3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
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 FFmpeg 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]));
256 avctx->channels = channels;
258 ac->output_configured = oc_type;
264 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
266 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
267 * @param sce_map mono (Single Channel Element) map
268 * @param type speaker type/position for these channels
270 static void decode_channel_map(enum ChannelPosition *cpe_map,
271 enum ChannelPosition *sce_map,
272 enum ChannelPosition type,
273 GetBitContext *gb, int n)
276 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
277 map[get_bits(gb, 4)] = type;
282 * Decode program configuration element; reference: table 4.2.
284 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
286 * @return Returns error status. 0 - OK, !0 - error
288 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
289 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
292 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
295 skip_bits(gb, 2); // object_type
297 sampling_index = get_bits(gb, 4);
298 if (m4ac->sampling_index != sampling_index)
299 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");
301 num_front = get_bits(gb, 4);
302 num_side = get_bits(gb, 4);
303 num_back = get_bits(gb, 4);
304 num_lfe = get_bits(gb, 2);
305 num_assoc_data = get_bits(gb, 3);
306 num_cc = get_bits(gb, 4);
309 skip_bits(gb, 4); // mono_mixdown_tag
311 skip_bits(gb, 4); // stereo_mixdown_tag
314 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
316 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
317 av_log(avctx, AV_LOG_ERROR, overread_err);
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
321 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
322 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
323 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
325 skip_bits_long(gb, 4 * num_assoc_data);
327 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
331 /* comment field, first byte is length */
332 comment_len = get_bits(gb, 8) * 8;
333 if (get_bits_left(gb) < comment_len) {
334 av_log(avctx, AV_LOG_ERROR, overread_err);
337 skip_bits_long(gb, comment_len);
342 * Set up channel positions based on a default channel configuration
343 * as specified in table 1.17.
345 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
347 * @return Returns error status. 0 - OK, !0 - error
349 static av_cold int set_default_channel_config(AVCodecContext *avctx,
350 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
353 if (channel_config < 1 || channel_config > 7) {
354 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
359 /* default channel configurations:
361 * 1ch : front center (mono)
362 * 2ch : L + R (stereo)
363 * 3ch : front center + L + R
364 * 4ch : front center + L + R + back center
365 * 5ch : front center + L + R + back stereo
366 * 6ch : front center + L + R + back stereo + LFE
367 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
370 if (channel_config != 2)
371 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
372 if (channel_config > 1)
373 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
374 if (channel_config == 4)
375 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
376 if (channel_config > 4)
377 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
378 = AAC_CHANNEL_BACK; // back stereo
379 if (channel_config > 5)
380 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
381 if (channel_config == 7)
382 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
388 * Decode GA "General Audio" specific configuration; reference: table 4.1.
390 * @param ac pointer to AACContext, may be null
391 * @param avctx pointer to AVCCodecContext, used for logging
393 * @return Returns error status. 0 - OK, !0 - error
395 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
397 MPEG4AudioConfig *m4ac,
400 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
401 int extension_flag, ret;
403 if (get_bits1(gb)) { // frameLengthFlag
404 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
408 if (get_bits1(gb)) // dependsOnCoreCoder
409 skip_bits(gb, 14); // coreCoderDelay
410 extension_flag = get_bits1(gb);
412 if (m4ac->object_type == AOT_AAC_SCALABLE ||
413 m4ac->object_type == AOT_ER_AAC_SCALABLE)
414 skip_bits(gb, 3); // layerNr
416 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
417 if (channel_config == 0) {
418 skip_bits(gb, 4); // element_instance_tag
419 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
422 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
425 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
428 if (extension_flag) {
429 switch (m4ac->object_type) {
431 skip_bits(gb, 5); // numOfSubFrame
432 skip_bits(gb, 11); // layer_length
436 case AOT_ER_AAC_SCALABLE:
438 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
439 * aacScalefactorDataResilienceFlag
440 * aacSpectralDataResilienceFlag
444 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
450 * Decode audio specific configuration; reference: table 1.13.
452 * @param ac pointer to AACContext, may be null
453 * @param avctx pointer to AVCCodecContext, used for logging
454 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
455 * @param data pointer to AVCodecContext extradata
456 * @param data_size size of AVCCodecContext extradata
458 * @return Returns error status or number of consumed bits. <0 - error
460 static int decode_audio_specific_config(AACContext *ac,
461 AVCodecContext *avctx,
462 MPEG4AudioConfig *m4ac,
463 const uint8_t *data, int data_size)
468 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
469 for (i = 0; i < avctx->extradata_size; i++)
470 av_dlog(avctx, "%02x ", avctx->extradata[i]);
471 av_dlog(avctx, "\n");
473 init_get_bits(&gb, data, data_size * 8);
475 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
477 if (m4ac->sampling_index > 12) {
478 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
481 if (m4ac->sbr == 1 && m4ac->ps == -1)
484 skip_bits_long(&gb, i);
486 switch (m4ac->object_type) {
490 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
494 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
495 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
499 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
500 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
501 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
503 return get_bits_count(&gb);
507 * linear congruential pseudorandom number generator
509 * @param previous_val pointer to the current state of the generator
511 * @return Returns a 32-bit pseudorandom integer
513 static av_always_inline int lcg_random(int previous_val)
515 return previous_val * 1664525 + 1013904223;
518 static av_always_inline void reset_predict_state(PredictorState *ps)
528 static void reset_all_predictors(PredictorState *ps)
531 for (i = 0; i < MAX_PREDICTORS; i++)
532 reset_predict_state(&ps[i]);
535 static int sample_rate_idx (int rate)
537 if (92017 <= rate) return 0;
538 else if (75132 <= rate) return 1;
539 else if (55426 <= rate) return 2;
540 else if (46009 <= rate) return 3;
541 else if (37566 <= rate) return 4;
542 else if (27713 <= rate) return 5;
543 else if (23004 <= rate) return 6;
544 else if (18783 <= rate) return 7;
545 else if (13856 <= rate) return 8;
546 else if (11502 <= rate) return 9;
547 else if (9391 <= rate) return 10;
551 static void reset_predictor_group(PredictorState *ps, int group_num)
554 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
555 reset_predict_state(&ps[i]);
558 #define AAC_INIT_VLC_STATIC(num, size) \
559 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
560 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
561 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
564 static av_cold int aac_decode_init(AVCodecContext *avctx)
566 AACContext *ac = avctx->priv_data;
567 float output_scale_factor;
570 ac->m4ac.sample_rate = avctx->sample_rate;
572 if (avctx->extradata_size > 0) {
573 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
575 avctx->extradata_size) < 0)
579 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
581 sr = sample_rate_idx(avctx->sample_rate);
582 ac->m4ac.sampling_index = sr;
583 ac->m4ac.channels = avctx->channels;
585 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
586 if (ff_mpeg4audio_channels[i] == avctx->channels)
588 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
591 ac->m4ac.chan_config = i;
593 if (ac->m4ac.chan_config) {
594 if (set_default_channel_config(avctx, new_che_pos, ac->m4ac.chan_config) < 0 &&
595 avctx->error_recognition >= FF_ER_EXPLODE)
596 return AVERROR_INVALIDDATA;
597 output_configure(ac, ac->che_pos, new_che_pos, ac->m4ac.chan_config, OC_GLOBAL_HDR);
601 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
602 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
603 output_scale_factor = 1.0 / 32768.0;
605 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
606 output_scale_factor = 1.0;
609 AAC_INIT_VLC_STATIC( 0, 304);
610 AAC_INIT_VLC_STATIC( 1, 270);
611 AAC_INIT_VLC_STATIC( 2, 550);
612 AAC_INIT_VLC_STATIC( 3, 300);
613 AAC_INIT_VLC_STATIC( 4, 328);
614 AAC_INIT_VLC_STATIC( 5, 294);
615 AAC_INIT_VLC_STATIC( 6, 306);
616 AAC_INIT_VLC_STATIC( 7, 268);
617 AAC_INIT_VLC_STATIC( 8, 510);
618 AAC_INIT_VLC_STATIC( 9, 366);
619 AAC_INIT_VLC_STATIC(10, 462);
623 dsputil_init(&ac->dsp, avctx);
624 ff_fmt_convert_init(&ac->fmt_conv, avctx);
626 ac->random_state = 0x1f2e3d4c;
630 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
631 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
632 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
635 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
636 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
637 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
638 // window initialization
639 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
640 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
641 ff_init_ff_sine_windows(10);
642 ff_init_ff_sine_windows( 7);
650 * Skip data_stream_element; reference: table 4.10.
652 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
654 int byte_align = get_bits1(gb);
655 int count = get_bits(gb, 8);
657 count += get_bits(gb, 8);
661 if (get_bits_left(gb) < 8 * count) {
662 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
665 skip_bits_long(gb, 8 * count);
669 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
674 ics->predictor_reset_group = get_bits(gb, 5);
675 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
676 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
680 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
681 ics->prediction_used[sfb] = get_bits1(gb);
687 * Decode Long Term Prediction data; reference: table 4.xx.
689 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
690 GetBitContext *gb, uint8_t max_sfb)
694 ltp->lag = get_bits(gb, 11);
695 ltp->coef = ltp_coef[get_bits(gb, 3)];
696 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
697 ltp->used[sfb] = get_bits1(gb);
701 * Decode Individual Channel Stream info; reference: table 4.6.
703 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
705 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
706 GetBitContext *gb, int common_window)
709 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
710 memset(ics, 0, sizeof(IndividualChannelStream));
713 ics->window_sequence[1] = ics->window_sequence[0];
714 ics->window_sequence[0] = get_bits(gb, 2);
715 ics->use_kb_window[1] = ics->use_kb_window[0];
716 ics->use_kb_window[0] = get_bits1(gb);
717 ics->num_window_groups = 1;
718 ics->group_len[0] = 1;
719 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
721 ics->max_sfb = get_bits(gb, 4);
722 for (i = 0; i < 7; i++) {
724 ics->group_len[ics->num_window_groups - 1]++;
726 ics->num_window_groups++;
727 ics->group_len[ics->num_window_groups - 1] = 1;
730 ics->num_windows = 8;
731 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
732 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
733 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
734 ics->predictor_present = 0;
736 ics->max_sfb = get_bits(gb, 6);
737 ics->num_windows = 1;
738 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
739 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
740 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
741 ics->predictor_present = get_bits1(gb);
742 ics->predictor_reset_group = 0;
743 if (ics->predictor_present) {
744 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
745 if (decode_prediction(ac, ics, gb)) {
746 memset(ics, 0, sizeof(IndividualChannelStream));
749 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
750 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
751 memset(ics, 0, sizeof(IndividualChannelStream));
754 if ((ics->ltp.present = get_bits(gb, 1)))
755 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
760 if (ics->max_sfb > ics->num_swb) {
761 av_log(ac->avctx, AV_LOG_ERROR,
762 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
763 ics->max_sfb, ics->num_swb);
764 memset(ics, 0, sizeof(IndividualChannelStream));
772 * Decode band types (section_data payload); reference: table 4.46.
774 * @param band_type array of the used band type
775 * @param band_type_run_end array of the last scalefactor band of a band type run
777 * @return Returns error status. 0 - OK, !0 - error
779 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
780 int band_type_run_end[120], GetBitContext *gb,
781 IndividualChannelStream *ics)
784 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
785 for (g = 0; g < ics->num_window_groups; g++) {
787 while (k < ics->max_sfb) {
788 uint8_t sect_end = k;
790 int sect_band_type = get_bits(gb, 4);
791 if (sect_band_type == 12) {
792 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
795 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
796 sect_end += sect_len_incr;
797 sect_end += sect_len_incr;
798 if (get_bits_left(gb) < 0) {
799 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
802 if (sect_end > ics->max_sfb) {
803 av_log(ac->avctx, AV_LOG_ERROR,
804 "Number of bands (%d) exceeds limit (%d).\n",
805 sect_end, ics->max_sfb);
808 for (; k < sect_end; k++) {
809 band_type [idx] = sect_band_type;
810 band_type_run_end[idx++] = sect_end;
818 * Decode scalefactors; reference: table 4.47.
820 * @param global_gain first scalefactor value as scalefactors are differentially coded
821 * @param band_type array of the used band type
822 * @param band_type_run_end array of the last scalefactor band of a band type run
823 * @param sf array of scalefactors or intensity stereo positions
825 * @return Returns error status. 0 - OK, !0 - error
827 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
828 unsigned int global_gain,
829 IndividualChannelStream *ics,
830 enum BandType band_type[120],
831 int band_type_run_end[120])
834 int offset[3] = { global_gain, global_gain - 90, 0 };
837 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
838 for (g = 0; g < ics->num_window_groups; g++) {
839 for (i = 0; i < ics->max_sfb;) {
840 int run_end = band_type_run_end[idx];
841 if (band_type[idx] == ZERO_BT) {
842 for (; i < run_end; i++, idx++)
844 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
845 for (; i < run_end; i++, idx++) {
846 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
847 clipped_offset = av_clip(offset[2], -155, 100);
848 if (offset[2] != clipped_offset) {
849 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
850 "position clipped (%d -> %d).\nIf you heard an "
851 "audible artifact, there may be a bug in the "
852 "decoder. ", offset[2], clipped_offset);
854 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
856 } else if (band_type[idx] == NOISE_BT) {
857 for (; i < run_end; i++, idx++) {
858 if (noise_flag-- > 0)
859 offset[1] += get_bits(gb, 9) - 256;
861 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
862 clipped_offset = av_clip(offset[1], -100, 155);
863 if (offset[1] != clipped_offset) {
864 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
865 "(%d -> %d).\nIf you heard an audible "
866 "artifact, there may be a bug in the decoder. ",
867 offset[1], clipped_offset);
869 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
872 for (; i < run_end; i++, idx++) {
873 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
874 if (offset[0] > 255U) {
875 av_log(ac->avctx, AV_LOG_ERROR,
876 "%s (%d) out of range.\n", sf_str[0], offset[0]);
879 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
888 * Decode pulse data; reference: table 4.7.
890 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
891 const uint16_t *swb_offset, int num_swb)
894 pulse->num_pulse = get_bits(gb, 2) + 1;
895 pulse_swb = get_bits(gb, 6);
896 if (pulse_swb >= num_swb)
898 pulse->pos[0] = swb_offset[pulse_swb];
899 pulse->pos[0] += get_bits(gb, 5);
900 if (pulse->pos[0] > 1023)
902 pulse->amp[0] = get_bits(gb, 4);
903 for (i = 1; i < pulse->num_pulse; i++) {
904 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
905 if (pulse->pos[i] > 1023)
907 pulse->amp[i] = get_bits(gb, 4);
913 * Decode Temporal Noise Shaping data; reference: table 4.48.
915 * @return Returns error status. 0 - OK, !0 - error
917 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
918 GetBitContext *gb, const IndividualChannelStream *ics)
920 int w, filt, i, coef_len, coef_res, coef_compress;
921 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
922 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
923 for (w = 0; w < ics->num_windows; w++) {
924 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
925 coef_res = get_bits1(gb);
927 for (filt = 0; filt < tns->n_filt[w]; filt++) {
929 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
931 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
932 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
933 tns->order[w][filt], tns_max_order);
934 tns->order[w][filt] = 0;
937 if (tns->order[w][filt]) {
938 tns->direction[w][filt] = get_bits1(gb);
939 coef_compress = get_bits1(gb);
940 coef_len = coef_res + 3 - coef_compress;
941 tmp2_idx = 2 * coef_compress + coef_res;
943 for (i = 0; i < tns->order[w][filt]; i++)
944 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
953 * Decode Mid/Side data; reference: table 4.54.
955 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
956 * [1] mask is decoded from bitstream; [2] mask is all 1s;
957 * [3] reserved for scalable AAC
959 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
963 if (ms_present == 1) {
964 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
965 cpe->ms_mask[idx] = get_bits1(gb);
966 } else if (ms_present == 2) {
967 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
972 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
976 *dst++ = v[idx & 15] * s;
977 *dst++ = v[idx>>4 & 15] * s;
983 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
987 *dst++ = v[idx & 3] * s;
988 *dst++ = v[idx>>2 & 3] * s;
989 *dst++ = v[idx>>4 & 3] * s;
990 *dst++ = v[idx>>6 & 3] * s;
996 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
997 unsigned sign, const float *scale)
999 union float754 s0, s1;
1001 s0.f = s1.f = *scale;
1002 s0.i ^= sign >> 1 << 31;
1005 *dst++ = v[idx & 15] * s0.f;
1006 *dst++ = v[idx>>4 & 15] * s1.f;
1013 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1014 unsigned sign, const float *scale)
1016 unsigned nz = idx >> 12;
1017 union float754 s = { .f = *scale };
1020 t.i = s.i ^ (sign & 1U<<31);
1021 *dst++ = v[idx & 3] * t.f;
1023 sign <<= nz & 1; nz >>= 1;
1024 t.i = s.i ^ (sign & 1U<<31);
1025 *dst++ = v[idx>>2 & 3] * t.f;
1027 sign <<= nz & 1; nz >>= 1;
1028 t.i = s.i ^ (sign & 1U<<31);
1029 *dst++ = v[idx>>4 & 3] * t.f;
1031 sign <<= nz & 1; nz >>= 1;
1032 t.i = s.i ^ (sign & 1U<<31);
1033 *dst++ = v[idx>>6 & 3] * t.f;
1040 * Decode spectral data; reference: table 4.50.
1041 * Dequantize and scale spectral data; reference: 4.6.3.3.
1043 * @param coef array of dequantized, scaled spectral data
1044 * @param sf array of scalefactors or intensity stereo positions
1045 * @param pulse_present set if pulses are present
1046 * @param pulse pointer to pulse data struct
1047 * @param band_type array of the used band type
1049 * @return Returns error status. 0 - OK, !0 - error
1051 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1052 GetBitContext *gb, const float sf[120],
1053 int pulse_present, const Pulse *pulse,
1054 const IndividualChannelStream *ics,
1055 enum BandType band_type[120])
1057 int i, k, g, idx = 0;
1058 const int c = 1024 / ics->num_windows;
1059 const uint16_t *offsets = ics->swb_offset;
1060 float *coef_base = coef;
1062 for (g = 0; g < ics->num_windows; g++)
1063 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1065 for (g = 0; g < ics->num_window_groups; g++) {
1066 unsigned g_len = ics->group_len[g];
1068 for (i = 0; i < ics->max_sfb; i++, idx++) {
1069 const unsigned cbt_m1 = band_type[idx] - 1;
1070 float *cfo = coef + offsets[i];
1071 int off_len = offsets[i + 1] - offsets[i];
1074 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1075 for (group = 0; group < g_len; group++, cfo+=128) {
1076 memset(cfo, 0, off_len * sizeof(float));
1078 } else if (cbt_m1 == NOISE_BT - 1) {
1079 for (group = 0; group < g_len; group++, cfo+=128) {
1083 for (k = 0; k < off_len; k++) {
1084 ac->random_state = lcg_random(ac->random_state);
1085 cfo[k] = ac->random_state;
1088 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1089 scale = sf[idx] / sqrtf(band_energy);
1090 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1093 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1094 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1095 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1096 OPEN_READER(re, gb);
1098 switch (cbt_m1 >> 1) {
1100 for (group = 0; group < g_len; group++, cfo+=128) {
1108 UPDATE_CACHE(re, gb);
1109 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1110 cb_idx = cb_vector_idx[code];
1111 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1117 for (group = 0; group < g_len; group++, cfo+=128) {
1127 UPDATE_CACHE(re, gb);
1128 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1129 cb_idx = cb_vector_idx[code];
1130 nnz = cb_idx >> 8 & 15;
1131 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1132 LAST_SKIP_BITS(re, gb, nnz);
1133 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1139 for (group = 0; group < g_len; group++, cfo+=128) {
1147 UPDATE_CACHE(re, gb);
1148 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1149 cb_idx = cb_vector_idx[code];
1150 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1157 for (group = 0; group < g_len; group++, cfo+=128) {
1167 UPDATE_CACHE(re, gb);
1168 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1169 cb_idx = cb_vector_idx[code];
1170 nnz = cb_idx >> 8 & 15;
1171 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1172 LAST_SKIP_BITS(re, gb, nnz);
1173 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1179 for (group = 0; group < g_len; group++, cfo+=128) {
1181 uint32_t *icf = (uint32_t *) cf;
1191 UPDATE_CACHE(re, gb);
1192 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1200 cb_idx = cb_vector_idx[code];
1203 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1204 LAST_SKIP_BITS(re, gb, nnz);
1206 for (j = 0; j < 2; j++) {
1210 /* The total length of escape_sequence must be < 22 bits according
1211 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1212 UPDATE_CACHE(re, gb);
1213 b = GET_CACHE(re, gb);
1214 b = 31 - av_log2(~b);
1217 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1221 SKIP_BITS(re, gb, b + 1);
1223 n = (1 << b) + SHOW_UBITS(re, gb, b);
1224 LAST_SKIP_BITS(re, gb, b);
1225 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1228 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1229 *icf++ = (bits & 1U<<31) | v;
1236 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1240 CLOSE_READER(re, gb);
1246 if (pulse_present) {
1248 for (i = 0; i < pulse->num_pulse; i++) {
1249 float co = coef_base[ pulse->pos[i] ];
1250 while (offsets[idx + 1] <= pulse->pos[i])
1252 if (band_type[idx] != NOISE_BT && sf[idx]) {
1253 float ico = -pulse->amp[i];
1256 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1258 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1265 static av_always_inline float flt16_round(float pf)
1269 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1273 static av_always_inline float flt16_even(float pf)
1277 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1281 static av_always_inline float flt16_trunc(float pf)
1285 pun.i &= 0xFFFF0000U;
1289 static av_always_inline void predict(PredictorState *ps, float *coef,
1292 const float a = 0.953125; // 61.0 / 64
1293 const float alpha = 0.90625; // 29.0 / 32
1297 float r0 = ps->r0, r1 = ps->r1;
1298 float cor0 = ps->cor0, cor1 = ps->cor1;
1299 float var0 = ps->var0, var1 = ps->var1;
1301 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1302 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1304 pv = flt16_round(k1 * r0 + k2 * r1);
1311 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1312 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1313 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1314 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1316 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1317 ps->r0 = flt16_trunc(a * e0);
1321 * Apply AAC-Main style frequency domain prediction.
1323 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1327 if (!sce->ics.predictor_initialized) {
1328 reset_all_predictors(sce->predictor_state);
1329 sce->ics.predictor_initialized = 1;
1332 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1333 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1334 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1335 predict(&sce->predictor_state[k], &sce->coeffs[k],
1336 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1339 if (sce->ics.predictor_reset_group)
1340 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1342 reset_all_predictors(sce->predictor_state);
1346 * Decode an individual_channel_stream payload; reference: table 4.44.
1348 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1349 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1351 * @return Returns error status. 0 - OK, !0 - error
1353 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1354 GetBitContext *gb, int common_window, int scale_flag)
1357 TemporalNoiseShaping *tns = &sce->tns;
1358 IndividualChannelStream *ics = &sce->ics;
1359 float *out = sce->coeffs;
1360 int global_gain, pulse_present = 0;
1362 /* This assignment is to silence a GCC warning about the variable being used
1363 * uninitialized when in fact it always is.
1365 pulse.num_pulse = 0;
1367 global_gain = get_bits(gb, 8);
1369 if (!common_window && !scale_flag) {
1370 if (decode_ics_info(ac, ics, gb, 0) < 0)
1374 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1376 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1381 if ((pulse_present = get_bits1(gb))) {
1382 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1383 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1386 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1387 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1391 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1393 if (get_bits1(gb)) {
1394 av_log_missing_feature(ac->avctx, "SSR", 1);
1399 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1402 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1403 apply_prediction(ac, sce);
1409 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1411 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1413 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1414 float *ch0 = cpe->ch[0].coeffs;
1415 float *ch1 = cpe->ch[1].coeffs;
1416 int g, i, group, idx = 0;
1417 const uint16_t *offsets = ics->swb_offset;
1418 for (g = 0; g < ics->num_window_groups; g++) {
1419 for (i = 0; i < ics->max_sfb; i++, idx++) {
1420 if (cpe->ms_mask[idx] &&
1421 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1422 for (group = 0; group < ics->group_len[g]; group++) {
1423 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1424 ch1 + group * 128 + offsets[i],
1425 offsets[i+1] - offsets[i]);
1429 ch0 += ics->group_len[g] * 128;
1430 ch1 += ics->group_len[g] * 128;
1435 * intensity stereo decoding; reference: 4.6.8.2.3
1437 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1438 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1439 * [3] reserved for scalable AAC
1441 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1443 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1444 SingleChannelElement *sce1 = &cpe->ch[1];
1445 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1446 const uint16_t *offsets = ics->swb_offset;
1447 int g, group, i, idx = 0;
1450 for (g = 0; g < ics->num_window_groups; g++) {
1451 for (i = 0; i < ics->max_sfb;) {
1452 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1453 const int bt_run_end = sce1->band_type_run_end[idx];
1454 for (; i < bt_run_end; i++, idx++) {
1455 c = -1 + 2 * (sce1->band_type[idx] - 14);
1457 c *= 1 - 2 * cpe->ms_mask[idx];
1458 scale = c * sce1->sf[idx];
1459 for (group = 0; group < ics->group_len[g]; group++)
1460 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1461 coef0 + group * 128 + offsets[i],
1463 offsets[i + 1] - offsets[i]);
1466 int bt_run_end = sce1->band_type_run_end[idx];
1467 idx += bt_run_end - i;
1471 coef0 += ics->group_len[g] * 128;
1472 coef1 += ics->group_len[g] * 128;
1477 * Decode a channel_pair_element; reference: table 4.4.
1479 * @return Returns error status. 0 - OK, !0 - error
1481 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1483 int i, ret, common_window, ms_present = 0;
1485 common_window = get_bits1(gb);
1486 if (common_window) {
1487 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1489 i = cpe->ch[1].ics.use_kb_window[0];
1490 cpe->ch[1].ics = cpe->ch[0].ics;
1491 cpe->ch[1].ics.use_kb_window[1] = i;
1492 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1493 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1494 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1495 ms_present = get_bits(gb, 2);
1496 if (ms_present == 3) {
1497 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1499 } else if (ms_present)
1500 decode_mid_side_stereo(cpe, gb, ms_present);
1502 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1504 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1507 if (common_window) {
1509 apply_mid_side_stereo(ac, cpe);
1510 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1511 apply_prediction(ac, &cpe->ch[0]);
1512 apply_prediction(ac, &cpe->ch[1]);
1516 apply_intensity_stereo(ac, cpe, ms_present);
1520 static const float cce_scale[] = {
1521 1.09050773266525765921, //2^(1/8)
1522 1.18920711500272106672, //2^(1/4)
1528 * Decode coupling_channel_element; reference: table 4.8.
1530 * @return Returns error status. 0 - OK, !0 - error
1532 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1538 SingleChannelElement *sce = &che->ch[0];
1539 ChannelCoupling *coup = &che->coup;
1541 coup->coupling_point = 2 * get_bits1(gb);
1542 coup->num_coupled = get_bits(gb, 3);
1543 for (c = 0; c <= coup->num_coupled; c++) {
1545 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1546 coup->id_select[c] = get_bits(gb, 4);
1547 if (coup->type[c] == TYPE_CPE) {
1548 coup->ch_select[c] = get_bits(gb, 2);
1549 if (coup->ch_select[c] == 3)
1552 coup->ch_select[c] = 2;
1554 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1556 sign = get_bits(gb, 1);
1557 scale = cce_scale[get_bits(gb, 2)];
1559 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1562 for (c = 0; c < num_gain; c++) {
1566 float gain_cache = 1.;
1568 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1569 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1570 gain_cache = powf(scale, -gain);
1572 if (coup->coupling_point == AFTER_IMDCT) {
1573 coup->gain[c][0] = gain_cache;
1575 for (g = 0; g < sce->ics.num_window_groups; g++) {
1576 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1577 if (sce->band_type[idx] != ZERO_BT) {
1579 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1587 gain_cache = powf(scale, -t) * s;
1590 coup->gain[c][idx] = gain_cache;
1600 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1602 * @return Returns number of bytes consumed.
1604 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1608 int num_excl_chan = 0;
1611 for (i = 0; i < 7; i++)
1612 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1613 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1615 return num_excl_chan / 7;
1619 * Decode dynamic range information; reference: table 4.52.
1621 * @param cnt length of TYPE_FIL syntactic element in bytes
1623 * @return Returns number of bytes consumed.
1625 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1626 GetBitContext *gb, int cnt)
1629 int drc_num_bands = 1;
1632 /* pce_tag_present? */
1633 if (get_bits1(gb)) {
1634 che_drc->pce_instance_tag = get_bits(gb, 4);
1635 skip_bits(gb, 4); // tag_reserved_bits
1639 /* excluded_chns_present? */
1640 if (get_bits1(gb)) {
1641 n += decode_drc_channel_exclusions(che_drc, gb);
1644 /* drc_bands_present? */
1645 if (get_bits1(gb)) {
1646 che_drc->band_incr = get_bits(gb, 4);
1647 che_drc->interpolation_scheme = get_bits(gb, 4);
1649 drc_num_bands += che_drc->band_incr;
1650 for (i = 0; i < drc_num_bands; i++) {
1651 che_drc->band_top[i] = get_bits(gb, 8);
1656 /* prog_ref_level_present? */
1657 if (get_bits1(gb)) {
1658 che_drc->prog_ref_level = get_bits(gb, 7);
1659 skip_bits1(gb); // prog_ref_level_reserved_bits
1663 for (i = 0; i < drc_num_bands; i++) {
1664 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1665 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1673 * Decode extension data (incomplete); reference: table 4.51.
1675 * @param cnt length of TYPE_FIL syntactic element in bytes
1677 * @return Returns number of bytes consumed
1679 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1680 ChannelElement *che, enum RawDataBlockType elem_type)
1684 switch (get_bits(gb, 4)) { // extension type
1685 case EXT_SBR_DATA_CRC:
1689 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1691 } else if (!ac->m4ac.sbr) {
1692 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1693 skip_bits_long(gb, 8 * cnt - 4);
1695 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1696 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1697 skip_bits_long(gb, 8 * cnt - 4);
1699 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1702 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1706 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1708 case EXT_DYNAMIC_RANGE:
1709 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1713 case EXT_DATA_ELEMENT:
1715 skip_bits_long(gb, 8 * cnt - 4);
1722 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1724 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1725 * @param coef spectral coefficients
1727 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1728 IndividualChannelStream *ics, int decode)
1730 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1732 int bottom, top, order, start, end, size, inc;
1733 float lpc[TNS_MAX_ORDER];
1734 float tmp[TNS_MAX_ORDER];
1736 for (w = 0; w < ics->num_windows; w++) {
1737 bottom = ics->num_swb;
1738 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1740 bottom = FFMAX(0, top - tns->length[w][filt]);
1741 order = tns->order[w][filt];
1746 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1748 start = ics->swb_offset[FFMIN(bottom, mmm)];
1749 end = ics->swb_offset[FFMIN( top, mmm)];
1750 if ((size = end - start) <= 0)
1752 if (tns->direction[w][filt]) {
1762 for (m = 0; m < size; m++, start += inc)
1763 for (i = 1; i <= FFMIN(m, order); i++)
1764 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1767 for (m = 0; m < size; m++, start += inc) {
1768 tmp[0] = coef[start];
1769 for (i = 1; i <= FFMIN(m, order); i++)
1770 coef[start] += tmp[i] * lpc[i - 1];
1771 for (i = order; i > 0; i--)
1772 tmp[i] = tmp[i - 1];
1780 * Apply windowing and MDCT to obtain the spectral
1781 * coefficient from the predicted sample by LTP.
1783 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1784 float *in, IndividualChannelStream *ics)
1786 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1787 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1788 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1789 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1791 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1792 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1794 memset(in, 0, 448 * sizeof(float));
1795 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1797 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1798 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1800 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1801 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1803 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1807 * Apply the long term prediction
1809 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1811 const LongTermPrediction *ltp = &sce->ics.ltp;
1812 const uint16_t *offsets = sce->ics.swb_offset;
1815 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1816 float *predTime = sce->ret;
1817 float *predFreq = ac->buf_mdct;
1818 int16_t num_samples = 2048;
1820 if (ltp->lag < 1024)
1821 num_samples = ltp->lag + 1024;
1822 for (i = 0; i < num_samples; i++)
1823 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1824 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1826 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1828 if (sce->tns.present)
1829 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1831 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1833 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1834 sce->coeffs[i] += predFreq[i];
1839 * Update the LTP buffer for next frame
1841 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1843 IndividualChannelStream *ics = &sce->ics;
1844 float *saved = sce->saved;
1845 float *saved_ltp = sce->coeffs;
1846 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1847 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1850 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1851 memcpy(saved_ltp, saved, 512 * sizeof(float));
1852 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1853 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1854 for (i = 0; i < 64; i++)
1855 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1856 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1857 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1858 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1859 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1860 for (i = 0; i < 64; i++)
1861 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1862 } else { // LONG_STOP or ONLY_LONG
1863 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1864 for (i = 0; i < 512; i++)
1865 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1868 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1869 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1870 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1874 * Conduct IMDCT and windowing.
1876 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1878 IndividualChannelStream *ics = &sce->ics;
1879 float *in = sce->coeffs;
1880 float *out = sce->ret;
1881 float *saved = sce->saved;
1882 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1883 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1884 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1885 float *buf = ac->buf_mdct;
1886 float *temp = ac->temp;
1890 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1891 for (i = 0; i < 1024; i += 128)
1892 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1894 ac->mdct.imdct_half(&ac->mdct, buf, in);
1896 /* window overlapping
1897 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1898 * and long to short transitions are considered to be short to short
1899 * transitions. This leaves just two cases (long to long and short to short)
1900 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1902 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1903 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1904 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1906 memcpy( out, saved, 448 * sizeof(float));
1908 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1909 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1910 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1911 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1912 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1913 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1914 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1916 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1917 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1922 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1923 memcpy( saved, temp + 64, 64 * sizeof(float));
1924 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1925 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1926 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1927 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1928 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1929 memcpy( saved, buf + 512, 448 * sizeof(float));
1930 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1931 } else { // LONG_STOP or ONLY_LONG
1932 memcpy( saved, buf + 512, 512 * sizeof(float));
1937 * Apply dependent channel coupling (applied before IMDCT).
1939 * @param index index into coupling gain array
1941 static void apply_dependent_coupling(AACContext *ac,
1942 SingleChannelElement *target,
1943 ChannelElement *cce, int index)
1945 IndividualChannelStream *ics = &cce->ch[0].ics;
1946 const uint16_t *offsets = ics->swb_offset;
1947 float *dest = target->coeffs;
1948 const float *src = cce->ch[0].coeffs;
1949 int g, i, group, k, idx = 0;
1950 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1951 av_log(ac->avctx, AV_LOG_ERROR,
1952 "Dependent coupling is not supported together with LTP\n");
1955 for (g = 0; g < ics->num_window_groups; g++) {
1956 for (i = 0; i < ics->max_sfb; i++, idx++) {
1957 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1958 const float gain = cce->coup.gain[index][idx];
1959 for (group = 0; group < ics->group_len[g]; group++) {
1960 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1962 dest[group * 128 + k] += gain * src[group * 128 + k];
1967 dest += ics->group_len[g] * 128;
1968 src += ics->group_len[g] * 128;
1973 * Apply independent channel coupling (applied after IMDCT).
1975 * @param index index into coupling gain array
1977 static void apply_independent_coupling(AACContext *ac,
1978 SingleChannelElement *target,
1979 ChannelElement *cce, int index)
1982 const float gain = cce->coup.gain[index][0];
1983 const float *src = cce->ch[0].ret;
1984 float *dest = target->ret;
1985 const int len = 1024 << (ac->m4ac.sbr == 1);
1987 for (i = 0; i < len; i++)
1988 dest[i] += gain * src[i];
1992 * channel coupling transformation interface
1994 * @param apply_coupling_method pointer to (in)dependent coupling function
1996 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1997 enum RawDataBlockType type, int elem_id,
1998 enum CouplingPoint coupling_point,
1999 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2003 for (i = 0; i < MAX_ELEM_ID; i++) {
2004 ChannelElement *cce = ac->che[TYPE_CCE][i];
2007 if (cce && cce->coup.coupling_point == coupling_point) {
2008 ChannelCoupling *coup = &cce->coup;
2010 for (c = 0; c <= coup->num_coupled; c++) {
2011 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2012 if (coup->ch_select[c] != 1) {
2013 apply_coupling_method(ac, &cc->ch[0], cce, index);
2014 if (coup->ch_select[c] != 0)
2017 if (coup->ch_select[c] != 2)
2018 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2020 index += 1 + (coup->ch_select[c] == 3);
2027 * Convert spectral data to float samples, applying all supported tools as appropriate.
2029 static void spectral_to_sample(AACContext *ac)
2032 for (type = 3; type >= 0; type--) {
2033 for (i = 0; i < MAX_ELEM_ID; i++) {
2034 ChannelElement *che = ac->che[type][i];
2036 if (type <= TYPE_CPE)
2037 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2038 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2039 if (che->ch[0].ics.predictor_present) {
2040 if (che->ch[0].ics.ltp.present)
2041 apply_ltp(ac, &che->ch[0]);
2042 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2043 apply_ltp(ac, &che->ch[1]);
2046 if (che->ch[0].tns.present)
2047 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2048 if (che->ch[1].tns.present)
2049 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2050 if (type <= TYPE_CPE)
2051 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2052 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2053 imdct_and_windowing(ac, &che->ch[0]);
2054 if (ac->m4ac.object_type == AOT_AAC_LTP)
2055 update_ltp(ac, &che->ch[0]);
2056 if (type == TYPE_CPE) {
2057 imdct_and_windowing(ac, &che->ch[1]);
2058 if (ac->m4ac.object_type == AOT_AAC_LTP)
2059 update_ltp(ac, &che->ch[1]);
2061 if (ac->m4ac.sbr > 0) {
2062 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2065 if (type <= TYPE_CCE)
2066 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2072 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2075 AACADTSHeaderInfo hdr_info;
2077 size = ff_aac_parse_header(gb, &hdr_info);
2079 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2080 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2081 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2082 ac->m4ac.chan_config = hdr_info.chan_config;
2083 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2085 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2087 } else if (ac->output_configured != OC_LOCKED) {
2088 ac->m4ac.chan_config = 0;
2089 ac->output_configured = OC_NONE;
2091 if (ac->output_configured != OC_LOCKED) {
2095 ac->m4ac.sample_rate = hdr_info.sample_rate;
2096 ac->m4ac.sampling_index = hdr_info.sampling_index;
2097 ac->m4ac.object_type = hdr_info.object_type;
2098 if (!ac->avctx->sample_rate)
2099 ac->avctx->sample_rate = hdr_info.sample_rate;
2100 if (hdr_info.num_aac_frames == 1) {
2101 if (!hdr_info.crc_absent)
2104 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2111 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2112 int *data_size, GetBitContext *gb)
2114 AACContext *ac = avctx->priv_data;
2115 ChannelElement *che = NULL, *che_prev = NULL;
2116 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2117 int err, elem_id, data_size_tmp;
2118 int samples = 0, multiplier;
2120 if (show_bits(gb, 12) == 0xfff) {
2121 if (parse_adts_frame_header(ac, gb) < 0) {
2122 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2125 if (ac->m4ac.sampling_index > 12) {
2126 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2131 ac->tags_mapped = 0;
2133 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2134 elem_id = get_bits(gb, 4);
2136 if (elem_type < TYPE_DSE) {
2137 if (!(che=get_che(ac, elem_type, elem_id))) {
2138 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2139 elem_type, elem_id);
2145 switch (elem_type) {
2148 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2152 err = decode_cpe(ac, gb, che);
2156 err = decode_cce(ac, gb, che);
2160 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2164 err = skip_data_stream_element(ac, gb);
2168 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2169 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2170 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2172 if (ac->output_configured > OC_TRIAL_PCE)
2173 av_log(avctx, AV_LOG_ERROR,
2174 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2176 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2182 elem_id += get_bits(gb, 8) - 1;
2183 if (get_bits_left(gb) < 8 * elem_id) {
2184 av_log(avctx, AV_LOG_ERROR, overread_err);
2188 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2189 err = 0; /* FIXME */
2193 err = -1; /* should not happen, but keeps compiler happy */
2198 elem_type_prev = elem_type;
2203 if (get_bits_left(gb) < 3) {
2204 av_log(avctx, AV_LOG_ERROR, overread_err);
2209 spectral_to_sample(ac);
2211 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2212 samples <<= multiplier;
2213 if (ac->output_configured < OC_LOCKED) {
2214 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2215 avctx->frame_size = samples;
2218 data_size_tmp = samples * avctx->channels *
2219 av_get_bytes_per_sample(avctx->sample_fmt);
2220 if (*data_size < data_size_tmp) {
2221 av_log(avctx, AV_LOG_ERROR,
2222 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2223 *data_size, data_size_tmp);
2226 *data_size = data_size_tmp;
2229 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2230 ac->fmt_conv.float_interleave(data, (const float **)ac->output_data,
2231 samples, avctx->channels);
2233 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data,
2234 samples, avctx->channels);
2237 if (ac->output_configured)
2238 ac->output_configured = OC_LOCKED;
2243 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2244 int *data_size, AVPacket *avpkt)
2246 const uint8_t *buf = avpkt->data;
2247 int buf_size = avpkt->size;
2253 init_get_bits(&gb, buf, buf_size * 8);
2255 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2258 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2259 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2260 if (buf[buf_offset])
2263 return buf_size > buf_offset ? buf_consumed : buf_size;
2266 static av_cold int aac_decode_close(AVCodecContext *avctx)
2268 AACContext *ac = avctx->priv_data;
2271 for (i = 0; i < MAX_ELEM_ID; i++) {
2272 for (type = 0; type < 4; type++) {
2273 if (ac->che[type][i])
2274 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2275 av_freep(&ac->che[type][i]);
2279 ff_mdct_end(&ac->mdct);
2280 ff_mdct_end(&ac->mdct_small);
2281 ff_mdct_end(&ac->mdct_ltp);
2286 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2288 struct LATMContext {
2289 AACContext aac_ctx; ///< containing AACContext
2290 int initialized; ///< initilized after a valid extradata was seen
2293 int audio_mux_version_A; ///< LATM syntax version
2294 int frame_length_type; ///< 0/1 variable/fixed frame length
2295 int frame_length; ///< frame length for fixed frame length
2298 static inline uint32_t latm_get_value(GetBitContext *b)
2300 int length = get_bits(b, 2);
2302 return get_bits_long(b, (length+1)*8);
2305 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2308 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2309 MPEG4AudioConfig m4ac;
2310 int config_start_bit = get_bits_count(gb);
2311 int bits_consumed, esize;
2313 if (config_start_bit % 8) {
2314 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2315 "config not byte aligned.\n", 1);
2316 return AVERROR_INVALIDDATA;
2319 decode_audio_specific_config(NULL, avctx, &m4ac,
2320 gb->buffer + (config_start_bit / 8),
2321 get_bits_left(gb) / 8);
2323 if (bits_consumed < 0)
2324 return AVERROR_INVALIDDATA;
2326 esize = (bits_consumed+7) / 8;
2328 if (avctx->extradata_size <= esize) {
2329 av_free(avctx->extradata);
2330 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2331 if (!avctx->extradata)
2332 return AVERROR(ENOMEM);
2335 avctx->extradata_size = esize;
2336 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2337 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2339 skip_bits_long(gb, bits_consumed);
2342 return bits_consumed;
2345 static int read_stream_mux_config(struct LATMContext *latmctx,
2348 int ret, audio_mux_version = get_bits(gb, 1);
2350 latmctx->audio_mux_version_A = 0;
2351 if (audio_mux_version)
2352 latmctx->audio_mux_version_A = get_bits(gb, 1);
2354 if (!latmctx->audio_mux_version_A) {
2356 if (audio_mux_version)
2357 latm_get_value(gb); // taraFullness
2359 skip_bits(gb, 1); // allStreamSameTimeFraming
2360 skip_bits(gb, 6); // numSubFrames
2362 if (get_bits(gb, 4)) { // numPrograms
2363 av_log_missing_feature(latmctx->aac_ctx.avctx,
2364 "multiple programs are not supported\n", 1);
2365 return AVERROR_PATCHWELCOME;
2368 // for each program (which there is only on in DVB)
2370 // for each layer (which there is only on in DVB)
2371 if (get_bits(gb, 3)) { // numLayer
2372 av_log_missing_feature(latmctx->aac_ctx.avctx,
2373 "multiple layers are not supported\n", 1);
2374 return AVERROR_PATCHWELCOME;
2377 // for all but first stream: use_same_config = get_bits(gb, 1);
2378 if (!audio_mux_version) {
2379 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2382 int ascLen = latm_get_value(gb);
2383 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2386 skip_bits_long(gb, ascLen);
2389 latmctx->frame_length_type = get_bits(gb, 3);
2390 switch (latmctx->frame_length_type) {
2392 skip_bits(gb, 8); // latmBufferFullness
2395 latmctx->frame_length = get_bits(gb, 9);
2400 skip_bits(gb, 6); // CELP frame length table index
2404 skip_bits(gb, 1); // HVXC frame length table index
2408 if (get_bits(gb, 1)) { // other data
2409 if (audio_mux_version) {
2410 latm_get_value(gb); // other_data_bits
2414 esc = get_bits(gb, 1);
2420 if (get_bits(gb, 1)) // crc present
2421 skip_bits(gb, 8); // config_crc
2427 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2431 if (ctx->frame_length_type == 0) {
2432 int mux_slot_length = 0;
2434 tmp = get_bits(gb, 8);
2435 mux_slot_length += tmp;
2436 } while (tmp == 255);
2437 return mux_slot_length;
2438 } else if (ctx->frame_length_type == 1) {
2439 return ctx->frame_length;
2440 } else if (ctx->frame_length_type == 3 ||
2441 ctx->frame_length_type == 5 ||
2442 ctx->frame_length_type == 7) {
2443 skip_bits(gb, 2); // mux_slot_length_coded
2448 static int read_audio_mux_element(struct LATMContext *latmctx,
2452 uint8_t use_same_mux = get_bits(gb, 1);
2453 if (!use_same_mux) {
2454 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2456 } else if (!latmctx->aac_ctx.avctx->extradata) {
2457 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2458 "no decoder config found\n");
2459 return AVERROR(EAGAIN);
2461 if (latmctx->audio_mux_version_A == 0) {
2462 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2463 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2464 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2465 return AVERROR_INVALIDDATA;
2466 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2467 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2468 "frame length mismatch %d << %d\n",
2469 mux_slot_length_bytes * 8, get_bits_left(gb));
2470 return AVERROR_INVALIDDATA;
2477 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2480 struct LATMContext *latmctx = avctx->priv_data;
2484 if (avpkt->size == 0)
2487 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2489 // check for LOAS sync word
2490 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2491 return AVERROR_INVALIDDATA;
2493 muxlength = get_bits(&gb, 13) + 3;
2494 // not enough data, the parser should have sorted this
2495 if (muxlength > avpkt->size)
2496 return AVERROR_INVALIDDATA;
2498 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2501 if (!latmctx->initialized) {
2502 if (!avctx->extradata) {
2506 aac_decode_close(avctx);
2507 if ((err = aac_decode_init(avctx)) < 0)
2509 latmctx->initialized = 1;
2513 if (show_bits(&gb, 12) == 0xfff) {
2514 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2515 "ADTS header detected, probably as result of configuration "
2517 return AVERROR_INVALIDDATA;
2520 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2526 av_cold static int latm_decode_init(AVCodecContext *avctx)
2528 struct LATMContext *latmctx = avctx->priv_data;
2531 ret = aac_decode_init(avctx);
2533 if (avctx->extradata_size > 0) {
2534 latmctx->initialized = !ret;
2536 latmctx->initialized = 0;
2543 AVCodec ff_aac_decoder = {
2545 .type = AVMEDIA_TYPE_AUDIO,
2547 .priv_data_size = sizeof(AACContext),
2548 .init = aac_decode_init,
2549 .close = aac_decode_close,
2550 .decode = aac_decode_frame,
2551 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2552 .sample_fmts = (const enum AVSampleFormat[]) {
2553 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2555 .capabilities = CODEC_CAP_CHANNEL_CONF,
2556 .channel_layouts = aac_channel_layout,
2560 Note: This decoder filter is intended to decode LATM streams transferred
2561 in MPEG transport streams which only contain one program.
2562 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2564 AVCodec ff_aac_latm_decoder = {
2566 .type = AVMEDIA_TYPE_AUDIO,
2567 .id = CODEC_ID_AAC_LATM,
2568 .priv_data_size = sizeof(struct LATMContext),
2569 .init = latm_decode_init,
2570 .close = aac_decode_close,
2571 .decode = latm_decode_frame,
2572 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2573 .sample_fmts = (const enum AVSampleFormat[]) {
2574 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2576 .capabilities = CODEC_CAP_CHANNEL_CONF,
2577 .channel_layouts = aac_channel_layout,