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 void reset_predictor_group(PredictorState *ps, int group_num)
538 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
539 reset_predict_state(&ps[i]);
542 #define AAC_INIT_VLC_STATIC(num, size) \
543 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
544 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
545 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
548 static av_cold int aac_decode_init(AVCodecContext *avctx)
550 AACContext *ac = avctx->priv_data;
551 float output_scale_factor;
554 ac->m4ac.sample_rate = avctx->sample_rate;
556 if (avctx->extradata_size > 0) {
557 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
559 avctx->extradata_size) < 0)
563 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
564 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
565 output_scale_factor = 1.0 / 32768.0;
567 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
568 output_scale_factor = 1.0;
571 AAC_INIT_VLC_STATIC( 0, 304);
572 AAC_INIT_VLC_STATIC( 1, 270);
573 AAC_INIT_VLC_STATIC( 2, 550);
574 AAC_INIT_VLC_STATIC( 3, 300);
575 AAC_INIT_VLC_STATIC( 4, 328);
576 AAC_INIT_VLC_STATIC( 5, 294);
577 AAC_INIT_VLC_STATIC( 6, 306);
578 AAC_INIT_VLC_STATIC( 7, 268);
579 AAC_INIT_VLC_STATIC( 8, 510);
580 AAC_INIT_VLC_STATIC( 9, 366);
581 AAC_INIT_VLC_STATIC(10, 462);
585 dsputil_init(&ac->dsp, avctx);
586 ff_fmt_convert_init(&ac->fmt_conv, avctx);
588 ac->random_state = 0x1f2e3d4c;
592 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
593 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
594 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
597 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
598 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
599 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
600 // window initialization
601 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
602 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
603 ff_init_ff_sine_windows(10);
604 ff_init_ff_sine_windows( 7);
612 * Skip data_stream_element; reference: table 4.10.
614 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
616 int byte_align = get_bits1(gb);
617 int count = get_bits(gb, 8);
619 count += get_bits(gb, 8);
623 if (get_bits_left(gb) < 8 * count) {
624 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
627 skip_bits_long(gb, 8 * count);
631 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
636 ics->predictor_reset_group = get_bits(gb, 5);
637 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
638 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
642 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
643 ics->prediction_used[sfb] = get_bits1(gb);
649 * Decode Long Term Prediction data; reference: table 4.xx.
651 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
652 GetBitContext *gb, uint8_t max_sfb)
656 ltp->lag = get_bits(gb, 11);
657 ltp->coef = ltp_coef[get_bits(gb, 3)];
658 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
659 ltp->used[sfb] = get_bits1(gb);
663 * Decode Individual Channel Stream info; reference: table 4.6.
665 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
667 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
668 GetBitContext *gb, int common_window)
671 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
672 memset(ics, 0, sizeof(IndividualChannelStream));
675 ics->window_sequence[1] = ics->window_sequence[0];
676 ics->window_sequence[0] = get_bits(gb, 2);
677 ics->use_kb_window[1] = ics->use_kb_window[0];
678 ics->use_kb_window[0] = get_bits1(gb);
679 ics->num_window_groups = 1;
680 ics->group_len[0] = 1;
681 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
683 ics->max_sfb = get_bits(gb, 4);
684 for (i = 0; i < 7; i++) {
686 ics->group_len[ics->num_window_groups - 1]++;
688 ics->num_window_groups++;
689 ics->group_len[ics->num_window_groups - 1] = 1;
692 ics->num_windows = 8;
693 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
694 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
695 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
696 ics->predictor_present = 0;
698 ics->max_sfb = get_bits(gb, 6);
699 ics->num_windows = 1;
700 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
701 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
702 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
703 ics->predictor_present = get_bits1(gb);
704 ics->predictor_reset_group = 0;
705 if (ics->predictor_present) {
706 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
707 if (decode_prediction(ac, ics, gb)) {
708 memset(ics, 0, sizeof(IndividualChannelStream));
711 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
712 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
713 memset(ics, 0, sizeof(IndividualChannelStream));
716 if ((ics->ltp.present = get_bits(gb, 1)))
717 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
722 if (ics->max_sfb > ics->num_swb) {
723 av_log(ac->avctx, AV_LOG_ERROR,
724 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
725 ics->max_sfb, ics->num_swb);
726 memset(ics, 0, sizeof(IndividualChannelStream));
734 * Decode band types (section_data payload); reference: table 4.46.
736 * @param band_type array of the used band type
737 * @param band_type_run_end array of the last scalefactor band of a band type run
739 * @return Returns error status. 0 - OK, !0 - error
741 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
742 int band_type_run_end[120], GetBitContext *gb,
743 IndividualChannelStream *ics)
746 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
747 for (g = 0; g < ics->num_window_groups; g++) {
749 while (k < ics->max_sfb) {
750 uint8_t sect_end = k;
752 int sect_band_type = get_bits(gb, 4);
753 if (sect_band_type == 12) {
754 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
757 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
758 sect_end += sect_len_incr;
759 sect_end += sect_len_incr;
760 if (get_bits_left(gb) < 0) {
761 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
764 if (sect_end > ics->max_sfb) {
765 av_log(ac->avctx, AV_LOG_ERROR,
766 "Number of bands (%d) exceeds limit (%d).\n",
767 sect_end, ics->max_sfb);
770 for (; k < sect_end; k++) {
771 band_type [idx] = sect_band_type;
772 band_type_run_end[idx++] = sect_end;
780 * Decode scalefactors; reference: table 4.47.
782 * @param global_gain first scalefactor value as scalefactors are differentially coded
783 * @param band_type array of the used band type
784 * @param band_type_run_end array of the last scalefactor band of a band type run
785 * @param sf array of scalefactors or intensity stereo positions
787 * @return Returns error status. 0 - OK, !0 - error
789 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
790 unsigned int global_gain,
791 IndividualChannelStream *ics,
792 enum BandType band_type[120],
793 int band_type_run_end[120])
796 int offset[3] = { global_gain, global_gain - 90, 0 };
799 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
800 for (g = 0; g < ics->num_window_groups; g++) {
801 for (i = 0; i < ics->max_sfb;) {
802 int run_end = band_type_run_end[idx];
803 if (band_type[idx] == ZERO_BT) {
804 for (; i < run_end; i++, idx++)
806 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
807 for (; i < run_end; i++, idx++) {
808 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
809 clipped_offset = av_clip(offset[2], -155, 100);
810 if (offset[2] != clipped_offset) {
811 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
812 "position clipped (%d -> %d).\nIf you heard an "
813 "audible artifact, there may be a bug in the "
814 "decoder. ", offset[2], clipped_offset);
816 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
818 } else if (band_type[idx] == NOISE_BT) {
819 for (; i < run_end; i++, idx++) {
820 if (noise_flag-- > 0)
821 offset[1] += get_bits(gb, 9) - 256;
823 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
824 clipped_offset = av_clip(offset[1], -100, 155);
825 if (offset[1] != clipped_offset) {
826 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
827 "(%d -> %d).\nIf you heard an audible "
828 "artifact, there may be a bug in the decoder. ",
829 offset[1], clipped_offset);
831 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
834 for (; i < run_end; i++, idx++) {
835 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
836 if (offset[0] > 255U) {
837 av_log(ac->avctx, AV_LOG_ERROR,
838 "%s (%d) out of range.\n", sf_str[0], offset[0]);
841 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
850 * Decode pulse data; reference: table 4.7.
852 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
853 const uint16_t *swb_offset, int num_swb)
856 pulse->num_pulse = get_bits(gb, 2) + 1;
857 pulse_swb = get_bits(gb, 6);
858 if (pulse_swb >= num_swb)
860 pulse->pos[0] = swb_offset[pulse_swb];
861 pulse->pos[0] += get_bits(gb, 5);
862 if (pulse->pos[0] > 1023)
864 pulse->amp[0] = get_bits(gb, 4);
865 for (i = 1; i < pulse->num_pulse; i++) {
866 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
867 if (pulse->pos[i] > 1023)
869 pulse->amp[i] = get_bits(gb, 4);
875 * Decode Temporal Noise Shaping data; reference: table 4.48.
877 * @return Returns error status. 0 - OK, !0 - error
879 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
880 GetBitContext *gb, const IndividualChannelStream *ics)
882 int w, filt, i, coef_len, coef_res, coef_compress;
883 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
884 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
885 for (w = 0; w < ics->num_windows; w++) {
886 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
887 coef_res = get_bits1(gb);
889 for (filt = 0; filt < tns->n_filt[w]; filt++) {
891 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
893 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
894 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
895 tns->order[w][filt], tns_max_order);
896 tns->order[w][filt] = 0;
899 if (tns->order[w][filt]) {
900 tns->direction[w][filt] = get_bits1(gb);
901 coef_compress = get_bits1(gb);
902 coef_len = coef_res + 3 - coef_compress;
903 tmp2_idx = 2 * coef_compress + coef_res;
905 for (i = 0; i < tns->order[w][filt]; i++)
906 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
915 * Decode Mid/Side data; reference: table 4.54.
917 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
918 * [1] mask is decoded from bitstream; [2] mask is all 1s;
919 * [3] reserved for scalable AAC
921 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
925 if (ms_present == 1) {
926 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
927 cpe->ms_mask[idx] = get_bits1(gb);
928 } else if (ms_present == 2) {
929 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
934 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
938 *dst++ = v[idx & 15] * s;
939 *dst++ = v[idx>>4 & 15] * s;
945 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
949 *dst++ = v[idx & 3] * s;
950 *dst++ = v[idx>>2 & 3] * s;
951 *dst++ = v[idx>>4 & 3] * s;
952 *dst++ = v[idx>>6 & 3] * s;
958 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
959 unsigned sign, const float *scale)
961 union float754 s0, s1;
963 s0.f = s1.f = *scale;
964 s0.i ^= sign >> 1 << 31;
967 *dst++ = v[idx & 15] * s0.f;
968 *dst++ = v[idx>>4 & 15] * s1.f;
975 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
976 unsigned sign, const float *scale)
978 unsigned nz = idx >> 12;
979 union float754 s = { .f = *scale };
982 t.i = s.i ^ (sign & 1U<<31);
983 *dst++ = v[idx & 3] * t.f;
985 sign <<= nz & 1; nz >>= 1;
986 t.i = s.i ^ (sign & 1U<<31);
987 *dst++ = v[idx>>2 & 3] * t.f;
989 sign <<= nz & 1; nz >>= 1;
990 t.i = s.i ^ (sign & 1U<<31);
991 *dst++ = v[idx>>4 & 3] * t.f;
993 sign <<= nz & 1; nz >>= 1;
994 t.i = s.i ^ (sign & 1U<<31);
995 *dst++ = v[idx>>6 & 3] * t.f;
1002 * Decode spectral data; reference: table 4.50.
1003 * Dequantize and scale spectral data; reference: 4.6.3.3.
1005 * @param coef array of dequantized, scaled spectral data
1006 * @param sf array of scalefactors or intensity stereo positions
1007 * @param pulse_present set if pulses are present
1008 * @param pulse pointer to pulse data struct
1009 * @param band_type array of the used band type
1011 * @return Returns error status. 0 - OK, !0 - error
1013 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1014 GetBitContext *gb, const float sf[120],
1015 int pulse_present, const Pulse *pulse,
1016 const IndividualChannelStream *ics,
1017 enum BandType band_type[120])
1019 int i, k, g, idx = 0;
1020 const int c = 1024 / ics->num_windows;
1021 const uint16_t *offsets = ics->swb_offset;
1022 float *coef_base = coef;
1024 for (g = 0; g < ics->num_windows; g++)
1025 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1027 for (g = 0; g < ics->num_window_groups; g++) {
1028 unsigned g_len = ics->group_len[g];
1030 for (i = 0; i < ics->max_sfb; i++, idx++) {
1031 const unsigned cbt_m1 = band_type[idx] - 1;
1032 float *cfo = coef + offsets[i];
1033 int off_len = offsets[i + 1] - offsets[i];
1036 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1037 for (group = 0; group < g_len; group++, cfo+=128) {
1038 memset(cfo, 0, off_len * sizeof(float));
1040 } else if (cbt_m1 == NOISE_BT - 1) {
1041 for (group = 0; group < g_len; group++, cfo+=128) {
1045 for (k = 0; k < off_len; k++) {
1046 ac->random_state = lcg_random(ac->random_state);
1047 cfo[k] = ac->random_state;
1050 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1051 scale = sf[idx] / sqrtf(band_energy);
1052 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1055 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1056 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1057 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1058 OPEN_READER(re, gb);
1060 switch (cbt_m1 >> 1) {
1062 for (group = 0; group < g_len; group++, cfo+=128) {
1070 UPDATE_CACHE(re, gb);
1071 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1072 cb_idx = cb_vector_idx[code];
1073 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1079 for (group = 0; group < g_len; group++, cfo+=128) {
1089 UPDATE_CACHE(re, gb);
1090 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1091 cb_idx = cb_vector_idx[code];
1092 nnz = cb_idx >> 8 & 15;
1093 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1094 LAST_SKIP_BITS(re, gb, nnz);
1095 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1101 for (group = 0; group < g_len; group++, cfo+=128) {
1109 UPDATE_CACHE(re, gb);
1110 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1111 cb_idx = cb_vector_idx[code];
1112 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1119 for (group = 0; group < g_len; group++, cfo+=128) {
1129 UPDATE_CACHE(re, gb);
1130 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1131 cb_idx = cb_vector_idx[code];
1132 nnz = cb_idx >> 8 & 15;
1133 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1134 LAST_SKIP_BITS(re, gb, nnz);
1135 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1141 for (group = 0; group < g_len; group++, cfo+=128) {
1143 uint32_t *icf = (uint32_t *) cf;
1153 UPDATE_CACHE(re, gb);
1154 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1162 cb_idx = cb_vector_idx[code];
1165 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1166 LAST_SKIP_BITS(re, gb, nnz);
1168 for (j = 0; j < 2; j++) {
1172 /* The total length of escape_sequence must be < 22 bits according
1173 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1174 UPDATE_CACHE(re, gb);
1175 b = GET_CACHE(re, gb);
1176 b = 31 - av_log2(~b);
1179 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1183 SKIP_BITS(re, gb, b + 1);
1185 n = (1 << b) + SHOW_UBITS(re, gb, b);
1186 LAST_SKIP_BITS(re, gb, b);
1187 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1190 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1191 *icf++ = (bits & 1U<<31) | v;
1198 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1202 CLOSE_READER(re, gb);
1208 if (pulse_present) {
1210 for (i = 0; i < pulse->num_pulse; i++) {
1211 float co = coef_base[ pulse->pos[i] ];
1212 while (offsets[idx + 1] <= pulse->pos[i])
1214 if (band_type[idx] != NOISE_BT && sf[idx]) {
1215 float ico = -pulse->amp[i];
1218 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1220 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1227 static av_always_inline float flt16_round(float pf)
1231 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1235 static av_always_inline float flt16_even(float pf)
1239 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1243 static av_always_inline float flt16_trunc(float pf)
1247 pun.i &= 0xFFFF0000U;
1251 static av_always_inline void predict(PredictorState *ps, float *coef,
1254 const float a = 0.953125; // 61.0 / 64
1255 const float alpha = 0.90625; // 29.0 / 32
1259 float r0 = ps->r0, r1 = ps->r1;
1260 float cor0 = ps->cor0, cor1 = ps->cor1;
1261 float var0 = ps->var0, var1 = ps->var1;
1263 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1264 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1266 pv = flt16_round(k1 * r0 + k2 * r1);
1273 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1274 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1275 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1276 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1278 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1279 ps->r0 = flt16_trunc(a * e0);
1283 * Apply AAC-Main style frequency domain prediction.
1285 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1289 if (!sce->ics.predictor_initialized) {
1290 reset_all_predictors(sce->predictor_state);
1291 sce->ics.predictor_initialized = 1;
1294 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1295 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1296 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1297 predict(&sce->predictor_state[k], &sce->coeffs[k],
1298 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1301 if (sce->ics.predictor_reset_group)
1302 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1304 reset_all_predictors(sce->predictor_state);
1308 * Decode an individual_channel_stream payload; reference: table 4.44.
1310 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1311 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1313 * @return Returns error status. 0 - OK, !0 - error
1315 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1316 GetBitContext *gb, int common_window, int scale_flag)
1319 TemporalNoiseShaping *tns = &sce->tns;
1320 IndividualChannelStream *ics = &sce->ics;
1321 float *out = sce->coeffs;
1322 int global_gain, pulse_present = 0;
1324 /* This assignment is to silence a GCC warning about the variable being used
1325 * uninitialized when in fact it always is.
1327 pulse.num_pulse = 0;
1329 global_gain = get_bits(gb, 8);
1331 if (!common_window && !scale_flag) {
1332 if (decode_ics_info(ac, ics, gb, 0) < 0)
1336 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1338 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1343 if ((pulse_present = get_bits1(gb))) {
1344 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1345 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1348 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1349 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1353 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1355 if (get_bits1(gb)) {
1356 av_log_missing_feature(ac->avctx, "SSR", 1);
1361 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1364 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1365 apply_prediction(ac, sce);
1371 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1373 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1375 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1376 float *ch0 = cpe->ch[0].coeffs;
1377 float *ch1 = cpe->ch[1].coeffs;
1378 int g, i, group, idx = 0;
1379 const uint16_t *offsets = ics->swb_offset;
1380 for (g = 0; g < ics->num_window_groups; g++) {
1381 for (i = 0; i < ics->max_sfb; i++, idx++) {
1382 if (cpe->ms_mask[idx] &&
1383 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1384 for (group = 0; group < ics->group_len[g]; group++) {
1385 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1386 ch1 + group * 128 + offsets[i],
1387 offsets[i+1] - offsets[i]);
1391 ch0 += ics->group_len[g] * 128;
1392 ch1 += ics->group_len[g] * 128;
1397 * intensity stereo decoding; reference: 4.6.8.2.3
1399 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1400 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1401 * [3] reserved for scalable AAC
1403 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1405 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1406 SingleChannelElement *sce1 = &cpe->ch[1];
1407 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1408 const uint16_t *offsets = ics->swb_offset;
1409 int g, group, i, idx = 0;
1412 for (g = 0; g < ics->num_window_groups; g++) {
1413 for (i = 0; i < ics->max_sfb;) {
1414 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1415 const int bt_run_end = sce1->band_type_run_end[idx];
1416 for (; i < bt_run_end; i++, idx++) {
1417 c = -1 + 2 * (sce1->band_type[idx] - 14);
1419 c *= 1 - 2 * cpe->ms_mask[idx];
1420 scale = c * sce1->sf[idx];
1421 for (group = 0; group < ics->group_len[g]; group++)
1422 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1423 coef0 + group * 128 + offsets[i],
1425 offsets[i + 1] - offsets[i]);
1428 int bt_run_end = sce1->band_type_run_end[idx];
1429 idx += bt_run_end - i;
1433 coef0 += ics->group_len[g] * 128;
1434 coef1 += ics->group_len[g] * 128;
1439 * Decode a channel_pair_element; reference: table 4.4.
1441 * @return Returns error status. 0 - OK, !0 - error
1443 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1445 int i, ret, common_window, ms_present = 0;
1447 common_window = get_bits1(gb);
1448 if (common_window) {
1449 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1451 i = cpe->ch[1].ics.use_kb_window[0];
1452 cpe->ch[1].ics = cpe->ch[0].ics;
1453 cpe->ch[1].ics.use_kb_window[1] = i;
1454 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1455 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1456 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1457 ms_present = get_bits(gb, 2);
1458 if (ms_present == 3) {
1459 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1461 } else if (ms_present)
1462 decode_mid_side_stereo(cpe, gb, ms_present);
1464 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1466 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1469 if (common_window) {
1471 apply_mid_side_stereo(ac, cpe);
1472 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1473 apply_prediction(ac, &cpe->ch[0]);
1474 apply_prediction(ac, &cpe->ch[1]);
1478 apply_intensity_stereo(ac, cpe, ms_present);
1482 static const float cce_scale[] = {
1483 1.09050773266525765921, //2^(1/8)
1484 1.18920711500272106672, //2^(1/4)
1490 * Decode coupling_channel_element; reference: table 4.8.
1492 * @return Returns error status. 0 - OK, !0 - error
1494 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1500 SingleChannelElement *sce = &che->ch[0];
1501 ChannelCoupling *coup = &che->coup;
1503 coup->coupling_point = 2 * get_bits1(gb);
1504 coup->num_coupled = get_bits(gb, 3);
1505 for (c = 0; c <= coup->num_coupled; c++) {
1507 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1508 coup->id_select[c] = get_bits(gb, 4);
1509 if (coup->type[c] == TYPE_CPE) {
1510 coup->ch_select[c] = get_bits(gb, 2);
1511 if (coup->ch_select[c] == 3)
1514 coup->ch_select[c] = 2;
1516 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1518 sign = get_bits(gb, 1);
1519 scale = cce_scale[get_bits(gb, 2)];
1521 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1524 for (c = 0; c < num_gain; c++) {
1528 float gain_cache = 1.;
1530 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1531 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1532 gain_cache = powf(scale, -gain);
1534 if (coup->coupling_point == AFTER_IMDCT) {
1535 coup->gain[c][0] = gain_cache;
1537 for (g = 0; g < sce->ics.num_window_groups; g++) {
1538 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1539 if (sce->band_type[idx] != ZERO_BT) {
1541 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1549 gain_cache = powf(scale, -t) * s;
1552 coup->gain[c][idx] = gain_cache;
1562 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1564 * @return Returns number of bytes consumed.
1566 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1570 int num_excl_chan = 0;
1573 for (i = 0; i < 7; i++)
1574 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1575 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1577 return num_excl_chan / 7;
1581 * Decode dynamic range information; reference: table 4.52.
1583 * @param cnt length of TYPE_FIL syntactic element in bytes
1585 * @return Returns number of bytes consumed.
1587 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1588 GetBitContext *gb, int cnt)
1591 int drc_num_bands = 1;
1594 /* pce_tag_present? */
1595 if (get_bits1(gb)) {
1596 che_drc->pce_instance_tag = get_bits(gb, 4);
1597 skip_bits(gb, 4); // tag_reserved_bits
1601 /* excluded_chns_present? */
1602 if (get_bits1(gb)) {
1603 n += decode_drc_channel_exclusions(che_drc, gb);
1606 /* drc_bands_present? */
1607 if (get_bits1(gb)) {
1608 che_drc->band_incr = get_bits(gb, 4);
1609 che_drc->interpolation_scheme = get_bits(gb, 4);
1611 drc_num_bands += che_drc->band_incr;
1612 for (i = 0; i < drc_num_bands; i++) {
1613 che_drc->band_top[i] = get_bits(gb, 8);
1618 /* prog_ref_level_present? */
1619 if (get_bits1(gb)) {
1620 che_drc->prog_ref_level = get_bits(gb, 7);
1621 skip_bits1(gb); // prog_ref_level_reserved_bits
1625 for (i = 0; i < drc_num_bands; i++) {
1626 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1627 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1635 * Decode extension data (incomplete); reference: table 4.51.
1637 * @param cnt length of TYPE_FIL syntactic element in bytes
1639 * @return Returns number of bytes consumed
1641 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1642 ChannelElement *che, enum RawDataBlockType elem_type)
1646 switch (get_bits(gb, 4)) { // extension type
1647 case EXT_SBR_DATA_CRC:
1651 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1653 } else if (!ac->m4ac.sbr) {
1654 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1655 skip_bits_long(gb, 8 * cnt - 4);
1657 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1658 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1659 skip_bits_long(gb, 8 * cnt - 4);
1661 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1664 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1668 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1670 case EXT_DYNAMIC_RANGE:
1671 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1675 case EXT_DATA_ELEMENT:
1677 skip_bits_long(gb, 8 * cnt - 4);
1684 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1686 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1687 * @param coef spectral coefficients
1689 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1690 IndividualChannelStream *ics, int decode)
1692 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1694 int bottom, top, order, start, end, size, inc;
1695 float lpc[TNS_MAX_ORDER];
1696 float tmp[TNS_MAX_ORDER];
1698 for (w = 0; w < ics->num_windows; w++) {
1699 bottom = ics->num_swb;
1700 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1702 bottom = FFMAX(0, top - tns->length[w][filt]);
1703 order = tns->order[w][filt];
1708 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1710 start = ics->swb_offset[FFMIN(bottom, mmm)];
1711 end = ics->swb_offset[FFMIN( top, mmm)];
1712 if ((size = end - start) <= 0)
1714 if (tns->direction[w][filt]) {
1724 for (m = 0; m < size; m++, start += inc)
1725 for (i = 1; i <= FFMIN(m, order); i++)
1726 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1729 for (m = 0; m < size; m++, start += inc) {
1730 tmp[0] = coef[start];
1731 for (i = 1; i <= FFMIN(m, order); i++)
1732 coef[start] += tmp[i] * lpc[i - 1];
1733 for (i = order; i > 0; i--)
1734 tmp[i] = tmp[i - 1];
1742 * Apply windowing and MDCT to obtain the spectral
1743 * coefficient from the predicted sample by LTP.
1745 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1746 float *in, IndividualChannelStream *ics)
1748 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1749 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1750 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1751 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1753 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1754 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1756 memset(in, 0, 448 * sizeof(float));
1757 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1758 memcpy(in + 576, in + 576, 448 * sizeof(float));
1760 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1761 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1763 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1764 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1765 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1767 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1771 * Apply the long term prediction
1773 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1775 const LongTermPrediction *ltp = &sce->ics.ltp;
1776 const uint16_t *offsets = sce->ics.swb_offset;
1779 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1780 float *predTime = sce->ret;
1781 float *predFreq = ac->buf_mdct;
1782 int16_t num_samples = 2048;
1784 if (ltp->lag < 1024)
1785 num_samples = ltp->lag + 1024;
1786 for (i = 0; i < num_samples; i++)
1787 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1788 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1790 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1792 if (sce->tns.present)
1793 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1795 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1797 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1798 sce->coeffs[i] += predFreq[i];
1803 * Update the LTP buffer for next frame
1805 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1807 IndividualChannelStream *ics = &sce->ics;
1808 float *saved = sce->saved;
1809 float *saved_ltp = sce->coeffs;
1810 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1811 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1814 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1815 memcpy(saved_ltp, saved, 512 * sizeof(float));
1816 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1817 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1818 for (i = 0; i < 64; i++)
1819 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1820 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1821 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1822 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1823 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1824 for (i = 0; i < 64; i++)
1825 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1826 } else { // LONG_STOP or ONLY_LONG
1827 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1828 for (i = 0; i < 512; i++)
1829 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1832 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1833 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1834 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1838 * Conduct IMDCT and windowing.
1840 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1842 IndividualChannelStream *ics = &sce->ics;
1843 float *in = sce->coeffs;
1844 float *out = sce->ret;
1845 float *saved = sce->saved;
1846 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1847 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1848 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1849 float *buf = ac->buf_mdct;
1850 float *temp = ac->temp;
1854 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1855 for (i = 0; i < 1024; i += 128)
1856 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1858 ac->mdct.imdct_half(&ac->mdct, buf, in);
1860 /* window overlapping
1861 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1862 * and long to short transitions are considered to be short to short
1863 * transitions. This leaves just two cases (long to long and short to short)
1864 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1866 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1867 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1868 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1870 memcpy( out, saved, 448 * sizeof(float));
1872 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1873 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1874 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1875 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1876 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1877 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1878 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1880 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1881 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1886 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1887 memcpy( saved, temp + 64, 64 * sizeof(float));
1888 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1889 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1890 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1891 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1892 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1893 memcpy( saved, buf + 512, 448 * sizeof(float));
1894 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1895 } else { // LONG_STOP or ONLY_LONG
1896 memcpy( saved, buf + 512, 512 * sizeof(float));
1901 * Apply dependent channel coupling (applied before IMDCT).
1903 * @param index index into coupling gain array
1905 static void apply_dependent_coupling(AACContext *ac,
1906 SingleChannelElement *target,
1907 ChannelElement *cce, int index)
1909 IndividualChannelStream *ics = &cce->ch[0].ics;
1910 const uint16_t *offsets = ics->swb_offset;
1911 float *dest = target->coeffs;
1912 const float *src = cce->ch[0].coeffs;
1913 int g, i, group, k, idx = 0;
1914 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1915 av_log(ac->avctx, AV_LOG_ERROR,
1916 "Dependent coupling is not supported together with LTP\n");
1919 for (g = 0; g < ics->num_window_groups; g++) {
1920 for (i = 0; i < ics->max_sfb; i++, idx++) {
1921 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1922 const float gain = cce->coup.gain[index][idx];
1923 for (group = 0; group < ics->group_len[g]; group++) {
1924 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1926 dest[group * 128 + k] += gain * src[group * 128 + k];
1931 dest += ics->group_len[g] * 128;
1932 src += ics->group_len[g] * 128;
1937 * Apply independent channel coupling (applied after IMDCT).
1939 * @param index index into coupling gain array
1941 static void apply_independent_coupling(AACContext *ac,
1942 SingleChannelElement *target,
1943 ChannelElement *cce, int index)
1946 const float gain = cce->coup.gain[index][0];
1947 const float *src = cce->ch[0].ret;
1948 float *dest = target->ret;
1949 const int len = 1024 << (ac->m4ac.sbr == 1);
1951 for (i = 0; i < len; i++)
1952 dest[i] += gain * src[i];
1956 * channel coupling transformation interface
1958 * @param apply_coupling_method pointer to (in)dependent coupling function
1960 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1961 enum RawDataBlockType type, int elem_id,
1962 enum CouplingPoint coupling_point,
1963 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1967 for (i = 0; i < MAX_ELEM_ID; i++) {
1968 ChannelElement *cce = ac->che[TYPE_CCE][i];
1971 if (cce && cce->coup.coupling_point == coupling_point) {
1972 ChannelCoupling *coup = &cce->coup;
1974 for (c = 0; c <= coup->num_coupled; c++) {
1975 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1976 if (coup->ch_select[c] != 1) {
1977 apply_coupling_method(ac, &cc->ch[0], cce, index);
1978 if (coup->ch_select[c] != 0)
1981 if (coup->ch_select[c] != 2)
1982 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1984 index += 1 + (coup->ch_select[c] == 3);
1991 * Convert spectral data to float samples, applying all supported tools as appropriate.
1993 static void spectral_to_sample(AACContext *ac)
1996 for (type = 3; type >= 0; type--) {
1997 for (i = 0; i < MAX_ELEM_ID; i++) {
1998 ChannelElement *che = ac->che[type][i];
2000 if (type <= TYPE_CPE)
2001 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2002 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2003 if (che->ch[0].ics.predictor_present) {
2004 if (che->ch[0].ics.ltp.present)
2005 apply_ltp(ac, &che->ch[0]);
2006 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2007 apply_ltp(ac, &che->ch[1]);
2010 if (che->ch[0].tns.present)
2011 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2012 if (che->ch[1].tns.present)
2013 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2014 if (type <= TYPE_CPE)
2015 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2016 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2017 imdct_and_windowing(ac, &che->ch[0]);
2018 if (ac->m4ac.object_type == AOT_AAC_LTP)
2019 update_ltp(ac, &che->ch[0]);
2020 if (type == TYPE_CPE) {
2021 imdct_and_windowing(ac, &che->ch[1]);
2022 if (ac->m4ac.object_type == AOT_AAC_LTP)
2023 update_ltp(ac, &che->ch[1]);
2025 if (ac->m4ac.sbr > 0) {
2026 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2029 if (type <= TYPE_CCE)
2030 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2036 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2039 AACADTSHeaderInfo hdr_info;
2041 size = ff_aac_parse_header(gb, &hdr_info);
2043 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2044 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2045 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2046 ac->m4ac.chan_config = hdr_info.chan_config;
2047 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2049 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2051 } else if (ac->output_configured != OC_LOCKED) {
2052 ac->output_configured = OC_NONE;
2054 if (ac->output_configured != OC_LOCKED) {
2058 ac->m4ac.sample_rate = hdr_info.sample_rate;
2059 ac->m4ac.sampling_index = hdr_info.sampling_index;
2060 ac->m4ac.object_type = hdr_info.object_type;
2061 if (!ac->avctx->sample_rate)
2062 ac->avctx->sample_rate = hdr_info.sample_rate;
2063 if (hdr_info.num_aac_frames == 1) {
2064 if (!hdr_info.crc_absent)
2067 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2074 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2075 int *data_size, GetBitContext *gb)
2077 AACContext *ac = avctx->priv_data;
2078 ChannelElement *che = NULL, *che_prev = NULL;
2079 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2080 int err, elem_id, data_size_tmp;
2081 int samples = 0, multiplier;
2083 if (show_bits(gb, 12) == 0xfff) {
2084 if (parse_adts_frame_header(ac, gb) < 0) {
2085 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2088 if (ac->m4ac.sampling_index > 12) {
2089 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2094 ac->tags_mapped = 0;
2096 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2097 elem_id = get_bits(gb, 4);
2099 if (elem_type < TYPE_DSE) {
2100 if (!(che=get_che(ac, elem_type, elem_id))) {
2101 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2102 elem_type, elem_id);
2108 switch (elem_type) {
2111 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2115 err = decode_cpe(ac, gb, che);
2119 err = decode_cce(ac, gb, che);
2123 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2127 err = skip_data_stream_element(ac, gb);
2131 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2132 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2133 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2135 if (ac->output_configured > OC_TRIAL_PCE)
2136 av_log(avctx, AV_LOG_ERROR,
2137 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2139 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2145 elem_id += get_bits(gb, 8) - 1;
2146 if (get_bits_left(gb) < 8 * elem_id) {
2147 av_log(avctx, AV_LOG_ERROR, overread_err);
2151 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2152 err = 0; /* FIXME */
2156 err = -1; /* should not happen, but keeps compiler happy */
2161 elem_type_prev = elem_type;
2166 if (get_bits_left(gb) < 3) {
2167 av_log(avctx, AV_LOG_ERROR, overread_err);
2172 spectral_to_sample(ac);
2174 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2175 samples <<= multiplier;
2176 if (ac->output_configured < OC_LOCKED) {
2177 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2178 avctx->frame_size = samples;
2181 data_size_tmp = samples * avctx->channels *
2182 av_get_bytes_per_sample(avctx->sample_fmt);
2183 if (*data_size < data_size_tmp) {
2184 av_log(avctx, AV_LOG_ERROR,
2185 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2186 *data_size, data_size_tmp);
2189 *data_size = data_size_tmp;
2192 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2193 ac->fmt_conv.float_interleave(data, (const float **)ac->output_data,
2194 samples, avctx->channels);
2196 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data,
2197 samples, avctx->channels);
2200 if (ac->output_configured)
2201 ac->output_configured = OC_LOCKED;
2206 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2207 int *data_size, AVPacket *avpkt)
2209 const uint8_t *buf = avpkt->data;
2210 int buf_size = avpkt->size;
2216 init_get_bits(&gb, buf, buf_size * 8);
2218 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2221 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2222 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2223 if (buf[buf_offset])
2226 return buf_size > buf_offset ? buf_consumed : buf_size;
2229 static av_cold int aac_decode_close(AVCodecContext *avctx)
2231 AACContext *ac = avctx->priv_data;
2234 for (i = 0; i < MAX_ELEM_ID; i++) {
2235 for (type = 0; type < 4; type++) {
2236 if (ac->che[type][i])
2237 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2238 av_freep(&ac->che[type][i]);
2242 ff_mdct_end(&ac->mdct);
2243 ff_mdct_end(&ac->mdct_small);
2244 ff_mdct_end(&ac->mdct_ltp);
2249 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2251 struct LATMContext {
2252 AACContext aac_ctx; ///< containing AACContext
2253 int initialized; ///< initilized after a valid extradata was seen
2256 int audio_mux_version_A; ///< LATM syntax version
2257 int frame_length_type; ///< 0/1 variable/fixed frame length
2258 int frame_length; ///< frame length for fixed frame length
2261 static inline uint32_t latm_get_value(GetBitContext *b)
2263 int length = get_bits(b, 2);
2265 return get_bits_long(b, (length+1)*8);
2268 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2271 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2272 MPEG4AudioConfig m4ac;
2273 int config_start_bit = get_bits_count(gb);
2274 int bits_consumed, esize;
2276 if (config_start_bit % 8) {
2277 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2278 "config not byte aligned.\n", 1);
2279 return AVERROR_INVALIDDATA;
2282 decode_audio_specific_config(NULL, avctx, &m4ac,
2283 gb->buffer + (config_start_bit / 8),
2284 get_bits_left(gb) / 8);
2286 if (bits_consumed < 0)
2287 return AVERROR_INVALIDDATA;
2289 esize = (bits_consumed+7) / 8;
2291 if (avctx->extradata_size <= esize) {
2292 av_free(avctx->extradata);
2293 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2294 if (!avctx->extradata)
2295 return AVERROR(ENOMEM);
2298 avctx->extradata_size = esize;
2299 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2300 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2302 skip_bits_long(gb, bits_consumed);
2305 return bits_consumed;
2308 static int read_stream_mux_config(struct LATMContext *latmctx,
2311 int ret, audio_mux_version = get_bits(gb, 1);
2313 latmctx->audio_mux_version_A = 0;
2314 if (audio_mux_version)
2315 latmctx->audio_mux_version_A = get_bits(gb, 1);
2317 if (!latmctx->audio_mux_version_A) {
2319 if (audio_mux_version)
2320 latm_get_value(gb); // taraFullness
2322 skip_bits(gb, 1); // allStreamSameTimeFraming
2323 skip_bits(gb, 6); // numSubFrames
2325 if (get_bits(gb, 4)) { // numPrograms
2326 av_log_missing_feature(latmctx->aac_ctx.avctx,
2327 "multiple programs are not supported\n", 1);
2328 return AVERROR_PATCHWELCOME;
2331 // for each program (which there is only on in DVB)
2333 // for each layer (which there is only on in DVB)
2334 if (get_bits(gb, 3)) { // numLayer
2335 av_log_missing_feature(latmctx->aac_ctx.avctx,
2336 "multiple layers are not supported\n", 1);
2337 return AVERROR_PATCHWELCOME;
2340 // for all but first stream: use_same_config = get_bits(gb, 1);
2341 if (!audio_mux_version) {
2342 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2345 int ascLen = latm_get_value(gb);
2346 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2349 skip_bits_long(gb, ascLen);
2352 latmctx->frame_length_type = get_bits(gb, 3);
2353 switch (latmctx->frame_length_type) {
2355 skip_bits(gb, 8); // latmBufferFullness
2358 latmctx->frame_length = get_bits(gb, 9);
2363 skip_bits(gb, 6); // CELP frame length table index
2367 skip_bits(gb, 1); // HVXC frame length table index
2371 if (get_bits(gb, 1)) { // other data
2372 if (audio_mux_version) {
2373 latm_get_value(gb); // other_data_bits
2377 esc = get_bits(gb, 1);
2383 if (get_bits(gb, 1)) // crc present
2384 skip_bits(gb, 8); // config_crc
2390 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2394 if (ctx->frame_length_type == 0) {
2395 int mux_slot_length = 0;
2397 tmp = get_bits(gb, 8);
2398 mux_slot_length += tmp;
2399 } while (tmp == 255);
2400 return mux_slot_length;
2401 } else if (ctx->frame_length_type == 1) {
2402 return ctx->frame_length;
2403 } else if (ctx->frame_length_type == 3 ||
2404 ctx->frame_length_type == 5 ||
2405 ctx->frame_length_type == 7) {
2406 skip_bits(gb, 2); // mux_slot_length_coded
2411 static int read_audio_mux_element(struct LATMContext *latmctx,
2415 uint8_t use_same_mux = get_bits(gb, 1);
2416 if (!use_same_mux) {
2417 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2419 } else if (!latmctx->aac_ctx.avctx->extradata) {
2420 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2421 "no decoder config found\n");
2422 return AVERROR(EAGAIN);
2424 if (latmctx->audio_mux_version_A == 0) {
2425 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2426 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2427 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2428 return AVERROR_INVALIDDATA;
2429 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2430 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2431 "frame length mismatch %d << %d\n",
2432 mux_slot_length_bytes * 8, get_bits_left(gb));
2433 return AVERROR_INVALIDDATA;
2440 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2443 struct LATMContext *latmctx = avctx->priv_data;
2447 if (avpkt->size == 0)
2450 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2452 // check for LOAS sync word
2453 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2454 return AVERROR_INVALIDDATA;
2456 muxlength = get_bits(&gb, 13) + 3;
2457 // not enough data, the parser should have sorted this
2458 if (muxlength > avpkt->size)
2459 return AVERROR_INVALIDDATA;
2461 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2464 if (!latmctx->initialized) {
2465 if (!avctx->extradata) {
2469 aac_decode_close(avctx);
2470 if ((err = aac_decode_init(avctx)) < 0)
2472 latmctx->initialized = 1;
2476 if (show_bits(&gb, 12) == 0xfff) {
2477 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2478 "ADTS header detected, probably as result of configuration "
2480 return AVERROR_INVALIDDATA;
2483 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2489 av_cold static int latm_decode_init(AVCodecContext *avctx)
2491 struct LATMContext *latmctx = avctx->priv_data;
2494 ret = aac_decode_init(avctx);
2496 if (avctx->extradata_size > 0) {
2497 latmctx->initialized = !ret;
2499 latmctx->initialized = 0;
2506 AVCodec ff_aac_decoder = {
2515 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2516 .sample_fmts = (const enum AVSampleFormat[]) {
2517 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2519 .channel_layouts = aac_channel_layout,
2523 Note: This decoder filter is intended to decode LATM streams transferred
2524 in MPEG transport streams which only contain one program.
2525 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2527 AVCodec ff_aac_latm_decoder = {
2529 .type = AVMEDIA_TYPE_AUDIO,
2530 .id = CODEC_ID_AAC_LATM,
2531 .priv_data_size = sizeof(struct LATMContext),
2532 .init = latm_decode_init,
2533 .close = aac_decode_close,
2534 .decode = latm_decode_frame,
2535 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2536 .sample_fmts = (const enum AVSampleFormat[]) {
2537 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2539 .channel_layouts = aac_channel_layout,