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"
101 #include "libavutil/intfloat.h"
109 # include "arm/aac.h"
112 static VLC vlc_scalefactors;
113 static VLC vlc_spectral[11];
115 static const char overread_err[] = "Input buffer exhausted before END element found\n";
117 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
119 // For PCE based channel configurations map the channels solely based on tags.
120 if (!ac->m4ac.chan_config) {
121 return ac->tag_che_map[type][elem_id];
123 // For indexed channel configurations map the channels solely based on position.
124 switch (ac->m4ac.chan_config) {
126 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
128 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
131 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
132 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
133 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
134 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
136 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
139 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
141 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
144 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
146 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
150 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
152 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
153 } else if (ac->m4ac.chan_config == 2) {
157 if (!ac->tags_mapped && type == TYPE_SCE) {
159 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
166 static int count_channels(enum ChannelPosition che_pos[4][MAX_ELEM_ID])
168 int i, type, sum = 0;
169 for (i = 0; i < MAX_ELEM_ID; i++) {
170 for (type = 0; type < 4; type++) {
171 sum += (1 + (type == TYPE_CPE)) *
172 (che_pos[type][i] != AAC_CHANNEL_OFF &&
173 che_pos[type][i] != AAC_CHANNEL_CC);
180 * Check for the channel element in the current channel position configuration.
181 * If it exists, make sure the appropriate element is allocated and map the
182 * channel order to match the internal FFmpeg channel layout.
184 * @param che_pos current channel position configuration
185 * @param type channel element type
186 * @param id channel element id
187 * @param channels count of the number of channels in the configuration
189 * @return Returns error status. 0 - OK, !0 - error
191 static av_cold int che_configure(AACContext *ac,
192 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
193 int type, int id, int *channels)
195 if (che_pos[type][id]) {
196 if (!ac->che[type][id]) {
197 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
198 return AVERROR(ENOMEM);
199 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
201 if (type != TYPE_CCE) {
202 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
203 if (type == TYPE_CPE ||
204 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
205 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
209 if (ac->che[type][id])
210 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
211 av_freep(&ac->che[type][id]);
217 * Configure output channel order based on the current program configuration element.
219 * @param che_pos current channel position configuration
220 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
222 * @return Returns error status. 0 - OK, !0 - error
224 static av_cold int output_configure(AACContext *ac,
225 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
226 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
227 int channel_config, enum OCStatus oc_type)
229 AVCodecContext *avctx = ac->avctx;
230 int i, type, channels = 0, ret;
232 if (new_che_pos != che_pos)
233 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
235 if (channel_config) {
236 for (i = 0; i < tags_per_config[channel_config]; i++) {
237 if ((ret = che_configure(ac, che_pos,
238 aac_channel_layout_map[channel_config - 1][i][0],
239 aac_channel_layout_map[channel_config - 1][i][1],
244 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
246 avctx->channel_layout = aac_channel_layout[channel_config - 1];
248 /* Allocate or free elements depending on if they are in the
249 * current program configuration.
251 * Set up default 1:1 output mapping.
253 * For a 5.1 stream the output order will be:
254 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
257 for (i = 0; i < MAX_ELEM_ID; i++) {
258 for (type = 0; type < 4; type++) {
259 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
264 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
267 avctx->channels = channels;
269 ac->output_configured = oc_type;
274 static void flush(AVCodecContext *avctx)
276 AACContext *ac= avctx->priv_data;
279 for (type = 3; type >= 0; type--) {
280 for (i = 0; i < MAX_ELEM_ID; i++) {
281 ChannelElement *che = ac->che[type][i];
283 for (j = 0; j <= 1; j++) {
284 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
292 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
294 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
295 * @param sce_map mono (Single Channel Element) map
296 * @param type speaker type/position for these channels
298 static void decode_channel_map(enum ChannelPosition *cpe_map,
299 enum ChannelPosition *sce_map,
300 enum ChannelPosition type,
301 GetBitContext *gb, int n)
304 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
305 map[get_bits(gb, 4)] = type;
310 * Decode program configuration element; reference: table 4.2.
312 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
314 * @return Returns error status. 0 - OK, !0 - error
316 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
317 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
320 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
323 skip_bits(gb, 2); // object_type
325 sampling_index = get_bits(gb, 4);
326 if (m4ac->sampling_index != sampling_index)
327 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");
329 num_front = get_bits(gb, 4);
330 num_side = get_bits(gb, 4);
331 num_back = get_bits(gb, 4);
332 num_lfe = get_bits(gb, 2);
333 num_assoc_data = get_bits(gb, 3);
334 num_cc = get_bits(gb, 4);
337 skip_bits(gb, 4); // mono_mixdown_tag
339 skip_bits(gb, 4); // stereo_mixdown_tag
342 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
344 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
345 av_log(avctx, AV_LOG_ERROR, overread_err);
348 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
349 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
350 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
351 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
353 skip_bits_long(gb, 4 * num_assoc_data);
355 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
359 /* comment field, first byte is length */
360 comment_len = get_bits(gb, 8) * 8;
361 if (get_bits_left(gb) < comment_len) {
362 av_log(avctx, AV_LOG_ERROR, overread_err);
365 skip_bits_long(gb, comment_len);
370 * Set up channel positions based on a default channel configuration
371 * as specified in table 1.17.
373 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
375 * @return Returns error status. 0 - OK, !0 - error
377 static av_cold int set_default_channel_config(AVCodecContext *avctx,
378 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
381 if (channel_config < 1 || channel_config > 7) {
382 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
387 /* default channel configurations:
389 * 1ch : front center (mono)
390 * 2ch : L + R (stereo)
391 * 3ch : front center + L + R
392 * 4ch : front center + L + R + back center
393 * 5ch : front center + L + R + back stereo
394 * 6ch : front center + L + R + back stereo + LFE
395 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
398 if (channel_config != 2)
399 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
400 if (channel_config > 1)
401 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
402 if (channel_config == 4)
403 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
404 if (channel_config > 4)
405 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
406 = AAC_CHANNEL_BACK; // back stereo
407 if (channel_config > 5)
408 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
409 if (channel_config == 7)
410 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
416 * Decode GA "General Audio" specific configuration; reference: table 4.1.
418 * @param ac pointer to AACContext, may be null
419 * @param avctx pointer to AVCCodecContext, used for logging
421 * @return Returns error status. 0 - OK, !0 - error
423 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
425 MPEG4AudioConfig *m4ac,
428 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
429 int extension_flag, ret;
431 if (get_bits1(gb)) { // frameLengthFlag
432 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
436 if (get_bits1(gb)) // dependsOnCoreCoder
437 skip_bits(gb, 14); // coreCoderDelay
438 extension_flag = get_bits1(gb);
440 if (m4ac->object_type == AOT_AAC_SCALABLE ||
441 m4ac->object_type == AOT_ER_AAC_SCALABLE)
442 skip_bits(gb, 3); // layerNr
444 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
445 if (channel_config == 0) {
446 skip_bits(gb, 4); // element_instance_tag
447 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
450 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
454 if (count_channels(new_che_pos) > 1) {
456 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
459 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
462 if (extension_flag) {
463 switch (m4ac->object_type) {
465 skip_bits(gb, 5); // numOfSubFrame
466 skip_bits(gb, 11); // layer_length
470 case AOT_ER_AAC_SCALABLE:
472 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
473 * aacScalefactorDataResilienceFlag
474 * aacSpectralDataResilienceFlag
478 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
484 * Decode audio specific configuration; reference: table 1.13.
486 * @param ac pointer to AACContext, may be null
487 * @param avctx pointer to AVCCodecContext, used for logging
488 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
489 * @param data pointer to buffer holding an audio specific config
490 * @param bit_size size of audio specific config or data in bits
491 * @param sync_extension look for an appended sync extension
493 * @return Returns error status or number of consumed bits. <0 - error
495 static int decode_audio_specific_config(AACContext *ac,
496 AVCodecContext *avctx,
497 MPEG4AudioConfig *m4ac,
498 const uint8_t *data, int bit_size,
504 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
505 for (i = 0; i < avctx->extradata_size; i++)
506 av_dlog(avctx, "%02x ", avctx->extradata[i]);
507 av_dlog(avctx, "\n");
509 init_get_bits(&gb, data, bit_size);
511 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
513 if (m4ac->sampling_index > 12) {
514 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
518 skip_bits_long(&gb, i);
520 switch (m4ac->object_type) {
524 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
528 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
529 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
533 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
534 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
535 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
537 return get_bits_count(&gb);
541 * linear congruential pseudorandom number generator
543 * @param previous_val pointer to the current state of the generator
545 * @return Returns a 32-bit pseudorandom integer
547 static av_always_inline int lcg_random(int previous_val)
549 return previous_val * 1664525 + 1013904223;
552 static av_always_inline void reset_predict_state(PredictorState *ps)
562 static void reset_all_predictors(PredictorState *ps)
565 for (i = 0; i < MAX_PREDICTORS; i++)
566 reset_predict_state(&ps[i]);
569 static int sample_rate_idx (int rate)
571 if (92017 <= rate) return 0;
572 else if (75132 <= rate) return 1;
573 else if (55426 <= rate) return 2;
574 else if (46009 <= rate) return 3;
575 else if (37566 <= rate) return 4;
576 else if (27713 <= rate) return 5;
577 else if (23004 <= rate) return 6;
578 else if (18783 <= rate) return 7;
579 else if (13856 <= rate) return 8;
580 else if (11502 <= rate) return 9;
581 else if (9391 <= rate) return 10;
585 static void reset_predictor_group(PredictorState *ps, int group_num)
588 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
589 reset_predict_state(&ps[i]);
592 #define AAC_INIT_VLC_STATIC(num, size) \
593 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
594 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
595 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
598 static av_cold int aac_decode_init(AVCodecContext *avctx)
600 AACContext *ac = avctx->priv_data;
601 float output_scale_factor;
604 ac->m4ac.sample_rate = avctx->sample_rate;
606 if (avctx->extradata_size > 0) {
607 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
609 avctx->extradata_size*8, 1) < 0)
613 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
615 sr = sample_rate_idx(avctx->sample_rate);
616 ac->m4ac.sampling_index = sr;
617 ac->m4ac.channels = avctx->channels;
621 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
622 if (ff_mpeg4audio_channels[i] == avctx->channels)
624 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
627 ac->m4ac.chan_config = i;
629 if (ac->m4ac.chan_config) {
630 int ret = set_default_channel_config(avctx, new_che_pos, ac->m4ac.chan_config);
632 output_configure(ac, ac->che_pos, new_che_pos, ac->m4ac.chan_config, OC_GLOBAL_HDR);
633 else if (avctx->err_recognition & AV_EF_EXPLODE)
634 return AVERROR_INVALIDDATA;
638 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
639 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
640 output_scale_factor = 1.0 / 32768.0;
642 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
643 output_scale_factor = 1.0;
646 AAC_INIT_VLC_STATIC( 0, 304);
647 AAC_INIT_VLC_STATIC( 1, 270);
648 AAC_INIT_VLC_STATIC( 2, 550);
649 AAC_INIT_VLC_STATIC( 3, 300);
650 AAC_INIT_VLC_STATIC( 4, 328);
651 AAC_INIT_VLC_STATIC( 5, 294);
652 AAC_INIT_VLC_STATIC( 6, 306);
653 AAC_INIT_VLC_STATIC( 7, 268);
654 AAC_INIT_VLC_STATIC( 8, 510);
655 AAC_INIT_VLC_STATIC( 9, 366);
656 AAC_INIT_VLC_STATIC(10, 462);
660 dsputil_init(&ac->dsp, avctx);
661 ff_fmt_convert_init(&ac->fmt_conv, avctx);
663 ac->random_state = 0x1f2e3d4c;
667 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
668 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
669 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
672 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
673 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
674 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
675 // window initialization
676 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
677 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
678 ff_init_ff_sine_windows(10);
679 ff_init_ff_sine_windows( 7);
683 avcodec_get_frame_defaults(&ac->frame);
684 avctx->coded_frame = &ac->frame;
690 * Skip data_stream_element; reference: table 4.10.
692 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
694 int byte_align = get_bits1(gb);
695 int count = get_bits(gb, 8);
697 count += get_bits(gb, 8);
701 if (get_bits_left(gb) < 8 * count) {
702 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
705 skip_bits_long(gb, 8 * count);
709 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
714 ics->predictor_reset_group = get_bits(gb, 5);
715 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
716 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
720 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
721 ics->prediction_used[sfb] = get_bits1(gb);
727 * Decode Long Term Prediction data; reference: table 4.xx.
729 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
730 GetBitContext *gb, uint8_t max_sfb)
734 ltp->lag = get_bits(gb, 11);
735 ltp->coef = ltp_coef[get_bits(gb, 3)];
736 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
737 ltp->used[sfb] = get_bits1(gb);
741 * Decode Individual Channel Stream info; reference: table 4.6.
743 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
747 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
748 return AVERROR_INVALIDDATA;
750 ics->window_sequence[1] = ics->window_sequence[0];
751 ics->window_sequence[0] = get_bits(gb, 2);
752 ics->use_kb_window[1] = ics->use_kb_window[0];
753 ics->use_kb_window[0] = get_bits1(gb);
754 ics->num_window_groups = 1;
755 ics->group_len[0] = 1;
756 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
758 ics->max_sfb = get_bits(gb, 4);
759 for (i = 0; i < 7; i++) {
761 ics->group_len[ics->num_window_groups - 1]++;
763 ics->num_window_groups++;
764 ics->group_len[ics->num_window_groups - 1] = 1;
767 ics->num_windows = 8;
768 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
769 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
770 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
771 ics->predictor_present = 0;
773 ics->max_sfb = get_bits(gb, 6);
774 ics->num_windows = 1;
775 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
776 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
777 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
778 ics->predictor_present = get_bits1(gb);
779 ics->predictor_reset_group = 0;
780 if (ics->predictor_present) {
781 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
782 if (decode_prediction(ac, ics, gb)) {
783 return AVERROR_INVALIDDATA;
785 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
786 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
787 return AVERROR_INVALIDDATA;
789 if ((ics->ltp.present = get_bits(gb, 1)))
790 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
795 if (ics->max_sfb > ics->num_swb) {
796 av_log(ac->avctx, AV_LOG_ERROR,
797 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
798 ics->max_sfb, ics->num_swb);
799 return AVERROR_INVALIDDATA;
806 * Decode band types (section_data payload); reference: table 4.46.
808 * @param band_type array of the used band type
809 * @param band_type_run_end array of the last scalefactor band of a band type run
811 * @return Returns error status. 0 - OK, !0 - error
813 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
814 int band_type_run_end[120], GetBitContext *gb,
815 IndividualChannelStream *ics)
818 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
819 for (g = 0; g < ics->num_window_groups; g++) {
821 while (k < ics->max_sfb) {
822 uint8_t sect_end = k;
824 int sect_band_type = get_bits(gb, 4);
825 if (sect_band_type == 12) {
826 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
830 sect_len_incr = get_bits(gb, bits);
831 sect_end += sect_len_incr;
832 if (get_bits_left(gb) < 0) {
833 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
836 if (sect_end > ics->max_sfb) {
837 av_log(ac->avctx, AV_LOG_ERROR,
838 "Number of bands (%d) exceeds limit (%d).\n",
839 sect_end, ics->max_sfb);
842 } while (sect_len_incr == (1 << bits) - 1);
843 for (; k < sect_end; k++) {
844 band_type [idx] = sect_band_type;
845 band_type_run_end[idx++] = sect_end;
853 * Decode scalefactors; reference: table 4.47.
855 * @param global_gain first scalefactor value as scalefactors are differentially coded
856 * @param band_type array of the used band type
857 * @param band_type_run_end array of the last scalefactor band of a band type run
858 * @param sf array of scalefactors or intensity stereo positions
860 * @return Returns error status. 0 - OK, !0 - error
862 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
863 unsigned int global_gain,
864 IndividualChannelStream *ics,
865 enum BandType band_type[120],
866 int band_type_run_end[120])
869 int offset[3] = { global_gain, global_gain - 90, 0 };
872 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
873 for (g = 0; g < ics->num_window_groups; g++) {
874 for (i = 0; i < ics->max_sfb;) {
875 int run_end = band_type_run_end[idx];
876 if (band_type[idx] == ZERO_BT) {
877 for (; i < run_end; i++, idx++)
879 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
880 for (; i < run_end; i++, idx++) {
881 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
882 clipped_offset = av_clip(offset[2], -155, 100);
883 if (offset[2] != clipped_offset) {
884 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
885 "position clipped (%d -> %d).\nIf you heard an "
886 "audible artifact, there may be a bug in the "
887 "decoder. ", offset[2], clipped_offset);
889 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
891 } else if (band_type[idx] == NOISE_BT) {
892 for (; i < run_end; i++, idx++) {
893 if (noise_flag-- > 0)
894 offset[1] += get_bits(gb, 9) - 256;
896 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
897 clipped_offset = av_clip(offset[1], -100, 155);
898 if (offset[1] != clipped_offset) {
899 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
900 "(%d -> %d).\nIf you heard an audible "
901 "artifact, there may be a bug in the decoder. ",
902 offset[1], clipped_offset);
904 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
907 for (; i < run_end; i++, idx++) {
908 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
909 if (offset[0] > 255U) {
910 av_log(ac->avctx, AV_LOG_ERROR,
911 "%s (%d) out of range.\n", sf_str[0], offset[0]);
914 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
923 * Decode pulse data; reference: table 4.7.
925 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
926 const uint16_t *swb_offset, int num_swb)
929 pulse->num_pulse = get_bits(gb, 2) + 1;
930 pulse_swb = get_bits(gb, 6);
931 if (pulse_swb >= num_swb)
933 pulse->pos[0] = swb_offset[pulse_swb];
934 pulse->pos[0] += get_bits(gb, 5);
935 if (pulse->pos[0] > 1023)
937 pulse->amp[0] = get_bits(gb, 4);
938 for (i = 1; i < pulse->num_pulse; i++) {
939 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
940 if (pulse->pos[i] > 1023)
942 pulse->amp[i] = get_bits(gb, 4);
948 * Decode Temporal Noise Shaping data; reference: table 4.48.
950 * @return Returns error status. 0 - OK, !0 - error
952 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
953 GetBitContext *gb, const IndividualChannelStream *ics)
955 int w, filt, i, coef_len, coef_res, coef_compress;
956 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
957 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
958 for (w = 0; w < ics->num_windows; w++) {
959 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
960 coef_res = get_bits1(gb);
962 for (filt = 0; filt < tns->n_filt[w]; filt++) {
964 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
966 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
967 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
968 tns->order[w][filt], tns_max_order);
969 tns->order[w][filt] = 0;
972 if (tns->order[w][filt]) {
973 tns->direction[w][filt] = get_bits1(gb);
974 coef_compress = get_bits1(gb);
975 coef_len = coef_res + 3 - coef_compress;
976 tmp2_idx = 2 * coef_compress + coef_res;
978 for (i = 0; i < tns->order[w][filt]; i++)
979 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
988 * Decode Mid/Side data; reference: table 4.54.
990 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
991 * [1] mask is decoded from bitstream; [2] mask is all 1s;
992 * [3] reserved for scalable AAC
994 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
998 if (ms_present == 1) {
999 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1000 cpe->ms_mask[idx] = get_bits1(gb);
1001 } else if (ms_present == 2) {
1002 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1007 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1011 *dst++ = v[idx & 15] * s;
1012 *dst++ = v[idx>>4 & 15] * s;
1018 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1022 *dst++ = v[idx & 3] * s;
1023 *dst++ = v[idx>>2 & 3] * s;
1024 *dst++ = v[idx>>4 & 3] * s;
1025 *dst++ = v[idx>>6 & 3] * s;
1031 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1032 unsigned sign, const float *scale)
1034 union av_intfloat32 s0, s1;
1036 s0.f = s1.f = *scale;
1037 s0.i ^= sign >> 1 << 31;
1040 *dst++ = v[idx & 15] * s0.f;
1041 *dst++ = v[idx>>4 & 15] * s1.f;
1048 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1049 unsigned sign, const float *scale)
1051 unsigned nz = idx >> 12;
1052 union av_intfloat32 s = { .f = *scale };
1053 union av_intfloat32 t;
1055 t.i = s.i ^ (sign & 1U<<31);
1056 *dst++ = v[idx & 3] * t.f;
1058 sign <<= nz & 1; nz >>= 1;
1059 t.i = s.i ^ (sign & 1U<<31);
1060 *dst++ = v[idx>>2 & 3] * t.f;
1062 sign <<= nz & 1; nz >>= 1;
1063 t.i = s.i ^ (sign & 1U<<31);
1064 *dst++ = v[idx>>4 & 3] * t.f;
1066 sign <<= nz & 1; nz >>= 1;
1067 t.i = s.i ^ (sign & 1U<<31);
1068 *dst++ = v[idx>>6 & 3] * t.f;
1075 * Decode spectral data; reference: table 4.50.
1076 * Dequantize and scale spectral data; reference: 4.6.3.3.
1078 * @param coef array of dequantized, scaled spectral data
1079 * @param sf array of scalefactors or intensity stereo positions
1080 * @param pulse_present set if pulses are present
1081 * @param pulse pointer to pulse data struct
1082 * @param band_type array of the used band type
1084 * @return Returns error status. 0 - OK, !0 - error
1086 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1087 GetBitContext *gb, const float sf[120],
1088 int pulse_present, const Pulse *pulse,
1089 const IndividualChannelStream *ics,
1090 enum BandType band_type[120])
1092 int i, k, g, idx = 0;
1093 const int c = 1024 / ics->num_windows;
1094 const uint16_t *offsets = ics->swb_offset;
1095 float *coef_base = coef;
1097 for (g = 0; g < ics->num_windows; g++)
1098 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1100 for (g = 0; g < ics->num_window_groups; g++) {
1101 unsigned g_len = ics->group_len[g];
1103 for (i = 0; i < ics->max_sfb; i++, idx++) {
1104 const unsigned cbt_m1 = band_type[idx] - 1;
1105 float *cfo = coef + offsets[i];
1106 int off_len = offsets[i + 1] - offsets[i];
1109 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1110 for (group = 0; group < g_len; group++, cfo+=128) {
1111 memset(cfo, 0, off_len * sizeof(float));
1113 } else if (cbt_m1 == NOISE_BT - 1) {
1114 for (group = 0; group < g_len; group++, cfo+=128) {
1118 for (k = 0; k < off_len; k++) {
1119 ac->random_state = lcg_random(ac->random_state);
1120 cfo[k] = ac->random_state;
1123 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1124 scale = sf[idx] / sqrtf(band_energy);
1125 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1128 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1129 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1130 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1131 OPEN_READER(re, gb);
1133 switch (cbt_m1 >> 1) {
1135 for (group = 0; group < g_len; group++, cfo+=128) {
1143 UPDATE_CACHE(re, gb);
1144 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1145 cb_idx = cb_vector_idx[code];
1146 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1152 for (group = 0; group < g_len; group++, cfo+=128) {
1162 UPDATE_CACHE(re, gb);
1163 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1164 cb_idx = cb_vector_idx[code];
1165 nnz = cb_idx >> 8 & 15;
1166 bits = nnz ? GET_CACHE(re, gb) : 0;
1167 LAST_SKIP_BITS(re, gb, nnz);
1168 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1174 for (group = 0; group < g_len; group++, cfo+=128) {
1182 UPDATE_CACHE(re, gb);
1183 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1184 cb_idx = cb_vector_idx[code];
1185 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1192 for (group = 0; group < g_len; group++, cfo+=128) {
1202 UPDATE_CACHE(re, gb);
1203 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1204 cb_idx = cb_vector_idx[code];
1205 nnz = cb_idx >> 8 & 15;
1206 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1207 LAST_SKIP_BITS(re, gb, nnz);
1208 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1214 for (group = 0; group < g_len; group++, cfo+=128) {
1216 uint32_t *icf = (uint32_t *) cf;
1226 UPDATE_CACHE(re, gb);
1227 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1235 cb_idx = cb_vector_idx[code];
1238 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1239 LAST_SKIP_BITS(re, gb, nnz);
1241 for (j = 0; j < 2; j++) {
1245 /* The total length of escape_sequence must be < 22 bits according
1246 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1247 UPDATE_CACHE(re, gb);
1248 b = GET_CACHE(re, gb);
1249 b = 31 - av_log2(~b);
1252 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1256 SKIP_BITS(re, gb, b + 1);
1258 n = (1 << b) + SHOW_UBITS(re, gb, b);
1259 LAST_SKIP_BITS(re, gb, b);
1260 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1263 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1264 *icf++ = (bits & 1U<<31) | v;
1271 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1275 CLOSE_READER(re, gb);
1281 if (pulse_present) {
1283 for (i = 0; i < pulse->num_pulse; i++) {
1284 float co = coef_base[ pulse->pos[i] ];
1285 while (offsets[idx + 1] <= pulse->pos[i])
1287 if (band_type[idx] != NOISE_BT && sf[idx]) {
1288 float ico = -pulse->amp[i];
1291 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1293 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1300 static av_always_inline float flt16_round(float pf)
1302 union av_intfloat32 tmp;
1304 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1308 static av_always_inline float flt16_even(float pf)
1310 union av_intfloat32 tmp;
1312 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1316 static av_always_inline float flt16_trunc(float pf)
1318 union av_intfloat32 pun;
1320 pun.i &= 0xFFFF0000U;
1324 static av_always_inline void predict(PredictorState *ps, float *coef,
1327 const float a = 0.953125; // 61.0 / 64
1328 const float alpha = 0.90625; // 29.0 / 32
1332 float r0 = ps->r0, r1 = ps->r1;
1333 float cor0 = ps->cor0, cor1 = ps->cor1;
1334 float var0 = ps->var0, var1 = ps->var1;
1336 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1337 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1339 pv = flt16_round(k1 * r0 + k2 * r1);
1346 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1347 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1348 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1349 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1351 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1352 ps->r0 = flt16_trunc(a * e0);
1356 * Apply AAC-Main style frequency domain prediction.
1358 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1362 if (!sce->ics.predictor_initialized) {
1363 reset_all_predictors(sce->predictor_state);
1364 sce->ics.predictor_initialized = 1;
1367 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1368 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1369 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1370 predict(&sce->predictor_state[k], &sce->coeffs[k],
1371 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1374 if (sce->ics.predictor_reset_group)
1375 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1377 reset_all_predictors(sce->predictor_state);
1381 * Decode an individual_channel_stream payload; reference: table 4.44.
1383 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1384 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1386 * @return Returns error status. 0 - OK, !0 - error
1388 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1389 GetBitContext *gb, int common_window, int scale_flag)
1392 TemporalNoiseShaping *tns = &sce->tns;
1393 IndividualChannelStream *ics = &sce->ics;
1394 float *out = sce->coeffs;
1395 int global_gain, pulse_present = 0;
1397 /* This assignment is to silence a GCC warning about the variable being used
1398 * uninitialized when in fact it always is.
1400 pulse.num_pulse = 0;
1402 global_gain = get_bits(gb, 8);
1404 if (!common_window && !scale_flag) {
1405 if (decode_ics_info(ac, ics, gb) < 0)
1406 return AVERROR_INVALIDDATA;
1409 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1411 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1416 if ((pulse_present = get_bits1(gb))) {
1417 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1418 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1421 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1422 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1426 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1428 if (get_bits1(gb)) {
1429 av_log_missing_feature(ac->avctx, "SSR", 1);
1434 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1437 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1438 apply_prediction(ac, sce);
1444 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1446 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1448 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1449 float *ch0 = cpe->ch[0].coeffs;
1450 float *ch1 = cpe->ch[1].coeffs;
1451 int g, i, group, idx = 0;
1452 const uint16_t *offsets = ics->swb_offset;
1453 for (g = 0; g < ics->num_window_groups; g++) {
1454 for (i = 0; i < ics->max_sfb; i++, idx++) {
1455 if (cpe->ms_mask[idx] &&
1456 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1457 for (group = 0; group < ics->group_len[g]; group++) {
1458 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1459 ch1 + group * 128 + offsets[i],
1460 offsets[i+1] - offsets[i]);
1464 ch0 += ics->group_len[g] * 128;
1465 ch1 += ics->group_len[g] * 128;
1470 * intensity stereo decoding; reference: 4.6.8.2.3
1472 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1473 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1474 * [3] reserved for scalable AAC
1476 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1478 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1479 SingleChannelElement *sce1 = &cpe->ch[1];
1480 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1481 const uint16_t *offsets = ics->swb_offset;
1482 int g, group, i, idx = 0;
1485 for (g = 0; g < ics->num_window_groups; g++) {
1486 for (i = 0; i < ics->max_sfb;) {
1487 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1488 const int bt_run_end = sce1->band_type_run_end[idx];
1489 for (; i < bt_run_end; i++, idx++) {
1490 c = -1 + 2 * (sce1->band_type[idx] - 14);
1492 c *= 1 - 2 * cpe->ms_mask[idx];
1493 scale = c * sce1->sf[idx];
1494 for (group = 0; group < ics->group_len[g]; group++)
1495 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1496 coef0 + group * 128 + offsets[i],
1498 offsets[i + 1] - offsets[i]);
1501 int bt_run_end = sce1->band_type_run_end[idx];
1502 idx += bt_run_end - i;
1506 coef0 += ics->group_len[g] * 128;
1507 coef1 += ics->group_len[g] * 128;
1512 * Decode a channel_pair_element; reference: table 4.4.
1514 * @return Returns error status. 0 - OK, !0 - error
1516 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1518 int i, ret, common_window, ms_present = 0;
1520 common_window = get_bits1(gb);
1521 if (common_window) {
1522 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1523 return AVERROR_INVALIDDATA;
1524 i = cpe->ch[1].ics.use_kb_window[0];
1525 cpe->ch[1].ics = cpe->ch[0].ics;
1526 cpe->ch[1].ics.use_kb_window[1] = i;
1527 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1528 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1529 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1530 ms_present = get_bits(gb, 2);
1531 if (ms_present == 3) {
1532 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1534 } else if (ms_present)
1535 decode_mid_side_stereo(cpe, gb, ms_present);
1537 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1539 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1542 if (common_window) {
1544 apply_mid_side_stereo(ac, cpe);
1545 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1546 apply_prediction(ac, &cpe->ch[0]);
1547 apply_prediction(ac, &cpe->ch[1]);
1551 apply_intensity_stereo(ac, cpe, ms_present);
1555 static const float cce_scale[] = {
1556 1.09050773266525765921, //2^(1/8)
1557 1.18920711500272106672, //2^(1/4)
1563 * Decode coupling_channel_element; reference: table 4.8.
1565 * @return Returns error status. 0 - OK, !0 - error
1567 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1573 SingleChannelElement *sce = &che->ch[0];
1574 ChannelCoupling *coup = &che->coup;
1576 coup->coupling_point = 2 * get_bits1(gb);
1577 coup->num_coupled = get_bits(gb, 3);
1578 for (c = 0; c <= coup->num_coupled; c++) {
1580 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1581 coup->id_select[c] = get_bits(gb, 4);
1582 if (coup->type[c] == TYPE_CPE) {
1583 coup->ch_select[c] = get_bits(gb, 2);
1584 if (coup->ch_select[c] == 3)
1587 coup->ch_select[c] = 2;
1589 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1591 sign = get_bits(gb, 1);
1592 scale = cce_scale[get_bits(gb, 2)];
1594 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1597 for (c = 0; c < num_gain; c++) {
1601 float gain_cache = 1.;
1603 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1604 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1605 gain_cache = powf(scale, -gain);
1607 if (coup->coupling_point == AFTER_IMDCT) {
1608 coup->gain[c][0] = gain_cache;
1610 for (g = 0; g < sce->ics.num_window_groups; g++) {
1611 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1612 if (sce->band_type[idx] != ZERO_BT) {
1614 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1622 gain_cache = powf(scale, -t) * s;
1625 coup->gain[c][idx] = gain_cache;
1635 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1637 * @return Returns number of bytes consumed.
1639 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1643 int num_excl_chan = 0;
1646 for (i = 0; i < 7; i++)
1647 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1648 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1650 return num_excl_chan / 7;
1654 * Decode dynamic range information; reference: table 4.52.
1656 * @param cnt length of TYPE_FIL syntactic element in bytes
1658 * @return Returns number of bytes consumed.
1660 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1661 GetBitContext *gb, int cnt)
1664 int drc_num_bands = 1;
1667 /* pce_tag_present? */
1668 if (get_bits1(gb)) {
1669 che_drc->pce_instance_tag = get_bits(gb, 4);
1670 skip_bits(gb, 4); // tag_reserved_bits
1674 /* excluded_chns_present? */
1675 if (get_bits1(gb)) {
1676 n += decode_drc_channel_exclusions(che_drc, gb);
1679 /* drc_bands_present? */
1680 if (get_bits1(gb)) {
1681 che_drc->band_incr = get_bits(gb, 4);
1682 che_drc->interpolation_scheme = get_bits(gb, 4);
1684 drc_num_bands += che_drc->band_incr;
1685 for (i = 0; i < drc_num_bands; i++) {
1686 che_drc->band_top[i] = get_bits(gb, 8);
1691 /* prog_ref_level_present? */
1692 if (get_bits1(gb)) {
1693 che_drc->prog_ref_level = get_bits(gb, 7);
1694 skip_bits1(gb); // prog_ref_level_reserved_bits
1698 for (i = 0; i < drc_num_bands; i++) {
1699 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1700 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1708 * Decode extension data (incomplete); reference: table 4.51.
1710 * @param cnt length of TYPE_FIL syntactic element in bytes
1712 * @return Returns number of bytes consumed
1714 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1715 ChannelElement *che, enum RawDataBlockType elem_type)
1719 switch (get_bits(gb, 4)) { // extension type
1720 case EXT_SBR_DATA_CRC:
1724 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1726 } else if (!ac->m4ac.sbr) {
1727 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1728 skip_bits_long(gb, 8 * cnt - 4);
1730 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1731 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1732 skip_bits_long(gb, 8 * cnt - 4);
1734 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1737 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1741 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1743 case EXT_DYNAMIC_RANGE:
1744 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1748 case EXT_DATA_ELEMENT:
1750 skip_bits_long(gb, 8 * cnt - 4);
1757 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1759 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1760 * @param coef spectral coefficients
1762 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1763 IndividualChannelStream *ics, int decode)
1765 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1767 int bottom, top, order, start, end, size, inc;
1768 float lpc[TNS_MAX_ORDER];
1769 float tmp[TNS_MAX_ORDER];
1771 for (w = 0; w < ics->num_windows; w++) {
1772 bottom = ics->num_swb;
1773 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1775 bottom = FFMAX(0, top - tns->length[w][filt]);
1776 order = tns->order[w][filt];
1781 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1783 start = ics->swb_offset[FFMIN(bottom, mmm)];
1784 end = ics->swb_offset[FFMIN( top, mmm)];
1785 if ((size = end - start) <= 0)
1787 if (tns->direction[w][filt]) {
1797 for (m = 0; m < size; m++, start += inc)
1798 for (i = 1; i <= FFMIN(m, order); i++)
1799 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1802 for (m = 0; m < size; m++, start += inc) {
1803 tmp[0] = coef[start];
1804 for (i = 1; i <= FFMIN(m, order); i++)
1805 coef[start] += tmp[i] * lpc[i - 1];
1806 for (i = order; i > 0; i--)
1807 tmp[i] = tmp[i - 1];
1815 * Apply windowing and MDCT to obtain the spectral
1816 * coefficient from the predicted sample by LTP.
1818 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1819 float *in, IndividualChannelStream *ics)
1821 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1822 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1823 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1824 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1826 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1827 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1829 memset(in, 0, 448 * sizeof(float));
1830 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1832 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1833 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1835 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1836 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1838 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1842 * Apply the long term prediction
1844 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1846 const LongTermPrediction *ltp = &sce->ics.ltp;
1847 const uint16_t *offsets = sce->ics.swb_offset;
1850 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1851 float *predTime = sce->ret;
1852 float *predFreq = ac->buf_mdct;
1853 int16_t num_samples = 2048;
1855 if (ltp->lag < 1024)
1856 num_samples = ltp->lag + 1024;
1857 for (i = 0; i < num_samples; i++)
1858 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1859 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1861 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1863 if (sce->tns.present)
1864 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1866 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1868 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1869 sce->coeffs[i] += predFreq[i];
1874 * Update the LTP buffer for next frame
1876 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1878 IndividualChannelStream *ics = &sce->ics;
1879 float *saved = sce->saved;
1880 float *saved_ltp = sce->coeffs;
1881 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1882 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1885 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1886 memcpy(saved_ltp, saved, 512 * sizeof(float));
1887 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1888 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1889 for (i = 0; i < 64; i++)
1890 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1891 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1892 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1893 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1894 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1895 for (i = 0; i < 64; i++)
1896 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1897 } else { // LONG_STOP or ONLY_LONG
1898 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1899 for (i = 0; i < 512; i++)
1900 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1903 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1904 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1905 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1909 * Conduct IMDCT and windowing.
1911 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1913 IndividualChannelStream *ics = &sce->ics;
1914 float *in = sce->coeffs;
1915 float *out = sce->ret;
1916 float *saved = sce->saved;
1917 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1918 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1919 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1920 float *buf = ac->buf_mdct;
1921 float *temp = ac->temp;
1925 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1926 for (i = 0; i < 1024; i += 128)
1927 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1929 ac->mdct.imdct_half(&ac->mdct, buf, in);
1931 /* window overlapping
1932 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1933 * and long to short transitions are considered to be short to short
1934 * transitions. This leaves just two cases (long to long and short to short)
1935 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1937 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1938 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1939 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1941 memcpy( out, saved, 448 * sizeof(float));
1943 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1944 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1945 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1946 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1947 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1948 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1949 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1951 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1952 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1957 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1958 memcpy( saved, temp + 64, 64 * sizeof(float));
1959 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1960 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1961 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1962 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1963 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1964 memcpy( saved, buf + 512, 448 * sizeof(float));
1965 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1966 } else { // LONG_STOP or ONLY_LONG
1967 memcpy( saved, buf + 512, 512 * sizeof(float));
1972 * Apply dependent channel coupling (applied before IMDCT).
1974 * @param index index into coupling gain array
1976 static void apply_dependent_coupling(AACContext *ac,
1977 SingleChannelElement *target,
1978 ChannelElement *cce, int index)
1980 IndividualChannelStream *ics = &cce->ch[0].ics;
1981 const uint16_t *offsets = ics->swb_offset;
1982 float *dest = target->coeffs;
1983 const float *src = cce->ch[0].coeffs;
1984 int g, i, group, k, idx = 0;
1985 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1986 av_log(ac->avctx, AV_LOG_ERROR,
1987 "Dependent coupling is not supported together with LTP\n");
1990 for (g = 0; g < ics->num_window_groups; g++) {
1991 for (i = 0; i < ics->max_sfb; i++, idx++) {
1992 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1993 const float gain = cce->coup.gain[index][idx];
1994 for (group = 0; group < ics->group_len[g]; group++) {
1995 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1997 dest[group * 128 + k] += gain * src[group * 128 + k];
2002 dest += ics->group_len[g] * 128;
2003 src += ics->group_len[g] * 128;
2008 * Apply independent channel coupling (applied after IMDCT).
2010 * @param index index into coupling gain array
2012 static void apply_independent_coupling(AACContext *ac,
2013 SingleChannelElement *target,
2014 ChannelElement *cce, int index)
2017 const float gain = cce->coup.gain[index][0];
2018 const float *src = cce->ch[0].ret;
2019 float *dest = target->ret;
2020 const int len = 1024 << (ac->m4ac.sbr == 1);
2022 for (i = 0; i < len; i++)
2023 dest[i] += gain * src[i];
2027 * channel coupling transformation interface
2029 * @param apply_coupling_method pointer to (in)dependent coupling function
2031 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2032 enum RawDataBlockType type, int elem_id,
2033 enum CouplingPoint coupling_point,
2034 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2038 for (i = 0; i < MAX_ELEM_ID; i++) {
2039 ChannelElement *cce = ac->che[TYPE_CCE][i];
2042 if (cce && cce->coup.coupling_point == coupling_point) {
2043 ChannelCoupling *coup = &cce->coup;
2045 for (c = 0; c <= coup->num_coupled; c++) {
2046 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2047 if (coup->ch_select[c] != 1) {
2048 apply_coupling_method(ac, &cc->ch[0], cce, index);
2049 if (coup->ch_select[c] != 0)
2052 if (coup->ch_select[c] != 2)
2053 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2055 index += 1 + (coup->ch_select[c] == 3);
2062 * Convert spectral data to float samples, applying all supported tools as appropriate.
2064 static void spectral_to_sample(AACContext *ac)
2067 for (type = 3; type >= 0; type--) {
2068 for (i = 0; i < MAX_ELEM_ID; i++) {
2069 ChannelElement *che = ac->che[type][i];
2071 if (type <= TYPE_CPE)
2072 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2073 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2074 if (che->ch[0].ics.predictor_present) {
2075 if (che->ch[0].ics.ltp.present)
2076 apply_ltp(ac, &che->ch[0]);
2077 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2078 apply_ltp(ac, &che->ch[1]);
2081 if (che->ch[0].tns.present)
2082 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2083 if (che->ch[1].tns.present)
2084 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2085 if (type <= TYPE_CPE)
2086 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2087 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2088 imdct_and_windowing(ac, &che->ch[0]);
2089 if (ac->m4ac.object_type == AOT_AAC_LTP)
2090 update_ltp(ac, &che->ch[0]);
2091 if (type == TYPE_CPE) {
2092 imdct_and_windowing(ac, &che->ch[1]);
2093 if (ac->m4ac.object_type == AOT_AAC_LTP)
2094 update_ltp(ac, &che->ch[1]);
2096 if (ac->m4ac.sbr > 0) {
2097 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2100 if (type <= TYPE_CCE)
2101 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2107 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2110 AACADTSHeaderInfo hdr_info;
2112 size = avpriv_aac_parse_header(gb, &hdr_info);
2114 if (hdr_info.chan_config) {
2115 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2116 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2117 ac->m4ac.chan_config = hdr_info.chan_config;
2118 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2120 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config,
2121 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2123 } else if (ac->output_configured != OC_LOCKED) {
2124 ac->m4ac.chan_config = 0;
2125 ac->output_configured = OC_NONE;
2127 if (ac->output_configured != OC_LOCKED) {
2130 ac->m4ac.sample_rate = hdr_info.sample_rate;
2131 ac->m4ac.sampling_index = hdr_info.sampling_index;
2132 ac->m4ac.object_type = hdr_info.object_type;
2134 if (!ac->avctx->sample_rate)
2135 ac->avctx->sample_rate = hdr_info.sample_rate;
2136 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2137 // This is 2 for "VLB " audio in NSV files.
2138 // See samples/nsv/vlb_audio.
2139 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2140 ac->warned_num_aac_frames = 1;
2142 if (!hdr_info.crc_absent)
2148 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2149 int *got_frame_ptr, GetBitContext *gb)
2151 AACContext *ac = avctx->priv_data;
2152 ChannelElement *che = NULL, *che_prev = NULL;
2153 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2155 int samples = 0, multiplier, audio_found = 0;
2157 if (show_bits(gb, 12) == 0xfff) {
2158 if (parse_adts_frame_header(ac, gb) < 0) {
2159 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2162 if (ac->m4ac.sampling_index > 12) {
2163 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2168 ac->tags_mapped = 0;
2170 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2171 elem_id = get_bits(gb, 4);
2173 if (elem_type < TYPE_DSE) {
2174 if (!ac->tags_mapped && elem_type == TYPE_CPE && ac->m4ac.chan_config==1) {
2175 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID]= {0};
2176 ac->m4ac.chan_config=2;
2178 if (set_default_channel_config(ac->avctx, new_che_pos, 2)<0)
2180 if (output_configure(ac, ac->che_pos, new_che_pos, 2, OC_TRIAL_FRAME)<0)
2183 if (!(che=get_che(ac, elem_type, elem_id))) {
2184 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2185 elem_type, elem_id);
2191 switch (elem_type) {
2194 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2199 err = decode_cpe(ac, gb, che);
2204 err = decode_cce(ac, gb, che);
2208 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2213 err = skip_data_stream_element(ac, gb);
2217 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2218 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2219 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2221 if (ac->output_configured > OC_TRIAL_PCE)
2222 av_log(avctx, AV_LOG_INFO,
2223 "Evaluating a further program_config_element.\n");
2224 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2226 ac->m4ac.chan_config = 0;
2232 elem_id += get_bits(gb, 8) - 1;
2233 if (get_bits_left(gb) < 8 * elem_id) {
2234 av_log(avctx, AV_LOG_ERROR, overread_err);
2238 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2239 err = 0; /* FIXME */
2243 err = -1; /* should not happen, but keeps compiler happy */
2248 elem_type_prev = elem_type;
2253 if (get_bits_left(gb) < 3) {
2254 av_log(avctx, AV_LOG_ERROR, overread_err);
2259 spectral_to_sample(ac);
2261 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2262 samples <<= multiplier;
2263 if (ac->output_configured < OC_LOCKED) {
2264 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2265 avctx->frame_size = samples;
2269 /* get output buffer */
2270 ac->frame.nb_samples = samples;
2271 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2272 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2276 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2277 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2278 (const float **)ac->output_data,
2279 samples, avctx->channels);
2281 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2282 (const float **)ac->output_data,
2283 samples, avctx->channels);
2285 *(AVFrame *)data = ac->frame;
2287 *got_frame_ptr = !!samples;
2289 if (ac->output_configured && audio_found)
2290 ac->output_configured = OC_LOCKED;
2295 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2296 int *got_frame_ptr, AVPacket *avpkt)
2298 AACContext *ac = avctx->priv_data;
2299 const uint8_t *buf = avpkt->data;
2300 int buf_size = avpkt->size;
2305 int new_extradata_size;
2306 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2307 AV_PKT_DATA_NEW_EXTRADATA,
2308 &new_extradata_size);
2310 if (new_extradata) {
2311 av_free(avctx->extradata);
2312 avctx->extradata = av_mallocz(new_extradata_size +
2313 FF_INPUT_BUFFER_PADDING_SIZE);
2314 if (!avctx->extradata)
2315 return AVERROR(ENOMEM);
2316 avctx->extradata_size = new_extradata_size;
2317 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2318 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2320 avctx->extradata_size*8, 1) < 0)
2321 return AVERROR_INVALIDDATA;
2324 init_get_bits(&gb, buf, buf_size * 8);
2326 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2329 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2330 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2331 if (buf[buf_offset])
2334 return buf_size > buf_offset ? buf_consumed : buf_size;
2337 static av_cold int aac_decode_close(AVCodecContext *avctx)
2339 AACContext *ac = avctx->priv_data;
2342 for (i = 0; i < MAX_ELEM_ID; i++) {
2343 for (type = 0; type < 4; type++) {
2344 if (ac->che[type][i])
2345 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2346 av_freep(&ac->che[type][i]);
2350 ff_mdct_end(&ac->mdct);
2351 ff_mdct_end(&ac->mdct_small);
2352 ff_mdct_end(&ac->mdct_ltp);
2357 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2359 struct LATMContext {
2360 AACContext aac_ctx; ///< containing AACContext
2361 int initialized; ///< initilized after a valid extradata was seen
2364 int audio_mux_version_A; ///< LATM syntax version
2365 int frame_length_type; ///< 0/1 variable/fixed frame length
2366 int frame_length; ///< frame length for fixed frame length
2369 static inline uint32_t latm_get_value(GetBitContext *b)
2371 int length = get_bits(b, 2);
2373 return get_bits_long(b, (length+1)*8);
2376 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2377 GetBitContext *gb, int asclen)
2379 AACContext *ac = &latmctx->aac_ctx;
2380 AVCodecContext *avctx = ac->avctx;
2381 MPEG4AudioConfig m4ac = {0};
2382 int config_start_bit = get_bits_count(gb);
2383 int sync_extension = 0;
2384 int bits_consumed, esize;
2388 asclen = FFMIN(asclen, get_bits_left(gb));
2390 asclen = get_bits_left(gb);
2392 if (config_start_bit % 8) {
2393 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2394 "config not byte aligned.\n", 1);
2395 return AVERROR_INVALIDDATA;
2398 return AVERROR_INVALIDDATA;
2399 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2400 gb->buffer + (config_start_bit / 8),
2401 asclen, sync_extension);
2403 if (bits_consumed < 0)
2404 return AVERROR_INVALIDDATA;
2406 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2407 ac->m4ac.chan_config != m4ac.chan_config) {
2409 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2410 latmctx->initialized = 0;
2412 esize = (bits_consumed+7) / 8;
2414 if (avctx->extradata_size < esize) {
2415 av_free(avctx->extradata);
2416 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2417 if (!avctx->extradata)
2418 return AVERROR(ENOMEM);
2421 avctx->extradata_size = esize;
2422 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2423 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2425 skip_bits_long(gb, bits_consumed);
2427 return bits_consumed;
2430 static int read_stream_mux_config(struct LATMContext *latmctx,
2433 int ret, audio_mux_version = get_bits(gb, 1);
2435 latmctx->audio_mux_version_A = 0;
2436 if (audio_mux_version)
2437 latmctx->audio_mux_version_A = get_bits(gb, 1);
2439 if (!latmctx->audio_mux_version_A) {
2441 if (audio_mux_version)
2442 latm_get_value(gb); // taraFullness
2444 skip_bits(gb, 1); // allStreamSameTimeFraming
2445 skip_bits(gb, 6); // numSubFrames
2447 if (get_bits(gb, 4)) { // numPrograms
2448 av_log_missing_feature(latmctx->aac_ctx.avctx,
2449 "multiple programs are not supported\n", 1);
2450 return AVERROR_PATCHWELCOME;
2453 // for each program (which there is only on in DVB)
2455 // for each layer (which there is only on in DVB)
2456 if (get_bits(gb, 3)) { // numLayer
2457 av_log_missing_feature(latmctx->aac_ctx.avctx,
2458 "multiple layers are not supported\n", 1);
2459 return AVERROR_PATCHWELCOME;
2462 // for all but first stream: use_same_config = get_bits(gb, 1);
2463 if (!audio_mux_version) {
2464 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2467 int ascLen = latm_get_value(gb);
2468 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2471 skip_bits_long(gb, ascLen);
2474 latmctx->frame_length_type = get_bits(gb, 3);
2475 switch (latmctx->frame_length_type) {
2477 skip_bits(gb, 8); // latmBufferFullness
2480 latmctx->frame_length = get_bits(gb, 9);
2485 skip_bits(gb, 6); // CELP frame length table index
2489 skip_bits(gb, 1); // HVXC frame length table index
2493 if (get_bits(gb, 1)) { // other data
2494 if (audio_mux_version) {
2495 latm_get_value(gb); // other_data_bits
2499 esc = get_bits(gb, 1);
2505 if (get_bits(gb, 1)) // crc present
2506 skip_bits(gb, 8); // config_crc
2512 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2516 if (ctx->frame_length_type == 0) {
2517 int mux_slot_length = 0;
2519 tmp = get_bits(gb, 8);
2520 mux_slot_length += tmp;
2521 } while (tmp == 255);
2522 return mux_slot_length;
2523 } else if (ctx->frame_length_type == 1) {
2524 return ctx->frame_length;
2525 } else if (ctx->frame_length_type == 3 ||
2526 ctx->frame_length_type == 5 ||
2527 ctx->frame_length_type == 7) {
2528 skip_bits(gb, 2); // mux_slot_length_coded
2533 static int read_audio_mux_element(struct LATMContext *latmctx,
2537 uint8_t use_same_mux = get_bits(gb, 1);
2538 if (!use_same_mux) {
2539 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2541 } else if (!latmctx->aac_ctx.avctx->extradata) {
2542 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2543 "no decoder config found\n");
2544 return AVERROR(EAGAIN);
2546 if (latmctx->audio_mux_version_A == 0) {
2547 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2548 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2549 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2550 return AVERROR_INVALIDDATA;
2551 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2552 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2553 "frame length mismatch %d << %d\n",
2554 mux_slot_length_bytes * 8, get_bits_left(gb));
2555 return AVERROR_INVALIDDATA;
2562 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2563 int *got_frame_ptr, AVPacket *avpkt)
2565 struct LATMContext *latmctx = avctx->priv_data;
2569 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2571 // check for LOAS sync word
2572 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2573 return AVERROR_INVALIDDATA;
2575 muxlength = get_bits(&gb, 13) + 3;
2576 // not enough data, the parser should have sorted this
2577 if (muxlength > avpkt->size)
2578 return AVERROR_INVALIDDATA;
2580 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2583 if (!latmctx->initialized) {
2584 if (!avctx->extradata) {
2588 if ((err = decode_audio_specific_config(
2589 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2590 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2592 latmctx->initialized = 1;
2596 if (show_bits(&gb, 12) == 0xfff) {
2597 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2598 "ADTS header detected, probably as result of configuration "
2600 return AVERROR_INVALIDDATA;
2603 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2609 av_cold static int latm_decode_init(AVCodecContext *avctx)
2611 struct LATMContext *latmctx = avctx->priv_data;
2612 int ret = aac_decode_init(avctx);
2614 if (avctx->extradata_size > 0)
2615 latmctx->initialized = !ret;
2621 AVCodec ff_aac_decoder = {
2623 .type = AVMEDIA_TYPE_AUDIO,
2625 .priv_data_size = sizeof(AACContext),
2626 .init = aac_decode_init,
2627 .close = aac_decode_close,
2628 .decode = aac_decode_frame,
2629 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2630 .sample_fmts = (const enum AVSampleFormat[]) {
2631 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2633 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2634 .channel_layouts = aac_channel_layout,
2638 Note: This decoder filter is intended to decode LATM streams transferred
2639 in MPEG transport streams which only contain one program.
2640 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2642 AVCodec ff_aac_latm_decoder = {
2644 .type = AVMEDIA_TYPE_AUDIO,
2645 .id = CODEC_ID_AAC_LATM,
2646 .priv_data_size = sizeof(struct LATMContext),
2647 .init = latm_decode_init,
2648 .close = aac_decode_close,
2649 .decode = latm_decode_frame,
2650 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2651 .sample_fmts = (const enum AVSampleFormat[]) {
2652 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2654 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2655 .channel_layouts = aac_channel_layout,