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
87 #include "aacdectab.h"
88 #include "mpeg4audio.h"
89 #include "aac_parser.h"
101 static VLC vlc_scalefactors
;
102 static VLC vlc_spectral
[11];
105 static ChannelElement
*get_che(AACContext
*ac
, int type
, int elem_id
)
107 static const int8_t tags_per_config
[16] = { 0, 1, 1, 2, 3, 3, 4, 5, 0, 0, 0, 0, 0, 0, 0, 0 };
108 if (ac
->tag_che_map
[type
][elem_id
]) {
109 return ac
->tag_che_map
[type
][elem_id
];
111 if (ac
->tags_mapped
>= tags_per_config
[ac
->m4ac
.chan_config
]) {
114 switch (ac
->m4ac
.chan_config
) {
116 if (ac
->tags_mapped
== 3 && type
== TYPE_CPE
) {
118 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][2];
121 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
122 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
123 encountered such a stream, transfer the LFE[0] element to SCE[1] */
124 if (ac
->tags_mapped
== tags_per_config
[ac
->m4ac
.chan_config
] - 1 && (type
== TYPE_LFE
|| type
== TYPE_SCE
)) {
126 return ac
->tag_che_map
[type
][elem_id
] = ac
->che
[TYPE_LFE
][0];
129 if (ac
->tags_mapped
== 2 && type
== TYPE_CPE
) {
131 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][1];
134 if (ac
->tags_mapped
== 2 && ac
->m4ac
.chan_config
== 4 && type
== TYPE_SCE
) {
136 return ac
->tag_che_map
[TYPE_SCE
][elem_id
] = ac
->che
[TYPE_SCE
][1];
140 if (ac
->tags_mapped
== (ac
->m4ac
.chan_config
!= 2) && type
== TYPE_CPE
) {
142 return ac
->tag_che_map
[TYPE_CPE
][elem_id
] = ac
->che
[TYPE_CPE
][0];
143 } else if (ac
->m4ac
.chan_config
== 2) {
147 if (!ac
->tags_mapped
&& type
== TYPE_SCE
) {
149 return ac
->tag_che_map
[TYPE_SCE
][elem_id
] = ac
->che
[TYPE_SCE
][0];
157 * Configure output channel order based on the current program configuration element.
159 * @param che_pos current channel position configuration
160 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
162 * @return Returns error status. 0 - OK, !0 - error
164 static int output_configure(AACContext
*ac
,
165 enum ChannelPosition che_pos
[4][MAX_ELEM_ID
],
166 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
169 AVCodecContext
*avctx
= ac
->avccontext
;
170 int i
, type
, channels
= 0;
172 memcpy(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
174 /* Allocate or free elements depending on if they are in the
175 * current program configuration.
177 * Set up default 1:1 output mapping.
179 * For a 5.1 stream the output order will be:
180 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
183 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
184 for (type
= 0; type
< 4; type
++) {
185 if (che_pos
[type
][i
]) {
186 if (!ac
->che
[type
][i
] && !(ac
->che
[type
][i
] = av_mallocz(sizeof(ChannelElement
))))
187 return AVERROR(ENOMEM
);
188 if (type
!= TYPE_CCE
) {
189 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[0].ret
;
190 if (type
== TYPE_CPE
) {
191 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[1].ret
;
195 av_freep(&ac
->che
[type
][i
]);
199 if (channel_config
) {
200 memset(ac
->tag_che_map
, 0, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
203 memcpy(ac
->tag_che_map
, ac
->che
, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
204 ac
->tags_mapped
= 4 * MAX_ELEM_ID
;
207 avctx
->channels
= channels
;
209 ac
->output_configured
= 1;
215 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
217 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
218 * @param sce_map mono (Single Channel Element) map
219 * @param type speaker type/position for these channels
221 static void decode_channel_map(enum ChannelPosition
*cpe_map
,
222 enum ChannelPosition
*sce_map
,
223 enum ChannelPosition type
,
224 GetBitContext
*gb
, int n
)
227 enum ChannelPosition
*map
= cpe_map
&& get_bits1(gb
) ? cpe_map
: sce_map
; // stereo or mono map
228 map
[get_bits(gb
, 4)] = type
;
233 * Decode program configuration element; reference: table 4.2.
235 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
237 * @return Returns error status. 0 - OK, !0 - error
239 static int decode_pce(AACContext
*ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
242 int num_front
, num_side
, num_back
, num_lfe
, num_assoc_data
, num_cc
, sampling_index
;
244 skip_bits(gb
, 2); // object_type
246 sampling_index
= get_bits(gb
, 4);
247 if (ac
->m4ac
.sampling_index
!= sampling_index
)
248 av_log(ac
->avccontext
, AV_LOG_WARNING
, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
250 num_front
= get_bits(gb
, 4);
251 num_side
= get_bits(gb
, 4);
252 num_back
= get_bits(gb
, 4);
253 num_lfe
= get_bits(gb
, 2);
254 num_assoc_data
= get_bits(gb
, 3);
255 num_cc
= get_bits(gb
, 4);
258 skip_bits(gb
, 4); // mono_mixdown_tag
260 skip_bits(gb
, 4); // stereo_mixdown_tag
263 skip_bits(gb
, 3); // mixdown_coeff_index and pseudo_surround
265 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_FRONT
, gb
, num_front
);
266 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_SIDE
, gb
, num_side
);
267 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_BACK
, gb
, num_back
);
268 decode_channel_map(NULL
, new_che_pos
[TYPE_LFE
], AAC_CHANNEL_LFE
, gb
, num_lfe
);
270 skip_bits_long(gb
, 4 * num_assoc_data
);
272 decode_channel_map(new_che_pos
[TYPE_CCE
], new_che_pos
[TYPE_CCE
], AAC_CHANNEL_CC
, gb
, num_cc
);
276 /* comment field, first byte is length */
277 skip_bits_long(gb
, 8 * get_bits(gb
, 8));
282 * Set up channel positions based on a default channel configuration
283 * as specified in table 1.17.
285 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
287 * @return Returns error status. 0 - OK, !0 - error
289 static int set_default_channel_config(AACContext
*ac
,
290 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
293 if (channel_config
< 1 || channel_config
> 7) {
294 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid default channel configuration (%d)\n",
299 /* default channel configurations:
301 * 1ch : front center (mono)
302 * 2ch : L + R (stereo)
303 * 3ch : front center + L + R
304 * 4ch : front center + L + R + back center
305 * 5ch : front center + L + R + back stereo
306 * 6ch : front center + L + R + back stereo + LFE
307 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
310 if (channel_config
!= 2)
311 new_che_pos
[TYPE_SCE
][0] = AAC_CHANNEL_FRONT
; // front center (or mono)
312 if (channel_config
> 1)
313 new_che_pos
[TYPE_CPE
][0] = AAC_CHANNEL_FRONT
; // L + R (or stereo)
314 if (channel_config
== 4)
315 new_che_pos
[TYPE_SCE
][1] = AAC_CHANNEL_BACK
; // back center
316 if (channel_config
> 4)
317 new_che_pos
[TYPE_CPE
][(channel_config
== 7) + 1]
318 = AAC_CHANNEL_BACK
; // back stereo
319 if (channel_config
> 5)
320 new_che_pos
[TYPE_LFE
][0] = AAC_CHANNEL_LFE
; // LFE
321 if (channel_config
== 7)
322 new_che_pos
[TYPE_CPE
][1] = AAC_CHANNEL_FRONT
; // outer front left + outer front right
328 * Decode GA "General Audio" specific configuration; reference: table 4.1.
330 * @return Returns error status. 0 - OK, !0 - error
332 static int decode_ga_specific_config(AACContext
*ac
, GetBitContext
*gb
,
335 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
336 int extension_flag
, ret
;
338 if (get_bits1(gb
)) { // frameLengthFlag
339 av_log_missing_feature(ac
->avccontext
, "960/120 MDCT window is", 1);
343 if (get_bits1(gb
)) // dependsOnCoreCoder
344 skip_bits(gb
, 14); // coreCoderDelay
345 extension_flag
= get_bits1(gb
);
347 if (ac
->m4ac
.object_type
== AOT_AAC_SCALABLE
||
348 ac
->m4ac
.object_type
== AOT_ER_AAC_SCALABLE
)
349 skip_bits(gb
, 3); // layerNr
351 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
352 if (channel_config
== 0) {
353 skip_bits(gb
, 4); // element_instance_tag
354 if ((ret
= decode_pce(ac
, new_che_pos
, gb
)))
357 if ((ret
= set_default_channel_config(ac
, new_che_pos
, channel_config
)))
360 if ((ret
= output_configure(ac
, ac
->che_pos
, new_che_pos
, channel_config
)))
363 if (extension_flag
) {
364 switch (ac
->m4ac
.object_type
) {
366 skip_bits(gb
, 5); // numOfSubFrame
367 skip_bits(gb
, 11); // layer_length
371 case AOT_ER_AAC_SCALABLE
:
373 skip_bits(gb
, 3); /* aacSectionDataResilienceFlag
374 * aacScalefactorDataResilienceFlag
375 * aacSpectralDataResilienceFlag
379 skip_bits1(gb
); // extensionFlag3 (TBD in version 3)
385 * Decode audio specific configuration; reference: table 1.13.
387 * @param data pointer to AVCodecContext extradata
388 * @param data_size size of AVCCodecContext extradata
390 * @return Returns error status. 0 - OK, !0 - error
392 static int decode_audio_specific_config(AACContext
*ac
, void *data
,
398 init_get_bits(&gb
, data
, data_size
* 8);
400 if ((i
= ff_mpeg4audio_get_config(&ac
->m4ac
, data
, data_size
)) < 0)
402 if (ac
->m4ac
.sampling_index
> 12) {
403 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
407 skip_bits_long(&gb
, i
);
409 switch (ac
->m4ac
.object_type
) {
412 if (decode_ga_specific_config(ac
, &gb
, ac
->m4ac
.chan_config
))
416 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Audio object type %s%d is not supported.\n",
417 ac
->m4ac
.sbr
== 1? "SBR+" : "", ac
->m4ac
.object_type
);
424 * linear congruential pseudorandom number generator
426 * @param previous_val pointer to the current state of the generator
428 * @return Returns a 32-bit pseudorandom integer
430 static av_always_inline
int lcg_random(int previous_val
)
432 return previous_val
* 1664525 + 1013904223;
435 static void reset_predict_state(PredictorState
*ps
)
445 static void reset_all_predictors(PredictorState
*ps
)
448 for (i
= 0; i
< MAX_PREDICTORS
; i
++)
449 reset_predict_state(&ps
[i
]);
452 static void reset_predictor_group(PredictorState
*ps
, int group_num
)
455 for (i
= group_num
- 1; i
< MAX_PREDICTORS
; i
+= 30)
456 reset_predict_state(&ps
[i
]);
459 static av_cold
int aac_decode_init(AVCodecContext
*avccontext
)
461 AACContext
*ac
= avccontext
->priv_data
;
464 ac
->avccontext
= avccontext
;
466 if (avccontext
->extradata_size
> 0) {
467 if (decode_audio_specific_config(ac
, avccontext
->extradata
, avccontext
->extradata_size
))
469 avccontext
->sample_rate
= ac
->m4ac
.sample_rate
;
470 } else if (avccontext
->channels
> 0) {
471 ac
->m4ac
.sample_rate
= avccontext
->sample_rate
;
474 avccontext
->sample_fmt
= SAMPLE_FMT_S16
;
475 avccontext
->frame_size
= 1024;
477 AAC_INIT_VLC_STATIC( 0, 144);
478 AAC_INIT_VLC_STATIC( 1, 114);
479 AAC_INIT_VLC_STATIC( 2, 188);
480 AAC_INIT_VLC_STATIC( 3, 180);
481 AAC_INIT_VLC_STATIC( 4, 172);
482 AAC_INIT_VLC_STATIC( 5, 140);
483 AAC_INIT_VLC_STATIC( 6, 168);
484 AAC_INIT_VLC_STATIC( 7, 114);
485 AAC_INIT_VLC_STATIC( 8, 262);
486 AAC_INIT_VLC_STATIC( 9, 248);
487 AAC_INIT_VLC_STATIC(10, 384);
489 dsputil_init(&ac
->dsp
, avccontext
);
491 ac
->random_state
= 0x1f2e3d4c;
493 // -1024 - Compensate wrong IMDCT method.
494 // 32768 - Required to scale values to the correct range for the bias method
495 // for float to int16 conversion.
497 if (ac
->dsp
.float_to_int16
== ff_float_to_int16_c
) {
498 ac
->add_bias
= 385.0f
;
499 ac
->sf_scale
= 1. / (-1024. * 32768.);
503 ac
->sf_scale
= 1. / -1024.;
507 #if !CONFIG_HARDCODED_TABLES
508 for (i
= 0; i
< 428; i
++)
509 ff_aac_pow2sf_tab
[i
] = pow(2, (i
- 200) / 4.);
510 #endif /* CONFIG_HARDCODED_TABLES */
512 INIT_VLC_STATIC(&vlc_scalefactors
,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code
),
513 ff_aac_scalefactor_bits
, sizeof(ff_aac_scalefactor_bits
[0]), sizeof(ff_aac_scalefactor_bits
[0]),
514 ff_aac_scalefactor_code
, sizeof(ff_aac_scalefactor_code
[0]), sizeof(ff_aac_scalefactor_code
[0]),
517 ff_mdct_init(&ac
->mdct
, 11, 1, 1.0);
518 ff_mdct_init(&ac
->mdct_small
, 8, 1, 1.0);
519 // window initialization
520 ff_kbd_window_init(ff_aac_kbd_long_1024
, 4.0, 1024);
521 ff_kbd_window_init(ff_aac_kbd_short_128
, 6.0, 128);
522 ff_sine_window_init(ff_sine_1024
, 1024);
523 ff_sine_window_init(ff_sine_128
, 128);
529 * Skip data_stream_element; reference: table 4.10.
531 static void skip_data_stream_element(GetBitContext
*gb
)
533 int byte_align
= get_bits1(gb
);
534 int count
= get_bits(gb
, 8);
536 count
+= get_bits(gb
, 8);
539 skip_bits_long(gb
, 8 * count
);
542 static int decode_prediction(AACContext
*ac
, IndividualChannelStream
*ics
,
547 ics
->predictor_reset_group
= get_bits(gb
, 5);
548 if (ics
->predictor_reset_group
== 0 || ics
->predictor_reset_group
> 30) {
549 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Invalid Predictor Reset Group.\n");
553 for (sfb
= 0; sfb
< FFMIN(ics
->max_sfb
, ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]); sfb
++) {
554 ics
->prediction_used
[sfb
] = get_bits1(gb
);
560 * Decode Individual Channel Stream info; reference: table 4.6.
562 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
564 static int decode_ics_info(AACContext
*ac
, IndividualChannelStream
*ics
,
565 GetBitContext
*gb
, int common_window
)
568 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Reserved bit set.\n");
569 memset(ics
, 0, sizeof(IndividualChannelStream
));
572 ics
->window_sequence
[1] = ics
->window_sequence
[0];
573 ics
->window_sequence
[0] = get_bits(gb
, 2);
574 ics
->use_kb_window
[1] = ics
->use_kb_window
[0];
575 ics
->use_kb_window
[0] = get_bits1(gb
);
576 ics
->num_window_groups
= 1;
577 ics
->group_len
[0] = 1;
578 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
580 ics
->max_sfb
= get_bits(gb
, 4);
581 for (i
= 0; i
< 7; i
++) {
583 ics
->group_len
[ics
->num_window_groups
- 1]++;
585 ics
->num_window_groups
++;
586 ics
->group_len
[ics
->num_window_groups
- 1] = 1;
589 ics
->num_windows
= 8;
590 ics
->swb_offset
= ff_swb_offset_128
[ac
->m4ac
.sampling_index
];
591 ics
->num_swb
= ff_aac_num_swb_128
[ac
->m4ac
.sampling_index
];
592 ics
->tns_max_bands
= ff_tns_max_bands_128
[ac
->m4ac
.sampling_index
];
593 ics
->predictor_present
= 0;
595 ics
->max_sfb
= get_bits(gb
, 6);
596 ics
->num_windows
= 1;
597 ics
->swb_offset
= ff_swb_offset_1024
[ac
->m4ac
.sampling_index
];
598 ics
->num_swb
= ff_aac_num_swb_1024
[ac
->m4ac
.sampling_index
];
599 ics
->tns_max_bands
= ff_tns_max_bands_1024
[ac
->m4ac
.sampling_index
];
600 ics
->predictor_present
= get_bits1(gb
);
601 ics
->predictor_reset_group
= 0;
602 if (ics
->predictor_present
) {
603 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
604 if (decode_prediction(ac
, ics
, gb
)) {
605 memset(ics
, 0, sizeof(IndividualChannelStream
));
608 } else if (ac
->m4ac
.object_type
== AOT_AAC_LC
) {
609 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Prediction is not allowed in AAC-LC.\n");
610 memset(ics
, 0, sizeof(IndividualChannelStream
));
613 av_log_missing_feature(ac
->avccontext
, "Predictor bit set but LTP is", 1);
614 memset(ics
, 0, sizeof(IndividualChannelStream
));
620 if (ics
->max_sfb
> ics
->num_swb
) {
621 av_log(ac
->avccontext
, AV_LOG_ERROR
,
622 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
623 ics
->max_sfb
, ics
->num_swb
);
624 memset(ics
, 0, sizeof(IndividualChannelStream
));
632 * Decode band types (section_data payload); reference: table 4.46.
634 * @param band_type array of the used band type
635 * @param band_type_run_end array of the last scalefactor band of a band type run
637 * @return Returns error status. 0 - OK, !0 - error
639 static int decode_band_types(AACContext
*ac
, enum BandType band_type
[120],
640 int band_type_run_end
[120], GetBitContext
*gb
,
641 IndividualChannelStream
*ics
)
644 const int bits
= (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) ? 3 : 5;
645 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
647 while (k
< ics
->max_sfb
) {
648 uint8_t sect_len
= k
;
650 int sect_band_type
= get_bits(gb
, 4);
651 if (sect_band_type
== 12) {
652 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid band type\n");
655 while ((sect_len_incr
= get_bits(gb
, bits
)) == (1 << bits
) - 1)
656 sect_len
+= sect_len_incr
;
657 sect_len
+= sect_len_incr
;
658 if (sect_len
> ics
->max_sfb
) {
659 av_log(ac
->avccontext
, AV_LOG_ERROR
,
660 "Number of bands (%d) exceeds limit (%d).\n",
661 sect_len
, ics
->max_sfb
);
664 for (; k
< sect_len
; k
++) {
665 band_type
[idx
] = sect_band_type
;
666 band_type_run_end
[idx
++] = sect_len
;
674 * Decode scalefactors; reference: table 4.47.
676 * @param global_gain first scalefactor value as scalefactors are differentially coded
677 * @param band_type array of the used band type
678 * @param band_type_run_end array of the last scalefactor band of a band type run
679 * @param sf array of scalefactors or intensity stereo positions
681 * @return Returns error status. 0 - OK, !0 - error
683 static int decode_scalefactors(AACContext
*ac
, float sf
[120], GetBitContext
*gb
,
684 unsigned int global_gain
,
685 IndividualChannelStream
*ics
,
686 enum BandType band_type
[120],
687 int band_type_run_end
[120])
689 const int sf_offset
= ac
->sf_offset
+ (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
? 12 : 0);
691 int offset
[3] = { global_gain
, global_gain
- 90, 100 };
693 static const char *sf_str
[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
694 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
695 for (i
= 0; i
< ics
->max_sfb
;) {
696 int run_end
= band_type_run_end
[idx
];
697 if (band_type
[idx
] == ZERO_BT
) {
698 for (; i
< run_end
; i
++, idx
++)
700 } else if ((band_type
[idx
] == INTENSITY_BT
) || (band_type
[idx
] == INTENSITY_BT2
)) {
701 for (; i
< run_end
; i
++, idx
++) {
702 offset
[2] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
703 if (offset
[2] > 255U) {
704 av_log(ac
->avccontext
, AV_LOG_ERROR
,
705 "%s (%d) out of range.\n", sf_str
[2], offset
[2]);
708 sf
[idx
] = ff_aac_pow2sf_tab
[-offset
[2] + 300];
710 } else if (band_type
[idx
] == NOISE_BT
) {
711 for (; i
< run_end
; i
++, idx
++) {
712 if (noise_flag
-- > 0)
713 offset
[1] += get_bits(gb
, 9) - 256;
715 offset
[1] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
716 if (offset
[1] > 255U) {
717 av_log(ac
->avccontext
, AV_LOG_ERROR
,
718 "%s (%d) out of range.\n", sf_str
[1], offset
[1]);
721 sf
[idx
] = -ff_aac_pow2sf_tab
[offset
[1] + sf_offset
+ 100];
724 for (; i
< run_end
; i
++, idx
++) {
725 offset
[0] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
726 if (offset
[0] > 255U) {
727 av_log(ac
->avccontext
, AV_LOG_ERROR
,
728 "%s (%d) out of range.\n", sf_str
[0], offset
[0]);
731 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[0] + sf_offset
];
740 * Decode pulse data; reference: table 4.7.
742 static int decode_pulses(Pulse
*pulse
, GetBitContext
*gb
,
743 const uint16_t *swb_offset
, int num_swb
)
746 pulse
->num_pulse
= get_bits(gb
, 2) + 1;
747 pulse_swb
= get_bits(gb
, 6);
748 if (pulse_swb
>= num_swb
)
750 pulse
->pos
[0] = swb_offset
[pulse_swb
];
751 pulse
->pos
[0] += get_bits(gb
, 5);
752 if (pulse
->pos
[0] > 1023)
754 pulse
->amp
[0] = get_bits(gb
, 4);
755 for (i
= 1; i
< pulse
->num_pulse
; i
++) {
756 pulse
->pos
[i
] = get_bits(gb
, 5) + pulse
->pos
[i
- 1];
757 if (pulse
->pos
[i
] > 1023)
759 pulse
->amp
[i
] = get_bits(gb
, 4);
765 * Decode Temporal Noise Shaping data; reference: table 4.48.
767 * @return Returns error status. 0 - OK, !0 - error
769 static int decode_tns(AACContext
*ac
, TemporalNoiseShaping
*tns
,
770 GetBitContext
*gb
, const IndividualChannelStream
*ics
)
772 int w
, filt
, i
, coef_len
, coef_res
, coef_compress
;
773 const int is8
= ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
;
774 const int tns_max_order
= is8
? 7 : ac
->m4ac
.object_type
== AOT_AAC_MAIN
? 20 : 12;
775 for (w
= 0; w
< ics
->num_windows
; w
++) {
776 if ((tns
->n_filt
[w
] = get_bits(gb
, 2 - is8
))) {
777 coef_res
= get_bits1(gb
);
779 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
781 tns
->length
[w
][filt
] = get_bits(gb
, 6 - 2 * is8
);
783 if ((tns
->order
[w
][filt
] = get_bits(gb
, 5 - 2 * is8
)) > tns_max_order
) {
784 av_log(ac
->avccontext
, AV_LOG_ERROR
, "TNS filter order %d is greater than maximum %d.",
785 tns
->order
[w
][filt
], tns_max_order
);
786 tns
->order
[w
][filt
] = 0;
789 if (tns
->order
[w
][filt
]) {
790 tns
->direction
[w
][filt
] = get_bits1(gb
);
791 coef_compress
= get_bits1(gb
);
792 coef_len
= coef_res
+ 3 - coef_compress
;
793 tmp2_idx
= 2 * coef_compress
+ coef_res
;
795 for (i
= 0; i
< tns
->order
[w
][filt
]; i
++)
796 tns
->coef
[w
][filt
][i
] = tns_tmp2_map
[tmp2_idx
][get_bits(gb
, coef_len
)];
805 * Decode Mid/Side data; reference: table 4.54.
807 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
808 * [1] mask is decoded from bitstream; [2] mask is all 1s;
809 * [3] reserved for scalable AAC
811 static void decode_mid_side_stereo(ChannelElement
*cpe
, GetBitContext
*gb
,
815 if (ms_present
== 1) {
816 for (idx
= 0; idx
< cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
; idx
++)
817 cpe
->ms_mask
[idx
] = get_bits1(gb
);
818 } else if (ms_present
== 2) {
819 memset(cpe
->ms_mask
, 1, cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
* sizeof(cpe
->ms_mask
[0]));
824 * Decode spectral data; reference: table 4.50.
825 * Dequantize and scale spectral data; reference: 4.6.3.3.
827 * @param coef array of dequantized, scaled spectral data
828 * @param sf array of scalefactors or intensity stereo positions
829 * @param pulse_present set if pulses are present
830 * @param pulse pointer to pulse data struct
831 * @param band_type array of the used band type
833 * @return Returns error status. 0 - OK, !0 - error
835 static int decode_spectrum_and_dequant(AACContext
*ac
, float coef
[1024],
836 GetBitContext
*gb
, float sf
[120],
837 int pulse_present
, const Pulse
*pulse
,
838 const IndividualChannelStream
*ics
,
839 enum BandType band_type
[120])
841 int i
, k
, g
, idx
= 0;
842 const int c
= 1024 / ics
->num_windows
;
843 const uint16_t *offsets
= ics
->swb_offset
;
844 float *coef_base
= coef
;
845 static const float sign_lookup
[] = { 1.0f
, -1.0f
};
847 for (g
= 0; g
< ics
->num_windows
; g
++)
848 memset(coef
+ g
* 128 + offsets
[ics
->max_sfb
], 0, sizeof(float) * (c
- offsets
[ics
->max_sfb
]));
850 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
851 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
852 const int cur_band_type
= band_type
[idx
];
853 const int dim
= cur_band_type
>= FIRST_PAIR_BT
? 2 : 4;
854 const int is_cb_unsigned
= IS_CODEBOOK_UNSIGNED(cur_band_type
);
856 if (cur_band_type
== ZERO_BT
|| cur_band_type
== INTENSITY_BT2
|| cur_band_type
== INTENSITY_BT
) {
857 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
858 memset(coef
+ group
* 128 + offsets
[i
], 0, (offsets
[i
+ 1] - offsets
[i
]) * sizeof(float));
860 } else if (cur_band_type
== NOISE_BT
) {
861 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
863 float band_energy
= 0;
864 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
++) {
865 ac
->random_state
= lcg_random(ac
->random_state
);
866 coef
[group
* 128 + k
] = ac
->random_state
;
867 band_energy
+= coef
[group
* 128 + k
] * coef
[group
* 128 + k
];
869 scale
= sf
[idx
] / sqrtf(band_energy
);
870 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
++) {
871 coef
[group
* 128 + k
] *= scale
;
875 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
876 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
+= dim
) {
877 const int index
= get_vlc2(gb
, vlc_spectral
[cur_band_type
- 1].table
, 6, 3);
878 const int coef_tmp_idx
= (group
<< 7) + k
;
881 if (index
>= ff_aac_spectral_sizes
[cur_band_type
- 1]) {
882 av_log(ac
->avccontext
, AV_LOG_ERROR
,
883 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
884 cur_band_type
- 1, index
, ff_aac_spectral_sizes
[cur_band_type
- 1]);
887 vq_ptr
= &ff_aac_codebook_vectors
[cur_band_type
- 1][index
* dim
];
888 if (is_cb_unsigned
) {
890 coef
[coef_tmp_idx
] = sign_lookup
[get_bits1(gb
)];
892 coef
[coef_tmp_idx
+ 1] = sign_lookup
[get_bits1(gb
)];
895 coef
[coef_tmp_idx
+ 2] = sign_lookup
[get_bits1(gb
)];
897 coef
[coef_tmp_idx
+ 3] = sign_lookup
[get_bits1(gb
)];
899 if (cur_band_type
== ESC_BT
) {
900 for (j
= 0; j
< 2; j
++) {
901 if (vq_ptr
[j
] == 64.0f
) {
903 /* The total length of escape_sequence must be < 22 bits according
904 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
905 while (get_bits1(gb
) && n
< 15) n
++;
907 av_log(ac
->avccontext
, AV_LOG_ERROR
, "error in spectral data, ESC overflow\n");
910 n
= (1 << n
) + get_bits(gb
, n
);
911 coef
[coef_tmp_idx
+ j
] *= cbrtf(n
) * n
;
913 coef
[coef_tmp_idx
+ j
] *= vq_ptr
[j
];
916 coef
[coef_tmp_idx
] *= vq_ptr
[0];
917 coef
[coef_tmp_idx
+ 1] *= vq_ptr
[1];
919 coef
[coef_tmp_idx
+ 2] *= vq_ptr
[2];
920 coef
[coef_tmp_idx
+ 3] *= vq_ptr
[3];
924 coef
[coef_tmp_idx
] = vq_ptr
[0];
925 coef
[coef_tmp_idx
+ 1] = vq_ptr
[1];
927 coef
[coef_tmp_idx
+ 2] = vq_ptr
[2];
928 coef
[coef_tmp_idx
+ 3] = vq_ptr
[3];
931 coef
[coef_tmp_idx
] *= sf
[idx
];
932 coef
[coef_tmp_idx
+ 1] *= sf
[idx
];
934 coef
[coef_tmp_idx
+ 2] *= sf
[idx
];
935 coef
[coef_tmp_idx
+ 3] *= sf
[idx
];
941 coef
+= ics
->group_len
[g
] << 7;
946 for (i
= 0; i
< pulse
->num_pulse
; i
++) {
947 float co
= coef_base
[ pulse
->pos
[i
] ];
948 while (offsets
[idx
+ 1] <= pulse
->pos
[i
])
950 if (band_type
[idx
] != NOISE_BT
&& sf
[idx
]) {
951 float ico
= -pulse
->amp
[i
];
954 ico
= co
/ sqrtf(sqrtf(fabsf(co
))) + (co
> 0 ? -ico
: ico
);
956 coef_base
[ pulse
->pos
[i
] ] = cbrtf(fabsf(ico
)) * ico
* sf
[idx
];
963 static av_always_inline
float flt16_round(float pf
)
967 tmp
.i
= (tmp
.i
+ 0x00008000U
) & 0xFFFF0000U
;
971 static av_always_inline
float flt16_even(float pf
)
975 tmp
.i
= (tmp
.i
+ 0x00007FFFU
+ (tmp
.i
& 0x00010000U
>> 16)) & 0xFFFF0000U
;
979 static av_always_inline
float flt16_trunc(float pf
)
983 pun
.i
&= 0xFFFF0000U
;
987 static void predict(AACContext
*ac
, PredictorState
*ps
, float *coef
,
990 const float a
= 0.953125; // 61.0 / 64
991 const float alpha
= 0.90625; // 29.0 / 32
996 k1
= ps
->var0
> 1 ? ps
->cor0
* flt16_even(a
/ ps
->var0
) : 0;
997 k2
= ps
->var1
> 1 ? ps
->cor1
* flt16_even(a
/ ps
->var1
) : 0;
999 pv
= flt16_round(k1
* ps
->r0
+ k2
* ps
->r1
);
1001 *coef
+= pv
* ac
->sf_scale
;
1003 e0
= *coef
/ ac
->sf_scale
;
1004 e1
= e0
- k1
* ps
->r0
;
1006 ps
->cor1
= flt16_trunc(alpha
* ps
->cor1
+ ps
->r1
* e1
);
1007 ps
->var1
= flt16_trunc(alpha
* ps
->var1
+ 0.5 * (ps
->r1
* ps
->r1
+ e1
* e1
));
1008 ps
->cor0
= flt16_trunc(alpha
* ps
->cor0
+ ps
->r0
* e0
);
1009 ps
->var0
= flt16_trunc(alpha
* ps
->var0
+ 0.5 * (ps
->r0
* ps
->r0
+ e0
* e0
));
1011 ps
->r1
= flt16_trunc(a
* (ps
->r0
- k1
* e0
));
1012 ps
->r0
= flt16_trunc(a
* e0
);
1016 * Apply AAC-Main style frequency domain prediction.
1018 static void apply_prediction(AACContext
*ac
, SingleChannelElement
*sce
)
1022 if (!sce
->ics
.predictor_initialized
) {
1023 reset_all_predictors(sce
->predictor_state
);
1024 sce
->ics
.predictor_initialized
= 1;
1027 if (sce
->ics
.window_sequence
[0] != EIGHT_SHORT_SEQUENCE
) {
1028 for (sfb
= 0; sfb
< ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]; sfb
++) {
1029 for (k
= sce
->ics
.swb_offset
[sfb
]; k
< sce
->ics
.swb_offset
[sfb
+ 1]; k
++) {
1030 predict(ac
, &sce
->predictor_state
[k
], &sce
->coeffs
[k
],
1031 sce
->ics
.predictor_present
&& sce
->ics
.prediction_used
[sfb
]);
1034 if (sce
->ics
.predictor_reset_group
)
1035 reset_predictor_group(sce
->predictor_state
, sce
->ics
.predictor_reset_group
);
1037 reset_all_predictors(sce
->predictor_state
);
1041 * Decode an individual_channel_stream payload; reference: table 4.44.
1043 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1044 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1046 * @return Returns error status. 0 - OK, !0 - error
1048 static int decode_ics(AACContext
*ac
, SingleChannelElement
*sce
,
1049 GetBitContext
*gb
, int common_window
, int scale_flag
)
1052 TemporalNoiseShaping
*tns
= &sce
->tns
;
1053 IndividualChannelStream
*ics
= &sce
->ics
;
1054 float *out
= sce
->coeffs
;
1055 int global_gain
, pulse_present
= 0;
1057 /* This assignment is to silence a GCC warning about the variable being used
1058 * uninitialized when in fact it always is.
1060 pulse
.num_pulse
= 0;
1062 global_gain
= get_bits(gb
, 8);
1064 if (!common_window
&& !scale_flag
) {
1065 if (decode_ics_info(ac
, ics
, gb
, 0) < 0)
1069 if (decode_band_types(ac
, sce
->band_type
, sce
->band_type_run_end
, gb
, ics
) < 0)
1071 if (decode_scalefactors(ac
, sce
->sf
, gb
, global_gain
, ics
, sce
->band_type
, sce
->band_type_run_end
) < 0)
1076 if ((pulse_present
= get_bits1(gb
))) {
1077 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1078 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse tool not allowed in eight short sequence.\n");
1081 if (decode_pulses(&pulse
, gb
, ics
->swb_offset
, ics
->num_swb
)) {
1082 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse data corrupt or invalid.\n");
1086 if ((tns
->present
= get_bits1(gb
)) && decode_tns(ac
, tns
, gb
, ics
))
1088 if (get_bits1(gb
)) {
1089 av_log_missing_feature(ac
->avccontext
, "SSR", 1);
1094 if (decode_spectrum_and_dequant(ac
, out
, gb
, sce
->sf
, pulse_present
, &pulse
, ics
, sce
->band_type
) < 0)
1097 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
&& !common_window
)
1098 apply_prediction(ac
, sce
);
1104 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1106 static void apply_mid_side_stereo(ChannelElement
*cpe
)
1108 const IndividualChannelStream
*ics
= &cpe
->ch
[0].ics
;
1109 float *ch0
= cpe
->ch
[0].coeffs
;
1110 float *ch1
= cpe
->ch
[1].coeffs
;
1111 int g
, i
, k
, group
, idx
= 0;
1112 const uint16_t *offsets
= ics
->swb_offset
;
1113 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1114 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1115 if (cpe
->ms_mask
[idx
] &&
1116 cpe
->ch
[0].band_type
[idx
] < NOISE_BT
&& cpe
->ch
[1].band_type
[idx
] < NOISE_BT
) {
1117 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1118 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
++) {
1119 float tmp
= ch0
[group
* 128 + k
] - ch1
[group
* 128 + k
];
1120 ch0
[group
* 128 + k
] += ch1
[group
* 128 + k
];
1121 ch1
[group
* 128 + k
] = tmp
;
1126 ch0
+= ics
->group_len
[g
] * 128;
1127 ch1
+= ics
->group_len
[g
] * 128;
1132 * intensity stereo decoding; reference: 4.6.8.2.3
1134 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1135 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1136 * [3] reserved for scalable AAC
1138 static void apply_intensity_stereo(ChannelElement
*cpe
, int ms_present
)
1140 const IndividualChannelStream
*ics
= &cpe
->ch
[1].ics
;
1141 SingleChannelElement
*sce1
= &cpe
->ch
[1];
1142 float *coef0
= cpe
->ch
[0].coeffs
, *coef1
= cpe
->ch
[1].coeffs
;
1143 const uint16_t *offsets
= ics
->swb_offset
;
1144 int g
, group
, i
, k
, idx
= 0;
1147 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1148 for (i
= 0; i
< ics
->max_sfb
;) {
1149 if (sce1
->band_type
[idx
] == INTENSITY_BT
|| sce1
->band_type
[idx
] == INTENSITY_BT2
) {
1150 const int bt_run_end
= sce1
->band_type_run_end
[idx
];
1151 for (; i
< bt_run_end
; i
++, idx
++) {
1152 c
= -1 + 2 * (sce1
->band_type
[idx
] - 14);
1154 c
*= 1 - 2 * cpe
->ms_mask
[idx
];
1155 scale
= c
* sce1
->sf
[idx
];
1156 for (group
= 0; group
< ics
->group_len
[g
]; group
++)
1157 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
++)
1158 coef1
[group
* 128 + k
] = scale
* coef0
[group
* 128 + k
];
1161 int bt_run_end
= sce1
->band_type_run_end
[idx
];
1162 idx
+= bt_run_end
- i
;
1166 coef0
+= ics
->group_len
[g
] * 128;
1167 coef1
+= ics
->group_len
[g
] * 128;
1172 * Decode a channel_pair_element; reference: table 4.4.
1174 * @param elem_id Identifies the instance of a syntax element.
1176 * @return Returns error status. 0 - OK, !0 - error
1178 static int decode_cpe(AACContext
*ac
, GetBitContext
*gb
, ChannelElement
*cpe
)
1180 int i
, ret
, common_window
, ms_present
= 0;
1182 common_window
= get_bits1(gb
);
1183 if (common_window
) {
1184 if (decode_ics_info(ac
, &cpe
->ch
[0].ics
, gb
, 1))
1186 i
= cpe
->ch
[1].ics
.use_kb_window
[0];
1187 cpe
->ch
[1].ics
= cpe
->ch
[0].ics
;
1188 cpe
->ch
[1].ics
.use_kb_window
[1] = i
;
1189 ms_present
= get_bits(gb
, 2);
1190 if (ms_present
== 3) {
1191 av_log(ac
->avccontext
, AV_LOG_ERROR
, "ms_present = 3 is reserved.\n");
1193 } else if (ms_present
)
1194 decode_mid_side_stereo(cpe
, gb
, ms_present
);
1196 if ((ret
= decode_ics(ac
, &cpe
->ch
[0], gb
, common_window
, 0)))
1198 if ((ret
= decode_ics(ac
, &cpe
->ch
[1], gb
, common_window
, 0)))
1201 if (common_window
) {
1203 apply_mid_side_stereo(cpe
);
1204 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
1205 apply_prediction(ac
, &cpe
->ch
[0]);
1206 apply_prediction(ac
, &cpe
->ch
[1]);
1210 apply_intensity_stereo(cpe
, ms_present
);
1215 * Decode coupling_channel_element; reference: table 4.8.
1217 * @param elem_id Identifies the instance of a syntax element.
1219 * @return Returns error status. 0 - OK, !0 - error
1221 static int decode_cce(AACContext
*ac
, GetBitContext
*gb
, ChannelElement
*che
)
1227 SingleChannelElement
*sce
= &che
->ch
[0];
1228 ChannelCoupling
*coup
= &che
->coup
;
1230 coup
->coupling_point
= 2 * get_bits1(gb
);
1231 coup
->num_coupled
= get_bits(gb
, 3);
1232 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1234 coup
->type
[c
] = get_bits1(gb
) ? TYPE_CPE
: TYPE_SCE
;
1235 coup
->id_select
[c
] = get_bits(gb
, 4);
1236 if (coup
->type
[c
] == TYPE_CPE
) {
1237 coup
->ch_select
[c
] = get_bits(gb
, 2);
1238 if (coup
->ch_select
[c
] == 3)
1241 coup
->ch_select
[c
] = 2;
1243 coup
->coupling_point
+= get_bits1(gb
) || (coup
->coupling_point
>> 1);
1245 sign
= get_bits(gb
, 1);
1246 scale
= pow(2., pow(2., (int)get_bits(gb
, 2) - 3));
1248 if ((ret
= decode_ics(ac
, sce
, gb
, 0, 0)))
1251 for (c
= 0; c
< num_gain
; c
++) {
1255 float gain_cache
= 1.;
1257 cge
= coup
->coupling_point
== AFTER_IMDCT
? 1 : get_bits1(gb
);
1258 gain
= cge
? get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60: 0;
1259 gain_cache
= pow(scale
, -gain
);
1261 if (coup
->coupling_point
== AFTER_IMDCT
) {
1262 coup
->gain
[c
][0] = gain_cache
;
1264 for (g
= 0; g
< sce
->ics
.num_window_groups
; g
++) {
1265 for (sfb
= 0; sfb
< sce
->ics
.max_sfb
; sfb
++, idx
++) {
1266 if (sce
->band_type
[idx
] != ZERO_BT
) {
1268 int t
= get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
1276 gain_cache
= pow(scale
, -t
) * s
;
1279 coup
->gain
[c
][idx
] = gain_cache
;
1289 * Decode Spectral Band Replication extension data; reference: table 4.55.
1291 * @param crc flag indicating the presence of CRC checksum
1292 * @param cnt length of TYPE_FIL syntactic element in bytes
1294 * @return Returns number of bytes consumed from the TYPE_FIL element.
1296 static int decode_sbr_extension(AACContext
*ac
, GetBitContext
*gb
,
1299 // TODO : sbr_extension implementation
1300 av_log_missing_feature(ac
->avccontext
, "SBR", 0);
1301 skip_bits_long(gb
, 8 * cnt
- 4); // -4 due to reading extension type
1306 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1308 * @return Returns number of bytes consumed.
1310 static int decode_drc_channel_exclusions(DynamicRangeControl
*che_drc
,
1314 int num_excl_chan
= 0;
1317 for (i
= 0; i
< 7; i
++)
1318 che_drc
->exclude_mask
[num_excl_chan
++] = get_bits1(gb
);
1319 } while (num_excl_chan
< MAX_CHANNELS
- 7 && get_bits1(gb
));
1321 return num_excl_chan
/ 7;
1325 * Decode dynamic range information; reference: table 4.52.
1327 * @param cnt length of TYPE_FIL syntactic element in bytes
1329 * @return Returns number of bytes consumed.
1331 static int decode_dynamic_range(DynamicRangeControl
*che_drc
,
1332 GetBitContext
*gb
, int cnt
)
1335 int drc_num_bands
= 1;
1338 /* pce_tag_present? */
1339 if (get_bits1(gb
)) {
1340 che_drc
->pce_instance_tag
= get_bits(gb
, 4);
1341 skip_bits(gb
, 4); // tag_reserved_bits
1345 /* excluded_chns_present? */
1346 if (get_bits1(gb
)) {
1347 n
+= decode_drc_channel_exclusions(che_drc
, gb
);
1350 /* drc_bands_present? */
1351 if (get_bits1(gb
)) {
1352 che_drc
->band_incr
= get_bits(gb
, 4);
1353 che_drc
->interpolation_scheme
= get_bits(gb
, 4);
1355 drc_num_bands
+= che_drc
->band_incr
;
1356 for (i
= 0; i
< drc_num_bands
; i
++) {
1357 che_drc
->band_top
[i
] = get_bits(gb
, 8);
1362 /* prog_ref_level_present? */
1363 if (get_bits1(gb
)) {
1364 che_drc
->prog_ref_level
= get_bits(gb
, 7);
1365 skip_bits1(gb
); // prog_ref_level_reserved_bits
1369 for (i
= 0; i
< drc_num_bands
; i
++) {
1370 che_drc
->dyn_rng_sgn
[i
] = get_bits1(gb
);
1371 che_drc
->dyn_rng_ctl
[i
] = get_bits(gb
, 7);
1379 * Decode extension data (incomplete); reference: table 4.51.
1381 * @param cnt length of TYPE_FIL syntactic element in bytes
1383 * @return Returns number of bytes consumed
1385 static int decode_extension_payload(AACContext
*ac
, GetBitContext
*gb
, int cnt
)
1389 switch (get_bits(gb
, 4)) { // extension type
1390 case EXT_SBR_DATA_CRC
:
1393 res
= decode_sbr_extension(ac
, gb
, crc_flag
, cnt
);
1395 case EXT_DYNAMIC_RANGE
:
1396 res
= decode_dynamic_range(&ac
->che_drc
, gb
, cnt
);
1400 case EXT_DATA_ELEMENT
:
1402 skip_bits_long(gb
, 8 * cnt
- 4);
1409 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1411 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1412 * @param coef spectral coefficients
1414 static void apply_tns(float coef
[1024], TemporalNoiseShaping
*tns
,
1415 IndividualChannelStream
*ics
, int decode
)
1417 const int mmm
= FFMIN(ics
->tns_max_bands
, ics
->max_sfb
);
1419 int bottom
, top
, order
, start
, end
, size
, inc
;
1420 float lpc
[TNS_MAX_ORDER
];
1422 for (w
= 0; w
< ics
->num_windows
; w
++) {
1423 bottom
= ics
->num_swb
;
1424 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
1426 bottom
= FFMAX(0, top
- tns
->length
[w
][filt
]);
1427 order
= tns
->order
[w
][filt
];
1432 compute_lpc_coefs(tns
->coef
[w
][filt
], order
, lpc
, 0, 0, 0);
1434 start
= ics
->swb_offset
[FFMIN(bottom
, mmm
)];
1435 end
= ics
->swb_offset
[FFMIN( top
, mmm
)];
1436 if ((size
= end
- start
) <= 0)
1438 if (tns
->direction
[w
][filt
]) {
1447 for (m
= 0; m
< size
; m
++, start
+= inc
)
1448 for (i
= 1; i
<= FFMIN(m
, order
); i
++)
1449 coef
[start
] -= coef
[start
- i
* inc
] * lpc
[i
- 1];
1455 * Conduct IMDCT and windowing.
1457 static void imdct_and_windowing(AACContext
*ac
, SingleChannelElement
*sce
)
1459 IndividualChannelStream
*ics
= &sce
->ics
;
1460 float *in
= sce
->coeffs
;
1461 float *out
= sce
->ret
;
1462 float *saved
= sce
->saved
;
1463 const float *swindow
= ics
->use_kb_window
[0] ? ff_aac_kbd_short_128
: ff_sine_128
;
1464 const float *lwindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_long_1024
: ff_sine_1024
;
1465 const float *swindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_short_128
: ff_sine_128
;
1466 float *buf
= ac
->buf_mdct
;
1467 float *temp
= ac
->temp
;
1471 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1472 if (ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
)
1473 av_log(ac
->avccontext
, AV_LOG_WARNING
,
1474 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1475 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1476 for (i
= 0; i
< 1024; i
+= 128)
1477 ff_imdct_half(&ac
->mdct_small
, buf
+ i
, in
+ i
);
1479 ff_imdct_half(&ac
->mdct
, buf
, in
);
1481 /* window overlapping
1482 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1483 * and long to short transitions are considered to be short to short
1484 * transitions. This leaves just two cases (long to long and short to short)
1485 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1487 if ((ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
) &&
1488 (ics
->window_sequence
[0] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[0] == LONG_START_SEQUENCE
)) {
1489 ac
->dsp
.vector_fmul_window( out
, saved
, buf
, lwindow_prev
, ac
->add_bias
, 512);
1491 for (i
= 0; i
< 448; i
++)
1492 out
[i
] = saved
[i
] + ac
->add_bias
;
1494 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1495 ac
->dsp
.vector_fmul_window(out
+ 448 + 0*128, saved
+ 448, buf
+ 0*128, swindow_prev
, ac
->add_bias
, 64);
1496 ac
->dsp
.vector_fmul_window(out
+ 448 + 1*128, buf
+ 0*128 + 64, buf
+ 1*128, swindow
, ac
->add_bias
, 64);
1497 ac
->dsp
.vector_fmul_window(out
+ 448 + 2*128, buf
+ 1*128 + 64, buf
+ 2*128, swindow
, ac
->add_bias
, 64);
1498 ac
->dsp
.vector_fmul_window(out
+ 448 + 3*128, buf
+ 2*128 + 64, buf
+ 3*128, swindow
, ac
->add_bias
, 64);
1499 ac
->dsp
.vector_fmul_window(temp
, buf
+ 3*128 + 64, buf
+ 4*128, swindow
, ac
->add_bias
, 64);
1500 memcpy( out
+ 448 + 4*128, temp
, 64 * sizeof(float));
1502 ac
->dsp
.vector_fmul_window(out
+ 448, saved
+ 448, buf
, swindow_prev
, ac
->add_bias
, 64);
1503 for (i
= 576; i
< 1024; i
++)
1504 out
[i
] = buf
[i
-512] + ac
->add_bias
;
1509 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1510 for (i
= 0; i
< 64; i
++)
1511 saved
[i
] = temp
[64 + i
] - ac
->add_bias
;
1512 ac
->dsp
.vector_fmul_window(saved
+ 64, buf
+ 4*128 + 64, buf
+ 5*128, swindow
, 0, 64);
1513 ac
->dsp
.vector_fmul_window(saved
+ 192, buf
+ 5*128 + 64, buf
+ 6*128, swindow
, 0, 64);
1514 ac
->dsp
.vector_fmul_window(saved
+ 320, buf
+ 6*128 + 64, buf
+ 7*128, swindow
, 0, 64);
1515 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1516 } else if (ics
->window_sequence
[0] == LONG_START_SEQUENCE
) {
1517 memcpy( saved
, buf
+ 512, 448 * sizeof(float));
1518 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1519 } else { // LONG_STOP or ONLY_LONG
1520 memcpy( saved
, buf
+ 512, 512 * sizeof(float));
1525 * Apply dependent channel coupling (applied before IMDCT).
1527 * @param index index into coupling gain array
1529 static void apply_dependent_coupling(AACContext
*ac
,
1530 SingleChannelElement
*target
,
1531 ChannelElement
*cce
, int index
)
1533 IndividualChannelStream
*ics
= &cce
->ch
[0].ics
;
1534 const uint16_t *offsets
= ics
->swb_offset
;
1535 float *dest
= target
->coeffs
;
1536 const float *src
= cce
->ch
[0].coeffs
;
1537 int g
, i
, group
, k
, idx
= 0;
1538 if (ac
->m4ac
.object_type
== AOT_AAC_LTP
) {
1539 av_log(ac
->avccontext
, AV_LOG_ERROR
,
1540 "Dependent coupling is not supported together with LTP\n");
1543 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1544 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1545 if (cce
->ch
[0].band_type
[idx
] != ZERO_BT
) {
1546 const float gain
= cce
->coup
.gain
[index
][idx
];
1547 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1548 for (k
= offsets
[i
]; k
< offsets
[i
+ 1]; k
++) {
1550 dest
[group
* 128 + k
] += gain
* src
[group
* 128 + k
];
1555 dest
+= ics
->group_len
[g
] * 128;
1556 src
+= ics
->group_len
[g
] * 128;
1561 * Apply independent channel coupling (applied after IMDCT).
1563 * @param index index into coupling gain array
1565 static void apply_independent_coupling(AACContext
*ac
,
1566 SingleChannelElement
*target
,
1567 ChannelElement
*cce
, int index
)
1570 const float gain
= cce
->coup
.gain
[index
][0];
1571 const float bias
= ac
->add_bias
;
1572 const float *src
= cce
->ch
[0].ret
;
1573 float *dest
= target
->ret
;
1575 for (i
= 0; i
< 1024; i
++)
1576 dest
[i
] += gain
* (src
[i
] - bias
);
1580 * channel coupling transformation interface
1582 * @param index index into coupling gain array
1583 * @param apply_coupling_method pointer to (in)dependent coupling function
1585 static void apply_channel_coupling(AACContext
*ac
, ChannelElement
*cc
,
1586 enum RawDataBlockType type
, int elem_id
,
1587 enum CouplingPoint coupling_point
,
1588 void (*apply_coupling_method
)(AACContext
*ac
, SingleChannelElement
*target
, ChannelElement
*cce
, int index
))
1592 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1593 ChannelElement
*cce
= ac
->che
[TYPE_CCE
][i
];
1596 if (cce
&& cce
->coup
.coupling_point
== coupling_point
) {
1597 ChannelCoupling
*coup
= &cce
->coup
;
1599 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1600 if (coup
->type
[c
] == type
&& coup
->id_select
[c
] == elem_id
) {
1601 if (coup
->ch_select
[c
] != 1) {
1602 apply_coupling_method(ac
, &cc
->ch
[0], cce
, index
);
1603 if (coup
->ch_select
[c
] != 0)
1606 if (coup
->ch_select
[c
] != 2)
1607 apply_coupling_method(ac
, &cc
->ch
[1], cce
, index
++);
1609 index
+= 1 + (coup
->ch_select
[c
] == 3);
1616 * Convert spectral data to float samples, applying all supported tools as appropriate.
1618 static void spectral_to_sample(AACContext
*ac
)
1621 for (type
= 3; type
>= 0; type
--) {
1622 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1623 ChannelElement
*che
= ac
->che
[type
][i
];
1625 if (type
<= TYPE_CPE
)
1626 apply_channel_coupling(ac
, che
, type
, i
, BEFORE_TNS
, apply_dependent_coupling
);
1627 if (che
->ch
[0].tns
.present
)
1628 apply_tns(che
->ch
[0].coeffs
, &che
->ch
[0].tns
, &che
->ch
[0].ics
, 1);
1629 if (che
->ch
[1].tns
.present
)
1630 apply_tns(che
->ch
[1].coeffs
, &che
->ch
[1].tns
, &che
->ch
[1].ics
, 1);
1631 if (type
<= TYPE_CPE
)
1632 apply_channel_coupling(ac
, che
, type
, i
, BETWEEN_TNS_AND_IMDCT
, apply_dependent_coupling
);
1633 if (type
!= TYPE_CCE
|| che
->coup
.coupling_point
== AFTER_IMDCT
)
1634 imdct_and_windowing(ac
, &che
->ch
[0]);
1635 if (type
== TYPE_CPE
)
1636 imdct_and_windowing(ac
, &che
->ch
[1]);
1637 if (type
<= TYPE_CCE
)
1638 apply_channel_coupling(ac
, che
, type
, i
, AFTER_IMDCT
, apply_independent_coupling
);
1644 static int parse_adts_frame_header(AACContext
*ac
, GetBitContext
*gb
)
1647 AACADTSHeaderInfo hdr_info
;
1649 size
= ff_aac_parse_header(gb
, &hdr_info
);
1651 if (!ac
->output_configured
&& hdr_info
.chan_config
) {
1652 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
1653 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
1654 ac
->m4ac
.chan_config
= hdr_info
.chan_config
;
1655 if (set_default_channel_config(ac
, new_che_pos
, hdr_info
.chan_config
))
1657 if (output_configure(ac
, ac
->che_pos
, new_che_pos
, 1))
1660 ac
->m4ac
.sample_rate
= hdr_info
.sample_rate
;
1661 ac
->m4ac
.sampling_index
= hdr_info
.sampling_index
;
1662 ac
->m4ac
.object_type
= hdr_info
.object_type
;
1663 if (hdr_info
.num_aac_frames
== 1) {
1664 if (!hdr_info
.crc_absent
)
1667 av_log_missing_feature(ac
->avccontext
, "More than one AAC RDB per ADTS frame is", 0);
1674 static int aac_decode_frame(AVCodecContext
*avccontext
, void *data
,
1675 int *data_size
, AVPacket
*avpkt
)
1677 const uint8_t *buf
= avpkt
->data
;
1678 int buf_size
= avpkt
->size
;
1679 AACContext
*ac
= avccontext
->priv_data
;
1680 ChannelElement
*che
= NULL
;
1682 enum RawDataBlockType elem_type
;
1683 int err
, elem_id
, data_size_tmp
;
1685 init_get_bits(&gb
, buf
, buf_size
* 8);
1687 if (show_bits(&gb
, 12) == 0xfff) {
1688 if (parse_adts_frame_header(ac
, &gb
) < 0) {
1689 av_log(avccontext
, AV_LOG_ERROR
, "Error decoding AAC frame header.\n");
1692 if (ac
->m4ac
.sampling_index
> 12) {
1693 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
1699 while ((elem_type
= get_bits(&gb
, 3)) != TYPE_END
) {
1700 elem_id
= get_bits(&gb
, 4);
1702 if (elem_type
< TYPE_DSE
&& !(che
=get_che(ac
, elem_type
, elem_id
))) {
1703 av_log(ac
->avccontext
, AV_LOG_ERROR
, "channel element %d.%d is not allocated\n", elem_type
, elem_id
);
1707 switch (elem_type
) {
1710 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1714 err
= decode_cpe(ac
, &gb
, che
);
1718 err
= decode_cce(ac
, &gb
, che
);
1722 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1726 skip_data_stream_element(&gb
);
1731 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
1732 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
1733 if ((err
= decode_pce(ac
, new_che_pos
, &gb
)))
1735 if (ac
->output_configured
)
1736 av_log(avccontext
, AV_LOG_ERROR
,
1737 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1739 err
= output_configure(ac
, ac
->che_pos
, new_che_pos
, 0);
1745 elem_id
+= get_bits(&gb
, 8) - 1;
1747 elem_id
-= decode_extension_payload(ac
, &gb
, elem_id
);
1748 err
= 0; /* FIXME */
1752 err
= -1; /* should not happen, but keeps compiler happy */
1760 spectral_to_sample(ac
);
1762 if (!ac
->is_saved
) {
1768 data_size_tmp
= 1024 * avccontext
->channels
* sizeof(int16_t);
1769 if (*data_size
< data_size_tmp
) {
1770 av_log(avccontext
, AV_LOG_ERROR
,
1771 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1772 *data_size
, data_size_tmp
);
1775 *data_size
= data_size_tmp
;
1777 ac
->dsp
.float_to_int16_interleave(data
, (const float **)ac
->output_data
, 1024, avccontext
->channels
);
1782 static av_cold
int aac_decode_close(AVCodecContext
*avccontext
)
1784 AACContext
*ac
= avccontext
->priv_data
;
1787 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1788 for (type
= 0; type
< 4; type
++)
1789 av_freep(&ac
->che
[type
][i
]);
1792 ff_mdct_end(&ac
->mdct
);
1793 ff_mdct_end(&ac
->mdct_small
);
1797 AVCodec aac_decoder
= {
1806 .long_name
= NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1807 .sample_fmts
= (enum SampleFormat
[]) {
1808 SAMPLE_FMT_S16
,SAMPLE_FMT_NONE