Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aac.c @ 0cb71412

History | View | Annotate | Download (64.5 KB)

1
/*
2
 * AAC decoder
3
 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4
 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22

    
23
/**
24
 * @file libavcodec/aac.c
25
 * AAC decoder
26
 * @author Oded Shimon  ( ods15 ods15 dyndns org )
27
 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
28
 */
29

    
30
/*
31
 * supported tools
32
 *
33
 * Support?             Name
34
 * N (code in SoC repo) gain control
35
 * Y                    block switching
36
 * Y                    window shapes - standard
37
 * N                    window shapes - Low Delay
38
 * Y                    filterbank - standard
39
 * N (code in SoC repo) filterbank - Scalable Sample Rate
40
 * Y                    Temporal Noise Shaping
41
 * N (code in SoC repo) Long Term Prediction
42
 * Y                    intensity stereo
43
 * Y                    channel coupling
44
 * Y                    frequency domain prediction
45
 * Y                    Perceptual Noise Substitution
46
 * Y                    Mid/Side stereo
47
 * N                    Scalable Inverse AAC Quantization
48
 * N                    Frequency Selective Switch
49
 * N                    upsampling filter
50
 * Y                    quantization & coding - AAC
51
 * N                    quantization & coding - TwinVQ
52
 * N                    quantization & coding - BSAC
53
 * N                    AAC Error Resilience tools
54
 * N                    Error Resilience payload syntax
55
 * N                    Error Protection tool
56
 * N                    CELP
57
 * N                    Silence Compression
58
 * N                    HVXC
59
 * N                    HVXC 4kbits/s VR
60
 * N                    Structured Audio tools
61
 * N                    Structured Audio Sample Bank Format
62
 * N                    MIDI
63
 * N                    Harmonic and Individual Lines plus Noise
64
 * N                    Text-To-Speech Interface
65
 * N (in progress)      Spectral Band Replication
66
 * Y (not in this code) Layer-1
67
 * Y (not in this code) Layer-2
68
 * Y (not in this code) Layer-3
69
 * N                    SinuSoidal Coding (Transient, Sinusoid, Noise)
70
 * N (planned)          Parametric Stereo
71
 * N                    Direct Stream Transfer
72
 *
73
 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
74
 *       - HE AAC v2 comprises LC AAC with Spectral Band Replication and
75
           Parametric Stereo.
76
 */
77

    
78

    
79
#include "avcodec.h"
80
#include "internal.h"
81
#include "get_bits.h"
82
#include "dsputil.h"
83
#include "lpc.h"
84

    
85
#include "aac.h"
86
#include "aactab.h"
87
#include "aacdectab.h"
88
#include "mpeg4audio.h"
89
#include "aac_parser.h"
90

    
91
#include <assert.h>
92
#include <errno.h>
93
#include <math.h>
94
#include <string.h>
95

    
96
union float754 {
97
    float f;
98
    uint32_t i;
99
};
100

    
101
static VLC vlc_scalefactors;
102
static VLC vlc_spectral[11];
103

    
104

    
105
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
106
{
107
    static const int8_t tags_per_config[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 };
108
    if (ac->tag_che_map[type][elem_id]) {
109
        return ac->tag_che_map[type][elem_id];
110
    }
111
    if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
112
        return NULL;
113
    }
114
    switch (ac->m4ac.chan_config) {
115
    case 7:
116
        if (ac->tags_mapped == 3 && type == TYPE_CPE) {
117
            ac->tags_mapped++;
118
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
119
        }
120
    case 6:
121
        /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
122
           instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
123
           encountered such a stream, transfer the LFE[0] element to SCE[1] */
124
        if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
125
            ac->tags_mapped++;
126
            return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
127
        }
128
    case 5:
129
        if (ac->tags_mapped == 2 && type == TYPE_CPE) {
130
            ac->tags_mapped++;
131
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
132
        }
133
    case 4:
134
        if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
135
            ac->tags_mapped++;
136
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
137
        }
138
    case 3:
139
    case 2:
140
        if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
141
            ac->tags_mapped++;
142
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
143
        } else if (ac->m4ac.chan_config == 2) {
144
            return NULL;
145
        }
146
    case 1:
147
        if (!ac->tags_mapped && type == TYPE_SCE) {
148
            ac->tags_mapped++;
149
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
150
        }
151
    default:
152
        return NULL;
153
    }
154
}
155

    
156
/**
157
 * Configure output channel order based on the current program configuration element.
158
 *
159
 * @param   che_pos current channel position configuration
160
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
161
 *
162
 * @return  Returns error status. 0 - OK, !0 - error
163
 */
164
static int output_configure(AACContext *ac,
165
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
166
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
167
                            int channel_config)
168
{
169
    AVCodecContext *avctx = ac->avccontext;
170
    int i, type, channels = 0;
171

    
172
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
173

    
174
    /* Allocate or free elements depending on if they are in the
175
     * current program configuration.
176
     *
177
     * Set up default 1:1 output mapping.
178
     *
179
     * For a 5.1 stream the output order will be:
180
     *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
181
     */
182

    
183
    for (i = 0; i < MAX_ELEM_ID; i++) {
184
        for (type = 0; type < 4; type++) {
185
            if (che_pos[type][i]) {
186
                if (!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
187
                    return AVERROR(ENOMEM);
188
                if (type != TYPE_CCE) {
189
                    ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
190
                    if (type == TYPE_CPE) {
191
                        ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
192
                    }
193
                }
194
            } else
195
                av_freep(&ac->che[type][i]);
196
        }
197
    }
198

    
199
    if (channel_config) {
200
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
201
        ac->tags_mapped = 0;
202
    } else {
203
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
204
        ac->tags_mapped = 4 * MAX_ELEM_ID;
205
    }
206

    
207
    avctx->channels = channels;
208

    
209
    ac->output_configured = 1;
210

    
211
    return 0;
212
}
213

    
214
/**
215
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
216
 *
217
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
218
 * @param sce_map mono (Single Channel Element) map
219
 * @param type speaker type/position for these channels
220
 */
221
static void decode_channel_map(enum ChannelPosition *cpe_map,
222
                               enum ChannelPosition *sce_map,
223
                               enum ChannelPosition type,
224
                               GetBitContext *gb, int n)
225
{
226
    while (n--) {
227
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
228
        map[get_bits(gb, 4)] = type;
229
    }
230
}
231

    
232
/**
233
 * Decode program configuration element; reference: table 4.2.
234
 *
235
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
236
 *
237
 * @return  Returns error status. 0 - OK, !0 - error
238
 */
239
static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
240
                      GetBitContext *gb)
241
{
242
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
243

    
244
    skip_bits(gb, 2);  // object_type
245

    
246
    sampling_index = get_bits(gb, 4);
247
    if (ac->m4ac.sampling_index != sampling_index)
248
        av_log(ac->avccontext, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
249

    
250
    num_front       = get_bits(gb, 4);
251
    num_side        = get_bits(gb, 4);
252
    num_back        = get_bits(gb, 4);
253
    num_lfe         = get_bits(gb, 2);
254
    num_assoc_data  = get_bits(gb, 3);
255
    num_cc          = get_bits(gb, 4);
256

    
257
    if (get_bits1(gb))
258
        skip_bits(gb, 4); // mono_mixdown_tag
259
    if (get_bits1(gb))
260
        skip_bits(gb, 4); // stereo_mixdown_tag
261

    
262
    if (get_bits1(gb))
263
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
264

    
265
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
266
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
267
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
268
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
269

    
270
    skip_bits_long(gb, 4 * num_assoc_data);
271

    
272
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
273

    
274
    align_get_bits(gb);
275

    
276
    /* comment field, first byte is length */
277
    skip_bits_long(gb, 8 * get_bits(gb, 8));
278
    return 0;
279
}
280

    
281
/**
282
 * Set up channel positions based on a default channel configuration
283
 * as specified in table 1.17.
284
 *
285
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
286
 *
287
 * @return  Returns error status. 0 - OK, !0 - error
288
 */
289
static int set_default_channel_config(AACContext *ac,
290
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
291
                                      int channel_config)
292
{
293
    if (channel_config < 1 || channel_config > 7) {
294
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
295
               channel_config);
296
        return -1;
297
    }
298

    
299
    /* default channel configurations:
300
     *
301
     * 1ch : front center (mono)
302
     * 2ch : L + R (stereo)
303
     * 3ch : front center + L + R
304
     * 4ch : front center + L + R + back center
305
     * 5ch : front center + L + R + back stereo
306
     * 6ch : front center + L + R + back stereo + LFE
307
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
308
     */
309

    
310
    if (channel_config != 2)
311
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
312
    if (channel_config > 1)
313
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
314
    if (channel_config == 4)
315
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
316
    if (channel_config > 4)
317
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
318
        = AAC_CHANNEL_BACK;  // back stereo
319
    if (channel_config > 5)
320
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
321
    if (channel_config == 7)
322
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
323

    
324
    return 0;
325
}
326

    
327
/**
328
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
329
 *
330
 * @return  Returns error status. 0 - OK, !0 - error
331
 */
332
static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
333
                                     int channel_config)
334
{
335
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
336
    int extension_flag, ret;
337

    
338
    if (get_bits1(gb)) { // frameLengthFlag
339
        av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
340
        return -1;
341
    }
342

    
343
    if (get_bits1(gb))       // dependsOnCoreCoder
344
        skip_bits(gb, 14);   // coreCoderDelay
345
    extension_flag = get_bits1(gb);
346

    
347
    if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
348
        ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
349
        skip_bits(gb, 3);     // layerNr
350

    
351
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
352
    if (channel_config == 0) {
353
        skip_bits(gb, 4);  // element_instance_tag
354
        if ((ret = decode_pce(ac, new_che_pos, gb)))
355
            return ret;
356
    } else {
357
        if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
358
            return ret;
359
    }
360
    if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config)))
361
        return ret;
362

    
363
    if (extension_flag) {
364
        switch (ac->m4ac.object_type) {
365
        case AOT_ER_BSAC:
366
            skip_bits(gb, 5);    // numOfSubFrame
367
            skip_bits(gb, 11);   // layer_length
368
            break;
369
        case AOT_ER_AAC_LC:
370
        case AOT_ER_AAC_LTP:
371
        case AOT_ER_AAC_SCALABLE:
372
        case AOT_ER_AAC_LD:
373
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
374
                                    * aacScalefactorDataResilienceFlag
375
                                    * aacSpectralDataResilienceFlag
376
                                    */
377
            break;
378
        }
379
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
380
    }
381
    return 0;
382
}
383

    
384
/**
385
 * Decode audio specific configuration; reference: table 1.13.
386
 *
387
 * @param   data        pointer to AVCodecContext extradata
388
 * @param   data_size   size of AVCCodecContext extradata
389
 *
390
 * @return  Returns error status. 0 - OK, !0 - error
391
 */
392
static int decode_audio_specific_config(AACContext *ac, void *data,
393
                                        int data_size)
394
{
395
    GetBitContext gb;
396
    int i;
397

    
398
    init_get_bits(&gb, data, data_size * 8);
399

    
400
    if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
401
        return -1;
402
    if (ac->m4ac.sampling_index > 12) {
403
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
404
        return -1;
405
    }
406

    
407
    skip_bits_long(&gb, i);
408

    
409
    switch (ac->m4ac.object_type) {
410
    case AOT_AAC_MAIN:
411
    case AOT_AAC_LC:
412
        if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
413
            return -1;
414
        break;
415
    default:
416
        av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
417
               ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
418
        return -1;
419
    }
420
    return 0;
421
}
422

    
423
/**
424
 * linear congruential pseudorandom number generator
425
 *
426
 * @param   previous_val    pointer to the current state of the generator
427
 *
428
 * @return  Returns a 32-bit pseudorandom integer
429
 */
430
static av_always_inline int lcg_random(int previous_val)
431
{
432
    return previous_val * 1664525 + 1013904223;
433
}
434

    
435
static void reset_predict_state(PredictorState *ps)
436
{
437
    ps->r0   = 0.0f;
438
    ps->r1   = 0.0f;
439
    ps->cor0 = 0.0f;
440
    ps->cor1 = 0.0f;
441
    ps->var0 = 1.0f;
442
    ps->var1 = 1.0f;
443
}
444

    
445
static void reset_all_predictors(PredictorState *ps)
446
{
447
    int i;
448
    for (i = 0; i < MAX_PREDICTORS; i++)
449
        reset_predict_state(&ps[i]);
450
}
451

    
452
static void reset_predictor_group(PredictorState *ps, int group_num)
453
{
454
    int i;
455
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
456
        reset_predict_state(&ps[i]);
457
}
458

    
459
static av_cold int aac_decode_init(AVCodecContext *avccontext)
460
{
461
    AACContext *ac = avccontext->priv_data;
462
    int i;
463

    
464
    ac->avccontext = avccontext;
465

    
466
    if (avccontext->extradata_size > 0) {
467
        if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
468
            return -1;
469
        avccontext->sample_rate = ac->m4ac.sample_rate;
470
    } else if (avccontext->channels > 0) {
471
        ac->m4ac.sample_rate = avccontext->sample_rate;
472
    }
473

    
474
    avccontext->sample_fmt = SAMPLE_FMT_S16;
475
    avccontext->frame_size = 1024;
476

    
477
    AAC_INIT_VLC_STATIC( 0, 144);
478
    AAC_INIT_VLC_STATIC( 1, 114);
479
    AAC_INIT_VLC_STATIC( 2, 188);
480
    AAC_INIT_VLC_STATIC( 3, 180);
481
    AAC_INIT_VLC_STATIC( 4, 172);
482
    AAC_INIT_VLC_STATIC( 5, 140);
483
    AAC_INIT_VLC_STATIC( 6, 168);
484
    AAC_INIT_VLC_STATIC( 7, 114);
485
    AAC_INIT_VLC_STATIC( 8, 262);
486
    AAC_INIT_VLC_STATIC( 9, 248);
487
    AAC_INIT_VLC_STATIC(10, 384);
488

    
489
    dsputil_init(&ac->dsp, avccontext);
490

    
491
    ac->random_state = 0x1f2e3d4c;
492

    
493
    // -1024 - Compensate wrong IMDCT method.
494
    // 32768 - Required to scale values to the correct range for the bias method
495
    //         for float to int16 conversion.
496

    
497
    if (ac->dsp.float_to_int16 == ff_float_to_int16_c) {
498
        ac->add_bias  = 385.0f;
499
        ac->sf_scale  = 1. / (-1024. * 32768.);
500
        ac->sf_offset = 0;
501
    } else {
502
        ac->add_bias  = 0.0f;
503
        ac->sf_scale  = 1. / -1024.;
504
        ac->sf_offset = 60;
505
    }
506

    
507
#if !CONFIG_HARDCODED_TABLES
508
    for (i = 0; i < 428; i++)
509
        ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
510
#endif /* CONFIG_HARDCODED_TABLES */
511

    
512
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
513
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
514
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
515
                    352);
516

    
517
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
518
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
519
    // window initialization
520
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
521
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
522
    ff_sine_window_init(ff_sine_1024, 1024);
523
    ff_sine_window_init(ff_sine_128, 128);
524

    
525
    return 0;
526
}
527

    
528
/**
529
 * Skip data_stream_element; reference: table 4.10.
530
 */
531
static void skip_data_stream_element(GetBitContext *gb)
532
{
533
    int byte_align = get_bits1(gb);
534
    int count = get_bits(gb, 8);
535
    if (count == 255)
536
        count += get_bits(gb, 8);
537
    if (byte_align)
538
        align_get_bits(gb);
539
    skip_bits_long(gb, 8 * count);
540
}
541

    
542
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
543
                             GetBitContext *gb)
544
{
545
    int sfb;
546
    if (get_bits1(gb)) {
547
        ics->predictor_reset_group = get_bits(gb, 5);
548
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
549
            av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
550
            return -1;
551
        }
552
    }
553
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
554
        ics->prediction_used[sfb] = get_bits1(gb);
555
    }
556
    return 0;
557
}
558

    
559
/**
560
 * Decode Individual Channel Stream info; reference: table 4.6.
561
 *
562
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
563
 */
564
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
565
                           GetBitContext *gb, int common_window)
566
{
567
    if (get_bits1(gb)) {
568
        av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
569
        memset(ics, 0, sizeof(IndividualChannelStream));
570
        return -1;
571
    }
572
    ics->window_sequence[1] = ics->window_sequence[0];
573
    ics->window_sequence[0] = get_bits(gb, 2);
574
    ics->use_kb_window[1]   = ics->use_kb_window[0];
575
    ics->use_kb_window[0]   = get_bits1(gb);
576
    ics->num_window_groups  = 1;
577
    ics->group_len[0]       = 1;
578
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
579
        int i;
580
        ics->max_sfb = get_bits(gb, 4);
581
        for (i = 0; i < 7; i++) {
582
            if (get_bits1(gb)) {
583
                ics->group_len[ics->num_window_groups - 1]++;
584
            } else {
585
                ics->num_window_groups++;
586
                ics->group_len[ics->num_window_groups - 1] = 1;
587
            }
588
        }
589
        ics->num_windows       = 8;
590
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
591
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
592
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
593
        ics->predictor_present = 0;
594
    } else {
595
        ics->max_sfb               = get_bits(gb, 6);
596
        ics->num_windows           = 1;
597
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
598
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
599
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
600
        ics->predictor_present     = get_bits1(gb);
601
        ics->predictor_reset_group = 0;
602
        if (ics->predictor_present) {
603
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
604
                if (decode_prediction(ac, ics, gb)) {
605
                    memset(ics, 0, sizeof(IndividualChannelStream));
606
                    return -1;
607
                }
608
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
609
                av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
610
                memset(ics, 0, sizeof(IndividualChannelStream));
611
                return -1;
612
            } else {
613
                av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
614
                memset(ics, 0, sizeof(IndividualChannelStream));
615
                return -1;
616
            }
617
        }
618
    }
619

    
620
    if (ics->max_sfb > ics->num_swb) {
621
        av_log(ac->avccontext, AV_LOG_ERROR,
622
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
623
               ics->max_sfb, ics->num_swb);
624
        memset(ics, 0, sizeof(IndividualChannelStream));
625
        return -1;
626
    }
627

    
628
    return 0;
629
}
630

    
631
/**
632
 * Decode band types (section_data payload); reference: table 4.46.
633
 *
634
 * @param   band_type           array of the used band type
635
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
636
 *
637
 * @return  Returns error status. 0 - OK, !0 - error
638
 */
639
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
640
                             int band_type_run_end[120], GetBitContext *gb,
641
                             IndividualChannelStream *ics)
642
{
643
    int g, idx = 0;
644
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
645
    for (g = 0; g < ics->num_window_groups; g++) {
646
        int k = 0;
647
        while (k < ics->max_sfb) {
648
            uint8_t sect_len = k;
649
            int sect_len_incr;
650
            int sect_band_type = get_bits(gb, 4);
651
            if (sect_band_type == 12) {
652
                av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
653
                return -1;
654
            }
655
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
656
                sect_len += sect_len_incr;
657
            sect_len += sect_len_incr;
658
            if (sect_len > ics->max_sfb) {
659
                av_log(ac->avccontext, AV_LOG_ERROR,
660
                       "Number of bands (%d) exceeds limit (%d).\n",
661
                       sect_len, ics->max_sfb);
662
                return -1;
663
            }
664
            for (; k < sect_len; k++) {
665
                band_type        [idx]   = sect_band_type;
666
                band_type_run_end[idx++] = sect_len;
667
            }
668
        }
669
    }
670
    return 0;
671
}
672

    
673
/**
674
 * Decode scalefactors; reference: table 4.47.
675
 *
676
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
677
 * @param   band_type           array of the used band type
678
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
679
 * @param   sf                  array of scalefactors or intensity stereo positions
680
 *
681
 * @return  Returns error status. 0 - OK, !0 - error
682
 */
683
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
684
                               unsigned int global_gain,
685
                               IndividualChannelStream *ics,
686
                               enum BandType band_type[120],
687
                               int band_type_run_end[120])
688
{
689
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
690
    int g, i, idx = 0;
691
    int offset[3] = { global_gain, global_gain - 90, 100 };
692
    int noise_flag = 1;
693
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
694
    for (g = 0; g < ics->num_window_groups; g++) {
695
        for (i = 0; i < ics->max_sfb;) {
696
            int run_end = band_type_run_end[idx];
697
            if (band_type[idx] == ZERO_BT) {
698
                for (; i < run_end; i++, idx++)
699
                    sf[idx] = 0.;
700
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
701
                for (; i < run_end; i++, idx++) {
702
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
703
                    if (offset[2] > 255U) {
704
                        av_log(ac->avccontext, AV_LOG_ERROR,
705
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
706
                        return -1;
707
                    }
708
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
709
                }
710
            } else if (band_type[idx] == NOISE_BT) {
711
                for (; i < run_end; i++, idx++) {
712
                    if (noise_flag-- > 0)
713
                        offset[1] += get_bits(gb, 9) - 256;
714
                    else
715
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
716
                    if (offset[1] > 255U) {
717
                        av_log(ac->avccontext, AV_LOG_ERROR,
718
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
719
                        return -1;
720
                    }
721
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
722
                }
723
            } else {
724
                for (; i < run_end; i++, idx++) {
725
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
726
                    if (offset[0] > 255U) {
727
                        av_log(ac->avccontext, AV_LOG_ERROR,
728
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
729
                        return -1;
730
                    }
731
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
732
                }
733
            }
734
        }
735
    }
736
    return 0;
737
}
738

    
739
/**
740
 * Decode pulse data; reference: table 4.7.
741
 */
742
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
743
                         const uint16_t *swb_offset, int num_swb)
744
{
745
    int i, pulse_swb;
746
    pulse->num_pulse = get_bits(gb, 2) + 1;
747
    pulse_swb        = get_bits(gb, 6);
748
    if (pulse_swb >= num_swb)
749
        return -1;
750
    pulse->pos[0]    = swb_offset[pulse_swb];
751
    pulse->pos[0]   += get_bits(gb, 5);
752
    if (pulse->pos[0] > 1023)
753
        return -1;
754
    pulse->amp[0]    = get_bits(gb, 4);
755
    for (i = 1; i < pulse->num_pulse; i++) {
756
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
757
        if (pulse->pos[i] > 1023)
758
            return -1;
759
        pulse->amp[i] = get_bits(gb, 4);
760
    }
761
    return 0;
762
}
763

    
764
/**
765
 * Decode Temporal Noise Shaping data; reference: table 4.48.
766
 *
767
 * @return  Returns error status. 0 - OK, !0 - error
768
 */
769
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
770
                      GetBitContext *gb, const IndividualChannelStream *ics)
771
{
772
    int w, filt, i, coef_len, coef_res, coef_compress;
773
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
774
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
775
    for (w = 0; w < ics->num_windows; w++) {
776
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
777
            coef_res = get_bits1(gb);
778

    
779
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
780
                int tmp2_idx;
781
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
782

    
783
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
784
                    av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
785
                           tns->order[w][filt], tns_max_order);
786
                    tns->order[w][filt] = 0;
787
                    return -1;
788
                }
789
                if (tns->order[w][filt]) {
790
                    tns->direction[w][filt] = get_bits1(gb);
791
                    coef_compress = get_bits1(gb);
792
                    coef_len = coef_res + 3 - coef_compress;
793
                    tmp2_idx = 2 * coef_compress + coef_res;
794

    
795
                    for (i = 0; i < tns->order[w][filt]; i++)
796
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
797
                }
798
            }
799
        }
800
    }
801
    return 0;
802
}
803

    
804
/**
805
 * Decode Mid/Side data; reference: table 4.54.
806
 *
807
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
808
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
809
 *                      [3] reserved for scalable AAC
810
 */
811
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
812
                                   int ms_present)
813
{
814
    int idx;
815
    if (ms_present == 1) {
816
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
817
            cpe->ms_mask[idx] = get_bits1(gb);
818
    } else if (ms_present == 2) {
819
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
820
    }
821
}
822

    
823
/**
824
 * Decode spectral data; reference: table 4.50.
825
 * Dequantize and scale spectral data; reference: 4.6.3.3.
826
 *
827
 * @param   coef            array of dequantized, scaled spectral data
828
 * @param   sf              array of scalefactors or intensity stereo positions
829
 * @param   pulse_present   set if pulses are present
830
 * @param   pulse           pointer to pulse data struct
831
 * @param   band_type       array of the used band type
832
 *
833
 * @return  Returns error status. 0 - OK, !0 - error
834
 */
835
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
836
                                       GetBitContext *gb, float sf[120],
837
                                       int pulse_present, const Pulse *pulse,
838
                                       const IndividualChannelStream *ics,
839
                                       enum BandType band_type[120])
840
{
841
    int i, k, g, idx = 0;
842
    const int c = 1024 / ics->num_windows;
843
    const uint16_t *offsets = ics->swb_offset;
844
    float *coef_base = coef;
845
    static const float sign_lookup[] = { 1.0f, -1.0f };
846

    
847
    for (g = 0; g < ics->num_windows; g++)
848
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
849

    
850
    for (g = 0; g < ics->num_window_groups; g++) {
851
        for (i = 0; i < ics->max_sfb; i++, idx++) {
852
            const int cur_band_type = band_type[idx];
853
            const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
854
            const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
855
            int group;
856
            if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
857
                for (group = 0; group < ics->group_len[g]; group++) {
858
                    memset(coef + group * 128 + offsets[i], 0, (offsets[i + 1] - offsets[i]) * sizeof(float));
859
                }
860
            } else if (cur_band_type == NOISE_BT) {
861
                for (group = 0; group < ics->group_len[g]; group++) {
862
                    float scale;
863
                    float band_energy = 0;
864
                    float *cf = coef + group * 128 + offsets[i];
865
                    int len = offsets[i+1] - offsets[i];
866

    
867
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
868
                        ac->random_state  = lcg_random(ac->random_state);
869
                        coef[group * 128 + k] = ac->random_state;
870
                    }
871

    
872
                    band_energy += ac->dsp.scalarproduct_float(cf, cf, len);
873
                    scale = sf[idx] / sqrtf(band_energy);
874
                    ac->dsp.vector_fmul_scalar(cf, cf, scale, len);
875
                }
876
            } else {
877
                for (group = 0; group < ics->group_len[g]; group++) {
878
                    const float *vq[96];
879
                    const float **vqp = vq;
880
                    float *cf = coef + (group << 7) + offsets[i];
881
                    int len = offsets[i + 1] - offsets[i];
882

    
883
                    for (k = offsets[i]; k < offsets[i + 1]; k += dim) {
884
                        const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
885
                        const int coef_tmp_idx = (group << 7) + k;
886
                        const float *vq_ptr;
887
                        int j;
888
                        if (index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
889
                            av_log(ac->avccontext, AV_LOG_ERROR,
890
                                   "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
891
                                   cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
892
                            return -1;
893
                        }
894
                        vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
895
                        *vqp++ = vq_ptr;
896
                        if (is_cb_unsigned) {
897
                            if (vq_ptr[0])
898
                                coef[coef_tmp_idx    ] = sign_lookup[get_bits1(gb)];
899
                            if (vq_ptr[1])
900
                                coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
901
                            if (dim == 4) {
902
                                if (vq_ptr[2])
903
                                    coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
904
                                if (vq_ptr[3])
905
                                    coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
906
                            }
907
                            if (cur_band_type == ESC_BT) {
908
                                for (j = 0; j < 2; j++) {
909
                                    if (vq_ptr[j] == 64.0f) {
910
                                        int n = 4;
911
                                        /* The total length of escape_sequence must be < 22 bits according
912
                                           to the specification (i.e. max is 11111111110xxxxxxxxxx). */
913
                                        while (get_bits1(gb) && n < 15) n++;
914
                                        if (n == 15) {
915
                                            av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
916
                                            return -1;
917
                                        }
918
                                        n = (1 << n) + get_bits(gb, n);
919
                                        coef[coef_tmp_idx + j] *= cbrtf(n) * n;
920
                                    } else
921
                                        coef[coef_tmp_idx + j] *= vq_ptr[j];
922
                                }
923
                            }
924
                        }
925
                    }
926

    
927
                    if (is_cb_unsigned && cur_band_type != ESC_BT) {
928
                        ac->dsp.vector_fmul_sv_scalar[dim>>2](
929
                            cf, cf, vq, sf[idx], len);
930
                    } else if (cur_band_type == ESC_BT) {
931
                        ac->dsp.vector_fmul_scalar(cf, cf, sf[idx], len);
932
                    } else {    /* !is_cb_unsigned */
933
                        ac->dsp.sv_fmul_scalar[dim>>2](cf, vq, sf[idx], len);
934
                    }
935
                }
936
            }
937
        }
938
        coef += ics->group_len[g] << 7;
939
    }
940

    
941
    if (pulse_present) {
942
        idx = 0;
943
        for (i = 0; i < pulse->num_pulse; i++) {
944
            float co = coef_base[ pulse->pos[i] ];
945
            while (offsets[idx + 1] <= pulse->pos[i])
946
                idx++;
947
            if (band_type[idx] != NOISE_BT && sf[idx]) {
948
                float ico = -pulse->amp[i];
949
                if (co) {
950
                    co /= sf[idx];
951
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
952
                }
953
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
954
            }
955
        }
956
    }
957
    return 0;
958
}
959

    
960
static av_always_inline float flt16_round(float pf)
961
{
962
    union float754 tmp;
963
    tmp.f = pf;
964
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
965
    return tmp.f;
966
}
967

    
968
static av_always_inline float flt16_even(float pf)
969
{
970
    union float754 tmp;
971
    tmp.f = pf;
972
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
973
    return tmp.f;
974
}
975

    
976
static av_always_inline float flt16_trunc(float pf)
977
{
978
    union float754 pun;
979
    pun.f = pf;
980
    pun.i &= 0xFFFF0000U;
981
    return pun.f;
982
}
983

    
984
static void predict(AACContext *ac, PredictorState *ps, float *coef,
985
                    int output_enable)
986
{
987
    const float a     = 0.953125; // 61.0 / 64
988
    const float alpha = 0.90625;  // 29.0 / 32
989
    float e0, e1;
990
    float pv;
991
    float k1, k2;
992

    
993
    k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
994
    k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
995

    
996
    pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
997
    if (output_enable)
998
        *coef += pv * ac->sf_scale;
999

    
1000
    e0 = *coef / ac->sf_scale;
1001
    e1 = e0 - k1 * ps->r0;
1002

    
1003
    ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1004
    ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1005
    ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1006
    ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1007

    
1008
    ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1009
    ps->r0 = flt16_trunc(a * e0);
1010
}
1011

    
1012
/**
1013
 * Apply AAC-Main style frequency domain prediction.
1014
 */
1015
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1016
{
1017
    int sfb, k;
1018

    
1019
    if (!sce->ics.predictor_initialized) {
1020
        reset_all_predictors(sce->predictor_state);
1021
        sce->ics.predictor_initialized = 1;
1022
    }
1023

    
1024
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1025
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1026
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1027
                predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1028
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1029
            }
1030
        }
1031
        if (sce->ics.predictor_reset_group)
1032
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1033
    } else
1034
        reset_all_predictors(sce->predictor_state);
1035
}
1036

    
1037
/**
1038
 * Decode an individual_channel_stream payload; reference: table 4.44.
1039
 *
1040
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1041
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1042
 *
1043
 * @return  Returns error status. 0 - OK, !0 - error
1044
 */
1045
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1046
                      GetBitContext *gb, int common_window, int scale_flag)
1047
{
1048
    Pulse pulse;
1049
    TemporalNoiseShaping    *tns = &sce->tns;
1050
    IndividualChannelStream *ics = &sce->ics;
1051
    float *out = sce->coeffs;
1052
    int global_gain, pulse_present = 0;
1053

    
1054
    /* This assignment is to silence a GCC warning about the variable being used
1055
     * uninitialized when in fact it always is.
1056
     */
1057
    pulse.num_pulse = 0;
1058

    
1059
    global_gain = get_bits(gb, 8);
1060

    
1061
    if (!common_window && !scale_flag) {
1062
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1063
            return -1;
1064
    }
1065

    
1066
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1067
        return -1;
1068
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1069
        return -1;
1070

    
1071
    pulse_present = 0;
1072
    if (!scale_flag) {
1073
        if ((pulse_present = get_bits1(gb))) {
1074
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1075
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1076
                return -1;
1077
            }
1078
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1079
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1080
                return -1;
1081
            }
1082
        }
1083
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1084
            return -1;
1085
        if (get_bits1(gb)) {
1086
            av_log_missing_feature(ac->avccontext, "SSR", 1);
1087
            return -1;
1088
        }
1089
    }
1090

    
1091
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1092
        return -1;
1093

    
1094
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1095
        apply_prediction(ac, sce);
1096

    
1097
    return 0;
1098
}
1099

    
1100
/**
1101
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1102
 */
1103
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1104
{
1105
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1106
    float *ch0 = cpe->ch[0].coeffs;
1107
    float *ch1 = cpe->ch[1].coeffs;
1108
    int g, i, group, idx = 0;
1109
    const uint16_t *offsets = ics->swb_offset;
1110
    for (g = 0; g < ics->num_window_groups; g++) {
1111
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1112
            if (cpe->ms_mask[idx] &&
1113
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1114
                for (group = 0; group < ics->group_len[g]; group++) {
1115
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1116
                                              ch1 + group * 128 + offsets[i],
1117
                                              offsets[i+1] - offsets[i]);
1118
                }
1119
            }
1120
        }
1121
        ch0 += ics->group_len[g] * 128;
1122
        ch1 += ics->group_len[g] * 128;
1123
    }
1124
}
1125

    
1126
/**
1127
 * intensity stereo decoding; reference: 4.6.8.2.3
1128
 *
1129
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1130
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1131
 *                      [3] reserved for scalable AAC
1132
 */
1133
static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1134
{
1135
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1136
    SingleChannelElement         *sce1 = &cpe->ch[1];
1137
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1138
    const uint16_t *offsets = ics->swb_offset;
1139
    int g, group, i, k, idx = 0;
1140
    int c;
1141
    float scale;
1142
    for (g = 0; g < ics->num_window_groups; g++) {
1143
        for (i = 0; i < ics->max_sfb;) {
1144
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1145
                const int bt_run_end = sce1->band_type_run_end[idx];
1146
                for (; i < bt_run_end; i++, idx++) {
1147
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1148
                    if (ms_present)
1149
                        c *= 1 - 2 * cpe->ms_mask[idx];
1150
                    scale = c * sce1->sf[idx];
1151
                    for (group = 0; group < ics->group_len[g]; group++)
1152
                        for (k = offsets[i]; k < offsets[i + 1]; k++)
1153
                            coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1154
                }
1155
            } else {
1156
                int bt_run_end = sce1->band_type_run_end[idx];
1157
                idx += bt_run_end - i;
1158
                i    = bt_run_end;
1159
            }
1160
        }
1161
        coef0 += ics->group_len[g] * 128;
1162
        coef1 += ics->group_len[g] * 128;
1163
    }
1164
}
1165

    
1166
/**
1167
 * Decode a channel_pair_element; reference: table 4.4.
1168
 *
1169
 * @param   elem_id Identifies the instance of a syntax element.
1170
 *
1171
 * @return  Returns error status. 0 - OK, !0 - error
1172
 */
1173
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1174
{
1175
    int i, ret, common_window, ms_present = 0;
1176

    
1177
    common_window = get_bits1(gb);
1178
    if (common_window) {
1179
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1180
            return -1;
1181
        i = cpe->ch[1].ics.use_kb_window[0];
1182
        cpe->ch[1].ics = cpe->ch[0].ics;
1183
        cpe->ch[1].ics.use_kb_window[1] = i;
1184
        ms_present = get_bits(gb, 2);
1185
        if (ms_present == 3) {
1186
            av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1187
            return -1;
1188
        } else if (ms_present)
1189
            decode_mid_side_stereo(cpe, gb, ms_present);
1190
    }
1191
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1192
        return ret;
1193
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1194
        return ret;
1195

    
1196
    if (common_window) {
1197
        if (ms_present)
1198
            apply_mid_side_stereo(ac, cpe);
1199
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1200
            apply_prediction(ac, &cpe->ch[0]);
1201
            apply_prediction(ac, &cpe->ch[1]);
1202
        }
1203
    }
1204

    
1205
    apply_intensity_stereo(cpe, ms_present);
1206
    return 0;
1207
}
1208

    
1209
/**
1210
 * Decode coupling_channel_element; reference: table 4.8.
1211
 *
1212
 * @param   elem_id Identifies the instance of a syntax element.
1213
 *
1214
 * @return  Returns error status. 0 - OK, !0 - error
1215
 */
1216
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1217
{
1218
    int num_gain = 0;
1219
    int c, g, sfb, ret;
1220
    int sign;
1221
    float scale;
1222
    SingleChannelElement *sce = &che->ch[0];
1223
    ChannelCoupling     *coup = &che->coup;
1224

    
1225
    coup->coupling_point = 2 * get_bits1(gb);
1226
    coup->num_coupled = get_bits(gb, 3);
1227
    for (c = 0; c <= coup->num_coupled; c++) {
1228
        num_gain++;
1229
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1230
        coup->id_select[c] = get_bits(gb, 4);
1231
        if (coup->type[c] == TYPE_CPE) {
1232
            coup->ch_select[c] = get_bits(gb, 2);
1233
            if (coup->ch_select[c] == 3)
1234
                num_gain++;
1235
        } else
1236
            coup->ch_select[c] = 2;
1237
    }
1238
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1239

    
1240
    sign  = get_bits(gb, 1);
1241
    scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1242

    
1243
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1244
        return ret;
1245

    
1246
    for (c = 0; c < num_gain; c++) {
1247
        int idx  = 0;
1248
        int cge  = 1;
1249
        int gain = 0;
1250
        float gain_cache = 1.;
1251
        if (c) {
1252
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1253
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1254
            gain_cache = pow(scale, -gain);
1255
        }
1256
        if (coup->coupling_point == AFTER_IMDCT) {
1257
            coup->gain[c][0] = gain_cache;
1258
        } else {
1259
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1260
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1261
                    if (sce->band_type[idx] != ZERO_BT) {
1262
                        if (!cge) {
1263
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1264
                            if (t) {
1265
                                int s = 1;
1266
                                t = gain += t;
1267
                                if (sign) {
1268
                                    s  -= 2 * (t & 0x1);
1269
                                    t >>= 1;
1270
                                }
1271
                                gain_cache = pow(scale, -t) * s;
1272
                            }
1273
                        }
1274
                        coup->gain[c][idx] = gain_cache;
1275
                    }
1276
                }
1277
            }
1278
        }
1279
    }
1280
    return 0;
1281
}
1282

    
1283
/**
1284
 * Decode Spectral Band Replication extension data; reference: table 4.55.
1285
 *
1286
 * @param   crc flag indicating the presence of CRC checksum
1287
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1288
 *
1289
 * @return  Returns number of bytes consumed from the TYPE_FIL element.
1290
 */
1291
static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1292
                                int crc, int cnt)
1293
{
1294
    // TODO : sbr_extension implementation
1295
    av_log_missing_feature(ac->avccontext, "SBR", 0);
1296
    skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1297
    return cnt;
1298
}
1299

    
1300
/**
1301
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1302
 *
1303
 * @return  Returns number of bytes consumed.
1304
 */
1305
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1306
                                         GetBitContext *gb)
1307
{
1308
    int i;
1309
    int num_excl_chan = 0;
1310

    
1311
    do {
1312
        for (i = 0; i < 7; i++)
1313
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1314
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1315

    
1316
    return num_excl_chan / 7;
1317
}
1318

    
1319
/**
1320
 * Decode dynamic range information; reference: table 4.52.
1321
 *
1322
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1323
 *
1324
 * @return  Returns number of bytes consumed.
1325
 */
1326
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1327
                                GetBitContext *gb, int cnt)
1328
{
1329
    int n             = 1;
1330
    int drc_num_bands = 1;
1331
    int i;
1332

    
1333
    /* pce_tag_present? */
1334
    if (get_bits1(gb)) {
1335
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1336
        skip_bits(gb, 4); // tag_reserved_bits
1337
        n++;
1338
    }
1339

    
1340
    /* excluded_chns_present? */
1341
    if (get_bits1(gb)) {
1342
        n += decode_drc_channel_exclusions(che_drc, gb);
1343
    }
1344

    
1345
    /* drc_bands_present? */
1346
    if (get_bits1(gb)) {
1347
        che_drc->band_incr            = get_bits(gb, 4);
1348
        che_drc->interpolation_scheme = get_bits(gb, 4);
1349
        n++;
1350
        drc_num_bands += che_drc->band_incr;
1351
        for (i = 0; i < drc_num_bands; i++) {
1352
            che_drc->band_top[i] = get_bits(gb, 8);
1353
            n++;
1354
        }
1355
    }
1356

    
1357
    /* prog_ref_level_present? */
1358
    if (get_bits1(gb)) {
1359
        che_drc->prog_ref_level = get_bits(gb, 7);
1360
        skip_bits1(gb); // prog_ref_level_reserved_bits
1361
        n++;
1362
    }
1363

    
1364
    for (i = 0; i < drc_num_bands; i++) {
1365
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1366
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1367
        n++;
1368
    }
1369

    
1370
    return n;
1371
}
1372

    
1373
/**
1374
 * Decode extension data (incomplete); reference: table 4.51.
1375
 *
1376
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1377
 *
1378
 * @return Returns number of bytes consumed
1379
 */
1380
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1381
{
1382
    int crc_flag = 0;
1383
    int res = cnt;
1384
    switch (get_bits(gb, 4)) { // extension type
1385
    case EXT_SBR_DATA_CRC:
1386
        crc_flag++;
1387
    case EXT_SBR_DATA:
1388
        res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1389
        break;
1390
    case EXT_DYNAMIC_RANGE:
1391
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1392
        break;
1393
    case EXT_FILL:
1394
    case EXT_FILL_DATA:
1395
    case EXT_DATA_ELEMENT:
1396
    default:
1397
        skip_bits_long(gb, 8 * cnt - 4);
1398
        break;
1399
    };
1400
    return res;
1401
}
1402

    
1403
/**
1404
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1405
 *
1406
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1407
 * @param   coef    spectral coefficients
1408
 */
1409
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1410
                      IndividualChannelStream *ics, int decode)
1411
{
1412
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1413
    int w, filt, m, i;
1414
    int bottom, top, order, start, end, size, inc;
1415
    float lpc[TNS_MAX_ORDER];
1416

    
1417
    for (w = 0; w < ics->num_windows; w++) {
1418
        bottom = ics->num_swb;
1419
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1420
            top    = bottom;
1421
            bottom = FFMAX(0, top - tns->length[w][filt]);
1422
            order  = tns->order[w][filt];
1423
            if (order == 0)
1424
                continue;
1425

    
1426
            // tns_decode_coef
1427
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1428

    
1429
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1430
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1431
            if ((size = end - start) <= 0)
1432
                continue;
1433
            if (tns->direction[w][filt]) {
1434
                inc = -1;
1435
                start = end - 1;
1436
            } else {
1437
                inc = 1;
1438
            }
1439
            start += w * 128;
1440

    
1441
            // ar filter
1442
            for (m = 0; m < size; m++, start += inc)
1443
                for (i = 1; i <= FFMIN(m, order); i++)
1444
                    coef[start] -= coef[start - i * inc] * lpc[i - 1];
1445
        }
1446
    }
1447
}
1448

    
1449
/**
1450
 * Conduct IMDCT and windowing.
1451
 */
1452
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1453
{
1454
    IndividualChannelStream *ics = &sce->ics;
1455
    float *in    = sce->coeffs;
1456
    float *out   = sce->ret;
1457
    float *saved = sce->saved;
1458
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1459
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1460
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1461
    float *buf  = ac->buf_mdct;
1462
    float *temp = ac->temp;
1463
    int i;
1464

    
1465
    // imdct
1466
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1467
        if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1468
            av_log(ac->avccontext, AV_LOG_WARNING,
1469
                   "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1470
                   "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1471
        for (i = 0; i < 1024; i += 128)
1472
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1473
    } else
1474
        ff_imdct_half(&ac->mdct, buf, in);
1475

    
1476
    /* window overlapping
1477
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1478
     * and long to short transitions are considered to be short to short
1479
     * transitions. This leaves just two cases (long to long and short to short)
1480
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1481
     */
1482
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1483
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1484
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, ac->add_bias, 512);
1485
    } else {
1486
        for (i = 0; i < 448; i++)
1487
            out[i] = saved[i] + ac->add_bias;
1488

    
1489
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1490
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, ac->add_bias, 64);
1491
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      ac->add_bias, 64);
1492
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      ac->add_bias, 64);
1493
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      ac->add_bias, 64);
1494
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      ac->add_bias, 64);
1495
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1496
        } else {
1497
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, ac->add_bias, 64);
1498
            for (i = 576; i < 1024; i++)
1499
                out[i] = buf[i-512] + ac->add_bias;
1500
        }
1501
    }
1502

    
1503
    // buffer update
1504
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1505
        for (i = 0; i < 64; i++)
1506
            saved[i] = temp[64 + i] - ac->add_bias;
1507
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1508
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1509
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1510
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1511
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1512
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1513
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1514
    } else { // LONG_STOP or ONLY_LONG
1515
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1516
    }
1517
}
1518

    
1519
/**
1520
 * Apply dependent channel coupling (applied before IMDCT).
1521
 *
1522
 * @param   index   index into coupling gain array
1523
 */
1524
static void apply_dependent_coupling(AACContext *ac,
1525
                                     SingleChannelElement *target,
1526
                                     ChannelElement *cce, int index)
1527
{
1528
    IndividualChannelStream *ics = &cce->ch[0].ics;
1529
    const uint16_t *offsets = ics->swb_offset;
1530
    float *dest = target->coeffs;
1531
    const float *src = cce->ch[0].coeffs;
1532
    int g, i, group, k, idx = 0;
1533
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1534
        av_log(ac->avccontext, AV_LOG_ERROR,
1535
               "Dependent coupling is not supported together with LTP\n");
1536
        return;
1537
    }
1538
    for (g = 0; g < ics->num_window_groups; g++) {
1539
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1540
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1541
                const float gain = cce->coup.gain[index][idx];
1542
                for (group = 0; group < ics->group_len[g]; group++) {
1543
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1544
                        // XXX dsputil-ize
1545
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1546
                    }
1547
                }
1548
            }
1549
        }
1550
        dest += ics->group_len[g] * 128;
1551
        src  += ics->group_len[g] * 128;
1552
    }
1553
}
1554

    
1555
/**
1556
 * Apply independent channel coupling (applied after IMDCT).
1557
 *
1558
 * @param   index   index into coupling gain array
1559
 */
1560
static void apply_independent_coupling(AACContext *ac,
1561
                                       SingleChannelElement *target,
1562
                                       ChannelElement *cce, int index)
1563
{
1564
    int i;
1565
    const float gain = cce->coup.gain[index][0];
1566
    const float bias = ac->add_bias;
1567
    const float *src = cce->ch[0].ret;
1568
    float *dest = target->ret;
1569

    
1570
    for (i = 0; i < 1024; i++)
1571
        dest[i] += gain * (src[i] - bias);
1572
}
1573

    
1574
/**
1575
 * channel coupling transformation interface
1576
 *
1577
 * @param   index   index into coupling gain array
1578
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1579
 */
1580
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1581
                                   enum RawDataBlockType type, int elem_id,
1582
                                   enum CouplingPoint coupling_point,
1583
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1584
{
1585
    int i, c;
1586

    
1587
    for (i = 0; i < MAX_ELEM_ID; i++) {
1588
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1589
        int index = 0;
1590

    
1591
        if (cce && cce->coup.coupling_point == coupling_point) {
1592
            ChannelCoupling *coup = &cce->coup;
1593

    
1594
            for (c = 0; c <= coup->num_coupled; c++) {
1595
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1596
                    if (coup->ch_select[c] != 1) {
1597
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1598
                        if (coup->ch_select[c] != 0)
1599
                            index++;
1600
                    }
1601
                    if (coup->ch_select[c] != 2)
1602
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1603
                } else
1604
                    index += 1 + (coup->ch_select[c] == 3);
1605
            }
1606
        }
1607
    }
1608
}
1609

    
1610
/**
1611
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1612
 */
1613
static void spectral_to_sample(AACContext *ac)
1614
{
1615
    int i, type;
1616
    for (type = 3; type >= 0; type--) {
1617
        for (i = 0; i < MAX_ELEM_ID; i++) {
1618
            ChannelElement *che = ac->che[type][i];
1619
            if (che) {
1620
                if (type <= TYPE_CPE)
1621
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1622
                if (che->ch[0].tns.present)
1623
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1624
                if (che->ch[1].tns.present)
1625
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1626
                if (type <= TYPE_CPE)
1627
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1628
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1629
                    imdct_and_windowing(ac, &che->ch[0]);
1630
                if (type == TYPE_CPE)
1631
                    imdct_and_windowing(ac, &che->ch[1]);
1632
                if (type <= TYPE_CCE)
1633
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1634
            }
1635
        }
1636
    }
1637
}
1638

    
1639
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1640
{
1641
    int size;
1642
    AACADTSHeaderInfo hdr_info;
1643

    
1644
    size = ff_aac_parse_header(gb, &hdr_info);
1645
    if (size > 0) {
1646
        if (!ac->output_configured && hdr_info.chan_config) {
1647
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1648
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1649
            ac->m4ac.chan_config = hdr_info.chan_config;
1650
            if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1651
                return -7;
1652
            if (output_configure(ac, ac->che_pos, new_che_pos, 1))
1653
                return -7;
1654
        }
1655
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1656
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1657
        ac->m4ac.object_type     = hdr_info.object_type;
1658
        if (hdr_info.num_aac_frames == 1) {
1659
            if (!hdr_info.crc_absent)
1660
                skip_bits(gb, 16);
1661
        } else {
1662
            av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1663
            return -1;
1664
        }
1665
    }
1666
    return size;
1667
}
1668

    
1669
static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1670
                            int *data_size, AVPacket *avpkt)
1671
{
1672
    const uint8_t *buf = avpkt->data;
1673
    int buf_size = avpkt->size;
1674
    AACContext *ac = avccontext->priv_data;
1675
    ChannelElement *che = NULL;
1676
    GetBitContext gb;
1677
    enum RawDataBlockType elem_type;
1678
    int err, elem_id, data_size_tmp;
1679

    
1680
    init_get_bits(&gb, buf, buf_size * 8);
1681

    
1682
    if (show_bits(&gb, 12) == 0xfff) {
1683
        if (parse_adts_frame_header(ac, &gb) < 0) {
1684
            av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1685
            return -1;
1686
        }
1687
        if (ac->m4ac.sampling_index > 12) {
1688
            av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1689
            return -1;
1690
        }
1691
    }
1692

    
1693
    // parse
1694
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1695
        elem_id = get_bits(&gb, 4);
1696

    
1697
        if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1698
            av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1699
            return -1;
1700
        }
1701

    
1702
        switch (elem_type) {
1703

    
1704
        case TYPE_SCE:
1705
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1706
            break;
1707

    
1708
        case TYPE_CPE:
1709
            err = decode_cpe(ac, &gb, che);
1710
            break;
1711

    
1712
        case TYPE_CCE:
1713
            err = decode_cce(ac, &gb, che);
1714
            break;
1715

    
1716
        case TYPE_LFE:
1717
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1718
            break;
1719

    
1720
        case TYPE_DSE:
1721
            skip_data_stream_element(&gb);
1722
            err = 0;
1723
            break;
1724

    
1725
        case TYPE_PCE: {
1726
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1727
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1728
            if ((err = decode_pce(ac, new_che_pos, &gb)))
1729
                break;
1730
            if (ac->output_configured)
1731
                av_log(avccontext, AV_LOG_ERROR,
1732
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1733
            else
1734
                err = output_configure(ac, ac->che_pos, new_che_pos, 0);
1735
            break;
1736
        }
1737

    
1738
        case TYPE_FIL:
1739
            if (elem_id == 15)
1740
                elem_id += get_bits(&gb, 8) - 1;
1741
            while (elem_id > 0)
1742
                elem_id -= decode_extension_payload(ac, &gb, elem_id);
1743
            err = 0; /* FIXME */
1744
            break;
1745

    
1746
        default:
1747
            err = -1; /* should not happen, but keeps compiler happy */
1748
            break;
1749
        }
1750

    
1751
        if (err)
1752
            return err;
1753
    }
1754

    
1755
    spectral_to_sample(ac);
1756

    
1757
    if (!ac->is_saved) {
1758
        ac->is_saved = 1;
1759
        *data_size = 0;
1760
        return buf_size;
1761
    }
1762

    
1763
    data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1764
    if (*data_size < data_size_tmp) {
1765
        av_log(avccontext, AV_LOG_ERROR,
1766
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1767
               *data_size, data_size_tmp);
1768
        return -1;
1769
    }
1770
    *data_size = data_size_tmp;
1771

    
1772
    ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1773

    
1774
    return buf_size;
1775
}
1776

    
1777
static av_cold int aac_decode_close(AVCodecContext *avccontext)
1778
{
1779
    AACContext *ac = avccontext->priv_data;
1780
    int i, type;
1781

    
1782
    for (i = 0; i < MAX_ELEM_ID; i++) {
1783
        for (type = 0; type < 4; type++)
1784
            av_freep(&ac->che[type][i]);
1785
    }
1786

    
1787
    ff_mdct_end(&ac->mdct);
1788
    ff_mdct_end(&ac->mdct_small);
1789
    return 0;
1790
}
1791

    
1792
AVCodec aac_decoder = {
1793
    "aac",
1794
    CODEC_TYPE_AUDIO,
1795
    CODEC_ID_AAC,
1796
    sizeof(AACContext),
1797
    aac_decode_init,
1798
    NULL,
1799
    aac_decode_close,
1800
    aac_decode_frame,
1801
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1802
    .sample_fmts = (const enum SampleFormat[]) {
1803
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
1804
    },
1805
};