3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * N (code in SoC repo) Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
93 #include "aacdectab.h"
94 #include "cbrt_tablegen.h"
97 #include "mpeg4audio.h"
98 #include "aacadtsdec.h"
106 # include "arm/aac.h"
114 static VLC vlc_scalefactors;
115 static VLC vlc_spectral[11];
117 static const char overread_err[] = "Input buffer exhausted before END element found\n";
119 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
121 // For PCE based channel configurations map the channels solely based on tags.
122 if (!ac->m4ac.chan_config) {
123 return ac->tag_che_map[type][elem_id];
125 // For indexed channel configurations map the channels solely based on position.
126 switch (ac->m4ac.chan_config) {
128 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
130 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
133 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
134 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
135 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
136 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
138 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
141 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
143 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
146 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
148 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
152 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
154 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
155 } else if (ac->m4ac.chan_config == 2) {
159 if (!ac->tags_mapped && type == TYPE_SCE) {
161 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
169 * Check for the channel element in the current channel position configuration.
170 * If it exists, make sure the appropriate element is allocated and map the
171 * channel order to match the internal FFmpeg channel layout.
173 * @param che_pos current channel position configuration
174 * @param type channel element type
175 * @param id channel element id
176 * @param channels count of the number of channels in the configuration
178 * @return Returns error status. 0 - OK, !0 - error
180 static av_cold int che_configure(AACContext *ac,
181 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
185 if (che_pos[type][id]) {
186 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
187 return AVERROR(ENOMEM);
188 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
189 if (type != TYPE_CCE) {
190 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
191 if (type == TYPE_CPE ||
192 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
193 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
197 if (ac->che[type][id])
198 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
199 av_freep(&ac->che[type][id]);
205 * Configure output channel order based on the current program configuration element.
207 * @param che_pos current channel position configuration
208 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
210 * @return Returns error status. 0 - OK, !0 - error
212 static av_cold int output_configure(AACContext *ac,
213 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
214 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
215 int channel_config, enum OCStatus oc_type)
217 AVCodecContext *avctx = ac->avctx;
218 int i, type, channels = 0, ret;
220 if (new_che_pos != che_pos)
221 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
223 if (channel_config) {
224 for (i = 0; i < tags_per_config[channel_config]; i++) {
225 if ((ret = che_configure(ac, che_pos,
226 aac_channel_layout_map[channel_config - 1][i][0],
227 aac_channel_layout_map[channel_config - 1][i][1],
232 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
234 avctx->channel_layout = aac_channel_layout[channel_config - 1];
236 /* Allocate or free elements depending on if they are in the
237 * current program configuration.
239 * Set up default 1:1 output mapping.
241 * For a 5.1 stream the output order will be:
242 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
245 for (i = 0; i < MAX_ELEM_ID; i++) {
246 for (type = 0; type < 4; type++) {
247 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
252 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
254 avctx->channel_layout = 0;
257 avctx->channels = channels;
259 ac->output_configured = oc_type;
265 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
267 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
268 * @param sce_map mono (Single Channel Element) map
269 * @param type speaker type/position for these channels
271 static void decode_channel_map(enum ChannelPosition *cpe_map,
272 enum ChannelPosition *sce_map,
273 enum ChannelPosition type,
274 GetBitContext *gb, int n)
277 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
278 map[get_bits(gb, 4)] = type;
283 * Decode program configuration element; reference: table 4.2.
285 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
287 * @return Returns error status. 0 - OK, !0 - error
289 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
290 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
293 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
296 skip_bits(gb, 2); // object_type
298 sampling_index = get_bits(gb, 4);
299 if (m4ac->sampling_index != sampling_index)
300 av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
302 num_front = get_bits(gb, 4);
303 num_side = get_bits(gb, 4);
304 num_back = get_bits(gb, 4);
305 num_lfe = get_bits(gb, 2);
306 num_assoc_data = get_bits(gb, 3);
307 num_cc = get_bits(gb, 4);
310 skip_bits(gb, 4); // mono_mixdown_tag
312 skip_bits(gb, 4); // stereo_mixdown_tag
315 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
317 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
318 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
320 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
322 skip_bits_long(gb, 4 * num_assoc_data);
324 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
328 /* comment field, first byte is length */
329 comment_len = get_bits(gb, 8) * 8;
330 if (get_bits_left(gb) < comment_len) {
331 av_log(avctx, AV_LOG_ERROR, overread_err);
334 skip_bits_long(gb, comment_len);
339 * Set up channel positions based on a default channel configuration
340 * as specified in table 1.17.
342 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
344 * @return Returns error status. 0 - OK, !0 - error
346 static av_cold int set_default_channel_config(AVCodecContext *avctx,
347 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
350 if (channel_config < 1 || channel_config > 7) {
351 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
356 /* default channel configurations:
358 * 1ch : front center (mono)
359 * 2ch : L + R (stereo)
360 * 3ch : front center + L + R
361 * 4ch : front center + L + R + back center
362 * 5ch : front center + L + R + back stereo
363 * 6ch : front center + L + R + back stereo + LFE
364 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
367 if (channel_config != 2)
368 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
369 if (channel_config > 1)
370 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
371 if (channel_config == 4)
372 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
373 if (channel_config > 4)
374 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
375 = AAC_CHANNEL_BACK; // back stereo
376 if (channel_config > 5)
377 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
378 if (channel_config == 7)
379 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
385 * Decode GA "General Audio" specific configuration; reference: table 4.1.
387 * @param ac pointer to AACContext, may be null
388 * @param avctx pointer to AVCCodecContext, used for logging
390 * @return Returns error status. 0 - OK, !0 - error
392 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
394 MPEG4AudioConfig *m4ac,
397 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
398 int extension_flag, ret;
400 if (get_bits1(gb)) { // frameLengthFlag
401 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
405 if (get_bits1(gb)) // dependsOnCoreCoder
406 skip_bits(gb, 14); // coreCoderDelay
407 extension_flag = get_bits1(gb);
409 if (m4ac->object_type == AOT_AAC_SCALABLE ||
410 m4ac->object_type == AOT_ER_AAC_SCALABLE)
411 skip_bits(gb, 3); // layerNr
413 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
414 if (channel_config == 0) {
415 skip_bits(gb, 4); // element_instance_tag
416 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
419 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
422 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
425 if (extension_flag) {
426 switch (m4ac->object_type) {
428 skip_bits(gb, 5); // numOfSubFrame
429 skip_bits(gb, 11); // layer_length
433 case AOT_ER_AAC_SCALABLE:
435 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
436 * aacScalefactorDataResilienceFlag
437 * aacSpectralDataResilienceFlag
441 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
447 * Decode audio specific configuration; reference: table 1.13.
449 * @param ac pointer to AACContext, may be null
450 * @param avctx pointer to AVCCodecContext, used for logging
451 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
452 * @param data pointer to AVCodecContext extradata
453 * @param data_size size of AVCCodecContext extradata
455 * @return Returns error status or number of consumed bits. <0 - error
457 static int decode_audio_specific_config(AACContext *ac,
458 AVCodecContext *avctx,
459 MPEG4AudioConfig *m4ac,
460 const uint8_t *data, int data_size)
465 init_get_bits(&gb, data, data_size * 8);
467 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
469 if (m4ac->sampling_index > 12) {
470 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
473 if (m4ac->sbr == 1 && m4ac->ps == -1)
476 skip_bits_long(&gb, i);
478 switch (m4ac->object_type) {
481 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
485 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
486 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
490 return get_bits_count(&gb);
494 * linear congruential pseudorandom number generator
496 * @param previous_val pointer to the current state of the generator
498 * @return Returns a 32-bit pseudorandom integer
500 static av_always_inline int lcg_random(int previous_val)
502 return previous_val * 1664525 + 1013904223;
505 static av_always_inline void reset_predict_state(PredictorState *ps)
515 static void reset_all_predictors(PredictorState *ps)
518 for (i = 0; i < MAX_PREDICTORS; i++)
519 reset_predict_state(&ps[i]);
522 static void reset_predictor_group(PredictorState *ps, int group_num)
525 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
526 reset_predict_state(&ps[i]);
529 #define AAC_INIT_VLC_STATIC(num, size) \
530 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
531 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
532 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
535 static av_cold int aac_decode_init(AVCodecContext *avctx)
537 AACContext *ac = avctx->priv_data;
540 ac->m4ac.sample_rate = avctx->sample_rate;
542 if (avctx->extradata_size > 0) {
543 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
545 avctx->extradata_size) < 0)
549 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
551 AAC_INIT_VLC_STATIC( 0, 304);
552 AAC_INIT_VLC_STATIC( 1, 270);
553 AAC_INIT_VLC_STATIC( 2, 550);
554 AAC_INIT_VLC_STATIC( 3, 300);
555 AAC_INIT_VLC_STATIC( 4, 328);
556 AAC_INIT_VLC_STATIC( 5, 294);
557 AAC_INIT_VLC_STATIC( 6, 306);
558 AAC_INIT_VLC_STATIC( 7, 268);
559 AAC_INIT_VLC_STATIC( 8, 510);
560 AAC_INIT_VLC_STATIC( 9, 366);
561 AAC_INIT_VLC_STATIC(10, 462);
565 dsputil_init(&ac->dsp, avctx);
566 ff_fmt_convert_init(&ac->fmt_conv, avctx);
568 ac->random_state = 0x1f2e3d4c;
570 // -1024 - Compensate wrong IMDCT method.
571 // 60 - Required to scale values to the correct range [-32768,32767]
572 // for float to int16 conversion. (1 << (60 / 4)) == 32768
573 ac->sf_scale = 1. / -1024.;
578 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
579 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
580 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
583 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
584 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
585 // window initialization
586 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
587 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
588 ff_init_ff_sine_windows(10);
589 ff_init_ff_sine_windows( 7);
597 * Skip data_stream_element; reference: table 4.10.
599 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
601 int byte_align = get_bits1(gb);
602 int count = get_bits(gb, 8);
604 count += get_bits(gb, 8);
608 if (get_bits_left(gb) < 8 * count) {
609 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
612 skip_bits_long(gb, 8 * count);
616 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
621 ics->predictor_reset_group = get_bits(gb, 5);
622 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
623 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
627 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
628 ics->prediction_used[sfb] = get_bits1(gb);
634 * Decode Individual Channel Stream info; reference: table 4.6.
636 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
638 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
639 GetBitContext *gb, int common_window)
642 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
643 memset(ics, 0, sizeof(IndividualChannelStream));
646 ics->window_sequence[1] = ics->window_sequence[0];
647 ics->window_sequence[0] = get_bits(gb, 2);
648 ics->use_kb_window[1] = ics->use_kb_window[0];
649 ics->use_kb_window[0] = get_bits1(gb);
650 ics->num_window_groups = 1;
651 ics->group_len[0] = 1;
652 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
654 ics->max_sfb = get_bits(gb, 4);
655 for (i = 0; i < 7; i++) {
657 ics->group_len[ics->num_window_groups - 1]++;
659 ics->num_window_groups++;
660 ics->group_len[ics->num_window_groups - 1] = 1;
663 ics->num_windows = 8;
664 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
665 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
666 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
667 ics->predictor_present = 0;
669 ics->max_sfb = get_bits(gb, 6);
670 ics->num_windows = 1;
671 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
672 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
673 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
674 ics->predictor_present = get_bits1(gb);
675 ics->predictor_reset_group = 0;
676 if (ics->predictor_present) {
677 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
678 if (decode_prediction(ac, ics, gb)) {
679 memset(ics, 0, sizeof(IndividualChannelStream));
682 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
683 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
684 memset(ics, 0, sizeof(IndividualChannelStream));
687 av_log_missing_feature(ac->avctx, "Predictor bit set but LTP is", 1);
688 memset(ics, 0, sizeof(IndividualChannelStream));
694 if (ics->max_sfb > ics->num_swb) {
695 av_log(ac->avctx, AV_LOG_ERROR,
696 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
697 ics->max_sfb, ics->num_swb);
698 memset(ics, 0, sizeof(IndividualChannelStream));
706 * Decode band types (section_data payload); reference: table 4.46.
708 * @param band_type array of the used band type
709 * @param band_type_run_end array of the last scalefactor band of a band type run
711 * @return Returns error status. 0 - OK, !0 - error
713 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
714 int band_type_run_end[120], GetBitContext *gb,
715 IndividualChannelStream *ics)
718 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
719 for (g = 0; g < ics->num_window_groups; g++) {
721 while (k < ics->max_sfb) {
722 uint8_t sect_end = k;
724 int sect_band_type = get_bits(gb, 4);
725 if (sect_band_type == 12) {
726 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
729 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
730 sect_end += sect_len_incr;
731 sect_end += sect_len_incr;
732 if (get_bits_left(gb) < 0) {
733 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
736 if (sect_end > ics->max_sfb) {
737 av_log(ac->avctx, AV_LOG_ERROR,
738 "Number of bands (%d) exceeds limit (%d).\n",
739 sect_end, ics->max_sfb);
742 for (; k < sect_end; k++) {
743 band_type [idx] = sect_band_type;
744 band_type_run_end[idx++] = sect_end;
752 * Decode scalefactors; reference: table 4.47.
754 * @param global_gain first scalefactor value as scalefactors are differentially coded
755 * @param band_type array of the used band type
756 * @param band_type_run_end array of the last scalefactor band of a band type run
757 * @param sf array of scalefactors or intensity stereo positions
759 * @return Returns error status. 0 - OK, !0 - error
761 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
762 unsigned int global_gain,
763 IndividualChannelStream *ics,
764 enum BandType band_type[120],
765 int band_type_run_end[120])
767 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
769 int offset[3] = { global_gain, global_gain - 90, 100 };
771 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
772 for (g = 0; g < ics->num_window_groups; g++) {
773 for (i = 0; i < ics->max_sfb;) {
774 int run_end = band_type_run_end[idx];
775 if (band_type[idx] == ZERO_BT) {
776 for (; i < run_end; i++, idx++)
778 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
779 for (; i < run_end; i++, idx++) {
780 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
781 if (offset[2] > 255U) {
782 av_log(ac->avctx, AV_LOG_ERROR,
783 "%s (%d) out of range.\n", sf_str[2], offset[2]);
786 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
788 } else if (band_type[idx] == NOISE_BT) {
789 for (; i < run_end; i++, idx++) {
790 if (noise_flag-- > 0)
791 offset[1] += get_bits(gb, 9) - 256;
793 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
794 if (offset[1] > 255U) {
795 av_log(ac->avctx, AV_LOG_ERROR,
796 "%s (%d) out of range.\n", sf_str[1], offset[1]);
799 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
802 for (; i < run_end; i++, idx++) {
803 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
804 if (offset[0] > 255U) {
805 av_log(ac->avctx, AV_LOG_ERROR,
806 "%s (%d) out of range.\n", sf_str[0], offset[0]);
809 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
818 * Decode pulse data; reference: table 4.7.
820 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
821 const uint16_t *swb_offset, int num_swb)
824 pulse->num_pulse = get_bits(gb, 2) + 1;
825 pulse_swb = get_bits(gb, 6);
826 if (pulse_swb >= num_swb)
828 pulse->pos[0] = swb_offset[pulse_swb];
829 pulse->pos[0] += get_bits(gb, 5);
830 if (pulse->pos[0] > 1023)
832 pulse->amp[0] = get_bits(gb, 4);
833 for (i = 1; i < pulse->num_pulse; i++) {
834 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
835 if (pulse->pos[i] > 1023)
837 pulse->amp[i] = get_bits(gb, 4);
843 * Decode Temporal Noise Shaping data; reference: table 4.48.
845 * @return Returns error status. 0 - OK, !0 - error
847 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
848 GetBitContext *gb, const IndividualChannelStream *ics)
850 int w, filt, i, coef_len, coef_res, coef_compress;
851 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
852 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
853 for (w = 0; w < ics->num_windows; w++) {
854 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
855 coef_res = get_bits1(gb);
857 for (filt = 0; filt < tns->n_filt[w]; filt++) {
859 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
861 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
862 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
863 tns->order[w][filt], tns_max_order);
864 tns->order[w][filt] = 0;
867 if (tns->order[w][filt]) {
868 tns->direction[w][filt] = get_bits1(gb);
869 coef_compress = get_bits1(gb);
870 coef_len = coef_res + 3 - coef_compress;
871 tmp2_idx = 2 * coef_compress + coef_res;
873 for (i = 0; i < tns->order[w][filt]; i++)
874 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
883 * Decode Mid/Side data; reference: table 4.54.
885 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
886 * [1] mask is decoded from bitstream; [2] mask is all 1s;
887 * [3] reserved for scalable AAC
889 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
893 if (ms_present == 1) {
894 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
895 cpe->ms_mask[idx] = get_bits1(gb);
896 } else if (ms_present == 2) {
897 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
902 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
906 *dst++ = v[idx & 15] * s;
907 *dst++ = v[idx>>4 & 15] * s;
913 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
917 *dst++ = v[idx & 3] * s;
918 *dst++ = v[idx>>2 & 3] * s;
919 *dst++ = v[idx>>4 & 3] * s;
920 *dst++ = v[idx>>6 & 3] * s;
926 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
927 unsigned sign, const float *scale)
929 union float754 s0, s1;
931 s0.f = s1.f = *scale;
932 s0.i ^= sign >> 1 << 31;
935 *dst++ = v[idx & 15] * s0.f;
936 *dst++ = v[idx>>4 & 15] * s1.f;
943 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
944 unsigned sign, const float *scale)
946 unsigned nz = idx >> 12;
947 union float754 s = { .f = *scale };
950 t.i = s.i ^ (sign & 1<<31);
951 *dst++ = v[idx & 3] * t.f;
953 sign <<= nz & 1; nz >>= 1;
954 t.i = s.i ^ (sign & 1<<31);
955 *dst++ = v[idx>>2 & 3] * t.f;
957 sign <<= nz & 1; nz >>= 1;
958 t.i = s.i ^ (sign & 1<<31);
959 *dst++ = v[idx>>4 & 3] * t.f;
961 sign <<= nz & 1; nz >>= 1;
962 t.i = s.i ^ (sign & 1<<31);
963 *dst++ = v[idx>>6 & 3] * t.f;
970 * Decode spectral data; reference: table 4.50.
971 * Dequantize and scale spectral data; reference: 4.6.3.3.
973 * @param coef array of dequantized, scaled spectral data
974 * @param sf array of scalefactors or intensity stereo positions
975 * @param pulse_present set if pulses are present
976 * @param pulse pointer to pulse data struct
977 * @param band_type array of the used band type
979 * @return Returns error status. 0 - OK, !0 - error
981 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
982 GetBitContext *gb, const float sf[120],
983 int pulse_present, const Pulse *pulse,
984 const IndividualChannelStream *ics,
985 enum BandType band_type[120])
987 int i, k, g, idx = 0;
988 const int c = 1024 / ics->num_windows;
989 const uint16_t *offsets = ics->swb_offset;
990 float *coef_base = coef;
992 for (g = 0; g < ics->num_windows; g++)
993 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
995 for (g = 0; g < ics->num_window_groups; g++) {
996 unsigned g_len = ics->group_len[g];
998 for (i = 0; i < ics->max_sfb; i++, idx++) {
999 const unsigned cbt_m1 = band_type[idx] - 1;
1000 float *cfo = coef + offsets[i];
1001 int off_len = offsets[i + 1] - offsets[i];
1004 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1005 for (group = 0; group < g_len; group++, cfo+=128) {
1006 memset(cfo, 0, off_len * sizeof(float));
1008 } else if (cbt_m1 == NOISE_BT - 1) {
1009 for (group = 0; group < g_len; group++, cfo+=128) {
1013 for (k = 0; k < off_len; k++) {
1014 ac->random_state = lcg_random(ac->random_state);
1015 cfo[k] = ac->random_state;
1018 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1019 scale = sf[idx] / sqrtf(band_energy);
1020 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1023 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1024 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1025 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1026 OPEN_READER(re, gb);
1028 switch (cbt_m1 >> 1) {
1030 for (group = 0; group < g_len; group++, cfo+=128) {
1038 UPDATE_CACHE(re, gb);
1039 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1040 cb_idx = cb_vector_idx[code];
1041 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1047 for (group = 0; group < g_len; group++, cfo+=128) {
1057 UPDATE_CACHE(re, gb);
1058 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1059 cb_idx = cb_vector_idx[code];
1060 nnz = cb_idx >> 8 & 15;
1061 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1062 LAST_SKIP_BITS(re, gb, nnz);
1063 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1069 for (group = 0; group < g_len; group++, cfo+=128) {
1077 UPDATE_CACHE(re, gb);
1078 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1079 cb_idx = cb_vector_idx[code];
1080 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1087 for (group = 0; group < g_len; group++, cfo+=128) {
1097 UPDATE_CACHE(re, gb);
1098 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1099 cb_idx = cb_vector_idx[code];
1100 nnz = cb_idx >> 8 & 15;
1101 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1102 LAST_SKIP_BITS(re, gb, nnz);
1103 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1109 for (group = 0; group < g_len; group++, cfo+=128) {
1111 uint32_t *icf = (uint32_t *) cf;
1121 UPDATE_CACHE(re, gb);
1122 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1130 cb_idx = cb_vector_idx[code];
1133 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1134 LAST_SKIP_BITS(re, gb, nnz);
1136 for (j = 0; j < 2; j++) {
1140 /* The total length of escape_sequence must be < 22 bits according
1141 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1142 UPDATE_CACHE(re, gb);
1143 b = GET_CACHE(re, gb);
1144 b = 31 - av_log2(~b);
1147 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1151 SKIP_BITS(re, gb, b + 1);
1153 n = (1 << b) + SHOW_UBITS(re, gb, b);
1154 LAST_SKIP_BITS(re, gb, b);
1155 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1158 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1159 *icf++ = (bits & 1<<31) | v;
1166 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1170 CLOSE_READER(re, gb);
1176 if (pulse_present) {
1178 for (i = 0; i < pulse->num_pulse; i++) {
1179 float co = coef_base[ pulse->pos[i] ];
1180 while (offsets[idx + 1] <= pulse->pos[i])
1182 if (band_type[idx] != NOISE_BT && sf[idx]) {
1183 float ico = -pulse->amp[i];
1186 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1188 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1195 static av_always_inline float flt16_round(float pf)
1199 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1203 static av_always_inline float flt16_even(float pf)
1207 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1211 static av_always_inline float flt16_trunc(float pf)
1215 pun.i &= 0xFFFF0000U;
1219 static av_always_inline void predict(PredictorState *ps, float *coef,
1220 float sf_scale, float inv_sf_scale,
1223 const float a = 0.953125; // 61.0 / 64
1224 const float alpha = 0.90625; // 29.0 / 32
1228 float r0 = ps->r0, r1 = ps->r1;
1229 float cor0 = ps->cor0, cor1 = ps->cor1;
1230 float var0 = ps->var0, var1 = ps->var1;
1232 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1233 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1235 pv = flt16_round(k1 * r0 + k2 * r1);
1237 *coef += pv * sf_scale;
1239 e0 = *coef * inv_sf_scale;
1242 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1243 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1244 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1245 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1247 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1248 ps->r0 = flt16_trunc(a * e0);
1252 * Apply AAC-Main style frequency domain prediction.
1254 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1257 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1259 if (!sce->ics.predictor_initialized) {
1260 reset_all_predictors(sce->predictor_state);
1261 sce->ics.predictor_initialized = 1;
1264 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1265 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1266 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1267 predict(&sce->predictor_state[k], &sce->coeffs[k],
1268 sf_scale, inv_sf_scale,
1269 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1272 if (sce->ics.predictor_reset_group)
1273 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1275 reset_all_predictors(sce->predictor_state);
1279 * Decode an individual_channel_stream payload; reference: table 4.44.
1281 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1282 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1284 * @return Returns error status. 0 - OK, !0 - error
1286 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1287 GetBitContext *gb, int common_window, int scale_flag)
1290 TemporalNoiseShaping *tns = &sce->tns;
1291 IndividualChannelStream *ics = &sce->ics;
1292 float *out = sce->coeffs;
1293 int global_gain, pulse_present = 0;
1295 /* This assignment is to silence a GCC warning about the variable being used
1296 * uninitialized when in fact it always is.
1298 pulse.num_pulse = 0;
1300 global_gain = get_bits(gb, 8);
1302 if (!common_window && !scale_flag) {
1303 if (decode_ics_info(ac, ics, gb, 0) < 0)
1307 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1309 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1314 if ((pulse_present = get_bits1(gb))) {
1315 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1316 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1319 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1320 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1324 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1326 if (get_bits1(gb)) {
1327 av_log_missing_feature(ac->avctx, "SSR", 1);
1332 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1335 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1336 apply_prediction(ac, sce);
1342 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1344 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1346 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1347 float *ch0 = cpe->ch[0].coeffs;
1348 float *ch1 = cpe->ch[1].coeffs;
1349 int g, i, group, idx = 0;
1350 const uint16_t *offsets = ics->swb_offset;
1351 for (g = 0; g < ics->num_window_groups; g++) {
1352 for (i = 0; i < ics->max_sfb; i++, idx++) {
1353 if (cpe->ms_mask[idx] &&
1354 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1355 for (group = 0; group < ics->group_len[g]; group++) {
1356 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1357 ch1 + group * 128 + offsets[i],
1358 offsets[i+1] - offsets[i]);
1362 ch0 += ics->group_len[g] * 128;
1363 ch1 += ics->group_len[g] * 128;
1368 * intensity stereo decoding; reference: 4.6.8.2.3
1370 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1371 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1372 * [3] reserved for scalable AAC
1374 static void apply_intensity_stereo(ChannelElement *cpe, int ms_present)
1376 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1377 SingleChannelElement *sce1 = &cpe->ch[1];
1378 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1379 const uint16_t *offsets = ics->swb_offset;
1380 int g, group, i, k, idx = 0;
1383 for (g = 0; g < ics->num_window_groups; g++) {
1384 for (i = 0; i < ics->max_sfb;) {
1385 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1386 const int bt_run_end = sce1->band_type_run_end[idx];
1387 for (; i < bt_run_end; i++, idx++) {
1388 c = -1 + 2 * (sce1->band_type[idx] - 14);
1390 c *= 1 - 2 * cpe->ms_mask[idx];
1391 scale = c * sce1->sf[idx];
1392 for (group = 0; group < ics->group_len[g]; group++)
1393 for (k = offsets[i]; k < offsets[i + 1]; k++)
1394 coef1[group * 128 + k] = scale * coef0[group * 128 + k];
1397 int bt_run_end = sce1->band_type_run_end[idx];
1398 idx += bt_run_end - i;
1402 coef0 += ics->group_len[g] * 128;
1403 coef1 += ics->group_len[g] * 128;
1408 * Decode a channel_pair_element; reference: table 4.4.
1410 * @return Returns error status. 0 - OK, !0 - error
1412 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1414 int i, ret, common_window, ms_present = 0;
1416 common_window = get_bits1(gb);
1417 if (common_window) {
1418 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1420 i = cpe->ch[1].ics.use_kb_window[0];
1421 cpe->ch[1].ics = cpe->ch[0].ics;
1422 cpe->ch[1].ics.use_kb_window[1] = i;
1423 ms_present = get_bits(gb, 2);
1424 if (ms_present == 3) {
1425 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1427 } else if (ms_present)
1428 decode_mid_side_stereo(cpe, gb, ms_present);
1430 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1432 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1435 if (common_window) {
1437 apply_mid_side_stereo(ac, cpe);
1438 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1439 apply_prediction(ac, &cpe->ch[0]);
1440 apply_prediction(ac, &cpe->ch[1]);
1444 apply_intensity_stereo(cpe, ms_present);
1448 static const float cce_scale[] = {
1449 1.09050773266525765921, //2^(1/8)
1450 1.18920711500272106672, //2^(1/4)
1456 * Decode coupling_channel_element; reference: table 4.8.
1458 * @return Returns error status. 0 - OK, !0 - error
1460 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1466 SingleChannelElement *sce = &che->ch[0];
1467 ChannelCoupling *coup = &che->coup;
1469 coup->coupling_point = 2 * get_bits1(gb);
1470 coup->num_coupled = get_bits(gb, 3);
1471 for (c = 0; c <= coup->num_coupled; c++) {
1473 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1474 coup->id_select[c] = get_bits(gb, 4);
1475 if (coup->type[c] == TYPE_CPE) {
1476 coup->ch_select[c] = get_bits(gb, 2);
1477 if (coup->ch_select[c] == 3)
1480 coup->ch_select[c] = 2;
1482 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1484 sign = get_bits(gb, 1);
1485 scale = cce_scale[get_bits(gb, 2)];
1487 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1490 for (c = 0; c < num_gain; c++) {
1494 float gain_cache = 1.;
1496 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1497 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1498 gain_cache = powf(scale, -gain);
1500 if (coup->coupling_point == AFTER_IMDCT) {
1501 coup->gain[c][0] = gain_cache;
1503 for (g = 0; g < sce->ics.num_window_groups; g++) {
1504 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1505 if (sce->band_type[idx] != ZERO_BT) {
1507 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1515 gain_cache = powf(scale, -t) * s;
1518 coup->gain[c][idx] = gain_cache;
1528 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1530 * @return Returns number of bytes consumed.
1532 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1536 int num_excl_chan = 0;
1539 for (i = 0; i < 7; i++)
1540 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1541 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1543 return num_excl_chan / 7;
1547 * Decode dynamic range information; reference: table 4.52.
1549 * @param cnt length of TYPE_FIL syntactic element in bytes
1551 * @return Returns number of bytes consumed.
1553 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1554 GetBitContext *gb, int cnt)
1557 int drc_num_bands = 1;
1560 /* pce_tag_present? */
1561 if (get_bits1(gb)) {
1562 che_drc->pce_instance_tag = get_bits(gb, 4);
1563 skip_bits(gb, 4); // tag_reserved_bits
1567 /* excluded_chns_present? */
1568 if (get_bits1(gb)) {
1569 n += decode_drc_channel_exclusions(che_drc, gb);
1572 /* drc_bands_present? */
1573 if (get_bits1(gb)) {
1574 che_drc->band_incr = get_bits(gb, 4);
1575 che_drc->interpolation_scheme = get_bits(gb, 4);
1577 drc_num_bands += che_drc->band_incr;
1578 for (i = 0; i < drc_num_bands; i++) {
1579 che_drc->band_top[i] = get_bits(gb, 8);
1584 /* prog_ref_level_present? */
1585 if (get_bits1(gb)) {
1586 che_drc->prog_ref_level = get_bits(gb, 7);
1587 skip_bits1(gb); // prog_ref_level_reserved_bits
1591 for (i = 0; i < drc_num_bands; i++) {
1592 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1593 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1601 * Decode extension data (incomplete); reference: table 4.51.
1603 * @param cnt length of TYPE_FIL syntactic element in bytes
1605 * @return Returns number of bytes consumed
1607 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1608 ChannelElement *che, enum RawDataBlockType elem_type)
1612 switch (get_bits(gb, 4)) { // extension type
1613 case EXT_SBR_DATA_CRC:
1617 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1619 } else if (!ac->m4ac.sbr) {
1620 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1621 skip_bits_long(gb, 8 * cnt - 4);
1623 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1624 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1625 skip_bits_long(gb, 8 * cnt - 4);
1627 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1630 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1634 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1636 case EXT_DYNAMIC_RANGE:
1637 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1641 case EXT_DATA_ELEMENT:
1643 skip_bits_long(gb, 8 * cnt - 4);
1650 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1652 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1653 * @param coef spectral coefficients
1655 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1656 IndividualChannelStream *ics, int decode)
1658 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1660 int bottom, top, order, start, end, size, inc;
1661 float lpc[TNS_MAX_ORDER];
1663 for (w = 0; w < ics->num_windows; w++) {
1664 bottom = ics->num_swb;
1665 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1667 bottom = FFMAX(0, top - tns->length[w][filt]);
1668 order = tns->order[w][filt];
1673 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1675 start = ics->swb_offset[FFMIN(bottom, mmm)];
1676 end = ics->swb_offset[FFMIN( top, mmm)];
1677 if ((size = end - start) <= 0)
1679 if (tns->direction[w][filt]) {
1688 for (m = 0; m < size; m++, start += inc)
1689 for (i = 1; i <= FFMIN(m, order); i++)
1690 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1696 * Conduct IMDCT and windowing.
1698 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1700 IndividualChannelStream *ics = &sce->ics;
1701 float *in = sce->coeffs;
1702 float *out = sce->ret;
1703 float *saved = sce->saved;
1704 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1705 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1706 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1707 float *buf = ac->buf_mdct;
1708 float *temp = ac->temp;
1712 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1713 for (i = 0; i < 1024; i += 128)
1714 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1716 ff_imdct_half(&ac->mdct, buf, in);
1718 /* window overlapping
1719 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1720 * and long to short transitions are considered to be short to short
1721 * transitions. This leaves just two cases (long to long and short to short)
1722 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1724 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1725 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1726 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1728 memcpy( out, saved, 448 * sizeof(float));
1730 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1731 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1732 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1733 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1734 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1735 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1736 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1738 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1739 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1744 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1745 memcpy( saved, temp + 64, 64 * sizeof(float));
1746 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1747 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1748 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1749 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1750 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1751 memcpy( saved, buf + 512, 448 * sizeof(float));
1752 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1753 } else { // LONG_STOP or ONLY_LONG
1754 memcpy( saved, buf + 512, 512 * sizeof(float));
1759 * Apply dependent channel coupling (applied before IMDCT).
1761 * @param index index into coupling gain array
1763 static void apply_dependent_coupling(AACContext *ac,
1764 SingleChannelElement *target,
1765 ChannelElement *cce, int index)
1767 IndividualChannelStream *ics = &cce->ch[0].ics;
1768 const uint16_t *offsets = ics->swb_offset;
1769 float *dest = target->coeffs;
1770 const float *src = cce->ch[0].coeffs;
1771 int g, i, group, k, idx = 0;
1772 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1773 av_log(ac->avctx, AV_LOG_ERROR,
1774 "Dependent coupling is not supported together with LTP\n");
1777 for (g = 0; g < ics->num_window_groups; g++) {
1778 for (i = 0; i < ics->max_sfb; i++, idx++) {
1779 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1780 const float gain = cce->coup.gain[index][idx];
1781 for (group = 0; group < ics->group_len[g]; group++) {
1782 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1784 dest[group * 128 + k] += gain * src[group * 128 + k];
1789 dest += ics->group_len[g] * 128;
1790 src += ics->group_len[g] * 128;
1795 * Apply independent channel coupling (applied after IMDCT).
1797 * @param index index into coupling gain array
1799 static void apply_independent_coupling(AACContext *ac,
1800 SingleChannelElement *target,
1801 ChannelElement *cce, int index)
1804 const float gain = cce->coup.gain[index][0];
1805 const float *src = cce->ch[0].ret;
1806 float *dest = target->ret;
1807 const int len = 1024 << (ac->m4ac.sbr == 1);
1809 for (i = 0; i < len; i++)
1810 dest[i] += gain * src[i];
1814 * channel coupling transformation interface
1816 * @param apply_coupling_method pointer to (in)dependent coupling function
1818 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1819 enum RawDataBlockType type, int elem_id,
1820 enum CouplingPoint coupling_point,
1821 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1825 for (i = 0; i < MAX_ELEM_ID; i++) {
1826 ChannelElement *cce = ac->che[TYPE_CCE][i];
1829 if (cce && cce->coup.coupling_point == coupling_point) {
1830 ChannelCoupling *coup = &cce->coup;
1832 for (c = 0; c <= coup->num_coupled; c++) {
1833 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1834 if (coup->ch_select[c] != 1) {
1835 apply_coupling_method(ac, &cc->ch[0], cce, index);
1836 if (coup->ch_select[c] != 0)
1839 if (coup->ch_select[c] != 2)
1840 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1842 index += 1 + (coup->ch_select[c] == 3);
1849 * Convert spectral data to float samples, applying all supported tools as appropriate.
1851 static void spectral_to_sample(AACContext *ac)
1854 for (type = 3; type >= 0; type--) {
1855 for (i = 0; i < MAX_ELEM_ID; i++) {
1856 ChannelElement *che = ac->che[type][i];
1858 if (type <= TYPE_CPE)
1859 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1860 if (che->ch[0].tns.present)
1861 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1862 if (che->ch[1].tns.present)
1863 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
1864 if (type <= TYPE_CPE)
1865 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
1866 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
1867 imdct_and_windowing(ac, &che->ch[0]);
1868 if (type == TYPE_CPE) {
1869 imdct_and_windowing(ac, &che->ch[1]);
1871 if (ac->m4ac.sbr > 0) {
1872 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
1875 if (type <= TYPE_CCE)
1876 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
1882 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
1885 AACADTSHeaderInfo hdr_info;
1887 size = ff_aac_parse_header(gb, &hdr_info);
1889 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
1890 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1891 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1892 ac->m4ac.chan_config = hdr_info.chan_config;
1893 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
1895 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
1897 } else if (ac->output_configured != OC_LOCKED) {
1898 ac->output_configured = OC_NONE;
1900 if (ac->output_configured != OC_LOCKED) {
1904 ac->m4ac.sample_rate = hdr_info.sample_rate;
1905 ac->m4ac.sampling_index = hdr_info.sampling_index;
1906 ac->m4ac.object_type = hdr_info.object_type;
1907 if (!ac->avctx->sample_rate)
1908 ac->avctx->sample_rate = hdr_info.sample_rate;
1909 if (hdr_info.num_aac_frames == 1) {
1910 if (!hdr_info.crc_absent)
1913 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
1920 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
1921 int *data_size, GetBitContext *gb)
1923 AACContext *ac = avctx->priv_data;
1924 ChannelElement *che = NULL, *che_prev = NULL;
1925 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
1926 int err, elem_id, data_size_tmp;
1927 int samples = 0, multiplier;
1929 if (show_bits(gb, 12) == 0xfff) {
1930 if (parse_adts_frame_header(ac, gb) < 0) {
1931 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
1934 if (ac->m4ac.sampling_index > 12) {
1935 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
1940 ac->tags_mapped = 0;
1942 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
1943 elem_id = get_bits(gb, 4);
1945 if (elem_type < TYPE_DSE) {
1946 if (!(che=get_che(ac, elem_type, elem_id))) {
1947 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
1948 elem_type, elem_id);
1954 switch (elem_type) {
1957 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
1961 err = decode_cpe(ac, gb, che);
1965 err = decode_cce(ac, gb, che);
1969 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
1973 err = skip_data_stream_element(ac, gb);
1977 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
1978 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
1979 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
1981 if (ac->output_configured > OC_TRIAL_PCE)
1982 av_log(avctx, AV_LOG_ERROR,
1983 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
1985 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
1991 elem_id += get_bits(gb, 8) - 1;
1992 if (get_bits_left(gb) < 8 * elem_id) {
1993 av_log(avctx, AV_LOG_ERROR, overread_err);
1997 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
1998 err = 0; /* FIXME */
2002 err = -1; /* should not happen, but keeps compiler happy */
2007 elem_type_prev = elem_type;
2012 if (get_bits_left(gb) < 3) {
2013 av_log(avctx, AV_LOG_ERROR, overread_err);
2018 spectral_to_sample(ac);
2020 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2021 samples <<= multiplier;
2022 if (ac->output_configured < OC_LOCKED) {
2023 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2024 avctx->frame_size = samples;
2027 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2028 if (*data_size < data_size_tmp) {
2029 av_log(avctx, AV_LOG_ERROR,
2030 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2031 *data_size, data_size_tmp);
2034 *data_size = data_size_tmp;
2037 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2039 if (ac->output_configured)
2040 ac->output_configured = OC_LOCKED;
2045 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2046 int *data_size, AVPacket *avpkt)
2048 const uint8_t *buf = avpkt->data;
2049 int buf_size = avpkt->size;
2055 init_get_bits(&gb, buf, buf_size * 8);
2057 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2060 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2061 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2062 if (buf[buf_offset])
2065 return buf_size > buf_offset ? buf_consumed : buf_size;
2068 static av_cold int aac_decode_close(AVCodecContext *avctx)
2070 AACContext *ac = avctx->priv_data;
2073 for (i = 0; i < MAX_ELEM_ID; i++) {
2074 for (type = 0; type < 4; type++) {
2075 if (ac->che[type][i])
2076 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2077 av_freep(&ac->che[type][i]);
2081 ff_mdct_end(&ac->mdct);
2082 ff_mdct_end(&ac->mdct_small);
2087 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2089 struct LATMContext {
2090 AACContext aac_ctx; ///< containing AACContext
2091 int initialized; ///< initilized after a valid extradata was seen
2094 int audio_mux_version_A; ///< LATM syntax version
2095 int frame_length_type; ///< 0/1 variable/fixed frame length
2096 int frame_length; ///< frame length for fixed frame length
2099 static inline uint32_t latm_get_value(GetBitContext *b)
2101 int length = get_bits(b, 2);
2103 return get_bits_long(b, (length+1)*8);
2106 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2109 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2110 MPEG4AudioConfig m4ac;
2111 int config_start_bit = get_bits_count(gb);
2112 int bits_consumed, esize;
2114 if (config_start_bit % 8) {
2115 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2116 "config not byte aligned.\n", 1);
2117 return AVERROR_INVALIDDATA;
2120 decode_audio_specific_config(NULL, avctx, &m4ac,
2121 gb->buffer + (config_start_bit / 8),
2122 get_bits_left(gb) / 8);
2124 if (bits_consumed < 0)
2125 return AVERROR_INVALIDDATA;
2127 esize = (bits_consumed+7) / 8;
2129 if (avctx->extradata_size <= esize) {
2130 av_free(avctx->extradata);
2131 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2132 if (!avctx->extradata)
2133 return AVERROR(ENOMEM);
2136 avctx->extradata_size = esize;
2137 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2138 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2140 skip_bits_long(gb, bits_consumed);
2143 return bits_consumed;
2146 static int read_stream_mux_config(struct LATMContext *latmctx,
2149 int ret, audio_mux_version = get_bits(gb, 1);
2151 latmctx->audio_mux_version_A = 0;
2152 if (audio_mux_version)
2153 latmctx->audio_mux_version_A = get_bits(gb, 1);
2155 if (!latmctx->audio_mux_version_A) {
2157 if (audio_mux_version)
2158 latm_get_value(gb); // taraFullness
2160 skip_bits(gb, 1); // allStreamSameTimeFraming
2161 skip_bits(gb, 6); // numSubFrames
2163 if (get_bits(gb, 4)) { // numPrograms
2164 av_log_missing_feature(latmctx->aac_ctx.avctx,
2165 "multiple programs are not supported\n", 1);
2166 return AVERROR_PATCHWELCOME;
2169 // for each program (which there is only on in DVB)
2171 // for each layer (which there is only on in DVB)
2172 if (get_bits(gb, 3)) { // numLayer
2173 av_log_missing_feature(latmctx->aac_ctx.avctx,
2174 "multiple layers are not supported\n", 1);
2175 return AVERROR_PATCHWELCOME;
2178 // for all but first stream: use_same_config = get_bits(gb, 1);
2179 if (!audio_mux_version) {
2180 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2183 int ascLen = latm_get_value(gb);
2184 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2187 skip_bits_long(gb, ascLen);
2190 latmctx->frame_length_type = get_bits(gb, 3);
2191 switch (latmctx->frame_length_type) {
2193 skip_bits(gb, 8); // latmBufferFullness
2196 latmctx->frame_length = get_bits(gb, 9);
2201 skip_bits(gb, 6); // CELP frame length table index
2205 skip_bits(gb, 1); // HVXC frame length table index
2209 if (get_bits(gb, 1)) { // other data
2210 if (audio_mux_version) {
2211 latm_get_value(gb); // other_data_bits
2215 esc = get_bits(gb, 1);
2221 if (get_bits(gb, 1)) // crc present
2222 skip_bits(gb, 8); // config_crc
2228 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2232 if (ctx->frame_length_type == 0) {
2233 int mux_slot_length = 0;
2235 tmp = get_bits(gb, 8);
2236 mux_slot_length += tmp;
2237 } while (tmp == 255);
2238 return mux_slot_length;
2239 } else if (ctx->frame_length_type == 1) {
2240 return ctx->frame_length;
2241 } else if (ctx->frame_length_type == 3 ||
2242 ctx->frame_length_type == 5 ||
2243 ctx->frame_length_type == 7) {
2244 skip_bits(gb, 2); // mux_slot_length_coded
2249 static int read_audio_mux_element(struct LATMContext *latmctx,
2253 uint8_t use_same_mux = get_bits(gb, 1);
2254 if (!use_same_mux) {
2255 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2257 } else if (!latmctx->aac_ctx.avctx->extradata) {
2258 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2259 "no decoder config found\n");
2260 return AVERROR(EAGAIN);
2262 if (latmctx->audio_mux_version_A == 0) {
2263 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2264 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2265 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2266 return AVERROR_INVALIDDATA;
2267 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2268 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2269 "frame length mismatch %d << %d\n",
2270 mux_slot_length_bytes * 8, get_bits_left(gb));
2271 return AVERROR_INVALIDDATA;
2278 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2281 struct LATMContext *latmctx = avctx->priv_data;
2285 if (avpkt->size == 0)
2288 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2290 // check for LOAS sync word
2291 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2292 return AVERROR_INVALIDDATA;
2294 muxlength = get_bits(&gb, 13) + 3;
2295 // not enough data, the parser should have sorted this
2296 if (muxlength > avpkt->size)
2297 return AVERROR_INVALIDDATA;
2299 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2302 if (!latmctx->initialized) {
2303 if (!avctx->extradata) {
2307 if ((err = aac_decode_init(avctx)) < 0)
2309 latmctx->initialized = 1;
2313 if (show_bits(&gb, 12) == 0xfff) {
2314 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2315 "ADTS header detected, probably as result of configuration "
2317 return AVERROR_INVALIDDATA;
2320 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2326 av_cold static int latm_decode_init(AVCodecContext *avctx)
2328 struct LATMContext *latmctx = avctx->priv_data;
2331 ret = aac_decode_init(avctx);
2333 if (avctx->extradata_size > 0) {
2334 latmctx->initialized = !ret;
2336 latmctx->initialized = 0;
2343 AVCodec ff_aac_decoder = {
2352 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2353 .sample_fmts = (const enum AVSampleFormat[]) {
2354 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2356 .channel_layouts = aac_channel_layout,
2360 Note: This decoder filter is intended to decode LATM streams transferred
2361 in MPEG transport streams which only contain one program.
2362 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2364 AVCodec ff_aac_latm_decoder = {
2366 .type = CODEC_TYPE_AUDIO,
2367 .id = CODEC_ID_AAC_LATM,
2368 .priv_data_size = sizeof(struct LATMContext),
2369 .init = latm_decode_init,
2370 .close = aac_decode_close,
2371 .decode = latm_decode_frame,
2372 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2373 .sample_fmts = (const enum AVSampleFormat[]) {
2374 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2376 .channel_layouts = aac_channel_layout,