Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aac.c @ 7d485f16

History | View | Annotate | Download (63.4 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 { float f; uint32_t i; };
97

    
98
static VLC vlc_scalefactors;
99
static VLC vlc_spectral[11];
100

    
101

    
102
static ChannelElement* get_che(AACContext *ac, int type, int elem_id) {
103
    static const int8_t tags_per_config[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 };
104
    if (ac->tag_che_map[type][elem_id]) {
105
        return ac->tag_che_map[type][elem_id];
106
    }
107
    if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
108
        return NULL;
109
    }
110
    switch (ac->m4ac.chan_config) {
111
        case 7:
112
            if (ac->tags_mapped == 3 && type == TYPE_CPE) {
113
                ac->tags_mapped++;
114
                return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
115
            }
116
        case 6:
117
            /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
118
               instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
119
               encountered such a stream, transfer the LFE[0] element to SCE[1] */
120
            if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
121
                ac->tags_mapped++;
122
                return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
123
            }
124
        case 5:
125
            if (ac->tags_mapped == 2 && type == TYPE_CPE) {
126
                ac->tags_mapped++;
127
                return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
128
            }
129
        case 4:
130
            if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
131
                ac->tags_mapped++;
132
                return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
133
            }
134
        case 3:
135
        case 2:
136
            if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
137
                ac->tags_mapped++;
138
                return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
139
            } else if (ac->m4ac.chan_config == 2) {
140
                return NULL;
141
            }
142
        case 1:
143
            if (!ac->tags_mapped && type == TYPE_SCE) {
144
                ac->tags_mapped++;
145
                return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
146
            }
147
        default:
148
            return NULL;
149
    }
150
}
151

    
152
/**
153
 * Configure output channel order based on the current program configuration element.
154
 *
155
 * @param   che_pos current channel position configuration
156
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
157
 *
158
 * @return  Returns error status. 0 - OK, !0 - error
159
 */
160
static int output_configure(AACContext *ac, enum ChannelPosition che_pos[4][MAX_ELEM_ID],
161
        enum ChannelPosition new_che_pos[4][MAX_ELEM_ID], int channel_config) {
162
    AVCodecContext *avctx = ac->avccontext;
163
    int i, type, channels = 0;
164

    
165
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
166

    
167
    /* Allocate or free elements depending on if they are in the
168
     * current program configuration.
169
     *
170
     * Set up default 1:1 output mapping.
171
     *
172
     * For a 5.1 stream the output order will be:
173
     *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
174
     */
175

    
176
    for(i = 0; i < MAX_ELEM_ID; i++) {
177
        for(type = 0; type < 4; type++) {
178
            if(che_pos[type][i]) {
179
                if(!ac->che[type][i] && !(ac->che[type][i] = av_mallocz(sizeof(ChannelElement))))
180
                    return AVERROR(ENOMEM);
181
                if(type != TYPE_CCE) {
182
                    ac->output_data[channels++] = ac->che[type][i]->ch[0].ret;
183
                    if(type == TYPE_CPE) {
184
                        ac->output_data[channels++] = ac->che[type][i]->ch[1].ret;
185
                    }
186
                }
187
            } else
188
                av_freep(&ac->che[type][i]);
189
        }
190
    }
191

    
192
    if (channel_config) {
193
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
194
        ac->tags_mapped = 0;
195
    } else {
196
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
197
        ac->tags_mapped = 4*MAX_ELEM_ID;
198
    }
199

    
200
    avctx->channels = channels;
201

    
202
    return 0;
203
}
204

    
205
/**
206
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
207
 *
208
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
209
 * @param sce_map mono (Single Channel Element) map
210
 * @param type speaker type/position for these channels
211
 */
212
static void decode_channel_map(enum ChannelPosition *cpe_map,
213
        enum ChannelPosition *sce_map, enum ChannelPosition type, GetBitContext * gb, int n) {
214
    while(n--) {
215
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
216
        map[get_bits(gb, 4)] = type;
217
    }
218
}
219

    
220
/**
221
 * Decode program configuration element; reference: table 4.2.
222
 *
223
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
224
 *
225
 * @return  Returns error status. 0 - OK, !0 - error
226
 */
227
static int decode_pce(AACContext * ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
228
        GetBitContext * gb) {
229
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
230

    
231
    skip_bits(gb, 2);  // object_type
232

    
233
    sampling_index = get_bits(gb, 4);
234
    if(sampling_index > 12) {
235
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
236
        return -1;
237
    }
238
    ac->m4ac.sampling_index = sampling_index;
239
    ac->m4ac.sample_rate = ff_mpeg4audio_sample_rates[ac->m4ac.sampling_index];
240
    num_front       = get_bits(gb, 4);
241
    num_side        = get_bits(gb, 4);
242
    num_back        = get_bits(gb, 4);
243
    num_lfe         = get_bits(gb, 2);
244
    num_assoc_data  = get_bits(gb, 3);
245
    num_cc          = get_bits(gb, 4);
246

    
247
    if (get_bits1(gb))
248
        skip_bits(gb, 4); // mono_mixdown_tag
249
    if (get_bits1(gb))
250
        skip_bits(gb, 4); // stereo_mixdown_tag
251

    
252
    if (get_bits1(gb))
253
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
254

    
255
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
256
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
257
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
258
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
259

    
260
    skip_bits_long(gb, 4 * num_assoc_data);
261

    
262
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
263

    
264
    align_get_bits(gb);
265

    
266
    /* comment field, first byte is length */
267
    skip_bits_long(gb, 8 * get_bits(gb, 8));
268
    return 0;
269
}
270

    
271
/**
272
 * Set up channel positions based on a default channel configuration
273
 * as specified in table 1.17.
274
 *
275
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
276
 *
277
 * @return  Returns error status. 0 - OK, !0 - error
278
 */
279
static int set_default_channel_config(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
280
        int channel_config)
281
{
282
    if(channel_config < 1 || channel_config > 7) {
283
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
284
               channel_config);
285
        return -1;
286
    }
287

    
288
    /* default channel configurations:
289
     *
290
     * 1ch : front center (mono)
291
     * 2ch : L + R (stereo)
292
     * 3ch : front center + L + R
293
     * 4ch : front center + L + R + back center
294
     * 5ch : front center + L + R + back stereo
295
     * 6ch : front center + L + R + back stereo + LFE
296
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
297
     */
298

    
299
    if(channel_config != 2)
300
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
301
    if(channel_config > 1)
302
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
303
    if(channel_config == 4)
304
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
305
    if(channel_config > 4)
306
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
307
                                 = AAC_CHANNEL_BACK;  // back stereo
308
    if(channel_config > 5)
309
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
310
    if(channel_config == 7)
311
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
312

    
313
    return 0;
314
}
315

    
316
/**
317
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
318
 *
319
 * @return  Returns error status. 0 - OK, !0 - error
320
 */
321
static int decode_ga_specific_config(AACContext * ac, GetBitContext * gb, int channel_config) {
322
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
323
    int extension_flag, ret;
324

    
325
    if(get_bits1(gb)) {  // frameLengthFlag
326
        ff_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
327
        return -1;
328
    }
329

    
330
    if (get_bits1(gb))       // dependsOnCoreCoder
331
        skip_bits(gb, 14);   // coreCoderDelay
332
    extension_flag = get_bits1(gb);
333

    
334
    if(ac->m4ac.object_type == AOT_AAC_SCALABLE ||
335
       ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
336
        skip_bits(gb, 3);     // layerNr
337

    
338
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
339
    if (channel_config == 0) {
340
        skip_bits(gb, 4);  // element_instance_tag
341
        if((ret = decode_pce(ac, new_che_pos, gb)))
342
            return ret;
343
    } else {
344
        if((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
345
            return ret;
346
    }
347
    if((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config)))
348
        return ret;
349

    
350
    if (extension_flag) {
351
        switch (ac->m4ac.object_type) {
352
            case AOT_ER_BSAC:
353
                skip_bits(gb, 5);    // numOfSubFrame
354
                skip_bits(gb, 11);   // layer_length
355
                break;
356
            case AOT_ER_AAC_LC:
357
            case AOT_ER_AAC_LTP:
358
            case AOT_ER_AAC_SCALABLE:
359
            case AOT_ER_AAC_LD:
360
                skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
361
                                    * aacScalefactorDataResilienceFlag
362
                                    * aacSpectralDataResilienceFlag
363
                                    */
364
                break;
365
        }
366
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
367
    }
368
    return 0;
369
}
370

    
371
/**
372
 * Decode audio specific configuration; reference: table 1.13.
373
 *
374
 * @param   data        pointer to AVCodecContext extradata
375
 * @param   data_size   size of AVCCodecContext extradata
376
 *
377
 * @return  Returns error status. 0 - OK, !0 - error
378
 */
379
static int decode_audio_specific_config(AACContext * ac, void *data, int data_size) {
380
    GetBitContext gb;
381
    int i;
382

    
383
    init_get_bits(&gb, data, data_size * 8);
384

    
385
    if((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
386
        return -1;
387
    if(ac->m4ac.sampling_index > 12) {
388
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
389
        return -1;
390
    }
391

    
392
    skip_bits_long(&gb, i);
393

    
394
    switch (ac->m4ac.object_type) {
395
    case AOT_AAC_MAIN:
396
    case AOT_AAC_LC:
397
        if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
398
            return -1;
399
        break;
400
    default:
401
        av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
402
               ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
403
        return -1;
404
    }
405
    return 0;
406
}
407

    
408
/**
409
 * linear congruential pseudorandom number generator
410
 *
411
 * @param   previous_val    pointer to the current state of the generator
412
 *
413
 * @return  Returns a 32-bit pseudorandom integer
414
 */
415
static av_always_inline int lcg_random(int previous_val) {
416
    return previous_val * 1664525 + 1013904223;
417
}
418

    
419
static void reset_predict_state(PredictorState * ps) {
420
    ps->r0 = 0.0f;
421
    ps->r1 = 0.0f;
422
    ps->cor0 = 0.0f;
423
    ps->cor1 = 0.0f;
424
    ps->var0 = 1.0f;
425
    ps->var1 = 1.0f;
426
}
427

    
428
static void reset_all_predictors(PredictorState * ps) {
429
    int i;
430
    for (i = 0; i < MAX_PREDICTORS; i++)
431
        reset_predict_state(&ps[i]);
432
}
433

    
434
static void reset_predictor_group(PredictorState * ps, int group_num) {
435
    int i;
436
    for (i = group_num-1; i < MAX_PREDICTORS; i+=30)
437
        reset_predict_state(&ps[i]);
438
}
439

    
440
static av_cold int aac_decode_init(AVCodecContext * avccontext) {
441
    AACContext * ac = avccontext->priv_data;
442
    int i;
443

    
444
    ac->avccontext = avccontext;
445

    
446
    if (avccontext->extradata_size > 0) {
447
        if(decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
448
            return -1;
449
        avccontext->sample_rate = ac->m4ac.sample_rate;
450
    } else if (avccontext->channels > 0) {
451
        enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
452
        memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
453
        if(set_default_channel_config(ac, new_che_pos, avccontext->channels - (avccontext->channels == 8)))
454
            return -1;
455
        if(output_configure(ac, ac->che_pos, new_che_pos, 1))
456
            return -1;
457
        ac->m4ac.sample_rate = avccontext->sample_rate;
458
    }
459

    
460
    avccontext->sample_fmt  = SAMPLE_FMT_S16;
461
    avccontext->frame_size  = 1024;
462

    
463
    AAC_INIT_VLC_STATIC( 0, 144);
464
    AAC_INIT_VLC_STATIC( 1, 114);
465
    AAC_INIT_VLC_STATIC( 2, 188);
466
    AAC_INIT_VLC_STATIC( 3, 180);
467
    AAC_INIT_VLC_STATIC( 4, 172);
468
    AAC_INIT_VLC_STATIC( 5, 140);
469
    AAC_INIT_VLC_STATIC( 6, 168);
470
    AAC_INIT_VLC_STATIC( 7, 114);
471
    AAC_INIT_VLC_STATIC( 8, 262);
472
    AAC_INIT_VLC_STATIC( 9, 248);
473
    AAC_INIT_VLC_STATIC(10, 384);
474

    
475
    dsputil_init(&ac->dsp, avccontext);
476

    
477
    ac->random_state = 0x1f2e3d4c;
478

    
479
    // -1024 - Compensate wrong IMDCT method.
480
    // 32768 - Required to scale values to the correct range for the bias method
481
    //         for float to int16 conversion.
482

    
483
    if(ac->dsp.float_to_int16 == ff_float_to_int16_c) {
484
        ac->add_bias = 385.0f;
485
        ac->sf_scale = 1. / (-1024. * 32768.);
486
        ac->sf_offset = 0;
487
    } else {
488
        ac->add_bias = 0.0f;
489
        ac->sf_scale = 1. / -1024.;
490
        ac->sf_offset = 60;
491
    }
492

    
493
#if !CONFIG_HARDCODED_TABLES
494
    for (i = 0; i < 428; i++)
495
        ff_aac_pow2sf_tab[i] = pow(2, (i - 200)/4.);
496
#endif /* CONFIG_HARDCODED_TABLES */
497

    
498
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
499
        ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
500
        ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
501
        352);
502

    
503
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
504
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
505
    // window initialization
506
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
507
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
508
    ff_sine_window_init(ff_sine_1024, 1024);
509
    ff_sine_window_init(ff_sine_128, 128);
510

    
511
    return 0;
512
}
513

    
514
/**
515
 * Skip data_stream_element; reference: table 4.10.
516
 */
517
static void skip_data_stream_element(GetBitContext * gb) {
518
    int byte_align = get_bits1(gb);
519
    int count = get_bits(gb, 8);
520
    if (count == 255)
521
        count += get_bits(gb, 8);
522
    if (byte_align)
523
        align_get_bits(gb);
524
    skip_bits_long(gb, 8 * count);
525
}
526

    
527
static int decode_prediction(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb) {
528
    int sfb;
529
    if (get_bits1(gb)) {
530
        ics->predictor_reset_group = get_bits(gb, 5);
531
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
532
            av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
533
            return -1;
534
        }
535
    }
536
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
537
        ics->prediction_used[sfb] = get_bits1(gb);
538
    }
539
    return 0;
540
}
541

    
542
/**
543
 * Decode Individual Channel Stream info; reference: table 4.6.
544
 *
545
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
546
 */
547
static int decode_ics_info(AACContext * ac, IndividualChannelStream * ics, GetBitContext * gb, int common_window) {
548
    if (get_bits1(gb)) {
549
        av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
550
        memset(ics, 0, sizeof(IndividualChannelStream));
551
        return -1;
552
    }
553
    ics->window_sequence[1] = ics->window_sequence[0];
554
    ics->window_sequence[0] = get_bits(gb, 2);
555
    ics->use_kb_window[1] = ics->use_kb_window[0];
556
    ics->use_kb_window[0] = get_bits1(gb);
557
    ics->num_window_groups = 1;
558
    ics->group_len[0] = 1;
559
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
560
        int i;
561
        ics->max_sfb = get_bits(gb, 4);
562
        for (i = 0; i < 7; i++) {
563
            if (get_bits1(gb)) {
564
                ics->group_len[ics->num_window_groups-1]++;
565
            } else {
566
                ics->num_window_groups++;
567
                ics->group_len[ics->num_window_groups-1] = 1;
568
            }
569
        }
570
        ics->num_windows   = 8;
571
        ics->swb_offset    =      swb_offset_128[ac->m4ac.sampling_index];
572
        ics->num_swb       =  ff_aac_num_swb_128[ac->m4ac.sampling_index];
573
        ics->tns_max_bands =   tns_max_bands_128[ac->m4ac.sampling_index];
574
        ics->predictor_present = 0;
575
    } else {
576
        ics->max_sfb       = get_bits(gb, 6);
577
        ics->num_windows   = 1;
578
        ics->swb_offset    =     swb_offset_1024[ac->m4ac.sampling_index];
579
        ics->num_swb       = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
580
        ics->tns_max_bands =  tns_max_bands_1024[ac->m4ac.sampling_index];
581
        ics->predictor_present = get_bits1(gb);
582
        ics->predictor_reset_group = 0;
583
        if (ics->predictor_present) {
584
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
585
                if (decode_prediction(ac, ics, gb)) {
586
                    memset(ics, 0, sizeof(IndividualChannelStream));
587
                    return -1;
588
                }
589
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
590
                av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
591
                memset(ics, 0, sizeof(IndividualChannelStream));
592
                return -1;
593
            } else {
594
                ff_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
595
                memset(ics, 0, sizeof(IndividualChannelStream));
596
                return -1;
597
            }
598
        }
599
    }
600

    
601
    if(ics->max_sfb > ics->num_swb) {
602
        av_log(ac->avccontext, AV_LOG_ERROR,
603
            "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
604
            ics->max_sfb, ics->num_swb);
605
        memset(ics, 0, sizeof(IndividualChannelStream));
606
        return -1;
607
    }
608

    
609
    return 0;
610
}
611

    
612
/**
613
 * Decode band types (section_data payload); reference: table 4.46.
614
 *
615
 * @param   band_type           array of the used band type
616
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
617
 *
618
 * @return  Returns error status. 0 - OK, !0 - error
619
 */
620
static int decode_band_types(AACContext * ac, enum BandType band_type[120],
621
        int band_type_run_end[120], GetBitContext * gb, IndividualChannelStream * ics) {
622
    int g, idx = 0;
623
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
624
    for (g = 0; g < ics->num_window_groups; g++) {
625
        int k = 0;
626
        while (k < ics->max_sfb) {
627
            uint8_t sect_len = k;
628
            int sect_len_incr;
629
            int sect_band_type = get_bits(gb, 4);
630
            if (sect_band_type == 12) {
631
                av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
632
                return -1;
633
            }
634
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits)-1)
635
                sect_len += sect_len_incr;
636
            sect_len += sect_len_incr;
637
            if (sect_len > ics->max_sfb) {
638
                av_log(ac->avccontext, AV_LOG_ERROR,
639
                    "Number of bands (%d) exceeds limit (%d).\n",
640
                    sect_len, ics->max_sfb);
641
                return -1;
642
            }
643
            for (; k < sect_len; k++) {
644
                band_type        [idx]   = sect_band_type;
645
                band_type_run_end[idx++] = sect_len;
646
            }
647
        }
648
    }
649
    return 0;
650
}
651

    
652
/**
653
 * Decode scalefactors; reference: table 4.47.
654
 *
655
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
656
 * @param   band_type           array of the used band type
657
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
658
 * @param   sf                  array of scalefactors or intensity stereo positions
659
 *
660
 * @return  Returns error status. 0 - OK, !0 - error
661
 */
662
static int decode_scalefactors(AACContext * ac, float sf[120], GetBitContext * gb,
663
        unsigned int global_gain, IndividualChannelStream * ics,
664
        enum BandType band_type[120], int band_type_run_end[120]) {
665
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
666
    int g, i, idx = 0;
667
    int offset[3] = { global_gain, global_gain - 90, 100 };
668
    int noise_flag = 1;
669
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
670
    for (g = 0; g < ics->num_window_groups; g++) {
671
        for (i = 0; i < ics->max_sfb;) {
672
            int run_end = band_type_run_end[idx];
673
            if (band_type[idx] == ZERO_BT) {
674
                for(; i < run_end; i++, idx++)
675
                    sf[idx] = 0.;
676
            }else if((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
677
                for(; i < run_end; i++, idx++) {
678
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
679
                    if(offset[2] > 255U) {
680
                        av_log(ac->avccontext, AV_LOG_ERROR,
681
                            "%s (%d) out of range.\n", sf_str[2], offset[2]);
682
                        return -1;
683
                    }
684
                    sf[idx]  = ff_aac_pow2sf_tab[-offset[2] + 300];
685
                }
686
            }else if(band_type[idx] == NOISE_BT) {
687
                for(; i < run_end; i++, idx++) {
688
                    if(noise_flag-- > 0)
689
                        offset[1] += get_bits(gb, 9) - 256;
690
                    else
691
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
692
                    if(offset[1] > 255U) {
693
                        av_log(ac->avccontext, AV_LOG_ERROR,
694
                            "%s (%d) out of range.\n", sf_str[1], offset[1]);
695
                        return -1;
696
                    }
697
                    sf[idx]  = -ff_aac_pow2sf_tab[ offset[1] + sf_offset + 100];
698
                }
699
            }else {
700
                for(; i < run_end; i++, idx++) {
701
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
702
                    if(offset[0] > 255U) {
703
                        av_log(ac->avccontext, AV_LOG_ERROR,
704
                            "%s (%d) out of range.\n", sf_str[0], offset[0]);
705
                        return -1;
706
                    }
707
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
708
                }
709
            }
710
        }
711
    }
712
    return 0;
713
}
714

    
715
/**
716
 * Decode pulse data; reference: table 4.7.
717
 */
718
static int decode_pulses(Pulse * pulse, GetBitContext * gb, const uint16_t * swb_offset, int num_swb) {
719
    int i, pulse_swb;
720
    pulse->num_pulse = get_bits(gb, 2) + 1;
721
    pulse_swb        = get_bits(gb, 6);
722
    if (pulse_swb >= num_swb)
723
        return -1;
724
    pulse->pos[0]    = swb_offset[pulse_swb];
725
    pulse->pos[0]   += get_bits(gb, 5);
726
    if (pulse->pos[0] > 1023)
727
        return -1;
728
    pulse->amp[0]    = get_bits(gb, 4);
729
    for (i = 1; i < pulse->num_pulse; i++) {
730
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i-1];
731
        if (pulse->pos[i] > 1023)
732
            return -1;
733
        pulse->amp[i] = get_bits(gb, 4);
734
    }
735
    return 0;
736
}
737

    
738
/**
739
 * Decode Temporal Noise Shaping data; reference: table 4.48.
740
 *
741
 * @return  Returns error status. 0 - OK, !0 - error
742
 */
743
static int decode_tns(AACContext * ac, TemporalNoiseShaping * tns,
744
        GetBitContext * gb, const IndividualChannelStream * ics) {
745
    int w, filt, i, coef_len, coef_res, coef_compress;
746
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
747
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
748
    for (w = 0; w < ics->num_windows; w++) {
749
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
750
            coef_res = get_bits1(gb);
751

    
752
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
753
                int tmp2_idx;
754
                tns->length[w][filt] = get_bits(gb, 6 - 2*is8);
755

    
756
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2*is8)) > tns_max_order) {
757
                    av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
758
                           tns->order[w][filt], tns_max_order);
759
                    tns->order[w][filt] = 0;
760
                    return -1;
761
                }
762
                if (tns->order[w][filt]) {
763
                    tns->direction[w][filt] = get_bits1(gb);
764
                    coef_compress = get_bits1(gb);
765
                    coef_len = coef_res + 3 - coef_compress;
766
                    tmp2_idx = 2*coef_compress + coef_res;
767

    
768
                    for (i = 0; i < tns->order[w][filt]; i++)
769
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
770
                }
771
            }
772
        }
773
    }
774
    return 0;
775
}
776

    
777
/**
778
 * Decode Mid/Side data; reference: table 4.54.
779
 *
780
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
781
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
782
 *                      [3] reserved for scalable AAC
783
 */
784
static void decode_mid_side_stereo(ChannelElement * cpe, GetBitContext * gb,
785
        int ms_present) {
786
    int idx;
787
    if (ms_present == 1) {
788
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
789
            cpe->ms_mask[idx] = get_bits1(gb);
790
    } else if (ms_present == 2) {
791
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
792
    }
793
}
794

    
795
/**
796
 * Decode spectral data; reference: table 4.50.
797
 * Dequantize and scale spectral data; reference: 4.6.3.3.
798
 *
799
 * @param   coef            array of dequantized, scaled spectral data
800
 * @param   sf              array of scalefactors or intensity stereo positions
801
 * @param   pulse_present   set if pulses are present
802
 * @param   pulse           pointer to pulse data struct
803
 * @param   band_type       array of the used band type
804
 *
805
 * @return  Returns error status. 0 - OK, !0 - error
806
 */
807
static int decode_spectrum_and_dequant(AACContext * ac, float coef[1024], GetBitContext * gb, float sf[120],
808
        int pulse_present, const Pulse * pulse, const IndividualChannelStream * ics, enum BandType band_type[120]) {
809
    int i, k, g, idx = 0;
810
    const int c = 1024/ics->num_windows;
811
    const uint16_t * offsets = ics->swb_offset;
812
    float *coef_base = coef;
813
    static const float sign_lookup[] = { 1.0f, -1.0f };
814

    
815
    for (g = 0; g < ics->num_windows; g++)
816
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float)*(c - offsets[ics->max_sfb]));
817

    
818
    for (g = 0; g < ics->num_window_groups; g++) {
819
        for (i = 0; i < ics->max_sfb; i++, idx++) {
820
            const int cur_band_type = band_type[idx];
821
            const int dim = cur_band_type >= FIRST_PAIR_BT ? 2 : 4;
822
            const int is_cb_unsigned = IS_CODEBOOK_UNSIGNED(cur_band_type);
823
            int group;
824
            if (cur_band_type == ZERO_BT || cur_band_type == INTENSITY_BT2 || cur_band_type == INTENSITY_BT) {
825
                for (group = 0; group < ics->group_len[g]; group++) {
826
                    memset(coef + group * 128 + offsets[i], 0, (offsets[i+1] - offsets[i])*sizeof(float));
827
                }
828
            }else if (cur_band_type == NOISE_BT) {
829
                for (group = 0; group < ics->group_len[g]; group++) {
830
                    float scale;
831
                    float band_energy = 0;
832
                    for (k = offsets[i]; k < offsets[i+1]; k++) {
833
                        ac->random_state  = lcg_random(ac->random_state);
834
                        coef[group*128+k] = ac->random_state;
835
                        band_energy += coef[group*128+k]*coef[group*128+k];
836
                    }
837
                    scale = sf[idx] / sqrtf(band_energy);
838
                    for (k = offsets[i]; k < offsets[i+1]; k++) {
839
                        coef[group*128+k] *= scale;
840
                    }
841
                }
842
            }else {
843
                for (group = 0; group < ics->group_len[g]; group++) {
844
                    for (k = offsets[i]; k < offsets[i+1]; k += dim) {
845
                        const int index = get_vlc2(gb, vlc_spectral[cur_band_type - 1].table, 6, 3);
846
                        const int coef_tmp_idx = (group << 7) + k;
847
                        const float *vq_ptr;
848
                        int j;
849
                        if(index >= ff_aac_spectral_sizes[cur_band_type - 1]) {
850
                            av_log(ac->avccontext, AV_LOG_ERROR,
851
                                "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
852
                                cur_band_type - 1, index, ff_aac_spectral_sizes[cur_band_type - 1]);
853
                            return -1;
854
                        }
855
                        vq_ptr = &ff_aac_codebook_vectors[cur_band_type - 1][index * dim];
856
                        if (is_cb_unsigned) {
857
                            if (vq_ptr[0]) coef[coef_tmp_idx    ] = sign_lookup[get_bits1(gb)];
858
                            if (vq_ptr[1]) coef[coef_tmp_idx + 1] = sign_lookup[get_bits1(gb)];
859
                            if (dim == 4) {
860
                                if (vq_ptr[2]) coef[coef_tmp_idx + 2] = sign_lookup[get_bits1(gb)];
861
                                if (vq_ptr[3]) coef[coef_tmp_idx + 3] = sign_lookup[get_bits1(gb)];
862
                            }
863
                            if (cur_band_type == ESC_BT) {
864
                                for (j = 0; j < 2; j++) {
865
                                    if (vq_ptr[j] == 64.0f) {
866
                                        int n = 4;
867
                                        /* The total length of escape_sequence must be < 22 bits according
868
                                           to the specification (i.e. max is 11111111110xxxxxxxxxx). */
869
                                        while (get_bits1(gb) && n < 15) n++;
870
                                        if(n == 15) {
871
                                            av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
872
                                            return -1;
873
                                        }
874
                                        n = (1<<n) + get_bits(gb, n);
875
                                        coef[coef_tmp_idx + j] *= cbrtf(n) * n;
876
                                    }else
877
                                        coef[coef_tmp_idx + j] *= vq_ptr[j];
878
                                }
879
                            }else
880
                            {
881
                                coef[coef_tmp_idx    ] *= vq_ptr[0];
882
                                coef[coef_tmp_idx + 1] *= vq_ptr[1];
883
                                if (dim == 4) {
884
                                    coef[coef_tmp_idx + 2] *= vq_ptr[2];
885
                                    coef[coef_tmp_idx + 3] *= vq_ptr[3];
886
                                }
887
                            }
888
                        }else {
889
                            coef[coef_tmp_idx    ] = vq_ptr[0];
890
                            coef[coef_tmp_idx + 1] = vq_ptr[1];
891
                            if (dim == 4) {
892
                                coef[coef_tmp_idx + 2] = vq_ptr[2];
893
                                coef[coef_tmp_idx + 3] = vq_ptr[3];
894
                            }
895
                        }
896
                        coef[coef_tmp_idx    ] *= sf[idx];
897
                        coef[coef_tmp_idx + 1] *= sf[idx];
898
                        if (dim == 4) {
899
                            coef[coef_tmp_idx + 2] *= sf[idx];
900
                            coef[coef_tmp_idx + 3] *= sf[idx];
901
                        }
902
                    }
903
                }
904
            }
905
        }
906
        coef += ics->group_len[g]<<7;
907
    }
908

    
909
    if (pulse_present) {
910
        idx = 0;
911
        for(i = 0; i < pulse->num_pulse; i++){
912
            float co  = coef_base[ pulse->pos[i] ];
913
            while(offsets[idx + 1] <= pulse->pos[i])
914
                idx++;
915
            if (band_type[idx] != NOISE_BT && sf[idx]) {
916
                float ico = -pulse->amp[i];
917
                if (co) {
918
                    co /= sf[idx];
919
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
920
                }
921
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
922
            }
923
        }
924
    }
925
    return 0;
926
}
927

    
928
static av_always_inline float flt16_round(float pf) {
929
    union float754 tmp;
930
    tmp.f = pf;
931
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
932
    return tmp.f;
933
}
934

    
935
static av_always_inline float flt16_even(float pf) {
936
    union float754 tmp;
937
    tmp.f = pf;
938
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U>>16)) & 0xFFFF0000U;
939
    return tmp.f;
940
}
941

    
942
static av_always_inline float flt16_trunc(float pf) {
943
    union float754 pun;
944
    pun.f = pf;
945
    pun.i &= 0xFFFF0000U;
946
    return pun.f;
947
}
948

    
949
static void predict(AACContext * ac, PredictorState * ps, float* coef, int output_enable) {
950
    const float a     = 0.953125; // 61.0/64
951
    const float alpha = 0.90625;  // 29.0/32
952
    float e0, e1;
953
    float pv;
954
    float k1, k2;
955

    
956
    k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
957
    k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
958

    
959
    pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
960
    if (output_enable)
961
        *coef += pv * ac->sf_scale;
962

    
963
    e0 = *coef / ac->sf_scale;
964
    e1 = e0 - k1 * ps->r0;
965

    
966
    ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
967
    ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
968
    ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
969
    ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
970

    
971
    ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
972
    ps->r0 = flt16_trunc(a * e0);
973
}
974

    
975
/**
976
 * Apply AAC-Main style frequency domain prediction.
977
 */
978
static void apply_prediction(AACContext * ac, SingleChannelElement * sce) {
979
    int sfb, k;
980

    
981
    if (!sce->ics.predictor_initialized) {
982
        reset_all_predictors(sce->predictor_state);
983
        sce->ics.predictor_initialized = 1;
984
    }
985

    
986
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
987
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
988
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
989
                predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
990
                    sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
991
            }
992
        }
993
        if (sce->ics.predictor_reset_group)
994
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
995
    } else
996
        reset_all_predictors(sce->predictor_state);
997
}
998

    
999
/**
1000
 * Decode an individual_channel_stream payload; reference: table 4.44.
1001
 *
1002
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1003
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1004
 *
1005
 * @return  Returns error status. 0 - OK, !0 - error
1006
 */
1007
static int decode_ics(AACContext * ac, SingleChannelElement * sce, GetBitContext * gb, int common_window, int scale_flag) {
1008
    Pulse pulse;
1009
    TemporalNoiseShaping * tns = &sce->tns;
1010
    IndividualChannelStream * ics = &sce->ics;
1011
    float * out = sce->coeffs;
1012
    int global_gain, pulse_present = 0;
1013

    
1014
    /* This assignment is to silence a GCC warning about the variable being used
1015
     * uninitialized when in fact it always is.
1016
     */
1017
    pulse.num_pulse = 0;
1018

    
1019
    global_gain = get_bits(gb, 8);
1020

    
1021
    if (!common_window && !scale_flag) {
1022
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1023
            return -1;
1024
    }
1025

    
1026
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1027
        return -1;
1028
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1029
        return -1;
1030

    
1031
    pulse_present = 0;
1032
    if (!scale_flag) {
1033
        if ((pulse_present = get_bits1(gb))) {
1034
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1035
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1036
                return -1;
1037
            }
1038
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1039
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1040
                return -1;
1041
            }
1042
        }
1043
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1044
            return -1;
1045
        if (get_bits1(gb)) {
1046
            ff_log_missing_feature(ac->avccontext, "SSR", 1);
1047
            return -1;
1048
        }
1049
    }
1050

    
1051
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1052
        return -1;
1053

    
1054
    if(ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1055
        apply_prediction(ac, sce);
1056

    
1057
    return 0;
1058
}
1059

    
1060
/**
1061
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1062
 */
1063
static void apply_mid_side_stereo(ChannelElement * cpe) {
1064
    const IndividualChannelStream * ics = &cpe->ch[0].ics;
1065
    float *ch0 = cpe->ch[0].coeffs;
1066
    float *ch1 = cpe->ch[1].coeffs;
1067
    int g, i, k, group, idx = 0;
1068
    const uint16_t * offsets = ics->swb_offset;
1069
    for (g = 0; g < ics->num_window_groups; g++) {
1070
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1071
            if (cpe->ms_mask[idx] &&
1072
                cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1073
                for (group = 0; group < ics->group_len[g]; group++) {
1074
                    for (k = offsets[i]; k < offsets[i+1]; k++) {
1075
                        float tmp = ch0[group*128 + k] - ch1[group*128 + k];
1076
                        ch0[group*128 + k] += ch1[group*128 + k];
1077
                        ch1[group*128 + k] = tmp;
1078
                    }
1079
                }
1080
            }
1081
        }
1082
        ch0 += ics->group_len[g]*128;
1083
        ch1 += ics->group_len[g]*128;
1084
    }
1085
}
1086

    
1087
/**
1088
 * intensity stereo decoding; reference: 4.6.8.2.3
1089
 *
1090
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1091
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1092
 *                      [3] reserved for scalable AAC
1093
 */
1094
static void apply_intensity_stereo(ChannelElement * cpe, int ms_present) {
1095
    const IndividualChannelStream * ics = &cpe->ch[1].ics;
1096
    SingleChannelElement * sce1 = &cpe->ch[1];
1097
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1098
    const uint16_t * offsets = ics->swb_offset;
1099
    int g, group, i, k, idx = 0;
1100
    int c;
1101
    float scale;
1102
    for (g = 0; g < ics->num_window_groups; g++) {
1103
        for (i = 0; i < ics->max_sfb;) {
1104
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1105
                const int bt_run_end = sce1->band_type_run_end[idx];
1106
                for (; i < bt_run_end; i++, idx++) {
1107
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1108
                    if (ms_present)
1109
                        c *= 1 - 2 * cpe->ms_mask[idx];
1110
                    scale = c * sce1->sf[idx];
1111
                    for (group = 0; group < ics->group_len[g]; group++)
1112
                        for (k = offsets[i]; k < offsets[i+1]; k++)
1113
                            coef1[group*128 + k] = scale * coef0[group*128 + k];
1114
                }
1115
            } else {
1116
                int bt_run_end = sce1->band_type_run_end[idx];
1117
                idx += bt_run_end - i;
1118
                i    = bt_run_end;
1119
            }
1120
        }
1121
        coef0 += ics->group_len[g]*128;
1122
        coef1 += ics->group_len[g]*128;
1123
    }
1124
}
1125

    
1126
/**
1127
 * Decode a channel_pair_element; reference: table 4.4.
1128
 *
1129
 * @param   elem_id Identifies the instance of a syntax element.
1130
 *
1131
 * @return  Returns error status. 0 - OK, !0 - error
1132
 */
1133
static int decode_cpe(AACContext * ac, GetBitContext * gb, ChannelElement * cpe) {
1134
    int i, ret, common_window, ms_present = 0;
1135

    
1136
    common_window = get_bits1(gb);
1137
    if (common_window) {
1138
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1139
            return -1;
1140
        i = cpe->ch[1].ics.use_kb_window[0];
1141
        cpe->ch[1].ics = cpe->ch[0].ics;
1142
        cpe->ch[1].ics.use_kb_window[1] = i;
1143
        ms_present = get_bits(gb, 2);
1144
        if(ms_present == 3) {
1145
            av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1146
            return -1;
1147
        } else if(ms_present)
1148
            decode_mid_side_stereo(cpe, gb, ms_present);
1149
    }
1150
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1151
        return ret;
1152
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1153
        return ret;
1154

    
1155
    if (common_window) {
1156
        if (ms_present)
1157
            apply_mid_side_stereo(cpe);
1158
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1159
            apply_prediction(ac, &cpe->ch[0]);
1160
            apply_prediction(ac, &cpe->ch[1]);
1161
        }
1162
    }
1163

    
1164
    apply_intensity_stereo(cpe, ms_present);
1165
    return 0;
1166
}
1167

    
1168
/**
1169
 * Decode coupling_channel_element; reference: table 4.8.
1170
 *
1171
 * @param   elem_id Identifies the instance of a syntax element.
1172
 *
1173
 * @return  Returns error status. 0 - OK, !0 - error
1174
 */
1175
static int decode_cce(AACContext * ac, GetBitContext * gb, ChannelElement * che) {
1176
    int num_gain = 0;
1177
    int c, g, sfb, ret;
1178
    int sign;
1179
    float scale;
1180
    SingleChannelElement * sce = &che->ch[0];
1181
    ChannelCoupling * coup     = &che->coup;
1182

    
1183
    coup->coupling_point = 2*get_bits1(gb);
1184
    coup->num_coupled = get_bits(gb, 3);
1185
    for (c = 0; c <= coup->num_coupled; c++) {
1186
        num_gain++;
1187
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1188
        coup->id_select[c] = get_bits(gb, 4);
1189
        if (coup->type[c] == TYPE_CPE) {
1190
            coup->ch_select[c] = get_bits(gb, 2);
1191
            if (coup->ch_select[c] == 3)
1192
                num_gain++;
1193
        } else
1194
            coup->ch_select[c] = 2;
1195
    }
1196
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point>>1);
1197

    
1198
    sign = get_bits(gb, 1);
1199
    scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1200

    
1201
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1202
        return ret;
1203

    
1204
    for (c = 0; c < num_gain; c++) {
1205
        int idx = 0;
1206
        int cge = 1;
1207
        int gain = 0;
1208
        float gain_cache = 1.;
1209
        if (c) {
1210
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1211
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1212
            gain_cache = pow(scale, -gain);
1213
        }
1214
        if (coup->coupling_point == AFTER_IMDCT) {
1215
            coup->gain[c][0] = gain_cache;
1216
        } else {
1217
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1218
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1219
                    if (sce->band_type[idx] != ZERO_BT) {
1220
                        if (!cge) {
1221
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1222
                                if (t) {
1223
                                int s = 1;
1224
                                t = gain += t;
1225
                                if (sign) {
1226
                                    s  -= 2 * (t & 0x1);
1227
                                    t >>= 1;
1228
                                }
1229
                                gain_cache = pow(scale, -t) * s;
1230
                            }
1231
                        }
1232
                        coup->gain[c][idx] = gain_cache;
1233
                    }
1234
                }
1235
            }
1236
        }
1237
    }
1238
    return 0;
1239
}
1240

    
1241
/**
1242
 * Decode Spectral Band Replication extension data; reference: table 4.55.
1243
 *
1244
 * @param   crc flag indicating the presence of CRC checksum
1245
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1246
 *
1247
 * @return  Returns number of bytes consumed from the TYPE_FIL element.
1248
 */
1249
static int decode_sbr_extension(AACContext * ac, GetBitContext * gb, int crc, int cnt) {
1250
    // TODO : sbr_extension implementation
1251
    ff_log_missing_feature(ac->avccontext, "SBR", 0);
1252
    skip_bits_long(gb, 8*cnt - 4); // -4 due to reading extension type
1253
    return cnt;
1254
}
1255

    
1256
/**
1257
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1258
 *
1259
 * @return  Returns number of bytes consumed.
1260
 */
1261
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc, GetBitContext * gb) {
1262
    int i;
1263
    int num_excl_chan = 0;
1264

    
1265
    do {
1266
        for (i = 0; i < 7; i++)
1267
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1268
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1269

    
1270
    return num_excl_chan / 7;
1271
}
1272

    
1273
/**
1274
 * Decode dynamic range information; reference: table 4.52.
1275
 *
1276
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1277
 *
1278
 * @return  Returns number of bytes consumed.
1279
 */
1280
static int decode_dynamic_range(DynamicRangeControl *che_drc, GetBitContext * gb, int cnt) {
1281
    int n = 1;
1282
    int drc_num_bands = 1;
1283
    int i;
1284

    
1285
    /* pce_tag_present? */
1286
    if(get_bits1(gb)) {
1287
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1288
        skip_bits(gb, 4); // tag_reserved_bits
1289
        n++;
1290
    }
1291

    
1292
    /* excluded_chns_present? */
1293
    if(get_bits1(gb)) {
1294
        n += decode_drc_channel_exclusions(che_drc, gb);
1295
    }
1296

    
1297
    /* drc_bands_present? */
1298
    if (get_bits1(gb)) {
1299
        che_drc->band_incr            = get_bits(gb, 4);
1300
        che_drc->interpolation_scheme = get_bits(gb, 4);
1301
        n++;
1302
        drc_num_bands += che_drc->band_incr;
1303
        for (i = 0; i < drc_num_bands; i++) {
1304
            che_drc->band_top[i] = get_bits(gb, 8);
1305
            n++;
1306
        }
1307
    }
1308

    
1309
    /* prog_ref_level_present? */
1310
    if (get_bits1(gb)) {
1311
        che_drc->prog_ref_level = get_bits(gb, 7);
1312
        skip_bits1(gb); // prog_ref_level_reserved_bits
1313
        n++;
1314
    }
1315

    
1316
    for (i = 0; i < drc_num_bands; i++) {
1317
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1318
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1319
        n++;
1320
    }
1321

    
1322
    return n;
1323
}
1324

    
1325
/**
1326
 * Decode extension data (incomplete); reference: table 4.51.
1327
 *
1328
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1329
 *
1330
 * @return Returns number of bytes consumed
1331
 */
1332
static int decode_extension_payload(AACContext * ac, GetBitContext * gb, int cnt) {
1333
    int crc_flag = 0;
1334
    int res = cnt;
1335
    switch (get_bits(gb, 4)) { // extension type
1336
        case EXT_SBR_DATA_CRC:
1337
            crc_flag++;
1338
        case EXT_SBR_DATA:
1339
            res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1340
            break;
1341
        case EXT_DYNAMIC_RANGE:
1342
            res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1343
            break;
1344
        case EXT_FILL:
1345
        case EXT_FILL_DATA:
1346
        case EXT_DATA_ELEMENT:
1347
        default:
1348
            skip_bits_long(gb, 8*cnt - 4);
1349
            break;
1350
    };
1351
    return res;
1352
}
1353

    
1354
/**
1355
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1356
 *
1357
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1358
 * @param   coef    spectral coefficients
1359
 */
1360
static void apply_tns(float coef[1024], TemporalNoiseShaping * tns, IndividualChannelStream * ics, int decode) {
1361
    const int mmm = FFMIN(ics->tns_max_bands,  ics->max_sfb);
1362
    int w, filt, m, i;
1363
    int bottom, top, order, start, end, size, inc;
1364
    float lpc[TNS_MAX_ORDER];
1365

    
1366
    for (w = 0; w < ics->num_windows; w++) {
1367
        bottom = ics->num_swb;
1368
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1369
            top    = bottom;
1370
            bottom = FFMAX(0, top - tns->length[w][filt]);
1371
            order  = tns->order[w][filt];
1372
            if (order == 0)
1373
                continue;
1374

    
1375
            // tns_decode_coef
1376
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1377

    
1378
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1379
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1380
            if ((size = end - start) <= 0)
1381
                continue;
1382
            if (tns->direction[w][filt]) {
1383
                inc = -1; start = end - 1;
1384
            } else {
1385
                inc = 1;
1386
            }
1387
            start += w * 128;
1388

    
1389
            // ar filter
1390
            for (m = 0; m < size; m++, start += inc)
1391
                for (i = 1; i <= FFMIN(m, order); i++)
1392
                    coef[start] -= coef[start - i*inc] * lpc[i-1];
1393
        }
1394
    }
1395
}
1396

    
1397
/**
1398
 * Conduct IMDCT and windowing.
1399
 */
1400
static void imdct_and_windowing(AACContext * ac, SingleChannelElement * sce) {
1401
    IndividualChannelStream * ics = &sce->ics;
1402
    float * in = sce->coeffs;
1403
    float * out = sce->ret;
1404
    float * saved = sce->saved;
1405
    const float * swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1406
    const float * lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1407
    const float * swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1408
    float * buf = ac->buf_mdct;
1409
    float * temp = ac->temp;
1410
    int i;
1411

    
1412
    // imdct
1413
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1414
        if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1415
            av_log(ac->avccontext, AV_LOG_WARNING,
1416
                   "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1417
                   "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1418
        for (i = 0; i < 1024; i += 128)
1419
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1420
    } else
1421
        ff_imdct_half(&ac->mdct, buf, in);
1422

    
1423
    /* window overlapping
1424
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1425
     * and long to short transitions are considered to be short to short
1426
     * transitions. This leaves just two cases (long to long and short to short)
1427
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1428
     */
1429
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1430
        (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1431
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, ac->add_bias, 512);
1432
    } else {
1433
        for (i = 0; i < 448; i++)
1434
            out[i] = saved[i] + ac->add_bias;
1435

    
1436
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1437
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, ac->add_bias, 64);
1438
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      ac->add_bias, 64);
1439
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      ac->add_bias, 64);
1440
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      ac->add_bias, 64);
1441
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      ac->add_bias, 64);
1442
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1443
        } else {
1444
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, ac->add_bias, 64);
1445
            for (i = 576; i < 1024; i++)
1446
                out[i] = buf[i-512] + ac->add_bias;
1447
        }
1448
    }
1449

    
1450
    // buffer update
1451
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1452
        for (i = 0; i < 64; i++)
1453
            saved[i] = temp[64 + i] - ac->add_bias;
1454
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1455
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1456
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1457
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1458
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1459
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1460
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1461
    } else { // LONG_STOP or ONLY_LONG
1462
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1463
    }
1464
}
1465

    
1466
/**
1467
 * Apply dependent channel coupling (applied before IMDCT).
1468
 *
1469
 * @param   index   index into coupling gain array
1470
 */
1471
static void apply_dependent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1472
    IndividualChannelStream * ics = &cce->ch[0].ics;
1473
    const uint16_t * offsets = ics->swb_offset;
1474
    float * dest = target->coeffs;
1475
    const float * src = cce->ch[0].coeffs;
1476
    int g, i, group, k, idx = 0;
1477
    if(ac->m4ac.object_type == AOT_AAC_LTP) {
1478
        av_log(ac->avccontext, AV_LOG_ERROR,
1479
               "Dependent coupling is not supported together with LTP\n");
1480
        return;
1481
    }
1482
    for (g = 0; g < ics->num_window_groups; g++) {
1483
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1484
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1485
                const float gain = cce->coup.gain[index][idx];
1486
                for (group = 0; group < ics->group_len[g]; group++) {
1487
                    for (k = offsets[i]; k < offsets[i+1]; k++) {
1488
                        // XXX dsputil-ize
1489
                        dest[group*128+k] += gain * src[group*128+k];
1490
                    }
1491
                }
1492
            }
1493
        }
1494
        dest += ics->group_len[g]*128;
1495
        src  += ics->group_len[g]*128;
1496
    }
1497
}
1498

    
1499
/**
1500
 * Apply independent channel coupling (applied after IMDCT).
1501
 *
1502
 * @param   index   index into coupling gain array
1503
 */
1504
static void apply_independent_coupling(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index) {
1505
    int i;
1506
    const float gain = cce->coup.gain[index][0];
1507
    const float bias = ac->add_bias;
1508
    const float* src = cce->ch[0].ret;
1509
    float* dest = target->ret;
1510

    
1511
    for (i = 0; i < 1024; i++)
1512
        dest[i] += gain * (src[i] - bias);
1513
}
1514

    
1515
/**
1516
 * channel coupling transformation interface
1517
 *
1518
 * @param   index   index into coupling gain array
1519
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1520
 */
1521
static void apply_channel_coupling(AACContext * ac, ChannelElement * cc,
1522
        enum RawDataBlockType type, int elem_id, enum CouplingPoint coupling_point,
1523
        void (*apply_coupling_method)(AACContext * ac, SingleChannelElement * target, ChannelElement * cce, int index))
1524
{
1525
    int i, c;
1526

    
1527
    for (i = 0; i < MAX_ELEM_ID; i++) {
1528
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1529
        int index = 0;
1530

    
1531
        if (cce && cce->coup.coupling_point == coupling_point) {
1532
            ChannelCoupling * coup = &cce->coup;
1533

    
1534
            for (c = 0; c <= coup->num_coupled; c++) {
1535
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1536
                    if (coup->ch_select[c] != 1) {
1537
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1538
                        if (coup->ch_select[c] != 0)
1539
                            index++;
1540
                    }
1541
                    if (coup->ch_select[c] != 2)
1542
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1543
                } else
1544
                    index += 1 + (coup->ch_select[c] == 3);
1545
            }
1546
        }
1547
    }
1548
}
1549

    
1550
/**
1551
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1552
 */
1553
static void spectral_to_sample(AACContext * ac) {
1554
    int i, type;
1555
    for(type = 3; type >= 0; type--) {
1556
        for (i = 0; i < MAX_ELEM_ID; i++) {
1557
            ChannelElement *che = ac->che[type][i];
1558
            if(che) {
1559
                if(type <= TYPE_CPE)
1560
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1561
                if(che->ch[0].tns.present)
1562
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1563
                if(che->ch[1].tns.present)
1564
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1565
                if(type <= TYPE_CPE)
1566
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1567
                if(type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1568
                    imdct_and_windowing(ac, &che->ch[0]);
1569
                if(type == TYPE_CPE)
1570
                    imdct_and_windowing(ac, &che->ch[1]);
1571
                if(type <= TYPE_CCE)
1572
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1573
            }
1574
        }
1575
    }
1576
}
1577

    
1578
static int parse_adts_frame_header(AACContext * ac, GetBitContext * gb) {
1579

    
1580
    int size;
1581
    AACADTSHeaderInfo hdr_info;
1582

    
1583
    size = ff_aac_parse_header(gb, &hdr_info);
1584
    if (size > 0) {
1585
        if (hdr_info.chan_config)
1586
            ac->m4ac.chan_config = hdr_info.chan_config;
1587
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1588
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1589
        ac->m4ac.object_type     = hdr_info.object_type;
1590
        if (hdr_info.num_aac_frames == 1) {
1591
            if (!hdr_info.crc_absent)
1592
                skip_bits(gb, 16);
1593
        } else {
1594
            ff_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1595
            return -1;
1596
        }
1597
    }
1598
    return size;
1599
}
1600

    
1601
static int aac_decode_frame(AVCodecContext * avccontext, void * data, int * data_size, AVPacket *avpkt) {
1602
    const uint8_t *buf = avpkt->data;
1603
    int buf_size = avpkt->size;
1604
    AACContext * ac = avccontext->priv_data;
1605
    ChannelElement * che = NULL;
1606
    GetBitContext gb;
1607
    enum RawDataBlockType elem_type;
1608
    int err, elem_id, data_size_tmp;
1609

    
1610
    init_get_bits(&gb, buf, buf_size*8);
1611

    
1612
    if (show_bits(&gb, 12) == 0xfff) {
1613
        if (parse_adts_frame_header(ac, &gb) < 0) {
1614
            av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1615
            return -1;
1616
        }
1617
        if (ac->m4ac.sampling_index > 12) {
1618
            av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1619
            return -1;
1620
        }
1621
    }
1622

    
1623
    // parse
1624
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1625
        elem_id = get_bits(&gb, 4);
1626

    
1627
        if(elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1628
            av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1629
            return -1;
1630
        }
1631

    
1632
        switch (elem_type) {
1633

    
1634
        case TYPE_SCE:
1635
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1636
            break;
1637

    
1638
        case TYPE_CPE:
1639
            err = decode_cpe(ac, &gb, che);
1640
            break;
1641

    
1642
        case TYPE_CCE:
1643
            err = decode_cce(ac, &gb, che);
1644
            break;
1645

    
1646
        case TYPE_LFE:
1647
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1648
            break;
1649

    
1650
        case TYPE_DSE:
1651
            skip_data_stream_element(&gb);
1652
            err = 0;
1653
            break;
1654

    
1655
        case TYPE_PCE:
1656
        {
1657
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1658
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1659
            if((err = decode_pce(ac, new_che_pos, &gb)))
1660
                break;
1661
            err = output_configure(ac, ac->che_pos, new_che_pos, 0);
1662
            break;
1663
        }
1664

    
1665
        case TYPE_FIL:
1666
            if (elem_id == 15)
1667
                elem_id += get_bits(&gb, 8) - 1;
1668
            while (elem_id > 0)
1669
                elem_id -= decode_extension_payload(ac, &gb, elem_id);
1670
            err = 0; /* FIXME */
1671
            break;
1672

    
1673
        default:
1674
            err = -1; /* should not happen, but keeps compiler happy */
1675
            break;
1676
        }
1677

    
1678
        if(err)
1679
            return err;
1680
    }
1681

    
1682
    spectral_to_sample(ac);
1683

    
1684
    if (!ac->is_saved) {
1685
        ac->is_saved = 1;
1686
        *data_size = 0;
1687
        return buf_size;
1688
    }
1689

    
1690
    data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1691
    if(*data_size < data_size_tmp) {
1692
        av_log(avccontext, AV_LOG_ERROR,
1693
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1694
               *data_size, data_size_tmp);
1695
        return -1;
1696
    }
1697
    *data_size = data_size_tmp;
1698

    
1699
    ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1700

    
1701
    return buf_size;
1702
}
1703

    
1704
static av_cold int aac_decode_close(AVCodecContext * avccontext) {
1705
    AACContext * ac = avccontext->priv_data;
1706
    int i, type;
1707

    
1708
    for (i = 0; i < MAX_ELEM_ID; i++) {
1709
        for(type = 0; type < 4; type++)
1710
            av_freep(&ac->che[type][i]);
1711
    }
1712

    
1713
    ff_mdct_end(&ac->mdct);
1714
    ff_mdct_end(&ac->mdct_small);
1715
    return 0 ;
1716
}
1717

    
1718
AVCodec aac_decoder = {
1719
    "aac",
1720
    CODEC_TYPE_AUDIO,
1721
    CODEC_ID_AAC,
1722
    sizeof(AACContext),
1723
    aac_decode_init,
1724
    NULL,
1725
    aac_decode_close,
1726
    aac_decode_frame,
1727
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1728
    .sample_fmts = (enum SampleFormat[]){SAMPLE_FMT_S16,SAMPLE_FMT_NONE},
1729
};