Statistics
| Branch: | Revision:

ffmpeg / libavcodec / sonic.c @ 5509bffa

History | View | Annotate | Download (24.3 KB)

1
/*
2
 * Simple free lossless/lossy audio codec
3
 * Copyright (c) 2004 Alex Beregszaszi
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
 */
19
#include "avcodec.h"
20
#include "bitstream.h"
21
#include "golomb.h"
22

    
23
/**
24
 * @file sonic.c
25
 * Simple free lossless/lossy audio codec
26
 * Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk)
27
 * Written and designed by Alex Beregszaszi
28
 *
29
 * TODO:
30
 *  - CABAC put/get_symbol
31
 *  - independent quantizer for channels
32
 *  - >2 channels support
33
 *  - more decorrelation types
34
 *  - more tap_quant tests
35
 *  - selectable intlist writers/readers (bonk-style, golomb, cabac)
36
 */
37

    
38
#define MAX_CHANNELS 2
39

    
40
#define MID_SIDE 0
41
#define LEFT_SIDE 1
42
#define RIGHT_SIDE 2
43

    
44
typedef struct SonicContext {
45
    int lossless, decorrelation;
46

    
47
    int num_taps, downsampling;
48
    double quantization;
49

    
50
    int channels, samplerate, block_align, frame_size;
51

    
52
    int *tap_quant;
53
    int *int_samples;
54
    int *coded_samples[MAX_CHANNELS];
55

    
56
    // for encoding
57
    int *tail;
58
    int tail_size;
59
    int *window;
60
    int window_size;
61

    
62
    // for decoding
63
    int *predictor_k;
64
    int *predictor_state[MAX_CHANNELS];
65
} SonicContext;
66

    
67
#define LATTICE_SHIFT   10
68
#define SAMPLE_SHIFT    4
69
#define LATTICE_FACTOR  (1 << LATTICE_SHIFT)
70
#define SAMPLE_FACTOR   (1 << SAMPLE_SHIFT)
71

    
72
#define BASE_QUANT      0.6
73
#define RATE_VARIATION  3.0
74

    
75
static inline int divide(int a, int b)
76
{
77
    if (a < 0)
78
        return -( (-a + b/2)/b );
79
    else
80
        return (a + b/2)/b;
81
}
82

    
83
static inline int shift(int a,int b)
84
{
85
    return (a+(1<<(b-1))) >> b;
86
}
87

    
88
static inline int shift_down(int a,int b)
89
{
90
    return (a>>b)+((a<0)?1:0);
91
}
92

    
93
#if 1
94
static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
95
{
96
    int i;
97

    
98
    for (i = 0; i < entries; i++)
99
        set_se_golomb(pb, buf[i]);
100

    
101
    return 1;
102
}
103

    
104
static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
105
{
106
    int i;
107

    
108
    for (i = 0; i < entries; i++)
109
        buf[i] = get_se_golomb(gb);
110

    
111
    return 1;
112
}
113

    
114
#else
115

    
116
#define ADAPT_LEVEL 8
117

    
118
static int bits_to_store(uint64_t x)
119
{
120
    int res = 0;
121

    
122
    while(x)
123
    {
124
        res++;
125
        x >>= 1;
126
    }
127
    return res;
128
}
129

    
130
static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
131
{
132
    int i, bits;
133

    
134
    if (!max)
135
        return;
136

    
137
    bits = bits_to_store(max);
138

    
139
    for (i = 0; i < bits-1; i++)
140
        put_bits(pb, 1, value & (1 << i));
141

    
142
    if ( (value | (1 << (bits-1))) <= max)
143
        put_bits(pb, 1, value & (1 << (bits-1)));
144
}
145

    
146
static unsigned int read_uint_max(GetBitContext *gb, int max)
147
{
148
    int i, bits, value = 0;
149

    
150
    if (!max)
151
        return 0;
152

    
153
    bits = bits_to_store(max);
154

    
155
    for (i = 0; i < bits-1; i++)
156
        if (get_bits1(gb))
157
            value += 1 << i;
158

    
159
    if ( (value | (1<<(bits-1))) <= max)
160
        if (get_bits1(gb))
161
            value += 1 << (bits-1);
162

    
163
    return value;
164
}
165

    
166
static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
167
{
168
    int i, j, x = 0, low_bits = 0, max = 0;
169
    int step = 256, pos = 0, dominant = 0, any = 0;
170
    int *copy, *bits;
171

    
172
    copy = av_mallocz(4* entries);
173
    if (!copy)
174
        return -1;
175

    
176
    if (base_2_part)
177
    {
178
        int energy = 0;
179

    
180
        for (i = 0; i < entries; i++)
181
            energy += abs(buf[i]);
182

    
183
        low_bits = bits_to_store(energy / (entries * 2));
184
        if (low_bits > 15)
185
            low_bits = 15;
186

    
187
        put_bits(pb, 4, low_bits);
188
    }
189

    
190
    for (i = 0; i < entries; i++)
191
    {
192
        put_bits(pb, low_bits, abs(buf[i]));
193
        copy[i] = abs(buf[i]) >> low_bits;
194
        if (copy[i] > max)
195
            max = abs(copy[i]);
196
    }
197

    
198
    bits = av_mallocz(4* entries*max);
199
    if (!bits)
200
    {
201
//        av_free(copy);
202
        return -1;
203
    }
204

    
205
    for (i = 0; i <= max; i++)
206
    {
207
        for (j = 0; j < entries; j++)
208
            if (copy[j] >= i)
209
                bits[x++] = copy[j] > i;
210
    }
211

    
212
    // store bitstream
213
    while (pos < x)
214
    {
215
        int steplet = step >> 8;
216

    
217
        if (pos + steplet > x)
218
            steplet = x - pos;
219

    
220
        for (i = 0; i < steplet; i++)
221
            if (bits[i+pos] != dominant)
222
                any = 1;
223

    
224
        put_bits(pb, 1, any);
225

    
226
        if (!any)
227
        {
228
            pos += steplet;
229
            step += step / ADAPT_LEVEL;
230
        }
231
        else
232
        {
233
            int interloper = 0;
234

    
235
            while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
236
                interloper++;
237

    
238
            // note change
239
            write_uint_max(pb, interloper, (step >> 8) - 1);
240

    
241
            pos += interloper + 1;
242
            step -= step / ADAPT_LEVEL;
243
        }
244

    
245
        if (step < 256)
246
        {
247
            step = 65536 / step;
248
            dominant = !dominant;
249
        }
250
    }
251

    
252
    // store signs
253
    for (i = 0; i < entries; i++)
254
        if (buf[i])
255
            put_bits(pb, 1, buf[i] < 0);
256

    
257
//    av_free(bits);
258
//    av_free(copy);
259

    
260
    return 0;
261
}
262

    
263
static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
264
{
265
    int i, low_bits = 0, x = 0;
266
    int n_zeros = 0, step = 256, dominant = 0;
267
    int pos = 0, level = 0;
268
    int *bits = av_mallocz(4* entries);
269

    
270
    if (!bits)
271
        return -1;
272

    
273
    if (base_2_part)
274
    {
275
        low_bits = get_bits(gb, 4);
276

    
277
        if (low_bits)
278
            for (i = 0; i < entries; i++)
279
                buf[i] = get_bits(gb, low_bits);
280
    }
281

    
282
//    av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);
283

    
284
    while (n_zeros < entries)
285
    {
286
        int steplet = step >> 8;
287

    
288
        if (!get_bits1(gb))
289
        {
290
            for (i = 0; i < steplet; i++)
291
                bits[x++] = dominant;
292

    
293
            if (!dominant)
294
                n_zeros += steplet;
295

    
296
            step += step / ADAPT_LEVEL;
297
        }
298
        else
299
        {
300
            int actual_run = read_uint_max(gb, steplet-1);
301

    
302
//            av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run);
303

    
304
            for (i = 0; i < actual_run; i++)
305
                bits[x++] = dominant;
306

    
307
            bits[x++] = !dominant;
308

    
309
            if (!dominant)
310
                n_zeros += actual_run;
311
            else
312
                n_zeros++;
313

    
314
            step -= step / ADAPT_LEVEL;
315
        }
316

    
317
        if (step < 256)
318
        {
319
            step = 65536 / step;
320
            dominant = !dominant;
321
        }
322
    }
323

    
324
    // reconstruct unsigned values
325
    n_zeros = 0;
326
    for (i = 0; n_zeros < entries; i++)
327
    {
328
        while(1)
329
        {
330
            if (pos >= entries)
331
            {
332
                pos = 0;
333
                level += 1 << low_bits;
334
            }
335

    
336
            if (buf[pos] >= level)
337
                break;
338

    
339
            pos++;
340
        }
341

    
342
        if (bits[i])
343
            buf[pos] += 1 << low_bits;
344
        else
345
            n_zeros++;
346

    
347
        pos++;
348
    }
349
//    av_free(bits);
350

    
351
    // read signs
352
    for (i = 0; i < entries; i++)
353
        if (buf[i] && get_bits1(gb))
354
            buf[i] = -buf[i];
355

    
356
//    av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);
357

    
358
    return 0;
359
}
360
#endif
361

    
362
static void predictor_init_state(int *k, int *state, int order)
363
{
364
    int i;
365

    
366
    for (i = order-2; i >= 0; i--)
367
    {
368
        int j, p, x = state[i];
369

    
370
        for (j = 0, p = i+1; p < order; j++,p++)
371
            {
372
            int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
373
            state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
374
            x = tmp;
375
        }
376
    }
377
}
378

    
379
static int predictor_calc_error(int *k, int *state, int order, int error)
380
{
381
    int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);
382

    
383
#if 1
384
    int *k_ptr = &(k[order-2]),
385
        *state_ptr = &(state[order-2]);
386
    for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
387
    {
388
        int k_value = *k_ptr, state_value = *state_ptr;
389
        x -= shift_down(k_value * state_value, LATTICE_SHIFT);
390
        state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT);
391
    }
392
#else
393
    for (i = order-2; i >= 0; i--)
394
    {
395
        x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
396
        state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT);
397
    }
398
#endif
399

    
400
    // don't drift too far, to avoid overflows
401
    if (x >  (SAMPLE_FACTOR<<16)) x =  (SAMPLE_FACTOR<<16);
402
    if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);
403

    
404
    state[0] = x;
405

    
406
    return x;
407
}
408

    
409
// Heavily modified Levinson-Durbin algorithm which
410
// copes better with quantization, and calculates the
411
// actual whitened result as it goes.
412

    
413
static void modified_levinson_durbin(int *window, int window_entries,
414
        int *out, int out_entries, int channels, int *tap_quant)
415
{
416
    int i;
417
    int *state = av_mallocz(4* window_entries);
418

    
419
    memcpy(state, window, 4* window_entries);
420

    
421
    for (i = 0; i < out_entries; i++)
422
    {
423
        int step = (i+1)*channels, k, j;
424
        double xx = 0.0, xy = 0.0;
425
#if 1
426
        int *x_ptr = &(window[step]), *state_ptr = &(state[0]);
427
        j = window_entries - step;
428
        for (;j>=0;j--,x_ptr++,state_ptr++)
429
        {
430
            double x_value = *x_ptr, state_value = *state_ptr;
431
            xx += state_value*state_value;
432
            xy += x_value*state_value;
433
        }
434
#else
435
        for (j = 0; j <= (window_entries - step); j++);
436
        {
437
            double stepval = window[step+j], stateval = window[j];
438
//            xx += (double)window[j]*(double)window[j];
439
//            xy += (double)window[step+j]*(double)window[j];
440
            xx += stateval*stateval;
441
            xy += stepval*stateval;
442
        }
443
#endif
444
        if (xx == 0.0)
445
            k = 0;
446
        else
447
            k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5));
448

    
449
        if (k > (LATTICE_FACTOR/tap_quant[i]))
450
            k = LATTICE_FACTOR/tap_quant[i];
451
        if (-k > (LATTICE_FACTOR/tap_quant[i]))
452
            k = -(LATTICE_FACTOR/tap_quant[i]);
453

    
454
        out[i] = k;
455
        k *= tap_quant[i];
456

    
457
#if 1
458
        x_ptr = &(window[step]);
459
        state_ptr = &(state[0]);
460
        j = window_entries - step;
461
        for (;j>=0;j--,x_ptr++,state_ptr++)
462
        {
463
            int x_value = *x_ptr, state_value = *state_ptr;
464
            *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
465
            *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
466
        }
467
#else
468
        for (j=0; j <= (window_entries - step); j++)
469
        {
470
            int stepval = window[step+j], stateval=state[j];
471
            window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
472
            state[j] += shift_down(k * stepval, LATTICE_SHIFT);
473
        }
474
#endif
475
    }
476

    
477
    av_free(state);
478
}
479

    
480
static int samplerate_table[] =
481
    { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };
482

    
483
#ifdef CONFIG_ENCODERS
484

    
485
static inline int code_samplerate(int samplerate)
486
{
487
    switch (samplerate)
488
    {
489
        case 44100: return 0;
490
        case 22050: return 1;
491
        case 11025: return 2;
492
        case 96000: return 3;
493
        case 48000: return 4;
494
        case 32000: return 5;
495
        case 24000: return 6;
496
        case 16000: return 7;
497
        case 8000: return 8;
498
    }
499
    return -1;
500
}
501

    
502
static int sonic_encode_init(AVCodecContext *avctx)
503
{
504
    SonicContext *s = avctx->priv_data;
505
    PutBitContext pb;
506
    int i, version = 0;
507

    
508
    if (avctx->channels > MAX_CHANNELS)
509
    {
510
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
511
        return -1; /* only stereo or mono for now */
512
    }
513

    
514
    if (avctx->channels == 2)
515
        s->decorrelation = MID_SIDE;
516

    
517
    if (avctx->codec->id == CODEC_ID_SONIC_LS)
518
    {
519
        s->lossless = 1;
520
        s->num_taps = 32;
521
        s->downsampling = 1;
522
        s->quantization = 0.0;
523
    }
524
    else
525
    {
526
        s->num_taps = 128;
527
        s->downsampling = 2;
528
        s->quantization = 1.0;
529
    }
530

    
531
    // max tap 2048
532
    if ((s->num_taps < 32) || (s->num_taps > 1024) ||
533
        ((s->num_taps>>5)<<5 != s->num_taps))
534
    {
535
        av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n");
536
        return -1;
537
    }
538

    
539
    // generate taps
540
    s->tap_quant = av_mallocz(4* s->num_taps);
541
    for (i = 0; i < s->num_taps; i++)
542
        s->tap_quant[i] = (int)(sqrt(i+1));
543

    
544
    s->channels = avctx->channels;
545
    s->samplerate = avctx->sample_rate;
546

    
547
    s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling;
548
    s->frame_size = s->channels*s->block_align*s->downsampling;
549

    
550
    s->tail = av_mallocz(4* s->num_taps*s->channels);
551
    if (!s->tail)
552
        return -1;
553
    s->tail_size = s->num_taps*s->channels;
554

    
555
    s->predictor_k = av_mallocz(4 * s->num_taps);
556
    if (!s->predictor_k)
557
        return -1;
558

    
559
    for (i = 0; i < s->channels; i++)
560
    {
561
        s->coded_samples[i] = av_mallocz(4* s->block_align);
562
        if (!s->coded_samples[i])
563
            return -1;
564
    }
565

    
566
    s->int_samples = av_mallocz(4* s->frame_size);
567

    
568
    s->window_size = ((2*s->tail_size)+s->frame_size);
569
    s->window = av_mallocz(4* s->window_size);
570
    if (!s->window)
571
        return -1;
572

    
573
    avctx->extradata = av_mallocz(16);
574
    if (!avctx->extradata)
575
        return -1;
576
    init_put_bits(&pb, avctx->extradata, 16*8);
577

    
578
    put_bits(&pb, 2, version); // version
579
    if (version == 1)
580
    {
581
        put_bits(&pb, 2, s->channels);
582
        put_bits(&pb, 4, code_samplerate(s->samplerate));
583
    }
584
    put_bits(&pb, 1, s->lossless);
585
    if (!s->lossless)
586
        put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
587
    put_bits(&pb, 2, s->decorrelation);
588
    put_bits(&pb, 2, s->downsampling);
589
    put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
590
    put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table
591

    
592
    flush_put_bits(&pb);
593
    avctx->extradata_size = put_bits_count(&pb)/8;
594

    
595
    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
596
        version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
597

    
598
    avctx->coded_frame = avcodec_alloc_frame();
599
    if (!avctx->coded_frame)
600
        return -ENOMEM;
601
    avctx->coded_frame->key_frame = 1;
602
    avctx->frame_size = s->block_align*s->downsampling;
603

    
604
    return 0;
605
}
606

    
607
static int sonic_encode_close(AVCodecContext *avctx)
608
{
609
    SonicContext *s = avctx->priv_data;
610
    int i;
611

    
612
    av_freep(&avctx->coded_frame);
613

    
614
    for (i = 0; i < s->channels; i++)
615
        av_free(s->coded_samples[i]);
616

    
617
    av_free(s->predictor_k);
618
    av_free(s->tail);
619
    av_free(s->tap_quant);
620
    av_free(s->window);
621
    av_free(s->int_samples);
622

    
623
    return 0;
624
}
625

    
626
static int sonic_encode_frame(AVCodecContext *avctx,
627
                            uint8_t *buf, int buf_size, void *data)
628
{
629
    SonicContext *s = avctx->priv_data;
630
    PutBitContext pb;
631
    int i, j, ch, quant = 0, x = 0;
632
    short *samples = data;
633

    
634
    init_put_bits(&pb, buf, buf_size*8);
635

    
636
    // short -> internal
637
    for (i = 0; i < s->frame_size; i++)
638
        s->int_samples[i] = samples[i];
639

    
640
    if (!s->lossless)
641
        for (i = 0; i < s->frame_size; i++)
642
            s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;
643

    
644
    switch(s->decorrelation)
645
    {
646
        case MID_SIDE:
647
            for (i = 0; i < s->frame_size; i += s->channels)
648
            {
649
                s->int_samples[i] += s->int_samples[i+1];
650
                s->int_samples[i+1] -= shift(s->int_samples[i], 1);
651
            }
652
            break;
653
        case LEFT_SIDE:
654
            for (i = 0; i < s->frame_size; i += s->channels)
655
                s->int_samples[i+1] -= s->int_samples[i];
656
            break;
657
        case RIGHT_SIDE:
658
            for (i = 0; i < s->frame_size; i += s->channels)
659
                s->int_samples[i] -= s->int_samples[i+1];
660
            break;
661
    }
662

    
663
    memset(s->window, 0, 4* s->window_size);
664

    
665
    for (i = 0; i < s->tail_size; i++)
666
        s->window[x++] = s->tail[i];
667

    
668
    for (i = 0; i < s->frame_size; i++)
669
        s->window[x++] = s->int_samples[i];
670

    
671
    for (i = 0; i < s->tail_size; i++)
672
        s->window[x++] = 0;
673

    
674
    for (i = 0; i < s->tail_size; i++)
675
        s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i];
676

    
677
    // generate taps
678
    modified_levinson_durbin(s->window, s->window_size,
679
                s->predictor_k, s->num_taps, s->channels, s->tap_quant);
680
    if (intlist_write(&pb, s->predictor_k, s->num_taps, 0) < 0)
681
        return -1;
682

    
683
    for (ch = 0; ch < s->channels; ch++)
684
    {
685
        x = s->tail_size+ch;
686
        for (i = 0; i < s->block_align; i++)
687
        {
688
            int sum = 0;
689
            for (j = 0; j < s->downsampling; j++, x += s->channels)
690
                sum += s->window[x];
691
            s->coded_samples[ch][i] = sum;
692
        }
693
    }
694

    
695
    // simple rate control code
696
    if (!s->lossless)
697
    {
698
        double energy1 = 0.0, energy2 = 0.0;
699
        for (ch = 0; ch < s->channels; ch++)
700
        {
701
            for (i = 0; i < s->block_align; i++)
702
            {
703
                double sample = s->coded_samples[ch][i];
704
                energy2 += sample*sample;
705
                energy1 += fabs(sample);
706
            }
707
        }
708

    
709
        energy2 = sqrt(energy2/(s->channels*s->block_align));
710
        energy1 = sqrt(2.0)*energy1/(s->channels*s->block_align);
711

    
712
        // increase bitrate when samples are like a gaussian distribution
713
        // reduce bitrate when samples are like a two-tailed exponential distribution
714

    
715
        if (energy2 > energy1)
716
            energy2 += (energy2-energy1)*RATE_VARIATION;
717

    
718
        quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR);
719
//        av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2);
720

    
721
        if (quant < 1)
722
            quant = 1;
723
        if (quant > 65535)
724
            quant = 65535;
725

    
726
        set_ue_golomb(&pb, quant);
727

    
728
        quant *= SAMPLE_FACTOR;
729
    }
730

    
731
    // write out coded samples
732
    for (ch = 0; ch < s->channels; ch++)
733
    {
734
        if (!s->lossless)
735
            for (i = 0; i < s->block_align; i++)
736
                s->coded_samples[ch][i] = divide(s->coded_samples[ch][i], quant);
737

    
738
        if (intlist_write(&pb, s->coded_samples[ch], s->block_align, 1) < 0)
739
            return -1;
740
    }
741

    
742
//    av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8);
743

    
744
    flush_put_bits(&pb);
745
    return (put_bits_count(&pb)+7)/8;
746
}
747
#endif //CONFIG_ENCODERS
748

    
749
static int sonic_decode_init(AVCodecContext *avctx)
750
{
751
    SonicContext *s = avctx->priv_data;
752
    GetBitContext gb;
753
    int i, version;
754

    
755
    s->channels = avctx->channels;
756
    s->samplerate = avctx->sample_rate;
757

    
758
    if (!avctx->extradata)
759
    {
760
        av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n");
761
        return -1;
762
    }
763

    
764
    init_get_bits(&gb, avctx->extradata, avctx->extradata_size);
765

    
766
    version = get_bits(&gb, 2);
767
    if (version > 1)
768
    {
769
        av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n");
770
        return -1;
771
    }
772

    
773
    if (version == 1)
774
    {
775
        s->channels = get_bits(&gb, 2);
776
        s->samplerate = samplerate_table[get_bits(&gb, 4)];
777
        av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n",
778
            s->channels, s->samplerate);
779
    }
780

    
781
    if (s->channels > MAX_CHANNELS)
782
    {
783
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
784
        return -1;
785
    }
786

    
787
    s->lossless = get_bits1(&gb);
788
    if (!s->lossless)
789
        skip_bits(&gb, 3); // XXX FIXME
790
    s->decorrelation = get_bits(&gb, 2);
791

    
792
    s->downsampling = get_bits(&gb, 2);
793
    s->num_taps = (get_bits(&gb, 5)+1)<<5;
794
    if (get_bits1(&gb)) // XXX FIXME
795
        av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
796

    
797
    s->block_align = (int)(2048.0*(s->samplerate/44100))/s->downsampling;
798
    s->frame_size = s->channels*s->block_align*s->downsampling;
799
//    avctx->frame_size = s->block_align;
800

    
801
    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
802
        version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
803

    
804
    // generate taps
805
    s->tap_quant = av_mallocz(4* s->num_taps);
806
    for (i = 0; i < s->num_taps; i++)
807
        s->tap_quant[i] = (int)(sqrt(i+1));
808

    
809
    s->predictor_k = av_mallocz(4* s->num_taps);
810

    
811
    for (i = 0; i < s->channels; i++)
812
    {
813
        s->predictor_state[i] = av_mallocz(4* s->num_taps);
814
        if (!s->predictor_state[i])
815
            return -1;
816
    }
817

    
818
    for (i = 0; i < s->channels; i++)
819
    {
820
        s->coded_samples[i] = av_mallocz(4* s->block_align);
821
        if (!s->coded_samples[i])
822
            return -1;
823
    }
824
    s->int_samples = av_mallocz(4* s->frame_size);
825

    
826
    return 0;
827
}
828

    
829
static int sonic_decode_close(AVCodecContext *avctx)
830
{
831
    SonicContext *s = avctx->priv_data;
832
    int i;
833

    
834
    av_free(s->int_samples);
835
    av_free(s->tap_quant);
836
    av_free(s->predictor_k);
837

    
838
    for (i = 0; i < s->channels; i++)
839
    {
840
        av_free(s->predictor_state[i]);
841
        av_free(s->coded_samples[i]);
842
    }
843

    
844
    return 0;
845
}
846

    
847
static int sonic_decode_frame(AVCodecContext *avctx,
848
                            void *data, int *data_size,
849
                            uint8_t *buf, int buf_size)
850
{
851
    SonicContext *s = avctx->priv_data;
852
    GetBitContext gb;
853
    int i, quant, ch, j;
854
    short *samples = data;
855

    
856
    if (buf_size == 0) return 0;
857

    
858
//    av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size);
859

    
860
    init_get_bits(&gb, buf, buf_size*8);
861

    
862
    intlist_read(&gb, s->predictor_k, s->num_taps, 0);
863

    
864
    // dequantize
865
    for (i = 0; i < s->num_taps; i++)
866
        s->predictor_k[i] *= s->tap_quant[i];
867

    
868
    if (s->lossless)
869
        quant = 1;
870
    else
871
        quant = get_ue_golomb(&gb) * SAMPLE_FACTOR;
872

    
873
//    av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant);
874

    
875
    for (ch = 0; ch < s->channels; ch++)
876
    {
877
        int x = ch;
878

    
879
        predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps);
880

    
881
        intlist_read(&gb, s->coded_samples[ch], s->block_align, 1);
882

    
883
        for (i = 0; i < s->block_align; i++)
884
        {
885
            for (j = 0; j < s->downsampling - 1; j++)
886
            {
887
                s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0);
888
                x += s->channels;
889
            }
890

    
891
            s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant);
892
            x += s->channels;
893
        }
894

    
895
        for (i = 0; i < s->num_taps; i++)
896
            s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
897
    }
898

    
899
    switch(s->decorrelation)
900
    {
901
        case MID_SIDE:
902
            for (i = 0; i < s->frame_size; i += s->channels)
903
            {
904
                s->int_samples[i+1] += shift(s->int_samples[i], 1);
905
                s->int_samples[i] -= s->int_samples[i+1];
906
            }
907
            break;
908
        case LEFT_SIDE:
909
            for (i = 0; i < s->frame_size; i += s->channels)
910
                s->int_samples[i+1] += s->int_samples[i];
911
            break;
912
        case RIGHT_SIDE:
913
            for (i = 0; i < s->frame_size; i += s->channels)
914
                s->int_samples[i] += s->int_samples[i+1];
915
            break;
916
    }
917

    
918
    if (!s->lossless)
919
        for (i = 0; i < s->frame_size; i++)
920
            s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);
921

    
922
    // internal -> short
923
    for (i = 0; i < s->frame_size; i++)
924
    {
925
        if (s->int_samples[i] > 32767)
926
            samples[i] = 32767;
927
        else if (s->int_samples[i] < -32768)
928
            samples[i] = -32768;
929
        else
930
            samples[i] = s->int_samples[i];
931
    }
932

    
933
    align_get_bits(&gb);
934

    
935
    *data_size = s->frame_size * 2;
936

    
937
    return (get_bits_count(&gb)+7)/8;
938
}
939

    
940
#ifdef CONFIG_ENCODERS
941
AVCodec sonic_encoder = {
942
    "sonic",
943
    CODEC_TYPE_AUDIO,
944
    CODEC_ID_SONIC,
945
    sizeof(SonicContext),
946
    sonic_encode_init,
947
    sonic_encode_frame,
948
    sonic_encode_close,
949
    NULL,
950
};
951

    
952
AVCodec sonic_ls_encoder = {
953
    "sonicls",
954
    CODEC_TYPE_AUDIO,
955
    CODEC_ID_SONIC_LS,
956
    sizeof(SonicContext),
957
    sonic_encode_init,
958
    sonic_encode_frame,
959
    sonic_encode_close,
960
    NULL,
961
};
962
#endif
963

    
964
#ifdef CONFIG_DECODERS
965
AVCodec sonic_decoder = {
966
    "sonic",
967
    CODEC_TYPE_AUDIO,
968
    CODEC_ID_SONIC,
969
    sizeof(SonicContext),
970
    sonic_decode_init,
971
    NULL,
972
    sonic_decode_close,
973
    sonic_decode_frame,
974
};
975
#endif