3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
108 # include "arm/aac.h"
116 static VLC vlc_scalefactors;
117 static VLC vlc_spectral[11];
119 static const char overread_err[] = "Input buffer exhausted before END element found\n";
121 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
123 // For PCE based channel configurations map the channels solely based on tags.
124 if (!ac->m4ac.chan_config) {
125 return ac->tag_che_map[type][elem_id];
127 // For indexed channel configurations map the channels solely based on position.
128 switch (ac->m4ac.chan_config) {
130 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
132 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
135 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
136 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
137 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
138 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
140 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
143 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
145 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
148 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
150 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
154 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
156 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
157 } else if (ac->m4ac.chan_config == 2) {
161 if (!ac->tags_mapped && type == TYPE_SCE) {
163 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
171 * Check for the channel element in the current channel position configuration.
172 * If it exists, make sure the appropriate element is allocated and map the
173 * channel order to match the internal FFmpeg channel layout.
175 * @param che_pos current channel position configuration
176 * @param type channel element type
177 * @param id channel element id
178 * @param channels count of the number of channels in the configuration
180 * @return Returns error status. 0 - OK, !0 - error
182 static av_cold int che_configure(AACContext *ac,
183 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
184 int type, int id, int *channels)
186 if (che_pos[type][id]) {
187 if (!ac->che[type][id] && !(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
188 return AVERROR(ENOMEM);
189 ff_aac_sbr_ctx_init(&ac->che[type][id]->sbr);
190 if (type != TYPE_CCE) {
191 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
192 if (type == TYPE_CPE ||
193 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
194 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
198 if (ac->che[type][id])
199 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
200 av_freep(&ac->che[type][id]);
206 * Configure output channel order based on the current program configuration element.
208 * @param che_pos current channel position configuration
209 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
211 * @return Returns error status. 0 - OK, !0 - error
213 static av_cold int output_configure(AACContext *ac,
214 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
215 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
216 int channel_config, enum OCStatus oc_type)
218 AVCodecContext *avctx = ac->avctx;
219 int i, type, channels = 0, ret;
221 if (new_che_pos != che_pos)
222 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
224 if (channel_config) {
225 for (i = 0; i < tags_per_config[channel_config]; i++) {
226 if ((ret = che_configure(ac, che_pos,
227 aac_channel_layout_map[channel_config - 1][i][0],
228 aac_channel_layout_map[channel_config - 1][i][1],
233 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
235 avctx->channel_layout = aac_channel_layout[channel_config - 1];
237 /* Allocate or free elements depending on if they are in the
238 * current program configuration.
240 * Set up default 1:1 output mapping.
242 * For a 5.1 stream the output order will be:
243 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
246 for (i = 0; i < MAX_ELEM_ID; i++) {
247 for (type = 0; type < 4; type++) {
248 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
253 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
255 avctx->channel_layout = 0;
258 avctx->channels = channels;
260 ac->output_configured = oc_type;
266 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
268 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
269 * @param sce_map mono (Single Channel Element) map
270 * @param type speaker type/position for these channels
272 static void decode_channel_map(enum ChannelPosition *cpe_map,
273 enum ChannelPosition *sce_map,
274 enum ChannelPosition type,
275 GetBitContext *gb, int n)
278 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
279 map[get_bits(gb, 4)] = type;
284 * Decode program configuration element; reference: table 4.2.
286 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
288 * @return Returns error status. 0 - OK, !0 - error
290 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
291 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
294 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
297 skip_bits(gb, 2); // object_type
299 sampling_index = get_bits(gb, 4);
300 if (m4ac->sampling_index != sampling_index)
301 av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
303 num_front = get_bits(gb, 4);
304 num_side = get_bits(gb, 4);
305 num_back = get_bits(gb, 4);
306 num_lfe = get_bits(gb, 2);
307 num_assoc_data = get_bits(gb, 3);
308 num_cc = get_bits(gb, 4);
311 skip_bits(gb, 4); // mono_mixdown_tag
313 skip_bits(gb, 4); // stereo_mixdown_tag
316 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
318 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
319 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
320 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
321 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
323 skip_bits_long(gb, 4 * num_assoc_data);
325 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
329 /* comment field, first byte is length */
330 comment_len = get_bits(gb, 8) * 8;
331 if (get_bits_left(gb) < comment_len) {
332 av_log(avctx, AV_LOG_ERROR, overread_err);
335 skip_bits_long(gb, comment_len);
340 * Set up channel positions based on a default channel configuration
341 * as specified in table 1.17.
343 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
345 * @return Returns error status. 0 - OK, !0 - error
347 static av_cold int set_default_channel_config(AVCodecContext *avctx,
348 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
351 if (channel_config < 1 || channel_config > 7) {
352 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
357 /* default channel configurations:
359 * 1ch : front center (mono)
360 * 2ch : L + R (stereo)
361 * 3ch : front center + L + R
362 * 4ch : front center + L + R + back center
363 * 5ch : front center + L + R + back stereo
364 * 6ch : front center + L + R + back stereo + LFE
365 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
368 if (channel_config != 2)
369 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
370 if (channel_config > 1)
371 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
372 if (channel_config == 4)
373 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
374 if (channel_config > 4)
375 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
376 = AAC_CHANNEL_BACK; // back stereo
377 if (channel_config > 5)
378 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
379 if (channel_config == 7)
380 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
386 * Decode GA "General Audio" specific configuration; reference: table 4.1.
388 * @param ac pointer to AACContext, may be null
389 * @param avctx pointer to AVCCodecContext, used for logging
391 * @return Returns error status. 0 - OK, !0 - error
393 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
395 MPEG4AudioConfig *m4ac,
398 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
399 int extension_flag, ret;
401 if (get_bits1(gb)) { // frameLengthFlag
402 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
406 if (get_bits1(gb)) // dependsOnCoreCoder
407 skip_bits(gb, 14); // coreCoderDelay
408 extension_flag = get_bits1(gb);
410 if (m4ac->object_type == AOT_AAC_SCALABLE ||
411 m4ac->object_type == AOT_ER_AAC_SCALABLE)
412 skip_bits(gb, 3); // layerNr
414 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
415 if (channel_config == 0) {
416 skip_bits(gb, 4); // element_instance_tag
417 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
420 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
423 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
426 if (extension_flag) {
427 switch (m4ac->object_type) {
429 skip_bits(gb, 5); // numOfSubFrame
430 skip_bits(gb, 11); // layer_length
434 case AOT_ER_AAC_SCALABLE:
436 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
437 * aacScalefactorDataResilienceFlag
438 * aacSpectralDataResilienceFlag
442 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
448 * Decode audio specific configuration; reference: table 1.13.
450 * @param ac pointer to AACContext, may be null
451 * @param avctx pointer to AVCCodecContext, used for logging
452 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
453 * @param data pointer to AVCodecContext extradata
454 * @param data_size size of AVCCodecContext extradata
456 * @return Returns error status or number of consumed bits. <0 - error
458 static int decode_audio_specific_config(AACContext *ac,
459 AVCodecContext *avctx,
460 MPEG4AudioConfig *m4ac,
461 const uint8_t *data, int data_size)
466 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
467 for (i = 0; i < avctx->extradata_size; i++)
468 av_dlog(avctx, "%02x ", avctx->extradata[i]);
469 av_dlog(avctx, "\n");
471 init_get_bits(&gb, data, data_size * 8);
473 if ((i = ff_mpeg4audio_get_config(m4ac, data, data_size)) < 0)
475 if (m4ac->sampling_index > 12) {
476 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
479 if (m4ac->sbr == 1 && m4ac->ps == -1)
482 skip_bits_long(&gb, i);
484 switch (m4ac->object_type) {
488 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
492 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
493 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
497 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
498 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
499 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
501 return get_bits_count(&gb);
505 * linear congruential pseudorandom number generator
507 * @param previous_val pointer to the current state of the generator
509 * @return Returns a 32-bit pseudorandom integer
511 static av_always_inline int lcg_random(int previous_val)
513 return previous_val * 1664525 + 1013904223;
516 static av_always_inline void reset_predict_state(PredictorState *ps)
526 static void reset_all_predictors(PredictorState *ps)
529 for (i = 0; i < MAX_PREDICTORS; i++)
530 reset_predict_state(&ps[i]);
533 static void reset_predictor_group(PredictorState *ps, int group_num)
536 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
537 reset_predict_state(&ps[i]);
540 #define AAC_INIT_VLC_STATIC(num, size) \
541 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
542 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
543 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
546 static av_cold int aac_decode_init(AVCodecContext *avctx)
548 AACContext *ac = avctx->priv_data;
551 ac->m4ac.sample_rate = avctx->sample_rate;
553 if (avctx->extradata_size > 0) {
554 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
556 avctx->extradata_size) < 0)
560 avctx->sample_fmt = avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT ?
561 AV_SAMPLE_FMT_FLT : AV_SAMPLE_FMT_S16;
563 AAC_INIT_VLC_STATIC( 0, 304);
564 AAC_INIT_VLC_STATIC( 1, 270);
565 AAC_INIT_VLC_STATIC( 2, 550);
566 AAC_INIT_VLC_STATIC( 3, 300);
567 AAC_INIT_VLC_STATIC( 4, 328);
568 AAC_INIT_VLC_STATIC( 5, 294);
569 AAC_INIT_VLC_STATIC( 6, 306);
570 AAC_INIT_VLC_STATIC( 7, 268);
571 AAC_INIT_VLC_STATIC( 8, 510);
572 AAC_INIT_VLC_STATIC( 9, 366);
573 AAC_INIT_VLC_STATIC(10, 462);
577 dsputil_init(&ac->dsp, avctx);
578 ff_fmt_convert_init(&ac->fmt_conv, avctx);
580 ac->random_state = 0x1f2e3d4c;
584 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
585 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
586 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
589 ff_mdct_init(&ac->mdct, 11, 1, 1.0/1024.0);
590 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0/128.0);
591 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0);
592 // window initialization
593 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
594 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
595 ff_init_ff_sine_windows(10);
596 ff_init_ff_sine_windows( 7);
604 * Skip data_stream_element; reference: table 4.10.
606 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
608 int byte_align = get_bits1(gb);
609 int count = get_bits(gb, 8);
611 count += get_bits(gb, 8);
615 if (get_bits_left(gb) < 8 * count) {
616 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
619 skip_bits_long(gb, 8 * count);
623 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
628 ics->predictor_reset_group = get_bits(gb, 5);
629 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
630 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
634 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
635 ics->prediction_used[sfb] = get_bits1(gb);
641 * Decode Long Term Prediction data; reference: table 4.xx.
643 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
644 GetBitContext *gb, uint8_t max_sfb)
648 ltp->lag = get_bits(gb, 11);
649 ltp->coef = ltp_coef[get_bits(gb, 3)];
650 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
651 ltp->used[sfb] = get_bits1(gb);
655 * Decode Individual Channel Stream info; reference: table 4.6.
657 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
659 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
660 GetBitContext *gb, int common_window)
663 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
664 memset(ics, 0, sizeof(IndividualChannelStream));
667 ics->window_sequence[1] = ics->window_sequence[0];
668 ics->window_sequence[0] = get_bits(gb, 2);
669 ics->use_kb_window[1] = ics->use_kb_window[0];
670 ics->use_kb_window[0] = get_bits1(gb);
671 ics->num_window_groups = 1;
672 ics->group_len[0] = 1;
673 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
675 ics->max_sfb = get_bits(gb, 4);
676 for (i = 0; i < 7; i++) {
678 ics->group_len[ics->num_window_groups - 1]++;
680 ics->num_window_groups++;
681 ics->group_len[ics->num_window_groups - 1] = 1;
684 ics->num_windows = 8;
685 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
686 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
687 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
688 ics->predictor_present = 0;
690 ics->max_sfb = get_bits(gb, 6);
691 ics->num_windows = 1;
692 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
693 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
694 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
695 ics->predictor_present = get_bits1(gb);
696 ics->predictor_reset_group = 0;
697 if (ics->predictor_present) {
698 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
699 if (decode_prediction(ac, ics, gb)) {
700 memset(ics, 0, sizeof(IndividualChannelStream));
703 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
704 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
705 memset(ics, 0, sizeof(IndividualChannelStream));
708 if ((ics->ltp.present = get_bits(gb, 1)))
709 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
714 if (ics->max_sfb > ics->num_swb) {
715 av_log(ac->avctx, AV_LOG_ERROR,
716 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
717 ics->max_sfb, ics->num_swb);
718 memset(ics, 0, sizeof(IndividualChannelStream));
726 * Decode band types (section_data payload); reference: table 4.46.
728 * @param band_type array of the used band type
729 * @param band_type_run_end array of the last scalefactor band of a band type run
731 * @return Returns error status. 0 - OK, !0 - error
733 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
734 int band_type_run_end[120], GetBitContext *gb,
735 IndividualChannelStream *ics)
738 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
739 for (g = 0; g < ics->num_window_groups; g++) {
741 while (k < ics->max_sfb) {
742 uint8_t sect_end = k;
744 int sect_band_type = get_bits(gb, 4);
745 if (sect_band_type == 12) {
746 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
749 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
750 sect_end += sect_len_incr;
751 sect_end += sect_len_incr;
752 if (get_bits_left(gb) < 0) {
753 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
756 if (sect_end > ics->max_sfb) {
757 av_log(ac->avctx, AV_LOG_ERROR,
758 "Number of bands (%d) exceeds limit (%d).\n",
759 sect_end, ics->max_sfb);
762 for (; k < sect_end; k++) {
763 band_type [idx] = sect_band_type;
764 band_type_run_end[idx++] = sect_end;
772 * Decode scalefactors; reference: table 4.47.
774 * @param global_gain first scalefactor value as scalefactors are differentially coded
775 * @param band_type array of the used band type
776 * @param band_type_run_end array of the last scalefactor band of a band type run
777 * @param sf array of scalefactors or intensity stereo positions
779 * @return Returns error status. 0 - OK, !0 - error
781 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
782 unsigned int global_gain,
783 IndividualChannelStream *ics,
784 enum BandType band_type[120],
785 int band_type_run_end[120])
788 int offset[3] = { global_gain, global_gain - 90, 0 };
791 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
792 for (g = 0; g < ics->num_window_groups; g++) {
793 for (i = 0; i < ics->max_sfb;) {
794 int run_end = band_type_run_end[idx];
795 if (band_type[idx] == ZERO_BT) {
796 for (; i < run_end; i++, idx++)
798 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
799 for (; i < run_end; i++, idx++) {
800 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
801 clipped_offset = av_clip(offset[2], -155, 100);
802 if (offset[2] != clipped_offset) {
803 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
804 "position clipped (%d -> %d).\nIf you heard an "
805 "audible artifact, there may be a bug in the "
806 "decoder. ", offset[2], clipped_offset);
808 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
810 } else if (band_type[idx] == NOISE_BT) {
811 for (; i < run_end; i++, idx++) {
812 if (noise_flag-- > 0)
813 offset[1] += get_bits(gb, 9) - 256;
815 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
816 clipped_offset = av_clip(offset[1], -100, 155);
817 if (offset[2] != clipped_offset) {
818 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
819 "(%d -> %d).\nIf you heard an audible "
820 "artifact, there may be a bug in the decoder. ",
821 offset[1], clipped_offset);
823 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
826 for (; i < run_end; i++, idx++) {
827 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
828 if (offset[0] > 255U) {
829 av_log(ac->avctx, AV_LOG_ERROR,
830 "%s (%d) out of range.\n", sf_str[0], offset[0]);
833 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
842 * Decode pulse data; reference: table 4.7.
844 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
845 const uint16_t *swb_offset, int num_swb)
848 pulse->num_pulse = get_bits(gb, 2) + 1;
849 pulse_swb = get_bits(gb, 6);
850 if (pulse_swb >= num_swb)
852 pulse->pos[0] = swb_offset[pulse_swb];
853 pulse->pos[0] += get_bits(gb, 5);
854 if (pulse->pos[0] > 1023)
856 pulse->amp[0] = get_bits(gb, 4);
857 for (i = 1; i < pulse->num_pulse; i++) {
858 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
859 if (pulse->pos[i] > 1023)
861 pulse->amp[i] = get_bits(gb, 4);
867 * Decode Temporal Noise Shaping data; reference: table 4.48.
869 * @return Returns error status. 0 - OK, !0 - error
871 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
872 GetBitContext *gb, const IndividualChannelStream *ics)
874 int w, filt, i, coef_len, coef_res, coef_compress;
875 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
876 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
877 for (w = 0; w < ics->num_windows; w++) {
878 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
879 coef_res = get_bits1(gb);
881 for (filt = 0; filt < tns->n_filt[w]; filt++) {
883 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
885 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
886 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
887 tns->order[w][filt], tns_max_order);
888 tns->order[w][filt] = 0;
891 if (tns->order[w][filt]) {
892 tns->direction[w][filt] = get_bits1(gb);
893 coef_compress = get_bits1(gb);
894 coef_len = coef_res + 3 - coef_compress;
895 tmp2_idx = 2 * coef_compress + coef_res;
897 for (i = 0; i < tns->order[w][filt]; i++)
898 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
907 * Decode Mid/Side data; reference: table 4.54.
909 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
910 * [1] mask is decoded from bitstream; [2] mask is all 1s;
911 * [3] reserved for scalable AAC
913 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
917 if (ms_present == 1) {
918 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
919 cpe->ms_mask[idx] = get_bits1(gb);
920 } else if (ms_present == 2) {
921 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
926 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
930 *dst++ = v[idx & 15] * s;
931 *dst++ = v[idx>>4 & 15] * s;
937 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
941 *dst++ = v[idx & 3] * s;
942 *dst++ = v[idx>>2 & 3] * s;
943 *dst++ = v[idx>>4 & 3] * s;
944 *dst++ = v[idx>>6 & 3] * s;
950 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
951 unsigned sign, const float *scale)
953 union float754 s0, s1;
955 s0.f = s1.f = *scale;
956 s0.i ^= sign >> 1 << 31;
959 *dst++ = v[idx & 15] * s0.f;
960 *dst++ = v[idx>>4 & 15] * s1.f;
967 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
968 unsigned sign, const float *scale)
970 unsigned nz = idx >> 12;
971 union float754 s = { .f = *scale };
974 t.i = s.i ^ (sign & 1U<<31);
975 *dst++ = v[idx & 3] * t.f;
977 sign <<= nz & 1; nz >>= 1;
978 t.i = s.i ^ (sign & 1U<<31);
979 *dst++ = v[idx>>2 & 3] * t.f;
981 sign <<= nz & 1; nz >>= 1;
982 t.i = s.i ^ (sign & 1U<<31);
983 *dst++ = v[idx>>4 & 3] * t.f;
985 sign <<= nz & 1; nz >>= 1;
986 t.i = s.i ^ (sign & 1U<<31);
987 *dst++ = v[idx>>6 & 3] * t.f;
994 * Decode spectral data; reference: table 4.50.
995 * Dequantize and scale spectral data; reference: 4.6.3.3.
997 * @param coef array of dequantized, scaled spectral data
998 * @param sf array of scalefactors or intensity stereo positions
999 * @param pulse_present set if pulses are present
1000 * @param pulse pointer to pulse data struct
1001 * @param band_type array of the used band type
1003 * @return Returns error status. 0 - OK, !0 - error
1005 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1006 GetBitContext *gb, const float sf[120],
1007 int pulse_present, const Pulse *pulse,
1008 const IndividualChannelStream *ics,
1009 enum BandType band_type[120])
1011 int i, k, g, idx = 0;
1012 const int c = 1024 / ics->num_windows;
1013 const uint16_t *offsets = ics->swb_offset;
1014 float *coef_base = coef;
1016 for (g = 0; g < ics->num_windows; g++)
1017 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1019 for (g = 0; g < ics->num_window_groups; g++) {
1020 unsigned g_len = ics->group_len[g];
1022 for (i = 0; i < ics->max_sfb; i++, idx++) {
1023 const unsigned cbt_m1 = band_type[idx] - 1;
1024 float *cfo = coef + offsets[i];
1025 int off_len = offsets[i + 1] - offsets[i];
1028 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1029 for (group = 0; group < g_len; group++, cfo+=128) {
1030 memset(cfo, 0, off_len * sizeof(float));
1032 } else if (cbt_m1 == NOISE_BT - 1) {
1033 for (group = 0; group < g_len; group++, cfo+=128) {
1037 for (k = 0; k < off_len; k++) {
1038 ac->random_state = lcg_random(ac->random_state);
1039 cfo[k] = ac->random_state;
1042 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1043 scale = sf[idx] / sqrtf(band_energy);
1044 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1047 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1048 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1049 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1050 OPEN_READER(re, gb);
1052 switch (cbt_m1 >> 1) {
1054 for (group = 0; group < g_len; group++, cfo+=128) {
1062 UPDATE_CACHE(re, gb);
1063 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1064 cb_idx = cb_vector_idx[code];
1065 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1071 for (group = 0; group < g_len; group++, cfo+=128) {
1081 UPDATE_CACHE(re, gb);
1082 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1083 cb_idx = cb_vector_idx[code];
1084 nnz = cb_idx >> 8 & 15;
1085 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1086 LAST_SKIP_BITS(re, gb, nnz);
1087 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1093 for (group = 0; group < g_len; group++, cfo+=128) {
1101 UPDATE_CACHE(re, gb);
1102 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1103 cb_idx = cb_vector_idx[code];
1104 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1111 for (group = 0; group < g_len; group++, cfo+=128) {
1121 UPDATE_CACHE(re, gb);
1122 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1123 cb_idx = cb_vector_idx[code];
1124 nnz = cb_idx >> 8 & 15;
1125 sign = SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12);
1126 LAST_SKIP_BITS(re, gb, nnz);
1127 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1133 for (group = 0; group < g_len; group++, cfo+=128) {
1135 uint32_t *icf = (uint32_t *) cf;
1145 UPDATE_CACHE(re, gb);
1146 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1154 cb_idx = cb_vector_idx[code];
1157 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1158 LAST_SKIP_BITS(re, gb, nnz);
1160 for (j = 0; j < 2; j++) {
1164 /* The total length of escape_sequence must be < 22 bits according
1165 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1166 UPDATE_CACHE(re, gb);
1167 b = GET_CACHE(re, gb);
1168 b = 31 - av_log2(~b);
1171 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1175 SKIP_BITS(re, gb, b + 1);
1177 n = (1 << b) + SHOW_UBITS(re, gb, b);
1178 LAST_SKIP_BITS(re, gb, b);
1179 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1182 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1183 *icf++ = (bits & 1U<<31) | v;
1190 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1194 CLOSE_READER(re, gb);
1200 if (pulse_present) {
1202 for (i = 0; i < pulse->num_pulse; i++) {
1203 float co = coef_base[ pulse->pos[i] ];
1204 while (offsets[idx + 1] <= pulse->pos[i])
1206 if (band_type[idx] != NOISE_BT && sf[idx]) {
1207 float ico = -pulse->amp[i];
1210 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1212 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1219 static av_always_inline float flt16_round(float pf)
1223 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1227 static av_always_inline float flt16_even(float pf)
1231 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1235 static av_always_inline float flt16_trunc(float pf)
1239 pun.i &= 0xFFFF0000U;
1243 static av_always_inline void predict(PredictorState *ps, float *coef,
1246 const float a = 0.953125; // 61.0 / 64
1247 const float alpha = 0.90625; // 29.0 / 32
1251 float r0 = ps->r0, r1 = ps->r1;
1252 float cor0 = ps->cor0, cor1 = ps->cor1;
1253 float var0 = ps->var0, var1 = ps->var1;
1255 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1256 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1258 pv = flt16_round(k1 * r0 + k2 * r1);
1265 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1266 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1267 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1268 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1270 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1271 ps->r0 = flt16_trunc(a * e0);
1275 * Apply AAC-Main style frequency domain prediction.
1277 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1281 if (!sce->ics.predictor_initialized) {
1282 reset_all_predictors(sce->predictor_state);
1283 sce->ics.predictor_initialized = 1;
1286 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1287 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1288 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1289 predict(&sce->predictor_state[k], &sce->coeffs[k],
1290 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1293 if (sce->ics.predictor_reset_group)
1294 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1296 reset_all_predictors(sce->predictor_state);
1300 * Decode an individual_channel_stream payload; reference: table 4.44.
1302 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1303 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1305 * @return Returns error status. 0 - OK, !0 - error
1307 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1308 GetBitContext *gb, int common_window, int scale_flag)
1311 TemporalNoiseShaping *tns = &sce->tns;
1312 IndividualChannelStream *ics = &sce->ics;
1313 float *out = sce->coeffs;
1314 int global_gain, pulse_present = 0;
1316 /* This assignment is to silence a GCC warning about the variable being used
1317 * uninitialized when in fact it always is.
1319 pulse.num_pulse = 0;
1321 global_gain = get_bits(gb, 8);
1323 if (!common_window && !scale_flag) {
1324 if (decode_ics_info(ac, ics, gb, 0) < 0)
1328 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1330 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1335 if ((pulse_present = get_bits1(gb))) {
1336 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1337 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1340 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1341 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1345 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1347 if (get_bits1(gb)) {
1348 av_log_missing_feature(ac->avctx, "SSR", 1);
1353 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1356 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1357 apply_prediction(ac, sce);
1363 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1365 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1367 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1368 float *ch0 = cpe->ch[0].coeffs;
1369 float *ch1 = cpe->ch[1].coeffs;
1370 int g, i, group, idx = 0;
1371 const uint16_t *offsets = ics->swb_offset;
1372 for (g = 0; g < ics->num_window_groups; g++) {
1373 for (i = 0; i < ics->max_sfb; i++, idx++) {
1374 if (cpe->ms_mask[idx] &&
1375 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1376 for (group = 0; group < ics->group_len[g]; group++) {
1377 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1378 ch1 + group * 128 + offsets[i],
1379 offsets[i+1] - offsets[i]);
1383 ch0 += ics->group_len[g] * 128;
1384 ch1 += ics->group_len[g] * 128;
1389 * intensity stereo decoding; reference: 4.6.8.2.3
1391 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1392 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1393 * [3] reserved for scalable AAC
1395 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1397 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1398 SingleChannelElement *sce1 = &cpe->ch[1];
1399 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1400 const uint16_t *offsets = ics->swb_offset;
1401 int g, group, i, idx = 0;
1404 for (g = 0; g < ics->num_window_groups; g++) {
1405 for (i = 0; i < ics->max_sfb;) {
1406 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1407 const int bt_run_end = sce1->band_type_run_end[idx];
1408 for (; i < bt_run_end; i++, idx++) {
1409 c = -1 + 2 * (sce1->band_type[idx] - 14);
1411 c *= 1 - 2 * cpe->ms_mask[idx];
1412 scale = c * sce1->sf[idx];
1413 for (group = 0; group < ics->group_len[g]; group++)
1414 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1415 coef0 + group * 128 + offsets[i],
1417 offsets[i + 1] - offsets[i]);
1420 int bt_run_end = sce1->band_type_run_end[idx];
1421 idx += bt_run_end - i;
1425 coef0 += ics->group_len[g] * 128;
1426 coef1 += ics->group_len[g] * 128;
1431 * Decode a channel_pair_element; reference: table 4.4.
1433 * @return Returns error status. 0 - OK, !0 - error
1435 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1437 int i, ret, common_window, ms_present = 0;
1439 common_window = get_bits1(gb);
1440 if (common_window) {
1441 if (decode_ics_info(ac, &cpe->ch[0].ics, gb, 1))
1443 i = cpe->ch[1].ics.use_kb_window[0];
1444 cpe->ch[1].ics = cpe->ch[0].ics;
1445 cpe->ch[1].ics.use_kb_window[1] = i;
1446 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1447 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1448 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1449 ms_present = get_bits(gb, 2);
1450 if (ms_present == 3) {
1451 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1453 } else if (ms_present)
1454 decode_mid_side_stereo(cpe, gb, ms_present);
1456 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1458 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1461 if (common_window) {
1463 apply_mid_side_stereo(ac, cpe);
1464 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1465 apply_prediction(ac, &cpe->ch[0]);
1466 apply_prediction(ac, &cpe->ch[1]);
1470 apply_intensity_stereo(ac, cpe, ms_present);
1474 static const float cce_scale[] = {
1475 1.09050773266525765921, //2^(1/8)
1476 1.18920711500272106672, //2^(1/4)
1482 * Decode coupling_channel_element; reference: table 4.8.
1484 * @return Returns error status. 0 - OK, !0 - error
1486 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1492 SingleChannelElement *sce = &che->ch[0];
1493 ChannelCoupling *coup = &che->coup;
1495 coup->coupling_point = 2 * get_bits1(gb);
1496 coup->num_coupled = get_bits(gb, 3);
1497 for (c = 0; c <= coup->num_coupled; c++) {
1499 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1500 coup->id_select[c] = get_bits(gb, 4);
1501 if (coup->type[c] == TYPE_CPE) {
1502 coup->ch_select[c] = get_bits(gb, 2);
1503 if (coup->ch_select[c] == 3)
1506 coup->ch_select[c] = 2;
1508 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1510 sign = get_bits(gb, 1);
1511 scale = cce_scale[get_bits(gb, 2)];
1513 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1516 for (c = 0; c < num_gain; c++) {
1520 float gain_cache = 1.;
1522 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1523 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1524 gain_cache = powf(scale, -gain);
1526 if (coup->coupling_point == AFTER_IMDCT) {
1527 coup->gain[c][0] = gain_cache;
1529 for (g = 0; g < sce->ics.num_window_groups; g++) {
1530 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1531 if (sce->band_type[idx] != ZERO_BT) {
1533 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1541 gain_cache = powf(scale, -t) * s;
1544 coup->gain[c][idx] = gain_cache;
1554 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1556 * @return Returns number of bytes consumed.
1558 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1562 int num_excl_chan = 0;
1565 for (i = 0; i < 7; i++)
1566 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1567 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1569 return num_excl_chan / 7;
1573 * Decode dynamic range information; reference: table 4.52.
1575 * @param cnt length of TYPE_FIL syntactic element in bytes
1577 * @return Returns number of bytes consumed.
1579 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1580 GetBitContext *gb, int cnt)
1583 int drc_num_bands = 1;
1586 /* pce_tag_present? */
1587 if (get_bits1(gb)) {
1588 che_drc->pce_instance_tag = get_bits(gb, 4);
1589 skip_bits(gb, 4); // tag_reserved_bits
1593 /* excluded_chns_present? */
1594 if (get_bits1(gb)) {
1595 n += decode_drc_channel_exclusions(che_drc, gb);
1598 /* drc_bands_present? */
1599 if (get_bits1(gb)) {
1600 che_drc->band_incr = get_bits(gb, 4);
1601 che_drc->interpolation_scheme = get_bits(gb, 4);
1603 drc_num_bands += che_drc->band_incr;
1604 for (i = 0; i < drc_num_bands; i++) {
1605 che_drc->band_top[i] = get_bits(gb, 8);
1610 /* prog_ref_level_present? */
1611 if (get_bits1(gb)) {
1612 che_drc->prog_ref_level = get_bits(gb, 7);
1613 skip_bits1(gb); // prog_ref_level_reserved_bits
1617 for (i = 0; i < drc_num_bands; i++) {
1618 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1619 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1627 * Decode extension data (incomplete); reference: table 4.51.
1629 * @param cnt length of TYPE_FIL syntactic element in bytes
1631 * @return Returns number of bytes consumed
1633 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1634 ChannelElement *che, enum RawDataBlockType elem_type)
1638 switch (get_bits(gb, 4)) { // extension type
1639 case EXT_SBR_DATA_CRC:
1643 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1645 } else if (!ac->m4ac.sbr) {
1646 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1647 skip_bits_long(gb, 8 * cnt - 4);
1649 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1650 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1651 skip_bits_long(gb, 8 * cnt - 4);
1653 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1656 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1660 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1662 case EXT_DYNAMIC_RANGE:
1663 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1667 case EXT_DATA_ELEMENT:
1669 skip_bits_long(gb, 8 * cnt - 4);
1676 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1678 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1679 * @param coef spectral coefficients
1681 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1682 IndividualChannelStream *ics, int decode)
1684 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1686 int bottom, top, order, start, end, size, inc;
1687 float lpc[TNS_MAX_ORDER];
1688 float tmp[TNS_MAX_ORDER];
1690 for (w = 0; w < ics->num_windows; w++) {
1691 bottom = ics->num_swb;
1692 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1694 bottom = FFMAX(0, top - tns->length[w][filt]);
1695 order = tns->order[w][filt];
1700 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1702 start = ics->swb_offset[FFMIN(bottom, mmm)];
1703 end = ics->swb_offset[FFMIN( top, mmm)];
1704 if ((size = end - start) <= 0)
1706 if (tns->direction[w][filt]) {
1716 for (m = 0; m < size; m++, start += inc)
1717 for (i = 1; i <= FFMIN(m, order); i++)
1718 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1721 for (m = 0; m < size; m++, start += inc) {
1722 tmp[0] = coef[start];
1723 for (i = 1; i <= FFMIN(m, order); i++)
1724 coef[start] += tmp[i] * lpc[i - 1];
1725 for (i = order; i > 0; i--)
1726 tmp[i] = tmp[i - 1];
1734 * Apply windowing and MDCT to obtain the spectral
1735 * coefficient from the predicted sample by LTP.
1737 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1738 float *in, IndividualChannelStream *ics)
1740 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1741 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1742 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1743 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1745 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1746 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1748 memset(in, 0, 448 * sizeof(float));
1749 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1750 memcpy(in + 576, in + 576, 448 * sizeof(float));
1752 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1753 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1755 memcpy(in + 1024, in + 1024, 448 * sizeof(float));
1756 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1757 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1759 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1763 * Apply the long term prediction
1765 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1767 const LongTermPrediction *ltp = &sce->ics.ltp;
1768 const uint16_t *offsets = sce->ics.swb_offset;
1771 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1772 float *predTime = sce->ret;
1773 float *predFreq = ac->buf_mdct;
1774 int16_t num_samples = 2048;
1776 if (ltp->lag < 1024)
1777 num_samples = ltp->lag + 1024;
1778 for (i = 0; i < num_samples; i++)
1779 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1780 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1782 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1784 if (sce->tns.present)
1785 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1787 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1789 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1790 sce->coeffs[i] += predFreq[i];
1795 * Update the LTP buffer for next frame
1797 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1799 IndividualChannelStream *ics = &sce->ics;
1800 float *saved = sce->saved;
1801 float *saved_ltp = sce->coeffs;
1802 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1803 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1806 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1807 memcpy(saved_ltp, saved, 512 * sizeof(float));
1808 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1809 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1810 for (i = 0; i < 64; i++)
1811 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1812 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1813 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1814 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1815 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1816 for (i = 0; i < 64; i++)
1817 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1818 } else { // LONG_STOP or ONLY_LONG
1819 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1820 for (i = 0; i < 512; i++)
1821 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1824 memcpy(sce->ltp_state, &sce->ltp_state[1024], 1024 * sizeof(int16_t));
1825 ac->fmt_conv.float_to_int16(&(sce->ltp_state[1024]), sce->ret, 1024);
1826 ac->fmt_conv.float_to_int16(&(sce->ltp_state[2048]), saved_ltp, 1024);
1830 * Conduct IMDCT and windowing.
1832 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1834 IndividualChannelStream *ics = &sce->ics;
1835 float *in = sce->coeffs;
1836 float *out = sce->ret;
1837 float *saved = sce->saved;
1838 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1839 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1840 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1841 float *buf = ac->buf_mdct;
1842 float *temp = ac->temp;
1846 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1847 for (i = 0; i < 1024; i += 128)
1848 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1850 ac->mdct.imdct_half(&ac->mdct, buf, in);
1852 /* window overlapping
1853 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1854 * and long to short transitions are considered to be short to short
1855 * transitions. This leaves just two cases (long to long and short to short)
1856 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1858 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1859 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1860 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1862 memcpy( out, saved, 448 * sizeof(float));
1864 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1865 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1866 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1867 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1868 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1869 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1870 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1872 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1873 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1878 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1879 memcpy( saved, temp + 64, 64 * sizeof(float));
1880 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1881 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1882 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1883 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1884 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1885 memcpy( saved, buf + 512, 448 * sizeof(float));
1886 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1887 } else { // LONG_STOP or ONLY_LONG
1888 memcpy( saved, buf + 512, 512 * sizeof(float));
1893 * Apply dependent channel coupling (applied before IMDCT).
1895 * @param index index into coupling gain array
1897 static void apply_dependent_coupling(AACContext *ac,
1898 SingleChannelElement *target,
1899 ChannelElement *cce, int index)
1901 IndividualChannelStream *ics = &cce->ch[0].ics;
1902 const uint16_t *offsets = ics->swb_offset;
1903 float *dest = target->coeffs;
1904 const float *src = cce->ch[0].coeffs;
1905 int g, i, group, k, idx = 0;
1906 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1907 av_log(ac->avctx, AV_LOG_ERROR,
1908 "Dependent coupling is not supported together with LTP\n");
1911 for (g = 0; g < ics->num_window_groups; g++) {
1912 for (i = 0; i < ics->max_sfb; i++, idx++) {
1913 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1914 const float gain = cce->coup.gain[index][idx];
1915 for (group = 0; group < ics->group_len[g]; group++) {
1916 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1918 dest[group * 128 + k] += gain * src[group * 128 + k];
1923 dest += ics->group_len[g] * 128;
1924 src += ics->group_len[g] * 128;
1929 * Apply independent channel coupling (applied after IMDCT).
1931 * @param index index into coupling gain array
1933 static void apply_independent_coupling(AACContext *ac,
1934 SingleChannelElement *target,
1935 ChannelElement *cce, int index)
1938 const float gain = cce->coup.gain[index][0];
1939 const float *src = cce->ch[0].ret;
1940 float *dest = target->ret;
1941 const int len = 1024 << (ac->m4ac.sbr == 1);
1943 for (i = 0; i < len; i++)
1944 dest[i] += gain * src[i];
1948 * channel coupling transformation interface
1950 * @param apply_coupling_method pointer to (in)dependent coupling function
1952 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
1953 enum RawDataBlockType type, int elem_id,
1954 enum CouplingPoint coupling_point,
1955 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
1959 for (i = 0; i < MAX_ELEM_ID; i++) {
1960 ChannelElement *cce = ac->che[TYPE_CCE][i];
1963 if (cce && cce->coup.coupling_point == coupling_point) {
1964 ChannelCoupling *coup = &cce->coup;
1966 for (c = 0; c <= coup->num_coupled; c++) {
1967 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
1968 if (coup->ch_select[c] != 1) {
1969 apply_coupling_method(ac, &cc->ch[0], cce, index);
1970 if (coup->ch_select[c] != 0)
1973 if (coup->ch_select[c] != 2)
1974 apply_coupling_method(ac, &cc->ch[1], cce, index++);
1976 index += 1 + (coup->ch_select[c] == 3);
1983 * Convert spectral data to float samples, applying all supported tools as appropriate.
1985 static void spectral_to_sample(AACContext *ac)
1988 for (type = 3; type >= 0; type--) {
1989 for (i = 0; i < MAX_ELEM_ID; i++) {
1990 ChannelElement *che = ac->che[type][i];
1992 if (type <= TYPE_CPE)
1993 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
1994 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1995 if (che->ch[0].ics.predictor_present) {
1996 if (che->ch[0].ics.ltp.present)
1997 apply_ltp(ac, &che->ch[0]);
1998 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
1999 apply_ltp(ac, &che->ch[1]);
2002 if (che->ch[0].tns.present)
2003 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2004 if (che->ch[1].tns.present)
2005 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2006 if (type <= TYPE_CPE)
2007 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2008 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2009 imdct_and_windowing(ac, &che->ch[0]);
2010 if (ac->m4ac.object_type == AOT_AAC_LTP)
2011 update_ltp(ac, &che->ch[0]);
2012 if (type == TYPE_CPE) {
2013 imdct_and_windowing(ac, &che->ch[1]);
2014 if (ac->m4ac.object_type == AOT_AAC_LTP)
2015 update_ltp(ac, &che->ch[1]);
2017 if (ac->m4ac.sbr > 0) {
2018 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2021 if (type <= TYPE_CCE)
2022 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2028 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2031 AACADTSHeaderInfo hdr_info;
2033 size = ff_aac_parse_header(gb, &hdr_info);
2035 if (ac->output_configured != OC_LOCKED && hdr_info.chan_config) {
2036 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2037 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2038 ac->m4ac.chan_config = hdr_info.chan_config;
2039 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2041 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config, OC_TRIAL_FRAME))
2043 } else if (ac->output_configured != OC_LOCKED) {
2044 ac->output_configured = OC_NONE;
2046 if (ac->output_configured != OC_LOCKED) {
2050 ac->m4ac.sample_rate = hdr_info.sample_rate;
2051 ac->m4ac.sampling_index = hdr_info.sampling_index;
2052 ac->m4ac.object_type = hdr_info.object_type;
2053 if (!ac->avctx->sample_rate)
2054 ac->avctx->sample_rate = hdr_info.sample_rate;
2055 if (hdr_info.num_aac_frames == 1) {
2056 if (!hdr_info.crc_absent)
2059 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2066 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2067 int *data_size, GetBitContext *gb)
2069 AACContext *ac = avctx->priv_data;
2070 ChannelElement *che = NULL, *che_prev = NULL;
2071 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2072 int err, elem_id, data_size_tmp;
2073 int samples = 0, multiplier;
2075 if (show_bits(gb, 12) == 0xfff) {
2076 if (parse_adts_frame_header(ac, gb) < 0) {
2077 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2080 if (ac->m4ac.sampling_index > 12) {
2081 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2086 ac->tags_mapped = 0;
2088 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2089 elem_id = get_bits(gb, 4);
2091 if (elem_type < TYPE_DSE) {
2092 if (!(che=get_che(ac, elem_type, elem_id))) {
2093 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2094 elem_type, elem_id);
2100 switch (elem_type) {
2103 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2107 err = decode_cpe(ac, gb, che);
2111 err = decode_cce(ac, gb, che);
2115 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2119 err = skip_data_stream_element(ac, gb);
2123 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2124 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2125 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2127 if (ac->output_configured > OC_TRIAL_PCE)
2128 av_log(avctx, AV_LOG_ERROR,
2129 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2131 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2137 elem_id += get_bits(gb, 8) - 1;
2138 if (get_bits_left(gb) < 8 * elem_id) {
2139 av_log(avctx, AV_LOG_ERROR, overread_err);
2143 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2144 err = 0; /* FIXME */
2148 err = -1; /* should not happen, but keeps compiler happy */
2153 elem_type_prev = elem_type;
2158 if (get_bits_left(gb) < 3) {
2159 av_log(avctx, AV_LOG_ERROR, overread_err);
2164 spectral_to_sample(ac);
2166 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2167 samples <<= multiplier;
2168 if (ac->output_configured < OC_LOCKED) {
2169 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2170 avctx->frame_size = samples;
2173 data_size_tmp = samples * avctx->channels;
2174 data_size_tmp *= avctx->sample_fmt == AV_SAMPLE_FMT_FLT ? sizeof(float) : sizeof(int16_t);
2175 if (*data_size < data_size_tmp) {
2176 av_log(avctx, AV_LOG_ERROR,
2177 "Output buffer too small (%d) or trying to output too many samples (%d) for this frame.\n",
2178 *data_size, data_size_tmp);
2181 *data_size = data_size_tmp;
2184 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
2185 float_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2187 ac->fmt_conv.float_to_int16_interleave(data, (const float **)ac->output_data, samples, avctx->channels);
2190 if (ac->output_configured)
2191 ac->output_configured = OC_LOCKED;
2196 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2197 int *data_size, AVPacket *avpkt)
2199 const uint8_t *buf = avpkt->data;
2200 int buf_size = avpkt->size;
2206 init_get_bits(&gb, buf, buf_size * 8);
2208 if ((err = aac_decode_frame_int(avctx, data, data_size, &gb)) < 0)
2211 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2212 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2213 if (buf[buf_offset])
2216 return buf_size > buf_offset ? buf_consumed : buf_size;
2219 static av_cold int aac_decode_close(AVCodecContext *avctx)
2221 AACContext *ac = avctx->priv_data;
2224 for (i = 0; i < MAX_ELEM_ID; i++) {
2225 for (type = 0; type < 4; type++) {
2226 if (ac->che[type][i])
2227 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2228 av_freep(&ac->che[type][i]);
2232 ff_mdct_end(&ac->mdct);
2233 ff_mdct_end(&ac->mdct_small);
2234 ff_mdct_end(&ac->mdct_ltp);
2239 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2241 struct LATMContext {
2242 AACContext aac_ctx; ///< containing AACContext
2243 int initialized; ///< initilized after a valid extradata was seen
2246 int audio_mux_version_A; ///< LATM syntax version
2247 int frame_length_type; ///< 0/1 variable/fixed frame length
2248 int frame_length; ///< frame length for fixed frame length
2251 static inline uint32_t latm_get_value(GetBitContext *b)
2253 int length = get_bits(b, 2);
2255 return get_bits_long(b, (length+1)*8);
2258 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2261 AVCodecContext *avctx = latmctx->aac_ctx.avctx;
2262 MPEG4AudioConfig m4ac;
2263 int config_start_bit = get_bits_count(gb);
2264 int bits_consumed, esize;
2266 if (config_start_bit % 8) {
2267 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2268 "config not byte aligned.\n", 1);
2269 return AVERROR_INVALIDDATA;
2272 decode_audio_specific_config(NULL, avctx, &m4ac,
2273 gb->buffer + (config_start_bit / 8),
2274 get_bits_left(gb) / 8);
2276 if (bits_consumed < 0)
2277 return AVERROR_INVALIDDATA;
2279 esize = (bits_consumed+7) / 8;
2281 if (avctx->extradata_size <= esize) {
2282 av_free(avctx->extradata);
2283 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2284 if (!avctx->extradata)
2285 return AVERROR(ENOMEM);
2288 avctx->extradata_size = esize;
2289 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2290 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2292 skip_bits_long(gb, bits_consumed);
2295 return bits_consumed;
2298 static int read_stream_mux_config(struct LATMContext *latmctx,
2301 int ret, audio_mux_version = get_bits(gb, 1);
2303 latmctx->audio_mux_version_A = 0;
2304 if (audio_mux_version)
2305 latmctx->audio_mux_version_A = get_bits(gb, 1);
2307 if (!latmctx->audio_mux_version_A) {
2309 if (audio_mux_version)
2310 latm_get_value(gb); // taraFullness
2312 skip_bits(gb, 1); // allStreamSameTimeFraming
2313 skip_bits(gb, 6); // numSubFrames
2315 if (get_bits(gb, 4)) { // numPrograms
2316 av_log_missing_feature(latmctx->aac_ctx.avctx,
2317 "multiple programs are not supported\n", 1);
2318 return AVERROR_PATCHWELCOME;
2321 // for each program (which there is only on in DVB)
2323 // for each layer (which there is only on in DVB)
2324 if (get_bits(gb, 3)) { // numLayer
2325 av_log_missing_feature(latmctx->aac_ctx.avctx,
2326 "multiple layers are not supported\n", 1);
2327 return AVERROR_PATCHWELCOME;
2330 // for all but first stream: use_same_config = get_bits(gb, 1);
2331 if (!audio_mux_version) {
2332 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2335 int ascLen = latm_get_value(gb);
2336 if ((ret = latm_decode_audio_specific_config(latmctx, gb)) < 0)
2339 skip_bits_long(gb, ascLen);
2342 latmctx->frame_length_type = get_bits(gb, 3);
2343 switch (latmctx->frame_length_type) {
2345 skip_bits(gb, 8); // latmBufferFullness
2348 latmctx->frame_length = get_bits(gb, 9);
2353 skip_bits(gb, 6); // CELP frame length table index
2357 skip_bits(gb, 1); // HVXC frame length table index
2361 if (get_bits(gb, 1)) { // other data
2362 if (audio_mux_version) {
2363 latm_get_value(gb); // other_data_bits
2367 esc = get_bits(gb, 1);
2373 if (get_bits(gb, 1)) // crc present
2374 skip_bits(gb, 8); // config_crc
2380 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2384 if (ctx->frame_length_type == 0) {
2385 int mux_slot_length = 0;
2387 tmp = get_bits(gb, 8);
2388 mux_slot_length += tmp;
2389 } while (tmp == 255);
2390 return mux_slot_length;
2391 } else if (ctx->frame_length_type == 1) {
2392 return ctx->frame_length;
2393 } else if (ctx->frame_length_type == 3 ||
2394 ctx->frame_length_type == 5 ||
2395 ctx->frame_length_type == 7) {
2396 skip_bits(gb, 2); // mux_slot_length_coded
2401 static int read_audio_mux_element(struct LATMContext *latmctx,
2405 uint8_t use_same_mux = get_bits(gb, 1);
2406 if (!use_same_mux) {
2407 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2409 } else if (!latmctx->aac_ctx.avctx->extradata) {
2410 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2411 "no decoder config found\n");
2412 return AVERROR(EAGAIN);
2414 if (latmctx->audio_mux_version_A == 0) {
2415 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2416 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2417 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2418 return AVERROR_INVALIDDATA;
2419 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2420 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2421 "frame length mismatch %d << %d\n",
2422 mux_slot_length_bytes * 8, get_bits_left(gb));
2423 return AVERROR_INVALIDDATA;
2430 static int latm_decode_frame(AVCodecContext *avctx, void *out, int *out_size,
2433 struct LATMContext *latmctx = avctx->priv_data;
2437 if (avpkt->size == 0)
2440 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2442 // check for LOAS sync word
2443 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2444 return AVERROR_INVALIDDATA;
2446 muxlength = get_bits(&gb, 13) + 3;
2447 // not enough data, the parser should have sorted this
2448 if (muxlength > avpkt->size)
2449 return AVERROR_INVALIDDATA;
2451 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2454 if (!latmctx->initialized) {
2455 if (!avctx->extradata) {
2459 if ((err = aac_decode_init(avctx)) < 0)
2461 latmctx->initialized = 1;
2465 if (show_bits(&gb, 12) == 0xfff) {
2466 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2467 "ADTS header detected, probably as result of configuration "
2469 return AVERROR_INVALIDDATA;
2472 if ((err = aac_decode_frame_int(avctx, out, out_size, &gb)) < 0)
2478 av_cold static int latm_decode_init(AVCodecContext *avctx)
2480 struct LATMContext *latmctx = avctx->priv_data;
2483 ret = aac_decode_init(avctx);
2485 if (avctx->extradata_size > 0) {
2486 latmctx->initialized = !ret;
2488 latmctx->initialized = 0;
2495 AVCodec ff_aac_decoder = {
2504 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2505 .sample_fmts = (const enum AVSampleFormat[]) {
2506 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2508 .channel_layouts = aac_channel_layout,
2512 Note: This decoder filter is intended to decode LATM streams transferred
2513 in MPEG transport streams which only contain one program.
2514 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2516 AVCodec ff_aac_latm_decoder = {
2518 .type = AVMEDIA_TYPE_AUDIO,
2519 .id = CODEC_ID_AAC_LATM,
2520 .priv_data_size = sizeof(struct LATMContext),
2521 .init = latm_decode_init,
2522 .close = aac_decode_close,
2523 .decode = latm_decode_frame,
2524 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2525 .sample_fmts = (const enum AVSampleFormat[]) {
2526 AV_SAMPLE_FMT_S16,AV_SAMPLE_FMT_FLT,AV_SAMPLE_FMT_NONE
2528 .channel_layouts = aac_channel_layout,