3
* Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4
* Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7
* Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8
* Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10
* This file is part of Libav.
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.
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.
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
30
* @author Oded Shimon ( ods15 ods15 dyndns org )
31
* @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38
* N (code in SoC repo) gain control
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
48
* Y frequency domain prediction
49
* Y Perceptual Noise Substitution
51
* N Scalable Inverse AAC Quantization
52
* N Frequency Selective Switch
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
61
* N Silence Compression
64
* N Structured Audio tools
65
* N Structured Audio Sample Bank Format
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)
75
* N Direct Stream Transfer
77
* Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78
* - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88
#include "fmtconvert.h"
95
#include "aacdectab.h"
96
#include "cbrt_tablegen.h"
99
#include "mpeg4audio.h"
100
#include "aacadtsdec.h"
108
# include "arm/aac.h"
116
static VLC vlc_scalefactors;
117
static VLC vlc_spectral[11];
119
static const char overread_err[] = "Input buffer exhausted before END element found\n";
121
static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
123
// For PCE based channel configurations map the channels solely based on tags.
124
if (!ac->m4ac.chan_config) {
125
return ac->tag_che_map[type][elem_id];
127
// For indexed channel configurations map the channels solely based on position.
128
switch (ac->m4ac.chan_config) {
130
if (ac->tags_mapped == 3 && type == TYPE_CPE) {
132
return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
135
/* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
136
instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
137
encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
138
if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
140
return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
143
if (ac->tags_mapped == 2 && type == TYPE_CPE) {
145
return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
148
if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
150
return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
154
if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
156
return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
157
} else if (ac->m4ac.chan_config == 2) {
161
if (!ac->tags_mapped && type == TYPE_SCE) {
163
return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
171
* Check for the channel element in the current channel position configuration.
172
* If it exists, make sure the appropriate element is allocated and map the
173
* channel order to match the internal Libav channel layout.
175
* @param che_pos current channel position configuration
176
* @param type channel element type
177
* @param id channel element id
178
* @param channels count of the number of channels in the configuration
180
* @return Returns error status. 0 - OK, !0 - error
182
static av_cold int che_configure(AACContext *ac,
183
enum ChannelPosition che_pos[4][MAX_ELEM_ID],
184
int type, int id, int *channels)
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, &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;
198
if (ac->che[type][id])
199
ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
200
av_freep(&ac->che[type][id]);
206
* Configure output channel order based on the current program configuration element.
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.
211
* @return Returns error status. 0 - OK, !0 - error
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)
218
AVCodecContext *avctx = ac->avctx;
219
int i, type, channels = 0, ret;
221
if (new_che_pos != che_pos)
222
memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
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],
233
memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
235
avctx->channel_layout = aac_channel_layout[channel_config - 1];
237
/* Allocate or free elements depending on if they are in the
238
* current program configuration.
240
* Set up default 1:1 output mapping.
242
* For a 5.1 stream the output order will be:
243
* [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
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)))
253
memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
255
avctx->channel_layout = 0;
258
avctx->channels = channels;
260
ac->output_configured = oc_type;
266
* Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
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
272
static void decode_channel_map(enum ChannelPosition *cpe_map,
273
enum ChannelPosition *sce_map,
274
enum ChannelPosition type,
275
GetBitContext *gb, int 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;
284
* Decode program configuration element; reference: table 4.2.
286
* @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
288
* @return Returns error status. 0 - OK, !0 - error
290
static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
291
enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
294
int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
297
skip_bits(gb, 2); // object_type
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");
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);
311
skip_bits(gb, 4); // mono_mixdown_tag
313
skip_bits(gb, 4); // stereo_mixdown_tag
316
skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
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 );
323
skip_bits_long(gb, 4 * num_assoc_data);
325
decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
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);
335
skip_bits_long(gb, comment_len);
340
* Set up channel positions based on a default channel configuration
341
* as specified in table 1.17.
343
* @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
345
* @return Returns error status. 0 - OK, !0 - error
347
static av_cold int set_default_channel_config(AVCodecContext *avctx,
348
enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
351
if (channel_config < 1 || channel_config > 7) {
352
av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
357
/* default channel configurations:
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
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
386
* Decode GA "General Audio" specific configuration; reference: table 4.1.
388
* @param ac pointer to AACContext, may be null
389
* @param avctx pointer to AVCCodecContext, used for logging
391
* @return Returns error status. 0 - OK, !0 - error
393
static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
395
MPEG4AudioConfig *m4ac,
398
enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
399
int extension_flag, ret;
401
if (get_bits1(gb)) { // frameLengthFlag
402
av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
406
if (get_bits1(gb)) // dependsOnCoreCoder
407
skip_bits(gb, 14); // coreCoderDelay
408
extension_flag = get_bits1(gb);
410
if (m4ac->object_type == AOT_AAC_SCALABLE ||
411
m4ac->object_type == AOT_ER_AAC_SCALABLE)
412
skip_bits(gb, 3); // layerNr
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)))
420
if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
423
if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
426
if (extension_flag) {
427
switch (m4ac->object_type) {
429
skip_bits(gb, 5); // numOfSubFrame
430
skip_bits(gb, 11); // layer_length
434
case AOT_ER_AAC_SCALABLE:
436
skip_bits(gb, 3); /* aacSectionDataResilienceFlag
437
* aacScalefactorDataResilienceFlag
438
* aacSpectralDataResilienceFlag
442
skip_bits1(gb); // extensionFlag3 (TBD in version 3)
448
* Decode audio specific configuration; reference: table 1.13.
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
456
* @return Returns error status or number of consumed bits. <0 - error
458
static int decode_audio_specific_config(AACContext *ac,
459
AVCodecContext *avctx,
460
MPEG4AudioConfig *m4ac,
461
const uint8_t *data, int data_size)
466
av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
467
for (i = 0; i < avctx->extradata_size; i++)
468
av_dlog(avctx, "%02x ", avctx->extradata[i]);
469
av_dlog(avctx, "\n");
471
init_get_bits(&gb, data, data_size * 8);
473
if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
475
if (m4ac->sampling_index > 12) {
476
av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
479
if (m4ac->sbr == 1 && m4ac->ps == -1)
482
skip_bits_long(&gb, i);
484
switch (m4ac->object_type) {
488
if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
492
av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
493
m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
497
av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
498
m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
499
m4ac->sample_rate, m4ac->sbr, m4ac->ps);
501
return get_bits_count(&gb);
505
* linear congruential pseudorandom number generator
507
* @param previous_val pointer to the current state of the generator
509
* @return Returns a 32-bit pseudorandom integer
511
static av_always_inline int lcg_random(int previous_val)
513
return previous_val * 1664525 + 1013904223;
516
static av_always_inline void reset_predict_state(PredictorState *ps)
526
static void reset_all_predictors(PredictorState *ps)
529
for (i = 0; i < MAX_PREDICTORS; i++)
530
reset_predict_state(&ps[i]);
533
static void reset_predictor_group(PredictorState *ps, int group_num)
536
for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
537
reset_predict_state(&ps[i]);
540
#define AAC_INIT_VLC_STATIC(num, size) \
541
INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
542
ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
543
ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
546
static av_cold int aac_decode_init(AVCodecContext *avctx)
548
AACContext *ac = avctx->priv_data;
549
float output_scale_factor;
552
ac->m4ac.sample_rate = avctx->sample_rate;
554
if (avctx->extradata_size > 0) {
555
if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
557
avctx->extradata_size) < 0)
561
if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
562
avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
563
output_scale_factor = 1.0 / 32768.0;
565
avctx->sample_fmt = AV_SAMPLE_FMT_S16;
566
output_scale_factor = 1.0;
569
AAC_INIT_VLC_STATIC( 0, 304);
570
AAC_INIT_VLC_STATIC( 1, 270);
571
AAC_INIT_VLC_STATIC( 2, 550);
572
AAC_INIT_VLC_STATIC( 3, 300);
573
AAC_INIT_VLC_STATIC( 4, 328);
574
AAC_INIT_VLC_STATIC( 5, 294);
575
AAC_INIT_VLC_STATIC( 6, 306);
576
AAC_INIT_VLC_STATIC( 7, 268);
577
AAC_INIT_VLC_STATIC( 8, 510);
578
AAC_INIT_VLC_STATIC( 9, 366);
579
AAC_INIT_VLC_STATIC(10, 462);
583
dsputil_init(&ac->dsp, avctx);
584
ff_fmt_convert_init(&ac->fmt_conv, avctx);
586
ac->random_state = 0x1f2e3d4c;
590
INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
591
ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
592
ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
595
ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
596
ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
597
ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
598
// window initialization
599
ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
600
ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
601
ff_init_ff_sine_windows(10);
602
ff_init_ff_sine_windows( 7);
610
* Skip data_stream_element; reference: table 4.10.
612
static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
614
int byte_align = get_bits1(gb);
615
int count = get_bits(gb, 8);
617
count += get_bits(gb, 8);
621
if (get_bits_left(gb) < 8 * count) {
622
av_log(ac->avctx, AV_LOG_ERROR, overread_err);
625
skip_bits_long(gb, 8 * count);
629
static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
634
ics->predictor_reset_group = get_bits(gb, 5);
635
if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
636
av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
640
for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
641
ics->prediction_used[sfb] = get_bits1(gb);
647
* Decode Long Term Prediction data; reference: table 4.xx.
649
static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
650
GetBitContext *gb, uint8_t max_sfb)
654
ltp->lag = get_bits(gb, 11);
655
ltp->coef = ltp_coef[get_bits(gb, 3)];
656
for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
657
ltp->used[sfb] = get_bits1(gb);
661
* Decode Individual Channel Stream info; reference: table 4.6.
663
* @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
665
static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
666
GetBitContext *gb, int common_window)
669
av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
670
memset(ics, 0, sizeof(IndividualChannelStream));
673
ics->window_sequence[1] = ics->window_sequence[0];
674
ics->window_sequence[0] = get_bits(gb, 2);
675
ics->use_kb_window[1] = ics->use_kb_window[0];
676
ics->use_kb_window[0] = get_bits1(gb);
677
ics->num_window_groups = 1;
678
ics->group_len[0] = 1;
679
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
681
ics->max_sfb = get_bits(gb, 4);
682
for (i = 0; i < 7; i++) {
684
ics->group_len[ics->num_window_groups - 1]++;
686
ics->num_window_groups++;
687
ics->group_len[ics->num_window_groups - 1] = 1;
690
ics->num_windows = 8;
691
ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
692
ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
693
ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
694
ics->predictor_present = 0;
696
ics->max_sfb = get_bits(gb, 6);
697
ics->num_windows = 1;
698
ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
699
ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
700
ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
701
ics->predictor_present = get_bits1(gb);
702
ics->predictor_reset_group = 0;
703
if (ics->predictor_present) {
704
if (ac->m4ac.object_type == AOT_AAC_MAIN) {
705
if (decode_prediction(ac, ics, gb)) {
706
memset(ics, 0, sizeof(IndividualChannelStream));
709
} else if (ac->m4ac.object_type == AOT_AAC_LC) {
710
av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
711
memset(ics, 0, sizeof(IndividualChannelStream));
714
if ((ics->ltp.present = get_bits(gb, 1)))
715
decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
720
if (ics->max_sfb > ics->num_swb) {
721
av_log(ac->avctx, AV_LOG_ERROR,
722
"Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
723
ics->max_sfb, ics->num_swb);
724
memset(ics, 0, sizeof(IndividualChannelStream));
732
* Decode band types (section_data payload); reference: table 4.46.
734
* @param band_type array of the used band type
735
* @param band_type_run_end array of the last scalefactor band of a band type run
737
* @return Returns error status. 0 - OK, !0 - error
739
static int decode_band_types(AACContext *ac, enum BandType band_type[120],
740
int band_type_run_end[120], GetBitContext *gb,
741
IndividualChannelStream *ics)
744
const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
745
for (g = 0; g < ics->num_window_groups; g++) {
747
while (k < ics->max_sfb) {
748
uint8_t sect_end = k;
750
int sect_band_type = get_bits(gb, 4);
751
if (sect_band_type == 12) {
752
av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
755
while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
756
sect_end += sect_len_incr;
757
sect_end += sect_len_incr;
758
if (get_bits_left(gb) < 0) {
759
av_log(ac->avctx, AV_LOG_ERROR, overread_err);
762
if (sect_end > ics->max_sfb) {
763
av_log(ac->avctx, AV_LOG_ERROR,
764
"Number of bands (%d) exceeds limit (%d).\n",
765
sect_end, ics->max_sfb);
768
for (; k < sect_end; k++) {
769
band_type [idx] = sect_band_type;
770
band_type_run_end[idx++] = sect_end;
778
* Decode scalefactors; reference: table 4.47.
780
* @param global_gain first scalefactor value as scalefactors are differentially coded
781
* @param band_type array of the used band type
782
* @param band_type_run_end array of the last scalefactor band of a band type run
783
* @param sf array of scalefactors or intensity stereo positions
785
* @return Returns error status. 0 - OK, !0 - error
787
static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
788
unsigned int global_gain,
789
IndividualChannelStream *ics,
790
enum BandType band_type[120],
791
int band_type_run_end[120])
794
int offset[3] = { global_gain, global_gain - 90, 0 };
797
static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
798
for (g = 0; g < ics->num_window_groups; g++) {
799
for (i = 0; i < ics->max_sfb;) {
800
int run_end = band_type_run_end[idx];
801
if (band_type[idx] == ZERO_BT) {
802
for (; i < run_end; i++, idx++)
804
} else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
805
for (; i < run_end; i++, idx++) {
806
offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
807
clipped_offset = av_clip(offset[2], -155, 100);
808
if (offset[2] != clipped_offset) {
809
av_log_ask_for_sample(ac->avctx, "Intensity stereo "
810
"position clipped (%d -> %d).\nIf you heard an "
811
"audible artifact, there may be a bug in the "
812
"decoder. ", offset[2], clipped_offset);
814
sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
816
} else if (band_type[idx] == NOISE_BT) {
817
for (; i < run_end; i++, idx++) {
818
if (noise_flag-- > 0)
819
offset[1] += get_bits(gb, 9) - 256;
821
offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
822
clipped_offset = av_clip(offset[1], -100, 155);
823
if (offset[1] != clipped_offset) {
824
av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
825
"(%d -> %d).\nIf you heard an audible "
826
"artifact, there may be a bug in the decoder. ",
827
offset[1], clipped_offset);
829
sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
832
for (; i < run_end; i++, idx++) {
833
offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
834
if (offset[0] > 255U) {
835
av_log(ac->avctx, AV_LOG_ERROR,
836
"%s (%d) out of range.\n", sf_str[0], offset[0]);
839
sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
848
* Decode pulse data; reference: table 4.7.
850
static int decode_pulses(Pulse *pulse, GetBitContext *gb,
851
const uint16_t *swb_offset, int num_swb)
854
pulse->num_pulse = get_bits(gb, 2) + 1;
855
pulse_swb = get_bits(gb, 6);
856
if (pulse_swb >= num_swb)
858
pulse->pos[0] = swb_offset[pulse_swb];
859
pulse->pos[0] += get_bits(gb, 5);
860
if (pulse->pos[0] > 1023)
862
pulse->amp[0] = get_bits(gb, 4);
863
for (i = 1; i < pulse->num_pulse; i++) {
864
pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
865
if (pulse->pos[i] > 1023)
867
pulse->amp[i] = get_bits(gb, 4);
873
* Decode Temporal Noise Shaping data; reference: table 4.48.
875
* @return Returns error status. 0 - OK, !0 - error
877
static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
878
GetBitContext *gb, const IndividualChannelStream *ics)
880
int w, filt, i, coef_len, coef_res, coef_compress;
881
const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
882
const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
883
for (w = 0; w < ics->num_windows; w++) {
884
if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
885
coef_res = get_bits1(gb);
887
for (filt = 0; filt < tns->n_filt[w]; filt++) {
889
tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
891
if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
892
av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
893
tns->order[w][filt], tns_max_order);
894
tns->order[w][filt] = 0;
897
if (tns->order[w][filt]) {
898
tns->direction[w][filt] = get_bits1(gb);
899
coef_compress = get_bits1(gb);
900
coef_len = coef_res + 3 - coef_compress;
901
tmp2_idx = 2 * coef_compress + coef_res;
903
for (i = 0; i < tns->order[w][filt]; i++)
904
tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
913
* Decode Mid/Side data; reference: table 4.54.
915
* @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
916
* [1] mask is decoded from bitstream; [2] mask is all 1s;
917
* [3] reserved for scalable AAC
919
static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
923
if (ms_present == 1) {
924
for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
925
cpe->ms_mask[idx] = get_bits1(gb);
926
} else if (ms_present == 2) {
927
memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
932
static inline float *VMUL2(float *dst, const float *v, unsigned idx,
936
*dst++ = v[idx & 15] * s;
937
*dst++ = v[idx>>4 & 15] * s;
943
static inline float *VMUL4(float *dst, const float *v, unsigned idx,
947
*dst++ = v[idx & 3] * s;
948
*dst++ = v[idx>>2 & 3] * s;
949
*dst++ = v[idx>>4 & 3] * s;
950
*dst++ = v[idx>>6 & 3] * s;
956
static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
957
unsigned sign, const float *scale)
959
union float754 s0, s1;
961
s0.f = s1.f = *scale;
962
s0.i ^= sign >> 1 << 31;
965
*dst++ = v[idx & 15] * s0.f;
966
*dst++ = v[idx>>4 & 15] * s1.f;
973
static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
974
unsigned sign, const float *scale)
976
unsigned nz = idx >> 12;
977
union float754 s = { .f = *scale };
980
t.i = s.i ^ (sign & 1U<<31);
981
*dst++ = v[idx & 3] * t.f;
983
sign <<= nz & 1; nz >>= 1;
984
t.i = s.i ^ (sign & 1U<<31);
985
*dst++ = v[idx>>2 & 3] * t.f;
987
sign <<= nz & 1; nz >>= 1;
988
t.i = s.i ^ (sign & 1U<<31);
989
*dst++ = v[idx>>4 & 3] * t.f;
991
sign <<= nz & 1; nz >>= 1;
992
t.i = s.i ^ (sign & 1U<<31);
993
*dst++ = v[idx>>6 & 3] * t.f;
1000
* Decode spectral data; reference: table 4.50.
1001
* Dequantize and scale spectral data; reference: 4.6.3.3.
1003
* @param coef array of dequantized, scaled spectral data
1004
* @param sf array of scalefactors or intensity stereo positions
1005
* @param pulse_present set if pulses are present
1006
* @param pulse pointer to pulse data struct
1007
* @param band_type array of the used band type
1009
* @return Returns error status. 0 - OK, !0 - error
1011
static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1012
GetBitContext *gb, const float sf[120],
1013
int pulse_present, const Pulse *pulse,
1014
const IndividualChannelStream *ics,
1015
enum BandType band_type[120])
1017
int i, k, g, idx = 0;
1018
const int c = 1024 / ics->num_windows;
1019
const uint16_t *offsets = ics->swb_offset;
1020
float *coef_base = coef;
1022
for (g = 0; g < ics->num_windows; g++)
1023
memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1025
for (g = 0; g < ics->num_window_groups; g++) {
1026
unsigned g_len = ics->group_len[g];
1028
for (i = 0; i < ics->max_sfb; i++, idx++) {
1029
const unsigned cbt_m1 = band_type[idx] - 1;
1030
float *cfo = coef + offsets[i];
1031
int off_len = offsets[i + 1] - offsets[i];
1034
if (cbt_m1 >= INTENSITY_BT2 - 1) {
1035
for (group = 0; group < g_len; group++, cfo+=128) {
1036
memset(cfo, 0, off_len * sizeof(float));
1038
} else if (cbt_m1 == NOISE_BT - 1) {
1039
for (group = 0; group < g_len; group++, cfo+=128) {
1043
for (k = 0; k < off_len; k++) {
1044
ac->random_state = lcg_random(ac->random_state);
1045
cfo[k] = ac->random_state;
1048
band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1049
scale = sf[idx] / sqrtf(band_energy);
1050
ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1053
const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1054
const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1055
VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1056
OPEN_READER(re, gb);
1058
switch (cbt_m1 >> 1) {
1060
for (group = 0; group < g_len; group++, cfo+=128) {
1068
UPDATE_CACHE(re, gb);
1069
GET_VLC(code, re, gb, vlc_tab, 8, 2);
1070
cb_idx = cb_vector_idx[code];
1071
cf = VMUL4(cf, vq, cb_idx, sf + idx);
1077
for (group = 0; group < g_len; group++, cfo+=128) {
1087
UPDATE_CACHE(re, gb);
1088
GET_VLC(code, re, gb, vlc_tab, 8, 2);
1089
cb_idx = cb_vector_idx[code];
1090
nnz = cb_idx >> 8 & 15;
1091
bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1092
LAST_SKIP_BITS(re, gb, nnz);
1093
cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1099
for (group = 0; group < g_len; group++, cfo+=128) {
1107
UPDATE_CACHE(re, gb);
1108
GET_VLC(code, re, gb, vlc_tab, 8, 2);
1109
cb_idx = cb_vector_idx[code];
1110
cf = VMUL2(cf, vq, cb_idx, sf + idx);
1117
for (group = 0; group < g_len; group++, cfo+=128) {
1127
UPDATE_CACHE(re, gb);
1128
GET_VLC(code, re, gb, vlc_tab, 8, 2);
1129
cb_idx = cb_vector_idx[code];
1130
nnz = cb_idx >> 8 & 15;
1131
sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1132
LAST_SKIP_BITS(re, gb, nnz);
1133
cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1139
for (group = 0; group < g_len; group++, cfo+=128) {
1141
uint32_t *icf = (uint32_t *) cf;
1151
UPDATE_CACHE(re, gb);
1152
GET_VLC(code, re, gb, vlc_tab, 8, 2);
1160
cb_idx = cb_vector_idx[code];
1163
bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1164
LAST_SKIP_BITS(re, gb, nnz);
1166
for (j = 0; j < 2; j++) {
1170
/* The total length of escape_sequence must be < 22 bits according
1171
to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1172
UPDATE_CACHE(re, gb);
1173
b = GET_CACHE(re, gb);
1174
b = 31 - av_log2(~b);
1177
av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1181
SKIP_BITS(re, gb, b + 1);
1183
n = (1 << b) + SHOW_UBITS(re, gb, b);
1184
LAST_SKIP_BITS(re, gb, b);
1185
*icf++ = cbrt_tab[n] | (bits & 1U<<31);
1188
unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1189
*icf++ = (bits & 1U<<31) | v;
1196
ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1200
CLOSE_READER(re, gb);
1206
if (pulse_present) {
1208
for (i = 0; i < pulse->num_pulse; i++) {
1209
float co = coef_base[ pulse->pos[i] ];
1210
while (offsets[idx + 1] <= pulse->pos[i])
1212
if (band_type[idx] != NOISE_BT && sf[idx]) {
1213
float ico = -pulse->amp[i];
1216
ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1218
coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1225
static av_always_inline float flt16_round(float pf)
1229
tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1233
static av_always_inline float flt16_even(float pf)
1237
tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1241
static av_always_inline float flt16_trunc(float pf)
1245
pun.i &= 0xFFFF0000U;
1249
static av_always_inline void predict(PredictorState *ps, float *coef,
1252
const float a = 0.953125; // 61.0 / 64
1253
const float alpha = 0.90625; // 29.0 / 32
1257
float r0 = ps->r0, r1 = ps->r1;
1258
float cor0 = ps->cor0, cor1 = ps->cor1;
1259
float var0 = ps->var0, var1 = ps->var1;
1261
k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1262
k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1264
pv = flt16_round(k1 * r0 + k2 * r1);
1271
ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1272
ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1273
ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1274
ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1276
ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1277
ps->r0 = flt16_trunc(a * e0);
1281
* Apply AAC-Main style frequency domain prediction.
1283
static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1287
if (!sce->ics.predictor_initialized) {
1288
reset_all_predictors(sce->predictor_state);
1289
sce->ics.predictor_initialized = 1;
1292
if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1293
for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1294
for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1295
predict(&sce->predictor_state[k], &sce->coeffs[k],
1296
sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1299
if (sce->ics.predictor_reset_group)
1300
reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1302
reset_all_predictors(sce->predictor_state);
1306
* Decode an individual_channel_stream payload; reference: table 4.44.
1308
* @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1309
* @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1311
* @return Returns error status. 0 - OK, !0 - error
1313
static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1314
GetBitContext *gb, int common_window, int scale_flag)
1317
TemporalNoiseShaping *tns = &sce->tns;
1318
IndividualChannelStream *ics = &sce->ics;
1319
float *out = sce->coeffs;
1320
int global_gain, pulse_present = 0;
1322
/* This assignment is to silence a GCC warning about the variable being used
1323
* uninitialized when in fact it always is.
1325
pulse.num_pulse = 0;
1327
global_gain = get_bits(gb, 8);
1329
if (!common_window && !scale_flag) {
1330
if (decode_ics_info(ac, ics, gb, 0) < 0)
1334
if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1336
if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1341
if ((pulse_present = get_bits1(gb))) {
1342
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1343
av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1346
if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1347
av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1351
if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1353
if (get_bits1(gb)) {
1354
av_log_missing_feature(ac->avctx, "SSR", 1);
1359
if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1362
if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1363
apply_prediction(ac, sce);
1369
* Mid/Side stereo decoding; reference: 4.6.8.1.3.
1371
static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1373
const IndividualChannelStream *ics = &cpe->ch[0].ics;
1374
float *ch0 = cpe->ch[0].coeffs;
1375
float *ch1 = cpe->ch[1].coeffs;
1376
int g, i, group, idx = 0;
1377
const uint16_t *offsets = ics->swb_offset;
1378
for (g = 0; g < ics->num_window_groups; g++) {
1379
for (i = 0; i < ics->max_sfb; i++, idx++) {
1380
if (cpe->ms_mask[idx] &&
1381
cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1382
for (group = 0; group < ics->group_len[g]; group++) {
1383
ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1384
ch1 + group * 128 + offsets[i],
1385
offsets[i+1] - offsets[i]);
1389
ch0 += ics->group_len[g] * 128;
1390
ch1 += ics->group_len[g] * 128;
1395
* intensity stereo decoding; reference: 4.6.8.2.3
1397
* @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1398
* [1] mask is decoded from bitstream; [2] mask is all 1s;
1399
* [3] reserved for scalable AAC
1401
static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1403
const IndividualChannelStream *ics = &cpe->ch[1].ics;
1404
SingleChannelElement *sce1 = &cpe->ch[1];
1405
float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1406
const uint16_t *offsets = ics->swb_offset;
1407
int g, group, i, idx = 0;
1410
for (g = 0; g < ics->num_window_groups; g++) {
1411
for (i = 0; i < ics->max_sfb;) {
1412
if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1413
const int bt_run_end = sce1->band_type_run_end[idx];
1414
for (; i < bt_run_end; i++, idx++) {
1415
c = -1 + 2 * (sce1->band_type[idx] - 14);
1417
c *= 1 - 2 * cpe->ms_mask[idx];
1418
scale = c * sce1->sf[idx];
1419
for (group = 0; group < ics->group_len[g]; group++)
1420
ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1421
coef0 + group * 128 + offsets[i],
1423
offsets[i + 1] - offsets[i]);
1426
int bt_run_end = sce1->band_type_run_end[idx];
1427
idx += bt_run_end - i;
1431
coef0 += ics->group_len[g] * 128;
1432
coef1 += ics->group_len[g] * 128;
1437
* Decode a channel_pair_element; reference: table 4.4.
1439
* @return Returns error status. 0 - OK, !0 - error
1441
static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1443
int i, ret, common_window, ms_present = 0;
1445
common_window = get_bits1(gb);
1446
if (common_window) {
1447
if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1449
i = cpe->ch[1].ics.use_kb_window[0];
1450
cpe->ch[1].ics = cpe->ch[0].ics;
1451
cpe->ch[1].ics.use_kb_window[1] = i;
1452
if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1453
if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1454
decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1455
ms_present = get_bits(gb, 2);
1456
if (ms_present == 3) {
1457
av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1459
} else if (ms_present)
1460
decode_mid_side_stereo(cpe, gb, ms_present);
1462
if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1464
if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1467
if (common_window) {
1469
apply_mid_side_stereo(ac, cpe);
1470
if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1471
apply_prediction(ac, &cpe->ch[0]);
1472
apply_prediction(ac, &cpe->ch[1]);
1476
apply_intensity_stereo(ac, cpe, ms_present);
1480
static const float cce_scale[] = {
1481
1.09050773266525765921, //2^(1/8)
1482
1.18920711500272106672, //2^(1/4)
1488
* Decode coupling_channel_element; reference: table 4.8.
1490
* @return Returns error status. 0 - OK, !0 - error
1492
static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1498
SingleChannelElement *sce = &che->ch[0];
1499
ChannelCoupling *coup = &che->coup;
1501
coup->coupling_point = 2 * get_bits1(gb);
1502
coup->num_coupled = get_bits(gb, 3);
1503
for (c = 0; c <= coup->num_coupled; c++) {
1505
coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1506
coup->id_select[c] = get_bits(gb, 4);
1507
if (coup->type[c] == TYPE_CPE) {
1508
coup->ch_select[c] = get_bits(gb, 2);
1509
if (coup->ch_select[c] == 3)
1512
coup->ch_select[c] = 2;
1514
coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1516
sign = get_bits(gb, 1);
1517
scale = cce_scale[get_bits(gb, 2)];
1519
if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1522
for (c = 0; c < num_gain; c++) {
1526
float gain_cache = 1.;
1528
cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1529
gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1530
gain_cache = powf(scale, -gain);
1532
if (coup->coupling_point == AFTER_IMDCT) {
1533
coup->gain[c][0] = gain_cache;
1535
for (g = 0; g < sce->ics.num_window_groups; g++) {
1536
for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1537
if (sce->band_type[idx] != ZERO_BT) {
1539
int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1547
gain_cache = powf(scale, -t) * s;
1550
coup->gain[c][idx] = gain_cache;
1560
* Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1562
* @return Returns number of bytes consumed.
1564
static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1568
int num_excl_chan = 0;
1571
for (i = 0; i < 7; i++)
1572
che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1573
} while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1575
return num_excl_chan / 7;
1579
* Decode dynamic range information; reference: table 4.52.
1581
* @param cnt length of TYPE_FIL syntactic element in bytes
1583
* @return Returns number of bytes consumed.
1585
static int decode_dynamic_range(DynamicRangeControl *che_drc,
1586
GetBitContext *gb, int cnt)
1589
int drc_num_bands = 1;
1592
/* pce_tag_present? */
1593
if (get_bits1(gb)) {
1594
che_drc->pce_instance_tag = get_bits(gb, 4);
1595
skip_bits(gb, 4); // tag_reserved_bits
1599
/* excluded_chns_present? */
1600
if (get_bits1(gb)) {
1601
n += decode_drc_channel_exclusions(che_drc, gb);
1604
/* drc_bands_present? */
1605
if (get_bits1(gb)) {
1606
che_drc->band_incr = get_bits(gb, 4);
1607
che_drc->interpolation_scheme = get_bits(gb, 4);
1609
drc_num_bands += che_drc->band_incr;
1610
for (i = 0; i < drc_num_bands; i++) {
1611
che_drc->band_top[i] = get_bits(gb, 8);
1616
/* prog_ref_level_present? */
1617
if (get_bits1(gb)) {
1618
che_drc->prog_ref_level = get_bits(gb, 7);
1619
skip_bits1(gb); // prog_ref_level_reserved_bits
1623
for (i = 0; i < drc_num_bands; i++) {
1624
che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1625
che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1633
* Decode extension data (incomplete); reference: table 4.51.
1635
* @param cnt length of TYPE_FIL syntactic element in bytes
1637
* @return Returns number of bytes consumed
1639
static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1640
ChannelElement *che, enum RawDataBlockType elem_type)
1644
switch (get_bits(gb, 4)) { // extension type
1645
case EXT_SBR_DATA_CRC:
1649
av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1651
} else if (!ac->m4ac.sbr) {
1652
av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1653
skip_bits_long(gb, 8 * cnt - 4);
1655
} else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1656
av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1657
skip_bits_long(gb, 8 * cnt - 4);
1659
} else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1662
output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1666
res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1668
case EXT_DYNAMIC_RANGE:
1669
res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1673
case EXT_DATA_ELEMENT:
1675
skip_bits_long(gb, 8 * cnt - 4);
1682
* Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1684
* @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1685
* @param coef spectral coefficients
1687
static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1688
IndividualChannelStream *ics, int decode)
1690
const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1692
int bottom, top, order, start, end, size, inc;
1693
float lpc[TNS_MAX_ORDER];
1694
float tmp[TNS_MAX_ORDER];
1696
for (w = 0; w < ics->num_windows; w++) {
1697
bottom = ics->num_swb;
1698
for (filt = 0; filt < tns->n_filt[w]; filt++) {
1700
bottom = FFMAX(0, top - tns->length[w][filt]);
1701
order = tns->order[w][filt];
1706
compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1708
start = ics->swb_offset[FFMIN(bottom, mmm)];
1709
end = ics->swb_offset[FFMIN( top, mmm)];
1710
if ((size = end - start) <= 0)
1712
if (tns->direction[w][filt]) {
1722
for (m = 0; m < size; m++, start += inc)
1723
for (i = 1; i <= FFMIN(m, order); i++)
1724
coef[start] -= coef[start - i * inc] * lpc[i - 1];
1727
for (m = 0; m < size; m++, start += inc) {
1728
tmp[0] = coef[start];
1729
for (i = 1; i <= FFMIN(m, order); i++)
1730
coef[start] += tmp[i] * lpc[i - 1];
1731
for (i = order; i > 0; i--)
1732
tmp[i] = tmp[i - 1];
1740
* Apply windowing and MDCT to obtain the spectral
1741
* coefficient from the predicted sample by LTP.
1743
static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1744
float *in, IndividualChannelStream *ics)
1746
const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1747
const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1748
const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1749
const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1751
if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1752
ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1754
memset(in, 0, 448 * sizeof(float));
1755
ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1756
memcpy(in + 576, in + 576, 448 * sizeof(float));
1758
if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1759
ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1761
memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1762
ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1763
memset(in + 1024 + 576, 0, 448 * sizeof(float));
1765
ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1769
* Apply the long term prediction
1771
static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1773
const LongTermPrediction *ltp = &sce->ics.ltp;
1774
const uint16_t *offsets = sce->ics.swb_offset;
1777
if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1778
float *predTime = sce->ret;
1779
float *predFreq = ac->buf_mdct;
1780
int16_t num_samples = 2048;
1782
if (ltp->lag < 1024)
1783
num_samples = ltp->lag + 1024;
1784
for (i = 0; i < num_samples; i++)
1785
predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1786
memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1788
windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1790
if (sce->tns.present)
1791
apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1793
for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1795
for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1796
sce->coeffs[i] += predFreq[i];
1801
* Update the LTP buffer for next frame
1803
static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1805
IndividualChannelStream *ics = &sce->ics;
1806
float *saved = sce->saved;
1807
float *saved_ltp = sce->coeffs;
1808
const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1809
const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1812
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1813
memcpy(saved_ltp, saved, 512 * sizeof(float));
1814
memset(saved_ltp + 576, 0, 448 * sizeof(float));
1815
ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1816
for (i = 0; i < 64; i++)
1817
saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1818
} else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1819
memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1820
memset(saved_ltp + 576, 0, 448 * sizeof(float));
1821
ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1822
for (i = 0; i < 64; i++)
1823
saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1824
} else { // LONG_STOP or ONLY_LONG
1825
ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1826
for (i = 0; i < 512; i++)
1827
saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1830
memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1831
memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1832
memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1836
* Conduct IMDCT and windowing.
1838
static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1840
IndividualChannelStream *ics = &sce->ics;
1841
float *in = sce->coeffs;
1842
float *out = sce->ret;
1843
float *saved = sce->saved;
1844
const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1845
const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1846
const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1847
float *buf = ac->buf_mdct;
1848
float *temp = ac->temp;
1852
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1853
for (i = 0; i < 1024; i += 128)
1854
ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1856
ac->mdct.imdct_half(&ac->mdct, buf, in);
1858
/* window overlapping
1859
* NOTE: To simplify the overlapping code, all 'meaningless' short to long
1860
* and long to short transitions are considered to be short to short
1861
* transitions. This leaves just two cases (long to long and short to short)
1862
* with a little special sauce for EIGHT_SHORT_SEQUENCE.
1864
if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1865
(ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1866
ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1868
memcpy( out, saved, 448 * sizeof(float));
1870
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1871
ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1872
ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1873
ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1874
ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1875
ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1876
memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1878
ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1879
memcpy( out + 576, buf + 64, 448 * sizeof(float));
1884
if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1885
memcpy( saved, temp + 64, 64 * sizeof(float));
1886
ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1887
ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1888
ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1889
memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1890
} else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1891
memcpy( saved, buf + 512, 448 * sizeof(float));
1892
memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1893
} else { // LONG_STOP or ONLY_LONG
1894
memcpy( saved, buf + 512, 512 * sizeof(float));
1899
* Apply dependent channel coupling (applied before IMDCT).
1901
* @param index index into coupling gain array
1903
static void apply_dependent_coupling(AACContext *ac,
1904
SingleChannelElement *target,
1905
ChannelElement *cce, int index)
1907
IndividualChannelStream *ics = &cce->ch[0].ics;
1908
const uint16_t *offsets = ics->swb_offset;
1909
float *dest = target->coeffs;
1910
const float *src = cce->ch[0].coeffs;
1911
int g, i, group, k, idx = 0;
1912
if (ac->m4ac.object_type == AOT_AAC_LTP) {
1913
av_log(ac->avctx, AV_LOG_ERROR,
1914
"Dependent coupling is not supported together with LTP\n");
1917
for (g = 0; g < ics->num_window_groups; g++) {
1918
for (i = 0; i < ics->max_sfb; i++, idx++) {
1919
if (cce->ch[0].band_type[idx] != ZERO_BT) {
1920
const float gain = cce->coup.gain[index][idx];
1921
for (group = 0; group < ics->group_len[g]; group++) {
1922
for (k = offsets[i]; k < offsets[i + 1]; k++) {
1924
dest[group * 128 + k] += gain * src[group * 128 + k];
1929
dest += ics->group_len[g] * 128;
1930
src += ics->group_len[g] * 128;
1935
* Apply independent channel coupling (applied after IMDCT).
1937
* @param index index into coupling gain array
1939
static void apply_independent_coupling(AACContext *ac,
1940
SingleChannelElement *target,
1941
ChannelElement *cce, int index)
1944
const float gain = cce->coup.gain[index][0];
1945
const float *src = cce->ch[0].ret;
1946
float *dest = target->ret;
1947
const int len = 1024 << (ac->m4ac.sbr == 1);
1949
for (i = 0; i < len; i++)
1950
dest[i] += gain * src[i];
1954
* channel coupling transformation interface
1956
* @param apply_coupling_method pointer to (in)dependent coupling function
1958
static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1959
enum RawDataBlockType type, int elem_id,
1960
enum CouplingPoint coupling_point,
1961
void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1965
for (i = 0; i < MAX_ELEM_ID; i++) {
1966
ChannelElement *cce = ac->che[TYPE_CCE][i];
1969
if (cce && cce->coup.coupling_point == coupling_point) {
1970
ChannelCoupling *coup = &cce->coup;
1972
for (c = 0; c <= coup->num_coupled; c++) {
1973
if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1974
if (coup->ch_select[c] != 1) {
1975
apply_coupling_method(ac, &cc->ch[0], cce, index);
1976
if (coup->ch_select[c] != 0)
1979
if (coup->ch_select[c] != 2)
1980
apply_coupling_method(ac, &cc->ch[1], cce, index++);
1982
index += 1 + (coup->ch_select[c] == 3);
1989
* Convert spectral data to float samples, applying all supported tools as appropriate.
1991
static void spectral_to_sample(AACContext *ac)
1994
for (type = 3; type >= 0; type--) {
1995
for (i = 0; i < MAX_ELEM_ID; i++) {
1996
ChannelElement *che = ac->che[type][i];
1998
if (type <= TYPE_CPE)
1999
apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2000
if (ac->m4ac.object_type == AOT_AAC_LTP) {
2001
if (che->ch[0].ics.predictor_present) {
2002
if (che->ch[0].ics.ltp.present)
2003
apply_ltp(ac, &che->ch[0]);
2004
if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2005
apply_ltp(ac, &che->ch[1]);
2008
if (che->ch[0].tns.present)
2009
apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2010
if (che->ch[1].tns.present)
2011
apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2012
if (type <= TYPE_CPE)
2013
apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2014
if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2015
imdct_and_windowing(ac, &che->ch[0]);
2016
if (ac->m4ac.object_type == AOT_AAC_LTP)
2017
update_ltp(ac, &che->ch[0]);
2018
if (type == TYPE_CPE) {
2019
imdct_and_windowing(ac, &che->ch[1]);
2020
if (ac->m4ac.object_type == AOT_AAC_LTP)
2021
update_ltp(ac, &che->ch[1]);
2023
if (ac->m4ac.sbr > 0) {
2024
ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2027
if (type <= TYPE_CCE)
2028
apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2034
static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2037
AACADTSHeaderInfo hdr_info;
2039
size = ff_aac_parse_header(gb, &hdr_info);
2041
if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2042
enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2043
memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2044
ac->m4ac.chan_config = hdr_info.chan_config;
2045
if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2047
if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2049
} else if (ac->output_configured != OC_LOCKED) {
2050
ac->output_configured = OC_NONE;
2052
if (ac->output_configured != OC_LOCKED) {
2056
ac->m4ac.sample_rate = hdr_info.sample_rate;
2057
ac->m4ac.sampling_index = hdr_info.sampling_index;
2058
ac->m4ac.object_type = hdr_info.object_type;
2059
if (!ac->avctx->sample_rate)
2060
ac->avctx->sample_rate = hdr_info.sample_rate;
2061
if (hdr_info.num_aac_frames == 1) {
2062
if (!hdr_info.crc_absent)
2065
av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2072
static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2073
int *data_size, GetBitContext *gb)
2075
AACContext *ac = avctx->priv_data;
2076
ChannelElement *che = NULL, *che_prev = NULL;
2077
enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2078
int err, elem_id, data_size_tmp;
2079
int samples = 0, multiplier;
2081
if (show_bits(gb, 12) == 0xfff) {
2082
if (parse_adts_frame_header(ac, gb) < 0) {
2083
av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2086
if (ac->m4ac.sampling_index > 12) {
2087
av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2092
ac->tags_mapped = 0;
2094
while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2095
elem_id = get_bits(gb, 4);
2097
if (elem_type < TYPE_DSE) {
2098
if (!(che=get_che(ac, elem_type, elem_id))) {
2099
av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2100
elem_type, elem_id);
2106
switch (elem_type) {
2109
err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2113
err = decode_cpe(ac, gb, che);
2117
err = decode_cce(ac, gb, che);
2121
err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2125
err = skip_data_stream_element(ac, gb);
2129
enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2130
memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2131
if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2133
if (ac->output_configured > OC_TRIAL_PCE)
2134
av_log(avctx, AV_LOG_ERROR,
2135
"Not evaluating a further program_config_element as this construct is dubious at best.\n");
2137
err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2143
elem_id += get_bits(gb, 8) - 1;
2144
if (get_bits_left(gb) < 8 * elem_id) {
2145
av_log(avctx, AV_LOG_ERROR, overread_err);
2149
elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2150
err = 0; /* FIXME */
2154
err = -1; /* should not happen, but keeps compiler happy */
2159
elem_type_prev = elem_type;
2164
if (get_bits_left(gb) < 3) {
2165
av_log(avctx, AV_LOG_ERROR, overread_err);
2170
spectral_to_sample(ac);
2172
multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2173
samples <<= multiplier;
2174
if (ac->output_configured < OC_LOCKED) {
2175
avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2176
avctx->frame_size = samples;
2179
data_size_tmp = samples * avctx->channels *
2180
(av_get_bits_per_sample_fmt(avctx->sample_fmt) / 8);
2181
if (*data_size < data_size_tmp) {
2182
av_log(avctx, AV_LOG_ERROR,
2183
"Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2184
*data_size, data_size_tmp);
2187
*data_size = data_size_tmp;
2190
if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2191
ac->fmt_conv.float_interleave(data, (const float **)ac->output_data,
2192
samples, avctx->channels);
2194
ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data,
2195
samples, avctx->channels);
2198
if (ac->output_configured)
2199
ac->output_configured = OC_LOCKED;
2204
static int aac_decode_frame(AVCodecContext *avctx, void *data,
2205
int *data_size, AVPacket *avpkt)
2207
const uint8_t *buf = avpkt->data;
2208
int buf_size = avpkt->size;
2214
init_get_bits(&gb, buf, buf_size * 8);
2216
if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2219
buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2220
for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2221
if (buf[buf_offset])
2224
return buf_size > buf_offset ? buf_consumed : buf_size;
2227
static av_cold int aac_decode_close(AVCodecContext *avctx)
2229
AACContext *ac = avctx->priv_data;
2232
for (i = 0; i < MAX_ELEM_ID; i++) {
2233
for (type = 0; type < 4; type++) {
2234
if (ac->che[type][i])
2235
ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2236
av_freep(&ac->che[type][i]);
2240
ff_mdct_end(&ac->mdct);
2241
ff_mdct_end(&ac->mdct_small);
2242
ff_mdct_end(&ac->mdct_ltp);
2247
#define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2249
struct LATMContext {
2250
AACContext aac_ctx; ///< containing AACContext
2251
int initialized; ///< initilized after a valid extradata was seen
2254
int audio_mux_version_A; ///< LATM syntax version
2255
int frame_length_type; ///< 0/1 variable/fixed frame length
2256
int frame_length; ///< frame length for fixed frame length
2259
static inline uint32_t latm_get_value(GetBitContext *b)
2261
int length = get_bits(b, 2);
2263
return get_bits_long(b, (length+1)*8);
2266
static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2269
AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2270
MPEG4AudioConfig m4ac;
2271
int config_start_bit = get_bits_count(gb);
2272
int bits_consumed, esize;
2274
if (config_start_bit % 8) {
2275
av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2276
"config not byte aligned.\n", 1);
2277
return AVERROR_INVALIDDATA;
2280
decode_audio_specific_config(NULL, avctx, &m4ac,
2281
gb->buffer + (config_start_bit / 8),
2282
get_bits_left(gb) / 8);
2284
if (bits_consumed < 0)
2285
return AVERROR_INVALIDDATA;
2287
esize = (bits_consumed+7) / 8;
2289
if (avctx->extradata_size <= esize) {
2290
av_free(avctx->extradata);
2291
avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2292
if (!avctx->extradata)
2293
return AVERROR(ENOMEM);
2296
avctx->extradata_size = esize;
2297
memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2298
memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2300
skip_bits_long(gb, bits_consumed);
2303
return bits_consumed;
2306
static int read_stream_mux_config(struct LATMContext *latmctx,
2309
int ret, audio_mux_version = get_bits(gb, 1);
2311
latmctx->audio_mux_version_A = 0;
2312
if (audio_mux_version)
2313
latmctx->audio_mux_version_A = get_bits(gb, 1);
2315
if (!latmctx->audio_mux_version_A) {
2317
if (audio_mux_version)
2318
latm_get_value(gb); // taraFullness
2320
skip_bits(gb, 1); // allStreamSameTimeFraming
2321
skip_bits(gb, 6); // numSubFrames
2323
if (get_bits(gb, 4)) { // numPrograms
2324
av_log_missing_feature(latmctx->aac_ctx.avctx,
2325
"multiple programs are not supported\n", 1);
2326
return AVERROR_PATCHWELCOME;
2329
// for each program (which there is only on in DVB)
2331
// for each layer (which there is only on in DVB)
2332
if (get_bits(gb, 3)) { // numLayer
2333
av_log_missing_feature(latmctx->aac_ctx.avctx,
2334
"multiple layers are not supported\n", 1);
2335
return AVERROR_PATCHWELCOME;
2338
// for all but first stream: use_same_config = get_bits(gb, 1);
2339
if (!audio_mux_version) {
2340
if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2343
int ascLen = latm_get_value(gb);
2344
if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2347
skip_bits_long(gb, ascLen);
2350
latmctx->frame_length_type = get_bits(gb, 3);
2351
switch (latmctx->frame_length_type) {
2353
skip_bits(gb, 8); // latmBufferFullness
2356
latmctx->frame_length = get_bits(gb, 9);
2361
skip_bits(gb, 6); // CELP frame length table index
2365
skip_bits(gb, 1); // HVXC frame length table index
2369
if (get_bits(gb, 1)) { // other data
2370
if (audio_mux_version) {
2371
latm_get_value(gb); // other_data_bits
2375
esc = get_bits(gb, 1);
2381
if (get_bits(gb, 1)) // crc present
2382
skip_bits(gb, 8); // config_crc
2388
static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2392
if (ctx->frame_length_type == 0) {
2393
int mux_slot_length = 0;
2395
tmp = get_bits(gb, 8);
2396
mux_slot_length += tmp;
2397
} while (tmp == 255);
2398
return mux_slot_length;
2399
} else if (ctx->frame_length_type == 1) {
2400
return ctx->frame_length;
2401
} else if (ctx->frame_length_type == 3 ||
2402
ctx->frame_length_type == 5 ||
2403
ctx->frame_length_type == 7) {
2404
skip_bits(gb, 2); // mux_slot_length_coded
2409
static int read_audio_mux_element(struct LATMContext *latmctx,
2413
uint8_t use_same_mux = get_bits(gb, 1);
2414
if (!use_same_mux) {
2415
if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2417
} else if (!latmctx->aac_ctx.avctx->extradata) {
2418
av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2419
"no decoder config found\n");
2420
return AVERROR(EAGAIN);
2422
if (latmctx->audio_mux_version_A == 0) {
2423
int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2424
if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2425
av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2426
return AVERROR_INVALIDDATA;
2427
} else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2428
av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2429
"frame length mismatch %d << %d\n",
2430
mux_slot_length_bytes * 8, get_bits_left(gb));
2431
return AVERROR_INVALIDDATA;
2438
static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2441
struct LATMContext *latmctx = avctx->priv_data;
2445
if (avpkt->size == 0)
2448
init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2450
// check for LOAS sync word
2451
if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2452
return AVERROR_INVALIDDATA;
2454
muxlength = get_bits(&gb, 13) + 3;
2455
// not enough data, the parser should have sorted this
2456
if (muxlength > avpkt->size)
2457
return AVERROR_INVALIDDATA;
2459
if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2462
if (!latmctx->initialized) {
2463
if (!avctx->extradata) {
2467
aac_decode_close(avctx);
2468
if ((err = aac_decode_init(avctx)) < 0)
2470
latmctx->initialized = 1;
2474
if (show_bits(&gb, 12) == 0xfff) {
2475
av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2476
"ADTS header detected, probably as result of configuration "
2478
return AVERROR_INVALIDDATA;
2481
if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2487
av_cold static int latm_decode_init(AVCodecContext *avctx)
2489
struct LATMContext *latmctx = avctx->priv_data;
2492
ret = aac_decode_init(avctx);
2494
if (avctx->extradata_size > 0) {
2495
latmctx->initialized = !ret;
2497
latmctx->initialized = 0;
2504
AVCodec ff_aac_decoder = {
2513
.long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2514
.sample_fmts = (const enum AVSampleFormat[]) {
2515
AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2517
.channel_layouts = aac_channel_layout,
2521
Note: This decoder filter is intended to decode LATM streams transferred
2522
in MPEG transport streams which only contain one program.
2523
To do a more complex LATM demuxing a separate LATM demuxer should be used.
2525
AVCodec ff_aac_latm_decoder = {
2527
.type = AVMEDIA_TYPE_AUDIO,
2528
.id = CODEC_ID_AAC_LATM,
2529
.priv_data_size = sizeof(struct LATMContext),
2530
.init = latm_decode_init,
2531
.close = aac_decode_close,
2532
.decode = latm_decode_frame,
2533
.long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2534
.sample_fmts = (const enum AVSampleFormat[]) {
2535
AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2537
.channel_layouts = aac_channel_layout,