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 * Y 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) {
482 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
486 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
487 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
491 return get_bits_count(&gb);
495 * linear congruential pseudorandom number generator
497 * @param previous_val pointer to the current state of the generator
499 * @return Returns a 32-bit pseudorandom integer
501 static av_always_inline int lcg_random(int previous_val)
503 return previous_val * 1664525 + 1013904223;
506 static av_always_inline void reset_predict_state(PredictorState *ps)
516 static void reset_all_predictors(PredictorState *ps)
519 for (i = 0; i < MAX_PREDICTORS; i++)
520 reset_predict_state(&ps[i]);
523 static void reset_predictor_group(PredictorState *ps, int group_num)
526 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
527 reset_predict_state(&ps[i]);
530 #define AAC_INIT_VLC_STATIC(num, size) \
531 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
532 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
533 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
536 static av_cold int aac_decode_init(AVCodecContext *avctx)
538 AACContext *ac = avctx->priv_data;
541 ac->m4ac.sample_rate = avctx->sample_rate;
543 if (avctx->extradata_size > 0) {
544 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
546 avctx->extradata_size) < 0)
550 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
552 AAC_INIT_VLC_STATIC( 0, 304);
553 AAC_INIT_VLC_STATIC( 1, 270);
554 AAC_INIT_VLC_STATIC( 2, 550);
555 AAC_INIT_VLC_STATIC( 3, 300);
556 AAC_INIT_VLC_STATIC( 4, 328);
557 AAC_INIT_VLC_STATIC( 5, 294);
558 AAC_INIT_VLC_STATIC( 6, 306);
559 AAC_INIT_VLC_STATIC( 7, 268);
560 AAC_INIT_VLC_STATIC( 8, 510);
561 AAC_INIT_VLC_STATIC( 9, 366);
562 AAC_INIT_VLC_STATIC(10, 462);
566 dsputil_init(&ac->dsp, avctx);
567 ff_fmt_convert_init(&ac->fmt_conv, avctx);
569 ac->random_state = 0x1f2e3d4c;
571 // -1024 - Compensate wrong IMDCT method.
572 // 60 - Required to scale values to the correct range [-32768,32767]
573 // for float to int16 conversion. (1 << (60 / 4)) == 32768
574 ac->sf_scale = 1. / -1024.;
579 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
580 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
581 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
584 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
585 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
586 ff_mdct_init(&ac->mdct_ltp, 11, 0, 1.0);
587 // window initialization
588 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
589 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
590 ff_init_ff_sine_windows(10);
591 ff_init_ff_sine_windows( 7);
599 * Skip data_stream_element; reference: table 4.10.
601 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
603 int byte_align = get_bits1(gb);
604 int count = get_bits(gb, 8);
606 count += get_bits(gb, 8);
610 if (get_bits_left(gb) < 8 * count) {
611 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
614 skip_bits_long(gb, 8 * count);
618 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
623 ics->predictor_reset_group = get_bits(gb, 5);
624 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
625 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
629 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
630 ics->prediction_used[sfb] = get_bits1(gb);
636 * Decode Long Term Prediction data; reference: table 4.xx.
638 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
639 GetBitContext *gb, uint8_t max_sfb)
643 ltp->lag = get_bits(gb, 11);
644 ltp->coef = ltp_coef[get_bits(gb, 3)] * ac->sf_scale;
645 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
646 ltp->used[sfb] = get_bits1(gb);
650 * Decode Individual Channel Stream info; reference: table 4.6.
652 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
654 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
655 GetBitContext *gb, int common_window)
658 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
659 memset(ics, 0, sizeof(IndividualChannelStream));
662 ics->window_sequence[1] = ics->window_sequence[0];
663 ics->window_sequence[0] = get_bits(gb, 2);
664 ics->use_kb_window[1] = ics->use_kb_window[0];
665 ics->use_kb_window[0] = get_bits1(gb);
666 ics->num_window_groups = 1;
667 ics->group_len[0] = 1;
668 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
670 ics->max_sfb = get_bits(gb, 4);
671 for (i = 0; i < 7; i++) {
673 ics->group_len[ics->num_window_groups - 1]++;
675 ics->num_window_groups++;
676 ics->group_len[ics->num_window_groups - 1] = 1;
679 ics->num_windows = 8;
680 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
681 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
682 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
683 ics->predictor_present = 0;
685 ics->max_sfb = get_bits(gb, 6);
686 ics->num_windows = 1;
687 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
688 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
689 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
690 ics->predictor_present = get_bits1(gb);
691 ics->predictor_reset_group = 0;
692 if (ics->predictor_present) {
693 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
694 if (decode_prediction(ac, ics, gb)) {
695 memset(ics, 0, sizeof(IndividualChannelStream));
698 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
699 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
700 memset(ics, 0, sizeof(IndividualChannelStream));
703 if ((ics->ltp.present = get_bits(gb, 1)))
704 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
709 if (ics->max_sfb > ics->num_swb) {
710 av_log(ac->avctx, AV_LOG_ERROR,
711 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
712 ics->max_sfb, ics->num_swb);
713 memset(ics, 0, sizeof(IndividualChannelStream));
721 * Decode band types (section_data payload); reference: table 4.46.
723 * @param band_type array of the used band type
724 * @param band_type_run_end array of the last scalefactor band of a band type run
726 * @return Returns error status. 0 - OK, !0 - error
728 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
729 int band_type_run_end[120], GetBitContext *gb,
730 IndividualChannelStream *ics)
733 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
734 for (g = 0; g < ics->num_window_groups; g++) {
736 while (k < ics->max_sfb) {
737 uint8_t sect_end = k;
739 int sect_band_type = get_bits(gb, 4);
740 if (sect_band_type == 12) {
741 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
744 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
745 sect_end += sect_len_incr;
746 sect_end += sect_len_incr;
747 if (get_bits_left(gb) < 0) {
748 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
751 if (sect_end > ics->max_sfb) {
752 av_log(ac->avctx, AV_LOG_ERROR,
753 "Number of bands (%d) exceeds limit (%d).\n",
754 sect_end, ics->max_sfb);
757 for (; k < sect_end; k++) {
758 band_type [idx] = sect_band_type;
759 band_type_run_end[idx++] = sect_end;
767 * Decode scalefactors; reference: table 4.47.
769 * @param global_gain first scalefactor value as scalefactors are differentially coded
770 * @param band_type array of the used band type
771 * @param band_type_run_end array of the last scalefactor band of a band type run
772 * @param sf array of scalefactors or intensity stereo positions
774 * @return Returns error status. 0 - OK, !0 - error
776 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
777 unsigned int global_gain,
778 IndividualChannelStream *ics,
779 enum BandType band_type[120],
780 int band_type_run_end[120])
782 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
784 int offset[3] = { global_gain, global_gain - 90, 100 };
786 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
787 for (g = 0; g < ics->num_window_groups; g++) {
788 for (i = 0; i < ics->max_sfb;) {
789 int run_end = band_type_run_end[idx];
790 if (band_type[idx] == ZERO_BT) {
791 for (; i < run_end; i++, idx++)
793 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
794 for (; i < run_end; i++, idx++) {
795 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
796 if (offset[2] > 255U) {
797 av_log(ac->avctx, AV_LOG_ERROR,
798 "%s (%d) out of range.\n", sf_str[2], offset[2]);
801 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
803 } else if (band_type[idx] == NOISE_BT) {
804 for (; i < run_end; i++, idx++) {
805 if (noise_flag-- > 0)
806 offset[1] += get_bits(gb, 9) - 256;
808 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
809 if (offset[1] > 255U) {
810 av_log(ac->avctx, AV_LOG_ERROR,
811 "%s (%d) out of range.\n", sf_str[1], offset[1]);
814 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
817 for (; i < run_end; i++, idx++) {
818 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
819 if (offset[0] > 255U) {
820 av_log(ac->avctx, AV_LOG_ERROR,
821 "%s (%d) out of range.\n", sf_str[0], offset[0]);
824 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
833 * Decode pulse data; reference: table 4.7.
835 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
836 const uint16_t *swb_offset, int num_swb)
839 pulse->num_pulse = get_bits(gb, 2) + 1;
840 pulse_swb = get_bits(gb, 6);
841 if (pulse_swb >= num_swb)
843 pulse->pos[0] = swb_offset[pulse_swb];
844 pulse->pos[0] += get_bits(gb, 5);
845 if (pulse->pos[0] > 1023)
847 pulse->amp[0] = get_bits(gb, 4);
848 for (i = 1; i < pulse->num_pulse; i++) {
849 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
850 if (pulse->pos[i] > 1023)
852 pulse->amp[i] = get_bits(gb, 4);
858 * Decode Temporal Noise Shaping data; reference: table 4.48.
860 * @return Returns error status. 0 - OK, !0 - error
862 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
863 GetBitContext *gb, const IndividualChannelStream *ics)
865 int w, filt, i, coef_len, coef_res, coef_compress;
866 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
867 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
868 for (w = 0; w < ics->num_windows; w++) {
869 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
870 coef_res = get_bits1(gb);
872 for (filt = 0; filt < tns->n_filt[w]; filt++) {
874 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
876 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
877 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
878 tns->order[w][filt], tns_max_order);
879 tns->order[w][filt] = 0;
882 if (tns->order[w][filt]) {
883 tns->direction[w][filt] = get_bits1(gb);
884 coef_compress = get_bits1(gb);
885 coef_len = coef_res + 3 - coef_compress;
886 tmp2_idx = 2 * coef_compress + coef_res;
888 for (i = 0; i < tns->order[w][filt]; i++)
889 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
898 * Decode Mid/Side data; reference: table 4.54.
900 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
901 * [1] mask is decoded from bitstream; [2] mask is all 1s;
902 * [3] reserved for scalable AAC
904 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
908 if (ms_present == 1) {
909 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
910 cpe->ms_mask[idx] = get_bits1(gb);
911 } else if (ms_present == 2) {
912 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
917 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
921 *dst++ = v[idx & 15] * s;
922 *dst++ = v[idx>>4 & 15] * s;
928 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
932 *dst++ = v[idx & 3] * s;
933 *dst++ = v[idx>>2 & 3] * s;
934 *dst++ = v[idx>>4 & 3] * s;
935 *dst++ = v[idx>>6 & 3] * s;
941 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
942 unsigned sign, const float *scale)
944 union float754 s0, s1;
946 s0.f = s1.f = *scale;
947 s0.i ^= sign >> 1 << 31;
950 *dst++ = v[idx & 15] * s0.f;
951 *dst++ = v[idx>>4 & 15] * s1.f;
958 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
959 unsigned sign, const float *scale)
961 unsigned nz = idx >> 12;
962 union float754 s = { .f = *scale };
965 t.i = s.i ^ (sign & 1<<31);
966 *dst++ = v[idx & 3] * t.f;
968 sign <<= nz & 1; nz >>= 1;
969 t.i = s.i ^ (sign & 1<<31);
970 *dst++ = v[idx>>2 & 3] * t.f;
972 sign <<= nz & 1; nz >>= 1;
973 t.i = s.i ^ (sign & 1<<31);
974 *dst++ = v[idx>>4 & 3] * t.f;
976 sign <<= nz & 1; nz >>= 1;
977 t.i = s.i ^ (sign & 1<<31);
978 *dst++ = v[idx>>6 & 3] * t.f;
985 * Decode spectral data; reference: table 4.50.
986 * Dequantize and scale spectral data; reference: 4.6.3.3.
988 * @param coef array of dequantized, scaled spectral data
989 * @param sf array of scalefactors or intensity stereo positions
990 * @param pulse_present set if pulses are present
991 * @param pulse pointer to pulse data struct
992 * @param band_type array of the used band type
994 * @return Returns error status. 0 - OK, !0 - error
996 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
997 GetBitContext *gb, const float sf[120],
998 int pulse_present, const Pulse *pulse,
999 const IndividualChannelStream *ics,
1000 enum BandType band_type[120])
1002 int i, k, g, idx = 0;
1003 const int c = 1024 / ics->num_windows;
1004 const uint16_t *offsets = ics->swb_offset;
1005 float *coef_base = coef;
1007 for (g = 0; g < ics->num_windows; g++)
1008 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1010 for (g = 0; g < ics->num_window_groups; g++) {
1011 unsigned g_len = ics->group_len[g];
1013 for (i = 0; i < ics->max_sfb; i++, idx++) {
1014 const unsigned cbt_m1 = band_type[idx] - 1;
1015 float *cfo = coef + offsets[i];
1016 int off_len = offsets[i + 1] - offsets[i];
1019 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1020 for (group = 0; group < g_len; group++, cfo+=128) {
1021 memset(cfo, 0, off_len * sizeof(float));
1023 } else if (cbt_m1 == NOISE_BT - 1) {
1024 for (group = 0; group < g_len; group++, cfo+=128) {
1028 for (k = 0; k < off_len; k++) {
1029 ac->random_state = lcg_random(ac->random_state);
1030 cfo[k] = ac->random_state;
1033 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1034 scale = sf[idx] / sqrtf(band_energy);
1035 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1038 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1039 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1040 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1041 OPEN_READER(re, gb);
1043 switch (cbt_m1 >> 1) {
1045 for (group = 0; group < g_len; group++, cfo+=128) {
1053 UPDATE_CACHE(re, gb);
1054 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1055 cb_idx = cb_vector_idx[code];
1056 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1062 for (group = 0; group < g_len; group++, cfo+=128) {
1072 UPDATE_CACHE(re, gb);
1073 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1074 cb_idx = cb_vector_idx[code];
1075 nnz = cb_idx >> 8 & 15;
1076 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1077 LAST_SKIP_BITS(re, gb, nnz);
1078 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1084 for (group = 0; group < g_len; group++, cfo+=128) {
1092 UPDATE_CACHE(re, gb);
1093 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1094 cb_idx = cb_vector_idx[code];
1095 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1102 for (group = 0; group < g_len; group++, cfo+=128) {
1112 UPDATE_CACHE(re, gb);
1113 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1114 cb_idx = cb_vector_idx[code];
1115 nnz = cb_idx >> 8 & 15;
1116 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1117 LAST_SKIP_BITS(re, gb, nnz);
1118 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1124 for (group = 0; group < g_len; group++, cfo+=128) {
1126 uint32_t *icf = (uint32_t *) cf;
1136 UPDATE_CACHE(re, gb);
1137 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1145 cb_idx = cb_vector_idx[code];
1148 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1149 LAST_SKIP_BITS(re, gb, nnz);
1151 for (j = 0; j < 2; j++) {
1155 /* The total length of escape_sequence must be < 22 bits according
1156 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1157 UPDATE_CACHE(re, gb);
1158 b = GET_CACHE(re, gb);
1159 b = 31 - av_log2(~b);
1162 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1166 SKIP_BITS(re, gb, b + 1);
1168 n = (1 << b) + SHOW_UBITS(re, gb, b);
1169 LAST_SKIP_BITS(re, gb, b);
1170 *icf++ = cbrt_tab[n] | (bits & 1<<31);
1173 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1174 *icf++ = (bits & 1<<31) | v;
1181 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1185 CLOSE_READER(re, gb);
1191 if (pulse_present) {
1193 for (i = 0; i < pulse->num_pulse; i++) {
1194 float co = coef_base[ pulse->pos[i] ];
1195 while (offsets[idx + 1] <= pulse->pos[i])
1197 if (band_type[idx] != NOISE_BT && sf[idx]) {
1198 float ico = -pulse->amp[i];
1201 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1203 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1210 static av_always_inline float flt16_round(float pf)
1214 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1218 static av_always_inline float flt16_even(float pf)
1222 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1226 static av_always_inline float flt16_trunc(float pf)
1230 pun.i &= 0xFFFF0000U;
1234 static av_always_inline void predict(PredictorState *ps, float *coef,
1235 float sf_scale, float inv_sf_scale,
1238 const float a = 0.953125; // 61.0 / 64
1239 const float alpha = 0.90625; // 29.0 / 32
1243 float r0 = ps->r0, r1 = ps->r1;
1244 float cor0 = ps->cor0, cor1 = ps->cor1;
1245 float var0 = ps->var0, var1 = ps->var1;
1247 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1248 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1250 pv = flt16_round(k1 * r0 + k2 * r1);
1252 *coef += pv * sf_scale;
1254 e0 = *coef * inv_sf_scale;
1257 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1258 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1259 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1260 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1262 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1263 ps->r0 = flt16_trunc(a * e0);
1267 * Apply AAC-Main style frequency domain prediction.
1269 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1272 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1274 if (!sce->ics.predictor_initialized) {
1275 reset_all_predictors(sce->predictor_state);
1276 sce->ics.predictor_initialized = 1;
1279 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1280 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1281 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1282 predict(&sce->predictor_state[k], &sce->coeffs[k],
1283 sf_scale, inv_sf_scale,
1284 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1287 if (sce->ics.predictor_reset_group)
1288 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1290 reset_all_predictors(sce->predictor_state);
1294 * Decode an individual_channel_stream payload; reference: table 4.44.
1296 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1297 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1299 * @return Returns error status. 0 - OK, !0 - error
1301 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1302 GetBitContext *gb, int common_window, int scale_flag)
1305 TemporalNoiseShaping *tns = &sce->tns;
1306 IndividualChannelStream *ics = &sce->ics;
1307 float *out = sce->coeffs;
1308 int global_gain, pulse_present = 0;
1310 /* This assignment is to silence a GCC warning about the variable being used
1311 * uninitialized when in fact it always is.
1313 pulse.num_pulse = 0;
1315 global_gain = get_bits(gb, 8);
1317 if (!common_window && !scale_flag) {
1318 if (decode_ics_info(ac, ics, gb, 0) < 0)
1322 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1324 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1329 if ((pulse_present = get_bits1(gb))) {
1330 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1331 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1334 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1335 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1339 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1341 if (get_bits1(gb)) {
1342 av_log_missing_feature(ac->avctx, "SSR", 1);
1347 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1350 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1351 apply_prediction(ac, sce);
1357 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1359 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1361 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1362 float *ch0 = cpe->ch[0].coeffs;
1363 float *ch1 = cpe->ch[1].coeffs;
1364 int g, i, group, idx = 0;
1365 const uint16_t *offsets = ics->swb_offset;
1366 for (g = 0; g < ics->num_window_groups; g++) {
1367 for (i = 0; i < ics->max_sfb; i++, idx++) {
1368 if (cpe->ms_mask[idx] &&
1369 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1370 for (group = 0; group < ics->group_len[g]; group++) {
1371 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1372 ch1 + group * 128 + offsets[i],
1373 offsets[i+1] - offsets[i]);
1377 ch0 += ics->group_len[g] * 128;
1378 ch1 += ics->group_len[g] * 128;
1383 * intensity stereo decoding; reference: 4.6.8.2.3
1385 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1386 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1387 * [3] reserved for scalable AAC
1389 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1391 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1392 SingleChannelElement *sce1 = &cpe->ch[1];
1393 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1394 const uint16_t *offsets = ics->swb_offset;
1395 int g, group, i, idx = 0;
1398 for (g = 0; g < ics->num_window_groups; g++) {
1399 for (i = 0; i < ics->max_sfb;) {
1400 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1401 const int bt_run_end = sce1->band_type_run_end[idx];
1402 for (; i < bt_run_end; i++, idx++) {
1403 c = -1 + 2 * (sce1->band_type[idx] - 14);
1405 c *= 1 - 2 * cpe->ms_mask[idx];
1406 scale = c * sce1->sf[idx];
1407 for (group = 0; group < ics->group_len[g]; group++)
1408 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1409 coef0 + group * 128 + offsets[i],
1411 offsets[i + 1] - offsets[i]);
1414 int bt_run_end = sce1->band_type_run_end[idx];
1415 idx += bt_run_end - i;
1419 coef0 += ics->group_len[g] * 128;
1420 coef1 += ics->group_len[g] * 128;
1425 * Decode a channel_pair_element; reference: table 4.4.
1427 * @return Returns error status. 0 - OK, !0 - error
1429 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1431 int i, ret, common_window, ms_present = 0;
1433 common_window = get_bits1(gb);
1434 if (common_window) {
1435 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1437 i = cpe->ch[1].ics.use_kb_window[0];
1438 cpe->ch[1].ics = cpe->ch[0].ics;
1439 cpe->ch[1].ics.use_kb_window[1] = i;
1440 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1441 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1442 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1443 ms_present = get_bits(gb, 2);
1444 if (ms_present == 3) {
1445 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1447 } else if (ms_present)
1448 decode_mid_side_stereo(cpe, gb, ms_present);
1450 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1452 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1455 if (common_window) {
1457 apply_mid_side_stereo(ac, cpe);
1458 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1459 apply_prediction(ac, &cpe->ch[0]);
1460 apply_prediction(ac, &cpe->ch[1]);
1464 apply_intensity_stereo(ac, cpe, ms_present);
1468 static const float cce_scale[] = {
1469 1.09050773266525765921, //2^(1/8)
1470 1.18920711500272106672, //2^(1/4)
1476 * Decode coupling_channel_element; reference: table 4.8.
1478 * @return Returns error status. 0 - OK, !0 - error
1480 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1486 SingleChannelElement *sce = &che->ch[0];
1487 ChannelCoupling *coup = &che->coup;
1489 coup->coupling_point = 2 * get_bits1(gb);
1490 coup->num_coupled = get_bits(gb, 3);
1491 for (c = 0; c <= coup->num_coupled; c++) {
1493 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1494 coup->id_select[c] = get_bits(gb, 4);
1495 if (coup->type[c] == TYPE_CPE) {
1496 coup->ch_select[c] = get_bits(gb, 2);
1497 if (coup->ch_select[c] == 3)
1500 coup->ch_select[c] = 2;
1502 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1504 sign = get_bits(gb, 1);
1505 scale = cce_scale[get_bits(gb, 2)];
1507 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1510 for (c = 0; c < num_gain; c++) {
1514 float gain_cache = 1.;
1516 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1517 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1518 gain_cache = powf(scale, -gain);
1520 if (coup->coupling_point == AFTER_IMDCT) {
1521 coup->gain[c][0] = gain_cache;
1523 for (g = 0; g < sce->ics.num_window_groups; g++) {
1524 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1525 if (sce->band_type[idx] != ZERO_BT) {
1527 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1535 gain_cache = powf(scale, -t) * s;
1538 coup->gain[c][idx] = gain_cache;
1548 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1550 * @return Returns number of bytes consumed.
1552 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1556 int num_excl_chan = 0;
1559 for (i = 0; i < 7; i++)
1560 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1561 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1563 return num_excl_chan / 7;
1567 * Decode dynamic range information; reference: table 4.52.
1569 * @param cnt length of TYPE_FIL syntactic element in bytes
1571 * @return Returns number of bytes consumed.
1573 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1574 GetBitContext *gb, int cnt)
1577 int drc_num_bands = 1;
1580 /* pce_tag_present? */
1581 if (get_bits1(gb)) {
1582 che_drc->pce_instance_tag = get_bits(gb, 4);
1583 skip_bits(gb, 4); // tag_reserved_bits
1587 /* excluded_chns_present? */
1588 if (get_bits1(gb)) {
1589 n += decode_drc_channel_exclusions(che_drc, gb);
1592 /* drc_bands_present? */
1593 if (get_bits1(gb)) {
1594 che_drc->band_incr = get_bits(gb, 4);
1595 che_drc->interpolation_scheme = get_bits(gb, 4);
1597 drc_num_bands += che_drc->band_incr;
1598 for (i = 0; i < drc_num_bands; i++) {
1599 che_drc->band_top[i] = get_bits(gb, 8);
1604 /* prog_ref_level_present? */
1605 if (get_bits1(gb)) {
1606 che_drc->prog_ref_level = get_bits(gb, 7);
1607 skip_bits1(gb); // prog_ref_level_reserved_bits
1611 for (i = 0; i < drc_num_bands; i++) {
1612 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1613 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1621 * Decode extension data (incomplete); reference: table 4.51.
1623 * @param cnt length of TYPE_FIL syntactic element in bytes
1625 * @return Returns number of bytes consumed
1627 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1628 ChannelElement *che, enum RawDataBlockType elem_type)
1632 switch (get_bits(gb, 4)) { // extension type
1633 case EXT_SBR_DATA_CRC:
1637 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1639 } else if (!ac->m4ac.sbr) {
1640 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1641 skip_bits_long(gb, 8 * cnt - 4);
1643 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1644 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1645 skip_bits_long(gb, 8 * cnt - 4);
1647 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1650 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1654 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1656 case EXT_DYNAMIC_RANGE:
1657 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1661 case EXT_DATA_ELEMENT:
1663 skip_bits_long(gb, 8 * cnt - 4);
1670 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1672 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1673 * @param coef spectral coefficients
1675 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1676 IndividualChannelStream *ics, int decode)
1678 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1680 int bottom, top, order, start, end, size, inc;
1681 float lpc[TNS_MAX_ORDER];
1682 float tmp[TNS_MAX_ORDER];
1684 for (w = 0; w < ics->num_windows; w++) {
1685 bottom = ics->num_swb;
1686 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1688 bottom = FFMAX(0, top - tns->length[w][filt]);
1689 order = tns->order[w][filt];
1694 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1696 start = ics->swb_offset[FFMIN(bottom, mmm)];
1697 end = ics->swb_offset[FFMIN( top, mmm)];
1698 if ((size = end - start) <= 0)
1700 if (tns->direction[w][filt]) {
1710 for (m = 0; m < size; m++, start += inc)
1711 for (i = 1; i <= FFMIN(m, order); i++)
1712 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1715 for (m = 0; m < size; m++, start += inc) {
1716 tmp[0] = coef[start];
1717 for (i = 1; i <= FFMIN(m, order); i++)
1718 coef[start] += tmp[i] * lpc[i - 1];
1719 for (i = order; i > 0; i--)
1720 tmp[i] = tmp[i - 1];
1728 * Apply windowing and MDCT to obtain the spectral
1729 * coefficient from the predicted sample by LTP.
1731 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1732 float *in, IndividualChannelStream *ics)
1734 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1735 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1736 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1737 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1739 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1740 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1742 memset(in, 0, 448 * sizeof(float));
1743 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1744 memcpy(in + 576, in + 576, 448 * sizeof(float));
1746 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1747 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1749 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1750 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1751 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1753 ff_mdct_calc(&ac->mdct_ltp, out, in);
1757 * Apply the long term prediction
1759 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1761 const LongTermPrediction *ltp = &sce->ics.ltp;
1762 const uint16_t *offsets = sce->ics.swb_offset;
1765 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1766 float *predTime = sce->ret;
1767 float *predFreq = ac->buf_mdct;
1768 int16_t num_samples = 2048;
1770 if (ltp->lag < 1024)
1771 num_samples = ltp->lag + 1024;
1772 for (i = 0; i < num_samples; i++)
1773 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1774 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1776 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1778 if (sce->tns.present)
1779 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1781 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1783 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1784 sce->coeffs[i] += predFreq[i];
1789 * Update the LTP buffer for next frame
1791 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1793 IndividualChannelStream *ics = &sce->ics;
1794 float *saved = sce->saved;
1795 float *saved_ltp = sce->coeffs;
1796 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1797 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1800 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1801 memcpy(saved_ltp, saved, 512 * sizeof(float));
1802 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1803 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1804 for (i = 0; i < 64; i++)
1805 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1806 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1807 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1808 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1809 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1810 for (i = 0; i < 64; i++)
1811 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1812 } else { // LONG_STOP or ONLY_LONG
1813 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1814 for (i = 0; i < 512; i++)
1815 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1818 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1819 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1820 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1824 * Conduct IMDCT and windowing.
1826 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1828 IndividualChannelStream *ics = &sce->ics;
1829 float *in = sce->coeffs;
1830 float *out = sce->ret;
1831 float *saved = sce->saved;
1832 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1833 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1834 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1835 float *buf = ac->buf_mdct;
1836 float *temp = ac->temp;
1840 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1841 for (i = 0; i < 1024; i += 128)
1842 ff_imdct_half(&ac->mdct_small, buf + i, in + i);
1844 ff_imdct_half(&ac->mdct, buf, in);
1846 /* window overlapping
1847 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1848 * and long to short transitions are considered to be short to short
1849 * transitions. This leaves just two cases (long to long and short to short)
1850 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1852 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1853 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1854 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1856 memcpy( out, saved, 448 * sizeof(float));
1858 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1859 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1860 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1861 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1862 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1863 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1864 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1866 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1867 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1872 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1873 memcpy( saved, temp + 64, 64 * sizeof(float));
1874 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1875 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1876 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1877 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1878 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1879 memcpy( saved, buf + 512, 448 * sizeof(float));
1880 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1881 } else { // LONG_STOP or ONLY_LONG
1882 memcpy( saved, buf + 512, 512 * sizeof(float));
1887 * Apply dependent channel coupling (applied before IMDCT).
1889 * @param index index into coupling gain array
1891 static void apply_dependent_coupling(AACContext *ac,
1892 SingleChannelElement *target,
1893 ChannelElement *cce, int index)
1895 IndividualChannelStream *ics = &cce->ch[0].ics;
1896 const uint16_t *offsets = ics->swb_offset;
1897 float *dest = target->coeffs;
1898 const float *src = cce->ch[0].coeffs;
1899 int g, i, group, k, idx = 0;
1900 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1901 av_log(ac->avctx, AV_LOG_ERROR,
1902 "Dependent coupling is not supported together with LTP\n");
1905 for (g = 0; g < ics->num_window_groups; g++) {
1906 for (i = 0; i < ics->max_sfb; i++, idx++) {
1907 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1908 const float gain = cce->coup.gain[index][idx];
1909 for (group = 0; group < ics->group_len[g]; group++) {
1910 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1912 dest[group * 128 + k] += gain * src[group * 128 + k];
1917 dest += ics->group_len[g] * 128;
1918 src += ics->group_len[g] * 128;
1923 * Apply independent channel coupling (applied after IMDCT).
1925 * @param index index into coupling gain array
1927 static void apply_independent_coupling(AACContext *ac,
1928 SingleChannelElement *target,
1929 ChannelElement *cce, int index)
1932 const float gain = cce->coup.gain[index][0];
1933 const float *src = cce->ch[0].ret;
1934 float *dest = target->ret;
1935 const int len = 1024 << (ac->m4ac.sbr == 1);
1937 for (i = 0; i < len; i++)
1938 dest[i] += gain * src[i];
1942 * channel coupling transformation interface
1944 * @param apply_coupling_method pointer to (in)dependent coupling function
1946 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1947 enum RawDataBlockType type, int elem_id,
1948 enum CouplingPoint coupling_point,
1949 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1953 for (i = 0; i < MAX_ELEM_ID; i++) {
1954 ChannelElement *cce = ac->che[TYPE_CCE][i];
1957 if (cce && cce->coup.coupling_point == coupling_point) {
1958 ChannelCoupling *coup = &cce->coup;
1960 for (c = 0; c <= coup->num_coupled; c++) {
1961 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1962 if (coup->ch_select[c] != 1) {
1963 apply_coupling_method(ac, &cc->ch[0], cce, index);
1964 if (coup->ch_select[c] != 0)
1967 if (coup->ch_select[c] != 2)
1968 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1970 index += 1 + (coup->ch_select[c] == 3);
1977 * Convert spectral data to float samples, applying all supported tools as appropriate.
1979 static void spectral_to_sample(AACContext *ac)
1982 for (type = 3; type >= 0; type--) {
1983 for (i = 0; i < MAX_ELEM_ID; i++) {
1984 ChannelElement *che = ac->che[type][i];
1986 if (type <= TYPE_CPE)
1987 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1988 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1989 if (che->ch[0].ics.predictor_present) {
1990 if (che->ch[0].ics.ltp.present)
1991 apply_ltp(ac, &che->ch[0]);
1992 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
1993 apply_ltp(ac, &che->ch[1]);
1996 if (che->ch[0].tns.present)
1997 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
1998 if (che->ch[1].tns.present)
1999 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2000 if (type <= TYPE_CPE)
2001 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2002 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2003 imdct_and_windowing(ac, &che->ch[0]);
2004 if (ac->m4ac.object_type == AOT_AAC_LTP)
2005 update_ltp(ac, &che->ch[0]);
2006 if (type == TYPE_CPE) {
2007 imdct_and_windowing(ac, &che->ch[1]);
2008 if (ac->m4ac.object_type == AOT_AAC_LTP)
2009 update_ltp(ac, &che->ch[1]);
2011 if (ac->m4ac.sbr > 0) {
2012 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2015 if (type <= TYPE_CCE)
2016 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2022 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2025 AACADTSHeaderInfo hdr_info;
2027 size = ff_aac_parse_header(gb, &hdr_info);
2029 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2030 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2031 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2032 ac->m4ac.chan_config = hdr_info.chan_config;
2033 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2035 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2037 } else if (ac->output_configured != OC_LOCKED) {
2038 ac->output_configured = OC_NONE;
2040 if (ac->output_configured != OC_LOCKED) {
2044 ac->m4ac.sample_rate = hdr_info.sample_rate;
2045 ac->m4ac.sampling_index = hdr_info.sampling_index;
2046 ac->m4ac.object_type = hdr_info.object_type;
2047 if (!ac->avctx->sample_rate)
2048 ac->avctx->sample_rate = hdr_info.sample_rate;
2049 if (hdr_info.num_aac_frames == 1) {
2050 if (!hdr_info.crc_absent)
2053 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2060 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2061 int *data_size, GetBitContext *gb)
2063 AACContext *ac = avctx->priv_data;
2064 ChannelElement *che = NULL, *che_prev = NULL;
2065 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2066 int err, elem_id, data_size_tmp;
2067 int samples = 0, multiplier;
2069 if (show_bits(gb, 12) == 0xfff) {
2070 if (parse_adts_frame_header(ac, gb) < 0) {
2071 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2074 if (ac->m4ac.sampling_index > 12) {
2075 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2080 ac->tags_mapped = 0;
2082 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2083 elem_id = get_bits(gb, 4);
2085 if (elem_type < TYPE_DSE) {
2086 if (!(che=get_che(ac, elem_type, elem_id))) {
2087 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2088 elem_type, elem_id);
2094 switch (elem_type) {
2097 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2101 err = decode_cpe(ac, gb, che);
2105 err = decode_cce(ac, gb, che);
2109 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2113 err = skip_data_stream_element(ac, gb);
2117 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2118 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2119 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2121 if (ac->output_configured > OC_TRIAL_PCE)
2122 av_log(avctx, AV_LOG_ERROR,
2123 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2125 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2131 elem_id += get_bits(gb, 8) - 1;
2132 if (get_bits_left(gb) < 8 * elem_id) {
2133 av_log(avctx, AV_LOG_ERROR, overread_err);
2137 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2138 err = 0; /* FIXME */
2142 err = -1; /* should not happen, but keeps compiler happy */
2147 elem_type_prev = elem_type;
2152 if (get_bits_left(gb) < 3) {
2153 av_log(avctx, AV_LOG_ERROR, overread_err);
2158 spectral_to_sample(ac);
2160 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2161 samples <<= multiplier;
2162 if (ac->output_configured < OC_LOCKED) {
2163 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2164 avctx->frame_size = samples;
2167 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2168 if (*data_size < data_size_tmp) {
2169 av_log(avctx, AV_LOG_ERROR,
2170 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2171 *data_size, data_size_tmp);
2174 *data_size = data_size_tmp;
2177 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2179 if (ac->output_configured)
2180 ac->output_configured = OC_LOCKED;
2185 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2186 int *data_size, AVPacket *avpkt)
2188 const uint8_t *buf = avpkt->data;
2189 int buf_size = avpkt->size;
2195 init_get_bits(&gb, buf, buf_size * 8);
2197 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2200 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2201 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2202 if (buf[buf_offset])
2205 return buf_size > buf_offset ? buf_consumed : buf_size;
2208 static av_cold int aac_decode_close(AVCodecContext *avctx)
2210 AACContext *ac = avctx->priv_data;
2213 for (i = 0; i < MAX_ELEM_ID; i++) {
2214 for (type = 0; type < 4; type++) {
2215 if (ac->che[type][i])
2216 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2217 av_freep(&ac->che[type][i]);
2221 ff_mdct_end(&ac->mdct);
2222 ff_mdct_end(&ac->mdct_small);
2223 ff_mdct_end(&ac->mdct_ltp);
2228 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2230 struct LATMContext {
2231 AACContext aac_ctx; ///< containing AACContext
2232 int initialized; ///< initilized after a valid extradata was seen
2235 int audio_mux_version_A; ///< LATM syntax version
2236 int frame_length_type; ///< 0/1 variable/fixed frame length
2237 int frame_length; ///< frame length for fixed frame length
2240 static inline uint32_t latm_get_value(GetBitContext *b)
2242 int length = get_bits(b, 2);
2244 return get_bits_long(b, (length+1)*8);
2247 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2250 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2251 int config_start_bit = get_bits_count(gb);
2252 int bits_consumed, esize;
2254 if (config_start_bit % 8) {
2255 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2256 "config not byte aligned.\n", 1);
2257 return AVERROR_INVALIDDATA;
2260 decode_audio_specific_config(&latmctx->aac_ctx, avctx,
2261 &latmctx->aac_ctx.m4ac,
2262 gb->buffer + (config_start_bit / 8),
2263 get_bits_left(gb) / 8);
2265 if (bits_consumed < 0)
2266 return AVERROR_INVALIDDATA;
2268 esize = (bits_consumed+7) / 8;
2270 if (avctx->extradata_size <= esize) {
2271 av_free(avctx->extradata);
2272 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2273 if (!avctx->extradata)
2274 return AVERROR(ENOMEM);
2277 avctx->extradata_size = esize;
2278 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2279 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2281 skip_bits_long(gb, bits_consumed);
2284 return bits_consumed;
2287 static int read_stream_mux_config(struct LATMContext *latmctx,
2290 int ret, audio_mux_version = get_bits(gb, 1);
2292 latmctx->audio_mux_version_A = 0;
2293 if (audio_mux_version)
2294 latmctx->audio_mux_version_A = get_bits(gb, 1);
2296 if (!latmctx->audio_mux_version_A) {
2298 if (audio_mux_version)
2299 latm_get_value(gb); // taraFullness
2301 skip_bits(gb, 1); // allStreamSameTimeFraming
2302 skip_bits(gb, 6); // numSubFrames
2304 if (get_bits(gb, 4)) { // numPrograms
2305 av_log_missing_feature(latmctx->aac_ctx.avctx,
2306 "multiple programs are not supported\n", 1);
2307 return AVERROR_PATCHWELCOME;
2310 // for each program (which there is only on in DVB)
2312 // for each layer (which there is only on in DVB)
2313 if (get_bits(gb, 3)) { // numLayer
2314 av_log_missing_feature(latmctx->aac_ctx.avctx,
2315 "multiple layers are not supported\n", 1);
2316 return AVERROR_PATCHWELCOME;
2319 // for all but first stream: use_same_config = get_bits(gb, 1);
2320 if (!audio_mux_version) {
2321 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2324 int ascLen = latm_get_value(gb);
2325 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2328 skip_bits_long(gb, ascLen);
2331 latmctx->frame_length_type = get_bits(gb, 3);
2332 switch (latmctx->frame_length_type) {
2334 skip_bits(gb, 8); // latmBufferFullness
2337 latmctx->frame_length = get_bits(gb, 9);
2342 skip_bits(gb, 6); // CELP frame length table index
2346 skip_bits(gb, 1); // HVXC frame length table index
2350 if (get_bits(gb, 1)) { // other data
2351 if (audio_mux_version) {
2352 latm_get_value(gb); // other_data_bits
2356 esc = get_bits(gb, 1);
2362 if (get_bits(gb, 1)) // crc present
2363 skip_bits(gb, 8); // config_crc
2369 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2373 if (ctx->frame_length_type == 0) {
2374 int mux_slot_length = 0;
2376 tmp = get_bits(gb, 8);
2377 mux_slot_length += tmp;
2378 } while (tmp == 255);
2379 return mux_slot_length;
2380 } else if (ctx->frame_length_type == 1) {
2381 return ctx->frame_length;
2382 } else if (ctx->frame_length_type == 3 ||
2383 ctx->frame_length_type == 5 ||
2384 ctx->frame_length_type == 7) {
2385 skip_bits(gb, 2); // mux_slot_length_coded
2390 static int read_audio_mux_element(struct LATMContext *latmctx,
2394 uint8_t use_same_mux = get_bits(gb, 1);
2395 if (!use_same_mux) {
2396 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2398 } else if (!latmctx->aac_ctx.avctx->extradata) {
2399 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2400 "no decoder config found\n");
2401 return AVERROR(EAGAIN);
2403 if (latmctx->audio_mux_version_A == 0) {
2404 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2405 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2406 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2407 return AVERROR_INVALIDDATA;
2408 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2409 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2410 "frame length mismatch %d << %d\n",
2411 mux_slot_length_bytes * 8, get_bits_left(gb));
2412 return AVERROR_INVALIDDATA;
2419 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2422 struct LATMContext *latmctx = avctx->priv_data;
2426 if (avpkt->size == 0)
2429 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2431 // check for LOAS sync word
2432 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2433 return AVERROR_INVALIDDATA;
2435 muxlength = get_bits(&gb, 13) + 3;
2436 // not enough data, the parser should have sorted this
2437 if (muxlength > avpkt->size)
2438 return AVERROR_INVALIDDATA;
2440 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2443 if (!latmctx->initialized) {
2444 if (!avctx->extradata) {
2448 if ((err = aac_decode_init(avctx)) < 0)
2450 latmctx->initialized = 1;
2454 if (show_bits(&gb, 12) == 0xfff) {
2455 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2456 "ADTS header detected, probably as result of configuration "
2458 return AVERROR_INVALIDDATA;
2461 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2467 av_cold static int latm_decode_init(AVCodecContext *avctx)
2469 struct LATMContext *latmctx = avctx->priv_data;
2472 ret = aac_decode_init(avctx);
2474 if (avctx->extradata_size > 0) {
2475 latmctx->initialized = !ret;
2477 latmctx->initialized = 0;
2484 AVCodec ff_aac_decoder = {
2493 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2494 .sample_fmts = (const enum AVSampleFormat[]) {
2495 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2497 .channel_layouts = aac_channel_layout,
2501 Note: This decoder filter is intended to decode LATM streams transferred
2502 in MPEG transport streams which only contain one program.
2503 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2505 AVCodec ff_aac_latm_decoder = {
2507 .type = AVMEDIA_TYPE_AUDIO,
2508 .id = CODEC_ID_AAC_LATM,
2509 .priv_data_size = sizeof(struct LATMContext),
2510 .init = latm_decode_init,
2511 .close = aac_decode_close,
2512 .decode = latm_decode_frame,
2513 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2514 .sample_fmts = (const enum AVSampleFormat[]) {
2515 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2517 .channel_layouts = aac_channel_layout,