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],
184 int type, int id, int *channels)
186 if (che_pos[type][id]) {
187 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
188 return AVERROR(ENOMEM);
189 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
190 if (type != TYPE_CCE) {
191 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
192 if (type == TYPE_CPE ||
193 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
194 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
198 if (ac->che[type][id])
199 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
200 av_freep(&ac->che[type][id]);
206 * Configure output channel order based on the current program configuration element.
208 * @param che_pos current channel position configuration
209 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
211 * @return Returns error status. 0 - OK, !0 - error
213 static av_cold int output_configure(AACContext *ac,
214 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
215 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
216 int channel_config, enum OCStatus oc_type)
218 AVCodecContext *avctx = ac->avctx;
219 int i, type, channels = 0, ret;
221 if (new_che_pos != che_pos)
222 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
224 if (channel_config) {
225 for (i = 0; i < tags_per_config[channel_config]; i++) {
226 if ((ret = che_configure(ac, che_pos,
227 aac_channel_layout_map[channel_config - 1][i][0],
228 aac_channel_layout_map[channel_config - 1][i][1],
233 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
235 avctx->channel_layout = aac_channel_layout[channel_config - 1];
237 /* Allocate or free elements depending on if they are in the
238 * current program configuration.
240 * Set up default 1:1 output mapping.
242 * For a 5.1 stream the output order will be:
243 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
246 for (i = 0; i < MAX_ELEM_ID; i++) {
247 for (type = 0; type < 4; type++) {
248 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
253 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
255 avctx->channel_layout = 0;
258 avctx->channels = channels;
260 ac->output_configured = oc_type;
266 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
268 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
269 * @param sce_map mono (Single Channel Element) map
270 * @param type speaker type/position for these channels
272 static void decode_channel_map(enum ChannelPosition *cpe_map,
273 enum ChannelPosition *sce_map,
274 enum ChannelPosition type,
275 GetBitContext *gb, int n)
278 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
279 map[get_bits(gb, 4)] = type;
284 * Decode program configuration element; reference: table 4.2.
286 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
288 * @return Returns error status. 0 - OK, !0 - error
290 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
291 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
294 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
297 skip_bits(gb, 2); // object_type
299 sampling_index = get_bits(gb, 4);
300 if (m4ac->sampling_index != sampling_index)
301 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");
303 num_front = get_bits(gb, 4);
304 num_side = get_bits(gb, 4);
305 num_back = get_bits(gb, 4);
306 num_lfe = get_bits(gb, 2);
307 num_assoc_data = get_bits(gb, 3);
308 num_cc = get_bits(gb, 4);
311 skip_bits(gb, 4); // mono_mixdown_tag
313 skip_bits(gb, 4); // stereo_mixdown_tag
316 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
318 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
321 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
323 skip_bits_long(gb, 4 * num_assoc_data);
325 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
329 /* comment field, first byte is length */
330 comment_len = get_bits(gb, 8) * 8;
331 if (get_bits_left(gb) < comment_len) {
332 av_log(avctx, AV_LOG_ERROR, overread_err);
335 skip_bits_long(gb, comment_len);
340 * Set up channel positions based on a default channel configuration
341 * as specified in table 1.17.
343 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
345 * @return Returns error status. 0 - OK, !0 - error
347 static av_cold int set_default_channel_config(AVCodecContext *avctx,
348 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
351 if (channel_config < 1 || channel_config > 7) {
352 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
357 /* default channel configurations:
359 * 1ch : front center (mono)
360 * 2ch : L + R (stereo)
361 * 3ch : front center + L + R
362 * 4ch : front center + L + R + back center
363 * 5ch : front center + L + R + back stereo
364 * 6ch : front center + L + R + back stereo + LFE
365 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
368 if (channel_config != 2)
369 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
370 if (channel_config > 1)
371 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
372 if (channel_config == 4)
373 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
374 if (channel_config > 4)
375 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
376 = AAC_CHANNEL_BACK; // back stereo
377 if (channel_config > 5)
378 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
379 if (channel_config == 7)
380 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
386 * Decode GA "General Audio" specific configuration; reference: table 4.1.
388 * @param ac pointer to AACContext, may be null
389 * @param avctx pointer to AVCCodecContext, used for logging
391 * @return Returns error status. 0 - OK, !0 - error
393 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
395 MPEG4AudioConfig *m4ac,
398 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
399 int extension_flag, ret;
401 if (get_bits1(gb)) { // frameLengthFlag
402 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
406 if (get_bits1(gb)) // dependsOnCoreCoder
407 skip_bits(gb, 14); // coreCoderDelay
408 extension_flag = get_bits1(gb);
410 if (m4ac->object_type == AOT_AAC_SCALABLE ||
411 m4ac->object_type == AOT_ER_AAC_SCALABLE)
412 skip_bits(gb, 3); // layerNr
414 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
415 if (channel_config == 0) {
416 skip_bits(gb, 4); // element_instance_tag
417 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
420 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
423 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
426 if (extension_flag) {
427 switch (m4ac->object_type) {
429 skip_bits(gb, 5); // numOfSubFrame
430 skip_bits(gb, 11); // layer_length
434 case AOT_ER_AAC_SCALABLE:
436 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
437 * aacScalefactorDataResilienceFlag
438 * aacSpectralDataResilienceFlag
442 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
448 * Decode audio specific configuration; reference: table 1.13.
450 * @param ac pointer to AACContext, may be null
451 * @param avctx pointer to AVCCodecContext, used for logging
452 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
453 * @param data pointer to AVCodecContext extradata
454 * @param data_size size of AVCCodecContext extradata
456 * @return Returns error status or number of consumed bits. <0 - error
458 static int decode_audio_specific_config(AACContext *ac,
459 AVCodecContext *avctx,
460 MPEG4AudioConfig *m4ac,
461 const uint8_t *data, int data_size)
466 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
467 for (i = 0; i < avctx->extradata_size; i++)
468 av_dlog(avctx, "%02x ", avctx->extradata[i]);
469 av_dlog(avctx, "\n");
471 init_get_bits(&gb, data, data_size * 8);
473 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
475 if (m4ac->sampling_index > 12) {
476 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
479 if (m4ac->sbr == 1 && m4ac->ps == -1)
482 skip_bits_long(&gb, i);
484 switch (m4ac->object_type) {
488 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
492 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
493 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
497 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
498 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
499 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
501 return get_bits_count(&gb);
505 * linear congruential pseudorandom number generator
507 * @param previous_val pointer to the current state of the generator
509 * @return Returns a 32-bit pseudorandom integer
511 static av_always_inline int lcg_random(int previous_val)
513 return previous_val * 1664525 + 1013904223;
516 static av_always_inline void reset_predict_state(PredictorState *ps)
526 static void reset_all_predictors(PredictorState *ps)
529 for (i = 0; i < MAX_PREDICTORS; i++)
530 reset_predict_state(&ps[i]);
533 static int sample_rate_idx (int rate)
535 if (92017 <= rate) return 0;
536 else if (75132 <= rate) return 1;
537 else if (55426 <= rate) return 2;
538 else if (46009 <= rate) return 3;
539 else if (37566 <= rate) return 4;
540 else if (27713 <= rate) return 5;
541 else if (23004 <= rate) return 6;
542 else if (18783 <= rate) return 7;
543 else if (13856 <= rate) return 8;
544 else if (11502 <= rate) return 9;
545 else if (9391 <= rate) return 10;
549 static void reset_predictor_group(PredictorState *ps, int group_num)
552 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
553 reset_predict_state(&ps[i]);
556 #define AAC_INIT_VLC_STATIC(num, size) \
557 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
558 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
559 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
562 static av_cold int aac_decode_init(AVCodecContext *avctx)
564 AACContext *ac = avctx->priv_data;
565 float output_scale_factor;
568 ac->m4ac.sample_rate = avctx->sample_rate;
570 if (avctx->extradata_size > 0) {
571 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
573 avctx->extradata_size) < 0)
577 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
579 sr = sample_rate_idx(avctx->sample_rate);
580 ac->m4ac.sampling_index = sr;
581 ac->m4ac.channels = avctx->channels;
583 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
584 if (ff_mpeg4audio_channels[i] == avctx->channels)
586 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
589 ac->m4ac.chan_config = i;
591 if (ac->m4ac.chan_config) {
592 set_default_channel_config(avctx, new_che_pos, ac->m4ac.chan_config);
593 output_configure(ac, ac->che_pos, new_che_pos, ac->m4ac.chan_config, OC_GLOBAL_HDR);
597 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
598 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
599 output_scale_factor = 1.0 / 32768.0;
601 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
602 output_scale_factor = 1.0;
605 AAC_INIT_VLC_STATIC( 0, 304);
606 AAC_INIT_VLC_STATIC( 1, 270);
607 AAC_INIT_VLC_STATIC( 2, 550);
608 AAC_INIT_VLC_STATIC( 3, 300);
609 AAC_INIT_VLC_STATIC( 4, 328);
610 AAC_INIT_VLC_STATIC( 5, 294);
611 AAC_INIT_VLC_STATIC( 6, 306);
612 AAC_INIT_VLC_STATIC( 7, 268);
613 AAC_INIT_VLC_STATIC( 8, 510);
614 AAC_INIT_VLC_STATIC( 9, 366);
615 AAC_INIT_VLC_STATIC(10, 462);
619 dsputil_init(&ac->dsp, avctx);
620 ff_fmt_convert_init(&ac->fmt_conv, avctx);
622 ac->random_state = 0x1f2e3d4c;
626 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
627 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
628 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
631 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
632 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
633 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
634 // window initialization
635 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
636 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
637 ff_init_ff_sine_windows(10);
638 ff_init_ff_sine_windows( 7);
646 * Skip data_stream_element; reference: table 4.10.
648 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
650 int byte_align = get_bits1(gb);
651 int count = get_bits(gb, 8);
653 count += get_bits(gb, 8);
657 if (get_bits_left(gb) < 8 * count) {
658 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
661 skip_bits_long(gb, 8 * count);
665 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
670 ics->predictor_reset_group = get_bits(gb, 5);
671 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
672 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
676 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
677 ics->prediction_used[sfb] = get_bits1(gb);
683 * Decode Long Term Prediction data; reference: table 4.xx.
685 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
686 GetBitContext *gb, uint8_t max_sfb)
690 ltp->lag = get_bits(gb, 11);
691 ltp->coef = ltp_coef[get_bits(gb, 3)];
692 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
693 ltp->used[sfb] = get_bits1(gb);
697 * Decode Individual Channel Stream info; reference: table 4.6.
699 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
701 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
702 GetBitContext *gb, int common_window)
705 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
706 memset(ics, 0, sizeof(IndividualChannelStream));
709 ics->window_sequence[1] = ics->window_sequence[0];
710 ics->window_sequence[0] = get_bits(gb, 2);
711 ics->use_kb_window[1] = ics->use_kb_window[0];
712 ics->use_kb_window[0] = get_bits1(gb);
713 ics->num_window_groups = 1;
714 ics->group_len[0] = 1;
715 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
717 ics->max_sfb = get_bits(gb, 4);
718 for (i = 0; i < 7; i++) {
720 ics->group_len[ics->num_window_groups - 1]++;
722 ics->num_window_groups++;
723 ics->group_len[ics->num_window_groups - 1] = 1;
726 ics->num_windows = 8;
727 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
728 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
729 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
730 ics->predictor_present = 0;
732 ics->max_sfb = get_bits(gb, 6);
733 ics->num_windows = 1;
734 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
735 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
736 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
737 ics->predictor_present = get_bits1(gb);
738 ics->predictor_reset_group = 0;
739 if (ics->predictor_present) {
740 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
741 if (decode_prediction(ac, ics, gb)) {
742 memset(ics, 0, sizeof(IndividualChannelStream));
745 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
746 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
747 memset(ics, 0, sizeof(IndividualChannelStream));
750 if ((ics->ltp.present = get_bits(gb, 1)))
751 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
756 if (ics->max_sfb > ics->num_swb) {
757 av_log(ac->avctx, AV_LOG_ERROR,
758 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
759 ics->max_sfb, ics->num_swb);
760 memset(ics, 0, sizeof(IndividualChannelStream));
768 * Decode band types (section_data payload); reference: table 4.46.
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
773 * @return Returns error status. 0 - OK, !0 - error
775 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
776 int band_type_run_end[120], GetBitContext *gb,
777 IndividualChannelStream *ics)
780 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
781 for (g = 0; g < ics->num_window_groups; g++) {
783 while (k < ics->max_sfb) {
784 uint8_t sect_end = k;
786 int sect_band_type = get_bits(gb, 4);
787 if (sect_band_type == 12) {
788 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
791 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
792 sect_end += sect_len_incr;
793 sect_end += sect_len_incr;
794 if (get_bits_left(gb) < 0) {
795 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
798 if (sect_end > ics->max_sfb) {
799 av_log(ac->avctx, AV_LOG_ERROR,
800 "Number of bands (%d) exceeds limit (%d).\n",
801 sect_end, ics->max_sfb);
804 for (; k < sect_end; k++) {
805 band_type [idx] = sect_band_type;
806 band_type_run_end[idx++] = sect_end;
814 * Decode scalefactors; reference: table 4.47.
816 * @param global_gain first scalefactor value as scalefactors are differentially coded
817 * @param band_type array of the used band type
818 * @param band_type_run_end array of the last scalefactor band of a band type run
819 * @param sf array of scalefactors or intensity stereo positions
821 * @return Returns error status. 0 - OK, !0 - error
823 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
824 unsigned int global_gain,
825 IndividualChannelStream *ics,
826 enum BandType band_type[120],
827 int band_type_run_end[120])
830 int offset[3] = { global_gain, global_gain - 90, 0 };
833 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
834 for (g = 0; g < ics->num_window_groups; g++) {
835 for (i = 0; i < ics->max_sfb;) {
836 int run_end = band_type_run_end[idx];
837 if (band_type[idx] == ZERO_BT) {
838 for (; i < run_end; i++, idx++)
840 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
841 for (; i < run_end; i++, idx++) {
842 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
843 clipped_offset = av_clip(offset[2], -155, 100);
844 if (offset[2] != clipped_offset) {
845 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
846 "position clipped (%d -> %d).\nIf you heard an "
847 "audible artifact, there may be a bug in the "
848 "decoder. ", offset[2], clipped_offset);
850 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
852 } else if (band_type[idx] == NOISE_BT) {
853 for (; i < run_end; i++, idx++) {
854 if (noise_flag-- > 0)
855 offset[1] += get_bits(gb, 9) - 256;
857 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
858 clipped_offset = av_clip(offset[1], -100, 155);
859 if (offset[1] != clipped_offset) {
860 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
861 "(%d -> %d).\nIf you heard an audible "
862 "artifact, there may be a bug in the decoder. ",
863 offset[1], clipped_offset);
865 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
868 for (; i < run_end; i++, idx++) {
869 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
870 if (offset[0] > 255U) {
871 av_log(ac->avctx, AV_LOG_ERROR,
872 "%s (%d) out of range.\n", sf_str[0], offset[0]);
875 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
884 * Decode pulse data; reference: table 4.7.
886 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
887 const uint16_t *swb_offset, int num_swb)
890 pulse->num_pulse = get_bits(gb, 2) + 1;
891 pulse_swb = get_bits(gb, 6);
892 if (pulse_swb >= num_swb)
894 pulse->pos[0] = swb_offset[pulse_swb];
895 pulse->pos[0] += get_bits(gb, 5);
896 if (pulse->pos[0] > 1023)
898 pulse->amp[0] = get_bits(gb, 4);
899 for (i = 1; i < pulse->num_pulse; i++) {
900 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
901 if (pulse->pos[i] > 1023)
903 pulse->amp[i] = get_bits(gb, 4);
909 * Decode Temporal Noise Shaping data; reference: table 4.48.
911 * @return Returns error status. 0 - OK, !0 - error
913 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
914 GetBitContext *gb, const IndividualChannelStream *ics)
916 int w, filt, i, coef_len, coef_res, coef_compress;
917 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
918 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
919 for (w = 0; w < ics->num_windows; w++) {
920 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
921 coef_res = get_bits1(gb);
923 for (filt = 0; filt < tns->n_filt[w]; filt++) {
925 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
927 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
928 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
929 tns->order[w][filt], tns_max_order);
930 tns->order[w][filt] = 0;
933 if (tns->order[w][filt]) {
934 tns->direction[w][filt] = get_bits1(gb);
935 coef_compress = get_bits1(gb);
936 coef_len = coef_res + 3 - coef_compress;
937 tmp2_idx = 2 * coef_compress + coef_res;
939 for (i = 0; i < tns->order[w][filt]; i++)
940 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
949 * Decode Mid/Side data; reference: table 4.54.
951 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
952 * [1] mask is decoded from bitstream; [2] mask is all 1s;
953 * [3] reserved for scalable AAC
955 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
959 if (ms_present == 1) {
960 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
961 cpe->ms_mask[idx] = get_bits1(gb);
962 } else if (ms_present == 2) {
963 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
968 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
972 *dst++ = v[idx & 15] * s;
973 *dst++ = v[idx>>4 & 15] * s;
979 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
983 *dst++ = v[idx & 3] * s;
984 *dst++ = v[idx>>2 & 3] * s;
985 *dst++ = v[idx>>4 & 3] * s;
986 *dst++ = v[idx>>6 & 3] * s;
992 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
993 unsigned sign, const float *scale)
995 union float754 s0, s1;
997 s0.f = s1.f = *scale;
998 s0.i ^= sign >> 1 << 31;
1001 *dst++ = v[idx & 15] * s0.f;
1002 *dst++ = v[idx>>4 & 15] * s1.f;
1009 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1010 unsigned sign, const float *scale)
1012 unsigned nz = idx >> 12;
1013 union float754 s = { .f = *scale };
1016 t.i = s.i ^ (sign & 1U<<31);
1017 *dst++ = v[idx & 3] * t.f;
1019 sign <<= nz & 1; nz >>= 1;
1020 t.i = s.i ^ (sign & 1U<<31);
1021 *dst++ = v[idx>>2 & 3] * t.f;
1023 sign <<= nz & 1; nz >>= 1;
1024 t.i = s.i ^ (sign & 1U<<31);
1025 *dst++ = v[idx>>4 & 3] * t.f;
1027 sign <<= nz & 1; nz >>= 1;
1028 t.i = s.i ^ (sign & 1U<<31);
1029 *dst++ = v[idx>>6 & 3] * t.f;
1036 * Decode spectral data; reference: table 4.50.
1037 * Dequantize and scale spectral data; reference: 4.6.3.3.
1039 * @param coef array of dequantized, scaled spectral data
1040 * @param sf array of scalefactors or intensity stereo positions
1041 * @param pulse_present set if pulses are present
1042 * @param pulse pointer to pulse data struct
1043 * @param band_type array of the used band type
1045 * @return Returns error status. 0 - OK, !0 - error
1047 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1048 GetBitContext *gb, const float sf[120],
1049 int pulse_present, const Pulse *pulse,
1050 const IndividualChannelStream *ics,
1051 enum BandType band_type[120])
1053 int i, k, g, idx = 0;
1054 const int c = 1024 / ics->num_windows;
1055 const uint16_t *offsets = ics->swb_offset;
1056 float *coef_base = coef;
1058 for (g = 0; g < ics->num_windows; g++)
1059 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1061 for (g = 0; g < ics->num_window_groups; g++) {
1062 unsigned g_len = ics->group_len[g];
1064 for (i = 0; i < ics->max_sfb; i++, idx++) {
1065 const unsigned cbt_m1 = band_type[idx] - 1;
1066 float *cfo = coef + offsets[i];
1067 int off_len = offsets[i + 1] - offsets[i];
1070 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1071 for (group = 0; group < g_len; group++, cfo+=128) {
1072 memset(cfo, 0, off_len * sizeof(float));
1074 } else if (cbt_m1 == NOISE_BT - 1) {
1075 for (group = 0; group < g_len; group++, cfo+=128) {
1079 for (k = 0; k < off_len; k++) {
1080 ac->random_state = lcg_random(ac->random_state);
1081 cfo[k] = ac->random_state;
1084 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1085 scale = sf[idx] / sqrtf(band_energy);
1086 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1089 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1090 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1091 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1092 OPEN_READER(re, gb);
1094 switch (cbt_m1 >> 1) {
1096 for (group = 0; group < g_len; group++, cfo+=128) {
1104 UPDATE_CACHE(re, gb);
1105 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1106 cb_idx = cb_vector_idx[code];
1107 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1113 for (group = 0; group < g_len; group++, cfo+=128) {
1123 UPDATE_CACHE(re, gb);
1124 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1125 cb_idx = cb_vector_idx[code];
1126 nnz = cb_idx >> 8 & 15;
1127 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1128 LAST_SKIP_BITS(re, gb, nnz);
1129 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1135 for (group = 0; group < g_len; group++, cfo+=128) {
1143 UPDATE_CACHE(re, gb);
1144 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1145 cb_idx = cb_vector_idx[code];
1146 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1153 for (group = 0; group < g_len; group++, cfo+=128) {
1163 UPDATE_CACHE(re, gb);
1164 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1165 cb_idx = cb_vector_idx[code];
1166 nnz = cb_idx >> 8 & 15;
1167 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1168 LAST_SKIP_BITS(re, gb, nnz);
1169 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1175 for (group = 0; group < g_len; group++, cfo+=128) {
1177 uint32_t *icf = (uint32_t *) cf;
1187 UPDATE_CACHE(re, gb);
1188 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1196 cb_idx = cb_vector_idx[code];
1199 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1200 LAST_SKIP_BITS(re, gb, nnz);
1202 for (j = 0; j < 2; j++) {
1206 /* The total length of escape_sequence must be < 22 bits according
1207 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1208 UPDATE_CACHE(re, gb);
1209 b = GET_CACHE(re, gb);
1210 b = 31 - av_log2(~b);
1213 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1217 SKIP_BITS(re, gb, b + 1);
1219 n = (1 << b) + SHOW_UBITS(re, gb, b);
1220 LAST_SKIP_BITS(re, gb, b);
1221 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1224 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1225 *icf++ = (bits & 1U<<31) | v;
1232 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1236 CLOSE_READER(re, gb);
1242 if (pulse_present) {
1244 for (i = 0; i < pulse->num_pulse; i++) {
1245 float co = coef_base[ pulse->pos[i] ];
1246 while (offsets[idx + 1] <= pulse->pos[i])
1248 if (band_type[idx] != NOISE_BT && sf[idx]) {
1249 float ico = -pulse->amp[i];
1252 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1254 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1261 static av_always_inline float flt16_round(float pf)
1265 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1269 static av_always_inline float flt16_even(float pf)
1273 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1277 static av_always_inline float flt16_trunc(float pf)
1281 pun.i &= 0xFFFF0000U;
1285 static av_always_inline void predict(PredictorState *ps, float *coef,
1288 const float a = 0.953125; // 61.0 / 64
1289 const float alpha = 0.90625; // 29.0 / 32
1293 float r0 = ps->r0, r1 = ps->r1;
1294 float cor0 = ps->cor0, cor1 = ps->cor1;
1295 float var0 = ps->var0, var1 = ps->var1;
1297 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1298 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1300 pv = flt16_round(k1 * r0 + k2 * r1);
1307 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1308 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1309 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1310 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1312 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1313 ps->r0 = flt16_trunc(a * e0);
1317 * Apply AAC-Main style frequency domain prediction.
1319 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1323 if (!sce->ics.predictor_initialized) {
1324 reset_all_predictors(sce->predictor_state);
1325 sce->ics.predictor_initialized = 1;
1328 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1329 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1330 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1331 predict(&sce->predictor_state[k], &sce->coeffs[k],
1332 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1335 if (sce->ics.predictor_reset_group)
1336 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1338 reset_all_predictors(sce->predictor_state);
1342 * Decode an individual_channel_stream payload; reference: table 4.44.
1344 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1345 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1347 * @return Returns error status. 0 - OK, !0 - error
1349 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1350 GetBitContext *gb, int common_window, int scale_flag)
1353 TemporalNoiseShaping *tns = &sce->tns;
1354 IndividualChannelStream *ics = &sce->ics;
1355 float *out = sce->coeffs;
1356 int global_gain, pulse_present = 0;
1358 /* This assignment is to silence a GCC warning about the variable being used
1359 * uninitialized when in fact it always is.
1361 pulse.num_pulse = 0;
1363 global_gain = get_bits(gb, 8);
1365 if (!common_window && !scale_flag) {
1366 if (decode_ics_info(ac, ics, gb, 0) < 0)
1370 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1372 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1377 if ((pulse_present = get_bits1(gb))) {
1378 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1379 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1382 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1383 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1387 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1389 if (get_bits1(gb)) {
1390 av_log_missing_feature(ac->avctx, "SSR", 1);
1395 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1398 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1399 apply_prediction(ac, sce);
1405 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1407 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1409 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1410 float *ch0 = cpe->ch[0].coeffs;
1411 float *ch1 = cpe->ch[1].coeffs;
1412 int g, i, group, idx = 0;
1413 const uint16_t *offsets = ics->swb_offset;
1414 for (g = 0; g < ics->num_window_groups; g++) {
1415 for (i = 0; i < ics->max_sfb; i++, idx++) {
1416 if (cpe->ms_mask[idx] &&
1417 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1418 for (group = 0; group < ics->group_len[g]; group++) {
1419 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1420 ch1 + group * 128 + offsets[i],
1421 offsets[i+1] - offsets[i]);
1425 ch0 += ics->group_len[g] * 128;
1426 ch1 += ics->group_len[g] * 128;
1431 * intensity stereo decoding; reference: 4.6.8.2.3
1433 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1434 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1435 * [3] reserved for scalable AAC
1437 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1439 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1440 SingleChannelElement *sce1 = &cpe->ch[1];
1441 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1442 const uint16_t *offsets = ics->swb_offset;
1443 int g, group, i, idx = 0;
1446 for (g = 0; g < ics->num_window_groups; g++) {
1447 for (i = 0; i < ics->max_sfb;) {
1448 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1449 const int bt_run_end = sce1->band_type_run_end[idx];
1450 for (; i < bt_run_end; i++, idx++) {
1451 c = -1 + 2 * (sce1->band_type[idx] - 14);
1453 c *= 1 - 2 * cpe->ms_mask[idx];
1454 scale = c * sce1->sf[idx];
1455 for (group = 0; group < ics->group_len[g]; group++)
1456 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1457 coef0 + group * 128 + offsets[i],
1459 offsets[i + 1] - offsets[i]);
1462 int bt_run_end = sce1->band_type_run_end[idx];
1463 idx += bt_run_end - i;
1467 coef0 += ics->group_len[g] * 128;
1468 coef1 += ics->group_len[g] * 128;
1473 * Decode a channel_pair_element; reference: table 4.4.
1475 * @return Returns error status. 0 - OK, !0 - error
1477 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1479 int i, ret, common_window, ms_present = 0;
1481 common_window = get_bits1(gb);
1482 if (common_window) {
1483 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1485 i = cpe->ch[1].ics.use_kb_window[0];
1486 cpe->ch[1].ics = cpe->ch[0].ics;
1487 cpe->ch[1].ics.use_kb_window[1] = i;
1488 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1489 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1490 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1491 ms_present = get_bits(gb, 2);
1492 if (ms_present == 3) {
1493 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1495 } else if (ms_present)
1496 decode_mid_side_stereo(cpe, gb, ms_present);
1498 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1500 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1503 if (common_window) {
1505 apply_mid_side_stereo(ac, cpe);
1506 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1507 apply_prediction(ac, &cpe->ch[0]);
1508 apply_prediction(ac, &cpe->ch[1]);
1512 apply_intensity_stereo(ac, cpe, ms_present);
1516 static const float cce_scale[] = {
1517 1.09050773266525765921, //2^(1/8)
1518 1.18920711500272106672, //2^(1/4)
1524 * Decode coupling_channel_element; reference: table 4.8.
1526 * @return Returns error status. 0 - OK, !0 - error
1528 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1534 SingleChannelElement *sce = &che->ch[0];
1535 ChannelCoupling *coup = &che->coup;
1537 coup->coupling_point = 2 * get_bits1(gb);
1538 coup->num_coupled = get_bits(gb, 3);
1539 for (c = 0; c <= coup->num_coupled; c++) {
1541 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1542 coup->id_select[c] = get_bits(gb, 4);
1543 if (coup->type[c] == TYPE_CPE) {
1544 coup->ch_select[c] = get_bits(gb, 2);
1545 if (coup->ch_select[c] == 3)
1548 coup->ch_select[c] = 2;
1550 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1552 sign = get_bits(gb, 1);
1553 scale = cce_scale[get_bits(gb, 2)];
1555 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1558 for (c = 0; c < num_gain; c++) {
1562 float gain_cache = 1.;
1564 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1565 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1566 gain_cache = powf(scale, -gain);
1568 if (coup->coupling_point == AFTER_IMDCT) {
1569 coup->gain[c][0] = gain_cache;
1571 for (g = 0; g < sce->ics.num_window_groups; g++) {
1572 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1573 if (sce->band_type[idx] != ZERO_BT) {
1575 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1583 gain_cache = powf(scale, -t) * s;
1586 coup->gain[c][idx] = gain_cache;
1596 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1598 * @return Returns number of bytes consumed.
1600 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1604 int num_excl_chan = 0;
1607 for (i = 0; i < 7; i++)
1608 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1609 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1611 return num_excl_chan / 7;
1615 * Decode dynamic range information; reference: table 4.52.
1617 * @param cnt length of TYPE_FIL syntactic element in bytes
1619 * @return Returns number of bytes consumed.
1621 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1622 GetBitContext *gb, int cnt)
1625 int drc_num_bands = 1;
1628 /* pce_tag_present? */
1629 if (get_bits1(gb)) {
1630 che_drc->pce_instance_tag = get_bits(gb, 4);
1631 skip_bits(gb, 4); // tag_reserved_bits
1635 /* excluded_chns_present? */
1636 if (get_bits1(gb)) {
1637 n += decode_drc_channel_exclusions(che_drc, gb);
1640 /* drc_bands_present? */
1641 if (get_bits1(gb)) {
1642 che_drc->band_incr = get_bits(gb, 4);
1643 che_drc->interpolation_scheme = get_bits(gb, 4);
1645 drc_num_bands += che_drc->band_incr;
1646 for (i = 0; i < drc_num_bands; i++) {
1647 che_drc->band_top[i] = get_bits(gb, 8);
1652 /* prog_ref_level_present? */
1653 if (get_bits1(gb)) {
1654 che_drc->prog_ref_level = get_bits(gb, 7);
1655 skip_bits1(gb); // prog_ref_level_reserved_bits
1659 for (i = 0; i < drc_num_bands; i++) {
1660 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1661 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1669 * Decode extension data (incomplete); reference: table 4.51.
1671 * @param cnt length of TYPE_FIL syntactic element in bytes
1673 * @return Returns number of bytes consumed
1675 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1676 ChannelElement *che, enum RawDataBlockType elem_type)
1680 switch (get_bits(gb, 4)) { // extension type
1681 case EXT_SBR_DATA_CRC:
1685 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1687 } else if (!ac->m4ac.sbr) {
1688 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1689 skip_bits_long(gb, 8 * cnt - 4);
1691 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1692 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1693 skip_bits_long(gb, 8 * cnt - 4);
1695 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1698 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1702 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1704 case EXT_DYNAMIC_RANGE:
1705 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1709 case EXT_DATA_ELEMENT:
1711 skip_bits_long(gb, 8 * cnt - 4);
1718 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1720 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1721 * @param coef spectral coefficients
1723 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1724 IndividualChannelStream *ics, int decode)
1726 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1728 int bottom, top, order, start, end, size, inc;
1729 float lpc[TNS_MAX_ORDER];
1730 float tmp[TNS_MAX_ORDER];
1732 for (w = 0; w < ics->num_windows; w++) {
1733 bottom = ics->num_swb;
1734 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1736 bottom = FFMAX(0, top - tns->length[w][filt]);
1737 order = tns->order[w][filt];
1742 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1744 start = ics->swb_offset[FFMIN(bottom, mmm)];
1745 end = ics->swb_offset[FFMIN( top, mmm)];
1746 if ((size = end - start) <= 0)
1748 if (tns->direction[w][filt]) {
1758 for (m = 0; m < size; m++, start += inc)
1759 for (i = 1; i <= FFMIN(m, order); i++)
1760 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1763 for (m = 0; m < size; m++, start += inc) {
1764 tmp[0] = coef[start];
1765 for (i = 1; i <= FFMIN(m, order); i++)
1766 coef[start] += tmp[i] * lpc[i - 1];
1767 for (i = order; i > 0; i--)
1768 tmp[i] = tmp[i - 1];
1776 * Apply windowing and MDCT to obtain the spectral
1777 * coefficient from the predicted sample by LTP.
1779 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1780 float *in, IndividualChannelStream *ics)
1782 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1783 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1784 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1785 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1787 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1788 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1790 memset(in, 0, 448 * sizeof(float));
1791 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1792 memcpy(in + 576, in + 576, 448 * sizeof(float));
1794 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1795 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1797 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1798 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1799 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1801 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1805 * Apply the long term prediction
1807 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1809 const LongTermPrediction *ltp = &sce->ics.ltp;
1810 const uint16_t *offsets = sce->ics.swb_offset;
1813 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1814 float *predTime = sce->ret;
1815 float *predFreq = ac->buf_mdct;
1816 int16_t num_samples = 2048;
1818 if (ltp->lag < 1024)
1819 num_samples = ltp->lag + 1024;
1820 for (i = 0; i < num_samples; i++)
1821 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1822 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1824 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1826 if (sce->tns.present)
1827 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1829 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1831 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1832 sce->coeffs[i] += predFreq[i];
1837 * Update the LTP buffer for next frame
1839 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1841 IndividualChannelStream *ics = &sce->ics;
1842 float *saved = sce->saved;
1843 float *saved_ltp = sce->coeffs;
1844 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1845 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1848 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1849 memcpy(saved_ltp, saved, 512 * sizeof(float));
1850 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1851 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1852 for (i = 0; i < 64; i++)
1853 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1854 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1855 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1856 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1857 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1858 for (i = 0; i < 64; i++)
1859 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1860 } else { // LONG_STOP or ONLY_LONG
1861 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1862 for (i = 0; i < 512; i++)
1863 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1866 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1867 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1868 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1872 * Conduct IMDCT and windowing.
1874 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1876 IndividualChannelStream *ics = &sce->ics;
1877 float *in = sce->coeffs;
1878 float *out = sce->ret;
1879 float *saved = sce->saved;
1880 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1881 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1882 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1883 float *buf = ac->buf_mdct;
1884 float *temp = ac->temp;
1888 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1889 for (i = 0; i < 1024; i += 128)
1890 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1892 ac->mdct.imdct_half(&ac->mdct, buf, in);
1894 /* window overlapping
1895 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1896 * and long to short transitions are considered to be short to short
1897 * transitions. This leaves just two cases (long to long and short to short)
1898 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1900 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1901 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1902 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1904 memcpy( out, saved, 448 * sizeof(float));
1906 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1907 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1908 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1909 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1910 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1911 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1912 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1914 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1915 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1920 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1921 memcpy( saved, temp + 64, 64 * sizeof(float));
1922 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1923 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1924 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1925 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1926 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1927 memcpy( saved, buf + 512, 448 * sizeof(float));
1928 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1929 } else { // LONG_STOP or ONLY_LONG
1930 memcpy( saved, buf + 512, 512 * sizeof(float));
1935 * Apply dependent channel coupling (applied before IMDCT).
1937 * @param index index into coupling gain array
1939 static void apply_dependent_coupling(AACContext *ac,
1940 SingleChannelElement *target,
1941 ChannelElement *cce, int index)
1943 IndividualChannelStream *ics = &cce->ch[0].ics;
1944 const uint16_t *offsets = ics->swb_offset;
1945 float *dest = target->coeffs;
1946 const float *src = cce->ch[0].coeffs;
1947 int g, i, group, k, idx = 0;
1948 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1949 av_log(ac->avctx, AV_LOG_ERROR,
1950 "Dependent coupling is not supported together with LTP\n");
1953 for (g = 0; g < ics->num_window_groups; g++) {
1954 for (i = 0; i < ics->max_sfb; i++, idx++) {
1955 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1956 const float gain = cce->coup.gain[index][idx];
1957 for (group = 0; group < ics->group_len[g]; group++) {
1958 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1960 dest[group * 128 + k] += gain * src[group * 128 + k];
1965 dest += ics->group_len[g] * 128;
1966 src += ics->group_len[g] * 128;
1971 * Apply independent channel coupling (applied after IMDCT).
1973 * @param index index into coupling gain array
1975 static void apply_independent_coupling(AACContext *ac,
1976 SingleChannelElement *target,
1977 ChannelElement *cce, int index)
1980 const float gain = cce->coup.gain[index][0];
1981 const float *src = cce->ch[0].ret;
1982 float *dest = target->ret;
1983 const int len = 1024 << (ac->m4ac.sbr == 1);
1985 for (i = 0; i < len; i++)
1986 dest[i] += gain * src[i];
1990 * channel coupling transformation interface
1992 * @param apply_coupling_method pointer to (in)dependent coupling function
1994 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1995 enum RawDataBlockType type, int elem_id,
1996 enum CouplingPoint coupling_point,
1997 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2001 for (i = 0; i < MAX_ELEM_ID; i++) {
2002 ChannelElement *cce = ac->che[TYPE_CCE][i];
2005 if (cce && cce->coup.coupling_point == coupling_point) {
2006 ChannelCoupling *coup = &cce->coup;
2008 for (c = 0; c <= coup->num_coupled; c++) {
2009 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2010 if (coup->ch_select[c] != 1) {
2011 apply_coupling_method(ac, &cc->ch[0], cce, index);
2012 if (coup->ch_select[c] != 0)
2015 if (coup->ch_select[c] != 2)
2016 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2018 index += 1 + (coup->ch_select[c] == 3);
2025 * Convert spectral data to float samples, applying all supported tools as appropriate.
2027 static void spectral_to_sample(AACContext *ac)
2030 for (type = 3; type >= 0; type--) {
2031 for (i = 0; i < MAX_ELEM_ID; i++) {
2032 ChannelElement *che = ac->che[type][i];
2034 if (type <= TYPE_CPE)
2035 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2036 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2037 if (che->ch[0].ics.predictor_present) {
2038 if (che->ch[0].ics.ltp.present)
2039 apply_ltp(ac, &che->ch[0]);
2040 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2041 apply_ltp(ac, &che->ch[1]);
2044 if (che->ch[0].tns.present)
2045 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2046 if (che->ch[1].tns.present)
2047 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2048 if (type <= TYPE_CPE)
2049 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2050 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2051 imdct_and_windowing(ac, &che->ch[0]);
2052 if (ac->m4ac.object_type == AOT_AAC_LTP)
2053 update_ltp(ac, &che->ch[0]);
2054 if (type == TYPE_CPE) {
2055 imdct_and_windowing(ac, &che->ch[1]);
2056 if (ac->m4ac.object_type == AOT_AAC_LTP)
2057 update_ltp(ac, &che->ch[1]);
2059 if (ac->m4ac.sbr > 0) {
2060 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2063 if (type <= TYPE_CCE)
2064 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2070 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2073 AACADTSHeaderInfo hdr_info;
2075 size = ff_aac_parse_header(gb, &hdr_info);
2077 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2078 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2079 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2080 ac->m4ac.chan_config = hdr_info.chan_config;
2081 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2083 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2085 } else if (ac->output_configured != OC_LOCKED) {
2086 ac->m4ac.chan_config = 0;
2087 ac->output_configured = OC_NONE;
2089 if (ac->output_configured != OC_LOCKED) {
2093 ac->m4ac.sample_rate = hdr_info.sample_rate;
2094 ac->m4ac.sampling_index = hdr_info.sampling_index;
2095 ac->m4ac.object_type = hdr_info.object_type;
2096 if (!ac->avctx->sample_rate)
2097 ac->avctx->sample_rate = hdr_info.sample_rate;
2098 if (hdr_info.num_aac_frames == 1) {
2099 if (!hdr_info.crc_absent)
2102 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2109 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2110 int *data_size, GetBitContext *gb)
2112 AACContext *ac = avctx->priv_data;
2113 ChannelElement *che = NULL, *che_prev = NULL;
2114 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2115 int err, elem_id, data_size_tmp;
2116 int samples = 0, multiplier;
2118 if (show_bits(gb, 12) == 0xfff) {
2119 if (parse_adts_frame_header(ac, gb) < 0) {
2120 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2123 if (ac->m4ac.sampling_index > 12) {
2124 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2129 ac->tags_mapped = 0;
2131 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2132 elem_id = get_bits(gb, 4);
2134 if (elem_type < TYPE_DSE) {
2135 if (!(che=get_che(ac, elem_type, elem_id))) {
2136 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2137 elem_type, elem_id);
2143 switch (elem_type) {
2146 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2150 err = decode_cpe(ac, gb, che);
2154 err = decode_cce(ac, gb, che);
2158 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2162 err = skip_data_stream_element(ac, gb);
2166 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2167 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2168 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2170 if (ac->output_configured > OC_TRIAL_PCE)
2171 av_log(avctx, AV_LOG_ERROR,
2172 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2174 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2180 elem_id += get_bits(gb, 8) - 1;
2181 if (get_bits_left(gb) < 8 * elem_id) {
2182 av_log(avctx, AV_LOG_ERROR, overread_err);
2186 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2187 err = 0; /* FIXME */
2191 err = -1; /* should not happen, but keeps compiler happy */
2196 elem_type_prev = elem_type;
2201 if (get_bits_left(gb) < 3) {
2202 av_log(avctx, AV_LOG_ERROR, overread_err);
2207 spectral_to_sample(ac);
2209 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2210 samples <<= multiplier;
2211 if (ac->output_configured < OC_LOCKED) {
2212 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2213 avctx->frame_size = samples;
2216 data_size_tmp = samples * avctx->channels *
2217 av_get_bytes_per_sample(avctx->sample_fmt);
2218 if (*data_size < data_size_tmp) {
2219 av_log(avctx, AV_LOG_ERROR,
2220 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2221 *data_size, data_size_tmp);
2224 *data_size = data_size_tmp;
2227 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2228 ac->fmt_conv.float_interleave(data, (const float **)ac->output_data,
2229 samples, avctx->channels);
2231 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data,
2232 samples, avctx->channels);
2235 if (ac->output_configured)
2236 ac->output_configured = OC_LOCKED;
2241 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2242 int *data_size, AVPacket *avpkt)
2244 const uint8_t *buf = avpkt->data;
2245 int buf_size = avpkt->size;
2251 init_get_bits(&gb, buf, buf_size * 8);
2253 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2256 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2257 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2258 if (buf[buf_offset])
2261 return buf_size > buf_offset ? buf_consumed : buf_size;
2264 static av_cold int aac_decode_close(AVCodecContext *avctx)
2266 AACContext *ac = avctx->priv_data;
2269 for (i = 0; i < MAX_ELEM_ID; i++) {
2270 for (type = 0; type < 4; type++) {
2271 if (ac->che[type][i])
2272 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2273 av_freep(&ac->che[type][i]);
2277 ff_mdct_end(&ac->mdct);
2278 ff_mdct_end(&ac->mdct_small);
2279 ff_mdct_end(&ac->mdct_ltp);
2284 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2286 struct LATMContext {
2287 AACContext aac_ctx; ///< containing AACContext
2288 int initialized; ///< initilized after a valid extradata was seen
2291 int audio_mux_version_A; ///< LATM syntax version
2292 int frame_length_type; ///< 0/1 variable/fixed frame length
2293 int frame_length; ///< frame length for fixed frame length
2296 static inline uint32_t latm_get_value(GetBitContext *b)
2298 int length = get_bits(b, 2);
2300 return get_bits_long(b, (length+1)*8);
2303 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2306 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2307 MPEG4AudioConfig m4ac;
2308 int config_start_bit = get_bits_count(gb);
2309 int bits_consumed, esize;
2311 if (config_start_bit % 8) {
2312 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2313 "config not byte aligned.\n", 1);
2314 return AVERROR_INVALIDDATA;
2317 decode_audio_specific_config(NULL, avctx, &m4ac,
2318 gb->buffer + (config_start_bit / 8),
2319 get_bits_left(gb) / 8);
2321 if (bits_consumed < 0)
2322 return AVERROR_INVALIDDATA;
2324 esize = (bits_consumed+7) / 8;
2326 if (avctx->extradata_size <= esize) {
2327 av_free(avctx->extradata);
2328 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2329 if (!avctx->extradata)
2330 return AVERROR(ENOMEM);
2333 avctx->extradata_size = esize;
2334 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2335 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2337 skip_bits_long(gb, bits_consumed);
2340 return bits_consumed;
2343 static int read_stream_mux_config(struct LATMContext *latmctx,
2346 int ret, audio_mux_version = get_bits(gb, 1);
2348 latmctx->audio_mux_version_A = 0;
2349 if (audio_mux_version)
2350 latmctx->audio_mux_version_A = get_bits(gb, 1);
2352 if (!latmctx->audio_mux_version_A) {
2354 if (audio_mux_version)
2355 latm_get_value(gb); // taraFullness
2357 skip_bits(gb, 1); // allStreamSameTimeFraming
2358 skip_bits(gb, 6); // numSubFrames
2360 if (get_bits(gb, 4)) { // numPrograms
2361 av_log_missing_feature(latmctx->aac_ctx.avctx,
2362 "multiple programs are not supported\n", 1);
2363 return AVERROR_PATCHWELCOME;
2366 // for each program (which there is only on in DVB)
2368 // for each layer (which there is only on in DVB)
2369 if (get_bits(gb, 3)) { // numLayer
2370 av_log_missing_feature(latmctx->aac_ctx.avctx,
2371 "multiple layers are not supported\n", 1);
2372 return AVERROR_PATCHWELCOME;
2375 // for all but first stream: use_same_config = get_bits(gb, 1);
2376 if (!audio_mux_version) {
2377 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2380 int ascLen = latm_get_value(gb);
2381 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2384 skip_bits_long(gb, ascLen);
2387 latmctx->frame_length_type = get_bits(gb, 3);
2388 switch (latmctx->frame_length_type) {
2390 skip_bits(gb, 8); // latmBufferFullness
2393 latmctx->frame_length = get_bits(gb, 9);
2398 skip_bits(gb, 6); // CELP frame length table index
2402 skip_bits(gb, 1); // HVXC frame length table index
2406 if (get_bits(gb, 1)) { // other data
2407 if (audio_mux_version) {
2408 latm_get_value(gb); // other_data_bits
2412 esc = get_bits(gb, 1);
2418 if (get_bits(gb, 1)) // crc present
2419 skip_bits(gb, 8); // config_crc
2425 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2429 if (ctx->frame_length_type == 0) {
2430 int mux_slot_length = 0;
2432 tmp = get_bits(gb, 8);
2433 mux_slot_length += tmp;
2434 } while (tmp == 255);
2435 return mux_slot_length;
2436 } else if (ctx->frame_length_type == 1) {
2437 return ctx->frame_length;
2438 } else if (ctx->frame_length_type == 3 ||
2439 ctx->frame_length_type == 5 ||
2440 ctx->frame_length_type == 7) {
2441 skip_bits(gb, 2); // mux_slot_length_coded
2446 static int read_audio_mux_element(struct LATMContext *latmctx,
2450 uint8_t use_same_mux = get_bits(gb, 1);
2451 if (!use_same_mux) {
2452 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2454 } else if (!latmctx->aac_ctx.avctx->extradata) {
2455 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2456 "no decoder config found\n");
2457 return AVERROR(EAGAIN);
2459 if (latmctx->audio_mux_version_A == 0) {
2460 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2461 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2462 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2463 return AVERROR_INVALIDDATA;
2464 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2465 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2466 "frame length mismatch %d << %d\n",
2467 mux_slot_length_bytes * 8, get_bits_left(gb));
2468 return AVERROR_INVALIDDATA;
2475 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2478 struct LATMContext *latmctx = avctx->priv_data;
2482 if (avpkt->size == 0)
2485 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2487 // check for LOAS sync word
2488 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2489 return AVERROR_INVALIDDATA;
2491 muxlength = get_bits(&gb, 13) + 3;
2492 // not enough data, the parser should have sorted this
2493 if (muxlength > avpkt->size)
2494 return AVERROR_INVALIDDATA;
2496 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2499 if (!latmctx->initialized) {
2500 if (!avctx->extradata) {
2504 aac_decode_close(avctx);
2505 if ((err = aac_decode_init(avctx)) < 0)
2507 latmctx->initialized = 1;
2511 if (show_bits(&gb, 12) == 0xfff) {
2512 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2513 "ADTS header detected, probably as result of configuration "
2515 return AVERROR_INVALIDDATA;
2518 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2524 av_cold static int latm_decode_init(AVCodecContext *avctx)
2526 struct LATMContext *latmctx = avctx->priv_data;
2529 ret = aac_decode_init(avctx);
2531 if (avctx->extradata_size > 0) {
2532 latmctx->initialized = !ret;
2534 latmctx->initialized = 0;
2541 AVCodec ff_aac_decoder = {
2543 .type = AVMEDIA_TYPE_AUDIO,
2545 .priv_data_size = sizeof(AACContext),
2546 .init = aac_decode_init,
2547 .close = aac_decode_close,
2548 .decode = aac_decode_frame,
2549 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2550 .sample_fmts = (const enum AVSampleFormat[]) {
2551 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2553 .capabilities = CODEC_CAP_CHANNEL_CONF,
2554 .channel_layouts = aac_channel_layout,
2558 Note: This decoder filter is intended to decode LATM streams transferred
2559 in MPEG transport streams which only contain one program.
2560 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2562 AVCodec ff_aac_latm_decoder = {
2564 .type = AVMEDIA_TYPE_AUDIO,
2565 .id = CODEC_ID_AAC_LATM,
2566 .priv_data_size = sizeof(struct LATMContext),
2567 .init = latm_decode_init,
2568 .close = aac_decode_close,
2569 .decode = latm_decode_frame,
2570 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2571 .sample_fmts = (const enum AVSampleFormat[]) {
2572 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2574 .capabilities = CODEC_CAP_CHANNEL_CONF,
2575 .channel_layouts = aac_channel_layout,