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 Libav.
12 * Libav 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 * Libav 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 Libav; 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"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
108 # include "arm/aac.h"
116 static VLC vlc_scalefactors;
117 static VLC vlc_spectral[11];
119 static const char overread_err[] = "Input buffer exhausted before END element found\n";
121 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
123 // For PCE based channel configurations map the channels solely based on tags.
124 if (!ac->m4ac.chan_config) {
125 return ac->tag_che_map[type][elem_id];
127 // For indexed channel configurations map the channels solely based on position.
128 switch (ac->m4ac.chan_config) {
130 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
132 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
135 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
136 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
137 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
138 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
140 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
143 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
145 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
148 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
150 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
154 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
156 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
157 } else if (ac->m4ac.chan_config == 2) {
161 if (!ac->tags_mapped && type == TYPE_SCE) {
163 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
171 * Check for the channel element in the current channel position configuration.
172 * If it exists, make sure the appropriate element is allocated and map the
173 * channel order to match the internal Libav channel layout.
175 * @param che_pos current channel position configuration
176 * @param type channel element type
177 * @param id channel element id
178 * @param channels count of the number of channels in the configuration
180 * @return Returns error status. 0 - OK, !0 - error
182 static av_cold int che_configure(AACContext *ac,
183 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
187 if (che_pos[type][id]) {
188 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
189 return AVERROR(ENOMEM);
190 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
191 if (type != TYPE_CCE) {
192 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
193 if (type == TYPE_CPE ||
194 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
195 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
199 if (ac->che[type][id])
200 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
201 av_freep(&ac->che[type][id]);
207 * Configure output channel order based on the current program configuration element.
209 * @param che_pos current channel position configuration
210 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
212 * @return Returns error status. 0 - OK, !0 - error
214 static av_cold int output_configure(AACContext *ac,
215 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
216 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
217 int channel_config, enum OCStatus oc_type)
219 AVCodecContext *avctx = ac->avctx;
220 int i, type, channels = 0, ret;
222 if (new_che_pos != che_pos)
223 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
225 if (channel_config) {
226 for (i = 0; i < tags_per_config[channel_config]; i++) {
227 if ((ret = che_configure(ac, che_pos,
228 aac_channel_layout_map[channel_config - 1][i][0],
229 aac_channel_layout_map[channel_config - 1][i][1],
234 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
236 avctx->channel_layout = aac_channel_layout[channel_config - 1];
238 /* Allocate or free elements depending on if they are in the
239 * current program configuration.
241 * Set up default 1:1 output mapping.
243 * For a 5.1 stream the output order will be:
244 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
247 for (i = 0; i < MAX_ELEM_ID; i++) {
248 for (type = 0; type < 4; type++) {
249 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
254 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
256 avctx->channel_layout = 0;
259 avctx->channels = channels;
261 ac->output_configured = oc_type;
267 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
269 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
270 * @param sce_map mono (Single Channel Element) map
271 * @param type speaker type/position for these channels
273 static void decode_channel_map(enum ChannelPosition *cpe_map,
274 enum ChannelPosition *sce_map,
275 enum ChannelPosition type,
276 GetBitContext *gb, int n)
279 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
280 map[get_bits(gb, 4)] = type;
285 * Decode program configuration element; reference: table 4.2.
287 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
289 * @return Returns error status. 0 - OK, !0 - error
291 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
292 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
295 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
298 skip_bits(gb, 2); // object_type
300 sampling_index = get_bits(gb, 4);
301 if (m4ac->sampling_index != sampling_index)
302 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");
304 num_front = get_bits(gb, 4);
305 num_side = get_bits(gb, 4);
306 num_back = get_bits(gb, 4);
307 num_lfe = get_bits(gb, 2);
308 num_assoc_data = get_bits(gb, 3);
309 num_cc = get_bits(gb, 4);
312 skip_bits(gb, 4); // mono_mixdown_tag
314 skip_bits(gb, 4); // stereo_mixdown_tag
317 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
321 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
322 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
324 skip_bits_long(gb, 4 * num_assoc_data);
326 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
330 /* comment field, first byte is length */
331 comment_len = get_bits(gb, 8) * 8;
332 if (get_bits_left(gb) < comment_len) {
333 av_log(avctx, AV_LOG_ERROR, overread_err);
336 skip_bits_long(gb, comment_len);
341 * Set up channel positions based on a default channel configuration
342 * as specified in table 1.17.
344 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
346 * @return Returns error status. 0 - OK, !0 - error
348 static av_cold int set_default_channel_config(AVCodecContext *avctx,
349 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
352 if (channel_config < 1 || channel_config > 7) {
353 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
358 /* default channel configurations:
360 * 1ch : front center (mono)
361 * 2ch : L + R (stereo)
362 * 3ch : front center + L + R
363 * 4ch : front center + L + R + back center
364 * 5ch : front center + L + R + back stereo
365 * 6ch : front center + L + R + back stereo + LFE
366 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
369 if (channel_config != 2)
370 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
371 if (channel_config > 1)
372 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
373 if (channel_config == 4)
374 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
375 if (channel_config > 4)
376 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
377 = AAC_CHANNEL_BACK; // back stereo
378 if (channel_config > 5)
379 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
380 if (channel_config == 7)
381 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
387 * Decode GA "General Audio" specific configuration; reference: table 4.1.
389 * @param ac pointer to AACContext, may be null
390 * @param avctx pointer to AVCCodecContext, used for logging
392 * @return Returns error status. 0 - OK, !0 - error
394 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
396 MPEG4AudioConfig *m4ac,
399 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
400 int extension_flag, ret;
402 if (get_bits1(gb)) { // frameLengthFlag
403 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
407 if (get_bits1(gb)) // dependsOnCoreCoder
408 skip_bits(gb, 14); // coreCoderDelay
409 extension_flag = get_bits1(gb);
411 if (m4ac->object_type == AOT_AAC_SCALABLE ||
412 m4ac->object_type == AOT_ER_AAC_SCALABLE)
413 skip_bits(gb, 3); // layerNr
415 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
416 if (channel_config == 0) {
417 skip_bits(gb, 4); // element_instance_tag
418 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
421 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
424 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
427 if (extension_flag) {
428 switch (m4ac->object_type) {
430 skip_bits(gb, 5); // numOfSubFrame
431 skip_bits(gb, 11); // layer_length
435 case AOT_ER_AAC_SCALABLE:
437 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
438 * aacScalefactorDataResilienceFlag
439 * aacSpectralDataResilienceFlag
443 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
449 * Decode audio specific configuration; reference: table 1.13.
451 * @param ac pointer to AACContext, may be null
452 * @param avctx pointer to AVCCodecContext, used for logging
453 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
454 * @param data pointer to AVCodecContext extradata
455 * @param data_size size of AVCCodecContext extradata
457 * @return Returns error status or number of consumed bits. <0 - error
459 static int decode_audio_specific_config(AACContext *ac,
460 AVCodecContext *avctx,
461 MPEG4AudioConfig *m4ac,
462 const uint8_t *data, int data_size)
467 init_get_bits(&gb, data, data_size * 8);
469 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
471 if (m4ac->sampling_index > 12) {
472 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
475 if (m4ac->sbr == 1 && m4ac->ps == -1)
478 skip_bits_long(&gb, i);
480 switch (m4ac->object_type) {
484 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
488 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
489 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
493 return get_bits_count(&gb);
497 * linear congruential pseudorandom number generator
499 * @param previous_val pointer to the current state of the generator
501 * @return Returns a 32-bit pseudorandom integer
503 static av_always_inline int lcg_random(int previous_val)
505 return previous_val * 1664525 + 1013904223;
508 static av_always_inline void reset_predict_state(PredictorState *ps)
518 static void reset_all_predictors(PredictorState *ps)
521 for (i = 0; i < MAX_PREDICTORS; i++)
522 reset_predict_state(&ps[i]);
525 static void reset_predictor_group(PredictorState *ps, int group_num)
528 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
529 reset_predict_state(&ps[i]);
532 #define AAC_INIT_VLC_STATIC(num, size) \
533 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
534 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
535 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
538 static av_cold int aac_decode_init(AVCodecContext *avctx)
540 AACContext *ac = avctx->priv_data;
543 ac->m4ac.sample_rate = avctx->sample_rate;
545 if (avctx->extradata_size > 0) {
546 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
548 avctx->extradata_size) < 0)
552 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
554 AAC_INIT_VLC_STATIC( 0, 304);
555 AAC_INIT_VLC_STATIC( 1, 270);
556 AAC_INIT_VLC_STATIC( 2, 550);
557 AAC_INIT_VLC_STATIC( 3, 300);
558 AAC_INIT_VLC_STATIC( 4, 328);
559 AAC_INIT_VLC_STATIC( 5, 294);
560 AAC_INIT_VLC_STATIC( 6, 306);
561 AAC_INIT_VLC_STATIC( 7, 268);
562 AAC_INIT_VLC_STATIC( 8, 510);
563 AAC_INIT_VLC_STATIC( 9, 366);
564 AAC_INIT_VLC_STATIC(10, 462);
568 dsputil_init(&ac->dsp, avctx);
569 ff_fmt_convert_init(&ac->fmt_conv, avctx);
571 ac->random_state = 0x1f2e3d4c;
573 // -1024 - Compensate wrong IMDCT method.
574 // 60 - Required to scale values to the correct range [-32768,32767]
575 // for float to int16 conversion. (1 << (60 / 4)) == 32768
576 ac->sf_scale = 1. / -1024.;
581 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
582 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
583 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
586 ff_mdct_init(&ac->mdct, 11, 1, 1.0);
587 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0);
588 ff_mdct_init(&ac->mdct_ltp, 11, 0, 1.0);
589 // window initialization
590 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
591 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
592 ff_init_ff_sine_windows(10);
593 ff_init_ff_sine_windows( 7);
601 * Skip data_stream_element; reference: table 4.10.
603 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
605 int byte_align = get_bits1(gb);
606 int count = get_bits(gb, 8);
608 count += get_bits(gb, 8);
612 if (get_bits_left(gb) < 8 * count) {
613 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
616 skip_bits_long(gb, 8 * count);
620 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
625 ics->predictor_reset_group = get_bits(gb, 5);
626 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
627 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
631 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
632 ics->prediction_used[sfb] = get_bits1(gb);
638 * Decode Long Term Prediction data; reference: table 4.xx.
640 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
641 GetBitContext *gb, uint8_t max_sfb)
645 ltp->lag = get_bits(gb, 11);
646 ltp->coef = ltp_coef[get_bits(gb, 3)] * ac->sf_scale;
647 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
648 ltp->used[sfb] = get_bits1(gb);
652 * Decode Individual Channel Stream info; reference: table 4.6.
654 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
656 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
657 GetBitContext *gb, int common_window)
660 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
661 memset(ics, 0, sizeof(IndividualChannelStream));
664 ics->window_sequence[1] = ics->window_sequence[0];
665 ics->window_sequence[0] = get_bits(gb, 2);
666 ics->use_kb_window[1] = ics->use_kb_window[0];
667 ics->use_kb_window[0] = get_bits1(gb);
668 ics->num_window_groups = 1;
669 ics->group_len[0] = 1;
670 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
672 ics->max_sfb = get_bits(gb, 4);
673 for (i = 0; i < 7; i++) {
675 ics->group_len[ics->num_window_groups - 1]++;
677 ics->num_window_groups++;
678 ics->group_len[ics->num_window_groups - 1] = 1;
681 ics->num_windows = 8;
682 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
683 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
684 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
685 ics->predictor_present = 0;
687 ics->max_sfb = get_bits(gb, 6);
688 ics->num_windows = 1;
689 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
690 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
691 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
692 ics->predictor_present = get_bits1(gb);
693 ics->predictor_reset_group = 0;
694 if (ics->predictor_present) {
695 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
696 if (decode_prediction(ac, ics, gb)) {
697 memset(ics, 0, sizeof(IndividualChannelStream));
700 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
701 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
702 memset(ics, 0, sizeof(IndividualChannelStream));
705 if ((ics->ltp.present = get_bits(gb, 1)))
706 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
711 if (ics->max_sfb > ics->num_swb) {
712 av_log(ac->avctx, AV_LOG_ERROR,
713 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
714 ics->max_sfb, ics->num_swb);
715 memset(ics, 0, sizeof(IndividualChannelStream));
723 * Decode band types (section_data payload); reference: table 4.46.
725 * @param band_type array of the used band type
726 * @param band_type_run_end array of the last scalefactor band of a band type run
728 * @return Returns error status. 0 - OK, !0 - error
730 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
731 int band_type_run_end[120], GetBitContext *gb,
732 IndividualChannelStream *ics)
735 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
736 for (g = 0; g < ics->num_window_groups; g++) {
738 while (k < ics->max_sfb) {
739 uint8_t sect_end = k;
741 int sect_band_type = get_bits(gb, 4);
742 if (sect_band_type == 12) {
743 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
746 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
747 sect_end += sect_len_incr;
748 sect_end += sect_len_incr;
749 if (get_bits_left(gb) < 0) {
750 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
753 if (sect_end > ics->max_sfb) {
754 av_log(ac->avctx, AV_LOG_ERROR,
755 "Number of bands (%d) exceeds limit (%d).\n",
756 sect_end, ics->max_sfb);
759 for (; k < sect_end; k++) {
760 band_type [idx] = sect_band_type;
761 band_type_run_end[idx++] = sect_end;
769 * Decode scalefactors; reference: table 4.47.
771 * @param global_gain first scalefactor value as scalefactors are differentially coded
772 * @param band_type array of the used band type
773 * @param band_type_run_end array of the last scalefactor band of a band type run
774 * @param sf array of scalefactors or intensity stereo positions
776 * @return Returns error status. 0 - OK, !0 - error
778 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
779 unsigned int global_gain,
780 IndividualChannelStream *ics,
781 enum BandType band_type[120],
782 int band_type_run_end[120])
784 const int sf_offset = ac->sf_offset + (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE ? 12 : 0);
786 int offset[3] = { global_gain, global_gain - 90, 100 };
788 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
789 for (g = 0; g < ics->num_window_groups; g++) {
790 for (i = 0; i < ics->max_sfb;) {
791 int run_end = band_type_run_end[idx];
792 if (band_type[idx] == ZERO_BT) {
793 for (; i < run_end; i++, idx++)
795 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
796 for (; i < run_end; i++, idx++) {
797 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
798 if (offset[2] > 255U) {
799 av_log(ac->avctx, AV_LOG_ERROR,
800 "%s (%d) out of range.\n", sf_str[2], offset[2]);
803 sf[idx] = ff_aac_pow2sf_tab[-offset[2] + 300];
805 } else if (band_type[idx] == NOISE_BT) {
806 for (; i < run_end; i++, idx++) {
807 if (noise_flag-- > 0)
808 offset[1] += get_bits(gb, 9) - 256;
810 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
811 if (offset[1] > 255U) {
812 av_log(ac->avctx, AV_LOG_ERROR,
813 "%s (%d) out of range.\n", sf_str[1], offset[1]);
816 sf[idx] = -ff_aac_pow2sf_tab[offset[1] + sf_offset + 100];
819 for (; i < run_end; i++, idx++) {
820 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
821 if (offset[0] > 255U) {
822 av_log(ac->avctx, AV_LOG_ERROR,
823 "%s (%d) out of range.\n", sf_str[0], offset[0]);
826 sf[idx] = -ff_aac_pow2sf_tab[ offset[0] + sf_offset];
835 * Decode pulse data; reference: table 4.7.
837 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
838 const uint16_t *swb_offset, int num_swb)
841 pulse->num_pulse = get_bits(gb, 2) + 1;
842 pulse_swb = get_bits(gb, 6);
843 if (pulse_swb >= num_swb)
845 pulse->pos[0] = swb_offset[pulse_swb];
846 pulse->pos[0] += get_bits(gb, 5);
847 if (pulse->pos[0] > 1023)
849 pulse->amp[0] = get_bits(gb, 4);
850 for (i = 1; i < pulse->num_pulse; i++) {
851 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
852 if (pulse->pos[i] > 1023)
854 pulse->amp[i] = get_bits(gb, 4);
860 * Decode Temporal Noise Shaping data; reference: table 4.48.
862 * @return Returns error status. 0 - OK, !0 - error
864 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
865 GetBitContext *gb, const IndividualChannelStream *ics)
867 int w, filt, i, coef_len, coef_res, coef_compress;
868 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
869 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
870 for (w = 0; w < ics->num_windows; w++) {
871 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
872 coef_res = get_bits1(gb);
874 for (filt = 0; filt < tns->n_filt[w]; filt++) {
876 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
878 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
879 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
880 tns->order[w][filt], tns_max_order);
881 tns->order[w][filt] = 0;
884 if (tns->order[w][filt]) {
885 tns->direction[w][filt] = get_bits1(gb);
886 coef_compress = get_bits1(gb);
887 coef_len = coef_res + 3 - coef_compress;
888 tmp2_idx = 2 * coef_compress + coef_res;
890 for (i = 0; i < tns->order[w][filt]; i++)
891 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
900 * Decode Mid/Side data; reference: table 4.54.
902 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
903 * [1] mask is decoded from bitstream; [2] mask is all 1s;
904 * [3] reserved for scalable AAC
906 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
910 if (ms_present == 1) {
911 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
912 cpe->ms_mask[idx] = get_bits1(gb);
913 } else if (ms_present == 2) {
914 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
919 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
923 *dst++ = v[idx & 15] * s;
924 *dst++ = v[idx>>4 & 15] * s;
930 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
934 *dst++ = v[idx & 3] * s;
935 *dst++ = v[idx>>2 & 3] * s;
936 *dst++ = v[idx>>4 & 3] * s;
937 *dst++ = v[idx>>6 & 3] * s;
943 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
944 unsigned sign, const float *scale)
946 union float754 s0, s1;
948 s0.f = s1.f = *scale;
949 s0.i ^= sign >> 1 << 31;
952 *dst++ = v[idx & 15] * s0.f;
953 *dst++ = v[idx>>4 & 15] * s1.f;
960 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
961 unsigned sign, const float *scale)
963 unsigned nz = idx >> 12;
964 union float754 s = { .f = *scale };
967 t.i = s.i ^ (sign & 1U<<31);
968 *dst++ = v[idx & 3] * t.f;
970 sign <<= nz & 1; nz >>= 1;
971 t.i = s.i ^ (sign & 1U<<31);
972 *dst++ = v[idx>>2 & 3] * t.f;
974 sign <<= nz & 1; nz >>= 1;
975 t.i = s.i ^ (sign & 1U<<31);
976 *dst++ = v[idx>>4 & 3] * t.f;
978 sign <<= nz & 1; nz >>= 1;
979 t.i = s.i ^ (sign & 1U<<31);
980 *dst++ = v[idx>>6 & 3] * t.f;
987 * Decode spectral data; reference: table 4.50.
988 * Dequantize and scale spectral data; reference: 4.6.3.3.
990 * @param coef array of dequantized, scaled spectral data
991 * @param sf array of scalefactors or intensity stereo positions
992 * @param pulse_present set if pulses are present
993 * @param pulse pointer to pulse data struct
994 * @param band_type array of the used band type
996 * @return Returns error status. 0 - OK, !0 - error
998 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
999 GetBitContext *gb, const float sf[120],
1000 int pulse_present, const Pulse *pulse,
1001 const IndividualChannelStream *ics,
1002 enum BandType band_type[120])
1004 int i, k, g, idx = 0;
1005 const int c = 1024 / ics->num_windows;
1006 const uint16_t *offsets = ics->swb_offset;
1007 float *coef_base = coef;
1009 for (g = 0; g < ics->num_windows; g++)
1010 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1012 for (g = 0; g < ics->num_window_groups; g++) {
1013 unsigned g_len = ics->group_len[g];
1015 for (i = 0; i < ics->max_sfb; i++, idx++) {
1016 const unsigned cbt_m1 = band_type[idx] - 1;
1017 float *cfo = coef + offsets[i];
1018 int off_len = offsets[i + 1] - offsets[i];
1021 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1022 for (group = 0; group < g_len; group++, cfo+=128) {
1023 memset(cfo, 0, off_len * sizeof(float));
1025 } else if (cbt_m1 == NOISE_BT - 1) {
1026 for (group = 0; group < g_len; group++, cfo+=128) {
1030 for (k = 0; k < off_len; k++) {
1031 ac->random_state = lcg_random(ac->random_state);
1032 cfo[k] = ac->random_state;
1035 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1036 scale = sf[idx] / sqrtf(band_energy);
1037 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1040 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1041 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1042 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1043 OPEN_READER(re, gb);
1045 switch (cbt_m1 >> 1) {
1047 for (group = 0; group < g_len; group++, cfo+=128) {
1055 UPDATE_CACHE(re, gb);
1056 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1057 cb_idx = cb_vector_idx[code];
1058 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1064 for (group = 0; group < g_len; group++, cfo+=128) {
1074 UPDATE_CACHE(re, gb);
1075 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1076 cb_idx = cb_vector_idx[code];
1077 nnz = cb_idx >> 8 & 15;
1078 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1079 LAST_SKIP_BITS(re, gb, nnz);
1080 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1086 for (group = 0; group < g_len; group++, cfo+=128) {
1094 UPDATE_CACHE(re, gb);
1095 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1096 cb_idx = cb_vector_idx[code];
1097 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1104 for (group = 0; group < g_len; group++, cfo+=128) {
1114 UPDATE_CACHE(re, gb);
1115 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1116 cb_idx = cb_vector_idx[code];
1117 nnz = cb_idx >> 8 & 15;
1118 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1119 LAST_SKIP_BITS(re, gb, nnz);
1120 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1126 for (group = 0; group < g_len; group++, cfo+=128) {
1128 uint32_t *icf = (uint32_t *) cf;
1138 UPDATE_CACHE(re, gb);
1139 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1147 cb_idx = cb_vector_idx[code];
1150 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1151 LAST_SKIP_BITS(re, gb, nnz);
1153 for (j = 0; j < 2; j++) {
1157 /* The total length of escape_sequence must be < 22 bits according
1158 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1159 UPDATE_CACHE(re, gb);
1160 b = GET_CACHE(re, gb);
1161 b = 31 - av_log2(~b);
1164 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1168 SKIP_BITS(re, gb, b + 1);
1170 n = (1 << b) + SHOW_UBITS(re, gb, b);
1171 LAST_SKIP_BITS(re, gb, b);
1172 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1175 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1176 *icf++ = (bits & 1U<<31) | v;
1183 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1187 CLOSE_READER(re, gb);
1193 if (pulse_present) {
1195 for (i = 0; i < pulse->num_pulse; i++) {
1196 float co = coef_base[ pulse->pos[i] ];
1197 while (offsets[idx + 1] <= pulse->pos[i])
1199 if (band_type[idx] != NOISE_BT && sf[idx]) {
1200 float ico = -pulse->amp[i];
1203 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1205 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1212 static av_always_inline float flt16_round(float pf)
1216 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1220 static av_always_inline float flt16_even(float pf)
1224 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1228 static av_always_inline float flt16_trunc(float pf)
1232 pun.i &= 0xFFFF0000U;
1236 static av_always_inline void predict(PredictorState *ps, float *coef,
1237 float sf_scale, float inv_sf_scale,
1240 const float a = 0.953125; // 61.0 / 64
1241 const float alpha = 0.90625; // 29.0 / 32
1245 float r0 = ps->r0, r1 = ps->r1;
1246 float cor0 = ps->cor0, cor1 = ps->cor1;
1247 float var0 = ps->var0, var1 = ps->var1;
1249 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1250 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1252 pv = flt16_round(k1 * r0 + k2 * r1);
1254 *coef += pv * sf_scale;
1256 e0 = *coef * inv_sf_scale;
1259 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1260 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1261 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1262 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1264 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1265 ps->r0 = flt16_trunc(a * e0);
1269 * Apply AAC-Main style frequency domain prediction.
1271 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1274 float sf_scale = ac->sf_scale, inv_sf_scale = 1 / ac->sf_scale;
1276 if (!sce->ics.predictor_initialized) {
1277 reset_all_predictors(sce->predictor_state);
1278 sce->ics.predictor_initialized = 1;
1281 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1282 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1283 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1284 predict(&sce->predictor_state[k], &sce->coeffs[k],
1285 sf_scale, inv_sf_scale,
1286 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1289 if (sce->ics.predictor_reset_group)
1290 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1292 reset_all_predictors(sce->predictor_state);
1296 * Decode an individual_channel_stream payload; reference: table 4.44.
1298 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1299 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1301 * @return Returns error status. 0 - OK, !0 - error
1303 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1304 GetBitContext *gb, int common_window, int scale_flag)
1307 TemporalNoiseShaping *tns = &sce->tns;
1308 IndividualChannelStream *ics = &sce->ics;
1309 float *out = sce->coeffs;
1310 int global_gain, pulse_present = 0;
1312 /* This assignment is to silence a GCC warning about the variable being used
1313 * uninitialized when in fact it always is.
1315 pulse.num_pulse = 0;
1317 global_gain = get_bits(gb, 8);
1319 if (!common_window && !scale_flag) {
1320 if (decode_ics_info(ac, ics, gb, 0) < 0)
1324 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1326 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1331 if ((pulse_present = get_bits1(gb))) {
1332 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1333 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1336 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1337 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1341 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1343 if (get_bits1(gb)) {
1344 av_log_missing_feature(ac->avctx, "SSR", 1);
1349 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1352 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1353 apply_prediction(ac, sce);
1359 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1361 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1363 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1364 float *ch0 = cpe->ch[0].coeffs;
1365 float *ch1 = cpe->ch[1].coeffs;
1366 int g, i, group, idx = 0;
1367 const uint16_t *offsets = ics->swb_offset;
1368 for (g = 0; g < ics->num_window_groups; g++) {
1369 for (i = 0; i < ics->max_sfb; i++, idx++) {
1370 if (cpe->ms_mask[idx] &&
1371 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1372 for (group = 0; group < ics->group_len[g]; group++) {
1373 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1374 ch1 + group * 128 + offsets[i],
1375 offsets[i+1] - offsets[i]);
1379 ch0 += ics->group_len[g] * 128;
1380 ch1 += ics->group_len[g] * 128;
1385 * intensity stereo decoding; reference: 4.6.8.2.3
1387 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1388 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1389 * [3] reserved for scalable AAC
1391 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1393 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1394 SingleChannelElement *sce1 = &cpe->ch[1];
1395 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1396 const uint16_t *offsets = ics->swb_offset;
1397 int g, group, i, idx = 0;
1400 for (g = 0; g < ics->num_window_groups; g++) {
1401 for (i = 0; i < ics->max_sfb;) {
1402 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1403 const int bt_run_end = sce1->band_type_run_end[idx];
1404 for (; i < bt_run_end; i++, idx++) {
1405 c = -1 + 2 * (sce1->band_type[idx] - 14);
1407 c *= 1 - 2 * cpe->ms_mask[idx];
1408 scale = c * sce1->sf[idx];
1409 for (group = 0; group < ics->group_len[g]; group++)
1410 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1411 coef0 + group * 128 + offsets[i],
1413 offsets[i + 1] - offsets[i]);
1416 int bt_run_end = sce1->band_type_run_end[idx];
1417 idx += bt_run_end - i;
1421 coef0 += ics->group_len[g] * 128;
1422 coef1 += ics->group_len[g] * 128;
1427 * Decode a channel_pair_element; reference: table 4.4.
1429 * @return Returns error status. 0 - OK, !0 - error
1431 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1433 int i, ret, common_window, ms_present = 0;
1435 common_window = get_bits1(gb);
1436 if (common_window) {
1437 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1439 i = cpe->ch[1].ics.use_kb_window[0];
1440 cpe->ch[1].ics = cpe->ch[0].ics;
1441 cpe->ch[1].ics.use_kb_window[1] = i;
1442 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1443 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1444 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1445 ms_present = get_bits(gb, 2);
1446 if (ms_present == 3) {
1447 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1449 } else if (ms_present)
1450 decode_mid_side_stereo(cpe, gb, ms_present);
1452 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1454 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1457 if (common_window) {
1459 apply_mid_side_stereo(ac, cpe);
1460 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1461 apply_prediction(ac, &cpe->ch[0]);
1462 apply_prediction(ac, &cpe->ch[1]);
1466 apply_intensity_stereo(ac, cpe, ms_present);
1470 static const float cce_scale[] = {
1471 1.09050773266525765921, //2^(1/8)
1472 1.18920711500272106672, //2^(1/4)
1478 * Decode coupling_channel_element; reference: table 4.8.
1480 * @return Returns error status. 0 - OK, !0 - error
1482 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1488 SingleChannelElement *sce = &che->ch[0];
1489 ChannelCoupling *coup = &che->coup;
1491 coup->coupling_point = 2 * get_bits1(gb);
1492 coup->num_coupled = get_bits(gb, 3);
1493 for (c = 0; c <= coup->num_coupled; c++) {
1495 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1496 coup->id_select[c] = get_bits(gb, 4);
1497 if (coup->type[c] == TYPE_CPE) {
1498 coup->ch_select[c] = get_bits(gb, 2);
1499 if (coup->ch_select[c] == 3)
1502 coup->ch_select[c] = 2;
1504 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1506 sign = get_bits(gb, 1);
1507 scale = cce_scale[get_bits(gb, 2)];
1509 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1512 for (c = 0; c < num_gain; c++) {
1516 float gain_cache = 1.;
1518 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1519 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1520 gain_cache = powf(scale, -gain);
1522 if (coup->coupling_point == AFTER_IMDCT) {
1523 coup->gain[c][0] = gain_cache;
1525 for (g = 0; g < sce->ics.num_window_groups; g++) {
1526 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1527 if (sce->band_type[idx] != ZERO_BT) {
1529 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1537 gain_cache = powf(scale, -t) * s;
1540 coup->gain[c][idx] = gain_cache;
1550 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1552 * @return Returns number of bytes consumed.
1554 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1558 int num_excl_chan = 0;
1561 for (i = 0; i < 7; i++)
1562 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1563 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1565 return num_excl_chan / 7;
1569 * Decode dynamic range information; reference: table 4.52.
1571 * @param cnt length of TYPE_FIL syntactic element in bytes
1573 * @return Returns number of bytes consumed.
1575 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1576 GetBitContext *gb, int cnt)
1579 int drc_num_bands = 1;
1582 /* pce_tag_present? */
1583 if (get_bits1(gb)) {
1584 che_drc->pce_instance_tag = get_bits(gb, 4);
1585 skip_bits(gb, 4); // tag_reserved_bits
1589 /* excluded_chns_present? */
1590 if (get_bits1(gb)) {
1591 n += decode_drc_channel_exclusions(che_drc, gb);
1594 /* drc_bands_present? */
1595 if (get_bits1(gb)) {
1596 che_drc->band_incr = get_bits(gb, 4);
1597 che_drc->interpolation_scheme = get_bits(gb, 4);
1599 drc_num_bands += che_drc->band_incr;
1600 for (i = 0; i < drc_num_bands; i++) {
1601 che_drc->band_top[i] = get_bits(gb, 8);
1606 /* prog_ref_level_present? */
1607 if (get_bits1(gb)) {
1608 che_drc->prog_ref_level = get_bits(gb, 7);
1609 skip_bits1(gb); // prog_ref_level_reserved_bits
1613 for (i = 0; i < drc_num_bands; i++) {
1614 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1615 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1623 * Decode extension data (incomplete); reference: table 4.51.
1625 * @param cnt length of TYPE_FIL syntactic element in bytes
1627 * @return Returns number of bytes consumed
1629 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1630 ChannelElement *che, enum RawDataBlockType elem_type)
1634 switch (get_bits(gb, 4)) { // extension type
1635 case EXT_SBR_DATA_CRC:
1639 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1641 } else if (!ac->m4ac.sbr) {
1642 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1643 skip_bits_long(gb, 8 * cnt - 4);
1645 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1646 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1647 skip_bits_long(gb, 8 * cnt - 4);
1649 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1652 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1656 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1658 case EXT_DYNAMIC_RANGE:
1659 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1663 case EXT_DATA_ELEMENT:
1665 skip_bits_long(gb, 8 * cnt - 4);
1672 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1674 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1675 * @param coef spectral coefficients
1677 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1678 IndividualChannelStream *ics, int decode)
1680 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1682 int bottom, top, order, start, end, size, inc;
1683 float lpc[TNS_MAX_ORDER];
1684 float tmp[TNS_MAX_ORDER];
1686 for (w = 0; w < ics->num_windows; w++) {
1687 bottom = ics->num_swb;
1688 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1690 bottom = FFMAX(0, top - tns->length[w][filt]);
1691 order = tns->order[w][filt];
1696 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1698 start = ics->swb_offset[FFMIN(bottom, mmm)];
1699 end = ics->swb_offset[FFMIN( top, mmm)];
1700 if ((size = end - start) <= 0)
1702 if (tns->direction[w][filt]) {
1712 for (m = 0; m < size; m++, start += inc)
1713 for (i = 1; i <= FFMIN(m, order); i++)
1714 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1717 for (m = 0; m < size; m++, start += inc) {
1718 tmp[0] = coef[start];
1719 for (i = 1; i <= FFMIN(m, order); i++)
1720 coef[start] += tmp[i] * lpc[i - 1];
1721 for (i = order; i > 0; i--)
1722 tmp[i] = tmp[i - 1];
1730 * Apply windowing and MDCT to obtain the spectral
1731 * coefficient from the predicted sample by LTP.
1733 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1734 float *in, IndividualChannelStream *ics)
1736 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1737 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1738 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1739 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1741 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1742 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1744 memset(in, 0, 448 * sizeof(float));
1745 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1746 memcpy(in + 576, in + 576, 448 * sizeof(float));
1748 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1749 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1751 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1752 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1753 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1755 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1759 * Apply the long term prediction
1761 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1763 const LongTermPrediction *ltp = &sce->ics.ltp;
1764 const uint16_t *offsets = sce->ics.swb_offset;
1767 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1768 float *predTime = sce->ret;
1769 float *predFreq = ac->buf_mdct;
1770 int16_t num_samples = 2048;
1772 if (ltp->lag < 1024)
1773 num_samples = ltp->lag + 1024;
1774 for (i = 0; i < num_samples; i++)
1775 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1776 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1778 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1780 if (sce->tns.present)
1781 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1783 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1785 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1786 sce->coeffs[i] += predFreq[i];
1791 * Update the LTP buffer for next frame
1793 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1795 IndividualChannelStream *ics = &sce->ics;
1796 float *saved = sce->saved;
1797 float *saved_ltp = sce->coeffs;
1798 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1799 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1802 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1803 memcpy(saved_ltp, saved, 512 * sizeof(float));
1804 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1805 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1806 for (i = 0; i < 64; i++)
1807 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1808 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1809 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1810 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1811 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1812 for (i = 0; i < 64; i++)
1813 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1814 } else { // LONG_STOP or ONLY_LONG
1815 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1816 for (i = 0; i < 512; i++)
1817 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1820 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1821 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1822 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1826 * Conduct IMDCT and windowing.
1828 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1830 IndividualChannelStream *ics = &sce->ics;
1831 float *in = sce->coeffs;
1832 float *out = sce->ret;
1833 float *saved = sce->saved;
1834 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1835 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1836 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1837 float *buf = ac->buf_mdct;
1838 float *temp = ac->temp;
1842 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1843 for (i = 0; i < 1024; i += 128)
1844 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1846 ac->mdct.imdct_half(&ac->mdct, buf, in);
1848 /* window overlapping
1849 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1850 * and long to short transitions are considered to be short to short
1851 * transitions. This leaves just two cases (long to long and short to short)
1852 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1854 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1855 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1856 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1858 memcpy( out, saved, 448 * sizeof(float));
1860 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1861 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1862 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1863 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1864 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1865 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1866 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1868 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1869 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1874 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1875 memcpy( saved, temp + 64, 64 * sizeof(float));
1876 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1877 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1878 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1879 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1880 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1881 memcpy( saved, buf + 512, 448 * sizeof(float));
1882 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1883 } else { // LONG_STOP or ONLY_LONG
1884 memcpy( saved, buf + 512, 512 * sizeof(float));
1889 * Apply dependent channel coupling (applied before IMDCT).
1891 * @param index index into coupling gain array
1893 static void apply_dependent_coupling(AACContext *ac,
1894 SingleChannelElement *target,
1895 ChannelElement *cce, int index)
1897 IndividualChannelStream *ics = &cce->ch[0].ics;
1898 const uint16_t *offsets = ics->swb_offset;
1899 float *dest = target->coeffs;
1900 const float *src = cce->ch[0].coeffs;
1901 int g, i, group, k, idx = 0;
1902 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1903 av_log(ac->avctx, AV_LOG_ERROR,
1904 "Dependent coupling is not supported together with LTP\n");
1907 for (g = 0; g < ics->num_window_groups; g++) {
1908 for (i = 0; i < ics->max_sfb; i++, idx++) {
1909 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1910 const float gain = cce->coup.gain[index][idx];
1911 for (group = 0; group < ics->group_len[g]; group++) {
1912 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1914 dest[group * 128 + k] += gain * src[group * 128 + k];
1919 dest += ics->group_len[g] * 128;
1920 src += ics->group_len[g] * 128;
1925 * Apply independent channel coupling (applied after IMDCT).
1927 * @param index index into coupling gain array
1929 static void apply_independent_coupling(AACContext *ac,
1930 SingleChannelElement *target,
1931 ChannelElement *cce, int index)
1934 const float gain = cce->coup.gain[index][0];
1935 const float *src = cce->ch[0].ret;
1936 float *dest = target->ret;
1937 const int len = 1024 << (ac->m4ac.sbr == 1);
1939 for (i = 0; i < len; i++)
1940 dest[i] += gain * src[i];
1944 * channel coupling transformation interface
1946 * @param apply_coupling_method pointer to (in)dependent coupling function
1948 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1949 enum RawDataBlockType type, int elem_id,
1950 enum CouplingPoint coupling_point,
1951 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1955 for (i = 0; i < MAX_ELEM_ID; i++) {
1956 ChannelElement *cce = ac->che[TYPE_CCE][i];
1959 if (cce && cce->coup.coupling_point == coupling_point) {
1960 ChannelCoupling *coup = &cce->coup;
1962 for (c = 0; c <= coup->num_coupled; c++) {
1963 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1964 if (coup->ch_select[c] != 1) {
1965 apply_coupling_method(ac, &cc->ch[0], cce, index);
1966 if (coup->ch_select[c] != 0)
1969 if (coup->ch_select[c] != 2)
1970 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1972 index += 1 + (coup->ch_select[c] == 3);
1979 * Convert spectral data to float samples, applying all supported tools as appropriate.
1981 static void spectral_to_sample(AACContext *ac)
1984 for (type = 3; type >= 0; type--) {
1985 for (i = 0; i < MAX_ELEM_ID; i++) {
1986 ChannelElement *che = ac->che[type][i];
1988 if (type <= TYPE_CPE)
1989 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1990 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1991 if (che->ch[0].ics.predictor_present) {
1992 if (che->ch[0].ics.ltp.present)
1993 apply_ltp(ac, &che->ch[0]);
1994 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
1995 apply_ltp(ac, &che->ch[1]);
1998 if (che->ch[0].tns.present)
1999 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2000 if (che->ch[1].tns.present)
2001 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2002 if (type <= TYPE_CPE)
2003 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2004 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2005 imdct_and_windowing(ac, &che->ch[0]);
2006 if (ac->m4ac.object_type == AOT_AAC_LTP)
2007 update_ltp(ac, &che->ch[0]);
2008 if (type == TYPE_CPE) {
2009 imdct_and_windowing(ac, &che->ch[1]);
2010 if (ac->m4ac.object_type == AOT_AAC_LTP)
2011 update_ltp(ac, &che->ch[1]);
2013 if (ac->m4ac.sbr > 0) {
2014 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2017 if (type <= TYPE_CCE)
2018 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2024 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2027 AACADTSHeaderInfo hdr_info;
2029 size = ff_aac_parse_header(gb, &hdr_info);
2031 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2032 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2033 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2034 ac->m4ac.chan_config = hdr_info.chan_config;
2035 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2037 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2039 } else if (ac->output_configured != OC_LOCKED) {
2040 ac->output_configured = OC_NONE;
2042 if (ac->output_configured != OC_LOCKED) {
2046 ac->m4ac.sample_rate = hdr_info.sample_rate;
2047 ac->m4ac.sampling_index = hdr_info.sampling_index;
2048 ac->m4ac.object_type = hdr_info.object_type;
2049 if (!ac->avctx->sample_rate)
2050 ac->avctx->sample_rate = hdr_info.sample_rate;
2051 if (hdr_info.num_aac_frames == 1) {
2052 if (!hdr_info.crc_absent)
2055 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2062 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2063 int *data_size, GetBitContext *gb)
2065 AACContext *ac = avctx->priv_data;
2066 ChannelElement *che = NULL, *che_prev = NULL;
2067 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2068 int err, elem_id, data_size_tmp;
2069 int samples = 0, multiplier;
2071 if (show_bits(gb, 12) == 0xfff) {
2072 if (parse_adts_frame_header(ac, gb) < 0) {
2073 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2076 if (ac->m4ac.sampling_index > 12) {
2077 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2082 ac->tags_mapped = 0;
2084 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2085 elem_id = get_bits(gb, 4);
2087 if (elem_type < TYPE_DSE) {
2088 if (!(che=get_che(ac, elem_type, elem_id))) {
2089 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2090 elem_type, elem_id);
2096 switch (elem_type) {
2099 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2103 err = decode_cpe(ac, gb, che);
2107 err = decode_cce(ac, gb, che);
2111 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2115 err = skip_data_stream_element(ac, gb);
2119 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2120 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2121 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2123 if (ac->output_configured > OC_TRIAL_PCE)
2124 av_log(avctx, AV_LOG_ERROR,
2125 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2127 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2133 elem_id += get_bits(gb, 8) - 1;
2134 if (get_bits_left(gb) < 8 * elem_id) {
2135 av_log(avctx, AV_LOG_ERROR, overread_err);
2139 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2140 err = 0; /* FIXME */
2144 err = -1; /* should not happen, but keeps compiler happy */
2149 elem_type_prev = elem_type;
2154 if (get_bits_left(gb) < 3) {
2155 av_log(avctx, AV_LOG_ERROR, overread_err);
2160 spectral_to_sample(ac);
2162 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2163 samples <<= multiplier;
2164 if (ac->output_configured < OC_LOCKED) {
2165 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2166 avctx->frame_size = samples;
2169 data_size_tmp = samples * avctx->channels * sizeof(int16_t);
2170 if (*data_size < data_size_tmp) {
2171 av_log(avctx, AV_LOG_ERROR,
2172 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2173 *data_size, data_size_tmp);
2176 *data_size = data_size_tmp;
2179 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2181 if (ac->output_configured)
2182 ac->output_configured = OC_LOCKED;
2187 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2188 int *data_size, AVPacket *avpkt)
2190 const uint8_t *buf = avpkt->data;
2191 int buf_size = avpkt->size;
2197 init_get_bits(&gb, buf, buf_size * 8);
2199 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2202 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2203 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2204 if (buf[buf_offset])
2207 return buf_size > buf_offset ? buf_consumed : buf_size;
2210 static av_cold int aac_decode_close(AVCodecContext *avctx)
2212 AACContext *ac = avctx->priv_data;
2215 for (i = 0; i < MAX_ELEM_ID; i++) {
2216 for (type = 0; type < 4; type++) {
2217 if (ac->che[type][i])
2218 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2219 av_freep(&ac->che[type][i]);
2223 ff_mdct_end(&ac->mdct);
2224 ff_mdct_end(&ac->mdct_small);
2225 ff_mdct_end(&ac->mdct_ltp);
2230 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2232 struct LATMContext {
2233 AACContext aac_ctx; ///< containing AACContext
2234 int initialized; ///< initilized after a valid extradata was seen
2237 int audio_mux_version_A; ///< LATM syntax version
2238 int frame_length_type; ///< 0/1 variable/fixed frame length
2239 int frame_length; ///< frame length for fixed frame length
2242 static inline uint32_t latm_get_value(GetBitContext *b)
2244 int length = get_bits(b, 2);
2246 return get_bits_long(b, (length+1)*8);
2249 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2252 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2253 MPEG4AudioConfig m4ac;
2254 int config_start_bit = get_bits_count(gb);
2255 int bits_consumed, esize;
2257 if (config_start_bit % 8) {
2258 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2259 "config not byte aligned.\n", 1);
2260 return AVERROR_INVALIDDATA;
2263 decode_audio_specific_config(NULL, avctx, &m4ac,
2264 gb->buffer + (config_start_bit / 8),
2265 get_bits_left(gb) / 8);
2267 if (bits_consumed < 0)
2268 return AVERROR_INVALIDDATA;
2270 esize = (bits_consumed+7) / 8;
2272 if (avctx->extradata_size <= esize) {
2273 av_free(avctx->extradata);
2274 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2275 if (!avctx->extradata)
2276 return AVERROR(ENOMEM);
2279 avctx->extradata_size = esize;
2280 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2281 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2283 skip_bits_long(gb, bits_consumed);
2286 return bits_consumed;
2289 static int read_stream_mux_config(struct LATMContext *latmctx,
2292 int ret, audio_mux_version = get_bits(gb, 1);
2294 latmctx->audio_mux_version_A = 0;
2295 if (audio_mux_version)
2296 latmctx->audio_mux_version_A = get_bits(gb, 1);
2298 if (!latmctx->audio_mux_version_A) {
2300 if (audio_mux_version)
2301 latm_get_value(gb); // taraFullness
2303 skip_bits(gb, 1); // allStreamSameTimeFraming
2304 skip_bits(gb, 6); // numSubFrames
2306 if (get_bits(gb, 4)) { // numPrograms
2307 av_log_missing_feature(latmctx->aac_ctx.avctx,
2308 "multiple programs are not supported\n", 1);
2309 return AVERROR_PATCHWELCOME;
2312 // for each program (which there is only on in DVB)
2314 // for each layer (which there is only on in DVB)
2315 if (get_bits(gb, 3)) { // numLayer
2316 av_log_missing_feature(latmctx->aac_ctx.avctx,
2317 "multiple layers are not supported\n", 1);
2318 return AVERROR_PATCHWELCOME;
2321 // for all but first stream: use_same_config = get_bits(gb, 1);
2322 if (!audio_mux_version) {
2323 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2326 int ascLen = latm_get_value(gb);
2327 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2330 skip_bits_long(gb, ascLen);
2333 latmctx->frame_length_type = get_bits(gb, 3);
2334 switch (latmctx->frame_length_type) {
2336 skip_bits(gb, 8); // latmBufferFullness
2339 latmctx->frame_length = get_bits(gb, 9);
2344 skip_bits(gb, 6); // CELP frame length table index
2348 skip_bits(gb, 1); // HVXC frame length table index
2352 if (get_bits(gb, 1)) { // other data
2353 if (audio_mux_version) {
2354 latm_get_value(gb); // other_data_bits
2358 esc = get_bits(gb, 1);
2364 if (get_bits(gb, 1)) // crc present
2365 skip_bits(gb, 8); // config_crc
2371 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2375 if (ctx->frame_length_type == 0) {
2376 int mux_slot_length = 0;
2378 tmp = get_bits(gb, 8);
2379 mux_slot_length += tmp;
2380 } while (tmp == 255);
2381 return mux_slot_length;
2382 } else if (ctx->frame_length_type == 1) {
2383 return ctx->frame_length;
2384 } else if (ctx->frame_length_type == 3 ||
2385 ctx->frame_length_type == 5 ||
2386 ctx->frame_length_type == 7) {
2387 skip_bits(gb, 2); // mux_slot_length_coded
2392 static int read_audio_mux_element(struct LATMContext *latmctx,
2396 uint8_t use_same_mux = get_bits(gb, 1);
2397 if (!use_same_mux) {
2398 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2400 } else if (!latmctx->aac_ctx.avctx->extradata) {
2401 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2402 "no decoder config found\n");
2403 return AVERROR(EAGAIN);
2405 if (latmctx->audio_mux_version_A == 0) {
2406 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2407 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2408 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2409 return AVERROR_INVALIDDATA;
2410 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2411 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2412 "frame length mismatch %d << %d\n",
2413 mux_slot_length_bytes * 8, get_bits_left(gb));
2414 return AVERROR_INVALIDDATA;
2421 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2424 struct LATMContext *latmctx = avctx->priv_data;
2428 if (avpkt->size == 0)
2431 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2433 // check for LOAS sync word
2434 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2435 return AVERROR_INVALIDDATA;
2437 muxlength = get_bits(&gb, 13) + 3;
2438 // not enough data, the parser should have sorted this
2439 if (muxlength > avpkt->size)
2440 return AVERROR_INVALIDDATA;
2442 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2445 if (!latmctx->initialized) {
2446 if (!avctx->extradata) {
2450 if ((err = aac_decode_init(avctx)) < 0)
2452 latmctx->initialized = 1;
2456 if (show_bits(&gb, 12) == 0xfff) {
2457 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2458 "ADTS header detected, probably as result of configuration "
2460 return AVERROR_INVALIDDATA;
2463 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2469 av_cold static int latm_decode_init(AVCodecContext *avctx)
2471 struct LATMContext *latmctx = avctx->priv_data;
2474 ret = aac_decode_init(avctx);
2476 if (avctx->extradata_size > 0) {
2477 latmctx->initialized = !ret;
2479 latmctx->initialized = 0;
2486 AVCodec ff_aac_decoder = {
2495 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2496 .sample_fmts = (const enum AVSampleFormat[]) {
2497 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2499 .channel_layouts = aac_channel_layout,
2503 Note: This decoder filter is intended to decode LATM streams transferred
2504 in MPEG transport streams which only contain one program.
2505 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2507 AVCodec ff_aac_latm_decoder = {
2509 .type = AVMEDIA_TYPE_AUDIO,
2510 .id = CODEC_ID_AAC_LATM,
2511 .priv_data_size = sizeof(struct LATMContext),
2512 .init = latm_decode_init,
2513 .close = aac_decode_close,
2514 .decode = latm_decode_frame,
2515 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2516 .sample_fmts = (const enum AVSampleFormat[]) {
2517 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_NONE
2519 .channel_layouts = aac_channel_layout,