Statistics
| Branch: | Revision:

ffmpeg / libswscale / utils.c @ 479e1e61

History | View | Annotate | Download (56.4 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
#define _SVID_SOURCE //needed for MAP_ANONYMOUS
25
#include <inttypes.h>
26
#include <string.h>
27
#include <math.h>
28
#include <stdio.h>
29
#include "config.h"
30
#include <assert.h>
31
#if HAVE_SYS_MMAN_H
32
#include <sys/mman.h>
33
#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
34
#define MAP_ANONYMOUS MAP_ANON
35
#endif
36
#endif
37
#if HAVE_VIRTUALALLOC
38
#define WIN32_LEAN_AND_MEAN
39
#include <windows.h>
40
#endif
41
#include "swscale.h"
42
#include "swscale_internal.h"
43
#include "rgb2rgb.h"
44
#include "libavutil/intreadwrite.h"
45
#include "libavutil/x86_cpu.h"
46
#include "libavutil/avutil.h"
47
#include "libavutil/bswap.h"
48
#include "libavutil/pixdesc.h"
49

    
50
unsigned swscale_version(void)
51
{
52
    return LIBSWSCALE_VERSION_INT;
53
}
54

    
55
const char *swscale_configuration(void)
56
{
57
    return FFMPEG_CONFIGURATION;
58
}
59

    
60
const char *swscale_license(void)
61
{
62
#define LICENSE_PREFIX "libswscale license: "
63
    return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
64
}
65

    
66
#define RET 0xC3 //near return opcode for x86
67

    
68
#define isSupportedIn(x)    (       \
69
           (x)==PIX_FMT_YUV420P     \
70
        || (x)==PIX_FMT_YUVA420P    \
71
        || (x)==PIX_FMT_YUYV422     \
72
        || (x)==PIX_FMT_UYVY422     \
73
        || (x)==PIX_FMT_RGB48BE     \
74
        || (x)==PIX_FMT_RGB48LE     \
75
        || (x)==PIX_FMT_RGB32       \
76
        || (x)==PIX_FMT_RGB32_1     \
77
        || (x)==PIX_FMT_BGR24       \
78
        || (x)==PIX_FMT_BGR565      \
79
        || (x)==PIX_FMT_BGR555      \
80
        || (x)==PIX_FMT_BGR32       \
81
        || (x)==PIX_FMT_BGR32_1     \
82
        || (x)==PIX_FMT_RGB24       \
83
        || (x)==PIX_FMT_RGB565      \
84
        || (x)==PIX_FMT_RGB555      \
85
        || (x)==PIX_FMT_GRAY8       \
86
        || (x)==PIX_FMT_YUV410P     \
87
        || (x)==PIX_FMT_YUV440P     \
88
        || (x)==PIX_FMT_NV12        \
89
        || (x)==PIX_FMT_NV21        \
90
        || (x)==PIX_FMT_GRAY16BE    \
91
        || (x)==PIX_FMT_GRAY16LE    \
92
        || (x)==PIX_FMT_YUV444P     \
93
        || (x)==PIX_FMT_YUV422P     \
94
        || (x)==PIX_FMT_YUV411P     \
95
        || (x)==PIX_FMT_YUVJ420P    \
96
        || (x)==PIX_FMT_YUVJ422P    \
97
        || (x)==PIX_FMT_YUVJ440P    \
98
        || (x)==PIX_FMT_YUVJ444P    \
99
        || (x)==PIX_FMT_PAL8        \
100
        || (x)==PIX_FMT_BGR8        \
101
        || (x)==PIX_FMT_RGB8        \
102
        || (x)==PIX_FMT_BGR4_BYTE   \
103
        || (x)==PIX_FMT_RGB4_BYTE   \
104
        || (x)==PIX_FMT_YUV440P     \
105
        || (x)==PIX_FMT_MONOWHITE   \
106
        || (x)==PIX_FMT_MONOBLACK   \
107
        || (x)==PIX_FMT_YUV420P16LE   \
108
        || (x)==PIX_FMT_YUV422P16LE   \
109
        || (x)==PIX_FMT_YUV444P16LE   \
110
        || (x)==PIX_FMT_YUV420P16BE   \
111
        || (x)==PIX_FMT_YUV422P16BE   \
112
        || (x)==PIX_FMT_YUV444P16BE   \
113
    )
114

    
115
int sws_isSupportedInput(enum PixelFormat pix_fmt)
116
{
117
    return isSupportedIn(pix_fmt);
118
}
119

    
120
#define isSupportedOut(x)   (       \
121
           (x)==PIX_FMT_YUV420P     \
122
        || (x)==PIX_FMT_YUVA420P    \
123
        || (x)==PIX_FMT_YUYV422     \
124
        || (x)==PIX_FMT_UYVY422     \
125
        || (x)==PIX_FMT_YUV444P     \
126
        || (x)==PIX_FMT_YUV422P     \
127
        || (x)==PIX_FMT_YUV411P     \
128
        || (x)==PIX_FMT_YUVJ420P    \
129
        || (x)==PIX_FMT_YUVJ422P    \
130
        || (x)==PIX_FMT_YUVJ440P    \
131
        || (x)==PIX_FMT_YUVJ444P    \
132
        || isAnyRGB(x)              \
133
        || (x)==PIX_FMT_NV12        \
134
        || (x)==PIX_FMT_NV21        \
135
        || (x)==PIX_FMT_GRAY16BE    \
136
        || (x)==PIX_FMT_GRAY16LE    \
137
        || (x)==PIX_FMT_GRAY8       \
138
        || (x)==PIX_FMT_YUV410P     \
139
        || (x)==PIX_FMT_YUV440P     \
140
        || (x)==PIX_FMT_YUV420P16LE   \
141
        || (x)==PIX_FMT_YUV422P16LE   \
142
        || (x)==PIX_FMT_YUV444P16LE   \
143
        || (x)==PIX_FMT_YUV420P16BE   \
144
        || (x)==PIX_FMT_YUV422P16BE   \
145
        || (x)==PIX_FMT_YUV444P16BE   \
146
    )
147

    
148
int sws_isSupportedOutput(enum PixelFormat pix_fmt)
149
{
150
    return isSupportedOut(pix_fmt);
151
}
152

    
153
#define usePal(x) (av_pix_fmt_descriptors[x].flags & PIX_FMT_PAL)
154

    
155
extern const int32_t ff_yuv2rgb_coeffs[8][4];
156

    
157
const char *sws_format_name(enum PixelFormat format)
158
{
159
    if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name)
160
        return av_pix_fmt_descriptors[format].name;
161
    else
162
        return "Unknown format";
163
}
164

    
165
static double getSplineCoeff(double a, double b, double c, double d, double dist)
166
{
167
//    printf("%f %f %f %f %f\n", a,b,c,d,dist);
168
    if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
169
    else           return getSplineCoeff(        0.0,
170
                                          b+ 2.0*c + 3.0*d,
171
                                                 c + 3.0*d,
172
                                         -b- 3.0*c - 6.0*d,
173
                                         dist-1.0);
174
}
175

    
176
static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
177
                      int srcW, int dstW, int filterAlign, int one, int flags,
178
                      SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
179
{
180
    int i;
181
    int filterSize;
182
    int filter2Size;
183
    int minFilterSize;
184
    int64_t *filter=NULL;
185
    int64_t *filter2=NULL;
186
    const int64_t fone= 1LL<<54;
187
    int ret= -1;
188
#if ARCH_X86
189
    if (flags & SWS_CPU_CAPS_MMX)
190
        __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
191
#endif
192

    
193
    // NOTE: the +1 is for the MMX scaler which reads over the end
194
    FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
195

    
196
    if (FFABS(xInc - 0x10000) <10) { // unscaled
197
        int i;
198
        filterSize= 1;
199
        FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
200

    
201
        for (i=0; i<dstW; i++) {
202
            filter[i*filterSize]= fone;
203
            (*filterPos)[i]=i;
204
        }
205

    
206
    } else if (flags&SWS_POINT) { // lame looking point sampling mode
207
        int i;
208
        int xDstInSrc;
209
        filterSize= 1;
210
        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
211

    
212
        xDstInSrc= xInc/2 - 0x8000;
213
        for (i=0; i<dstW; i++) {
214
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
215

    
216
            (*filterPos)[i]= xx;
217
            filter[i]= fone;
218
            xDstInSrc+= xInc;
219
        }
220
    } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
221
        int i;
222
        int xDstInSrc;
223
        filterSize= 2;
224
        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
225

    
226
        xDstInSrc= xInc/2 - 0x8000;
227
        for (i=0; i<dstW; i++) {
228
            int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
229
            int j;
230

    
231
            (*filterPos)[i]= xx;
232
            //bilinear upscale / linear interpolate / area averaging
233
            for (j=0; j<filterSize; j++) {
234
                int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
235
                if (coeff<0) coeff=0;
236
                filter[i*filterSize + j]= coeff;
237
                xx++;
238
            }
239
            xDstInSrc+= xInc;
240
        }
241
    } else {
242
        int xDstInSrc;
243
        int sizeFactor;
244

    
245
        if      (flags&SWS_BICUBIC)      sizeFactor=  4;
246
        else if (flags&SWS_X)            sizeFactor=  8;
247
        else if (flags&SWS_AREA)         sizeFactor=  1; //downscale only, for upscale it is bilinear
248
        else if (flags&SWS_GAUSS)        sizeFactor=  8;   // infinite ;)
249
        else if (flags&SWS_LANCZOS)      sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
250
        else if (flags&SWS_SINC)         sizeFactor= 20; // infinite ;)
251
        else if (flags&SWS_SPLINE)       sizeFactor= 20;  // infinite ;)
252
        else if (flags&SWS_BILINEAR)     sizeFactor=  2;
253
        else {
254
            sizeFactor= 0; //GCC warning killer
255
            assert(0);
256
        }
257

    
258
        if (xInc <= 1<<16)      filterSize= 1 + sizeFactor; // upscale
259
        else                    filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
260

    
261
        if (filterSize > srcW-2) filterSize=srcW-2;
262

    
263
        FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
264

    
265
        xDstInSrc= xInc - 0x10000;
266
        for (i=0; i<dstW; i++) {
267
            int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
268
            int j;
269
            (*filterPos)[i]= xx;
270
            for (j=0; j<filterSize; j++) {
271
                int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
272
                double floatd;
273
                int64_t coeff;
274

    
275
                if (xInc > 1<<16)
276
                    d= d*dstW/srcW;
277
                floatd= d * (1.0/(1<<30));
278

    
279
                if (flags & SWS_BICUBIC) {
280
                    int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] :   0) * (1<<24);
281
                    int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
282
                    int64_t dd = ( d*d)>>30;
283
                    int64_t ddd= (dd*d)>>30;
284

    
285
                    if      (d < 1LL<<30)
286
                        coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
287
                    else if (d < 1LL<<31)
288
                        coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
289
                    else
290
                        coeff=0.0;
291
                    coeff *= fone>>(30+24);
292
                }
293
/*                else if (flags & SWS_X) {
294
                    double p= param ? param*0.01 : 0.3;
295
                    coeff = d ? sin(d*PI)/(d*PI) : 1.0;
296
                    coeff*= pow(2.0, - p*d*d);
297
                }*/
298
                else if (flags & SWS_X) {
299
                    double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
300
                    double c;
301

    
302
                    if (floatd<1.0)
303
                        c = cos(floatd*M_PI);
304
                    else
305
                        c=-1.0;
306
                    if (c<0.0)      c= -pow(-c, A);
307
                    else            c=  pow( c, A);
308
                    coeff= (c*0.5 + 0.5)*fone;
309
                } else if (flags & SWS_AREA) {
310
                    int64_t d2= d - (1<<29);
311
                    if      (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
312
                    else if (d2*xInc <  (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
313
                    else coeff=0.0;
314
                    coeff *= fone>>(30+16);
315
                } else if (flags & SWS_GAUSS) {
316
                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
317
                    coeff = (pow(2.0, - p*floatd*floatd))*fone;
318
                } else if (flags & SWS_SINC) {
319
                    coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
320
                } else if (flags & SWS_LANCZOS) {
321
                    double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
322
                    coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
323
                    if (floatd>p) coeff=0;
324
                } else if (flags & SWS_BILINEAR) {
325
                    coeff= (1<<30) - d;
326
                    if (coeff<0) coeff=0;
327
                    coeff *= fone >> 30;
328
                } else if (flags & SWS_SPLINE) {
329
                    double p=-2.196152422706632;
330
                    coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
331
                } else {
332
                    coeff= 0.0; //GCC warning killer
333
                    assert(0);
334
                }
335

    
336
                filter[i*filterSize + j]= coeff;
337
                xx++;
338
            }
339
            xDstInSrc+= 2*xInc;
340
        }
341
    }
342

    
343
    /* apply src & dst Filter to filter -> filter2
344
       av_free(filter);
345
    */
346
    assert(filterSize>0);
347
    filter2Size= filterSize;
348
    if (srcFilter) filter2Size+= srcFilter->length - 1;
349
    if (dstFilter) filter2Size+= dstFilter->length - 1;
350
    assert(filter2Size>0);
351
    FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
352

    
353
    for (i=0; i<dstW; i++) {
354
        int j, k;
355

    
356
        if(srcFilter) {
357
            for (k=0; k<srcFilter->length; k++) {
358
                for (j=0; j<filterSize; j++)
359
                    filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
360
            }
361
        } else {
362
            for (j=0; j<filterSize; j++)
363
                filter2[i*filter2Size + j]= filter[i*filterSize + j];
364
        }
365
        //FIXME dstFilter
366

    
367
        (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
368
    }
369
    av_freep(&filter);
370

    
371
    /* try to reduce the filter-size (step1 find size and shift left) */
372
    // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
373
    minFilterSize= 0;
374
    for (i=dstW-1; i>=0; i--) {
375
        int min= filter2Size;
376
        int j;
377
        int64_t cutOff=0.0;
378

    
379
        /* get rid of near zero elements on the left by shifting left */
380
        for (j=0; j<filter2Size; j++) {
381
            int k;
382
            cutOff += FFABS(filter2[i*filter2Size]);
383

    
384
            if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
385

    
386
            /* preserve monotonicity because the core can't handle the filter otherwise */
387
            if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
388

    
389
            // move filter coefficients left
390
            for (k=1; k<filter2Size; k++)
391
                filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
392
            filter2[i*filter2Size + k - 1]= 0;
393
            (*filterPos)[i]++;
394
        }
395

    
396
        cutOff=0;
397
        /* count near zeros on the right */
398
        for (j=filter2Size-1; j>0; j--) {
399
            cutOff += FFABS(filter2[i*filter2Size + j]);
400

    
401
            if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
402
            min--;
403
        }
404

    
405
        if (min>minFilterSize) minFilterSize= min;
406
    }
407

    
408
    if (flags & SWS_CPU_CAPS_ALTIVEC) {
409
        // we can handle the special case 4,
410
        // so we don't want to go to the full 8
411
        if (minFilterSize < 5)
412
            filterAlign = 4;
413

    
414
        // We really don't want to waste our time
415
        // doing useless computation, so fall back on
416
        // the scalar C code for very small filters.
417
        // Vectorizing is worth it only if you have a
418
        // decent-sized vector.
419
        if (minFilterSize < 3)
420
            filterAlign = 1;
421
    }
422

    
423
    if (flags & SWS_CPU_CAPS_MMX) {
424
        // special case for unscaled vertical filtering
425
        if (minFilterSize == 1 && filterAlign == 2)
426
            filterAlign= 1;
427
    }
428

    
429
    assert(minFilterSize > 0);
430
    filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
431
    assert(filterSize > 0);
432
    filter= av_malloc(filterSize*dstW*sizeof(*filter));
433
    if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
434
        goto fail;
435
    *outFilterSize= filterSize;
436

    
437
    if (flags&SWS_PRINT_INFO)
438
        av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
439
    /* try to reduce the filter-size (step2 reduce it) */
440
    for (i=0; i<dstW; i++) {
441
        int j;
442

    
443
        for (j=0; j<filterSize; j++) {
444
            if (j>=filter2Size) filter[i*filterSize + j]= 0;
445
            else               filter[i*filterSize + j]= filter2[i*filter2Size + j];
446
            if((flags & SWS_BITEXACT) && j>=minFilterSize)
447
                filter[i*filterSize + j]= 0;
448
        }
449
    }
450

    
451
    //FIXME try to align filterPos if possible
452

    
453
    //fix borders
454
    for (i=0; i<dstW; i++) {
455
        int j;
456
        if ((*filterPos)[i] < 0) {
457
            // move filter coefficients left to compensate for filterPos
458
            for (j=1; j<filterSize; j++) {
459
                int left= FFMAX(j + (*filterPos)[i], 0);
460
                filter[i*filterSize + left] += filter[i*filterSize + j];
461
                filter[i*filterSize + j]=0;
462
            }
463
            (*filterPos)[i]= 0;
464
        }
465

    
466
        if ((*filterPos)[i] + filterSize > srcW) {
467
            int shift= (*filterPos)[i] + filterSize - srcW;
468
            // move filter coefficients right to compensate for filterPos
469
            for (j=filterSize-2; j>=0; j--) {
470
                int right= FFMIN(j + shift, filterSize-1);
471
                filter[i*filterSize +right] += filter[i*filterSize +j];
472
                filter[i*filterSize +j]=0;
473
            }
474
            (*filterPos)[i]= srcW - filterSize;
475
        }
476
    }
477

    
478
    // Note the +1 is for the MMX scaler which reads over the end
479
    /* align at 16 for AltiVec (needed by hScale_altivec_real) */
480
    FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
481

    
482
    /* normalize & store in outFilter */
483
    for (i=0; i<dstW; i++) {
484
        int j;
485
        int64_t error=0;
486
        int64_t sum=0;
487

    
488
        for (j=0; j<filterSize; j++) {
489
            sum+= filter[i*filterSize + j];
490
        }
491
        sum= (sum + one/2)/ one;
492
        for (j=0; j<*outFilterSize; j++) {
493
            int64_t v= filter[i*filterSize + j] + error;
494
            int intV= ROUNDED_DIV(v, sum);
495
            (*outFilter)[i*(*outFilterSize) + j]= intV;
496
            error= v - intV*sum;
497
        }
498
    }
499

    
500
    (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
501
    for (i=0; i<*outFilterSize; i++) {
502
        int j= dstW*(*outFilterSize);
503
        (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
504
    }
505

    
506
    ret=0;
507
fail:
508
    av_free(filter);
509
    av_free(filter2);
510
    return ret;
511
}
512

    
513
#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
514
static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
515
{
516
    uint8_t *fragmentA;
517
    x86_reg imm8OfPShufW1A;
518
    x86_reg imm8OfPShufW2A;
519
    x86_reg fragmentLengthA;
520
    uint8_t *fragmentB;
521
    x86_reg imm8OfPShufW1B;
522
    x86_reg imm8OfPShufW2B;
523
    x86_reg fragmentLengthB;
524
    int fragmentPos;
525

    
526
    int xpos, i;
527

    
528
    // create an optimized horizontal scaling routine
529
    /* This scaler is made of runtime-generated MMX2 code using specially
530
     * tuned pshufw instructions. For every four output pixels, if four
531
     * input pixels are enough for the fast bilinear scaling, then a chunk
532
     * of fragmentB is used. If five input pixels are needed, then a chunk
533
     * of fragmentA is used.
534
     */
535

    
536
    //code fragment
537

    
538
    __asm__ volatile(
539
        "jmp                         9f                 \n\t"
540
    // Begin
541
        "0:                                             \n\t"
542
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
543
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
544
        "movd   1(%%"REG_c", %%"REG_S"), %%mm1          \n\t"
545
        "punpcklbw                %%mm7, %%mm1          \n\t"
546
        "punpcklbw                %%mm7, %%mm0          \n\t"
547
        "pshufw                   $0xFF, %%mm1, %%mm1   \n\t"
548
        "1:                                             \n\t"
549
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
550
        "2:                                             \n\t"
551
        "psubw                    %%mm1, %%mm0          \n\t"
552
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
553
        "pmullw                   %%mm3, %%mm0          \n\t"
554
        "psllw                       $7, %%mm1          \n\t"
555
        "paddw                    %%mm1, %%mm0          \n\t"
556

    
557
        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
558

    
559
        "add                         $8, %%"REG_a"      \n\t"
560
    // End
561
        "9:                                             \n\t"
562
//        "int $3                                         \n\t"
563
        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
564
        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
565
        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
566
        "dec                         %1                 \n\t"
567
        "dec                         %2                 \n\t"
568
        "sub                         %0, %1             \n\t"
569
        "sub                         %0, %2             \n\t"
570
        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
571
        "sub                         %0, %3             \n\t"
572

    
573

    
574
        :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
575
        "=r" (fragmentLengthA)
576
    );
577

    
578
    __asm__ volatile(
579
        "jmp                         9f                 \n\t"
580
    // Begin
581
        "0:                                             \n\t"
582
        "movq    (%%"REG_d", %%"REG_a"), %%mm3          \n\t"
583
        "movd    (%%"REG_c", %%"REG_S"), %%mm0          \n\t"
584
        "punpcklbw                %%mm7, %%mm0          \n\t"
585
        "pshufw                   $0xFF, %%mm0, %%mm1   \n\t"
586
        "1:                                             \n\t"
587
        "pshufw                   $0xFF, %%mm0, %%mm0   \n\t"
588
        "2:                                             \n\t"
589
        "psubw                    %%mm1, %%mm0          \n\t"
590
        "movl   8(%%"REG_b", %%"REG_a"), %%esi          \n\t"
591
        "pmullw                   %%mm3, %%mm0          \n\t"
592
        "psllw                       $7, %%mm1          \n\t"
593
        "paddw                    %%mm1, %%mm0          \n\t"
594

    
595
        "movq                     %%mm0, (%%"REG_D", %%"REG_a") \n\t"
596

    
597
        "add                         $8, %%"REG_a"      \n\t"
598
    // End
599
        "9:                                             \n\t"
600
//        "int                       $3                   \n\t"
601
        "lea                 " LOCAL_MANGLE(0b) ", %0   \n\t"
602
        "lea                 " LOCAL_MANGLE(1b) ", %1   \n\t"
603
        "lea                 " LOCAL_MANGLE(2b) ", %2   \n\t"
604
        "dec                         %1                 \n\t"
605
        "dec                         %2                 \n\t"
606
        "sub                         %0, %1             \n\t"
607
        "sub                         %0, %2             \n\t"
608
        "lea                 " LOCAL_MANGLE(9b) ", %3   \n\t"
609
        "sub                         %0, %3             \n\t"
610

    
611

    
612
        :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
613
        "=r" (fragmentLengthB)
614
    );
615

    
616
    xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
617
    fragmentPos=0;
618

    
619
    for (i=0; i<dstW/numSplits; i++) {
620
        int xx=xpos>>16;
621

    
622
        if ((i&3) == 0) {
623
            int a=0;
624
            int b=((xpos+xInc)>>16) - xx;
625
            int c=((xpos+xInc*2)>>16) - xx;
626
            int d=((xpos+xInc*3)>>16) - xx;
627
            int inc                = (d+1<4);
628
            uint8_t *fragment      = (d+1<4) ? fragmentB       : fragmentA;
629
            x86_reg imm8OfPShufW1  = (d+1<4) ? imm8OfPShufW1B  : imm8OfPShufW1A;
630
            x86_reg imm8OfPShufW2  = (d+1<4) ? imm8OfPShufW2B  : imm8OfPShufW2A;
631
            x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
632
            int maxShift= 3-(d+inc);
633
            int shift=0;
634

    
635
            if (filterCode) {
636
                filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
637
                filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
638
                filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
639
                filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
640
                filterPos[i/2]= xx;
641

    
642
                memcpy(filterCode + fragmentPos, fragment, fragmentLength);
643

    
644
                filterCode[fragmentPos + imm8OfPShufW1]=
645
                    (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
646
                filterCode[fragmentPos + imm8OfPShufW2]=
647
                    a | (b<<2) | (c<<4) | (d<<6);
648

    
649
                if (i+4-inc>=dstW) shift=maxShift; //avoid overread
650
                else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
651

    
652
                if (shift && i>=shift) {
653
                    filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
654
                    filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
655
                    filterPos[i/2]-=shift;
656
                }
657
            }
658

    
659
            fragmentPos+= fragmentLength;
660

    
661
            if (filterCode)
662
                filterCode[fragmentPos]= RET;
663
        }
664
        xpos+=xInc;
665
    }
666
    if (filterCode)
667
        filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
668

    
669
    return fragmentPos + 1;
670
}
671
#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL */
672

    
673
static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
674
{
675
    *h = av_pix_fmt_descriptors[format].log2_chroma_w;
676
    *v = av_pix_fmt_descriptors[format].log2_chroma_h;
677
}
678

    
679
static uint16_t roundToInt16(int64_t f)
680
{
681
    int r= (f + (1<<15))>>16;
682
         if (r<-0x7FFF) return 0x8000;
683
    else if (r> 0x7FFF) return 0x7FFF;
684
    else                return r;
685
}
686

    
687
int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
688
{
689
    int64_t crv =  inv_table[0];
690
    int64_t cbu =  inv_table[1];
691
    int64_t cgu = -inv_table[2];
692
    int64_t cgv = -inv_table[3];
693
    int64_t cy  = 1<<16;
694
    int64_t oy  = 0;
695

    
696
    memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
697
    memcpy(c->dstColorspaceTable,     table, sizeof(int)*4);
698

    
699
    c->brightness= brightness;
700
    c->contrast  = contrast;
701
    c->saturation= saturation;
702
    c->srcRange  = srcRange;
703
    c->dstRange  = dstRange;
704
    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
705

    
706
    c->uOffset=   0x0400040004000400LL;
707
    c->vOffset=   0x0400040004000400LL;
708

    
709
    if (!srcRange) {
710
        cy= (cy*255) / 219;
711
        oy= 16<<16;
712
    } else {
713
        crv= (crv*224) / 255;
714
        cbu= (cbu*224) / 255;
715
        cgu= (cgu*224) / 255;
716
        cgv= (cgv*224) / 255;
717
    }
718

    
719
    cy = (cy *contrast             )>>16;
720
    crv= (crv*contrast * saturation)>>32;
721
    cbu= (cbu*contrast * saturation)>>32;
722
    cgu= (cgu*contrast * saturation)>>32;
723
    cgv= (cgv*contrast * saturation)>>32;
724

    
725
    oy -= 256*brightness;
726

    
727
    c->yCoeff=    roundToInt16(cy *8192) * 0x0001000100010001ULL;
728
    c->vrCoeff=   roundToInt16(crv*8192) * 0x0001000100010001ULL;
729
    c->ubCoeff=   roundToInt16(cbu*8192) * 0x0001000100010001ULL;
730
    c->vgCoeff=   roundToInt16(cgv*8192) * 0x0001000100010001ULL;
731
    c->ugCoeff=   roundToInt16(cgu*8192) * 0x0001000100010001ULL;
732
    c->yOffset=   roundToInt16(oy *   8) * 0x0001000100010001ULL;
733

    
734
    c->yuv2rgb_y_coeff  = (int16_t)roundToInt16(cy <<13);
735
    c->yuv2rgb_y_offset = (int16_t)roundToInt16(oy << 9);
736
    c->yuv2rgb_v2r_coeff= (int16_t)roundToInt16(crv<<13);
737
    c->yuv2rgb_v2g_coeff= (int16_t)roundToInt16(cgv<<13);
738
    c->yuv2rgb_u2g_coeff= (int16_t)roundToInt16(cgu<<13);
739
    c->yuv2rgb_u2b_coeff= (int16_t)roundToInt16(cbu<<13);
740

    
741
    ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
742
    //FIXME factorize
743

    
744
#if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)
745
    if (c->flags & SWS_CPU_CAPS_ALTIVEC)
746
        ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
747
#endif
748
    return 0;
749
}
750

    
751
int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation)
752
{
753
    if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
754

    
755
    *inv_table = c->srcColorspaceTable;
756
    *table     = c->dstColorspaceTable;
757
    *srcRange  = c->srcRange;
758
    *dstRange  = c->dstRange;
759
    *brightness= c->brightness;
760
    *contrast  = c->contrast;
761
    *saturation= c->saturation;
762

    
763
    return 0;
764
}
765

    
766
static int handle_jpeg(enum PixelFormat *format)
767
{
768
    switch (*format) {
769
    case PIX_FMT_YUVJ420P:
770
        *format = PIX_FMT_YUV420P;
771
        return 1;
772
    case PIX_FMT_YUVJ422P:
773
        *format = PIX_FMT_YUV422P;
774
        return 1;
775
    case PIX_FMT_YUVJ444P:
776
        *format = PIX_FMT_YUV444P;
777
        return 1;
778
    case PIX_FMT_YUVJ440P:
779
        *format = PIX_FMT_YUV440P;
780
        return 1;
781
    default:
782
        return 0;
783
    }
784
}
785

    
786
SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
787
                           int dstW, int dstH, enum PixelFormat dstFormat, int flags,
788
                           SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
789
{
790
    SwsContext *c;
791
    int i;
792
    int usesVFilter, usesHFilter;
793
    int unscaled;
794
    int srcRange, dstRange;
795
    SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
796
#if ARCH_X86
797
    if (flags & SWS_CPU_CAPS_MMX)
798
        __asm__ volatile("emms\n\t"::: "memory");
799
#endif
800

    
801
#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
802
    flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
803
    flags |= ff_hardcodedcpuflags();
804
#endif /* CONFIG_RUNTIME_CPUDETECT */
805
    if (!rgb15to16) sws_rgb2rgb_init(flags);
806

    
807
    unscaled = (srcW == dstW && srcH == dstH);
808

    
809
    srcRange = handle_jpeg(&srcFormat);
810
    dstRange = handle_jpeg(&dstFormat);
811

    
812
    if (!isSupportedIn(srcFormat)) {
813
        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat));
814
        return NULL;
815
    }
816
    if (!isSupportedOut(dstFormat)) {
817
        av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat));
818
        return NULL;
819
    }
820

    
821
    i= flags & ( SWS_POINT
822
                |SWS_AREA
823
                |SWS_BILINEAR
824
                |SWS_FAST_BILINEAR
825
                |SWS_BICUBIC
826
                |SWS_X
827
                |SWS_GAUSS
828
                |SWS_LANCZOS
829
                |SWS_SINC
830
                |SWS_SPLINE
831
                |SWS_BICUBLIN);
832
    if(!i || (i & (i-1))) {
833
        av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n");
834
        return NULL;
835
    }
836

    
837
    /* sanity check */
838
    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
839
        av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n",
840
               srcW, srcH, dstW, dstH);
841
        return NULL;
842
    }
843
    if(srcW > VOFW || dstW > VOFW) {
844
        av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n");
845
        return NULL;
846
    }
847

    
848
    if (!dstFilter) dstFilter= &dummyFilter;
849
    if (!srcFilter) srcFilter= &dummyFilter;
850

    
851
    FF_ALLOCZ_OR_GOTO(NULL, c, sizeof(SwsContext), fail);
852

    
853
    c->av_class = &sws_context_class;
854
    c->srcW= srcW;
855
    c->srcH= srcH;
856
    c->dstW= dstW;
857
    c->dstH= dstH;
858
    c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
859
    c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
860
    c->flags= flags;
861
    c->dstFormat= dstFormat;
862
    c->srcFormat= srcFormat;
863
    c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
864
    c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
865
    c->vRounder= 4* 0x0001000100010001ULL;
866

    
867
    usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
868
                  (srcFilter->chrV && srcFilter->chrV->length>1) ||
869
                  (dstFilter->lumV && dstFilter->lumV->length>1) ||
870
                  (dstFilter->chrV && dstFilter->chrV->length>1);
871
    usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
872
                  (srcFilter->chrH && srcFilter->chrH->length>1) ||
873
                  (dstFilter->lumH && dstFilter->lumH->length>1) ||
874
                  (dstFilter->chrH && dstFilter->chrH->length>1);
875

    
876
    getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
877
    getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
878

    
879
    // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
880
    if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
881

    
882
    // drop some chroma lines if the user wants it
883
    c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
884
    c->chrSrcVSubSample+= c->vChrDrop;
885

    
886
    // drop every other pixel for chroma calculation unless user wants full chroma
887
    if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
888
      && srcFormat!=PIX_FMT_RGB8      && srcFormat!=PIX_FMT_BGR8
889
      && srcFormat!=PIX_FMT_RGB4      && srcFormat!=PIX_FMT_BGR4
890
      && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
891
      && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&(SWS_FAST_BILINEAR|SWS_POINT))))
892
        c->chrSrcHSubSample=1;
893

    
894
    if (param) {
895
        c->param[0] = param[0];
896
        c->param[1] = param[1];
897
    } else {
898
        c->param[0] =
899
        c->param[1] = SWS_PARAM_DEFAULT;
900
    }
901

    
902
    // Note the -((-x)>>y) is so that we always round toward +inf.
903
    c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
904
    c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
905
    c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
906
    c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
907

    
908
    sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, dstRange, 0, 1<<16, 1<<16);
909

    
910
    /* unscaled special cases */
911
    if (unscaled && !usesHFilter && !usesVFilter && (srcRange == dstRange || isAnyRGB(dstFormat))) {
912
        ff_get_unscaled_swscale(c);
913

    
914
        if (c->swScale) {
915
            if (flags&SWS_PRINT_INFO)
916
                av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
917
                       sws_format_name(srcFormat), sws_format_name(dstFormat));
918
            return c;
919
        }
920
    }
921

    
922
    if (flags & SWS_CPU_CAPS_MMX2) {
923
        c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
924
        if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
925
            if (flags&SWS_PRINT_INFO)
926
                av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
927
        }
928
        if (usesHFilter) c->canMMX2BeUsed=0;
929
    }
930
    else
931
        c->canMMX2BeUsed=0;
932

    
933
    c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
934
    c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
935

    
936
    // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
937
    // but only for the FAST_BILINEAR mode otherwise do correct scaling
938
    // n-2 is the last chrominance sample available
939
    // this is not perfect, but no one should notice the difference, the more correct variant
940
    // would be like the vertical one, but that would require some special code for the
941
    // first and last pixel
942
    if (flags&SWS_FAST_BILINEAR) {
943
        if (c->canMMX2BeUsed) {
944
            c->lumXInc+= 20;
945
            c->chrXInc+= 20;
946
        }
947
        //we don't use the x86 asm scaler if MMX is available
948
        else if (flags & SWS_CPU_CAPS_MMX) {
949
            c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
950
            c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
951
        }
952
    }
953

    
954
    /* precalculate horizontal scaler filter coefficients */
955
    {
956
#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
957
// can't downscale !!!
958
        if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
959
            c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);
960
            c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
961

    
962
#ifdef MAP_ANONYMOUS
963
            c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
964
            c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
965
#elif HAVE_VIRTUALALLOC
966
            c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
967
            c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
968
#else
969
            c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
970
            c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
971
#endif
972

    
973
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilter   , (dstW        /8+8)*sizeof(int16_t), fail);
974
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilter   , (c->chrDstW  /4+8)*sizeof(int16_t), fail);
975
            FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW      /2/8+8)*sizeof(int32_t), fail);
976
            FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
977

    
978
            initMMX2HScaler(      dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
979
            initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
980

    
981
#ifdef MAP_ANONYMOUS
982
            mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
983
            mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
984
#endif
985
        } else
986
#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL */
987
        {
988
            const int filterAlign=
989
                (flags & SWS_CPU_CAPS_MMX) ? 4 :
990
                (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
991
                1;
992

    
993
            if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
994
                           srcW      ,       dstW, filterAlign, 1<<14,
995
                           (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
996
                           srcFilter->lumH, dstFilter->lumH, c->param) < 0)
997
                goto fail;
998
            if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
999
                           c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
1000
                           (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1001
                           srcFilter->chrH, dstFilter->chrH, c->param) < 0)
1002
                goto fail;
1003
        }
1004
    } // initialize horizontal stuff
1005

    
1006
    /* precalculate vertical scaler filter coefficients */
1007
    {
1008
        const int filterAlign=
1009
            (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
1010
            (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
1011
            1;
1012

    
1013
        if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
1014
                       srcH      ,        dstH, filterAlign, (1<<12),
1015
                       (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
1016
                       srcFilter->lumV, dstFilter->lumV, c->param) < 0)
1017
            goto fail;
1018
        if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
1019
                       c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
1020
                       (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
1021
                       srcFilter->chrV, dstFilter->chrV, c->param) < 0)
1022
            goto fail;
1023

    
1024
#if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)
1025
        FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
1026
        FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
1027

    
1028
        for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
1029
            int j;
1030
            short *p = (short *)&c->vYCoeffsBank[i];
1031
            for (j=0;j<8;j++)
1032
                p[j] = c->vLumFilter[i];
1033
        }
1034

    
1035
        for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
1036
            int j;
1037
            short *p = (short *)&c->vCCoeffsBank[i];
1038
            for (j=0;j<8;j++)
1039
                p[j] = c->vChrFilter[i];
1040
        }
1041
#endif
1042
    }
1043

    
1044
    // calculate buffer sizes so that they won't run out while handling these damn slices
1045
    c->vLumBufSize= c->vLumFilterSize;
1046
    c->vChrBufSize= c->vChrFilterSize;
1047
    for (i=0; i<dstH; i++) {
1048
        int chrI= i*c->chrDstH / dstH;
1049
        int nextSlice= FFMAX(c->vLumFilterPos[i   ] + c->vLumFilterSize - 1,
1050
                           ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
1051

    
1052
        nextSlice>>= c->chrSrcVSubSample;
1053
        nextSlice<<= c->chrSrcVSubSample;
1054
        if (c->vLumFilterPos[i   ] + c->vLumBufSize < nextSlice)
1055
            c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1056
        if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1057
            c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1058
    }
1059

    
1060
    // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1061
    // allocate several megabytes to handle all possible cases)
1062
    FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1063
    FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1064
    if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1065
        FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1066
    //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)
1067
    /* align at 16 bytes for AltiVec */
1068
    for (i=0; i<c->vLumBufSize; i++) {
1069
        FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
1070
        c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1071
    }
1072
    for (i=0; i<c->vChrBufSize; i++) {
1073
        FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail);
1074
        c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize];
1075
    }
1076
    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1077
        for (i=0; i<c->vLumBufSize; i++) {
1078
            FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
1079
            c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1080
        }
1081

    
1082
    //try to avoid drawing green stuff between the right end and the stride end
1083
    for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2);
1084

    
1085
    assert(2*VOFW == VOF);
1086

    
1087
    assert(c->chrDstH <= dstH);
1088

    
1089
    if (flags&SWS_PRINT_INFO) {
1090
        if (flags&SWS_FAST_BILINEAR)
1091
            av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1092
        else if (flags&SWS_BILINEAR)
1093
            av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1094
        else if (flags&SWS_BICUBIC)
1095
            av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1096
        else if (flags&SWS_X)
1097
            av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1098
        else if (flags&SWS_POINT)
1099
            av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1100
        else if (flags&SWS_AREA)
1101
            av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1102
        else if (flags&SWS_BICUBLIN)
1103
            av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1104
        else if (flags&SWS_GAUSS)
1105
            av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1106
        else if (flags&SWS_SINC)
1107
            av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1108
        else if (flags&SWS_LANCZOS)
1109
            av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1110
        else if (flags&SWS_SPLINE)
1111
            av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1112
        else
1113
            av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1114

    
1115
        av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1116
               sws_format_name(srcFormat),
1117
#ifdef DITHER1XBPP
1118
               dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ? "dithered " : "",
1119
#else
1120
               "",
1121
#endif
1122
               sws_format_name(dstFormat));
1123

    
1124
        if (flags & SWS_CPU_CAPS_MMX2)
1125
            av_log(c, AV_LOG_INFO, "using MMX2\n");
1126
        else if (flags & SWS_CPU_CAPS_3DNOW)
1127
            av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1128
        else if (flags & SWS_CPU_CAPS_MMX)
1129
            av_log(c, AV_LOG_INFO, "using MMX\n");
1130
        else if (flags & SWS_CPU_CAPS_ALTIVEC)
1131
            av_log(c, AV_LOG_INFO, "using AltiVec\n");
1132
        else
1133
            av_log(c, AV_LOG_INFO, "using C\n");
1134

    
1135
        if (flags & SWS_CPU_CAPS_MMX) {
1136
            if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
1137
                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
1138
            else {
1139
                if (c->hLumFilterSize==4)
1140
                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
1141
                else if (c->hLumFilterSize==8)
1142
                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
1143
                else
1144
                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
1145

    
1146
                if (c->hChrFilterSize==4)
1147
                    av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
1148
                else if (c->hChrFilterSize==8)
1149
                    av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
1150
                else
1151
                    av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
1152
            }
1153
        } else {
1154
#if ARCH_X86
1155
            av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
1156
#else
1157
            if (flags & SWS_FAST_BILINEAR)
1158
                av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
1159
            else
1160
                av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
1161
#endif
1162
        }
1163
        if (isPlanarYUV(dstFormat)) {
1164
            if (c->vLumFilterSize==1)
1165
                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1166
            else
1167
                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1168
        } else {
1169
            if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
1170
                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
1171
                       "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1172
            else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
1173
                av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1174
            else
1175
                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1176
        }
1177

    
1178
        if (dstFormat==PIX_FMT_BGR24)
1179
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
1180
                   (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
1181
        else if (dstFormat==PIX_FMT_RGB32)
1182
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1183
        else if (dstFormat==PIX_FMT_BGR565)
1184
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1185
        else if (dstFormat==PIX_FMT_BGR555)
1186
            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
1187

    
1188
        av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1189
        av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1190
               c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1191
        av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1192
               c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1193
    }
1194

    
1195
    c->swScale= ff_getSwsFunc(c);
1196
    return c;
1197

    
1198
fail:
1199
    sws_freeContext(c);
1200
    return NULL;
1201
}
1202

    
1203
SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1204
                                float lumaSharpen, float chromaSharpen,
1205
                                float chromaHShift, float chromaVShift,
1206
                                int verbose)
1207
{
1208
    SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1209
    if (!filter)
1210
        return NULL;
1211

    
1212
    if (lumaGBlur!=0.0) {
1213
        filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1214
        filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1215
    } else {
1216
        filter->lumH= sws_getIdentityVec();
1217
        filter->lumV= sws_getIdentityVec();
1218
    }
1219

    
1220
    if (chromaGBlur!=0.0) {
1221
        filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1222
        filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1223
    } else {
1224
        filter->chrH= sws_getIdentityVec();
1225
        filter->chrV= sws_getIdentityVec();
1226
    }
1227

    
1228
    if (chromaSharpen!=0.0) {
1229
        SwsVector *id= sws_getIdentityVec();
1230
        sws_scaleVec(filter->chrH, -chromaSharpen);
1231
        sws_scaleVec(filter->chrV, -chromaSharpen);
1232
        sws_addVec(filter->chrH, id);
1233
        sws_addVec(filter->chrV, id);
1234
        sws_freeVec(id);
1235
    }
1236

    
1237
    if (lumaSharpen!=0.0) {
1238
        SwsVector *id= sws_getIdentityVec();
1239
        sws_scaleVec(filter->lumH, -lumaSharpen);
1240
        sws_scaleVec(filter->lumV, -lumaSharpen);
1241
        sws_addVec(filter->lumH, id);
1242
        sws_addVec(filter->lumV, id);
1243
        sws_freeVec(id);
1244
    }
1245

    
1246
    if (chromaHShift != 0.0)
1247
        sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1248

    
1249
    if (chromaVShift != 0.0)
1250
        sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1251

    
1252
    sws_normalizeVec(filter->chrH, 1.0);
1253
    sws_normalizeVec(filter->chrV, 1.0);
1254
    sws_normalizeVec(filter->lumH, 1.0);
1255
    sws_normalizeVec(filter->lumV, 1.0);
1256

    
1257
    if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1258
    if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1259

    
1260
    return filter;
1261
}
1262

    
1263
SwsVector *sws_allocVec(int length)
1264
{
1265
    SwsVector *vec = av_malloc(sizeof(SwsVector));
1266
    if (!vec)
1267
        return NULL;
1268
    vec->length = length;
1269
    vec->coeff  = av_malloc(sizeof(double) * length);
1270
    if (!vec->coeff)
1271
        av_freep(&vec);
1272
    return vec;
1273
}
1274

    
1275
SwsVector *sws_getGaussianVec(double variance, double quality)
1276
{
1277
    const int length= (int)(variance*quality + 0.5) | 1;
1278
    int i;
1279
    double middle= (length-1)*0.5;
1280
    SwsVector *vec= sws_allocVec(length);
1281

    
1282
    if (!vec)
1283
        return NULL;
1284

    
1285
    for (i=0; i<length; i++) {
1286
        double dist= i-middle;
1287
        vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1288
    }
1289

    
1290
    sws_normalizeVec(vec, 1.0);
1291

    
1292
    return vec;
1293
}
1294

    
1295
SwsVector *sws_getConstVec(double c, int length)
1296
{
1297
    int i;
1298
    SwsVector *vec= sws_allocVec(length);
1299

    
1300
    if (!vec)
1301
        return NULL;
1302

    
1303
    for (i=0; i<length; i++)
1304
        vec->coeff[i]= c;
1305

    
1306
    return vec;
1307
}
1308

    
1309
SwsVector *sws_getIdentityVec(void)
1310
{
1311
    return sws_getConstVec(1.0, 1);
1312
}
1313

    
1314
double sws_dcVec(SwsVector *a)
1315
{
1316
    int i;
1317
    double sum=0;
1318

    
1319
    for (i=0; i<a->length; i++)
1320
        sum+= a->coeff[i];
1321

    
1322
    return sum;
1323
}
1324

    
1325
void sws_scaleVec(SwsVector *a, double scalar)
1326
{
1327
    int i;
1328

    
1329
    for (i=0; i<a->length; i++)
1330
        a->coeff[i]*= scalar;
1331
}
1332

    
1333
void sws_normalizeVec(SwsVector *a, double height)
1334
{
1335
    sws_scaleVec(a, height/sws_dcVec(a));
1336
}
1337

    
1338
static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1339
{
1340
    int length= a->length + b->length - 1;
1341
    int i, j;
1342
    SwsVector *vec= sws_getConstVec(0.0, length);
1343

    
1344
    if (!vec)
1345
        return NULL;
1346

    
1347
    for (i=0; i<a->length; i++) {
1348
        for (j=0; j<b->length; j++) {
1349
            vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1350
        }
1351
    }
1352

    
1353
    return vec;
1354
}
1355

    
1356
static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1357
{
1358
    int length= FFMAX(a->length, b->length);
1359
    int i;
1360
    SwsVector *vec= sws_getConstVec(0.0, length);
1361

    
1362
    if (!vec)
1363
        return NULL;
1364

    
1365
    for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1366
    for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1367

    
1368
    return vec;
1369
}
1370

    
1371
static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1372
{
1373
    int length= FFMAX(a->length, b->length);
1374
    int i;
1375
    SwsVector *vec= sws_getConstVec(0.0, length);
1376

    
1377
    if (!vec)
1378
        return NULL;
1379

    
1380
    for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1381
    for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1382

    
1383
    return vec;
1384
}
1385

    
1386
/* shift left / or right if "shift" is negative */
1387
static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1388
{
1389
    int length= a->length + FFABS(shift)*2;
1390
    int i;
1391
    SwsVector *vec= sws_getConstVec(0.0, length);
1392

    
1393
    if (!vec)
1394
        return NULL;
1395

    
1396
    for (i=0; i<a->length; i++) {
1397
        vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1398
    }
1399

    
1400
    return vec;
1401
}
1402

    
1403
void sws_shiftVec(SwsVector *a, int shift)
1404
{
1405
    SwsVector *shifted= sws_getShiftedVec(a, shift);
1406
    av_free(a->coeff);
1407
    a->coeff= shifted->coeff;
1408
    a->length= shifted->length;
1409
    av_free(shifted);
1410
}
1411

    
1412
void sws_addVec(SwsVector *a, SwsVector *b)
1413
{
1414
    SwsVector *sum= sws_sumVec(a, b);
1415
    av_free(a->coeff);
1416
    a->coeff= sum->coeff;
1417
    a->length= sum->length;
1418
    av_free(sum);
1419
}
1420

    
1421
void sws_subVec(SwsVector *a, SwsVector *b)
1422
{
1423
    SwsVector *diff= sws_diffVec(a, b);
1424
    av_free(a->coeff);
1425
    a->coeff= diff->coeff;
1426
    a->length= diff->length;
1427
    av_free(diff);
1428
}
1429

    
1430
void sws_convVec(SwsVector *a, SwsVector *b)
1431
{
1432
    SwsVector *conv= sws_getConvVec(a, b);
1433
    av_free(a->coeff);
1434
    a->coeff= conv->coeff;
1435
    a->length= conv->length;
1436
    av_free(conv);
1437
}
1438

    
1439
SwsVector *sws_cloneVec(SwsVector *a)
1440
{
1441
    int i;
1442
    SwsVector *vec= sws_allocVec(a->length);
1443

    
1444
    if (!vec)
1445
        return NULL;
1446

    
1447
    for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1448

    
1449
    return vec;
1450
}
1451

    
1452
void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1453
{
1454
    int i;
1455
    double max=0;
1456
    double min=0;
1457
    double range;
1458

    
1459
    for (i=0; i<a->length; i++)
1460
        if (a->coeff[i]>max) max= a->coeff[i];
1461

    
1462
    for (i=0; i<a->length; i++)
1463
        if (a->coeff[i]<min) min= a->coeff[i];
1464

    
1465
    range= max - min;
1466

    
1467
    for (i=0; i<a->length; i++) {
1468
        int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1469
        av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1470
        for (;x>0; x--) av_log(log_ctx, log_level, " ");
1471
        av_log(log_ctx, log_level, "|\n");
1472
    }
1473
}
1474

    
1475
#if LIBSWSCALE_VERSION_MAJOR < 1
1476
void sws_printVec(SwsVector *a)
1477
{
1478
    sws_printVec2(a, NULL, AV_LOG_DEBUG);
1479
}
1480
#endif
1481

    
1482
void sws_freeVec(SwsVector *a)
1483
{
1484
    if (!a) return;
1485
    av_freep(&a->coeff);
1486
    a->length=0;
1487
    av_free(a);
1488
}
1489

    
1490
void sws_freeFilter(SwsFilter *filter)
1491
{
1492
    if (!filter) return;
1493

    
1494
    if (filter->lumH) sws_freeVec(filter->lumH);
1495
    if (filter->lumV) sws_freeVec(filter->lumV);
1496
    if (filter->chrH) sws_freeVec(filter->chrH);
1497
    if (filter->chrV) sws_freeVec(filter->chrV);
1498
    av_free(filter);
1499
}
1500

    
1501
void sws_freeContext(SwsContext *c)
1502
{
1503
    int i;
1504
    if (!c) return;
1505

    
1506
    if (c->lumPixBuf) {
1507
        for (i=0; i<c->vLumBufSize; i++)
1508
            av_freep(&c->lumPixBuf[i]);
1509
        av_freep(&c->lumPixBuf);
1510
    }
1511

    
1512
    if (c->chrPixBuf) {
1513
        for (i=0; i<c->vChrBufSize; i++)
1514
            av_freep(&c->chrPixBuf[i]);
1515
        av_freep(&c->chrPixBuf);
1516
    }
1517

    
1518
    if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1519
        for (i=0; i<c->vLumBufSize; i++)
1520
            av_freep(&c->alpPixBuf[i]);
1521
        av_freep(&c->alpPixBuf);
1522
    }
1523

    
1524
    av_freep(&c->vLumFilter);
1525
    av_freep(&c->vChrFilter);
1526
    av_freep(&c->hLumFilter);
1527
    av_freep(&c->hChrFilter);
1528
#if ARCH_PPC && (HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT)
1529
    av_freep(&c->vYCoeffsBank);
1530
    av_freep(&c->vCCoeffsBank);
1531
#endif
1532

    
1533
    av_freep(&c->vLumFilterPos);
1534
    av_freep(&c->vChrFilterPos);
1535
    av_freep(&c->hLumFilterPos);
1536
    av_freep(&c->hChrFilterPos);
1537

    
1538
#if ARCH_X86 && CONFIG_GPL
1539
#ifdef MAP_ANONYMOUS
1540
    if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1541
    if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1542
#elif HAVE_VIRTUALALLOC
1543
    if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, MEM_RELEASE);
1544
    if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, MEM_RELEASE);
1545
#else
1546
    av_free(c->lumMmx2FilterCode);
1547
    av_free(c->chrMmx2FilterCode);
1548
#endif
1549
    c->lumMmx2FilterCode=NULL;
1550
    c->chrMmx2FilterCode=NULL;
1551
#endif /* ARCH_X86 && CONFIG_GPL */
1552

    
1553
    av_freep(&c->yuvTable);
1554

    
1555
    av_free(c);
1556
}
1557

    
1558
struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1559
                                        int srcW, int srcH, enum PixelFormat srcFormat,
1560
                                        int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1561
                                        SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1562
{
1563
    static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1564

    
1565
    if (!param)
1566
        param = default_param;
1567

    
1568
    if (context &&
1569
        (context->srcW      != srcW      ||
1570
         context->srcH      != srcH      ||
1571
         context->srcFormat != srcFormat ||
1572
         context->dstW      != dstW      ||
1573
         context->dstH      != dstH      ||
1574
         context->dstFormat != dstFormat ||
1575
         context->flags     != flags     ||
1576
         context->param[0]  != param[0]  ||
1577
         context->param[1]  != param[1])) {
1578
        sws_freeContext(context);
1579
        context = NULL;
1580
    }
1581

    
1582
    if (!context) {
1583
        return sws_getContext(srcW, srcH, srcFormat,
1584
                              dstW, dstH, dstFormat, flags,
1585
                              srcFilter, dstFilter, param);
1586
    }
1587
    return context;
1588
}
1589