Statistics
| Branch: | Revision:

ffmpeg / libavcodec / sonic.c @ 72415b2a

History | View | Annotate | Download (24.6 KB)

1
/*
2
 * Simple free lossless/lossy audio codec
3
 * Copyright (c) 2004 Alex Beregszaszi
4
 *
5
 * This file is part of FFmpeg.
6
 *
7
 * FFmpeg is free software; you can redistribute it and/or
8
 * modify it under the terms of the GNU Lesser General Public
9
 * License as published by the Free Software Foundation; either
10
 * version 2.1 of the License, or (at your option) any later version.
11
 *
12
 * FFmpeg is distributed in the hope that it will be useful,
13
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15
 * Lesser General Public License for more details.
16
 *
17
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with FFmpeg; if not, write to the Free Software
19
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20
 */
21
#include "avcodec.h"
22
#include "get_bits.h"
23
#include "golomb.h"
24

    
25
/**
26
 * @file libavcodec/sonic.c
27
 * Simple free lossless/lossy audio codec
28
 * Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk)
29
 * Written and designed by Alex Beregszaszi
30
 *
31
 * TODO:
32
 *  - CABAC put/get_symbol
33
 *  - independent quantizer for channels
34
 *  - >2 channels support
35
 *  - more decorrelation types
36
 *  - more tap_quant tests
37
 *  - selectable intlist writers/readers (bonk-style, golomb, cabac)
38
 */
39

    
40
#define MAX_CHANNELS 2
41

    
42
#define MID_SIDE 0
43
#define LEFT_SIDE 1
44
#define RIGHT_SIDE 2
45

    
46
typedef struct SonicContext {
47
    int lossless, decorrelation;
48

    
49
    int num_taps, downsampling;
50
    double quantization;
51

    
52
    int channels, samplerate, block_align, frame_size;
53

    
54
    int *tap_quant;
55
    int *int_samples;
56
    int *coded_samples[MAX_CHANNELS];
57

    
58
    // for encoding
59
    int *tail;
60
    int tail_size;
61
    int *window;
62
    int window_size;
63

    
64
    // for decoding
65
    int *predictor_k;
66
    int *predictor_state[MAX_CHANNELS];
67
} SonicContext;
68

    
69
#define LATTICE_SHIFT   10
70
#define SAMPLE_SHIFT    4
71
#define LATTICE_FACTOR  (1 << LATTICE_SHIFT)
72
#define SAMPLE_FACTOR   (1 << SAMPLE_SHIFT)
73

    
74
#define BASE_QUANT      0.6
75
#define RATE_VARIATION  3.0
76

    
77
static inline int divide(int a, int b)
78
{
79
    if (a < 0)
80
        return -( (-a + b/2)/b );
81
    else
82
        return (a + b/2)/b;
83
}
84

    
85
static inline int shift(int a,int b)
86
{
87
    return (a+(1<<(b-1))) >> b;
88
}
89

    
90
static inline int shift_down(int a,int b)
91
{
92
    return (a>>b)+((a<0)?1:0);
93
}
94

    
95
#if 1
96
static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
97
{
98
    int i;
99

    
100
    for (i = 0; i < entries; i++)
101
        set_se_golomb(pb, buf[i]);
102

    
103
    return 1;
104
}
105

    
106
static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
107
{
108
    int i;
109

    
110
    for (i = 0; i < entries; i++)
111
        buf[i] = get_se_golomb(gb);
112

    
113
    return 1;
114
}
115

    
116
#else
117

    
118
#define ADAPT_LEVEL 8
119

    
120
static int bits_to_store(uint64_t x)
121
{
122
    int res = 0;
123

    
124
    while(x)
125
    {
126
        res++;
127
        x >>= 1;
128
    }
129
    return res;
130
}
131

    
132
static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
133
{
134
    int i, bits;
135

    
136
    if (!max)
137
        return;
138

    
139
    bits = bits_to_store(max);
140

    
141
    for (i = 0; i < bits-1; i++)
142
        put_bits(pb, 1, value & (1 << i));
143

    
144
    if ( (value | (1 << (bits-1))) <= max)
145
        put_bits(pb, 1, value & (1 << (bits-1)));
146
}
147

    
148
static unsigned int read_uint_max(GetBitContext *gb, int max)
149
{
150
    int i, bits, value = 0;
151

    
152
    if (!max)
153
        return 0;
154

    
155
    bits = bits_to_store(max);
156

    
157
    for (i = 0; i < bits-1; i++)
158
        if (get_bits1(gb))
159
            value += 1 << i;
160

    
161
    if ( (value | (1<<(bits-1))) <= max)
162
        if (get_bits1(gb))
163
            value += 1 << (bits-1);
164

    
165
    return value;
166
}
167

    
168
static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
169
{
170
    int i, j, x = 0, low_bits = 0, max = 0;
171
    int step = 256, pos = 0, dominant = 0, any = 0;
172
    int *copy, *bits;
173

    
174
    copy = av_mallocz(4* entries);
175
    if (!copy)
176
        return -1;
177

    
178
    if (base_2_part)
179
    {
180
        int energy = 0;
181

    
182
        for (i = 0; i < entries; i++)
183
            energy += abs(buf[i]);
184

    
185
        low_bits = bits_to_store(energy / (entries * 2));
186
        if (low_bits > 15)
187
            low_bits = 15;
188

    
189
        put_bits(pb, 4, low_bits);
190
    }
191

    
192
    for (i = 0; i < entries; i++)
193
    {
194
        put_bits(pb, low_bits, abs(buf[i]));
195
        copy[i] = abs(buf[i]) >> low_bits;
196
        if (copy[i] > max)
197
            max = abs(copy[i]);
198
    }
199

    
200
    bits = av_mallocz(4* entries*max);
201
    if (!bits)
202
    {
203
//        av_free(copy);
204
        return -1;
205
    }
206

    
207
    for (i = 0; i <= max; i++)
208
    {
209
        for (j = 0; j < entries; j++)
210
            if (copy[j] >= i)
211
                bits[x++] = copy[j] > i;
212
    }
213

    
214
    // store bitstream
215
    while (pos < x)
216
    {
217
        int steplet = step >> 8;
218

    
219
        if (pos + steplet > x)
220
            steplet = x - pos;
221

    
222
        for (i = 0; i < steplet; i++)
223
            if (bits[i+pos] != dominant)
224
                any = 1;
225

    
226
        put_bits(pb, 1, any);
227

    
228
        if (!any)
229
        {
230
            pos += steplet;
231
            step += step / ADAPT_LEVEL;
232
        }
233
        else
234
        {
235
            int interloper = 0;
236

    
237
            while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
238
                interloper++;
239

    
240
            // note change
241
            write_uint_max(pb, interloper, (step >> 8) - 1);
242

    
243
            pos += interloper + 1;
244
            step -= step / ADAPT_LEVEL;
245
        }
246

    
247
        if (step < 256)
248
        {
249
            step = 65536 / step;
250
            dominant = !dominant;
251
        }
252
    }
253

    
254
    // store signs
255
    for (i = 0; i < entries; i++)
256
        if (buf[i])
257
            put_bits(pb, 1, buf[i] < 0);
258

    
259
//    av_free(bits);
260
//    av_free(copy);
261

    
262
    return 0;
263
}
264

    
265
static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
266
{
267
    int i, low_bits = 0, x = 0;
268
    int n_zeros = 0, step = 256, dominant = 0;
269
    int pos = 0, level = 0;
270
    int *bits = av_mallocz(4* entries);
271

    
272
    if (!bits)
273
        return -1;
274

    
275
    if (base_2_part)
276
    {
277
        low_bits = get_bits(gb, 4);
278

    
279
        if (low_bits)
280
            for (i = 0; i < entries; i++)
281
                buf[i] = get_bits(gb, low_bits);
282
    }
283

    
284
//    av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);
285

    
286
    while (n_zeros < entries)
287
    {
288
        int steplet = step >> 8;
289

    
290
        if (!get_bits1(gb))
291
        {
292
            for (i = 0; i < steplet; i++)
293
                bits[x++] = dominant;
294

    
295
            if (!dominant)
296
                n_zeros += steplet;
297

    
298
            step += step / ADAPT_LEVEL;
299
        }
300
        else
301
        {
302
            int actual_run = read_uint_max(gb, steplet-1);
303

    
304
//            av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run);
305

    
306
            for (i = 0; i < actual_run; i++)
307
                bits[x++] = dominant;
308

    
309
            bits[x++] = !dominant;
310

    
311
            if (!dominant)
312
                n_zeros += actual_run;
313
            else
314
                n_zeros++;
315

    
316
            step -= step / ADAPT_LEVEL;
317
        }
318

    
319
        if (step < 256)
320
        {
321
            step = 65536 / step;
322
            dominant = !dominant;
323
        }
324
    }
325

    
326
    // reconstruct unsigned values
327
    n_zeros = 0;
328
    for (i = 0; n_zeros < entries; i++)
329
    {
330
        while(1)
331
        {
332
            if (pos >= entries)
333
            {
334
                pos = 0;
335
                level += 1 << low_bits;
336
            }
337

    
338
            if (buf[pos] >= level)
339
                break;
340

    
341
            pos++;
342
        }
343

    
344
        if (bits[i])
345
            buf[pos] += 1 << low_bits;
346
        else
347
            n_zeros++;
348

    
349
        pos++;
350
    }
351
//    av_free(bits);
352

    
353
    // read signs
354
    for (i = 0; i < entries; i++)
355
        if (buf[i] && get_bits1(gb))
356
            buf[i] = -buf[i];
357

    
358
//    av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);
359

    
360
    return 0;
361
}
362
#endif
363

    
364
static void predictor_init_state(int *k, int *state, int order)
365
{
366
    int i;
367

    
368
    for (i = order-2; i >= 0; i--)
369
    {
370
        int j, p, x = state[i];
371

    
372
        for (j = 0, p = i+1; p < order; j++,p++)
373
            {
374
            int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
375
            state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
376
            x = tmp;
377
        }
378
    }
379
}
380

    
381
static int predictor_calc_error(int *k, int *state, int order, int error)
382
{
383
    int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);
384

    
385
#if 1
386
    int *k_ptr = &(k[order-2]),
387
        *state_ptr = &(state[order-2]);
388
    for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
389
    {
390
        int k_value = *k_ptr, state_value = *state_ptr;
391
        x -= shift_down(k_value * state_value, LATTICE_SHIFT);
392
        state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT);
393
    }
394
#else
395
    for (i = order-2; i >= 0; i--)
396
    {
397
        x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
398
        state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT);
399
    }
400
#endif
401

    
402
    // don't drift too far, to avoid overflows
403
    if (x >  (SAMPLE_FACTOR<<16)) x =  (SAMPLE_FACTOR<<16);
404
    if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);
405

    
406
    state[0] = x;
407

    
408
    return x;
409
}
410

    
411
#if CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER
412
// Heavily modified Levinson-Durbin algorithm which
413
// copes better with quantization, and calculates the
414
// actual whitened result as it goes.
415

    
416
static void modified_levinson_durbin(int *window, int window_entries,
417
        int *out, int out_entries, int channels, int *tap_quant)
418
{
419
    int i;
420
    int *state = av_mallocz(4* window_entries);
421

    
422
    memcpy(state, window, 4* window_entries);
423

    
424
    for (i = 0; i < out_entries; i++)
425
    {
426
        int step = (i+1)*channels, k, j;
427
        double xx = 0.0, xy = 0.0;
428
#if 1
429
        int *x_ptr = &(window[step]), *state_ptr = &(state[0]);
430
        j = window_entries - step;
431
        for (;j>=0;j--,x_ptr++,state_ptr++)
432
        {
433
            double x_value = *x_ptr, state_value = *state_ptr;
434
            xx += state_value*state_value;
435
            xy += x_value*state_value;
436
        }
437
#else
438
        for (j = 0; j <= (window_entries - step); j++);
439
        {
440
            double stepval = window[step+j], stateval = window[j];
441
//            xx += (double)window[j]*(double)window[j];
442
//            xy += (double)window[step+j]*(double)window[j];
443
            xx += stateval*stateval;
444
            xy += stepval*stateval;
445
        }
446
#endif
447
        if (xx == 0.0)
448
            k = 0;
449
        else
450
            k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5));
451

    
452
        if (k > (LATTICE_FACTOR/tap_quant[i]))
453
            k = LATTICE_FACTOR/tap_quant[i];
454
        if (-k > (LATTICE_FACTOR/tap_quant[i]))
455
            k = -(LATTICE_FACTOR/tap_quant[i]);
456

    
457
        out[i] = k;
458
        k *= tap_quant[i];
459

    
460
#if 1
461
        x_ptr = &(window[step]);
462
        state_ptr = &(state[0]);
463
        j = window_entries - step;
464
        for (;j>=0;j--,x_ptr++,state_ptr++)
465
        {
466
            int x_value = *x_ptr, state_value = *state_ptr;
467
            *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
468
            *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
469
        }
470
#else
471
        for (j=0; j <= (window_entries - step); j++)
472
        {
473
            int stepval = window[step+j], stateval=state[j];
474
            window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
475
            state[j] += shift_down(k * stepval, LATTICE_SHIFT);
476
        }
477
#endif
478
    }
479

    
480
    av_free(state);
481
}
482

    
483
static inline int code_samplerate(int samplerate)
484
{
485
    switch (samplerate)
486
    {
487
        case 44100: return 0;
488
        case 22050: return 1;
489
        case 11025: return 2;
490
        case 96000: return 3;
491
        case 48000: return 4;
492
        case 32000: return 5;
493
        case 24000: return 6;
494
        case 16000: return 7;
495
        case 8000: return 8;
496
    }
497
    return -1;
498
}
499

    
500
static av_cold int sonic_encode_init(AVCodecContext *avctx)
501
{
502
    SonicContext *s = avctx->priv_data;
503
    PutBitContext pb;
504
    int i, version = 0;
505

    
506
    if (avctx->channels > MAX_CHANNELS)
507
    {
508
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
509
        return -1; /* only stereo or mono for now */
510
    }
511

    
512
    if (avctx->channels == 2)
513
        s->decorrelation = MID_SIDE;
514

    
515
    if (avctx->codec->id == CODEC_ID_SONIC_LS)
516
    {
517
        s->lossless = 1;
518
        s->num_taps = 32;
519
        s->downsampling = 1;
520
        s->quantization = 0.0;
521
    }
522
    else
523
    {
524
        s->num_taps = 128;
525
        s->downsampling = 2;
526
        s->quantization = 1.0;
527
    }
528

    
529
    // max tap 2048
530
    if ((s->num_taps < 32) || (s->num_taps > 1024) ||
531
        ((s->num_taps>>5)<<5 != s->num_taps))
532
    {
533
        av_log(avctx, AV_LOG_ERROR, "Invalid number of taps\n");
534
        return -1;
535
    }
536

    
537
    // generate taps
538
    s->tap_quant = av_mallocz(4* s->num_taps);
539
    for (i = 0; i < s->num_taps; i++)
540
        s->tap_quant[i] = (int)(sqrt(i+1));
541

    
542
    s->channels = avctx->channels;
543
    s->samplerate = avctx->sample_rate;
544

    
545
    s->block_align = (int)(2048.0*s->samplerate/44100)/s->downsampling;
546
    s->frame_size = s->channels*s->block_align*s->downsampling;
547

    
548
    s->tail = av_mallocz(4* s->num_taps*s->channels);
549
    if (!s->tail)
550
        return -1;
551
    s->tail_size = s->num_taps*s->channels;
552

    
553
    s->predictor_k = av_mallocz(4 * s->num_taps);
554
    if (!s->predictor_k)
555
        return -1;
556

    
557
    for (i = 0; i < s->channels; i++)
558
    {
559
        s->coded_samples[i] = av_mallocz(4* s->block_align);
560
        if (!s->coded_samples[i])
561
            return -1;
562
    }
563

    
564
    s->int_samples = av_mallocz(4* s->frame_size);
565

    
566
    s->window_size = ((2*s->tail_size)+s->frame_size);
567
    s->window = av_mallocz(4* s->window_size);
568
    if (!s->window)
569
        return -1;
570

    
571
    avctx->extradata = av_mallocz(16);
572
    if (!avctx->extradata)
573
        return -1;
574
    init_put_bits(&pb, avctx->extradata, 16*8);
575

    
576
    put_bits(&pb, 2, version); // version
577
    if (version == 1)
578
    {
579
        put_bits(&pb, 2, s->channels);
580
        put_bits(&pb, 4, code_samplerate(s->samplerate));
581
    }
582
    put_bits(&pb, 1, s->lossless);
583
    if (!s->lossless)
584
        put_bits(&pb, 3, SAMPLE_SHIFT); // XXX FIXME: sample precision
585
    put_bits(&pb, 2, s->decorrelation);
586
    put_bits(&pb, 2, s->downsampling);
587
    put_bits(&pb, 5, (s->num_taps >> 5)-1); // 32..1024
588
    put_bits(&pb, 1, 0); // XXX FIXME: no custom tap quant table
589

    
590
    flush_put_bits(&pb);
591
    avctx->extradata_size = put_bits_count(&pb)/8;
592

    
593
    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
594
        version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
595

    
596
    avctx->coded_frame = avcodec_alloc_frame();
597
    if (!avctx->coded_frame)
598
        return AVERROR(ENOMEM);
599
    avctx->coded_frame->key_frame = 1;
600
    avctx->frame_size = s->block_align*s->downsampling;
601

    
602
    return 0;
603
}
604

    
605
static av_cold int sonic_encode_close(AVCodecContext *avctx)
606
{
607
    SonicContext *s = avctx->priv_data;
608
    int i;
609

    
610
    av_freep(&avctx->coded_frame);
611

    
612
    for (i = 0; i < s->channels; i++)
613
        av_free(s->coded_samples[i]);
614

    
615
    av_free(s->predictor_k);
616
    av_free(s->tail);
617
    av_free(s->tap_quant);
618
    av_free(s->window);
619
    av_free(s->int_samples);
620

    
621
    return 0;
622
}
623

    
624
static int sonic_encode_frame(AVCodecContext *avctx,
625
                            uint8_t *buf, int buf_size, void *data)
626
{
627
    SonicContext *s = avctx->priv_data;
628
    PutBitContext pb;
629
    int i, j, ch, quant = 0, x = 0;
630
    short *samples = data;
631

    
632
    init_put_bits(&pb, buf, buf_size*8);
633

    
634
    // short -> internal
635
    for (i = 0; i < s->frame_size; i++)
636
        s->int_samples[i] = samples[i];
637

    
638
    if (!s->lossless)
639
        for (i = 0; i < s->frame_size; i++)
640
            s->int_samples[i] = s->int_samples[i] << SAMPLE_SHIFT;
641

    
642
    switch(s->decorrelation)
643
    {
644
        case MID_SIDE:
645
            for (i = 0; i < s->frame_size; i += s->channels)
646
            {
647
                s->int_samples[i] += s->int_samples[i+1];
648
                s->int_samples[i+1] -= shift(s->int_samples[i], 1);
649
            }
650
            break;
651
        case LEFT_SIDE:
652
            for (i = 0; i < s->frame_size; i += s->channels)
653
                s->int_samples[i+1] -= s->int_samples[i];
654
            break;
655
        case RIGHT_SIDE:
656
            for (i = 0; i < s->frame_size; i += s->channels)
657
                s->int_samples[i] -= s->int_samples[i+1];
658
            break;
659
    }
660

    
661
    memset(s->window, 0, 4* s->window_size);
662

    
663
    for (i = 0; i < s->tail_size; i++)
664
        s->window[x++] = s->tail[i];
665

    
666
    for (i = 0; i < s->frame_size; i++)
667
        s->window[x++] = s->int_samples[i];
668

    
669
    for (i = 0; i < s->tail_size; i++)
670
        s->window[x++] = 0;
671

    
672
    for (i = 0; i < s->tail_size; i++)
673
        s->tail[i] = s->int_samples[s->frame_size - s->tail_size + i];
674

    
675
    // generate taps
676
    modified_levinson_durbin(s->window, s->window_size,
677
                s->predictor_k, s->num_taps, s->channels, s->tap_quant);
678
    if (intlist_write(&pb, s->predictor_k, s->num_taps, 0) < 0)
679
        return -1;
680

    
681
    for (ch = 0; ch < s->channels; ch++)
682
    {
683
        x = s->tail_size+ch;
684
        for (i = 0; i < s->block_align; i++)
685
        {
686
            int sum = 0;
687
            for (j = 0; j < s->downsampling; j++, x += s->channels)
688
                sum += s->window[x];
689
            s->coded_samples[ch][i] = sum;
690
        }
691
    }
692

    
693
    // simple rate control code
694
    if (!s->lossless)
695
    {
696
        double energy1 = 0.0, energy2 = 0.0;
697
        for (ch = 0; ch < s->channels; ch++)
698
        {
699
            for (i = 0; i < s->block_align; i++)
700
            {
701
                double sample = s->coded_samples[ch][i];
702
                energy2 += sample*sample;
703
                energy1 += fabs(sample);
704
            }
705
        }
706

    
707
        energy2 = sqrt(energy2/(s->channels*s->block_align));
708
        energy1 = sqrt(2.0)*energy1/(s->channels*s->block_align);
709

    
710
        // increase bitrate when samples are like a gaussian distribution
711
        // reduce bitrate when samples are like a two-tailed exponential distribution
712

    
713
        if (energy2 > energy1)
714
            energy2 += (energy2-energy1)*RATE_VARIATION;
715

    
716
        quant = (int)(BASE_QUANT*s->quantization*energy2/SAMPLE_FACTOR);
717
//        av_log(avctx, AV_LOG_DEBUG, "quant: %d energy: %f / %f\n", quant, energy1, energy2);
718

    
719
        if (quant < 1)
720
            quant = 1;
721
        if (quant > 65535)
722
            quant = 65535;
723

    
724
        set_ue_golomb(&pb, quant);
725

    
726
        quant *= SAMPLE_FACTOR;
727
    }
728

    
729
    // write out coded samples
730
    for (ch = 0; ch < s->channels; ch++)
731
    {
732
        if (!s->lossless)
733
            for (i = 0; i < s->block_align; i++)
734
                s->coded_samples[ch][i] = divide(s->coded_samples[ch][i], quant);
735

    
736
        if (intlist_write(&pb, s->coded_samples[ch], s->block_align, 1) < 0)
737
            return -1;
738
    }
739

    
740
//    av_log(avctx, AV_LOG_DEBUG, "used bytes: %d\n", (put_bits_count(&pb)+7)/8);
741

    
742
    flush_put_bits(&pb);
743
    return (put_bits_count(&pb)+7)/8;
744
}
745
#endif /* CONFIG_SONIC_ENCODER || CONFIG_SONIC_LS_ENCODER */
746

    
747
#if CONFIG_SONIC_DECODER
748
static const int samplerate_table[] =
749
    { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };
750

    
751
static av_cold int sonic_decode_init(AVCodecContext *avctx)
752
{
753
    SonicContext *s = avctx->priv_data;
754
    GetBitContext gb;
755
    int i, version;
756

    
757
    s->channels = avctx->channels;
758
    s->samplerate = avctx->sample_rate;
759

    
760
    if (!avctx->extradata)
761
    {
762
        av_log(avctx, AV_LOG_ERROR, "No mandatory headers present\n");
763
        return -1;
764
    }
765

    
766
    init_get_bits(&gb, avctx->extradata, avctx->extradata_size);
767

    
768
    version = get_bits(&gb, 2);
769
    if (version > 1)
770
    {
771
        av_log(avctx, AV_LOG_ERROR, "Unsupported Sonic version, please report\n");
772
        return -1;
773
    }
774

    
775
    if (version == 1)
776
    {
777
        s->channels = get_bits(&gb, 2);
778
        s->samplerate = samplerate_table[get_bits(&gb, 4)];
779
        av_log(avctx, AV_LOG_INFO, "Sonicv2 chans: %d samprate: %d\n",
780
            s->channels, s->samplerate);
781
    }
782

    
783
    if (s->channels > MAX_CHANNELS)
784
    {
785
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo streams are supported by now\n");
786
        return -1;
787
    }
788

    
789
    s->lossless = get_bits1(&gb);
790
    if (!s->lossless)
791
        skip_bits(&gb, 3); // XXX FIXME
792
    s->decorrelation = get_bits(&gb, 2);
793

    
794
    s->downsampling = get_bits(&gb, 2);
795
    s->num_taps = (get_bits(&gb, 5)+1)<<5;
796
    if (get_bits1(&gb)) // XXX FIXME
797
        av_log(avctx, AV_LOG_INFO, "Custom quant table\n");
798

    
799
    s->block_align = (int)(2048.0*(s->samplerate/44100))/s->downsampling;
800
    s->frame_size = s->channels*s->block_align*s->downsampling;
801
//    avctx->frame_size = s->block_align;
802

    
803
    av_log(avctx, AV_LOG_INFO, "Sonic: ver: %d ls: %d dr: %d taps: %d block: %d frame: %d downsamp: %d\n",
804
        version, s->lossless, s->decorrelation, s->num_taps, s->block_align, s->frame_size, s->downsampling);
805

    
806
    // generate taps
807
    s->tap_quant = av_mallocz(4* s->num_taps);
808
    for (i = 0; i < s->num_taps; i++)
809
        s->tap_quant[i] = (int)(sqrt(i+1));
810

    
811
    s->predictor_k = av_mallocz(4* s->num_taps);
812

    
813
    for (i = 0; i < s->channels; i++)
814
    {
815
        s->predictor_state[i] = av_mallocz(4* s->num_taps);
816
        if (!s->predictor_state[i])
817
            return -1;
818
    }
819

    
820
    for (i = 0; i < s->channels; i++)
821
    {
822
        s->coded_samples[i] = av_mallocz(4* s->block_align);
823
        if (!s->coded_samples[i])
824
            return -1;
825
    }
826
    s->int_samples = av_mallocz(4* s->frame_size);
827

    
828
    avctx->sample_fmt = SAMPLE_FMT_S16;
829
    return 0;
830
}
831

    
832
static av_cold int sonic_decode_close(AVCodecContext *avctx)
833
{
834
    SonicContext *s = avctx->priv_data;
835
    int i;
836

    
837
    av_free(s->int_samples);
838
    av_free(s->tap_quant);
839
    av_free(s->predictor_k);
840

    
841
    for (i = 0; i < s->channels; i++)
842
    {
843
        av_free(s->predictor_state[i]);
844
        av_free(s->coded_samples[i]);
845
    }
846

    
847
    return 0;
848
}
849

    
850
static int sonic_decode_frame(AVCodecContext *avctx,
851
                            void *data, int *data_size,
852
                            AVPacket *avpkt)
853
{
854
    const uint8_t *buf = avpkt->data;
855
    int buf_size = avpkt->size;
856
    SonicContext *s = avctx->priv_data;
857
    GetBitContext gb;
858
    int i, quant, ch, j;
859
    short *samples = data;
860

    
861
    if (buf_size == 0) return 0;
862

    
863
//    av_log(NULL, AV_LOG_INFO, "buf_size: %d\n", buf_size);
864

    
865
    init_get_bits(&gb, buf, buf_size*8);
866

    
867
    intlist_read(&gb, s->predictor_k, s->num_taps, 0);
868

    
869
    // dequantize
870
    for (i = 0; i < s->num_taps; i++)
871
        s->predictor_k[i] *= s->tap_quant[i];
872

    
873
    if (s->lossless)
874
        quant = 1;
875
    else
876
        quant = get_ue_golomb(&gb) * SAMPLE_FACTOR;
877

    
878
//    av_log(NULL, AV_LOG_INFO, "quant: %d\n", quant);
879

    
880
    for (ch = 0; ch < s->channels; ch++)
881
    {
882
        int x = ch;
883

    
884
        predictor_init_state(s->predictor_k, s->predictor_state[ch], s->num_taps);
885

    
886
        intlist_read(&gb, s->coded_samples[ch], s->block_align, 1);
887

    
888
        for (i = 0; i < s->block_align; i++)
889
        {
890
            for (j = 0; j < s->downsampling - 1; j++)
891
            {
892
                s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, 0);
893
                x += s->channels;
894
            }
895

    
896
            s->int_samples[x] = predictor_calc_error(s->predictor_k, s->predictor_state[ch], s->num_taps, s->coded_samples[ch][i] * quant);
897
            x += s->channels;
898
        }
899

    
900
        for (i = 0; i < s->num_taps; i++)
901
            s->predictor_state[ch][i] = s->int_samples[s->frame_size - s->channels + ch - i*s->channels];
902
    }
903

    
904
    switch(s->decorrelation)
905
    {
906
        case MID_SIDE:
907
            for (i = 0; i < s->frame_size; i += s->channels)
908
            {
909
                s->int_samples[i+1] += shift(s->int_samples[i], 1);
910
                s->int_samples[i] -= s->int_samples[i+1];
911
            }
912
            break;
913
        case LEFT_SIDE:
914
            for (i = 0; i < s->frame_size; i += s->channels)
915
                s->int_samples[i+1] += s->int_samples[i];
916
            break;
917
        case RIGHT_SIDE:
918
            for (i = 0; i < s->frame_size; i += s->channels)
919
                s->int_samples[i] += s->int_samples[i+1];
920
            break;
921
    }
922

    
923
    if (!s->lossless)
924
        for (i = 0; i < s->frame_size; i++)
925
            s->int_samples[i] = shift(s->int_samples[i], SAMPLE_SHIFT);
926

    
927
    // internal -> short
928
    for (i = 0; i < s->frame_size; i++)
929
        samples[i] = av_clip_int16(s->int_samples[i]);
930

    
931
    align_get_bits(&gb);
932

    
933
    *data_size = s->frame_size * 2;
934

    
935
    return (get_bits_count(&gb)+7)/8;
936
}
937

    
938
AVCodec sonic_decoder = {
939
    "sonic",
940
    AVMEDIA_TYPE_AUDIO,
941
    CODEC_ID_SONIC,
942
    sizeof(SonicContext),
943
    sonic_decode_init,
944
    NULL,
945
    sonic_decode_close,
946
    sonic_decode_frame,
947
    .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
948
};
949
#endif /* CONFIG_SONIC_DECODER */
950

    
951
#if CONFIG_SONIC_ENCODER
952
AVCodec sonic_encoder = {
953
    "sonic",
954
    AVMEDIA_TYPE_AUDIO,
955
    CODEC_ID_SONIC,
956
    sizeof(SonicContext),
957
    sonic_encode_init,
958
    sonic_encode_frame,
959
    sonic_encode_close,
960
    NULL,
961
    .long_name = NULL_IF_CONFIG_SMALL("Sonic"),
962
};
963
#endif
964

    
965
#if CONFIG_SONIC_LS_ENCODER
966
AVCodec sonic_ls_encoder = {
967
    "sonicls",
968
    AVMEDIA_TYPE_AUDIO,
969
    CODEC_ID_SONIC_LS,
970
    sizeof(SonicContext),
971
    sonic_encode_init,
972
    sonic_encode_frame,
973
    sonic_encode_close,
974
    NULL,
975
    .long_name = NULL_IF_CONFIG_SMALL("Sonic lossless"),
976
};
977
#endif