Replace MIN() and MAX() with FFMIN() and FFMAX()
[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 #if defined(ARCH_X86) || defined(ARCH_X86_64)
155 static uint64_t attribute_used __attribute__((aligned(8))) bF8=       0xF8F8F8F8F8F8F8F8LL;
156 static uint64_t attribute_used __attribute__((aligned(8))) bFC=       0xFCFCFCFCFCFCFCFCLL;
157 static uint64_t __attribute__((aligned(8))) w10=       0x0010001000100010LL;
158 static uint64_t attribute_used __attribute__((aligned(8))) w02=       0x0002000200020002LL;
159 static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
160 static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
161 static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
162 static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
163
164 static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
165 static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
166 static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
167 static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
168
169 static uint64_t __attribute__((aligned(8))) dither4[2]={
170         0x0103010301030103LL,
171         0x0200020002000200LL,};
172
173 static uint64_t __attribute__((aligned(8))) dither8[2]={
174         0x0602060206020602LL,
175         0x0004000400040004LL,};
176
177 static uint64_t __attribute__((aligned(8))) b16Mask=   0x001F001F001F001FLL;
178 static uint64_t attribute_used __attribute__((aligned(8))) g16Mask=   0x07E007E007E007E0LL;
179 static uint64_t attribute_used __attribute__((aligned(8))) r16Mask=   0xF800F800F800F800LL;
180 static uint64_t __attribute__((aligned(8))) b15Mask=   0x001F001F001F001FLL;
181 static uint64_t attribute_used __attribute__((aligned(8))) g15Mask=   0x03E003E003E003E0LL;
182 static uint64_t attribute_used __attribute__((aligned(8))) r15Mask=   0x7C007C007C007C00LL;
183
184 static uint64_t attribute_used __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
185 static uint64_t attribute_used __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
186 static uint64_t attribute_used __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
187
188 #ifdef FAST_BGR2YV12
189 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
190 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
191 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
192 #else
193 static const uint64_t bgr2YCoeff  attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
194 static const uint64_t bgr2UCoeff  attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
195 static const uint64_t bgr2VCoeff  attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
196 #endif
197 static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
198 static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
199 static const uint64_t w1111       attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
200 #endif
201
202 // clipping helper table for C implementations:
203 static unsigned char clip_table[768];
204
205 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
206                   
207 extern const uint8_t dither_2x2_4[2][8];
208 extern const uint8_t dither_2x2_8[2][8];
209 extern const uint8_t dither_8x8_32[8][8];
210 extern const uint8_t dither_8x8_73[8][8];
211 extern const uint8_t dither_8x8_220[8][8];
212
213 char *sws_format_name(int format)
214 {
215     static char fmt_name[64];
216     char *res;
217     static int buffer;
218
219     res = fmt_name + buffer * 32;
220     buffer = 1 - buffer;
221     snprintf(res, 32, "0x%x (%c%c%c%c)", format,
222                     format >> 24, (format >> 16) & 0xFF,
223                     (format >> 8) & 0xFF,
224                     format & 0xFF);
225
226     return res;
227 }
228
229 #if defined(ARCH_X86) || defined(ARCH_X86_64)
230 void in_asm_used_var_warning_killer()
231 {
232  volatile int i= bF8+bFC+w10+
233  bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
234  M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0]+bm01010101;
235  if(i) i=0;
236 }
237 #endif
238
239 static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
240                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
241                                     uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
242 {
243         //FIXME Optimize (just quickly writen not opti..)
244         int i;
245         for(i=0; i<dstW; i++)
246         {
247                 int val=1<<18;
248                 int j;
249                 for(j=0; j<lumFilterSize; j++)
250                         val += lumSrc[j][i] * lumFilter[j];
251
252                 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
253         }
254
255         if(uDest != NULL)
256                 for(i=0; i<chrDstW; i++)
257                 {
258                         int u=1<<18;
259                         int v=1<<18;
260                         int j;
261                         for(j=0; j<chrFilterSize; j++)
262                         {
263                                 u += chrSrc[j][i] * chrFilter[j];
264                                 v += chrSrc[j][i + 2048] * chrFilter[j];
265                         }
266
267                         uDest[i]= FFMIN(FFMAX(u>>19, 0), 255);
268                         vDest[i]= FFMIN(FFMAX(v>>19, 0), 255);
269                 }
270 }
271
272 static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
273                                 int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
274                                 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
275 {
276         //FIXME Optimize (just quickly writen not opti..)
277         int i;
278         for(i=0; i<dstW; i++)
279         {
280                 int val=1<<18;
281                 int j;
282                 for(j=0; j<lumFilterSize; j++)
283                         val += lumSrc[j][i] * lumFilter[j];
284
285                 dest[i]= FFMIN(FFMAX(val>>19, 0), 255);
286         }
287
288         if(uDest == NULL)
289                 return;
290
291         if(dstFormat == IMGFMT_NV12)
292                 for(i=0; i<chrDstW; i++)
293                 {
294                         int u=1<<18;
295                         int v=1<<18;
296                         int j;
297                         for(j=0; j<chrFilterSize; j++)
298                         {
299                                 u += chrSrc[j][i] * chrFilter[j];
300                                 v += chrSrc[j][i + 2048] * chrFilter[j];
301                         }
302
303                         uDest[2*i]= FFMIN(FFMAX(u>>19, 0), 255);
304                         uDest[2*i+1]= FFMIN(FFMAX(v>>19, 0), 255);
305                 }
306         else
307                 for(i=0; i<chrDstW; i++)
308                 {
309                         int u=1<<18;
310                         int v=1<<18;
311                         int j;
312                         for(j=0; j<chrFilterSize; j++)
313                         {
314                                 u += chrSrc[j][i] * chrFilter[j];
315                                 v += chrSrc[j][i + 2048] * chrFilter[j];
316                         }
317
318                         uDest[2*i]= FFMIN(FFMAX(v>>19, 0), 255);
319                         uDest[2*i+1]= FFMIN(FFMAX(u>>19, 0), 255);
320                 }
321 }
322
323 #define YSCALE_YUV_2_PACKEDX_C(type) \
324                 for(i=0; i<(dstW>>1); i++){\
325                         int j;\
326                         int Y1=1<<18;\
327                         int Y2=1<<18;\
328                         int U=1<<18;\
329                         int V=1<<18;\
330                         type *r, *b, *g;\
331                         const int i2= 2*i;\
332                         \
333                         for(j=0; j<lumFilterSize; j++)\
334                         {\
335                                 Y1 += lumSrc[j][i2] * lumFilter[j];\
336                                 Y2 += lumSrc[j][i2+1] * lumFilter[j];\
337                         }\
338                         for(j=0; j<chrFilterSize; j++)\
339                         {\
340                                 U += chrSrc[j][i] * chrFilter[j];\
341                                 V += chrSrc[j][i+2048] * chrFilter[j];\
342                         }\
343                         Y1>>=19;\
344                         Y2>>=19;\
345                         U >>=19;\
346                         V >>=19;\
347                         if((Y1|Y2|U|V)&256)\
348                         {\
349                                 if(Y1>255)   Y1=255;\
350                                 else if(Y1<0)Y1=0;\
351                                 if(Y2>255)   Y2=255;\
352                                 else if(Y2<0)Y2=0;\
353                                 if(U>255)    U=255;\
354                                 else if(U<0) U=0;\
355                                 if(V>255)    V=255;\
356                                 else if(V<0) V=0;\
357                         }
358                         
359 #define YSCALE_YUV_2_RGBX_C(type) \
360                         YSCALE_YUV_2_PACKEDX_C(type)\
361                         r = c->table_rV[V];\
362                         g = c->table_gU[U] + c->table_gV[V];\
363                         b = c->table_bU[U];\
364
365 #define YSCALE_YUV_2_PACKED2_C \
366                 for(i=0; i<(dstW>>1); i++){\
367                         const int i2= 2*i;\
368                         int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;\
369                         int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;\
370                         int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;\
371                         int V= (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;\
372
373 #define YSCALE_YUV_2_RGB2_C(type) \
374                         YSCALE_YUV_2_PACKED2_C\
375                         type *r, *b, *g;\
376                         r = c->table_rV[V];\
377                         g = c->table_gU[U] + c->table_gV[V];\
378                         b = c->table_bU[U];\
379
380 #define YSCALE_YUV_2_PACKED1_C \
381                 for(i=0; i<(dstW>>1); i++){\
382                         const int i2= 2*i;\
383                         int Y1= buf0[i2  ]>>7;\
384                         int Y2= buf0[i2+1]>>7;\
385                         int U= (uvbuf1[i     ])>>7;\
386                         int V= (uvbuf1[i+2048])>>7;\
387
388 #define YSCALE_YUV_2_RGB1_C(type) \
389                         YSCALE_YUV_2_PACKED1_C\
390                         type *r, *b, *g;\
391                         r = c->table_rV[V];\
392                         g = c->table_gU[U] + c->table_gV[V];\
393                         b = c->table_bU[U];\
394
395 #define YSCALE_YUV_2_PACKED1B_C \
396                 for(i=0; i<(dstW>>1); i++){\
397                         const int i2= 2*i;\
398                         int Y1= buf0[i2  ]>>7;\
399                         int Y2= buf0[i2+1]>>7;\
400                         int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
401                         int V= (uvbuf0[i+2048] + uvbuf1[i+2048])>>8;\
402
403 #define YSCALE_YUV_2_RGB1B_C(type) \
404                         YSCALE_YUV_2_PACKED1B_C\
405                         type *r, *b, *g;\
406                         r = c->table_rV[V];\
407                         g = c->table_gU[U] + c->table_gV[V];\
408                         b = c->table_bU[U];\
409
410 #define YSCALE_YUV_2_ANYRGB_C(func, func2)\
411         switch(c->dstFormat)\
412         {\
413         case IMGFMT_BGR32:\
414         case IMGFMT_RGB32:\
415                 func(uint32_t)\
416                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
417                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
418                 }               \
419                 break;\
420         case IMGFMT_RGB24:\
421                 func(uint8_t)\
422                         ((uint8_t*)dest)[0]= r[Y1];\
423                         ((uint8_t*)dest)[1]= g[Y1];\
424                         ((uint8_t*)dest)[2]= b[Y1];\
425                         ((uint8_t*)dest)[3]= r[Y2];\
426                         ((uint8_t*)dest)[4]= g[Y2];\
427                         ((uint8_t*)dest)[5]= b[Y2];\
428                         dest+=6;\
429                 }\
430                 break;\
431         case IMGFMT_BGR24:\
432                 func(uint8_t)\
433                         ((uint8_t*)dest)[0]= b[Y1];\
434                         ((uint8_t*)dest)[1]= g[Y1];\
435                         ((uint8_t*)dest)[2]= r[Y1];\
436                         ((uint8_t*)dest)[3]= b[Y2];\
437                         ((uint8_t*)dest)[4]= g[Y2];\
438                         ((uint8_t*)dest)[5]= r[Y2];\
439                         dest+=6;\
440                 }\
441                 break;\
442         case IMGFMT_RGB16:\
443         case IMGFMT_BGR16:\
444                 {\
445                         const int dr1= dither_2x2_8[y&1    ][0];\
446                         const int dg1= dither_2x2_4[y&1    ][0];\
447                         const int db1= dither_2x2_8[(y&1)^1][0];\
448                         const int dr2= dither_2x2_8[y&1    ][1];\
449                         const int dg2= dither_2x2_4[y&1    ][1];\
450                         const int db2= dither_2x2_8[(y&1)^1][1];\
451                         func(uint16_t)\
452                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
453                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
454                         }\
455                 }\
456                 break;\
457         case IMGFMT_RGB15:\
458         case IMGFMT_BGR15:\
459                 {\
460                         const int dr1= dither_2x2_8[y&1    ][0];\
461                         const int dg1= dither_2x2_8[y&1    ][1];\
462                         const int db1= dither_2x2_8[(y&1)^1][0];\
463                         const int dr2= dither_2x2_8[y&1    ][1];\
464                         const int dg2= dither_2x2_8[y&1    ][0];\
465                         const int db2= dither_2x2_8[(y&1)^1][1];\
466                         func(uint16_t)\
467                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
468                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
469                         }\
470                 }\
471                 break;\
472         case IMGFMT_RGB8:\
473         case IMGFMT_BGR8:\
474                 {\
475                         const uint8_t * const d64= dither_8x8_73[y&7];\
476                         const uint8_t * const d32= dither_8x8_32[y&7];\
477                         func(uint8_t)\
478                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
479                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
480                         }\
481                 }\
482                 break;\
483         case IMGFMT_RGB4:\
484         case IMGFMT_BGR4:\
485                 {\
486                         const uint8_t * const d64= dither_8x8_73 [y&7];\
487                         const uint8_t * const d128=dither_8x8_220[y&7];\
488                         func(uint8_t)\
489                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
490                                                  + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
491                         }\
492                 }\
493                 break;\
494         case IMGFMT_RG4B:\
495         case IMGFMT_BG4B:\
496                 {\
497                         const uint8_t * const d64= dither_8x8_73 [y&7];\
498                         const uint8_t * const d128=dither_8x8_220[y&7];\
499                         func(uint8_t)\
500                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
501                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
502                         }\
503                 }\
504                 break;\
505         case IMGFMT_RGB1:\
506         case IMGFMT_BGR1:\
507                 {\
508                         const uint8_t * const d128=dither_8x8_220[y&7];\
509                         uint8_t *g= c->table_gU[128] + c->table_gV[128];\
510                         for(i=0; i<dstW-7; i+=8){\
511                                 int acc;\
512                                 acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
513                                 acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
514                                 acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
515                                 acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
516                                 acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
517                                 acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
518                                 acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
519                                 acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
520                                 ((uint8_t*)dest)[0]= acc;\
521                                 dest++;\
522                         }\
523 \
524 /*\
525 ((uint8_t*)dest)-= dstW>>4;\
526 {\
527                         int acc=0;\
528                         int left=0;\
529                         static int top[1024];\
530                         static int last_new[1024][1024];\
531                         static int last_in3[1024][1024];\
532                         static int drift[1024][1024];\
533                         int topLeft=0;\
534                         int shift=0;\
535                         int count=0;\
536                         const uint8_t * const d128=dither_8x8_220[y&7];\
537                         int error_new=0;\
538                         int error_in3=0;\
539                         int f=0;\
540                         \
541                         for(i=dstW>>1; i<dstW; i++){\
542                                 int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
543                                 int in2 = (76309 * (in - 16) + 32768) >> 16;\
544                                 int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
545                                 int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
546                                         + (last_new[y][i] - in3)*f/256;\
547                                 int new= old> 128 ? 255 : 0;\
548 \
549                                 error_new+= ABS(last_new[y][i] - new);\
550                                 error_in3+= ABS(last_in3[y][i] - in3);\
551                                 f= error_new - error_in3*4;\
552                                 if(f<0) f=0;\
553                                 if(f>256) f=256;\
554 \
555                                 topLeft= top[i];\
556                                 left= top[i]= old - new;\
557                                 last_new[y][i]= new;\
558                                 last_in3[y][i]= in3;\
559 \
560                                 acc+= acc + (new&1);\
561                                 if((i&7)==6){\
562                                         ((uint8_t*)dest)[0]= acc;\
563                                         ((uint8_t*)dest)++;\
564                                 }\
565                         }\
566 }\
567 */\
568                 }\
569                 break;\
570         case IMGFMT_YUY2:\
571                 func2\
572                         ((uint8_t*)dest)[2*i2+0]= Y1;\
573                         ((uint8_t*)dest)[2*i2+1]= U;\
574                         ((uint8_t*)dest)[2*i2+2]= Y2;\
575                         ((uint8_t*)dest)[2*i2+3]= V;\
576                 }               \
577                 break;\
578         case IMGFMT_UYVY:\
579                 func2\
580                         ((uint8_t*)dest)[2*i2+0]= U;\
581                         ((uint8_t*)dest)[2*i2+1]= Y1;\
582                         ((uint8_t*)dest)[2*i2+2]= V;\
583                         ((uint8_t*)dest)[2*i2+3]= Y2;\
584                 }               \
585                 break;\
586         }\
587
588
589 static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
590                                     int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
591                                     uint8_t *dest, int dstW, int y)
592 {
593         int i;
594         switch(c->dstFormat)
595         {
596         case IMGFMT_RGB32:
597         case IMGFMT_BGR32:
598                 YSCALE_YUV_2_RGBX_C(uint32_t)
599                         ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
600                         ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
601                 }
602                 break;
603         case IMGFMT_RGB24:
604                 YSCALE_YUV_2_RGBX_C(uint8_t)
605                         ((uint8_t*)dest)[0]= r[Y1];
606                         ((uint8_t*)dest)[1]= g[Y1];
607                         ((uint8_t*)dest)[2]= b[Y1];
608                         ((uint8_t*)dest)[3]= r[Y2];
609                         ((uint8_t*)dest)[4]= g[Y2];
610                         ((uint8_t*)dest)[5]= b[Y2];
611                         dest+=6;
612                 }
613                 break;
614         case IMGFMT_BGR24:
615                 YSCALE_YUV_2_RGBX_C(uint8_t)
616                         ((uint8_t*)dest)[0]= b[Y1];
617                         ((uint8_t*)dest)[1]= g[Y1];
618                         ((uint8_t*)dest)[2]= r[Y1];
619                         ((uint8_t*)dest)[3]= b[Y2];
620                         ((uint8_t*)dest)[4]= g[Y2];
621                         ((uint8_t*)dest)[5]= r[Y2];
622                         dest+=6;
623                 }
624                 break;
625         case IMGFMT_RGB16:
626         case IMGFMT_BGR16:
627                 {
628                         const int dr1= dither_2x2_8[y&1    ][0];
629                         const int dg1= dither_2x2_4[y&1    ][0];
630                         const int db1= dither_2x2_8[(y&1)^1][0];
631                         const int dr2= dither_2x2_8[y&1    ][1];
632                         const int dg2= dither_2x2_4[y&1    ][1];
633                         const int db2= dither_2x2_8[(y&1)^1][1];
634                         YSCALE_YUV_2_RGBX_C(uint16_t)
635                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
636                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
637                         }
638                 }
639                 break;
640         case IMGFMT_RGB15:
641         case IMGFMT_BGR15:
642                 {
643                         const int dr1= dither_2x2_8[y&1    ][0];
644                         const int dg1= dither_2x2_8[y&1    ][1];
645                         const int db1= dither_2x2_8[(y&1)^1][0];
646                         const int dr2= dither_2x2_8[y&1    ][1];
647                         const int dg2= dither_2x2_8[y&1    ][0];
648                         const int db2= dither_2x2_8[(y&1)^1][1];
649                         YSCALE_YUV_2_RGBX_C(uint16_t)
650                                 ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
651                                 ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
652                         }
653                 }
654                 break;
655         case IMGFMT_RGB8:
656         case IMGFMT_BGR8:
657                 {
658                         const uint8_t * const d64= dither_8x8_73[y&7];
659                         const uint8_t * const d32= dither_8x8_32[y&7];
660                         YSCALE_YUV_2_RGBX_C(uint8_t)
661                                 ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
662                                 ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
663                         }
664                 }
665                 break;
666         case IMGFMT_RGB4:
667         case IMGFMT_BGR4:
668                 {
669                         const uint8_t * const d64= dither_8x8_73 [y&7];
670                         const uint8_t * const d128=dither_8x8_220[y&7];
671                         YSCALE_YUV_2_RGBX_C(uint8_t)
672                                 ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
673                                                   +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
674                         }
675                 }
676                 break;
677         case IMGFMT_RG4B:
678         case IMGFMT_BG4B:
679                 {
680                         const uint8_t * const d64= dither_8x8_73 [y&7];
681                         const uint8_t * const d128=dither_8x8_220[y&7];
682                         YSCALE_YUV_2_RGBX_C(uint8_t)
683                                 ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
684                                 ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
685                         }
686                 }
687                 break;
688         case IMGFMT_RGB1:
689         case IMGFMT_BGR1:
690                 {
691                         const uint8_t * const d128=dither_8x8_220[y&7];
692                         uint8_t *g= c->table_gU[128] + c->table_gV[128];
693                         int acc=0;
694                         for(i=0; i<dstW-1; i+=2){
695                                 int j;
696                                 int Y1=1<<18;
697                                 int Y2=1<<18;
698
699                                 for(j=0; j<lumFilterSize; j++)
700                                 {
701                                         Y1 += lumSrc[j][i] * lumFilter[j];
702                                         Y2 += lumSrc[j][i+1] * lumFilter[j];
703                                 }
704                                 Y1>>=19;
705                                 Y2>>=19;
706                                 if((Y1|Y2)&256)
707                                 {
708                                         if(Y1>255)   Y1=255;
709                                         else if(Y1<0)Y1=0;
710                                         if(Y2>255)   Y2=255;
711                                         else if(Y2<0)Y2=0;
712                                 }
713                                 acc+= acc + g[Y1+d128[(i+0)&7]];
714                                 acc+= acc + g[Y2+d128[(i+1)&7]];
715                                 if((i&7)==6){
716                                         ((uint8_t*)dest)[0]= acc;
717                                         dest++;
718                                 }
719                         }
720                 }
721                 break;
722         case IMGFMT_YUY2:
723                 YSCALE_YUV_2_PACKEDX_C(void)
724                         ((uint8_t*)dest)[2*i2+0]= Y1;
725                         ((uint8_t*)dest)[2*i2+1]= U;
726                         ((uint8_t*)dest)[2*i2+2]= Y2;
727                         ((uint8_t*)dest)[2*i2+3]= V;
728                 }
729                 break;
730         case IMGFMT_UYVY:
731                 YSCALE_YUV_2_PACKEDX_C(void)
732                         ((uint8_t*)dest)[2*i2+0]= U;
733                         ((uint8_t*)dest)[2*i2+1]= Y1;
734                         ((uint8_t*)dest)[2*i2+2]= V;
735                         ((uint8_t*)dest)[2*i2+3]= Y2;
736                 }
737                 break;
738         }
739 }
740
741
742 //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
743 //Plain C versions
744 #if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
745 #define COMPILE_C
746 #endif
747
748 #ifdef ARCH_POWERPC
749 #if defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
750 #define COMPILE_ALTIVEC
751 #endif //HAVE_ALTIVEC
752 #endif //ARCH_POWERPC
753
754 #if defined(ARCH_X86) || defined(ARCH_X86_64)
755
756 #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
757 #define COMPILE_MMX
758 #endif
759
760 #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
761 #define COMPILE_MMX2
762 #endif
763
764 #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
765 #define COMPILE_3DNOW
766 #endif
767 #endif //ARCH_X86 || ARCH_X86_64
768
769 #undef HAVE_MMX
770 #undef HAVE_MMX2
771 #undef HAVE_3DNOW
772
773 #ifdef COMPILE_C
774 #undef HAVE_MMX
775 #undef HAVE_MMX2
776 #undef HAVE_3DNOW
777 #undef HAVE_ALTIVEC
778 #define RENAME(a) a ## _C
779 #include "swscale_template.c"
780 #endif
781
782 #ifdef ARCH_POWERPC
783 #ifdef COMPILE_ALTIVEC
784 #undef RENAME
785 #define HAVE_ALTIVEC
786 #define RENAME(a) a ## _altivec
787 #include "swscale_template.c"
788 #endif
789 #endif //ARCH_POWERPC
790
791 #if defined(ARCH_X86) || defined(ARCH_X86_64)
792
793 //X86 versions
794 /*
795 #undef RENAME
796 #undef HAVE_MMX
797 #undef HAVE_MMX2
798 #undef HAVE_3DNOW
799 #define ARCH_X86
800 #define RENAME(a) a ## _X86
801 #include "swscale_template.c"
802 */
803 //MMX versions
804 #ifdef COMPILE_MMX
805 #undef RENAME
806 #define HAVE_MMX
807 #undef HAVE_MMX2
808 #undef HAVE_3DNOW
809 #define RENAME(a) a ## _MMX
810 #include "swscale_template.c"
811 #endif
812
813 //MMX2 versions
814 #ifdef COMPILE_MMX2
815 #undef RENAME
816 #define HAVE_MMX
817 #define HAVE_MMX2
818 #undef HAVE_3DNOW
819 #define RENAME(a) a ## _MMX2
820 #include "swscale_template.c"
821 #endif
822
823 //3DNOW versions
824 #ifdef COMPILE_3DNOW
825 #undef RENAME
826 #define HAVE_MMX
827 #undef HAVE_MMX2
828 #define HAVE_3DNOW
829 #define RENAME(a) a ## _3DNow
830 #include "swscale_template.c"
831 #endif
832
833 #endif //ARCH_X86 || ARCH_X86_64
834
835 // minor note: the HAVE_xyz is messed up after that line so don't use it
836
837 static double getSplineCoeff(double a, double b, double c, double d, double dist)
838 {
839 //      printf("%f %f %f %f %f\n", a,b,c,d,dist);
840         if(dist<=1.0)   return ((d*dist + c)*dist + b)*dist +a;
841         else            return getSplineCoeff(  0.0, 
842                                                  b+ 2.0*c + 3.0*d,
843                                                         c + 3.0*d,
844                                                 -b- 3.0*c - 6.0*d,
845                                                 dist-1.0);
846 }
847
848 static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
849                               int srcW, int dstW, int filterAlign, int one, int flags,
850                               SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
851 {
852         int i;
853         int filterSize;
854         int filter2Size;
855         int minFilterSize;
856         double *filter=NULL;
857         double *filter2=NULL;
858 #if defined(ARCH_X86) || defined(ARCH_X86_64)
859         if(flags & SWS_CPU_CAPS_MMX)
860                 asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
861 #endif
862
863         // Note the +1 is for the MMXscaler which reads over the end
864         *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
865
866         if(ABS(xInc - 0x10000) <10) // unscaled
867         {
868                 int i;
869                 filterSize= 1;
870                 filter= av_malloc(dstW*sizeof(double)*filterSize);
871                 for(i=0; i<dstW*filterSize; i++) filter[i]=0;
872
873                 for(i=0; i<dstW; i++)
874                 {
875                         filter[i*filterSize]=1;
876                         (*filterPos)[i]=i;
877                 }
878
879         }
880         else if(flags&SWS_POINT) // lame looking point sampling mode
881         {
882                 int i;
883                 int xDstInSrc;
884                 filterSize= 1;
885                 filter= av_malloc(dstW*sizeof(double)*filterSize);
886                 
887                 xDstInSrc= xInc/2 - 0x8000;
888                 for(i=0; i<dstW; i++)
889                 {
890                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
891
892                         (*filterPos)[i]= xx;
893                         filter[i]= 1.0;
894                         xDstInSrc+= xInc;
895                 }
896         }
897         else if((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
898         {
899                 int i;
900                 int xDstInSrc;
901                 if     (flags&SWS_BICUBIC) filterSize= 4;
902                 else if(flags&SWS_X      ) filterSize= 4;
903                 else                       filterSize= 2; // SWS_BILINEAR / SWS_AREA 
904                 filter= av_malloc(dstW*sizeof(double)*filterSize);
905
906                 xDstInSrc= xInc/2 - 0x8000;
907                 for(i=0; i<dstW; i++)
908                 {
909                         int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
910                         int j;
911
912                         (*filterPos)[i]= xx;
913                                 //Bilinear upscale / linear interpolate / Area averaging
914                                 for(j=0; j<filterSize; j++)
915                                 {
916                                         double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
917                                         double coeff= 1.0 - d;
918                                         if(coeff<0) coeff=0;
919                                         filter[i*filterSize + j]= coeff;
920                                         xx++;
921                                 }
922                         xDstInSrc+= xInc;
923                 }
924         }
925         else
926         {
927                 double xDstInSrc;
928                 double sizeFactor, filterSizeInSrc;
929                 const double xInc1= (double)xInc / (double)(1<<16);
930
931                 if     (flags&SWS_BICUBIC)      sizeFactor= 4.0;
932                 else if(flags&SWS_X)            sizeFactor= 8.0;
933                 else if(flags&SWS_AREA)         sizeFactor= 1.0; //downscale only, for upscale it is bilinear
934                 else if(flags&SWS_GAUSS)        sizeFactor= 8.0;   // infinite ;)
935                 else if(flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
936                 else if(flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
937                 else if(flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
938                 else if(flags&SWS_BILINEAR)     sizeFactor= 2.0;
939                 else {
940                         sizeFactor= 0.0; //GCC warning killer
941                         ASSERT(0)
942                 }
943                 
944                 if(xInc1 <= 1.0)        filterSizeInSrc= sizeFactor; // upscale
945                 else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
946
947                 filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
948                 if(filterSize > srcW-2) filterSize=srcW-2;
949
950                 filter= av_malloc(dstW*sizeof(double)*filterSize);
951
952                 xDstInSrc= xInc1 / 2.0 - 0.5;
953                 for(i=0; i<dstW; i++)
954                 {
955                         int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
956                         int j;
957                         (*filterPos)[i]= xx;
958                         for(j=0; j<filterSize; j++)
959                         {
960                                 double d= ABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
961                                 double coeff;
962                                 if(flags & SWS_BICUBIC)
963                                 {
964                                         double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
965                                         double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
966
967                                         if(d<1.0) 
968                                                 coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
969                                         else if(d<2.0)
970                                                 coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
971                                         else
972                                                 coeff=0.0;
973                                 }
974 /*                              else if(flags & SWS_X)
975                                 {
976                                         double p= param ? param*0.01 : 0.3;
977                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
978                                         coeff*= pow(2.0, - p*d*d);
979                                 }*/
980                                 else if(flags & SWS_X)
981                                 {
982                                         double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
983                                         
984                                         if(d<1.0)
985                                                 coeff = cos(d*PI);
986                                         else
987                                                 coeff=-1.0;
988                                         if(coeff<0.0)   coeff= -pow(-coeff, A);
989                                         else            coeff=  pow( coeff, A);
990                                         coeff= coeff*0.5 + 0.5;
991                                 }
992                                 else if(flags & SWS_AREA)
993                                 {
994                                         double srcPixelSize= 1.0/xInc1;
995                                         if(d + srcPixelSize/2 < 0.5) coeff= 1.0;
996                                         else if(d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
997                                         else coeff=0.0;
998                                 }
999                                 else if(flags & SWS_GAUSS)
1000                                 {
1001                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1002                                         coeff = pow(2.0, - p*d*d);
1003                                 }
1004                                 else if(flags & SWS_SINC)
1005                                 {
1006                                         coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1007                                 }
1008                                 else if(flags & SWS_LANCZOS)
1009                                 {
1010                                         double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; 
1011                                         coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1012                                         if(d>p) coeff=0;
1013                                 }
1014                                 else if(flags & SWS_BILINEAR)
1015                                 {
1016                                         coeff= 1.0 - d;
1017                                         if(coeff<0) coeff=0;
1018                                 }
1019                                 else if(flags & SWS_SPLINE)
1020                                 {
1021                                         double p=-2.196152422706632;
1022                                         coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1023                                 }
1024                                 else {
1025                                         coeff= 0.0; //GCC warning killer
1026                                         ASSERT(0)
1027                                 }
1028
1029                                 filter[i*filterSize + j]= coeff;
1030                                 xx++;
1031                         }
1032                         xDstInSrc+= xInc1;
1033                 }
1034         }
1035
1036         /* apply src & dst Filter to filter -> filter2
1037            av_free(filter);
1038         */
1039         ASSERT(filterSize>0)
1040         filter2Size= filterSize;
1041         if(srcFilter) filter2Size+= srcFilter->length - 1;
1042         if(dstFilter) filter2Size+= dstFilter->length - 1;
1043         ASSERT(filter2Size>0)
1044         filter2= av_malloc(filter2Size*dstW*sizeof(double));
1045
1046         for(i=0; i<dstW; i++)
1047         {
1048                 int j;
1049                 SwsVector scaleFilter;
1050                 SwsVector *outVec;
1051
1052                 scaleFilter.coeff= filter + i*filterSize;
1053                 scaleFilter.length= filterSize;
1054
1055                 if(srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1056                 else          outVec= &scaleFilter;
1057
1058                 ASSERT(outVec->length == filter2Size)
1059                 //FIXME dstFilter
1060
1061                 for(j=0; j<outVec->length; j++)
1062                 {
1063                         filter2[i*filter2Size + j]= outVec->coeff[j];
1064                 }
1065
1066                 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1067
1068                 if(outVec != &scaleFilter) sws_freeVec(outVec);
1069         }
1070         av_free(filter); filter=NULL;
1071
1072         /* try to reduce the filter-size (step1 find size and shift left) */
1073         // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
1074         minFilterSize= 0;
1075         for(i=dstW-1; i>=0; i--)
1076         {
1077                 int min= filter2Size;
1078                 int j;
1079                 double cutOff=0.0;
1080
1081                 /* get rid off near zero elements on the left by shifting left */
1082                 for(j=0; j<filter2Size; j++)
1083                 {
1084                         int k;
1085                         cutOff += ABS(filter2[i*filter2Size]);
1086
1087                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1088
1089                         /* preserve Monotonicity because the core can't handle the filter otherwise */
1090                         if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1091
1092                         // Move filter coeffs left
1093                         for(k=1; k<filter2Size; k++)
1094                                 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1095                         filter2[i*filter2Size + k - 1]= 0.0;
1096                         (*filterPos)[i]++;
1097                 }
1098
1099                 cutOff=0.0;
1100                 /* count near zeros on the right */
1101                 for(j=filter2Size-1; j>0; j--)
1102                 {
1103                         cutOff += ABS(filter2[i*filter2Size + j]);
1104
1105                         if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1106                         min--;
1107                 }
1108
1109                 if(min>minFilterSize) minFilterSize= min;
1110         }
1111
1112         if (flags & SWS_CPU_CAPS_ALTIVEC) {
1113           // we can handle the special case 4,
1114           // so we don't want to go to the full 8
1115           if (minFilterSize < 5)
1116             filterAlign = 4;
1117
1118           // we really don't want to waste our time
1119           // doing useless computation, so fall-back on
1120           // the scalar C code for very small filter.
1121           // vectorizing is worth it only if you have
1122           // decent-sized vector.
1123           if (minFilterSize < 3)
1124             filterAlign = 1;
1125         }
1126
1127         if (flags & SWS_CPU_CAPS_MMX) {
1128                 // special case for unscaled vertical filtering
1129                 if(minFilterSize == 1 && filterAlign == 2)
1130                         filterAlign= 1;
1131         }
1132
1133         ASSERT(minFilterSize > 0)
1134         filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1135         ASSERT(filterSize > 0)
1136         filter= av_malloc(filterSize*dstW*sizeof(double));
1137         if(filterSize >= MAX_FILTER_SIZE)
1138                 return -1;
1139         *outFilterSize= filterSize;
1140
1141         if(flags&SWS_PRINT_INFO)
1142                 MSG_V("SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1143         /* try to reduce the filter-size (step2 reduce it) */
1144         for(i=0; i<dstW; i++)
1145         {
1146                 int j;
1147
1148                 for(j=0; j<filterSize; j++)
1149                 {
1150                         if(j>=filter2Size) filter[i*filterSize + j]= 0.0;
1151                         else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1152                 }
1153         }
1154         av_free(filter2); filter2=NULL;
1155         
1156
1157         //FIXME try to align filterpos if possible
1158
1159         //fix borders
1160         for(i=0; i<dstW; i++)
1161         {
1162                 int j;
1163                 if((*filterPos)[i] < 0)
1164                 {
1165                         // Move filter coeffs left to compensate for filterPos
1166                         for(j=1; j<filterSize; j++)
1167                         {
1168                                 int left= FFMAX(j + (*filterPos)[i], 0);
1169                                 filter[i*filterSize + left] += filter[i*filterSize + j];
1170                                 filter[i*filterSize + j]=0;
1171                         }
1172                         (*filterPos)[i]= 0;
1173                 }
1174
1175                 if((*filterPos)[i] + filterSize > srcW)
1176                 {
1177                         int shift= (*filterPos)[i] + filterSize - srcW;
1178                         // Move filter coeffs right to compensate for filterPos
1179                         for(j=filterSize-2; j>=0; j--)
1180                         {
1181                                 int right= FFMIN(j + shift, filterSize-1);
1182                                 filter[i*filterSize +right] += filter[i*filterSize +j];
1183                                 filter[i*filterSize +j]=0;
1184                         }
1185                         (*filterPos)[i]= srcW - filterSize;
1186                 }
1187         }
1188
1189         // Note the +1 is for the MMXscaler which reads over the end
1190         /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1191         *outFilter= av_malloc(*outFilterSize*(dstW+1)*sizeof(int16_t));
1192         memset(*outFilter, 0, *outFilterSize*(dstW+1)*sizeof(int16_t));
1193
1194         /* Normalize & Store in outFilter */
1195         for(i=0; i<dstW; i++)
1196         {
1197                 int j;
1198                 double error=0;
1199                 double sum=0;
1200                 double scale= one;
1201
1202                 for(j=0; j<filterSize; j++)
1203                 {
1204                         sum+= filter[i*filterSize + j];
1205                 }
1206                 scale/= sum;
1207                 for(j=0; j<*outFilterSize; j++)
1208                 {
1209                         double v= filter[i*filterSize + j]*scale + error;
1210                         int intV= floor(v + 0.5);
1211                         (*outFilter)[i*(*outFilterSize) + j]= intV;
1212                         error = v - intV;
1213                 }
1214         }
1215         
1216         (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1217         for(i=0; i<*outFilterSize; i++)
1218         {
1219                 int j= dstW*(*outFilterSize);
1220                 (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1221         }
1222
1223         av_free(filter);
1224         return 0;
1225 }
1226
1227 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1228 static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1229 {
1230         uint8_t *fragmentA;
1231         long imm8OfPShufW1A;
1232         long imm8OfPShufW2A;
1233         long fragmentLengthA;
1234         uint8_t *fragmentB;
1235         long imm8OfPShufW1B;
1236         long imm8OfPShufW2B;
1237         long fragmentLengthB;
1238         int fragmentPos;
1239
1240         int xpos, i;
1241
1242         // create an optimized horizontal scaling routine
1243
1244         //code fragment
1245
1246         asm volatile(
1247                 "jmp 9f                         \n\t"
1248         // Begin
1249                 "0:                             \n\t"
1250                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1251                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1252                 "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
1253                 "punpcklbw %%mm7, %%mm1         \n\t"
1254                 "punpcklbw %%mm7, %%mm0         \n\t"
1255                 "pshufw $0xFF, %%mm1, %%mm1     \n\t"
1256                 "1:                             \n\t"
1257                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1258                 "2:                             \n\t"
1259                 "psubw %%mm1, %%mm0             \n\t"
1260                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1261                 "pmullw %%mm3, %%mm0            \n\t"
1262                 "psllw $7, %%mm1                \n\t"
1263                 "paddw %%mm1, %%mm0             \n\t"
1264
1265                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1266
1267                 "add $8, %%"REG_a"              \n\t"
1268         // End
1269                 "9:                             \n\t"
1270 //              "int $3\n\t"
1271                 "lea 0b, %0                     \n\t"
1272                 "lea 1b, %1                     \n\t"
1273                 "lea 2b, %2                     \n\t"
1274                 "dec %1                         \n\t"
1275                 "dec %2                         \n\t"
1276                 "sub %0, %1                     \n\t"
1277                 "sub %0, %2                     \n\t"
1278                 "lea 9b, %3                     \n\t"
1279                 "sub %0, %3                     \n\t"
1280
1281
1282                 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1283                 "=r" (fragmentLengthA)
1284         );
1285
1286         asm volatile(
1287                 "jmp 9f                         \n\t"
1288         // Begin
1289                 "0:                             \n\t"
1290                 "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t" 
1291                 "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t" 
1292                 "punpcklbw %%mm7, %%mm0         \n\t"
1293                 "pshufw $0xFF, %%mm0, %%mm1     \n\t"
1294                 "1:                             \n\t"
1295                 "pshufw $0xFF, %%mm0, %%mm0     \n\t"
1296                 "2:                             \n\t"
1297                 "psubw %%mm1, %%mm0             \n\t"
1298                 "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
1299                 "pmullw %%mm3, %%mm0            \n\t"
1300                 "psllw $7, %%mm1                \n\t"
1301                 "paddw %%mm1, %%mm0             \n\t"
1302
1303                 "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
1304
1305                 "add $8, %%"REG_a"              \n\t"
1306         // End
1307                 "9:                             \n\t"
1308 //              "int $3\n\t"
1309                 "lea 0b, %0                     \n\t"
1310                 "lea 1b, %1                     \n\t"
1311                 "lea 2b, %2                     \n\t"
1312                 "dec %1                         \n\t"
1313                 "dec %2                         \n\t"
1314                 "sub %0, %1                     \n\t"
1315                 "sub %0, %2                     \n\t"
1316                 "lea 9b, %3                     \n\t"
1317                 "sub %0, %3                     \n\t"
1318
1319
1320                 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1321                 "=r" (fragmentLengthB)
1322         );
1323
1324         xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1325         fragmentPos=0;
1326         
1327         for(i=0; i<dstW/numSplits; i++)
1328         {
1329                 int xx=xpos>>16;
1330
1331                 if((i&3) == 0)
1332                 {
1333                         int a=0;
1334                         int b=((xpos+xInc)>>16) - xx;
1335                         int c=((xpos+xInc*2)>>16) - xx;
1336                         int d=((xpos+xInc*3)>>16) - xx;
1337
1338                         filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1339                         filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1340                         filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1341                         filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1342                         filterPos[i/2]= xx;
1343
1344                         if(d+1<4)
1345                         {
1346                                 int maxShift= 3-(d+1);
1347                                 int shift=0;
1348
1349                                 memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1350
1351                                 funnyCode[fragmentPos + imm8OfPShufW1B]=
1352                                         (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1353                                 funnyCode[fragmentPos + imm8OfPShufW2B]=
1354                                         a | (b<<2) | (c<<4) | (d<<6);
1355
1356                                 if(i+3>=dstW) shift=maxShift; //avoid overread
1357                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1358
1359                                 if(shift && i>=shift)
1360                                 {
1361                                         funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1362                                         funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1363                                         filterPos[i/2]-=shift;
1364                                 }
1365
1366                                 fragmentPos+= fragmentLengthB;
1367                         }
1368                         else
1369                         {
1370                                 int maxShift= 3-d;
1371                                 int shift=0;
1372
1373                                 memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1374
1375                                 funnyCode[fragmentPos + imm8OfPShufW1A]=
1376                                 funnyCode[fragmentPos + imm8OfPShufW2A]=
1377                                         a | (b<<2) | (c<<4) | (d<<6);
1378
1379                                 if(i+4>=dstW) shift=maxShift; //avoid overread
1380                                 else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1381
1382                                 if(shift && i>=shift)
1383                                 {
1384                                         funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1385                                         funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1386                                         filterPos[i/2]-=shift;
1387                                 }
1388
1389                                 fragmentPos+= fragmentLengthA;
1390                         }
1391
1392                         funnyCode[fragmentPos]= RET;
1393                 }
1394                 xpos+=xInc;
1395         }
1396         filterPos[i/2]= xpos>>16; // needed to jump to the next part
1397 }
1398 #endif // ARCH_X86 || ARCH_X86_64
1399
1400 static void globalInit(void){
1401     // generating tables:
1402     int i;
1403     for(i=0; i<768; i++){
1404         int c= FFMIN(FFMAX(i-256, 0), 255);
1405         clip_table[i]=c;
1406     }
1407 }
1408
1409 static SwsFunc getSwsFunc(int flags){
1410     
1411 #ifdef RUNTIME_CPUDETECT
1412 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1413         // ordered per speed fasterst first
1414         if(flags & SWS_CPU_CAPS_MMX2)
1415                 return swScale_MMX2;
1416         else if(flags & SWS_CPU_CAPS_3DNOW)
1417                 return swScale_3DNow;
1418         else if(flags & SWS_CPU_CAPS_MMX)
1419                 return swScale_MMX;
1420         else
1421                 return swScale_C;
1422
1423 #else
1424 #ifdef ARCH_POWERPC
1425         if(flags & SWS_CPU_CAPS_ALTIVEC)
1426           return swScale_altivec;
1427         else
1428           return swScale_C;
1429 #endif
1430         return swScale_C;
1431 #endif
1432 #else //RUNTIME_CPUDETECT
1433 #ifdef HAVE_MMX2
1434         return swScale_MMX2;
1435 #elif defined (HAVE_3DNOW)
1436         return swScale_3DNow;
1437 #elif defined (HAVE_MMX)
1438         return swScale_MMX;
1439 #elif defined (HAVE_ALTIVEC)
1440         return swScale_altivec;
1441 #else
1442         return swScale_C;
1443 #endif
1444 #endif //!RUNTIME_CPUDETECT
1445 }
1446
1447 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1448              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1449         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1450         /* Copy Y plane */
1451         if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1452                 memcpy(dst, src[0], srcSliceH*dstStride[0]);
1453         else
1454         {
1455                 int i;
1456                 uint8_t *srcPtr= src[0];
1457                 uint8_t *dstPtr= dst;
1458                 for(i=0; i<srcSliceH; i++)
1459                 {
1460                         memcpy(dstPtr, srcPtr, c->srcW);
1461                         srcPtr+= srcStride[0];
1462                         dstPtr+= dstStride[0];
1463                 }
1464         }
1465         dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1466         if (c->dstFormat == IMGFMT_NV12)
1467                 interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
1468         else
1469                 interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
1470
1471         return srcSliceH;
1472 }
1473
1474 static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1475              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1476         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1477
1478         yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1479
1480         return srcSliceH;
1481 }
1482
1483 static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1484              int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1485         uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1486
1487         yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );
1488
1489         return srcSliceH;
1490 }
1491
1492 /* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1493 static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1494                            int srcSliceH, uint8_t* dst[], int dstStride[]){
1495         const int srcFormat= c->srcFormat;
1496         const int dstFormat= c->dstFormat;
1497         const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
1498         const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
1499         const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 
1500         const int dstId= (dstFormat&0xFF)>>2;
1501         void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1502
1503         /* BGR -> BGR */
1504         if(   (isBGR(srcFormat) && isBGR(dstFormat))
1505            || (isRGB(srcFormat) && isRGB(dstFormat))){
1506                 switch(srcId | (dstId<<4)){
1507                 case 0x34: conv= rgb16to15; break;
1508                 case 0x36: conv= rgb24to15; break;
1509                 case 0x38: conv= rgb32to15; break;
1510                 case 0x43: conv= rgb15to16; break;
1511                 case 0x46: conv= rgb24to16; break;
1512                 case 0x48: conv= rgb32to16; break;
1513                 case 0x63: conv= rgb15to24; break;
1514                 case 0x64: conv= rgb16to24; break;
1515                 case 0x68: conv= rgb32to24; break;
1516                 case 0x83: conv= rgb15to32; break;
1517                 case 0x84: conv= rgb16to32; break;
1518                 case 0x86: conv= rgb24to32; break;
1519                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1520                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1521                 }
1522         }else if(   (isBGR(srcFormat) && isRGB(dstFormat))
1523                  || (isRGB(srcFormat) && isBGR(dstFormat))){
1524                 switch(srcId | (dstId<<4)){
1525                 case 0x33: conv= rgb15tobgr15; break;
1526                 case 0x34: conv= rgb16tobgr15; break;
1527                 case 0x36: conv= rgb24tobgr15; break;
1528                 case 0x38: conv= rgb32tobgr15; break;
1529                 case 0x43: conv= rgb15tobgr16; break;
1530                 case 0x44: conv= rgb16tobgr16; break;
1531                 case 0x46: conv= rgb24tobgr16; break;
1532                 case 0x48: conv= rgb32tobgr16; break;
1533                 case 0x63: conv= rgb15tobgr24; break;
1534                 case 0x64: conv= rgb16tobgr24; break;
1535                 case 0x66: conv= rgb24tobgr24; break;
1536                 case 0x68: conv= rgb32tobgr24; break;
1537                 case 0x83: conv= rgb15tobgr32; break;
1538                 case 0x84: conv= rgb16tobgr32; break;
1539                 case 0x86: conv= rgb24tobgr32; break;
1540                 case 0x88: conv= rgb32tobgr32; break;
1541                 default: MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1542                                  sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1543                 }
1544         }else{
1545                 MSG_ERR("swScaler: internal error %s -> %s converter\n", 
1546                          sws_format_name(srcFormat), sws_format_name(dstFormat));
1547         }
1548
1549         if(dstStride[0]*srcBpp == srcStride[0]*dstBpp)
1550                 conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1551         else
1552         {
1553                 int i;
1554                 uint8_t *srcPtr= src[0];
1555                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1556
1557                 for(i=0; i<srcSliceH; i++)
1558                 {
1559                         conv(srcPtr, dstPtr, c->srcW*srcBpp);
1560                         srcPtr+= srcStride[0];
1561                         dstPtr+= dstStride[0];
1562                 }
1563         }     
1564         return srcSliceH;
1565 }
1566
1567 static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1568              int srcSliceH, uint8_t* dst[], int dstStride[]){
1569
1570         rgb24toyv12(
1571                 src[0], 
1572                 dst[0]+ srcSliceY    *dstStride[0], 
1573                 dst[1]+(srcSliceY>>1)*dstStride[1], 
1574                 dst[2]+(srcSliceY>>1)*dstStride[2],
1575                 c->srcW, srcSliceH, 
1576                 dstStride[0], dstStride[1], srcStride[0]);
1577         return srcSliceH;
1578 }
1579
1580 static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1581              int srcSliceH, uint8_t* dst[], int dstStride[]){
1582         int i;
1583
1584         /* copy Y */
1585         if(srcStride[0]==dstStride[0] && srcStride[0] > 0) 
1586                 memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1587         else{
1588                 uint8_t *srcPtr= src[0];
1589                 uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1590
1591                 for(i=0; i<srcSliceH; i++)
1592                 {
1593                         memcpy(dstPtr, srcPtr, c->srcW);
1594                         srcPtr+= srcStride[0];
1595                         dstPtr+= dstStride[0];
1596                 }
1597         }
1598
1599         if(c->dstFormat==IMGFMT_YV12){
1600                 planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1601                 planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1602         }else{
1603                 planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1604                 planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1605         }
1606         return srcSliceH;
1607 }
1608
1609 /**
1610  * bring pointers in YUV order instead of YVU
1611  */
1612 static inline void sws_orderYUV(int format, uint8_t * sortedP[], int sortedStride[], uint8_t * p[], int stride[]){
1613         if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
1614            || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
1615                 sortedP[0]= p[0];
1616                 sortedP[1]= p[2];
1617                 sortedP[2]= p[1];
1618                 sortedStride[0]= stride[0];
1619                 sortedStride[1]= stride[2];
1620                 sortedStride[2]= stride[1];
1621         }
1622         else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
1623         {
1624                 sortedP[0]= p[0];
1625                 sortedP[1]= 
1626                 sortedP[2]= NULL;
1627                 sortedStride[0]= stride[0];
1628                 sortedStride[1]= 
1629                 sortedStride[2]= 0;
1630         }
1631         else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
1632         {
1633                 sortedP[0]= p[0];
1634                 sortedP[1]= p[1];
1635                 sortedP[2]= p[2];
1636                 sortedStride[0]= stride[0];
1637                 sortedStride[1]= stride[1];
1638                 sortedStride[2]= stride[2];
1639         }
1640         else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
1641         {
1642                 sortedP[0]= p[0];
1643                 sortedP[1]= p[1];
1644                 sortedP[2]= NULL;
1645                 sortedStride[0]= stride[0];
1646                 sortedStride[1]= stride[1];
1647                 sortedStride[2]= 0;
1648         }else{
1649                 MSG_ERR("internal error in orderYUV\n");
1650         }
1651 }
1652
1653 /* unscaled copy like stuff (assumes nearly identical formats) */
1654 static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1655              int srcSliceH, uint8_t* dst[], int dstStride[]){
1656
1657         if(isPacked(c->srcFormat))
1658         {
1659                 if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
1660                         memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1661                 else
1662                 {
1663                         int i;
1664                         uint8_t *srcPtr= src[0];
1665                         uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1666                         int length=0;
1667
1668                         /* universal length finder */
1669                         while(length+c->srcW <= ABS(dstStride[0]) 
1670                            && length+c->srcW <= ABS(srcStride[0])) length+= c->srcW;
1671                         ASSERT(length!=0);
1672
1673                         for(i=0; i<srcSliceH; i++)
1674                         {
1675                                 memcpy(dstPtr, srcPtr, length);
1676                                 srcPtr+= srcStride[0];
1677                                 dstPtr+= dstStride[0];
1678                         }
1679                 }
1680         }
1681         else 
1682         { /* Planar YUV or gray */
1683                 int plane;
1684                 for(plane=0; plane<3; plane++)
1685                 {
1686                         int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1687                         int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1688                         int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1689
1690                         if((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1691                         {
1692                                 if(!isGray(c->dstFormat))
1693                                         memset(dst[plane], 128, dstStride[plane]*height);
1694                         }
1695                         else
1696                         {
1697                                 if(dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1698                                         memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1699                                 else
1700                                 {
1701                                         int i;
1702                                         uint8_t *srcPtr= src[plane];
1703                                         uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1704                                         for(i=0; i<height; i++)
1705                                         {
1706                                                 memcpy(dstPtr, srcPtr, length);
1707                                                 srcPtr+= srcStride[plane];
1708                                                 dstPtr+= dstStride[plane];
1709                                         }
1710                                 }
1711                         }
1712                 }
1713         }
1714         return srcSliceH;
1715 }
1716
1717 static int remove_dup_fourcc(int fourcc)
1718 {
1719         switch(fourcc)
1720         {
1721             case IMGFMT_I420:
1722             case IMGFMT_IYUV: return IMGFMT_YV12;
1723             case IMGFMT_Y8  : return IMGFMT_Y800;
1724             case IMGFMT_IF09: return IMGFMT_YVU9;
1725             default: return fourcc;
1726         }
1727 }
1728
1729 static void getSubSampleFactors(int *h, int *v, int format){
1730         switch(format){
1731         case IMGFMT_UYVY:
1732         case IMGFMT_YUY2:
1733                 *h=1;
1734                 *v=0;
1735                 break;
1736         case IMGFMT_YV12:
1737         case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
1738         case IMGFMT_NV12:
1739         case IMGFMT_NV21:
1740                 *h=1;
1741                 *v=1;
1742                 break;
1743         case IMGFMT_YVU9:
1744                 *h=2;
1745                 *v=2;
1746                 break;
1747         case IMGFMT_444P:
1748                 *h=0;
1749                 *v=0;
1750                 break;
1751         case IMGFMT_422P:
1752                 *h=1;
1753                 *v=0;
1754                 break;
1755         case IMGFMT_411P:
1756                 *h=2;
1757                 *v=0;
1758                 break;
1759         default:
1760                 *h=0;
1761                 *v=0;
1762                 break;
1763         }
1764 }
1765
1766 static uint16_t roundToInt16(int64_t f){
1767         int r= (f + (1<<15))>>16;
1768              if(r<-0x7FFF) return 0x8000;
1769         else if(r> 0x7FFF) return 0x7FFF;
1770         else               return r;
1771 }
1772
1773 /**
1774  * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1775  * @param fullRange if 1 then the luma range is 0..255 if 0 its 16..235
1776  * @return -1 if not supported
1777  */
1778 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1779         int64_t crv =  inv_table[0];
1780         int64_t cbu =  inv_table[1];
1781         int64_t cgu = -inv_table[2];
1782         int64_t cgv = -inv_table[3];
1783         int64_t cy  = 1<<16;
1784         int64_t oy  = 0;
1785
1786         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1787         memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1788         memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1789
1790         c->brightness= brightness;
1791         c->contrast  = contrast;
1792         c->saturation= saturation;
1793         c->srcRange  = srcRange;
1794         c->dstRange  = dstRange;
1795
1796         c->uOffset=   0x0400040004000400LL;
1797         c->vOffset=   0x0400040004000400LL;
1798
1799         if(!srcRange){
1800                 cy= (cy*255) / 219;
1801                 oy= 16<<16;
1802         }
1803
1804         cy = (cy *contrast             )>>16;
1805         crv= (crv*contrast * saturation)>>32;
1806         cbu= (cbu*contrast * saturation)>>32;
1807         cgu= (cgu*contrast * saturation)>>32;
1808         cgv= (cgv*contrast * saturation)>>32;
1809
1810         oy -= 256*brightness;
1811
1812         c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1813         c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1814         c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1815         c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1816         c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1817         c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1818
1819         yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1820         //FIXME factorize
1821
1822 #ifdef COMPILE_ALTIVEC
1823         if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1824             yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1825 #endif  
1826         return 0;
1827 }
1828
1829 /**
1830  * @return -1 if not supported
1831  */
1832 int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1833         if(isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1834
1835         *inv_table = c->srcColorspaceTable;
1836         *table     = c->dstColorspaceTable;
1837         *srcRange  = c->srcRange;
1838         *dstRange  = c->dstRange;
1839         *brightness= c->brightness;
1840         *contrast  = c->contrast;
1841         *saturation= c->saturation;
1842         
1843         return 0;       
1844 }
1845
1846 SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
1847                          SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
1848
1849         SwsContext *c;
1850         int i;
1851         int usesVFilter, usesHFilter;
1852         int unscaled, needsDither;
1853         int srcFormat, dstFormat;
1854         SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
1855 #if defined(ARCH_X86) || defined(ARCH_X86_64)
1856         if(flags & SWS_CPU_CAPS_MMX)
1857                 asm volatile("emms\n\t"::: "memory");
1858 #endif
1859
1860 #ifndef RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
1861         flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
1862 #ifdef HAVE_MMX2
1863         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
1864 #elif defined (HAVE_3DNOW)
1865         flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
1866 #elif defined (HAVE_MMX)
1867         flags |= SWS_CPU_CAPS_MMX;
1868 #elif defined (HAVE_ALTIVEC)
1869         flags |= SWS_CPU_CAPS_ALTIVEC;
1870 #endif
1871 #endif
1872         if(clip_table[512] != 255) globalInit();
1873         if(rgb15to16 == NULL) sws_rgb2rgb_init(flags);
1874
1875         /* avoid duplicate Formats, so we don't need to check to much */
1876         srcFormat = remove_dup_fourcc(origSrcFormat);
1877         dstFormat = remove_dup_fourcc(origDstFormat);
1878
1879         unscaled = (srcW == dstW && srcH == dstH);
1880         needsDither= (isBGR(dstFormat) || isRGB(dstFormat)) 
1881                      && (dstFormat&0xFF)<24
1882                      && ((dstFormat&0xFF)<(srcFormat&0xFF) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
1883
1884         if(!isSupportedIn(srcFormat)) 
1885         {
1886                 MSG_ERR("swScaler: %s is not supported as input format\n", sws_format_name(srcFormat));
1887                 return NULL;
1888         }
1889         if(!isSupportedOut(dstFormat))
1890         {
1891                 MSG_ERR("swScaler: %s is not supported as output format\n", sws_format_name(dstFormat));
1892                 return NULL;
1893         }
1894
1895         /* sanity check */
1896         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
1897         {
1898                  MSG_ERR("swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", 
1899                         srcW, srcH, dstW, dstH);
1900                 return NULL;
1901         }
1902
1903         if(!dstFilter) dstFilter= &dummyFilter;
1904         if(!srcFilter) srcFilter= &dummyFilter;
1905
1906         c= av_malloc(sizeof(SwsContext));
1907         memset(c, 0, sizeof(SwsContext));
1908
1909         c->srcW= srcW;
1910         c->srcH= srcH;
1911         c->dstW= dstW;
1912         c->dstH= dstH;
1913         c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
1914         c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
1915         c->flags= flags;
1916         c->dstFormat= dstFormat;
1917         c->srcFormat= srcFormat;
1918         c->origDstFormat= origDstFormat;
1919         c->origSrcFormat= origSrcFormat;
1920         c->vRounder= 4* 0x0001000100010001ULL;
1921
1922         usesHFilter= usesVFilter= 0;
1923         if(dstFilter->lumV!=NULL && dstFilter->lumV->length>1) usesVFilter=1;
1924         if(dstFilter->lumH!=NULL && dstFilter->lumH->length>1) usesHFilter=1;
1925         if(dstFilter->chrV!=NULL && dstFilter->chrV->length>1) usesVFilter=1;
1926         if(dstFilter->chrH!=NULL && dstFilter->chrH->length>1) usesHFilter=1;
1927         if(srcFilter->lumV!=NULL && srcFilter->lumV->length>1) usesVFilter=1;
1928         if(srcFilter->lumH!=NULL && srcFilter->lumH->length>1) usesHFilter=1;
1929         if(srcFilter->chrV!=NULL && srcFilter->chrV->length>1) usesVFilter=1;
1930         if(srcFilter->chrH!=NULL && srcFilter->chrH->length>1) usesHFilter=1;
1931
1932         getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
1933         getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
1934
1935         // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
1936         if((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
1937
1938         // drop some chroma lines if the user wants it
1939         c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
1940         c->chrSrcVSubSample+= c->vChrDrop;
1941
1942         // drop every 2. pixel for chroma calculation unless user wants full chroma
1943         if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)) 
1944                 c->chrSrcHSubSample=1;
1945
1946         if(param){
1947                 c->param[0] = param[0];
1948                 c->param[1] = param[1];
1949         }else{
1950                 c->param[0] =
1951                 c->param[1] = SWS_PARAM_DEFAULT;
1952         }
1953
1954         c->chrIntHSubSample= c->chrDstHSubSample;
1955         c->chrIntVSubSample= c->chrSrcVSubSample;
1956
1957         // note the -((-x)>>y) is so that we allways round toward +inf
1958         c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
1959         c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
1960         c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
1961         c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
1962
1963         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); 
1964
1965         /* unscaled special Cases */
1966         if(unscaled && !usesHFilter && !usesVFilter)
1967         {
1968                 /* yv12_to_nv12 */
1969                 if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
1970                 {
1971                         c->swScale= PlanarToNV12Wrapper;
1972                 }
1973                 /* yuv2bgr */
1974                 if((srcFormat==IMGFMT_YV12 || srcFormat==IMGFMT_422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
1975                 {
1976                         c->swScale= yuv2rgb_get_func_ptr(c);
1977                 }
1978                 
1979                 if( srcFormat==IMGFMT_YVU9 && dstFormat==IMGFMT_YV12 )
1980                 {
1981                         c->swScale= yvu9toyv12Wrapper;
1982                 }
1983
1984                 /* bgr24toYV12 */
1985                 if(srcFormat==IMGFMT_BGR24 && dstFormat==IMGFMT_YV12)
1986                         c->swScale= bgr24toyv12Wrapper;
1987                 
1988                 /* rgb/bgr -> rgb/bgr (no dither needed forms) */
1989                 if(   (isBGR(srcFormat) || isRGB(srcFormat))
1990                    && (isBGR(dstFormat) || isRGB(dstFormat)) 
1991                    && !needsDither)
1992                         c->swScale= rgb2rgbWrapper;
1993
1994                 /* LQ converters if -sws 0 or -sws 4*/
1995                 if(c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
1996                         /* rgb/bgr -> rgb/bgr (dither needed forms) */
1997                         if(  (isBGR(srcFormat) || isRGB(srcFormat))
1998                           && (isBGR(dstFormat) || isRGB(dstFormat)) 
1999                           && needsDither)
2000                                 c->swScale= rgb2rgbWrapper;
2001
2002                         /* yv12_to_yuy2 */
2003                         if(srcFormat == IMGFMT_YV12 && 
2004                             (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY))
2005                         {
2006                                 if (dstFormat == IMGFMT_YUY2)
2007                                     c->swScale= PlanarToYuy2Wrapper;
2008                                 else
2009                                     c->swScale= PlanarToUyvyWrapper;
2010                         }
2011                 }
2012
2013 #ifdef COMPILE_ALTIVEC
2014                 if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2015                     ((srcFormat == IMGFMT_YV12 && 
2016                       (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
2017                   // unscaled YV12 -> packed YUV, we want speed
2018                   if (dstFormat == IMGFMT_YUY2)
2019                     c->swScale= yv12toyuy2_unscaled_altivec;
2020                   else
2021                     c->swScale= yv12touyvy_unscaled_altivec;
2022                 }
2023 #endif
2024
2025                 /* simple copy */
2026                 if(   srcFormat == dstFormat
2027                    || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2028                    || (isPlanarYUV(dstFormat) && isGray(srcFormat))
2029                   )
2030                 {
2031                         c->swScale= simpleCopy;
2032                 }
2033
2034                 if(c->swScale){
2035                         if(flags&SWS_PRINT_INFO)
2036                                 MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
2037                                         sws_format_name(srcFormat), sws_format_name(dstFormat));
2038                         return c;
2039                 }
2040         }
2041
2042         if(flags & SWS_CPU_CAPS_MMX2)
2043         {
2044                 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2045                 if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2046                 {
2047                         if(flags&SWS_PRINT_INFO)
2048                                 MSG_INFO("SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
2049                 }
2050                 if(usesHFilter) c->canMMX2BeUsed=0;
2051         }
2052         else
2053                 c->canMMX2BeUsed=0;
2054
2055         c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2056         c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2057
2058         // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2059         // but only for the FAST_BILINEAR mode otherwise do correct scaling
2060         // n-2 is the last chrominance sample available
2061         // this is not perfect, but noone shuld notice the difference, the more correct variant
2062         // would be like the vertical one, but that would require some special code for the
2063         // first and last pixel
2064         if(flags&SWS_FAST_BILINEAR)
2065         {
2066                 if(c->canMMX2BeUsed)
2067                 {
2068                         c->lumXInc+= 20;
2069                         c->chrXInc+= 20;
2070                 }
2071                 //we don't use the x86asm scaler if mmx is available
2072                 else if(flags & SWS_CPU_CAPS_MMX)
2073                 {
2074                         c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2075                         c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2076                 }
2077         }
2078
2079         /* precalculate horizontal scaler filter coefficients */
2080         {
2081                 const int filterAlign=
2082                   (flags & SWS_CPU_CAPS_MMX) ? 4 :
2083                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2084                   1;
2085
2086                 initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2087                                  srcW      ,       dstW, filterAlign, 1<<14,
2088                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2089                                  srcFilter->lumH, dstFilter->lumH, c->param);
2090                 initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2091                                  c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2092                                  (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2093                                  srcFilter->chrH, dstFilter->chrH, c->param);
2094
2095 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2096 // can't downscale !!!
2097                 if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2098                 {
2099 #define MAX_FUNNY_CODE_SIZE 10000
2100 #ifdef MAP_ANONYMOUS
2101                         c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2102                         c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2103 #else
2104                         c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2105                         c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2106 #endif
2107
2108                         c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2109                         c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2110                         c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2111                         c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2112
2113                         initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2114                         initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2115                 }
2116 #endif
2117         } // Init Horizontal stuff
2118
2119
2120
2121         /* precalculate vertical scaler filter coefficients */
2122         {
2123                 const int filterAlign=
2124                   (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2125                   (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2126                   1;
2127
2128                 initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2129                                 srcH      ,        dstH, filterAlign, (1<<12)-4,
2130                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2131                                 srcFilter->lumV, dstFilter->lumV, c->param);
2132                 initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2133                                 c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2134                                 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2135                                 srcFilter->chrV, dstFilter->chrV, c->param);
2136
2137 #ifdef HAVE_ALTIVEC
2138                 c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2139                 c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2140
2141                 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2142                   int j;
2143                   short *p = (short *)&c->vYCoeffsBank[i];
2144                   for (j=0;j<8;j++)
2145                     p[j] = c->vLumFilter[i];
2146                 }
2147
2148                 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2149                   int j;
2150                   short *p = (short *)&c->vCCoeffsBank[i];
2151                   for (j=0;j<8;j++)
2152                     p[j] = c->vChrFilter[i];
2153                 }
2154 #endif
2155         }
2156
2157         // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2158         c->vLumBufSize= c->vLumFilterSize;
2159         c->vChrBufSize= c->vChrFilterSize;
2160         for(i=0; i<dstH; i++)
2161         {
2162                 int chrI= i*c->chrDstH / dstH;
2163                 int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2164                                  ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2165
2166                 nextSlice>>= c->chrSrcVSubSample;
2167                 nextSlice<<= c->chrSrcVSubSample;
2168                 if(c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2169                         c->vLumBufSize= nextSlice - c->vLumFilterPos[i   ];
2170                 if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2171                         c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2172         }
2173
2174         // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2175         c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2176         c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2177         //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)
2178         /* align at 16 bytes for AltiVec */
2179         for(i=0; i<c->vLumBufSize; i++)
2180                 c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_malloc(4000);
2181         for(i=0; i<c->vChrBufSize; i++)
2182                 c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc(8000);
2183
2184         //try to avoid drawing green stuff between the right end and the stride end
2185         for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
2186         for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
2187
2188         ASSERT(c->chrDstH <= dstH)
2189
2190         if(flags&SWS_PRINT_INFO)
2191         {
2192 #ifdef DITHER1XBPP
2193                 char *dither= " dithered";
2194 #else
2195                 char *dither= "";
2196 #endif
2197                 if(flags&SWS_FAST_BILINEAR)
2198                         MSG_INFO("\nSwScaler: FAST_BILINEAR scaler, ");
2199                 else if(flags&SWS_BILINEAR)
2200                         MSG_INFO("\nSwScaler: BILINEAR scaler, ");
2201                 else if(flags&SWS_BICUBIC)
2202                         MSG_INFO("\nSwScaler: BICUBIC scaler, ");
2203                 else if(flags&SWS_X)
2204                         MSG_INFO("\nSwScaler: Experimental scaler, ");
2205                 else if(flags&SWS_POINT)
2206                         MSG_INFO("\nSwScaler: Nearest Neighbor / POINT scaler, ");
2207                 else if(flags&SWS_AREA)
2208                         MSG_INFO("\nSwScaler: Area Averageing scaler, ");
2209                 else if(flags&SWS_BICUBLIN)
2210                         MSG_INFO("\nSwScaler: luma BICUBIC / chroma BILINEAR scaler, ");
2211                 else if(flags&SWS_GAUSS)
2212                         MSG_INFO("\nSwScaler: Gaussian scaler, ");
2213                 else if(flags&SWS_SINC)
2214                         MSG_INFO("\nSwScaler: Sinc scaler, ");
2215                 else if(flags&SWS_LANCZOS)
2216                         MSG_INFO("\nSwScaler: Lanczos scaler, ");
2217                 else if(flags&SWS_SPLINE)
2218                         MSG_INFO("\nSwScaler: Bicubic spline scaler, ");
2219                 else
2220                         MSG_INFO("\nSwScaler: ehh flags invalid?! ");
2221
2222                 if(dstFormat==IMGFMT_BGR15 || dstFormat==IMGFMT_BGR16)
2223                         MSG_INFO("from %s to%s %s ", 
2224                                 sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2225                 else
2226                         MSG_INFO("from %s to %s ", 
2227                                 sws_format_name(srcFormat), sws_format_name(dstFormat));
2228
2229                 if(flags & SWS_CPU_CAPS_MMX2)
2230                         MSG_INFO("using MMX2\n");
2231                 else if(flags & SWS_CPU_CAPS_3DNOW)
2232                         MSG_INFO("using 3DNOW\n");
2233                 else if(flags & SWS_CPU_CAPS_MMX)
2234                         MSG_INFO("using MMX\n");
2235                 else if(flags & SWS_CPU_CAPS_ALTIVEC)
2236                         MSG_INFO("using AltiVec\n");
2237                 else 
2238                         MSG_INFO("using C\n");
2239         }
2240
2241         if(flags & SWS_PRINT_INFO)
2242         {
2243                 if(flags & SWS_CPU_CAPS_MMX)
2244                 {
2245                         if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2246                                 MSG_V("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2247                         else
2248                         {
2249                                 if(c->hLumFilterSize==4)
2250                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
2251                                 else if(c->hLumFilterSize==8)
2252                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
2253                                 else
2254                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
2255
2256                                 if(c->hChrFilterSize==4)
2257                                         MSG_V("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
2258                                 else if(c->hChrFilterSize==8)
2259                                         MSG_V("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
2260                                 else
2261                                         MSG_V("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
2262                         }
2263                 }
2264                 else
2265                 {
2266 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2267                         MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
2268 #else
2269                         if(flags & SWS_FAST_BILINEAR)
2270                                 MSG_V("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
2271                         else
2272                                 MSG_V("SwScaler: using C scaler for horizontal scaling\n");
2273 #endif
2274                 }
2275                 if(isPlanarYUV(dstFormat))
2276                 {
2277                         if(c->vLumFilterSize==1)
2278                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2279                         else
2280                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2281                 }
2282                 else
2283                 {
2284                         if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
2285                                 MSG_V("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2286                                        "SwScaler:       2-tap scaler for vertical chrominance scaling (BGR)\n",(flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2287                         else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
2288                                 MSG_V("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2289                         else
2290                                 MSG_V("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2291                 }
2292
2293                 if(dstFormat==IMGFMT_BGR24)
2294                         MSG_V("SwScaler: using %s YV12->BGR24 Converter\n",
2295                                 (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2296                 else if(dstFormat==IMGFMT_BGR32)
2297                         MSG_V("SwScaler: using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2298                 else if(dstFormat==IMGFMT_BGR16)
2299                         MSG_V("SwScaler: using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2300                 else if(dstFormat==IMGFMT_BGR15)
2301                         MSG_V("SwScaler: using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2302
2303                 MSG_V("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2304         }
2305         if(flags & SWS_PRINT_INFO)
2306         {
2307                 MSG_DBG2("SwScaler:Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2308                         c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2309                 MSG_DBG2("SwScaler:Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2310                         c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2311         }
2312
2313         c->swScale= getSwsFunc(flags);
2314         return c;
2315 }
2316
2317 /**
2318  * swscale warper, so we don't need to export the SwsContext.
2319  * assumes planar YUV to be in YUV order instead of YVU
2320  */
2321 int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2322                            int srcSliceH, uint8_t* dst[], int dstStride[]){
2323         if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2324             MSG_ERR("swScaler: slices start in the middle!\n");
2325             return 0;
2326         }
2327         if (c->sliceDir == 0) {
2328             if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2329         }
2330
2331         // copy strides, so they can safely be modified
2332         if (c->sliceDir == 1) {
2333             // slices go from top to bottom
2334             int srcStride2[3]= {srcStride[0], srcStride[1], srcStride[2]};
2335             int dstStride2[3]= {dstStride[0], dstStride[1], dstStride[2]};
2336             return c->swScale(c, src, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2337         } else {
2338             // slices go from bottom to top => we flip the image internally
2339             uint8_t* src2[3]= {src[0] + (srcSliceH-1)*srcStride[0],
2340                                src[1] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1],
2341                                src[2] + ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2]
2342             };
2343             uint8_t* dst2[3]= {dst[0] + (c->dstH-1)*dstStride[0],
2344                                dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2345                                dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2346             int srcStride2[3]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2347             int dstStride2[3]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2348             
2349             return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2350         }
2351 }
2352
2353 /**
2354  * swscale warper, so we don't need to export the SwsContext
2355  */
2356 int sws_scale(SwsContext *c, uint8_t* srcParam[], int srcStrideParam[], int srcSliceY,
2357                            int srcSliceH, uint8_t* dstParam[], int dstStrideParam[]){
2358         int srcStride[3];
2359         int dstStride[3];
2360         uint8_t *src[3];
2361         uint8_t *dst[3];
2362         sws_orderYUV(c->origSrcFormat, src, srcStride, srcParam, srcStrideParam);
2363         sws_orderYUV(c->origDstFormat, dst, dstStride, dstParam, dstStrideParam);
2364 //printf("sws: slice %d %d\n", srcSliceY, srcSliceH);
2365
2366         return c->swScale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2367 }
2368
2369 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, 
2370                                 float lumaSharpen, float chromaSharpen,
2371                                 float chromaHShift, float chromaVShift,
2372                                 int verbose)
2373 {
2374         SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2375
2376         if(lumaGBlur!=0.0){
2377                 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2378                 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2379         }else{
2380                 filter->lumH= sws_getIdentityVec();
2381                 filter->lumV= sws_getIdentityVec();
2382         }
2383
2384         if(chromaGBlur!=0.0){
2385                 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2386                 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2387         }else{
2388                 filter->chrH= sws_getIdentityVec();
2389                 filter->chrV= sws_getIdentityVec();
2390         }
2391
2392         if(chromaSharpen!=0.0){
2393                 SwsVector *id= sws_getIdentityVec();
2394                 sws_scaleVec(filter->chrH, -chromaSharpen);
2395                 sws_scaleVec(filter->chrV, -chromaSharpen);
2396                 sws_addVec(filter->chrH, id);
2397                 sws_addVec(filter->chrV, id);
2398                 sws_freeVec(id);
2399         }
2400
2401         if(lumaSharpen!=0.0){
2402                 SwsVector *id= sws_getIdentityVec();
2403                 sws_scaleVec(filter->lumH, -lumaSharpen);
2404                 sws_scaleVec(filter->lumV, -lumaSharpen);
2405                 sws_addVec(filter->lumH, id);
2406                 sws_addVec(filter->lumV, id);
2407                 sws_freeVec(id);
2408         }
2409
2410         if(chromaHShift != 0.0)
2411                 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2412
2413         if(chromaVShift != 0.0)
2414                 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2415
2416         sws_normalizeVec(filter->chrH, 1.0);
2417         sws_normalizeVec(filter->chrV, 1.0);
2418         sws_normalizeVec(filter->lumH, 1.0);
2419         sws_normalizeVec(filter->lumV, 1.0);
2420
2421         if(verbose) sws_printVec(filter->chrH);
2422         if(verbose) sws_printVec(filter->lumH);
2423
2424         return filter;
2425 }
2426
2427 /**
2428  * returns a normalized gaussian curve used to filter stuff
2429  * quality=3 is high quality, lowwer is lowwer quality
2430  */
2431 SwsVector *sws_getGaussianVec(double variance, double quality){
2432         const int length= (int)(variance*quality + 0.5) | 1;
2433         int i;
2434         double *coeff= av_malloc(length*sizeof(double));
2435         double middle= (length-1)*0.5;
2436         SwsVector *vec= av_malloc(sizeof(SwsVector));
2437
2438         vec->coeff= coeff;
2439         vec->length= length;
2440
2441         for(i=0; i<length; i++)
2442         {
2443                 double dist= i-middle;
2444                 coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
2445         }
2446
2447         sws_normalizeVec(vec, 1.0);
2448
2449         return vec;
2450 }
2451
2452 SwsVector *sws_getConstVec(double c, int length){
2453         int i;
2454         double *coeff= av_malloc(length*sizeof(double));
2455         SwsVector *vec= av_malloc(sizeof(SwsVector));
2456
2457         vec->coeff= coeff;
2458         vec->length= length;
2459
2460         for(i=0; i<length; i++)
2461                 coeff[i]= c;
2462
2463         return vec;
2464 }
2465
2466
2467 SwsVector *sws_getIdentityVec(void){
2468         return sws_getConstVec(1.0, 1);
2469 }
2470
2471 double sws_dcVec(SwsVector *a){
2472         int i;
2473         double sum=0;
2474
2475         for(i=0; i<a->length; i++)
2476                 sum+= a->coeff[i];
2477
2478         return sum;
2479 }
2480
2481 void sws_scaleVec(SwsVector *a, double scalar){
2482         int i;
2483
2484         for(i=0; i<a->length; i++)
2485                 a->coeff[i]*= scalar;
2486 }
2487
2488 void sws_normalizeVec(SwsVector *a, double height){
2489         sws_scaleVec(a, height/sws_dcVec(a));
2490 }
2491
2492 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2493         int length= a->length + b->length - 1;
2494         double *coeff= av_malloc(length*sizeof(double));
2495         int i, j;
2496         SwsVector *vec= av_malloc(sizeof(SwsVector));
2497
2498         vec->coeff= coeff;
2499         vec->length= length;
2500
2501         for(i=0; i<length; i++) coeff[i]= 0.0;
2502
2503         for(i=0; i<a->length; i++)
2504         {
2505                 for(j=0; j<b->length; j++)
2506                 {
2507                         coeff[i+j]+= a->coeff[i]*b->coeff[j];
2508                 }
2509         }
2510
2511         return vec;
2512 }
2513
2514 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2515         int length= FFMAX(a->length, b->length);
2516         double *coeff= av_malloc(length*sizeof(double));
2517         int i;
2518         SwsVector *vec= av_malloc(sizeof(SwsVector));
2519
2520         vec->coeff= coeff;
2521         vec->length= length;
2522
2523         for(i=0; i<length; i++) coeff[i]= 0.0;
2524
2525         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2526         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2527
2528         return vec;
2529 }
2530
2531 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2532         int length= FFMAX(a->length, b->length);
2533         double *coeff= av_malloc(length*sizeof(double));
2534         int i;
2535         SwsVector *vec= av_malloc(sizeof(SwsVector));
2536
2537         vec->coeff= coeff;
2538         vec->length= length;
2539
2540         for(i=0; i<length; i++) coeff[i]= 0.0;
2541
2542         for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2543         for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2544
2545         return vec;
2546 }
2547
2548 /* shift left / or right if "shift" is negative */
2549 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2550         int length= a->length + ABS(shift)*2;
2551         double *coeff= av_malloc(length*sizeof(double));
2552         int i;
2553         SwsVector *vec= av_malloc(sizeof(SwsVector));
2554
2555         vec->coeff= coeff;
2556         vec->length= length;
2557
2558         for(i=0; i<length; i++) coeff[i]= 0.0;
2559
2560         for(i=0; i<a->length; i++)
2561         {
2562                 coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2563         }
2564
2565         return vec;
2566 }
2567
2568 void sws_shiftVec(SwsVector *a, int shift){
2569         SwsVector *shifted= sws_getShiftedVec(a, shift);
2570         av_free(a->coeff);
2571         a->coeff= shifted->coeff;
2572         a->length= shifted->length;
2573         av_free(shifted);
2574 }
2575
2576 void sws_addVec(SwsVector *a, SwsVector *b){
2577         SwsVector *sum= sws_sumVec(a, b);
2578         av_free(a->coeff);
2579         a->coeff= sum->coeff;
2580         a->length= sum->length;
2581         av_free(sum);
2582 }
2583
2584 void sws_subVec(SwsVector *a, SwsVector *b){
2585         SwsVector *diff= sws_diffVec(a, b);
2586         av_free(a->coeff);
2587         a->coeff= diff->coeff;
2588         a->length= diff->length;
2589         av_free(diff);
2590 }
2591
2592 void sws_convVec(SwsVector *a, SwsVector *b){
2593         SwsVector *conv= sws_getConvVec(a, b);
2594         av_free(a->coeff);  
2595         a->coeff= conv->coeff;
2596         a->length= conv->length;
2597         av_free(conv);
2598 }
2599
2600 SwsVector *sws_cloneVec(SwsVector *a){
2601         double *coeff= av_malloc(a->length*sizeof(double));
2602         int i;
2603         SwsVector *vec= av_malloc(sizeof(SwsVector));
2604
2605         vec->coeff= coeff;
2606         vec->length= a->length;
2607
2608         for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2609
2610         return vec;
2611 }
2612
2613 void sws_printVec(SwsVector *a){
2614         int i;
2615         double max=0;
2616         double min=0;
2617         double range;
2618
2619         for(i=0; i<a->length; i++)
2620                 if(a->coeff[i]>max) max= a->coeff[i];
2621
2622         for(i=0; i<a->length; i++)
2623                 if(a->coeff[i]<min) min= a->coeff[i];
2624
2625         range= max - min;
2626
2627         for(i=0; i<a->length; i++)
2628         {
2629                 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2630                 MSG_DBG2("%1.3f ", a->coeff[i]);
2631                 for(;x>0; x--) MSG_DBG2(" ");
2632                 MSG_DBG2("|\n");
2633         }
2634 }
2635
2636 void sws_freeVec(SwsVector *a){
2637         if(!a) return;
2638         av_free(a->coeff);
2639         a->coeff=NULL;
2640         a->length=0;
2641         av_free(a);
2642 }
2643
2644 void sws_freeFilter(SwsFilter *filter){
2645         if(!filter) return;
2646
2647         if(filter->lumH) sws_freeVec(filter->lumH);
2648         if(filter->lumV) sws_freeVec(filter->lumV);
2649         if(filter->chrH) sws_freeVec(filter->chrH);
2650         if(filter->chrV) sws_freeVec(filter->chrV);
2651         av_free(filter);
2652 }
2653
2654
2655 void sws_freeContext(SwsContext *c){
2656         int i;
2657         if(!c) return;
2658
2659         if(c->lumPixBuf)
2660         {
2661                 for(i=0; i<c->vLumBufSize; i++)
2662                 {
2663                         av_free(c->lumPixBuf[i]);
2664                         c->lumPixBuf[i]=NULL;
2665                 }
2666                 av_free(c->lumPixBuf);
2667                 c->lumPixBuf=NULL;
2668         }
2669
2670         if(c->chrPixBuf)
2671         {
2672                 for(i=0; i<c->vChrBufSize; i++)
2673                 {
2674                         av_free(c->chrPixBuf[i]);
2675                         c->chrPixBuf[i]=NULL;
2676                 }
2677                 av_free(c->chrPixBuf);
2678                 c->chrPixBuf=NULL;
2679         }
2680
2681         av_free(c->vLumFilter);
2682         c->vLumFilter = NULL;
2683         av_free(c->vChrFilter);
2684         c->vChrFilter = NULL;
2685         av_free(c->hLumFilter);
2686         c->hLumFilter = NULL;
2687         av_free(c->hChrFilter);
2688         c->hChrFilter = NULL;
2689 #ifdef HAVE_ALTIVEC
2690         av_free(c->vYCoeffsBank);
2691         c->vYCoeffsBank = NULL;
2692         av_free(c->vCCoeffsBank);
2693         c->vCCoeffsBank = NULL;
2694 #endif
2695
2696         av_free(c->vLumFilterPos);
2697         c->vLumFilterPos = NULL;
2698         av_free(c->vChrFilterPos);
2699         c->vChrFilterPos = NULL;
2700         av_free(c->hLumFilterPos);
2701         c->hLumFilterPos = NULL;
2702         av_free(c->hChrFilterPos);
2703         c->hChrFilterPos = NULL;
2704
2705 #if defined(ARCH_X86) || defined(ARCH_X86_64)
2706 #ifdef MAP_ANONYMOUS
2707         if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2708         if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2709 #else
2710         av_free(c->funnyYCode);
2711         av_free(c->funnyUVCode);
2712 #endif
2713         c->funnyYCode=NULL;
2714         c->funnyUVCode=NULL;
2715 #endif
2716
2717         av_free(c->lumMmx2Filter);
2718         c->lumMmx2Filter=NULL;
2719         av_free(c->chrMmx2Filter);
2720         c->chrMmx2Filter=NULL;
2721         av_free(c->lumMmx2FilterPos);
2722         c->lumMmx2FilterPos=NULL;
2723         av_free(c->chrMmx2FilterPos);
2724         c->chrMmx2FilterPos=NULL;
2725         av_free(c->yuvTable);
2726         c->yuvTable=NULL;
2727
2728         av_free(c);
2729 }
2730