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->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 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
319 av_log(avctx, AV_LOG_ERROR, overread_err);
322 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
323 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
324 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
325 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
327 skip_bits_long(gb, 4 * num_assoc_data);
329 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
333 /* comment field, first byte is length */
334 comment_len = get_bits(gb, 8) * 8;
335 if (get_bits_left(gb) < comment_len) {
336 av_log(avctx, AV_LOG_ERROR, overread_err);
339 skip_bits_long(gb, comment_len);
344 * Set up channel positions based on a default channel configuration
345 * as specified in table 1.17.
347 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
349 * @return Returns error status. 0 - OK, !0 - error
351 static av_cold int set_default_channel_config(AVCodecContext *avctx,
352 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
355 if (channel_config < 1 || channel_config > 7) {
356 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
361 /* default channel configurations:
363 * 1ch : front center (mono)
364 * 2ch : L + R (stereo)
365 * 3ch : front center + L + R
366 * 4ch : front center + L + R + back center
367 * 5ch : front center + L + R + back stereo
368 * 6ch : front center + L + R + back stereo + LFE
369 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
372 if (channel_config != 2)
373 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
374 if (channel_config > 1)
375 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
376 if (channel_config == 4)
377 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
378 if (channel_config > 4)
379 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
380 = AAC_CHANNEL_BACK; // back stereo
381 if (channel_config > 5)
382 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
383 if (channel_config == 7)
384 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
390 * Decode GA "General Audio" specific configuration; reference: table 4.1.
392 * @param ac pointer to AACContext, may be null
393 * @param avctx pointer to AVCCodecContext, used for logging
395 * @return Returns error status. 0 - OK, !0 - error
397 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
399 MPEG4AudioConfig *m4ac,
402 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
403 int extension_flag, ret;
405 if (get_bits1(gb)) { // frameLengthFlag
406 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
410 if (get_bits1(gb)) // dependsOnCoreCoder
411 skip_bits(gb, 14); // coreCoderDelay
412 extension_flag = get_bits1(gb);
414 if (m4ac->object_type == AOT_AAC_SCALABLE ||
415 m4ac->object_type == AOT_ER_AAC_SCALABLE)
416 skip_bits(gb, 3); // layerNr
418 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
419 if (channel_config == 0) {
420 skip_bits(gb, 4); // element_instance_tag
421 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
424 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
427 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
430 if (extension_flag) {
431 switch (m4ac->object_type) {
433 skip_bits(gb, 5); // numOfSubFrame
434 skip_bits(gb, 11); // layer_length
438 case AOT_ER_AAC_SCALABLE:
440 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
441 * aacScalefactorDataResilienceFlag
442 * aacSpectralDataResilienceFlag
446 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
452 * Decode audio specific configuration; reference: table 1.13.
454 * @param ac pointer to AACContext, may be null
455 * @param avctx pointer to AVCCodecContext, used for logging
456 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
457 * @param data pointer to AVCodecContext extradata
458 * @param data_size size of AVCCodecContext extradata
460 * @return Returns error status or number of consumed bits. <0 - error
462 static int decode_audio_specific_config(AACContext *ac,
463 AVCodecContext *avctx,
464 MPEG4AudioConfig *m4ac,
465 const uint8_t *data, int data_size)
470 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
471 for (i = 0; i < avctx->extradata_size; i++)
472 av_dlog(avctx, "%02x ", avctx->extradata[i]);
473 av_dlog(avctx, "\n");
475 init_get_bits(&gb, data, data_size * 8);
477 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
479 if (m4ac->sampling_index > 12) {
480 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
483 if (m4ac->sbr == 1 && m4ac->ps == -1)
486 skip_bits_long(&gb, i);
488 switch (m4ac->object_type) {
492 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
496 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
497 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
501 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
502 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
503 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
505 return get_bits_count(&gb);
509 * linear congruential pseudorandom number generator
511 * @param previous_val pointer to the current state of the generator
513 * @return Returns a 32-bit pseudorandom integer
515 static av_always_inline int lcg_random(int previous_val)
517 return previous_val * 1664525 + 1013904223;
520 static av_always_inline void reset_predict_state(PredictorState *ps)
530 static void reset_all_predictors(PredictorState *ps)
533 for (i = 0; i < MAX_PREDICTORS; i++)
534 reset_predict_state(&ps[i]);
537 static void reset_predictor_group(PredictorState *ps, int group_num)
540 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
541 reset_predict_state(&ps[i]);
544 #define AAC_INIT_VLC_STATIC(num, size) \
545 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
546 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
547 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
550 static av_cold int aac_decode_init(AVCodecContext *avctx)
552 AACContext *ac = avctx->priv_data;
555 ac->m4ac.sample_rate = avctx->sample_rate;
557 if (avctx->extradata_size > 0) {
558 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
560 avctx->extradata_size) < 0)
564 avctx->sample_fmt = avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT ?
565 AV_SAMPLE_FMT_FLT : AV_SAMPLE_FMT_S16;
567 AAC_INIT_VLC_STATIC( 0, 304);
568 AAC_INIT_VLC_STATIC( 1, 270);
569 AAC_INIT_VLC_STATIC( 2, 550);
570 AAC_INIT_VLC_STATIC( 3, 300);
571 AAC_INIT_VLC_STATIC( 4, 328);
572 AAC_INIT_VLC_STATIC( 5, 294);
573 AAC_INIT_VLC_STATIC( 6, 306);
574 AAC_INIT_VLC_STATIC( 7, 268);
575 AAC_INIT_VLC_STATIC( 8, 510);
576 AAC_INIT_VLC_STATIC( 9, 366);
577 AAC_INIT_VLC_STATIC(10, 462);
581 dsputil_init(&ac->dsp, avctx);
582 ff_fmt_convert_init(&ac->fmt_conv, avctx);
584 ac->random_state = 0x1f2e3d4c;
588 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
589 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
590 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
593 ff_mdct_init(&ac->mdct, 11, 1, 1.0/1024.0);
594 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0/128.0);
595 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0);
596 // window initialization
597 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
598 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
599 ff_init_ff_sine_windows(10);
600 ff_init_ff_sine_windows( 7);
608 * Skip data_stream_element; reference: table 4.10.
610 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
612 int byte_align = get_bits1(gb);
613 int count = get_bits(gb, 8);
615 count += get_bits(gb, 8);
619 if (get_bits_left(gb) < 8 * count) {
620 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
623 skip_bits_long(gb, 8 * count);
627 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
632 ics->predictor_reset_group = get_bits(gb, 5);
633 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
634 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
638 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
639 ics->prediction_used[sfb] = get_bits1(gb);
645 * Decode Long Term Prediction data; reference: table 4.xx.
647 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
648 GetBitContext *gb, uint8_t max_sfb)
652 ltp->lag = get_bits(gb, 11);
653 ltp->coef = ltp_coef[get_bits(gb, 3)];
654 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
655 ltp->used[sfb] = get_bits1(gb);
659 * Decode Individual Channel Stream info; reference: table 4.6.
661 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
663 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
664 GetBitContext *gb, int common_window)
667 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
668 memset(ics, 0, sizeof(IndividualChannelStream));
671 ics->window_sequence[1] = ics->window_sequence[0];
672 ics->window_sequence[0] = get_bits(gb, 2);
673 ics->use_kb_window[1] = ics->use_kb_window[0];
674 ics->use_kb_window[0] = get_bits1(gb);
675 ics->num_window_groups = 1;
676 ics->group_len[0] = 1;
677 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
679 ics->max_sfb = get_bits(gb, 4);
680 for (i = 0; i < 7; i++) {
682 ics->group_len[ics->num_window_groups - 1]++;
684 ics->num_window_groups++;
685 ics->group_len[ics->num_window_groups - 1] = 1;
688 ics->num_windows = 8;
689 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
690 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
691 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
692 ics->predictor_present = 0;
694 ics->max_sfb = get_bits(gb, 6);
695 ics->num_windows = 1;
696 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
697 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
698 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
699 ics->predictor_present = get_bits1(gb);
700 ics->predictor_reset_group = 0;
701 if (ics->predictor_present) {
702 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
703 if (decode_prediction(ac, ics, gb)) {
704 memset(ics, 0, sizeof(IndividualChannelStream));
707 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
708 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
709 memset(ics, 0, sizeof(IndividualChannelStream));
712 if ((ics->ltp.present = get_bits(gb, 1)))
713 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
718 if (ics->max_sfb > ics->num_swb) {
719 av_log(ac->avctx, AV_LOG_ERROR,
720 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
721 ics->max_sfb, ics->num_swb);
722 memset(ics, 0, sizeof(IndividualChannelStream));
730 * Decode band types (section_data payload); reference: table 4.46.
732 * @param band_type array of the used band type
733 * @param band_type_run_end array of the last scalefactor band of a band type run
735 * @return Returns error status. 0 - OK, !0 - error
737 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
738 int band_type_run_end[120], GetBitContext *gb,
739 IndividualChannelStream *ics)
742 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
743 for (g = 0; g < ics->num_window_groups; g++) {
745 while (k < ics->max_sfb) {
746 uint8_t sect_end = k;
748 int sect_band_type = get_bits(gb, 4);
749 if (sect_band_type == 12) {
750 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
753 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
754 sect_end += sect_len_incr;
755 sect_end += sect_len_incr;
756 if (get_bits_left(gb) < 0) {
757 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
760 if (sect_end > ics->max_sfb) {
761 av_log(ac->avctx, AV_LOG_ERROR,
762 "Number of bands (%d) exceeds limit (%d).\n",
763 sect_end, ics->max_sfb);
766 for (; k < sect_end; k++) {
767 band_type [idx] = sect_band_type;
768 band_type_run_end[idx++] = sect_end;
776 * Decode scalefactors; reference: table 4.47.
778 * @param global_gain first scalefactor value as scalefactors are differentially coded
779 * @param band_type array of the used band type
780 * @param band_type_run_end array of the last scalefactor band of a band type run
781 * @param sf array of scalefactors or intensity stereo positions
783 * @return Returns error status. 0 - OK, !0 - error
785 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
786 unsigned int global_gain,
787 IndividualChannelStream *ics,
788 enum BandType band_type[120],
789 int band_type_run_end[120])
792 int offset[3] = { global_gain, global_gain - 90, 0 };
795 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
796 for (g = 0; g < ics->num_window_groups; g++) {
797 for (i = 0; i < ics->max_sfb;) {
798 int run_end = band_type_run_end[idx];
799 if (band_type[idx] == ZERO_BT) {
800 for (; i < run_end; i++, idx++)
802 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
803 for (; i < run_end; i++, idx++) {
804 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
805 clipped_offset = av_clip(offset[2], -155, 100);
806 if (offset[2] != clipped_offset) {
807 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
808 "position clipped (%d -> %d).\nIf you heard an "
809 "audible artifact, there may be a bug in the "
810 "decoder. ", offset[2], clipped_offset);
812 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
814 } else if (band_type[idx] == NOISE_BT) {
815 for (; i < run_end; i++, idx++) {
816 if (noise_flag-- > 0)
817 offset[1] += get_bits(gb, 9) - 256;
819 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
820 clipped_offset = av_clip(offset[1], -100, 155);
821 if (offset[2] != clipped_offset) {
822 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
823 "(%d -> %d).\nIf you heard an audible "
824 "artifact, there may be a bug in the decoder. ",
825 offset[1], clipped_offset);
827 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
830 for (; i < run_end; i++, idx++) {
831 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
832 if (offset[0] > 255U) {
833 av_log(ac->avctx, AV_LOG_ERROR,
834 "%s (%d) out of range.\n", sf_str[0], offset[0]);
837 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
846 * Decode pulse data; reference: table 4.7.
848 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
849 const uint16_t *swb_offset, int num_swb)
852 pulse->num_pulse = get_bits(gb, 2) + 1;
853 pulse_swb = get_bits(gb, 6);
854 if (pulse_swb >= num_swb)
856 pulse->pos[0] = swb_offset[pulse_swb];
857 pulse->pos[0] += get_bits(gb, 5);
858 if (pulse->pos[0] > 1023)
860 pulse->amp[0] = get_bits(gb, 4);
861 for (i = 1; i < pulse->num_pulse; i++) {
862 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
863 if (pulse->pos[i] > 1023)
865 pulse->amp[i] = get_bits(gb, 4);
871 * Decode Temporal Noise Shaping data; reference: table 4.48.
873 * @return Returns error status. 0 - OK, !0 - error
875 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
876 GetBitContext *gb, const IndividualChannelStream *ics)
878 int w, filt, i, coef_len, coef_res, coef_compress;
879 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
880 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
881 for (w = 0; w < ics->num_windows; w++) {
882 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
883 coef_res = get_bits1(gb);
885 for (filt = 0; filt < tns->n_filt[w]; filt++) {
887 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
889 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
890 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
891 tns->order[w][filt], tns_max_order);
892 tns->order[w][filt] = 0;
895 if (tns->order[w][filt]) {
896 tns->direction[w][filt] = get_bits1(gb);
897 coef_compress = get_bits1(gb);
898 coef_len = coef_res + 3 - coef_compress;
899 tmp2_idx = 2 * coef_compress + coef_res;
901 for (i = 0; i < tns->order[w][filt]; i++)
902 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
911 * Decode Mid/Side data; reference: table 4.54.
913 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
914 * [1] mask is decoded from bitstream; [2] mask is all 1s;
915 * [3] reserved for scalable AAC
917 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
921 if (ms_present == 1) {
922 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
923 cpe->ms_mask[idx] = get_bits1(gb);
924 } else if (ms_present == 2) {
925 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
930 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
934 *dst++ = v[idx & 15] * s;
935 *dst++ = v[idx>>4 & 15] * s;
941 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
945 *dst++ = v[idx & 3] * s;
946 *dst++ = v[idx>>2 & 3] * s;
947 *dst++ = v[idx>>4 & 3] * s;
948 *dst++ = v[idx>>6 & 3] * s;
954 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
955 unsigned sign, const float *scale)
957 union float754 s0, s1;
959 s0.f = s1.f = *scale;
960 s0.i ^= sign >> 1 << 31;
963 *dst++ = v[idx & 15] * s0.f;
964 *dst++ = v[idx>>4 & 15] * s1.f;
971 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
972 unsigned sign, const float *scale)
974 unsigned nz = idx >> 12;
975 union float754 s = { .f = *scale };
978 t.i = s.i ^ (sign & 1U<<31);
979 *dst++ = v[idx & 3] * t.f;
981 sign <<= nz & 1; nz >>= 1;
982 t.i = s.i ^ (sign & 1U<<31);
983 *dst++ = v[idx>>2 & 3] * t.f;
985 sign <<= nz & 1; nz >>= 1;
986 t.i = s.i ^ (sign & 1U<<31);
987 *dst++ = v[idx>>4 & 3] * t.f;
989 sign <<= nz & 1; nz >>= 1;
990 t.i = s.i ^ (sign & 1U<<31);
991 *dst++ = v[idx>>6 & 3] * t.f;
998 * Decode spectral data; reference: table 4.50.
999 * Dequantize and scale spectral data; reference: 4.6.3.3.
1001 * @param coef array of dequantized, scaled spectral data
1002 * @param sf array of scalefactors or intensity stereo positions
1003 * @param pulse_present set if pulses are present
1004 * @param pulse pointer to pulse data struct
1005 * @param band_type array of the used band type
1007 * @return Returns error status. 0 - OK, !0 - error
1009 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1010 GetBitContext *gb, const float sf[120],
1011 int pulse_present, const Pulse *pulse,
1012 const IndividualChannelStream *ics,
1013 enum BandType band_type[120])
1015 int i, k, g, idx = 0;
1016 const int c = 1024 / ics->num_windows;
1017 const uint16_t *offsets = ics->swb_offset;
1018 float *coef_base = coef;
1020 for (g = 0; g < ics->num_windows; g++)
1021 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1023 for (g = 0; g < ics->num_window_groups; g++) {
1024 unsigned g_len = ics->group_len[g];
1026 for (i = 0; i < ics->max_sfb; i++, idx++) {
1027 const unsigned cbt_m1 = band_type[idx] - 1;
1028 float *cfo = coef + offsets[i];
1029 int off_len = offsets[i + 1] - offsets[i];
1032 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1033 for (group = 0; group < g_len; group++, cfo+=128) {
1034 memset(cfo, 0, off_len * sizeof(float));
1036 } else if (cbt_m1 == NOISE_BT - 1) {
1037 for (group = 0; group < g_len; group++, cfo+=128) {
1041 for (k = 0; k < off_len; k++) {
1042 ac->random_state = lcg_random(ac->random_state);
1043 cfo[k] = ac->random_state;
1046 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1047 scale = sf[idx] / sqrtf(band_energy);
1048 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1051 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1052 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1053 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1054 OPEN_READER(re, gb);
1056 switch (cbt_m1 >> 1) {
1058 for (group = 0; group < g_len; group++, cfo+=128) {
1066 UPDATE_CACHE(re, gb);
1067 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1068 cb_idx = cb_vector_idx[code];
1069 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1075 for (group = 0; group < g_len; group++, cfo+=128) {
1085 UPDATE_CACHE(re, gb);
1086 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1087 cb_idx = cb_vector_idx[code];
1088 nnz = cb_idx >> 8 & 15;
1089 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1090 LAST_SKIP_BITS(re, gb, nnz);
1091 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1097 for (group = 0; group < g_len; group++, cfo+=128) {
1105 UPDATE_CACHE(re, gb);
1106 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1107 cb_idx = cb_vector_idx[code];
1108 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1115 for (group = 0; group < g_len; group++, cfo+=128) {
1125 UPDATE_CACHE(re, gb);
1126 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1127 cb_idx = cb_vector_idx[code];
1128 nnz = cb_idx >> 8 & 15;
1129 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1130 LAST_SKIP_BITS(re, gb, nnz);
1131 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1137 for (group = 0; group < g_len; group++, cfo+=128) {
1139 uint32_t *icf = (uint32_t *) cf;
1149 UPDATE_CACHE(re, gb);
1150 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1158 cb_idx = cb_vector_idx[code];
1161 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1162 LAST_SKIP_BITS(re, gb, nnz);
1164 for (j = 0; j < 2; j++) {
1168 /* The total length of escape_sequence must be < 22 bits according
1169 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1170 UPDATE_CACHE(re, gb);
1171 b = GET_CACHE(re, gb);
1172 b = 31 - av_log2(~b);
1175 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1179 SKIP_BITS(re, gb, b + 1);
1181 n = (1 << b) + SHOW_UBITS(re, gb, b);
1182 LAST_SKIP_BITS(re, gb, b);
1183 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1186 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1187 *icf++ = (bits & 1U<<31) | v;
1194 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1198 CLOSE_READER(re, gb);
1204 if (pulse_present) {
1206 for (i = 0; i < pulse->num_pulse; i++) {
1207 float co = coef_base[ pulse->pos[i] ];
1208 while (offsets[idx + 1] <= pulse->pos[i])
1210 if (band_type[idx] != NOISE_BT && sf[idx]) {
1211 float ico = -pulse->amp[i];
1214 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1216 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1223 static av_always_inline float flt16_round(float pf)
1227 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1231 static av_always_inline float flt16_even(float pf)
1235 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1239 static av_always_inline float flt16_trunc(float pf)
1243 pun.i &= 0xFFFF0000U;
1247 static av_always_inline void predict(PredictorState *ps, float *coef,
1250 const float a = 0.953125; // 61.0 / 64
1251 const float alpha = 0.90625; // 29.0 / 32
1255 float r0 = ps->r0, r1 = ps->r1;
1256 float cor0 = ps->cor0, cor1 = ps->cor1;
1257 float var0 = ps->var0, var1 = ps->var1;
1259 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1260 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1262 pv = flt16_round(k1 * r0 + k2 * r1);
1269 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1270 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1271 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1272 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1274 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1275 ps->r0 = flt16_trunc(a * e0);
1279 * Apply AAC-Main style frequency domain prediction.
1281 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1285 if (!sce->ics.predictor_initialized) {
1286 reset_all_predictors(sce->predictor_state);
1287 sce->ics.predictor_initialized = 1;
1290 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1291 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1292 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1293 predict(&sce->predictor_state[k], &sce->coeffs[k],
1294 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1297 if (sce->ics.predictor_reset_group)
1298 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1300 reset_all_predictors(sce->predictor_state);
1304 * Decode an individual_channel_stream payload; reference: table 4.44.
1306 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1307 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1309 * @return Returns error status. 0 - OK, !0 - error
1311 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1312 GetBitContext *gb, int common_window, int scale_flag)
1315 TemporalNoiseShaping *tns = &sce->tns;
1316 IndividualChannelStream *ics = &sce->ics;
1317 float *out = sce->coeffs;
1318 int global_gain, pulse_present = 0;
1320 /* This assignment is to silence a GCC warning about the variable being used
1321 * uninitialized when in fact it always is.
1323 pulse.num_pulse = 0;
1325 global_gain = get_bits(gb, 8);
1327 if (!common_window && !scale_flag) {
1328 if (decode_ics_info(ac, ics, gb, 0) < 0)
1332 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1334 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1339 if ((pulse_present = get_bits1(gb))) {
1340 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1341 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1344 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1345 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1349 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1351 if (get_bits1(gb)) {
1352 av_log_missing_feature(ac->avctx, "SSR", 1);
1357 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1360 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1361 apply_prediction(ac, sce);
1367 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1369 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1371 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1372 float *ch0 = cpe->ch[0].coeffs;
1373 float *ch1 = cpe->ch[1].coeffs;
1374 int g, i, group, idx = 0;
1375 const uint16_t *offsets = ics->swb_offset;
1376 for (g = 0; g < ics->num_window_groups; g++) {
1377 for (i = 0; i < ics->max_sfb; i++, idx++) {
1378 if (cpe->ms_mask[idx] &&
1379 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1380 for (group = 0; group < ics->group_len[g]; group++) {
1381 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1382 ch1 + group * 128 + offsets[i],
1383 offsets[i+1] - offsets[i]);
1387 ch0 += ics->group_len[g] * 128;
1388 ch1 += ics->group_len[g] * 128;
1393 * intensity stereo decoding; reference: 4.6.8.2.3
1395 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1396 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1397 * [3] reserved for scalable AAC
1399 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1401 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1402 SingleChannelElement *sce1 = &cpe->ch[1];
1403 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1404 const uint16_t *offsets = ics->swb_offset;
1405 int g, group, i, idx = 0;
1408 for (g = 0; g < ics->num_window_groups; g++) {
1409 for (i = 0; i < ics->max_sfb;) {
1410 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1411 const int bt_run_end = sce1->band_type_run_end[idx];
1412 for (; i < bt_run_end; i++, idx++) {
1413 c = -1 + 2 * (sce1->band_type[idx] - 14);
1415 c *= 1 - 2 * cpe->ms_mask[idx];
1416 scale = c * sce1->sf[idx];
1417 for (group = 0; group < ics->group_len[g]; group++)
1418 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1419 coef0 + group * 128 + offsets[i],
1421 offsets[i + 1] - offsets[i]);
1424 int bt_run_end = sce1->band_type_run_end[idx];
1425 idx += bt_run_end - i;
1429 coef0 += ics->group_len[g] * 128;
1430 coef1 += ics->group_len[g] * 128;
1435 * Decode a channel_pair_element; reference: table 4.4.
1437 * @return Returns error status. 0 - OK, !0 - error
1439 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1441 int i, ret, common_window, ms_present = 0;
1443 common_window = get_bits1(gb);
1444 if (common_window) {
1445 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1447 i = cpe->ch[1].ics.use_kb_window[0];
1448 cpe->ch[1].ics = cpe->ch[0].ics;
1449 cpe->ch[1].ics.use_kb_window[1] = i;
1450 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1451 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1452 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1453 ms_present = get_bits(gb, 2);
1454 if (ms_present == 3) {
1455 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1457 } else if (ms_present)
1458 decode_mid_side_stereo(cpe, gb, ms_present);
1460 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1462 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1465 if (common_window) {
1467 apply_mid_side_stereo(ac, cpe);
1468 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1469 apply_prediction(ac, &cpe->ch[0]);
1470 apply_prediction(ac, &cpe->ch[1]);
1474 apply_intensity_stereo(ac, cpe, ms_present);
1478 static const float cce_scale[] = {
1479 1.09050773266525765921, //2^(1/8)
1480 1.18920711500272106672, //2^(1/4)
1486 * Decode coupling_channel_element; reference: table 4.8.
1488 * @return Returns error status. 0 - OK, !0 - error
1490 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1496 SingleChannelElement *sce = &che->ch[0];
1497 ChannelCoupling *coup = &che->coup;
1499 coup->coupling_point = 2 * get_bits1(gb);
1500 coup->num_coupled = get_bits(gb, 3);
1501 for (c = 0; c <= coup->num_coupled; c++) {
1503 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1504 coup->id_select[c] = get_bits(gb, 4);
1505 if (coup->type[c] == TYPE_CPE) {
1506 coup->ch_select[c] = get_bits(gb, 2);
1507 if (coup->ch_select[c] == 3)
1510 coup->ch_select[c] = 2;
1512 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1514 sign = get_bits(gb, 1);
1515 scale = cce_scale[get_bits(gb, 2)];
1517 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1520 for (c = 0; c < num_gain; c++) {
1524 float gain_cache = 1.;
1526 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1527 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1528 gain_cache = powf(scale, -gain);
1530 if (coup->coupling_point == AFTER_IMDCT) {
1531 coup->gain[c][0] = gain_cache;
1533 for (g = 0; g < sce->ics.num_window_groups; g++) {
1534 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1535 if (sce->band_type[idx] != ZERO_BT) {
1537 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1545 gain_cache = powf(scale, -t) * s;
1548 coup->gain[c][idx] = gain_cache;
1558 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1560 * @return Returns number of bytes consumed.
1562 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1566 int num_excl_chan = 0;
1569 for (i = 0; i < 7; i++)
1570 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1571 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1573 return num_excl_chan / 7;
1577 * Decode dynamic range information; reference: table 4.52.
1579 * @param cnt length of TYPE_FIL syntactic element in bytes
1581 * @return Returns number of bytes consumed.
1583 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1584 GetBitContext *gb, int cnt)
1587 int drc_num_bands = 1;
1590 /* pce_tag_present? */
1591 if (get_bits1(gb)) {
1592 che_drc->pce_instance_tag = get_bits(gb, 4);
1593 skip_bits(gb, 4); // tag_reserved_bits
1597 /* excluded_chns_present? */
1598 if (get_bits1(gb)) {
1599 n += decode_drc_channel_exclusions(che_drc, gb);
1602 /* drc_bands_present? */
1603 if (get_bits1(gb)) {
1604 che_drc->band_incr = get_bits(gb, 4);
1605 che_drc->interpolation_scheme = get_bits(gb, 4);
1607 drc_num_bands += che_drc->band_incr;
1608 for (i = 0; i < drc_num_bands; i++) {
1609 che_drc->band_top[i] = get_bits(gb, 8);
1614 /* prog_ref_level_present? */
1615 if (get_bits1(gb)) {
1616 che_drc->prog_ref_level = get_bits(gb, 7);
1617 skip_bits1(gb); // prog_ref_level_reserved_bits
1621 for (i = 0; i < drc_num_bands; i++) {
1622 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1623 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1631 * Decode extension data (incomplete); reference: table 4.51.
1633 * @param cnt length of TYPE_FIL syntactic element in bytes
1635 * @return Returns number of bytes consumed
1637 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1638 ChannelElement *che, enum RawDataBlockType elem_type)
1642 switch (get_bits(gb, 4)) { // extension type
1643 case EXT_SBR_DATA_CRC:
1647 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1649 } else if (!ac->m4ac.sbr) {
1650 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1651 skip_bits_long(gb, 8 * cnt - 4);
1653 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1654 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1655 skip_bits_long(gb, 8 * cnt - 4);
1657 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1660 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1664 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1666 case EXT_DYNAMIC_RANGE:
1667 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1671 case EXT_DATA_ELEMENT:
1673 skip_bits_long(gb, 8 * cnt - 4);
1680 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1682 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1683 * @param coef spectral coefficients
1685 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1686 IndividualChannelStream *ics, int decode)
1688 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1690 int bottom, top, order, start, end, size, inc;
1691 float lpc[TNS_MAX_ORDER];
1692 float tmp[TNS_MAX_ORDER];
1694 for (w = 0; w < ics->num_windows; w++) {
1695 bottom = ics->num_swb;
1696 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1698 bottom = FFMAX(0, top - tns->length[w][filt]);
1699 order = tns->order[w][filt];
1704 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1706 start = ics->swb_offset[FFMIN(bottom, mmm)];
1707 end = ics->swb_offset[FFMIN( top, mmm)];
1708 if ((size = end - start) <= 0)
1710 if (tns->direction[w][filt]) {
1720 for (m = 0; m < size; m++, start += inc)
1721 for (i = 1; i <= FFMIN(m, order); i++)
1722 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1725 for (m = 0; m < size; m++, start += inc) {
1726 tmp[0] = coef[start];
1727 for (i = 1; i <= FFMIN(m, order); i++)
1728 coef[start] += tmp[i] * lpc[i - 1];
1729 for (i = order; i > 0; i--)
1730 tmp[i] = tmp[i - 1];
1738 * Apply windowing and MDCT to obtain the spectral
1739 * coefficient from the predicted sample by LTP.
1741 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1742 float *in, IndividualChannelStream *ics)
1744 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1745 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1746 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1747 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1749 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1750 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1752 memset(in, 0, 448 * sizeof(float));
1753 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1754 memcpy(in + 576, in + 576, 448 * sizeof(float));
1756 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1757 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1759 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1760 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1761 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1763 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1767 * Apply the long term prediction
1769 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1771 const LongTermPrediction *ltp = &sce->ics.ltp;
1772 const uint16_t *offsets = sce->ics.swb_offset;
1775 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1776 float *predTime = sce->ret;
1777 float *predFreq = ac->buf_mdct;
1778 int16_t num_samples = 2048;
1780 if (ltp->lag < 1024)
1781 num_samples = ltp->lag + 1024;
1782 for (i = 0; i < num_samples; i++)
1783 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1784 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1786 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1788 if (sce->tns.present)
1789 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1791 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1793 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1794 sce->coeffs[i] += predFreq[i];
1799 * Update the LTP buffer for next frame
1801 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1803 IndividualChannelStream *ics = &sce->ics;
1804 float *saved = sce->saved;
1805 float *saved_ltp = sce->coeffs;
1806 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1807 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1810 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1811 memcpy(saved_ltp, saved, 512 * sizeof(float));
1812 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1813 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1814 for (i = 0; i < 64; i++)
1815 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1816 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1817 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1818 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1819 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1820 for (i = 0; i < 64; i++)
1821 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1822 } else { // LONG_STOP or ONLY_LONG
1823 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1824 for (i = 0; i < 512; i++)
1825 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1828 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1829 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1830 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1834 * Conduct IMDCT and windowing.
1836 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1838 IndividualChannelStream *ics = &sce->ics;
1839 float *in = sce->coeffs;
1840 float *out = sce->ret;
1841 float *saved = sce->saved;
1842 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1843 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1844 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1845 float *buf = ac->buf_mdct;
1846 float *temp = ac->temp;
1850 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1851 for (i = 0; i < 1024; i += 128)
1852 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1854 ac->mdct.imdct_half(&ac->mdct, buf, in);
1856 /* window overlapping
1857 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1858 * and long to short transitions are considered to be short to short
1859 * transitions. This leaves just two cases (long to long and short to short)
1860 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1862 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1863 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1864 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1866 memcpy( out, saved, 448 * sizeof(float));
1868 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1869 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1870 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1871 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1872 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1873 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1874 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1876 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1877 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1882 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1883 memcpy( saved, temp + 64, 64 * sizeof(float));
1884 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1885 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1886 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1887 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1888 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1889 memcpy( saved, buf + 512, 448 * sizeof(float));
1890 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1891 } else { // LONG_STOP or ONLY_LONG
1892 memcpy( saved, buf + 512, 512 * sizeof(float));
1897 * Apply dependent channel coupling (applied before IMDCT).
1899 * @param index index into coupling gain array
1901 static void apply_dependent_coupling(AACContext *ac,
1902 SingleChannelElement *target,
1903 ChannelElement *cce, int index)
1905 IndividualChannelStream *ics = &cce->ch[0].ics;
1906 const uint16_t *offsets = ics->swb_offset;
1907 float *dest = target->coeffs;
1908 const float *src = cce->ch[0].coeffs;
1909 int g, i, group, k, idx = 0;
1910 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1911 av_log(ac->avctx, AV_LOG_ERROR,
1912 "Dependent coupling is not supported together with LTP\n");
1915 for (g = 0; g < ics->num_window_groups; g++) {
1916 for (i = 0; i < ics->max_sfb; i++, idx++) {
1917 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1918 const float gain = cce->coup.gain[index][idx];
1919 for (group = 0; group < ics->group_len[g]; group++) {
1920 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1922 dest[group * 128 + k] += gain * src[group * 128 + k];
1927 dest += ics->group_len[g] * 128;
1928 src += ics->group_len[g] * 128;
1933 * Apply independent channel coupling (applied after IMDCT).
1935 * @param index index into coupling gain array
1937 static void apply_independent_coupling(AACContext *ac,
1938 SingleChannelElement *target,
1939 ChannelElement *cce, int index)
1942 const float gain = cce->coup.gain[index][0];
1943 const float *src = cce->ch[0].ret;
1944 float *dest = target->ret;
1945 const int len = 1024 << (ac->m4ac.sbr == 1);
1947 for (i = 0; i < len; i++)
1948 dest[i] += gain * src[i];
1952 * channel coupling transformation interface
1954 * @param apply_coupling_method pointer to (in)dependent coupling function
1956 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1957 enum RawDataBlockType type, int elem_id,
1958 enum CouplingPoint coupling_point,
1959 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1963 for (i = 0; i < MAX_ELEM_ID; i++) {
1964 ChannelElement *cce = ac->che[TYPE_CCE][i];
1967 if (cce && cce->coup.coupling_point == coupling_point) {
1968 ChannelCoupling *coup = &cce->coup;
1970 for (c = 0; c <= coup->num_coupled; c++) {
1971 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1972 if (coup->ch_select[c] != 1) {
1973 apply_coupling_method(ac, &cc->ch[0], cce, index);
1974 if (coup->ch_select[c] != 0)
1977 if (coup->ch_select[c] != 2)
1978 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1980 index += 1 + (coup->ch_select[c] == 3);
1987 * Convert spectral data to float samples, applying all supported tools as appropriate.
1989 static void spectral_to_sample(AACContext *ac)
1992 for (type = 3; type >= 0; type--) {
1993 for (i = 0; i < MAX_ELEM_ID; i++) {
1994 ChannelElement *che = ac->che[type][i];
1996 if (type <= TYPE_CPE)
1997 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1998 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1999 if (che->ch[0].ics.predictor_present) {
2000 if (che->ch[0].ics.ltp.present)
2001 apply_ltp(ac, &che->ch[0]);
2002 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2003 apply_ltp(ac, &che->ch[1]);
2006 if (che->ch[0].tns.present)
2007 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2008 if (che->ch[1].tns.present)
2009 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2010 if (type <= TYPE_CPE)
2011 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2012 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2013 imdct_and_windowing(ac, &che->ch[0]);
2014 if (ac->m4ac.object_type == AOT_AAC_LTP)
2015 update_ltp(ac, &che->ch[0]);
2016 if (type == TYPE_CPE) {
2017 imdct_and_windowing(ac, &che->ch[1]);
2018 if (ac->m4ac.object_type == AOT_AAC_LTP)
2019 update_ltp(ac, &che->ch[1]);
2021 if (ac->m4ac.sbr > 0) {
2022 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2025 if (type <= TYPE_CCE)
2026 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2032 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2035 AACADTSHeaderInfo hdr_info;
2037 size = ff_aac_parse_header(gb, &hdr_info);
2039 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2040 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2041 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2042 ac->m4ac.chan_config = hdr_info.chan_config;
2043 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2045 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2047 } else if (ac->output_configured != OC_LOCKED) {
2048 ac->output_configured = OC_NONE;
2050 if (ac->output_configured != OC_LOCKED) {
2054 ac->m4ac.sample_rate = hdr_info.sample_rate;
2055 ac->m4ac.sampling_index = hdr_info.sampling_index;
2056 ac->m4ac.object_type = hdr_info.object_type;
2057 if (!ac->avctx->sample_rate)
2058 ac->avctx->sample_rate = hdr_info.sample_rate;
2059 if (hdr_info.num_aac_frames == 1) {
2060 if (!hdr_info.crc_absent)
2063 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2070 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2071 int *data_size, GetBitContext *gb)
2073 AACContext *ac = avctx->priv_data;
2074 ChannelElement *che = NULL, *che_prev = NULL;
2075 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2076 int err, elem_id, data_size_tmp;
2077 int samples = 0, multiplier;
2079 if (show_bits(gb, 12) == 0xfff) {
2080 if (parse_adts_frame_header(ac, gb) < 0) {
2081 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2084 if (ac->m4ac.sampling_index > 12) {
2085 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2090 ac->tags_mapped = 0;
2092 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2093 elem_id = get_bits(gb, 4);
2095 if (elem_type < TYPE_DSE) {
2096 if (!(che=get_che(ac, elem_type, elem_id))) {
2097 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2098 elem_type, elem_id);
2104 switch (elem_type) {
2107 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2111 err = decode_cpe(ac, gb, che);
2115 err = decode_cce(ac, gb, che);
2119 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2123 err = skip_data_stream_element(ac, gb);
2127 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2128 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2129 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2131 if (ac->output_configured > OC_TRIAL_PCE)
2132 av_log(avctx, AV_LOG_ERROR,
2133 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2135 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2141 elem_id += get_bits(gb, 8) - 1;
2142 if (get_bits_left(gb) < 8 * elem_id) {
2143 av_log(avctx, AV_LOG_ERROR, overread_err);
2147 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2148 err = 0; /* FIXME */
2152 err = -1; /* should not happen, but keeps compiler happy */
2157 elem_type_prev = elem_type;
2162 if (get_bits_left(gb) < 3) {
2163 av_log(avctx, AV_LOG_ERROR, overread_err);
2168 spectral_to_sample(ac);
2170 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2171 samples <<= multiplier;
2172 if (ac->output_configured < OC_LOCKED) {
2173 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2174 avctx->frame_size = samples;
2177 data_size_tmp = samples * avctx->channels;
2178 data_size_tmp *= avctx->sample_fmt == AV_SAMPLE_FMT_FLT ? sizeof(float) : sizeof(int16_t);
2179 if (*data_size < data_size_tmp) {
2180 av_log(avctx, AV_LOG_ERROR,
2181 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2182 *data_size, data_size_tmp);
2185 *data_size = data_size_tmp;
2188 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
2189 float_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2191 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2194 if (ac->output_configured)
2195 ac->output_configured = OC_LOCKED;
2200 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2201 int *data_size, AVPacket *avpkt)
2203 const uint8_t *buf = avpkt->data;
2204 int buf_size = avpkt->size;
2210 init_get_bits(&gb, buf, buf_size * 8);
2212 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2215 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2216 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2217 if (buf[buf_offset])
2220 return buf_size > buf_offset ? buf_consumed : buf_size;
2223 static av_cold int aac_decode_close(AVCodecContext *avctx)
2225 AACContext *ac = avctx->priv_data;
2228 for (i = 0; i < MAX_ELEM_ID; i++) {
2229 for (type = 0; type < 4; type++) {
2230 if (ac->che[type][i])
2231 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2232 av_freep(&ac->che[type][i]);
2236 ff_mdct_end(&ac->mdct);
2237 ff_mdct_end(&ac->mdct_small);
2238 ff_mdct_end(&ac->mdct_ltp);
2243 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2245 struct LATMContext {
2246 AACContext aac_ctx; ///< containing AACContext
2247 int initialized; ///< initilized after a valid extradata was seen
2250 int audio_mux_version_A; ///< LATM syntax version
2251 int frame_length_type; ///< 0/1 variable/fixed frame length
2252 int frame_length; ///< frame length for fixed frame length
2255 static inline uint32_t latm_get_value(GetBitContext *b)
2257 int length = get_bits(b, 2);
2259 return get_bits_long(b, (length+1)*8);
2262 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2265 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2266 MPEG4AudioConfig m4ac;
2267 int config_start_bit = get_bits_count(gb);
2268 int bits_consumed, esize;
2270 if (config_start_bit % 8) {
2271 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2272 "config not byte aligned.\n", 1);
2273 return AVERROR_INVALIDDATA;
2276 decode_audio_specific_config(NULL, avctx, &m4ac,
2277 gb->buffer + (config_start_bit / 8),
2278 get_bits_left(gb) / 8);
2280 if (bits_consumed < 0)
2281 return AVERROR_INVALIDDATA;
2283 esize = (bits_consumed+7) / 8;
2285 if (avctx->extradata_size <= esize) {
2286 av_free(avctx->extradata);
2287 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2288 if (!avctx->extradata)
2289 return AVERROR(ENOMEM);
2292 avctx->extradata_size = esize;
2293 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2294 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2296 skip_bits_long(gb, bits_consumed);
2299 return bits_consumed;
2302 static int read_stream_mux_config(struct LATMContext *latmctx,
2305 int ret, audio_mux_version = get_bits(gb, 1);
2307 latmctx->audio_mux_version_A = 0;
2308 if (audio_mux_version)
2309 latmctx->audio_mux_version_A = get_bits(gb, 1);
2311 if (!latmctx->audio_mux_version_A) {
2313 if (audio_mux_version)
2314 latm_get_value(gb); // taraFullness
2316 skip_bits(gb, 1); // allStreamSameTimeFraming
2317 skip_bits(gb, 6); // numSubFrames
2319 if (get_bits(gb, 4)) { // numPrograms
2320 av_log_missing_feature(latmctx->aac_ctx.avctx,
2321 "multiple programs are not supported\n", 1);
2322 return AVERROR_PATCHWELCOME;
2325 // for each program (which there is only on in DVB)
2327 // for each layer (which there is only on in DVB)
2328 if (get_bits(gb, 3)) { // numLayer
2329 av_log_missing_feature(latmctx->aac_ctx.avctx,
2330 "multiple layers are not supported\n", 1);
2331 return AVERROR_PATCHWELCOME;
2334 // for all but first stream: use_same_config = get_bits(gb, 1);
2335 if (!audio_mux_version) {
2336 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2339 int ascLen = latm_get_value(gb);
2340 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2343 skip_bits_long(gb, ascLen);
2346 latmctx->frame_length_type = get_bits(gb, 3);
2347 switch (latmctx->frame_length_type) {
2349 skip_bits(gb, 8); // latmBufferFullness
2352 latmctx->frame_length = get_bits(gb, 9);
2357 skip_bits(gb, 6); // CELP frame length table index
2361 skip_bits(gb, 1); // HVXC frame length table index
2365 if (get_bits(gb, 1)) { // other data
2366 if (audio_mux_version) {
2367 latm_get_value(gb); // other_data_bits
2371 esc = get_bits(gb, 1);
2377 if (get_bits(gb, 1)) // crc present
2378 skip_bits(gb, 8); // config_crc
2384 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2388 if (ctx->frame_length_type == 0) {
2389 int mux_slot_length = 0;
2391 tmp = get_bits(gb, 8);
2392 mux_slot_length += tmp;
2393 } while (tmp == 255);
2394 return mux_slot_length;
2395 } else if (ctx->frame_length_type == 1) {
2396 return ctx->frame_length;
2397 } else if (ctx->frame_length_type == 3 ||
2398 ctx->frame_length_type == 5 ||
2399 ctx->frame_length_type == 7) {
2400 skip_bits(gb, 2); // mux_slot_length_coded
2405 static int read_audio_mux_element(struct LATMContext *latmctx,
2409 uint8_t use_same_mux = get_bits(gb, 1);
2410 if (!use_same_mux) {
2411 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2413 } else if (!latmctx->aac_ctx.avctx->extradata) {
2414 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2415 "no decoder config found\n");
2416 return AVERROR(EAGAIN);
2418 if (latmctx->audio_mux_version_A == 0) {
2419 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2420 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2421 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2422 return AVERROR_INVALIDDATA;
2423 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2424 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2425 "frame length mismatch %d << %d\n",
2426 mux_slot_length_bytes * 8, get_bits_left(gb));
2427 return AVERROR_INVALIDDATA;
2434 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2437 struct LATMContext *latmctx = avctx->priv_data;
2441 if (avpkt->size == 0)
2444 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2446 // check for LOAS sync word
2447 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2448 return AVERROR_INVALIDDATA;
2450 muxlength = get_bits(&gb, 13) + 3;
2451 // not enough data, the parser should have sorted this
2452 if (muxlength > avpkt->size)
2453 return AVERROR_INVALIDDATA;
2455 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2458 if (!latmctx->initialized) {
2459 if (!avctx->extradata) {
2463 if ((err = aac_decode_init(avctx)) < 0)
2465 latmctx->initialized = 1;
2469 if (show_bits(&gb, 12) == 0xfff) {
2470 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2471 "ADTS header detected, probably as result of configuration "
2473 return AVERROR_INVALIDDATA;
2476 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2482 av_cold static int latm_decode_init(AVCodecContext *avctx)
2484 struct LATMContext *latmctx = avctx->priv_data;
2487 ret = aac_decode_init(avctx);
2489 if (avctx->extradata_size > 0) {
2490 latmctx->initialized = !ret;
2492 latmctx->initialized = 0;
2499 AVCodec ff_aac_decoder = {
2508 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2509 .sample_fmts = (const enum AVSampleFormat[]) {
2510 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2512 .channel_layouts = aac_channel_layout,
2516 Note: This decoder filter is intended to decode LATM streams transferred
2517 in MPEG transport streams which only contain one program.
2518 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2520 AVCodec ff_aac_latm_decoder = {
2522 .type = AVMEDIA_TYPE_AUDIO,
2523 .id = CODEC_ID_AAC_LATM,
2524 .priv_data_size = sizeof(struct LATMContext),
2525 .init = latm_decode_init,
2526 .close = aac_decode_close,
2527 .decode = latm_decode_frame,
2528 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2529 .sample_fmts = (const enum AVSampleFormat[]) {
2530 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2532 .channel_layouts = aac_channel_layout,