Statistics
| Branch: | Revision:

ffmpeg / libavcodec / aacdec.c @ a45fbda9

History | View | Annotate | Download (88.1 KB)

1
/*
2
 * AAC decoder
3
 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4
 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5
 *
6
 * AAC LATM decoder
7
 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8
 * Copyright (c) 2010      Janne Grunau <janne-ffmpeg@jannau.net>
9
 *
10
 * This file is part of Libav.
11
 *
12
 * Libav is free software; you can redistribute it and/or
13
 * modify it under the terms of the GNU Lesser General Public
14
 * License as published by the Free Software Foundation; either
15
 * version 2.1 of the License, or (at your option) any later version.
16
 *
17
 * Libav is distributed in the hope that it will be useful,
18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
20
 * Lesser General Public License for more details.
21
 *
22
 * You should have received a copy of the GNU Lesser General Public
23
 * License along with Libav; if not, write to the Free Software
24
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25
 */
26

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

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

    
82

    
83
#include "avcodec.h"
84
#include "internal.h"
85
#include "get_bits.h"
86
#include "dsputil.h"
87
#include "fft.h"
88
#include "fmtconvert.h"
89
#include "lpc.h"
90
#include "kbdwin.h"
91

    
92
#include "aac.h"
93
#include "aactab.h"
94
#include "aacdectab.h"
95
#include "cbrt_tablegen.h"
96
#include "sbr.h"
97
#include "aacsbr.h"
98
#include "mpeg4audio.h"
99
#include "aacadtsdec.h"
100

    
101
#include <assert.h>
102
#include <errno.h>
103
#include <math.h>
104
#include <string.h>
105

    
106
#if ARCH_ARM
107
#   include "arm/aac.h"
108
#endif
109

    
110
union float754 {
111
    float f;
112
    uint32_t i;
113
};
114

    
115
static VLC vlc_scalefactors;
116
static VLC vlc_spectral[11];
117

    
118
static const char overread_err[] = "Input buffer exhausted before END element found\n";
119

    
120
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
121
{
122
    // For PCE based channel configurations map the channels solely based on tags.
123
    if (!ac->m4ac.chan_config) {
124
        return ac->tag_che_map[type][elem_id];
125
    }
126
    // For indexed channel configurations map the channels solely based on position.
127
    switch (ac->m4ac.chan_config) {
128
    case 7:
129
        if (ac->tags_mapped == 3 && type == TYPE_CPE) {
130
            ac->tags_mapped++;
131
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
132
        }
133
    case 6:
134
        /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
135
           instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
136
           encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
137
        if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
138
            ac->tags_mapped++;
139
            return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
140
        }
141
    case 5:
142
        if (ac->tags_mapped == 2 && type == TYPE_CPE) {
143
            ac->tags_mapped++;
144
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
145
        }
146
    case 4:
147
        if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
148
            ac->tags_mapped++;
149
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
150
        }
151
    case 3:
152
    case 2:
153
        if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
154
            ac->tags_mapped++;
155
            return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
156
        } else if (ac->m4ac.chan_config == 2) {
157
            return NULL;
158
        }
159
    case 1:
160
        if (!ac->tags_mapped && type == TYPE_SCE) {
161
            ac->tags_mapped++;
162
            return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
163
        }
164
    default:
165
        return NULL;
166
    }
167
}
168

    
169
/**
170
 * Check for the channel element in the current channel position configuration.
171
 * If it exists, make sure the appropriate element is allocated and map the
172
 * channel order to match the internal Libav channel layout.
173
 *
174
 * @param   che_pos current channel position configuration
175
 * @param   type channel element type
176
 * @param   id channel element id
177
 * @param   channels count of the number of channels in the configuration
178
 *
179
 * @return  Returns error status. 0 - OK, !0 - error
180
 */
181
static av_cold int che_configure(AACContext *ac,
182
                         enum ChannelPosition che_pos[4][MAX_ELEM_ID],
183
                         int type, int id,
184
                         int *channels)
185
{
186
    if (che_pos[type][id]) {
187
        if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
188
            return AVERROR(ENOMEM);
189
        ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
190
        if (type != TYPE_CCE) {
191
            ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
192
            if (type == TYPE_CPE ||
193
                (type == TYPE_SCE && ac->m4ac.ps == 1)) {
194
                ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
195
            }
196
        }
197
    } else {
198
        if (ac->che[type][id])
199
            ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
200
        av_freep(&ac->che[type][id]);
201
    }
202
    return 0;
203
}
204

    
205
/**
206
 * Configure output channel order based on the current program configuration element.
207
 *
208
 * @param   che_pos current channel position configuration
209
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
210
 *
211
 * @return  Returns error status. 0 - OK, !0 - error
212
 */
213
static av_cold int output_configure(AACContext *ac,
214
                            enum ChannelPosition che_pos[4][MAX_ELEM_ID],
215
                            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
216
                            int channel_config, enum OCStatus oc_type)
217
{
218
    AVCodecContext *avctx = ac->avctx;
219
    int i, type, channels = 0, ret;
220

    
221
    if (new_che_pos != che_pos)
222
    memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
223

    
224
    if (channel_config) {
225
        for (i = 0; i < tags_per_config[channel_config]; i++) {
226
            if ((ret = che_configure(ac, che_pos,
227
                                     aac_channel_layout_map[channel_config - 1][i][0],
228
                                     aac_channel_layout_map[channel_config - 1][i][1],
229
                                     &channels)))
230
                return ret;
231
        }
232

    
233
        memset(ac->tag_che_map, 0,       4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
234

    
235
        avctx->channel_layout = aac_channel_layout[channel_config - 1];
236
    } else {
237
        /* Allocate or free elements depending on if they are in the
238
         * current program configuration.
239
         *
240
         * Set up default 1:1 output mapping.
241
         *
242
         * For a 5.1 stream the output order will be:
243
         *    [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
244
         */
245

    
246
        for (i = 0; i < MAX_ELEM_ID; i++) {
247
            for (type = 0; type < 4; type++) {
248
                if ((ret = che_configure(ac, che_pos, type, i, &channels)))
249
                    return ret;
250
            }
251
        }
252

    
253
        memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
254

    
255
        avctx->channel_layout = 0;
256
    }
257

    
258
    avctx->channels = channels;
259

    
260
    ac->output_configured = oc_type;
261

    
262
    return 0;
263
}
264

    
265
/**
266
 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
267
 *
268
 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
269
 * @param sce_map mono (Single Channel Element) map
270
 * @param type speaker type/position for these channels
271
 */
272
static void decode_channel_map(enum ChannelPosition *cpe_map,
273
                               enum ChannelPosition *sce_map,
274
                               enum ChannelPosition type,
275
                               GetBitContext *gb, int n)
276
{
277
    while (n--) {
278
        enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
279
        map[get_bits(gb, 4)] = type;
280
    }
281
}
282

    
283
/**
284
 * Decode program configuration element; reference: table 4.2.
285
 *
286
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
287
 *
288
 * @return  Returns error status. 0 - OK, !0 - error
289
 */
290
static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
291
                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
292
                      GetBitContext *gb)
293
{
294
    int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
295
    int comment_len;
296

    
297
    skip_bits(gb, 2);  // object_type
298

    
299
    sampling_index = get_bits(gb, 4);
300
    if (m4ac->sampling_index != sampling_index)
301
        av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
302

    
303
    num_front       = get_bits(gb, 4);
304
    num_side        = get_bits(gb, 4);
305
    num_back        = get_bits(gb, 4);
306
    num_lfe         = get_bits(gb, 2);
307
    num_assoc_data  = get_bits(gb, 3);
308
    num_cc          = get_bits(gb, 4);
309

    
310
    if (get_bits1(gb))
311
        skip_bits(gb, 4); // mono_mixdown_tag
312
    if (get_bits1(gb))
313
        skip_bits(gb, 4); // stereo_mixdown_tag
314

    
315
    if (get_bits1(gb))
316
        skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
317

    
318
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
319
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE,  gb, num_side );
320
    decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK,  gb, num_back );
321
    decode_channel_map(NULL,                  new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE,   gb, num_lfe  );
322

    
323
    skip_bits_long(gb, 4 * num_assoc_data);
324

    
325
    decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC,    gb, num_cc   );
326

    
327
    align_get_bits(gb);
328

    
329
    /* comment field, first byte is length */
330
    comment_len = get_bits(gb, 8) * 8;
331
    if (get_bits_left(gb) < comment_len) {
332
        av_log(avctx, AV_LOG_ERROR, overread_err);
333
        return -1;
334
    }
335
    skip_bits_long(gb, comment_len);
336
    return 0;
337
}
338

    
339
/**
340
 * Set up channel positions based on a default channel configuration
341
 * as specified in table 1.17.
342
 *
343
 * @param   new_che_pos New channel position configuration - we only do something if it differs from the current one.
344
 *
345
 * @return  Returns error status. 0 - OK, !0 - error
346
 */
347
static av_cold int set_default_channel_config(AVCodecContext *avctx,
348
                                      enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
349
                                      int channel_config)
350
{
351
    if (channel_config < 1 || channel_config > 7) {
352
        av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
353
               channel_config);
354
        return -1;
355
    }
356

    
357
    /* default channel configurations:
358
     *
359
     * 1ch : front center (mono)
360
     * 2ch : L + R (stereo)
361
     * 3ch : front center + L + R
362
     * 4ch : front center + L + R + back center
363
     * 5ch : front center + L + R + back stereo
364
     * 6ch : front center + L + R + back stereo + LFE
365
     * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
366
     */
367

    
368
    if (channel_config != 2)
369
        new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
370
    if (channel_config > 1)
371
        new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
372
    if (channel_config == 4)
373
        new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK;  // back center
374
    if (channel_config > 4)
375
        new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
376
        = AAC_CHANNEL_BACK;  // back stereo
377
    if (channel_config > 5)
378
        new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE;   // LFE
379
    if (channel_config == 7)
380
        new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
381

    
382
    return 0;
383
}
384

    
385
/**
386
 * Decode GA "General Audio" specific configuration; reference: table 4.1.
387
 *
388
 * @param   ac          pointer to AACContext, may be null
389
 * @param   avctx       pointer to AVCCodecContext, used for logging
390
 *
391
 * @return  Returns error status. 0 - OK, !0 - error
392
 */
393
static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
394
                                     GetBitContext *gb,
395
                                     MPEG4AudioConfig *m4ac,
396
                                     int channel_config)
397
{
398
    enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
399
    int extension_flag, ret;
400

    
401
    if (get_bits1(gb)) { // frameLengthFlag
402
        av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
403
        return -1;
404
    }
405

    
406
    if (get_bits1(gb))       // dependsOnCoreCoder
407
        skip_bits(gb, 14);   // coreCoderDelay
408
    extension_flag = get_bits1(gb);
409

    
410
    if (m4ac->object_type == AOT_AAC_SCALABLE ||
411
        m4ac->object_type == AOT_ER_AAC_SCALABLE)
412
        skip_bits(gb, 3);     // layerNr
413

    
414
    memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
415
    if (channel_config == 0) {
416
        skip_bits(gb, 4);  // element_instance_tag
417
        if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
418
            return ret;
419
    } else {
420
        if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
421
            return ret;
422
    }
423
    if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
424
        return ret;
425

    
426
    if (extension_flag) {
427
        switch (m4ac->object_type) {
428
        case AOT_ER_BSAC:
429
            skip_bits(gb, 5);    // numOfSubFrame
430
            skip_bits(gb, 11);   // layer_length
431
            break;
432
        case AOT_ER_AAC_LC:
433
        case AOT_ER_AAC_LTP:
434
        case AOT_ER_AAC_SCALABLE:
435
        case AOT_ER_AAC_LD:
436
            skip_bits(gb, 3);  /* aacSectionDataResilienceFlag
437
                                    * aacScalefactorDataResilienceFlag
438
                                    * aacSpectralDataResilienceFlag
439
                                    */
440
            break;
441
        }
442
        skip_bits1(gb);    // extensionFlag3 (TBD in version 3)
443
    }
444
    return 0;
445
}
446

    
447
/**
448
 * Decode audio specific configuration; reference: table 1.13.
449
 *
450
 * @param   ac          pointer to AACContext, may be null
451
 * @param   avctx       pointer to AVCCodecContext, used for logging
452
 * @param   m4ac        pointer to MPEG4AudioConfig, used for parsing
453
 * @param   data        pointer to AVCodecContext extradata
454
 * @param   data_size   size of AVCCodecContext extradata
455
 *
456
 * @return  Returns error status or number of consumed bits. <0 - error
457
 */
458
static int decode_audio_specific_config(AACContext *ac,
459
                                        AVCodecContext *avctx,
460
                                        MPEG4AudioConfig *m4ac,
461
                                        const uint8_t *data, int data_size)
462
{
463
    GetBitContext gb;
464
    int i;
465

    
466
    init_get_bits(&gb, data, data_size * 8);
467

    
468
    if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
469
        return -1;
470
    if (m4ac->sampling_index > 12) {
471
        av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
472
        return -1;
473
    }
474
    if (m4ac->sbr == 1 && m4ac->ps == -1)
475
        m4ac->ps = 1;
476

    
477
    skip_bits_long(&gb, i);
478

    
479
    switch (m4ac->object_type) {
480
    case AOT_AAC_MAIN:
481
    case AOT_AAC_LC:
482
    case AOT_AAC_LTP:
483
        if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
484
            return -1;
485
        break;
486
    default:
487
        av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
488
               m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
489
        return -1;
490
    }
491

    
492
    return get_bits_count(&gb);
493
}
494

    
495
/**
496
 * linear congruential pseudorandom number generator
497
 *
498
 * @param   previous_val    pointer to the current state of the generator
499
 *
500
 * @return  Returns a 32-bit pseudorandom integer
501
 */
502
static av_always_inline int lcg_random(int previous_val)
503
{
504
    return previous_val * 1664525 + 1013904223;
505
}
506

    
507
static av_always_inline void reset_predict_state(PredictorState *ps)
508
{
509
    ps->r0   = 0.0f;
510
    ps->r1   = 0.0f;
511
    ps->cor0 = 0.0f;
512
    ps->cor1 = 0.0f;
513
    ps->var0 = 1.0f;
514
    ps->var1 = 1.0f;
515
}
516

    
517
static void reset_all_predictors(PredictorState *ps)
518
{
519
    int i;
520
    for (i = 0; i < MAX_PREDICTORS; i++)
521
        reset_predict_state(&ps[i]);
522
}
523

    
524
static void reset_predictor_group(PredictorState *ps, int group_num)
525
{
526
    int i;
527
    for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
528
        reset_predict_state(&ps[i]);
529
}
530

    
531
#define AAC_INIT_VLC_STATIC(num, size) \
532
    INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
533
         ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
534
        ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
535
        size);
536

    
537
static av_cold int aac_decode_init(AVCodecContext *avctx)
538
{
539
    AACContext *ac = avctx->priv_data;
540

    
541
    ac->avctx = avctx;
542
    ac->m4ac.sample_rate = avctx->sample_rate;
543

    
544
    if (avctx->extradata_size > 0) {
545
        if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
546
                                         avctx->extradata,
547
                                         avctx->extradata_size) < 0)
548
            return -1;
549
    }
550

    
551
    avctx->sample_fmt = AV_SAMPLE_FMT_S16;
552

    
553
    AAC_INIT_VLC_STATIC( 0, 304);
554
    AAC_INIT_VLC_STATIC( 1, 270);
555
    AAC_INIT_VLC_STATIC( 2, 550);
556
    AAC_INIT_VLC_STATIC( 3, 300);
557
    AAC_INIT_VLC_STATIC( 4, 328);
558
    AAC_INIT_VLC_STATIC( 5, 294);
559
    AAC_INIT_VLC_STATIC( 6, 306);
560
    AAC_INIT_VLC_STATIC( 7, 268);
561
    AAC_INIT_VLC_STATIC( 8, 510);
562
    AAC_INIT_VLC_STATIC( 9, 366);
563
    AAC_INIT_VLC_STATIC(10, 462);
564

    
565
    ff_aac_sbr_init();
566

    
567
    dsputil_init(&ac->dsp, avctx);
568
    ff_fmt_convert_init(&ac->fmt_conv, avctx);
569

    
570
    ac->random_state = 0x1f2e3d4c;
571

    
572
    // -1024 - Compensate wrong IMDCT method.
573
    // 60    - Required to scale values to the correct range [-32768,32767]
574
    //         for float to int16 conversion. (1 << (60 / 4)) == 32768
575
    ac->sf_scale  = 1. / -1024.;
576
    ac->sf_offset = 60;
577

    
578
    ff_aac_tableinit();
579

    
580
    INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
581
                    ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
582
                    ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
583
                    352);
584

    
585
    ff_mdct_init(&ac->mdct,       11, 1, 1.0);
586
    ff_mdct_init(&ac->mdct_small,  8, 1, 1.0);
587
    ff_mdct_init(&ac->mdct_ltp,   11, 0, 1.0);
588
    // window initialization
589
    ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
590
    ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
591
    ff_init_ff_sine_windows(10);
592
    ff_init_ff_sine_windows( 7);
593

    
594
    cbrt_tableinit();
595

    
596
    return 0;
597
}
598

    
599
/**
600
 * Skip data_stream_element; reference: table 4.10.
601
 */
602
static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
603
{
604
    int byte_align = get_bits1(gb);
605
    int count = get_bits(gb, 8);
606
    if (count == 255)
607
        count += get_bits(gb, 8);
608
    if (byte_align)
609
        align_get_bits(gb);
610

    
611
    if (get_bits_left(gb) < 8 * count) {
612
        av_log(ac->avctx, AV_LOG_ERROR, overread_err);
613
        return -1;
614
    }
615
    skip_bits_long(gb, 8 * count);
616
    return 0;
617
}
618

    
619
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
620
                             GetBitContext *gb)
621
{
622
    int sfb;
623
    if (get_bits1(gb)) {
624
        ics->predictor_reset_group = get_bits(gb, 5);
625
        if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
626
            av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
627
            return -1;
628
        }
629
    }
630
    for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
631
        ics->prediction_used[sfb] = get_bits1(gb);
632
    }
633
    return 0;
634
}
635

    
636
/**
637
 * Decode Long Term Prediction data; reference: table 4.xx.
638
 */
639
static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
640
                       GetBitContext *gb, uint8_t max_sfb)
641
{
642
    int sfb;
643

    
644
    ltp->lag  = get_bits(gb, 11);
645
    ltp->coef = ltp_coef[get_bits(gb, 3)] * ac->sf_scale;
646
    for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
647
        ltp->used[sfb] = get_bits1(gb);
648
}
649

    
650
/**
651
 * Decode Individual Channel Stream info; reference: table 4.6.
652
 *
653
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
654
 */
655
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
656
                           GetBitContext *gb, int common_window)
657
{
658
    if (get_bits1(gb)) {
659
        av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
660
        memset(ics, 0, sizeof(IndividualChannelStream));
661
        return -1;
662
    }
663
    ics->window_sequence[1] = ics->window_sequence[0];
664
    ics->window_sequence[0] = get_bits(gb, 2);
665
    ics->use_kb_window[1]   = ics->use_kb_window[0];
666
    ics->use_kb_window[0]   = get_bits1(gb);
667
    ics->num_window_groups  = 1;
668
    ics->group_len[0]       = 1;
669
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
670
        int i;
671
        ics->max_sfb = get_bits(gb, 4);
672
        for (i = 0; i < 7; i++) {
673
            if (get_bits1(gb)) {
674
                ics->group_len[ics->num_window_groups - 1]++;
675
            } else {
676
                ics->num_window_groups++;
677
                ics->group_len[ics->num_window_groups - 1] = 1;
678
            }
679
        }
680
        ics->num_windows       = 8;
681
        ics->swb_offset        =    ff_swb_offset_128[ac->m4ac.sampling_index];
682
        ics->num_swb           =   ff_aac_num_swb_128[ac->m4ac.sampling_index];
683
        ics->tns_max_bands     = ff_tns_max_bands_128[ac->m4ac.sampling_index];
684
        ics->predictor_present = 0;
685
    } else {
686
        ics->max_sfb               = get_bits(gb, 6);
687
        ics->num_windows           = 1;
688
        ics->swb_offset            =    ff_swb_offset_1024[ac->m4ac.sampling_index];
689
        ics->num_swb               =   ff_aac_num_swb_1024[ac->m4ac.sampling_index];
690
        ics->tns_max_bands         = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
691
        ics->predictor_present     = get_bits1(gb);
692
        ics->predictor_reset_group = 0;
693
        if (ics->predictor_present) {
694
            if (ac->m4ac.object_type == AOT_AAC_MAIN) {
695
                if (decode_prediction(ac, ics, gb)) {
696
                    memset(ics, 0, sizeof(IndividualChannelStream));
697
                    return -1;
698
                }
699
            } else if (ac->m4ac.object_type == AOT_AAC_LC) {
700
                av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
701
                memset(ics, 0, sizeof(IndividualChannelStream));
702
                return -1;
703
            } else {
704
                if ((ics->ltp.present = get_bits(gb, 1)))
705
                    decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
706
            }
707
        }
708
    }
709

    
710
    if (ics->max_sfb > ics->num_swb) {
711
        av_log(ac->avctx, AV_LOG_ERROR,
712
               "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
713
               ics->max_sfb, ics->num_swb);
714
        memset(ics, 0, sizeof(IndividualChannelStream));
715
        return -1;
716
    }
717

    
718
    return 0;
719
}
720

    
721
/**
722
 * Decode band types (section_data payload); reference: table 4.46.
723
 *
724
 * @param   band_type           array of the used band type
725
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
726
 *
727
 * @return  Returns error status. 0 - OK, !0 - error
728
 */
729
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
730
                             int band_type_run_end[120], GetBitContext *gb,
731
                             IndividualChannelStream *ics)
732
{
733
    int g, idx = 0;
734
    const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
735
    for (g = 0; g < ics->num_window_groups; g++) {
736
        int k = 0;
737
        while (k < ics->max_sfb) {
738
            uint8_t sect_end = k;
739
            int sect_len_incr;
740
            int sect_band_type = get_bits(gb, 4);
741
            if (sect_band_type == 12) {
742
                av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
743
                return -1;
744
            }
745
            while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
746
                sect_end += sect_len_incr;
747
            sect_end += sect_len_incr;
748
            if (get_bits_left(gb) < 0) {
749
                av_log(ac->avctx, AV_LOG_ERROR, overread_err);
750
                return -1;
751
            }
752
            if (sect_end > ics->max_sfb) {
753
                av_log(ac->avctx, AV_LOG_ERROR,
754
                       "Number of bands (%d) exceeds limit (%d).\n",
755
                       sect_end, ics->max_sfb);
756
                return -1;
757
            }
758
            for (; k < sect_end; k++) {
759
                band_type        [idx]   = sect_band_type;
760
                band_type_run_end[idx++] = sect_end;
761
            }
762
        }
763
    }
764
    return 0;
765
}
766

    
767
/**
768
 * Decode scalefactors; reference: table 4.47.
769
 *
770
 * @param   global_gain         first scalefactor value as scalefactors are differentially coded
771
 * @param   band_type           array of the used band type
772
 * @param   band_type_run_end   array of the last scalefactor band of a band type run
773
 * @param   sf                  array of scalefactors or intensity stereo positions
774
 *
775
 * @return  Returns error status. 0 - OK, !0 - error
776
 */
777
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
778
                               unsigned int global_gain,
779
                               IndividualChannelStream *ics,
780
                               enum BandType band_type[120],
781
                               int band_type_run_end[120])
782
{
783
    const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
784
    int g, i, idx = 0;
785
    int offset[3] = { global_gain, global_gain - 90, 100 };
786
    int noise_flag = 1;
787
    static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
788
    for (g = 0; g < ics->num_window_groups; g++) {
789
        for (i = 0; i < ics->max_sfb;) {
790
            int run_end = band_type_run_end[idx];
791
            if (band_type[idx] == ZERO_BT) {
792
                for (; i < run_end; i++, idx++)
793
                    sf[idx] = 0.;
794
            } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
795
                for (; i < run_end; i++, idx++) {
796
                    offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
797
                    if (offset[2] > 255U) {
798
                        av_log(ac->avctx, AV_LOG_ERROR,
799
                               "%s (%d) out of range.\n", sf_str[2], offset[2]);
800
                        return -1;
801
                    }
802
                    sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
803
                }
804
            } else if (band_type[idx] == NOISE_BT) {
805
                for (; i < run_end; i++, idx++) {
806
                    if (noise_flag-- > 0)
807
                        offset[1] += get_bits(gb, 9) - 256;
808
                    else
809
                        offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
810
                    if (offset[1] > 255U) {
811
                        av_log(ac->avctx, AV_LOG_ERROR,
812
                               "%s (%d) out of range.\n", sf_str[1], offset[1]);
813
                        return -1;
814
                    }
815
                    sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
816
                }
817
            } else {
818
                for (; i < run_end; i++, idx++) {
819
                    offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
820
                    if (offset[0] > 255U) {
821
                        av_log(ac->avctx, AV_LOG_ERROR,
822
                               "%s (%d) out of range.\n", sf_str[0], offset[0]);
823
                        return -1;
824
                    }
825
                    sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
826
                }
827
            }
828
        }
829
    }
830
    return 0;
831
}
832

    
833
/**
834
 * Decode pulse data; reference: table 4.7.
835
 */
836
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
837
                         const uint16_t *swb_offset, int num_swb)
838
{
839
    int i, pulse_swb;
840
    pulse->num_pulse = get_bits(gb, 2) + 1;
841
    pulse_swb        = get_bits(gb, 6);
842
    if (pulse_swb >= num_swb)
843
        return -1;
844
    pulse->pos[0]    = swb_offset[pulse_swb];
845
    pulse->pos[0]   += get_bits(gb, 5);
846
    if (pulse->pos[0] > 1023)
847
        return -1;
848
    pulse->amp[0]    = get_bits(gb, 4);
849
    for (i = 1; i < pulse->num_pulse; i++) {
850
        pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
851
        if (pulse->pos[i] > 1023)
852
            return -1;
853
        pulse->amp[i] = get_bits(gb, 4);
854
    }
855
    return 0;
856
}
857

    
858
/**
859
 * Decode Temporal Noise Shaping data; reference: table 4.48.
860
 *
861
 * @return  Returns error status. 0 - OK, !0 - error
862
 */
863
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
864
                      GetBitContext *gb, const IndividualChannelStream *ics)
865
{
866
    int w, filt, i, coef_len, coef_res, coef_compress;
867
    const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
868
    const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
869
    for (w = 0; w < ics->num_windows; w++) {
870
        if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
871
            coef_res = get_bits1(gb);
872

    
873
            for (filt = 0; filt < tns->n_filt[w]; filt++) {
874
                int tmp2_idx;
875
                tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
876

    
877
                if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
878
                    av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
879
                           tns->order[w][filt], tns_max_order);
880
                    tns->order[w][filt] = 0;
881
                    return -1;
882
                }
883
                if (tns->order[w][filt]) {
884
                    tns->direction[w][filt] = get_bits1(gb);
885
                    coef_compress = get_bits1(gb);
886
                    coef_len = coef_res + 3 - coef_compress;
887
                    tmp2_idx = 2 * coef_compress + coef_res;
888

    
889
                    for (i = 0; i < tns->order[w][filt]; i++)
890
                        tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
891
                }
892
            }
893
        }
894
    }
895
    return 0;
896
}
897

    
898
/**
899
 * Decode Mid/Side data; reference: table 4.54.
900
 *
901
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
902
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
903
 *                      [3] reserved for scalable AAC
904
 */
905
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
906
                                   int ms_present)
907
{
908
    int idx;
909
    if (ms_present == 1) {
910
        for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
911
            cpe->ms_mask[idx] = get_bits1(gb);
912
    } else if (ms_present == 2) {
913
        memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
914
    }
915
}
916

    
917
#ifndef VMUL2
918
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
919
                           const float *scale)
920
{
921
    float s = *scale;
922
    *dst++ = v[idx    & 15] * s;
923
    *dst++ = v[idx>>4 & 15] * s;
924
    return dst;
925
}
926
#endif
927

    
928
#ifndef VMUL4
929
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
930
                           const float *scale)
931
{
932
    float s = *scale;
933
    *dst++ = v[idx    & 3] * s;
934
    *dst++ = v[idx>>2 & 3] * s;
935
    *dst++ = v[idx>>4 & 3] * s;
936
    *dst++ = v[idx>>6 & 3] * s;
937
    return dst;
938
}
939
#endif
940

    
941
#ifndef VMUL2S
942
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
943
                            unsigned sign, const float *scale)
944
{
945
    union float754 s0, s1;
946

    
947
    s0.f = s1.f = *scale;
948
    s0.i ^= sign >> 1 << 31;
949
    s1.i ^= sign      << 31;
950

    
951
    *dst++ = v[idx    & 15] * s0.f;
952
    *dst++ = v[idx>>4 & 15] * s1.f;
953

    
954
    return dst;
955
}
956
#endif
957

    
958
#ifndef VMUL4S
959
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
960
                            unsigned sign, const float *scale)
961
{
962
    unsigned nz = idx >> 12;
963
    union float754 s = { .f = *scale };
964
    union float754 t;
965

    
966
    t.i = s.i ^ (sign & 1<<31);
967
    *dst++ = v[idx    & 3] * t.f;
968

    
969
    sign <<= nz & 1; nz >>= 1;
970
    t.i = s.i ^ (sign & 1<<31);
971
    *dst++ = v[idx>>2 & 3] * t.f;
972

    
973
    sign <<= nz & 1; nz >>= 1;
974
    t.i = s.i ^ (sign & 1<<31);
975
    *dst++ = v[idx>>4 & 3] * t.f;
976

    
977
    sign <<= nz & 1; nz >>= 1;
978
    t.i = s.i ^ (sign & 1<<31);
979
    *dst++ = v[idx>>6 & 3] * t.f;
980

    
981
    return dst;
982
}
983
#endif
984

    
985
/**
986
 * Decode spectral data; reference: table 4.50.
987
 * Dequantize and scale spectral data; reference: 4.6.3.3.
988
 *
989
 * @param   coef            array of dequantized, scaled spectral data
990
 * @param   sf              array of scalefactors or intensity stereo positions
991
 * @param   pulse_present   set if pulses are present
992
 * @param   pulse           pointer to pulse data struct
993
 * @param   band_type       array of the used band type
994
 *
995
 * @return  Returns error status. 0 - OK, !0 - error
996
 */
997
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
998
                                       GetBitContext *gb, const float sf[120],
999
                                       int pulse_present, const Pulse *pulse,
1000
                                       const IndividualChannelStream *ics,
1001
                                       enum BandType band_type[120])
1002
{
1003
    int i, k, g, idx = 0;
1004
    const int c = 1024 / ics->num_windows;
1005
    const uint16_t *offsets = ics->swb_offset;
1006
    float *coef_base = coef;
1007

    
1008
    for (g = 0; g < ics->num_windows; g++)
1009
        memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1010

    
1011
    for (g = 0; g < ics->num_window_groups; g++) {
1012
        unsigned g_len = ics->group_len[g];
1013

    
1014
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1015
            const unsigned cbt_m1 = band_type[idx] - 1;
1016
            float *cfo = coef + offsets[i];
1017
            int off_len = offsets[i + 1] - offsets[i];
1018
            int group;
1019

    
1020
            if (cbt_m1 >= INTENSITY_BT2 - 1) {
1021
                for (group = 0; group < g_len; group++, cfo+=128) {
1022
                    memset(cfo, 0, off_len * sizeof(float));
1023
                }
1024
            } else if (cbt_m1 == NOISE_BT - 1) {
1025
                for (group = 0; group < g_len; group++, cfo+=128) {
1026
                    float scale;
1027
                    float band_energy;
1028

    
1029
                    for (k = 0; k < off_len; k++) {
1030
                        ac->random_state  = lcg_random(ac->random_state);
1031
                        cfo[k] = ac->random_state;
1032
                    }
1033

    
1034
                    band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1035
                    scale = sf[idx] / sqrtf(band_energy);
1036
                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1037
                }
1038
            } else {
1039
                const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1040
                const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1041
                VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1042
                OPEN_READER(re, gb);
1043

    
1044
                switch (cbt_m1 >> 1) {
1045
                case 0:
1046
                    for (group = 0; group < g_len; group++, cfo+=128) {
1047
                        float *cf = cfo;
1048
                        int len = off_len;
1049

    
1050
                        do {
1051
                            int code;
1052
                            unsigned cb_idx;
1053

    
1054
                            UPDATE_CACHE(re, gb);
1055
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1056
                            cb_idx = cb_vector_idx[code];
1057
                            cf = VMUL4(cf, vq, cb_idx, sf + idx);
1058
                        } while (len -= 4);
1059
                    }
1060
                    break;
1061

    
1062
                case 1:
1063
                    for (group = 0; group < g_len; group++, cfo+=128) {
1064
                        float *cf = cfo;
1065
                        int len = off_len;
1066

    
1067
                        do {
1068
                            int code;
1069
                            unsigned nnz;
1070
                            unsigned cb_idx;
1071
                            uint32_t bits;
1072

    
1073
                            UPDATE_CACHE(re, gb);
1074
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1075
                            cb_idx = cb_vector_idx[code];
1076
                            nnz = cb_idx >> 8 & 15;
1077
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1078
                            LAST_SKIP_BITS(re, gb, nnz);
1079
                            cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1080
                        } while (len -= 4);
1081
                    }
1082
                    break;
1083

    
1084
                case 2:
1085
                    for (group = 0; group < g_len; group++, cfo+=128) {
1086
                        float *cf = cfo;
1087
                        int len = off_len;
1088

    
1089
                        do {
1090
                            int code;
1091
                            unsigned cb_idx;
1092

    
1093
                            UPDATE_CACHE(re, gb);
1094
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1095
                            cb_idx = cb_vector_idx[code];
1096
                            cf = VMUL2(cf, vq, cb_idx, sf + idx);
1097
                        } while (len -= 2);
1098
                    }
1099
                    break;
1100

    
1101
                case 3:
1102
                case 4:
1103
                    for (group = 0; group < g_len; group++, cfo+=128) {
1104
                        float *cf = cfo;
1105
                        int len = off_len;
1106

    
1107
                        do {
1108
                            int code;
1109
                            unsigned nnz;
1110
                            unsigned cb_idx;
1111
                            unsigned sign;
1112

    
1113
                            UPDATE_CACHE(re, gb);
1114
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1115
                            cb_idx = cb_vector_idx[code];
1116
                            nnz = cb_idx >> 8 & 15;
1117
                            sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1118
                            LAST_SKIP_BITS(re, gb, nnz);
1119
                            cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1120
                        } while (len -= 2);
1121
                    }
1122
                    break;
1123

    
1124
                default:
1125
                    for (group = 0; group < g_len; group++, cfo+=128) {
1126
                        float *cf = cfo;
1127
                        uint32_t *icf = (uint32_t *) cf;
1128
                        int len = off_len;
1129

    
1130
                        do {
1131
                            int code;
1132
                            unsigned nzt, nnz;
1133
                            unsigned cb_idx;
1134
                            uint32_t bits;
1135
                            int j;
1136

    
1137
                            UPDATE_CACHE(re, gb);
1138
                            GET_VLC(code, re, gb, vlc_tab, 8, 2);
1139

    
1140
                            if (!code) {
1141
                                *icf++ = 0;
1142
                                *icf++ = 0;
1143
                                continue;
1144
                            }
1145

    
1146
                            cb_idx = cb_vector_idx[code];
1147
                            nnz = cb_idx >> 12;
1148
                            nzt = cb_idx >> 8;
1149
                            bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1150
                            LAST_SKIP_BITS(re, gb, nnz);
1151

    
1152
                            for (j = 0; j < 2; j++) {
1153
                                if (nzt & 1<<j) {
1154
                                    uint32_t b;
1155
                                    int n;
1156
                                    /* The total length of escape_sequence must be < 22 bits according
1157
                                       to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1158
                                    UPDATE_CACHE(re, gb);
1159
                                    b = GET_CACHE(re, gb);
1160
                                    b = 31 - av_log2(~b);
1161

    
1162
                                    if (b > 8) {
1163
                                        av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1164
                                        return -1;
1165
                                    }
1166

    
1167
                                    SKIP_BITS(re, gb, b + 1);
1168
                                    b += 4;
1169
                                    n = (1 << b) + SHOW_UBITS(re, gb, b);
1170
                                    LAST_SKIP_BITS(re, gb, b);
1171
                                    *icf++ = cbrt_tab[n] | (bits & 1<<31);
1172
                                    bits <<= 1;
1173
                                } else {
1174
                                    unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1175
                                    *icf++ = (bits & 1<<31) | v;
1176
                                    bits <<= !!v;
1177
                                }
1178
                                cb_idx >>= 4;
1179
                            }
1180
                        } while (len -= 2);
1181

    
1182
                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1183
                    }
1184
                }
1185

    
1186
                CLOSE_READER(re, gb);
1187
            }
1188
        }
1189
        coef += g_len << 7;
1190
    }
1191

    
1192
    if (pulse_present) {
1193
        idx = 0;
1194
        for (i = 0; i < pulse->num_pulse; i++) {
1195
            float co = coef_base[ pulse->pos[i] ];
1196
            while (offsets[idx + 1] <= pulse->pos[i])
1197
                idx++;
1198
            if (band_type[idx] != NOISE_BT && sf[idx]) {
1199
                float ico = -pulse->amp[i];
1200
                if (co) {
1201
                    co /= sf[idx];
1202
                    ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1203
                }
1204
                coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1205
            }
1206
        }
1207
    }
1208
    return 0;
1209
}
1210

    
1211
static av_always_inline float flt16_round(float pf)
1212
{
1213
    union float754 tmp;
1214
    tmp.f = pf;
1215
    tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1216
    return tmp.f;
1217
}
1218

    
1219
static av_always_inline float flt16_even(float pf)
1220
{
1221
    union float754 tmp;
1222
    tmp.f = pf;
1223
    tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1224
    return tmp.f;
1225
}
1226

    
1227
static av_always_inline float flt16_trunc(float pf)
1228
{
1229
    union float754 pun;
1230
    pun.f = pf;
1231
    pun.i &= 0xFFFF0000U;
1232
    return pun.f;
1233
}
1234

    
1235
static av_always_inline void predict(PredictorState *ps, float *coef,
1236
                                     float sf_scale, float inv_sf_scale,
1237
                    int output_enable)
1238
{
1239
    const float a     = 0.953125; // 61.0 / 64
1240
    const float alpha = 0.90625;  // 29.0 / 32
1241
    float e0, e1;
1242
    float pv;
1243
    float k1, k2;
1244
    float   r0 = ps->r0,     r1 = ps->r1;
1245
    float cor0 = ps->cor0, cor1 = ps->cor1;
1246
    float var0 = ps->var0, var1 = ps->var1;
1247

    
1248
    k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1249
    k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1250

    
1251
    pv = flt16_round(k1 * r0 + k2 * r1);
1252
    if (output_enable)
1253
        *coef += pv * sf_scale;
1254

    
1255
    e0 = *coef * inv_sf_scale;
1256
    e1 = e0 - k1 * r0;
1257

    
1258
    ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1259
    ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1260
    ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1261
    ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1262

    
1263
    ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1264
    ps->r0 = flt16_trunc(a * e0);
1265
}
1266

    
1267
/**
1268
 * Apply AAC-Main style frequency domain prediction.
1269
 */
1270
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1271
{
1272
    int sfb, k;
1273
    float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1274

    
1275
    if (!sce->ics.predictor_initialized) {
1276
        reset_all_predictors(sce->predictor_state);
1277
        sce->ics.predictor_initialized = 1;
1278
    }
1279

    
1280
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1281
        for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1282
            for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1283
                predict(&sce->predictor_state[k], &sce->coeffs[k],
1284
                        sf_scale, inv_sf_scale,
1285
                        sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1286
            }
1287
        }
1288
        if (sce->ics.predictor_reset_group)
1289
            reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1290
    } else
1291
        reset_all_predictors(sce->predictor_state);
1292
}
1293

    
1294
/**
1295
 * Decode an individual_channel_stream payload; reference: table 4.44.
1296
 *
1297
 * @param   common_window   Channels have independent [0], or shared [1], Individual Channel Stream information.
1298
 * @param   scale_flag      scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1299
 *
1300
 * @return  Returns error status. 0 - OK, !0 - error
1301
 */
1302
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1303
                      GetBitContext *gb, int common_window, int scale_flag)
1304
{
1305
    Pulse pulse;
1306
    TemporalNoiseShaping    *tns = &sce->tns;
1307
    IndividualChannelStream *ics = &sce->ics;
1308
    float *out = sce->coeffs;
1309
    int global_gain, pulse_present = 0;
1310

    
1311
    /* This assignment is to silence a GCC warning about the variable being used
1312
     * uninitialized when in fact it always is.
1313
     */
1314
    pulse.num_pulse = 0;
1315

    
1316
    global_gain = get_bits(gb, 8);
1317

    
1318
    if (!common_window && !scale_flag) {
1319
        if (decode_ics_info(ac, ics, gb, 0) < 0)
1320
            return -1;
1321
    }
1322

    
1323
    if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1324
        return -1;
1325
    if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1326
        return -1;
1327

    
1328
    pulse_present = 0;
1329
    if (!scale_flag) {
1330
        if ((pulse_present = get_bits1(gb))) {
1331
            if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1332
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1333
                return -1;
1334
            }
1335
            if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1336
                av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1337
                return -1;
1338
            }
1339
        }
1340
        if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1341
            return -1;
1342
        if (get_bits1(gb)) {
1343
            av_log_missing_feature(ac->avctx, "SSR", 1);
1344
            return -1;
1345
        }
1346
    }
1347

    
1348
    if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1349
        return -1;
1350

    
1351
    if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1352
        apply_prediction(ac, sce);
1353

    
1354
    return 0;
1355
}
1356

    
1357
/**
1358
 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1359
 */
1360
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1361
{
1362
    const IndividualChannelStream *ics = &cpe->ch[0].ics;
1363
    float *ch0 = cpe->ch[0].coeffs;
1364
    float *ch1 = cpe->ch[1].coeffs;
1365
    int g, i, group, idx = 0;
1366
    const uint16_t *offsets = ics->swb_offset;
1367
    for (g = 0; g < ics->num_window_groups; g++) {
1368
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1369
            if (cpe->ms_mask[idx] &&
1370
                    cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1371
                for (group = 0; group < ics->group_len[g]; group++) {
1372
                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1373
                                              ch1 + group * 128 + offsets[i],
1374
                                              offsets[i+1] - offsets[i]);
1375
                }
1376
            }
1377
        }
1378
        ch0 += ics->group_len[g] * 128;
1379
        ch1 += ics->group_len[g] * 128;
1380
    }
1381
}
1382

    
1383
/**
1384
 * intensity stereo decoding; reference: 4.6.8.2.3
1385
 *
1386
 * @param   ms_present  Indicates mid/side stereo presence. [0] mask is all 0s;
1387
 *                      [1] mask is decoded from bitstream; [2] mask is all 1s;
1388
 *                      [3] reserved for scalable AAC
1389
 */
1390
static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1391
{
1392
    const IndividualChannelStream *ics = &cpe->ch[1].ics;
1393
    SingleChannelElement         *sce1 = &cpe->ch[1];
1394
    float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1395
    const uint16_t *offsets = ics->swb_offset;
1396
    int g, group, i, idx = 0;
1397
    int c;
1398
    float scale;
1399
    for (g = 0; g < ics->num_window_groups; g++) {
1400
        for (i = 0; i < ics->max_sfb;) {
1401
            if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1402
                const int bt_run_end = sce1->band_type_run_end[idx];
1403
                for (; i < bt_run_end; i++, idx++) {
1404
                    c = -1 + 2 * (sce1->band_type[idx] - 14);
1405
                    if (ms_present)
1406
                        c *= 1 - 2 * cpe->ms_mask[idx];
1407
                    scale = c * sce1->sf[idx];
1408
                    for (group = 0; group < ics->group_len[g]; group++)
1409
                        ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1410
                                                   coef0 + group * 128 + offsets[i],
1411
                                                   scale,
1412
                                                   offsets[i + 1] - offsets[i]);
1413
                }
1414
            } else {
1415
                int bt_run_end = sce1->band_type_run_end[idx];
1416
                idx += bt_run_end - i;
1417
                i    = bt_run_end;
1418
            }
1419
        }
1420
        coef0 += ics->group_len[g] * 128;
1421
        coef1 += ics->group_len[g] * 128;
1422
    }
1423
}
1424

    
1425
/**
1426
 * Decode a channel_pair_element; reference: table 4.4.
1427
 *
1428
 * @return  Returns error status. 0 - OK, !0 - error
1429
 */
1430
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1431
{
1432
    int i, ret, common_window, ms_present = 0;
1433

    
1434
    common_window = get_bits1(gb);
1435
    if (common_window) {
1436
        if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1437
            return -1;
1438
        i = cpe->ch[1].ics.use_kb_window[0];
1439
        cpe->ch[1].ics = cpe->ch[0].ics;
1440
        cpe->ch[1].ics.use_kb_window[1] = i;
1441
        if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1442
            if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1443
                decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1444
        ms_present = get_bits(gb, 2);
1445
        if (ms_present == 3) {
1446
            av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1447
            return -1;
1448
        } else if (ms_present)
1449
            decode_mid_side_stereo(cpe, gb, ms_present);
1450
    }
1451
    if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1452
        return ret;
1453
    if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1454
        return ret;
1455

    
1456
    if (common_window) {
1457
        if (ms_present)
1458
            apply_mid_side_stereo(ac, cpe);
1459
        if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1460
            apply_prediction(ac, &cpe->ch[0]);
1461
            apply_prediction(ac, &cpe->ch[1]);
1462
        }
1463
    }
1464

    
1465
    apply_intensity_stereo(ac, cpe, ms_present);
1466
    return 0;
1467
}
1468

    
1469
static const float cce_scale[] = {
1470
    1.09050773266525765921, //2^(1/8)
1471
    1.18920711500272106672, //2^(1/4)
1472
    M_SQRT2,
1473
    2,
1474
};
1475

    
1476
/**
1477
 * Decode coupling_channel_element; reference: table 4.8.
1478
 *
1479
 * @return  Returns error status. 0 - OK, !0 - error
1480
 */
1481
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1482
{
1483
    int num_gain = 0;
1484
    int c, g, sfb, ret;
1485
    int sign;
1486
    float scale;
1487
    SingleChannelElement *sce = &che->ch[0];
1488
    ChannelCoupling     *coup = &che->coup;
1489

    
1490
    coup->coupling_point = 2 * get_bits1(gb);
1491
    coup->num_coupled = get_bits(gb, 3);
1492
    for (c = 0; c <= coup->num_coupled; c++) {
1493
        num_gain++;
1494
        coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1495
        coup->id_select[c] = get_bits(gb, 4);
1496
        if (coup->type[c] == TYPE_CPE) {
1497
            coup->ch_select[c] = get_bits(gb, 2);
1498
            if (coup->ch_select[c] == 3)
1499
                num_gain++;
1500
        } else
1501
            coup->ch_select[c] = 2;
1502
    }
1503
    coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1504

    
1505
    sign  = get_bits(gb, 1);
1506
    scale = cce_scale[get_bits(gb, 2)];
1507

    
1508
    if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1509
        return ret;
1510

    
1511
    for (c = 0; c < num_gain; c++) {
1512
        int idx  = 0;
1513
        int cge  = 1;
1514
        int gain = 0;
1515
        float gain_cache = 1.;
1516
        if (c) {
1517
            cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1518
            gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1519
            gain_cache = powf(scale, -gain);
1520
        }
1521
        if (coup->coupling_point == AFTER_IMDCT) {
1522
            coup->gain[c][0] = gain_cache;
1523
        } else {
1524
            for (g = 0; g < sce->ics.num_window_groups; g++) {
1525
                for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1526
                    if (sce->band_type[idx] != ZERO_BT) {
1527
                        if (!cge) {
1528
                            int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1529
                            if (t) {
1530
                                int s = 1;
1531
                                t = gain += t;
1532
                                if (sign) {
1533
                                    s  -= 2 * (t & 0x1);
1534
                                    t >>= 1;
1535
                                }
1536
                                gain_cache = powf(scale, -t) * s;
1537
                            }
1538
                        }
1539
                        coup->gain[c][idx] = gain_cache;
1540
                    }
1541
                }
1542
            }
1543
        }
1544
    }
1545
    return 0;
1546
}
1547

    
1548
/**
1549
 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1550
 *
1551
 * @return  Returns number of bytes consumed.
1552
 */
1553
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1554
                                         GetBitContext *gb)
1555
{
1556
    int i;
1557
    int num_excl_chan = 0;
1558

    
1559
    do {
1560
        for (i = 0; i < 7; i++)
1561
            che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1562
    } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1563

    
1564
    return num_excl_chan / 7;
1565
}
1566

    
1567
/**
1568
 * Decode dynamic range information; reference: table 4.52.
1569
 *
1570
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1571
 *
1572
 * @return  Returns number of bytes consumed.
1573
 */
1574
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1575
                                GetBitContext *gb, int cnt)
1576
{
1577
    int n             = 1;
1578
    int drc_num_bands = 1;
1579
    int i;
1580

    
1581
    /* pce_tag_present? */
1582
    if (get_bits1(gb)) {
1583
        che_drc->pce_instance_tag  = get_bits(gb, 4);
1584
        skip_bits(gb, 4); // tag_reserved_bits
1585
        n++;
1586
    }
1587

    
1588
    /* excluded_chns_present? */
1589
    if (get_bits1(gb)) {
1590
        n += decode_drc_channel_exclusions(che_drc, gb);
1591
    }
1592

    
1593
    /* drc_bands_present? */
1594
    if (get_bits1(gb)) {
1595
        che_drc->band_incr            = get_bits(gb, 4);
1596
        che_drc->interpolation_scheme = get_bits(gb, 4);
1597
        n++;
1598
        drc_num_bands += che_drc->band_incr;
1599
        for (i = 0; i < drc_num_bands; i++) {
1600
            che_drc->band_top[i] = get_bits(gb, 8);
1601
            n++;
1602
        }
1603
    }
1604

    
1605
    /* prog_ref_level_present? */
1606
    if (get_bits1(gb)) {
1607
        che_drc->prog_ref_level = get_bits(gb, 7);
1608
        skip_bits1(gb); // prog_ref_level_reserved_bits
1609
        n++;
1610
    }
1611

    
1612
    for (i = 0; i < drc_num_bands; i++) {
1613
        che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1614
        che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1615
        n++;
1616
    }
1617

    
1618
    return n;
1619
}
1620

    
1621
/**
1622
 * Decode extension data (incomplete); reference: table 4.51.
1623
 *
1624
 * @param   cnt length of TYPE_FIL syntactic element in bytes
1625
 *
1626
 * @return Returns number of bytes consumed
1627
 */
1628
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1629
                                    ChannelElement *che, enum RawDataBlockType elem_type)
1630
{
1631
    int crc_flag = 0;
1632
    int res = cnt;
1633
    switch (get_bits(gb, 4)) { // extension type
1634
    case EXT_SBR_DATA_CRC:
1635
        crc_flag++;
1636
    case EXT_SBR_DATA:
1637
        if (!che) {
1638
            av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1639
            return res;
1640
        } else if (!ac->m4ac.sbr) {
1641
            av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1642
            skip_bits_long(gb, 8 * cnt - 4);
1643
            return res;
1644
        } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1645
            av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1646
            skip_bits_long(gb, 8 * cnt - 4);
1647
            return res;
1648
        } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1649
            ac->m4ac.sbr = 1;
1650
            ac->m4ac.ps = 1;
1651
            output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1652
        } else {
1653
            ac->m4ac.sbr = 1;
1654
        }
1655
        res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1656
        break;
1657
    case EXT_DYNAMIC_RANGE:
1658
        res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1659
        break;
1660
    case EXT_FILL:
1661
    case EXT_FILL_DATA:
1662
    case EXT_DATA_ELEMENT:
1663
    default:
1664
        skip_bits_long(gb, 8 * cnt - 4);
1665
        break;
1666
    };
1667
    return res;
1668
}
1669

    
1670
/**
1671
 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1672
 *
1673
 * @param   decode  1 if tool is used normally, 0 if tool is used in LTP.
1674
 * @param   coef    spectral coefficients
1675
 */
1676
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1677
                      IndividualChannelStream *ics, int decode)
1678
{
1679
    const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1680
    int w, filt, m, i;
1681
    int bottom, top, order, start, end, size, inc;
1682
    float lpc[TNS_MAX_ORDER];
1683
    float tmp[TNS_MAX_ORDER];
1684

    
1685
    for (w = 0; w < ics->num_windows; w++) {
1686
        bottom = ics->num_swb;
1687
        for (filt = 0; filt < tns->n_filt[w]; filt++) {
1688
            top    = bottom;
1689
            bottom = FFMAX(0, top - tns->length[w][filt]);
1690
            order  = tns->order[w][filt];
1691
            if (order == 0)
1692
                continue;
1693

    
1694
            // tns_decode_coef
1695
            compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1696

    
1697
            start = ics->swb_offset[FFMIN(bottom, mmm)];
1698
            end   = ics->swb_offset[FFMIN(   top, mmm)];
1699
            if ((size = end - start) <= 0)
1700
                continue;
1701
            if (tns->direction[w][filt]) {
1702
                inc = -1;
1703
                start = end - 1;
1704
            } else {
1705
                inc = 1;
1706
            }
1707
            start += w * 128;
1708

    
1709
            if (decode) {
1710
                // ar filter
1711
                for (m = 0; m < size; m++, start += inc)
1712
                    for (i = 1; i <= FFMIN(m, order); i++)
1713
                        coef[start] -= coef[start - i * inc] * lpc[i - 1];
1714
            } else {
1715
                // ma filter
1716
                for (m = 0; m < size; m++, start += inc) {
1717
                    tmp[0] = coef[start];
1718
                    for (i = 1; i <= FFMIN(m, order); i++)
1719
                        coef[start] += tmp[i] * lpc[i - 1];
1720
                    for (i = order; i > 0; i--)
1721
                        tmp[i] = tmp[i - 1];
1722
                }
1723
            }
1724
        }
1725
    }
1726
}
1727

    
1728
/**
1729
 *  Apply windowing and MDCT to obtain the spectral
1730
 *  coefficient from the predicted sample by LTP.
1731
 */
1732
static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1733
                                   float *in, IndividualChannelStream *ics)
1734
{
1735
    const float *lwindow      = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1736
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1737
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1738
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1739

    
1740
    if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1741
        ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1742
    } else {
1743
        memset(in, 0, 448 * sizeof(float));
1744
        ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1745
        memcpy(in + 576, in + 576, 448 * sizeof(float));
1746
    }
1747
    if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1748
        ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1749
    } else {
1750
        memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1751
        ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1752
        memset(in + 1024 + 576, 0, 448 * sizeof(float));
1753
    }
1754
    ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1755
}
1756

    
1757
/**
1758
 * Apply the long term prediction
1759
 */
1760
static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1761
{
1762
    const LongTermPrediction *ltp = &sce->ics.ltp;
1763
    const uint16_t *offsets = sce->ics.swb_offset;
1764
    int i, sfb;
1765

    
1766
    if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1767
        float *predTime = sce->ret;
1768
        float *predFreq = ac->buf_mdct;
1769
        int16_t num_samples = 2048;
1770

    
1771
        if (ltp->lag < 1024)
1772
            num_samples = ltp->lag + 1024;
1773
        for (i = 0; i < num_samples; i++)
1774
            predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1775
        memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1776

    
1777
        windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1778

    
1779
        if (sce->tns.present)
1780
            apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1781

    
1782
        for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1783
            if (ltp->used[sfb])
1784
                for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1785
                    sce->coeffs[i] += predFreq[i];
1786
    }
1787
}
1788

    
1789
/**
1790
 * Update the LTP buffer for next frame
1791
 */
1792
static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1793
{
1794
    IndividualChannelStream *ics = &sce->ics;
1795
    float *saved     = sce->saved;
1796
    float *saved_ltp = sce->coeffs;
1797
    const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1798
    const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1799
    int i;
1800

    
1801
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1802
        memcpy(saved_ltp,       saved, 512 * sizeof(float));
1803
        memset(saved_ltp + 576, 0,     448 * sizeof(float));
1804
        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
1805
        for (i = 0; i < 64; i++)
1806
            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1807
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1808
        memcpy(saved_ltp,       ac->buf_mdct + 512, 448 * sizeof(float));
1809
        memset(saved_ltp + 576, 0,                  448 * sizeof(float));
1810
        ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960,     &swindow[64],      64);
1811
        for (i = 0; i < 64; i++)
1812
            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1813
    } else { // LONG_STOP or ONLY_LONG
1814
        ac->dsp.vector_fmul_reverse(saved_ltp,       ac->buf_mdct + 512,     &lwindow[512],     512);
1815
        for (i = 0; i < 512; i++)
1816
            saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1817
    }
1818

    
1819
    memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1820
    ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret,  1024);
1821
    ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1822
}
1823

    
1824
/**
1825
 * Conduct IMDCT and windowing.
1826
 */
1827
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1828
{
1829
    IndividualChannelStream *ics = &sce->ics;
1830
    float *in    = sce->coeffs;
1831
    float *out   = sce->ret;
1832
    float *saved = sce->saved;
1833
    const float *swindow      = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1834
    const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1835
    const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1836
    float *buf  = ac->buf_mdct;
1837
    float *temp = ac->temp;
1838
    int i;
1839

    
1840
    // imdct
1841
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1842
        for (i = 0; i < 1024; i += 128)
1843
            ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1844
    } else
1845
        ac->mdct.imdct_half(&ac->mdct, buf, in);
1846

    
1847
    /* window overlapping
1848
     * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1849
     * and long to short transitions are considered to be short to short
1850
     * transitions. This leaves just two cases (long to long and short to short)
1851
     * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1852
     */
1853
    if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1854
            (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1855
        ac->dsp.vector_fmul_window(    out,               saved,            buf,         lwindow_prev, 512);
1856
    } else {
1857
        memcpy(                        out,               saved,            448 * sizeof(float));
1858

    
1859
        if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1860
            ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448,      buf + 0*128, swindow_prev, 64);
1861
            ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow,      64);
1862
            ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow,      64);
1863
            ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow,      64);
1864
            ac->dsp.vector_fmul_window(temp,              buf + 3*128 + 64, buf + 4*128, swindow,      64);
1865
            memcpy(                    out + 448 + 4*128, temp, 64 * sizeof(float));
1866
        } else {
1867
            ac->dsp.vector_fmul_window(out + 448,         saved + 448,      buf,         swindow_prev, 64);
1868
            memcpy(                    out + 576,         buf + 64,         448 * sizeof(float));
1869
        }
1870
    }
1871

    
1872
    // buffer update
1873
    if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1874
        memcpy(                    saved,       temp + 64,         64 * sizeof(float));
1875
        ac->dsp.vector_fmul_window(saved + 64,  buf + 4*128 + 64, buf + 5*128, swindow, 64);
1876
        ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1877
        ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1878
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1879
    } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1880
        memcpy(                    saved,       buf + 512,        448 * sizeof(float));
1881
        memcpy(                    saved + 448, buf + 7*128 + 64,  64 * sizeof(float));
1882
    } else { // LONG_STOP or ONLY_LONG
1883
        memcpy(                    saved,       buf + 512,        512 * sizeof(float));
1884
    }
1885
}
1886

    
1887
/**
1888
 * Apply dependent channel coupling (applied before IMDCT).
1889
 *
1890
 * @param   index   index into coupling gain array
1891
 */
1892
static void apply_dependent_coupling(AACContext *ac,
1893
                                     SingleChannelElement *target,
1894
                                     ChannelElement *cce, int index)
1895
{
1896
    IndividualChannelStream *ics = &cce->ch[0].ics;
1897
    const uint16_t *offsets = ics->swb_offset;
1898
    float *dest = target->coeffs;
1899
    const float *src = cce->ch[0].coeffs;
1900
    int g, i, group, k, idx = 0;
1901
    if (ac->m4ac.object_type == AOT_AAC_LTP) {
1902
        av_log(ac->avctx, AV_LOG_ERROR,
1903
               "Dependent coupling is not supported together with LTP\n");
1904
        return;
1905
    }
1906
    for (g = 0; g < ics->num_window_groups; g++) {
1907
        for (i = 0; i < ics->max_sfb; i++, idx++) {
1908
            if (cce->ch[0].band_type[idx] != ZERO_BT) {
1909
                const float gain = cce->coup.gain[index][idx];
1910
                for (group = 0; group < ics->group_len[g]; group++) {
1911
                    for (k = offsets[i]; k < offsets[i + 1]; k++) {
1912
                        // XXX dsputil-ize
1913
                        dest[group * 128 + k] += gain * src[group * 128 + k];
1914
                    }
1915
                }
1916
            }
1917
        }
1918
        dest += ics->group_len[g] * 128;
1919
        src  += ics->group_len[g] * 128;
1920
    }
1921
}
1922

    
1923
/**
1924
 * Apply independent channel coupling (applied after IMDCT).
1925
 *
1926
 * @param   index   index into coupling gain array
1927
 */
1928
static void apply_independent_coupling(AACContext *ac,
1929
                                       SingleChannelElement *target,
1930
                                       ChannelElement *cce, int index)
1931
{
1932
    int i;
1933
    const float gain = cce->coup.gain[index][0];
1934
    const float *src = cce->ch[0].ret;
1935
    float *dest = target->ret;
1936
    const int len = 1024 << (ac->m4ac.sbr == 1);
1937

    
1938
    for (i = 0; i < len; i++)
1939
        dest[i] += gain * src[i];
1940
}
1941

    
1942
/**
1943
 * channel coupling transformation interface
1944
 *
1945
 * @param   apply_coupling_method   pointer to (in)dependent coupling function
1946
 */
1947
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1948
                                   enum RawDataBlockType type, int elem_id,
1949
                                   enum CouplingPoint coupling_point,
1950
                                   void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1951
{
1952
    int i, c;
1953

    
1954
    for (i = 0; i < MAX_ELEM_ID; i++) {
1955
        ChannelElement *cce = ac->che[TYPE_CCE][i];
1956
        int index = 0;
1957

    
1958
        if (cce && cce->coup.coupling_point == coupling_point) {
1959
            ChannelCoupling *coup = &cce->coup;
1960

    
1961
            for (c = 0; c <= coup->num_coupled; c++) {
1962
                if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1963
                    if (coup->ch_select[c] != 1) {
1964
                        apply_coupling_method(ac, &cc->ch[0], cce, index);
1965
                        if (coup->ch_select[c] != 0)
1966
                            index++;
1967
                    }
1968
                    if (coup->ch_select[c] != 2)
1969
                        apply_coupling_method(ac, &cc->ch[1], cce, index++);
1970
                } else
1971
                    index += 1 + (coup->ch_select[c] == 3);
1972
            }
1973
        }
1974
    }
1975
}
1976

    
1977
/**
1978
 * Convert spectral data to float samples, applying all supported tools as appropriate.
1979
 */
1980
static void spectral_to_sample(AACContext *ac)
1981
{
1982
    int i, type;
1983
    for (type = 3; type >= 0; type--) {
1984
        for (i = 0; i < MAX_ELEM_ID; i++) {
1985
            ChannelElement *che = ac->che[type][i];
1986
            if (che) {
1987
                if (type <= TYPE_CPE)
1988
                    apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1989
                if (ac->m4ac.object_type == AOT_AAC_LTP) {
1990
                    if (che->ch[0].ics.predictor_present) {
1991
                        if (che->ch[0].ics.ltp.present)
1992
                            apply_ltp(ac, &che->ch[0]);
1993
                        if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
1994
                            apply_ltp(ac, &che->ch[1]);
1995
                    }
1996
                }
1997
                if (che->ch[0].tns.present)
1998
                    apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1999
                if (che->ch[1].tns.present)
2000
                    apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2001
                if (type <= TYPE_CPE)
2002
                    apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2003
                if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2004
                    imdct_and_windowing(ac, &che->ch[0]);
2005
                    if (ac->m4ac.object_type == AOT_AAC_LTP)
2006
                        update_ltp(ac, &che->ch[0]);
2007
                    if (type == TYPE_CPE) {
2008
                        imdct_and_windowing(ac, &che->ch[1]);
2009
                        if (ac->m4ac.object_type == AOT_AAC_LTP)
2010
                            update_ltp(ac, &che->ch[1]);
2011
                    }
2012
                    if (ac->m4ac.sbr > 0) {
2013
                        ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2014
                    }
2015
                }
2016
                if (type <= TYPE_CCE)
2017
                    apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2018
            }
2019
        }
2020
    }
2021
}
2022

    
2023
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2024
{
2025
    int size;
2026
    AACADTSHeaderInfo hdr_info;
2027

    
2028
    size = ff_aac_parse_header(gb, &hdr_info);
2029
    if (size > 0) {
2030
        if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2031
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2032
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2033
            ac->m4ac.chan_config = hdr_info.chan_config;
2034
            if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2035
                return -7;
2036
            if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2037
                return -7;
2038
        } else if (ac->output_configured != OC_LOCKED) {
2039
            ac->output_configured = OC_NONE;
2040
        }
2041
        if (ac->output_configured != OC_LOCKED) {
2042
            ac->m4ac.sbr = -1;
2043
            ac->m4ac.ps  = -1;
2044
        }
2045
        ac->m4ac.sample_rate     = hdr_info.sample_rate;
2046
        ac->m4ac.sampling_index  = hdr_info.sampling_index;
2047
        ac->m4ac.object_type     = hdr_info.object_type;
2048
        if (!ac->avctx->sample_rate)
2049
            ac->avctx->sample_rate = hdr_info.sample_rate;
2050
        if (hdr_info.num_aac_frames == 1) {
2051
            if (!hdr_info.crc_absent)
2052
                skip_bits(gb, 16);
2053
        } else {
2054
            av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2055
            return -1;
2056
        }
2057
    }
2058
    return size;
2059
}
2060

    
2061
static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2062
                                int *data_size, GetBitContext *gb)
2063
{
2064
    AACContext *ac = avctx->priv_data;
2065
    ChannelElement *che = NULL, *che_prev = NULL;
2066
    enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2067
    int err, elem_id, data_size_tmp;
2068
    int samples = 0, multiplier;
2069

    
2070
    if (show_bits(gb, 12) == 0xfff) {
2071
        if (parse_adts_frame_header(ac, gb) < 0) {
2072
            av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2073
            return -1;
2074
        }
2075
        if (ac->m4ac.sampling_index > 12) {
2076
            av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2077
            return -1;
2078
        }
2079
    }
2080

    
2081
    ac->tags_mapped = 0;
2082
    // parse
2083
    while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2084
        elem_id = get_bits(gb, 4);
2085

    
2086
        if (elem_type < TYPE_DSE) {
2087
            if (!(che=get_che(ac, elem_type, elem_id))) {
2088
                av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2089
                       elem_type, elem_id);
2090
                return -1;
2091
            }
2092
            samples = 1024;
2093
        }
2094

    
2095
        switch (elem_type) {
2096

    
2097
        case TYPE_SCE:
2098
            err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2099
            break;
2100

    
2101
        case TYPE_CPE:
2102
            err = decode_cpe(ac, gb, che);
2103
            break;
2104

    
2105
        case TYPE_CCE:
2106
            err = decode_cce(ac, gb, che);
2107
            break;
2108

    
2109
        case TYPE_LFE:
2110
            err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2111
            break;
2112

    
2113
        case TYPE_DSE:
2114
            err = skip_data_stream_element(ac, gb);
2115
            break;
2116

    
2117
        case TYPE_PCE: {
2118
            enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2119
            memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2120
            if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2121
                break;
2122
            if (ac->output_configured > OC_TRIAL_PCE)
2123
                av_log(avctx, AV_LOG_ERROR,
2124
                       "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2125
            else
2126
                err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2127
            break;
2128
        }
2129

    
2130
        case TYPE_FIL:
2131
            if (elem_id == 15)
2132
                elem_id += get_bits(gb, 8) - 1;
2133
            if (get_bits_left(gb) < 8 * elem_id) {
2134
                    av_log(avctx, AV_LOG_ERROR, overread_err);
2135
                    return -1;
2136
            }
2137
            while (elem_id > 0)
2138
                elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2139
            err = 0; /* FIXME */
2140
            break;
2141

    
2142
        default:
2143
            err = -1; /* should not happen, but keeps compiler happy */
2144
            break;
2145
        }
2146

    
2147
        che_prev       = che;
2148
        elem_type_prev = elem_type;
2149

    
2150
        if (err)
2151
            return err;
2152

    
2153
        if (get_bits_left(gb) < 3) {
2154
            av_log(avctx, AV_LOG_ERROR, overread_err);
2155
            return -1;
2156
        }
2157
    }
2158

    
2159
    spectral_to_sample(ac);
2160

    
2161
    multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2162
    samples <<= multiplier;
2163
    if (ac->output_configured < OC_LOCKED) {
2164
        avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2165
        avctx->frame_size = samples;
2166
    }
2167

    
2168
    data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2169
    if (*data_size < data_size_tmp) {
2170
        av_log(avctx, AV_LOG_ERROR,
2171
               "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2172
               *data_size, data_size_tmp);
2173
        return -1;
2174
    }
2175
    *data_size = data_size_tmp;
2176

    
2177
    if (samples)
2178
        ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2179

    
2180
    if (ac->output_configured)
2181
        ac->output_configured = OC_LOCKED;
2182

    
2183
    return 0;
2184
}
2185

    
2186
static int aac_decode_frame(AVCodecContext *avctx, void *data,
2187
                            int *data_size, AVPacket *avpkt)
2188
{
2189
    const uint8_t *buf = avpkt->data;
2190
    int buf_size = avpkt->size;
2191
    GetBitContext gb;
2192
    int buf_consumed;
2193
    int buf_offset;
2194
    int err;
2195

    
2196
    init_get_bits(&gb, buf, buf_size * 8);
2197

    
2198
    if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2199
        return err;
2200

    
2201
    buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2202
    for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2203
        if (buf[buf_offset])
2204
            break;
2205

    
2206
    return buf_size > buf_offset ? buf_consumed : buf_size;
2207
}
2208

    
2209
static av_cold int aac_decode_close(AVCodecContext *avctx)
2210
{
2211
    AACContext *ac = avctx->priv_data;
2212
    int i, type;
2213

    
2214
    for (i = 0; i < MAX_ELEM_ID; i++) {
2215
        for (type = 0; type < 4; type++) {
2216
            if (ac->che[type][i])
2217
                ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2218
            av_freep(&ac->che[type][i]);
2219
        }
2220
    }
2221

    
2222
    ff_mdct_end(&ac->mdct);
2223
    ff_mdct_end(&ac->mdct_small);
2224
    ff_mdct_end(&ac->mdct_ltp);
2225
    return 0;
2226
}
2227

    
2228

    
2229
#define LOAS_SYNC_WORD   0x2b7       ///< 11 bits LOAS sync word
2230

    
2231
struct LATMContext {
2232
    AACContext      aac_ctx;             ///< containing AACContext
2233
    int             initialized;         ///< initilized after a valid extradata was seen
2234

    
2235
    // parser data
2236
    int             audio_mux_version_A; ///< LATM syntax version
2237
    int             frame_length_type;   ///< 0/1 variable/fixed frame length
2238
    int             frame_length;        ///< frame length for fixed frame length
2239
};
2240

    
2241
static inline uint32_t latm_get_value(GetBitContext *b)
2242
{
2243
    int length = get_bits(b, 2);
2244

    
2245
    return get_bits_long(b, (length+1)*8);
2246
}
2247

    
2248
static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2249
                                             GetBitContext *gb)
2250
{
2251
    AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2252
    int  config_start_bit = get_bits_count(gb);
2253
    int     bits_consumed, esize;
2254

    
2255
    if (config_start_bit % 8) {
2256
        av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2257
                               "config not byte aligned.\n", 1);
2258
        return AVERROR_INVALIDDATA;
2259
    } else {
2260
        bits_consumed =
2261
            decode_audio_specific_config(&latmctx->aac_ctx, avctx,
2262
                                         &latmctx->aac_ctx.m4ac,
2263
                                         gb->buffer + (config_start_bit / 8),
2264
                                         get_bits_left(gb) / 8);
2265

    
2266
        if (bits_consumed < 0)
2267
            return AVERROR_INVALIDDATA;
2268

    
2269
        esize = (bits_consumed+7) / 8;
2270

    
2271
        if (avctx->extradata_size <= esize) {
2272
            av_free(avctx->extradata);
2273
            avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2274
            if (!avctx->extradata)
2275
                return AVERROR(ENOMEM);
2276
        }
2277

    
2278
        avctx->extradata_size = esize;
2279
        memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2280
        memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2281

    
2282
        skip_bits_long(gb, bits_consumed);
2283
    }
2284

    
2285
    return bits_consumed;
2286
}
2287

    
2288
static int read_stream_mux_config(struct LATMContext *latmctx,
2289
                                  GetBitContext *gb)
2290
{
2291
    int ret, audio_mux_version = get_bits(gb, 1);
2292

    
2293
    latmctx->audio_mux_version_A = 0;
2294
    if (audio_mux_version)
2295
        latmctx->audio_mux_version_A = get_bits(gb, 1);
2296

    
2297
    if (!latmctx->audio_mux_version_A) {
2298

    
2299
        if (audio_mux_version)
2300
            latm_get_value(gb);                 // taraFullness
2301

    
2302
        skip_bits(gb, 1);                       // allStreamSameTimeFraming
2303
        skip_bits(gb, 6);                       // numSubFrames
2304
        // numPrograms
2305
        if (get_bits(gb, 4)) {                  // numPrograms
2306
            av_log_missing_feature(latmctx->aac_ctx.avctx,
2307
                                   "multiple programs are not supported\n", 1);
2308
            return AVERROR_PATCHWELCOME;
2309
        }
2310

    
2311
        // for each program (which there is only on in DVB)
2312

    
2313
        // for each layer (which there is only on in DVB)
2314
        if (get_bits(gb, 3)) {                   // numLayer
2315
            av_log_missing_feature(latmctx->aac_ctx.avctx,
2316
                                   "multiple layers are not supported\n", 1);
2317
            return AVERROR_PATCHWELCOME;
2318
        }
2319

    
2320
        // for all but first stream: use_same_config = get_bits(gb, 1);
2321
        if (!audio_mux_version) {
2322
            if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2323
                return ret;
2324
        } else {
2325
            int ascLen = latm_get_value(gb);
2326
            if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2327
                return ret;
2328
            ascLen -= ret;
2329
            skip_bits_long(gb, ascLen);
2330
        }
2331

    
2332
        latmctx->frame_length_type = get_bits(gb, 3);
2333
        switch (latmctx->frame_length_type) {
2334
        case 0:
2335
            skip_bits(gb, 8);       // latmBufferFullness
2336
            break;
2337
        case 1:
2338
            latmctx->frame_length = get_bits(gb, 9);
2339
            break;
2340
        case 3:
2341
        case 4:
2342
        case 5:
2343
            skip_bits(gb, 6);       // CELP frame length table index
2344
            break;
2345
        case 6:
2346
        case 7:
2347
            skip_bits(gb, 1);       // HVXC frame length table index
2348
            break;
2349
        }
2350

    
2351
        if (get_bits(gb, 1)) {                  // other data
2352
            if (audio_mux_version) {
2353
                latm_get_value(gb);             // other_data_bits
2354
            } else {
2355
                int esc;
2356
                do {
2357
                    esc = get_bits(gb, 1);
2358
                    skip_bits(gb, 8);
2359
                } while (esc);
2360
            }
2361
        }
2362

    
2363
        if (get_bits(gb, 1))                     // crc present
2364
            skip_bits(gb, 8);                    // config_crc
2365
    }
2366

    
2367
    return 0;
2368
}
2369

    
2370
static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2371
{
2372
    uint8_t tmp;
2373

    
2374
    if (ctx->frame_length_type == 0) {
2375
        int mux_slot_length = 0;
2376
        do {
2377
            tmp = get_bits(gb, 8);
2378
            mux_slot_length += tmp;
2379
        } while (tmp == 255);
2380
        return mux_slot_length;
2381
    } else if (ctx->frame_length_type == 1) {
2382
        return ctx->frame_length;
2383
    } else if (ctx->frame_length_type == 3 ||
2384
               ctx->frame_length_type == 5 ||
2385
               ctx->frame_length_type == 7) {
2386
        skip_bits(gb, 2);          // mux_slot_length_coded
2387
    }
2388
    return 0;
2389
}
2390

    
2391
static int read_audio_mux_element(struct LATMContext *latmctx,
2392
                                  GetBitContext *gb)
2393
{
2394
    int err;
2395
    uint8_t use_same_mux = get_bits(gb, 1);
2396
    if (!use_same_mux) {
2397
        if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2398
            return err;
2399
    } else if (!latmctx->aac_ctx.avctx->extradata) {
2400
        av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2401
               "no decoder config found\n");
2402
        return AVERROR(EAGAIN);
2403
    }
2404
    if (latmctx->audio_mux_version_A == 0) {
2405
        int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2406
        if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2407
            av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2408
            return AVERROR_INVALIDDATA;
2409
        } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2410
            av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2411
                   "frame length mismatch %d << %d\n",
2412
                   mux_slot_length_bytes * 8, get_bits_left(gb));
2413
            return AVERROR_INVALIDDATA;
2414
        }
2415
    }
2416
    return 0;
2417
}
2418

    
2419

    
2420
static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2421
                             AVPacket *avpkt)
2422
{
2423
    struct LATMContext *latmctx = avctx->priv_data;
2424
    int                 muxlength, err;
2425
    GetBitContext       gb;
2426

    
2427
    if (avpkt->size == 0)
2428
        return 0;
2429

    
2430
    init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2431

    
2432
    // check for LOAS sync word
2433
    if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2434
        return AVERROR_INVALIDDATA;
2435

    
2436
    muxlength = get_bits(&gb, 13) + 3;
2437
    // not enough data, the parser should have sorted this
2438
    if (muxlength > avpkt->size)
2439
        return AVERROR_INVALIDDATA;
2440

    
2441
    if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2442
        return err;
2443

    
2444
    if (!latmctx->initialized) {
2445
        if (!avctx->extradata) {
2446
            *out_size = 0;
2447
            return avpkt->size;
2448
        } else {
2449
            if ((err = aac_decode_init(avctx)) < 0)
2450
                return err;
2451
            latmctx->initialized = 1;
2452
        }
2453
    }
2454

    
2455
    if (show_bits(&gb, 12) == 0xfff) {
2456
        av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2457
               "ADTS header detected, probably as result of configuration "
2458
               "misparsing\n");
2459
        return AVERROR_INVALIDDATA;
2460
    }
2461

    
2462
    if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2463
        return err;
2464

    
2465
    return muxlength;
2466
}
2467

    
2468
av_cold static int latm_decode_init(AVCodecContext *avctx)
2469
{
2470
    struct LATMContext *latmctx = avctx->priv_data;
2471
    int ret;
2472

    
2473
    ret = aac_decode_init(avctx);
2474

    
2475
    if (avctx->extradata_size > 0) {
2476
        latmctx->initialized = !ret;
2477
    } else {
2478
        latmctx->initialized = 0;
2479
    }
2480

    
2481
    return ret;
2482
}
2483

    
2484

    
2485
AVCodec ff_aac_decoder = {
2486
    "aac",
2487
    AVMEDIA_TYPE_AUDIO,
2488
    CODEC_ID_AAC,
2489
    sizeof(AACContext),
2490
    aac_decode_init,
2491
    NULL,
2492
    aac_decode_close,
2493
    aac_decode_frame,
2494
    .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2495
    .sample_fmts = (const enum AVSampleFormat[]) {
2496
        AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2497
    },
2498
    .channel_layouts = aac_channel_layout,
2499
};
2500

    
2501
/*
2502
    Note: This decoder filter is intended to decode LATM streams transferred
2503
    in MPEG transport streams which only contain one program.
2504
    To do a more complex LATM demuxing a separate LATM demuxer should be used.
2505
*/
2506
AVCodec ff_aac_latm_decoder = {
2507
    .name = "aac_latm",
2508
    .type = AVMEDIA_TYPE_AUDIO,
2509
    .id   = CODEC_ID_AAC_LATM,
2510
    .priv_data_size = sizeof(struct LATMContext),
2511
    .init   = latm_decode_init,
2512
    .close  = aac_decode_close,
2513
    .decode = latm_decode_frame,
2514
    .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2515
    .sample_fmts = (const enum AVSampleFormat[]) {
2516
        AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2517
    },
2518
    .channel_layouts = aac_channel_layout,
2519
};