Statistics
| Branch: | Revision:

ffmpeg / libswscale / swscale.c @ 86593486

History | View | Annotate | Download (101 KB)

1
/*
2
 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or modify
7
 * it under the terms of the GNU General Public License as published by
8
 * the Free Software Foundation; either version 2 of the License, or
9
 * (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14
 * GNU General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU General Public License
17
 * along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 *
20
 * the C code (not assembly, mmx, ...) of this file can be used
21
 * under the LGPL license too
22
 */
23

    
24
/*
25
  supported Input formats: YV12, I420/IYUV, YUY2, UYVY, BGR32, BGR24, BGR16, BGR15, RGB32, RGB24, Y8/Y800, YVU9/IF09, PAL8
26
  supported output formats: YV12, I420/IYUV, YUY2, UYVY, {BGR,RGB}{1,4,8,15,16,24,32}, Y8/Y800, YVU9/IF09
27
  {BGR,RGB}{1,4,8,15,16} support dithering
28

29
  unscaled special converters (YV12=I420=IYUV, Y800=Y8)
30
  YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
31
  x -> x
32
  YUV9 -> YV12
33
  YUV9/YV12 -> Y800
34
  Y800 -> YUV9/YV12
35
  BGR24 -> BGR32 & RGB24 -> RGB32
36
  BGR32 -> BGR24 & RGB32 -> RGB24
37
  BGR15 -> BGR16
38
*/
39

    
40
/*
41
tested special converters (most are tested actually, but I did not write it down ...)
42
 YV12 -> BGR16
43
 YV12 -> YV12
44
 BGR15 -> BGR16
45
 BGR16 -> BGR16
46
 YVU9 -> YV12
47

48
untested special converters
49
  YV12/I420 -> BGR15/BGR24/BGR32 (it is the yuv2rgb stuff, so it should be ok)
50
  YV12/I420 -> YV12/I420
51
  YUY2/BGR15/BGR24/BGR32/RGB24/RGB32 -> same format
52
  BGR24 -> BGR32 & RGB24 -> RGB32
53
  BGR32 -> BGR24 & RGB32 -> RGB24
54
  BGR24 -> YV12
55
*/
56

    
57
#include <inttypes.h>
58
#include <string.h>
59
#include <math.h>
60
#include <stdio.h>
61
#include <unistd.h>
62
#include "config.h"
63
#include <assert.h>
64
#ifdef HAVE_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 "rgb2rgb.h"
75
#include "libavcodec/opt.h"
76

    
77
#undef MOVNTQ
78
#undef PAVGB
79

    
80
//#undef HAVE_MMX2
81
//#define HAVE_3DNOW
82
//#undef HAVE_MMX
83
//#undef ARCH_X86
84
//#define WORDS_BIGENDIAN
85
#define DITHER1XBPP
86

    
87
#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
88

    
89
#define RET 0xC3 //near return opcode for X86
90

    
91
#ifdef MP_DEBUG
92
#define ASSERT(x) assert(x);
93
#else
94
#define ASSERT(x) ;
95
#endif
96

    
97
#ifdef M_PI
98
#define PI M_PI
99
#else
100
#define PI 3.14159265358979323846
101
#endif
102

    
103
#define isSupportedIn(x)    (       \
104
           (x)==PIX_FMT_YUV420P     \
105
        || (x)==PIX_FMT_YUVA420P    \
106
        || (x)==PIX_FMT_YUYV422     \
107
        || (x)==PIX_FMT_UYVY422     \
108
        || (x)==PIX_FMT_RGB32       \
109
        || (x)==PIX_FMT_BGR24       \
110
        || (x)==PIX_FMT_BGR565      \
111
        || (x)==PIX_FMT_BGR555      \
112
        || (x)==PIX_FMT_BGR32       \
113
        || (x)==PIX_FMT_RGB24       \
114
        || (x)==PIX_FMT_RGB565      \
115
        || (x)==PIX_FMT_RGB555      \
116
        || (x)==PIX_FMT_GRAY8       \
117
        || (x)==PIX_FMT_YUV410P     \
118
        || (x)==PIX_FMT_GRAY16BE    \
119
        || (x)==PIX_FMT_GRAY16LE    \
120
        || (x)==PIX_FMT_YUV444P     \
121
        || (x)==PIX_FMT_YUV422P     \
122
        || (x)==PIX_FMT_YUV411P     \
123
        || (x)==PIX_FMT_PAL8        \
124
        || (x)==PIX_FMT_BGR8        \
125
        || (x)==PIX_FMT_RGB8        \
126
        || (x)==PIX_FMT_BGR4_BYTE   \
127
        || (x)==PIX_FMT_RGB4_BYTE   \
128
        || (x)==PIX_FMT_YUV440P     \
129
    )
130
#define isSupportedOut(x)   (       \
131
           (x)==PIX_FMT_YUV420P     \
132
        || (x)==PIX_FMT_YUYV422     \
133
        || (x)==PIX_FMT_UYVY422     \
134
        || (x)==PIX_FMT_YUV444P     \
135
        || (x)==PIX_FMT_YUV422P     \
136
        || (x)==PIX_FMT_YUV411P     \
137
        || isRGB(x)                 \
138
        || isBGR(x)                 \
139
        || (x)==PIX_FMT_NV12        \
140
        || (x)==PIX_FMT_NV21        \
141
        || (x)==PIX_FMT_GRAY16BE    \
142
        || (x)==PIX_FMT_GRAY16LE    \
143
        || (x)==PIX_FMT_GRAY8       \
144
        || (x)==PIX_FMT_YUV410P     \
145
    )
146
#define isPacked(x)         (       \
147
           (x)==PIX_FMT_PAL8        \
148
        || (x)==PIX_FMT_YUYV422     \
149
        || (x)==PIX_FMT_UYVY422     \
150
        || isRGB(x)                 \
151
        || isBGR(x)                 \
152
    )
153

    
154
#define RGB2YUV_SHIFT 16
155
#define BY ((int)( 0.098*(1<<RGB2YUV_SHIFT)+0.5))
156
#define BV ((int)(-0.071*(1<<RGB2YUV_SHIFT)+0.5))
157
#define BU ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
158
#define GY ((int)( 0.504*(1<<RGB2YUV_SHIFT)+0.5))
159
#define GV ((int)(-0.368*(1<<RGB2YUV_SHIFT)+0.5))
160
#define GU ((int)(-0.291*(1<<RGB2YUV_SHIFT)+0.5))
161
#define RY ((int)( 0.257*(1<<RGB2YUV_SHIFT)+0.5))
162
#define RV ((int)( 0.439*(1<<RGB2YUV_SHIFT)+0.5))
163
#define RU ((int)(-0.148*(1<<RGB2YUV_SHIFT)+0.5))
164

    
165
extern const int32_t Inverse_Table_6_9[8][4];
166

    
167
/*
168
NOTES
169
Special versions: fast Y 1:1 scaling (no interpolation in y direction)
170

171
TODO
172
more intelligent misalignment avoidance for the horizontal scaler
173
write special vertical cubic upscale version
174
Optimize C code (yv12 / minmax)
175
add support for packed pixel yuv input & output
176
add support for Y8 output
177
optimize bgr24 & bgr32
178
add BGR4 output support
179
write special BGR->BGR scaler
180
*/
181

    
182
#if defined(ARCH_X86) && defined (CONFIG_GPL)
183
DECLARE_ASM_CONST(8, uint64_t, bF8)=       0xF8F8F8F8F8F8F8F8LL;
184
DECLARE_ASM_CONST(8, uint64_t, bFC)=       0xFCFCFCFCFCFCFCFCLL;
185
DECLARE_ASM_CONST(8, uint64_t, w10)=       0x0010001000100010LL;
186
DECLARE_ASM_CONST(8, uint64_t, w02)=       0x0002000200020002LL;
187
DECLARE_ASM_CONST(8, uint64_t, bm00001111)=0x00000000FFFFFFFFLL;
188
DECLARE_ASM_CONST(8, uint64_t, bm00000111)=0x0000000000FFFFFFLL;
189
DECLARE_ASM_CONST(8, uint64_t, bm11111000)=0xFFFFFFFFFF000000LL;
190
DECLARE_ASM_CONST(8, uint64_t, bm01010101)=0x00FF00FF00FF00FFLL;
191

    
192
static volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
193
static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
194
static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
195
static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
196

    
197
const DECLARE_ALIGNED(8, uint64_t, ff_dither4[2]) = {
198
        0x0103010301030103LL,
199
        0x0200020002000200LL,};
200

    
201
const DECLARE_ALIGNED(8, uint64_t, ff_dither8[2]) = {
202
        0x0602060206020602LL,
203
        0x0004000400040004LL,};
204

    
205
DECLARE_ASM_CONST(8, uint64_t, b16Mask)=   0x001F001F001F001FLL;
206
DECLARE_ASM_CONST(8, uint64_t, g16Mask)=   0x07E007E007E007E0LL;
207
DECLARE_ASM_CONST(8, uint64_t, r16Mask)=   0xF800F800F800F800LL;
208
DECLARE_ASM_CONST(8, uint64_t, b15Mask)=   0x001F001F001F001FLL;
209
DECLARE_ASM_CONST(8, uint64_t, g15Mask)=   0x03E003E003E003E0LL;
210
DECLARE_ASM_CONST(8, uint64_t, r15Mask)=   0x7C007C007C007C00LL;
211

    
212
DECLARE_ALIGNED(8, const uint64_t, ff_M24A)         = 0x00FF0000FF0000FFLL;
213
DECLARE_ALIGNED(8, const uint64_t, ff_M24B)         = 0xFF0000FF0000FF00LL;
214
DECLARE_ALIGNED(8, const uint64_t, ff_M24C)         = 0x0000FF0000FF0000LL;
215

    
216
#ifdef FAST_BGR2YV12
217
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000000210041000DULL;
218
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000FFEEFFDC0038ULL;
219
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00000038FFD2FFF8ULL;
220
#else
221
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YCoeff)   = 0x000020E540830C8BULL;
222
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UCoeff)   = 0x0000ED0FDAC23831ULL;
223
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2VCoeff)   = 0x00003831D0E6F6EAULL;
224
#endif /* FAST_BGR2YV12 */
225
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2YOffset)  = 0x1010101010101010ULL;
226
DECLARE_ALIGNED(8, const uint64_t, ff_bgr2UVOffset) = 0x8080808080808080ULL;
227
DECLARE_ALIGNED(8, const uint64_t, ff_w1111)        = 0x0001000100010001ULL;
228
#endif /* defined(ARCH_X86) */
229

    
230
// clipping helper table for C implementations:
231
static unsigned char clip_table[768];
232

    
233
static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b);
234

    
235
extern const uint8_t dither_2x2_4[2][8];
236
extern const uint8_t dither_2x2_8[2][8];
237
extern const uint8_t dither_8x8_32[8][8];
238
extern const uint8_t dither_8x8_73[8][8];
239
extern const uint8_t dither_8x8_220[8][8];
240

    
241
static const char * sws_context_to_name(void * ptr) {
242
    return "swscaler";
243
}
244

    
245
#define OFFSET(x) offsetof(SwsContext, x)
246
#define DEFAULT 0
247
#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
248

    
249
static const AVOption options[] = {
250
    { "sws_flags", "scaler/cpu flags", OFFSET(flags), FF_OPT_TYPE_FLAGS, DEFAULT, 0, UINT_MAX, VE, "sws_flags" },
251
    { "fast_bilinear", "fast bilinear", 0, FF_OPT_TYPE_CONST, SWS_FAST_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
252
    { "bilinear", "bilinear", 0, FF_OPT_TYPE_CONST, SWS_BILINEAR, INT_MIN, INT_MAX, VE, "sws_flags" },
253
    { "bicubic", "bicubic", 0, FF_OPT_TYPE_CONST, SWS_BICUBIC, INT_MIN, INT_MAX, VE, "sws_flags" },
254
    { "experimental", "experimental", 0, FF_OPT_TYPE_CONST, SWS_X, INT_MIN, INT_MAX, VE, "sws_flags" },
255
    { "neighbor", "nearest neighbor", 0, FF_OPT_TYPE_CONST, SWS_POINT, INT_MIN, INT_MAX, VE, "sws_flags" },
256
    { "area", "averaging area", 0, FF_OPT_TYPE_CONST, SWS_AREA, INT_MIN, INT_MAX, VE, "sws_flags" },
257
    { "bicublin", "luma bicubic, chroma bilinear", 0, FF_OPT_TYPE_CONST, SWS_BICUBLIN, INT_MIN, INT_MAX, VE, "sws_flags" },
258
    { "gauss", "gaussian", 0, FF_OPT_TYPE_CONST, SWS_GAUSS, INT_MIN, INT_MAX, VE, "sws_flags" },
259
    { "sinc", "sinc", 0, FF_OPT_TYPE_CONST, SWS_SINC, INT_MIN, INT_MAX, VE, "sws_flags" },
260
    { "lanczos", "lanczos", 0, FF_OPT_TYPE_CONST, SWS_LANCZOS, INT_MIN, INT_MAX, VE, "sws_flags" },
261
    { "spline", "natural bicubic spline", 0, FF_OPT_TYPE_CONST, SWS_SPLINE, INT_MIN, INT_MAX, VE, "sws_flags" },
262
    { "print_info", "print info", 0, FF_OPT_TYPE_CONST, SWS_PRINT_INFO, INT_MIN, INT_MAX, VE, "sws_flags" },
263
    { "accurate_rnd", "accurate rounding", 0, FF_OPT_TYPE_CONST, SWS_ACCURATE_RND, INT_MIN, INT_MAX, VE, "sws_flags" },
264
    { "mmx", "MMX SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX, INT_MIN, INT_MAX, VE, "sws_flags" },
265
    { "mmx2", "MMX2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_MMX2, INT_MIN, INT_MAX, VE, "sws_flags" },
266
    { "3dnow", "3DNOW SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_3DNOW, INT_MIN, INT_MAX, VE, "sws_flags" },
267
    { "altivec", "AltiVec SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_ALTIVEC, INT_MIN, INT_MAX, VE, "sws_flags" },
268
    { "bfin", "Blackfin SIMD acceleration", 0, FF_OPT_TYPE_CONST, SWS_CPU_CAPS_BFIN, INT_MIN, INT_MAX, VE, "sws_flags" },
269
    { "full_chroma_int", "full chroma interpolation", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INT, INT_MIN, INT_MAX, VE, "sws_flags" },
270
    { "full_chroma_inp", "full chroma input", 0 , FF_OPT_TYPE_CONST, SWS_FULL_CHR_H_INP, INT_MIN, INT_MAX, VE, "sws_flags" },
271
    { NULL }
272
};
273

    
274
#undef VE
275
#undef DEFAULT
276

    
277
static const AVClass sws_context_class = { "SWScaler", sws_context_to_name, options };
278

    
279
const char *sws_format_name(enum PixelFormat format)
280
{
281
    switch (format) {
282
        case PIX_FMT_YUV420P:
283
            return "yuv420p";
284
        case PIX_FMT_YUVA420P:
285
            return "yuva420p";
286
        case PIX_FMT_YUYV422:
287
            return "yuyv422";
288
        case PIX_FMT_RGB24:
289
            return "rgb24";
290
        case PIX_FMT_BGR24:
291
            return "bgr24";
292
        case PIX_FMT_YUV422P:
293
            return "yuv422p";
294
        case PIX_FMT_YUV444P:
295
            return "yuv444p";
296
        case PIX_FMT_RGB32:
297
            return "rgb32";
298
        case PIX_FMT_YUV410P:
299
            return "yuv410p";
300
        case PIX_FMT_YUV411P:
301
            return "yuv411p";
302
        case PIX_FMT_RGB565:
303
            return "rgb565";
304
        case PIX_FMT_RGB555:
305
            return "rgb555";
306
        case PIX_FMT_GRAY16BE:
307
            return "gray16be";
308
        case PIX_FMT_GRAY16LE:
309
            return "gray16le";
310
        case PIX_FMT_GRAY8:
311
            return "gray8";
312
        case PIX_FMT_MONOWHITE:
313
            return "mono white";
314
        case PIX_FMT_MONOBLACK:
315
            return "mono black";
316
        case PIX_FMT_PAL8:
317
            return "Palette";
318
        case PIX_FMT_YUVJ420P:
319
            return "yuvj420p";
320
        case PIX_FMT_YUVJ422P:
321
            return "yuvj422p";
322
        case PIX_FMT_YUVJ444P:
323
            return "yuvj444p";
324
        case PIX_FMT_XVMC_MPEG2_MC:
325
            return "xvmc_mpeg2_mc";
326
        case PIX_FMT_XVMC_MPEG2_IDCT:
327
            return "xvmc_mpeg2_idct";
328
        case PIX_FMT_UYVY422:
329
            return "uyvy422";
330
        case PIX_FMT_UYYVYY411:
331
            return "uyyvyy411";
332
        case PIX_FMT_RGB32_1:
333
            return "rgb32x";
334
        case PIX_FMT_BGR32_1:
335
            return "bgr32x";
336
        case PIX_FMT_BGR32:
337
            return "bgr32";
338
        case PIX_FMT_BGR565:
339
            return "bgr565";
340
        case PIX_FMT_BGR555:
341
            return "bgr555";
342
        case PIX_FMT_BGR8:
343
            return "bgr8";
344
        case PIX_FMT_BGR4:
345
            return "bgr4";
346
        case PIX_FMT_BGR4_BYTE:
347
            return "bgr4 byte";
348
        case PIX_FMT_RGB8:
349
            return "rgb8";
350
        case PIX_FMT_RGB4:
351
            return "rgb4";
352
        case PIX_FMT_RGB4_BYTE:
353
            return "rgb4 byte";
354
        case PIX_FMT_NV12:
355
            return "nv12";
356
        case PIX_FMT_NV21:
357
            return "nv21";
358
        case PIX_FMT_YUV440P:
359
            return "yuv440p";
360
        default:
361
            return "Unknown format";
362
    }
363
}
364

    
365
#if defined(ARCH_X86) && defined (CONFIG_GPL)
366
void in_asm_used_var_warning_killer()
367
{
368
    volatile int i= bF8+bFC+w10+
369
    bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+
370
    ff_M24A+ff_M24B+ff_M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+ff_dither4[0]+ff_dither8[0]+bm01010101;
371
    if (i) i=0;
372
}
373
#endif
374

    
375
static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
376
                               int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
377
                               uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW, int chrDstW)
378
{
379
    //FIXME Optimize (just quickly writen not opti..)
380
    int i;
381
    for (i=0; i<dstW; i++)
382
    {
383
        int val=1<<18;
384
        int j;
385
        for (j=0; j<lumFilterSize; j++)
386
            val += lumSrc[j][i] * lumFilter[j];
387

    
388
        dest[i]= av_clip_uint8(val>>19);
389
    }
390

    
391
    if (uDest)
392
        for (i=0; i<chrDstW; i++)
393
        {
394
            int u=1<<18;
395
            int v=1<<18;
396
            int j;
397
            for (j=0; j<chrFilterSize; j++)
398
            {
399
                u += chrSrc[j][i] * chrFilter[j];
400
                v += chrSrc[j][i + VOFW] * chrFilter[j];
401
            }
402

    
403
            uDest[i]= av_clip_uint8(u>>19);
404
            vDest[i]= av_clip_uint8(v>>19);
405
        }
406
}
407

    
408
static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
409
                                int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
410
                                uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
411
{
412
    //FIXME Optimize (just quickly writen not opti..)
413
    int i;
414
    for (i=0; i<dstW; i++)
415
    {
416
        int val=1<<18;
417
        int j;
418
        for (j=0; j<lumFilterSize; j++)
419
            val += lumSrc[j][i] * lumFilter[j];
420

    
421
        dest[i]= av_clip_uint8(val>>19);
422
    }
423

    
424
    if (!uDest)
425
        return;
426

    
427
    if (dstFormat == PIX_FMT_NV12)
428
        for (i=0; i<chrDstW; i++)
429
        {
430
            int u=1<<18;
431
            int v=1<<18;
432
            int j;
433
            for (j=0; j<chrFilterSize; j++)
434
            {
435
                u += chrSrc[j][i] * chrFilter[j];
436
                v += chrSrc[j][i + VOFW] * chrFilter[j];
437
            }
438

    
439
            uDest[2*i]= av_clip_uint8(u>>19);
440
            uDest[2*i+1]= av_clip_uint8(v>>19);
441
        }
442
    else
443
        for (i=0; i<chrDstW; i++)
444
        {
445
            int u=1<<18;
446
            int v=1<<18;
447
            int j;
448
            for (j=0; j<chrFilterSize; j++)
449
            {
450
                u += chrSrc[j][i] * chrFilter[j];
451
                v += chrSrc[j][i + VOFW] * chrFilter[j];
452
            }
453

    
454
            uDest[2*i]= av_clip_uint8(v>>19);
455
            uDest[2*i+1]= av_clip_uint8(u>>19);
456
        }
457
}
458

    
459
#define YSCALE_YUV_2_PACKEDX_C(type) \
460
    for (i=0; i<(dstW>>1); i++){\
461
        int j;\
462
        int Y1 = 1<<18;\
463
        int Y2 = 1<<18;\
464
        int U  = 1<<18;\
465
        int V  = 1<<18;\
466
        type av_unused *r, *b, *g;\
467
        const int i2= 2*i;\
468
        \
469
        for (j=0; j<lumFilterSize; j++)\
470
        {\
471
            Y1 += lumSrc[j][i2] * lumFilter[j];\
472
            Y2 += lumSrc[j][i2+1] * lumFilter[j];\
473
        }\
474
        for (j=0; j<chrFilterSize; j++)\
475
        {\
476
            U += chrSrc[j][i] * chrFilter[j];\
477
            V += chrSrc[j][i+VOFW] * chrFilter[j];\
478
        }\
479
        Y1>>=19;\
480
        Y2>>=19;\
481
        U >>=19;\
482
        V >>=19;\
483
        if ((Y1|Y2|U|V)&256)\
484
        {\
485
            if (Y1>255)   Y1=255; \
486
            else if (Y1<0)Y1=0;   \
487
            if (Y2>255)   Y2=255; \
488
            else if (Y2<0)Y2=0;   \
489
            if (U>255)    U=255;  \
490
            else if (U<0) U=0;    \
491
            if (V>255)    V=255;  \
492
            else if (V<0) V=0;    \
493
        }
494

    
495
#define YSCALE_YUV_2_RGBX_C(type) \
496
    YSCALE_YUV_2_PACKEDX_C(type)  \
497
    r = (type *)c->table_rV[V];   \
498
    g = (type *)(c->table_gU[U] + c->table_gV[V]); \
499
    b = (type *)c->table_bU[U];   \
500

    
501
#define YSCALE_YUV_2_PACKED2_C   \
502
    for (i=0; i<(dstW>>1); i++){ \
503
        const int i2= 2*i;       \
504
        int Y1= (buf0[i2  ]*yalpha1+buf1[i2  ]*yalpha)>>19;           \
505
        int Y2= (buf0[i2+1]*yalpha1+buf1[i2+1]*yalpha)>>19;           \
506
        int U= (uvbuf0[i     ]*uvalpha1+uvbuf1[i     ]*uvalpha)>>19;  \
507
        int V= (uvbuf0[i+VOFW]*uvalpha1+uvbuf1[i+VOFW]*uvalpha)>>19;  \
508

    
509
#define YSCALE_YUV_2_RGB2_C(type) \
510
    YSCALE_YUV_2_PACKED2_C\
511
    type *r, *b, *g;\
512
    r = (type *)c->table_rV[V];\
513
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
514
    b = (type *)c->table_bU[U];\
515

    
516
#define YSCALE_YUV_2_PACKED1_C \
517
    for (i=0; i<(dstW>>1); i++){\
518
        const int i2= 2*i;\
519
        int Y1= buf0[i2  ]>>7;\
520
        int Y2= buf0[i2+1]>>7;\
521
        int U= (uvbuf1[i     ])>>7;\
522
        int V= (uvbuf1[i+VOFW])>>7;\
523

    
524
#define YSCALE_YUV_2_RGB1_C(type) \
525
    YSCALE_YUV_2_PACKED1_C\
526
    type *r, *b, *g;\
527
    r = (type *)c->table_rV[V];\
528
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
529
    b = (type *)c->table_bU[U];\
530

    
531
#define YSCALE_YUV_2_PACKED1B_C \
532
    for (i=0; i<(dstW>>1); i++){\
533
        const int i2= 2*i;\
534
        int Y1= buf0[i2  ]>>7;\
535
        int Y2= buf0[i2+1]>>7;\
536
        int U= (uvbuf0[i     ] + uvbuf1[i     ])>>8;\
537
        int V= (uvbuf0[i+VOFW] + uvbuf1[i+VOFW])>>8;\
538

    
539
#define YSCALE_YUV_2_RGB1B_C(type) \
540
    YSCALE_YUV_2_PACKED1B_C\
541
    type *r, *b, *g;\
542
    r = (type *)c->table_rV[V];\
543
    g = (type *)(c->table_gU[U] + c->table_gV[V]);\
544
    b = (type *)c->table_bU[U];\
545

    
546
#define YSCALE_YUV_2_ANYRGB_C(func, func2)\
547
    switch(c->dstFormat)\
548
    {\
549
    case PIX_FMT_RGB32:\
550
    case PIX_FMT_BGR32:\
551
        func(uint32_t)\
552
            ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];\
553
            ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];\
554
        }                \
555
        break;\
556
    case PIX_FMT_RGB24:\
557
        func(uint8_t)\
558
            ((uint8_t*)dest)[0]= r[Y1];\
559
            ((uint8_t*)dest)[1]= g[Y1];\
560
            ((uint8_t*)dest)[2]= b[Y1];\
561
            ((uint8_t*)dest)[3]= r[Y2];\
562
            ((uint8_t*)dest)[4]= g[Y2];\
563
            ((uint8_t*)dest)[5]= b[Y2];\
564
            dest+=6;\
565
        }\
566
        break;\
567
    case PIX_FMT_BGR24:\
568
        func(uint8_t)\
569
            ((uint8_t*)dest)[0]= b[Y1];\
570
            ((uint8_t*)dest)[1]= g[Y1];\
571
            ((uint8_t*)dest)[2]= r[Y1];\
572
            ((uint8_t*)dest)[3]= b[Y2];\
573
            ((uint8_t*)dest)[4]= g[Y2];\
574
            ((uint8_t*)dest)[5]= r[Y2];\
575
            dest+=6;\
576
        }\
577
        break;\
578
    case PIX_FMT_RGB565:\
579
    case PIX_FMT_BGR565:\
580
        {\
581
            const int dr1= dither_2x2_8[y&1    ][0];\
582
            const int dg1= dither_2x2_4[y&1    ][0];\
583
            const int db1= dither_2x2_8[(y&1)^1][0];\
584
            const int dr2= dither_2x2_8[y&1    ][1];\
585
            const int dg2= dither_2x2_4[y&1    ][1];\
586
            const int db2= dither_2x2_8[(y&1)^1][1];\
587
            func(uint16_t)\
588
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
589
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
590
            }\
591
        }\
592
        break;\
593
    case PIX_FMT_RGB555:\
594
    case PIX_FMT_BGR555:\
595
        {\
596
            const int dr1= dither_2x2_8[y&1    ][0];\
597
            const int dg1= dither_2x2_8[y&1    ][1];\
598
            const int db1= dither_2x2_8[(y&1)^1][0];\
599
            const int dr2= dither_2x2_8[y&1    ][1];\
600
            const int dg2= dither_2x2_8[y&1    ][0];\
601
            const int db2= dither_2x2_8[(y&1)^1][1];\
602
            func(uint16_t)\
603
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];\
604
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];\
605
            }\
606
        }\
607
        break;\
608
    case PIX_FMT_RGB8:\
609
    case PIX_FMT_BGR8:\
610
        {\
611
            const uint8_t * const d64= dither_8x8_73[y&7];\
612
            const uint8_t * const d32= dither_8x8_32[y&7];\
613
            func(uint8_t)\
614
                ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];\
615
                ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];\
616
            }\
617
        }\
618
        break;\
619
    case PIX_FMT_RGB4:\
620
    case PIX_FMT_BGR4:\
621
        {\
622
            const uint8_t * const d64= dither_8x8_73 [y&7];\
623
            const uint8_t * const d128=dither_8x8_220[y&7];\
624
            func(uint8_t)\
625
                ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]\
626
                                 + ((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);\
627
            }\
628
        }\
629
        break;\
630
    case PIX_FMT_RGB4_BYTE:\
631
    case PIX_FMT_BGR4_BYTE:\
632
        {\
633
            const uint8_t * const d64= dither_8x8_73 [y&7];\
634
            const uint8_t * const d128=dither_8x8_220[y&7];\
635
            func(uint8_t)\
636
                ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];\
637
                ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];\
638
            }\
639
        }\
640
        break;\
641
    case PIX_FMT_MONOBLACK:\
642
        {\
643
            const uint8_t * const d128=dither_8x8_220[y&7];\
644
            uint8_t *g= c->table_gU[128] + c->table_gV[128];\
645
            for (i=0; i<dstW-7; i+=8){\
646
                int acc;\
647
                acc =       g[((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19) + d128[0]];\
648
                acc+= acc + g[((buf0[i+1]*yalpha1+buf1[i+1]*yalpha)>>19) + d128[1]];\
649
                acc+= acc + g[((buf0[i+2]*yalpha1+buf1[i+2]*yalpha)>>19) + d128[2]];\
650
                acc+= acc + g[((buf0[i+3]*yalpha1+buf1[i+3]*yalpha)>>19) + d128[3]];\
651
                acc+= acc + g[((buf0[i+4]*yalpha1+buf1[i+4]*yalpha)>>19) + d128[4]];\
652
                acc+= acc + g[((buf0[i+5]*yalpha1+buf1[i+5]*yalpha)>>19) + d128[5]];\
653
                acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
654
                acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
655
                ((uint8_t*)dest)[0]= acc;\
656
                dest++;\
657
            }\
658
\
659
/*\
660
((uint8_t*)dest)-= dstW>>4;\
661
{\
662
            int acc=0;\
663
            int left=0;\
664
            static int top[1024];\
665
            static int last_new[1024][1024];\
666
            static int last_in3[1024][1024];\
667
            static int drift[1024][1024];\
668
            int topLeft=0;\
669
            int shift=0;\
670
            int count=0;\
671
            const uint8_t * const d128=dither_8x8_220[y&7];\
672
            int error_new=0;\
673
            int error_in3=0;\
674
            int f=0;\
675
            \
676
            for (i=dstW>>1; i<dstW; i++){\
677
                int in= ((buf0[i  ]*yalpha1+buf1[i  ]*yalpha)>>19);\
678
                int in2 = (76309 * (in - 16) + 32768) >> 16;\
679
                int in3 = (in2 < 0) ? 0 : ((in2 > 255) ? 255 : in2);\
680
                int old= (left*7 + topLeft + top[i]*5 + top[i+1]*3)/20 + in3\
681
                         + (last_new[y][i] - in3)*f/256;\
682
                int new= old> 128 ? 255 : 0;\
683
\
684
                error_new+= FFABS(last_new[y][i] - new);\
685
                error_in3+= FFABS(last_in3[y][i] - in3);\
686
                f= error_new - error_in3*4;\
687
                if (f<0) f=0;\
688
                if (f>256) f=256;\
689
\
690
                topLeft= top[i];\
691
                left= top[i]= old - new;\
692
                last_new[y][i]= new;\
693
                last_in3[y][i]= in3;\
694
\
695
                acc+= acc + (new&1);\
696
                if ((i&7)==6){\
697
                    ((uint8_t*)dest)[0]= acc;\
698
                    ((uint8_t*)dest)++;\
699
                }\
700
            }\
701
}\
702
*/\
703
        }\
704
        break;\
705
    case PIX_FMT_YUYV422:\
706
        func2\
707
            ((uint8_t*)dest)[2*i2+0]= Y1;\
708
            ((uint8_t*)dest)[2*i2+1]= U;\
709
            ((uint8_t*)dest)[2*i2+2]= Y2;\
710
            ((uint8_t*)dest)[2*i2+3]= V;\
711
        }                \
712
        break;\
713
    case PIX_FMT_UYVY422:\
714
        func2\
715
            ((uint8_t*)dest)[2*i2+0]= U;\
716
            ((uint8_t*)dest)[2*i2+1]= Y1;\
717
            ((uint8_t*)dest)[2*i2+2]= V;\
718
            ((uint8_t*)dest)[2*i2+3]= Y2;\
719
        }                \
720
        break;\
721
    }\
722

    
723

    
724
static inline void yuv2packedXinC(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
725
                                  int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
726
                                  uint8_t *dest, int dstW, int y)
727
{
728
    int i;
729
    switch(c->dstFormat)
730
    {
731
    case PIX_FMT_BGR32:
732
    case PIX_FMT_RGB32:
733
        YSCALE_YUV_2_RGBX_C(uint32_t)
734
            ((uint32_t*)dest)[i2+0]= r[Y1] + g[Y1] + b[Y1];
735
            ((uint32_t*)dest)[i2+1]= r[Y2] + g[Y2] + b[Y2];
736
        }
737
        break;
738
    case PIX_FMT_RGB24:
739
        YSCALE_YUV_2_RGBX_C(uint8_t)
740
            ((uint8_t*)dest)[0]= r[Y1];
741
            ((uint8_t*)dest)[1]= g[Y1];
742
            ((uint8_t*)dest)[2]= b[Y1];
743
            ((uint8_t*)dest)[3]= r[Y2];
744
            ((uint8_t*)dest)[4]= g[Y2];
745
            ((uint8_t*)dest)[5]= b[Y2];
746
            dest+=6;
747
        }
748
        break;
749
    case PIX_FMT_BGR24:
750
        YSCALE_YUV_2_RGBX_C(uint8_t)
751
            ((uint8_t*)dest)[0]= b[Y1];
752
            ((uint8_t*)dest)[1]= g[Y1];
753
            ((uint8_t*)dest)[2]= r[Y1];
754
            ((uint8_t*)dest)[3]= b[Y2];
755
            ((uint8_t*)dest)[4]= g[Y2];
756
            ((uint8_t*)dest)[5]= r[Y2];
757
            dest+=6;
758
        }
759
        break;
760
    case PIX_FMT_RGB565:
761
    case PIX_FMT_BGR565:
762
        {
763
            const int dr1= dither_2x2_8[y&1    ][0];
764
            const int dg1= dither_2x2_4[y&1    ][0];
765
            const int db1= dither_2x2_8[(y&1)^1][0];
766
            const int dr2= dither_2x2_8[y&1    ][1];
767
            const int dg2= dither_2x2_4[y&1    ][1];
768
            const int db2= dither_2x2_8[(y&1)^1][1];
769
            YSCALE_YUV_2_RGBX_C(uint16_t)
770
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
771
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
772
            }
773
        }
774
        break;
775
    case PIX_FMT_RGB555:
776
    case PIX_FMT_BGR555:
777
        {
778
            const int dr1= dither_2x2_8[y&1    ][0];
779
            const int dg1= dither_2x2_8[y&1    ][1];
780
            const int db1= dither_2x2_8[(y&1)^1][0];
781
            const int dr2= dither_2x2_8[y&1    ][1];
782
            const int dg2= dither_2x2_8[y&1    ][0];
783
            const int db2= dither_2x2_8[(y&1)^1][1];
784
            YSCALE_YUV_2_RGBX_C(uint16_t)
785
                ((uint16_t*)dest)[i2+0]= r[Y1+dr1] + g[Y1+dg1] + b[Y1+db1];
786
                ((uint16_t*)dest)[i2+1]= r[Y2+dr2] + g[Y2+dg2] + b[Y2+db2];
787
            }
788
        }
789
        break;
790
    case PIX_FMT_RGB8:
791
    case PIX_FMT_BGR8:
792
        {
793
            const uint8_t * const d64= dither_8x8_73[y&7];
794
            const uint8_t * const d32= dither_8x8_32[y&7];
795
            YSCALE_YUV_2_RGBX_C(uint8_t)
796
                ((uint8_t*)dest)[i2+0]= r[Y1+d32[(i2+0)&7]] + g[Y1+d32[(i2+0)&7]] + b[Y1+d64[(i2+0)&7]];
797
                ((uint8_t*)dest)[i2+1]= r[Y2+d32[(i2+1)&7]] + g[Y2+d32[(i2+1)&7]] + b[Y2+d64[(i2+1)&7]];
798
            }
799
        }
800
        break;
801
    case PIX_FMT_RGB4:
802
    case PIX_FMT_BGR4:
803
        {
804
            const uint8_t * const d64= dither_8x8_73 [y&7];
805
            const uint8_t * const d128=dither_8x8_220[y&7];
806
            YSCALE_YUV_2_RGBX_C(uint8_t)
807
                ((uint8_t*)dest)[i]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]]
808
                                  +((r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]])<<4);
809
            }
810
        }
811
        break;
812
    case PIX_FMT_RGB4_BYTE:
813
    case PIX_FMT_BGR4_BYTE:
814
        {
815
            const uint8_t * const d64= dither_8x8_73 [y&7];
816
            const uint8_t * const d128=dither_8x8_220[y&7];
817
            YSCALE_YUV_2_RGBX_C(uint8_t)
818
                ((uint8_t*)dest)[i2+0]= r[Y1+d128[(i2+0)&7]] + g[Y1+d64[(i2+0)&7]] + b[Y1+d128[(i2+0)&7]];
819
                ((uint8_t*)dest)[i2+1]= r[Y2+d128[(i2+1)&7]] + g[Y2+d64[(i2+1)&7]] + b[Y2+d128[(i2+1)&7]];
820
            }
821
        }
822
        break;
823
    case PIX_FMT_MONOBLACK:
824
        {
825
            const uint8_t * const d128=dither_8x8_220[y&7];
826
            uint8_t *g= c->table_gU[128] + c->table_gV[128];
827
            int acc=0;
828
            for (i=0; i<dstW-1; i+=2){
829
                int j;
830
                int Y1=1<<18;
831
                int Y2=1<<18;
832

    
833
                for (j=0; j<lumFilterSize; j++)
834
                {
835
                    Y1 += lumSrc[j][i] * lumFilter[j];
836
                    Y2 += lumSrc[j][i+1] * lumFilter[j];
837
                }
838
                Y1>>=19;
839
                Y2>>=19;
840
                if ((Y1|Y2)&256)
841
                {
842
                    if (Y1>255)   Y1=255;
843
                    else if (Y1<0)Y1=0;
844
                    if (Y2>255)   Y2=255;
845
                    else if (Y2<0)Y2=0;
846
                }
847
                acc+= acc + g[Y1+d128[(i+0)&7]];
848
                acc+= acc + g[Y2+d128[(i+1)&7]];
849
                if ((i&7)==6){
850
                    ((uint8_t*)dest)[0]= acc;
851
                    dest++;
852
                }
853
            }
854
        }
855
        break;
856
    case PIX_FMT_YUYV422:
857
        YSCALE_YUV_2_PACKEDX_C(void)
858
            ((uint8_t*)dest)[2*i2+0]= Y1;
859
            ((uint8_t*)dest)[2*i2+1]= U;
860
            ((uint8_t*)dest)[2*i2+2]= Y2;
861
            ((uint8_t*)dest)[2*i2+3]= V;
862
        }
863
        break;
864
    case PIX_FMT_UYVY422:
865
        YSCALE_YUV_2_PACKEDX_C(void)
866
            ((uint8_t*)dest)[2*i2+0]= U;
867
            ((uint8_t*)dest)[2*i2+1]= Y1;
868
            ((uint8_t*)dest)[2*i2+2]= V;
869
            ((uint8_t*)dest)[2*i2+3]= Y2;
870
        }
871
        break;
872
    }
873
}
874

    
875

    
876
//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
877
//Plain C versions
878
#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT) || !defined(CONFIG_GPL)
879
#define COMPILE_C
880
#endif
881

    
882
#ifdef ARCH_POWERPC
883
#if (defined (HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
884
#define COMPILE_ALTIVEC
885
#endif //HAVE_ALTIVEC
886
#endif //ARCH_POWERPC
887

    
888
#if defined(ARCH_X86)
889

    
890
#if ((defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
891
#define COMPILE_MMX
892
#endif
893

    
894
#if (defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
895
#define COMPILE_MMX2
896
#endif
897

    
898
#if ((defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)) && defined (CONFIG_GPL)
899
#define COMPILE_3DNOW
900
#endif
901
#endif //ARCH_X86 || ARCH_X86_64
902

    
903
#undef HAVE_MMX
904
#undef HAVE_MMX2
905
#undef HAVE_3DNOW
906

    
907
#ifdef COMPILE_C
908
#undef HAVE_MMX
909
#undef HAVE_MMX2
910
#undef HAVE_3DNOW
911
#undef HAVE_ALTIVEC
912
#define RENAME(a) a ## _C
913
#include "swscale_template.c"
914
#endif
915

    
916
#ifdef COMPILE_ALTIVEC
917
#undef RENAME
918
#define HAVE_ALTIVEC
919
#define RENAME(a) a ## _altivec
920
#include "swscale_template.c"
921
#endif
922

    
923
#if defined(ARCH_X86)
924

    
925
//X86 versions
926
/*
927
#undef RENAME
928
#undef HAVE_MMX
929
#undef HAVE_MMX2
930
#undef HAVE_3DNOW
931
#define ARCH_X86
932
#define RENAME(a) a ## _X86
933
#include "swscale_template.c"
934
*/
935
//MMX versions
936
#ifdef COMPILE_MMX
937
#undef RENAME
938
#define HAVE_MMX
939
#undef HAVE_MMX2
940
#undef HAVE_3DNOW
941
#define RENAME(a) a ## _MMX
942
#include "swscale_template.c"
943
#endif
944

    
945
//MMX2 versions
946
#ifdef COMPILE_MMX2
947
#undef RENAME
948
#define HAVE_MMX
949
#define HAVE_MMX2
950
#undef HAVE_3DNOW
951
#define RENAME(a) a ## _MMX2
952
#include "swscale_template.c"
953
#endif
954

    
955
//3DNOW versions
956
#ifdef COMPILE_3DNOW
957
#undef RENAME
958
#define HAVE_MMX
959
#undef HAVE_MMX2
960
#define HAVE_3DNOW
961
#define RENAME(a) a ## _3DNow
962
#include "swscale_template.c"
963
#endif
964

    
965
#endif //ARCH_X86 || ARCH_X86_64
966

    
967
// minor note: the HAVE_xyz is messed up after that line so don't use it
968

    
969
static double getSplineCoeff(double a, double b, double c, double d, double dist)
970
{
971
//    printf("%f %f %f %f %f\n", a,b,c,d,dist);
972
    if (dist<=1.0)      return ((d*dist + c)*dist + b)*dist +a;
973
    else                return getSplineCoeff(        0.0,
974
                                             b+ 2.0*c + 3.0*d,
975
                                                    c + 3.0*d,
976
                                            -b- 3.0*c - 6.0*d,
977
                                            dist-1.0);
978
}
979

    
980
static inline int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
981
                             int srcW, int dstW, int filterAlign, int one, int flags,
982
                             SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
983
{
984
    int i;
985
    int filterSize;
986
    int filter2Size;
987
    int minFilterSize;
988
    double *filter=NULL;
989
    double *filter2=NULL;
990
#if defined(ARCH_X86)
991
    if (flags & SWS_CPU_CAPS_MMX)
992
        asm volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
993
#endif
994

    
995
    // Note the +1 is for the MMXscaler which reads over the end
996
    *filterPos = av_malloc((dstW+1)*sizeof(int16_t));
997

    
998
    if (FFABS(xInc - 0x10000) <10) // unscaled
999
    {
1000
        int i;
1001
        filterSize= 1;
1002
        filter= av_malloc(dstW*sizeof(double)*filterSize);
1003
        for (i=0; i<dstW*filterSize; i++) filter[i]=0;
1004

    
1005
        for (i=0; i<dstW; i++)
1006
        {
1007
            filter[i*filterSize]=1;
1008
            (*filterPos)[i]=i;
1009
        }
1010

    
1011
    }
1012
    else if (flags&SWS_POINT) // lame looking point sampling mode
1013
    {
1014
        int i;
1015
        int xDstInSrc;
1016
        filterSize= 1;
1017
        filter= av_malloc(dstW*sizeof(double)*filterSize);
1018

    
1019
        xDstInSrc= xInc/2 - 0x8000;
1020
        for (i=0; i<dstW; i++)
1021
        {
1022
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1023

    
1024
            (*filterPos)[i]= xx;
1025
            filter[i]= 1.0;
1026
            xDstInSrc+= xInc;
1027
        }
1028
    }
1029
    else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) // bilinear upscale
1030
    {
1031
        int i;
1032
        int xDstInSrc;
1033
        if      (flags&SWS_BICUBIC) filterSize= 4;
1034
        else if (flags&SWS_X      ) filterSize= 4;
1035
        else                        filterSize= 2; // SWS_BILINEAR / SWS_AREA
1036
        filter= av_malloc(dstW*sizeof(double)*filterSize);
1037

    
1038
        xDstInSrc= xInc/2 - 0x8000;
1039
        for (i=0; i<dstW; i++)
1040
        {
1041
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
1042
            int j;
1043

    
1044
            (*filterPos)[i]= xx;
1045
                //Bilinear upscale / linear interpolate / Area averaging
1046
                for (j=0; j<filterSize; j++)
1047
                {
1048
                    double d= FFABS((xx<<16) - xDstInSrc)/(double)(1<<16);
1049
                    double coeff= 1.0 - d;
1050
                    if (coeff<0) coeff=0;
1051
                    filter[i*filterSize + j]= coeff;
1052
                    xx++;
1053
                }
1054
            xDstInSrc+= xInc;
1055
        }
1056
    }
1057
    else
1058
    {
1059
        double xDstInSrc;
1060
        double sizeFactor, filterSizeInSrc;
1061
        const double xInc1= (double)xInc / (double)(1<<16);
1062

    
1063
        if      (flags&SWS_BICUBIC)      sizeFactor=  4.0;
1064
        else if (flags&SWS_X)            sizeFactor=  8.0;
1065
        else if (flags&SWS_AREA)         sizeFactor=  1.0; //downscale only, for upscale it is bilinear
1066
        else if (flags&SWS_GAUSS)        sizeFactor=  8.0;   // infinite ;)
1067
        else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
1068
        else if (flags&SWS_SINC)         sizeFactor= 20.0; // infinite ;)
1069
        else if (flags&SWS_SPLINE)       sizeFactor= 20.0;  // infinite ;)
1070
        else if (flags&SWS_BILINEAR)     sizeFactor=  2.0;
1071
        else {
1072
            sizeFactor= 0.0; //GCC warning killer
1073
            ASSERT(0)
1074
        }
1075

    
1076
        if (xInc1 <= 1.0)       filterSizeInSrc= sizeFactor; // upscale
1077
        else                    filterSizeInSrc= sizeFactor*srcW / (double)dstW;
1078

    
1079
        filterSize= (int)ceil(1 + filterSizeInSrc); // will be reduced later if possible
1080
        if (filterSize > srcW-2) filterSize=srcW-2;
1081

    
1082
        filter= av_malloc(dstW*sizeof(double)*filterSize);
1083

    
1084
        xDstInSrc= xInc1 / 2.0 - 0.5;
1085
        for (i=0; i<dstW; i++)
1086
        {
1087
            int xx= (int)(xDstInSrc - (filterSize-1)*0.5 + 0.5);
1088
            int j;
1089
            (*filterPos)[i]= xx;
1090
            for (j=0; j<filterSize; j++)
1091
            {
1092
                double d= FFABS(xx - xDstInSrc)/filterSizeInSrc*sizeFactor;
1093
                double coeff;
1094
                if (flags & SWS_BICUBIC)
1095
                {
1096
                    double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
1097
                    double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
1098

    
1099
                    if (d<1.0)
1100
                        coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
1101
                    else if (d<2.0)
1102
                        coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
1103
                    else
1104
                        coeff=0.0;
1105
                }
1106
/*                else if (flags & SWS_X)
1107
                {
1108
                    double p= param ? param*0.01 : 0.3;
1109
                    coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1110
                    coeff*= pow(2.0, - p*d*d);
1111
                }*/
1112
                else if (flags & SWS_X)
1113
                {
1114
                    double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
1115

    
1116
                    if (d<1.0)
1117
                        coeff = cos(d*PI);
1118
                    else
1119
                        coeff=-1.0;
1120
                    if (coeff<0.0)      coeff= -pow(-coeff, A);
1121
                    else                coeff=  pow( coeff, A);
1122
                    coeff= coeff*0.5 + 0.5;
1123
                }
1124
                else if (flags & SWS_AREA)
1125
                {
1126
                    double srcPixelSize= 1.0/xInc1;
1127
                    if      (d + srcPixelSize/2 < 0.5) coeff= 1.0;
1128
                    else if (d - srcPixelSize/2 < 0.5) coeff= (0.5-d)/srcPixelSize + 0.5;
1129
                    else coeff=0.0;
1130
                }
1131
                else if (flags & SWS_GAUSS)
1132
                {
1133
                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1134
                    coeff = pow(2.0, - p*d*d);
1135
                }
1136
                else if (flags & SWS_SINC)
1137
                {
1138
                    coeff = d ? sin(d*PI)/(d*PI) : 1.0;
1139
                }
1140
                else if (flags & SWS_LANCZOS)
1141
                {
1142
                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
1143
                    coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
1144
                    if (d>p) coeff=0;
1145
                }
1146
                else if (flags & SWS_BILINEAR)
1147
                {
1148
                    coeff= 1.0 - d;
1149
                    if (coeff<0) coeff=0;
1150
                }
1151
                else if (flags & SWS_SPLINE)
1152
                {
1153
                    double p=-2.196152422706632;
1154
                    coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, d);
1155
                }
1156
                else {
1157
                    coeff= 0.0; //GCC warning killer
1158
                    ASSERT(0)
1159
                }
1160

    
1161
                filter[i*filterSize + j]= coeff;
1162
                xx++;
1163
            }
1164
            xDstInSrc+= xInc1;
1165
        }
1166
    }
1167

    
1168
    /* apply src & dst Filter to filter -> filter2
1169
       av_free(filter);
1170
    */
1171
    ASSERT(filterSize>0)
1172
    filter2Size= filterSize;
1173
    if (srcFilter) filter2Size+= srcFilter->length - 1;
1174
    if (dstFilter) filter2Size+= dstFilter->length - 1;
1175
    ASSERT(filter2Size>0)
1176
    filter2= av_malloc(filter2Size*dstW*sizeof(double));
1177

    
1178
    for (i=0; i<dstW; i++)
1179
    {
1180
        int j;
1181
        SwsVector scaleFilter;
1182
        SwsVector *outVec;
1183

    
1184
        scaleFilter.coeff= filter + i*filterSize;
1185
        scaleFilter.length= filterSize;
1186

    
1187
        if (srcFilter) outVec= sws_getConvVec(srcFilter, &scaleFilter);
1188
        else           outVec= &scaleFilter;
1189

    
1190
        ASSERT(outVec->length == filter2Size)
1191
        //FIXME dstFilter
1192

    
1193
        for (j=0; j<outVec->length; j++)
1194
        {
1195
            filter2[i*filter2Size + j]= outVec->coeff[j];
1196
        }
1197

    
1198
        (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
1199

    
1200
        if (outVec != &scaleFilter) sws_freeVec(outVec);
1201
    }
1202
    av_free(filter); filter=NULL;
1203

    
1204
    /* try to reduce the filter-size (step1 find size and shift left) */
1205
    // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
1206
    minFilterSize= 0;
1207
    for (i=dstW-1; i>=0; i--)
1208
    {
1209
        int min= filter2Size;
1210
        int j;
1211
        double cutOff=0.0;
1212

    
1213
        /* get rid off near zero elements on the left by shifting left */
1214
        for (j=0; j<filter2Size; j++)
1215
        {
1216
            int k;
1217
            cutOff += FFABS(filter2[i*filter2Size]);
1218

    
1219
            if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1220

    
1221
            /* preserve monotonicity because the core can't handle the filter otherwise */
1222
            if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
1223

    
1224
            // Move filter coeffs left
1225
            for (k=1; k<filter2Size; k++)
1226
                filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
1227
            filter2[i*filter2Size + k - 1]= 0.0;
1228
            (*filterPos)[i]++;
1229
        }
1230

    
1231
        cutOff=0.0;
1232
        /* count near zeros on the right */
1233
        for (j=filter2Size-1; j>0; j--)
1234
        {
1235
            cutOff += FFABS(filter2[i*filter2Size + j]);
1236

    
1237
            if (cutOff > SWS_MAX_REDUCE_CUTOFF) break;
1238
            min--;
1239
        }
1240

    
1241
        if (min>minFilterSize) minFilterSize= min;
1242
    }
1243

    
1244
    if (flags & SWS_CPU_CAPS_ALTIVEC) {
1245
        // we can handle the special case 4,
1246
        // so we don't want to go to the full 8
1247
        if (minFilterSize < 5)
1248
            filterAlign = 4;
1249

    
1250
        // we really don't want to waste our time
1251
        // doing useless computation, so fall-back on
1252
        // the scalar C code for very small filter.
1253
        // vectorizing is worth it only if you have
1254
        // decent-sized vector.
1255
        if (minFilterSize < 3)
1256
            filterAlign = 1;
1257
    }
1258

    
1259
    if (flags & SWS_CPU_CAPS_MMX) {
1260
        // special case for unscaled vertical filtering
1261
        if (minFilterSize == 1 && filterAlign == 2)
1262
            filterAlign= 1;
1263
    }
1264

    
1265
    ASSERT(minFilterSize > 0)
1266
    filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
1267
    ASSERT(filterSize > 0)
1268
    filter= av_malloc(filterSize*dstW*sizeof(double));
1269
    if (filterSize >= MAX_FILTER_SIZE)
1270
        return -1;
1271
    *outFilterSize= filterSize;
1272

    
1273
    if (flags&SWS_PRINT_INFO)
1274
        av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
1275
    /* try to reduce the filter-size (step2 reduce it) */
1276
    for (i=0; i<dstW; i++)
1277
    {
1278
        int j;
1279

    
1280
        for (j=0; j<filterSize; j++)
1281
        {
1282
            if (j>=filter2Size) filter[i*filterSize + j]= 0.0;
1283
            else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
1284
        }
1285
    }
1286
    av_free(filter2); filter2=NULL;
1287

    
1288

    
1289
    //FIXME try to align filterpos if possible
1290

    
1291
    //fix borders
1292
    for (i=0; i<dstW; i++)
1293
    {
1294
        int j;
1295
        if ((*filterPos)[i] < 0)
1296
        {
1297
            // Move filter coeffs left to compensate for filterPos
1298
            for (j=1; j<filterSize; j++)
1299
            {
1300
                int left= FFMAX(j + (*filterPos)[i], 0);
1301
                filter[i*filterSize + left] += filter[i*filterSize + j];
1302
                filter[i*filterSize + j]=0;
1303
            }
1304
            (*filterPos)[i]= 0;
1305
        }
1306

    
1307
        if ((*filterPos)[i] + filterSize > srcW)
1308
        {
1309
            int shift= (*filterPos)[i] + filterSize - srcW;
1310
            // Move filter coeffs right to compensate for filterPos
1311
            for (j=filterSize-2; j>=0; j--)
1312
            {
1313
                int right= FFMIN(j + shift, filterSize-1);
1314
                filter[i*filterSize +right] += filter[i*filterSize +j];
1315
                filter[i*filterSize +j]=0;
1316
            }
1317
            (*filterPos)[i]= srcW - filterSize;
1318
        }
1319
    }
1320

    
1321
    // Note the +1 is for the MMXscaler which reads over the end
1322
    /* align at 16 for AltiVec (needed by hScale_altivec_real) */
1323
    *outFilter= av_mallocz(*outFilterSize*(dstW+1)*sizeof(int16_t));
1324

    
1325
    /* Normalize & Store in outFilter */
1326
    for (i=0; i<dstW; i++)
1327
    {
1328
        int j;
1329
        double error=0;
1330
        double sum=0;
1331
        double scale= one;
1332

    
1333
        for (j=0; j<filterSize; j++)
1334
        {
1335
            sum+= filter[i*filterSize + j];
1336
        }
1337
        scale/= sum;
1338
        for (j=0; j<*outFilterSize; j++)
1339
        {
1340
            double v= filter[i*filterSize + j]*scale + error;
1341
            int intV= floor(v + 0.5);
1342
            (*outFilter)[i*(*outFilterSize) + j]= intV;
1343
            error = v - intV;
1344
        }
1345
    }
1346

    
1347
    (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
1348
    for (i=0; i<*outFilterSize; i++)
1349
    {
1350
        int j= dstW*(*outFilterSize);
1351
        (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
1352
    }
1353

    
1354
    av_free(filter);
1355
    return 0;
1356
}
1357

    
1358
#ifdef COMPILE_MMX2
1359
static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
1360
{
1361
    uint8_t *fragmentA;
1362
    long imm8OfPShufW1A;
1363
    long imm8OfPShufW2A;
1364
    long fragmentLengthA;
1365
    uint8_t *fragmentB;
1366
    long imm8OfPShufW1B;
1367
    long imm8OfPShufW2B;
1368
    long fragmentLengthB;
1369
    int fragmentPos;
1370

    
1371
    int xpos, i;
1372

    
1373
    // create an optimized horizontal scaling routine
1374

    
1375
    //code fragment
1376

    
1377
    asm volatile(
1378
        "jmp                         9f                 \n\t"
1379
    // Begin
1380
        "0:                                             \n\t"
1381
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
1382
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
1383
        "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
1384
        "punpcklbw                %%mm7, %%mm1          \n\t"
1385
        "punpcklbw                %%mm7, %%mm0          \n\t"
1386
        "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
1387
        "1:                                             \n\t"
1388
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
1389
        "2:                                             \n\t"
1390
        "psubw                    %%mm1, %%mm0          \n\t"
1391
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
1392
        "pmullw                   %%mm3, %%mm0          \n\t"
1393
        "psllw                       $7, %%mm1          \n\t"
1394
        "paddw                    %%mm1, %%mm0          \n\t"
1395

    
1396
        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1397

    
1398
        "add                         $8, %%"REG_a"      \n\t"
1399
    // End
1400
        "9:                                             \n\t"
1401
//        "int $3                                         \n\t"
1402
        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
1403
        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
1404
        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
1405
        "dec                         %1                 \n\t"
1406
        "dec                         %2                 \n\t"
1407
        "sub                         %0, %1             \n\t"
1408
        "sub                         %0, %2             \n\t"
1409
        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
1410
        "sub                         %0, %3             \n\t"
1411

    
1412

    
1413
        :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
1414
        "=r" (fragmentLengthA)
1415
    );
1416

    
1417
    asm volatile(
1418
        "jmp                         9f                 \n\t"
1419
    // Begin
1420
        "0:                                             \n\t"
1421
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
1422
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
1423
        "punpcklbw                %%mm7, %%mm0          \n\t"
1424
        "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
1425
        "1:                                             \n\t"
1426
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
1427
        "2:                                             \n\t"
1428
        "psubw                    %%mm1, %%mm0          \n\t"
1429
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
1430
        "pmullw                   %%mm3, %%mm0          \n\t"
1431
        "psllw                       $7, %%mm1          \n\t"
1432
        "paddw                    %%mm1, %%mm0          \n\t"
1433

    
1434
        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
1435

    
1436
        "add                         $8, %%"REG_a"      \n\t"
1437
    // End
1438
        "9:                                             \n\t"
1439
//        "int                       $3                   \n\t"
1440
        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
1441
        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
1442
        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
1443
        "dec                         %1                 \n\t"
1444
        "dec                         %2                 \n\t"
1445
        "sub                         %0, %1             \n\t"
1446
        "sub                         %0, %2             \n\t"
1447
        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
1448
        "sub                         %0, %3             \n\t"
1449

    
1450

    
1451
        :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
1452
        "=r" (fragmentLengthB)
1453
    );
1454

    
1455
    xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
1456
    fragmentPos=0;
1457

    
1458
    for (i=0; i<dstW/numSplits; i++)
1459
    {
1460
        int xx=xpos>>16;
1461

    
1462
        if ((i&3) == 0)
1463
        {
1464
            int a=0;
1465
            int b=((xpos+xInc)>>16) - xx;
1466
            int c=((xpos+xInc*2)>>16) - xx;
1467
            int d=((xpos+xInc*3)>>16) - xx;
1468

    
1469
            filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
1470
            filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
1471
            filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
1472
            filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
1473
            filterPos[i/2]= xx;
1474

    
1475
            if (d+1<4)
1476
            {
1477
                int maxShift= 3-(d+1);
1478
                int shift=0;
1479

    
1480
                memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);
1481

    
1482
                funnyCode[fragmentPos + imm8OfPShufW1B]=
1483
                    (a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
1484
                funnyCode[fragmentPos + imm8OfPShufW2B]=
1485
                    a | (b<<2) | (c<<4) | (d<<6);
1486

    
1487
                if (i+3>=dstW) shift=maxShift; //avoid overread
1488
                else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
1489

    
1490
                if (shift && i>=shift)
1491
                {
1492
                    funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
1493
                    funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
1494
                    filterPos[i/2]-=shift;
1495
                }
1496

    
1497
                fragmentPos+= fragmentLengthB;
1498
            }
1499
            else
1500
            {
1501
                int maxShift= 3-d;
1502
                int shift=0;
1503

    
1504
                memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);
1505

    
1506
                funnyCode[fragmentPos + imm8OfPShufW1A]=
1507
                funnyCode[fragmentPos + imm8OfPShufW2A]=
1508
                    a | (b<<2) | (c<<4) | (d<<6);
1509

    
1510
                if (i+4>=dstW) shift=maxShift; //avoid overread
1511
                else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align
1512

    
1513
                if (shift && i>=shift)
1514
                {
1515
                    funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
1516
                    funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
1517
                    filterPos[i/2]-=shift;
1518
                }
1519

    
1520
                fragmentPos+= fragmentLengthA;
1521
            }
1522

    
1523
            funnyCode[fragmentPos]= RET;
1524
        }
1525
        xpos+=xInc;
1526
    }
1527
    filterPos[i/2]= xpos>>16; // needed to jump to the next part
1528
}
1529
#endif /* COMPILE_MMX2 */
1530

    
1531
static void globalInit(void){
1532
    // generating tables:
1533
    int i;
1534
    for (i=0; i<768; i++){
1535
        int c= av_clip_uint8(i-256);
1536
        clip_table[i]=c;
1537
    }
1538
}
1539

    
1540
static SwsFunc getSwsFunc(int flags){
1541

    
1542
#if defined(RUNTIME_CPUDETECT) && defined (CONFIG_GPL)
1543
#if defined(ARCH_X86)
1544
    // ordered per speed fastest first
1545
    if (flags & SWS_CPU_CAPS_MMX2)
1546
        return swScale_MMX2;
1547
    else if (flags & SWS_CPU_CAPS_3DNOW)
1548
        return swScale_3DNow;
1549
    else if (flags & SWS_CPU_CAPS_MMX)
1550
        return swScale_MMX;
1551
    else
1552
        return swScale_C;
1553

    
1554
#else
1555
#ifdef ARCH_POWERPC
1556
    if (flags & SWS_CPU_CAPS_ALTIVEC)
1557
        return swScale_altivec;
1558
    else
1559
        return swScale_C;
1560
#endif
1561
    return swScale_C;
1562
#endif /* defined(ARCH_X86) */
1563
#else //RUNTIME_CPUDETECT
1564
#ifdef HAVE_MMX2
1565
    return swScale_MMX2;
1566
#elif defined (HAVE_3DNOW)
1567
    return swScale_3DNow;
1568
#elif defined (HAVE_MMX)
1569
    return swScale_MMX;
1570
#elif defined (HAVE_ALTIVEC)
1571
    return swScale_altivec;
1572
#else
1573
    return swScale_C;
1574
#endif
1575
#endif //!RUNTIME_CPUDETECT
1576
}
1577

    
1578
static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1579
                               int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1580
    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1581
    /* Copy Y plane */
1582
    if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1583
        memcpy(dst, src[0], srcSliceH*dstStride[0]);
1584
    else
1585
    {
1586
        int i;
1587
        uint8_t *srcPtr= src[0];
1588
        uint8_t *dstPtr= dst;
1589
        for (i=0; i<srcSliceH; i++)
1590
        {
1591
            memcpy(dstPtr, srcPtr, c->srcW);
1592
            srcPtr+= srcStride[0];
1593
            dstPtr+= dstStride[0];
1594
        }
1595
    }
1596
    dst = dstParam[1] + dstStride[1]*srcSliceY/2;
1597
    if (c->dstFormat == PIX_FMT_NV12)
1598
        interleaveBytes(src[1], src[2], dst, c->srcW/2, srcSliceH/2, srcStride[1], srcStride[2], dstStride[0]);
1599
    else
1600
        interleaveBytes(src[2], src[1], dst, c->srcW/2, srcSliceH/2, srcStride[2], srcStride[1], dstStride[0]);
1601

    
1602
    return srcSliceH;
1603
}
1604

    
1605
static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1606
                               int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1607
    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1608

    
1609
    yv12toyuy2(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1610

    
1611
    return srcSliceH;
1612
}
1613

    
1614
static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1615
                               int srcSliceH, uint8_t* dstParam[], int dstStride[]){
1616
    uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
1617

    
1618
    yv12touyvy(src[0], src[1], src[2], dst, c->srcW, srcSliceH, srcStride[0], srcStride[1], dstStride[0]);
1619

    
1620
    return srcSliceH;
1621
}
1622

    
1623
/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
1624
static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1625
                          int srcSliceH, uint8_t* dst[], int dstStride[]){
1626
    const int srcFormat= c->srcFormat;
1627
    const int dstFormat= c->dstFormat;
1628
    const int srcBpp= (fmt_depth(srcFormat) + 7) >> 3;
1629
    const int dstBpp= (fmt_depth(dstFormat) + 7) >> 3;
1630
    const int srcId= fmt_depth(srcFormat) >> 2; /* 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8 */
1631
    const int dstId= fmt_depth(dstFormat) >> 2;
1632
    void (*conv)(const uint8_t *src, uint8_t *dst, long src_size)=NULL;
1633

    
1634
    /* BGR -> BGR */
1635
    if (  (isBGR(srcFormat) && isBGR(dstFormat))
1636
       || (isRGB(srcFormat) && isRGB(dstFormat))){
1637
        switch(srcId | (dstId<<4)){
1638
        case 0x34: conv= rgb16to15; break;
1639
        case 0x36: conv= rgb24to15; break;
1640
        case 0x38: conv= rgb32to15; break;
1641
        case 0x43: conv= rgb15to16; break;
1642
        case 0x46: conv= rgb24to16; break;
1643
        case 0x48: conv= rgb32to16; break;
1644
        case 0x63: conv= rgb15to24; break;
1645
        case 0x64: conv= rgb16to24; break;
1646
        case 0x68: conv= rgb32to24; break;
1647
        case 0x83: conv= rgb15to32; break;
1648
        case 0x84: conv= rgb16to32; break;
1649
        case 0x86: conv= rgb24to32; break;
1650
        default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1651
                        sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1652
        }
1653
    }else if (  (isBGR(srcFormat) && isRGB(dstFormat))
1654
             || (isRGB(srcFormat) && isBGR(dstFormat))){
1655
        switch(srcId | (dstId<<4)){
1656
        case 0x33: conv= rgb15tobgr15; break;
1657
        case 0x34: conv= rgb16tobgr15; break;
1658
        case 0x36: conv= rgb24tobgr15; break;
1659
        case 0x38: conv= rgb32tobgr15; break;
1660
        case 0x43: conv= rgb15tobgr16; break;
1661
        case 0x44: conv= rgb16tobgr16; break;
1662
        case 0x46: conv= rgb24tobgr16; break;
1663
        case 0x48: conv= rgb32tobgr16; break;
1664
        case 0x63: conv= rgb15tobgr24; break;
1665
        case 0x64: conv= rgb16tobgr24; break;
1666
        case 0x66: conv= rgb24tobgr24; break;
1667
        case 0x68: conv= rgb32tobgr24; break;
1668
        case 0x83: conv= rgb15tobgr32; break;
1669
        case 0x84: conv= rgb16tobgr32; break;
1670
        case 0x86: conv= rgb24tobgr32; break;
1671
        case 0x88: conv= rgb32tobgr32; break;
1672
        default: av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1673
                        sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
1674
        }
1675
    }else{
1676
        av_log(c, AV_LOG_ERROR, "internal error %s -> %s converter\n",
1677
               sws_format_name(srcFormat), sws_format_name(dstFormat));
1678
    }
1679

    
1680
    if(conv)
1681
    {
1682
        if (dstStride[0]*srcBpp == srcStride[0]*dstBpp && srcStride[0] > 0)
1683
            conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
1684
        else
1685
        {
1686
            int i;
1687
            uint8_t *srcPtr= src[0];
1688
            uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1689

    
1690
            for (i=0; i<srcSliceH; i++)
1691
            {
1692
                conv(srcPtr, dstPtr, c->srcW*srcBpp);
1693
                srcPtr+= srcStride[0];
1694
                dstPtr+= dstStride[0];
1695
            }
1696
        }
1697
    }
1698
    return srcSliceH;
1699
}
1700

    
1701
static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1702
                              int srcSliceH, uint8_t* dst[], int dstStride[]){
1703

    
1704
    rgb24toyv12(
1705
        src[0],
1706
        dst[0]+ srcSliceY    *dstStride[0],
1707
        dst[1]+(srcSliceY>>1)*dstStride[1],
1708
        dst[2]+(srcSliceY>>1)*dstStride[2],
1709
        c->srcW, srcSliceH,
1710
        dstStride[0], dstStride[1], srcStride[0]);
1711
    return srcSliceH;
1712
}
1713

    
1714
static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1715
                             int srcSliceH, uint8_t* dst[], int dstStride[]){
1716
    int i;
1717

    
1718
    /* copy Y */
1719
    if (srcStride[0]==dstStride[0] && srcStride[0] > 0)
1720
        memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
1721
    else{
1722
        uint8_t *srcPtr= src[0];
1723
        uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1724

    
1725
        for (i=0; i<srcSliceH; i++)
1726
        {
1727
            memcpy(dstPtr, srcPtr, c->srcW);
1728
            srcPtr+= srcStride[0];
1729
            dstPtr+= dstStride[0];
1730
        }
1731
    }
1732

    
1733
    if (c->dstFormat==PIX_FMT_YUV420P){
1734
        planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
1735
        planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
1736
    }else{
1737
        planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
1738
        planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
1739
    }
1740
    return srcSliceH;
1741
}
1742

    
1743
/* unscaled copy like stuff (assumes nearly identical formats) */
1744
static int simpleCopy(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1745
                      int srcSliceH, uint8_t* dst[], int dstStride[]){
1746

    
1747
    if (isPacked(c->srcFormat))
1748
    {
1749
        if (dstStride[0]==srcStride[0] && srcStride[0] > 0)
1750
            memcpy(dst[0] + dstStride[0]*srcSliceY, src[0], srcSliceH*dstStride[0]);
1751
        else
1752
        {
1753
            int i;
1754
            uint8_t *srcPtr= src[0];
1755
            uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;
1756
            int length=0;
1757

    
1758
            /* universal length finder */
1759
            while(length+c->srcW <= FFABS(dstStride[0])
1760
               && length+c->srcW <= FFABS(srcStride[0])) length+= c->srcW;
1761
            ASSERT(length!=0);
1762

    
1763
            for (i=0; i<srcSliceH; i++)
1764
            {
1765
                memcpy(dstPtr, srcPtr, length);
1766
                srcPtr+= srcStride[0];
1767
                dstPtr+= dstStride[0];
1768
            }
1769
        }
1770
    }
1771
    else
1772
    { /* Planar YUV or gray */
1773
        int plane;
1774
        for (plane=0; plane<3; plane++)
1775
        {
1776
            int length= plane==0 ? c->srcW  : -((-c->srcW  )>>c->chrDstHSubSample);
1777
            int y=      plane==0 ? srcSliceY: -((-srcSliceY)>>c->chrDstVSubSample);
1778
            int height= plane==0 ? srcSliceH: -((-srcSliceH)>>c->chrDstVSubSample);
1779

    
1780
            if ((isGray(c->srcFormat) || isGray(c->dstFormat)) && plane>0)
1781
            {
1782
                if (!isGray(c->dstFormat))
1783
                    memset(dst[plane], 128, dstStride[plane]*height);
1784
            }
1785
            else
1786
            {
1787
                if (dstStride[plane]==srcStride[plane] && srcStride[plane] > 0)
1788
                    memcpy(dst[plane] + dstStride[plane]*y, src[plane], height*dstStride[plane]);
1789
                else
1790
                {
1791
                    int i;
1792
                    uint8_t *srcPtr= src[plane];
1793
                    uint8_t *dstPtr= dst[plane] + dstStride[plane]*y;
1794
                    for (i=0; i<height; i++)
1795
                    {
1796
                        memcpy(dstPtr, srcPtr, length);
1797
                        srcPtr+= srcStride[plane];
1798
                        dstPtr+= dstStride[plane];
1799
                    }
1800
                }
1801
            }
1802
        }
1803
    }
1804
    return srcSliceH;
1805
}
1806

    
1807
static int gray16togray(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1808
                        int srcSliceH, uint8_t* dst[], int dstStride[]){
1809

    
1810
    int length= c->srcW;
1811
    int y=      srcSliceY;
1812
    int height= srcSliceH;
1813
    int i, j;
1814
    uint8_t *srcPtr= src[0];
1815
    uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1816

    
1817
    if (!isGray(c->dstFormat)){
1818
        int height= -((-srcSliceH)>>c->chrDstVSubSample);
1819
        memset(dst[1], 128, dstStride[1]*height);
1820
        memset(dst[2], 128, dstStride[2]*height);
1821
    }
1822
    if (c->srcFormat == PIX_FMT_GRAY16LE) srcPtr++;
1823
    for (i=0; i<height; i++)
1824
    {
1825
        for (j=0; j<length; j++) dstPtr[j] = srcPtr[j<<1];
1826
        srcPtr+= srcStride[0];
1827
        dstPtr+= dstStride[0];
1828
    }
1829
    return srcSliceH;
1830
}
1831

    
1832
static int graytogray16(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1833
                        int srcSliceH, uint8_t* dst[], int dstStride[]){
1834

    
1835
    int length= c->srcW;
1836
    int y=      srcSliceY;
1837
    int height= srcSliceH;
1838
    int i, j;
1839
    uint8_t *srcPtr= src[0];
1840
    uint8_t *dstPtr= dst[0] + dstStride[0]*y;
1841
    for (i=0; i<height; i++)
1842
    {
1843
        for (j=0; j<length; j++)
1844
        {
1845
            dstPtr[j<<1] = srcPtr[j];
1846
            dstPtr[(j<<1)+1] = srcPtr[j];
1847
        }
1848
        srcPtr+= srcStride[0];
1849
        dstPtr+= dstStride[0];
1850
    }
1851
    return srcSliceH;
1852
}
1853

    
1854
static int gray16swap(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
1855
                      int srcSliceH, uint8_t* dst[], int dstStride[]){
1856

    
1857
    int length= c->srcW;
1858
    int y=      srcSliceY;
1859
    int height= srcSliceH;
1860
    int i, j;
1861
    uint16_t *srcPtr= src[0];
1862
    uint16_t *dstPtr= dst[0] + dstStride[0]*y/2;
1863
    for (i=0; i<height; i++)
1864
    {
1865
        for (j=0; j<length; j++) dstPtr[j] = bswap_16(srcPtr[j]);
1866
        srcPtr+= srcStride[0]/2;
1867
        dstPtr+= dstStride[0]/2;
1868
    }
1869
    return srcSliceH;
1870
}
1871

    
1872

    
1873
static void getSubSampleFactors(int *h, int *v, int format){
1874
    switch(format){
1875
    case PIX_FMT_UYVY422:
1876
    case PIX_FMT_YUYV422:
1877
        *h=1;
1878
        *v=0;
1879
        break;
1880
    case PIX_FMT_YUV420P:
1881
    case PIX_FMT_YUVA420P:
1882
    case PIX_FMT_GRAY16BE:
1883
    case PIX_FMT_GRAY16LE:
1884
    case PIX_FMT_GRAY8: //FIXME remove after different subsamplings are fully implemented
1885
    case PIX_FMT_NV12:
1886
    case PIX_FMT_NV21:
1887
        *h=1;
1888
        *v=1;
1889
        break;
1890
    case PIX_FMT_YUV440P:
1891
        *h=0;
1892
        *v=1;
1893
        break;
1894
    case PIX_FMT_YUV410P:
1895
        *h=2;
1896
        *v=2;
1897
        break;
1898
    case PIX_FMT_YUV444P:
1899
        *h=0;
1900
        *v=0;
1901
        break;
1902
    case PIX_FMT_YUV422P:
1903
        *h=1;
1904
        *v=0;
1905
        break;
1906
    case PIX_FMT_YUV411P:
1907
        *h=2;
1908
        *v=0;
1909
        break;
1910
    default:
1911
        *h=0;
1912
        *v=0;
1913
        break;
1914
    }
1915
}
1916

    
1917
static uint16_t roundToInt16(int64_t f){
1918
    int r= (f + (1<<15))>>16;
1919
         if (r<-0x7FFF) return 0x8000;
1920
    else if (r> 0x7FFF) return 0x7FFF;
1921
    else                return r;
1922
}
1923

    
1924
/**
1925
 * @param inv_table the yuv2rgb coeffs, normally Inverse_Table_6_9[x]
1926
 * @param fullRange if 1 then the luma range is 0..255 if 0 it is 16..235
1927
 * @return -1 if not supported
1928
 */
1929
int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation){
1930
    int64_t crv =  inv_table[0];
1931
    int64_t cbu =  inv_table[1];
1932
    int64_t cgu = -inv_table[2];
1933
    int64_t cgv = -inv_table[3];
1934
    int64_t cy  = 1<<16;
1935
    int64_t oy  = 0;
1936

    
1937
    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1938
    memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
1939
    memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
1940

    
1941
    c->brightness= brightness;
1942
    c->contrast  = contrast;
1943
    c->saturation= saturation;
1944
    c->srcRange  = srcRange;
1945
    c->dstRange  = dstRange;
1946

    
1947
    c->uOffset=   0x0400040004000400LL;
1948
    c->vOffset=   0x0400040004000400LL;
1949

    
1950
    if (!srcRange){
1951
        cy= (cy*255) / 219;
1952
        oy= 16<<16;
1953
    }else{
1954
        crv= (crv*224) / 255;
1955
        cbu= (cbu*224) / 255;
1956
        cgu= (cgu*224) / 255;
1957
        cgv= (cgv*224) / 255;
1958
    }
1959

    
1960
    cy = (cy *contrast             )>>16;
1961
    crv= (crv*contrast * saturation)>>32;
1962
    cbu= (cbu*contrast * saturation)>>32;
1963
    cgu= (cgu*contrast * saturation)>>32;
1964
    cgv= (cgv*contrast * saturation)>>32;
1965

    
1966
    oy -= 256*brightness;
1967

    
1968
    c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
1969
    c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
1970
    c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
1971
    c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
1972
    c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
1973
    c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
1974

    
1975
    yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
1976
    //FIXME factorize
1977

    
1978
#ifdef COMPILE_ALTIVEC
1979
    if (c->flags & SWS_CPU_CAPS_ALTIVEC)
1980
        yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
1981
#endif
1982
    return 0;
1983
}
1984

    
1985
/**
1986
 * @return -1 if not supported
1987
 */
1988
int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation){
1989
    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
1990

    
1991
    *inv_table = c->srcColorspaceTable;
1992
    *table     = c->dstColorspaceTable;
1993
    *srcRange  = c->srcRange;
1994
    *dstRange  = c->dstRange;
1995
    *brightness= c->brightness;
1996
    *contrast  = c->contrast;
1997
    *saturation= c->saturation;
1998

    
1999
    return 0;
2000
}
2001

    
2002
static int handle_jpeg(int *format)
2003
{
2004
    switch (*format) {
2005
        case PIX_FMT_YUVJ420P:
2006
            *format = PIX_FMT_YUV420P;
2007
            return 1;
2008
        case PIX_FMT_YUVJ422P:
2009
            *format = PIX_FMT_YUV422P;
2010
            return 1;
2011
        case PIX_FMT_YUVJ444P:
2012
            *format = PIX_FMT_YUV444P;
2013
            return 1;
2014
        case PIX_FMT_YUVJ440P:
2015
            *format = PIX_FMT_YUV440P;
2016
            return 1;
2017
        default:
2018
            return 0;
2019
    }
2020
}
2021

    
2022
SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
2023
                           SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
2024

    
2025
    SwsContext *c;
2026
    int i;
2027
    int usesVFilter, usesHFilter;
2028
    int unscaled, needsDither;
2029
    int srcRange, dstRange;
2030
    SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
2031
#if defined(ARCH_X86)
2032
    if (flags & SWS_CPU_CAPS_MMX)
2033
        asm volatile("emms\n\t"::: "memory");
2034
#endif
2035

    
2036
#if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
2037
    flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
2038
#ifdef HAVE_MMX2
2039
    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
2040
#elif defined (HAVE_3DNOW)
2041
    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
2042
#elif defined (HAVE_MMX)
2043
    flags |= SWS_CPU_CAPS_MMX;
2044
#elif defined (HAVE_ALTIVEC)
2045
    flags |= SWS_CPU_CAPS_ALTIVEC;
2046
#elif defined (ARCH_BFIN)
2047
    flags |= SWS_CPU_CAPS_BFIN;
2048
#endif
2049
#endif /* RUNTIME_CPUDETECT */
2050
    if (clip_table[512] != 255) globalInit();
2051
    if (!rgb15to16) sws_rgb2rgb_init(flags);
2052

    
2053
    unscaled = (srcW == dstW && srcH == dstH);
2054
    needsDither= (isBGR(dstFormat) || isRGB(dstFormat))
2055
        && (fmt_depth(dstFormat))<24
2056
        && ((fmt_depth(dstFormat))<(fmt_depth(srcFormat)) || (!(isRGB(srcFormat) || isBGR(srcFormat))));
2057

    
2058
    srcRange = handle_jpeg(&srcFormat);
2059
    dstRange = handle_jpeg(&dstFormat);
2060

    
2061
    if (!isSupportedIn(srcFormat))
2062
    {
2063
        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
2064
        return NULL;
2065
    }
2066
    if (!isSupportedOut(dstFormat))
2067
    {
2068
        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
2069
        return NULL;
2070
    }
2071

    
2072
    /* sanity check */
2073
    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
2074
    {
2075
        av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
2076
               srcW, srcH, dstW, dstH);
2077
        return NULL;
2078
    }
2079
    if(srcW > VOFW || dstW > VOFW){
2080
        av_log(NULL, AV_LOG_ERROR, "swScaler: Compile time max width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
2081
        return NULL;
2082
    }
2083

    
2084
    if (!dstFilter) dstFilter= &dummyFilter;
2085
    if (!srcFilter) srcFilter= &dummyFilter;
2086

    
2087
    c= av_mallocz(sizeof(SwsContext));
2088

    
2089
    c->av_class = &sws_context_class;
2090
    c->srcW= srcW;
2091
    c->srcH= srcH;
2092
    c->dstW= dstW;
2093
    c->dstH= dstH;
2094
    c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
2095
    c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
2096
    c->flags= flags;
2097
    c->dstFormat= dstFormat;
2098
    c->srcFormat= srcFormat;
2099
    c->vRounder= 4* 0x0001000100010001ULL;
2100

    
2101
    usesHFilter= usesVFilter= 0;
2102
    if (dstFilter->lumV && dstFilter->lumV->length>1) usesVFilter=1;
2103
    if (dstFilter->lumH && dstFilter->lumH->length>1) usesHFilter=1;
2104
    if (dstFilter->chrV && dstFilter->chrV->length>1) usesVFilter=1;
2105
    if (dstFilter->chrH && dstFilter->chrH->length>1) usesHFilter=1;
2106
    if (srcFilter->lumV && srcFilter->lumV->length>1) usesVFilter=1;
2107
    if (srcFilter->lumH && srcFilter->lumH->length>1) usesHFilter=1;
2108
    if (srcFilter->chrV && srcFilter->chrV->length>1) usesVFilter=1;
2109
    if (srcFilter->chrH && srcFilter->chrH->length>1) usesHFilter=1;
2110

    
2111
    getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
2112
    getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
2113

    
2114
    // reuse chroma for 2 pixles rgb/bgr unless user wants full chroma interpolation
2115
    if ((isBGR(dstFormat) || isRGB(dstFormat)) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
2116

    
2117
    // drop some chroma lines if the user wants it
2118
    c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
2119
    c->chrSrcVSubSample+= c->vChrDrop;
2120

    
2121
    // drop every 2. pixel for chroma calculation unless user wants full chroma
2122
    if ((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP)
2123
      && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
2124
      && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
2125
      && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE)
2126
        c->chrSrcHSubSample=1;
2127

    
2128
    if (param){
2129
        c->param[0] = param[0];
2130
        c->param[1] = param[1];
2131
    }else{
2132
        c->param[0] =
2133
        c->param[1] = SWS_PARAM_DEFAULT;
2134
    }
2135

    
2136
    c->chrIntHSubSample= c->chrDstHSubSample;
2137
    c->chrIntVSubSample= c->chrSrcVSubSample;
2138

    
2139
    // Note the -((-x)>>y) is so that we always round toward +inf.
2140
    c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
2141
    c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
2142
    c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
2143
    c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
2144

    
2145
    sws_setColorspaceDetails(c, Inverse_Table_6_9[SWS_CS_DEFAULT], srcRange, Inverse_Table_6_9[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
2146

    
2147
    /* unscaled special Cases */
2148
    if (unscaled && !usesHFilter && !usesVFilter)
2149
    {
2150
        /* yv12_to_nv12 */
2151
        if (srcFormat == PIX_FMT_YUV420P && (dstFormat == PIX_FMT_NV12 || dstFormat == PIX_FMT_NV21))
2152
        {
2153
            c->swScale= PlanarToNV12Wrapper;
2154
        }
2155
#ifdef CONFIG_GPL
2156
        /* yuv2bgr */
2157
        if ((srcFormat==PIX_FMT_YUV420P || srcFormat==PIX_FMT_YUV422P) && (isBGR(dstFormat) || isRGB(dstFormat)))
2158
        {
2159
            c->swScale= yuv2rgb_get_func_ptr(c);
2160
        }
2161
#endif
2162

    
2163
        if (srcFormat==PIX_FMT_YUV410P && dstFormat==PIX_FMT_YUV420P)
2164
        {
2165
            c->swScale= yvu9toyv12Wrapper;
2166
        }
2167

    
2168
        /* bgr24toYV12 */
2169
        if (srcFormat==PIX_FMT_BGR24 && dstFormat==PIX_FMT_YUV420P)
2170
            c->swScale= bgr24toyv12Wrapper;
2171

    
2172
        /* rgb/bgr -> rgb/bgr (no dither needed forms) */
2173
        if (  (isBGR(srcFormat) || isRGB(srcFormat))
2174
           && (isBGR(dstFormat) || isRGB(dstFormat))
2175
           && srcFormat != PIX_FMT_BGR8      && dstFormat != PIX_FMT_BGR8
2176
           && srcFormat != PIX_FMT_RGB8      && dstFormat != PIX_FMT_RGB8
2177
           && srcFormat != PIX_FMT_BGR4      && dstFormat != PIX_FMT_BGR4
2178
           && srcFormat != PIX_FMT_RGB4      && dstFormat != PIX_FMT_RGB4
2179
           && srcFormat != PIX_FMT_BGR4_BYTE && dstFormat != PIX_FMT_BGR4_BYTE
2180
           && srcFormat != PIX_FMT_RGB4_BYTE && dstFormat != PIX_FMT_RGB4_BYTE
2181
           && srcFormat != PIX_FMT_MONOBLACK && dstFormat != PIX_FMT_MONOBLACK
2182
           && !needsDither)
2183
             c->swScale= rgb2rgbWrapper;
2184

    
2185
        /* LQ converters if -sws 0 or -sws 4*/
2186
        if (c->flags&(SWS_FAST_BILINEAR|SWS_POINT)){
2187
            /* rgb/bgr -> rgb/bgr (dither needed forms) */
2188
            if ( (isBGR(srcFormat) || isRGB(srcFormat))
2189
              && (isBGR(dstFormat) || isRGB(dstFormat))
2190
              && needsDither)
2191
                c->swScale= rgb2rgbWrapper;
2192

    
2193
            /* yv12_to_yuy2 */
2194
            if (srcFormat == PIX_FMT_YUV420P &&
2195
                (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422))
2196
            {
2197
                if (dstFormat == PIX_FMT_YUYV422)
2198
                    c->swScale= PlanarToYuy2Wrapper;
2199
                else
2200
                    c->swScale= PlanarToUyvyWrapper;
2201
            }
2202
        }
2203

    
2204
#ifdef COMPILE_ALTIVEC
2205
        if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
2206
            ((srcFormat == PIX_FMT_YUV420P &&
2207
             (dstFormat == PIX_FMT_YUYV422 || dstFormat == PIX_FMT_UYVY422)))) {
2208
          // unscaled YV12 -> packed YUV, we want speed
2209
          if (dstFormat == PIX_FMT_YUYV422)
2210
              c->swScale= yv12toyuy2_unscaled_altivec;
2211
          else
2212
              c->swScale= yv12touyvy_unscaled_altivec;
2213
        }
2214
#endif
2215

    
2216
        /* simple copy */
2217
        if (  srcFormat == dstFormat
2218
            || (isPlanarYUV(srcFormat) && isGray(dstFormat))
2219
            || (isPlanarYUV(dstFormat) && isGray(srcFormat)))
2220
        {
2221
            c->swScale= simpleCopy;
2222
        }
2223

    
2224
        /* gray16{le,be} conversions */
2225
        if (isGray16(srcFormat) && (isPlanarYUV(dstFormat) || (dstFormat == PIX_FMT_GRAY8)))
2226
        {
2227
            c->swScale= gray16togray;
2228
        }
2229
        if ((isPlanarYUV(srcFormat) || (srcFormat == PIX_FMT_GRAY8)) && isGray16(dstFormat))
2230
        {
2231
            c->swScale= graytogray16;
2232
        }
2233
        if (srcFormat != dstFormat && isGray16(srcFormat) && isGray16(dstFormat))
2234
        {
2235
            c->swScale= gray16swap;
2236
        }
2237

    
2238
#ifdef ARCH_BFIN
2239
        if (flags & SWS_CPU_CAPS_BFIN)
2240
            ff_bfin_get_unscaled_swscale (c);
2241
#endif
2242

    
2243
        if (c->swScale){
2244
            if (flags&SWS_PRINT_INFO)
2245
                av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
2246
                                sws_format_name(srcFormat), sws_format_name(dstFormat));
2247
            return c;
2248
        }
2249
    }
2250

    
2251
    if (flags & SWS_CPU_CAPS_MMX2)
2252
    {
2253
        c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
2254
        if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
2255
        {
2256
            if (flags&SWS_PRINT_INFO)
2257
                av_log(c, AV_LOG_INFO, "output Width is not a multiple of 32 -> no MMX2 scaler\n");
2258
        }
2259
        if (usesHFilter) c->canMMX2BeUsed=0;
2260
    }
2261
    else
2262
        c->canMMX2BeUsed=0;
2263

    
2264
    c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
2265
    c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
2266

    
2267
    // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
2268
    // but only for the FAST_BILINEAR mode otherwise do correct scaling
2269
    // n-2 is the last chrominance sample available
2270
    // this is not perfect, but no one should notice the difference, the more correct variant
2271
    // would be like the vertical one, but that would require some special code for the
2272
    // first and last pixel
2273
    if (flags&SWS_FAST_BILINEAR)
2274
    {
2275
        if (c->canMMX2BeUsed)
2276
        {
2277
            c->lumXInc+= 20;
2278
            c->chrXInc+= 20;
2279
        }
2280
        //we don't use the x86asm scaler if mmx is available
2281
        else if (flags & SWS_CPU_CAPS_MMX)
2282
        {
2283
            c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
2284
            c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
2285
        }
2286
    }
2287

    
2288
    /* precalculate horizontal scaler filter coefficients */
2289
    {
2290
        const int filterAlign=
2291
            (flags & SWS_CPU_CAPS_MMX) ? 4 :
2292
            (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2293
            1;
2294

    
2295
        initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
2296
                   srcW      ,       dstW, filterAlign, 1<<14,
2297
                   (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2298
                   srcFilter->lumH, dstFilter->lumH, c->param);
2299
        initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
2300
                   c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
2301
                   (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2302
                   srcFilter->chrH, dstFilter->chrH, c->param);
2303

    
2304
#define MAX_FUNNY_CODE_SIZE 10000
2305
#if defined(COMPILE_MMX2)
2306
// can't downscale !!!
2307
        if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
2308
        {
2309
#ifdef MAP_ANONYMOUS
2310
            c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2311
            c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
2312
#else
2313
            c->funnyYCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2314
            c->funnyUVCode = av_malloc(MAX_FUNNY_CODE_SIZE);
2315
#endif
2316

    
2317
            c->lumMmx2Filter   = av_malloc((dstW        /8+8)*sizeof(int16_t));
2318
            c->chrMmx2Filter   = av_malloc((c->chrDstW  /4+8)*sizeof(int16_t));
2319
            c->lumMmx2FilterPos= av_malloc((dstW      /2/8+8)*sizeof(int32_t));
2320
            c->chrMmx2FilterPos= av_malloc((c->chrDstW/2/4+8)*sizeof(int32_t));
2321

    
2322
            initMMX2HScaler(      dstW, c->lumXInc, c->funnyYCode , c->lumMmx2Filter, c->lumMmx2FilterPos, 8);
2323
            initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode, c->chrMmx2Filter, c->chrMmx2FilterPos, 4);
2324
        }
2325
#endif /* defined(COMPILE_MMX2) */
2326
    } // Init Horizontal stuff
2327

    
2328

    
2329

    
2330
    /* precalculate vertical scaler filter coefficients */
2331
    {
2332
        const int filterAlign=
2333
            (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
2334
            (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
2335
            1;
2336

    
2337
        initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
2338
                   srcH      ,        dstH, filterAlign, (1<<12)-4,
2339
                   (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
2340
                   srcFilter->lumV, dstFilter->lumV, c->param);
2341
        initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
2342
                   c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
2343
                   (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
2344
                   srcFilter->chrV, dstFilter->chrV, c->param);
2345

    
2346
#ifdef HAVE_ALTIVEC
2347
        c->vYCoeffsBank = av_malloc(sizeof (vector signed short)*c->vLumFilterSize*c->dstH);
2348
        c->vCCoeffsBank = av_malloc(sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH);
2349

    
2350
        for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
2351
            int j;
2352
            short *p = (short *)&c->vYCoeffsBank[i];
2353
            for (j=0;j<8;j++)
2354
                p[j] = c->vLumFilter[i];
2355
        }
2356

    
2357
        for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
2358
            int j;
2359
            short *p = (short *)&c->vCCoeffsBank[i];
2360
            for (j=0;j<8;j++)
2361
                p[j] = c->vChrFilter[i];
2362
        }
2363
#endif
2364
    }
2365

    
2366
    // Calculate Buffer Sizes so that they won't run out while handling these damn slices
2367
    c->vLumBufSize= c->vLumFilterSize;
2368
    c->vChrBufSize= c->vChrFilterSize;
2369
    for (i=0; i<dstH; i++)
2370
    {
2371
        int chrI= i*c->chrDstH / dstH;
2372
        int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
2373
                           ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
2374

    
2375
        nextSlice>>= c->chrSrcVSubSample;
2376
        nextSlice<<= c->chrSrcVSubSample;
2377
        if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
2378
            c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
2379
        if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
2380
            c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
2381
    }
2382

    
2383
    // allocate pixbufs (we use dynamic allocation because otherwise we would need to
2384
    c->lumPixBuf= av_malloc(c->vLumBufSize*2*sizeof(int16_t*));
2385
    c->chrPixBuf= av_malloc(c->vChrBufSize*2*sizeof(int16_t*));
2386
    //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)
2387
    /* align at 16 bytes for AltiVec */
2388
    for (i=0; i<c->vLumBufSize; i++)
2389
        c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= av_mallocz(VOF+1);
2390
    for (i=0; i<c->vChrBufSize; i++)
2391
        c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= av_malloc((VOF+1)*2);
2392

    
2393
    //try to avoid drawing green stuff between the right end and the stride end
2394
    for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
2395

    
2396
    assert(2*VOFW == VOF);
2397

    
2398
    ASSERT(c->chrDstH <= dstH)
2399

    
2400
    if (flags&SWS_PRINT_INFO)
2401
    {
2402
#ifdef DITHER1XBPP
2403
        char *dither= " dithered";
2404
#else
2405
        char *dither= "";
2406
#endif
2407
        if (flags&SWS_FAST_BILINEAR)
2408
            av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
2409
        else if (flags&SWS_BILINEAR)
2410
            av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
2411
        else if (flags&SWS_BICUBIC)
2412
            av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
2413
        else if (flags&SWS_X)
2414
            av_log(c, AV_LOG_INFO, "Experimental scaler, ");
2415
        else if (flags&SWS_POINT)
2416
            av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
2417
        else if (flags&SWS_AREA)
2418
            av_log(c, AV_LOG_INFO, "Area Averageing scaler, ");
2419
        else if (flags&SWS_BICUBLIN)
2420
            av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
2421
        else if (flags&SWS_GAUSS)
2422
            av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
2423
        else if (flags&SWS_SINC)
2424
            av_log(c, AV_LOG_INFO, "Sinc scaler, ");
2425
        else if (flags&SWS_LANCZOS)
2426
            av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
2427
        else if (flags&SWS_SPLINE)
2428
            av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
2429
        else
2430
            av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
2431

    
2432
        if (dstFormat==PIX_FMT_BGR555 || dstFormat==PIX_FMT_BGR565)
2433
            av_log(c, AV_LOG_INFO, "from %s to%s %s ",
2434
                   sws_format_name(srcFormat), dither, sws_format_name(dstFormat));
2435
        else
2436
            av_log(c, AV_LOG_INFO, "from %s to %s ",
2437
                   sws_format_name(srcFormat), sws_format_name(dstFormat));
2438

    
2439
        if (flags & SWS_CPU_CAPS_MMX2)
2440
            av_log(c, AV_LOG_INFO, "using MMX2\n");
2441
        else if (flags & SWS_CPU_CAPS_3DNOW)
2442
            av_log(c, AV_LOG_INFO, "using 3DNOW\n");
2443
        else if (flags & SWS_CPU_CAPS_MMX)
2444
            av_log(c, AV_LOG_INFO, "using MMX\n");
2445
        else if (flags & SWS_CPU_CAPS_ALTIVEC)
2446
            av_log(c, AV_LOG_INFO, "using AltiVec\n");
2447
        else
2448
            av_log(c, AV_LOG_INFO, "using C\n");
2449
    }
2450

    
2451
    if (flags & SWS_PRINT_INFO)
2452
    {
2453
        if (flags & SWS_CPU_CAPS_MMX)
2454
        {
2455
            if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
2456
                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
2457
            else
2458
            {
2459
                if (c->hLumFilterSize==4)
2460
                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
2461
                else if (c->hLumFilterSize==8)
2462
                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
2463
                else
2464
                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
2465

    
2466
                if (c->hChrFilterSize==4)
2467
                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
2468
                else if (c->hChrFilterSize==8)
2469
                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
2470
                else
2471
                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
2472
            }
2473
        }
2474
        else
2475
        {
2476
#if defined(ARCH_X86)
2477
            av_log(c, AV_LOG_VERBOSE, "using X86-Asm scaler for horizontal scaling\n");
2478
#else
2479
            if (flags & SWS_FAST_BILINEAR)
2480
                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
2481
            else
2482
                av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
2483
#endif
2484
        }
2485
        if (isPlanarYUV(dstFormat))
2486
        {
2487
            if (c->vLumFilterSize==1)
2488
                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2489
            else
2490
                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2491
        }
2492
        else
2493
        {
2494
            if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
2495
                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
2496
                       "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2497
            else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
2498
                av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2499
            else
2500
                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2501
        }
2502

    
2503
        if (dstFormat==PIX_FMT_BGR24)
2504
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 Converter\n",
2505
                   (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
2506
        else if (dstFormat==PIX_FMT_RGB32)
2507
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2508
        else if (dstFormat==PIX_FMT_BGR565)
2509
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2510
        else if (dstFormat==PIX_FMT_BGR555)
2511
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 Converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
2512

    
2513
        av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
2514
    }
2515
    if (flags & SWS_PRINT_INFO)
2516
    {
2517
        av_log(c, AV_LOG_DEBUG, "Lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2518
               c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
2519
        av_log(c, AV_LOG_DEBUG, "Chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
2520
               c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
2521
    }
2522

    
2523
    c->swScale= getSwsFunc(flags);
2524
    return c;
2525
}
2526

    
2527
/**
2528
 * swscale wrapper, so we don't need to export the SwsContext.
2529
 * assumes planar YUV to be in YUV order instead of YVU
2530
 */
2531
int sws_scale(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2532
              int srcSliceH, uint8_t* dst[], int dstStride[]){
2533
    int i;
2534
    uint8_t* src2[4]= {src[0], src[1], src[2]};
2535
    uint32_t pal[256];
2536
    if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {
2537
        av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");
2538
        return 0;
2539
    }
2540
    if (c->sliceDir == 0) {
2541
        if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;
2542
    }
2543

    
2544
    if (c->srcFormat == PIX_FMT_PAL8){
2545
        for (i=0; i<256; i++){
2546
            int p= ((uint32_t*)(src[1]))[i];
2547
            int r= (p>>16)&0xFF;
2548
            int g= (p>> 8)&0xFF;
2549
            int b=  p     &0xFF;
2550
            int y= av_clip_uint8(((RY*r + GY*g + BY*b)>>RGB2YUV_SHIFT) + 16 );
2551
            int u= av_clip_uint8(((RU*r + GU*g + BU*b)>>RGB2YUV_SHIFT) + 128);
2552
            int v= av_clip_uint8(((RV*r + GV*g + BV*b)>>RGB2YUV_SHIFT) + 128);
2553
            pal[i]= y + (u<<8) + (v<<16);
2554
        }
2555
        src2[1]= pal;
2556
    }
2557

    
2558
    // copy strides, so they can safely be modified
2559
    if (c->sliceDir == 1) {
2560
        // slices go from top to bottom
2561
        int srcStride2[4]= {srcStride[0], srcStride[1], srcStride[2]};
2562
        int dstStride2[4]= {dstStride[0], dstStride[1], dstStride[2]};
2563
        return c->swScale(c, src2, srcStride2, srcSliceY, srcSliceH, dst, dstStride2);
2564
    } else {
2565
        // slices go from bottom to top => we flip the image internally
2566
        uint8_t* dst2[4]= {dst[0] + (c->dstH-1)*dstStride[0],
2567
                           dst[1] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[1],
2568
                           dst[2] + ((c->dstH>>c->chrDstVSubSample)-1)*dstStride[2]};
2569
        int srcStride2[4]= {-srcStride[0], -srcStride[1], -srcStride[2]};
2570
        int dstStride2[4]= {-dstStride[0], -dstStride[1], -dstStride[2]};
2571

    
2572
        src2[0] += (srcSliceH-1)*srcStride[0];
2573
        if (c->srcFormat != PIX_FMT_PAL8)
2574
            src2[1] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[1];
2575
        src2[2] += ((srcSliceH>>c->chrSrcVSubSample)-1)*srcStride[2];
2576

    
2577
        return c->swScale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH, srcSliceH, dst2, dstStride2);
2578
    }
2579
}
2580

    
2581
/**
2582
 * swscale wrapper, so we don't need to export the SwsContext
2583
 */
2584
int sws_scale_ordered(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
2585
                      int srcSliceH, uint8_t* dst[], int dstStride[]){
2586
    return sws_scale(c, src, srcStride, srcSliceY, srcSliceH, dst, dstStride);
2587
}
2588

    
2589
SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
2590
                                float lumaSharpen, float chromaSharpen,
2591
                                float chromaHShift, float chromaVShift,
2592
                                int verbose)
2593
{
2594
    SwsFilter *filter= av_malloc(sizeof(SwsFilter));
2595

    
2596
    if (lumaGBlur!=0.0){
2597
        filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
2598
        filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
2599
    }else{
2600
        filter->lumH= sws_getIdentityVec();
2601
        filter->lumV= sws_getIdentityVec();
2602
    }
2603

    
2604
    if (chromaGBlur!=0.0){
2605
        filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
2606
        filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
2607
    }else{
2608
        filter->chrH= sws_getIdentityVec();
2609
        filter->chrV= sws_getIdentityVec();
2610
    }
2611

    
2612
    if (chromaSharpen!=0.0){
2613
        SwsVector *id= sws_getIdentityVec();
2614
        sws_scaleVec(filter->chrH, -chromaSharpen);
2615
        sws_scaleVec(filter->chrV, -chromaSharpen);
2616
        sws_addVec(filter->chrH, id);
2617
        sws_addVec(filter->chrV, id);
2618
        sws_freeVec(id);
2619
    }
2620

    
2621
    if (lumaSharpen!=0.0){
2622
        SwsVector *id= sws_getIdentityVec();
2623
        sws_scaleVec(filter->lumH, -lumaSharpen);
2624
        sws_scaleVec(filter->lumV, -lumaSharpen);
2625
        sws_addVec(filter->lumH, id);
2626
        sws_addVec(filter->lumV, id);
2627
        sws_freeVec(id);
2628
    }
2629

    
2630
    if (chromaHShift != 0.0)
2631
        sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
2632

    
2633
    if (chromaVShift != 0.0)
2634
        sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
2635

    
2636
    sws_normalizeVec(filter->chrH, 1.0);
2637
    sws_normalizeVec(filter->chrV, 1.0);
2638
    sws_normalizeVec(filter->lumH, 1.0);
2639
    sws_normalizeVec(filter->lumV, 1.0);
2640

    
2641
    if (verbose) sws_printVec(filter->chrH);
2642
    if (verbose) sws_printVec(filter->lumH);
2643

    
2644
    return filter;
2645
}
2646

    
2647
/**
2648
 * returns a normalized gaussian curve used to filter stuff
2649
 * quality=3 is high quality, lowwer is lowwer quality
2650
 */
2651
SwsVector *sws_getGaussianVec(double variance, double quality){
2652
    const int length= (int)(variance*quality + 0.5) | 1;
2653
    int i;
2654
    double *coeff= av_malloc(length*sizeof(double));
2655
    double middle= (length-1)*0.5;
2656
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2657

    
2658
    vec->coeff= coeff;
2659
    vec->length= length;
2660

    
2661
    for (i=0; i<length; i++)
2662
    {
2663
        double dist= i-middle;
2664
        coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*PI);
2665
    }
2666

    
2667
    sws_normalizeVec(vec, 1.0);
2668

    
2669
    return vec;
2670
}
2671

    
2672
SwsVector *sws_getConstVec(double c, int length){
2673
    int i;
2674
    double *coeff= av_malloc(length*sizeof(double));
2675
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2676

    
2677
    vec->coeff= coeff;
2678
    vec->length= length;
2679

    
2680
    for (i=0; i<length; i++)
2681
        coeff[i]= c;
2682

    
2683
    return vec;
2684
}
2685

    
2686

    
2687
SwsVector *sws_getIdentityVec(void){
2688
    return sws_getConstVec(1.0, 1);
2689
}
2690

    
2691
double sws_dcVec(SwsVector *a){
2692
    int i;
2693
    double sum=0;
2694

    
2695
    for (i=0; i<a->length; i++)
2696
        sum+= a->coeff[i];
2697

    
2698
    return sum;
2699
}
2700

    
2701
void sws_scaleVec(SwsVector *a, double scalar){
2702
    int i;
2703

    
2704
    for (i=0; i<a->length; i++)
2705
        a->coeff[i]*= scalar;
2706
}
2707

    
2708
void sws_normalizeVec(SwsVector *a, double height){
2709
    sws_scaleVec(a, height/sws_dcVec(a));
2710
}
2711

    
2712
static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b){
2713
    int length= a->length + b->length - 1;
2714
    double *coeff= av_malloc(length*sizeof(double));
2715
    int i, j;
2716
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2717

    
2718
    vec->coeff= coeff;
2719
    vec->length= length;
2720

    
2721
    for (i=0; i<length; i++) coeff[i]= 0.0;
2722

    
2723
    for (i=0; i<a->length; i++)
2724
    {
2725
        for (j=0; j<b->length; j++)
2726
        {
2727
            coeff[i+j]+= a->coeff[i]*b->coeff[j];
2728
        }
2729
    }
2730

    
2731
    return vec;
2732
}
2733

    
2734
static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b){
2735
    int length= FFMAX(a->length, b->length);
2736
    double *coeff= av_malloc(length*sizeof(double));
2737
    int i;
2738
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2739

    
2740
    vec->coeff= coeff;
2741
    vec->length= length;
2742

    
2743
    for (i=0; i<length; i++) coeff[i]= 0.0;
2744

    
2745
    for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2746
    for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
2747

    
2748
    return vec;
2749
}
2750

    
2751
static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b){
2752
    int length= FFMAX(a->length, b->length);
2753
    double *coeff= av_malloc(length*sizeof(double));
2754
    int i;
2755
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2756

    
2757
    vec->coeff= coeff;
2758
    vec->length= length;
2759

    
2760
    for (i=0; i<length; i++) coeff[i]= 0.0;
2761

    
2762
    for (i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
2763
    for (i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
2764

    
2765
    return vec;
2766
}
2767

    
2768
/* shift left / or right if "shift" is negative */
2769
static SwsVector *sws_getShiftedVec(SwsVector *a, int shift){
2770
    int length= a->length + FFABS(shift)*2;
2771
    double *coeff= av_malloc(length*sizeof(double));
2772
    int i;
2773
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2774

    
2775
    vec->coeff= coeff;
2776
    vec->length= length;
2777

    
2778
    for (i=0; i<length; i++) coeff[i]= 0.0;
2779

    
2780
    for (i=0; i<a->length; i++)
2781
    {
2782
        coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
2783
    }
2784

    
2785
    return vec;
2786
}
2787

    
2788
void sws_shiftVec(SwsVector *a, int shift){
2789
    SwsVector *shifted= sws_getShiftedVec(a, shift);
2790
    av_free(a->coeff);
2791
    a->coeff= shifted->coeff;
2792
    a->length= shifted->length;
2793
    av_free(shifted);
2794
}
2795

    
2796
void sws_addVec(SwsVector *a, SwsVector *b){
2797
    SwsVector *sum= sws_sumVec(a, b);
2798
    av_free(a->coeff);
2799
    a->coeff= sum->coeff;
2800
    a->length= sum->length;
2801
    av_free(sum);
2802
}
2803

    
2804
void sws_subVec(SwsVector *a, SwsVector *b){
2805
    SwsVector *diff= sws_diffVec(a, b);
2806
    av_free(a->coeff);
2807
    a->coeff= diff->coeff;
2808
    a->length= diff->length;
2809
    av_free(diff);
2810
}
2811

    
2812
void sws_convVec(SwsVector *a, SwsVector *b){
2813
    SwsVector *conv= sws_getConvVec(a, b);
2814
    av_free(a->coeff);
2815
    a->coeff= conv->coeff;
2816
    a->length= conv->length;
2817
    av_free(conv);
2818
}
2819

    
2820
SwsVector *sws_cloneVec(SwsVector *a){
2821
    double *coeff= av_malloc(a->length*sizeof(double));
2822
    int i;
2823
    SwsVector *vec= av_malloc(sizeof(SwsVector));
2824

    
2825
    vec->coeff= coeff;
2826
    vec->length= a->length;
2827

    
2828
    for (i=0; i<a->length; i++) coeff[i]= a->coeff[i];
2829

    
2830
    return vec;
2831
}
2832

    
2833
void sws_printVec(SwsVector *a){
2834
    int i;
2835
    double max=0;
2836
    double min=0;
2837
    double range;
2838

    
2839
    for (i=0; i<a->length; i++)
2840
        if (a->coeff[i]>max) max= a->coeff[i];
2841

    
2842
    for (i=0; i<a->length; i++)
2843
        if (a->coeff[i]<min) min= a->coeff[i];
2844

    
2845
    range= max - min;
2846

    
2847
    for (i=0; i<a->length; i++)
2848
    {
2849
        int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
2850
        av_log(NULL, AV_LOG_DEBUG, "%1.3f ", a->coeff[i]);
2851
        for (;x>0; x--) av_log(NULL, AV_LOG_DEBUG, " ");
2852
        av_log(NULL, AV_LOG_DEBUG, "|\n");
2853
    }
2854
}
2855

    
2856
void sws_freeVec(SwsVector *a){
2857
    if (!a) return;
2858
    av_free(a->coeff);
2859
    a->coeff=NULL;
2860
    a->length=0;
2861
    av_free(a);
2862
}
2863

    
2864
void sws_freeFilter(SwsFilter *filter){
2865
    if (!filter) return;
2866

    
2867
    if (filter->lumH) sws_freeVec(filter->lumH);
2868
    if (filter->lumV) sws_freeVec(filter->lumV);
2869
    if (filter->chrH) sws_freeVec(filter->chrH);
2870
    if (filter->chrV) sws_freeVec(filter->chrV);
2871
    av_free(filter);
2872
}
2873

    
2874

    
2875
void sws_freeContext(SwsContext *c){
2876
    int i;
2877
    if (!c) return;
2878

    
2879
    if (c->lumPixBuf)
2880
    {
2881
        for (i=0; i<c->vLumBufSize; i++)
2882
        {
2883
            av_free(c->lumPixBuf[i]);
2884
            c->lumPixBuf[i]=NULL;
2885
        }
2886
        av_free(c->lumPixBuf);
2887
        c->lumPixBuf=NULL;
2888
    }
2889

    
2890
    if (c->chrPixBuf)
2891
    {
2892
        for (i=0; i<c->vChrBufSize; i++)
2893
        {
2894
            av_free(c->chrPixBuf[i]);
2895
            c->chrPixBuf[i]=NULL;
2896
        }
2897
        av_free(c->chrPixBuf);
2898
        c->chrPixBuf=NULL;
2899
    }
2900

    
2901
    av_free(c->vLumFilter);
2902
    c->vLumFilter = NULL;
2903
    av_free(c->vChrFilter);
2904
    c->vChrFilter = NULL;
2905
    av_free(c->hLumFilter);
2906
    c->hLumFilter = NULL;
2907
    av_free(c->hChrFilter);
2908
    c->hChrFilter = NULL;
2909
#ifdef HAVE_ALTIVEC
2910
    av_free(c->vYCoeffsBank);
2911
    c->vYCoeffsBank = NULL;
2912
    av_free(c->vCCoeffsBank);
2913
    c->vCCoeffsBank = NULL;
2914
#endif
2915

    
2916
    av_free(c->vLumFilterPos);
2917
    c->vLumFilterPos = NULL;
2918
    av_free(c->vChrFilterPos);
2919
    c->vChrFilterPos = NULL;
2920
    av_free(c->hLumFilterPos);
2921
    c->hLumFilterPos = NULL;
2922
    av_free(c->hChrFilterPos);
2923
    c->hChrFilterPos = NULL;
2924

    
2925
#if defined(ARCH_X86) && defined(CONFIG_GPL)
2926
#ifdef MAP_ANONYMOUS
2927
    if (c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
2928
    if (c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
2929
#else
2930
    av_free(c->funnyYCode);
2931
    av_free(c->funnyUVCode);
2932
#endif
2933
    c->funnyYCode=NULL;
2934
    c->funnyUVCode=NULL;
2935
#endif /* defined(ARCH_X86) */
2936

    
2937
    av_free(c->lumMmx2Filter);
2938
    c->lumMmx2Filter=NULL;
2939
    av_free(c->chrMmx2Filter);
2940
    c->chrMmx2Filter=NULL;
2941
    av_free(c->lumMmx2FilterPos);
2942
    c->lumMmx2FilterPos=NULL;
2943
    av_free(c->chrMmx2FilterPos);
2944
    c->chrMmx2FilterPos=NULL;
2945
    av_free(c->yuvTable);
2946
    c->yuvTable=NULL;
2947

    
2948
    av_free(c);
2949
}
2950

    
2951
/**
2952
 * Checks if context is valid or reallocs a new one instead.
2953
 * If context is NULL, just calls sws_getContext() to get a new one.
2954
 * Otherwise, checks if the parameters are the same already saved in context.
2955
 * If that is the case, returns the current context.
2956
 * Otherwise, frees context and gets a new one.
2957
 *
2958
 * Be warned that srcFilter, dstFilter are not checked, they are
2959
 * asumed to remain valid.
2960
 */
2961
struct SwsContext *sws_getCachedContext(struct SwsContext *context,
2962
                                        int srcW, int srcH, int srcFormat,
2963
                                        int dstW, int dstH, int dstFormat, int flags,
2964
                                        SwsFilter *srcFilter, SwsFilter *dstFilter, double *param)
2965
{
2966
    static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
2967

    
2968
    if (!param)
2969
        param = default_param;
2970

    
2971
    if (context) {
2972
        if (context->srcW != srcW || context->srcH != srcH ||
2973
            context->srcFormat != srcFormat ||
2974
            context->dstW != dstW || context->dstH != dstH ||
2975
            context->dstFormat != dstFormat || context->flags != flags ||
2976
            context->param[0] != param[0] || context->param[1] != param[1])
2977
        {
2978
            sws_freeContext(context);
2979
            context = NULL;
2980
        }
2981
    }
2982
    if (!context) {
2983
        return sws_getContext(srcW, srcH, srcFormat,
2984
                              dstW, dstH, dstFormat, flags,
2985
                              srcFilter, dstFilter, param);
2986
    }
2987
    return context;
2988
}
2989