Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aac.c @ f8a93a20

History | View | Annotate | Download (74 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
 * Y                    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 "fft.h"
84
#include "lpc.h"
85

    
86
#include "aac.h"
87
#include "aactab.h"
88
#include "aacdectab.h"
89
#include "sbr.h"
90
#include "aacsbr.h"
91
#include "mpeg4audio.h"
92
#include "aac_parser.h"
93

    
94
#include <assert.h>
95
#include <errno.h>
96
#include <math.h>
97
#include <string.h>
98

    
99
#if ARCH_ARM
100
#   include "arm/aac.h"
101
#endif
102

    
103
union float754 {
104
    float f;
105
    uint32_t i;
106
};
107

    
108
static VLC vlc_scalefactors;
109
static VLC vlc_spectral[11];
110

    
111
static uint32_t cbrt_tab[1<<13];
112

    
113
static const char overread_err[] = "Input buffer exhausted before END element found\n";
114

    
115
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
116
{
117
    if (ac->tag_che_map[type][elem_id]) {
118
        return ac->tag_che_map[type][elem_id];
119
    }
120
    if (ac->tags_mapped >= tags_per_config[ac->m4ac.chan_config]) {
121
        return NULL;
122
    }
123
    switch (ac->m4ac.chan_config) {
124
    case 7:
125
        if (ac->tags_mapped == 3 && type == TYPE_CPE) {
126
            ac->tags_mapped++;
127
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
128
        }
129
    case 6:
130
        /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
131
           instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
132
           encountered such a stream, transfer the LFE[0] element to SCE[1] */
133
        if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
134
            ac->tags_mapped++;
135
            return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
136
        }
137
    case 5:
138
        if (ac->tags_mapped == 2 && type == TYPE_CPE) {
139
            ac->tags_mapped++;
140
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
141
        }
142
    case 4:
143
        if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
144
            ac->tags_mapped++;
145
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
146
        }
147
    case 3:
148
    case 2:
149
        if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
150
            ac->tags_mapped++;
151
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
152
        } else if (ac->m4ac.chan_config == 2) {
153
            return NULL;
154
        }
155
    case 1:
156
        if (!ac->tags_mapped && type == TYPE_SCE) {
157
            ac->tags_mapped++;
158
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
159
        }
160
    default:
161
        return NULL;
162
    }
163
}
164

    
165
/**
166
 * Check for the channel element in the current channel position configuration.
167
 * If it exists, make sure the appropriate element is allocated and map the
168
 * channel order to match the internal FFmpeg channel layout.
169
 *
170
 * @param   che_pos current channel position configuration
171
 * @param   type channel element type
172
 * @param   id channel element id
173
 * @param   channels count of the number of channels in the configuration
174
 *
175
 * @return  Returns error status. 0 - OK, !0 - error
176
 */
177
static av_cold int che_configure(AACContext *ac,
178
                         enum ChannelPosition che_pos[4][MAX_ELEM_ID],
179
                         int type, int id,
180
                         int *channels)
181
{
182
    if (che_pos[type][id]) {
183
        if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
184
            return AVERROR(ENOMEM);
185
        ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
186
        if (type != TYPE_CCE) {
187
            ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
188
            if (type == TYPE_CPE) {
189
                ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
190
            }
191
        }
192
    } else {
193
        if (ac->che[type][id])
194
            ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
195
        av_freep(&ac->che[type][id]);
196
    }
197
    return 0;
198
}
199

    
200
/**
201
 * Configure output channel order based on the current program configuration element.
202
 *
203
 * @param   che_pos current channel position configuration
204
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
205
 *
206
 * @return  Returns error status. 0 - OK, !0 - error
207
 */
208
static av_cold int output_configure(AACContext *ac,
209
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
210
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
211
                            int channel_config, enum OCStatus oc_type)
212
{
213
    AVCodecContext *avctx = ac->avccontext;
214
    int i, type, channels = 0, ret;
215

    
216
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
217

    
218
    if (channel_config) {
219
        for (i = 0; i < tags_per_config[channel_config]; i++) {
220
            if ((ret = che_configure(ac, che_pos,
221
                                     aac_channel_layout_map[channel_config - 1][i][0],
222
                                     aac_channel_layout_map[channel_config - 1][i][1],
223
                                     &channels)))
224
                return ret;
225
        }
226

    
227
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
228
        ac->tags_mapped = 0;
229

    
230
        avctx->channel_layout = aac_channel_layout[channel_config - 1];
231
    } else {
232
        /* Allocate or free elements depending on if they are in the
233
         * current program configuration.
234
         *
235
         * Set up default 1:1 output mapping.
236
         *
237
         * For a 5.1 stream the output order will be:
238
         *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
239
         */
240

    
241
        for (i = 0; i < MAX_ELEM_ID; i++) {
242
            for (type = 0; type < 4; type++) {
243
                if ((ret = che_configure(ac, che_pos, type, i, &channels)))
244
                    return ret;
245
            }
246
        }
247

    
248
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
249
        ac->tags_mapped = 4 * MAX_ELEM_ID;
250

    
251
        avctx->channel_layout = 0;
252
    }
253

    
254
    avctx->channels = channels;
255

    
256
    ac->output_configured = oc_type;
257

    
258
    return 0;
259
}
260

    
261
/**
262
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
263
 *
264
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
265
 * @param sce_map mono (Single Channel Element) map
266
 * @param type speaker type/position for these channels
267
 */
268
static void decode_channel_map(enum ChannelPosition *cpe_map,
269
                               enum ChannelPosition *sce_map,
270
                               enum ChannelPosition type,
271
                               GetBitContext *gb, int n)
272
{
273
    while (n--) {
274
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
275
        map[get_bits(gb, 4)] = type;
276
    }
277
}
278

    
279
/**
280
 * Decode program configuration element; reference: table 4.2.
281
 *
282
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
283
 *
284
 * @return  Returns error status. 0 - OK, !0 - error
285
 */
286
static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
287
                      GetBitContext *gb)
288
{
289
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
290
    int comment_len;
291

    
292
    skip_bits(gb, 2);  // object_type
293

    
294
    sampling_index = get_bits(gb, 4);
295
    if (ac->m4ac.sampling_index != sampling_index)
296
        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");
297

    
298
    num_front       = get_bits(gb, 4);
299
    num_side        = get_bits(gb, 4);
300
    num_back        = get_bits(gb, 4);
301
    num_lfe         = get_bits(gb, 2);
302
    num_assoc_data  = get_bits(gb, 3);
303
    num_cc          = get_bits(gb, 4);
304

    
305
    if (get_bits1(gb))
306
        skip_bits(gb, 4); // mono_mixdown_tag
307
    if (get_bits1(gb))
308
        skip_bits(gb, 4); // stereo_mixdown_tag
309

    
310
    if (get_bits1(gb))
311
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
312

    
313
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
314
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
315
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
316
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
317

    
318
    skip_bits_long(gb, 4 * num_assoc_data);
319

    
320
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
321

    
322
    align_get_bits(gb);
323

    
324
    /* comment field, first byte is length */
325
    comment_len = get_bits(gb, 8) * 8;
326
    if (get_bits_left(gb) < comment_len) {
327
        av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
328
        return -1;
329
    }
330
    skip_bits_long(gb, comment_len);
331
    return 0;
332
}
333

    
334
/**
335
 * Set up channel positions based on a default channel configuration
336
 * as specified in table 1.17.
337
 *
338
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
339
 *
340
 * @return  Returns error status. 0 - OK, !0 - error
341
 */
342
static av_cold int set_default_channel_config(AACContext *ac,
343
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
344
                                      int channel_config)
345
{
346
    if (channel_config < 1 || channel_config > 7) {
347
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
348
               channel_config);
349
        return -1;
350
    }
351

    
352
    /* default channel configurations:
353
     *
354
     * 1ch : front center (mono)
355
     * 2ch : L + R (stereo)
356
     * 3ch : front center + L + R
357
     * 4ch : front center + L + R + back center
358
     * 5ch : front center + L + R + back stereo
359
     * 6ch : front center + L + R + back stereo + LFE
360
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
361
     */
362

    
363
    if (channel_config != 2)
364
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
365
    if (channel_config > 1)
366
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
367
    if (channel_config == 4)
368
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
369
    if (channel_config > 4)
370
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
371
        = AAC_CHANNEL_BACK;  // back stereo
372
    if (channel_config > 5)
373
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
374
    if (channel_config == 7)
375
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
376

    
377
    return 0;
378
}
379

    
380
/**
381
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
382
 *
383
 * @return  Returns error status. 0 - OK, !0 - error
384
 */
385
static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
386
                                     int channel_config)
387
{
388
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
389
    int extension_flag, ret;
390

    
391
    if (get_bits1(gb)) { // frameLengthFlag
392
        av_log_missing_feature(ac->avccontext, "960/120 MDCT window is", 1);
393
        return -1;
394
    }
395

    
396
    if (get_bits1(gb))       // dependsOnCoreCoder
397
        skip_bits(gb, 14);   // coreCoderDelay
398
    extension_flag = get_bits1(gb);
399

    
400
    if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
401
        ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
402
        skip_bits(gb, 3);     // layerNr
403

    
404
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
405
    if (channel_config == 0) {
406
        skip_bits(gb, 4);  // element_instance_tag
407
        if ((ret = decode_pce(ac, new_che_pos, gb)))
408
            return ret;
409
    } else {
410
        if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
411
            return ret;
412
    }
413
    if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
414
        return ret;
415

    
416
    if (extension_flag) {
417
        switch (ac->m4ac.object_type) {
418
        case AOT_ER_BSAC:
419
            skip_bits(gb, 5);    // numOfSubFrame
420
            skip_bits(gb, 11);   // layer_length
421
            break;
422
        case AOT_ER_AAC_LC:
423
        case AOT_ER_AAC_LTP:
424
        case AOT_ER_AAC_SCALABLE:
425
        case AOT_ER_AAC_LD:
426
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
427
                                    * aacScalefactorDataResilienceFlag
428
                                    * aacSpectralDataResilienceFlag
429
                                    */
430
            break;
431
        }
432
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
433
    }
434
    return 0;
435
}
436

    
437
/**
438
 * Decode audio specific configuration; reference: table 1.13.
439
 *
440
 * @param   data        pointer to AVCodecContext extradata
441
 * @param   data_size   size of AVCCodecContext extradata
442
 *
443
 * @return  Returns error status. 0 - OK, !0 - error
444
 */
445
static int decode_audio_specific_config(AACContext *ac, void *data,
446
                                        int data_size)
447
{
448
    GetBitContext gb;
449
    int i;
450

    
451
    init_get_bits(&gb, data, data_size * 8);
452

    
453
    if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
454
        return -1;
455
    if (ac->m4ac.sampling_index > 12) {
456
        av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
457
        return -1;
458
    }
459

    
460
    skip_bits_long(&gb, i);
461

    
462
    switch (ac->m4ac.object_type) {
463
    case AOT_AAC_MAIN:
464
    case AOT_AAC_LC:
465
        if (decode_ga_specific_config(ac, &gb, ac->m4ac.chan_config))
466
            return -1;
467
        break;
468
    default:
469
        av_log(ac->avccontext, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
470
               ac->m4ac.sbr == 1? "SBR+" : "", ac->m4ac.object_type);
471
        return -1;
472
    }
473
    return 0;
474
}
475

    
476
/**
477
 * linear congruential pseudorandom number generator
478
 *
479
 * @param   previous_val    pointer to the current state of the generator
480
 *
481
 * @return  Returns a 32-bit pseudorandom integer
482
 */
483
static av_always_inline int lcg_random(int previous_val)
484
{
485
    return previous_val * 1664525 + 1013904223;
486
}
487

    
488
static av_always_inline void reset_predict_state(PredictorState *ps)
489
{
490
    ps->r0   = 0.0f;
491
    ps->r1   = 0.0f;
492
    ps->cor0 = 0.0f;
493
    ps->cor1 = 0.0f;
494
    ps->var0 = 1.0f;
495
    ps->var1 = 1.0f;
496
}
497

    
498
static void reset_all_predictors(PredictorState *ps)
499
{
500
    int i;
501
    for (i = 0; i < MAX_PREDICTORS; i++)
502
        reset_predict_state(&ps[i]);
503
}
504

    
505
static void reset_predictor_group(PredictorState *ps, int group_num)
506
{
507
    int i;
508
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
509
        reset_predict_state(&ps[i]);
510
}
511

    
512
static av_cold int aac_decode_init(AVCodecContext *avccontext)
513
{
514
    AACContext *ac = avccontext->priv_data;
515
    int i;
516

    
517
    ac->avccontext = avccontext;
518
    ac->m4ac.sample_rate = avccontext->sample_rate;
519

    
520
    if (avccontext->extradata_size > 0) {
521
        if (decode_audio_specific_config(ac, avccontext->extradata, avccontext->extradata_size))
522
            return -1;
523
    }
524

    
525
    avccontext->sample_fmt = SAMPLE_FMT_S16;
526

    
527
    AAC_INIT_VLC_STATIC( 0, 304);
528
    AAC_INIT_VLC_STATIC( 1, 270);
529
    AAC_INIT_VLC_STATIC( 2, 550);
530
    AAC_INIT_VLC_STATIC( 3, 300);
531
    AAC_INIT_VLC_STATIC( 4, 328);
532
    AAC_INIT_VLC_STATIC( 5, 294);
533
    AAC_INIT_VLC_STATIC( 6, 306);
534
    AAC_INIT_VLC_STATIC( 7, 268);
535
    AAC_INIT_VLC_STATIC( 8, 510);
536
    AAC_INIT_VLC_STATIC( 9, 366);
537
    AAC_INIT_VLC_STATIC(10, 462);
538

    
539
    ff_aac_sbr_init();
540

    
541
    dsputil_init(&ac->dsp, avccontext);
542

    
543
    ac->random_state = 0x1f2e3d4c;
544

    
545
    // -1024 - Compensate wrong IMDCT method.
546
    // 32768 - Required to scale values to the correct range for the bias method
547
    //         for float to int16 conversion.
548

    
549
    if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
550
        ac->add_bias  = 385.0f;
551
        ac->sf_scale  = 1. / (-1024. * 32768.);
552
        ac->sf_offset = 0;
553
    } else {
554
        ac->add_bias  = 0.0f;
555
        ac->sf_scale  = 1. / -1024.;
556
        ac->sf_offset = 60;
557
    }
558

    
559
#if !CONFIG_HARDCODED_TABLES
560
    for (i = 0; i < 428; i++)
561
        ff_aac_pow2sf_tab[i] = pow(2, (i - 200) / 4.);
562
#endif /* CONFIG_HARDCODED_TABLES */
563

    
564
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
565
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
566
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
567
                    352);
568

    
569
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
570
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
571
    // window initialization
572
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
573
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
574
    ff_init_ff_sine_windows(10);
575
    ff_init_ff_sine_windows( 7);
576

    
577
    if (!cbrt_tab[(1<<13) - 1]) {
578
        for (i = 0; i < 1<<13; i++) {
579
            union float754 f;
580
            f.f = cbrtf(i) * i;
581
            cbrt_tab[i] = f.i;
582
        }
583
    }
584

    
585
    return 0;
586
}
587

    
588
/**
589
 * Skip data_stream_element; reference: table 4.10.
590
 */
591
static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
592
{
593
    int byte_align = get_bits1(gb);
594
    int count = get_bits(gb, 8);
595
    if (count == 255)
596
        count += get_bits(gb, 8);
597
    if (byte_align)
598
        align_get_bits(gb);
599

    
600
    if (get_bits_left(gb) < 8 * count) {
601
        av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
602
        return -1;
603
    }
604
    skip_bits_long(gb, 8 * count);
605
    return 0;
606
}
607

    
608
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
609
                             GetBitContext *gb)
610
{
611
    int sfb;
612
    if (get_bits1(gb)) {
613
        ics->predictor_reset_group = get_bits(gb, 5);
614
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
615
            av_log(ac->avccontext, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
616
            return -1;
617
        }
618
    }
619
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
620
        ics->prediction_used[sfb] = get_bits1(gb);
621
    }
622
    return 0;
623
}
624

    
625
/**
626
 * Decode Individual Channel Stream info; reference: table 4.6.
627
 *
628
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
629
 */
630
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
631
                           GetBitContext *gb, int common_window)
632
{
633
    if (get_bits1(gb)) {
634
        av_log(ac->avccontext, AV_LOG_ERROR, "Reserved bit set.\n");
635
        memset(ics, 0, sizeof(IndividualChannelStream));
636
        return -1;
637
    }
638
    ics->window_sequence[1] = ics->window_sequence[0];
639
    ics->window_sequence[0] = get_bits(gb, 2);
640
    ics->use_kb_window[1]   = ics->use_kb_window[0];
641
    ics->use_kb_window[0]   = get_bits1(gb);
642
    ics->num_window_groups  = 1;
643
    ics->group_len[0]       = 1;
644
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
645
        int i;
646
        ics->max_sfb = get_bits(gb, 4);
647
        for (i = 0; i < 7; i++) {
648
            if (get_bits1(gb)) {
649
                ics->group_len[ics->num_window_groups - 1]++;
650
            } else {
651
                ics->num_window_groups++;
652
                ics->group_len[ics->num_window_groups - 1] = 1;
653
            }
654
        }
655
        ics->num_windows       = 8;
656
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
657
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
658
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
659
        ics->predictor_present = 0;
660
    } else {
661
        ics->max_sfb               = get_bits(gb, 6);
662
        ics->num_windows           = 1;
663
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
664
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
665
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
666
        ics->predictor_present     = get_bits1(gb);
667
        ics->predictor_reset_group = 0;
668
        if (ics->predictor_present) {
669
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
670
                if (decode_prediction(ac, ics, gb)) {
671
                    memset(ics, 0, sizeof(IndividualChannelStream));
672
                    return -1;
673
                }
674
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
675
                av_log(ac->avccontext, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
676
                memset(ics, 0, sizeof(IndividualChannelStream));
677
                return -1;
678
            } else {
679
                av_log_missing_feature(ac->avccontext, "Predictor bit set but LTP is", 1);
680
                memset(ics, 0, sizeof(IndividualChannelStream));
681
                return -1;
682
            }
683
        }
684
    }
685

    
686
    if (ics->max_sfb > ics->num_swb) {
687
        av_log(ac->avccontext, AV_LOG_ERROR,
688
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
689
               ics->max_sfb, ics->num_swb);
690
        memset(ics, 0, sizeof(IndividualChannelStream));
691
        return -1;
692
    }
693

    
694
    return 0;
695
}
696

    
697
/**
698
 * Decode band types (section_data payload); reference: table 4.46.
699
 *
700
 * @param   band_type           array of the used band type
701
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
702
 *
703
 * @return  Returns error status. 0 - OK, !0 - error
704
 */
705
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
706
                             int band_type_run_end[120], GetBitContext *gb,
707
                             IndividualChannelStream *ics)
708
{
709
    int g, idx = 0;
710
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
711
    for (g = 0; g < ics->num_window_groups; g++) {
712
        int k = 0;
713
        while (k < ics->max_sfb) {
714
            uint8_t sect_end = k;
715
            int sect_len_incr;
716
            int sect_band_type = get_bits(gb, 4);
717
            if (sect_band_type == 12) {
718
                av_log(ac->avccontext, AV_LOG_ERROR, "invalid band type\n");
719
                return -1;
720
            }
721
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
722
                sect_end += sect_len_incr;
723
            sect_end += sect_len_incr;
724
            if (get_bits_left(gb) < 0) {
725
                av_log(ac->avccontext, AV_LOG_ERROR, overread_err);
726
                return -1;
727
            }
728
            if (sect_end > ics->max_sfb) {
729
                av_log(ac->avccontext, AV_LOG_ERROR,
730
                       "Number of bands (%d) exceeds limit (%d).\n",
731
                       sect_end, ics->max_sfb);
732
                return -1;
733
            }
734
            for (; k < sect_end; k++) {
735
                band_type        [idx]   = sect_band_type;
736
                band_type_run_end[idx++] = sect_end;
737
            }
738
        }
739
    }
740
    return 0;
741
}
742

    
743
/**
744
 * Decode scalefactors; reference: table 4.47.
745
 *
746
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
747
 * @param   band_type           array of the used band type
748
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
749
 * @param   sf                  array of scalefactors or intensity stereo positions
750
 *
751
 * @return  Returns error status. 0 - OK, !0 - error
752
 */
753
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
754
                               unsigned int global_gain,
755
                               IndividualChannelStream *ics,
756
                               enum BandType band_type[120],
757
                               int band_type_run_end[120])
758
{
759
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
760
    int g, i, idx = 0;
761
    int offset[3] = { global_gain, global_gain - 90, 100 };
762
    int noise_flag = 1;
763
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
764
    for (g = 0; g < ics->num_window_groups; g++) {
765
        for (i = 0; i < ics->max_sfb;) {
766
            int run_end = band_type_run_end[idx];
767
            if (band_type[idx] == ZERO_BT) {
768
                for (; i < run_end; i++, idx++)
769
                    sf[idx] = 0.;
770
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
771
                for (; i < run_end; i++, idx++) {
772
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
773
                    if (offset[2] > 255U) {
774
                        av_log(ac->avccontext, AV_LOG_ERROR,
775
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
776
                        return -1;
777
                    }
778
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
779
                }
780
            } else if (band_type[idx] == NOISE_BT) {
781
                for (; i < run_end; i++, idx++) {
782
                    if (noise_flag-- > 0)
783
                        offset[1] += get_bits(gb, 9) - 256;
784
                    else
785
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
786
                    if (offset[1] > 255U) {
787
                        av_log(ac->avccontext, AV_LOG_ERROR,
788
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
789
                        return -1;
790
                    }
791
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
792
                }
793
            } else {
794
                for (; i < run_end; i++, idx++) {
795
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
796
                    if (offset[0] > 255U) {
797
                        av_log(ac->avccontext, AV_LOG_ERROR,
798
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
799
                        return -1;
800
                    }
801
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
802
                }
803
            }
804
        }
805
    }
806
    return 0;
807
}
808

    
809
/**
810
 * Decode pulse data; reference: table 4.7.
811
 */
812
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
813
                         const uint16_t *swb_offset, int num_swb)
814
{
815
    int i, pulse_swb;
816
    pulse->num_pulse = get_bits(gb, 2) + 1;
817
    pulse_swb        = get_bits(gb, 6);
818
    if (pulse_swb >= num_swb)
819
        return -1;
820
    pulse->pos[0]    = swb_offset[pulse_swb];
821
    pulse->pos[0]   += get_bits(gb, 5);
822
    if (pulse->pos[0] > 1023)
823
        return -1;
824
    pulse->amp[0]    = get_bits(gb, 4);
825
    for (i = 1; i < pulse->num_pulse; i++) {
826
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
827
        if (pulse->pos[i] > 1023)
828
            return -1;
829
        pulse->amp[i] = get_bits(gb, 4);
830
    }
831
    return 0;
832
}
833

    
834
/**
835
 * Decode Temporal Noise Shaping data; reference: table 4.48.
836
 *
837
 * @return  Returns error status. 0 - OK, !0 - error
838
 */
839
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
840
                      GetBitContext *gb, const IndividualChannelStream *ics)
841
{
842
    int w, filt, i, coef_len, coef_res, coef_compress;
843
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
844
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
845
    for (w = 0; w < ics->num_windows; w++) {
846
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
847
            coef_res = get_bits1(gb);
848

    
849
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
850
                int tmp2_idx;
851
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
852

    
853
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
854
                    av_log(ac->avccontext, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
855
                           tns->order[w][filt], tns_max_order);
856
                    tns->order[w][filt] = 0;
857
                    return -1;
858
                }
859
                if (tns->order[w][filt]) {
860
                    tns->direction[w][filt] = get_bits1(gb);
861
                    coef_compress = get_bits1(gb);
862
                    coef_len = coef_res + 3 - coef_compress;
863
                    tmp2_idx = 2 * coef_compress + coef_res;
864

    
865
                    for (i = 0; i < tns->order[w][filt]; i++)
866
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
867
                }
868
            }
869
        }
870
    }
871
    return 0;
872
}
873

    
874
/**
875
 * Decode Mid/Side data; reference: table 4.54.
876
 *
877
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
878
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
879
 *                      [3] reserved for scalable AAC
880
 */
881
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
882
                                   int ms_present)
883
{
884
    int idx;
885
    if (ms_present == 1) {
886
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
887
            cpe->ms_mask[idx] = get_bits1(gb);
888
    } else if (ms_present == 2) {
889
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
890
    }
891
}
892

    
893
#ifndef VMUL2
894
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
895
                           const float *scale)
896
{
897
    float s = *scale;
898
    *dst++ = v[idx    & 15] * s;
899
    *dst++ = v[idx>>4 & 15] * s;
900
    return dst;
901
}
902
#endif
903

    
904
#ifndef VMUL4
905
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
906
                           const float *scale)
907
{
908
    float s = *scale;
909
    *dst++ = v[idx    & 3] * s;
910
    *dst++ = v[idx>>2 & 3] * s;
911
    *dst++ = v[idx>>4 & 3] * s;
912
    *dst++ = v[idx>>6 & 3] * s;
913
    return dst;
914
}
915
#endif
916

    
917
#ifndef VMUL2S
918
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
919
                            unsigned sign, const float *scale)
920
{
921
    union float754 s0, s1;
922

    
923
    s0.f = s1.f = *scale;
924
    s0.i ^= sign >> 1 << 31;
925
    s1.i ^= sign      << 31;
926

    
927
    *dst++ = v[idx    & 15] * s0.f;
928
    *dst++ = v[idx>>4 & 15] * s1.f;
929

    
930
    return dst;
931
}
932
#endif
933

    
934
#ifndef VMUL4S
935
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
936
                            unsigned sign, const float *scale)
937
{
938
    unsigned nz = idx >> 12;
939
    union float754 s = { .f = *scale };
940
    union float754 t;
941

    
942
    t.i = s.i ^ (sign & 1<<31);
943
    *dst++ = v[idx    & 3] * t.f;
944

    
945
    sign <<= nz & 1; nz >>= 1;
946
    t.i = s.i ^ (sign & 1<<31);
947
    *dst++ = v[idx>>2 & 3] * t.f;
948

    
949
    sign <<= nz & 1; nz >>= 1;
950
    t.i = s.i ^ (sign & 1<<31);
951
    *dst++ = v[idx>>4 & 3] * t.f;
952

    
953
    sign <<= nz & 1; nz >>= 1;
954
    t.i = s.i ^ (sign & 1<<31);
955
    *dst++ = v[idx>>6 & 3] * t.f;
956

    
957
    return dst;
958
}
959
#endif
960

    
961
/**
962
 * Decode spectral data; reference: table 4.50.
963
 * Dequantize and scale spectral data; reference: 4.6.3.3.
964
 *
965
 * @param   coef            array of dequantized, scaled spectral data
966
 * @param   sf              array of scalefactors or intensity stereo positions
967
 * @param   pulse_present   set if pulses are present
968
 * @param   pulse           pointer to pulse data struct
969
 * @param   band_type       array of the used band type
970
 *
971
 * @return  Returns error status. 0 - OK, !0 - error
972
 */
973
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
974
                                       GetBitContext *gb, const float sf[120],
975
                                       int pulse_present, const Pulse *pulse,
976
                                       const IndividualChannelStream *ics,
977
                                       enum BandType band_type[120])
978
{
979
    int i, k, g, idx = 0;
980
    const int c = 1024 / ics->num_windows;
981
    const uint16_t *offsets = ics->swb_offset;
982
    float *coef_base = coef;
983
    int err_idx;
984

    
985
    for (g = 0; g < ics->num_windows; g++)
986
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
987

    
988
    for (g = 0; g < ics->num_window_groups; g++) {
989
        unsigned g_len = ics->group_len[g];
990

    
991
        for (i = 0; i < ics->max_sfb; i++, idx++) {
992
            const unsigned cbt_m1 = band_type[idx] - 1;
993
            float *cfo = coef + offsets[i];
994
            int off_len = offsets[i + 1] - offsets[i];
995
            int group;
996

    
997
            if (cbt_m1 >= INTENSITY_BT2 - 1) {
998
                for (group = 0; group < g_len; group++, cfo+=128) {
999
                    memset(cfo, 0, off_len * sizeof(float));
1000
                }
1001
            } else if (cbt_m1 == NOISE_BT - 1) {
1002
                for (group = 0; group < g_len; group++, cfo+=128) {
1003
                    float scale;
1004
                    float band_energy;
1005

    
1006
                    for (k = 0; k < off_len; k++) {
1007
                        ac->random_state  = lcg_random(ac->random_state);
1008
                        cfo[k] = ac->random_state;
1009
                    }
1010

    
1011
                    band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1012
                    scale = sf[idx] / sqrtf(band_energy);
1013
                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1014
                }
1015
            } else {
1016
                const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1017
                const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1018
                VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1019
                const int cb_size = ff_aac_spectral_sizes[cbt_m1];
1020
                OPEN_READER(re, gb);
1021

    
1022
                switch (cbt_m1 >> 1) {
1023
                case 0:
1024
                    for (group = 0; group < g_len; group++, cfo+=128) {
1025
                        float *cf = cfo;
1026
                        int len = off_len;
1027

    
1028
                        do {
1029
                            int code;
1030
                            unsigned cb_idx;
1031

    
1032
                            UPDATE_CACHE(re, gb);
1033
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1034

    
1035
                            if (code >= cb_size) {
1036
                                err_idx = code;
1037
                                goto err_cb_overflow;
1038
                            }
1039

    
1040
                            cb_idx = cb_vector_idx[code];
1041
                            cf = VMUL4(cf, vq, cb_idx, sf + idx);
1042
                        } while (len -= 4);
1043
                    }
1044
                    break;
1045

    
1046
                case 1:
1047
                    for (group = 0; group < g_len; group++, cfo+=128) {
1048
                        float *cf = cfo;
1049
                        int len = off_len;
1050

    
1051
                        do {
1052
                            int code;
1053
                            unsigned nnz;
1054
                            unsigned cb_idx;
1055
                            uint32_t bits;
1056

    
1057
                            UPDATE_CACHE(re, gb);
1058
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1059

    
1060
                            if (code >= cb_size) {
1061
                                err_idx = code;
1062
                                goto err_cb_overflow;
1063
                            }
1064

    
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

    
1089
                            if (code >= cb_size) {
1090
                                err_idx = code;
1091
                                goto err_cb_overflow;
1092
                            }
1093

    
1094
                            cb_idx = cb_vector_idx[code];
1095
                            cf = VMUL2(cf, vq, cb_idx, sf + idx);
1096
                        } while (len -= 2);
1097
                    }
1098
                    break;
1099

    
1100
                case 3:
1101
                case 4:
1102
                    for (group = 0; group < g_len; group++, cfo+=128) {
1103
                        float *cf = cfo;
1104
                        int len = off_len;
1105

    
1106
                        do {
1107
                            int code;
1108
                            unsigned nnz;
1109
                            unsigned cb_idx;
1110
                            unsigned sign;
1111

    
1112
                            UPDATE_CACHE(re, gb);
1113
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1114

    
1115
                            if (code >= cb_size) {
1116
                                err_idx = code;
1117
                                goto err_cb_overflow;
1118
                            }
1119

    
1120
                            cb_idx = cb_vector_idx[code];
1121
                            nnz = cb_idx >> 8 & 15;
1122
                            sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1123
                            LAST_SKIP_BITS(re, gb, nnz);
1124
                            cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1125
                        } while (len -= 2);
1126
                    }
1127
                    break;
1128

    
1129
                default:
1130
                    for (group = 0; group < g_len; group++, cfo+=128) {
1131
                        float *cf = cfo;
1132
                        uint32_t *icf = (uint32_t *) cf;
1133
                        int len = off_len;
1134

    
1135
                        do {
1136
                            int code;
1137
                            unsigned nzt, nnz;
1138
                            unsigned cb_idx;
1139
                            uint32_t bits;
1140
                            int j;
1141

    
1142
                            UPDATE_CACHE(re, gb);
1143
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1144

    
1145
                            if (!code) {
1146
                                *icf++ = 0;
1147
                                *icf++ = 0;
1148
                                continue;
1149
                            }
1150

    
1151
                            if (code >= cb_size) {
1152
                                err_idx = code;
1153
                                goto err_cb_overflow;
1154
                            }
1155

    
1156
                            cb_idx = cb_vector_idx[code];
1157
                            nnz = cb_idx >> 12;
1158
                            nzt = cb_idx >> 8;
1159
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1160
                            LAST_SKIP_BITS(re, gb, nnz);
1161

    
1162
                            for (j = 0; j < 2; j++) {
1163
                                if (nzt & 1<<j) {
1164
                                    uint32_t b;
1165
                                    int n;
1166
                                    /* The total length of escape_sequence must be < 22 bits according
1167
                                       to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1168
                                    UPDATE_CACHE(re, gb);
1169
                                    b = GET_CACHE(re, gb);
1170
                                    b = 31 - av_log2(~b);
1171

    
1172
                                    if (b > 8) {
1173
                                        av_log(ac->avccontext, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1174
                                        return -1;
1175
                                    }
1176

    
1177
#if MIN_CACHE_BITS < 21
1178
                                    LAST_SKIP_BITS(re, gb, b + 1);
1179
                                    UPDATE_CACHE(re, gb);
1180
#else
1181
                                    SKIP_BITS(re, gb, b + 1);
1182
#endif
1183
                                    b += 4;
1184
                                    n = (1 << b) + SHOW_UBITS(re, gb, b);
1185
                                    LAST_SKIP_BITS(re, gb, b);
1186
                                    *icf++ = cbrt_tab[n] | (bits & 1<<31);
1187
                                    bits <<= 1;
1188
                                } else {
1189
                                    unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1190
                                    *icf++ = (bits & 1<<31) | v;
1191
                                    bits <<= !!v;
1192
                                }
1193
                                cb_idx >>= 4;
1194
                            }
1195
                        } while (len -= 2);
1196

    
1197
                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1198
                    }
1199
                }
1200

    
1201
                CLOSE_READER(re, gb);
1202
            }
1203
        }
1204
        coef += g_len << 7;
1205
    }
1206

    
1207
    if (pulse_present) {
1208
        idx = 0;
1209
        for (i = 0; i < pulse->num_pulse; i++) {
1210
            float co = coef_base[ pulse->pos[i] ];
1211
            while (offsets[idx + 1] <= pulse->pos[i])
1212
                idx++;
1213
            if (band_type[idx] != NOISE_BT && sf[idx]) {
1214
                float ico = -pulse->amp[i];
1215
                if (co) {
1216
                    co /= sf[idx];
1217
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1218
                }
1219
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1220
            }
1221
        }
1222
    }
1223
    return 0;
1224

    
1225
err_cb_overflow:
1226
    av_log(ac->avccontext, AV_LOG_ERROR,
1227
           "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
1228
           band_type[idx], err_idx, ff_aac_spectral_sizes[band_type[idx]]);
1229
    return -1;
1230
}
1231

    
1232
static av_always_inline float flt16_round(float pf)
1233
{
1234
    union float754 tmp;
1235
    tmp.f = pf;
1236
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1237
    return tmp.f;
1238
}
1239

    
1240
static av_always_inline float flt16_even(float pf)
1241
{
1242
    union float754 tmp;
1243
    tmp.f = pf;
1244
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1245
    return tmp.f;
1246
}
1247

    
1248
static av_always_inline float flt16_trunc(float pf)
1249
{
1250
    union float754 pun;
1251
    pun.f = pf;
1252
    pun.i &= 0xFFFF0000U;
1253
    return pun.f;
1254
}
1255

    
1256
static av_always_inline void predict(AACContext *ac, PredictorState *ps, float *coef,
1257
                    int output_enable)
1258
{
1259
    const float a     = 0.953125; // 61.0 / 64
1260
    const float alpha = 0.90625;  // 29.0 / 32
1261
    float e0, e1;
1262
    float pv;
1263
    float k1, k2;
1264

    
1265
    k1 = ps->var0 > 1 ? ps->cor0 * flt16_even(a / ps->var0) : 0;
1266
    k2 = ps->var1 > 1 ? ps->cor1 * flt16_even(a / ps->var1) : 0;
1267

    
1268
    pv = flt16_round(k1 * ps->r0 + k2 * ps->r1);
1269
    if (output_enable)
1270
        *coef += pv * ac->sf_scale;
1271

    
1272
    e0 = *coef / ac->sf_scale;
1273
    e1 = e0 - k1 * ps->r0;
1274

    
1275
    ps->cor1 = flt16_trunc(alpha * ps->cor1 + ps->r1 * e1);
1276
    ps->var1 = flt16_trunc(alpha * ps->var1 + 0.5 * (ps->r1 * ps->r1 + e1 * e1));
1277
    ps->cor0 = flt16_trunc(alpha * ps->cor0 + ps->r0 * e0);
1278
    ps->var0 = flt16_trunc(alpha * ps->var0 + 0.5 * (ps->r0 * ps->r0 + e0 * e0));
1279

    
1280
    ps->r1 = flt16_trunc(a * (ps->r0 - k1 * e0));
1281
    ps->r0 = flt16_trunc(a * e0);
1282
}
1283

    
1284
/**
1285
 * Apply AAC-Main style frequency domain prediction.
1286
 */
1287
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1288
{
1289
    int sfb, k;
1290

    
1291
    if (!sce->ics.predictor_initialized) {
1292
        reset_all_predictors(sce->predictor_state);
1293
        sce->ics.predictor_initialized = 1;
1294
    }
1295

    
1296
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1297
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1298
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1299
                predict(ac, &sce->predictor_state[k], &sce->coeffs[k],
1300
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1301
            }
1302
        }
1303
        if (sce->ics.predictor_reset_group)
1304
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1305
    } else
1306
        reset_all_predictors(sce->predictor_state);
1307
}
1308

    
1309
/**
1310
 * Decode an individual_channel_stream payload; reference: table 4.44.
1311
 *
1312
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1313
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1314
 *
1315
 * @return  Returns error status. 0 - OK, !0 - error
1316
 */
1317
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1318
                      GetBitContext *gb, int common_window, int scale_flag)
1319
{
1320
    Pulse pulse;
1321
    TemporalNoiseShaping    *tns = &sce->tns;
1322
    IndividualChannelStream *ics = &sce->ics;
1323
    float *out = sce->coeffs;
1324
    int global_gain, pulse_present = 0;
1325

    
1326
    /* This assignment is to silence a GCC warning about the variable being used
1327
     * uninitialized when in fact it always is.
1328
     */
1329
    pulse.num_pulse = 0;
1330

    
1331
    global_gain = get_bits(gb, 8);
1332

    
1333
    if (!common_window && !scale_flag) {
1334
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1335
            return -1;
1336
    }
1337

    
1338
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1339
        return -1;
1340
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1341
        return -1;
1342

    
1343
    pulse_present = 0;
1344
    if (!scale_flag) {
1345
        if ((pulse_present = get_bits1(gb))) {
1346
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1347
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1348
                return -1;
1349
            }
1350
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1351
                av_log(ac->avccontext, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1352
                return -1;
1353
            }
1354
        }
1355
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1356
            return -1;
1357
        if (get_bits1(gb)) {
1358
            av_log_missing_feature(ac->avccontext, "SSR", 1);
1359
            return -1;
1360
        }
1361
    }
1362

    
1363
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1364
        return -1;
1365

    
1366
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1367
        apply_prediction(ac, sce);
1368

    
1369
    return 0;
1370
}
1371

    
1372
/**
1373
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1374
 */
1375
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1376
{
1377
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1378
    float *ch0 = cpe->ch[0].coeffs;
1379
    float *ch1 = cpe->ch[1].coeffs;
1380
    int g, i, group, idx = 0;
1381
    const uint16_t *offsets = ics->swb_offset;
1382
    for (g = 0; g < ics->num_window_groups; g++) {
1383
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1384
            if (cpe->ms_mask[idx] &&
1385
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1386
                for (group = 0; group < ics->group_len[g]; group++) {
1387
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1388
                                              ch1 + group * 128 + offsets[i],
1389
                                              offsets[i+1] - offsets[i]);
1390
                }
1391
            }
1392
        }
1393
        ch0 += ics->group_len[g] * 128;
1394
        ch1 += ics->group_len[g] * 128;
1395
    }
1396
}
1397

    
1398
/**
1399
 * intensity stereo decoding; reference: 4.6.8.2.3
1400
 *
1401
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1402
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1403
 *                      [3] reserved for scalable AAC
1404
 */
1405
static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1406
{
1407
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1408
    SingleChannelElement         *sce1 = &cpe->ch[1];
1409
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1410
    const uint16_t *offsets = ics->swb_offset;
1411
    int g, group, i, k, idx = 0;
1412
    int c;
1413
    float scale;
1414
    for (g = 0; g < ics->num_window_groups; g++) {
1415
        for (i = 0; i < ics->max_sfb;) {
1416
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1417
                const int bt_run_end = sce1->band_type_run_end[idx];
1418
                for (; i < bt_run_end; i++, idx++) {
1419
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1420
                    if (ms_present)
1421
                        c *= 1 - 2 * cpe->ms_mask[idx];
1422
                    scale = c * sce1->sf[idx];
1423
                    for (group = 0; group < ics->group_len[g]; group++)
1424
                        for (k = offsets[i]; k < offsets[i + 1]; k++)
1425
                            coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1426
                }
1427
            } else {
1428
                int bt_run_end = sce1->band_type_run_end[idx];
1429
                idx += bt_run_end - i;
1430
                i    = bt_run_end;
1431
            }
1432
        }
1433
        coef0 += ics->group_len[g] * 128;
1434
        coef1 += ics->group_len[g] * 128;
1435
    }
1436
}
1437

    
1438
/**
1439
 * Decode a channel_pair_element; reference: table 4.4.
1440
 *
1441
 * @param   elem_id Identifies the instance of a syntax element.
1442
 *
1443
 * @return  Returns error status. 0 - OK, !0 - error
1444
 */
1445
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1446
{
1447
    int i, ret, common_window, ms_present = 0;
1448

    
1449
    common_window = get_bits1(gb);
1450
    if (common_window) {
1451
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1452
            return -1;
1453
        i = cpe->ch[1].ics.use_kb_window[0];
1454
        cpe->ch[1].ics = cpe->ch[0].ics;
1455
        cpe->ch[1].ics.use_kb_window[1] = i;
1456
        ms_present = get_bits(gb, 2);
1457
        if (ms_present == 3) {
1458
            av_log(ac->avccontext, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1459
            return -1;
1460
        } else if (ms_present)
1461
            decode_mid_side_stereo(cpe, gb, ms_present);
1462
    }
1463
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1464
        return ret;
1465
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1466
        return ret;
1467

    
1468
    if (common_window) {
1469
        if (ms_present)
1470
            apply_mid_side_stereo(ac, cpe);
1471
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1472
            apply_prediction(ac, &cpe->ch[0]);
1473
            apply_prediction(ac, &cpe->ch[1]);
1474
        }
1475
    }
1476

    
1477
    apply_intensity_stereo(cpe, ms_present);
1478
    return 0;
1479
}
1480

    
1481
/**
1482
 * Decode coupling_channel_element; reference: table 4.8.
1483
 *
1484
 * @param   elem_id Identifies the instance of a syntax element.
1485
 *
1486
 * @return  Returns error status. 0 - OK, !0 - error
1487
 */
1488
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1489
{
1490
    int num_gain = 0;
1491
    int c, g, sfb, ret;
1492
    int sign;
1493
    float scale;
1494
    SingleChannelElement *sce = &che->ch[0];
1495
    ChannelCoupling     *coup = &che->coup;
1496

    
1497
    coup->coupling_point = 2 * get_bits1(gb);
1498
    coup->num_coupled = get_bits(gb, 3);
1499
    for (c = 0; c <= coup->num_coupled; c++) {
1500
        num_gain++;
1501
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1502
        coup->id_select[c] = get_bits(gb, 4);
1503
        if (coup->type[c] == TYPE_CPE) {
1504
            coup->ch_select[c] = get_bits(gb, 2);
1505
            if (coup->ch_select[c] == 3)
1506
                num_gain++;
1507
        } else
1508
            coup->ch_select[c] = 2;
1509
    }
1510
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1511

    
1512
    sign  = get_bits(gb, 1);
1513
    scale = pow(2., pow(2., (int)get_bits(gb, 2) - 3));
1514

    
1515
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1516
        return ret;
1517

    
1518
    for (c = 0; c < num_gain; c++) {
1519
        int idx  = 0;
1520
        int cge  = 1;
1521
        int gain = 0;
1522
        float gain_cache = 1.;
1523
        if (c) {
1524
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1525
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1526
            gain_cache = pow(scale, -gain);
1527
        }
1528
        if (coup->coupling_point == AFTER_IMDCT) {
1529
            coup->gain[c][0] = gain_cache;
1530
        } else {
1531
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1532
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1533
                    if (sce->band_type[idx] != ZERO_BT) {
1534
                        if (!cge) {
1535
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1536
                            if (t) {
1537
                                int s = 1;
1538
                                t = gain += t;
1539
                                if (sign) {
1540
                                    s  -= 2 * (t & 0x1);
1541
                                    t >>= 1;
1542
                                }
1543
                                gain_cache = pow(scale, -t) * s;
1544
                            }
1545
                        }
1546
                        coup->gain[c][idx] = gain_cache;
1547
                    }
1548
                }
1549
            }
1550
        }
1551
    }
1552
    return 0;
1553
}
1554

    
1555
/**
1556
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1557
 *
1558
 * @return  Returns number of bytes consumed.
1559
 */
1560
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1561
                                         GetBitContext *gb)
1562
{
1563
    int i;
1564
    int num_excl_chan = 0;
1565

    
1566
    do {
1567
        for (i = 0; i < 7; i++)
1568
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1569
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1570

    
1571
    return num_excl_chan / 7;
1572
}
1573

    
1574
/**
1575
 * Decode dynamic range information; reference: table 4.52.
1576
 *
1577
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1578
 *
1579
 * @return  Returns number of bytes consumed.
1580
 */
1581
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1582
                                GetBitContext *gb, int cnt)
1583
{
1584
    int n             = 1;
1585
    int drc_num_bands = 1;
1586
    int i;
1587

    
1588
    /* pce_tag_present? */
1589
    if (get_bits1(gb)) {
1590
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1591
        skip_bits(gb, 4); // tag_reserved_bits
1592
        n++;
1593
    }
1594

    
1595
    /* excluded_chns_present? */
1596
    if (get_bits1(gb)) {
1597
        n += decode_drc_channel_exclusions(che_drc, gb);
1598
    }
1599

    
1600
    /* drc_bands_present? */
1601
    if (get_bits1(gb)) {
1602
        che_drc->band_incr            = get_bits(gb, 4);
1603
        che_drc->interpolation_scheme = get_bits(gb, 4);
1604
        n++;
1605
        drc_num_bands += che_drc->band_incr;
1606
        for (i = 0; i < drc_num_bands; i++) {
1607
            che_drc->band_top[i] = get_bits(gb, 8);
1608
            n++;
1609
        }
1610
    }
1611

    
1612
    /* prog_ref_level_present? */
1613
    if (get_bits1(gb)) {
1614
        che_drc->prog_ref_level = get_bits(gb, 7);
1615
        skip_bits1(gb); // prog_ref_level_reserved_bits
1616
        n++;
1617
    }
1618

    
1619
    for (i = 0; i < drc_num_bands; i++) {
1620
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1621
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1622
        n++;
1623
    }
1624

    
1625
    return n;
1626
}
1627

    
1628
/**
1629
 * Decode extension data (incomplete); reference: table 4.51.
1630
 *
1631
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1632
 *
1633
 * @return Returns number of bytes consumed
1634
 */
1635
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1636
                                    ChannelElement *che, enum RawDataBlockType elem_type)
1637
{
1638
    int crc_flag = 0;
1639
    int res = cnt;
1640
    switch (get_bits(gb, 4)) { // extension type
1641
    case EXT_SBR_DATA_CRC:
1642
        crc_flag++;
1643
    case EXT_SBR_DATA:
1644
        if (!che) {
1645
            av_log(ac->avccontext, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1646
            return res;
1647
        } else if (!ac->m4ac.sbr) {
1648
            av_log(ac->avccontext, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1649
            skip_bits_long(gb, 8 * cnt - 4);
1650
            return res;
1651
        } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1652
            av_log(ac->avccontext, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1653
            skip_bits_long(gb, 8 * cnt - 4);
1654
            return res;
1655
        } else {
1656
            ac->m4ac.sbr = 1;
1657
        }
1658
        res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1659
        break;
1660
    case EXT_DYNAMIC_RANGE:
1661
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1662
        break;
1663
    case EXT_FILL:
1664
    case EXT_FILL_DATA:
1665
    case EXT_DATA_ELEMENT:
1666
    default:
1667
        skip_bits_long(gb, 8 * cnt - 4);
1668
        break;
1669
    };
1670
    return res;
1671
}
1672

    
1673
/**
1674
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1675
 *
1676
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1677
 * @param   coef    spectral coefficients
1678
 */
1679
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1680
                      IndividualChannelStream *ics, int decode)
1681
{
1682
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1683
    int w, filt, m, i;
1684
    int bottom, top, order, start, end, size, inc;
1685
    float lpc[TNS_MAX_ORDER];
1686

    
1687
    for (w = 0; w < ics->num_windows; w++) {
1688
        bottom = ics->num_swb;
1689
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1690
            top    = bottom;
1691
            bottom = FFMAX(0, top - tns->length[w][filt]);
1692
            order  = tns->order[w][filt];
1693
            if (order == 0)
1694
                continue;
1695

    
1696
            // tns_decode_coef
1697
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1698

    
1699
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1700
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1701
            if ((size = end - start) <= 0)
1702
                continue;
1703
            if (tns->direction[w][filt]) {
1704
                inc = -1;
1705
                start = end - 1;
1706
            } else {
1707
                inc = 1;
1708
            }
1709
            start += w * 128;
1710

    
1711
            // ar filter
1712
            for (m = 0; m < size; m++, start += inc)
1713
                for (i = 1; i <= FFMIN(m, order); i++)
1714
                    coef[start] -= coef[start - i * inc] * lpc[i - 1];
1715
        }
1716
    }
1717
}
1718

    
1719
/**
1720
 * Conduct IMDCT and windowing.
1721
 */
1722
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
1723
{
1724
    IndividualChannelStream *ics = &sce->ics;
1725
    float *in    = sce->coeffs;
1726
    float *out   = sce->ret;
1727
    float *saved = sce->saved;
1728
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1729
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1730
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1731
    float *buf  = ac->buf_mdct;
1732
    float *temp = ac->temp;
1733
    int i;
1734

    
1735
    // imdct
1736
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1737
        if (ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE)
1738
            av_log(ac->avccontext, AV_LOG_WARNING,
1739
                   "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1740
                   "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1741
        for (i = 0; i < 1024; i += 128)
1742
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1743
    } else
1744
        ff_imdct_half(&ac->mdct, buf, in);
1745

    
1746
    /* window overlapping
1747
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1748
     * and long to short transitions are considered to be short to short
1749
     * transitions. This leaves just two cases (long to long and short to short)
1750
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1751
     */
1752
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1753
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1754
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, bias, 512);
1755
    } else {
1756
        for (i = 0; i < 448; i++)
1757
            out[i] = saved[i] + bias;
1758

    
1759
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1760
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, bias, 64);
1761
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      bias, 64);
1762
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      bias, 64);
1763
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      bias, 64);
1764
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      bias, 64);
1765
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1766
        } else {
1767
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, bias, 64);
1768
            for (i = 576; i < 1024; i++)
1769
                out[i] = buf[i-512] + bias;
1770
        }
1771
    }
1772

    
1773
    // buffer update
1774
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1775
        for (i = 0; i < 64; i++)
1776
            saved[i] = temp[64 + i] - bias;
1777
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1778
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1779
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1780
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1781
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1782
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1783
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1784
    } else { // LONG_STOP or ONLY_LONG
1785
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1786
    }
1787
}
1788

    
1789
/**
1790
 * Apply dependent channel coupling (applied before IMDCT).
1791
 *
1792
 * @param   index   index into coupling gain array
1793
 */
1794
static void apply_dependent_coupling(AACContext *ac,
1795
                                     SingleChannelElement *target,
1796
                                     ChannelElement *cce, int index)
1797
{
1798
    IndividualChannelStream *ics = &cce->ch[0].ics;
1799
    const uint16_t *offsets = ics->swb_offset;
1800
    float *dest = target->coeffs;
1801
    const float *src = cce->ch[0].coeffs;
1802
    int g, i, group, k, idx = 0;
1803
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1804
        av_log(ac->avccontext, AV_LOG_ERROR,
1805
               "Dependent coupling is not supported together with LTP\n");
1806
        return;
1807
    }
1808
    for (g = 0; g < ics->num_window_groups; g++) {
1809
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1810
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1811
                const float gain = cce->coup.gain[index][idx];
1812
                for (group = 0; group < ics->group_len[g]; group++) {
1813
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1814
                        // XXX dsputil-ize
1815
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1816
                    }
1817
                }
1818
            }
1819
        }
1820
        dest += ics->group_len[g] * 128;
1821
        src  += ics->group_len[g] * 128;
1822
    }
1823
}
1824

    
1825
/**
1826
 * Apply independent channel coupling (applied after IMDCT).
1827
 *
1828
 * @param   index   index into coupling gain array
1829
 */
1830
static void apply_independent_coupling(AACContext *ac,
1831
                                       SingleChannelElement *target,
1832
                                       ChannelElement *cce, int index)
1833
{
1834
    int i;
1835
    const float gain = cce->coup.gain[index][0];
1836
    const float bias = ac->add_bias;
1837
    const float *src = cce->ch[0].ret;
1838
    float *dest = target->ret;
1839
    const int len = 1024 << (ac->m4ac.sbr == 1);
1840

    
1841
    for (i = 0; i < len; i++)
1842
        dest[i] += gain * (src[i] - bias);
1843
}
1844

    
1845
/**
1846
 * channel coupling transformation interface
1847
 *
1848
 * @param   index   index into coupling gain array
1849
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1850
 */
1851
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1852
                                   enum RawDataBlockType type, int elem_id,
1853
                                   enum CouplingPoint coupling_point,
1854
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1855
{
1856
    int i, c;
1857

    
1858
    for (i = 0; i < MAX_ELEM_ID; i++) {
1859
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1860
        int index = 0;
1861

    
1862
        if (cce && cce->coup.coupling_point == coupling_point) {
1863
            ChannelCoupling *coup = &cce->coup;
1864

    
1865
            for (c = 0; c <= coup->num_coupled; c++) {
1866
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1867
                    if (coup->ch_select[c] != 1) {
1868
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1869
                        if (coup->ch_select[c] != 0)
1870
                            index++;
1871
                    }
1872
                    if (coup->ch_select[c] != 2)
1873
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1874
                } else
1875
                    index += 1 + (coup->ch_select[c] == 3);
1876
            }
1877
        }
1878
    }
1879
}
1880

    
1881
/**
1882
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1883
 */
1884
static void spectral_to_sample(AACContext *ac)
1885
{
1886
    int i, type;
1887
    float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
1888
    for (type = 3; type >= 0; type--) {
1889
        for (i = 0; i < MAX_ELEM_ID; i++) {
1890
            ChannelElement *che = ac->che[type][i];
1891
            if (che) {
1892
                if (type <= TYPE_CPE)
1893
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1894
                if (che->ch[0].tns.present)
1895
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1896
                if (che->ch[1].tns.present)
1897
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1898
                if (type <= TYPE_CPE)
1899
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1900
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1901
                    imdct_and_windowing(ac, &che->ch[0], imdct_bias);
1902
                    if (ac->m4ac.sbr > 0) {
1903
                        ff_sbr_dequant(ac, &che->sbr, type == TYPE_CPE ? TYPE_CPE : TYPE_SCE);
1904
                        ff_sbr_apply(ac, &che->sbr, 0, che->ch[0].ret, che->ch[0].ret);
1905
                    }
1906
                }
1907
                if (type == TYPE_CPE) {
1908
                    imdct_and_windowing(ac, &che->ch[1], imdct_bias);
1909
                    if (ac->m4ac.sbr > 0)
1910
                        ff_sbr_apply(ac, &che->sbr, 1, che->ch[1].ret, che->ch[1].ret);
1911
                }
1912
                if (type <= TYPE_CCE)
1913
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1914
            }
1915
        }
1916
    }
1917
}
1918

    
1919
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1920
{
1921
    int size;
1922
    AACADTSHeaderInfo hdr_info;
1923

    
1924
    size = ff_aac_parse_header(gb, &hdr_info);
1925
    if (size > 0) {
1926
        if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1927
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1928
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1929
            ac->m4ac.chan_config = hdr_info.chan_config;
1930
            if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1931
                return -7;
1932
            if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1933
                return -7;
1934
        } else if (ac->output_configured != OC_LOCKED) {
1935
            ac->output_configured = OC_NONE;
1936
        }
1937
        if (ac->output_configured != OC_LOCKED)
1938
            ac->m4ac.sbr = -1;
1939
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1940
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1941
        ac->m4ac.object_type     = hdr_info.object_type;
1942
        if (!ac->avccontext->sample_rate)
1943
            ac->avccontext->sample_rate = hdr_info.sample_rate;
1944
        if (hdr_info.num_aac_frames == 1) {
1945
            if (!hdr_info.crc_absent)
1946
                skip_bits(gb, 16);
1947
        } else {
1948
            av_log_missing_feature(ac->avccontext, "More than one AAC RDB per ADTS frame is", 0);
1949
            return -1;
1950
        }
1951
    }
1952
    return size;
1953
}
1954

    
1955
static int aac_decode_frame(AVCodecContext *avccontext, void *data,
1956
                            int *data_size, AVPacket *avpkt)
1957
{
1958
    const uint8_t *buf = avpkt->data;
1959
    int buf_size = avpkt->size;
1960
    AACContext *ac = avccontext->priv_data;
1961
    ChannelElement *che = NULL, *che_prev = NULL;
1962
    GetBitContext gb;
1963
    enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1964
    int err, elem_id, data_size_tmp;
1965
    int buf_consumed;
1966
    int samples = 1024, multiplier;
1967

    
1968
    init_get_bits(&gb, buf, buf_size * 8);
1969

    
1970
    if (show_bits(&gb, 12) == 0xfff) {
1971
        if (parse_adts_frame_header(ac, &gb) < 0) {
1972
            av_log(avccontext, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1973
            return -1;
1974
        }
1975
        if (ac->m4ac.sampling_index > 12) {
1976
            av_log(ac->avccontext, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1977
            return -1;
1978
        }
1979
    }
1980

    
1981
    // parse
1982
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1983
        elem_id = get_bits(&gb, 4);
1984

    
1985
        if (elem_type < TYPE_DSE && !(che=get_che(ac, elem_type, elem_id))) {
1986
            av_log(ac->avccontext, AV_LOG_ERROR, "channel element %d.%d is not allocated\n", elem_type, elem_id);
1987
            return -1;
1988
        }
1989

    
1990
        switch (elem_type) {
1991

    
1992
        case TYPE_SCE:
1993
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1994
            break;
1995

    
1996
        case TYPE_CPE:
1997
            err = decode_cpe(ac, &gb, che);
1998
            break;
1999

    
2000
        case TYPE_CCE:
2001
            err = decode_cce(ac, &gb, che);
2002
            break;
2003

    
2004
        case TYPE_LFE:
2005
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
2006
            break;
2007

    
2008
        case TYPE_DSE:
2009
            err = skip_data_stream_element(ac, &gb);
2010
            break;
2011

    
2012
        case TYPE_PCE: {
2013
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2014
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2015
            if ((err = decode_pce(ac, new_che_pos, &gb)))
2016
                break;
2017
            if (ac->output_configured > OC_TRIAL_PCE)
2018
                av_log(avccontext, AV_LOG_ERROR,
2019
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2020
            else
2021
                err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2022
            break;
2023
        }
2024

    
2025
        case TYPE_FIL:
2026
            if (elem_id == 15)
2027
                elem_id += get_bits(&gb, 8) - 1;
2028
            if (get_bits_left(&gb) < 8 * elem_id) {
2029
                    av_log(avccontext, AV_LOG_ERROR, overread_err);
2030
                    return -1;
2031
            }
2032
            while (elem_id > 0)
2033
                elem_id -= decode_extension_payload(ac, &gb, elem_id, che_prev, elem_type_prev);
2034
            err = 0; /* FIXME */
2035
            break;
2036

    
2037
        default:
2038
            err = -1; /* should not happen, but keeps compiler happy */
2039
            break;
2040
        }
2041

    
2042
        che_prev       = che;
2043
        elem_type_prev = elem_type;
2044

    
2045
        if (err)
2046
            return err;
2047

    
2048
        if (get_bits_left(&gb) < 3) {
2049
            av_log(avccontext, AV_LOG_ERROR, overread_err);
2050
            return -1;
2051
        }
2052
    }
2053

    
2054
    spectral_to_sample(ac);
2055

    
2056
    multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2057
    samples <<= multiplier;
2058
    if (ac->output_configured < OC_LOCKED) {
2059
        avccontext->sample_rate = ac->m4ac.sample_rate << multiplier;
2060
        avccontext->frame_size = samples;
2061
    }
2062

    
2063
    data_size_tmp = samples * avccontext->channels * sizeof(int16_t);
2064
    if (*data_size < data_size_tmp) {
2065
        av_log(avccontext, AV_LOG_ERROR,
2066
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2067
               *data_size, data_size_tmp);
2068
        return -1;
2069
    }
2070
    *data_size = data_size_tmp;
2071

    
2072
    ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avccontext->channels);
2073

    
2074
    if (ac->output_configured)
2075
        ac->output_configured = OC_LOCKED;
2076

    
2077
    buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2078
    return buf_size > buf_consumed ? buf_consumed : buf_size;
2079
}
2080

    
2081
static av_cold int aac_decode_close(AVCodecContext *avccontext)
2082
{
2083
    AACContext *ac = avccontext->priv_data;
2084
    int i, type;
2085

    
2086
    for (i = 0; i < MAX_ELEM_ID; i++) {
2087
        for (type = 0; type < 4; type++) {
2088
            if (ac->che[type][i])
2089
                ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2090
            av_freep(&ac->che[type][i]);
2091
        }
2092
    }
2093

    
2094
    ff_mdct_end(&ac->mdct);
2095
    ff_mdct_end(&ac->mdct_small);
2096
    return 0;
2097
}
2098

    
2099
AVCodec aac_decoder = {
2100
    "aac",
2101
    CODEC_TYPE_AUDIO,
2102
    CODEC_ID_AAC,
2103
    sizeof(AACContext),
2104
    aac_decode_init,
2105
    NULL,
2106
    aac_decode_close,
2107
    aac_decode_frame,
2108
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2109
    .sample_fmts = (const enum SampleFormat[]) {
2110
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2111
    },
2112
    .channel_layouts = aac_channel_layout,
2113
};