Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aacdec.c @ bb2d8e9f

History | View | Annotate | Download (72.9 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
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
 * Y                    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 "cbrt_tablegen.h"
90
#include "sbr.h"
91
#include "aacsbr.h"
92
#include "mpeg4audio.h"
93
#include "aacadtsdec.h"
94

    
95
#include <assert.h>
96
#include <errno.h>
97
#include <math.h>
98
#include <string.h>
99

    
100
#if ARCH_ARM
101
#   include "arm/aac.h"
102
#endif
103

    
104
union float754 {
105
    float f;
106
    uint32_t i;
107
};
108

    
109
static VLC vlc_scalefactors;
110
static VLC vlc_spectral[11];
111

    
112
static const char overread_err[] = "Input buffer exhausted before END element found\n";
113

    
114
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
115
{
116
    // For PCE based channel configurations map the channels solely based on tags.
117
    if (!ac->m4ac.chan_config) {
118
        return ac->tag_che_map[type][elem_id];
119
    }
120
    // For indexed channel configurations map the channels solely based on position.
121
    switch (ac->m4ac.chan_config) {
122
    case 7:
123
        if (ac->tags_mapped == 3 && type == TYPE_CPE) {
124
            ac->tags_mapped++;
125
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
126
        }
127
    case 6:
128
        /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
129
           instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
130
           encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
131
        if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
132
            ac->tags_mapped++;
133
            return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
134
        }
135
    case 5:
136
        if (ac->tags_mapped == 2 && type == TYPE_CPE) {
137
            ac->tags_mapped++;
138
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
139
        }
140
    case 4:
141
        if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
142
            ac->tags_mapped++;
143
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
144
        }
145
    case 3:
146
    case 2:
147
        if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
148
            ac->tags_mapped++;
149
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
150
        } else if (ac->m4ac.chan_config == 2) {
151
            return NULL;
152
        }
153
    case 1:
154
        if (!ac->tags_mapped && type == TYPE_SCE) {
155
            ac->tags_mapped++;
156
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
157
        }
158
    default:
159
        return NULL;
160
    }
161
}
162

    
163
/**
164
 * Check for the channel element in the current channel position configuration.
165
 * If it exists, make sure the appropriate element is allocated and map the
166
 * channel order to match the internal FFmpeg channel layout.
167
 *
168
 * @param   che_pos current channel position configuration
169
 * @param   type channel element type
170
 * @param   id channel element id
171
 * @param   channels count of the number of channels in the configuration
172
 *
173
 * @return  Returns error status. 0 - OK, !0 - error
174
 */
175
static av_cold int che_configure(AACContext *ac,
176
                         enum ChannelPosition che_pos[4][MAX_ELEM_ID],
177
                         int type, int id,
178
                         int *channels)
179
{
180
    if (che_pos[type][id]) {
181
        if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
182
            return AVERROR(ENOMEM);
183
        ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
184
        if (type != TYPE_CCE) {
185
            ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
186
            if (type == TYPE_CPE ||
187
                (type == TYPE_SCE && ac->m4ac.ps == 1)) {
188
                ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
189
            }
190
        }
191
    } else {
192
        if (ac->che[type][id])
193
            ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
194
        av_freep(&ac->che[type][id]);
195
    }
196
    return 0;
197
}
198

    
199
/**
200
 * Configure output channel order based on the current program configuration element.
201
 *
202
 * @param   che_pos current channel position configuration
203
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
204
 *
205
 * @return  Returns error status. 0 - OK, !0 - error
206
 */
207
static av_cold int output_configure(AACContext *ac,
208
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
209
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
210
                            int channel_config, enum OCStatus oc_type)
211
{
212
    AVCodecContext *avctx = ac->avctx;
213
    int i, type, channels = 0, ret;
214

    
215
    if (new_che_pos != che_pos)
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

    
229
        avctx->channel_layout = aac_channel_layout[channel_config - 1];
230
    } else {
231
        /* Allocate or free elements depending on if they are in the
232
         * current program configuration.
233
         *
234
         * Set up default 1:1 output mapping.
235
         *
236
         * For a 5.1 stream the output order will be:
237
         *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
238
         */
239

    
240
        for (i = 0; i < MAX_ELEM_ID; i++) {
241
            for (type = 0; type < 4; type++) {
242
                if ((ret = che_configure(ac, che_pos, type, i, &channels)))
243
                    return ret;
244
            }
245
        }
246

    
247
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
248

    
249
        avctx->channel_layout = 0;
250
    }
251

    
252
    avctx->channels = channels;
253

    
254
    ac->output_configured = oc_type;
255

    
256
    return 0;
257
}
258

    
259
/**
260
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
261
 *
262
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
263
 * @param sce_map mono (Single Channel Element) map
264
 * @param type speaker type/position for these channels
265
 */
266
static void decode_channel_map(enum ChannelPosition *cpe_map,
267
                               enum ChannelPosition *sce_map,
268
                               enum ChannelPosition type,
269
                               GetBitContext *gb, int n)
270
{
271
    while (n--) {
272
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
273
        map[get_bits(gb, 4)] = type;
274
    }
275
}
276

    
277
/**
278
 * Decode program configuration element; reference: table 4.2.
279
 *
280
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
281
 *
282
 * @return  Returns error status. 0 - OK, !0 - error
283
 */
284
static int decode_pce(AACContext *ac, enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
285
                      GetBitContext *gb)
286
{
287
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
288
    int comment_len;
289

    
290
    skip_bits(gb, 2);  // object_type
291

    
292
    sampling_index = get_bits(gb, 4);
293
    if (ac->m4ac.sampling_index != sampling_index)
294
        av_log(ac->avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
295

    
296
    num_front       = get_bits(gb, 4);
297
    num_side        = get_bits(gb, 4);
298
    num_back        = get_bits(gb, 4);
299
    num_lfe         = get_bits(gb, 2);
300
    num_assoc_data  = get_bits(gb, 3);
301
    num_cc          = get_bits(gb, 4);
302

    
303
    if (get_bits1(gb))
304
        skip_bits(gb, 4); // mono_mixdown_tag
305
    if (get_bits1(gb))
306
        skip_bits(gb, 4); // stereo_mixdown_tag
307

    
308
    if (get_bits1(gb))
309
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
310

    
311
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
312
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
313
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
314
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
315

    
316
    skip_bits_long(gb, 4 * num_assoc_data);
317

    
318
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
319

    
320
    align_get_bits(gb);
321

    
322
    /* comment field, first byte is length */
323
    comment_len = get_bits(gb, 8) * 8;
324
    if (get_bits_left(gb) < comment_len) {
325
        av_log(ac->avctx, AV_LOG_ERROR, overread_err);
326
        return -1;
327
    }
328
    skip_bits_long(gb, comment_len);
329
    return 0;
330
}
331

    
332
/**
333
 * Set up channel positions based on a default channel configuration
334
 * as specified in table 1.17.
335
 *
336
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
337
 *
338
 * @return  Returns error status. 0 - OK, !0 - error
339
 */
340
static av_cold int set_default_channel_config(AACContext *ac,
341
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
342
                                      int channel_config)
343
{
344
    if (channel_config < 1 || channel_config > 7) {
345
        av_log(ac->avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
346
               channel_config);
347
        return -1;
348
    }
349

    
350
    /* default channel configurations:
351
     *
352
     * 1ch : front center (mono)
353
     * 2ch : L + R (stereo)
354
     * 3ch : front center + L + R
355
     * 4ch : front center + L + R + back center
356
     * 5ch : front center + L + R + back stereo
357
     * 6ch : front center + L + R + back stereo + LFE
358
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
359
     */
360

    
361
    if (channel_config != 2)
362
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
363
    if (channel_config > 1)
364
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
365
    if (channel_config == 4)
366
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
367
    if (channel_config > 4)
368
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
369
        = AAC_CHANNEL_BACK;  // back stereo
370
    if (channel_config > 5)
371
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
372
    if (channel_config == 7)
373
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
374

    
375
    return 0;
376
}
377

    
378
/**
379
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
380
 *
381
 * @return  Returns error status. 0 - OK, !0 - error
382
 */
383
static int decode_ga_specific_config(AACContext *ac, GetBitContext *gb,
384
                                     int channel_config)
385
{
386
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
387
    int extension_flag, ret;
388

    
389
    if (get_bits1(gb)) { // frameLengthFlag
390
        av_log_missing_feature(ac->avctx, "960/120 MDCT window is", 1);
391
        return -1;
392
    }
393

    
394
    if (get_bits1(gb))       // dependsOnCoreCoder
395
        skip_bits(gb, 14);   // coreCoderDelay
396
    extension_flag = get_bits1(gb);
397

    
398
    if (ac->m4ac.object_type == AOT_AAC_SCALABLE ||
399
        ac->m4ac.object_type == AOT_ER_AAC_SCALABLE)
400
        skip_bits(gb, 3);     // layerNr
401

    
402
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
403
    if (channel_config == 0) {
404
        skip_bits(gb, 4);  // element_instance_tag
405
        if ((ret = decode_pce(ac, new_che_pos, gb)))
406
            return ret;
407
    } else {
408
        if ((ret = set_default_channel_config(ac, new_che_pos, channel_config)))
409
            return ret;
410
    }
411
    if ((ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
412
        return ret;
413

    
414
    if (extension_flag) {
415
        switch (ac->m4ac.object_type) {
416
        case AOT_ER_BSAC:
417
            skip_bits(gb, 5);    // numOfSubFrame
418
            skip_bits(gb, 11);   // layer_length
419
            break;
420
        case AOT_ER_AAC_LC:
421
        case AOT_ER_AAC_LTP:
422
        case AOT_ER_AAC_SCALABLE:
423
        case AOT_ER_AAC_LD:
424
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
425
                                    * aacScalefactorDataResilienceFlag
426
                                    * aacSpectralDataResilienceFlag
427
                                    */
428
            break;
429
        }
430
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
431
    }
432
    return 0;
433
}
434

    
435
/**
436
 * Decode audio specific configuration; reference: table 1.13.
437
 *
438
 * @param   data        pointer to AVCodecContext extradata
439
 * @param   data_size   size of AVCCodecContext extradata
440
 *
441
 * @return  Returns error status. 0 - OK, !0 - error
442
 */
443
static int decode_audio_specific_config(AACContext *ac, void *data,
444
                                        int data_size)
445
{
446
    GetBitContext gb;
447
    int i;
448

    
449
    init_get_bits(&gb, data, data_size * 8);
450

    
451
    if ((i = ff_mpeg4audio_get_config(&ac->m4ac, data, data_size)) < 0)
452
        return -1;
453
    if (ac->m4ac.sampling_index > 12) {
454
        av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
455
        return -1;
456
    }
457
    if (ac->m4ac.sbr == 1 && ac->m4ac.ps == -1)
458
        ac->m4ac.ps = 1;
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->avctx, 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
#define AAC_INIT_VLC_STATIC(num, size) \
513
    INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
514
         ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
515
        ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
516
        size);
517

    
518
static av_cold int aac_decode_init(AVCodecContext *avctx)
519
{
520
    AACContext *ac = avctx->priv_data;
521

    
522
    ac->avctx = avctx;
523
    ac->m4ac.sample_rate = avctx->sample_rate;
524

    
525
    if (avctx->extradata_size > 0) {
526
        if (decode_audio_specific_config(ac, avctx->extradata, avctx->extradata_size))
527
            return -1;
528
    }
529

    
530
    avctx->sample_fmt = SAMPLE_FMT_S16;
531

    
532
    AAC_INIT_VLC_STATIC( 0, 304);
533
    AAC_INIT_VLC_STATIC( 1, 270);
534
    AAC_INIT_VLC_STATIC( 2, 550);
535
    AAC_INIT_VLC_STATIC( 3, 300);
536
    AAC_INIT_VLC_STATIC( 4, 328);
537
    AAC_INIT_VLC_STATIC( 5, 294);
538
    AAC_INIT_VLC_STATIC( 6, 306);
539
    AAC_INIT_VLC_STATIC( 7, 268);
540
    AAC_INIT_VLC_STATIC( 8, 510);
541
    AAC_INIT_VLC_STATIC( 9, 366);
542
    AAC_INIT_VLC_STATIC(10, 462);
543

    
544
    ff_aac_sbr_init();
545

    
546
    dsputil_init(&ac->dsp, avctx);
547

    
548
    ac->random_state = 0x1f2e3d4c;
549

    
550
    // -1024 - Compensate wrong IMDCT method.
551
    // 32768 - Required to scale values to the correct range for the bias method
552
    //         for float to int16 conversion.
553

    
554
    if (ac->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
555
        ac->add_bias  = 385.0f;
556
        ac->sf_scale  = 1. / (-1024. * 32768.);
557
        ac->sf_offset = 0;
558
    } else {
559
        ac->add_bias  = 0.0f;
560
        ac->sf_scale  = 1. / -1024.;
561
        ac->sf_offset = 60;
562
    }
563

    
564
    ff_aac_tableinit();
565

    
566
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
567
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
568
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
569
                    352);
570

    
571
    ff_mdct_init(&ac->mdct, 11, 1, 1.0);
572
    ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
573
    // window initialization
574
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
575
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
576
    ff_init_ff_sine_windows(10);
577
    ff_init_ff_sine_windows( 7);
578

    
579
    cbrt_tableinit();
580

    
581
    return 0;
582
}
583

    
584
/**
585
 * Skip data_stream_element; reference: table 4.10.
586
 */
587
static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
588
{
589
    int byte_align = get_bits1(gb);
590
    int count = get_bits(gb, 8);
591
    if (count == 255)
592
        count += get_bits(gb, 8);
593
    if (byte_align)
594
        align_get_bits(gb);
595

    
596
    if (get_bits_left(gb) < 8 * count) {
597
        av_log(ac->avctx, AV_LOG_ERROR, overread_err);
598
        return -1;
599
    }
600
    skip_bits_long(gb, 8 * count);
601
    return 0;
602
}
603

    
604
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
605
                             GetBitContext *gb)
606
{
607
    int sfb;
608
    if (get_bits1(gb)) {
609
        ics->predictor_reset_group = get_bits(gb, 5);
610
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
611
            av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
612
            return -1;
613
        }
614
    }
615
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
616
        ics->prediction_used[sfb] = get_bits1(gb);
617
    }
618
    return 0;
619
}
620

    
621
/**
622
 * Decode Individual Channel Stream info; reference: table 4.6.
623
 *
624
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
625
 */
626
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
627
                           GetBitContext *gb, int common_window)
628
{
629
    if (get_bits1(gb)) {
630
        av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
631
        memset(ics, 0, sizeof(IndividualChannelStream));
632
        return -1;
633
    }
634
    ics->window_sequence[1] = ics->window_sequence[0];
635
    ics->window_sequence[0] = get_bits(gb, 2);
636
    ics->use_kb_window[1]   = ics->use_kb_window[0];
637
    ics->use_kb_window[0]   = get_bits1(gb);
638
    ics->num_window_groups  = 1;
639
    ics->group_len[0]       = 1;
640
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
641
        int i;
642
        ics->max_sfb = get_bits(gb, 4);
643
        for (i = 0; i < 7; i++) {
644
            if (get_bits1(gb)) {
645
                ics->group_len[ics->num_window_groups - 1]++;
646
            } else {
647
                ics->num_window_groups++;
648
                ics->group_len[ics->num_window_groups - 1] = 1;
649
            }
650
        }
651
        ics->num_windows       = 8;
652
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
653
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
654
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
655
        ics->predictor_present = 0;
656
    } else {
657
        ics->max_sfb               = get_bits(gb, 6);
658
        ics->num_windows           = 1;
659
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
660
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
661
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
662
        ics->predictor_present     = get_bits1(gb);
663
        ics->predictor_reset_group = 0;
664
        if (ics->predictor_present) {
665
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
666
                if (decode_prediction(ac, ics, gb)) {
667
                    memset(ics, 0, sizeof(IndividualChannelStream));
668
                    return -1;
669
                }
670
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
671
                av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
672
                memset(ics, 0, sizeof(IndividualChannelStream));
673
                return -1;
674
            } else {
675
                av_log_missing_feature(ac->avctx, "Predictor bit set but LTP is", 1);
676
                memset(ics, 0, sizeof(IndividualChannelStream));
677
                return -1;
678
            }
679
        }
680
    }
681

    
682
    if (ics->max_sfb > ics->num_swb) {
683
        av_log(ac->avctx, AV_LOG_ERROR,
684
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
685
               ics->max_sfb, ics->num_swb);
686
        memset(ics, 0, sizeof(IndividualChannelStream));
687
        return -1;
688
    }
689

    
690
    return 0;
691
}
692

    
693
/**
694
 * Decode band types (section_data payload); reference: table 4.46.
695
 *
696
 * @param   band_type           array of the used band type
697
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
698
 *
699
 * @return  Returns error status. 0 - OK, !0 - error
700
 */
701
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
702
                             int band_type_run_end[120], GetBitContext *gb,
703
                             IndividualChannelStream *ics)
704
{
705
    int g, idx = 0;
706
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
707
    for (g = 0; g < ics->num_window_groups; g++) {
708
        int k = 0;
709
        while (k < ics->max_sfb) {
710
            uint8_t sect_end = k;
711
            int sect_len_incr;
712
            int sect_band_type = get_bits(gb, 4);
713
            if (sect_band_type == 12) {
714
                av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
715
                return -1;
716
            }
717
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
718
                sect_end += sect_len_incr;
719
            sect_end += sect_len_incr;
720
            if (get_bits_left(gb) < 0) {
721
                av_log(ac->avctx, AV_LOG_ERROR, overread_err);
722
                return -1;
723
            }
724
            if (sect_end > ics->max_sfb) {
725
                av_log(ac->avctx, AV_LOG_ERROR,
726
                       "Number of bands (%d) exceeds limit (%d).\n",
727
                       sect_end, ics->max_sfb);
728
                return -1;
729
            }
730
            for (; k < sect_end; k++) {
731
                band_type        [idx]   = sect_band_type;
732
                band_type_run_end[idx++] = sect_end;
733
            }
734
        }
735
    }
736
    return 0;
737
}
738

    
739
/**
740
 * Decode scalefactors; reference: table 4.47.
741
 *
742
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
743
 * @param   band_type           array of the used band type
744
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
745
 * @param   sf                  array of scalefactors or intensity stereo positions
746
 *
747
 * @return  Returns error status. 0 - OK, !0 - error
748
 */
749
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
750
                               unsigned int global_gain,
751
                               IndividualChannelStream *ics,
752
                               enum BandType band_type[120],
753
                               int band_type_run_end[120])
754
{
755
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
756
    int g, i, idx = 0;
757
    int offset[3] = { global_gain, global_gain - 90, 100 };
758
    int noise_flag = 1;
759
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
760
    for (g = 0; g < ics->num_window_groups; g++) {
761
        for (i = 0; i < ics->max_sfb;) {
762
            int run_end = band_type_run_end[idx];
763
            if (band_type[idx] == ZERO_BT) {
764
                for (; i < run_end; i++, idx++)
765
                    sf[idx] = 0.;
766
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
767
                for (; i < run_end; i++, idx++) {
768
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
769
                    if (offset[2] > 255U) {
770
                        av_log(ac->avctx, AV_LOG_ERROR,
771
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
772
                        return -1;
773
                    }
774
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
775
                }
776
            } else if (band_type[idx] == NOISE_BT) {
777
                for (; i < run_end; i++, idx++) {
778
                    if (noise_flag-- > 0)
779
                        offset[1] += get_bits(gb, 9) - 256;
780
                    else
781
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
782
                    if (offset[1] > 255U) {
783
                        av_log(ac->avctx, AV_LOG_ERROR,
784
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
785
                        return -1;
786
                    }
787
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
788
                }
789
            } else {
790
                for (; i < run_end; i++, idx++) {
791
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
792
                    if (offset[0] > 255U) {
793
                        av_log(ac->avctx, AV_LOG_ERROR,
794
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
795
                        return -1;
796
                    }
797
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
798
                }
799
            }
800
        }
801
    }
802
    return 0;
803
}
804

    
805
/**
806
 * Decode pulse data; reference: table 4.7.
807
 */
808
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
809
                         const uint16_t *swb_offset, int num_swb)
810
{
811
    int i, pulse_swb;
812
    pulse->num_pulse = get_bits(gb, 2) + 1;
813
    pulse_swb        = get_bits(gb, 6);
814
    if (pulse_swb >= num_swb)
815
        return -1;
816
    pulse->pos[0]    = swb_offset[pulse_swb];
817
    pulse->pos[0]   += get_bits(gb, 5);
818
    if (pulse->pos[0] > 1023)
819
        return -1;
820
    pulse->amp[0]    = get_bits(gb, 4);
821
    for (i = 1; i < pulse->num_pulse; i++) {
822
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
823
        if (pulse->pos[i] > 1023)
824
            return -1;
825
        pulse->amp[i] = get_bits(gb, 4);
826
    }
827
    return 0;
828
}
829

    
830
/**
831
 * Decode Temporal Noise Shaping data; reference: table 4.48.
832
 *
833
 * @return  Returns error status. 0 - OK, !0 - error
834
 */
835
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
836
                      GetBitContext *gb, const IndividualChannelStream *ics)
837
{
838
    int w, filt, i, coef_len, coef_res, coef_compress;
839
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
840
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
841
    for (w = 0; w < ics->num_windows; w++) {
842
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
843
            coef_res = get_bits1(gb);
844

    
845
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
846
                int tmp2_idx;
847
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
848

    
849
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
850
                    av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
851
                           tns->order[w][filt], tns_max_order);
852
                    tns->order[w][filt] = 0;
853
                    return -1;
854
                }
855
                if (tns->order[w][filt]) {
856
                    tns->direction[w][filt] = get_bits1(gb);
857
                    coef_compress = get_bits1(gb);
858
                    coef_len = coef_res + 3 - coef_compress;
859
                    tmp2_idx = 2 * coef_compress + coef_res;
860

    
861
                    for (i = 0; i < tns->order[w][filt]; i++)
862
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
863
                }
864
            }
865
        }
866
    }
867
    return 0;
868
}
869

    
870
/**
871
 * Decode Mid/Side data; reference: table 4.54.
872
 *
873
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
874
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
875
 *                      [3] reserved for scalable AAC
876
 */
877
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
878
                                   int ms_present)
879
{
880
    int idx;
881
    if (ms_present == 1) {
882
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
883
            cpe->ms_mask[idx] = get_bits1(gb);
884
    } else if (ms_present == 2) {
885
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
886
    }
887
}
888

    
889
#ifndef VMUL2
890
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
891
                           const float *scale)
892
{
893
    float s = *scale;
894
    *dst++ = v[idx    & 15] * s;
895
    *dst++ = v[idx>>4 & 15] * s;
896
    return dst;
897
}
898
#endif
899

    
900
#ifndef VMUL4
901
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
902
                           const float *scale)
903
{
904
    float s = *scale;
905
    *dst++ = v[idx    & 3] * s;
906
    *dst++ = v[idx>>2 & 3] * s;
907
    *dst++ = v[idx>>4 & 3] * s;
908
    *dst++ = v[idx>>6 & 3] * s;
909
    return dst;
910
}
911
#endif
912

    
913
#ifndef VMUL2S
914
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
915
                            unsigned sign, const float *scale)
916
{
917
    union float754 s0, s1;
918

    
919
    s0.f = s1.f = *scale;
920
    s0.i ^= sign >> 1 << 31;
921
    s1.i ^= sign      << 31;
922

    
923
    *dst++ = v[idx    & 15] * s0.f;
924
    *dst++ = v[idx>>4 & 15] * s1.f;
925

    
926
    return dst;
927
}
928
#endif
929

    
930
#ifndef VMUL4S
931
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
932
                            unsigned sign, const float *scale)
933
{
934
    unsigned nz = idx >> 12;
935
    union float754 s = { .f = *scale };
936
    union float754 t;
937

    
938
    t.i = s.i ^ (sign & 1<<31);
939
    *dst++ = v[idx    & 3] * t.f;
940

    
941
    sign <<= nz & 1; nz >>= 1;
942
    t.i = s.i ^ (sign & 1<<31);
943
    *dst++ = v[idx>>2 & 3] * t.f;
944

    
945
    sign <<= nz & 1; nz >>= 1;
946
    t.i = s.i ^ (sign & 1<<31);
947
    *dst++ = v[idx>>4 & 3] * t.f;
948

    
949
    sign <<= nz & 1; nz >>= 1;
950
    t.i = s.i ^ (sign & 1<<31);
951
    *dst++ = v[idx>>6 & 3] * t.f;
952

    
953
    return dst;
954
}
955
#endif
956

    
957
/**
958
 * Decode spectral data; reference: table 4.50.
959
 * Dequantize and scale spectral data; reference: 4.6.3.3.
960
 *
961
 * @param   coef            array of dequantized, scaled spectral data
962
 * @param   sf              array of scalefactors or intensity stereo positions
963
 * @param   pulse_present   set if pulses are present
964
 * @param   pulse           pointer to pulse data struct
965
 * @param   band_type       array of the used band type
966
 *
967
 * @return  Returns error status. 0 - OK, !0 - error
968
 */
969
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
970
                                       GetBitContext *gb, const float sf[120],
971
                                       int pulse_present, const Pulse *pulse,
972
                                       const IndividualChannelStream *ics,
973
                                       enum BandType band_type[120])
974
{
975
    int i, k, g, idx = 0;
976
    const int c = 1024 / ics->num_windows;
977
    const uint16_t *offsets = ics->swb_offset;
978
    float *coef_base = coef;
979

    
980
    for (g = 0; g < ics->num_windows; g++)
981
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
982

    
983
    for (g = 0; g < ics->num_window_groups; g++) {
984
        unsigned g_len = ics->group_len[g];
985

    
986
        for (i = 0; i < ics->max_sfb; i++, idx++) {
987
            const unsigned cbt_m1 = band_type[idx] - 1;
988
            float *cfo = coef + offsets[i];
989
            int off_len = offsets[i + 1] - offsets[i];
990
            int group;
991

    
992
            if (cbt_m1 >= INTENSITY_BT2 - 1) {
993
                for (group = 0; group < g_len; group++, cfo+=128) {
994
                    memset(cfo, 0, off_len * sizeof(float));
995
                }
996
            } else if (cbt_m1 == NOISE_BT - 1) {
997
                for (group = 0; group < g_len; group++, cfo+=128) {
998
                    float scale;
999
                    float band_energy;
1000

    
1001
                    for (k = 0; k < off_len; k++) {
1002
                        ac->random_state  = lcg_random(ac->random_state);
1003
                        cfo[k] = ac->random_state;
1004
                    }
1005

    
1006
                    band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1007
                    scale = sf[idx] / sqrtf(band_energy);
1008
                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1009
                }
1010
            } else {
1011
                const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1012
                const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1013
                VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1014
                OPEN_READER(re, gb);
1015

    
1016
                switch (cbt_m1 >> 1) {
1017
                case 0:
1018
                    for (group = 0; group < g_len; group++, cfo+=128) {
1019
                        float *cf = cfo;
1020
                        int len = off_len;
1021

    
1022
                        do {
1023
                            int code;
1024
                            unsigned cb_idx;
1025

    
1026
                            UPDATE_CACHE(re, gb);
1027
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1028
                            cb_idx = cb_vector_idx[code];
1029
                            cf = VMUL4(cf, vq, cb_idx, sf + idx);
1030
                        } while (len -= 4);
1031
                    }
1032
                    break;
1033

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

    
1039
                        do {
1040
                            int code;
1041
                            unsigned nnz;
1042
                            unsigned cb_idx;
1043
                            uint32_t bits;
1044

    
1045
                            UPDATE_CACHE(re, gb);
1046
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1047
#if MIN_CACHE_BITS < 20
1048
                            UPDATE_CACHE(re, gb);
1049
#endif
1050
                            cb_idx = cb_vector_idx[code];
1051
                            nnz = cb_idx >> 8 & 15;
1052
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1053
                            LAST_SKIP_BITS(re, gb, nnz);
1054
                            cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1055
                        } while (len -= 4);
1056
                    }
1057
                    break;
1058

    
1059
                case 2:
1060
                    for (group = 0; group < g_len; group++, cfo+=128) {
1061
                        float *cf = cfo;
1062
                        int len = off_len;
1063

    
1064
                        do {
1065
                            int code;
1066
                            unsigned cb_idx;
1067

    
1068
                            UPDATE_CACHE(re, gb);
1069
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1070
                            cb_idx = cb_vector_idx[code];
1071
                            cf = VMUL2(cf, vq, cb_idx, sf + idx);
1072
                        } while (len -= 2);
1073
                    }
1074
                    break;
1075

    
1076
                case 3:
1077
                case 4:
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 nnz;
1085
                            unsigned cb_idx;
1086
                            unsigned sign;
1087

    
1088
                            UPDATE_CACHE(re, gb);
1089
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1090
                            cb_idx = cb_vector_idx[code];
1091
                            nnz = cb_idx >> 8 & 15;
1092
                            sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1093
                            LAST_SKIP_BITS(re, gb, nnz);
1094
                            cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1095
                        } while (len -= 2);
1096
                    }
1097
                    break;
1098

    
1099
                default:
1100
                    for (group = 0; group < g_len; group++, cfo+=128) {
1101
                        float *cf = cfo;
1102
                        uint32_t *icf = (uint32_t *) cf;
1103
                        int len = off_len;
1104

    
1105
                        do {
1106
                            int code;
1107
                            unsigned nzt, nnz;
1108
                            unsigned cb_idx;
1109
                            uint32_t bits;
1110
                            int j;
1111

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

    
1115
                            if (!code) {
1116
                                *icf++ = 0;
1117
                                *icf++ = 0;
1118
                                continue;
1119
                            }
1120

    
1121
                            cb_idx = cb_vector_idx[code];
1122
                            nnz = cb_idx >> 12;
1123
                            nzt = cb_idx >> 8;
1124
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1125
                            LAST_SKIP_BITS(re, gb, nnz);
1126

    
1127
                            for (j = 0; j < 2; j++) {
1128
                                if (nzt & 1<<j) {
1129
                                    uint32_t b;
1130
                                    int n;
1131
                                    /* The total length of escape_sequence must be < 22 bits according
1132
                                       to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1133
                                    UPDATE_CACHE(re, gb);
1134
                                    b = GET_CACHE(re, gb);
1135
                                    b = 31 - av_log2(~b);
1136

    
1137
                                    if (b > 8) {
1138
                                        av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1139
                                        return -1;
1140
                                    }
1141

    
1142
#if MIN_CACHE_BITS < 21
1143
                                    LAST_SKIP_BITS(re, gb, b + 1);
1144
                                    UPDATE_CACHE(re, gb);
1145
#else
1146
                                    SKIP_BITS(re, gb, b + 1);
1147
#endif
1148
                                    b += 4;
1149
                                    n = (1 << b) + SHOW_UBITS(re, gb, b);
1150
                                    LAST_SKIP_BITS(re, gb, b);
1151
                                    *icf++ = cbrt_tab[n] | (bits & 1<<31);
1152
                                    bits <<= 1;
1153
                                } else {
1154
                                    unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1155
                                    *icf++ = (bits & 1<<31) | v;
1156
                                    bits <<= !!v;
1157
                                }
1158
                                cb_idx >>= 4;
1159
                            }
1160
                        } while (len -= 2);
1161

    
1162
                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1163
                    }
1164
                }
1165

    
1166
                CLOSE_READER(re, gb);
1167
            }
1168
        }
1169
        coef += g_len << 7;
1170
    }
1171

    
1172
    if (pulse_present) {
1173
        idx = 0;
1174
        for (i = 0; i < pulse->num_pulse; i++) {
1175
            float co = coef_base[ pulse->pos[i] ];
1176
            while (offsets[idx + 1] <= pulse->pos[i])
1177
                idx++;
1178
            if (band_type[idx] != NOISE_BT && sf[idx]) {
1179
                float ico = -pulse->amp[i];
1180
                if (co) {
1181
                    co /= sf[idx];
1182
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1183
                }
1184
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1185
            }
1186
        }
1187
    }
1188
    return 0;
1189
}
1190

    
1191
static av_always_inline float flt16_round(float pf)
1192
{
1193
    union float754 tmp;
1194
    tmp.f = pf;
1195
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1196
    return tmp.f;
1197
}
1198

    
1199
static av_always_inline float flt16_even(float pf)
1200
{
1201
    union float754 tmp;
1202
    tmp.f = pf;
1203
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1204
    return tmp.f;
1205
}
1206

    
1207
static av_always_inline float flt16_trunc(float pf)
1208
{
1209
    union float754 pun;
1210
    pun.f = pf;
1211
    pun.i &= 0xFFFF0000U;
1212
    return pun.f;
1213
}
1214

    
1215
static av_always_inline void predict(PredictorState *ps, float *coef,
1216
                                     float sf_scale, float inv_sf_scale,
1217
                    int output_enable)
1218
{
1219
    const float a     = 0.953125; // 61.0 / 64
1220
    const float alpha = 0.90625;  // 29.0 / 32
1221
    float e0, e1;
1222
    float pv;
1223
    float k1, k2;
1224
    float   r0 = ps->r0,     r1 = ps->r1;
1225
    float cor0 = ps->cor0, cor1 = ps->cor1;
1226
    float var0 = ps->var0, var1 = ps->var1;
1227

    
1228
    k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1229
    k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1230

    
1231
    pv = flt16_round(k1 * r0 + k2 * r1);
1232
    if (output_enable)
1233
        *coef += pv * sf_scale;
1234

    
1235
    e0 = *coef * inv_sf_scale;
1236
    e1 = e0 - k1 * r0;
1237

    
1238
    ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1239
    ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1240
    ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1241
    ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1242

    
1243
    ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1244
    ps->r0 = flt16_trunc(a * e0);
1245
}
1246

    
1247
/**
1248
 * Apply AAC-Main style frequency domain prediction.
1249
 */
1250
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1251
{
1252
    int sfb, k;
1253
    float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1254

    
1255
    if (!sce->ics.predictor_initialized) {
1256
        reset_all_predictors(sce->predictor_state);
1257
        sce->ics.predictor_initialized = 1;
1258
    }
1259

    
1260
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1261
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1262
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1263
                predict(&sce->predictor_state[k], &sce->coeffs[k],
1264
                        sf_scale, inv_sf_scale,
1265
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1266
            }
1267
        }
1268
        if (sce->ics.predictor_reset_group)
1269
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1270
    } else
1271
        reset_all_predictors(sce->predictor_state);
1272
}
1273

    
1274
/**
1275
 * Decode an individual_channel_stream payload; reference: table 4.44.
1276
 *
1277
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1278
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1279
 *
1280
 * @return  Returns error status. 0 - OK, !0 - error
1281
 */
1282
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1283
                      GetBitContext *gb, int common_window, int scale_flag)
1284
{
1285
    Pulse pulse;
1286
    TemporalNoiseShaping    *tns = &sce->tns;
1287
    IndividualChannelStream *ics = &sce->ics;
1288
    float *out = sce->coeffs;
1289
    int global_gain, pulse_present = 0;
1290

    
1291
    /* This assignment is to silence a GCC warning about the variable being used
1292
     * uninitialized when in fact it always is.
1293
     */
1294
    pulse.num_pulse = 0;
1295

    
1296
    global_gain = get_bits(gb, 8);
1297

    
1298
    if (!common_window && !scale_flag) {
1299
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1300
            return -1;
1301
    }
1302

    
1303
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1304
        return -1;
1305
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1306
        return -1;
1307

    
1308
    pulse_present = 0;
1309
    if (!scale_flag) {
1310
        if ((pulse_present = get_bits1(gb))) {
1311
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1312
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1313
                return -1;
1314
            }
1315
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1316
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1317
                return -1;
1318
            }
1319
        }
1320
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1321
            return -1;
1322
        if (get_bits1(gb)) {
1323
            av_log_missing_feature(ac->avctx, "SSR", 1);
1324
            return -1;
1325
        }
1326
    }
1327

    
1328
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1329
        return -1;
1330

    
1331
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1332
        apply_prediction(ac, sce);
1333

    
1334
    return 0;
1335
}
1336

    
1337
/**
1338
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1339
 */
1340
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1341
{
1342
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1343
    float *ch0 = cpe->ch[0].coeffs;
1344
    float *ch1 = cpe->ch[1].coeffs;
1345
    int g, i, group, idx = 0;
1346
    const uint16_t *offsets = ics->swb_offset;
1347
    for (g = 0; g < ics->num_window_groups; g++) {
1348
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1349
            if (cpe->ms_mask[idx] &&
1350
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1351
                for (group = 0; group < ics->group_len[g]; group++) {
1352
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1353
                                              ch1 + group * 128 + offsets[i],
1354
                                              offsets[i+1] - offsets[i]);
1355
                }
1356
            }
1357
        }
1358
        ch0 += ics->group_len[g] * 128;
1359
        ch1 += ics->group_len[g] * 128;
1360
    }
1361
}
1362

    
1363
/**
1364
 * intensity stereo decoding; reference: 4.6.8.2.3
1365
 *
1366
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1367
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1368
 *                      [3] reserved for scalable AAC
1369
 */
1370
static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1371
{
1372
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1373
    SingleChannelElement         *sce1 = &cpe->ch[1];
1374
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1375
    const uint16_t *offsets = ics->swb_offset;
1376
    int g, group, i, k, idx = 0;
1377
    int c;
1378
    float scale;
1379
    for (g = 0; g < ics->num_window_groups; g++) {
1380
        for (i = 0; i < ics->max_sfb;) {
1381
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1382
                const int bt_run_end = sce1->band_type_run_end[idx];
1383
                for (; i < bt_run_end; i++, idx++) {
1384
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1385
                    if (ms_present)
1386
                        c *= 1 - 2 * cpe->ms_mask[idx];
1387
                    scale = c * sce1->sf[idx];
1388
                    for (group = 0; group < ics->group_len[g]; group++)
1389
                        for (k = offsets[i]; k < offsets[i + 1]; k++)
1390
                            coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1391
                }
1392
            } else {
1393
                int bt_run_end = sce1->band_type_run_end[idx];
1394
                idx += bt_run_end - i;
1395
                i    = bt_run_end;
1396
            }
1397
        }
1398
        coef0 += ics->group_len[g] * 128;
1399
        coef1 += ics->group_len[g] * 128;
1400
    }
1401
}
1402

    
1403
/**
1404
 * Decode a channel_pair_element; reference: table 4.4.
1405
 *
1406
 * @return  Returns error status. 0 - OK, !0 - error
1407
 */
1408
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1409
{
1410
    int i, ret, common_window, ms_present = 0;
1411

    
1412
    common_window = get_bits1(gb);
1413
    if (common_window) {
1414
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1415
            return -1;
1416
        i = cpe->ch[1].ics.use_kb_window[0];
1417
        cpe->ch[1].ics = cpe->ch[0].ics;
1418
        cpe->ch[1].ics.use_kb_window[1] = i;
1419
        ms_present = get_bits(gb, 2);
1420
        if (ms_present == 3) {
1421
            av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1422
            return -1;
1423
        } else if (ms_present)
1424
            decode_mid_side_stereo(cpe, gb, ms_present);
1425
    }
1426
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1427
        return ret;
1428
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1429
        return ret;
1430

    
1431
    if (common_window) {
1432
        if (ms_present)
1433
            apply_mid_side_stereo(ac, cpe);
1434
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1435
            apply_prediction(ac, &cpe->ch[0]);
1436
            apply_prediction(ac, &cpe->ch[1]);
1437
        }
1438
    }
1439

    
1440
    apply_intensity_stereo(cpe, ms_present);
1441
    return 0;
1442
}
1443

    
1444
static const float cce_scale[] = {
1445
    1.09050773266525765921, //2^(1/8)
1446
    1.18920711500272106672, //2^(1/4)
1447
    M_SQRT2,
1448
    2,
1449
};
1450

    
1451
/**
1452
 * Decode coupling_channel_element; reference: table 4.8.
1453
 *
1454
 * @return  Returns error status. 0 - OK, !0 - error
1455
 */
1456
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1457
{
1458
    int num_gain = 0;
1459
    int c, g, sfb, ret;
1460
    int sign;
1461
    float scale;
1462
    SingleChannelElement *sce = &che->ch[0];
1463
    ChannelCoupling     *coup = &che->coup;
1464

    
1465
    coup->coupling_point = 2 * get_bits1(gb);
1466
    coup->num_coupled = get_bits(gb, 3);
1467
    for (c = 0; c <= coup->num_coupled; c++) {
1468
        num_gain++;
1469
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1470
        coup->id_select[c] = get_bits(gb, 4);
1471
        if (coup->type[c] == TYPE_CPE) {
1472
            coup->ch_select[c] = get_bits(gb, 2);
1473
            if (coup->ch_select[c] == 3)
1474
                num_gain++;
1475
        } else
1476
            coup->ch_select[c] = 2;
1477
    }
1478
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1479

    
1480
    sign  = get_bits(gb, 1);
1481
    scale = cce_scale[get_bits(gb, 2)];
1482

    
1483
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1484
        return ret;
1485

    
1486
    for (c = 0; c < num_gain; c++) {
1487
        int idx  = 0;
1488
        int cge  = 1;
1489
        int gain = 0;
1490
        float gain_cache = 1.;
1491
        if (c) {
1492
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1493
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1494
            gain_cache = powf(scale, -gain);
1495
        }
1496
        if (coup->coupling_point == AFTER_IMDCT) {
1497
            coup->gain[c][0] = gain_cache;
1498
        } else {
1499
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1500
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1501
                    if (sce->band_type[idx] != ZERO_BT) {
1502
                        if (!cge) {
1503
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1504
                            if (t) {
1505
                                int s = 1;
1506
                                t = gain += t;
1507
                                if (sign) {
1508
                                    s  -= 2 * (t & 0x1);
1509
                                    t >>= 1;
1510
                                }
1511
                                gain_cache = powf(scale, -t) * s;
1512
                            }
1513
                        }
1514
                        coup->gain[c][idx] = gain_cache;
1515
                    }
1516
                }
1517
            }
1518
        }
1519
    }
1520
    return 0;
1521
}
1522

    
1523
/**
1524
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1525
 *
1526
 * @return  Returns number of bytes consumed.
1527
 */
1528
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1529
                                         GetBitContext *gb)
1530
{
1531
    int i;
1532
    int num_excl_chan = 0;
1533

    
1534
    do {
1535
        for (i = 0; i < 7; i++)
1536
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1537
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1538

    
1539
    return num_excl_chan / 7;
1540
}
1541

    
1542
/**
1543
 * Decode dynamic range information; reference: table 4.52.
1544
 *
1545
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1546
 *
1547
 * @return  Returns number of bytes consumed.
1548
 */
1549
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1550
                                GetBitContext *gb, int cnt)
1551
{
1552
    int n             = 1;
1553
    int drc_num_bands = 1;
1554
    int i;
1555

    
1556
    /* pce_tag_present? */
1557
    if (get_bits1(gb)) {
1558
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1559
        skip_bits(gb, 4); // tag_reserved_bits
1560
        n++;
1561
    }
1562

    
1563
    /* excluded_chns_present? */
1564
    if (get_bits1(gb)) {
1565
        n += decode_drc_channel_exclusions(che_drc, gb);
1566
    }
1567

    
1568
    /* drc_bands_present? */
1569
    if (get_bits1(gb)) {
1570
        che_drc->band_incr            = get_bits(gb, 4);
1571
        che_drc->interpolation_scheme = get_bits(gb, 4);
1572
        n++;
1573
        drc_num_bands += che_drc->band_incr;
1574
        for (i = 0; i < drc_num_bands; i++) {
1575
            che_drc->band_top[i] = get_bits(gb, 8);
1576
            n++;
1577
        }
1578
    }
1579

    
1580
    /* prog_ref_level_present? */
1581
    if (get_bits1(gb)) {
1582
        che_drc->prog_ref_level = get_bits(gb, 7);
1583
        skip_bits1(gb); // prog_ref_level_reserved_bits
1584
        n++;
1585
    }
1586

    
1587
    for (i = 0; i < drc_num_bands; i++) {
1588
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1589
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1590
        n++;
1591
    }
1592

    
1593
    return n;
1594
}
1595

    
1596
/**
1597
 * Decode extension data (incomplete); reference: table 4.51.
1598
 *
1599
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1600
 *
1601
 * @return Returns number of bytes consumed
1602
 */
1603
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1604
                                    ChannelElement *che, enum RawDataBlockType elem_type)
1605
{
1606
    int crc_flag = 0;
1607
    int res = cnt;
1608
    switch (get_bits(gb, 4)) { // extension type
1609
    case EXT_SBR_DATA_CRC:
1610
        crc_flag++;
1611
    case EXT_SBR_DATA:
1612
        if (!che) {
1613
            av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1614
            return res;
1615
        } else if (!ac->m4ac.sbr) {
1616
            av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1617
            skip_bits_long(gb, 8 * cnt - 4);
1618
            return res;
1619
        } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1620
            av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1621
            skip_bits_long(gb, 8 * cnt - 4);
1622
            return res;
1623
        } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1624
            ac->m4ac.sbr = 1;
1625
            ac->m4ac.ps = 1;
1626
            output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1627
        } else {
1628
            ac->m4ac.sbr = 1;
1629
        }
1630
        res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1631
        break;
1632
    case EXT_DYNAMIC_RANGE:
1633
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1634
        break;
1635
    case EXT_FILL:
1636
    case EXT_FILL_DATA:
1637
    case EXT_DATA_ELEMENT:
1638
    default:
1639
        skip_bits_long(gb, 8 * cnt - 4);
1640
        break;
1641
    };
1642
    return res;
1643
}
1644

    
1645
/**
1646
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1647
 *
1648
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1649
 * @param   coef    spectral coefficients
1650
 */
1651
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1652
                      IndividualChannelStream *ics, int decode)
1653
{
1654
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1655
    int w, filt, m, i;
1656
    int bottom, top, order, start, end, size, inc;
1657
    float lpc[TNS_MAX_ORDER];
1658

    
1659
    for (w = 0; w < ics->num_windows; w++) {
1660
        bottom = ics->num_swb;
1661
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1662
            top    = bottom;
1663
            bottom = FFMAX(0, top - tns->length[w][filt]);
1664
            order  = tns->order[w][filt];
1665
            if (order == 0)
1666
                continue;
1667

    
1668
            // tns_decode_coef
1669
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1670

    
1671
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1672
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1673
            if ((size = end - start) <= 0)
1674
                continue;
1675
            if (tns->direction[w][filt]) {
1676
                inc = -1;
1677
                start = end - 1;
1678
            } else {
1679
                inc = 1;
1680
            }
1681
            start += w * 128;
1682

    
1683
            // ar filter
1684
            for (m = 0; m < size; m++, start += inc)
1685
                for (i = 1; i <= FFMIN(m, order); i++)
1686
                    coef[start] -= coef[start - i * inc] * lpc[i - 1];
1687
        }
1688
    }
1689
}
1690

    
1691
/**
1692
 * Conduct IMDCT and windowing.
1693
 */
1694
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce, float bias)
1695
{
1696
    IndividualChannelStream *ics = &sce->ics;
1697
    float *in    = sce->coeffs;
1698
    float *out   = sce->ret;
1699
    float *saved = sce->saved;
1700
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1701
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1702
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1703
    float *buf  = ac->buf_mdct;
1704
    float *temp = ac->temp;
1705
    int i;
1706

    
1707
    // imdct
1708
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1709
        for (i = 0; i < 1024; i += 128)
1710
            ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1711
    } else
1712
        ff_imdct_half(&ac->mdct, buf, in);
1713

    
1714
    /* window overlapping
1715
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1716
     * and long to short transitions are considered to be short to short
1717
     * transitions. This leaves just two cases (long to long and short to short)
1718
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1719
     */
1720
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1721
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1722
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, bias, 512);
1723
    } else {
1724
        for (i = 0; i < 448; i++)
1725
            out[i] = saved[i] + bias;
1726

    
1727
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1728
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, bias, 64);
1729
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      bias, 64);
1730
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      bias, 64);
1731
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      bias, 64);
1732
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      bias, 64);
1733
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1734
        } else {
1735
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, bias, 64);
1736
            for (i = 576; i < 1024; i++)
1737
                out[i] = buf[i-512] + bias;
1738
        }
1739
    }
1740

    
1741
    // buffer update
1742
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1743
        for (i = 0; i < 64; i++)
1744
            saved[i] = temp[64 + i] - bias;
1745
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 0, 64);
1746
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 0, 64);
1747
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 0, 64);
1748
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1749
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1750
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1751
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1752
    } else { // LONG_STOP or ONLY_LONG
1753
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1754
    }
1755
}
1756

    
1757
/**
1758
 * Apply dependent channel coupling (applied before IMDCT).
1759
 *
1760
 * @param   index   index into coupling gain array
1761
 */
1762
static void apply_dependent_coupling(AACContext *ac,
1763
                                     SingleChannelElement *target,
1764
                                     ChannelElement *cce, int index)
1765
{
1766
    IndividualChannelStream *ics = &cce->ch[0].ics;
1767
    const uint16_t *offsets = ics->swb_offset;
1768
    float *dest = target->coeffs;
1769
    const float *src = cce->ch[0].coeffs;
1770
    int g, i, group, k, idx = 0;
1771
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1772
        av_log(ac->avctx, AV_LOG_ERROR,
1773
               "Dependent coupling is not supported together with LTP\n");
1774
        return;
1775
    }
1776
    for (g = 0; g < ics->num_window_groups; g++) {
1777
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1778
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1779
                const float gain = cce->coup.gain[index][idx];
1780
                for (group = 0; group < ics->group_len[g]; group++) {
1781
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1782
                        // XXX dsputil-ize
1783
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1784
                    }
1785
                }
1786
            }
1787
        }
1788
        dest += ics->group_len[g] * 128;
1789
        src  += ics->group_len[g] * 128;
1790
    }
1791
}
1792

    
1793
/**
1794
 * Apply independent channel coupling (applied after IMDCT).
1795
 *
1796
 * @param   index   index into coupling gain array
1797
 */
1798
static void apply_independent_coupling(AACContext *ac,
1799
                                       SingleChannelElement *target,
1800
                                       ChannelElement *cce, int index)
1801
{
1802
    int i;
1803
    const float gain = cce->coup.gain[index][0];
1804
    const float bias = ac->add_bias;
1805
    const float *src = cce->ch[0].ret;
1806
    float *dest = target->ret;
1807
    const int len = 1024 << (ac->m4ac.sbr == 1);
1808

    
1809
    for (i = 0; i < len; i++)
1810
        dest[i] += gain * (src[i] - bias);
1811
}
1812

    
1813
/**
1814
 * channel coupling transformation interface
1815
 *
1816
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1817
 */
1818
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1819
                                   enum RawDataBlockType type, int elem_id,
1820
                                   enum CouplingPoint coupling_point,
1821
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1822
{
1823
    int i, c;
1824

    
1825
    for (i = 0; i < MAX_ELEM_ID; i++) {
1826
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1827
        int index = 0;
1828

    
1829
        if (cce && cce->coup.coupling_point == coupling_point) {
1830
            ChannelCoupling *coup = &cce->coup;
1831

    
1832
            for (c = 0; c <= coup->num_coupled; c++) {
1833
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1834
                    if (coup->ch_select[c] != 1) {
1835
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1836
                        if (coup->ch_select[c] != 0)
1837
                            index++;
1838
                    }
1839
                    if (coup->ch_select[c] != 2)
1840
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1841
                } else
1842
                    index += 1 + (coup->ch_select[c] == 3);
1843
            }
1844
        }
1845
    }
1846
}
1847

    
1848
/**
1849
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1850
 */
1851
static void spectral_to_sample(AACContext *ac)
1852
{
1853
    int i, type;
1854
    float imdct_bias = (ac->m4ac.sbr <= 0) ? ac->add_bias : 0.0f;
1855
    for (type = 3; type >= 0; type--) {
1856
        for (i = 0; i < MAX_ELEM_ID; i++) {
1857
            ChannelElement *che = ac->che[type][i];
1858
            if (che) {
1859
                if (type <= TYPE_CPE)
1860
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1861
                if (che->ch[0].tns.present)
1862
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1863
                if (che->ch[1].tns.present)
1864
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1865
                if (type <= TYPE_CPE)
1866
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1867
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1868
                    imdct_and_windowing(ac, &che->ch[0], imdct_bias);
1869
                    if (type == TYPE_CPE) {
1870
                        imdct_and_windowing(ac, &che->ch[1], imdct_bias);
1871
                    }
1872
                    if (ac->m4ac.sbr > 0) {
1873
                        ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
1874
                    }
1875
                }
1876
                if (type <= TYPE_CCE)
1877
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1878
            }
1879
        }
1880
    }
1881
}
1882

    
1883
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1884
{
1885
    int size;
1886
    AACADTSHeaderInfo hdr_info;
1887

    
1888
    size = ff_aac_parse_header(gb, &hdr_info);
1889
    if (size > 0) {
1890
        if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1891
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1892
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1893
            ac->m4ac.chan_config = hdr_info.chan_config;
1894
            if (set_default_channel_config(ac, new_che_pos, hdr_info.chan_config))
1895
                return -7;
1896
            if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1897
                return -7;
1898
        } else if (ac->output_configured != OC_LOCKED) {
1899
            ac->output_configured = OC_NONE;
1900
        }
1901
        if (ac->output_configured != OC_LOCKED) {
1902
            ac->m4ac.sbr = -1;
1903
            ac->m4ac.ps  = -1;
1904
        }
1905
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
1906
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
1907
        ac->m4ac.object_type     = hdr_info.object_type;
1908
        if (!ac->avctx->sample_rate)
1909
            ac->avctx->sample_rate = hdr_info.sample_rate;
1910
        if (hdr_info.num_aac_frames == 1) {
1911
            if (!hdr_info.crc_absent)
1912
                skip_bits(gb, 16);
1913
        } else {
1914
            av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
1915
            return -1;
1916
        }
1917
    }
1918
    return size;
1919
}
1920

    
1921
static int aac_decode_frame(AVCodecContext *avctx, void *data,
1922
                            int *data_size, AVPacket *avpkt)
1923
{
1924
    const uint8_t *buf = avpkt->data;
1925
    int buf_size = avpkt->size;
1926
    AACContext *ac = avctx->priv_data;
1927
    ChannelElement *che = NULL, *che_prev = NULL;
1928
    GetBitContext gb;
1929
    enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1930
    int err, elem_id, data_size_tmp;
1931
    int buf_consumed;
1932
    int samples = 0, multiplier;
1933
    int buf_offset;
1934

    
1935
    init_get_bits(&gb, buf, buf_size * 8);
1936

    
1937
    if (show_bits(&gb, 12) == 0xfff) {
1938
        if (parse_adts_frame_header(ac, &gb) < 0) {
1939
            av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1940
            return -1;
1941
        }
1942
        if (ac->m4ac.sampling_index > 12) {
1943
            av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1944
            return -1;
1945
        }
1946
    }
1947

    
1948
    ac->tags_mapped = 0;
1949
    // parse
1950
    while ((elem_type = get_bits(&gb, 3)) != TYPE_END) {
1951
        elem_id = get_bits(&gb, 4);
1952

    
1953
        if (elem_type < TYPE_DSE) {
1954
            if (!(che=get_che(ac, elem_type, elem_id))) {
1955
                av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
1956
                       elem_type, elem_id);
1957
                return -1;
1958
            }
1959
            samples = 1024;
1960
        }
1961

    
1962
        switch (elem_type) {
1963

    
1964
        case TYPE_SCE:
1965
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1966
            break;
1967

    
1968
        case TYPE_CPE:
1969
            err = decode_cpe(ac, &gb, che);
1970
            break;
1971

    
1972
        case TYPE_CCE:
1973
            err = decode_cce(ac, &gb, che);
1974
            break;
1975

    
1976
        case TYPE_LFE:
1977
            err = decode_ics(ac, &che->ch[0], &gb, 0, 0);
1978
            break;
1979

    
1980
        case TYPE_DSE:
1981
            err = skip_data_stream_element(ac, &gb);
1982
            break;
1983

    
1984
        case TYPE_PCE: {
1985
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1986
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1987
            if ((err = decode_pce(ac, new_che_pos, &gb)))
1988
                break;
1989
            if (ac->output_configured > OC_TRIAL_PCE)
1990
                av_log(avctx, AV_LOG_ERROR,
1991
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1992
            else
1993
                err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1994
            break;
1995
        }
1996

    
1997
        case TYPE_FIL:
1998
            if (elem_id == 15)
1999
                elem_id += get_bits(&gb, 8) - 1;
2000
            if (get_bits_left(&gb) < 8 * elem_id) {
2001
                    av_log(avctx, AV_LOG_ERROR, overread_err);
2002
                    return -1;
2003
            }
2004
            while (elem_id > 0)
2005
                elem_id -= decode_extension_payload(ac, &gb, elem_id, che_prev, elem_type_prev);
2006
            err = 0; /* FIXME */
2007
            break;
2008

    
2009
        default:
2010
            err = -1; /* should not happen, but keeps compiler happy */
2011
            break;
2012
        }
2013

    
2014
        che_prev       = che;
2015
        elem_type_prev = elem_type;
2016

    
2017
        if (err)
2018
            return err;
2019

    
2020
        if (get_bits_left(&gb) < 3) {
2021
            av_log(avctx, AV_LOG_ERROR, overread_err);
2022
            return -1;
2023
        }
2024
    }
2025

    
2026
    spectral_to_sample(ac);
2027

    
2028
    multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2029
    samples <<= multiplier;
2030
    if (ac->output_configured < OC_LOCKED) {
2031
        avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2032
        avctx->frame_size = samples;
2033
    }
2034

    
2035
    data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2036
    if (*data_size < data_size_tmp) {
2037
        av_log(avctx, AV_LOG_ERROR,
2038
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2039
               *data_size, data_size_tmp);
2040
        return -1;
2041
    }
2042
    *data_size = data_size_tmp;
2043

    
2044
    if (samples)
2045
        ac->dsp.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2046

    
2047
    if (ac->output_configured)
2048
        ac->output_configured = OC_LOCKED;
2049

    
2050
    buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2051
    for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2052
        if (buf[buf_offset])
2053
            break;
2054

    
2055
    return buf_size > buf_offset ? buf_consumed : buf_size;
2056
}
2057

    
2058
static av_cold int aac_decode_close(AVCodecContext *avctx)
2059
{
2060
    AACContext *ac = avctx->priv_data;
2061
    int i, type;
2062

    
2063
    for (i = 0; i < MAX_ELEM_ID; i++) {
2064
        for (type = 0; type < 4; type++) {
2065
            if (ac->che[type][i])
2066
                ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2067
            av_freep(&ac->che[type][i]);
2068
        }
2069
    }
2070

    
2071
    ff_mdct_end(&ac->mdct);
2072
    ff_mdct_end(&ac->mdct_small);
2073
    return 0;
2074
}
2075

    
2076
AVCodec aac_decoder = {
2077
    "aac",
2078
    AVMEDIA_TYPE_AUDIO,
2079
    CODEC_ID_AAC,
2080
    sizeof(AACContext),
2081
    aac_decode_init,
2082
    NULL,
2083
    aac_decode_close,
2084
    aac_decode_frame,
2085
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2086
    .sample_fmts = (const enum SampleFormat[]) {
2087
        SAMPLE_FMT_S16,SAMPLE_FMT_NONE
2088
    },
2089
    .channel_layouts = aac_channel_layout,
2090
};