Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aacdec.c @ ebb7f7de

History | View | Annotate | Download (82.6 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
 * AAC LATM decoder
7
 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8
 * Copyright (c) 2010      Janne Grunau <janne-ffmpeg@jannau.net>
9
 *
10
 * This file is part of FFmpeg.
11
 *
12
 * FFmpeg is free software; you can redistribute it and/or
13
 * modify it under the terms of the GNU Lesser General Public
14
 * License as published by the Free Software Foundation; either
15
 * version 2.1 of the License, or (at your option) any later version.
16
 *
17
 * FFmpeg is distributed in the hope that it will be useful,
18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
20
 * Lesser General Public License for more details.
21
 *
22
 * You should have received a copy of the GNU Lesser General Public
23
 * License along with FFmpeg; if not, write to the Free Software
24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
 */
26

    
27
/**
28
 * @file
29
 * AAC decoder
30
 * @author Oded Shimon  ( ods15 ods15 dyndns org )
31
 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
32
 */
33

    
34
/*
35
 * supported tools
36
 *
37
 * Support?             Name
38
 * N (code in SoC repo) gain control
39
 * Y                    block switching
40
 * Y                    window shapes - standard
41
 * N                    window shapes - Low Delay
42
 * Y                    filterbank - standard
43
 * N (code in SoC repo) filterbank - Scalable Sample Rate
44
 * Y                    Temporal Noise Shaping
45
 * N (code in SoC repo) Long Term Prediction
46
 * Y                    intensity stereo
47
 * Y                    channel coupling
48
 * Y                    frequency domain prediction
49
 * Y                    Perceptual Noise Substitution
50
 * Y                    Mid/Side stereo
51
 * N                    Scalable Inverse AAC Quantization
52
 * N                    Frequency Selective Switch
53
 * N                    upsampling filter
54
 * Y                    quantization & coding - AAC
55
 * N                    quantization & coding - TwinVQ
56
 * N                    quantization & coding - BSAC
57
 * N                    AAC Error Resilience tools
58
 * N                    Error Resilience payload syntax
59
 * N                    Error Protection tool
60
 * N                    CELP
61
 * N                    Silence Compression
62
 * N                    HVXC
63
 * N                    HVXC 4kbits/s VR
64
 * N                    Structured Audio tools
65
 * N                    Structured Audio Sample Bank Format
66
 * N                    MIDI
67
 * N                    Harmonic and Individual Lines plus Noise
68
 * N                    Text-To-Speech Interface
69
 * Y                    Spectral Band Replication
70
 * Y (not in this code) Layer-1
71
 * Y (not in this code) Layer-2
72
 * Y (not in this code) Layer-3
73
 * N                    SinuSoidal Coding (Transient, Sinusoid, Noise)
74
 * Y                    Parametric Stereo
75
 * N                    Direct Stream Transfer
76
 *
77
 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78
 *       - HE AAC v2 comprises LC AAC with Spectral Band Replication and
79
           Parametric Stereo.
80
 */
81

    
82

    
83
#include "avcodec.h"
84
#include "internal.h"
85
#include "get_bits.h"
86
#include "dsputil.h"
87
#include "fft.h"
88
#include "lpc.h"
89

    
90
#include "aac.h"
91
#include "aactab.h"
92
#include "aacdectab.h"
93
#include "cbrt_tablegen.h"
94
#include "sbr.h"
95
#include "aacsbr.h"
96
#include "mpeg4audio.h"
97
#include "aacadtsdec.h"
98

    
99
#include <assert.h>
100
#include <errno.h>
101
#include <math.h>
102
#include <string.h>
103

    
104
#if ARCH_ARM
105
#   include "arm/aac.h"
106
#endif
107

    
108
union float754 {
109
    float f;
110
    uint32_t i;
111
};
112

    
113
static VLC vlc_scalefactors;
114
static VLC vlc_spectral[11];
115

    
116
static const char overread_err[] = "Input buffer exhausted before END element found\n";
117

    
118
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
119
{
120
    // For PCE based channel configurations map the channels solely based on tags.
121
    if (!ac->m4ac.chan_config) {
122
        return ac->tag_che_map[type][elem_id];
123
    }
124
    // For indexed channel configurations map the channels solely based on position.
125
    switch (ac->m4ac.chan_config) {
126
    case 7:
127
        if (ac->tags_mapped == 3 && type == TYPE_CPE) {
128
            ac->tags_mapped++;
129
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
130
        }
131
    case 6:
132
        /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
133
           instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
134
           encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
135
        if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
136
            ac->tags_mapped++;
137
            return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
138
        }
139
    case 5:
140
        if (ac->tags_mapped == 2 && type == TYPE_CPE) {
141
            ac->tags_mapped++;
142
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
143
        }
144
    case 4:
145
        if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
146
            ac->tags_mapped++;
147
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
148
        }
149
    case 3:
150
    case 2:
151
        if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
152
            ac->tags_mapped++;
153
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
154
        } else if (ac->m4ac.chan_config == 2) {
155
            return NULL;
156
        }
157
    case 1:
158
        if (!ac->tags_mapped && type == TYPE_SCE) {
159
            ac->tags_mapped++;
160
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
161
        }
162
    default:
163
        return NULL;
164
    }
165
}
166

    
167
/**
168
 * Check for the channel element in the current channel position configuration.
169
 * If it exists, make sure the appropriate element is allocated and map the
170
 * channel order to match the internal FFmpeg channel layout.
171
 *
172
 * @param   che_pos current channel position configuration
173
 * @param   type channel element type
174
 * @param   id channel element id
175
 * @param   channels count of the number of channels in the configuration
176
 *
177
 * @return  Returns error status. 0 - OK, !0 - error
178
 */
179
static av_cold int che_configure(AACContext *ac,
180
                         enum ChannelPosition che_pos[4][MAX_ELEM_ID],
181
                         int type, int id,
182
                         int *channels)
183
{
184
    if (che_pos[type][id]) {
185
        if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
186
            return AVERROR(ENOMEM);
187
        ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
188
        if (type != TYPE_CCE) {
189
            ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
190
            if (type == TYPE_CPE ||
191
                (type == TYPE_SCE && ac->m4ac.ps == 1)) {
192
                ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
193
            }
194
        }
195
    } else {
196
        if (ac->che[type][id])
197
            ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
198
        av_freep(&ac->che[type][id]);
199
    }
200
    return 0;
201
}
202

    
203
/**
204
 * Configure output channel order based on the current program configuration element.
205
 *
206
 * @param   che_pos current channel position configuration
207
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
208
 *
209
 * @return  Returns error status. 0 - OK, !0 - error
210
 */
211
static av_cold int output_configure(AACContext *ac,
212
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
213
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
214
                            int channel_config, enum OCStatus oc_type)
215
{
216
    AVCodecContext *avctx = ac->avctx;
217
    int i, type, channels = 0, ret;
218

    
219
    if (new_che_pos != che_pos)
220
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
221

    
222
    if (channel_config) {
223
        for (i = 0; i < tags_per_config[channel_config]; i++) {
224
            if ((ret = che_configure(ac, che_pos,
225
                                     aac_channel_layout_map[channel_config - 1][i][0],
226
                                     aac_channel_layout_map[channel_config - 1][i][1],
227
                                     &channels)))
228
                return ret;
229
        }
230

    
231
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
232

    
233
        avctx->channel_layout = aac_channel_layout[channel_config - 1];
234
    } else {
235
        /* Allocate or free elements depending on if they are in the
236
         * current program configuration.
237
         *
238
         * Set up default 1:1 output mapping.
239
         *
240
         * For a 5.1 stream the output order will be:
241
         *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
242
         */
243

    
244
        for (i = 0; i < MAX_ELEM_ID; i++) {
245
            for (type = 0; type < 4; type++) {
246
                if ((ret = che_configure(ac, che_pos, type, i, &channels)))
247
                    return ret;
248
            }
249
        }
250

    
251
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
252

    
253
        avctx->channel_layout = 0;
254
    }
255

    
256
    avctx->channels = channels;
257

    
258
    ac->output_configured = oc_type;
259

    
260
    return 0;
261
}
262

    
263
/**
264
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
265
 *
266
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
267
 * @param sce_map mono (Single Channel Element) map
268
 * @param type speaker type/position for these channels
269
 */
270
static void decode_channel_map(enum ChannelPosition *cpe_map,
271
                               enum ChannelPosition *sce_map,
272
                               enum ChannelPosition type,
273
                               GetBitContext *gb, int n)
274
{
275
    while (n--) {
276
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
277
        map[get_bits(gb, 4)] = type;
278
    }
279
}
280

    
281
/**
282
 * Decode program configuration element; reference: table 4.2.
283
 *
284
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
285
 *
286
 * @return  Returns error status. 0 - OK, !0 - error
287
 */
288
static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
289
                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
290
                      GetBitContext *gb)
291
{
292
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
293
    int comment_len;
294

    
295
    skip_bits(gb, 2);  // object_type
296

    
297
    sampling_index = get_bits(gb, 4);
298
    if (m4ac->sampling_index != sampling_index)
299
        av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
300

    
301
    num_front       = get_bits(gb, 4);
302
    num_side        = get_bits(gb, 4);
303
    num_back        = get_bits(gb, 4);
304
    num_lfe         = get_bits(gb, 2);
305
    num_assoc_data  = get_bits(gb, 3);
306
    num_cc          = get_bits(gb, 4);
307

    
308
    if (get_bits1(gb))
309
        skip_bits(gb, 4); // mono_mixdown_tag
310
    if (get_bits1(gb))
311
        skip_bits(gb, 4); // stereo_mixdown_tag
312

    
313
    if (get_bits1(gb))
314
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
315

    
316
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
317
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
318
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
319
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
320

    
321
    skip_bits_long(gb, 4 * num_assoc_data);
322

    
323
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
324

    
325
    align_get_bits(gb);
326

    
327
    /* comment field, first byte is length */
328
    comment_len = get_bits(gb, 8) * 8;
329
    if (get_bits_left(gb) < comment_len) {
330
        av_log(avctx, AV_LOG_ERROR, overread_err);
331
        return -1;
332
    }
333
    skip_bits_long(gb, comment_len);
334
    return 0;
335
}
336

    
337
/**
338
 * Set up channel positions based on a default channel configuration
339
 * as specified in table 1.17.
340
 *
341
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
342
 *
343
 * @return  Returns error status. 0 - OK, !0 - error
344
 */
345
static av_cold int set_default_channel_config(AVCodecContext *avctx,
346
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
347
                                      int channel_config)
348
{
349
    if (channel_config < 1 || channel_config > 7) {
350
        av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
351
               channel_config);
352
        return -1;
353
    }
354

    
355
    /* default channel configurations:
356
     *
357
     * 1ch : front center (mono)
358
     * 2ch : L + R (stereo)
359
     * 3ch : front center + L + R
360
     * 4ch : front center + L + R + back center
361
     * 5ch : front center + L + R + back stereo
362
     * 6ch : front center + L + R + back stereo + LFE
363
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
364
     */
365

    
366
    if (channel_config != 2)
367
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
368
    if (channel_config > 1)
369
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
370
    if (channel_config == 4)
371
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
372
    if (channel_config > 4)
373
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
374
        = AAC_CHANNEL_BACK;  // back stereo
375
    if (channel_config > 5)
376
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
377
    if (channel_config == 7)
378
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
379

    
380
    return 0;
381
}
382

    
383
/**
384
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
385
 *
386
 * @param   ac          pointer to AACContext, may be null
387
 * @param   avctx       pointer to AVCCodecContext, used for logging
388
 *
389
 * @return  Returns error status. 0 - OK, !0 - error
390
 */
391
static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
392
                                     GetBitContext *gb,
393
                                     MPEG4AudioConfig *m4ac,
394
                                     int channel_config)
395
{
396
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
397
    int extension_flag, ret;
398

    
399
    if (get_bits1(gb)) { // frameLengthFlag
400
        av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
401
        return -1;
402
    }
403

    
404
    if (get_bits1(gb))       // dependsOnCoreCoder
405
        skip_bits(gb, 14);   // coreCoderDelay
406
    extension_flag = get_bits1(gb);
407

    
408
    if (m4ac->object_type == AOT_AAC_SCALABLE ||
409
        m4ac->object_type == AOT_ER_AAC_SCALABLE)
410
        skip_bits(gb, 3);     // layerNr
411

    
412
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
413
    if (channel_config == 0) {
414
        skip_bits(gb, 4);  // element_instance_tag
415
        if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
416
            return ret;
417
    } else {
418
        if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
419
            return ret;
420
    }
421
    if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
422
        return ret;
423

    
424
    if (extension_flag) {
425
        switch (m4ac->object_type) {
426
        case AOT_ER_BSAC:
427
            skip_bits(gb, 5);    // numOfSubFrame
428
            skip_bits(gb, 11);   // layer_length
429
            break;
430
        case AOT_ER_AAC_LC:
431
        case AOT_ER_AAC_LTP:
432
        case AOT_ER_AAC_SCALABLE:
433
        case AOT_ER_AAC_LD:
434
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
435
                                    * aacScalefactorDataResilienceFlag
436
                                    * aacSpectralDataResilienceFlag
437
                                    */
438
            break;
439
        }
440
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
441
    }
442
    return 0;
443
}
444

    
445
/**
446
 * Decode audio specific configuration; reference: table 1.13.
447
 *
448
 * @param   ac          pointer to AACContext, may be null
449
 * @param   avctx       pointer to AVCCodecContext, used for logging
450
 * @param   m4ac        pointer to MPEG4AudioConfig, used for parsing
451
 * @param   data        pointer to AVCodecContext extradata
452
 * @param   data_size   size of AVCCodecContext extradata
453
 *
454
 * @return  Returns error status or number of consumed bits. <0 - error
455
 */
456
static int decode_audio_specific_config(AACContext *ac,
457
                                        AVCodecContext *avctx,
458
                                        MPEG4AudioConfig *m4ac,
459
                                        const uint8_t *data, int data_size)
460
{
461
    GetBitContext gb;
462
    int i;
463

    
464
    init_get_bits(&gb, data, data_size * 8);
465

    
466
    if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
467
        return -1;
468
    if (m4ac->sampling_index > 12) {
469
        av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
470
        return -1;
471
    }
472
    if (m4ac->sbr == 1 && m4ac->ps == -1)
473
        m4ac->ps = 1;
474

    
475
    skip_bits_long(&gb, i);
476

    
477
    switch (m4ac->object_type) {
478
    case AOT_AAC_MAIN:
479
    case AOT_AAC_LC:
480
        if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
481
            return -1;
482
        break;
483
    default:
484
        av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
485
               m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
486
        return -1;
487
    }
488

    
489
    return get_bits_count(&gb);
490
}
491

    
492
/**
493
 * linear congruential pseudorandom number generator
494
 *
495
 * @param   previous_val    pointer to the current state of the generator
496
 *
497
 * @return  Returns a 32-bit pseudorandom integer
498
 */
499
static av_always_inline int lcg_random(int previous_val)
500
{
501
    return previous_val * 1664525 + 1013904223;
502
}
503

    
504
static av_always_inline void reset_predict_state(PredictorState *ps)
505
{
506
    ps->r0   = 0.0f;
507
    ps->r1   = 0.0f;
508
    ps->cor0 = 0.0f;
509
    ps->cor1 = 0.0f;
510
    ps->var0 = 1.0f;
511
    ps->var1 = 1.0f;
512
}
513

    
514
static void reset_all_predictors(PredictorState *ps)
515
{
516
    int i;
517
    for (i = 0; i < MAX_PREDICTORS; i++)
518
        reset_predict_state(&ps[i]);
519
}
520

    
521
static void reset_predictor_group(PredictorState *ps, int group_num)
522
{
523
    int i;
524
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
525
        reset_predict_state(&ps[i]);
526
}
527

    
528
#define AAC_INIT_VLC_STATIC(num, size) \
529
    INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
530
         ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
531
        ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
532
        size);
533

    
534
static av_cold int aac_decode_init(AVCodecContext *avctx)
535
{
536
    AACContext *ac = avctx->priv_data;
537

    
538
    ac->avctx = avctx;
539
    ac->m4ac.sample_rate = avctx->sample_rate;
540

    
541
    if (avctx->extradata_size > 0) {
542
        if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
543
                                         avctx->extradata,
544
                                         avctx->extradata_size) < 0)
545
            return -1;
546
    }
547

    
548
    avctx->sample_fmt = SAMPLE_FMT_S16;
549

    
550
    AAC_INIT_VLC_STATIC( 0, 304);
551
    AAC_INIT_VLC_STATIC( 1, 270);
552
    AAC_INIT_VLC_STATIC( 2, 550);
553
    AAC_INIT_VLC_STATIC( 3, 300);
554
    AAC_INIT_VLC_STATIC( 4, 328);
555
    AAC_INIT_VLC_STATIC( 5, 294);
556
    AAC_INIT_VLC_STATIC( 6, 306);
557
    AAC_INIT_VLC_STATIC( 7, 268);
558
    AAC_INIT_VLC_STATIC( 8, 510);
559
    AAC_INIT_VLC_STATIC( 9, 366);
560
    AAC_INIT_VLC_STATIC(10, 462);
561

    
562
    ff_aac_sbr_init();
563

    
564
    dsputil_init(&ac->dsp, avctx);
565

    
566
    ac->random_state = 0x1f2e3d4c;
567

    
568
    // -1024 - Compensate wrong IMDCT method.
569
    // 32768 - Required to scale values to the correct range for the bias method
570
    //         for float to int16 conversion.
571

    
572
    if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
573
        ac->add_bias  = 385.0f;
574
        ac->sf_scale  = 1. / (-1024. * 32768.);
575
        ac->sf_offset = 0;
576
    } else {
577
        ac->add_bias  = 0.0f;
578
        ac->sf_scale  = 1. / -1024.;
579
        ac->sf_offset = 60;
580
    }
581

    
582
    ff_aac_tableinit();
583

    
584
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
585
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
586
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
587
                    352);
588

    
589
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
590
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
591
    // window initialization
592
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
593
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
594
    ff_init_ff_sine_windows(10);
595
    ff_init_ff_sine_windows( 7);
596

    
597
    cbrt_tableinit();
598

    
599
    return 0;
600
}
601

    
602
/**
603
 * Skip data_stream_element; reference: table 4.10.
604
 */
605
static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
606
{
607
    int byte_align = get_bits1(gb);
608
    int count = get_bits(gb, 8);
609
    if (count == 255)
610
        count += get_bits(gb, 8);
611
    if (byte_align)
612
        align_get_bits(gb);
613

    
614
    if (get_bits_left(gb) < 8 * count) {
615
        av_log(ac->avctx, AV_LOG_ERROR, overread_err);
616
        return -1;
617
    }
618
    skip_bits_long(gb, 8 * count);
619
    return 0;
620
}
621

    
622
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
623
                             GetBitContext *gb)
624
{
625
    int sfb;
626
    if (get_bits1(gb)) {
627
        ics->predictor_reset_group = get_bits(gb, 5);
628
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
629
            av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
630
            return -1;
631
        }
632
    }
633
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
634
        ics->prediction_used[sfb] = get_bits1(gb);
635
    }
636
    return 0;
637
}
638

    
639
/**
640
 * Decode Individual Channel Stream info; reference: table 4.6.
641
 *
642
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
643
 */
644
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
645
                           GetBitContext *gb, int common_window)
646
{
647
    if (get_bits1(gb)) {
648
        av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
649
        memset(ics, 0, sizeof(IndividualChannelStream));
650
        return -1;
651
    }
652
    ics->window_sequence[1] = ics->window_sequence[0];
653
    ics->window_sequence[0] = get_bits(gb, 2);
654
    ics->use_kb_window[1]   = ics->use_kb_window[0];
655
    ics->use_kb_window[0]   = get_bits1(gb);
656
    ics->num_window_groups  = 1;
657
    ics->group_len[0]       = 1;
658
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
659
        int i;
660
        ics->max_sfb = get_bits(gb, 4);
661
        for (i = 0; i < 7; i++) {
662
            if (get_bits1(gb)) {
663
                ics->group_len[ics->num_window_groups - 1]++;
664
            } else {
665
                ics->num_window_groups++;
666
                ics->group_len[ics->num_window_groups - 1] = 1;
667
            }
668
        }
669
        ics->num_windows       = 8;
670
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
671
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
672
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
673
        ics->predictor_present = 0;
674
    } else {
675
        ics->max_sfb               = get_bits(gb, 6);
676
        ics->num_windows           = 1;
677
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
678
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
679
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
680
        ics->predictor_present     = get_bits1(gb);
681
        ics->predictor_reset_group = 0;
682
        if (ics->predictor_present) {
683
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
684
                if (decode_prediction(ac, ics, gb)) {
685
                    memset(ics, 0, sizeof(IndividualChannelStream));
686
                    return -1;
687
                }
688
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
689
                av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
690
                memset(ics, 0, sizeof(IndividualChannelStream));
691
                return -1;
692
            } else {
693
                av_log_missing_feature(ac->avctx, "Predictor bit set but LTP is", 1);
694
                memset(ics, 0, sizeof(IndividualChannelStream));
695
                return -1;
696
            }
697
        }
698
    }
699

    
700
    if (ics->max_sfb > ics->num_swb) {
701
        av_log(ac->avctx, AV_LOG_ERROR,
702
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
703
               ics->max_sfb, ics->num_swb);
704
        memset(ics, 0, sizeof(IndividualChannelStream));
705
        return -1;
706
    }
707

    
708
    return 0;
709
}
710

    
711
/**
712
 * Decode band types (section_data payload); reference: table 4.46.
713
 *
714
 * @param   band_type           array of the used band type
715
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
716
 *
717
 * @return  Returns error status. 0 - OK, !0 - error
718
 */
719
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
720
                             int band_type_run_end[120], GetBitContext *gb,
721
                             IndividualChannelStream *ics)
722
{
723
    int g, idx = 0;
724
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
725
    for (g = 0; g < ics->num_window_groups; g++) {
726
        int k = 0;
727
        while (k < ics->max_sfb) {
728
            uint8_t sect_end = k;
729
            int sect_len_incr;
730
            int sect_band_type = get_bits(gb, 4);
731
            if (sect_band_type == 12) {
732
                av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
733
                return -1;
734
            }
735
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
736
                sect_end += sect_len_incr;
737
            sect_end += sect_len_incr;
738
            if (get_bits_left(gb) < 0) {
739
                av_log(ac->avctx, AV_LOG_ERROR, overread_err);
740
                return -1;
741
            }
742
            if (sect_end > ics->max_sfb) {
743
                av_log(ac->avctx, AV_LOG_ERROR,
744
                       "Number of bands (%d) exceeds limit (%d).\n",
745
                       sect_end, ics->max_sfb);
746
                return -1;
747
            }
748
            for (; k < sect_end; k++) {
749
                band_type        [idx]   = sect_band_type;
750
                band_type_run_end[idx++] = sect_end;
751
            }
752
        }
753
    }
754
    return 0;
755
}
756

    
757
/**
758
 * Decode scalefactors; reference: table 4.47.
759
 *
760
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
761
 * @param   band_type           array of the used band type
762
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
763
 * @param   sf                  array of scalefactors or intensity stereo positions
764
 *
765
 * @return  Returns error status. 0 - OK, !0 - error
766
 */
767
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
768
                               unsigned int global_gain,
769
                               IndividualChannelStream *ics,
770
                               enum BandType band_type[120],
771
                               int band_type_run_end[120])
772
{
773
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
774
    int g, i, idx = 0;
775
    int offset[3] = { global_gain, global_gain - 90, 100 };
776
    int noise_flag = 1;
777
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
778
    for (g = 0; g < ics->num_window_groups; g++) {
779
        for (i = 0; i < ics->max_sfb;) {
780
            int run_end = band_type_run_end[idx];
781
            if (band_type[idx] == ZERO_BT) {
782
                for (; i < run_end; i++, idx++)
783
                    sf[idx] = 0.;
784
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
785
                for (; i < run_end; i++, idx++) {
786
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
787
                    if (offset[2] > 255U) {
788
                        av_log(ac->avctx, AV_LOG_ERROR,
789
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
790
                        return -1;
791
                    }
792
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
793
                }
794
            } else if (band_type[idx] == NOISE_BT) {
795
                for (; i < run_end; i++, idx++) {
796
                    if (noise_flag-- > 0)
797
                        offset[1] += get_bits(gb, 9) - 256;
798
                    else
799
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
800
                    if (offset[1] > 255U) {
801
                        av_log(ac->avctx, AV_LOG_ERROR,
802
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
803
                        return -1;
804
                    }
805
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
806
                }
807
            } else {
808
                for (; i < run_end; i++, idx++) {
809
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
810
                    if (offset[0] > 255U) {
811
                        av_log(ac->avctx, AV_LOG_ERROR,
812
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
813
                        return -1;
814
                    }
815
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
816
                }
817
            }
818
        }
819
    }
820
    return 0;
821
}
822

    
823
/**
824
 * Decode pulse data; reference: table 4.7.
825
 */
826
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
827
                         const uint16_t *swb_offset, int num_swb)
828
{
829
    int i, pulse_swb;
830
    pulse->num_pulse = get_bits(gb, 2) + 1;
831
    pulse_swb        = get_bits(gb, 6);
832
    if (pulse_swb >= num_swb)
833
        return -1;
834
    pulse->pos[0]    = swb_offset[pulse_swb];
835
    pulse->pos[0]   += get_bits(gb, 5);
836
    if (pulse->pos[0] > 1023)
837
        return -1;
838
    pulse->amp[0]    = get_bits(gb, 4);
839
    for (i = 1; i < pulse->num_pulse; i++) {
840
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
841
        if (pulse->pos[i] > 1023)
842
            return -1;
843
        pulse->amp[i] = get_bits(gb, 4);
844
    }
845
    return 0;
846
}
847

    
848
/**
849
 * Decode Temporal Noise Shaping data; reference: table 4.48.
850
 *
851
 * @return  Returns error status. 0 - OK, !0 - error
852
 */
853
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
854
                      GetBitContext *gb, const IndividualChannelStream *ics)
855
{
856
    int w, filt, i, coef_len, coef_res, coef_compress;
857
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
858
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
859
    for (w = 0; w < ics->num_windows; w++) {
860
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
861
            coef_res = get_bits1(gb);
862

    
863
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
864
                int tmp2_idx;
865
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
866

    
867
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
868
                    av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
869
                           tns->order[w][filt], tns_max_order);
870
                    tns->order[w][filt] = 0;
871
                    return -1;
872
                }
873
                if (tns->order[w][filt]) {
874
                    tns->direction[w][filt] = get_bits1(gb);
875
                    coef_compress = get_bits1(gb);
876
                    coef_len = coef_res + 3 - coef_compress;
877
                    tmp2_idx = 2 * coef_compress + coef_res;
878

    
879
                    for (i = 0; i < tns->order[w][filt]; i++)
880
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
881
                }
882
            }
883
        }
884
    }
885
    return 0;
886
}
887

    
888
/**
889
 * Decode Mid/Side data; reference: table 4.54.
890
 *
891
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
892
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
893
 *                      [3] reserved for scalable AAC
894
 */
895
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
896
                                   int ms_present)
897
{
898
    int idx;
899
    if (ms_present == 1) {
900
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
901
            cpe->ms_mask[idx] = get_bits1(gb);
902
    } else if (ms_present == 2) {
903
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
904
    }
905
}
906

    
907
#ifndef VMUL2
908
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
909
                           const float *scale)
910
{
911
    float s = *scale;
912
    *dst++ = v[idx    & 15] * s;
913
    *dst++ = v[idx>>4 & 15] * s;
914
    return dst;
915
}
916
#endif
917

    
918
#ifndef VMUL4
919
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
920
                           const float *scale)
921
{
922
    float s = *scale;
923
    *dst++ = v[idx    & 3] * s;
924
    *dst++ = v[idx>>2 & 3] * s;
925
    *dst++ = v[idx>>4 & 3] * s;
926
    *dst++ = v[idx>>6 & 3] * s;
927
    return dst;
928
}
929
#endif
930

    
931
#ifndef VMUL2S
932
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
933
                            unsigned sign, const float *scale)
934
{
935
    union float754 s0, s1;
936

    
937
    s0.f = s1.f = *scale;
938
    s0.i ^= sign >> 1 << 31;
939
    s1.i ^= sign      << 31;
940

    
941
    *dst++ = v[idx    & 15] * s0.f;
942
    *dst++ = v[idx>>4 & 15] * s1.f;
943

    
944
    return dst;
945
}
946
#endif
947

    
948
#ifndef VMUL4S
949
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
950
                            unsigned sign, const float *scale)
951
{
952
    unsigned nz = idx >> 12;
953
    union float754 s = { .f = *scale };
954
    union float754 t;
955

    
956
    t.i = s.i ^ (sign & 1<<31);
957
    *dst++ = v[idx    & 3] * t.f;
958

    
959
    sign <<= nz & 1; nz >>= 1;
960
    t.i = s.i ^ (sign & 1<<31);
961
    *dst++ = v[idx>>2 & 3] * t.f;
962

    
963
    sign <<= nz & 1; nz >>= 1;
964
    t.i = s.i ^ (sign & 1<<31);
965
    *dst++ = v[idx>>4 & 3] * t.f;
966

    
967
    sign <<= nz & 1; nz >>= 1;
968
    t.i = s.i ^ (sign & 1<<31);
969
    *dst++ = v[idx>>6 & 3] * t.f;
970

    
971
    return dst;
972
}
973
#endif
974

    
975
/**
976
 * Decode spectral data; reference: table 4.50.
977
 * Dequantize and scale spectral data; reference: 4.6.3.3.
978
 *
979
 * @param   coef            array of dequantized, scaled spectral data
980
 * @param   sf              array of scalefactors or intensity stereo positions
981
 * @param   pulse_present   set if pulses are present
982
 * @param   pulse           pointer to pulse data struct
983
 * @param   band_type       array of the used band type
984
 *
985
 * @return  Returns error status. 0 - OK, !0 - error
986
 */
987
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
988
                                       GetBitContext *gb, const float sf[120],
989
                                       int pulse_present, const Pulse *pulse,
990
                                       const IndividualChannelStream *ics,
991
                                       enum BandType band_type[120])
992
{
993
    int i, k, g, idx = 0;
994
    const int c = 1024 / ics->num_windows;
995
    const uint16_t *offsets = ics->swb_offset;
996
    float *coef_base = coef;
997

    
998
    for (g = 0; g < ics->num_windows; g++)
999
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1000

    
1001
    for (g = 0; g < ics->num_window_groups; g++) {
1002
        unsigned g_len = ics->group_len[g];
1003

    
1004
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1005
            const unsigned cbt_m1 = band_type[idx] - 1;
1006
            float *cfo = coef + offsets[i];
1007
            int off_len = offsets[i + 1] - offsets[i];
1008
            int group;
1009

    
1010
            if (cbt_m1 >= INTENSITY_BT2 - 1) {
1011
                for (group = 0; group < g_len; group++, cfo+=128) {
1012
                    memset(cfo, 0, off_len * sizeof(float));
1013
                }
1014
            } else if (cbt_m1 == NOISE_BT - 1) {
1015
                for (group = 0; group < g_len; group++, cfo+=128) {
1016
                    float scale;
1017
                    float band_energy;
1018

    
1019
                    for (k = 0; k < off_len; k++) {
1020
                        ac->random_state  = lcg_random(ac->random_state);
1021
                        cfo[k] = ac->random_state;
1022
                    }
1023

    
1024
                    band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1025
                    scale = sf[idx] / sqrtf(band_energy);
1026
                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1027
                }
1028
            } else {
1029
                const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1030
                const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1031
                VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1032
                OPEN_READER(re, gb);
1033

    
1034
                switch (cbt_m1 >> 1) {
1035
                case 0:
1036
                    for (group = 0; group < g_len; group++, cfo+=128) {
1037
                        float *cf = cfo;
1038
                        int len = off_len;
1039

    
1040
                        do {
1041
                            int code;
1042
                            unsigned cb_idx;
1043

    
1044
                            UPDATE_CACHE(re, gb);
1045
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1046
                            cb_idx = cb_vector_idx[code];
1047
                            cf = VMUL4(cf, vq, cb_idx, sf + idx);
1048
                        } while (len -= 4);
1049
                    }
1050
                    break;
1051

    
1052
                case 1:
1053
                    for (group = 0; group < g_len; group++, cfo+=128) {
1054
                        float *cf = cfo;
1055
                        int len = off_len;
1056

    
1057
                        do {
1058
                            int code;
1059
                            unsigned nnz;
1060
                            unsigned cb_idx;
1061
                            uint32_t bits;
1062

    
1063
                            UPDATE_CACHE(re, gb);
1064
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1065
#if MIN_CACHE_BITS < 20
1066
                            UPDATE_CACHE(re, gb);
1067
#endif
1068
                            cb_idx = cb_vector_idx[code];
1069
                            nnz = cb_idx >> 8 & 15;
1070
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1071
                            LAST_SKIP_BITS(re, gb, nnz);
1072
                            cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1073
                        } while (len -= 4);
1074
                    }
1075
                    break;
1076

    
1077
                case 2:
1078
                    for (group = 0; group < g_len; group++, cfo+=128) {
1079
                        float *cf = cfo;
1080
                        int len = off_len;
1081

    
1082
                        do {
1083
                            int code;
1084
                            unsigned cb_idx;
1085

    
1086
                            UPDATE_CACHE(re, gb);
1087
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1088
                            cb_idx = cb_vector_idx[code];
1089
                            cf = VMUL2(cf, vq, cb_idx, sf + idx);
1090
                        } while (len -= 2);
1091
                    }
1092
                    break;
1093

    
1094
                case 3:
1095
                case 4:
1096
                    for (group = 0; group < g_len; group++, cfo+=128) {
1097
                        float *cf = cfo;
1098
                        int len = off_len;
1099

    
1100
                        do {
1101
                            int code;
1102
                            unsigned nnz;
1103
                            unsigned cb_idx;
1104
                            unsigned sign;
1105

    
1106
                            UPDATE_CACHE(re, gb);
1107
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1108
                            cb_idx = cb_vector_idx[code];
1109
                            nnz = cb_idx >> 8 & 15;
1110
                            sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1111
                            LAST_SKIP_BITS(re, gb, nnz);
1112
                            cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1113
                        } while (len -= 2);
1114
                    }
1115
                    break;
1116

    
1117
                default:
1118
                    for (group = 0; group < g_len; group++, cfo+=128) {
1119
                        float *cf = cfo;
1120
                        uint32_t *icf = (uint32_t *) cf;
1121
                        int len = off_len;
1122

    
1123
                        do {
1124
                            int code;
1125
                            unsigned nzt, nnz;
1126
                            unsigned cb_idx;
1127
                            uint32_t bits;
1128
                            int j;
1129

    
1130
                            UPDATE_CACHE(re, gb);
1131
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1132

    
1133
                            if (!code) {
1134
                                *icf++ = 0;
1135
                                *icf++ = 0;
1136
                                continue;
1137
                            }
1138

    
1139
                            cb_idx = cb_vector_idx[code];
1140
                            nnz = cb_idx >> 12;
1141
                            nzt = cb_idx >> 8;
1142
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1143
                            LAST_SKIP_BITS(re, gb, nnz);
1144

    
1145
                            for (j = 0; j < 2; j++) {
1146
                                if (nzt & 1<<j) {
1147
                                    uint32_t b;
1148
                                    int n;
1149
                                    /* The total length of escape_sequence must be < 22 bits according
1150
                                       to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1151
                                    UPDATE_CACHE(re, gb);
1152
                                    b = GET_CACHE(re, gb);
1153
                                    b = 31 - av_log2(~b);
1154

    
1155
                                    if (b > 8) {
1156
                                        av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1157
                                        return -1;
1158
                                    }
1159

    
1160
#if MIN_CACHE_BITS < 21
1161
                                    LAST_SKIP_BITS(re, gb, b + 1);
1162
                                    UPDATE_CACHE(re, gb);
1163
#else
1164
                                    SKIP_BITS(re, gb, b + 1);
1165
#endif
1166
                                    b += 4;
1167
                                    n = (1 << b) + SHOW_UBITS(re, gb, b);
1168
                                    LAST_SKIP_BITS(re, gb, b);
1169
                                    *icf++ = cbrt_tab[n] | (bits & 1<<31);
1170
                                    bits <<= 1;
1171
                                } else {
1172
                                    unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1173
                                    *icf++ = (bits & 1<<31) | v;
1174
                                    bits <<= !!v;
1175
                                }
1176
                                cb_idx >>= 4;
1177
                            }
1178
                        } while (len -= 2);
1179

    
1180
                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1181
                    }
1182
                }
1183

    
1184
                CLOSE_READER(re, gb);
1185
            }
1186
        }
1187
        coef += g_len << 7;
1188
    }
1189

    
1190
    if (pulse_present) {
1191
        idx = 0;
1192
        for (i = 0; i < pulse->num_pulse; i++) {
1193
            float co = coef_base[ pulse->pos[i] ];
1194
            while (offsets[idx + 1] <= pulse->pos[i])
1195
                idx++;
1196
            if (band_type[idx] != NOISE_BT && sf[idx]) {
1197
                float ico = -pulse->amp[i];
1198
                if (co) {
1199
                    co /= sf[idx];
1200
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1201
                }
1202
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1203
            }
1204
        }
1205
    }
1206
    return 0;
1207
}
1208

    
1209
static av_always_inline float flt16_round(float pf)
1210
{
1211
    union float754 tmp;
1212
    tmp.f = pf;
1213
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1214
    return tmp.f;
1215
}
1216

    
1217
static av_always_inline float flt16_even(float pf)
1218
{
1219
    union float754 tmp;
1220
    tmp.f = pf;
1221
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1222
    return tmp.f;
1223
}
1224

    
1225
static av_always_inline float flt16_trunc(float pf)
1226
{
1227
    union float754 pun;
1228
    pun.f = pf;
1229
    pun.i &= 0xFFFF0000U;
1230
    return pun.f;
1231
}
1232

    
1233
static av_always_inline void predict(PredictorState *ps, float *coef,
1234
                                     float sf_scale, float inv_sf_scale,
1235
                    int output_enable)
1236
{
1237
    const float a     = 0.953125; // 61.0 / 64
1238
    const float alpha = 0.90625;  // 29.0 / 32
1239
    float e0, e1;
1240
    float pv;
1241
    float k1, k2;
1242
    float   r0 = ps->r0,     r1 = ps->r1;
1243
    float cor0 = ps->cor0, cor1 = ps->cor1;
1244
    float var0 = ps->var0, var1 = ps->var1;
1245

    
1246
    k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1247
    k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1248

    
1249
    pv = flt16_round(k1 * r0 + k2 * r1);
1250
    if (output_enable)
1251
        *coef += pv * sf_scale;
1252

    
1253
    e0 = *coef * inv_sf_scale;
1254
    e1 = e0 - k1 * r0;
1255

    
1256
    ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1257
    ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1258
    ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1259
    ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1260

    
1261
    ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1262
    ps->r0 = flt16_trunc(a * e0);
1263
}
1264

    
1265
/**
1266
 * Apply AAC-Main style frequency domain prediction.
1267
 */
1268
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1269
{
1270
    int sfb, k;
1271
    float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1272

    
1273
    if (!sce->ics.predictor_initialized) {
1274
        reset_all_predictors(sce->predictor_state);
1275
        sce->ics.predictor_initialized = 1;
1276
    }
1277

    
1278
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1279
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1280
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1281
                predict(&sce->predictor_state[k], &sce->coeffs[k],
1282
                        sf_scale, inv_sf_scale,
1283
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1284
            }
1285
        }
1286
        if (sce->ics.predictor_reset_group)
1287
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1288
    } else
1289
        reset_all_predictors(sce->predictor_state);
1290
}
1291

    
1292
/**
1293
 * Decode an individual_channel_stream payload; reference: table 4.44.
1294
 *
1295
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1296
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1297
 *
1298
 * @return  Returns error status. 0 - OK, !0 - error
1299
 */
1300
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1301
                      GetBitContext *gb, int common_window, int scale_flag)
1302
{
1303
    Pulse pulse;
1304
    TemporalNoiseShaping    *tns = &sce->tns;
1305
    IndividualChannelStream *ics = &sce->ics;
1306
    float *out = sce->coeffs;
1307
    int global_gain, pulse_present = 0;
1308

    
1309
    /* This assignment is to silence a GCC warning about the variable being used
1310
     * uninitialized when in fact it always is.
1311
     */
1312
    pulse.num_pulse = 0;
1313

    
1314
    global_gain = get_bits(gb, 8);
1315

    
1316
    if (!common_window && !scale_flag) {
1317
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1318
            return -1;
1319
    }
1320

    
1321
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1322
        return -1;
1323
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1324
        return -1;
1325

    
1326
    pulse_present = 0;
1327
    if (!scale_flag) {
1328
        if ((pulse_present = get_bits1(gb))) {
1329
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1330
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1331
                return -1;
1332
            }
1333
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1334
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1335
                return -1;
1336
            }
1337
        }
1338
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1339
            return -1;
1340
        if (get_bits1(gb)) {
1341
            av_log_missing_feature(ac->avctx, "SSR", 1);
1342
            return -1;
1343
        }
1344
    }
1345

    
1346
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1347
        return -1;
1348

    
1349
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1350
        apply_prediction(ac, sce);
1351

    
1352
    return 0;
1353
}
1354

    
1355
/**
1356
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1357
 */
1358
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1359
{
1360
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1361
    float *ch0 = cpe->ch[0].coeffs;
1362
    float *ch1 = cpe->ch[1].coeffs;
1363
    int g, i, group, idx = 0;
1364
    const uint16_t *offsets = ics->swb_offset;
1365
    for (g = 0; g < ics->num_window_groups; g++) {
1366
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1367
            if (cpe->ms_mask[idx] &&
1368
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1369
                for (group = 0; group < ics->group_len[g]; group++) {
1370
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1371
                                              ch1 + group * 128 + offsets[i],
1372
                                              offsets[i+1] - offsets[i]);
1373
                }
1374
            }
1375
        }
1376
        ch0 += ics->group_len[g] * 128;
1377
        ch1 += ics->group_len[g] * 128;
1378
    }
1379
}
1380

    
1381
/**
1382
 * intensity stereo decoding; reference: 4.6.8.2.3
1383
 *
1384
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1385
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1386
 *                      [3] reserved for scalable AAC
1387
 */
1388
static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1389
{
1390
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1391
    SingleChannelElement         *sce1 = &cpe->ch[1];
1392
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1393
    const uint16_t *offsets = ics->swb_offset;
1394
    int g, group, i, k, idx = 0;
1395
    int c;
1396
    float scale;
1397
    for (g = 0; g < ics->num_window_groups; g++) {
1398
        for (i = 0; i < ics->max_sfb;) {
1399
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1400
                const int bt_run_end = sce1->band_type_run_end[idx];
1401
                for (; i < bt_run_end; i++, idx++) {
1402
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1403
                    if (ms_present)
1404
                        c *= 1 - 2 * cpe->ms_mask[idx];
1405
                    scale = c * sce1->sf[idx];
1406
                    for (group = 0; group < ics->group_len[g]; group++)
1407
                        for (k = offsets[i]; k < offsets[i + 1]; k++)
1408
                            coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1409
                }
1410
            } else {
1411
                int bt_run_end = sce1->band_type_run_end[idx];
1412
                idx += bt_run_end - i;
1413
                i    = bt_run_end;
1414
            }
1415
        }
1416
        coef0 += ics->group_len[g] * 128;
1417
        coef1 += ics->group_len[g] * 128;
1418
    }
1419
}
1420

    
1421
/**
1422
 * Decode a channel_pair_element; reference: table 4.4.
1423
 *
1424
 * @return  Returns error status. 0 - OK, !0 - error
1425
 */
1426
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1427
{
1428
    int i, ret, common_window, ms_present = 0;
1429

    
1430
    common_window = get_bits1(gb);
1431
    if (common_window) {
1432
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1433
            return -1;
1434
        i = cpe->ch[1].ics.use_kb_window[0];
1435
        cpe->ch[1].ics = cpe->ch[0].ics;
1436
        cpe->ch[1].ics.use_kb_window[1] = i;
1437
        ms_present = get_bits(gb, 2);
1438
        if (ms_present == 3) {
1439
            av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1440
            return -1;
1441
        } else if (ms_present)
1442
            decode_mid_side_stereo(cpe, gb, ms_present);
1443
    }
1444
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1445
        return ret;
1446
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1447
        return ret;
1448

    
1449
    if (common_window) {
1450
        if (ms_present)
1451
            apply_mid_side_stereo(ac, cpe);
1452
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1453
            apply_prediction(ac, &cpe->ch[0]);
1454
            apply_prediction(ac, &cpe->ch[1]);
1455
        }
1456
    }
1457

    
1458
    apply_intensity_stereo(cpe, ms_present);
1459
    return 0;
1460
}
1461

    
1462
static const float cce_scale[] = {
1463
    1.09050773266525765921, //2^(1/8)
1464
    1.18920711500272106672, //2^(1/4)
1465
    M_SQRT2,
1466
    2,
1467
};
1468

    
1469
/**
1470
 * Decode coupling_channel_element; reference: table 4.8.
1471
 *
1472
 * @return  Returns error status. 0 - OK, !0 - error
1473
 */
1474
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1475
{
1476
    int num_gain = 0;
1477
    int c, g, sfb, ret;
1478
    int sign;
1479
    float scale;
1480
    SingleChannelElement *sce = &che->ch[0];
1481
    ChannelCoupling     *coup = &che->coup;
1482

    
1483
    coup->coupling_point = 2 * get_bits1(gb);
1484
    coup->num_coupled = get_bits(gb, 3);
1485
    for (c = 0; c <= coup->num_coupled; c++) {
1486
        num_gain++;
1487
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1488
        coup->id_select[c] = get_bits(gb, 4);
1489
        if (coup->type[c] == TYPE_CPE) {
1490
            coup->ch_select[c] = get_bits(gb, 2);
1491
            if (coup->ch_select[c] == 3)
1492
                num_gain++;
1493
        } else
1494
            coup->ch_select[c] = 2;
1495
    }
1496
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1497

    
1498
    sign  = get_bits(gb, 1);
1499
    scale = cce_scale[get_bits(gb, 2)];
1500

    
1501
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1502
        return ret;
1503

    
1504
    for (c = 0; c < num_gain; c++) {
1505
        int idx  = 0;
1506
        int cge  = 1;
1507
        int gain = 0;
1508
        float gain_cache = 1.;
1509
        if (c) {
1510
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1511
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1512
            gain_cache = powf(scale, -gain);
1513
        }
1514
        if (coup->coupling_point == AFTER_IMDCT) {
1515
            coup->gain[c][0] = gain_cache;
1516
        } else {
1517
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1518
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1519
                    if (sce->band_type[idx] != ZERO_BT) {
1520
                        if (!cge) {
1521
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1522
                            if (t) {
1523
                                int s = 1;
1524
                                t = gain += t;
1525
                                if (sign) {
1526
                                    s  -= 2 * (t & 0x1);
1527
                                    t >>= 1;
1528
                                }
1529
                                gain_cache = powf(scale, -t) * s;
1530
                            }
1531
                        }
1532
                        coup->gain[c][idx] = gain_cache;
1533
                    }
1534
                }
1535
            }
1536
        }
1537
    }
1538
    return 0;
1539
}
1540

    
1541
/**
1542
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1543
 *
1544
 * @return  Returns number of bytes consumed.
1545
 */
1546
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1547
                                         GetBitContext *gb)
1548
{
1549
    int i;
1550
    int num_excl_chan = 0;
1551

    
1552
    do {
1553
        for (i = 0; i < 7; i++)
1554
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1555
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1556

    
1557
    return num_excl_chan / 7;
1558
}
1559

    
1560
/**
1561
 * Decode dynamic range information; reference: table 4.52.
1562
 *
1563
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1564
 *
1565
 * @return  Returns number of bytes consumed.
1566
 */
1567
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1568
                                GetBitContext *gb, int cnt)
1569
{
1570
    int n             = 1;
1571
    int drc_num_bands = 1;
1572
    int i;
1573

    
1574
    /* pce_tag_present? */
1575
    if (get_bits1(gb)) {
1576
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1577
        skip_bits(gb, 4); // tag_reserved_bits
1578
        n++;
1579
    }
1580

    
1581
    /* excluded_chns_present? */
1582
    if (get_bits1(gb)) {
1583
        n += decode_drc_channel_exclusions(che_drc, gb);
1584
    }
1585

    
1586
    /* drc_bands_present? */
1587
    if (get_bits1(gb)) {
1588
        che_drc->band_incr            = get_bits(gb, 4);
1589
        che_drc->interpolation_scheme = get_bits(gb, 4);
1590
        n++;
1591
        drc_num_bands += che_drc->band_incr;
1592
        for (i = 0; i < drc_num_bands; i++) {
1593
            che_drc->band_top[i] = get_bits(gb, 8);
1594
            n++;
1595
        }
1596
    }
1597

    
1598
    /* prog_ref_level_present? */
1599
    if (get_bits1(gb)) {
1600
        che_drc->prog_ref_level = get_bits(gb, 7);
1601
        skip_bits1(gb); // prog_ref_level_reserved_bits
1602
        n++;
1603
    }
1604

    
1605
    for (i = 0; i < drc_num_bands; i++) {
1606
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1607
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1608
        n++;
1609
    }
1610

    
1611
    return n;
1612
}
1613

    
1614
/**
1615
 * Decode extension data (incomplete); reference: table 4.51.
1616
 *
1617
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1618
 *
1619
 * @return Returns number of bytes consumed
1620
 */
1621
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1622
                                    ChannelElement *che, enum RawDataBlockType elem_type)
1623
{
1624
    int crc_flag = 0;
1625
    int res = cnt;
1626
    switch (get_bits(gb, 4)) { // extension type
1627
    case EXT_SBR_DATA_CRC:
1628
        crc_flag++;
1629
    case EXT_SBR_DATA:
1630
        if (!che) {
1631
            av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1632
            return res;
1633
        } else if (!ac->m4ac.sbr) {
1634
            av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1635
            skip_bits_long(gb, 8 * cnt - 4);
1636
            return res;
1637
        } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1638
            av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1639
            skip_bits_long(gb, 8 * cnt - 4);
1640
            return res;
1641
        } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1642
            ac->m4ac.sbr = 1;
1643
            ac->m4ac.ps = 1;
1644
            output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1645
        } else {
1646
            ac->m4ac.sbr = 1;
1647
        }
1648
        res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1649
        break;
1650
    case EXT_DYNAMIC_RANGE:
1651
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1652
        break;
1653
    case EXT_FILL:
1654
    case EXT_FILL_DATA:
1655
    case EXT_DATA_ELEMENT:
1656
    default:
1657
        skip_bits_long(gb, 8 * cnt - 4);
1658
        break;
1659
    };
1660
    return res;
1661
}
1662

    
1663
/**
1664
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1665
 *
1666
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1667
 * @param   coef    spectral coefficients
1668
 */
1669
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1670
                      IndividualChannelStream *ics, int decode)
1671
{
1672
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1673
    int w, filt, m, i;
1674
    int bottom, top, order, start, end, size, inc;
1675
    float lpc[TNS_MAX_ORDER];
1676

    
1677
    for (w = 0; w < ics->num_windows; w++) {
1678
        bottom = ics->num_swb;
1679
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1680
            top    = bottom;
1681
            bottom = FFMAX(0, top - tns->length[w][filt]);
1682
            order  = tns->order[w][filt];
1683
            if (order == 0)
1684
                continue;
1685

    
1686
            // tns_decode_coef
1687
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1688

    
1689
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1690
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1691
            if ((size = end - start) <= 0)
1692
                continue;
1693
            if (tns->direction[w][filt]) {
1694
                inc = -1;
1695
                start = end - 1;
1696
            } else {
1697
                inc = 1;
1698
            }
1699
            start += w * 128;
1700

    
1701
            // ar filter
1702
            for (m = 0; m < size; m++, start += inc)
1703
                for (i = 1; i <= FFMIN(m, order); i++)
1704
                    coef[start] -= coef[start - i * inc] * lpc[i - 1];
1705
        }
1706
    }
1707
}
1708

    
1709
/**
1710
 * Conduct IMDCT and windowing.
1711
 */
1712
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
1713
{
1714
    IndividualChannelStream *ics = &sce->ics;
1715
    float *in    = sce->coeffs;
1716
    float *out   = sce->ret;
1717
    float *saved = sce->saved;
1718
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1719
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1720
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1721
    float *buf  = ac->buf_mdct;
1722
    float *temp = ac->temp;
1723
    int i;
1724

    
1725
    // imdct
1726
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1727
        for (i = 0; i < 1024; i += 128)
1728
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1729
    } else
1730
        ff_imdct_half(&ac->mdct, buf, in);
1731

    
1732
    /* window overlapping
1733
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1734
     * and long to short transitions are considered to be short to short
1735
     * transitions. This leaves just two cases (long to long and short to short)
1736
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1737
     */
1738
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1739
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1740
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, bias, 512);
1741
    } else {
1742
        for (i = 0; i < 448; i++)
1743
            out[i] = saved[i] + bias;
1744

    
1745
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1746
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, bias, 64);
1747
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      bias, 64);
1748
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      bias, 64);
1749
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      bias, 64);
1750
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      bias, 64);
1751
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1752
        } else {
1753
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, bias, 64);
1754
            for (i = 576; i < 1024; i++)
1755
                out[i] = buf[i-512] + bias;
1756
        }
1757
    }
1758

    
1759
    // buffer update
1760
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1761
        for (i = 0; i < 64; i++)
1762
            saved[i] = temp[64 + i] - bias;
1763
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1764
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1765
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1766
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1767
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1768
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1769
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1770
    } else { // LONG_STOP or ONLY_LONG
1771
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1772
    }
1773
}
1774

    
1775
/**
1776
 * Apply dependent channel coupling (applied before IMDCT).
1777
 *
1778
 * @param   index   index into coupling gain array
1779
 */
1780
static void apply_dependent_coupling(AACContext *ac,
1781
                                     SingleChannelElement *target,
1782
                                     ChannelElement *cce, int index)
1783
{
1784
    IndividualChannelStream *ics = &cce->ch[0].ics;
1785
    const uint16_t *offsets = ics->swb_offset;
1786
    float *dest = target->coeffs;
1787
    const float *src = cce->ch[0].coeffs;
1788
    int g, i, group, k, idx = 0;
1789
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1790
        av_log(ac->avctx, AV_LOG_ERROR,
1791
               "Dependent coupling is not supported together with LTP\n");
1792
        return;
1793
    }
1794
    for (g = 0; g < ics->num_window_groups; g++) {
1795
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1796
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1797
                const float gain = cce->coup.gain[index][idx];
1798
                for (group = 0; group < ics->group_len[g]; group++) {
1799
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1800
                        // XXX dsputil-ize
1801
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1802
                    }
1803
                }
1804
            }
1805
        }
1806
        dest += ics->group_len[g] * 128;
1807
        src  += ics->group_len[g] * 128;
1808
    }
1809
}
1810

    
1811
/**
1812
 * Apply independent channel coupling (applied after IMDCT).
1813
 *
1814
 * @param   index   index into coupling gain array
1815
 */
1816
static void apply_independent_coupling(AACContext *ac,
1817
                                       SingleChannelElement *target,
1818
                                       ChannelElement *cce, int index)
1819
{
1820
    int i;
1821
    const float gain = cce->coup.gain[index][0];
1822
    const float bias = ac->add_bias;
1823
    const float *src = cce->ch[0].ret;
1824
    float *dest = target->ret;
1825
    const int len = 1024 << (ac->m4ac.sbr == 1);
1826

    
1827
    for (i = 0; i < len; i++)
1828
        dest[i] += gain * (src[i] - bias);
1829
}
1830

    
1831
/**
1832
 * channel coupling transformation interface
1833
 *
1834
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1835
 */
1836
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1837
                                   enum RawDataBlockType type, int elem_id,
1838
                                   enum CouplingPoint coupling_point,
1839
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1840
{
1841
    int i, c;
1842

    
1843
    for (i = 0; i < MAX_ELEM_ID; i++) {
1844
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1845
        int index = 0;
1846

    
1847
        if (cce && cce->coup.coupling_point == coupling_point) {
1848
            ChannelCoupling *coup = &cce->coup;
1849

    
1850
            for (c = 0; c <= coup->num_coupled; c++) {
1851
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1852
                    if (coup->ch_select[c] != 1) {
1853
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1854
                        if (coup->ch_select[c] != 0)
1855
                            index++;
1856
                    }
1857
                    if (coup->ch_select[c] != 2)
1858
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1859
                } else
1860
                    index += 1 + (coup->ch_select[c] == 3);
1861
            }
1862
        }
1863
    }
1864
}
1865

    
1866
/**
1867
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1868
 */
1869
static void spectral_to_sample(AACContext *ac)
1870
{
1871
    int i, type;
1872
    float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
1873
    for (type = 3; type >= 0; type--) {
1874
        for (i = 0; i < MAX_ELEM_ID; i++) {
1875
            ChannelElement *che = ac->che[type][i];
1876
            if (che) {
1877
                if (type <= TYPE_CPE)
1878
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1879
                if (che->ch[0].tns.present)
1880
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1881
                if (che->ch[1].tns.present)
1882
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1883
                if (type <= TYPE_CPE)
1884
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1885
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1886
                    imdct_and_windowing(ac, &che->ch[0], imdct_bias);
1887
                    if (type == TYPE_CPE) {
1888
                        imdct_and_windowing(ac, &che->ch[1], imdct_bias);
1889
                    }
1890
                    if (ac->m4ac.sbr > 0) {
1891
                        ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
1892
                    }
1893
                }
1894
                if (type <= TYPE_CCE)
1895
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1896
            }
1897
        }
1898
    }
1899
}
1900

    
1901
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1902
{
1903
    int size;
1904
    AACADTSHeaderInfo hdr_info;
1905

    
1906
    size = ff_aac_parse_header(gb, &hdr_info);
1907
    if (size > 0) {
1908
        if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1909
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1910
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1911
            ac->m4ac.chan_config = hdr_info.chan_config;
1912
            if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
1913
                return -7;
1914
            if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1915
                return -7;
1916
        } else if (ac->output_configured != OC_LOCKED) {
1917
            ac->output_configured = OC_NONE;
1918
        }
1919
        if (ac->output_configured != OC_LOCKED) {
1920
            ac->m4ac.sbr = -1;
1921
            ac->m4ac.ps  = -1;
1922
        }
1923
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1924
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1925
        ac->m4ac.object_type     = hdr_info.object_type;
1926
        if (!ac->avctx->sample_rate)
1927
            ac->avctx->sample_rate = hdr_info.sample_rate;
1928
        if (hdr_info.num_aac_frames == 1) {
1929
            if (!hdr_info.crc_absent)
1930
                skip_bits(gb, 16);
1931
        } else {
1932
            av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
1933
            return -1;
1934
        }
1935
    }
1936
    return size;
1937
}
1938

    
1939
static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
1940
                                int *data_size, GetBitContext *gb)
1941
{
1942
    AACContext *ac = avctx->priv_data;
1943
    ChannelElement *che = NULL, *che_prev = NULL;
1944
    enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1945
    int err, elem_id, data_size_tmp;
1946
    int samples = 0, multiplier;
1947

    
1948
    if (show_bits(gb, 12) == 0xfff) {
1949
        if (parse_adts_frame_header(ac, gb) < 0) {
1950
            av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1951
            return -1;
1952
        }
1953
        if (ac->m4ac.sampling_index > 12) {
1954
            av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1955
            return -1;
1956
        }
1957
    }
1958

    
1959
    ac->tags_mapped = 0;
1960
    // parse
1961
    while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
1962
        elem_id = get_bits(gb, 4);
1963

    
1964
        if (elem_type < TYPE_DSE) {
1965
            if (!(che=get_che(ac, elem_type, elem_id))) {
1966
                av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
1967
                       elem_type, elem_id);
1968
                return -1;
1969
            }
1970
            samples = 1024;
1971
        }
1972

    
1973
        switch (elem_type) {
1974

    
1975
        case TYPE_SCE:
1976
            err = decode_ics(ac, &che->ch[0], gb, 0, 0);
1977
            break;
1978

    
1979
        case TYPE_CPE:
1980
            err = decode_cpe(ac, gb, che);
1981
            break;
1982

    
1983
        case TYPE_CCE:
1984
            err = decode_cce(ac, gb, che);
1985
            break;
1986

    
1987
        case TYPE_LFE:
1988
            err = decode_ics(ac, &che->ch[0], gb, 0, 0);
1989
            break;
1990

    
1991
        case TYPE_DSE:
1992
            err = skip_data_stream_element(ac, gb);
1993
            break;
1994

    
1995
        case TYPE_PCE: {
1996
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1997
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1998
            if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
1999
                break;
2000
            if (ac->output_configured > OC_TRIAL_PCE)
2001
                av_log(avctx, AV_LOG_ERROR,
2002
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2003
            else
2004
                err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2005
            break;
2006
        }
2007

    
2008
        case TYPE_FIL:
2009
            if (elem_id == 15)
2010
                elem_id += get_bits(gb, 8) - 1;
2011
            if (get_bits_left(gb) < 8 * elem_id) {
2012
                    av_log(avctx, AV_LOG_ERROR, overread_err);
2013
                    return -1;
2014
            }
2015
            while (elem_id > 0)
2016
                elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2017
            err = 0; /* FIXME */
2018
            break;
2019

    
2020
        default:
2021
            err = -1; /* should not happen, but keeps compiler happy */
2022
            break;
2023
        }
2024

    
2025
        che_prev       = che;
2026
        elem_type_prev = elem_type;
2027

    
2028
        if (err)
2029
            return err;
2030

    
2031
        if (get_bits_left(gb) < 3) {
2032
            av_log(avctx, AV_LOG_ERROR, overread_err);
2033
            return -1;
2034
        }
2035
    }
2036

    
2037
    spectral_to_sample(ac);
2038

    
2039
    multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2040
    samples <<= multiplier;
2041
    if (ac->output_configured < OC_LOCKED) {
2042
        avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2043
        avctx->frame_size = samples;
2044
    }
2045

    
2046
    data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2047
    if (*data_size < data_size_tmp) {
2048
        av_log(avctx, AV_LOG_ERROR,
2049
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2050
               *data_size, data_size_tmp);
2051
        return -1;
2052
    }
2053
    *data_size = data_size_tmp;
2054

    
2055
    if (samples)
2056
        ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2057

    
2058
    if (ac->output_configured)
2059
        ac->output_configured = OC_LOCKED;
2060

    
2061
    return 0;
2062
}
2063

    
2064
static int aac_decode_frame(AVCodecContext *avctx, void *data,
2065
                            int *data_size, AVPacket *avpkt)
2066
{
2067
    const uint8_t *buf = avpkt->data;
2068
    int buf_size = avpkt->size;
2069
    GetBitContext gb;
2070
    int buf_consumed;
2071
    int buf_offset;
2072
    int err;
2073

    
2074
    init_get_bits(&gb, buf, buf_size * 8);
2075

    
2076
    if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2077
        return err;
2078

    
2079
    buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2080
    for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2081
        if (buf[buf_offset])
2082
            break;
2083

    
2084
    return buf_size > buf_offset ? buf_consumed : buf_size;
2085
}
2086

    
2087
static av_cold int aac_decode_close(AVCodecContext *avctx)
2088
{
2089
    AACContext *ac = avctx->priv_data;
2090
    int i, type;
2091

    
2092
    for (i = 0; i < MAX_ELEM_ID; i++) {
2093
        for (type = 0; type < 4; type++) {
2094
            if (ac->che[type][i])
2095
                ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2096
            av_freep(&ac->che[type][i]);
2097
        }
2098
    }
2099

    
2100
    ff_mdct_end(&ac->mdct);
2101
    ff_mdct_end(&ac->mdct_small);
2102
    return 0;
2103
}
2104

    
2105

    
2106
#define LOAS_SYNC_WORD   0x2b7       ///< 11 bits LOAS sync word
2107

    
2108
struct LATMContext {
2109
    AACContext      aac_ctx;             ///< containing AACContext
2110
    int             initialized;         ///< initilized after a valid extradata was seen
2111

    
2112
    // parser data
2113
    int             audio_mux_version_A; ///< LATM syntax version
2114
    int             frame_length_type;   ///< 0/1 variable/fixed frame length
2115
    int             frame_length;        ///< frame length for fixed frame length
2116
};
2117

    
2118
static inline uint32_t latm_get_value(GetBitContext *b)
2119
{
2120
    int length = get_bits(b, 2);
2121

    
2122
    return get_bits_long(b, (length+1)*8);
2123
}
2124

    
2125
static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2126
                                             GetBitContext *gb)
2127
{
2128
    AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2129
    MPEG4AudioConfig m4ac;
2130
    int  config_start_bit = get_bits_count(gb);
2131
    int     bits_consumed, esize;
2132

    
2133
    if (config_start_bit % 8) {
2134
        av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2135
                               "config not byte aligned.\n", 1);
2136
        return AVERROR_INVALIDDATA;
2137
    } else {
2138
        bits_consumed =
2139
            decode_audio_specific_config(NULL, avctx, &m4ac,
2140
                                         gb->buffer + (config_start_bit / 8),
2141
                                         get_bits_left(gb) / 8);
2142

    
2143
        if (bits_consumed < 0)
2144
            return AVERROR_INVALIDDATA;
2145

    
2146
        esize = (bits_consumed+7) / 8;
2147

    
2148
        if (avctx->extradata_size <= esize) {
2149
            av_free(avctx->extradata);
2150
            avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2151
            if (!avctx->extradata)
2152
                return AVERROR(ENOMEM);
2153
        }
2154

    
2155
        avctx->extradata_size = esize;
2156
        memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2157
        memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2158

    
2159
        skip_bits_long(gb, bits_consumed);
2160
    }
2161

    
2162
    return bits_consumed;
2163
}
2164

    
2165
static int read_stream_mux_config(struct LATMContext *latmctx,
2166
                                  GetBitContext *gb)
2167
{
2168
    int ret, audio_mux_version = get_bits(gb, 1);
2169

    
2170
    latmctx->audio_mux_version_A = 0;
2171
    if (audio_mux_version)
2172
        latmctx->audio_mux_version_A = get_bits(gb, 1);
2173

    
2174
    if (!latmctx->audio_mux_version_A) {
2175

    
2176
        if (audio_mux_version)
2177
            latm_get_value(gb);                 // taraFullness
2178

    
2179
        skip_bits(gb, 1);                       // allStreamSameTimeFraming
2180
        skip_bits(gb, 6);                       // numSubFrames
2181
        // numPrograms
2182
        if (get_bits(gb, 4)) {                  // numPrograms
2183
            av_log_missing_feature(latmctx->aac_ctx.avctx,
2184
                                   "multiple programs are not supported\n", 1);
2185
            return AVERROR_PATCHWELCOME;
2186
        }
2187

    
2188
        // for each program (which there is only on in DVB)
2189

    
2190
        // for each layer (which there is only on in DVB)
2191
        if (get_bits(gb, 3)) {                   // numLayer
2192
            av_log_missing_feature(latmctx->aac_ctx.avctx,
2193
                                   "multiple layers are not supported\n", 1);
2194
            return AVERROR_PATCHWELCOME;
2195
        }
2196

    
2197
        // for all but first stream: use_same_config = get_bits(gb, 1);
2198
        if (!audio_mux_version) {
2199
            if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2200
                return ret;
2201
        } else {
2202
            int ascLen = latm_get_value(gb);
2203
            if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2204
                return ret;
2205
            ascLen -= ret;
2206
            skip_bits_long(gb, ascLen);
2207
        }
2208

    
2209
        latmctx->frame_length_type = get_bits(gb, 3);
2210
        switch (latmctx->frame_length_type) {
2211
        case 0:
2212
            skip_bits(gb, 8);       // latmBufferFullness
2213
            break;
2214
        case 1:
2215
            latmctx->frame_length = get_bits(gb, 9);
2216
            break;
2217
        case 3:
2218
        case 4:
2219
        case 5:
2220
            skip_bits(gb, 6);       // CELP frame length table index
2221
            break;
2222
        case 6:
2223
        case 7:
2224
            skip_bits(gb, 1);       // HVXC frame length table index
2225
            break;
2226
        }
2227

    
2228
        if (get_bits(gb, 1)) {                  // other data
2229
            if (audio_mux_version) {
2230
                latm_get_value(gb);             // other_data_bits
2231
            } else {
2232
                int esc;
2233
                do {
2234
                    esc = get_bits(gb, 1);
2235
                    skip_bits(gb, 8);
2236
                } while (esc);
2237
            }
2238
        }
2239

    
2240
        if (get_bits(gb, 1))                     // crc present
2241
            skip_bits(gb, 8);                    // config_crc
2242
    }
2243

    
2244
    return 0;
2245
}
2246

    
2247
static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2248
{
2249
    uint8_t tmp;
2250

    
2251
    if (ctx->frame_length_type == 0) {
2252
        int mux_slot_length = 0;
2253
        do {
2254
            tmp = get_bits(gb, 8);
2255
            mux_slot_length += tmp;
2256
        } while (tmp == 255);
2257
        return mux_slot_length;
2258
    } else if (ctx->frame_length_type == 1) {
2259
        return ctx->frame_length;
2260
    } else if (ctx->frame_length_type == 3 ||
2261
               ctx->frame_length_type == 5 ||
2262
               ctx->frame_length_type == 7) {
2263
        skip_bits(gb, 2);          // mux_slot_length_coded
2264
    }
2265
    return 0;
2266
}
2267

    
2268
static int read_audio_mux_element(struct LATMContext *latmctx,
2269
                                  GetBitContext *gb)
2270
{
2271
    int err;
2272
    uint8_t use_same_mux = get_bits(gb, 1);
2273
    if (!use_same_mux) {
2274
        if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2275
            return err;
2276
    } else if (!latmctx->aac_ctx.avctx->extradata) {
2277
        av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2278
               "no decoder config found\n");
2279
        return AVERROR(EAGAIN);
2280
    }
2281
    if (latmctx->audio_mux_version_A == 0) {
2282
        int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2283
        if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2284
            av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2285
            return AVERROR_INVALIDDATA;
2286
        } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2287
            av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2288
                   "frame length mismatch %d << %d\n",
2289
                   mux_slot_length_bytes * 8, get_bits_left(gb));
2290
            return AVERROR_INVALIDDATA;
2291
        }
2292
    }
2293
    return 0;
2294
}
2295

    
2296

    
2297
static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2298
                             AVPacket *avpkt)
2299
{
2300
    struct LATMContext *latmctx = avctx->priv_data;
2301
    int                 muxlength, err;
2302
    GetBitContext       gb;
2303

    
2304
    if (avpkt->size == 0)
2305
        return 0;
2306

    
2307
    init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2308

    
2309
    // check for LOAS sync word
2310
    if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2311
        return AVERROR_INVALIDDATA;
2312

    
2313
    muxlength = get_bits(&gb, 13) + 3;
2314
    // not enough data, the parser should have sorted this
2315
    if (muxlength > avpkt->size)
2316
        return AVERROR_INVALIDDATA;
2317

    
2318
    if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2319
        return err;
2320

    
2321
    if (!latmctx->initialized) {
2322
        if (!avctx->extradata) {
2323
            *out_size = 0;
2324
            return avpkt->size;
2325
        } else {
2326
            if ((err = aac_decode_init(avctx)) < 0)
2327
                return err;
2328
            latmctx->initialized = 1;
2329
        }
2330
    }
2331

    
2332
    if (show_bits(&gb, 12) == 0xfff) {
2333
        av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2334
               "ADTS header detected, probably as result of configuration "
2335
               "misparsing\n");
2336
        return AVERROR_INVALIDDATA;
2337
    }
2338

    
2339
    if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2340
        return err;
2341

    
2342
    return muxlength;
2343
}
2344

    
2345
av_cold static int latm_decode_init(AVCodecContext *avctx)
2346
{
2347
    struct LATMContext *latmctx = avctx->priv_data;
2348
    int ret;
2349

    
2350
    ret = aac_decode_init(avctx);
2351

    
2352
    if (avctx->extradata_size > 0) {
2353
        latmctx->initialized = !ret;
2354
    } else {
2355
        latmctx->initialized = 0;
2356
    }
2357

    
2358
    return ret;
2359
}
2360

    
2361

    
2362
AVCodec aac_decoder = {
2363
    "aac",
2364
    AVMEDIA_TYPE_AUDIO,
2365
    CODEC_ID_AAC,
2366
    sizeof(AACContext),
2367
    aac_decode_init,
2368
    NULL,
2369
    aac_decode_close,
2370
    aac_decode_frame,
2371
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2372
    .sample_fmts = (const enum SampleFormat[]) {
2373
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2374
    },
2375
    .channel_layouts = aac_channel_layout,
2376
};
2377

    
2378
/*
2379
    Note: This decoder filter is intended to decode LATM streams transferred
2380
    in MPEG transport streams which only contain one program.
2381
    To do a more complex LATM demuxing a separate LATM demuxer should be used.
2382
*/
2383
AVCodec aac_latm_decoder = {
2384
    .name = "aac_latm",
2385
    .type = CODEC_TYPE_AUDIO,
2386
    .id   = CODEC_ID_AAC_LATM,
2387
    .priv_data_size = sizeof(struct LATMContext),
2388
    .init   = latm_decode_init,
2389
    .close  = aac_decode_close,
2390
    .decode = latm_decode_frame,
2391
    .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2392
    .sample_fmts = (const enum SampleFormat[]) {
2393
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2394
    },
2395
    .channel_layouts = aac_channel_layout,
2396
};