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
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
80 #include "bitstream.h"
86 #include "aacdectab.h"
87 #include "mpeg4audio.h"
94 static VLC vlc_scalefactors
;
95 static VLC vlc_spectral
[11];
99 * Configure output channel order based on the current program configuration element.
101 * @param che_pos current channel position configuration
102 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
104 * @return Returns error status. 0 - OK, !0 - error
106 static int output_configure(AACContext
*ac
, enum ChannelPosition che_pos
[4][MAX_ELEM_ID
],
107 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
]) {
108 AVCodecContext
*avctx
= ac
->avccontext
;
109 int i
, type
, channels
= 0;
111 if(!memcmp(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0])))
112 return 0; /* no change */
114 memcpy(che_pos
, new_che_pos
, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
116 /* Allocate or free elements depending on if they are in the
117 * current program configuration.
119 * Set up default 1:1 output mapping.
121 * For a 5.1 stream the output order will be:
122 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
125 for(i
= 0; i
< MAX_ELEM_ID
; i
++) {
126 for(type
= 0; type
< 4; type
++) {
127 if(che_pos
[type
][i
]) {
128 if(!ac
->che
[type
][i
] && !(ac
->che
[type
][i
] = av_mallocz(sizeof(ChannelElement
))))
129 return AVERROR(ENOMEM
);
130 if(type
!= TYPE_CCE
) {
131 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[0].ret
;
132 if(type
== TYPE_CPE
) {
133 ac
->output_data
[channels
++] = ac
->che
[type
][i
]->ch
[1].ret
;
137 av_freep(&ac
->che
[type
][i
]);
141 avctx
->channels
= channels
;
146 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
148 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
149 * @param sce_map mono (Single Channel Element) map
150 * @param type speaker type/position for these channels
152 static void decode_channel_map(enum ChannelPosition
*cpe_map
,
153 enum ChannelPosition
*sce_map
, enum ChannelPosition type
, GetBitContext
* gb
, int n
) {
155 enum ChannelPosition
*map
= cpe_map
&& get_bits1(gb
) ? cpe_map
: sce_map
; // stereo or mono map
156 map
[get_bits(gb
, 4)] = type
;
161 * Decode program configuration element; reference: table 4.2.
163 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
165 * @return Returns error status. 0 - OK, !0 - error
167 static int decode_pce(AACContext
* ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
168 GetBitContext
* gb
) {
169 int num_front
, num_side
, num_back
, num_lfe
, num_assoc_data
, num_cc
;
171 skip_bits(gb
, 2); // object_type
173 ac
->m4ac
.sampling_index
= get_bits(gb
, 4);
174 if(ac
->m4ac
.sampling_index
> 11) {
175 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
178 ac
->m4ac
.sample_rate
= ff_mpeg4audio_sample_rates
[ac
->m4ac
.sampling_index
];
179 num_front
= get_bits(gb
, 4);
180 num_side
= get_bits(gb
, 4);
181 num_back
= get_bits(gb
, 4);
182 num_lfe
= get_bits(gb
, 2);
183 num_assoc_data
= get_bits(gb
, 3);
184 num_cc
= get_bits(gb
, 4);
187 skip_bits(gb
, 4); // mono_mixdown_tag
189 skip_bits(gb
, 4); // stereo_mixdown_tag
192 skip_bits(gb
, 3); // mixdown_coeff_index and pseudo_surround
194 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_FRONT
, gb
, num_front
);
195 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_SIDE
, gb
, num_side
);
196 decode_channel_map(new_che_pos
[TYPE_CPE
], new_che_pos
[TYPE_SCE
], AAC_CHANNEL_BACK
, gb
, num_back
);
197 decode_channel_map(NULL
, new_che_pos
[TYPE_LFE
], AAC_CHANNEL_LFE
, gb
, num_lfe
);
199 skip_bits_long(gb
, 4 * num_assoc_data
);
201 decode_channel_map(new_che_pos
[TYPE_CCE
], new_che_pos
[TYPE_CCE
], AAC_CHANNEL_CC
, gb
, num_cc
);
205 /* comment field, first byte is length */
206 skip_bits_long(gb
, 8 * get_bits(gb
, 8));
211 * Set up channel positions based on a default channel configuration
212 * as specified in table 1.17.
214 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
216 * @return Returns error status. 0 - OK, !0 - error
218 static int set_default_channel_config(AACContext
*ac
, enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
],
221 if(channel_config
< 1 || channel_config
> 7) {
222 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid default channel configuration (%d)\n",
227 /* default channel configurations:
229 * 1ch : front center (mono)
230 * 2ch : L + R (stereo)
231 * 3ch : front center + L + R
232 * 4ch : front center + L + R + back center
233 * 5ch : front center + L + R + back stereo
234 * 6ch : front center + L + R + back stereo + LFE
235 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
238 if(channel_config
!= 2)
239 new_che_pos
[TYPE_SCE
][0] = AAC_CHANNEL_FRONT
; // front center (or mono)
240 if(channel_config
> 1)
241 new_che_pos
[TYPE_CPE
][0] = AAC_CHANNEL_FRONT
; // L + R (or stereo)
242 if(channel_config
== 4)
243 new_che_pos
[TYPE_SCE
][1] = AAC_CHANNEL_BACK
; // back center
244 if(channel_config
> 4)
245 new_che_pos
[TYPE_CPE
][(channel_config
== 7) + 1]
246 = AAC_CHANNEL_BACK
; // back stereo
247 if(channel_config
> 5)
248 new_che_pos
[TYPE_LFE
][0] = AAC_CHANNEL_LFE
; // LFE
249 if(channel_config
== 7)
250 new_che_pos
[TYPE_CPE
][1] = AAC_CHANNEL_FRONT
; // outer front left + outer front right
256 * Decode GA "General Audio" specific configuration; reference: table 4.1.
258 * @return Returns error status. 0 - OK, !0 - error
260 static int decode_ga_specific_config(AACContext
* ac
, GetBitContext
* gb
, int channel_config
) {
261 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
262 int extension_flag
, ret
;
264 if(get_bits1(gb
)) { // frameLengthFlag
265 av_log_missing_feature(ac
->avccontext
, "960/120 MDCT window is", 1);
269 if (get_bits1(gb
)) // dependsOnCoreCoder
270 skip_bits(gb
, 14); // coreCoderDelay
271 extension_flag
= get_bits1(gb
);
273 if(ac
->m4ac
.object_type
== AOT_AAC_SCALABLE
||
274 ac
->m4ac
.object_type
== AOT_ER_AAC_SCALABLE
)
275 skip_bits(gb
, 3); // layerNr
277 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
278 if (channel_config
== 0) {
279 skip_bits(gb
, 4); // element_instance_tag
280 if((ret
= decode_pce(ac
, new_che_pos
, gb
)))
283 if((ret
= set_default_channel_config(ac
, new_che_pos
, channel_config
)))
286 if((ret
= output_configure(ac
, ac
->che_pos
, new_che_pos
)))
289 if (extension_flag
) {
290 switch (ac
->m4ac
.object_type
) {
292 skip_bits(gb
, 5); // numOfSubFrame
293 skip_bits(gb
, 11); // layer_length
297 case AOT_ER_AAC_SCALABLE
:
299 skip_bits(gb
, 3); /* aacSectionDataResilienceFlag
300 * aacScalefactorDataResilienceFlag
301 * aacSpectralDataResilienceFlag
305 skip_bits1(gb
); // extensionFlag3 (TBD in version 3)
311 * Decode audio specific configuration; reference: table 1.13.
313 * @param data pointer to AVCodecContext extradata
314 * @param data_size size of AVCCodecContext extradata
316 * @return Returns error status. 0 - OK, !0 - error
318 static int decode_audio_specific_config(AACContext
* ac
, void *data
, int data_size
) {
322 init_get_bits(&gb
, data
, data_size
* 8);
324 if((i
= ff_mpeg4audio_get_config(&ac
->m4ac
, data
, data_size
)) < 0)
326 if(ac
->m4ac
.sampling_index
> 11) {
327 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid sampling rate index %d\n", ac
->m4ac
.sampling_index
);
331 skip_bits_long(&gb
, i
);
333 switch (ac
->m4ac
.object_type
) {
336 if (decode_ga_specific_config(ac
, &gb
, ac
->m4ac
.chan_config
))
340 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Audio object type %s%d is not supported.\n",
341 ac
->m4ac
.sbr
== 1? "SBR+" : "", ac
->m4ac
.object_type
);
348 * linear congruential pseudorandom number generator
350 * @param previous_val pointer to the current state of the generator
352 * @return Returns a 32-bit pseudorandom integer
354 static av_always_inline
int lcg_random(int previous_val
) {
355 return previous_val
* 1664525 + 1013904223;
358 static void reset_predict_state(PredictorState
* ps
) {
367 static void reset_all_predictors(PredictorState
* ps
) {
369 for (i
= 0; i
< MAX_PREDICTORS
; i
++)
370 reset_predict_state(&ps
[i
]);
373 static void reset_predictor_group(PredictorState
* ps
, int group_num
) {
375 for (i
= group_num
-1; i
< MAX_PREDICTORS
; i
+=30)
376 reset_predict_state(&ps
[i
]);
379 static av_cold
int aac_decode_init(AVCodecContext
* avccontext
) {
380 AACContext
* ac
= avccontext
->priv_data
;
383 ac
->avccontext
= avccontext
;
385 if (avccontext
->extradata_size
<= 0 ||
386 decode_audio_specific_config(ac
, avccontext
->extradata
, avccontext
->extradata_size
))
389 avccontext
->sample_fmt
= SAMPLE_FMT_S16
;
390 avccontext
->sample_rate
= ac
->m4ac
.sample_rate
;
391 avccontext
->frame_size
= 1024;
393 AAC_INIT_VLC_STATIC( 0, 144);
394 AAC_INIT_VLC_STATIC( 1, 114);
395 AAC_INIT_VLC_STATIC( 2, 188);
396 AAC_INIT_VLC_STATIC( 3, 180);
397 AAC_INIT_VLC_STATIC( 4, 172);
398 AAC_INIT_VLC_STATIC( 5, 140);
399 AAC_INIT_VLC_STATIC( 6, 168);
400 AAC_INIT_VLC_STATIC( 7, 114);
401 AAC_INIT_VLC_STATIC( 8, 262);
402 AAC_INIT_VLC_STATIC( 9, 248);
403 AAC_INIT_VLC_STATIC(10, 384);
405 dsputil_init(&ac
->dsp
, avccontext
);
407 ac
->random_state
= 0x1f2e3d4c;
409 // -1024 - Compensate wrong IMDCT method.
410 // 32768 - Required to scale values to the correct range for the bias method
411 // for float to int16 conversion.
413 if(ac
->dsp
.float_to_int16
== ff_float_to_int16_c
) {
414 ac
->add_bias
= 385.0f
;
415 ac
->sf_scale
= 1. / (-1024. * 32768.);
419 ac
->sf_scale
= 1. / -1024.;
423 #ifndef CONFIG_HARDCODED_TABLES
424 for (i
= 0; i
< 428; i
++)
425 ff_aac_pow2sf_tab
[i
] = pow(2, (i
- 200)/4.);
426 #endif /* CONFIG_HARDCODED_TABLES */
428 INIT_VLC_STATIC(&vlc_scalefactors
,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code
),
429 ff_aac_scalefactor_bits
, sizeof(ff_aac_scalefactor_bits
[0]), sizeof(ff_aac_scalefactor_bits
[0]),
430 ff_aac_scalefactor_code
, sizeof(ff_aac_scalefactor_code
[0]), sizeof(ff_aac_scalefactor_code
[0]),
433 ff_mdct_init(&ac
->mdct
, 11, 1);
434 ff_mdct_init(&ac
->mdct_small
, 8, 1);
435 // window initialization
436 ff_kbd_window_init(ff_aac_kbd_long_1024
, 4.0, 1024);
437 ff_kbd_window_init(ff_aac_kbd_short_128
, 6.0, 128);
438 ff_sine_window_init(ff_sine_1024
, 1024);
439 ff_sine_window_init(ff_sine_128
, 128);
445 * Skip data_stream_element; reference: table 4.10.
447 static void skip_data_stream_element(GetBitContext
* gb
) {
448 int byte_align
= get_bits1(gb
);
449 int count
= get_bits(gb
, 8);
451 count
+= get_bits(gb
, 8);
454 skip_bits_long(gb
, 8 * count
);
457 static int decode_prediction(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
) {
460 ics
->predictor_reset_group
= get_bits(gb
, 5);
461 if (ics
->predictor_reset_group
== 0 || ics
->predictor_reset_group
> 30) {
462 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Invalid Predictor Reset Group.\n");
466 for (sfb
= 0; sfb
< FFMIN(ics
->max_sfb
, ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]); sfb
++) {
467 ics
->prediction_used
[sfb
] = get_bits1(gb
);
473 * Decode Individual Channel Stream info; reference: table 4.6.
475 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
477 static int decode_ics_info(AACContext
* ac
, IndividualChannelStream
* ics
, GetBitContext
* gb
, int common_window
) {
479 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Reserved bit set.\n");
480 memset(ics
, 0, sizeof(IndividualChannelStream
));
483 ics
->window_sequence
[1] = ics
->window_sequence
[0];
484 ics
->window_sequence
[0] = get_bits(gb
, 2);
485 ics
->use_kb_window
[1] = ics
->use_kb_window
[0];
486 ics
->use_kb_window
[0] = get_bits1(gb
);
487 ics
->num_window_groups
= 1;
488 ics
->group_len
[0] = 1;
489 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
491 ics
->max_sfb
= get_bits(gb
, 4);
492 for (i
= 0; i
< 7; i
++) {
494 ics
->group_len
[ics
->num_window_groups
-1]++;
496 ics
->num_window_groups
++;
497 ics
->group_len
[ics
->num_window_groups
-1] = 1;
500 ics
->num_windows
= 8;
501 ics
->swb_offset
= swb_offset_128
[ac
->m4ac
.sampling_index
];
502 ics
->num_swb
= ff_aac_num_swb_128
[ac
->m4ac
.sampling_index
];
503 ics
->tns_max_bands
= tns_max_bands_128
[ac
->m4ac
.sampling_index
];
504 ics
->predictor_present
= 0;
506 ics
->max_sfb
= get_bits(gb
, 6);
507 ics
->num_windows
= 1;
508 ics
->swb_offset
= swb_offset_1024
[ac
->m4ac
.sampling_index
];
509 ics
->num_swb
= ff_aac_num_swb_1024
[ac
->m4ac
.sampling_index
];
510 ics
->tns_max_bands
= tns_max_bands_1024
[ac
->m4ac
.sampling_index
];
511 ics
->predictor_present
= get_bits1(gb
);
512 ics
->predictor_reset_group
= 0;
513 if (ics
->predictor_present
) {
514 if (ac
->m4ac
.object_type
== AOT_AAC_MAIN
) {
515 if (decode_prediction(ac
, ics
, gb
)) {
516 memset(ics
, 0, sizeof(IndividualChannelStream
));
519 } else if (ac
->m4ac
.object_type
== AOT_AAC_LC
) {
520 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Prediction is not allowed in AAC-LC.\n");
521 memset(ics
, 0, sizeof(IndividualChannelStream
));
524 av_log_missing_feature(ac
->avccontext
, "Predictor bit set but LTP is", 1);
525 memset(ics
, 0, sizeof(IndividualChannelStream
));
531 if(ics
->max_sfb
> ics
->num_swb
) {
532 av_log(ac
->avccontext
, AV_LOG_ERROR
,
533 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
534 ics
->max_sfb
, ics
->num_swb
);
535 memset(ics
, 0, sizeof(IndividualChannelStream
));
543 * Decode band types (section_data payload); reference: table 4.46.
545 * @param band_type array of the used band type
546 * @param band_type_run_end array of the last scalefactor band of a band type run
548 * @return Returns error status. 0 - OK, !0 - error
550 static int decode_band_types(AACContext
* ac
, enum BandType band_type
[120],
551 int band_type_run_end
[120], GetBitContext
* gb
, IndividualChannelStream
* ics
) {
553 const int bits
= (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) ? 3 : 5;
554 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
556 while (k
< ics
->max_sfb
) {
557 uint8_t sect_len
= k
;
559 int sect_band_type
= get_bits(gb
, 4);
560 if (sect_band_type
== 12) {
561 av_log(ac
->avccontext
, AV_LOG_ERROR
, "invalid band type\n");
564 while ((sect_len_incr
= get_bits(gb
, bits
)) == (1 << bits
)-1)
565 sect_len
+= sect_len_incr
;
566 sect_len
+= sect_len_incr
;
567 if (sect_len
> ics
->max_sfb
) {
568 av_log(ac
->avccontext
, AV_LOG_ERROR
,
569 "Number of bands (%d) exceeds limit (%d).\n",
570 sect_len
, ics
->max_sfb
);
573 for (; k
< sect_len
; k
++) {
574 band_type
[idx
] = sect_band_type
;
575 band_type_run_end
[idx
++] = sect_len
;
583 * Decode scalefactors; reference: table 4.47.
585 * @param global_gain first scalefactor value as scalefactors are differentially coded
586 * @param band_type array of the used band type
587 * @param band_type_run_end array of the last scalefactor band of a band type run
588 * @param sf array of scalefactors or intensity stereo positions
590 * @return Returns error status. 0 - OK, !0 - error
592 static int decode_scalefactors(AACContext
* ac
, float sf
[120], GetBitContext
* gb
,
593 unsigned int global_gain
, IndividualChannelStream
* ics
,
594 enum BandType band_type
[120], int band_type_run_end
[120]) {
595 const int sf_offset
= ac
->sf_offset
+ (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
? 12 : 0);
597 int offset
[3] = { global_gain
, global_gain
- 90, 100 };
599 static const char *sf_str
[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
600 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
601 for (i
= 0; i
< ics
->max_sfb
;) {
602 int run_end
= band_type_run_end
[idx
];
603 if (band_type
[idx
] == ZERO_BT
) {
604 for(; i
< run_end
; i
++, idx
++)
606 }else if((band_type
[idx
] == INTENSITY_BT
) || (band_type
[idx
] == INTENSITY_BT2
)) {
607 for(; i
< run_end
; i
++, idx
++) {
608 offset
[2] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
609 if(offset
[2] > 255U) {
610 av_log(ac
->avccontext
, AV_LOG_ERROR
,
611 "%s (%d) out of range.\n", sf_str
[2], offset
[2]);
614 sf
[idx
] = ff_aac_pow2sf_tab
[-offset
[2] + 300];
616 }else if(band_type
[idx
] == NOISE_BT
) {
617 for(; i
< run_end
; i
++, idx
++) {
619 offset
[1] += get_bits(gb
, 9) - 256;
621 offset
[1] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
622 if(offset
[1] > 255U) {
623 av_log(ac
->avccontext
, AV_LOG_ERROR
,
624 "%s (%d) out of range.\n", sf_str
[1], offset
[1]);
627 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[1] + sf_offset
+ 100];
630 for(; i
< run_end
; i
++, idx
++) {
631 offset
[0] += get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
632 if(offset
[0] > 255U) {
633 av_log(ac
->avccontext
, AV_LOG_ERROR
,
634 "%s (%d) out of range.\n", sf_str
[0], offset
[0]);
637 sf
[idx
] = -ff_aac_pow2sf_tab
[ offset
[0] + sf_offset
];
646 * Decode pulse data; reference: table 4.7.
648 static int decode_pulses(Pulse
* pulse
, GetBitContext
* gb
, const uint16_t * swb_offset
, int num_swb
) {
650 pulse
->num_pulse
= get_bits(gb
, 2) + 1;
651 pulse_swb
= get_bits(gb
, 6);
652 if (pulse_swb
>= num_swb
)
654 pulse
->pos
[0] = swb_offset
[pulse_swb
];
655 pulse
->pos
[0] += get_bits(gb
, 5);
656 if (pulse
->pos
[0] > 1023)
658 pulse
->amp
[0] = get_bits(gb
, 4);
659 for (i
= 1; i
< pulse
->num_pulse
; i
++) {
660 pulse
->pos
[i
] = get_bits(gb
, 5) + pulse
->pos
[i
-1];
661 if (pulse
->pos
[i
] > 1023)
663 pulse
->amp
[i
] = get_bits(gb
, 4);
669 * Decode Temporal Noise Shaping data; reference: table 4.48.
671 * @return Returns error status. 0 - OK, !0 - error
673 static int decode_tns(AACContext
* ac
, TemporalNoiseShaping
* tns
,
674 GetBitContext
* gb
, const IndividualChannelStream
* ics
) {
675 int w
, filt
, i
, coef_len
, coef_res
, coef_compress
;
676 const int is8
= ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
;
677 const int tns_max_order
= is8
? 7 : ac
->m4ac
.object_type
== AOT_AAC_MAIN
? 20 : 12;
678 for (w
= 0; w
< ics
->num_windows
; w
++) {
679 if ((tns
->n_filt
[w
] = get_bits(gb
, 2 - is8
))) {
680 coef_res
= get_bits1(gb
);
682 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
684 tns
->length
[w
][filt
] = get_bits(gb
, 6 - 2*is8
);
686 if ((tns
->order
[w
][filt
] = get_bits(gb
, 5 - 2*is8
)) > tns_max_order
) {
687 av_log(ac
->avccontext
, AV_LOG_ERROR
, "TNS filter order %d is greater than maximum %d.",
688 tns
->order
[w
][filt
], tns_max_order
);
689 tns
->order
[w
][filt
] = 0;
692 if (tns
->order
[w
][filt
]) {
693 tns
->direction
[w
][filt
] = get_bits1(gb
);
694 coef_compress
= get_bits1(gb
);
695 coef_len
= coef_res
+ 3 - coef_compress
;
696 tmp2_idx
= 2*coef_compress
+ coef_res
;
698 for (i
= 0; i
< tns
->order
[w
][filt
]; i
++)
699 tns
->coef
[w
][filt
][i
] = tns_tmp2_map
[tmp2_idx
][get_bits(gb
, coef_len
)];
708 * Decode Mid/Side data; reference: table 4.54.
710 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
711 * [1] mask is decoded from bitstream; [2] mask is all 1s;
712 * [3] reserved for scalable AAC
714 static void decode_mid_side_stereo(ChannelElement
* cpe
, GetBitContext
* gb
,
717 if (ms_present
== 1) {
718 for (idx
= 0; idx
< cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
; idx
++)
719 cpe
->ms_mask
[idx
] = get_bits1(gb
);
720 } else if (ms_present
== 2) {
721 memset(cpe
->ms_mask
, 1, cpe
->ch
[0].ics
.num_window_groups
* cpe
->ch
[0].ics
.max_sfb
* sizeof(cpe
->ms_mask
[0]));
726 * Decode spectral data; reference: table 4.50.
727 * Dequantize and scale spectral data; reference: 4.6.3.3.
729 * @param coef array of dequantized, scaled spectral data
730 * @param sf array of scalefactors or intensity stereo positions
731 * @param pulse_present set if pulses are present
732 * @param pulse pointer to pulse data struct
733 * @param band_type array of the used band type
735 * @return Returns error status. 0 - OK, !0 - error
737 static int decode_spectrum_and_dequant(AACContext
* ac
, float coef
[1024], GetBitContext
* gb
, float sf
[120],
738 int pulse_present
, const Pulse
* pulse
, const IndividualChannelStream
* ics
, enum BandType band_type
[120]) {
739 int i
, k
, g
, idx
= 0;
740 const int c
= 1024/ics
->num_windows
;
741 const uint16_t * offsets
= ics
->swb_offset
;
742 float *coef_base
= coef
;
744 for (g
= 0; g
< ics
->num_windows
; g
++)
745 memset(coef
+ g
* 128 + offsets
[ics
->max_sfb
], 0, sizeof(float)*(c
- offsets
[ics
->max_sfb
]));
747 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
748 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
749 const int cur_band_type
= band_type
[idx
];
750 const int dim
= cur_band_type
>= FIRST_PAIR_BT
? 2 : 4;
751 const int is_cb_unsigned
= IS_CODEBOOK_UNSIGNED(cur_band_type
);
753 if (cur_band_type
== ZERO_BT
) {
754 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
755 memset(coef
+ group
* 128 + offsets
[i
], 0, (offsets
[i
+1] - offsets
[i
])*sizeof(float));
757 }else if (cur_band_type
== NOISE_BT
) {
758 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
760 float band_energy
= 0;
761 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
762 ac
->random_state
= lcg_random(ac
->random_state
);
763 coef
[group
*128+k
] = ac
->random_state
;
764 band_energy
+= coef
[group
*128+k
]*coef
[group
*128+k
];
766 scale
= sf
[idx
] / sqrtf(band_energy
);
767 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
768 coef
[group
*128+k
] *= scale
;
771 }else if (cur_band_type
!= INTENSITY_BT2
&& cur_band_type
!= INTENSITY_BT
) {
772 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
773 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
+= dim
) {
774 const int index
= get_vlc2(gb
, vlc_spectral
[cur_band_type
- 1].table
, 6, 3);
775 const int coef_tmp_idx
= (group
<< 7) + k
;
778 if(index
>= ff_aac_spectral_sizes
[cur_band_type
- 1]) {
779 av_log(ac
->avccontext
, AV_LOG_ERROR
,
780 "Read beyond end of ff_aac_codebook_vectors[%d][]. index %d >= %d\n",
781 cur_band_type
- 1, index
, ff_aac_spectral_sizes
[cur_band_type
- 1]);
784 vq_ptr
= &ff_aac_codebook_vectors
[cur_band_type
- 1][index
* dim
];
785 if (is_cb_unsigned
) {
786 for (j
= 0; j
< dim
; j
++)
788 coef
[coef_tmp_idx
+ j
] = 1 - 2*(int)get_bits1(gb
);
790 for (j
= 0; j
< dim
; j
++)
791 coef
[coef_tmp_idx
+ j
] = 1.0f
;
793 if (cur_band_type
== ESC_BT
) {
794 for (j
= 0; j
< 2; j
++) {
795 if (vq_ptr
[j
] == 64.0f
) {
797 /* The total length of escape_sequence must be < 22 bits according
798 to the specification (i.e. max is 11111111110xxxxxxxxxx). */
799 while (get_bits1(gb
) && n
< 15) n
++;
801 av_log(ac
->avccontext
, AV_LOG_ERROR
, "error in spectral data, ESC overflow\n");
804 n
= (1<<n
) + get_bits(gb
, n
);
805 coef
[coef_tmp_idx
+ j
] *= cbrtf(fabsf(n
)) * n
;
807 coef
[coef_tmp_idx
+ j
] *= vq_ptr
[j
];
810 for (j
= 0; j
< dim
; j
++)
811 coef
[coef_tmp_idx
+ j
] *= vq_ptr
[j
];
812 for (j
= 0; j
< dim
; j
++)
813 coef
[coef_tmp_idx
+ j
] *= sf
[idx
];
818 coef
+= ics
->group_len
[g
]<<7;
823 for(i
= 0; i
< pulse
->num_pulse
; i
++){
824 float co
= coef_base
[ pulse
->pos
[i
] ];
825 while(offsets
[idx
+ 1] <= pulse
->pos
[i
])
827 if (band_type
[idx
] != NOISE_BT
&& sf
[idx
]) {
828 float ico
= -pulse
->amp
[i
];
831 ico
= co
/ sqrtf(sqrtf(fabsf(co
))) + (co
> 0 ? -ico
: ico
);
833 coef_base
[ pulse
->pos
[i
] ] = cbrtf(fabsf(ico
)) * ico
* sf
[idx
];
840 static av_always_inline
float flt16_round(float pf
) {
842 pf
= frexpf(pf
, &exp
);
843 pf
= ldexpf(roundf(ldexpf(pf
, 8)), exp
-8);
847 static av_always_inline
float flt16_even(float pf
) {
849 pf
= frexpf(pf
, &exp
);
850 pf
= ldexpf(rintf(ldexpf(pf
, 8)), exp
-8);
854 static av_always_inline
float flt16_trunc(float pf
) {
856 pf
= frexpf(pf
, &exp
);
857 pf
= ldexpf(truncf(ldexpf(pf
, 8)), exp
-8);
861 static void predict(AACContext
* ac
, PredictorState
* ps
, float* coef
, int output_enable
) {
862 const float a
= 0.953125; // 61.0/64
863 const float alpha
= 0.90625; // 29.0/32
868 k1
= ps
->var0
> 1 ? ps
->cor0
* flt16_even(a
/ ps
->var0
) : 0;
869 k2
= ps
->var1
> 1 ? ps
->cor1
* flt16_even(a
/ ps
->var1
) : 0;
871 pv
= flt16_round(k1
* ps
->r0
+ k2
* ps
->r1
);
873 *coef
+= pv
* ac
->sf_scale
;
875 e0
= *coef
/ ac
->sf_scale
;
876 e1
= e0
- k1
* ps
->r0
;
878 ps
->cor1
= flt16_trunc(alpha
* ps
->cor1
+ ps
->r1
* e1
);
879 ps
->var1
= flt16_trunc(alpha
* ps
->var1
+ 0.5 * (ps
->r1
* ps
->r1
+ e1
* e1
));
880 ps
->cor0
= flt16_trunc(alpha
* ps
->cor0
+ ps
->r0
* e0
);
881 ps
->var0
= flt16_trunc(alpha
* ps
->var0
+ 0.5 * (ps
->r0
* ps
->r0
+ e0
* e0
));
883 ps
->r1
= flt16_trunc(a
* (ps
->r0
- k1
* e0
));
884 ps
->r0
= flt16_trunc(a
* e0
);
888 * Apply AAC-Main style frequency domain prediction.
890 static void apply_prediction(AACContext
* ac
, SingleChannelElement
* sce
) {
893 if (!sce
->ics
.predictor_initialized
) {
894 reset_all_predictors(sce
->ics
.predictor_state
);
895 sce
->ics
.predictor_initialized
= 1;
898 if (sce
->ics
.window_sequence
[0] != EIGHT_SHORT_SEQUENCE
) {
899 for (sfb
= 0; sfb
< ff_aac_pred_sfb_max
[ac
->m4ac
.sampling_index
]; sfb
++) {
900 for (k
= sce
->ics
.swb_offset
[sfb
]; k
< sce
->ics
.swb_offset
[sfb
+ 1]; k
++) {
901 predict(ac
, &sce
->ics
.predictor_state
[k
], &sce
->coeffs
[k
],
902 sce
->ics
.predictor_present
&& sce
->ics
.prediction_used
[sfb
]);
905 if (sce
->ics
.predictor_reset_group
)
906 reset_predictor_group(sce
->ics
.predictor_state
, sce
->ics
.predictor_reset_group
);
908 reset_all_predictors(sce
->ics
.predictor_state
);
912 * Decode an individual_channel_stream payload; reference: table 4.44.
914 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
915 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
917 * @return Returns error status. 0 - OK, !0 - error
919 static int decode_ics(AACContext
* ac
, SingleChannelElement
* sce
, GetBitContext
* gb
, int common_window
, int scale_flag
) {
921 TemporalNoiseShaping
* tns
= &sce
->tns
;
922 IndividualChannelStream
* ics
= &sce
->ics
;
923 float * out
= sce
->coeffs
;
924 int global_gain
, pulse_present
= 0;
926 /* This assignment is to silence a GCC warning about the variable being used
927 * uninitialized when in fact it always is.
931 global_gain
= get_bits(gb
, 8);
933 if (!common_window
&& !scale_flag
) {
934 if (decode_ics_info(ac
, ics
, gb
, 0) < 0)
938 if (decode_band_types(ac
, sce
->band_type
, sce
->band_type_run_end
, gb
, ics
) < 0)
940 if (decode_scalefactors(ac
, sce
->sf
, gb
, global_gain
, ics
, sce
->band_type
, sce
->band_type_run_end
) < 0)
945 if ((pulse_present
= get_bits1(gb
))) {
946 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
947 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse tool not allowed in eight short sequence.\n");
950 if (decode_pulses(&pulse
, gb
, ics
->swb_offset
, ics
->num_swb
)) {
951 av_log(ac
->avccontext
, AV_LOG_ERROR
, "Pulse data corrupt or invalid.\n");
955 if ((tns
->present
= get_bits1(gb
)) && decode_tns(ac
, tns
, gb
, ics
))
958 av_log_missing_feature(ac
->avccontext
, "SSR", 1);
963 if (decode_spectrum_and_dequant(ac
, out
, gb
, sce
->sf
, pulse_present
, &pulse
, ics
, sce
->band_type
) < 0)
966 if(ac
->m4ac
.object_type
== AOT_AAC_MAIN
)
967 apply_prediction(ac
, sce
);
973 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
975 static void apply_mid_side_stereo(ChannelElement
* cpe
) {
976 const IndividualChannelStream
* ics
= &cpe
->ch
[0].ics
;
977 float *ch0
= cpe
->ch
[0].coeffs
;
978 float *ch1
= cpe
->ch
[1].coeffs
;
979 int g
, i
, k
, group
, idx
= 0;
980 const uint16_t * offsets
= ics
->swb_offset
;
981 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
982 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
983 if (cpe
->ms_mask
[idx
] &&
984 cpe
->ch
[0].band_type
[idx
] < NOISE_BT
&& cpe
->ch
[1].band_type
[idx
] < NOISE_BT
) {
985 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
986 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
987 float tmp
= ch0
[group
*128 + k
] - ch1
[group
*128 + k
];
988 ch0
[group
*128 + k
] += ch1
[group
*128 + k
];
989 ch1
[group
*128 + k
] = tmp
;
994 ch0
+= ics
->group_len
[g
]*128;
995 ch1
+= ics
->group_len
[g
]*128;
1000 * intensity stereo decoding; reference: 4.6.8.2.3
1002 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1003 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1004 * [3] reserved for scalable AAC
1006 static void apply_intensity_stereo(ChannelElement
* cpe
, int ms_present
) {
1007 const IndividualChannelStream
* ics
= &cpe
->ch
[1].ics
;
1008 SingleChannelElement
* sce1
= &cpe
->ch
[1];
1009 float *coef0
= cpe
->ch
[0].coeffs
, *coef1
= cpe
->ch
[1].coeffs
;
1010 const uint16_t * offsets
= ics
->swb_offset
;
1011 int g
, group
, i
, k
, idx
= 0;
1014 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1015 for (i
= 0; i
< ics
->max_sfb
;) {
1016 if (sce1
->band_type
[idx
] == INTENSITY_BT
|| sce1
->band_type
[idx
] == INTENSITY_BT2
) {
1017 const int bt_run_end
= sce1
->band_type_run_end
[idx
];
1018 for (; i
< bt_run_end
; i
++, idx
++) {
1019 c
= -1 + 2 * (sce1
->band_type
[idx
] - 14);
1021 c
*= 1 - 2 * cpe
->ms_mask
[idx
];
1022 scale
= c
* sce1
->sf
[idx
];
1023 for (group
= 0; group
< ics
->group_len
[g
]; group
++)
1024 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++)
1025 coef1
[group
*128 + k
] = scale
* coef0
[group
*128 + k
];
1028 int bt_run_end
= sce1
->band_type_run_end
[idx
];
1029 idx
+= bt_run_end
- i
;
1033 coef0
+= ics
->group_len
[g
]*128;
1034 coef1
+= ics
->group_len
[g
]*128;
1039 * Decode a channel_pair_element; reference: table 4.4.
1041 * @param elem_id Identifies the instance of a syntax element.
1043 * @return Returns error status. 0 - OK, !0 - error
1045 static int decode_cpe(AACContext
* ac
, GetBitContext
* gb
, int elem_id
) {
1046 int i
, ret
, common_window
, ms_present
= 0;
1047 ChannelElement
* cpe
;
1049 cpe
= ac
->che
[TYPE_CPE
][elem_id
];
1050 common_window
= get_bits1(gb
);
1051 if (common_window
) {
1052 if (decode_ics_info(ac
, &cpe
->ch
[0].ics
, gb
, 1))
1054 i
= cpe
->ch
[1].ics
.use_kb_window
[0];
1055 cpe
->ch
[1].ics
= cpe
->ch
[0].ics
;
1056 cpe
->ch
[1].ics
.use_kb_window
[1] = i
;
1057 ms_present
= get_bits(gb
, 2);
1058 if(ms_present
== 3) {
1059 av_log(ac
->avccontext
, AV_LOG_ERROR
, "ms_present = 3 is reserved.\n");
1061 } else if(ms_present
)
1062 decode_mid_side_stereo(cpe
, gb
, ms_present
);
1064 if ((ret
= decode_ics(ac
, &cpe
->ch
[0], gb
, common_window
, 0)))
1066 if ((ret
= decode_ics(ac
, &cpe
->ch
[1], gb
, common_window
, 0)))
1069 if (common_window
&& ms_present
)
1070 apply_mid_side_stereo(cpe
);
1072 apply_intensity_stereo(cpe
, ms_present
);
1077 * Decode coupling_channel_element; reference: table 4.8.
1079 * @param elem_id Identifies the instance of a syntax element.
1081 * @return Returns error status. 0 - OK, !0 - error
1083 static int decode_cce(AACContext
* ac
, GetBitContext
* gb
, ChannelElement
* che
) {
1088 SingleChannelElement
* sce
= &che
->ch
[0];
1089 ChannelCoupling
* coup
= &che
->coup
;
1091 coup
->coupling_point
= 2*get_bits1(gb
);
1092 coup
->num_coupled
= get_bits(gb
, 3);
1093 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1095 coup
->type
[c
] = get_bits1(gb
) ? TYPE_CPE
: TYPE_SCE
;
1096 coup
->id_select
[c
] = get_bits(gb
, 4);
1097 if (coup
->type
[c
] == TYPE_CPE
) {
1098 coup
->ch_select
[c
] = get_bits(gb
, 2);
1099 if (coup
->ch_select
[c
] == 3)
1102 coup
->ch_select
[c
] = 2;
1104 coup
->coupling_point
+= get_bits1(gb
);
1106 if (coup
->coupling_point
== 2) {
1107 av_log(ac
->avccontext
, AV_LOG_ERROR
,
1108 "Independently switched CCE with 'invalid' domain signalled.\n");
1109 memset(coup
, 0, sizeof(ChannelCoupling
));
1113 sign
= get_bits(gb
, 1);
1114 scale
= pow(2., pow(2., (int)get_bits(gb
, 2) - 3));
1116 if ((ret
= decode_ics(ac
, sce
, gb
, 0, 0)))
1119 for (c
= 0; c
< num_gain
; c
++) {
1123 float gain_cache
= 1.;
1125 cge
= coup
->coupling_point
== AFTER_IMDCT
? 1 : get_bits1(gb
);
1126 gain
= cge
? get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60: 0;
1127 gain_cache
= pow(scale
, -gain
);
1129 for (g
= 0; g
< sce
->ics
.num_window_groups
; g
++) {
1130 for (sfb
= 0; sfb
< sce
->ics
.max_sfb
; sfb
++, idx
++) {
1131 if (sce
->band_type
[idx
] != ZERO_BT
) {
1133 int t
= get_vlc2(gb
, vlc_scalefactors
.table
, 7, 3) - 60;
1141 gain_cache
= pow(scale
, -t
) * s
;
1144 coup
->gain
[c
][idx
] = gain_cache
;
1153 * Decode Spectral Band Replication extension data; reference: table 4.55.
1155 * @param crc flag indicating the presence of CRC checksum
1156 * @param cnt length of TYPE_FIL syntactic element in bytes
1158 * @return Returns number of bytes consumed from the TYPE_FIL element.
1160 static int decode_sbr_extension(AACContext
* ac
, GetBitContext
* gb
, int crc
, int cnt
) {
1161 // TODO : sbr_extension implementation
1162 av_log_missing_feature(ac
->avccontext
, "SBR", 0);
1163 skip_bits_long(gb
, 8*cnt
- 4); // -4 due to reading extension type
1168 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1170 * @return Returns number of bytes consumed.
1172 static int decode_drc_channel_exclusions(DynamicRangeControl
*che_drc
, GetBitContext
* gb
) {
1174 int num_excl_chan
= 0;
1177 for (i
= 0; i
< 7; i
++)
1178 che_drc
->exclude_mask
[num_excl_chan
++] = get_bits1(gb
);
1179 } while (num_excl_chan
< MAX_CHANNELS
- 7 && get_bits1(gb
));
1181 return num_excl_chan
/ 7;
1185 * Decode dynamic range information; reference: table 4.52.
1187 * @param cnt length of TYPE_FIL syntactic element in bytes
1189 * @return Returns number of bytes consumed.
1191 static int decode_dynamic_range(DynamicRangeControl
*che_drc
, GetBitContext
* gb
, int cnt
) {
1193 int drc_num_bands
= 1;
1196 /* pce_tag_present? */
1198 che_drc
->pce_instance_tag
= get_bits(gb
, 4);
1199 skip_bits(gb
, 4); // tag_reserved_bits
1203 /* excluded_chns_present? */
1205 n
+= decode_drc_channel_exclusions(che_drc
, gb
);
1208 /* drc_bands_present? */
1209 if (get_bits1(gb
)) {
1210 che_drc
->band_incr
= get_bits(gb
, 4);
1211 che_drc
->interpolation_scheme
= get_bits(gb
, 4);
1213 drc_num_bands
+= che_drc
->band_incr
;
1214 for (i
= 0; i
< drc_num_bands
; i
++) {
1215 che_drc
->band_top
[i
] = get_bits(gb
, 8);
1220 /* prog_ref_level_present? */
1221 if (get_bits1(gb
)) {
1222 che_drc
->prog_ref_level
= get_bits(gb
, 7);
1223 skip_bits1(gb
); // prog_ref_level_reserved_bits
1227 for (i
= 0; i
< drc_num_bands
; i
++) {
1228 che_drc
->dyn_rng_sgn
[i
] = get_bits1(gb
);
1229 che_drc
->dyn_rng_ctl
[i
] = get_bits(gb
, 7);
1237 * Decode extension data (incomplete); reference: table 4.51.
1239 * @param cnt length of TYPE_FIL syntactic element in bytes
1241 * @return Returns number of bytes consumed
1243 static int decode_extension_payload(AACContext
* ac
, GetBitContext
* gb
, int cnt
) {
1246 switch (get_bits(gb
, 4)) { // extension type
1247 case EXT_SBR_DATA_CRC
:
1250 res
= decode_sbr_extension(ac
, gb
, crc_flag
, cnt
);
1252 case EXT_DYNAMIC_RANGE
:
1253 res
= decode_dynamic_range(&ac
->che_drc
, gb
, cnt
);
1257 case EXT_DATA_ELEMENT
:
1259 skip_bits_long(gb
, 8*cnt
- 4);
1266 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1268 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1269 * @param coef spectral coefficients
1271 static void apply_tns(float coef
[1024], TemporalNoiseShaping
* tns
, IndividualChannelStream
* ics
, int decode
) {
1272 const int mmm
= FFMIN(ics
->tns_max_bands
, ics
->max_sfb
);
1274 int bottom
, top
, order
, start
, end
, size
, inc
;
1275 float lpc
[TNS_MAX_ORDER
];
1277 for (w
= 0; w
< ics
->num_windows
; w
++) {
1278 bottom
= ics
->num_swb
;
1279 for (filt
= 0; filt
< tns
->n_filt
[w
]; filt
++) {
1281 bottom
= FFMAX(0, top
- tns
->length
[w
][filt
]);
1282 order
= tns
->order
[w
][filt
];
1287 compute_lpc_coefs(tns
->coef
[w
][filt
], order
, lpc
, 0, 0, 0);
1289 start
= ics
->swb_offset
[FFMIN(bottom
, mmm
)];
1290 end
= ics
->swb_offset
[FFMIN( top
, mmm
)];
1291 if ((size
= end
- start
) <= 0)
1293 if (tns
->direction
[w
][filt
]) {
1294 inc
= -1; start
= end
- 1;
1301 for (m
= 0; m
< size
; m
++, start
+= inc
)
1302 for (i
= 1; i
<= FFMIN(m
, order
); i
++)
1303 coef
[start
] -= coef
[start
- i
*inc
] * lpc
[i
-1];
1309 * Conduct IMDCT and windowing.
1311 static void imdct_and_windowing(AACContext
* ac
, SingleChannelElement
* sce
) {
1312 IndividualChannelStream
* ics
= &sce
->ics
;
1313 float * in
= sce
->coeffs
;
1314 float * out
= sce
->ret
;
1315 float * saved
= sce
->saved
;
1316 const float * swindow
= ics
->use_kb_window
[0] ? ff_aac_kbd_short_128
: ff_sine_128
;
1317 const float * lwindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_long_1024
: ff_sine_1024
;
1318 const float * swindow_prev
= ics
->use_kb_window
[1] ? ff_aac_kbd_short_128
: ff_sine_128
;
1319 float * buf
= ac
->buf_mdct
;
1320 DECLARE_ALIGNED(16, float, temp
[128]);
1324 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1325 if (ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
)
1326 av_log(ac
->avccontext
, AV_LOG_WARNING
,
1327 "Transition from an ONLY_LONG or LONG_STOP to an EIGHT_SHORT sequence detected. "
1328 "If you heard an audible artifact, please submit the sample to the FFmpeg developers.\n");
1329 for (i
= 0; i
< 1024; i
+= 128)
1330 ff_imdct_half(&ac
->mdct_small
, buf
+ i
, in
+ i
);
1332 ff_imdct_half(&ac
->mdct
, buf
, in
);
1334 /* window overlapping
1335 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1336 * and long to short transitions are considered to be short to short
1337 * transitions. This leaves just two cases (long to long and short to short)
1338 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1340 if ((ics
->window_sequence
[1] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[1] == LONG_STOP_SEQUENCE
) &&
1341 (ics
->window_sequence
[0] == ONLY_LONG_SEQUENCE
|| ics
->window_sequence
[0] == LONG_START_SEQUENCE
)) {
1342 ac
->dsp
.vector_fmul_window( out
, saved
, buf
, lwindow_prev
, ac
->add_bias
, 512);
1344 for (i
= 0; i
< 448; i
++)
1345 out
[i
] = saved
[i
] + ac
->add_bias
;
1347 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1348 ac
->dsp
.vector_fmul_window(out
+ 448 + 0*128, saved
+ 448, buf
+ 0*128, swindow_prev
, ac
->add_bias
, 64);
1349 ac
->dsp
.vector_fmul_window(out
+ 448 + 1*128, buf
+ 0*128 + 64, buf
+ 1*128, swindow
, ac
->add_bias
, 64);
1350 ac
->dsp
.vector_fmul_window(out
+ 448 + 2*128, buf
+ 1*128 + 64, buf
+ 2*128, swindow
, ac
->add_bias
, 64);
1351 ac
->dsp
.vector_fmul_window(out
+ 448 + 3*128, buf
+ 2*128 + 64, buf
+ 3*128, swindow
, ac
->add_bias
, 64);
1352 ac
->dsp
.vector_fmul_window(temp
, buf
+ 3*128 + 64, buf
+ 4*128, swindow
, ac
->add_bias
, 64);
1353 memcpy( out
+ 448 + 4*128, temp
, 64 * sizeof(float));
1355 ac
->dsp
.vector_fmul_window(out
+ 448, saved
+ 448, buf
, swindow_prev
, ac
->add_bias
, 64);
1356 for (i
= 576; i
< 1024; i
++)
1357 out
[i
] = buf
[i
-512] + ac
->add_bias
;
1362 if (ics
->window_sequence
[0] == EIGHT_SHORT_SEQUENCE
) {
1363 for (i
= 0; i
< 64; i
++)
1364 saved
[i
] = temp
[64 + i
] - ac
->add_bias
;
1365 ac
->dsp
.vector_fmul_window(saved
+ 64, buf
+ 4*128 + 64, buf
+ 5*128, swindow
, 0, 64);
1366 ac
->dsp
.vector_fmul_window(saved
+ 192, buf
+ 5*128 + 64, buf
+ 6*128, swindow
, 0, 64);
1367 ac
->dsp
.vector_fmul_window(saved
+ 320, buf
+ 6*128 + 64, buf
+ 7*128, swindow
, 0, 64);
1368 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1369 } else if (ics
->window_sequence
[0] == LONG_START_SEQUENCE
) {
1370 memcpy( saved
, buf
+ 512, 448 * sizeof(float));
1371 memcpy( saved
+ 448, buf
+ 7*128 + 64, 64 * sizeof(float));
1372 } else { // LONG_STOP or ONLY_LONG
1373 memcpy( saved
, buf
+ 512, 512 * sizeof(float));
1378 * Apply dependent channel coupling (applied before IMDCT).
1380 * @param index index into coupling gain array
1382 static void apply_dependent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1383 IndividualChannelStream
* ics
= &cce
->ch
[0].ics
;
1384 const uint16_t * offsets
= ics
->swb_offset
;
1385 float * dest
= target
->coeffs
;
1386 const float * src
= cce
->ch
[0].coeffs
;
1387 int g
, i
, group
, k
, idx
= 0;
1388 if(ac
->m4ac
.object_type
== AOT_AAC_LTP
) {
1389 av_log(ac
->avccontext
, AV_LOG_ERROR
,
1390 "Dependent coupling is not supported together with LTP\n");
1393 for (g
= 0; g
< ics
->num_window_groups
; g
++) {
1394 for (i
= 0; i
< ics
->max_sfb
; i
++, idx
++) {
1395 if (cce
->ch
[0].band_type
[idx
] != ZERO_BT
) {
1396 for (group
= 0; group
< ics
->group_len
[g
]; group
++) {
1397 for (k
= offsets
[i
]; k
< offsets
[i
+1]; k
++) {
1399 dest
[group
*128+k
] += cce
->coup
.gain
[index
][idx
] * src
[group
*128+k
];
1404 dest
+= ics
->group_len
[g
]*128;
1405 src
+= ics
->group_len
[g
]*128;
1410 * Apply independent channel coupling (applied after IMDCT).
1412 * @param index index into coupling gain array
1414 static void apply_independent_coupling(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
) {
1416 for (i
= 0; i
< 1024; i
++)
1417 target
->ret
[i
] += cce
->coup
.gain
[index
][0] * (cce
->ch
[0].ret
[i
] - ac
->add_bias
);
1421 * channel coupling transformation interface
1423 * @param index index into coupling gain array
1424 * @param apply_coupling_method pointer to (in)dependent coupling function
1426 static void apply_channel_coupling(AACContext
* ac
, ChannelElement
* cc
,
1427 enum RawDataBlockType type
, int elem_id
, enum CouplingPoint coupling_point
,
1428 void (*apply_coupling_method
)(AACContext
* ac
, SingleChannelElement
* target
, ChannelElement
* cce
, int index
))
1432 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1433 ChannelElement
*cce
= ac
->che
[TYPE_CCE
][i
];
1436 if (cce
&& cce
->coup
.coupling_point
== coupling_point
) {
1437 ChannelCoupling
* coup
= &cce
->coup
;
1439 for (c
= 0; c
<= coup
->num_coupled
; c
++) {
1440 if (coup
->type
[c
] == type
&& coup
->id_select
[c
] == elem_id
) {
1441 if (coup
->ch_select
[c
] != 1) {
1442 apply_coupling_method(ac
, &cc
->ch
[0], cce
, index
);
1443 if (coup
->ch_select
[c
] != 0)
1446 if (coup
->ch_select
[c
] != 2)
1447 apply_coupling_method(ac
, &cc
->ch
[1], cce
, index
++);
1449 index
+= 1 + (coup
->ch_select
[c
] == 3);
1456 * Convert spectral data to float samples, applying all supported tools as appropriate.
1458 static void spectral_to_sample(AACContext
* ac
) {
1460 for(type
= 3; type
>= 0; type
--) {
1461 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1462 ChannelElement
*che
= ac
->che
[type
][i
];
1464 if(type
<= TYPE_CPE
)
1465 apply_channel_coupling(ac
, che
, type
, i
, BEFORE_TNS
, apply_dependent_coupling
);
1466 if(che
->ch
[0].tns
.present
)
1467 apply_tns(che
->ch
[0].coeffs
, &che
->ch
[0].tns
, &che
->ch
[0].ics
, 1);
1468 if(che
->ch
[1].tns
.present
)
1469 apply_tns(che
->ch
[1].coeffs
, &che
->ch
[1].tns
, &che
->ch
[1].ics
, 1);
1470 if(type
<= TYPE_CPE
)
1471 apply_channel_coupling(ac
, che
, type
, i
, BETWEEN_TNS_AND_IMDCT
, apply_dependent_coupling
);
1472 if(type
!= TYPE_CCE
|| che
->coup
.coupling_point
== AFTER_IMDCT
)
1473 imdct_and_windowing(ac
, &che
->ch
[0]);
1474 if(type
== TYPE_CPE
)
1475 imdct_and_windowing(ac
, &che
->ch
[1]);
1476 if(type
<= TYPE_CCE
)
1477 apply_channel_coupling(ac
, che
, type
, i
, AFTER_IMDCT
, apply_independent_coupling
);
1483 static int aac_decode_frame(AVCodecContext
* avccontext
, void * data
, int * data_size
, const uint8_t * buf
, int buf_size
) {
1484 AACContext
* ac
= avccontext
->priv_data
;
1486 enum RawDataBlockType elem_type
;
1487 int err
, elem_id
, data_size_tmp
;
1489 init_get_bits(&gb
, buf
, buf_size
*8);
1492 while ((elem_type
= get_bits(&gb
, 3)) != TYPE_END
) {
1493 elem_id
= get_bits(&gb
, 4);
1496 if(elem_type
== TYPE_SCE
&& elem_id
== 1 &&
1497 !ac
->che
[TYPE_SCE
][elem_id
] && ac
->che
[TYPE_LFE
][0]) {
1498 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
1499 instead of SCE[0] CPE[0] CPE[0] LFE[0]. If we seem to have
1500 encountered such a stream, transfer the LFE[0] element to SCE[1] */
1501 ac
->che
[TYPE_SCE
][elem_id
] = ac
->che
[TYPE_LFE
][0];
1502 ac
->che
[TYPE_LFE
][0] = NULL
;
1504 if(elem_type
< TYPE_DSE
) {
1505 if(!ac
->che
[elem_type
][elem_id
])
1507 if(elem_type
!= TYPE_CCE
)
1508 ac
->che
[elem_type
][elem_id
]->coup
.coupling_point
= 4;
1511 switch (elem_type
) {
1514 err
= decode_ics(ac
, &ac
->che
[TYPE_SCE
][elem_id
]->ch
[0], &gb
, 0, 0);
1518 err
= decode_cpe(ac
, &gb
, elem_id
);
1522 err
= decode_cce(ac
, &gb
, ac
->che
[TYPE_CCE
][elem_id
]);
1526 err
= decode_ics(ac
, &ac
->che
[TYPE_LFE
][elem_id
]->ch
[0], &gb
, 0, 0);
1530 skip_data_stream_element(&gb
);
1536 enum ChannelPosition new_che_pos
[4][MAX_ELEM_ID
];
1537 memset(new_che_pos
, 0, 4 * MAX_ELEM_ID
* sizeof(new_che_pos
[0][0]));
1538 if((err
= decode_pce(ac
, new_che_pos
, &gb
)))
1540 err
= output_configure(ac
, ac
->che_pos
, new_che_pos
);
1546 elem_id
+= get_bits(&gb
, 8) - 1;
1548 elem_id
-= decode_extension_payload(ac
, &gb
, elem_id
);
1549 err
= 0; /* FIXME */
1553 err
= -1; /* should not happen, but keeps compiler happy */
1561 spectral_to_sample(ac
);
1563 if (!ac
->is_saved
) {
1569 data_size_tmp
= 1024 * avccontext
->channels
* sizeof(int16_t);
1570 if(*data_size
< data_size_tmp
) {
1571 av_log(avccontext
, AV_LOG_ERROR
,
1572 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
1573 *data_size
, data_size_tmp
);
1576 *data_size
= data_size_tmp
;
1578 ac
->dsp
.float_to_int16_interleave(data
, (const float **)ac
->output_data
, 1024, avccontext
->channels
);
1583 static av_cold
int aac_decode_close(AVCodecContext
* avccontext
) {
1584 AACContext
* ac
= avccontext
->priv_data
;
1587 for (i
= 0; i
< MAX_ELEM_ID
; i
++) {
1588 for(type
= 0; type
< 4; type
++)
1589 av_freep(&ac
->che
[type
][i
]);
1592 ff_mdct_end(&ac
->mdct
);
1593 ff_mdct_end(&ac
->mdct_small
);
1597 AVCodec aac_decoder
= {
1606 .long_name
= NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
1607 .sample_fmts
= (enum SampleFormat
[]){SAMPLE_FMT_S16
,SAMPLE_FMT_NONE
},