Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aac.c @ 2ef21b91

History | View | Annotate | Download (70.4 KB)

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

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

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

    
78

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

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

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

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

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

    
104
static uint32_t cbrt_tab[1<<13];
105

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

    
156
/**
157
 * Check for the channel element in the current channel position configuration.
158
 * If it exists, make sure the appropriate element is allocated and map the
159
 * channel order to match the internal FFmpeg channel layout.
160
 *
161
 * @param   che_pos current channel position configuration
162
 * @param   type channel element type
163
 * @param   id channel element id
164
 * @param   channels count of the number of channels in the configuration
165
 *
166
 * @return  Returns error status. 0 - OK, !0 - error
167
 */
168
static int che_configure(AACContext *ac,
169
                         enum ChannelPosition che_pos[4][MAX_ELEM_ID],
170
                         int type, int id,
171
                         int *channels)
172
{
173
    if (che_pos[type][id]) {
174
        if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
175
            return AVERROR(ENOMEM);
176
        if (type != TYPE_CCE) {
177
            ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
178
            if (type == TYPE_CPE) {
179
                ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
180
            }
181
        }
182
    } else
183
        av_freep(&ac->che[type][id]);
184
    return 0;
185
}
186

    
187
/**
188
 * Configure output channel order based on the current program configuration element.
189
 *
190
 * @param   che_pos current channel position configuration
191
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
192
 *
193
 * @return  Returns error status. 0 - OK, !0 - error
194
 */
195
static int output_configure(AACContext *ac,
196
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
197
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
198
                            int channel_config, enum OCStatus oc_type)
199
{
200
    AVCodecContext *avctx = ac->avccontext;
201
    int i, type, channels = 0, ret;
202

    
203
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
204

    
205
    if (channel_config) {
206
        for (i = 0; i < tags_per_config[channel_config]; i++) {
207
            if ((ret = che_configure(ac, che_pos,
208
                                     aac_channel_layout_map[channel_config - 1][i][0],
209
                                     aac_channel_layout_map[channel_config - 1][i][1],
210
                                     &channels)))
211
                return ret;
212
        }
213

    
214
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
215
        ac->tags_mapped = 0;
216

    
217
        avctx->channel_layout = aac_channel_layout[channel_config - 1];
218
    } else {
219
        /* Allocate or free elements depending on if they are in the
220
         * current program configuration.
221
         *
222
         * Set up default 1:1 output mapping.
223
         *
224
         * For a 5.1 stream the output order will be:
225
         *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
226
         */
227

    
228
        for (i = 0; i < MAX_ELEM_ID; i++) {
229
            for (type = 0; type < 4; type++) {
230
                if ((ret = che_configure(ac, che_pos, type, i, &channels)))
231
                    return ret;
232
            }
233
        }
234

    
235
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
236
        ac->tags_mapped = 4 * MAX_ELEM_ID;
237

    
238
        avctx->channel_layout = 0;
239
    }
240

    
241
    avctx->channels = channels;
242

    
243
    ac->output_configured = oc_type;
244

    
245
    return 0;
246
}
247

    
248
/**
249
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
250
 *
251
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
252
 * @param sce_map mono (Single Channel Element) map
253
 * @param type speaker type/position for these channels
254
 */
255
static void decode_channel_map(enum ChannelPosition *cpe_map,
256
                               enum ChannelPosition *sce_map,
257
                               enum ChannelPosition type,
258
                               GetBitContext *gb, int n)
259
{
260
    while (n--) {
261
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
262
        map[get_bits(gb, 4)] = type;
263
    }
264
}
265

    
266
/**
267
 * Decode program configuration element; reference: table 4.2.
268
 *
269
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
270
 *
271
 * @return  Returns error status. 0 - OK, !0 - error
272
 */
273
static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
274
                      GetBitContext *gb)
275
{
276
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
277

    
278
    skip_bits(gb, 2);  // object_type
279

    
280
    sampling_index = get_bits(gb, 4);
281
    if (ac->m4ac.sampling_index != sampling_index)
282
        av_log(ac->avccontext, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
283

    
284
    num_front       = get_bits(gb, 4);
285
    num_side        = get_bits(gb, 4);
286
    num_back        = get_bits(gb, 4);
287
    num_lfe         = get_bits(gb, 2);
288
    num_assoc_data  = get_bits(gb, 3);
289
    num_cc          = get_bits(gb, 4);
290

    
291
    if (get_bits1(gb))
292
        skip_bits(gb, 4); // mono_mixdown_tag
293
    if (get_bits1(gb))
294
        skip_bits(gb, 4); // stereo_mixdown_tag
295

    
296
    if (get_bits1(gb))
297
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
298

    
299
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
300
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
301
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
302
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
303

    
304
    skip_bits_long(gb, 4 * num_assoc_data);
305

    
306
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
307

    
308
    align_get_bits(gb);
309

    
310
    /* comment field, first byte is length */
311
    skip_bits_long(gb, 8 * get_bits(gb, 8));
312
    return 0;
313
}
314

    
315
/**
316
 * Set up channel positions based on a default channel configuration
317
 * as specified in table 1.17.
318
 *
319
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
320
 *
321
 * @return  Returns error status. 0 - OK, !0 - error
322
 */
323
static int set_default_channel_config(AACContext *ac,
324
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
325
                                      int channel_config)
326
{
327
    if (channel_config < 1 || channel_config > 7) {
328
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
329
               channel_config);
330
        return -1;
331
    }
332

    
333
    /* default channel configurations:
334
     *
335
     * 1ch : front center (mono)
336
     * 2ch : L + R (stereo)
337
     * 3ch : front center + L + R
338
     * 4ch : front center + L + R + back center
339
     * 5ch : front center + L + R + back stereo
340
     * 6ch : front center + L + R + back stereo + LFE
341
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
342
     */
343

    
344
    if (channel_config != 2)
345
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
346
    if (channel_config > 1)
347
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
348
    if (channel_config == 4)
349
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
350
    if (channel_config > 4)
351
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
352
        = AAC_CHANNEL_BACK;  // back stereo
353
    if (channel_config > 5)
354
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
355
    if (channel_config == 7)
356
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
357

    
358
    return 0;
359
}
360

    
361
/**
362
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
363
 *
364
 * @return  Returns error status. 0 - OK, !0 - error
365
 */
366
static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
367
                                     int channel_config)
368
{
369
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
370
    int extension_flag, ret;
371

    
372
    if (get_bits1(gb)) { // frameLengthFlag
373
        av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
374
        return -1;
375
    }
376

    
377
    if (get_bits1(gb))       // dependsOnCoreCoder
378
        skip_bits(gb, 14);   // coreCoderDelay
379
    extension_flag = get_bits1(gb);
380

    
381
    if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
382
        ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
383
        skip_bits(gb, 3);     // layerNr
384

    
385
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
386
    if (channel_config == 0) {
387
        skip_bits(gb, 4);  // element_instance_tag
388
        if ((ret = decode_pce(ac, new_che_pos, gb)))
389
            return ret;
390
    } else {
391
        if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
392
            return ret;
393
    }
394
    if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
395
        return ret;
396

    
397
    if (extension_flag) {
398
        switch (ac->m4ac.object_type) {
399
        case AOT_ER_BSAC:
400
            skip_bits(gb, 5);    // numOfSubFrame
401
            skip_bits(gb, 11);   // layer_length
402
            break;
403
        case AOT_ER_AAC_LC:
404
        case AOT_ER_AAC_LTP:
405
        case AOT_ER_AAC_SCALABLE:
406
        case AOT_ER_AAC_LD:
407
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
408
                                    * aacScalefactorDataResilienceFlag
409
                                    * aacSpectralDataResilienceFlag
410
                                    */
411
            break;
412
        }
413
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
414
    }
415
    return 0;
416
}
417

    
418
/**
419
 * Decode audio specific configuration; reference: table 1.13.
420
 *
421
 * @param   data        pointer to AVCodecContext extradata
422
 * @param   data_size   size of AVCCodecContext extradata
423
 *
424
 * @return  Returns error status. 0 - OK, !0 - error
425
 */
426
static int decode_audio_specific_config(AACContext *ac, void *data,
427
                                        int data_size)
428
{
429
    GetBitContext gb;
430
    int i;
431

    
432
    init_get_bits(&gb, data, data_size * 8);
433

    
434
    if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
435
        return -1;
436
    if (ac->m4ac.sampling_index > 12) {
437
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
438
        return -1;
439
    }
440

    
441
    skip_bits_long(&gb, i);
442

    
443
    switch (ac->m4ac.object_type) {
444
    case AOT_AAC_MAIN:
445
    case AOT_AAC_LC:
446
        if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
447
            return -1;
448
        break;
449
    default:
450
        av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
451
               ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
452
        return -1;
453
    }
454
    return 0;
455
}
456

    
457
/**
458
 * linear congruential pseudorandom number generator
459
 *
460
 * @param   previous_val    pointer to the current state of the generator
461
 *
462
 * @return  Returns a 32-bit pseudorandom integer
463
 */
464
static av_always_inline int lcg_random(int previous_val)
465
{
466
    return previous_val * 1664525 + 1013904223;
467
}
468

    
469
static void reset_predict_state(PredictorState *ps)
470
{
471
    ps->r0   = 0.0f;
472
    ps->r1   = 0.0f;
473
    ps->cor0 = 0.0f;
474
    ps->cor1 = 0.0f;
475
    ps->var0 = 1.0f;
476
    ps->var1 = 1.0f;
477
}
478

    
479
static void reset_all_predictors(PredictorState *ps)
480
{
481
    int i;
482
    for (i = 0; i < MAX_PREDICTORS; i++)
483
        reset_predict_state(&ps[i]);
484
}
485

    
486
static void reset_predictor_group(PredictorState *ps, int group_num)
487
{
488
    int i;
489
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
490
        reset_predict_state(&ps[i]);
491
}
492

    
493
static av_cold int aac_decode_init(AVCodecContext *avccontext)
494
{
495
    AACContext *ac = avccontext->priv_data;
496
    int i;
497

    
498
    ac->avccontext = avccontext;
499

    
500
    if (avccontext->extradata_size > 0) {
501
        if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
502
            return -1;
503
        avccontext->sample_rate = ac->m4ac.sample_rate;
504
    } else if (avccontext->channels > 0) {
505
        ac->m4ac.sample_rate = avccontext->sample_rate;
506
    }
507

    
508
    avccontext->sample_fmt = SAMPLE_FMT_S16;
509
    avccontext->frame_size = 1024;
510

    
511
    AAC_INIT_VLC_STATIC( 0, 304);
512
    AAC_INIT_VLC_STATIC( 1, 270);
513
    AAC_INIT_VLC_STATIC( 2, 550);
514
    AAC_INIT_VLC_STATIC( 3, 300);
515
    AAC_INIT_VLC_STATIC( 4, 328);
516
    AAC_INIT_VLC_STATIC( 5, 294);
517
    AAC_INIT_VLC_STATIC( 6, 306);
518
    AAC_INIT_VLC_STATIC( 7, 268);
519
    AAC_INIT_VLC_STATIC( 8, 510);
520
    AAC_INIT_VLC_STATIC( 9, 366);
521
    AAC_INIT_VLC_STATIC(10, 462);
522

    
523
    dsputil_init(&ac->dsp, avccontext);
524

    
525
    ac->random_state = 0x1f2e3d4c;
526

    
527
    // -1024 - Compensate wrong IMDCT method.
528
    // 32768 - Required to scale values to the correct range for the bias method
529
    //         for float to int16 conversion.
530

    
531
    if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
532
        ac->add_bias  = 385.0f;
533
        ac->sf_scale  = 1. / (-1024. * 32768.);
534
        ac->sf_offset = 0;
535
    } else {
536
        ac->add_bias  = 0.0f;
537
        ac->sf_scale  = 1. / -1024.;
538
        ac->sf_offset = 60;
539
    }
540

    
541
#if !CONFIG_HARDCODED_TABLES
542
    for (i = 0; i < 428; i++)
543
        ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
544
#endif /* CONFIG_HARDCODED_TABLES */
545

    
546
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
547
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
548
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
549
                    352);
550

    
551
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
552
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
553
    // window initialization
554
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
555
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
556
    ff_init_ff_sine_windows(10);
557
    ff_init_ff_sine_windows( 7);
558

    
559
    if (!cbrt_tab[(1<<13) - 1]) {
560
        for (i = 0; i < 1<<13; i++) {
561
            union float754 f;
562
            f.f = cbrtf(i) * i;
563
            cbrt_tab[i] = f.i;
564
        }
565
    }
566

    
567
    return 0;
568
}
569

    
570
/**
571
 * Skip data_stream_element; reference: table 4.10.
572
 */
573
static void skip_data_stream_element(GetBitContext *gb)
574
{
575
    int byte_align = get_bits1(gb);
576
    int count = get_bits(gb, 8);
577
    if (count == 255)
578
        count += get_bits(gb, 8);
579
    if (byte_align)
580
        align_get_bits(gb);
581
    skip_bits_long(gb, 8 * count);
582
}
583

    
584
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
585
                             GetBitContext *gb)
586
{
587
    int sfb;
588
    if (get_bits1(gb)) {
589
        ics->predictor_reset_group = get_bits(gb, 5);
590
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
591
            av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
592
            return -1;
593
        }
594
    }
595
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
596
        ics->prediction_used[sfb] = get_bits1(gb);
597
    }
598
    return 0;
599
}
600

    
601
/**
602
 * Decode Individual Channel Stream info; reference: table 4.6.
603
 *
604
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
605
 */
606
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
607
                           GetBitContext *gb, int common_window)
608
{
609
    if (get_bits1(gb)) {
610
        av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
611
        memset(ics, 0, sizeof(IndividualChannelStream));
612
        return -1;
613
    }
614
    ics->window_sequence[1] = ics->window_sequence[0];
615
    ics->window_sequence[0] = get_bits(gb, 2);
616
    ics->use_kb_window[1]   = ics->use_kb_window[0];
617
    ics->use_kb_window[0]   = get_bits1(gb);
618
    ics->num_window_groups  = 1;
619
    ics->group_len[0]       = 1;
620
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
621
        int i;
622
        ics->max_sfb = get_bits(gb, 4);
623
        for (i = 0; i < 7; i++) {
624
            if (get_bits1(gb)) {
625
                ics->group_len[ics->num_window_groups - 1]++;
626
            } else {
627
                ics->num_window_groups++;
628
                ics->group_len[ics->num_window_groups - 1] = 1;
629
            }
630
        }
631
        ics->num_windows       = 8;
632
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
633
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
634
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
635
        ics->predictor_present = 0;
636
    } else {
637
        ics->max_sfb               = get_bits(gb, 6);
638
        ics->num_windows           = 1;
639
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
640
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
641
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
642
        ics->predictor_present     = get_bits1(gb);
643
        ics->predictor_reset_group = 0;
644
        if (ics->predictor_present) {
645
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
646
                if (decode_prediction(ac, ics, gb)) {
647
                    memset(ics, 0, sizeof(IndividualChannelStream));
648
                    return -1;
649
                }
650
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
651
                av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
652
                memset(ics, 0, sizeof(IndividualChannelStream));
653
                return -1;
654
            } else {
655
                av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
656
                memset(ics, 0, sizeof(IndividualChannelStream));
657
                return -1;
658
            }
659
        }
660
    }
661

    
662
    if (ics->max_sfb > ics->num_swb) {
663
        av_log(ac->avccontext, AV_LOG_ERROR,
664
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
665
               ics->max_sfb, ics->num_swb);
666
        memset(ics, 0, sizeof(IndividualChannelStream));
667
        return -1;
668
    }
669

    
670
    return 0;
671
}
672

    
673
/**
674
 * Decode band types (section_data payload); reference: table 4.46.
675
 *
676
 * @param   band_type           array of the used band type
677
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
678
 *
679
 * @return  Returns error status. 0 - OK, !0 - error
680
 */
681
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
682
                             int band_type_run_end[120], GetBitContext *gb,
683
                             IndividualChannelStream *ics)
684
{
685
    int g, idx = 0;
686
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
687
    for (g = 0; g < ics->num_window_groups; g++) {
688
        int k = 0;
689
        while (k < ics->max_sfb) {
690
            uint8_t sect_end = k;
691
            int sect_len_incr;
692
            int sect_band_type = get_bits(gb, 4);
693
            if (sect_band_type == 12) {
694
                av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
695
                return -1;
696
            }
697
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
698
                sect_end += sect_len_incr;
699
            sect_end += sect_len_incr;
700
            if (sect_end > ics->max_sfb) {
701
                av_log(ac->avccontext, AV_LOG_ERROR,
702
                       "Number of bands (%d) exceeds limit (%d).\n",
703
                       sect_end, ics->max_sfb);
704
                return -1;
705
            }
706
            for (; k < sect_end; k++) {
707
                band_type        [idx]   = sect_band_type;
708
                band_type_run_end[idx++] = sect_end;
709
            }
710
        }
711
    }
712
    return 0;
713
}
714

    
715
/**
716
 * Decode scalefactors; reference: table 4.47.
717
 *
718
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
719
 * @param   band_type           array of the used band type
720
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
721
 * @param   sf                  array of scalefactors or intensity stereo positions
722
 *
723
 * @return  Returns error status. 0 - OK, !0 - error
724
 */
725
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
726
                               unsigned int global_gain,
727
                               IndividualChannelStream *ics,
728
                               enum BandType band_type[120],
729
                               int band_type_run_end[120])
730
{
731
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
732
    int g, i, idx = 0;
733
    int offset[3] = { global_gain, global_gain - 90, 100 };
734
    int noise_flag = 1;
735
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
736
    for (g = 0; g < ics->num_window_groups; g++) {
737
        for (i = 0; i < ics->max_sfb;) {
738
            int run_end = band_type_run_end[idx];
739
            if (band_type[idx] == ZERO_BT) {
740
                for (; i < run_end; i++, idx++)
741
                    sf[idx] = 0.;
742
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
743
                for (; i < run_end; i++, idx++) {
744
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
745
                    if (offset[2] > 255U) {
746
                        av_log(ac->avccontext, AV_LOG_ERROR,
747
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
748
                        return -1;
749
                    }
750
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
751
                }
752
            } else if (band_type[idx] == NOISE_BT) {
753
                for (; i < run_end; i++, idx++) {
754
                    if (noise_flag-- > 0)
755
                        offset[1] += get_bits(gb, 9) - 256;
756
                    else
757
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
758
                    if (offset[1] > 255U) {
759
                        av_log(ac->avccontext, AV_LOG_ERROR,
760
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
761
                        return -1;
762
                    }
763
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
764
                }
765
            } else {
766
                for (; i < run_end; i++, idx++) {
767
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
768
                    if (offset[0] > 255U) {
769
                        av_log(ac->avccontext, AV_LOG_ERROR,
770
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
771
                        return -1;
772
                    }
773
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
774
                }
775
            }
776
        }
777
    }
778
    return 0;
779
}
780

    
781
/**
782
 * Decode pulse data; reference: table 4.7.
783
 */
784
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
785
                         const uint16_t *swb_offset, int num_swb)
786
{
787
    int i, pulse_swb;
788
    pulse->num_pulse = get_bits(gb, 2) + 1;
789
    pulse_swb        = get_bits(gb, 6);
790
    if (pulse_swb >= num_swb)
791
        return -1;
792
    pulse->pos[0]    = swb_offset[pulse_swb];
793
    pulse->pos[0]   += get_bits(gb, 5);
794
    if (pulse->pos[0] > 1023)
795
        return -1;
796
    pulse->amp[0]    = get_bits(gb, 4);
797
    for (i = 1; i < pulse->num_pulse; i++) {
798
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
799
        if (pulse->pos[i] > 1023)
800
            return -1;
801
        pulse->amp[i] = get_bits(gb, 4);
802
    }
803
    return 0;
804
}
805

    
806
/**
807
 * Decode Temporal Noise Shaping data; reference: table 4.48.
808
 *
809
 * @return  Returns error status. 0 - OK, !0 - error
810
 */
811
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
812
                      GetBitContext *gb, const IndividualChannelStream *ics)
813
{
814
    int w, filt, i, coef_len, coef_res, coef_compress;
815
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
816
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
817
    for (w = 0; w < ics->num_windows; w++) {
818
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
819
            coef_res = get_bits1(gb);
820

    
821
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
822
                int tmp2_idx;
823
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
824

    
825
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
826
                    av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.",
827
                           tns->order[w][filt], tns_max_order);
828
                    tns->order[w][filt] = 0;
829
                    return -1;
830
                }
831
                if (tns->order[w][filt]) {
832
                    tns->direction[w][filt] = get_bits1(gb);
833
                    coef_compress = get_bits1(gb);
834
                    coef_len = coef_res + 3 - coef_compress;
835
                    tmp2_idx = 2 * coef_compress + coef_res;
836

    
837
                    for (i = 0; i < tns->order[w][filt]; i++)
838
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
839
                }
840
            }
841
        }
842
    }
843
    return 0;
844
}
845

    
846
/**
847
 * Decode Mid/Side data; reference: table 4.54.
848
 *
849
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
850
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
851
 *                      [3] reserved for scalable AAC
852
 */
853
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
854
                                   int ms_present)
855
{
856
    int idx;
857
    if (ms_present == 1) {
858
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
859
            cpe->ms_mask[idx] = get_bits1(gb);
860
    } else if (ms_present == 2) {
861
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
862
    }
863
}
864

    
865
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
866
                           const float *scale)
867
{
868
    float s = *scale;
869
    *dst++ = v[idx    & 15] * s;
870
    *dst++ = v[idx>>4 & 15] * s;
871
    return dst;
872
}
873

    
874
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
875
                           const float *scale)
876
{
877
    float s = *scale;
878
    *dst++ = v[idx    & 3] * s;
879
    *dst++ = v[idx>>2 & 3] * s;
880
    *dst++ = v[idx>>4 & 3] * s;
881
    *dst++ = v[idx>>6 & 3] * s;
882
    return dst;
883
}
884

    
885
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
886
                            unsigned sign, const float *scale)
887
{
888
    union float754 s0, s1;
889

    
890
    s0.f = s1.f = *scale;
891
    s0.i ^= sign >> 1 << 31;
892
    s1.i ^= sign      << 31;
893

    
894
    *dst++ = v[idx    & 15] * s0.f;
895
    *dst++ = v[idx>>4 & 15] * s1.f;
896

    
897
    return dst;
898
}
899

    
900
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
901
                            unsigned sign, const float *scale)
902
{
903
    unsigned nz = idx >> 12;
904
    union float754 s = { .f = *scale };
905
    union float754 t;
906

    
907
    t.i = s.i ^ (sign & 1<<31);
908
    *dst++ = v[idx    & 3] * t.f;
909

    
910
    sign <<= nz & 1; nz >>= 1;
911
    t.i = s.i ^ (sign & 1<<31);
912
    *dst++ = v[idx>>2 & 3] * t.f;
913

    
914
    sign <<= nz & 1; nz >>= 1;
915
    t.i = s.i ^ (sign & 1<<31);
916
    *dst++ = v[idx>>4 & 3] * t.f;
917

    
918
    sign <<= nz & 1; nz >>= 1;
919
    t.i = s.i ^ (sign & 1<<31);
920
    *dst++ = v[idx>>6 & 3] * t.f;
921

    
922
    return dst;
923
}
924

    
925
/**
926
 * Decode spectral data; reference: table 4.50.
927
 * Dequantize and scale spectral data; reference: 4.6.3.3.
928
 *
929
 * @param   coef            array of dequantized, scaled spectral data
930
 * @param   sf              array of scalefactors or intensity stereo positions
931
 * @param   pulse_present   set if pulses are present
932
 * @param   pulse           pointer to pulse data struct
933
 * @param   band_type       array of the used band type
934
 *
935
 * @return  Returns error status. 0 - OK, !0 - error
936
 */
937
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
938
                                       GetBitContext *gb, const float sf[120],
939
                                       int pulse_present, const Pulse *pulse,
940
                                       const IndividualChannelStream *ics,
941
                                       enum BandType band_type[120])
942
{
943
    int i, k, g, idx = 0;
944
    const int c = 1024 / ics->num_windows;
945
    const uint16_t *offsets = ics->swb_offset;
946
    float *coef_base = coef;
947
    int err_idx;
948

    
949
    for (g = 0; g < ics->num_windows; g++)
950
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
951

    
952
    for (g = 0; g < ics->num_window_groups; g++) {
953
        unsigned g_len = ics->group_len[g];
954

    
955
        for (i = 0; i < ics->max_sfb; i++, idx++) {
956
            const unsigned cbt_m1 = band_type[idx] - 1;
957
            float *cfo = coef + offsets[i];
958
            int off_len = offsets[i + 1] - offsets[i];
959
            int group;
960

    
961
            if (cbt_m1 >= INTENSITY_BT2 - 1) {
962
                for (group = 0; group < g_len; group++, cfo+=128) {
963
                    memset(cfo, 0, off_len * sizeof(float));
964
                }
965
            } else if (cbt_m1 == NOISE_BT - 1) {
966
                for (group = 0; group < g_len; group++, cfo+=128) {
967
                    float scale;
968
                    float band_energy;
969

    
970
                    for (k = 0; k < off_len; k++) {
971
                        ac->random_state  = lcg_random(ac->random_state);
972
                        cfo[k] = ac->random_state;
973
                    }
974

    
975
                    band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
976
                    scale = sf[idx] / sqrtf(band_energy);
977
                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
978
                }
979
            } else {
980
                const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
981
                const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
982
                VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
983
                const int cb_size = ff_aac_spectral_sizes[cbt_m1];
984

    
985
                switch (cbt_m1 >> 1) {
986
                case 0:
987
                    for (group = 0; group < g_len; group++, cfo+=128) {
988
                        float *cf = cfo;
989
                        int len = off_len;
990

    
991
                        do {
992
                            const int index = get_vlc2(gb, vlc_tab, 8, 2);
993
                            unsigned cb_idx;
994

    
995
                            if (index >= cb_size) {
996
                                err_idx = index;
997
                                goto err_cb_overflow;
998
                            }
999

    
1000
                            cb_idx = cb_vector_idx[index];
1001
                            cf = VMUL4(cf, vq, cb_idx, sf + idx);
1002
                        } while (len -= 4);
1003
                    }
1004
                    break;
1005

    
1006
                case 1:
1007
                    for (group = 0; group < g_len; group++, cfo+=128) {
1008
                        float *cf = cfo;
1009
                        int len = off_len;
1010

    
1011
                        do {
1012
                            const int index = get_vlc2(gb, vlc_tab, 8, 2);
1013
                            unsigned nnz;
1014
                            unsigned cb_idx;
1015
                            uint32_t bits;
1016

    
1017
                            if (index >= cb_size) {
1018
                                err_idx = index;
1019
                                goto err_cb_overflow;
1020
                            }
1021

    
1022
                            cb_idx = cb_vector_idx[index];
1023
                            nnz = cb_idx >> 8 & 15;
1024
                            bits = get_bits(gb, nnz) << (32-nnz);
1025
                            cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1026
                        } while (len -= 4);
1027
                    }
1028
                    break;
1029

    
1030
                case 2:
1031
                    for (group = 0; group < g_len; group++, cfo+=128) {
1032
                        float *cf = cfo;
1033
                        int len = off_len;
1034

    
1035
                        do {
1036
                            const int index = get_vlc2(gb, vlc_tab, 8, 2);
1037
                            unsigned cb_idx;
1038

    
1039
                            if (index >= cb_size) {
1040
                                err_idx = index;
1041
                                goto err_cb_overflow;
1042
                            }
1043

    
1044
                            cb_idx = cb_vector_idx[index];
1045
                            cf = VMUL2(cf, vq, cb_idx, sf + idx);
1046
                        } while (len -= 2);
1047
                    }
1048
                    break;
1049

    
1050
                case 3:
1051
                case 4:
1052
                    for (group = 0; group < g_len; group++, cfo+=128) {
1053
                        float *cf = cfo;
1054
                        int len = off_len;
1055

    
1056
                        do {
1057
                            const int index = get_vlc2(gb, vlc_tab, 8, 2);
1058
                            unsigned nnz;
1059
                            unsigned cb_idx;
1060
                            unsigned sign;
1061

    
1062
                            if (index >= cb_size) {
1063
                                err_idx = index;
1064
                                goto err_cb_overflow;
1065
                            }
1066

    
1067
                            cb_idx = cb_vector_idx[index];
1068
                            nnz = cb_idx >> 8 & 15;
1069
                            sign = get_bits(gb, nnz) << (cb_idx >> 12);
1070
                            cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1071
                        } while (len -= 2);
1072
                    }
1073
                    break;
1074

    
1075
                default:
1076
                    for (group = 0; group < g_len; group++, cfo+=128) {
1077
                        float *cf = cfo;
1078
                        uint32_t *icf = (uint32_t *) cf;
1079
                        int len = off_len;
1080

    
1081
                        do {
1082
                            const int index = get_vlc2(gb, vlc_tab, 8, 2);
1083
                            unsigned nzt, nnz;
1084
                            unsigned cb_idx;
1085
                            uint32_t bits;
1086
                            int j;
1087

    
1088
                            if (!index) {
1089
                                *icf++ = 0;
1090
                                *icf++ = 0;
1091
                                continue;
1092
                            }
1093

    
1094
                            if (index >= cb_size) {
1095
                                err_idx = index;
1096
                                goto err_cb_overflow;
1097
                            }
1098

    
1099
                            cb_idx = cb_vector_idx[index];
1100
                            nnz = cb_idx >> 12;
1101
                            nzt = cb_idx >> 8;
1102
                            bits = get_bits(gb, nnz) << (32-nnz);
1103

    
1104
                            for (j = 0; j < 2; j++) {
1105
                                if (nzt & 1<<j) {
1106
                                    int n = 4;
1107
                                    /* The total length of escape_sequence must be < 22 bits according
1108
                                       to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1109
                                    while (get_bits1(gb) && n < 13) n++;
1110
                                    if (n == 13) {
1111
                                        av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1112
                                        return -1;
1113
                                    }
1114
                                    n = (1 << n) + get_bits(gb, n);
1115
                                    *icf++ = cbrt_tab[n] | (bits & 1<<31);
1116
                                    bits <<= 1;
1117
                                } else {
1118
                                    unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1119
                                    *icf++ = (bits & 1<<31) | v;
1120
                                    bits <<= !!v;
1121
                                }
1122
                                cb_idx >>= 4;
1123
                            }
1124
                        } while (len -= 2);
1125

    
1126
                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1127
                    }
1128
                }
1129
            }
1130
        }
1131
        coef += g_len << 7;
1132
    }
1133

    
1134
    if (pulse_present) {
1135
        idx = 0;
1136
        for (i = 0; i < pulse->num_pulse; i++) {
1137
            float co = coef_base[ pulse->pos[i] ];
1138
            while (offsets[idx + 1] <= pulse->pos[i])
1139
                idx++;
1140
            if (band_type[idx] != NOISE_BT && sf[idx]) {
1141
                float ico = -pulse->amp[i];
1142
                if (co) {
1143
                    co /= sf[idx];
1144
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1145
                }
1146
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1147
            }
1148
        }
1149
    }
1150
    return 0;
1151

    
1152
err_cb_overflow:
1153
    av_log(ac->avccontext, AV_LOG_ERROR,
1154
           "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1155
           band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1156
    return -1;
1157
}
1158

    
1159
static av_always_inline float flt16_round(float pf)
1160
{
1161
    union float754 tmp;
1162
    tmp.f = pf;
1163
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1164
    return tmp.f;
1165
}
1166

    
1167
static av_always_inline float flt16_even(float pf)
1168
{
1169
    union float754 tmp;
1170
    tmp.f = pf;
1171
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1172
    return tmp.f;
1173
}
1174

    
1175
static av_always_inline float flt16_trunc(float pf)
1176
{
1177
    union float754 pun;
1178
    pun.f = pf;
1179
    pun.i &= 0xFFFF0000U;
1180
    return pun.f;
1181
}
1182

    
1183
static void predict(AACContext *ac, PredictorState *ps, float *coef,
1184
                    int output_enable)
1185
{
1186
    const float a     = 0.953125; // 61.0 / 64
1187
    const float alpha = 0.90625;  // 29.0 / 32
1188
    float e0, e1;
1189
    float pv;
1190
    float k1, k2;
1191

    
1192
    k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1193
    k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1194

    
1195
    pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1196
    if (output_enable)
1197
        *coef += pv * ac->sf_scale;
1198

    
1199
    e0 = *coef / ac->sf_scale;
1200
    e1 = e0 - k1 * ps->r0;
1201

    
1202
    ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1203
    ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1204
    ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1205
    ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1206

    
1207
    ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1208
    ps->r0 = flt16_trunc(a * e0);
1209
}
1210

    
1211
/**
1212
 * Apply AAC-Main style frequency domain prediction.
1213
 */
1214
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1215
{
1216
    int sfb, k;
1217

    
1218
    if (!sce->ics.predictor_initialized) {
1219
        reset_all_predictors(sce->predictor_state);
1220
        sce->ics.predictor_initialized = 1;
1221
    }
1222

    
1223
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1224
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1225
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1226
                predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1227
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1228
            }
1229
        }
1230
        if (sce->ics.predictor_reset_group)
1231
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1232
    } else
1233
        reset_all_predictors(sce->predictor_state);
1234
}
1235

    
1236
/**
1237
 * Decode an individual_channel_stream payload; reference: table 4.44.
1238
 *
1239
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1240
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1241
 *
1242
 * @return  Returns error status. 0 - OK, !0 - error
1243
 */
1244
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1245
                      GetBitContext *gb, int common_window, int scale_flag)
1246
{
1247
    Pulse pulse;
1248
    TemporalNoiseShaping    *tns = &sce->tns;
1249
    IndividualChannelStream *ics = &sce->ics;
1250
    float *out = sce->coeffs;
1251
    int global_gain, pulse_present = 0;
1252

    
1253
    /* This assignment is to silence a GCC warning about the variable being used
1254
     * uninitialized when in fact it always is.
1255
     */
1256
    pulse.num_pulse = 0;
1257

    
1258
    global_gain = get_bits(gb, 8);
1259

    
1260
    if (!common_window && !scale_flag) {
1261
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1262
            return -1;
1263
    }
1264

    
1265
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1266
        return -1;
1267
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1268
        return -1;
1269

    
1270
    pulse_present = 0;
1271
    if (!scale_flag) {
1272
        if ((pulse_present = get_bits1(gb))) {
1273
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1274
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1275
                return -1;
1276
            }
1277
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1278
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1279
                return -1;
1280
            }
1281
        }
1282
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1283
            return -1;
1284
        if (get_bits1(gb)) {
1285
            av_log_missing_feature(ac->avccontext, "SSR", 1);
1286
            return -1;
1287
        }
1288
    }
1289

    
1290
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1291
        return -1;
1292

    
1293
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1294
        apply_prediction(ac, sce);
1295

    
1296
    return 0;
1297
}
1298

    
1299
/**
1300
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1301
 */
1302
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1303
{
1304
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1305
    float *ch0 = cpe->ch[0].coeffs;
1306
    float *ch1 = cpe->ch[1].coeffs;
1307
    int g, i, group, idx = 0;
1308
    const uint16_t *offsets = ics->swb_offset;
1309
    for (g = 0; g < ics->num_window_groups; g++) {
1310
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1311
            if (cpe->ms_mask[idx] &&
1312
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1313
                for (group = 0; group < ics->group_len[g]; group++) {
1314
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1315
                                              ch1 + group * 128 + offsets[i],
1316
                                              offsets[i+1] - offsets[i]);
1317
                }
1318
            }
1319
        }
1320
        ch0 += ics->group_len[g] * 128;
1321
        ch1 += ics->group_len[g] * 128;
1322
    }
1323
}
1324

    
1325
/**
1326
 * intensity stereo decoding; reference: 4.6.8.2.3
1327
 *
1328
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1329
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1330
 *                      [3] reserved for scalable AAC
1331
 */
1332
static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1333
{
1334
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1335
    SingleChannelElement         *sce1 = &cpe->ch[1];
1336
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1337
    const uint16_t *offsets = ics->swb_offset;
1338
    int g, group, i, k, idx = 0;
1339
    int c;
1340
    float scale;
1341
    for (g = 0; g < ics->num_window_groups; g++) {
1342
        for (i = 0; i < ics->max_sfb;) {
1343
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1344
                const int bt_run_end = sce1->band_type_run_end[idx];
1345
                for (; i < bt_run_end; i++, idx++) {
1346
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1347
                    if (ms_present)
1348
                        c *= 1 - 2 * cpe->ms_mask[idx];
1349
                    scale = c * sce1->sf[idx];
1350
                    for (group = 0; group < ics->group_len[g]; group++)
1351
                        for (k = offsets[i]; k < offsets[i + 1]; k++)
1352
                            coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1353
                }
1354
            } else {
1355
                int bt_run_end = sce1->band_type_run_end[idx];
1356
                idx += bt_run_end - i;
1357
                i    = bt_run_end;
1358
            }
1359
        }
1360
        coef0 += ics->group_len[g] * 128;
1361
        coef1 += ics->group_len[g] * 128;
1362
    }
1363
}
1364

    
1365
/**
1366
 * Decode a channel_pair_element; reference: table 4.4.
1367
 *
1368
 * @param   elem_id Identifies the instance of a syntax element.
1369
 *
1370
 * @return  Returns error status. 0 - OK, !0 - error
1371
 */
1372
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1373
{
1374
    int i, ret, common_window, ms_present = 0;
1375

    
1376
    common_window = get_bits1(gb);
1377
    if (common_window) {
1378
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1379
            return -1;
1380
        i = cpe->ch[1].ics.use_kb_window[0];
1381
        cpe->ch[1].ics = cpe->ch[0].ics;
1382
        cpe->ch[1].ics.use_kb_window[1] = i;
1383
        ms_present = get_bits(gb, 2);
1384
        if (ms_present == 3) {
1385
            av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1386
            return -1;
1387
        } else if (ms_present)
1388
            decode_mid_side_stereo(cpe, gb, ms_present);
1389
    }
1390
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1391
        return ret;
1392
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1393
        return ret;
1394

    
1395
    if (common_window) {
1396
        if (ms_present)
1397
            apply_mid_side_stereo(ac, cpe);
1398
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1399
            apply_prediction(ac, &cpe->ch[0]);
1400
            apply_prediction(ac, &cpe->ch[1]);
1401
        }
1402
    }
1403

    
1404
    apply_intensity_stereo(cpe, ms_present);
1405
    return 0;
1406
}
1407

    
1408
/**
1409
 * Decode coupling_channel_element; reference: table 4.8.
1410
 *
1411
 * @param   elem_id Identifies the instance of a syntax element.
1412
 *
1413
 * @return  Returns error status. 0 - OK, !0 - error
1414
 */
1415
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1416
{
1417
    int num_gain = 0;
1418
    int c, g, sfb, ret;
1419
    int sign;
1420
    float scale;
1421
    SingleChannelElement *sce = &che->ch[0];
1422
    ChannelCoupling     *coup = &che->coup;
1423

    
1424
    coup->coupling_point = 2 * get_bits1(gb);
1425
    coup->num_coupled = get_bits(gb, 3);
1426
    for (c = 0; c <= coup->num_coupled; c++) {
1427
        num_gain++;
1428
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1429
        coup->id_select[c] = get_bits(gb, 4);
1430
        if (coup->type[c] == TYPE_CPE) {
1431
            coup->ch_select[c] = get_bits(gb, 2);
1432
            if (coup->ch_select[c] == 3)
1433
                num_gain++;
1434
        } else
1435
            coup->ch_select[c] = 2;
1436
    }
1437
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1438

    
1439
    sign  = get_bits(gb, 1);
1440
    scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1441

    
1442
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1443
        return ret;
1444

    
1445
    for (c = 0; c < num_gain; c++) {
1446
        int idx  = 0;
1447
        int cge  = 1;
1448
        int gain = 0;
1449
        float gain_cache = 1.;
1450
        if (c) {
1451
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1452
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1453
            gain_cache = pow(scale, -gain);
1454
        }
1455
        if (coup->coupling_point == AFTER_IMDCT) {
1456
            coup->gain[c][0] = gain_cache;
1457
        } else {
1458
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1459
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1460
                    if (sce->band_type[idx] != ZERO_BT) {
1461
                        if (!cge) {
1462
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1463
                            if (t) {
1464
                                int s = 1;
1465
                                t = gain += t;
1466
                                if (sign) {
1467
                                    s  -= 2 * (t & 0x1);
1468
                                    t >>= 1;
1469
                                }
1470
                                gain_cache = pow(scale, -t) * s;
1471
                            }
1472
                        }
1473
                        coup->gain[c][idx] = gain_cache;
1474
                    }
1475
                }
1476
            }
1477
        }
1478
    }
1479
    return 0;
1480
}
1481

    
1482
/**
1483
 * Decode Spectral Band Replication extension data; reference: table 4.55.
1484
 *
1485
 * @param   crc flag indicating the presence of CRC checksum
1486
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1487
 *
1488
 * @return  Returns number of bytes consumed from the TYPE_FIL element.
1489
 */
1490
static int decode_sbr_extension(AACContext *ac, GetBitContext *gb,
1491
                                int crc, int cnt)
1492
{
1493
    // TODO : sbr_extension implementation
1494
    av_log_missing_feature(ac->avccontext, "SBR", 0);
1495
    skip_bits_long(gb, 8 * cnt - 4); // -4 due to reading extension type
1496
    return cnt;
1497
}
1498

    
1499
/**
1500
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1501
 *
1502
 * @return  Returns number of bytes consumed.
1503
 */
1504
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1505
                                         GetBitContext *gb)
1506
{
1507
    int i;
1508
    int num_excl_chan = 0;
1509

    
1510
    do {
1511
        for (i = 0; i < 7; i++)
1512
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1513
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1514

    
1515
    return num_excl_chan / 7;
1516
}
1517

    
1518
/**
1519
 * Decode dynamic range information; reference: table 4.52.
1520
 *
1521
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1522
 *
1523
 * @return  Returns number of bytes consumed.
1524
 */
1525
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1526
                                GetBitContext *gb, int cnt)
1527
{
1528
    int n             = 1;
1529
    int drc_num_bands = 1;
1530
    int i;
1531

    
1532
    /* pce_tag_present? */
1533
    if (get_bits1(gb)) {
1534
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1535
        skip_bits(gb, 4); // tag_reserved_bits
1536
        n++;
1537
    }
1538

    
1539
    /* excluded_chns_present? */
1540
    if (get_bits1(gb)) {
1541
        n += decode_drc_channel_exclusions(che_drc, gb);
1542
    }
1543

    
1544
    /* drc_bands_present? */
1545
    if (get_bits1(gb)) {
1546
        che_drc->band_incr            = get_bits(gb, 4);
1547
        che_drc->interpolation_scheme = get_bits(gb, 4);
1548
        n++;
1549
        drc_num_bands += che_drc->band_incr;
1550
        for (i = 0; i < drc_num_bands; i++) {
1551
            che_drc->band_top[i] = get_bits(gb, 8);
1552
            n++;
1553
        }
1554
    }
1555

    
1556
    /* prog_ref_level_present? */
1557
    if (get_bits1(gb)) {
1558
        che_drc->prog_ref_level = get_bits(gb, 7);
1559
        skip_bits1(gb); // prog_ref_level_reserved_bits
1560
        n++;
1561
    }
1562

    
1563
    for (i = 0; i < drc_num_bands; i++) {
1564
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1565
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1566
        n++;
1567
    }
1568

    
1569
    return n;
1570
}
1571

    
1572
/**
1573
 * Decode extension data (incomplete); reference: table 4.51.
1574
 *
1575
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1576
 *
1577
 * @return Returns number of bytes consumed
1578
 */
1579
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt)
1580
{
1581
    int crc_flag = 0;
1582
    int res = cnt;
1583
    switch (get_bits(gb, 4)) { // extension type
1584
    case EXT_SBR_DATA_CRC:
1585
        crc_flag++;
1586
    case EXT_SBR_DATA:
1587
        res = decode_sbr_extension(ac, gb, crc_flag, cnt);
1588
        break;
1589
    case EXT_DYNAMIC_RANGE:
1590
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1591
        break;
1592
    case EXT_FILL:
1593
    case EXT_FILL_DATA:
1594
    case EXT_DATA_ELEMENT:
1595
    default:
1596
        skip_bits_long(gb, 8 * cnt - 4);
1597
        break;
1598
    };
1599
    return res;
1600
}
1601

    
1602
/**
1603
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1604
 *
1605
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1606
 * @param   coef    spectral coefficients
1607
 */
1608
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1609
                      IndividualChannelStream *ics, int decode)
1610
{
1611
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1612
    int w, filt, m, i;
1613
    int bottom, top, order, start, end, size, inc;
1614
    float lpc[TNS_MAX_ORDER];
1615

    
1616
    for (w = 0; w < ics->num_windows; w++) {
1617
        bottom = ics->num_swb;
1618
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1619
            top    = bottom;
1620
            bottom = FFMAX(0, top - tns->length[w][filt]);
1621
            order  = tns->order[w][filt];
1622
            if (order == 0)
1623
                continue;
1624

    
1625
            // tns_decode_coef
1626
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1627

    
1628
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1629
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1630
            if ((size = end - start) <= 0)
1631
                continue;
1632
            if (tns->direction[w][filt]) {
1633
                inc = -1;
1634
                start = end - 1;
1635
            } else {
1636
                inc = 1;
1637
            }
1638
            start += w * 128;
1639

    
1640
            // ar filter
1641
            for (m = 0; m < size; m++, start += inc)
1642
                for (i = 1; i <= FFMIN(m, order); i++)
1643
                    coef[start] -= coef[start - i * inc] * lpc[i - 1];
1644
        }
1645
    }
1646
}
1647

    
1648
/**
1649
 * Conduct IMDCT and windowing.
1650
 */
1651
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1652
{
1653
    IndividualChannelStream *ics = &sce->ics;
1654
    float *in    = sce->coeffs;
1655
    float *out   = sce->ret;
1656
    float *saved = sce->saved;
1657
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1658
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1659
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1660
    float *buf  = ac->buf_mdct;
1661
    float *temp = ac->temp;
1662
    int i;
1663

    
1664
    // imdct
1665
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1666
        if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1667
            av_log(ac->avccontext, AV_LOG_WARNING,
1668
                   "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1669
                   "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1670
        for (i = 0; i < 1024; i += 128)
1671
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1672
    } else
1673
        ff_imdct_half(&ac->mdct, buf, in);
1674

    
1675
    /* window overlapping
1676
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1677
     * and long to short transitions are considered to be short to short
1678
     * transitions. This leaves just two cases (long to long and short to short)
1679
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1680
     */
1681
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1682
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1683
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, ac->add_bias, 512);
1684
    } else {
1685
        for (i = 0; i < 448; i++)
1686
            out[i] = saved[i] + ac->add_bias;
1687

    
1688
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1689
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, ac->add_bias, 64);
1690
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      ac->add_bias, 64);
1691
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      ac->add_bias, 64);
1692
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      ac->add_bias, 64);
1693
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      ac->add_bias, 64);
1694
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1695
        } else {
1696
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, ac->add_bias, 64);
1697
            for (i = 576; i < 1024; i++)
1698
                out[i] = buf[i-512] + ac->add_bias;
1699
        }
1700
    }
1701

    
1702
    // buffer update
1703
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1704
        for (i = 0; i < 64; i++)
1705
            saved[i] = temp[64 + i] - ac->add_bias;
1706
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1707
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1708
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1709
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1710
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1711
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1712
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1713
    } else { // LONG_STOP or ONLY_LONG
1714
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1715
    }
1716
}
1717

    
1718
/**
1719
 * Apply dependent channel coupling (applied before IMDCT).
1720
 *
1721
 * @param   index   index into coupling gain array
1722
 */
1723
static void apply_dependent_coupling(AACContext *ac,
1724
                                     SingleChannelElement *target,
1725
                                     ChannelElement *cce, int index)
1726
{
1727
    IndividualChannelStream *ics = &cce->ch[0].ics;
1728
    const uint16_t *offsets = ics->swb_offset;
1729
    float *dest = target->coeffs;
1730
    const float *src = cce->ch[0].coeffs;
1731
    int g, i, group, k, idx = 0;
1732
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1733
        av_log(ac->avccontext, AV_LOG_ERROR,
1734
               "Dependent coupling is not supported together with LTP\n");
1735
        return;
1736
    }
1737
    for (g = 0; g < ics->num_window_groups; g++) {
1738
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1739
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1740
                const float gain = cce->coup.gain[index][idx];
1741
                for (group = 0; group < ics->group_len[g]; group++) {
1742
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1743
                        // XXX dsputil-ize
1744
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1745
                    }
1746
                }
1747
            }
1748
        }
1749
        dest += ics->group_len[g] * 128;
1750
        src  += ics->group_len[g] * 128;
1751
    }
1752
}
1753

    
1754
/**
1755
 * Apply independent channel coupling (applied after IMDCT).
1756
 *
1757
 * @param   index   index into coupling gain array
1758
 */
1759
static void apply_independent_coupling(AACContext *ac,
1760
                                       SingleChannelElement *target,
1761
                                       ChannelElement *cce, int index)
1762
{
1763
    int i;
1764
    const float gain = cce->coup.gain[index][0];
1765
    const float bias = ac->add_bias;
1766
    const float *src = cce->ch[0].ret;
1767
    float *dest = target->ret;
1768

    
1769
    for (i = 0; i < 1024; i++)
1770
        dest[i] += gain * (src[i] - bias);
1771
}
1772

    
1773
/**
1774
 * channel coupling transformation interface
1775
 *
1776
 * @param   index   index into coupling gain array
1777
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1778
 */
1779
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1780
                                   enum RawDataBlockType type, int elem_id,
1781
                                   enum CouplingPoint coupling_point,
1782
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1783
{
1784
    int i, c;
1785

    
1786
    for (i = 0; i < MAX_ELEM_ID; i++) {
1787
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1788
        int index = 0;
1789

    
1790
        if (cce && cce->coup.coupling_point == coupling_point) {
1791
            ChannelCoupling *coup = &cce->coup;
1792

    
1793
            for (c = 0; c <= coup->num_coupled; c++) {
1794
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1795
                    if (coup->ch_select[c] != 1) {
1796
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1797
                        if (coup->ch_select[c] != 0)
1798
                            index++;
1799
                    }
1800
                    if (coup->ch_select[c] != 2)
1801
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1802
                } else
1803
                    index += 1 + (coup->ch_select[c] == 3);
1804
            }
1805
        }
1806
    }
1807
}
1808

    
1809
/**
1810
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1811
 */
1812
static void spectral_to_sample(AACContext *ac)
1813
{
1814
    int i, type;
1815
    for (type = 3; type >= 0; type--) {
1816
        for (i = 0; i < MAX_ELEM_ID; i++) {
1817
            ChannelElement *che = ac->che[type][i];
1818
            if (che) {
1819
                if (type <= TYPE_CPE)
1820
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1821
                if (che->ch[0].tns.present)
1822
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1823
                if (che->ch[1].tns.present)
1824
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1825
                if (type <= TYPE_CPE)
1826
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1827
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT)
1828
                    imdct_and_windowing(ac, &che->ch[0]);
1829
                if (type == TYPE_CPE)
1830
                    imdct_and_windowing(ac, &che->ch[1]);
1831
                if (type <= TYPE_CCE)
1832
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1833
            }
1834
        }
1835
    }
1836
}
1837

    
1838
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1839
{
1840
    int size;
1841
    AACADTSHeaderInfo hdr_info;
1842

    
1843
    size = ff_aac_parse_header(gb, &hdr_info);
1844
    if (size > 0) {
1845
        if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1846
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1847
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1848
            ac->m4ac.chan_config = hdr_info.chan_config;
1849
            if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1850
                return -7;
1851
            if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1852
                return -7;
1853
        } else if (ac->output_configured != OC_LOCKED) {
1854
            ac->output_configured = OC_NONE;
1855
        }
1856
        if (ac->output_configured != OC_LOCKED)
1857
            ac->m4ac.sbr = -1;
1858
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1859
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1860
        ac->m4ac.object_type     = hdr_info.object_type;
1861
        if (!ac->avccontext->sample_rate)
1862
            ac->avccontext->sample_rate = hdr_info.sample_rate;
1863
        if (hdr_info.num_aac_frames == 1) {
1864
            if (!hdr_info.crc_absent)
1865
                skip_bits(gb, 16);
1866
        } else {
1867
            av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1868
            return -1;
1869
        }
1870
    }
1871
    return size;
1872
}
1873

    
1874
static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1875
                            int *data_size, AVPacket *avpkt)
1876
{
1877
    const uint8_t *buf = avpkt->data;
1878
    int buf_size = avpkt->size;
1879
    AACContext *ac = avccontext->priv_data;
1880
    ChannelElement *che = NULL;
1881
    GetBitContext gb;
1882
    enum RawDataBlockType elem_type;
1883
    int err, elem_id, data_size_tmp;
1884

    
1885
    init_get_bits(&gb, buf, buf_size * 8);
1886

    
1887
    if (show_bits(&gb, 12) == 0xfff) {
1888
        if (parse_adts_frame_header(ac, &gb) < 0) {
1889
            av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1890
            return -1;
1891
        }
1892
        if (ac->m4ac.sampling_index > 12) {
1893
            av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1894
            return -1;
1895
        }
1896
    }
1897

    
1898
    // parse
1899
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1900
        elem_id = get_bits(&gb, 4);
1901

    
1902
        if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1903
            av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1904
            return -1;
1905
        }
1906

    
1907
        switch (elem_type) {
1908

    
1909
        case TYPE_SCE:
1910
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1911
            break;
1912

    
1913
        case TYPE_CPE:
1914
            err = decode_cpe(ac, &gb, che);
1915
            break;
1916

    
1917
        case TYPE_CCE:
1918
            err = decode_cce(ac, &gb, che);
1919
            break;
1920

    
1921
        case TYPE_LFE:
1922
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1923
            break;
1924

    
1925
        case TYPE_DSE:
1926
            skip_data_stream_element(&gb);
1927
            err = 0;
1928
            break;
1929

    
1930
        case TYPE_PCE: {
1931
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1932
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1933
            if ((err = decode_pce(ac, new_che_pos, &gb)))
1934
                break;
1935
            if (ac->output_configured > OC_TRIAL_PCE)
1936
                av_log(avccontext, AV_LOG_ERROR,
1937
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1938
            else
1939
                err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1940
            break;
1941
        }
1942

    
1943
        case TYPE_FIL:
1944
            if (elem_id == 15)
1945
                elem_id += get_bits(&gb, 8) - 1;
1946
            while (elem_id > 0)
1947
                elem_id -= decode_extension_payload(ac, &gb, elem_id);
1948
            err = 0; /* FIXME */
1949
            break;
1950

    
1951
        default:
1952
            err = -1; /* should not happen, but keeps compiler happy */
1953
            break;
1954
        }
1955

    
1956
        if (err)
1957
            return err;
1958
    }
1959

    
1960
    spectral_to_sample(ac);
1961

    
1962
    if (!ac->is_saved) {
1963
        ac->is_saved = 1;
1964
        *data_size = 0;
1965
        return buf_size;
1966
    }
1967

    
1968
    data_size_tmp = 1024 * avccontext->channels * sizeof(int16_t);
1969
    if (*data_size < data_size_tmp) {
1970
        av_log(avccontext, AV_LOG_ERROR,
1971
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1972
               *data_size, data_size_tmp);
1973
        return -1;
1974
    }
1975
    *data_size = data_size_tmp;
1976

    
1977
    ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, 1024, avccontext->channels);
1978

    
1979
    if (ac->output_configured)
1980
        ac->output_configured = OC_LOCKED;
1981

    
1982
    return buf_size;
1983
}
1984

    
1985
static av_cold int aac_decode_close(AVCodecContext *avccontext)
1986
{
1987
    AACContext *ac = avccontext->priv_data;
1988
    int i, type;
1989

    
1990
    for (i = 0; i < MAX_ELEM_ID; i++) {
1991
        for (type = 0; type < 4; type++)
1992
            av_freep(&ac->che[type][i]);
1993
    }
1994

    
1995
    ff_mdct_end(&ac->mdct);
1996
    ff_mdct_end(&ac->mdct_small);
1997
    return 0;
1998
}
1999

    
2000
AVCodec aac_decoder = {
2001
    "aac",
2002
    CODEC_TYPE_AUDIO,
2003
    CODEC_ID_AAC,
2004
    sizeof(AACContext),
2005
    aac_decode_init,
2006
    NULL,
2007
    aac_decode_close,
2008
    aac_decode_frame,
2009
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2010
    .sample_fmts = (const enum SampleFormat[]) {
2011
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2012
    },
2013
    .channel_layouts = aac_channel_layout,
2014
};