
1507                 case 0x63: conv= rgb15to24; break;
1508                 case 0x64: conv= rgb16to24; break;
1509                 case 0x68: conv= rgb32to24; break;
1510                 case 0x83: conv= rgb15to32; break;
1511                 case 0x84: conv= rgb16to32; break;
1512                 case 0x86: conv= rgb24to32; break;
1513                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1514                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1515                 }
1516         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1517                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1518                 switch(srcId | (dstId<<4)){
1519                 case 0x33: conv= rgb15tobgr15; break;
1520                 case 0x34: conv= rgb16tobgr15; break;
1521                 case 0x36: conv= rgb24tobgr15; break;
1522                 case 0x38: conv= rgb32tobgr15; break;
1523                 case 0x43: conv= rgb15tobgr16; break;
1524                 case 0x44: conv= rgb16tobgr16; break;
1525                 case 0x46: conv= rgb24tobgr16; break;
1526                 case 0x48: conv= rgb32tobgr16; break;
1527                 case 0x63: conv= rgb15tobgr24; break;
1528                 case 0x64: conv= rgb16tobgr24; break;
1529                 case 0x66: conv= rgb24tobgr24; break;
1530                 case 0x68: conv= rgb32tobgr24; break;
1531                 case 0x83: conv= rgb15tobgr32; break;
1532                 case 0x84: conv= rgb16tobgr32; break;
1533                 case 0x86: conv= rgb24tobgr32; break;
1534                 case 0x88: conv= rgb32tobgr32; break;
1535                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1536                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1537                 }
1538         }else{
1539                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1540                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1541         }
1542
1543         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1544                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1545         else
1546         {
1547                 int i;
1548                 uint8_t *srcPtr= src[0];
1549                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1550
1551                 for(i=0; i<srcSliceH; i++)
1552                 {
1553                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1554                         srcPtr+= srcStride[0];
1555                         dstPtr+= dstStride[0];
1556                 }
1557         }     
1558         return srcSliceH;
1559 }
1560
1561 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1562              int srcSliceH, uint8_t* dst[], int dstStride[]){
1563
1564         rgb24toyv12(
1565                 src[0], 
1566                 dst[0]+ srcSliceY    *dstStride[0], 
1567                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1568                 dst[2]+(srcSliceY>>1)*dstStride[2],
1569                 c->srcW, srcSliceH, 
1570                 dstStride[0], dstStride[1], srcStride[0]);
1571         return srcSliceH;
1572 }
1573
1574 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1575              int srcSliceH, uint8_t* dst[], int dstStride[]){
1576         int i;
1577
1578         /* copy Y */
1579         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1580                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1581         else{
1582                 uint8_t *srcPtr= src[0];
1583                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1584
1585                 for(i=0; i<srcSliceH; i++)
1586                 {
1587                         memcpy(dstPtr, srcPtr, c->srcW);
1588                         srcPtr+= srcStride[0];
1589                         dstPtr+= dstStride[0];
1590                 }
1591         }
1592
1593         if(c->dstFormat==IMGFMT_YV12){
1594                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1595                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1596         }else{
1597                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1598                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1599         }
1600         return srcSliceH;
1601 }
1602
1603 /**
1604  * bring pointers in YUV order instead of YVU
1605  */
1606 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1607         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1608            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1609                 sortedP[0]= p[0];
1610                 sortedP[1]= p[2];
1611                 sortedP[2]= p[1];
1612                 sortedStride[0]= stride[0];
1613                 sortedStride[1]= stride[2];
1614                 sortedStride[2]= stride[1];
1615         }
1616         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1617         {
1618                 sortedP[0]= p[0];
1619                 sortedP[1]= 
1620                 sortedP[2]= NULL;
1621                 sortedStride[0]= stride[0];
1622                 sortedStride[1]= 
1623                 sortedStride[2]= 0;
1624         }
1625         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1626         {
1627                 sortedP[0]= p[0];
1628                 sortedP[1]= p[1];
1629                 sortedP[2]= p[2];
1630                 sortedStride[0]= stride[0];
1631                 sortedStride[1]= stride[1];
1632                 sortedStride[2]= stride[2];
1633         }
1634         else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
1635         {
1636                 sortedP[0]= p[0];
1637                 sortedP[1]= p[1];
1638                 sortedP[2]= NULL;
1639                 sortedStride[0]= stride[0];
1640                 sortedStride[1]= stride[1];
1641                 sortedStride[2]= 0;
1642         }else{
1643                 MSG_ERR("internal error in orderYUV\n");
1644         }
1645 }
1646
1647 /* unscaled copy like stuff (assumes nearly identical formats) */
1648 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1649              int srcSliceH, uint8_t* dst[], int dstStride[]){
1650
1651         if(isPacked(c->srcFormat))
1652         {
1653                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1654                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1655                 else
1656                 {
1657                         int i;
1658                         uint8_t *srcPtr= src[0];
1659                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1660                         int length=0;
1661
1662                         /* universal length finder */
1663                         while(length+c->srcW <= ABS(dstStride[0]) 
1664                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1665                         ASSERT(length!=0);
1666
1667                         for(i=0; i<srcSliceH; i++)
1668                         {
1669                                 memcpy(dstPtr, srcPtr, length);
1670                                 srcPtr+= srcStride[0];
1671                                 dstPtr+= dstStride[0];
1672                         }
1673                 }
1674         }
1675         else 
1676         { /* Planar YUV or gray */
1677                 int plane;
1678                 for(plane=0; plane<3; plane++)
1679                 {
1680                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1681                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1682                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1683
1684                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1685                         {
1686                                 if(!isGray(c->dstFormat))
1687                                         memset(dst[plane], 128, dstStride[plane]*height);
1688                         }
1689                         else
1690                         {
1691                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1692                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1693                                 else
1694                                 {
1695                                         int i;
1696                                         uint8_t *srcPtr= src[plane];
1697                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1698                                         for(i=0; i<height; i++)
1699                                         {
1700                                                 memcpy(dstPtr, srcPtr, length);
1701                                                 srcPtr+= srcStride[plane];
1702                                                 dstPtr+= dstStride[plane];
1703                                         }
1704                                 }
1705                         }
1706                 }
1707         }
1708         return srcSliceH;
1709 }
1710
1711 static int remove_dup_fourcc(int fourcc)
1712 {
1713         switch(fourcc)
1714         {
1715             case IMGFMT_I420:
1716             case IMGFMT_IYUV: return IMGFMT_YV12;
1717             case IMGFMT_Y8  : return IMGFMT_Y800;
1718             case IMGFMT_IF09: return IMGFMT_YVU9;
1719             default: return fourcc;
1720         }
1721 }
1722
1723 static void getSubSampleFactors(int *h, int *v, int format){
1724         switch(format){
1725         case IMGFMT_UYVY:
1726         case IMGFMT_YUY2:
1727                 *h=1;
1728                 *v=0;
1729                 break;
1730         case IMGFMT_YV12:
1731         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1732         case IMGFMT_NV12:
1733         case IMGFMT_NV21:
1734                 *h=1;
1735                 *v=1;
1736                 break;
1737         case IMGFMT_YVU9:
1738                 *h=2;
1739                 *v=2;
1740                 break;
1741         case IMGFMT_444P:
1742                 *h=0;
1743                 *v=0;
1744                 break;
1745         case IMGFMT_422P:
1746                 *h=1;
1747                 *v=0;
1748                 break;
1749         case IMGFMT_411P:
1750                 *h=2;
1751                 *v=0;
1752                 break;
1753         default:
1754                 *h=0;
1755                 *v=0;
1756                 break;
1757         }
1758 }
1759
1760 static uint16_t roundToInt16(int64_t f){
1761         int r= (f + (1<<15))>>16;
1762              if(r<-0x7FFF) return 0x8000;
1763         else if(r> 0x7FFF) return 0x7FFF;
1764         else               return r;
1765 }
1766
1767 /**
1768  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1769  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1770  * @return -1 if not supported
1771  */
1772 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1773         int64_t crv =  inv_table[0];
1774         int64_t cbu =  inv_table[1];
1775         int64_t cgu = -inv_table[2];
1776         int64_t cgv = -inv_table[3];
1777         int64_t cy  = 1<<16;
1778         int64_t oy  = 0;
1779
1780         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1781         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1782         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1783
1784         c->brightness= brightness;
1785         c->contrast  = contrast;
1786         c->saturation= saturation;
1787         c->srcRange  = srcRange;
1788         c->dstRange  = dstRange;
1789
1790         c->uOffset=   0x0400040004000400LL;
1791         c->vOffset=   0x0400040004000400LL;
1792
1793         if(!srcRange){
1794                 cy= (cy*255) / 219;
1795                 oy= 16<<16;
1796         }
1797
1798         cy = (cy *contrast             )>>16;
1799         crv= (crv*contrast * saturation)>>32;
1800         cbu= (cbu*contrast * saturation)>>32;
1801         cgu= (cgu*contrast * saturation)>>32;
1802         cgv= (cgv*contrast * saturation)>>32;
1803
1804         oy -= 256*brightness;
1805
1806         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1807         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1808         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1809         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1810         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1811         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1812
1813         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1814         //FIXME factorize
1815
1816 #ifdef COMPILE_ALTIVEC
1817         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1818             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1819 #endif  
1820         return 0;
1821 }
1822
1823 /**
1824  * @return -1 if not supported
1825  */
1826 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1827         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1828
1829         *inv_table = c->srcColorspaceTable;
1830         *table     = c->dstColorspaceTable;
1831         *srcRange  = c->srcRange;
1832         *dstRange  = c->dstRange;
1833         *brightness= c->brightness;
1834         *contrast  = c->contrast;
1835         *saturation= c->saturation;
1836         
1837         return 0;       
1838 }
1839
1840 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1841                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1842
1843         SwsContext *c;
1844         int i;
1845         int usesVFilter, usesHFilter;
1846         int unscaled, needsDither;
1847         int srcFormat, dstFormat;
1848         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1849 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1850         if(flags & SWS_CPU_CAPS_MMX)
1851                 asm volatile("emms\n\t"::: "memory");
1852 #endif
1853
1854 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1855         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1856 #ifdef HAVE_MMX2
1857         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1858 #elif defined (HAVE_3DNOW)
1859         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1860 #elif defined (HAVE_MMX)
1861         flags |= SWS_CPU_CAPS_MMX;
1862 #elif defined (HAVE_ALTIVEC)
1863         flags |= SWS_CPU_CAPS_ALTIVEC;
1864 #endif
1865 #endif
1866         if(clip_table[512] != 255) globalInit();
1867         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1868
1869         /* avoid duplicate Formats, so we don't need to check to much */
1870         srcFormat = remove_dup_fourcc(origSrcFormat);
1871         dstFormat = remove_dup_fourcc(origDstFormat);
1872
1873         unscaled = (srcW == dstW && srcH == dstH);
1874         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1875                      && (dstFormat&0xFF)<24
1876                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1877
1878         if(!isSupportedIn(srcFormat)) 
1879         {
1880                 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1881                 return NULL;
1882         }
1883         if(!isSupportedOut(dstFormat))
1884         {
1885                 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1886                 return NULL;
1887         }
1888
1889         /* sanity check */
1890         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
1891         {
1892                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1893                         srcW, srcH, dstW, dstH);
1894                 return NULL;
1895         }
1896
1897         if(!dstFilter) dstFilter= &dummyFilter;
1898         if(!srcFilter) srcFilter= &dummyFilter;
1899
1900         c= memalign(64, sizeof(SwsContext));
1901         memset(c, 0, sizeof(SwsContext));
1902
1903         c->srcW= srcW;
1904         c->srcH= srcH;
1905         c->dstW= dstW;
1906         c->dstH= dstH;
1907         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1908         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1909         c->flags= flags;
1910         c->dstFormat= dstFormat;
1911         c->srcFormat= srcFormat;
1912         c->origDstFormat= origDstFormat;
1913         c->origSrcFormat= origSrcFormat;
1914         c->vRounder= 4* 0x0001000100010001ULL;
1915
1916         usesHFilter= usesVFilter= 0;
1917         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1918         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1919         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1920         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1921         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1922         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1923         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1924         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1925
1926         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1927         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1928
1929         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1930         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1931
1932         // drop some chroma lines if the user wants it
1933         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1934         c->chrSrcVSubSample+= c->vChrDrop;
1935
1936         // drop every 2. pixel for chroma calculation unless user wants full chroma
1937         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1938                 c->chrSrcHSubSample=1;
1939
1940         if(param){
1941                 c->param[0] = param[0];
1942                 c->param[1] = param[1];
1943         }else{
1944                 c->param[0] =
1945                 c->param[1] = SWS_PARAM_DEFAULT;
1946         }
1947
1948         c->chrIntHSubSample= c->chrDstHSubSample;
1949         c->chrIntVSubSample= c->chrSrcVSubSample;
1950
1951         // note the -((-x)>>y) is so that we allways round toward +inf
1952         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1953         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1954         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1955         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1956
1957         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); 
1958
1959         /* unscaled special Cases */
1960         if(unscaled && !usesHFilter && !usesVFilter)
1961         {
1962                 /* yv12_to_nv12 */
1963                 if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
1964                 {
1965                         c->swScale= PlanarToNV12Wrapper;
1966                 }
1967                 /* yuv2bgr */
1968                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1969                 {
1970                         c->swScale= yuv2rgb_get_func_ptr(c);
1971                 }
1972                 
1973                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1974                 {
1975                         c->swScale= yvu9toyv12Wrapper;
1976                 }
1977
1978                 /* bgr24toYV12 */
1979                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1980                         c->swScale= bgr24toyv12Wrapper;
1981                 
1982                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1983                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1984                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
1985                    && !needsDither)
1986                         c->swScale= rgb2rgbWrapper;
1987
1988                 /* LQ converters if -sws 0 or -sws 4*/
1989                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1990                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1991                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1992                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
1993                           && needsDither)
1994                                 c->swScale= rgb2rgbWrapper;
1995
1996                         /* yv12_to_yuy2 */
1997                         if(srcFormat == IMGFMT_YV12 && 
1998                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
1999                         {
2000                                 if (dstFormat == IMGFMT_YUY2)
2001                                     c->swScale= PlanarToYuy2Wrapper;
2002                                 else
2003                                     c->swScale= PlanarToUyvyWrapper;
2004                         }
2005                 }
2006
2007 #ifdef COMPILE_ALTIVEC
2008                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2009                     ((srcFormat == IMGFMT_YV12 && 
2010                       (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
2011                   // unscaled YV12 -> packed YUV, we want speed
2012                   if (dstFormat == IMGFMT_YUY2)
2013                     c->swScale= yv12toyuy2_unscaled_altivec;
2014                   else
2015                     c->swScale= yv12touyvy_unscaled_altivec;
2016                 }
2017 #endif
2018
2019                 /* simple copy */
2020                 if(   srcFormat == dstFormat
2021                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2022                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2023                   )
2024                 {
2025                         c->swScale= simpleCopy;
2026                 }
2027
2028                 if(c->swScale){
2029                         if(flags&SWS_PRINT_INFO)
2030                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2031                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2032                         return c;
2033                 }
2034         }
2035
2036         if(flags & SWS_CPU_CAPS_MMX2)
2037         {
2038                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2039                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2040                 {
2041                         if(flags&SWS_PRINT_INFO)
2042                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2043                 }
2044                 if(usesHFilter) c->canMMX2BeUsed=0;
2045         }
2046         else
2047                 c->canMMX2BeUsed=0;
2048
2049         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2050         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2051
2052         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2053         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2054         // n-2 is the last chrominance sample available
2055         // this is not perfect, but noone shuld notice the difference, the more correct variant
2056         // would be like the vertical one, but that would require some special code for the
2057         // first and last pixel
2058         if(flags&SWS_FAST_BILINEAR)
2059         {
2060                 if(c->canMMX2BeUsed)
2061                 {
2062                         c->lumXInc+= 20;
2063                         c->chrXInc+= 20;
2064                 }
2065                 //we don't use the x86asm scaler if mmx is available
2066                 else if(flags & SWS_CPU_CAPS_MMX)
2067                 {
2068                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2069                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2070                 }
2071         }
2072
2073         /* precalculate horizontal scaler filter coefficients */
2074         {
2075                 const int filterAlign=
2076                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2077                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2078                   1;
2079
2080                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2081                                  srcW      ,       dstW, filterAlign, 1<<14,
2082                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2083                                  srcFilter->lumH, dstFilter->lumH, c->param);
2084                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2085                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2086                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2087                                  srcFilter->chrH, dstFilter->chrH, c->param);
2088
2089 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2090 // can't downscale !!!
2091                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2092                 {
2093 #define MAX_FUNNY_CODE_SIZE 10000
2094 #ifdef MAP_ANONYMOUS
2095                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2096                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2097 #else
2098                         c->funnyYCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2099                         c->funnyUVCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
2100 #endif
2101
2102                         c->lumMmx2Filter   = (int16_t*)memalign(8, (dstW        /8+8)*sizeof(int16_t));
2103                         c->chrMmx2Filter   = (int16_t*)memalign(8, (c->chrDstW  /4+8)*sizeof(int16_t));
2104                         c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW      /2/8+8)*sizeof(int32_t));
2105                         c->chrMmx2FilterPos= (int32_t*)memalign(8, (c->chrDstW/2/4+8)*sizeof(int32_t));
2106
2107                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2108                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2109                 }
2110 #endif
2111         } // Init Horizontal stuff
2112
2113
2114
2115         /* precalculate vertical scaler filter coefficients */
2116         {
2117                 const int filterAlign=
2118                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2119                   1;
2120
2121                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2122                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2123                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2124                                 srcFilter->lumV, dstFilter->lumV, c->param);
2125                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2126                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2127                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2128                                 srcFilter->chrV, dstFilter->chrV, c->param);
2129
2130 #ifdef HAVE_ALTIVEC
2131                 c->vYCoeffsBank = memalign (16, sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2132                 c->vCCoeffsBank = memalign (16, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2133
2134                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2135                   int j;
2136                   short *p = (short *)&c->vYCoeffsBank[i];
2137                   for (j=0;j<8;j++)
2138                     p[j] = c->vLumFilter[i];
2139                 }
2140
2141                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2142                   int j;
2143                   short *p = (short *)&c->vCCoeffsBank[i];
2144                   for (j=0;j<8;j++)
2145                     p[j] = c->vChrFilter[i];
2146                 }
2147 #endif
2148         }
2149
2150         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2151         c->vLumBufSize= c->vLumFilterSize;
2152         c->vChrBufSize= c->vChrFilterSize;
2153         for(i=0; i<dstH; i++)
2154         {
2155                 int chrI= i*c->chrDstH / dstH;
2156                 int nextSlice= MAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2157                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2158
2159                 nextSlice>>= c->chrSrcVSubSample;
2160                 nextSlice<<= c->chrSrcVSubSample;
2161                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2162                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2163                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2164                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2165         }
2166
2167         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2168         c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
2169         c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
2170         //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)
2171         /* align at 16 bytes for AltiVec */
2172         for(i=0; i<c->vLumBufSize; i++)
2173                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(16, 4000);
2174         for(i=0; i<c->vChrBufSize; i++)
2175                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(16, 8000);
2176
2177         //try to avoid drawing green stuff between the right end and the stride end
2178         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2179         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2180
2181         ASSERT(c->chrDstH <= dstH)
2182
2183         if(flags&SWS_PRINT_INFO)
2184         {
2185 #ifdef DITHER1XBPP
2186                 char *dither= " dithered";
2187 #else
2188                 char *dither= "";
2189 #endif
2190                 if(flags&SWS_FAST_BILINEAR)
2191                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2192                 else if(flags&SWS_BILINEAR)
2193                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2194                 else if(flags&SWS_BICUBIC)
2195                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2196                 else if(flags&SWS_X)
2197                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2198                 else if(flags&SWS_POINT)
2199                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2200                 else if(flags&SWS_AREA)
2201                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2202                 else if(flags&SWS_BICUBLIN)
2203                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2204                 else if(flags&SWS_GAUSS)
2205                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2206                 else if(flags&SWS_SINC)
2207                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2208                 else if(flags&SWS_LANCZOS)
2209                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2210                 else if(flags&SWS_SPLINE)
2211                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2212                 else
2213                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2214
2215                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2216                         MSG_INFO("from %s to%s %s ", 
2217                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2218                 else
2219                         MSG_INFO("from %s to %s ", 
2220                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2221
2222                 if(flags & SWS_CPU_CAPS_MMX2)
2223                         MSG_INFO("using MMX2\n");
2224                 else if(flags & SWS_CPU_CAPS_3DNOW)
2225                         MSG_INFO("using 3DNOW\n");
2226                 else if(flags & SWS_CPU_CAPS_MMX)
2227                         MSG_INFO("using MMX\n");
2228                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2229                         MSG_INFO("using AltiVec\n");
2230                 else 
2231                         MSG_INFO("using C\n");
2232         }
2233
2234         if(flags & SWS_PRINT_INFO)
2235         {
2236                 if(flags & SWS_CPU_CAPS_MMX)
2237                 {
2238                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2239                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2240                         else
2241                         {
2242                                 if(c->hLumFilterSize==4)
2243                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2244                                 else if(c->hLumFilterSize==8)
2245                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2246                                 else
2247                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2248
2249                                 if(c->hChrFilterSize==4)
2250                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2251                                 else if(c->hChrFilterSize==8)
2252                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2253                                 else
2254                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2255                         }
2256                 }
2257                 else
2258                 {
2259 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2260                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2261 #else
2262                         if(flags & SWS_FAST_BILINEAR)
2263                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2264                         else
2265                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2266 #endif
2267                 }
2268                 if(isPlanarYUV(dstFormat))
2269                 {
2270                         if(c->vLumFilterSize==1)
2271                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2272                         else
2273                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2274                 }
2275                 else
2276                 {
2277                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2278                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2279                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2280                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2281                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2282                         else
2283                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2284                 }
2285
2286                 if(dstFormat==IMGFMT_BGR24)
2287                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2288                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2289                 else if(dstFormat==IMGFMT_BGR32)
2290                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2291                 else if(dstFormat==IMGFMT_BGR16)
2292                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2293                 else if(dstFormat==IMGFMT_BGR15)
2294                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2295
2296                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2297         }
2298         if(flags & SWS_PRINT_INFO)
2299         {
2300                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2301                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2302                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2303                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2304         }
2305
2306         c->swScale= getSwsFunc(flags);
2307         return c;
2308 }
2309
2310 /**
2311  * swscale warper, so we don't need to export the SwsContext.
2312  * assumes planar YUV to be in YUV order instead of YVU
2313  */
2314 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2315                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2316         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2317             MSG_ERR("swScaler: slices start in the middle!\n");
2318             return 0;
2319         }
2320         if (c->sliceDir == 0) {
2321             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2322         }
2323
2324         // copy strides, so they can safely be modified
2325         if (c->sliceDir == 1) {
2326             // slices go from top to bottom
2327             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2328             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2329             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2330         } else {
2331             // slices go from bottom to top => we flip the image internally
2332             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2333                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2334                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2335             };
2336             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2337                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2338                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2339             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2340             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2341             
2342             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2343         }
2344 }
2345
2346 /**
2347  * swscale warper, so we don't need to export the SwsContext
2348  */
2349 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2350                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2351         int srcStride[3];
2352         int dstStride[3];
2353         uint8_t *src[3];
2354         uint8_t *dst[3];
2355         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2356         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2357 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2358
2359         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2360 }
2361
2362 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2363                                 float lumaSharpen, float chromaSharpen,
2364                                 float chromaHShift, float chromaVShift,
2365                                 int verbose)
2366 {
2367         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2368
2369         if(lumaGBlur!=0.0){
2370                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2371                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2372         }else{
2373                 filter->lumH= sws_getIdentityVec();
2374                 filter->lumV= sws_getIdentityVec();
2375         }
2376
2377         if(chromaGBlur!=0.0){
2378                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2379                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2380         }else{
2381                 filter->chrH= sws_getIdentityVec();
2382                 filter->chrV= sws_getIdentityVec();
2383         }
2384
2385         if(chromaSharpen!=0.0){
2386                 SwsVector *id= sws_getIdentityVec();
2387                 sws_scaleVec(filter->chrH, -chromaSharpen);
2388                 sws_scaleVec(filter->chrV, -chromaSharpen);
2389                 sws_addVec(filter->chrH, id);
2390                 sws_addVec(filter->chrV, id);
2391                 sws_freeVec(id);
2392         }
2393
2394         if(lumaSharpen!=0.0){
2395                 SwsVector *id= sws_getIdentityVec();
2396                 sws_scaleVec(filter->lumH, -lumaSharpen);
2397                 sws_scaleVec(filter->lumV, -lumaSharpen);
2398                 sws_addVec(filter->lumH, id);
2399                 sws_addVec(filter->lumV, id);
2400                 sws_freeVec(id);
2401         }
2402
2403         if(chromaHShift != 0.0)
2404                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2405
2406         if(chromaVShift != 0.0)
2407                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2408
2409         sws_normalizeVec(filter->chrH, 1.0);
2410         sws_normalizeVec(filter->chrV, 1.0);
2411         sws_normalizeVec(filter->lumH, 1.0);
2412         sws_normalizeVec(filter->lumV, 1.0);
2413
2414         if(verbose) sws_printVec(filter->chrH);
2415         if(verbose) sws_printVec(filter->lumH);
2416
2417         return filter;
2418 }
2419
2420 /**
2421  * returns a normalized gaussian curve used to filter stuff
2422  * quality=3 is high quality, lowwer is lowwer quality
2423  */
2424 SwsVector *sws_getGaussianVec(double variance, double quality){
2425         const int length= (int)(variance*quality + 0.5) | 1;
2426         int i;
2427         double *coeff= memalign(sizeof(double), length*sizeof(double));
2428         double middle= (length-1)*0.5;
2429         SwsVector *vec= av_malloc(sizeof(SwsVector));
2430
2431         vec->coeff= coeff;
2432         vec->length= length;
2433
2434         for(i=0; i<length; i++)
2435         {
2436                 double dist= i-middle;
2437                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2438         }
2439
2440         sws_normalizeVec(vec, 1.0);
2441
2442         return vec;
2443 }
2444
2445 SwsVector *sws_getConstVec(double c, int length){
2446         int i;
2447         double *coeff= memalign(sizeof(double), length*sizeof(double));
2448         SwsVector *vec= av_malloc(sizeof(SwsVector));
2449
2450         vec->coeff= coeff;
2451         vec->length= length;
2452
2453         for(i=0; i<length; i++)
2454                 coeff[i]= c;
2455
2456         return vec;
2457 }
2458
2459
2460 SwsVector *sws_getIdentityVec(void){
2461         return sws_getConstVec(1.0, 1);
2462 }
2463
2464 double sws_dcVec(SwsVector *a){
2465         int i;
2466         double sum=0;
2467
2468         for(i=0; i<a->length; i++)
2469                 sum+= a->coeff[i];
2470
2471         return sum;
2472 }
2473
2474 void sws_scaleVec(SwsVector *a, double scalar){
2475         int i;
2476
2477         for(i=0; i<a->length; i++)
2478                 a->coeff[i]*= scalar;
2479 }
2480
2481 void sws_normalizeVec(SwsVector *a, double height){
2482         sws_scaleVec(a, height/sws_dcVec(a));
2483 }
2484
2485 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2486         int length= a->length + b->length - 1;
2487         double *coeff= memalign(sizeof(double), length*sizeof(double));
2488         int i, j;
2489         SwsVector *vec= av_malloc(sizeof(SwsVector));
2490
2491         vec->coeff= coeff;
2492         vec->length= length;
2493
2494         for(i=0; i<length; i++) coeff[i]= 0.0;
2495
2496         for(i=0; i<a->length; i++)
2497         {
2498                 for(j=0; j<b->length; j++)
2499                 {
2500                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2501                 }
2502         }
2503
2504         return vec;
2505 }
2506
2507 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2508         int length= MAX(a->length, b->length);
2509         double *coeff= memalign(sizeof(double), length*sizeof(double));
2510         int i;
2511         SwsVector *vec= av_malloc(sizeof(SwsVector));
2512
2513         vec->coeff= coeff;
2514         vec->length= length;
2515
2516         for(i=0; i<length; i++) coeff[i]= 0.0;
2517
2518         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2519         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2520
2521         return vec;
2522 }
2523
2524 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2525         int length= MAX(a->length, b->length);
2526         double *coeff= memalign(sizeof(double), length*sizeof(double));
2527         int i;
2528         SwsVector *vec= av_malloc(sizeof(SwsVector));
2529
2530         vec->coeff= coeff;
2531         vec->length= length;
2532
2533         for(i=0; i<length; i++) coeff[i]= 0.0;
2534
2535         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2536         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2537
2538         return vec;
2539 }
2540
2541 /* shift left / or right if "shift" is negative */
2542 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2543         int length= a->length + ABS(shift)*2;
2544         double *coeff= memalign(sizeof(double), length*sizeof(double));
2545         int i;
2546         SwsVector *vec= av_malloc(sizeof(SwsVector));
2547
2548         vec->coeff= coeff;
2549         vec->length= length;
2550
2551         for(i=0; i<length; i++) coeff[i]= 0.0;
2552
2553         for(i=0; i<a->length; i++)
2554         {
2555                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2556         }
2557
2558         return vec;
2559 }
2560
2561 void sws_shiftVec(SwsVector *a, int shift){
2562         SwsVector *shifted= sws_getShiftedVec(a, shift);
2563         av_free(a->coeff);
2564         a->coeff= shifted->coeff;
2565         a->length= shifted->length;
2566         av_free(shifted);
2567 }
2568
2569 void sws_addVec(SwsVector *a, SwsVector *b){
2570         SwsVector *sum= sws_sumVec(a, b);
2571         av_free(a->coeff);
2572         a->coeff= sum->coeff;
2573         a->length= sum->length;
2574         av_free(sum);
2575 }
2576
2577 void sws_subVec(SwsVector *a, SwsVector *b){
2578         SwsVector *diff= sws_diffVec(a, b);
2579         av_free(a->coeff);
2580         a->coeff= diff->coeff;
2581         a->length= diff->length;
2582         av_free(diff);
2583 }
2584
2585 void sws_convVec(SwsVector *a, SwsVector *b){
2586         SwsVector *conv= sws_getConvVec(a, b);
2587         av_free(a->coeff);  
2588         a->coeff= conv->coeff;
2589         a->length= conv->length;
2590         av_free(conv);
2591 }
2592
2593 SwsVector *sws_cloneVec(SwsVector *a){
2594         double *coeff= memalign(sizeof(double), a->length*sizeof(double));
2595         int i;
2596         SwsVector *vec= av_malloc(sizeof(SwsVector));
2597
2598         vec->coeff= coeff;
2599         vec->length= a->length;
2600
2601         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2602
2603         return vec;
2604 }
2605
2606 void sws_printVec(SwsVector *a){
2607         int i;
2608         double max=0;
2609         double min=0;
2610         double range;
2611
2612         for(i=0; i<a->length; i++)
2613                 if(a->coeff[i]>max) max= a->coeff[i];
2614
2615         for(i=0; i<a->length; i++)
2616                 if(a->coeff[i]<min) min= a->coeff[i];
2617
2618         range= max - min;
2619
2620         for(i=0; i<a->length; i++)
2621         {
2622                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2623                 MSG_DBG2("%1.3f ", a->coeff[i]);
2624                 for(;x>0; x--) MSG_DBG2(" ");
2625                 MSG_DBG2("|\n");
2626         }
2627 }
2628
2629 void sws_freeVec(SwsVector *a){
2630         if(!a) return;
2631         av_free(a->coeff);
2632         a->coeff=NULL;
2633         a->length=0;
2634         av_free(a);
2635 }
2636
2637 void sws_freeFilter(SwsFilter *filter){
2638         if(!filter) return;
2639
2640         if(filter->lumH) sws_freeVec(filter->lumH);
2641         if(filter->lumV) sws_freeVec(filter->lumV);
2642         if(filter->chrH) sws_freeVec(filter->chrH);
2643         if(filter->chrV) sws_freeVec(filter->chrV);
2644         av_free(filter);
2645 }
2646
2647
2648 void sws_freeContext(SwsContext *c){
2649         int i;
2650         if(!c) return;
2651
2652         if(c->lumPixBuf)
2653         {
2654                 for(i=0; i<c->vLumBufSize; i++)
2655                 {
2656                         av_free(c->lumPixBuf[i]);
2657                         c->lumPixBuf[i]=NULL;
2658                 }
2659                 av_free(c->lumPixBuf);
2660                 c->lumPixBuf=NULL;
2661         }
2662
2663         if(c->chrPixBuf)
2664         {
2665                 for(i=0; i<c->vChrBufSize; i++)
2666                 {
2667                         av_free(c->chrPixBuf[i]);
2668                         c->chrPixBuf[i]=NULL;
2669                 }
2670                 av_free(c->chrPixBuf);
2671                 c->chrPixBuf=NULL;
2672         }
2673
2674         av_free(c->vLumFilter);
2675         c->vLumFilter = NULL;
2676         av_free(c->vChrFilter);
2677         c->vChrFilter = NULL;
2678         av_free(c->hLumFilter);
2679         c->hLumFilter = NULL;
2680         av_free(c->hChrFilter);
2681         c->hChrFilter = NULL;
2682 #ifdef HAVE_ALTIVEC
2683         av_free(c->vYCoeffsBank);
2684         c->vYCoeffsBank = NULL;
2685         av_free(c->vCCoeffsBank);
2686         c->vCCoeffsBank = NULL;
2687 #endif
2688
2689         av_free(c->vLumFilterPos);
2690         c->vLumFilterPos = NULL;
2691         av_free(c->vChrFilterPos);
2692         c->vChrFilterPos = NULL;
2693         av_free(c->hLumFilterPos);
2694         c->hLumFilterPos = NULL;
2695         av_free(c->hChrFilterPos);
2696         c->hChrFilterPos = NULL;
2697
2698 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2699 #ifdef MAP_ANONYMOUS
2700         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2701         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2702 #else
2703         av_free(c->funnyYCode);
2704         av_free(c->funnyUVCode);
2705 #endif
2706         c->funnyYCode=NULL;
2707         c->funnyUVCode=NULL;
2708 #endif
2709
2710         av_free(c->lumMmx2Filter);
2711         c->lumMmx2Filter=NULL;
2712         av_free(c->chrMmx2Filter);
2713         c->chrMmx2Filter=NULL;
2714         av_free(c->lumMmx2FilterPos);
2715         c->lumMmx2FilterPos=NULL;
2716         av_free(c->chrMmx2FilterPos);
2717         c->chrMmx2FilterPos=NULL;
2718         av_free(c->yuvTable);
2719         c->yuvTable=NULL;
2720
2721         av_free(c);
2722 }
2723