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"
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 memcpy(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
167 /* Allocate or free elements depending on if they are in the
168 * current program configuration.
170 * Set up default 1:1 output mapping.
172 * For a 5.1 stream the output order will be:
173 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
176 for(i
= 0; i
< MAX_ELEM_ID
; i
++) {
177 for(type
= 0; type
< 4; type
++) {
178 if(che_pos
[type
][i
]) {
179 if(!ac
->che
[type
][i
] && !(ac
->che
[type
][i
] = av_mallocz(sizeof(ChannelElement
))))
180 return AVERROR(ENOMEM
);
181 if(type
!= TYPE_CCE
) {
182 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[0].ret
;
183 if(type
== TYPE_CPE
) {
184 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[1].ret
;
188 av_freep(&ac
->che
[type
][i
]);
192 if (channel_config
) {
193 memset(ac
->tag_che_map
, 0, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
196 memcpy(ac
->tag_che_map
, ac
->che
, 4 * MAX_ELEM_ID
* sizeof(ac
->che
[0][0]));
197 ac
->tags_mapped
= 4*MAX_ELEM_ID
;
200 avctx
->channels
= channels
;
206 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
208 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
209 * @param sce_map mono (Single Channel Element) map
210 * @param type speaker type/position for these channels
212 static void decode_channel_map(enum ChannelPosition
*cpe_map
,
213 enum ChannelPosition
*sce_map
, enum ChannelPosition type
, GetBitContext
* gb
, int n
) {
215 enum ChannelPosition
*map
= cpe_map
&& get_bits1(gb
) ? cpe_map
: sce_map
; // stereo or mono map
216 map
[get_bits(gb
, 4)] = type
;
221 * Decode program configuration element; reference: table 4.2.
223 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
225 * @return Returns error status. 0 - OK, !0 - error
227 static int decode_pce(AACContext
* ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
228 GetBitContext
* gb
) {
229 int num_front
, num_side
, num_back
, num_lfe
, num_assoc_data
, num_cc
, sampling_index
;
231 skip_bits(gb
, 2); // object_type
233 sampling_index
= get_bits(gb
, 4);
234 if (ac
->m4ac
.sampling_index
!= sampling_index
)
235 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");
237 num_front
= get_bits(gb
, 4);
238 num_side
= get_bits(gb
, 4);
239 num_back
= get_bits(gb
, 4);
240 num_lfe
= get_bits(gb
, 2);
241 num_assoc_data
= get_bits(gb
, 3);
242 num_cc
= get_bits(gb
, 4);
245 skip_bits(gb
, 4); // mono_mixdown_tag
247 skip_bits(gb
, 4); // stereo_mixdown_tag
250 skip_bits(gb
, 3); // mixdown_coeff_index and pseudo_surround
252 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_FRONT
, gb
, num_front
);
253 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_SIDE
, gb
, num_side
);
254 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_BACK
, gb
, num_back
);
255 decode_channel_map(NULL
, new_che_pos
[TYPE_LFE
], AAC_CHANNEL_LFE
, gb
, num_lfe
);
257 skip_bits_long(gb
, 4 * num_assoc_data
);
259 decode_channel_map(new_che_pos
[TYPE_CCE
], new_che_pos
[TYPE_CCE
], AAC_CHANNEL_CC
, gb
, num_cc
);
263 /* comment field, first byte is length */
264 skip_bits_long(gb
, 8 * get_bits(gb
, 8));
269 * Set up channel positions based on a default channel configuration
270 * as specified in table 1.17.
272 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
274 * @return Returns error status. 0 - OK, !0 - error
276 static int set_default_channel_config(AACContext
*ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
279 if(channel_config
< 1 || channel_config
> 7) {
280 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid default channel configuration (%d)\n",
285 /* default channel configurations:
287 * 1ch : front center (mono)
288 * 2ch : L + R (stereo)
289 * 3ch : front center + L + R
290 * 4ch : front center + L + R + back center
291 * 5ch : front center + L + R + back stereo
292 * 6ch : front center + L + R + back stereo + LFE
293 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
296 if(channel_config
!= 2)
297 new_che_pos
[TYPE_SCE
][0] = AAC_CHANNEL_FRONT
; // front center (or mono)
298 if(channel_config
> 1)
299 new_che_pos
[TYPE_CPE
][0] = AAC_CHANNEL_FRONT
; // L + R (or stereo)
300 if(channel_config
== 4)
301 new_che_pos
[TYPE_SCE
][1] = AAC_CHANNEL_BACK
; // back center
302 if(channel_config
> 4)
303 new_che_pos
[TYPE_CPE
][(channel_config
== 7) + 1]
304 = AAC_CHANNEL_BACK
; // back stereo
305 if(channel_config
> 5)
306 new_che_pos
[TYPE_LFE
][0] = AAC_CHANNEL_LFE
; // LFE
307 if(channel_config
== 7)
308 new_che_pos
[TYPE_CPE
][1] = AAC_CHANNEL_FRONT
; // outer front left + outer front right
314 * Decode GA "General Audio" specific configuration; reference: table 4.1.
316 * @return Returns error status. 0 - OK, !0 - error
318 static int decode_ga_specific_config(AACContext
* ac
, GetBitContext
* gb
, int channel_config
) {
319 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
320 int extension_flag
, ret
;
322 if(get_bits1(gb
)) { // frameLengthFlag
323 ff_log_missing_feature(ac
->avccontext
, "960/120 MDCT window is", 1);
327 if (get_bits1(gb
)) // dependsOnCoreCoder
328 skip_bits(gb
, 14); // coreCoderDelay
329 extension_flag
= get_bits1(gb
);
331 if(ac
->m4ac
.object_type
== AOT_AAC_SCALABLE
||
332 ac
->m4ac
.object_type
== AOT_ER_AAC_SCALABLE
)
333 skip_bits(gb
, 3); // layerNr
335 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
336 if (channel_config
== 0) {
337 skip_bits(gb
, 4); // element_instance_tag
338 if((ret
= decode_pce(ac
, new_che_pos
, gb
)))
341 if((ret
= set_default_channel_config(ac
, new_che_pos
, channel_config
)))
344 if((ret
= output_configure(ac
, ac
->che_pos
, new_che_pos
, channel_config
)))
347 if (extension_flag
) {
348 switch (ac
->m4ac
.object_type
) {
350 skip_bits(gb
, 5); // numOfSubFrame
351 skip_bits(gb
, 11); // layer_length
355 case AOT_ER_AAC_SCALABLE
:
357 skip_bits(gb
, 3); /* aacSectionDataResilienceFlag
358 * aacScalefactorDataResilienceFlag
359 * aacSpectralDataResilienceFlag
363 skip_bits1(gb
); // extensionFlag3 (TBD in version 3)
369 * Decode audio specific configuration; reference: table 1.13.
371 * @param data pointer to AVCodecContext extradata
372 * @param data_size size of AVCCodecContext extradata
374 * @return Returns error status. 0 - OK, !0 - error
376 static int decode_audio_specific_config(AACContext
* ac
, void *data
, int data_size
) {
380 init_get_bits(&gb
, data
, data_size
* 8);
382 if((i
= ff_mpeg4audio_get_config(&ac
->m4ac
, data
, data_size
)) < 0)
384 if(ac
->m4ac
.sampling_index
> 12) {
385 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
389 skip_bits_long(&gb
, i
);
391 switch (ac
->m4ac
.object_type
) {
394 if (decode_ga_specific_config(ac
, &gb
, ac
->m4ac
.chan_config
))
398 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Audio object type %s%d is not supported.\n",
399 ac
->m4ac
.sbr
== 1? "SBR+" : "", ac
->m4ac
.object_type
);
406 * linear congruential pseudorandom number generator
408 * @param previous_val pointer to the current state of the generator
410 * @return Returns a 32-bit pseudorandom integer
412 static av_always_inline
int lcg_random(int previous_val
) {
413 return previous_val
* 1664525 + 1013904223;
416 static void reset_predict_state(PredictorState
* ps
) {
425 static void reset_all_predictors(PredictorState
* ps
) {
427 for (i
= 0; i
< MAX_PREDICTORS
; i
++)
428 reset_predict_state(&ps
[i
]);
431 static void reset_predictor_group(PredictorState
* ps
, int group_num
) {
433 for (i
= group_num
-1; i
< MAX_PREDICTORS
; i
+=30)
434 reset_predict_state(&ps
[i
]);
437 static av_cold
int aac_decode_init(AVCodecContext
* avccontext
) {
438 AACContext
* ac
= avccontext
->priv_data
;
441 ac
->avccontext
= avccontext
;
443 if (avccontext
->extradata_size
> 0) {
444 if(decode_audio_specific_config(ac
, avccontext
->extradata
, avccontext
->extradata_size
))
446 avccontext
->sample_rate
= ac
->m4ac
.sample_rate
;
447 } else if (avccontext
->channels
> 0) {
448 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
449 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
450 if(set_default_channel_config(ac
, new_che_pos
, avccontext
->channels
- (avccontext
->channels
== 8)))
452 if(output_configure(ac
, ac
->che_pos
, new_che_pos
, 1))
454 ac
->m4ac
.sample_rate
= avccontext
->sample_rate
;
457 avccontext
->sample_fmt
= SAMPLE_FMT_S16
;
458 avccontext
->frame_size
= 1024;
460 AAC_INIT_VLC_STATIC( 0, 144);
461 AAC_INIT_VLC_STATIC( 1, 114);
462 AAC_INIT_VLC_STATIC( 2, 188);
463 AAC_INIT_VLC_STATIC( 3, 180);
464 AAC_INIT_VLC_STATIC( 4, 172);
465 AAC_INIT_VLC_STATIC( 5, 140);
466 AAC_INIT_VLC_STATIC( 6, 168);
467 AAC_INIT_VLC_STATIC( 7, 114);
468 AAC_INIT_VLC_STATIC( 8, 262);
469 AAC_INIT_VLC_STATIC( 9, 248);
470 AAC_INIT_VLC_STATIC(10, 384);
472 dsputil_init(&ac
->dsp
, avccontext
);
474 ac
->random_state
= 0x1f2e3d4c;
476 // -1024 - Compensate wrong IMDCT method.
477 // 32768 - Required to scale values to the correct range for the bias method
478 // for float to int16 conversion.
480 if(ac
->dsp
.float_to_int16
== ff_float_to_int16_c
) {
481 ac
->add_bias
= 385.0f
;
482 ac
->sf_scale
= 1. / (-1024. * 32768.);
486 ac
->sf_scale
= 1. / -1024.;
490 #if !CONFIG_HARDCODED_TABLES
491 for (i
= 0; i
< 428; i
++)
492 ff_aac_pow2sf_tab
[i
] = pow(2, (i
- 200)/4.);
493 #endif /* CONFIG_HARDCODED_TABLES */
495 INIT_VLC_STATIC(&vlc_scalefactors
,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code
),
496 ff_aac_scalefactor_bits
, sizeof(ff_aac_scalefactor_bits
[0]), sizeof(ff_aac_scalefactor_bits
[0]),
497 ff_aac_scalefactor_code
, sizeof(ff_aac_scalefactor_code
[0]), sizeof(ff_aac_scalefactor_code
[0]),
500 ff_mdct_init(&ac
->mdct
, 11, 1, 1.0);
501 ff_mdct_init(&ac
->mdct_small
, 8, 1, 1.0);
502 // window initialization
503 ff_kbd_window_init(ff_aac_kbd_long_1024
, 4.0, 1024);
504 ff_kbd_window_init(ff_aac_kbd_short_128
, 6.0, 128);
505 ff_sine_window_init(ff_sine_1024
, 1024);
506 ff_sine_window_init(ff_sine_128
, 128);
512 * Skip data_stream_element; reference: table 4.10.
514 static void skip_data_stream_element(GetBitContext
* gb
) {
515 int byte_align
= get_bits1(gb
);
516 int count
= get_bits(gb
, 8);
518 count
+= get_bits(gb
, 8);
521 skip_bits_long(gb
, 8 * count
);
524 static int decode_prediction(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
) {
527 ics
->predictor_reset_group
= get_bits(gb
, 5);
528 if (ics
->predictor_reset_group
== 0 || ics
->predictor_reset_group
> 30) {
529 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Invalid Predictor Reset Group.\n");
533 for (sfb
= 0; sfb
< FFMIN(ics
->max_sfb
, ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]); sfb
++) {
534 ics
->prediction_used
[sfb
] = get_bits1(gb
);
540 * Decode Individual Channel Stream info; reference: table 4.6.
542 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
544 static int decode_ics_info(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
, int common_window
) {
546 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Reserved bit set.\n");
547 memset(ics
, 0, sizeof(IndividualChannelStream
));
550 ics
->window_sequence
[1] = ics
->window_sequence
[0];
551 ics
->window_sequence
[0] = get_bits(gb
, 2);
552 ics
->use_kb_window
[1] = ics
->use_kb_window
[0];
553 ics
->use_kb_window
[0] = get_bits1(gb
);
554 ics
->num_window_groups
= 1;
555 ics
->group_len
[0] = 1;
556 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
558 ics
->max_sfb
= get_bits(gb
, 4);
559 for (i
= 0; i
< 7; i
++) {
561 ics
->group_len
[ics
->num_window_groups
-1]++;
563 ics
->num_window_groups
++;
564 ics
->group_len
[ics
->num_window_groups
-1] = 1;
567 ics
->num_windows
= 8;
568 ics
->swb_offset
= ff_swb_offset_128
[ac
->m4ac
.sampling_index
];
569 ics
->num_swb
= ff_aac_num_swb_128
[ac
->m4ac
.sampling_index
];
570 ics
->tns_max_bands
= ff_tns_max_bands_128
[ac
->m4ac
.sampling_index
];
571 ics
->predictor_present
= 0;
573 ics
->max_sfb
= get_bits(gb
, 6);
574 ics
->num_windows
= 1;
575 ics
->swb_offset
= ff_swb_offset_1024
[ac
->m4ac
.sampling_index
];
576 ics
->num_swb
= ff_aac_num_swb_1024
[ac
->m4ac
.sampling_index
];
577 ics
->tns_max_bands
= ff_tns_max_bands_1024
[ac
->m4ac
.sampling_index
];
578 ics
->predictor_present
= get_bits1(gb
);
579 ics
->predictor_reset_group
= 0;
580 if (ics
->predictor_present
) {
581 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
582 if (decode_prediction(ac
, ics
, gb
)) {
583 memset(ics
, 0, sizeof(IndividualChannelStream
));
586 } else if (ac
->m4ac
.object_type
== AOT_AAC_LC
) {
587 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Prediction is not allowed in AAC-LC.\n");
588 memset(ics
, 0, sizeof(IndividualChannelStream
));
591 ff_log_missing_feature(ac
->avccontext
, "Predictor bit set but LTP is", 1);
592 memset(ics
, 0, sizeof(IndividualChannelStream
));
598 if(ics
->max_sfb
> ics
->num_swb
) {
599 av_log(ac
->avccontext
, AV_LOG_ERROR
,
600 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
601 ics
->max_sfb
, ics
->num_swb
);
602 memset(ics
, 0, sizeof(IndividualChannelStream
));
610 * Decode band types (section_data payload); reference: table 4.46.
612 * @param band_type array of the used band type
613 * @param band_type_run_end array of the last scalefactor band of a band type run
615 * @return Returns error status. 0 - OK, !0 - error
617 static int decode_band_types(AACContext
* ac
, enum BandType band_type
[120],
618 int band_type_run_end
[120], GetBitContext
* gb
, IndividualChannelStream
* ics
) {
620 const int bits
= (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) ? 3 : 5;
621 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
623 while (k
< ics
->max_sfb
) {
624 uint8_t sect_len
= k
;
626 int sect_band_type
= get_bits(gb
, 4);
627 if (sect_band_type
== 12) {
628 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid band type\n");
631 while ((sect_len_incr
= get_bits(gb
, bits
)) == (1 << bits
)-1)
632 sect_len
+= sect_len_incr
;
633 sect_len
+= sect_len_incr
;
634 if (sect_len
> ics
->max_sfb
) {
635 av_log(ac
->avccontext
, AV_LOG_ERROR
,
636 "Number of bands (%d) exceeds limit (%d).\n",
637 sect_len
, ics
->max_sfb
);
640 for (; k
< sect_len
; k
++) {
641 band_type
[idx
] = sect_band_type
;
642 band_type_run_end
[idx
++] = sect_len
;
650 * Decode scalefactors; reference: table 4.47.
652 * @param global_gain first scalefactor value as scalefactors are differentially coded
653 * @param band_type array of the used band type
654 * @param band_type_run_end array of the last scalefactor band of a band type run
655 * @param sf array of scalefactors or intensity stereo positions
657 * @return Returns error status. 0 - OK, !0 - error
659 static int decode_scalefactors(AACContext
* ac
, float sf
[120], GetBitContext
* gb
,
660 unsigned int global_gain
, IndividualChannelStream
* ics
,
661 enum BandType band_type
[120], int band_type_run_end
[120]) {
662 const int sf_offset
= ac
->sf_offset
+ (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
? 12 : 0);
664 int offset
[3] = { global_gain
, global_gain
- 90, 100 };
666 static const char *sf_str
[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
667 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
668 for (i
= 0; i
< ics
->max_sfb
;) {
669 int run_end
= band_type_run_end
[idx
];
670 if (band_type
[idx
] == ZERO_BT
) {
671 for(; i
< run_end
; i
++, idx
++)
673 }else if((band_type
[idx
] == INTENSITY_BT
) || (band_type
[idx
] == INTENSITY_BT2
)) {
674 for(; i
< run_end
; i
++, idx
++) {
675 offset
[2] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
676 if(offset
[2] > 255U) {
677 av_log(ac
->avccontext
, AV_LOG_ERROR
,
678 "%s (%d) out of range.\n", sf_str
[2], offset
[2]);
681 sf
[idx
] = ff_aac_pow2sf_tab
[-offset
[2] + 300];
683 }else if(band_type
[idx
] == NOISE_BT
) {
684 for(; i
< run_end
; i
++, idx
++) {
686 offset
[1] += get_bits(gb
, 9) - 256;
688 offset
[1] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
689 if(offset
[1] > 255U) {
690 av_log(ac
->avccontext
, AV_LOG_ERROR
,
691 "%s (%d) out of range.\n", sf_str
[1], offset
[1]);
694 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[1] + sf_offset
+ 100];
697 for(; i
< run_end
; i
++, idx
++) {
698 offset
[0] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
699 if(offset
[0] > 255U) {
700 av_log(ac
->avccontext
, AV_LOG_ERROR
,
701 "%s (%d) out of range.\n", sf_str
[0], offset
[0]);
704 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[0] + sf_offset
];
713 * Decode pulse data; reference: table 4.7.
715 static int decode_pulses(Pulse
* pulse
, GetBitContext
* gb
, const uint16_t * swb_offset
, int num_swb
) {
717 pulse
->num_pulse
= get_bits(gb
, 2) + 1;
718 pulse_swb
= get_bits(gb
, 6);
719 if (pulse_swb
>= num_swb
)
721 pulse
->pos
[0] = swb_offset
[pulse_swb
];
722 pulse
->pos
[0] += get_bits(gb
, 5);
723 if (pulse
->pos
[0] > 1023)
725 pulse
->amp
[0] = get_bits(gb
, 4);
726 for (i
= 1; i
< pulse
->num_pulse
; i
++) {
727 pulse
->pos
[i
] = get_bits(gb
, 5) + pulse
->pos
[i
-1];
728 if (pulse
->pos
[i
] > 1023)
730 pulse
->amp
[i
] = get_bits(gb
, 4);
736 * Decode Temporal Noise Shaping data; reference: table 4.48.
738 * @return Returns error status. 0 - OK, !0 - error
740 static int decode_tns(AACContext
* ac
, TemporalNoiseShaping
* tns
,
741 GetBitContext
* gb
, const IndividualChannelStream
* ics
) {
742 int w
, filt
, i
, coef_len
, coef_res
, coef_compress
;
743 const int is8
= ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
;
744 const int tns_max_order
= is8
? 7 : ac
->m4ac
.object_type
== AOT_AAC_MAIN
? 20 : 12;
745 for (w
= 0; w
< ics
->num_windows
; w
++) {
746 if ((tns
->n_filt
[w
] = get_bits(gb
, 2 - is8
))) {
747 coef_res
= get_bits1(gb
);
749 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
751 tns
->length
[w
][filt
] = get_bits(gb
, 6 - 2*is8
);
753 if ((tns
->order
[w
][filt
] = get_bits(gb
, 5 - 2*is8
)) > tns_max_order
) {
754 av_log(ac
->avccontext
, AV_LOG_ERROR
, "TNS filter order %d is greater than maximum %d.",
755 tns
->order
[w
][filt
], tns_max_order
);
756 tns
->order
[w
][filt
] = 0;
759 if (tns
->order
[w
][filt
]) {
760 tns
->direction
[w
][filt
] = get_bits1(gb
);
761 coef_compress
= get_bits1(gb
);
762 coef_len
= coef_res
+ 3 - coef_compress
;
763 tmp2_idx
= 2*coef_compress
+ coef_res
;
765 for (i
= 0; i
< tns
->order
[w
][filt
]; i
++)
766 tns
->coef
[w
][filt
][i
] = tns_tmp2_map
[tmp2_idx
][get_bits(gb
, coef_len
)];
775 * Decode Mid/Side data; reference: table 4.54.
777 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
778 * [1] mask is decoded from bitstream; [2] mask is all 1s;
779 * [3] reserved for scalable AAC
781 static void decode_mid_side_stereo(ChannelElement
* cpe
, GetBitContext
* gb
,
784 if (ms_present
== 1) {
785 for (idx
= 0; idx
< cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
; idx
++)
786 cpe
->ms_mask
[idx
] = get_bits1(gb
);
787 } else if (ms_present
== 2) {
788 memset(cpe
->ms_mask
, 1, cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
* sizeof(cpe
->ms_mask
[0]));
793 * Decode spectral data; reference: table 4.50.
794 * Dequantize and scale spectral data; reference: 4.6.3.3.
796 * @param coef array of dequantized, scaled spectral data
797 * @param sf array of scalefactors or intensity stereo positions
798 * @param pulse_present set if pulses are present
799 * @param pulse pointer to pulse data struct
800 * @param band_type array of the used band type
802 * @return Returns error status. 0 - OK, !0 - error
804 static int decode_spectrum_and_dequant(AACContext
* ac
, float coef
[1024], GetBitContext
* gb
, float sf
[120],
805 int pulse_present
, const Pulse
* pulse
, const IndividualChannelStream
* ics
, enum BandType band_type
[120]) {
806 int i
, k
, g
, idx
= 0;
807 const int c
= 1024/ics
->num_windows
;
808 const uint16_t * offsets
= ics
->swb_offset
;
809 float *coef_base
= coef
;
810 static const float sign_lookup
[] = { 1.0f
, -1.0f
};
812 for (g
= 0; g
< ics
->num_windows
; g
++)
813 memset(coef
+ g
* 128 + offsets
[ics
->max_sfb
], 0, sizeof(float)*(c
- offsets
[ics
->max_sfb
]));
815 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
816 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
817 const int cur_band_type
= band_type
[idx
];
818 const int dim
= cur_band_type
>= FIRST_PAIR_BT
? 2 : 4;
819 const int is_cb_unsigned
= IS_CODEBOOK_UNSIGNED(cur_band_type
);
821 if (cur_band_type
== ZERO_BT
|| cur_band_type
== INTENSITY_BT2
|| cur_band_type
== INTENSITY_BT
) {
822 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
823 memset(coef
+ group
* 128 + offsets
[i
], 0, (offsets
[i
+1] - offsets
[i
])*sizeof(float));
825 }else if (cur_band_type
== NOISE_BT
) {
826 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
828 float band_energy
= 0;
829 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
830 ac
->random_state
= lcg_random(ac
->random_state
);
831 coef
[group
*128+k
] = ac
->random_state
;
832 band_energy
+= coef
[group
*128+k
]*coef
[group
*128+k
];
834 scale
= sf
[idx
] / sqrtf(band_energy
);
835 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
836 coef
[group
*128+k
] *= scale
;
840 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
841 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
+= dim
) {
842 const int index
= get_vlc2(gb
, vlc_spectral
[cur_band_type
- 1].table
, 6, 3);
843 const int coef_tmp_idx
= (group
<< 7) + k
;
846 if(index
>= ff_aac_spectral_sizes
[cur_band_type
- 1]) {
847 av_log(ac
->avccontext
, AV_LOG_ERROR
,
848 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
849 cur_band_type
- 1, index
, ff_aac_spectral_sizes
[cur_band_type
- 1]);
852 vq_ptr
= &ff_aac_codebook_vectors
[cur_band_type
- 1][index
* dim
];
853 if (is_cb_unsigned
) {
854 if (vq_ptr
[0]) coef
[coef_tmp_idx
] = sign_lookup
[get_bits1(gb
)];
855 if (vq_ptr
[1]) coef
[coef_tmp_idx
+ 1] = sign_lookup
[get_bits1(gb
)];
857 if (vq_ptr
[2]) coef
[coef_tmp_idx
+ 2] = sign_lookup
[get_bits1(gb
)];
858 if (vq_ptr
[3]) coef
[coef_tmp_idx
+ 3] = sign_lookup
[get_bits1(gb
)];
860 if (cur_band_type
== ESC_BT
) {
861 for (j
= 0; j
< 2; j
++) {
862 if (vq_ptr
[j
] == 64.0f
) {
864 /* The total length of escape_sequence must be < 22 bits according
865 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
866 while (get_bits1(gb
) && n
< 15) n
++;
868 av_log(ac
->avccontext
, AV_LOG_ERROR
, "error in spectral data, ESC overflow\n");
871 n
= (1<<n
) + get_bits(gb
, n
);
872 coef
[coef_tmp_idx
+ j
] *= cbrtf(n
) * n
;
874 coef
[coef_tmp_idx
+ j
] *= vq_ptr
[j
];
878 coef
[coef_tmp_idx
] *= vq_ptr
[0];
879 coef
[coef_tmp_idx
+ 1] *= vq_ptr
[1];
881 coef
[coef_tmp_idx
+ 2] *= vq_ptr
[2];
882 coef
[coef_tmp_idx
+ 3] *= vq_ptr
[3];
886 coef
[coef_tmp_idx
] = vq_ptr
[0];
887 coef
[coef_tmp_idx
+ 1] = vq_ptr
[1];
889 coef
[coef_tmp_idx
+ 2] = vq_ptr
[2];
890 coef
[coef_tmp_idx
+ 3] = vq_ptr
[3];
893 coef
[coef_tmp_idx
] *= sf
[idx
];
894 coef
[coef_tmp_idx
+ 1] *= sf
[idx
];
896 coef
[coef_tmp_idx
+ 2] *= sf
[idx
];
897 coef
[coef_tmp_idx
+ 3] *= sf
[idx
];
903 coef
+= ics
->group_len
[g
]<<7;
908 for(i
= 0; i
< pulse
->num_pulse
; i
++){
909 float co
= coef_base
[ pulse
->pos
[i
] ];
910 while(offsets
[idx
+ 1] <= pulse
->pos
[i
])
912 if (band_type
[idx
] != NOISE_BT
&& sf
[idx
]) {
913 float ico
= -pulse
->amp
[i
];
916 ico
= co
/ sqrtf(sqrtf(fabsf(co
))) + (co
> 0 ? -ico
: ico
);
918 coef_base
[ pulse
->pos
[i
] ] = cbrtf(fabsf(ico
)) * ico
* sf
[idx
];
925 static av_always_inline
float flt16_round(float pf
) {
928 tmp
.i
= (tmp
.i
+ 0x00008000U
) & 0xFFFF0000U
;
932 static av_always_inline
float flt16_even(float pf
) {
935 tmp
.i
= (tmp
.i
+ 0x00007FFFU
+ (tmp
.i
& 0x00010000U
>>16)) & 0xFFFF0000U
;
939 static av_always_inline
float flt16_trunc(float pf
) {
942 pun
.i
&= 0xFFFF0000U
;
946 static void predict(AACContext
* ac
, PredictorState
* ps
, float* coef
, int output_enable
) {
947 const float a
= 0.953125; // 61.0/64
948 const float alpha
= 0.90625; // 29.0/32
953 k1
= ps
->var0
> 1 ? ps
->cor0
* flt16_even(a
/ ps
->var0
) : 0;
954 k2
= ps
->var1
> 1 ? ps
->cor1
* flt16_even(a
/ ps
->var1
) : 0;
956 pv
= flt16_round(k1
* ps
->r0
+ k2
* ps
->r1
);
958 *coef
+= pv
* ac
->sf_scale
;
960 e0
= *coef
/ ac
->sf_scale
;
961 e1
= e0
- k1
* ps
->r0
;
963 ps
->cor1
= flt16_trunc(alpha
* ps
->cor1
+ ps
->r1
* e1
);
964 ps
->var1
= flt16_trunc(alpha
* ps
->var1
+ 0.5 * (ps
->r1
* ps
->r1
+ e1
* e1
));
965 ps
->cor0
= flt16_trunc(alpha
* ps
->cor0
+ ps
->r0
* e0
);
966 ps
->var0
= flt16_trunc(alpha
* ps
->var0
+ 0.5 * (ps
->r0
* ps
->r0
+ e0
* e0
));
968 ps
->r1
= flt16_trunc(a
* (ps
->r0
- k1
* e0
));
969 ps
->r0
= flt16_trunc(a
* e0
);
973 * Apply AAC-Main style frequency domain prediction.
975 static void apply_prediction(AACContext
* ac
, SingleChannelElement
* sce
) {
978 if (!sce
->ics
.predictor_initialized
) {
979 reset_all_predictors(sce
->predictor_state
);
980 sce
->ics
.predictor_initialized
= 1;
983 if (sce
->ics
.window_sequence
[0] != EIGHT_SHORT_SEQUENCE
) {
984 for (sfb
= 0; sfb
< ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]; sfb
++) {
985 for (k
= sce
->ics
.swb_offset
[sfb
]; k
< sce
->ics
.swb_offset
[sfb
+ 1]; k
++) {
986 predict(ac
, &sce
->predictor_state
[k
], &sce
->coeffs
[k
],
987 sce
->ics
.predictor_present
&& sce
->ics
.prediction_used
[sfb
]);
990 if (sce
->ics
.predictor_reset_group
)
991 reset_predictor_group(sce
->predictor_state
, sce
->ics
.predictor_reset_group
);
993 reset_all_predictors(sce
->predictor_state
);
997 * Decode an individual_channel_stream payload; reference: table 4.44.
999 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1000 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1002 * @return Returns error status. 0 - OK, !0 - error
1004 static int decode_ics(AACContext
* ac
, SingleChannelElement
* sce
, GetBitContext
* gb
, int common_window
, int scale_flag
) {
1006 TemporalNoiseShaping
* tns
= &sce
->tns
;
1007 IndividualChannelStream
* ics
= &sce
->ics
;
1008 float * out
= sce
->coeffs
;
1009 int global_gain
, pulse_present
= 0;
1011 /* This assignment is to silence a GCC warning about the variable being used
1012 * uninitialized when in fact it always is.
1014 pulse
.num_pulse
= 0;
1016 global_gain
= get_bits(gb
, 8);
1018 if (!common_window
&& !scale_flag
) {
1019 if (decode_ics_info(ac
, ics
, gb
, 0) < 0)
1023 if (decode_band_types(ac
, sce
->band_type
, sce
->band_type_run_end
, gb
, ics
) < 0)
1025 if (decode_scalefactors(ac
, sce
->sf
, gb
, global_gain
, ics
, sce
->band_type
, sce
->band_type_run_end
) < 0)
1030 if ((pulse_present
= get_bits1(gb
))) {
1031 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1032 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse tool not allowed in eight short sequence.\n");
1035 if (decode_pulses(&pulse
, gb
, ics
->swb_offset
, ics
->num_swb
)) {
1036 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse data corrupt or invalid.\n");
1040 if ((tns
->present
= get_bits1(gb
)) && decode_tns(ac
, tns
, gb
, ics
))
1042 if (get_bits1(gb
)) {
1043 ff_log_missing_feature(ac
->avccontext
, "SSR", 1);
1048 if (decode_spectrum_and_dequant(ac
, out
, gb
, sce
->sf
, pulse_present
, &pulse
, ics
, sce
->band_type
) < 0)
1051 if(ac
->m4ac
.object_type
== AOT_AAC_MAIN
&& !common_window
)
1052 apply_prediction(ac
, sce
);
1058 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1060 static void apply_mid_side_stereo(ChannelElement
* cpe
) {
1061 const IndividualChannelStream
* ics
= &cpe
->ch
[0].ics
;
1062 float *ch0
= cpe
->ch
[0].coeffs
;
1063 float *ch1
= cpe
->ch
[1].coeffs
;
1064 int g
, i
, k
, group
, idx
= 0;
1065 const uint16_t * offsets
= ics
->swb_offset
;
1066 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1067 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1068 if (cpe
->ms_mask
[idx
] &&
1069 cpe
->ch
[0].band_type
[idx
] < NOISE_BT
&& cpe
->ch
[1].band_type
[idx
] < NOISE_BT
) {
1070 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1071 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
1072 float tmp
= ch0
[group
*128 + k
] - ch1
[group
*128 + k
];
1073 ch0
[group
*128 + k
] += ch1
[group
*128 + k
];
1074 ch1
[group
*128 + k
] = tmp
;
1079 ch0
+= ics
->group_len
[g
]*128;
1080 ch1
+= ics
->group_len
[g
]*128;
1085 * intensity stereo decoding; reference: 4.6.8.2.3
1087 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1088 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1089 * [3] reserved for scalable AAC
1091 static void apply_intensity_stereo(ChannelElement
* cpe
, int ms_present
) {
1092 const IndividualChannelStream
* ics
= &cpe
->ch
[1].ics
;
1093 SingleChannelElement
* sce1
= &cpe
->ch
[1];
1094 float *coef0
= cpe
->ch
[0].coeffs
, *coef1
= cpe
->ch
[1].coeffs
;
1095 const uint16_t * offsets
= ics
->swb_offset
;
1096 int g
, group
, i
, k
, idx
= 0;
1099 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1100 for (i
= 0; i
< ics
->max_sfb
;) {
1101 if (sce1
->band_type
[idx
] == INTENSITY_BT
|| sce1
->band_type
[idx
] == INTENSITY_BT2
) {
1102 const int bt_run_end
= sce1
->band_type_run_end
[idx
];
1103 for (; i
< bt_run_end
; i
++, idx
++) {
1104 c
= -1 + 2 * (sce1
->band_type
[idx
] - 14);
1106 c
*= 1 - 2 * cpe
->ms_mask
[idx
];
1107 scale
= c
* sce1
->sf
[idx
];
1108 for (group
= 0; group
< ics
->group_len
[g
]; group
++)
1109 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++)
1110 coef1
[group
*128 + k
] = scale
* coef0
[group
*128 + k
];
1113 int bt_run_end
= sce1
->band_type_run_end
[idx
];
1114 idx
+= bt_run_end
- i
;
1118 coef0
+= ics
->group_len
[g
]*128;
1119 coef1
+= ics
->group_len
[g
]*128;
1124 * Decode a channel_pair_element; reference: table 4.4.
1126 * @param elem_id Identifies the instance of a syntax element.
1128 * @return Returns error status. 0 - OK, !0 - error
1130 static int decode_cpe(AACContext
* ac
, GetBitContext
* gb
, ChannelElement
* cpe
) {
1131 int i
, ret
, common_window
, ms_present
= 0;
1133 common_window
= get_bits1(gb
);
1134 if (common_window
) {
1135 if (decode_ics_info(ac
, &cpe
->ch
[0].ics
, gb
, 1))
1137 i
= cpe
->ch
[1].ics
.use_kb_window
[0];
1138 cpe
->ch
[1].ics
= cpe
->ch
[0].ics
;
1139 cpe
->ch
[1].ics
.use_kb_window
[1] = i
;
1140 ms_present
= get_bits(gb
, 2);
1141 if(ms_present
== 3) {
1142 av_log(ac
->avccontext
, AV_LOG_ERROR
, "ms_present = 3 is reserved.\n");
1144 } else if(ms_present
)
1145 decode_mid_side_stereo(cpe
, gb
, ms_present
);
1147 if ((ret
= decode_ics(ac
, &cpe
->ch
[0], gb
, common_window
, 0)))
1149 if ((ret
= decode_ics(ac
, &cpe
->ch
[1], gb
, common_window
, 0)))
1152 if (common_window
) {
1154 apply_mid_side_stereo(cpe
);
1155 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
1156 apply_prediction(ac
, &cpe
->ch
[0]);
1157 apply_prediction(ac
, &cpe
->ch
[1]);
1161 apply_intensity_stereo(cpe
, ms_present
);
1166 * Decode coupling_channel_element; reference: table 4.8.
1168 * @param elem_id Identifies the instance of a syntax element.
1170 * @return Returns error status. 0 - OK, !0 - error
1172 static int decode_cce(AACContext
* ac
, GetBitContext
* gb
, ChannelElement
* che
) {
1177 SingleChannelElement
* sce
= &che
->ch
[0];
1178 ChannelCoupling
* coup
= &che
->coup
;
1180 coup
->coupling_point
= 2*get_bits1(gb
);
1181 coup
->num_coupled
= get_bits(gb
, 3);
1182 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1184 coup
->type
[c
] = get_bits1(gb
) ? TYPE_CPE
: TYPE_SCE
;
1185 coup
->id_select
[c
] = get_bits(gb
, 4);
1186 if (coup
->type
[c
] == TYPE_CPE
) {
1187 coup
->ch_select
[c
] = get_bits(gb
, 2);
1188 if (coup
->ch_select
[c
] == 3)
1191 coup
->ch_select
[c
] = 2;
1193 coup
->coupling_point
+= get_bits1(gb
) || (coup
->coupling_point
>>1);
1195 sign
= get_bits(gb
, 1);
1196 scale
= pow(2., pow(2., (int)get_bits(gb
, 2) - 3));
1198 if ((ret
= decode_ics(ac
, sce
, gb
, 0, 0)))
1201 for (c
= 0; c
< num_gain
; c
++) {
1205 float gain_cache
= 1.;
1207 cge
= coup
->coupling_point
== AFTER_IMDCT
? 1 : get_bits1(gb
);
1208 gain
= cge
? get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60: 0;
1209 gain_cache
= pow(scale
, -gain
);
1211 if (coup
->coupling_point
== AFTER_IMDCT
) {
1212 coup
->gain
[c
][0] = gain_cache
;
1214 for (g
= 0; g
< sce
->ics
.num_window_groups
; g
++) {
1215 for (sfb
= 0; sfb
< sce
->ics
.max_sfb
; sfb
++, idx
++) {
1216 if (sce
->band_type
[idx
] != ZERO_BT
) {
1218 int t
= get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
1226 gain_cache
= pow(scale
, -t
) * s
;
1229 coup
->gain
[c
][idx
] = gain_cache
;
1239 * Decode Spectral Band Replication extension data; reference: table 4.55.
1241 * @param crc flag indicating the presence of CRC checksum
1242 * @param cnt length of TYPE_FIL syntactic element in bytes
1244 * @return Returns number of bytes consumed from the TYPE_FIL element.
1246 static int decode_sbr_extension(AACContext
* ac
, GetBitContext
* gb
, int crc
, int cnt
) {
1247 // TODO : sbr_extension implementation
1248 ff_log_missing_feature(ac
->avccontext
, "SBR", 0);
1249 skip_bits_long(gb
, 8*cnt
- 4); // -4 due to reading extension type
1254 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1256 * @return Returns number of bytes consumed.
1258 static int decode_drc_channel_exclusions(DynamicRangeControl
*che_drc
, GetBitContext
* gb
) {
1260 int num_excl_chan
= 0;
1263 for (i
= 0; i
< 7; i
++)
1264 che_drc
->exclude_mask
[num_excl_chan
++] = get_bits1(gb
);
1265 } while (num_excl_chan
< MAX_CHANNELS
- 7 && get_bits1(gb
));
1267 return num_excl_chan
/ 7;
1271 * Decode dynamic range information; reference: table 4.52.
1273 * @param cnt length of TYPE_FIL syntactic element in bytes
1275 * @return Returns number of bytes consumed.
1277 static int decode_dynamic_range(DynamicRangeControl
*che_drc
, GetBitContext
* gb
, int cnt
) {
1279 int drc_num_bands
= 1;
1282 /* pce_tag_present? */
1284 che_drc
->pce_instance_tag
= get_bits(gb
, 4);
1285 skip_bits(gb
, 4); // tag_reserved_bits
1289 /* excluded_chns_present? */
1291 n
+= decode_drc_channel_exclusions(che_drc
, gb
);
1294 /* drc_bands_present? */
1295 if (get_bits1(gb
)) {
1296 che_drc
->band_incr
= get_bits(gb
, 4);
1297 che_drc
->interpolation_scheme
= get_bits(gb
, 4);
1299 drc_num_bands
+= che_drc
->band_incr
;
1300 for (i
= 0; i
< drc_num_bands
; i
++) {
1301 che_drc
->band_top
[i
] = get_bits(gb
, 8);
1306 /* prog_ref_level_present? */
1307 if (get_bits1(gb
)) {
1308 che_drc
->prog_ref_level
= get_bits(gb
, 7);
1309 skip_bits1(gb
); // prog_ref_level_reserved_bits
1313 for (i
= 0; i
< drc_num_bands
; i
++) {
1314 che_drc
->dyn_rng_sgn
[i
] = get_bits1(gb
);
1315 che_drc
->dyn_rng_ctl
[i
] = get_bits(gb
, 7);
1323 * Decode extension data (incomplete); reference: table 4.51.
1325 * @param cnt length of TYPE_FIL syntactic element in bytes
1327 * @return Returns number of bytes consumed
1329 static int decode_extension_payload(AACContext
* ac
, GetBitContext
* gb
, int cnt
) {
1332 switch (get_bits(gb
, 4)) { // extension type
1333 case EXT_SBR_DATA_CRC
:
1336 res
= decode_sbr_extension(ac
, gb
, crc_flag
, cnt
);
1338 case EXT_DYNAMIC_RANGE
:
1339 res
= decode_dynamic_range(&ac
->che_drc
, gb
, cnt
);
1343 case EXT_DATA_ELEMENT
:
1345 skip_bits_long(gb
, 8*cnt
- 4);
1352 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1354 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1355 * @param coef spectral coefficients
1357 static void apply_tns(float coef
[1024], TemporalNoiseShaping
* tns
, IndividualChannelStream
* ics
, int decode
) {
1358 const int mmm
= FFMIN(ics
->tns_max_bands
, ics
->max_sfb
);
1360 int bottom
, top
, order
, start
, end
, size
, inc
;
1361 float lpc
[TNS_MAX_ORDER
];
1363 for (w
= 0; w
< ics
->num_windows
; w
++) {
1364 bottom
= ics
->num_swb
;
1365 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
1367 bottom
= FFMAX(0, top
- tns
->length
[w
][filt
]);
1368 order
= tns
->order
[w
][filt
];
1373 compute_lpc_coefs(tns
->coef
[w
][filt
], order
, lpc
, 0, 0, 0);
1375 start
= ics
->swb_offset
[FFMIN(bottom
, mmm
)];
1376 end
= ics
->swb_offset
[FFMIN( top
, mmm
)];
1377 if ((size
= end
- start
) <= 0)
1379 if (tns
->direction
[w
][filt
]) {
1380 inc
= -1; start
= end
- 1;
1387 for (m
= 0; m
< size
; m
++, start
+= inc
)
1388 for (i
= 1; i
<= FFMIN(m
, order
); i
++)
1389 coef
[start
] -= coef
[start
- i
*inc
] * lpc
[i
-1];
1395 * Conduct IMDCT and windowing.
1397 static void imdct_and_windowing(AACContext
* ac
, SingleChannelElement
* sce
) {
1398 IndividualChannelStream
* ics
= &sce
->ics
;
1399 float * in
= sce
->coeffs
;
1400 float * out
= sce
->ret
;
1401 float * saved
= sce
->saved
;
1402 const float * swindow
= ics
->use_kb_window
[0] ? ff_aac_kbd_short_128
: ff_sine_128
;
1403 const float * lwindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_long_1024
: ff_sine_1024
;
1404 const float * swindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_short_128
: ff_sine_128
;
1405 float * buf
= ac
->buf_mdct
;
1406 float * temp
= ac
->temp
;
1410 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1411 if (ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
)
1412 av_log(ac
->avccontext
, AV_LOG_WARNING
,
1413 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1414 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1415 for (i
= 0; i
< 1024; i
+= 128)
1416 ff_imdct_half(&ac
->mdct_small
, buf
+ i
, in
+ i
);
1418 ff_imdct_half(&ac
->mdct
, buf
, in
);
1420 /* window overlapping
1421 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1422 * and long to short transitions are considered to be short to short
1423 * transitions. This leaves just two cases (long to long and short to short)
1424 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1426 if ((ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
) &&
1427 (ics
->window_sequence
[0] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[0] == LONG_START_SEQUENCE
)) {
1428 ac
->dsp
.vector_fmul_window( out
, saved
, buf
, lwindow_prev
, ac
->add_bias
, 512);
1430 for (i
= 0; i
< 448; i
++)
1431 out
[i
] = saved
[i
] + ac
->add_bias
;
1433 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1434 ac
->dsp
.vector_fmul_window(out
+ 448 + 0*128, saved
+ 448, buf
+ 0*128, swindow_prev
, ac
->add_bias
, 64);
1435 ac
->dsp
.vector_fmul_window(out
+ 448 + 1*128, buf
+ 0*128 + 64, buf
+ 1*128, swindow
, ac
->add_bias
, 64);
1436 ac
->dsp
.vector_fmul_window(out
+ 448 + 2*128, buf
+ 1*128 + 64, buf
+ 2*128, swindow
, ac
->add_bias
, 64);
1437 ac
->dsp
.vector_fmul_window(out
+ 448 + 3*128, buf
+ 2*128 + 64, buf
+ 3*128, swindow
, ac
->add_bias
, 64);
1438 ac
->dsp
.vector_fmul_window(temp
, buf
+ 3*128 + 64, buf
+ 4*128, swindow
, ac
->add_bias
, 64);
1439 memcpy( out
+ 448 + 4*128, temp
, 64 * sizeof(float));
1441 ac
->dsp
.vector_fmul_window(out
+ 448, saved
+ 448, buf
, swindow_prev
, ac
->add_bias
, 64);
1442 for (i
= 576; i
< 1024; i
++)
1443 out
[i
] = buf
[i
-512] + ac
->add_bias
;
1448 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1449 for (i
= 0; i
< 64; i
++)
1450 saved
[i
] = temp
[64 + i
] - ac
->add_bias
;
1451 ac
->dsp
.vector_fmul_window(saved
+ 64, buf
+ 4*128 + 64, buf
+ 5*128, swindow
, 0, 64);
1452 ac
->dsp
.vector_fmul_window(saved
+ 192, buf
+ 5*128 + 64, buf
+ 6*128, swindow
, 0, 64);
1453 ac
->dsp
.vector_fmul_window(saved
+ 320, buf
+ 6*128 + 64, buf
+ 7*128, swindow
, 0, 64);
1454 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1455 } else if (ics
->window_sequence
[0] == LONG_START_SEQUENCE
) {
1456 memcpy( saved
, buf
+ 512, 448 * sizeof(float));
1457 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1458 } else { // LONG_STOP or ONLY_LONG
1459 memcpy( saved
, buf
+ 512, 512 * sizeof(float));
1464 * Apply dependent channel coupling (applied before IMDCT).
1466 * @param index index into coupling gain array
1468 static void apply_dependent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1469 IndividualChannelStream
* ics
= &cce
->ch
[0].ics
;
1470 const uint16_t * offsets
= ics
->swb_offset
;
1471 float * dest
= target
->coeffs
;
1472 const float * src
= cce
->ch
[0].coeffs
;
1473 int g
, i
, group
, k
, idx
= 0;
1474 if(ac
->m4ac
.object_type
== AOT_AAC_LTP
) {
1475 av_log(ac
->avccontext
, AV_LOG_ERROR
,
1476 "Dependent coupling is not supported together with LTP\n");
1479 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1480 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1481 if (cce
->ch
[0].band_type
[idx
] != ZERO_BT
) {
1482 const float gain
= cce
->coup
.gain
[index
][idx
];
1483 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1484 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
1486 dest
[group
*128+k
] += gain
* src
[group
*128+k
];
1491 dest
+= ics
->group_len
[g
]*128;
1492 src
+= ics
->group_len
[g
]*128;
1497 * Apply independent channel coupling (applied after IMDCT).
1499 * @param index index into coupling gain array
1501 static void apply_independent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1503 const float gain
= cce
->coup
.gain
[index
][0];
1504 const float bias
= ac
->add_bias
;
1505 const float* src
= cce
->ch
[0].ret
;
1506 float* dest
= target
->ret
;
1508 for (i
= 0; i
< 1024; i
++)
1509 dest
[i
] += gain
* (src
[i
] - bias
);
1513 * channel coupling transformation interface
1515 * @param index index into coupling gain array
1516 * @param apply_coupling_method pointer to (in)dependent coupling function
1518 static void apply_channel_coupling(AACContext
* ac
, ChannelElement
* cc
,
1519 enum RawDataBlockType type
, int elem_id
, enum CouplingPoint coupling_point
,
1520 void (*apply_coupling_method
)(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
))
1524 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1525 ChannelElement
*cce
= ac
->che
[TYPE_CCE
][i
];
1528 if (cce
&& cce
->coup
.coupling_point
== coupling_point
) {
1529 ChannelCoupling
* coup
= &cce
->coup
;
1531 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1532 if (coup
->type
[c
] == type
&& coup
->id_select
[c
] == elem_id
) {
1533 if (coup
->ch_select
[c
] != 1) {
1534 apply_coupling_method(ac
, &cc
->ch
[0], cce
, index
);
1535 if (coup
->ch_select
[c
] != 0)
1538 if (coup
->ch_select
[c
] != 2)
1539 apply_coupling_method(ac
, &cc
->ch
[1], cce
, index
++);
1541 index
+= 1 + (coup
->ch_select
[c
] == 3);
1548 * Convert spectral data to float samples, applying all supported tools as appropriate.
1550 static void spectral_to_sample(AACContext
* ac
) {
1552 for(type
= 3; type
>= 0; type
--) {
1553 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1554 ChannelElement
*che
= ac
->che
[type
][i
];
1556 if(type
<= TYPE_CPE
)
1557 apply_channel_coupling(ac
, che
, type
, i
, BEFORE_TNS
, apply_dependent_coupling
);
1558 if(che
->ch
[0].tns
.present
)
1559 apply_tns(che
->ch
[0].coeffs
, &che
->ch
[0].tns
, &che
->ch
[0].ics
, 1);
1560 if(che
->ch
[1].tns
.present
)
1561 apply_tns(che
->ch
[1].coeffs
, &che
->ch
[1].tns
, &che
->ch
[1].ics
, 1);
1562 if(type
<= TYPE_CPE
)
1563 apply_channel_coupling(ac
, che
, type
, i
, BETWEEN_TNS_AND_IMDCT
, apply_dependent_coupling
);
1564 if(type
!= TYPE_CCE
|| che
->coup
.coupling_point
== AFTER_IMDCT
)
1565 imdct_and_windowing(ac
, &che
->ch
[0]);
1566 if(type
== TYPE_CPE
)
1567 imdct_and_windowing(ac
, &che
->ch
[1]);
1568 if(type
<= TYPE_CCE
)
1569 apply_channel_coupling(ac
, che
, type
, i
, AFTER_IMDCT
, apply_independent_coupling
);
1575 static int parse_adts_frame_header(AACContext
* ac
, GetBitContext
* gb
) {
1578 AACADTSHeaderInfo hdr_info
;
1580 size
= ff_aac_parse_header(gb
, &hdr_info
);
1582 if (hdr_info
.chan_config
)
1583 ac
->m4ac
.chan_config
= hdr_info
.chan_config
;
1584 ac
->m4ac
.sample_rate
= hdr_info
.sample_rate
;
1585 ac
->m4ac
.sampling_index
= hdr_info
.sampling_index
;
1586 ac
->m4ac
.object_type
= hdr_info
.object_type
;
1587 if (hdr_info
.num_aac_frames
== 1) {
1588 if (!hdr_info
.crc_absent
)
1591 ff_log_missing_feature(ac
->avccontext
, "More than one AAC RDB per ADTS frame is", 0);
1598 static int aac_decode_frame(AVCodecContext
* avccontext
, void * data
, int * data_size
, AVPacket
*avpkt
) {
1599 const uint8_t *buf
= avpkt
->data
;
1600 int buf_size
= avpkt
->size
;
1601 AACContext
* ac
= avccontext
->priv_data
;
1602 ChannelElement
* che
= NULL
;
1604 enum RawDataBlockType elem_type
;
1605 int err
, elem_id
, data_size_tmp
;
1607 init_get_bits(&gb
, buf
, buf_size
*8);
1609 if (show_bits(&gb
, 12) == 0xfff) {
1610 if (parse_adts_frame_header(ac
, &gb
) < 0) {
1611 av_log(avccontext
, AV_LOG_ERROR
, "Error decoding AAC frame header.\n");
1614 if (ac
->m4ac
.sampling_index
> 12) {
1615 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
1621 while ((elem_type
= get_bits(&gb
, 3)) != TYPE_END
) {
1622 elem_id
= get_bits(&gb
, 4);
1624 if(elem_type
< TYPE_DSE
&& !(che
=get_che(ac
, elem_type
, elem_id
))) {
1625 av_log(ac
->avccontext
, AV_LOG_ERROR
, "channel element %d.%d is not allocated\n", elem_type
, elem_id
);
1629 switch (elem_type
) {
1632 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1636 err
= decode_cpe(ac
, &gb
, che
);
1640 err
= decode_cce(ac
, &gb
, che
);
1644 err
= decode_ics(ac
, &che
->ch
[0], &gb
, 0, 0);
1648 skip_data_stream_element(&gb
);
1654 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
1655 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
1656 if((err
= decode_pce(ac
, new_che_pos
, &gb
)))
1658 err
= output_configure(ac
, ac
->che_pos
, new_che_pos
, 0);
1664 elem_id
+= get_bits(&gb
, 8) - 1;
1666 elem_id
-= decode_extension_payload(ac
, &gb
, elem_id
);
1667 err
= 0; /* FIXME */
1671 err
= -1; /* should not happen, but keeps compiler happy */
1679 spectral_to_sample(ac
);
1681 if (!ac
->is_saved
) {
1687 data_size_tmp
= 1024 * avccontext
->channels
* sizeof(int16_t);
1688 if(*data_size
< data_size_tmp
) {
1689 av_log(avccontext
, AV_LOG_ERROR
,
1690 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1691 *data_size
, data_size_tmp
);
1694 *data_size
= data_size_tmp
;
1696 ac
->dsp
.float_to_int16_interleave(data
, (const float **)ac
->output_data
, 1024, avccontext
->channels
);
1701 static av_cold
int aac_decode_close(AVCodecContext
* avccontext
) {
1702 AACContext
* ac
= avccontext
->priv_data
;
1705 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1706 for(type
= 0; type
< 4; type
++)
1707 av_freep(&ac
->che
[type
][i
]);
1710 ff_mdct_end(&ac
->mdct
);
1711 ff_mdct_end(&ac
->mdct_small
);
1715 AVCodec aac_decoder
= {
1724 .long_name
= NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1725 .sample_fmts
= (enum SampleFormat
[]){SAMPLE_FMT_S16
,SAMPLE_FMT_NONE
},