Statistics
| Branch: | Revision:

ffmpeg / libavcodec / alac.c @ 9106a698

History | View | Annotate | Download (20 KB)

1
/*
2
 * ALAC (Apple Lossless Audio Codec) decoder
3
 * Copyright (c) 2005 David Hammerton
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

    
22
/**
23
 * @file libavcodec/alac.c
24
 * ALAC (Apple Lossless Audio Codec) decoder
25
 * @author 2005 David Hammerton
26
 *
27
 * For more information on the ALAC format, visit:
28
 *  http://crazney.net/programs/itunes/alac.html
29
 *
30
 * Note: This decoder expects a 36- (0x24-)byte QuickTime atom to be
31
 * passed through the extradata[_size] fields. This atom is tacked onto
32
 * the end of an 'alac' stsd atom and has the following format:
33
 *  bytes 0-3   atom size (0x24), big-endian
34
 *  bytes 4-7   atom type ('alac', not the 'alac' tag from start of stsd)
35
 *  bytes 8-35  data bytes needed by decoder
36
 *
37
 * Extradata:
38
 * 32bit  size
39
 * 32bit  tag (=alac)
40
 * 32bit  zero?
41
 * 32bit  max sample per frame
42
 *  8bit  ?? (zero?)
43
 *  8bit  sample size
44
 *  8bit  history mult
45
 *  8bit  initial history
46
 *  8bit  kmodifier
47
 *  8bit  channels?
48
 * 16bit  ??
49
 * 32bit  max coded frame size
50
 * 32bit  bitrate?
51
 * 32bit  samplerate
52
 */
53

    
54

    
55
#include "avcodec.h"
56
#include "get_bits.h"
57
#include "bytestream.h"
58
#include "unary.h"
59
#include "mathops.h"
60

    
61
#define ALAC_EXTRADATA_SIZE 36
62
#define MAX_CHANNELS 2
63

    
64
typedef struct {
65

    
66
    AVCodecContext *avctx;
67
    GetBitContext gb;
68
    /* init to 0; first frame decode should initialize from extradata and
69
     * set this to 1 */
70
    int context_initialized;
71

    
72
    int numchannels;
73
    int bytespersample;
74

    
75
    /* buffers */
76
    int32_t *predicterror_buffer[MAX_CHANNELS];
77

    
78
    int32_t *outputsamples_buffer[MAX_CHANNELS];
79

    
80
    /* stuff from setinfo */
81
    uint32_t setinfo_max_samples_per_frame; /* 0x1000 = 4096 */    /* max samples per frame? */
82
    uint8_t setinfo_sample_size; /* 0x10 */
83
    uint8_t setinfo_rice_historymult; /* 0x28 */
84
    uint8_t setinfo_rice_initialhistory; /* 0x0a */
85
    uint8_t setinfo_rice_kmodifier; /* 0x0e */
86
    /* end setinfo stuff */
87

    
88
} ALACContext;
89

    
90
static void allocate_buffers(ALACContext *alac)
91
{
92
    int chan;
93
    for (chan = 0; chan < MAX_CHANNELS; chan++) {
94
        alac->predicterror_buffer[chan] =
95
            av_malloc(alac->setinfo_max_samples_per_frame * 4);
96

    
97
        alac->outputsamples_buffer[chan] =
98
            av_malloc(alac->setinfo_max_samples_per_frame * 4);
99
    }
100
}
101

    
102
static int alac_set_info(ALACContext *alac)
103
{
104
    const unsigned char *ptr = alac->avctx->extradata;
105

    
106
    ptr += 4; /* size */
107
    ptr += 4; /* alac */
108
    ptr += 4; /* 0 ? */
109

    
110
    if(AV_RB32(ptr) >= UINT_MAX/4){
111
        av_log(alac->avctx, AV_LOG_ERROR, "setinfo_max_samples_per_frame too large\n");
112
        return -1;
113
    }
114

    
115
    /* buffer size / 2 ? */
116
    alac->setinfo_max_samples_per_frame = bytestream_get_be32(&ptr);
117
    ptr++;                          /* ??? */
118
    alac->setinfo_sample_size           = *ptr++;
119
    if (alac->setinfo_sample_size > 32) {
120
        av_log(alac->avctx, AV_LOG_ERROR, "setinfo_sample_size too large\n");
121
        return -1;
122
    }
123
    alac->setinfo_rice_historymult      = *ptr++;
124
    alac->setinfo_rice_initialhistory   = *ptr++;
125
    alac->setinfo_rice_kmodifier        = *ptr++;
126
    ptr++;                         /* channels? */
127
    bytestream_get_be16(&ptr);      /* ??? */
128
    bytestream_get_be32(&ptr);      /* max coded frame size */
129
    bytestream_get_be32(&ptr);      /* bitrate ? */
130
    bytestream_get_be32(&ptr);      /* samplerate */
131

    
132
    allocate_buffers(alac);
133

    
134
    return 0;
135
}
136

    
137
static inline int decode_scalar(GetBitContext *gb, int k, int limit, int readsamplesize){
138
    /* read x - number of 1s before 0 represent the rice */
139
    int x = get_unary_0_9(gb);
140

    
141
    if (x > 8) { /* RICE THRESHOLD */
142
        /* use alternative encoding */
143
        x = get_bits(gb, readsamplesize);
144
    } else {
145
        if (k >= limit)
146
            k = limit;
147

    
148
        if (k != 1) {
149
            int extrabits = show_bits(gb, k);
150

    
151
            /* multiply x by 2^k - 1, as part of their strange algorithm */
152
            x = (x << k) - x;
153

    
154
            if (extrabits > 1) {
155
                x += extrabits - 1;
156
                skip_bits(gb, k);
157
            } else
158
                skip_bits(gb, k - 1);
159
        }
160
    }
161
    return x;
162
}
163

    
164
static void bastardized_rice_decompress(ALACContext *alac,
165
                                 int32_t *output_buffer,
166
                                 int output_size,
167
                                 int readsamplesize, /* arg_10 */
168
                                 int rice_initialhistory, /* arg424->b */
169
                                 int rice_kmodifier, /* arg424->d */
170
                                 int rice_historymult, /* arg424->c */
171
                                 int rice_kmodifier_mask /* arg424->e */
172
        )
173
{
174
    int output_count;
175
    unsigned int history = rice_initialhistory;
176
    int sign_modifier = 0;
177

    
178
    for (output_count = 0; output_count < output_size; output_count++) {
179
        int32_t x;
180
        int32_t x_modified;
181
        int32_t final_val;
182

    
183
        /* standard rice encoding */
184
        int k; /* size of extra bits */
185

    
186
        /* read k, that is bits as is */
187
        k = av_log2((history >> 9) + 3);
188
        x= decode_scalar(&alac->gb, k, rice_kmodifier, readsamplesize);
189

    
190
        x_modified = sign_modifier + x;
191
        final_val = (x_modified + 1) / 2;
192
        if (x_modified & 1) final_val *= -1;
193

    
194
        output_buffer[output_count] = final_val;
195

    
196
        sign_modifier = 0;
197

    
198
        /* now update the history */
199
        history += x_modified * rice_historymult
200
                   - ((history * rice_historymult) >> 9);
201

    
202
        if (x_modified > 0xffff)
203
            history = 0xffff;
204

    
205
        /* special case: there may be compressed blocks of 0 */
206
        if ((history < 128) && (output_count+1 < output_size)) {
207
            int k;
208
            unsigned int block_size;
209

    
210
            sign_modifier = 1;
211

    
212
            k = 7 - av_log2(history) + ((history + 16) >> 6 /* / 64 */);
213

    
214
            block_size= decode_scalar(&alac->gb, k, rice_kmodifier, 16);
215

    
216
            if (block_size > 0) {
217
                if(block_size >= output_size - output_count){
218
                    av_log(alac->avctx, AV_LOG_ERROR, "invalid zero block size of %d %d %d\n", block_size, output_size, output_count);
219
                    block_size= output_size - output_count - 1;
220
                }
221
                memset(&output_buffer[output_count+1], 0, block_size * 4);
222
                output_count += block_size;
223
            }
224

    
225
            if (block_size > 0xffff)
226
                sign_modifier = 0;
227

    
228
            history = 0;
229
        }
230
    }
231
}
232

    
233
static inline int sign_only(int v)
234
{
235
    return v ? FFSIGN(v) : 0;
236
}
237

    
238
static void predictor_decompress_fir_adapt(int32_t *error_buffer,
239
                                           int32_t *buffer_out,
240
                                           int output_size,
241
                                           int readsamplesize,
242
                                           int16_t *predictor_coef_table,
243
                                           int predictor_coef_num,
244
                                           int predictor_quantitization)
245
{
246
    int i;
247

    
248
    /* first sample always copies */
249
    *buffer_out = *error_buffer;
250

    
251
    if (!predictor_coef_num) {
252
        if (output_size <= 1)
253
            return;
254

    
255
        memcpy(buffer_out+1, error_buffer+1, (output_size-1) * 4);
256
        return;
257
    }
258

    
259
    if (predictor_coef_num == 0x1f) { /* 11111 - max value of predictor_coef_num */
260
      /* second-best case scenario for fir decompression,
261
       * error describes a small difference from the previous sample only
262
       */
263
        if (output_size <= 1)
264
            return;
265
        for (i = 0; i < output_size - 1; i++) {
266
            int32_t prev_value;
267
            int32_t error_value;
268

    
269
            prev_value = buffer_out[i];
270
            error_value = error_buffer[i+1];
271
            buffer_out[i+1] =
272
                sign_extend((prev_value + error_value), readsamplesize);
273
        }
274
        return;
275
    }
276

    
277
    /* read warm-up samples */
278
    if (predictor_coef_num > 0)
279
        for (i = 0; i < predictor_coef_num; i++) {
280
            int32_t val;
281

    
282
            val = buffer_out[i] + error_buffer[i+1];
283
            val = sign_extend(val, readsamplesize);
284
            buffer_out[i+1] = val;
285
        }
286

    
287
#if 0
288
    /* 4 and 8 are very common cases (the only ones i've seen). these
289
     * should be unrolled and optimized
290
     */
291
    if (predictor_coef_num == 4) {
292
        /* FIXME: optimized general case */
293
        return;
294
    }
295

296
    if (predictor_coef_table == 8) {
297
        /* FIXME: optimized general case */
298
        return;
299
    }
300
#endif
301

    
302
    /* general case */
303
    if (predictor_coef_num > 0) {
304
        for (i = predictor_coef_num + 1; i < output_size; i++) {
305
            int j;
306
            int sum = 0;
307
            int outval;
308
            int error_val = error_buffer[i];
309

    
310
            for (j = 0; j < predictor_coef_num; j++) {
311
                sum += (buffer_out[predictor_coef_num-j] - buffer_out[0]) *
312
                       predictor_coef_table[j];
313
            }
314

    
315
            outval = (1 << (predictor_quantitization-1)) + sum;
316
            outval = outval >> predictor_quantitization;
317
            outval = outval + buffer_out[0] + error_val;
318
            outval = sign_extend(outval, readsamplesize);
319

    
320
            buffer_out[predictor_coef_num+1] = outval;
321

    
322
            if (error_val > 0) {
323
                int predictor_num = predictor_coef_num - 1;
324

    
325
                while (predictor_num >= 0 && error_val > 0) {
326
                    int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
327
                    int sign = sign_only(val);
328

    
329
                    predictor_coef_table[predictor_num] -= sign;
330

    
331
                    val *= sign; /* absolute value */
332

    
333
                    error_val -= ((val >> predictor_quantitization) *
334
                                  (predictor_coef_num - predictor_num));
335

    
336
                    predictor_num--;
337
                }
338
            } else if (error_val < 0) {
339
                int predictor_num = predictor_coef_num - 1;
340

    
341
                while (predictor_num >= 0 && error_val < 0) {
342
                    int val = buffer_out[0] - buffer_out[predictor_coef_num - predictor_num];
343
                    int sign = - sign_only(val);
344

    
345
                    predictor_coef_table[predictor_num] -= sign;
346

    
347
                    val *= sign; /* neg value */
348

    
349
                    error_val -= ((val >> predictor_quantitization) *
350
                                  (predictor_coef_num - predictor_num));
351

    
352
                    predictor_num--;
353
                }
354
            }
355

    
356
            buffer_out++;
357
        }
358
    }
359
}
360

    
361
static void reconstruct_stereo_16(int32_t *buffer[MAX_CHANNELS],
362
                                  int16_t *buffer_out,
363
                                  int numchannels, int numsamples,
364
                                  uint8_t interlacing_shift,
365
                                  uint8_t interlacing_leftweight)
366
{
367
    int i;
368
    if (numsamples <= 0)
369
        return;
370

    
371
    /* weighted interlacing */
372
    if (interlacing_leftweight) {
373
        for (i = 0; i < numsamples; i++) {
374
            int32_t a, b;
375

    
376
            a = buffer[0][i];
377
            b = buffer[1][i];
378

    
379
            a -= (b * interlacing_leftweight) >> interlacing_shift;
380
            b += a;
381

    
382
            buffer_out[i*numchannels] = b;
383
            buffer_out[i*numchannels + 1] = a;
384
        }
385

    
386
        return;
387
    }
388

    
389
    /* otherwise basic interlacing took place */
390
    for (i = 0; i < numsamples; i++) {
391
        int16_t left, right;
392

    
393
        left = buffer[0][i];
394
        right = buffer[1][i];
395

    
396
        buffer_out[i*numchannels] = left;
397
        buffer_out[i*numchannels + 1] = right;
398
    }
399
}
400

    
401
static int alac_decode_frame(AVCodecContext *avctx,
402
                             void *outbuffer, int *outputsize,
403
                             AVPacket *avpkt)
404
{
405
    const uint8_t *inbuffer = avpkt->data;
406
    int input_buffer_size = avpkt->size;
407
    ALACContext *alac = avctx->priv_data;
408

    
409
    int channels;
410
    unsigned int outputsamples;
411
    int hassize;
412
    unsigned int readsamplesize;
413
    int wasted_bytes;
414
    int isnotcompressed;
415
    uint8_t interlacing_shift;
416
    uint8_t interlacing_leftweight;
417

    
418
    /* short-circuit null buffers */
419
    if (!inbuffer || !input_buffer_size)
420
        return input_buffer_size;
421

    
422
    /* initialize from the extradata */
423
    if (!alac->context_initialized) {
424
        if (alac->avctx->extradata_size != ALAC_EXTRADATA_SIZE) {
425
            av_log(avctx, AV_LOG_ERROR, "alac: expected %d extradata bytes\n",
426
                ALAC_EXTRADATA_SIZE);
427
            return input_buffer_size;
428
        }
429
        if (alac_set_info(alac)) {
430
            av_log(avctx, AV_LOG_ERROR, "alac: set_info failed\n");
431
            return input_buffer_size;
432
        }
433
        alac->context_initialized = 1;
434
    }
435

    
436
    init_get_bits(&alac->gb, inbuffer, input_buffer_size * 8);
437

    
438
    channels = get_bits(&alac->gb, 3) + 1;
439
    if (channels > MAX_CHANNELS) {
440
        av_log(avctx, AV_LOG_ERROR, "channels > %d not supported\n",
441
               MAX_CHANNELS);
442
        return input_buffer_size;
443
    }
444

    
445
    /* 2^result = something to do with output waiting.
446
     * perhaps matters if we read > 1 frame in a pass?
447
     */
448
    skip_bits(&alac->gb, 4);
449

    
450
    skip_bits(&alac->gb, 12); /* unknown, skip 12 bits */
451

    
452
    /* the output sample size is stored soon */
453
    hassize = get_bits1(&alac->gb);
454

    
455
    wasted_bytes = get_bits(&alac->gb, 2); /* unknown ? */
456

    
457
    /* whether the frame is compressed */
458
    isnotcompressed = get_bits1(&alac->gb);
459

    
460
    if (hassize) {
461
        /* now read the number of samples as a 32bit integer */
462
        outputsamples = get_bits_long(&alac->gb, 32);
463
        if(outputsamples > alac->setinfo_max_samples_per_frame){
464
            av_log(avctx, AV_LOG_ERROR, "outputsamples %d > %d\n", outputsamples, alac->setinfo_max_samples_per_frame);
465
            return -1;
466
        }
467
    } else
468
        outputsamples = alac->setinfo_max_samples_per_frame;
469

    
470
    if(outputsamples > *outputsize / alac->bytespersample){
471
        av_log(avctx, AV_LOG_ERROR, "sample buffer too small\n");
472
        return -1;
473
    }
474

    
475
    *outputsize = outputsamples * alac->bytespersample;
476
    readsamplesize = alac->setinfo_sample_size - (wasted_bytes * 8) + channels - 1;
477
    if (readsamplesize > MIN_CACHE_BITS) {
478
        av_log(avctx, AV_LOG_ERROR, "readsamplesize too big (%d)\n", readsamplesize);
479
        return -1;
480
    }
481

    
482
    if (!isnotcompressed) {
483
        /* so it is compressed */
484
        int16_t predictor_coef_table[channels][32];
485
        int predictor_coef_num[channels];
486
        int prediction_type[channels];
487
        int prediction_quantitization[channels];
488
        int ricemodifier[channels];
489
        int i, chan;
490

    
491
        interlacing_shift = get_bits(&alac->gb, 8);
492
        interlacing_leftweight = get_bits(&alac->gb, 8);
493

    
494
        for (chan = 0; chan < channels; chan++) {
495
            prediction_type[chan] = get_bits(&alac->gb, 4);
496
            prediction_quantitization[chan] = get_bits(&alac->gb, 4);
497

    
498
            ricemodifier[chan] = get_bits(&alac->gb, 3);
499
            predictor_coef_num[chan] = get_bits(&alac->gb, 5);
500

    
501
            /* read the predictor table */
502
            for (i = 0; i < predictor_coef_num[chan]; i++)
503
                predictor_coef_table[chan][i] = (int16_t)get_bits(&alac->gb, 16);
504
        }
505

    
506
        if (wasted_bytes)
507
            av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented, unhandling of wasted_bytes\n");
508

    
509
        for (chan = 0; chan < channels; chan++) {
510
            bastardized_rice_decompress(alac,
511
                                        alac->predicterror_buffer[chan],
512
                                        outputsamples,
513
                                        readsamplesize,
514
                                        alac->setinfo_rice_initialhistory,
515
                                        alac->setinfo_rice_kmodifier,
516
                                        ricemodifier[chan] * alac->setinfo_rice_historymult / 4,
517
                                        (1 << alac->setinfo_rice_kmodifier) - 1);
518

    
519
            if (prediction_type[chan] == 0) {
520
                /* adaptive fir */
521
                predictor_decompress_fir_adapt(alac->predicterror_buffer[chan],
522
                                               alac->outputsamples_buffer[chan],
523
                                               outputsamples,
524
                                               readsamplesize,
525
                                               predictor_coef_table[chan],
526
                                               predictor_coef_num[chan],
527
                                               prediction_quantitization[chan]);
528
            } else {
529
                av_log(avctx, AV_LOG_ERROR, "FIXME: unhandled prediction type: %i\n", prediction_type[chan]);
530
                /* I think the only other prediction type (or perhaps this is
531
                 * just a boolean?) runs adaptive fir twice.. like:
532
                 * predictor_decompress_fir_adapt(predictor_error, tempout, ...)
533
                 * predictor_decompress_fir_adapt(predictor_error, outputsamples ...)
534
                 * little strange..
535
                 */
536
            }
537
        }
538
    } else {
539
        /* not compressed, easy case */
540
        int i, chan;
541
        for (i = 0; i < outputsamples; i++)
542
            for (chan = 0; chan < channels; chan++) {
543
                int32_t audiobits;
544

    
545
                audiobits = get_sbits_long(&alac->gb, alac->setinfo_sample_size);
546

    
547
                alac->outputsamples_buffer[chan][i] = audiobits;
548
            }
549
        /* wasted_bytes = 0; */
550
        interlacing_shift = 0;
551
        interlacing_leftweight = 0;
552
    }
553
    if (get_bits(&alac->gb, 3) != 7)
554
        av_log(avctx, AV_LOG_ERROR, "Error : Wrong End Of Frame\n");
555

    
556
    switch(alac->setinfo_sample_size) {
557
    case 16:
558
        if (channels == 2) {
559
            reconstruct_stereo_16(alac->outputsamples_buffer,
560
                                  (int16_t*)outbuffer,
561
                                  alac->numchannels,
562
                                  outputsamples,
563
                                  interlacing_shift,
564
                                  interlacing_leftweight);
565
        } else {
566
            int i;
567
            for (i = 0; i < outputsamples; i++) {
568
                int16_t sample = alac->outputsamples_buffer[0][i];
569
                ((int16_t*)outbuffer)[i * alac->numchannels] = sample;
570
            }
571
        }
572
        break;
573
    case 20:
574
    case 24:
575
        // It is not clear if there exist any encoder that creates 24 bit ALAC
576
        // files. iTunes convert 24 bit raw files to 16 bit before encoding.
577
    case 32:
578
        av_log(avctx, AV_LOG_ERROR, "FIXME: unimplemented sample size %i\n", alac->setinfo_sample_size);
579
        break;
580
    default:
581
        break;
582
    }
583

    
584
    if (input_buffer_size * 8 - get_bits_count(&alac->gb) > 8)
585
        av_log(avctx, AV_LOG_ERROR, "Error : %d bits left\n", input_buffer_size * 8 - get_bits_count(&alac->gb));
586

    
587
    return input_buffer_size;
588
}
589

    
590
static av_cold int alac_decode_init(AVCodecContext * avctx)
591
{
592
    ALACContext *alac = avctx->priv_data;
593
    alac->avctx = avctx;
594
    alac->context_initialized = 0;
595

    
596
    alac->numchannels = alac->avctx->channels;
597
    alac->bytespersample = 2 * alac->numchannels;
598
    avctx->sample_fmt = SAMPLE_FMT_S16;
599

    
600
    return 0;
601
}
602

    
603
static av_cold int alac_decode_close(AVCodecContext *avctx)
604
{
605
    ALACContext *alac = avctx->priv_data;
606

    
607
    int chan;
608
    for (chan = 0; chan < MAX_CHANNELS; chan++) {
609
        av_free(alac->predicterror_buffer[chan]);
610
        av_free(alac->outputsamples_buffer[chan]);
611
    }
612

    
613
    return 0;
614
}
615

    
616
AVCodec alac_decoder = {
617
    "alac",
618
    CODEC_TYPE_AUDIO,
619
    CODEC_ID_ALAC,
620
    sizeof(ALACContext),
621
    alac_decode_init,
622
    NULL,
623
    alac_decode_close,
624
    alac_decode_frame,
625
    .long_name = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
626
};