3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * @file libavcodec/aac.c
26 * @author Oded Shimon ( ods15 ods15 dyndns org )
27 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
34 * N (code in SoC repo) gain control
36 * Y window shapes - standard
37 * N window shapes - Low Delay
38 * Y filterbank - standard
39 * N (code in SoC repo) filterbank - Scalable Sample Rate
40 * Y Temporal Noise Shaping
41 * N (code in SoC repo) Long Term Prediction
44 * Y frequency domain prediction
45 * Y Perceptual Noise Substitution
47 * N Scalable Inverse AAC Quantization
48 * N Frequency Selective Switch
50 * Y quantization & coding - AAC
51 * N quantization & coding - TwinVQ
52 * N quantization & coding - BSAC
53 * N AAC Error Resilience tools
54 * N Error Resilience payload syntax
55 * N Error Protection tool
57 * N Silence Compression
60 * N Structured Audio tools
61 * N Structured Audio Sample Bank Format
63 * N Harmonic and Individual Lines plus Noise
64 * N Text-To-Speech Interface
65 * N (in progress) Spectral Band Replication
66 * Y (not in this code) Layer-1
67 * Y (not in this code) Layer-2
68 * Y (not in this code) Layer-3
69 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
70 * N (planned) Parametric Stereo
71 * N Direct Stream Transfer
73 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
74 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
81 #include "bitstream.h"
87 #include "aacdectab.h"
88 #include "mpeg4audio.h"
89 #include "aac_parser.h"
96 union float754
{ float f
; uint32_t i
; };
98 static VLC vlc_scalefactors
;
99 static VLC vlc_spectral
[11];
102 static ChannelElement
* get_che(AACContext
*ac
, int type
, int elem_id
) {
103 static const int8_t tags_per_config
[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 };
104 if (ac
->tag_che_map
[type
][elem_id
]) {
105 return ac
->tag_che_map
[type
][elem_id
];
107 if (ac
->tags_mapped
>= tags_per_config
[ac
->m4ac
.chan_config
]) {
110 switch (ac
->m4ac
.chan_config
) {
112 if (ac
->tags_mapped
== 3 && type
== TYPE_CPE
) {
114 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][2];
117 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
118 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
119 encountered such a stream, transfer the LFE[0] element to SCE[1] */
120 if (ac
->tags_mapped
== tags_per_config
[ac
->m4ac
.chan_config
] - 1 && (type
== TYPE_LFE
|| type
== TYPE_SCE
)) {
122 return ac
->tag_che_map
[type
][elem_id
] = ac
->che
[TYPE_LFE
][0];
125 if (ac
->tags_mapped
== 2 && type
== TYPE_CPE
) {
127 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][1];
130 if (ac
->tags_mapped
== 2 && ac
->m4ac
.chan_config
== 4 && type
== TYPE_SCE
) {
132 return ac
->tag_che_map
[TYPE_SCE
][elem_id
] = ac
->che
[TYPE_SCE
][1];
136 if (ac
->tags_mapped
== (ac
->m4ac
.chan_config
!= 2) && type
== TYPE_CPE
) {
138 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][0];
139 } else if (ac
->m4ac
.chan_config
== 2) {
143 if (!ac
->tags_mapped
&& type
== TYPE_SCE
) {
145 return ac
->tag_che_map
[TYPE_SCE
][elem_id
] = ac
->che
[TYPE_SCE
][0];
153 * Configure output channel order based on the current program configuration element.
155 * @param che_pos current channel position configuration
156 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
158 * @return Returns error status. 0 - OK, !0 - error
160 static int output_configure(AACContext
*ac
, enum ChannelPosition che_pos
[4][MAX_ELEM_ID
],
161 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
], int channel_config
) {
162 AVCodecContext
*avctx
= ac
->avccontext
;
163 int i
, type
, channels
= 0;
165 if(!memcmp(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0])))
166 return 0; /* no change */
168 memcpy(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
170 /* Allocate or free elements depending on if they are in the
171 * current program configuration.
173 * Set up default 1:1 output mapping.
175 * For a 5.1 stream the output order will be:
176 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
179 for(i
= 0; i
< MAX_ELEM_ID
; i
++) {
180 for(type
= 0; type
< 4; type
++) {
181 if(che_pos
[type
][i
]) {
182 if(!ac
->che
[type
][i
] && !(ac
->che
[type
][i
] = av_mallocz(sizeof(ChannelElement
))))
183 return AVERROR(ENOMEM
);
184 if(type
!= TYPE_CCE
) {
185 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[0].ret
;
186 if(type
== TYPE_CPE
) {
187 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[1].ret
;
191 av_freep(&ac
->che
[type
][i
]);
195 if (channel_config
) {
196 memset(ac
->tag_che_map
, 0, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
199 memcpy(ac
->tag_che_map
, ac
->che
, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
200 ac
->tags_mapped
= 4*MAX_ELEM_ID
;
203 avctx
->channels
= channels
;
209 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
211 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
212 * @param sce_map mono (Single Channel Element) map
213 * @param type speaker type/position for these channels
215 static void decode_channel_map(enum ChannelPosition
*cpe_map
,
216 enum ChannelPosition
*sce_map
, enum ChannelPosition type
, GetBitContext
* gb
, int n
) {
218 enum ChannelPosition
*map
= cpe_map
&& get_bits1(gb
) ? cpe_map
: sce_map
; // stereo or mono map
219 map
[get_bits(gb
, 4)] = type
;
224 * Decode program configuration element; reference: table 4.2.
226 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
228 * @return Returns error status. 0 - OK, !0 - error
230 static int decode_pce(AACContext
* ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
231 GetBitContext
* gb
) {
232 int num_front
, num_side
, num_back
, num_lfe
, num_assoc_data
, num_cc
, sampling_index
;
234 skip_bits(gb
, 2); // object_type
236 sampling_index
= get_bits(gb
, 4);
237 if(sampling_index
> 12) {
238 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
241 ac
->m4ac
.sampling_index
= sampling_index
;
242 ac
->m4ac
.sample_rate
= ff_mpeg4audio_sample_rates
[ac
->m4ac
.sampling_index
];
243 num_front
= get_bits(gb
, 4);
244 num_side
= get_bits(gb
, 4);
245 num_back
= get_bits(gb
, 4);
246 num_lfe
= get_bits(gb
, 2);
247 num_assoc_data
= get_bits(gb
, 3);
248 num_cc
= get_bits(gb
, 4);
251 skip_bits(gb
, 4); // mono_mixdown_tag
253 skip_bits(gb
, 4); // stereo_mixdown_tag
256 skip_bits(gb
, 3); // mixdown_coeff_index and pseudo_surround
258 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_FRONT
, gb
, num_front
);
259 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_SIDE
, gb
, num_side
);
260 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_BACK
, gb
, num_back
);
261 decode_channel_map(NULL
, new_che_pos
[TYPE_LFE
], AAC_CHANNEL_LFE
, gb
, num_lfe
);
263 skip_bits_long(gb
, 4 * num_assoc_data
);
265 decode_channel_map(new_che_pos
[TYPE_CCE
], new_che_pos
[TYPE_CCE
], AAC_CHANNEL_CC
, gb
, num_cc
);
269 /* comment field, first byte is length */
270 skip_bits_long(gb
, 8 * get_bits(gb
, 8));
275 * Set up channel positions based on a default channel configuration
276 * as specified in table 1.17.
278 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
280 * @return Returns error status. 0 - OK, !0 - error
282 static int set_default_channel_config(AACContext
*ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
285 if(channel_config
< 1 || channel_config
> 7) {
286 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid default channel configuration (%d)\n",
291 /* default channel configurations:
293 * 1ch : front center (mono)
294 * 2ch : L + R (stereo)
295 * 3ch : front center + L + R
296 * 4ch : front center + L + R + back center
297 * 5ch : front center + L + R + back stereo
298 * 6ch : front center + L + R + back stereo + LFE
299 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
302 if(channel_config
!= 2)
303 new_che_pos
[TYPE_SCE
][0] = AAC_CHANNEL_FRONT
; // front center (or mono)
304 if(channel_config
> 1)
305 new_che_pos
[TYPE_CPE
][0] = AAC_CHANNEL_FRONT
; // L + R (or stereo)
306 if(channel_config
== 4)
307 new_che_pos
[TYPE_SCE
][1] = AAC_CHANNEL_BACK
; // back center
308 if(channel_config
> 4)
309 new_che_pos
[TYPE_CPE
][(channel_config
== 7) + 1]
310 = AAC_CHANNEL_BACK
; // back stereo
311 if(channel_config
> 5)
312 new_che_pos
[TYPE_LFE
][0] = AAC_CHANNEL_LFE
; // LFE
313 if(channel_config
== 7)
314 new_che_pos
[TYPE_CPE
][1] = AAC_CHANNEL_FRONT
; // outer front left + outer front right
320 * Decode GA "General Audio" specific configuration; reference: table 4.1.
322 * @return Returns error status. 0 - OK, !0 - error
324 static int decode_ga_specific_config(AACContext
* ac
, GetBitContext
* gb
, int channel_config
) {
325 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
326 int extension_flag
, ret
;
328 if(get_bits1(gb
)) { // frameLengthFlag
329 ff_log_missing_feature(ac
->avccontext
, "960/120 MDCT window is", 1);
333 if (get_bits1(gb
)) // dependsOnCoreCoder
334 skip_bits(gb
, 14); // coreCoderDelay
335 extension_flag
= get_bits1(gb
);
337 if(ac
->m4ac
.object_type
== AOT_AAC_SCALABLE
||
338 ac
->m4ac
.object_type
== AOT_ER_AAC_SCALABLE
)
339 skip_bits(gb
, 3); // layerNr
341 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
342 if (channel_config
== 0) {
343 skip_bits(gb
, 4); // element_instance_tag
344 if((ret
= decode_pce(ac
, new_che_pos
, gb
)))
347 if((ret
= set_default_channel_config(ac
, new_che_pos
, channel_config
)))
350 if((ret
= output_configure(ac
, ac
->che_pos
, new_che_pos
, channel_config
)))
353 if (extension_flag
) {
354 switch (ac
->m4ac
.object_type
) {
356 skip_bits(gb
, 5); // numOfSubFrame
357 skip_bits(gb
, 11); // layer_length
361 case AOT_ER_AAC_SCALABLE
:
363 skip_bits(gb
, 3); /* aacSectionDataResilienceFlag
364 * aacScalefactorDataResilienceFlag
365 * aacSpectralDataResilienceFlag
369 skip_bits1(gb
); // extensionFlag3 (TBD in version 3)
375 * Decode audio specific configuration; reference: table 1.13.
377 * @param data pointer to AVCodecContext extradata
378 * @param data_size size of AVCCodecContext extradata
380 * @return Returns error status. 0 - OK, !0 - error
382 static int decode_audio_specific_config(AACContext
* ac
, void *data
, int data_size
) {
386 init_get_bits(&gb
, data
, data_size
* 8);
388 if((i
= ff_mpeg4audio_get_config(&ac
->m4ac
, data
, data_size
)) < 0)
390 if(ac
->m4ac
.sampling_index
> 12) {
391 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
395 skip_bits_long(&gb
, i
);
397 switch (ac
->m4ac
.object_type
) {
400 if (decode_ga_specific_config(ac
, &gb
, ac
->m4ac
.chan_config
))
404 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Audio object type %s%d is not supported.\n",
405 ac
->m4ac
.sbr
== 1? "SBR+" : "", ac
->m4ac
.object_type
);
412 * linear congruential pseudorandom number generator
414 * @param previous_val pointer to the current state of the generator
416 * @return Returns a 32-bit pseudorandom integer
418 static av_always_inline
int lcg_random(int previous_val
) {
419 return previous_val
* 1664525 + 1013904223;
422 static void reset_predict_state(PredictorState
* ps
) {
431 static void reset_all_predictors(PredictorState
* ps
) {
433 for (i
= 0; i
< MAX_PREDICTORS
; i
++)
434 reset_predict_state(&ps
[i
]);
437 static void reset_predictor_group(PredictorState
* ps
, int group_num
) {
439 for (i
= group_num
-1; i
< MAX_PREDICTORS
; i
+=30)
440 reset_predict_state(&ps
[i
]);
443 static av_cold
int aac_decode_init(AVCodecContext
* avccontext
) {
444 AACContext
* ac
= avccontext
->priv_data
;
447 ac
->avccontext
= avccontext
;
449 if (avccontext
->extradata_size
> 0) {
450 if(decode_audio_specific_config(ac
, avccontext
->extradata
, avccontext
->extradata_size
))
452 avccontext
->sample_rate
= ac
->m4ac
.sample_rate
;
453 } else if (avccontext
->channels
> 0) {
454 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
455 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
456 if(set_default_channel_config(ac
, new_che_pos
, avccontext
->channels
- (avccontext
->channels
== 8)))
458 if(output_configure(ac
, ac
->che_pos
, new_che_pos
, 1))
460 ac
->m4ac
.sample_rate
= avccontext
->sample_rate
;
462 ff_log_missing_feature(ac
->avccontext
, "Implicit channel configuration is", 0);
466 avccontext
->sample_fmt
= SAMPLE_FMT_S16
;
467 avccontext
->frame_size
= 1024;
469 AAC_INIT_VLC_STATIC( 0, 144);
470 AAC_INIT_VLC_STATIC( 1, 114);
471 AAC_INIT_VLC_STATIC( 2, 188);
472 AAC_INIT_VLC_STATIC( 3, 180);
473 AAC_INIT_VLC_STATIC( 4, 172);
474 AAC_INIT_VLC_STATIC( 5, 140);
475 AAC_INIT_VLC_STATIC( 6, 168);
476 AAC_INIT_VLC_STATIC( 7, 114);
477 AAC_INIT_VLC_STATIC( 8, 262);
478 AAC_INIT_VLC_STATIC( 9, 248);
479 AAC_INIT_VLC_STATIC(10, 384);
481 dsputil_init(&ac
->dsp
, avccontext
);
483 ac
->random_state
= 0x1f2e3d4c;
485 // -1024 - Compensate wrong IMDCT method.
486 // 32768 - Required to scale values to the correct range for the bias method
487 // for float to int16 conversion.
489 if(ac
->dsp
.float_to_int16
== ff_float_to_int16_c
) {
490 ac
->add_bias
= 385.0f
;
491 ac
->sf_scale
= 1. / (-1024. * 32768.);
495 ac
->sf_scale
= 1. / -1024.;
499 #if !CONFIG_HARDCODED_TABLES
500 for (i
= 0; i
< 428; i
++)
501 ff_aac_pow2sf_tab
[i
] = pow(2, (i
- 200)/4.);
502 #endif /* CONFIG_HARDCODED_TABLES */
504 INIT_VLC_STATIC(&vlc_scalefactors
,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code
),
505 ff_aac_scalefactor_bits
, sizeof(ff_aac_scalefactor_bits
[0]), sizeof(ff_aac_scalefactor_bits
[0]),
506 ff_aac_scalefactor_code
, sizeof(ff_aac_scalefactor_code
[0]), sizeof(ff_aac_scalefactor_code
[0]),
509 ff_mdct_init(&ac
->mdct
, 11, 1);
510 ff_mdct_init(&ac
->mdct_small
, 8, 1);
511 // window initialization
512 ff_kbd_window_init(ff_aac_kbd_long_1024
, 4.0, 1024);
513 ff_kbd_window_init(ff_aac_kbd_short_128
, 6.0, 128);
514 ff_sine_window_init(ff_sine_1024
, 1024);
515 ff_sine_window_init(ff_sine_128
, 128);
521 * Skip data_stream_element; reference: table 4.10.
523 static void skip_data_stream_element(GetBitContext
* gb
) {
524 int byte_align
= get_bits1(gb
);
525 int count
= get_bits(gb
, 8);
527 count
+= get_bits(gb
, 8);
530 skip_bits_long(gb
, 8 * count
);
533 static int decode_prediction(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
) {
536 ics
->predictor_reset_group
= get_bits(gb
, 5);
537 if (ics
->predictor_reset_group
== 0 || ics
->predictor_reset_group
> 30) {
538 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Invalid Predictor Reset Group.\n");
542 for (sfb
= 0; sfb
< FFMIN(ics
->max_sfb
, ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]); sfb
++) {
543 ics
->prediction_used
[sfb
] = get_bits1(gb
);
549 * Decode Individual Channel Stream info; reference: table 4.6.
551 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
553 static int decode_ics_info(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
, int common_window
) {
555 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Reserved bit set.\n");
556 memset(ics
, 0, sizeof(IndividualChannelStream
));
559 ics
->window_sequence
[1] = ics
->window_sequence
[0];
560 ics
->window_sequence
[0] = get_bits(gb
, 2);
561 ics
->use_kb_window
[1] = ics
->use_kb_window
[0];
562 ics
->use_kb_window
[0] = get_bits1(gb
);
563 ics
->num_window_groups
= 1;
564 ics
->group_len
[0] = 1;
565 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
567 ics
->max_sfb
= get_bits(gb
, 4);
568 for (i
= 0; i
< 7; i
++) {
570 ics
->group_len
[ics
->num_window_groups
-1]++;
572 ics
->num_window_groups
++;
573 ics
->group_len
[ics
->num_window_groups
-1] = 1;
576 ics
->num_windows
= 8;
577 ics
->swb_offset
= swb_offset_128
[ac
->m4ac
.sampling_index
];
578 ics
->num_swb
= ff_aac_num_swb_128
[ac
->m4ac
.sampling_index
];
579 ics
->tns_max_bands
= tns_max_bands_128
[ac
->m4ac
.sampling_index
];
580 ics
->predictor_present
= 0;
582 ics
->max_sfb
= get_bits(gb
, 6);
583 ics
->num_windows
= 1;
584 ics
->swb_offset
= swb_offset_1024
[ac
->m4ac
.sampling_index
];
585 ics
->num_swb
= ff_aac_num_swb_1024
[ac
->m4ac
.sampling_index
];
586 ics
->tns_max_bands
= tns_max_bands_1024
[ac
->m4ac
.sampling_index
];
587 ics
->predictor_present
= get_bits1(gb
);
588 ics
->predictor_reset_group
= 0;
589 if (ics
->predictor_present
) {
590 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
591 if (decode_prediction(ac
, ics
, gb
)) {
592 memset(ics
, 0, sizeof(IndividualChannelStream
));
595 } else if (ac
->m4ac
.object_type
== AOT_AAC_LC
) {
596 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Prediction is not allowed in AAC-LC.\n");
597 memset(ics
, 0, sizeof(IndividualChannelStream
));
600 ff_log_missing_feature(ac
->avccontext
, "Predictor bit set but LTP is", 1);
601 memset(ics
, 0, sizeof(IndividualChannelStream
));
607 if(ics
->max_sfb
> ics
->num_swb
) {
608 av_log(ac
->avccontext
, AV_LOG_ERROR
,
609 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
610 ics
->max_sfb
, ics
->num_swb
);
611 memset(ics
, 0, sizeof(IndividualChannelStream
));
619 * Decode band types (section_data payload); reference: table 4.46.
621 * @param band_type array of the used band type
622 * @param band_type_run_end array of the last scalefactor band of a band type run
624 * @return Returns error status. 0 - OK, !0 - error
626 static int decode_band_types(AACContext
* ac
, enum BandType band_type
[120],
627 int band_type_run_end
[120], GetBitContext
* gb
, IndividualChannelStream
* ics
) {
629 const int bits
= (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) ? 3 : 5;
630 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
632 while (k
< ics
->max_sfb
) {
633 uint8_t sect_len
= k
;
635 int sect_band_type
= get_bits(gb
, 4);
636 if (sect_band_type
== 12) {
637 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid band type\n");
640 while ((sect_len_incr
= get_bits(gb
, bits
)) == (1 << bits
)-1)
641 sect_len
+= sect_len_incr
;
642 sect_len
+= sect_len_incr
;
643 if (sect_len
> ics
->max_sfb
) {
644 av_log(ac
->avccontext
, AV_LOG_ERROR
,
645 "Number of bands (%d) exceeds limit (%d).\n",
646 sect_len
, ics
->max_sfb
);
649 for (; k
< sect_len
; k
++) {
650 band_type
[idx
] = sect_band_type
;
651 band_type_run_end
[idx
++] = sect_len
;
659 * Decode scalefactors; reference: table 4.47.
661 * @param global_gain first scalefactor value as scalefactors are differentially coded
662 * @param band_type array of the used band type
663 * @param band_type_run_end array of the last scalefactor band of a band type run
664 * @param sf array of scalefactors or intensity stereo positions
666 * @return Returns error status. 0 - OK, !0 - error
668 static int decode_scalefactors(AACContext
* ac
, float sf
[120], GetBitContext
* gb
,
669 unsigned int global_gain
, IndividualChannelStream
* ics
,
670 enum BandType band_type
[120], int band_type_run_end
[120]) {
671 const int sf_offset
= ac
->sf_offset
+ (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
? 12 : 0);
673 int offset
[3] = { global_gain
, global_gain
- 90, 100 };
675 static const char *sf_str
[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
676 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
677 for (i
= 0; i
< ics
->max_sfb
;) {
678 int run_end
= band_type_run_end
[idx
];
679 if (band_type
[idx
] == ZERO_BT
) {
680 for(; i
< run_end
; i
++, idx
++)
682 }else if((band_type
[idx
] == INTENSITY_BT
) || (band_type
[idx
] == INTENSITY_BT2
)) {
683 for(; i
< run_end
; i
++, idx
++) {
684 offset
[2] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
685 if(offset
[2] > 255U) {
686 av_log(ac
->avccontext
, AV_LOG_ERROR
,
687 "%s (%d) out of range.\n", sf_str
[2], offset
[2]);
690 sf
[idx
] = ff_aac_pow2sf_tab
[-offset
[2] + 300];
692 }else if(band_type
[idx
] == NOISE_BT
) {
693 for(; i
< run_end
; i
++, idx
++) {
695 offset
[1] += get_bits(gb
, 9) - 256;
697 offset
[1] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
698 if(offset
[1] > 255U) {
699 av_log(ac
->avccontext
, AV_LOG_ERROR
,
700 "%s (%d) out of range.\n", sf_str
[1], offset
[1]);
703 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[1] + sf_offset
+ 100];
706 for(; i
< run_end
; i
++, idx
++) {
707 offset
[0] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
708 if(offset
[0] > 255U) {
709 av_log(ac
->avccontext
, AV_LOG_ERROR
,
710 "%s (%d) out of range.\n", sf_str
[0], offset
[0]);
713 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[0] + sf_offset
];
722 * Decode pulse data; reference: table 4.7.
724 static int decode_pulses(Pulse
* pulse
, GetBitContext
* gb
, const uint16_t * swb_offset
, int num_swb
) {
726 pulse
->num_pulse
= get_bits(gb
, 2) + 1;
727 pulse_swb
= get_bits(gb
, 6);
728 if (pulse_swb
>= num_swb
)
730 pulse
->pos
[0] = swb_offset
[pulse_swb
];
731 pulse
->pos
[0] += get_bits(gb
, 5);
732 if (pulse
->pos
[0] > 1023)
734 pulse
->amp
[0] = get_bits(gb
, 4);
735 for (i
= 1; i
< pulse
->num_pulse
; i
++) {
736 pulse
->pos
[i
] = get_bits(gb
, 5) + pulse
->pos
[i
-1];
737 if (pulse
->pos
[i
] > 1023)
739 pulse
->amp
[i
] = get_bits(gb
, 4);
745 * Decode Temporal Noise Shaping data; reference: table 4.48.
747 * @return Returns error status. 0 - OK, !0 - error
749 static int decode_tns(AACContext
* ac
, TemporalNoiseShaping
* tns
,
750 GetBitContext
* gb
, const IndividualChannelStream
* ics
) {
751 int w
, filt
, i
, coef_len
, coef_res
, coef_compress
;
752 const int is8
= ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
;
753 const int tns_max_order
= is8
? 7 : ac
->m4ac
.object_type
== AOT_AAC_MAIN
? 20 : 12;
754 for (w
= 0; w
< ics
->num_windows
; w
++) {
755 if ((tns
->n_filt
[w
] = get_bits(gb
, 2 - is8
))) {
756 coef_res
= get_bits1(gb
);
758 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
760 tns
->length
[w
][filt
] = get_bits(gb
, 6 - 2*is8
);
762 if ((tns
->order
[w
][filt
] = get_bits(gb
, 5 - 2*is8
)) > tns_max_order
) {
763 av_log(ac
->avccontext
, AV_LOG_ERROR
, "TNS filter order %d is greater than maximum %d.",
764 tns
->order
[w
][filt
], tns_max_order
);
765 tns
->order
[w
][filt
] = 0;
768 if (tns
->order
[w
][filt
]) {
769 tns
->direction
[w
][filt
] = get_bits1(gb
);
770 coef_compress
= get_bits1(gb
);
771 coef_len
= coef_res
+ 3 - coef_compress
;
772 tmp2_idx
= 2*coef_compress
+ coef_res
;
774 for (i
= 0; i
< tns
->order
[w
][filt
]; i
++)
775 tns
->coef
[w
][filt
][i
] = tns_tmp2_map
[tmp2_idx
][get_bits(gb
, coef_len
)];
784 * Decode Mid/Side data; reference: table 4.54.
786 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
787 * [1] mask is decoded from bitstream; [2] mask is all 1s;
788 * [3] reserved for scalable AAC
790 static void decode_mid_side_stereo(ChannelElement
* cpe
, GetBitContext
* gb
,
793 if (ms_present
== 1) {
794 for (idx
= 0; idx
< cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
; idx
++)
795 cpe
->ms_mask
[idx
] = get_bits1(gb
);
796 } else if (ms_present
== 2) {
797 memset(cpe
->ms_mask
, 1, cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
* sizeof(cpe
->ms_mask
[0]));
802 * Decode spectral data; reference: table 4.50.
803 * Dequantize and scale spectral data; reference: 4.6.3.3.
805 * @param coef array of dequantized, scaled spectral data
806 * @param sf array of scalefactors or intensity stereo positions
807 * @param pulse_present set if pulses are present
808 * @param pulse pointer to pulse data struct
809 * @param band_type array of the used band type
811 * @return Returns error status. 0 - OK, !0 - error
813 static int decode_spectrum_and_dequant(AACContext
* ac
, float coef
[1024], GetBitContext
* gb
, float sf
[120],
814 int pulse_present
, const Pulse
* pulse
, const IndividualChannelStream
* ics
, enum BandType band_type
[120]) {
815 int i
, k
, g
, idx
= 0;
816 const int c
= 1024/ics
->num_windows
;
817 const uint16_t * offsets
= ics
->swb_offset
;
818 float *coef_base
= coef
;
819 static const float sign_lookup
[] = { 1.0f
, -1.0f
};
821 for (g
= 0; g
< ics
->num_windows
; g
++)
822 memset(coef
+ g
* 128 + offsets
[ics
->max_sfb
], 0, sizeof(float)*(c
- offsets
[ics
->max_sfb
]));
824 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
825 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
826 const int cur_band_type
= band_type
[idx
];
827 const int dim
= cur_band_type
>= FIRST_PAIR_BT
? 2 : 4;
828 const int is_cb_unsigned
= IS_CODEBOOK_UNSIGNED(cur_band_type
);
830 if (cur_band_type
== ZERO_BT
|| cur_band_type
== INTENSITY_BT2
|| cur_band_type
== INTENSITY_BT
) {
831 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
832 memset(coef
+ group
* 128 + offsets
[i
], 0, (offsets
[i
+1] - offsets
[i
])*sizeof(float));
834 }else if (cur_band_type
== NOISE_BT
) {
835 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
837 float band_energy
= 0;
838 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
839 ac
->random_state
= lcg_random(ac
->random_state
);
840 coef
[group
*128+k
] = ac
->random_state
;
841 band_energy
+= coef
[group
*128+k
]*coef
[group
*128+k
];
843 scale
= sf
[idx
] / sqrtf(band_energy
);
844 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
845 coef
[group
*128+k
] *= scale
;
849 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
850 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
+= dim
) {
851 const int index
= get_vlc2(gb
, vlc_spectral
[cur_band_type
- 1].table
, 6, 3);
852 const int coef_tmp_idx
= (group
<< 7) + k
;
855 if(index
>= ff_aac_spectral_sizes
[cur_band_type
- 1]) {
856 av_log(ac
->avccontext
, AV_LOG_ERROR
,
857 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
858 cur_band_type
- 1, index
, ff_aac_spectral_sizes
[cur_band_type
- 1]);
861 vq_ptr
= &ff_aac_codebook_vectors
[cur_band_type
- 1][index
* dim
];
862 if (is_cb_unsigned
) {
863 if (vq_ptr
[0]) coef
[coef_tmp_idx
] = sign_lookup
[get_bits1(gb
)];
864 if (vq_ptr
[1]) coef
[coef_tmp_idx
+ 1] = sign_lookup
[get_bits1(gb
)];
866 if (vq_ptr
[2]) coef
[coef_tmp_idx
+ 2] = sign_lookup
[get_bits1(gb
)];
867 if (vq_ptr
[3]) coef
[coef_tmp_idx
+ 3] = sign_lookup
[get_bits1(gb
)];
869 if (cur_band_type
== ESC_BT
) {
870 for (j
= 0; j
< 2; j
++) {
871 if (vq_ptr
[j
] == 64.0f
) {
873 /* The total length of escape_sequence must be < 22 bits according
874 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
875 while (get_bits1(gb
) && n
< 15) n
++;
877 av_log(ac
->avccontext
, AV_LOG_ERROR
, "error in spectral data, ESC overflow\n");
880 n
= (1<<n
) + get_bits(gb
, n
);
881 coef
[coef_tmp_idx
+ j
] *= cbrtf(n
) * n
;
883 coef
[coef_tmp_idx
+ j
] *= vq_ptr
[j
];
887 coef
[coef_tmp_idx
] *= vq_ptr
[0];
888 coef
[coef_tmp_idx
+ 1] *= vq_ptr
[1];
890 coef
[coef_tmp_idx
+ 2] *= vq_ptr
[2];
891 coef
[coef_tmp_idx
+ 3] *= vq_ptr
[3];
895 coef
[coef_tmp_idx
] = vq_ptr
[0];
896 coef
[coef_tmp_idx
+ 1] = vq_ptr
[1];
898 coef
[coef_tmp_idx
+ 2] = vq_ptr
[2];
899 coef
[coef_tmp_idx
+ 3] = vq_ptr
[3];
902 coef
[coef_tmp_idx
] *= sf
[idx
];
903 coef
[coef_tmp_idx
+ 1] *= sf
[idx
];
905 coef
[coef_tmp_idx
+ 2] *= sf
[idx
];
906 coef
[coef_tmp_idx
+ 3] *= sf
[idx
];
912 coef
+= ics
->group_len
[g
]<<7;
917 for(i
= 0; i
< pulse
->num_pulse
; i
++){
918 float co
= coef_base
[ pulse
->pos
[i
] ];
919 while(offsets
[idx
+ 1] <= pulse
->pos
[i
])
921 if (band_type
[idx
] != NOISE_BT
&& sf
[idx
]) {
922 float ico
= -pulse
->amp
[i
];
925 ico
= co
/ sqrtf(sqrtf(fabsf(co
))) + (co
> 0 ? -ico
: ico
);
927 coef_base
[ pulse
->pos
[i
] ] = cbrtf(fabsf(ico
)) * ico
* sf
[idx
];
934 static av_always_inline
float flt16_round(float pf
) {
937 tmp
.i
= (tmp
.i
+ 0x00008000U
) & 0xFFFF0000U
;
941 static av_always_inline
float flt16_even(float pf
) {
944 tmp
.i
= (tmp
.i
+ 0x00007FFFU
+ (tmp
.i
& 0x00010000U
>>16)) & 0xFFFF0000U
;
948 static av_always_inline
float flt16_trunc(float pf
) {
951 pun
.i
&= 0xFFFF0000U
;
955 static void predict(AACContext
* ac
, PredictorState
* ps
, float* coef
, int output_enable
) {
956 const float a
= 0.953125; // 61.0/64
957 const float alpha
= 0.90625; // 29.0/32
962 k1
= ps
->var0
> 1 ? ps
->cor0
* flt16_even(a
/ ps
->var0
) : 0;
963 k2
= ps
->var1
> 1 ? ps
->cor1
* flt16_even(a
/ ps
->var1
) : 0;
965 pv
= flt16_round(k1
* ps
->r0
+ k2
* ps
->r1
);
967 *coef
+= pv
* ac
->sf_scale
;
969 e0
= *coef
/ ac
->sf_scale
;
970 e1
= e0
- k1
* ps
->r0
;
972 ps
->cor1
= flt16_trunc(alpha
* ps
->cor1
+ ps
->r1
* e1
);
973 ps
->var1
= flt16_trunc(alpha
* ps
->var1
+ 0.5 * (ps
->r1
* ps
->r1
+ e1
* e1
));
974 ps
->cor0
= flt16_trunc(alpha
* ps
->cor0
+ ps
->r0
* e0
);
975 ps
->var0
= flt16_trunc(alpha
* ps
->var0
+ 0.5 * (ps
->r0
* ps
->r0
+ e0
* e0
));
977 ps
->r1
= flt16_trunc(a
* (ps
->r0
- k1
* e0
));
978 ps
->r0
= flt16_trunc(a
* e0
);
982 * Apply AAC-Main style frequency domain prediction.
984 static void apply_prediction(AACContext
* ac
, SingleChannelElement
* sce
) {
987 if (!sce
->ics
.predictor_initialized
) {
988 reset_all_predictors(sce
->predictor_state
);
989 sce
->ics
.predictor_initialized
= 1;
992 if (sce
->ics
.window_sequence
[0] != EIGHT_SHORT_SEQUENCE
) {
993 for (sfb
= 0; sfb
< ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]; sfb
++) {
994 for (k
= sce
->ics
.swb_offset
[sfb
]; k
< sce
->ics
.swb_offset
[sfb
+ 1]; k
++) {
995 predict(ac
, &sce
->predictor_state
[k
], &sce
->coeffs
[k
],
996 sce
->ics
.predictor_present
&& sce
->ics
.prediction_used
[sfb
]);
999 if (sce
->ics
.predictor_reset_group
)
1000 reset_predictor_group(sce
->predictor_state
, sce
->ics
.predictor_reset_group
);
1002 reset_all_predictors(sce
->predictor_state
);
1006 * Decode an individual_channel_stream payload; reference: table 4.44.
1008 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1009 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1011 * @return Returns error status. 0 - OK, !0 - error
1013 static int decode_ics(AACContext
* ac
, SingleChannelElement
* sce
, GetBitContext
* gb
, int common_window
, int scale_flag
) {
1015 TemporalNoiseShaping
* tns
= &sce
->tns
;
1016 IndividualChannelStream
* ics
= &sce
->ics
;
1017 float * out
= sce
->coeffs
;
1018 int global_gain
, pulse_present
= 0;
1020 /* This assignment is to silence a GCC warning about the variable being used
1021 * uninitialized when in fact it always is.
1023 pulse
.num_pulse
= 0;
1025 global_gain
= get_bits(gb
, 8);
1027 if (!common_window
&& !scale_flag
) {
1028 if (decode_ics_info(ac
, ics
, gb
, 0) < 0)
1032 if (decode_band_types(ac
, sce
->band_type
, sce
->band_type_run_end
, gb
, ics
) < 0)
1034 if (decode_scalefactors(ac
, sce
->sf
, gb
, global_gain
, ics
, sce
->band_type
, sce
->band_type_run_end
) < 0)
1039 if ((pulse_present
= get_bits1(gb
))) {
1040 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1041 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse tool not allowed in eight short sequence.\n");
1044 if (decode_pulses(&pulse
, gb
, ics
->swb_offset
, ics
->num_swb
)) {
1045 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse data corrupt or invalid.\n");
1049 if ((tns
->present
= get_bits1(gb
)) && decode_tns(ac
, tns
, gb
, ics
))
1051 if (get_bits1(gb
)) {
1052 ff_log_missing_feature(ac
->avccontext
, "SSR", 1);
1057 if (decode_spectrum_and_dequant(ac
, out
, gb
, sce
->sf
, pulse_present
, &pulse
, ics
, sce
->band_type
) < 0)
1060 if(ac
->m4ac
.object_type
== AOT_AAC_MAIN
&& !common_window
)
1061 apply_prediction(ac
, sce
);
1067 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1069 static void apply_mid_side_stereo(ChannelElement
* cpe
) {
1070 const IndividualChannelStream
* ics
= &cpe
->ch
[0].ics
;
1071 float *ch0
= cpe
->ch
[0].coeffs
;
1072 float *ch1
= cpe
->ch
[1].coeffs
;
1073 int g
, i
, k
, group
, idx
= 0;
1074 const uint16_t * offsets
= ics
->swb_offset
;
1075 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1076 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1077 if (cpe
->ms_mask
[idx
] &&
1078 cpe
->ch
[0].band_type
[idx
] < NOISE_BT
&& cpe
->ch
[1].band_type
[idx
] < NOISE_BT
) {
1079 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1080 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
1081 float tmp
= ch0
[group
*128 + k
] - ch1
[group
*128 + k
];
1082 ch0
[group
*128 + k
] += ch1
[group
*128 + k
];
1083 ch1
[group
*128 + k
] = tmp
;
1088 ch0
+= ics
->group_len
[g
]*128;
1089 ch1
+= ics
->group_len
[g
]*128;
1094 * intensity stereo decoding; reference: 4.6.8.2.3
1096 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1097 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1098 * [3] reserved for scalable AAC
1100 static void apply_intensity_stereo(ChannelElement
* cpe
, int ms_present
) {
1101 const IndividualChannelStream
* ics
= &cpe
->ch
[1].ics
;
1102 SingleChannelElement
* sce1
= &cpe
->ch
[1];
1103 float *coef0
= cpe
->ch
[0].coeffs
, *coef1
= cpe
->ch
[1].coeffs
;
1104 const uint16_t * offsets
= ics
->swb_offset
;
1105 int g
, group
, i
, k
, idx
= 0;
1108 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1109 for (i
= 0; i
< ics
->max_sfb
;) {
1110 if (sce1
->band_type
[idx
] == INTENSITY_BT
|| sce1
->band_type
[idx
] == INTENSITY_BT2
) {
1111 const int bt_run_end
= sce1
->band_type_run_end
[idx
];
1112 for (; i
< bt_run_end
; i
++, idx
++) {
1113 c
= -1 + 2 * (sce1
->band_type
[idx
] - 14);
1115 c
*= 1 - 2 * cpe
->ms_mask
[idx
];
1116 scale
= c
* sce1
->sf
[idx
];
1117 for (group
= 0; group
< ics
->group_len
[g
]; group
++)
1118 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++)
1119 coef1
[group
*128 + k
] = scale
* coef0
[group
*128 + k
];
1122 int bt_run_end
= sce1
->band_type_run_end
[idx
];
1123 idx
+= bt_run_end
- i
;
1127 coef0
+= ics
->group_len
[g
]*128;
1128 coef1
+= ics
->group_len
[g
]*128;
1133 * Decode a channel_pair_element; reference: table 4.4.
1135 * @param elem_id Identifies the instance of a syntax element.
1137 * @return Returns error status. 0 - OK, !0 - error
1139 static int decode_cpe(AACContext
* ac
, GetBitContext
* gb
, ChannelElement
* cpe
) {
1140 int i
, ret
, common_window
, ms_present
= 0;
1142 common_window
= get_bits1(gb
);
1143 if (common_window
) {
1144 if (decode_ics_info(ac
, &cpe
->ch
[0].ics
, gb
, 1))
1146 i
= cpe
->ch
[1].ics
.use_kb_window
[0];
1147 cpe
->ch
[1].ics
= cpe
->ch
[0].ics
;
1148 cpe
->ch
[1].ics
.use_kb_window
[1] = i
;
1149 ms_present
= get_bits(gb
, 2);
1150 if(ms_present
== 3) {
1151 av_log(ac
->avccontext
, AV_LOG_ERROR
, "ms_present = 3 is reserved.\n");
1153 } else if(ms_present
)
1154 decode_mid_side_stereo(cpe
, gb
, ms_present
);
1156 if ((ret
= decode_ics(ac
, &cpe
->ch
[0], gb
, common_window
, 0)))
1158 if ((ret
= decode_ics(ac
, &cpe
->ch
[1], gb
, common_window
, 0)))
1161 if (common_window
) {
1163 apply_mid_side_stereo(cpe
);
1164 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
1165 apply_prediction(ac
, &cpe
->ch
[0]);
1166 apply_prediction(ac
, &cpe
->ch
[1]);
1170 apply_intensity_stereo(cpe
, ms_present
);
1175 * Decode coupling_channel_element; reference: table 4.8.
1177 * @param elem_id Identifies the instance of a syntax element.
1179 * @return Returns error status. 0 - OK, !0 - error
1181 static int decode_cce(AACContext
* ac
, GetBitContext
* gb
, ChannelElement
* che
) {
1186 SingleChannelElement
* sce
= &che
->ch
[0];
1187 ChannelCoupling
* coup
= &che
->coup
;
1189 coup
->coupling_point
= 2*get_bits1(gb
);
1190 coup
->num_coupled
= get_bits(gb
, 3);
1191 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1193 coup
->type
[c
] = get_bits1(gb
) ? TYPE_CPE
: TYPE_SCE
;
1194 coup
->id_select
[c
] = get_bits(gb
, 4);
1195 if (coup
->type
[c
] == TYPE_CPE
) {
1196 coup
->ch_select
[c
] = get_bits(gb
, 2);
1197 if (coup
->ch_select
[c
] == 3)
1200 coup
->ch_select
[c
] = 2;
1202 coup
->coupling_point
+= get_bits1(gb
) || (coup
->coupling_point
>>1);
1204 sign
= get_bits(gb
, 1);
1205 scale
= pow(2., pow(2., (int)get_bits(gb
, 2) - 3));
1207 if ((ret
= decode_ics(ac
, sce
, gb
, 0, 0)))
1210 for (c
= 0; c
< num_gain
; c
++) {
1214 float gain_cache
= 1.;
1216 cge
= coup
->coupling_point
== AFTER_IMDCT
? 1 : get_bits1(gb
);
1217 gain
= cge
? get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60: 0;
1218 gain_cache
= pow(scale
, -gain
);
1220 if (coup
->coupling_point
== AFTER_IMDCT
) {
1221 coup
->gain
[c
][0] = gain_cache
;
1223 for (g
= 0; g
< sce
->ics
.num_window_groups
; g
++) {
1224 for (sfb
= 0; sfb
< sce
->ics
.max_sfb
; sfb
++, idx
++) {
1225 if (sce
->band_type
[idx
] != ZERO_BT
) {
1227 int t
= get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
1235 gain_cache
= pow(scale
, -t
) * s
;
1238 coup
->gain
[c
][idx
] = gain_cache
;
1248 * Decode Spectral Band Replication extension data; reference: table 4.55.
1250 * @param crc flag indicating the presence of CRC checksum
1251 * @param cnt length of TYPE_FIL syntactic element in bytes
1253 * @return Returns number of bytes consumed from the TYPE_FIL element.
1255 static int decode_sbr_extension(AACContext
* ac
, GetBitContext
* gb
, int crc
, int cnt
) {
1256 // TODO : sbr_extension implementation
1257 ff_log_missing_feature(ac
->avccontext
, "SBR", 0);
1258 skip_bits_long(gb
, 8*cnt
- 4); // -4 due to reading extension type
1263 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1265 * @return Returns number of bytes consumed.
1267 static int decode_drc_channel_exclusions(DynamicRangeControl
*che_drc
, GetBitContext
* gb
) {
1269 int num_excl_chan
= 0;
1272 for (i
= 0; i
< 7; i
++)
1273 che_drc
->exclude_mask
[num_excl_chan
++] = get_bits1(gb
);
1274 } while (num_excl_chan
< MAX_CHANNELS
- 7 && get_bits1(gb
));
1276 return num_excl_chan
/ 7;
1280 * Decode dynamic range information; reference: table 4.52.
1282 * @param cnt length of TYPE_FIL syntactic element in bytes
1284 * @return Returns number of bytes consumed.
1286 static int decode_dynamic_range(DynamicRangeControl
*che_drc
, GetBitContext
* gb
, int cnt
) {
1288 int drc_num_bands
= 1;
1291 /* pce_tag_present? */
1293 che_drc
->pce_instance_tag
= get_bits(gb
, 4);
1294 skip_bits(gb
, 4); // tag_reserved_bits
1298 /* excluded_chns_present? */
1300 n
+= decode_drc_channel_exclusions(che_drc
, gb
);
1303 /* drc_bands_present? */
1304 if (get_bits1(gb
)) {
1305 che_drc
->band_incr
= get_bits(gb
, 4);
1306 che_drc
->interpolation_scheme
= get_bits(gb
, 4);
1308 drc_num_bands
+= che_drc
->band_incr
;
1309 for (i
= 0; i
< drc_num_bands
; i
++) {
1310 che_drc
->band_top
[i
] = get_bits(gb
, 8);
1315 /* prog_ref_level_present? */
1316 if (get_bits1(gb
)) {
1317 che_drc
->prog_ref_level
= get_bits(gb
, 7);
1318 skip_bits1(gb
); // prog_ref_level_reserved_bits
1322 for (i
= 0; i
< drc_num_bands
; i
++) {
1323 che_drc
->dyn_rng_sgn
[i
] = get_bits1(gb
);
1324 che_drc
->dyn_rng_ctl
[i
] = get_bits(gb
, 7);
1332 * Decode extension data (incomplete); reference: table 4.51.
1334 * @param cnt length of TYPE_FIL syntactic element in bytes
1336 * @return Returns number of bytes consumed
1338 static int decode_extension_payload(AACContext
* ac
, GetBitContext
* gb
, int cnt
) {
1341 switch (get_bits(gb
, 4)) { // extension type
1342 case EXT_SBR_DATA_CRC
:
1345 res
= decode_sbr_extension(ac
, gb
, crc_flag
, cnt
);
1347 case EXT_DYNAMIC_RANGE
:
1348 res
= decode_dynamic_range(&ac
->che_drc
, gb
, cnt
);
1352 case EXT_DATA_ELEMENT
:
1354 skip_bits_long(gb
, 8*cnt
- 4);
1361 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1363 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1364 * @param coef spectral coefficients
1366 static void apply_tns(float coef
[1024], TemporalNoiseShaping
* tns
, IndividualChannelStream
* ics
, int decode
) {
1367 const int mmm
= FFMIN(ics
->tns_max_bands
, ics
->max_sfb
);
1369 int bottom
, top
, order
, start
, end
, size
, inc
;
1370 float lpc
[TNS_MAX_ORDER
];
1372 for (w
= 0; w
< ics
->num_windows
; w
++) {
1373 bottom
= ics
->num_swb
;
1374 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
1376 bottom
= FFMAX(0, top
- tns
->length
[w
][filt
]);
1377 order
= tns
->order
[w
][filt
];
1382 compute_lpc_coefs(tns
->coef
[w
][filt
], order
, lpc
, 0, 0, 0);
1384 start
= ics
->swb_offset
[FFMIN(bottom
, mmm
)];
1385 end
= ics
->swb_offset
[FFMIN( top
, mmm
)];
1386 if ((size
= end
- start
) <= 0)
1388 if (tns
->direction
[w
][filt
]) {
1389 inc
= -1; start
= end
- 1;
1396 for (m
= 0; m
< size
; m
++, start
+= inc
)
1397 for (i
= 1; i
<= FFMIN(m
, order
); i
++)
1398 coef
[start
] -= coef
[start
- i
*inc
] * lpc
[i
-1];
1404 * Conduct IMDCT and windowing.
1406 static void imdct_and_windowing(AACContext
* ac
, SingleChannelElement
* sce
) {
1407 IndividualChannelStream
* ics
= &sce
->ics
;
1408 float * in
= sce
->coeffs
;
1409 float * out
= sce
->ret
;
1410 float * saved
= sce
->saved
;
1411 const float * swindow
= ics
->use_kb_window
[0] ? ff_aac_kbd_short_128
: ff_sine_128
;
1412 const float * lwindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_long_1024
: ff_sine_1024
;
1413 const float * swindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_short_128
: ff_sine_128
;
1414 float * buf
= ac
->buf_mdct
;
1415 float * temp
= ac
->temp
;
1419 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1420 if (ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
)
1421 av_log(ac
->avccontext
, AV_LOG_WARNING
,
1422 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1423 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1424 for (i
= 0; i
< 1024; i
+= 128)
1425 ff_imdct_half(&ac
->mdct_small
, buf
+ i
, in
+ i
);
1427 ff_imdct_half(&ac
->mdct
, buf
, in
);
1429 /* window overlapping
1430 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1431 * and long to short transitions are considered to be short to short
1432 * transitions. This leaves just two cases (long to long and short to short)
1433 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1435 if ((ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
) &&
1436 (ics
->window_sequence
[0] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[0] == LONG_START_SEQUENCE
)) {
1437 ac
->dsp
.vector_fmul_window( out
, saved
, buf
, lwindow_prev
, ac
->add_bias
, 512);
1439 for (i
= 0; i
< 448; i
++)
1440 out
[i
] = saved
[i
] + ac
->add_bias
;
1442 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1443 ac
->dsp
.vector_fmul_window(out
+ 448 + 0*128, saved
+ 448, buf
+ 0*128, swindow_prev
, ac
->add_bias
, 64);
1444 ac
->dsp
.vector_fmul_window(out
+ 448 + 1*128, buf
+ 0*128 + 64, buf
+ 1*128, swindow
, ac
->add_bias
, 64);
1445 ac
->dsp
.vector_fmul_window(out
+ 448 + 2*128, buf
+ 1*128 + 64, buf
+ 2*128, swindow
, ac
->add_bias
, 64);
1446 ac
->dsp
.vector_fmul_window(out
+ 448 + 3*128, buf
+ 2*128 + 64, buf
+ 3*128, swindow
, ac
->add_bias
, 64);
1447 ac
->dsp
.vector_fmul_window(temp
, buf
+ 3*128 + 64, buf
+ 4*128, swindow
, ac
->add_bias
, 64);
1448 memcpy( out
+ 448 + 4*128, temp
, 64 * sizeof(float));
1450 ac
->dsp
.vector_fmul_window(out
+ 448, saved
+ 448, buf
, swindow_prev
, ac
->add_bias
, 64);
1451 for (i
= 576; i
< 1024; i
++)
1452 out
[i
] = buf
[i
-512] + ac
->add_bias
;
1457 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1458 for (i
= 0; i
< 64; i
++)
1459 saved
[i
] = temp
[64 + i
] - ac
->add_bias
;
1460 ac
->dsp
.vector_fmul_window(saved
+ 64, buf
+ 4*128 + 64, buf
+ 5*128, swindow
, 0, 64);
1461 ac
->dsp
.vector_fmul_window(saved
+ 192, buf
+ 5*128 + 64, buf
+ 6*128, swindow
, 0, 64);
1462 ac
->dsp
.vector_fmul_window(saved
+ 320, buf
+ 6*128 + 64, buf
+ 7*128, swindow
, 0, 64);
1463 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1464 } else if (ics
->window_sequence
[0] == LONG_START_SEQUENCE
) {
1465 memcpy( saved
, buf
+ 512, 448 * sizeof(float));
1466 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1467 } else { // LONG_STOP or ONLY_LONG
1468 memcpy( saved
, buf
+ 512, 512 * sizeof(float));
1473 * Apply dependent channel coupling (applied before IMDCT).
1475 * @param index index into coupling gain array
1477 static void apply_dependent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1478 IndividualChannelStream
* ics
= &cce
->ch
[0].ics
;
1479 const uint16_t * offsets
= ics
->swb_offset
;
1480 float * dest
= target
->coeffs
;
1481 const float * src
= cce
->ch
[0].coeffs
;
1482 int g
, i
, group
, k
, idx
= 0;
1483 if(ac
->m4ac
.object_type
== AOT_AAC_LTP
) {
1484 av_log(ac
->avccontext
, AV_LOG_ERROR
,
1485 "Dependent coupling is not supported together with LTP\n");
1488 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1489 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1490 if (cce
->ch
[0].band_type
[idx
] != ZERO_BT
) {
1491 const float gain
= cce
->coup
.gain
[index
][idx
];
1492 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1493 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
1495 dest
[group
*128+k
] += gain
* src
[group
*128+k
];
1500 dest
+= ics
->group_len
[g
]*128;
1501 src
+= ics
->group_len
[g
]*128;
1506 * Apply independent channel coupling (applied after IMDCT).
1508 * @param index index into coupling gain array
1510 static void apply_independent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1512 const float gain
= cce
->coup
.gain
[index
][0];
1513 const float bias
= ac
->add_bias
;
1514 const float* src
= cce
->ch
[0].ret
;
1515 float* dest
= target
->ret
;
1517 for (i
= 0; i
< 1024; i
++)
1518 dest
[i
] += gain
* (src
[i
] - bias
);
1522 * channel coupling transformation interface
1524 * @param index index into coupling gain array
1525 * @param apply_coupling_method pointer to (in)dependent coupling function
1527 static void apply_channel_coupling(AACContext
* ac
, ChannelElement
* cc
,
1528 enum RawDataBlockType type
, int elem_id
, enum CouplingPoint coupling_point
,
1529 void (*apply_coupling_method
)(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
))
1533 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1534 ChannelElement
*cce
= ac
->che
[TYPE_CCE
][i
];
1537 if (cce
&& cce
->coup
.coupling_point
== coupling_point
) {
1538 ChannelCoupling
* coup
= &cce
->coup
;
1540 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1541 if (coup
->type
[c
] == type
&& coup
->id_select
[c
] == elem_id
) {
1542 if (coup
->ch_select
[c
] != 1) {
1543 apply_coupling_method(ac
, &cc
->ch
[0], cce
, index
);
1544 if (coup
->ch_select
[c
] != 0)
1547 if (coup
->ch_select
[c
] != 2)
1548 apply_coupling_method(ac
, &cc
->ch
[1], cce
, index
++);
1550 index
+= 1 + (coup
->ch_select
[c
] == 3);
1557 * Convert spectral data to float samples, applying all supported tools as appropriate.
1559 static void spectral_to_sample(AACContext
* ac
) {
1561 for(type
= 3; type
>= 0; type
--) {
1562 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1563 ChannelElement
*che
= ac
->che
[type
][i
];
1565 if(type
<= TYPE_CPE
)
1566 apply_channel_coupling(ac
, che
, type
, i
, BEFORE_TNS
, apply_dependent_coupling
);
1567 if(che
->ch
[0].tns
.present
)
1568 apply_tns(che
->ch
[0].coeffs
, &che
->ch
[0].tns
, &che
->ch
[0].ics
, 1);
1569 if(che
->ch
[1].tns
.present
)
1570 apply_tns(che
->ch
[1].coeffs
, &che
->ch
[1].tns
, &che
->ch
[1].ics
, 1);
1571 if(type
<= TYPE_CPE
)
1572 apply_channel_coupling(ac
, che
, type
, i
, BETWEEN_TNS_AND_IMDCT
, apply_dependent_coupling
);
1573 if(type
!= TYPE_CCE
|| che
->coup
.coupling_point
== AFTER_IMDCT
)
1574 imdct_and_windowing(ac
, &che
->ch
[0]);
1575 if(type
== TYPE_CPE
)
1576 imdct_and_windowing(ac
, &che
->ch
[1]);
1577 if(type
<= TYPE_CCE
)
1578 apply_channel_coupling(ac
, che
, type
, i
, AFTER_IMDCT
, apply_independent_coupling
);
1584 static int parse_adts_frame_header(AACContext
* ac
, GetBitContext
* gb
) {
1587 AACADTSHeaderInfo hdr_info
;
1589 size
= ff_aac_parse_header(gb
, &hdr_info
);
1591 if (hdr_info
.chan_config
)
1592 ac
->m4ac
.chan_config
= hdr_info
.chan_config
;
1593 ac
->m4ac
.sample_rate
= hdr_info
.sample_rate
;
1594 ac
->m4ac
.sampling_index
= hdr_info
.sampling_index
;
1595 ac
->m4ac
.object_type
= hdr_info
.object_type
;
1596 if (hdr_info
.num_aac_frames
== 1) {
1597 if (!hdr_info
.crc_absent
)
1600 ff_log_missing_feature(ac
->avccontext
, "More than one AAC RDB per ADTS frame is", 0);
1607 static int aac_decode_frame(AVCodecContext
* avccontext
, void * data
, int * data_size
, AVPacket
*avpkt
) {
1608 const uint8_t *buf
= avpkt
->data
;
1609 int buf_size
= avpkt
->size
;
1610 AACContext
* ac
= avccontext
->priv_data
;
1611 ChannelElement
* che
= NULL
;
1613 enum RawDataBlockType elem_type
;
1614 int err
, elem_id
, data_size_tmp
;
1616 init_get_bits(&gb
, buf
, buf_size
*8);
1618 if (show_bits(&gb
, 12) == 0xfff) {
1619 if ((err
= parse_adts_frame_header(ac
, &gb
)) < 0) {
1620 av_log(avccontext
, AV_LOG_ERROR
, "Error decoding AAC frame header.\n");
1623 if (ac
->m4ac
.sampling_index
> 12) {
1624 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
1630 while ((elem_type
= get_bits(&gb
, 3)) != TYPE_END
) {
1631 elem_id
= get_bits(&gb
, 4);
1634 if(elem_type
< TYPE_DSE
&& !(che
=get_che(ac
, elem_type
, elem_id
))) {
1635 av_log(ac
->avccontext
, AV_LOG_ERROR
, "channel element %d.%d is not allocated\n", elem_type
, elem_id
);
1639 switch (elem_type
) {
1642 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1646 err
= decode_cpe(ac
, &gb
, che
);
1650 err
= decode_cce(ac
, &gb
, che
);
1654 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1658 skip_data_stream_element(&gb
);
1664 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
1665 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
1666 if((err
= decode_pce(ac
, new_che_pos
, &gb
)))
1668 err
= output_configure(ac
, ac
->che_pos
, new_che_pos
, 0);
1674 elem_id
+= get_bits(&gb
, 8) - 1;
1676 elem_id
-= decode_extension_payload(ac
, &gb
, elem_id
);
1677 err
= 0; /* FIXME */
1681 err
= -1; /* should not happen, but keeps compiler happy */
1689 spectral_to_sample(ac
);
1691 if (!ac
->is_saved
) {
1697 data_size_tmp
= 1024 * avccontext
->channels
* sizeof(int16_t);
1698 if(*data_size
< data_size_tmp
) {
1699 av_log(avccontext
, AV_LOG_ERROR
,
1700 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1701 *data_size
, data_size_tmp
);
1704 *data_size
= data_size_tmp
;
1706 ac
->dsp
.float_to_int16_interleave(data
, (const float **)ac
->output_data
, 1024, avccontext
->channels
);
1711 static av_cold
int aac_decode_close(AVCodecContext
* avccontext
) {
1712 AACContext
* ac
= avccontext
->priv_data
;
1715 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1716 for(type
= 0; type
< 4; type
++)
1717 av_freep(&ac
->che
[type
][i
]);
1720 ff_mdct_end(&ac
->mdct
);
1721 ff_mdct_end(&ac
->mdct_small
);
1725 AVCodec aac_decoder
= {
1734 .long_name
= NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1735 .sample_fmts
= (enum SampleFormat
[]){SAMPLE_FMT_S16
,SAMPLE_FMT_NONE
},