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 Libav.
12 * Libav 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 * Libav 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 Libav; 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 int count_channels(uint8_t (*layout)[3], int tags)
120 for (i = 0; i < tags; i++) {
121 int syn_ele = layout[i][0];
122 int pos = layout[i][2];
123 sum += (1 + (syn_ele == TYPE_CPE)) *
124 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
130 * Check for the channel element in the current channel position configuration.
131 * If it exists, make sure the appropriate element is allocated and map the
132 * channel order to match the internal Libav channel layout.
134 * @param che_pos current channel position configuration
135 * @param type channel element type
136 * @param id channel element id
137 * @param channels count of the number of channels in the configuration
139 * @return Returns error status. 0 - OK, !0 - error
141 static av_cold int che_configure(AACContext *ac,
142 enum ChannelPosition che_pos,
143 int type, int id, int *channels)
146 if (!ac->che[type][id]) {
147 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
148 return AVERROR(ENOMEM);
149 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
151 if (type != TYPE_CCE) {
152 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
153 if (type == TYPE_CPE ||
154 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
155 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
159 if (ac->che[type][id])
160 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
161 av_freep(&ac->che[type][id]);
166 struct elem_to_channel {
167 uint64_t av_position;
170 uint8_t aac_position;
173 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
174 uint8_t (*layout_map)[3], int offset, int tags, uint64_t left,
175 uint64_t right, int pos)
177 if (layout_map[offset][0] == TYPE_CPE) {
178 e2c_vec[offset] = (struct elem_to_channel) {
179 .av_position = left | right, .syn_ele = TYPE_CPE,
180 .elem_id = layout_map[offset ][1], .aac_position = pos };
183 e2c_vec[offset] = (struct elem_to_channel) {
184 .av_position = left, .syn_ele = TYPE_SCE,
185 .elem_id = layout_map[offset ][1], .aac_position = pos };
186 e2c_vec[offset + 1] = (struct elem_to_channel) {
187 .av_position = right, .syn_ele = TYPE_SCE,
188 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
193 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
194 int num_pos_channels = 0;
198 for (i = *current; i < tags; i++) {
199 if (layout_map[i][2] != pos)
201 if (layout_map[i][0] == TYPE_CPE) {
203 if (pos == AAC_CHANNEL_FRONT || !first_cpe) {
209 num_pos_channels += 2;
217 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
220 return num_pos_channels;
223 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
225 int i, n, total_non_cc_elements;
226 struct elem_to_channel e2c_vec[MAX_ELEM_ID] = {{ 0 }};
227 int num_front_channels, num_side_channels, num_back_channels;
232 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
233 if (num_front_channels < 0)
236 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
237 if (num_side_channels < 0)
240 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
241 if (num_back_channels < 0)
245 if (num_front_channels & 1) {
246 e2c_vec[i] = (struct elem_to_channel) {
247 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
248 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
250 num_front_channels--;
252 if (num_front_channels >= 4) {
253 i += assign_pair(e2c_vec, layout_map, i, tags,
254 AV_CH_FRONT_LEFT_OF_CENTER,
255 AV_CH_FRONT_RIGHT_OF_CENTER,
257 num_front_channels -= 2;
259 if (num_front_channels >= 2) {
260 i += assign_pair(e2c_vec, layout_map, i, tags,
264 num_front_channels -= 2;
266 while (num_front_channels >= 2) {
267 i += assign_pair(e2c_vec, layout_map, i, tags,
271 num_front_channels -= 2;
274 if (num_side_channels >= 2) {
275 i += assign_pair(e2c_vec, layout_map, i, tags,
279 num_side_channels -= 2;
281 while (num_side_channels >= 2) {
282 i += assign_pair(e2c_vec, layout_map, i, tags,
286 num_side_channels -= 2;
289 while (num_back_channels >= 4) {
290 i += assign_pair(e2c_vec, layout_map, i, tags,
294 num_back_channels -= 2;
296 if (num_back_channels >= 2) {
297 i += assign_pair(e2c_vec, layout_map, i, tags,
301 num_back_channels -= 2;
303 if (num_back_channels) {
304 e2c_vec[i] = (struct elem_to_channel) {
305 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
306 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
311 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
312 e2c_vec[i] = (struct elem_to_channel) {
313 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
314 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
317 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
318 e2c_vec[i] = (struct elem_to_channel) {
319 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
320 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
324 // Must choose a stable sort
325 total_non_cc_elements = n = i;
328 for (i = 1; i < n; i++) {
329 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
330 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
338 for (i = 0; i < total_non_cc_elements; i++) {
339 layout_map[i][0] = e2c_vec[i].syn_ele;
340 layout_map[i][1] = e2c_vec[i].elem_id;
341 layout_map[i][2] = e2c_vec[i].aac_position;
342 if (e2c_vec[i].av_position != UINT64_MAX) {
343 layout |= e2c_vec[i].av_position;
351 * Configure output channel order based on the current program configuration element.
353 * @return Returns error status. 0 - OK, !0 - error
355 static av_cold int output_configure(AACContext *ac,
356 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
357 int channel_config, enum OCStatus oc_type)
359 AVCodecContext *avctx = ac->avctx;
360 int i, channels = 0, ret;
363 if (ac->layout_map != layout_map) {
364 memcpy(ac->layout_map, layout_map, tags * sizeof(layout_map[0]));
365 ac->layout_map_tags = tags;
368 // Try to sniff a reasonable channel order, otherwise output the
369 // channels in the order the PCE declared them.
370 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
371 layout = sniff_channel_order(layout_map, tags);
372 for (i = 0; i < tags; i++) {
373 int type = layout_map[i][0];
374 int id = layout_map[i][1];
375 int position = layout_map[i][2];
376 // Allocate or free elements depending on if they are in the
377 // current program configuration.
378 ret = che_configure(ac, position, type, id, &channels);
383 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
384 avctx->channel_layout = layout;
385 avctx->channels = channels;
386 ac->output_configured = oc_type;
392 * Set up channel positions based on a default channel configuration
393 * as specified in table 1.17.
395 * @return Returns error status. 0 - OK, !0 - error
397 static av_cold int set_default_channel_config(AVCodecContext *avctx,
398 uint8_t (*layout_map)[3],
402 if (channel_config < 1 || channel_config > 7) {
403 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
407 *tags = tags_per_config[channel_config];
408 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
412 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
414 // For PCE based channel configurations map the channels solely based on tags.
415 if (!ac->m4ac.chan_config) {
416 return ac->tag_che_map[type][elem_id];
418 // Allow single CPE stereo files to be signalled with mono configuration.
419 if (!ac->tags_mapped && type == TYPE_CPE && ac->m4ac.chan_config == 1) {
420 uint8_t layout_map[MAX_ELEM_ID*4][3];
423 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
426 if (output_configure(ac, layout_map, layout_map_tags,
427 2, OC_TRIAL_FRAME) < 0)
430 ac->m4ac.chan_config = 2;
432 // For indexed channel configurations map the channels solely based on position.
433 switch (ac->m4ac.chan_config) {
435 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
437 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
440 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
441 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
442 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
443 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
445 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
448 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
450 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
453 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
455 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
459 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
461 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
462 } else if (ac->m4ac.chan_config == 2) {
466 if (!ac->tags_mapped && type == TYPE_SCE) {
468 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
476 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
478 * @param type speaker type/position for these channels
480 static void decode_channel_map(uint8_t layout_map[][3],
481 enum ChannelPosition type,
482 GetBitContext *gb, int n)
485 enum RawDataBlockType syn_ele;
487 case AAC_CHANNEL_FRONT:
488 case AAC_CHANNEL_BACK:
489 case AAC_CHANNEL_SIDE:
490 syn_ele = get_bits1(gb);
496 case AAC_CHANNEL_LFE:
500 layout_map[0][0] = syn_ele;
501 layout_map[0][1] = get_bits(gb, 4);
502 layout_map[0][2] = type;
508 * Decode program configuration element; reference: table 4.2.
510 * @return Returns error status. 0 - OK, !0 - error
512 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
513 uint8_t (*layout_map)[3],
516 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
520 skip_bits(gb, 2); // object_type
522 sampling_index = get_bits(gb, 4);
523 if (m4ac->sampling_index != sampling_index)
524 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");
526 num_front = get_bits(gb, 4);
527 num_side = get_bits(gb, 4);
528 num_back = get_bits(gb, 4);
529 num_lfe = get_bits(gb, 2);
530 num_assoc_data = get_bits(gb, 3);
531 num_cc = get_bits(gb, 4);
534 skip_bits(gb, 4); // mono_mixdown_tag
536 skip_bits(gb, 4); // stereo_mixdown_tag
539 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
541 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
543 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
545 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
547 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
550 skip_bits_long(gb, 4 * num_assoc_data);
552 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
557 /* comment field, first byte is length */
558 comment_len = get_bits(gb, 8) * 8;
559 if (get_bits_left(gb) < comment_len) {
560 av_log(avctx, AV_LOG_ERROR, overread_err);
563 skip_bits_long(gb, comment_len);
568 * Decode GA "General Audio" specific configuration; reference: table 4.1.
570 * @param ac pointer to AACContext, may be null
571 * @param avctx pointer to AVCCodecContext, used for logging
573 * @return Returns error status. 0 - OK, !0 - error
575 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
577 MPEG4AudioConfig *m4ac,
580 int extension_flag, ret;
581 uint8_t layout_map[MAX_ELEM_ID*4][3];
584 if (get_bits1(gb)) { // frameLengthFlag
585 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
589 if (get_bits1(gb)) // dependsOnCoreCoder
590 skip_bits(gb, 14); // coreCoderDelay
591 extension_flag = get_bits1(gb);
593 if (m4ac->object_type == AOT_AAC_SCALABLE ||
594 m4ac->object_type == AOT_ER_AAC_SCALABLE)
595 skip_bits(gb, 3); // layerNr
597 if (channel_config == 0) {
598 skip_bits(gb, 4); // element_instance_tag
599 tags = decode_pce(avctx, m4ac, layout_map, gb);
603 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
607 if (count_channels(layout_map, tags) > 1) {
609 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
612 if (ac && (ret = output_configure(ac, layout_map, tags,
613 channel_config, OC_GLOBAL_HDR)))
616 if (extension_flag) {
617 switch (m4ac->object_type) {
619 skip_bits(gb, 5); // numOfSubFrame
620 skip_bits(gb, 11); // layer_length
624 case AOT_ER_AAC_SCALABLE:
626 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
627 * aacScalefactorDataResilienceFlag
628 * aacSpectralDataResilienceFlag
632 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
638 * Decode audio specific configuration; reference: table 1.13.
640 * @param ac pointer to AACContext, may be null
641 * @param avctx pointer to AVCCodecContext, used for logging
642 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
643 * @param data pointer to buffer holding an audio specific config
644 * @param bit_size size of audio specific config or data in bits
645 * @param sync_extension look for an appended sync extension
647 * @return Returns error status or number of consumed bits. <0 - error
649 static int decode_audio_specific_config(AACContext *ac,
650 AVCodecContext *avctx,
651 MPEG4AudioConfig *m4ac,
652 const uint8_t *data, int bit_size,
658 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
659 for (i = 0; i < avctx->extradata_size; i++)
660 av_dlog(avctx, "%02x ", avctx->extradata[i]);
661 av_dlog(avctx, "\n");
663 init_get_bits(&gb, data, bit_size);
665 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
667 if (m4ac->sampling_index > 12) {
668 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
672 skip_bits_long(&gb, i);
674 switch (m4ac->object_type) {
678 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
682 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
683 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
687 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
688 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
689 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
691 return get_bits_count(&gb);
695 * linear congruential pseudorandom number generator
697 * @param previous_val pointer to the current state of the generator
699 * @return Returns a 32-bit pseudorandom integer
701 static av_always_inline int lcg_random(int previous_val)
703 return previous_val * 1664525 + 1013904223;
706 static av_always_inline void reset_predict_state(PredictorState *ps)
716 static void reset_all_predictors(PredictorState *ps)
719 for (i = 0; i < MAX_PREDICTORS; i++)
720 reset_predict_state(&ps[i]);
723 static int sample_rate_idx (int rate)
725 if (92017 <= rate) return 0;
726 else if (75132 <= rate) return 1;
727 else if (55426 <= rate) return 2;
728 else if (46009 <= rate) return 3;
729 else if (37566 <= rate) return 4;
730 else if (27713 <= rate) return 5;
731 else if (23004 <= rate) return 6;
732 else if (18783 <= rate) return 7;
733 else if (13856 <= rate) return 8;
734 else if (11502 <= rate) return 9;
735 else if (9391 <= rate) return 10;
739 static void reset_predictor_group(PredictorState *ps, int group_num)
742 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
743 reset_predict_state(&ps[i]);
746 #define AAC_INIT_VLC_STATIC(num, size) \
747 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
748 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
749 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
752 static av_cold int aac_decode_init(AVCodecContext *avctx)
754 AACContext *ac = avctx->priv_data;
755 float output_scale_factor;
758 ac->m4ac.sample_rate = avctx->sample_rate;
760 if (avctx->extradata_size > 0) {
761 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
763 avctx->extradata_size*8, 1) < 0)
767 uint8_t layout_map[MAX_ELEM_ID*4][3];
770 sr = sample_rate_idx(avctx->sample_rate);
771 ac->m4ac.sampling_index = sr;
772 ac->m4ac.channels = avctx->channels;
776 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
777 if (ff_mpeg4audio_channels[i] == avctx->channels)
779 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
782 ac->m4ac.chan_config = i;
784 if (ac->m4ac.chan_config) {
785 int ret = set_default_channel_config(avctx, layout_map,
786 &layout_map_tags, ac->m4ac.chan_config);
788 output_configure(ac, layout_map, layout_map_tags,
789 ac->m4ac.chan_config, OC_GLOBAL_HDR);
790 else if (avctx->err_recognition & AV_EF_EXPLODE)
791 return AVERROR_INVALIDDATA;
795 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
796 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
797 output_scale_factor = 1.0 / 32768.0;
799 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
800 output_scale_factor = 1.0;
803 AAC_INIT_VLC_STATIC( 0, 304);
804 AAC_INIT_VLC_STATIC( 1, 270);
805 AAC_INIT_VLC_STATIC( 2, 550);
806 AAC_INIT_VLC_STATIC( 3, 300);
807 AAC_INIT_VLC_STATIC( 4, 328);
808 AAC_INIT_VLC_STATIC( 5, 294);
809 AAC_INIT_VLC_STATIC( 6, 306);
810 AAC_INIT_VLC_STATIC( 7, 268);
811 AAC_INIT_VLC_STATIC( 8, 510);
812 AAC_INIT_VLC_STATIC( 9, 366);
813 AAC_INIT_VLC_STATIC(10, 462);
817 ff_dsputil_init(&ac->dsp, avctx);
818 ff_fmt_convert_init(&ac->fmt_conv, avctx);
820 ac->random_state = 0x1f2e3d4c;
824 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
825 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
826 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
829 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
830 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
831 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
832 // window initialization
833 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
834 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
835 ff_init_ff_sine_windows(10);
836 ff_init_ff_sine_windows( 7);
840 avcodec_get_frame_defaults(&ac->frame);
841 avctx->coded_frame = &ac->frame;
847 * Skip data_stream_element; reference: table 4.10.
849 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
851 int byte_align = get_bits1(gb);
852 int count = get_bits(gb, 8);
854 count += get_bits(gb, 8);
858 if (get_bits_left(gb) < 8 * count) {
859 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
862 skip_bits_long(gb, 8 * count);
866 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
871 ics->predictor_reset_group = get_bits(gb, 5);
872 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
873 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
877 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
878 ics->prediction_used[sfb] = get_bits1(gb);
884 * Decode Long Term Prediction data; reference: table 4.xx.
886 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
887 GetBitContext *gb, uint8_t max_sfb)
891 ltp->lag = get_bits(gb, 11);
892 ltp->coef = ltp_coef[get_bits(gb, 3)];
893 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
894 ltp->used[sfb] = get_bits1(gb);
898 * Decode Individual Channel Stream info; reference: table 4.6.
900 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
904 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
905 return AVERROR_INVALIDDATA;
907 ics->window_sequence[1] = ics->window_sequence[0];
908 ics->window_sequence[0] = get_bits(gb, 2);
909 ics->use_kb_window[1] = ics->use_kb_window[0];
910 ics->use_kb_window[0] = get_bits1(gb);
911 ics->num_window_groups = 1;
912 ics->group_len[0] = 1;
913 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
915 ics->max_sfb = get_bits(gb, 4);
916 for (i = 0; i < 7; i++) {
918 ics->group_len[ics->num_window_groups - 1]++;
920 ics->num_window_groups++;
921 ics->group_len[ics->num_window_groups - 1] = 1;
924 ics->num_windows = 8;
925 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
926 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
927 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
928 ics->predictor_present = 0;
930 ics->max_sfb = get_bits(gb, 6);
931 ics->num_windows = 1;
932 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
933 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
934 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
935 ics->predictor_present = get_bits1(gb);
936 ics->predictor_reset_group = 0;
937 if (ics->predictor_present) {
938 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
939 if (decode_prediction(ac, ics, gb)) {
940 return AVERROR_INVALIDDATA;
942 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
943 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
944 return AVERROR_INVALIDDATA;
946 if ((ics->ltp.present = get_bits(gb, 1)))
947 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
952 if (ics->max_sfb > ics->num_swb) {
953 av_log(ac->avctx, AV_LOG_ERROR,
954 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
955 ics->max_sfb, ics->num_swb);
956 return AVERROR_INVALIDDATA;
963 * Decode band types (section_data payload); reference: table 4.46.
965 * @param band_type array of the used band type
966 * @param band_type_run_end array of the last scalefactor band of a band type run
968 * @return Returns error status. 0 - OK, !0 - error
970 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
971 int band_type_run_end[120], GetBitContext *gb,
972 IndividualChannelStream *ics)
975 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
976 for (g = 0; g < ics->num_window_groups; g++) {
978 while (k < ics->max_sfb) {
979 uint8_t sect_end = k;
981 int sect_band_type = get_bits(gb, 4);
982 if (sect_band_type == 12) {
983 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
987 sect_len_incr = get_bits(gb, bits);
988 sect_end += sect_len_incr;
989 if (get_bits_left(gb) < 0) {
990 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
993 if (sect_end > ics->max_sfb) {
994 av_log(ac->avctx, AV_LOG_ERROR,
995 "Number of bands (%d) exceeds limit (%d).\n",
996 sect_end, ics->max_sfb);
999 } while (sect_len_incr == (1 << bits) - 1);
1000 for (; k < sect_end; k++) {
1001 band_type [idx] = sect_band_type;
1002 band_type_run_end[idx++] = sect_end;
1010 * Decode scalefactors; reference: table 4.47.
1012 * @param global_gain first scalefactor value as scalefactors are differentially coded
1013 * @param band_type array of the used band type
1014 * @param band_type_run_end array of the last scalefactor band of a band type run
1015 * @param sf array of scalefactors or intensity stereo positions
1017 * @return Returns error status. 0 - OK, !0 - error
1019 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1020 unsigned int global_gain,
1021 IndividualChannelStream *ics,
1022 enum BandType band_type[120],
1023 int band_type_run_end[120])
1026 int offset[3] = { global_gain, global_gain - 90, 0 };
1029 for (g = 0; g < ics->num_window_groups; g++) {
1030 for (i = 0; i < ics->max_sfb;) {
1031 int run_end = band_type_run_end[idx];
1032 if (band_type[idx] == ZERO_BT) {
1033 for (; i < run_end; i++, idx++)
1035 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1036 for (; i < run_end; i++, idx++) {
1037 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1038 clipped_offset = av_clip(offset[2], -155, 100);
1039 if (offset[2] != clipped_offset) {
1040 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1041 "position clipped (%d -> %d).\nIf you heard an "
1042 "audible artifact, there may be a bug in the "
1043 "decoder. ", offset[2], clipped_offset);
1045 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1047 } else if (band_type[idx] == NOISE_BT) {
1048 for (; i < run_end; i++, idx++) {
1049 if (noise_flag-- > 0)
1050 offset[1] += get_bits(gb, 9) - 256;
1052 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1053 clipped_offset = av_clip(offset[1], -100, 155);
1054 if (offset[1] != clipped_offset) {
1055 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1056 "(%d -> %d).\nIf you heard an audible "
1057 "artifact, there may be a bug in the decoder. ",
1058 offset[1], clipped_offset);
1060 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1063 for (; i < run_end; i++, idx++) {
1064 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1065 if (offset[0] > 255U) {
1066 av_log(ac->avctx, AV_LOG_ERROR,
1067 "Scalefactor (%d) out of range.\n", offset[0]);
1070 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1079 * Decode pulse data; reference: table 4.7.
1081 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1082 const uint16_t *swb_offset, int num_swb)
1085 pulse->num_pulse = get_bits(gb, 2) + 1;
1086 pulse_swb = get_bits(gb, 6);
1087 if (pulse_swb >= num_swb)
1089 pulse->pos[0] = swb_offset[pulse_swb];
1090 pulse->pos[0] += get_bits(gb, 5);
1091 if (pulse->pos[0] > 1023)
1093 pulse->amp[0] = get_bits(gb, 4);
1094 for (i = 1; i < pulse->num_pulse; i++) {
1095 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1096 if (pulse->pos[i] > 1023)
1098 pulse->amp[i] = get_bits(gb, 4);
1104 * Decode Temporal Noise Shaping data; reference: table 4.48.
1106 * @return Returns error status. 0 - OK, !0 - error
1108 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1109 GetBitContext *gb, const IndividualChannelStream *ics)
1111 int w, filt, i, coef_len, coef_res, coef_compress;
1112 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1113 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1114 for (w = 0; w < ics->num_windows; w++) {
1115 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1116 coef_res = get_bits1(gb);
1118 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1120 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1122 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1123 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1124 tns->order[w][filt], tns_max_order);
1125 tns->order[w][filt] = 0;
1128 if (tns->order[w][filt]) {
1129 tns->direction[w][filt] = get_bits1(gb);
1130 coef_compress = get_bits1(gb);
1131 coef_len = coef_res + 3 - coef_compress;
1132 tmp2_idx = 2 * coef_compress + coef_res;
1134 for (i = 0; i < tns->order[w][filt]; i++)
1135 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1144 * Decode Mid/Side data; reference: table 4.54.
1146 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1147 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1148 * [3] reserved for scalable AAC
1150 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1154 if (ms_present == 1) {
1155 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1156 cpe->ms_mask[idx] = get_bits1(gb);
1157 } else if (ms_present == 2) {
1158 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1163 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1167 *dst++ = v[idx & 15] * s;
1168 *dst++ = v[idx>>4 & 15] * s;
1174 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1178 *dst++ = v[idx & 3] * s;
1179 *dst++ = v[idx>>2 & 3] * s;
1180 *dst++ = v[idx>>4 & 3] * s;
1181 *dst++ = v[idx>>6 & 3] * s;
1187 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1188 unsigned sign, const float *scale)
1190 union av_intfloat32 s0, s1;
1192 s0.f = s1.f = *scale;
1193 s0.i ^= sign >> 1 << 31;
1196 *dst++ = v[idx & 15] * s0.f;
1197 *dst++ = v[idx>>4 & 15] * s1.f;
1204 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1205 unsigned sign, const float *scale)
1207 unsigned nz = idx >> 12;
1208 union av_intfloat32 s = { .f = *scale };
1209 union av_intfloat32 t;
1211 t.i = s.i ^ (sign & 1U<<31);
1212 *dst++ = v[idx & 3] * t.f;
1214 sign <<= nz & 1; nz >>= 1;
1215 t.i = s.i ^ (sign & 1U<<31);
1216 *dst++ = v[idx>>2 & 3] * t.f;
1218 sign <<= nz & 1; nz >>= 1;
1219 t.i = s.i ^ (sign & 1U<<31);
1220 *dst++ = v[idx>>4 & 3] * t.f;
1222 sign <<= nz & 1; nz >>= 1;
1223 t.i = s.i ^ (sign & 1U<<31);
1224 *dst++ = v[idx>>6 & 3] * t.f;
1231 * Decode spectral data; reference: table 4.50.
1232 * Dequantize and scale spectral data; reference: 4.6.3.3.
1234 * @param coef array of dequantized, scaled spectral data
1235 * @param sf array of scalefactors or intensity stereo positions
1236 * @param pulse_present set if pulses are present
1237 * @param pulse pointer to pulse data struct
1238 * @param band_type array of the used band type
1240 * @return Returns error status. 0 - OK, !0 - error
1242 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1243 GetBitContext *gb, const float sf[120],
1244 int pulse_present, const Pulse *pulse,
1245 const IndividualChannelStream *ics,
1246 enum BandType band_type[120])
1248 int i, k, g, idx = 0;
1249 const int c = 1024 / ics->num_windows;
1250 const uint16_t *offsets = ics->swb_offset;
1251 float *coef_base = coef;
1253 for (g = 0; g < ics->num_windows; g++)
1254 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1256 for (g = 0; g < ics->num_window_groups; g++) {
1257 unsigned g_len = ics->group_len[g];
1259 for (i = 0; i < ics->max_sfb; i++, idx++) {
1260 const unsigned cbt_m1 = band_type[idx] - 1;
1261 float *cfo = coef + offsets[i];
1262 int off_len = offsets[i + 1] - offsets[i];
1265 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1266 for (group = 0; group < g_len; group++, cfo+=128) {
1267 memset(cfo, 0, off_len * sizeof(float));
1269 } else if (cbt_m1 == NOISE_BT - 1) {
1270 for (group = 0; group < g_len; group++, cfo+=128) {
1274 for (k = 0; k < off_len; k++) {
1275 ac->random_state = lcg_random(ac->random_state);
1276 cfo[k] = ac->random_state;
1279 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1280 scale = sf[idx] / sqrtf(band_energy);
1281 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1284 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1285 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1286 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1287 OPEN_READER(re, gb);
1289 switch (cbt_m1 >> 1) {
1291 for (group = 0; group < g_len; group++, cfo+=128) {
1299 UPDATE_CACHE(re, gb);
1300 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1301 cb_idx = cb_vector_idx[code];
1302 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1308 for (group = 0; group < g_len; group++, cfo+=128) {
1318 UPDATE_CACHE(re, gb);
1319 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1320 cb_idx = cb_vector_idx[code];
1321 nnz = cb_idx >> 8 & 15;
1322 bits = nnz ? GET_CACHE(re, gb) : 0;
1323 LAST_SKIP_BITS(re, gb, nnz);
1324 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1330 for (group = 0; group < g_len; group++, cfo+=128) {
1338 UPDATE_CACHE(re, gb);
1339 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1340 cb_idx = cb_vector_idx[code];
1341 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1348 for (group = 0; group < g_len; group++, cfo+=128) {
1358 UPDATE_CACHE(re, gb);
1359 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1360 cb_idx = cb_vector_idx[code];
1361 nnz = cb_idx >> 8 & 15;
1362 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1363 LAST_SKIP_BITS(re, gb, nnz);
1364 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1370 for (group = 0; group < g_len; group++, cfo+=128) {
1372 uint32_t *icf = (uint32_t *) cf;
1382 UPDATE_CACHE(re, gb);
1383 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1391 cb_idx = cb_vector_idx[code];
1394 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1395 LAST_SKIP_BITS(re, gb, nnz);
1397 for (j = 0; j < 2; j++) {
1401 /* The total length of escape_sequence must be < 22 bits according
1402 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1403 UPDATE_CACHE(re, gb);
1404 b = GET_CACHE(re, gb);
1405 b = 31 - av_log2(~b);
1408 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1412 SKIP_BITS(re, gb, b + 1);
1414 n = (1 << b) + SHOW_UBITS(re, gb, b);
1415 LAST_SKIP_BITS(re, gb, b);
1416 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1419 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1420 *icf++ = (bits & 1U<<31) | v;
1427 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1431 CLOSE_READER(re, gb);
1437 if (pulse_present) {
1439 for (i = 0; i < pulse->num_pulse; i++) {
1440 float co = coef_base[ pulse->pos[i] ];
1441 while (offsets[idx + 1] <= pulse->pos[i])
1443 if (band_type[idx] != NOISE_BT && sf[idx]) {
1444 float ico = -pulse->amp[i];
1447 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1449 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1456 static av_always_inline float flt16_round(float pf)
1458 union av_intfloat32 tmp;
1460 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1464 static av_always_inline float flt16_even(float pf)
1466 union av_intfloat32 tmp;
1468 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1472 static av_always_inline float flt16_trunc(float pf)
1474 union av_intfloat32 pun;
1476 pun.i &= 0xFFFF0000U;
1480 static av_always_inline void predict(PredictorState *ps, float *coef,
1483 const float a = 0.953125; // 61.0 / 64
1484 const float alpha = 0.90625; // 29.0 / 32
1488 float r0 = ps->r0, r1 = ps->r1;
1489 float cor0 = ps->cor0, cor1 = ps->cor1;
1490 float var0 = ps->var0, var1 = ps->var1;
1492 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1493 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1495 pv = flt16_round(k1 * r0 + k2 * r1);
1502 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1503 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1504 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1505 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1507 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1508 ps->r0 = flt16_trunc(a * e0);
1512 * Apply AAC-Main style frequency domain prediction.
1514 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1518 if (!sce->ics.predictor_initialized) {
1519 reset_all_predictors(sce->predictor_state);
1520 sce->ics.predictor_initialized = 1;
1523 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1524 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1525 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1526 predict(&sce->predictor_state[k], &sce->coeffs[k],
1527 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1530 if (sce->ics.predictor_reset_group)
1531 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1533 reset_all_predictors(sce->predictor_state);
1537 * Decode an individual_channel_stream payload; reference: table 4.44.
1539 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1540 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1542 * @return Returns error status. 0 - OK, !0 - error
1544 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1545 GetBitContext *gb, int common_window, int scale_flag)
1548 TemporalNoiseShaping *tns = &sce->tns;
1549 IndividualChannelStream *ics = &sce->ics;
1550 float *out = sce->coeffs;
1551 int global_gain, pulse_present = 0;
1553 /* This assignment is to silence a GCC warning about the variable being used
1554 * uninitialized when in fact it always is.
1556 pulse.num_pulse = 0;
1558 global_gain = get_bits(gb, 8);
1560 if (!common_window && !scale_flag) {
1561 if (decode_ics_info(ac, ics, gb) < 0)
1562 return AVERROR_INVALIDDATA;
1565 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1567 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1572 if ((pulse_present = get_bits1(gb))) {
1573 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1574 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1577 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1578 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1582 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1584 if (get_bits1(gb)) {
1585 av_log_missing_feature(ac->avctx, "SSR", 1);
1590 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1593 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1594 apply_prediction(ac, sce);
1600 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1602 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1604 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1605 float *ch0 = cpe->ch[0].coeffs;
1606 float *ch1 = cpe->ch[1].coeffs;
1607 int g, i, group, idx = 0;
1608 const uint16_t *offsets = ics->swb_offset;
1609 for (g = 0; g < ics->num_window_groups; g++) {
1610 for (i = 0; i < ics->max_sfb; i++, idx++) {
1611 if (cpe->ms_mask[idx] &&
1612 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1613 for (group = 0; group < ics->group_len[g]; group++) {
1614 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1615 ch1 + group * 128 + offsets[i],
1616 offsets[i+1] - offsets[i]);
1620 ch0 += ics->group_len[g] * 128;
1621 ch1 += ics->group_len[g] * 128;
1626 * intensity stereo decoding; reference: 4.6.8.2.3
1628 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1629 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1630 * [3] reserved for scalable AAC
1632 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1634 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1635 SingleChannelElement *sce1 = &cpe->ch[1];
1636 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1637 const uint16_t *offsets = ics->swb_offset;
1638 int g, group, i, idx = 0;
1641 for (g = 0; g < ics->num_window_groups; g++) {
1642 for (i = 0; i < ics->max_sfb;) {
1643 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1644 const int bt_run_end = sce1->band_type_run_end[idx];
1645 for (; i < bt_run_end; i++, idx++) {
1646 c = -1 + 2 * (sce1->band_type[idx] - 14);
1648 c *= 1 - 2 * cpe->ms_mask[idx];
1649 scale = c * sce1->sf[idx];
1650 for (group = 0; group < ics->group_len[g]; group++)
1651 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1652 coef0 + group * 128 + offsets[i],
1654 offsets[i + 1] - offsets[i]);
1657 int bt_run_end = sce1->band_type_run_end[idx];
1658 idx += bt_run_end - i;
1662 coef0 += ics->group_len[g] * 128;
1663 coef1 += ics->group_len[g] * 128;
1668 * Decode a channel_pair_element; reference: table 4.4.
1670 * @return Returns error status. 0 - OK, !0 - error
1672 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1674 int i, ret, common_window, ms_present = 0;
1676 common_window = get_bits1(gb);
1677 if (common_window) {
1678 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1679 return AVERROR_INVALIDDATA;
1680 i = cpe->ch[1].ics.use_kb_window[0];
1681 cpe->ch[1].ics = cpe->ch[0].ics;
1682 cpe->ch[1].ics.use_kb_window[1] = i;
1683 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1684 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1685 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1686 ms_present = get_bits(gb, 2);
1687 if (ms_present == 3) {
1688 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1690 } else if (ms_present)
1691 decode_mid_side_stereo(cpe, gb, ms_present);
1693 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1695 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1698 if (common_window) {
1700 apply_mid_side_stereo(ac, cpe);
1701 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1702 apply_prediction(ac, &cpe->ch[0]);
1703 apply_prediction(ac, &cpe->ch[1]);
1707 apply_intensity_stereo(ac, cpe, ms_present);
1711 static const float cce_scale[] = {
1712 1.09050773266525765921, //2^(1/8)
1713 1.18920711500272106672, //2^(1/4)
1719 * Decode coupling_channel_element; reference: table 4.8.
1721 * @return Returns error status. 0 - OK, !0 - error
1723 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1729 SingleChannelElement *sce = &che->ch[0];
1730 ChannelCoupling *coup = &che->coup;
1732 coup->coupling_point = 2 * get_bits1(gb);
1733 coup->num_coupled = get_bits(gb, 3);
1734 for (c = 0; c <= coup->num_coupled; c++) {
1736 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1737 coup->id_select[c] = get_bits(gb, 4);
1738 if (coup->type[c] == TYPE_CPE) {
1739 coup->ch_select[c] = get_bits(gb, 2);
1740 if (coup->ch_select[c] == 3)
1743 coup->ch_select[c] = 2;
1745 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1747 sign = get_bits(gb, 1);
1748 scale = cce_scale[get_bits(gb, 2)];
1750 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1753 for (c = 0; c < num_gain; c++) {
1757 float gain_cache = 1.;
1759 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1760 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1761 gain_cache = powf(scale, -gain);
1763 if (coup->coupling_point == AFTER_IMDCT) {
1764 coup->gain[c][0] = gain_cache;
1766 for (g = 0; g < sce->ics.num_window_groups; g++) {
1767 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1768 if (sce->band_type[idx] != ZERO_BT) {
1770 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1778 gain_cache = powf(scale, -t) * s;
1781 coup->gain[c][idx] = gain_cache;
1791 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1793 * @return Returns number of bytes consumed.
1795 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1799 int num_excl_chan = 0;
1802 for (i = 0; i < 7; i++)
1803 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1804 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1806 return num_excl_chan / 7;
1810 * Decode dynamic range information; reference: table 4.52.
1812 * @param cnt length of TYPE_FIL syntactic element in bytes
1814 * @return Returns number of bytes consumed.
1816 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1817 GetBitContext *gb, int cnt)
1820 int drc_num_bands = 1;
1823 /* pce_tag_present? */
1824 if (get_bits1(gb)) {
1825 che_drc->pce_instance_tag = get_bits(gb, 4);
1826 skip_bits(gb, 4); // tag_reserved_bits
1830 /* excluded_chns_present? */
1831 if (get_bits1(gb)) {
1832 n += decode_drc_channel_exclusions(che_drc, gb);
1835 /* drc_bands_present? */
1836 if (get_bits1(gb)) {
1837 che_drc->band_incr = get_bits(gb, 4);
1838 che_drc->interpolation_scheme = get_bits(gb, 4);
1840 drc_num_bands += che_drc->band_incr;
1841 for (i = 0; i < drc_num_bands; i++) {
1842 che_drc->band_top[i] = get_bits(gb, 8);
1847 /* prog_ref_level_present? */
1848 if (get_bits1(gb)) {
1849 che_drc->prog_ref_level = get_bits(gb, 7);
1850 skip_bits1(gb); // prog_ref_level_reserved_bits
1854 for (i = 0; i < drc_num_bands; i++) {
1855 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1856 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1864 * Decode extension data (incomplete); reference: table 4.51.
1866 * @param cnt length of TYPE_FIL syntactic element in bytes
1868 * @return Returns number of bytes consumed
1870 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1871 ChannelElement *che, enum RawDataBlockType elem_type)
1875 switch (get_bits(gb, 4)) { // extension type
1876 case EXT_SBR_DATA_CRC:
1880 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1882 } else if (!ac->m4ac.sbr) {
1883 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1884 skip_bits_long(gb, 8 * cnt - 4);
1886 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1887 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1888 skip_bits_long(gb, 8 * cnt - 4);
1890 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1893 output_configure(ac, ac->layout_map, ac->layout_map_tags,
1894 ac->m4ac.chan_config, ac->output_configured);
1898 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1900 case EXT_DYNAMIC_RANGE:
1901 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1905 case EXT_DATA_ELEMENT:
1907 skip_bits_long(gb, 8 * cnt - 4);
1914 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1916 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1917 * @param coef spectral coefficients
1919 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1920 IndividualChannelStream *ics, int decode)
1922 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1924 int bottom, top, order, start, end, size, inc;
1925 float lpc[TNS_MAX_ORDER];
1926 float tmp[TNS_MAX_ORDER];
1928 for (w = 0; w < ics->num_windows; w++) {
1929 bottom = ics->num_swb;
1930 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1932 bottom = FFMAX(0, top - tns->length[w][filt]);
1933 order = tns->order[w][filt];
1938 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1940 start = ics->swb_offset[FFMIN(bottom, mmm)];
1941 end = ics->swb_offset[FFMIN( top, mmm)];
1942 if ((size = end - start) <= 0)
1944 if (tns->direction[w][filt]) {
1954 for (m = 0; m < size; m++, start += inc)
1955 for (i = 1; i <= FFMIN(m, order); i++)
1956 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1959 for (m = 0; m < size; m++, start += inc) {
1960 tmp[0] = coef[start];
1961 for (i = 1; i <= FFMIN(m, order); i++)
1962 coef[start] += tmp[i] * lpc[i - 1];
1963 for (i = order; i > 0; i--)
1964 tmp[i] = tmp[i - 1];
1972 * Apply windowing and MDCT to obtain the spectral
1973 * coefficient from the predicted sample by LTP.
1975 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1976 float *in, IndividualChannelStream *ics)
1978 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1979 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1980 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1981 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1983 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1984 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1986 memset(in, 0, 448 * sizeof(float));
1987 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1989 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1990 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1992 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1993 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1995 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1999 * Apply the long term prediction
2001 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2003 const LongTermPrediction *ltp = &sce->ics.ltp;
2004 const uint16_t *offsets = sce->ics.swb_offset;
2007 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2008 float *predTime = sce->ret;
2009 float *predFreq = ac->buf_mdct;
2010 int16_t num_samples = 2048;
2012 if (ltp->lag < 1024)
2013 num_samples = ltp->lag + 1024;
2014 for (i = 0; i < num_samples; i++)
2015 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2016 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2018 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2020 if (sce->tns.present)
2021 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2023 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2025 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2026 sce->coeffs[i] += predFreq[i];
2031 * Update the LTP buffer for next frame
2033 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2035 IndividualChannelStream *ics = &sce->ics;
2036 float *saved = sce->saved;
2037 float *saved_ltp = sce->coeffs;
2038 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2039 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2042 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2043 memcpy(saved_ltp, saved, 512 * sizeof(float));
2044 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2045 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2046 for (i = 0; i < 64; i++)
2047 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2048 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2049 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2050 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2051 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2052 for (i = 0; i < 64; i++)
2053 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2054 } else { // LONG_STOP or ONLY_LONG
2055 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2056 for (i = 0; i < 512; i++)
2057 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2060 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2061 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2062 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2066 * Conduct IMDCT and windowing.
2068 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2070 IndividualChannelStream *ics = &sce->ics;
2071 float *in = sce->coeffs;
2072 float *out = sce->ret;
2073 float *saved = sce->saved;
2074 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2075 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2076 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2077 float *buf = ac->buf_mdct;
2078 float *temp = ac->temp;
2082 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2083 for (i = 0; i < 1024; i += 128)
2084 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2086 ac->mdct.imdct_half(&ac->mdct, buf, in);
2088 /* window overlapping
2089 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2090 * and long to short transitions are considered to be short to short
2091 * transitions. This leaves just two cases (long to long and short to short)
2092 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2094 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2095 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2096 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2098 memcpy( out, saved, 448 * sizeof(float));
2100 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2101 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2102 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2103 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2104 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2105 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2106 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2108 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2109 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2114 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2115 memcpy( saved, temp + 64, 64 * sizeof(float));
2116 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2117 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2118 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2119 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2120 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2121 memcpy( saved, buf + 512, 448 * sizeof(float));
2122 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2123 } else { // LONG_STOP or ONLY_LONG
2124 memcpy( saved, buf + 512, 512 * sizeof(float));
2129 * Apply dependent channel coupling (applied before IMDCT).
2131 * @param index index into coupling gain array
2133 static void apply_dependent_coupling(AACContext *ac,
2134 SingleChannelElement *target,
2135 ChannelElement *cce, int index)
2137 IndividualChannelStream *ics = &cce->ch[0].ics;
2138 const uint16_t *offsets = ics->swb_offset;
2139 float *dest = target->coeffs;
2140 const float *src = cce->ch[0].coeffs;
2141 int g, i, group, k, idx = 0;
2142 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2143 av_log(ac->avctx, AV_LOG_ERROR,
2144 "Dependent coupling is not supported together with LTP\n");
2147 for (g = 0; g < ics->num_window_groups; g++) {
2148 for (i = 0; i < ics->max_sfb; i++, idx++) {
2149 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2150 const float gain = cce->coup.gain[index][idx];
2151 for (group = 0; group < ics->group_len[g]; group++) {
2152 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2154 dest[group * 128 + k] += gain * src[group * 128 + k];
2159 dest += ics->group_len[g] * 128;
2160 src += ics->group_len[g] * 128;
2165 * Apply independent channel coupling (applied after IMDCT).
2167 * @param index index into coupling gain array
2169 static void apply_independent_coupling(AACContext *ac,
2170 SingleChannelElement *target,
2171 ChannelElement *cce, int index)
2174 const float gain = cce->coup.gain[index][0];
2175 const float *src = cce->ch[0].ret;
2176 float *dest = target->ret;
2177 const int len = 1024 << (ac->m4ac.sbr == 1);
2179 for (i = 0; i < len; i++)
2180 dest[i] += gain * src[i];
2184 * channel coupling transformation interface
2186 * @param apply_coupling_method pointer to (in)dependent coupling function
2188 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2189 enum RawDataBlockType type, int elem_id,
2190 enum CouplingPoint coupling_point,
2191 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2195 for (i = 0; i < MAX_ELEM_ID; i++) {
2196 ChannelElement *cce = ac->che[TYPE_CCE][i];
2199 if (cce && cce->coup.coupling_point == coupling_point) {
2200 ChannelCoupling *coup = &cce->coup;
2202 for (c = 0; c <= coup->num_coupled; c++) {
2203 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2204 if (coup->ch_select[c] != 1) {
2205 apply_coupling_method(ac, &cc->ch[0], cce, index);
2206 if (coup->ch_select[c] != 0)
2209 if (coup->ch_select[c] != 2)
2210 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2212 index += 1 + (coup->ch_select[c] == 3);
2219 * Convert spectral data to float samples, applying all supported tools as appropriate.
2221 static void spectral_to_sample(AACContext *ac)
2224 for (type = 3; type >= 0; type--) {
2225 for (i = 0; i < MAX_ELEM_ID; i++) {
2226 ChannelElement *che = ac->che[type][i];
2228 if (type <= TYPE_CPE)
2229 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2230 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2231 if (che->ch[0].ics.predictor_present) {
2232 if (che->ch[0].ics.ltp.present)
2233 apply_ltp(ac, &che->ch[0]);
2234 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2235 apply_ltp(ac, &che->ch[1]);
2238 if (che->ch[0].tns.present)
2239 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2240 if (che->ch[1].tns.present)
2241 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2242 if (type <= TYPE_CPE)
2243 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2244 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2245 imdct_and_windowing(ac, &che->ch[0]);
2246 if (ac->m4ac.object_type == AOT_AAC_LTP)
2247 update_ltp(ac, &che->ch[0]);
2248 if (type == TYPE_CPE) {
2249 imdct_and_windowing(ac, &che->ch[1]);
2250 if (ac->m4ac.object_type == AOT_AAC_LTP)
2251 update_ltp(ac, &che->ch[1]);
2253 if (ac->m4ac.sbr > 0) {
2254 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2257 if (type <= TYPE_CCE)
2258 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2264 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2267 AACADTSHeaderInfo hdr_info;
2268 uint8_t layout_map[MAX_ELEM_ID*4][3];
2269 int layout_map_tags;
2271 size = avpriv_aac_parse_header(gb, &hdr_info);
2273 if (hdr_info.chan_config) {
2274 ac->m4ac.chan_config = hdr_info.chan_config;
2275 if (set_default_channel_config(ac->avctx, layout_map,
2276 &layout_map_tags, hdr_info.chan_config))
2278 if (output_configure(ac, layout_map, layout_map_tags,
2279 hdr_info.chan_config,
2280 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2282 } else if (ac->output_configured != OC_LOCKED) {
2283 ac->m4ac.chan_config = 0;
2284 ac->output_configured = OC_NONE;
2286 if (ac->output_configured != OC_LOCKED) {
2289 ac->m4ac.sample_rate = hdr_info.sample_rate;
2290 ac->m4ac.sampling_index = hdr_info.sampling_index;
2291 ac->m4ac.object_type = hdr_info.object_type;
2293 if (!ac->avctx->sample_rate)
2294 ac->avctx->sample_rate = hdr_info.sample_rate;
2295 if (hdr_info.num_aac_frames == 1) {
2296 if (!hdr_info.crc_absent)
2299 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2306 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2307 int *got_frame_ptr, GetBitContext *gb)
2309 AACContext *ac = avctx->priv_data;
2310 ChannelElement *che = NULL, *che_prev = NULL;
2311 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2313 int samples = 0, multiplier, audio_found = 0;
2315 if (show_bits(gb, 12) == 0xfff) {
2316 if (parse_adts_frame_header(ac, gb) < 0) {
2317 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2320 if (ac->m4ac.sampling_index > 12) {
2321 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2326 ac->tags_mapped = 0;
2328 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2329 elem_id = get_bits(gb, 4);
2331 if (elem_type < TYPE_DSE) {
2332 if (!(che=get_che(ac, elem_type, elem_id))) {
2333 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2334 elem_type, elem_id);
2340 switch (elem_type) {
2343 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2348 err = decode_cpe(ac, gb, che);
2353 err = decode_cce(ac, gb, che);
2357 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2362 err = skip_data_stream_element(ac, gb);
2366 uint8_t layout_map[MAX_ELEM_ID*4][3];
2368 tags = decode_pce(avctx, &ac->m4ac, layout_map, gb);
2373 if (ac->output_configured > OC_TRIAL_PCE)
2374 av_log(avctx, AV_LOG_ERROR,
2375 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2377 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2383 elem_id += get_bits(gb, 8) - 1;
2384 if (get_bits_left(gb) < 8 * elem_id) {
2385 av_log(avctx, AV_LOG_ERROR, overread_err);
2389 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2390 err = 0; /* FIXME */
2394 err = -1; /* should not happen, but keeps compiler happy */
2399 elem_type_prev = elem_type;
2404 if (get_bits_left(gb) < 3) {
2405 av_log(avctx, AV_LOG_ERROR, overread_err);
2410 spectral_to_sample(ac);
2412 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2413 samples <<= multiplier;
2414 if (ac->output_configured < OC_LOCKED) {
2415 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2416 avctx->frame_size = samples;
2420 /* get output buffer */
2421 ac->frame.nb_samples = samples;
2422 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2423 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2427 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2428 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2429 (const float **)ac->output_data,
2430 samples, avctx->channels);
2432 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2433 (const float **)ac->output_data,
2434 samples, avctx->channels);
2436 *(AVFrame *)data = ac->frame;
2438 *got_frame_ptr = !!samples;
2440 if (ac->output_configured && audio_found)
2441 ac->output_configured = OC_LOCKED;
2446 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2447 int *got_frame_ptr, AVPacket *avpkt)
2449 AACContext *ac = avctx->priv_data;
2450 const uint8_t *buf = avpkt->data;
2451 int buf_size = avpkt->size;
2456 int new_extradata_size;
2457 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2458 AV_PKT_DATA_NEW_EXTRADATA,
2459 &new_extradata_size);
2461 if (new_extradata) {
2462 av_free(avctx->extradata);
2463 avctx->extradata = av_mallocz(new_extradata_size +
2464 FF_INPUT_BUFFER_PADDING_SIZE);
2465 if (!avctx->extradata)
2466 return AVERROR(ENOMEM);
2467 avctx->extradata_size = new_extradata_size;
2468 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2469 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2471 avctx->extradata_size*8, 1) < 0)
2472 return AVERROR_INVALIDDATA;
2475 init_get_bits(&gb, buf, buf_size * 8);
2477 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2480 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2481 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2482 if (buf[buf_offset])
2485 return buf_size > buf_offset ? buf_consumed : buf_size;
2488 static av_cold int aac_decode_close(AVCodecContext *avctx)
2490 AACContext *ac = avctx->priv_data;
2493 for (i = 0; i < MAX_ELEM_ID; i++) {
2494 for (type = 0; type < 4; type++) {
2495 if (ac->che[type][i])
2496 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2497 av_freep(&ac->che[type][i]);
2501 ff_mdct_end(&ac->mdct);
2502 ff_mdct_end(&ac->mdct_small);
2503 ff_mdct_end(&ac->mdct_ltp);
2508 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2510 struct LATMContext {
2511 AACContext aac_ctx; ///< containing AACContext
2512 int initialized; ///< initilized after a valid extradata was seen
2515 int audio_mux_version_A; ///< LATM syntax version
2516 int frame_length_type; ///< 0/1 variable/fixed frame length
2517 int frame_length; ///< frame length for fixed frame length
2520 static inline uint32_t latm_get_value(GetBitContext *b)
2522 int length = get_bits(b, 2);
2524 return get_bits_long(b, (length+1)*8);
2527 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2528 GetBitContext *gb, int asclen)
2530 AACContext *ac = &latmctx->aac_ctx;
2531 AVCodecContext *avctx = ac->avctx;
2532 MPEG4AudioConfig m4ac = {0};
2533 int config_start_bit = get_bits_count(gb);
2534 int sync_extension = 0;
2535 int bits_consumed, esize;
2539 asclen = FFMIN(asclen, get_bits_left(gb));
2541 asclen = get_bits_left(gb);
2543 if (config_start_bit % 8) {
2544 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2545 "config not byte aligned.\n", 1);
2546 return AVERROR_INVALIDDATA;
2549 return AVERROR_INVALIDDATA;
2550 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2551 gb->buffer + (config_start_bit / 8),
2552 asclen, sync_extension);
2554 if (bits_consumed < 0)
2555 return AVERROR_INVALIDDATA;
2557 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2558 ac->m4ac.chan_config != m4ac.chan_config) {
2560 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2561 latmctx->initialized = 0;
2563 esize = (bits_consumed+7) / 8;
2565 if (avctx->extradata_size < esize) {
2566 av_free(avctx->extradata);
2567 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2568 if (!avctx->extradata)
2569 return AVERROR(ENOMEM);
2572 avctx->extradata_size = esize;
2573 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2574 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2576 skip_bits_long(gb, bits_consumed);
2578 return bits_consumed;
2581 static int read_stream_mux_config(struct LATMContext *latmctx,
2584 int ret, audio_mux_version = get_bits(gb, 1);
2586 latmctx->audio_mux_version_A = 0;
2587 if (audio_mux_version)
2588 latmctx->audio_mux_version_A = get_bits(gb, 1);
2590 if (!latmctx->audio_mux_version_A) {
2592 if (audio_mux_version)
2593 latm_get_value(gb); // taraFullness
2595 skip_bits(gb, 1); // allStreamSameTimeFraming
2596 skip_bits(gb, 6); // numSubFrames
2598 if (get_bits(gb, 4)) { // numPrograms
2599 av_log_missing_feature(latmctx->aac_ctx.avctx,
2600 "multiple programs are not supported\n", 1);
2601 return AVERROR_PATCHWELCOME;
2604 // for each program (which there is only on in DVB)
2606 // for each layer (which there is only on in DVB)
2607 if (get_bits(gb, 3)) { // numLayer
2608 av_log_missing_feature(latmctx->aac_ctx.avctx,
2609 "multiple layers are not supported\n", 1);
2610 return AVERROR_PATCHWELCOME;
2613 // for all but first stream: use_same_config = get_bits(gb, 1);
2614 if (!audio_mux_version) {
2615 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2618 int ascLen = latm_get_value(gb);
2619 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2622 skip_bits_long(gb, ascLen);
2625 latmctx->frame_length_type = get_bits(gb, 3);
2626 switch (latmctx->frame_length_type) {
2628 skip_bits(gb, 8); // latmBufferFullness
2631 latmctx->frame_length = get_bits(gb, 9);
2636 skip_bits(gb, 6); // CELP frame length table index
2640 skip_bits(gb, 1); // HVXC frame length table index
2644 if (get_bits(gb, 1)) { // other data
2645 if (audio_mux_version) {
2646 latm_get_value(gb); // other_data_bits
2650 esc = get_bits(gb, 1);
2656 if (get_bits(gb, 1)) // crc present
2657 skip_bits(gb, 8); // config_crc
2663 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2667 if (ctx->frame_length_type == 0) {
2668 int mux_slot_length = 0;
2670 tmp = get_bits(gb, 8);
2671 mux_slot_length += tmp;
2672 } while (tmp == 255);
2673 return mux_slot_length;
2674 } else if (ctx->frame_length_type == 1) {
2675 return ctx->frame_length;
2676 } else if (ctx->frame_length_type == 3 ||
2677 ctx->frame_length_type == 5 ||
2678 ctx->frame_length_type == 7) {
2679 skip_bits(gb, 2); // mux_slot_length_coded
2684 static int read_audio_mux_element(struct LATMContext *latmctx,
2688 uint8_t use_same_mux = get_bits(gb, 1);
2689 if (!use_same_mux) {
2690 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2692 } else if (!latmctx->aac_ctx.avctx->extradata) {
2693 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2694 "no decoder config found\n");
2695 return AVERROR(EAGAIN);
2697 if (latmctx->audio_mux_version_A == 0) {
2698 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2699 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2700 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2701 return AVERROR_INVALIDDATA;
2702 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2703 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2704 "frame length mismatch %d << %d\n",
2705 mux_slot_length_bytes * 8, get_bits_left(gb));
2706 return AVERROR_INVALIDDATA;
2713 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2714 int *got_frame_ptr, AVPacket *avpkt)
2716 struct LATMContext *latmctx = avctx->priv_data;
2720 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2722 // check for LOAS sync word
2723 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2724 return AVERROR_INVALIDDATA;
2726 muxlength = get_bits(&gb, 13) + 3;
2727 // not enough data, the parser should have sorted this
2728 if (muxlength > avpkt->size)
2729 return AVERROR_INVALIDDATA;
2731 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2734 if (!latmctx->initialized) {
2735 if (!avctx->extradata) {
2739 if ((err = decode_audio_specific_config(
2740 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2741 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2743 latmctx->initialized = 1;
2747 if (show_bits(&gb, 12) == 0xfff) {
2748 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2749 "ADTS header detected, probably as result of configuration "
2751 return AVERROR_INVALIDDATA;
2754 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2760 av_cold static int latm_decode_init(AVCodecContext *avctx)
2762 struct LATMContext *latmctx = avctx->priv_data;
2763 int ret = aac_decode_init(avctx);
2765 if (avctx->extradata_size > 0)
2766 latmctx->initialized = !ret;
2772 AVCodec ff_aac_decoder = {
2774 .type = AVMEDIA_TYPE_AUDIO,
2776 .priv_data_size = sizeof(AACContext),
2777 .init = aac_decode_init,
2778 .close = aac_decode_close,
2779 .decode = aac_decode_frame,
2780 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2781 .sample_fmts = (const enum AVSampleFormat[]) {
2782 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2784 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2785 .channel_layouts = aac_channel_layout,
2789 Note: This decoder filter is intended to decode LATM streams transferred
2790 in MPEG transport streams which only contain one program.
2791 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2793 AVCodec ff_aac_latm_decoder = {
2795 .type = AVMEDIA_TYPE_AUDIO,
2796 .id = CODEC_ID_AAC_LATM,
2797 .priv_data_size = sizeof(struct LATMContext),
2798 .init = latm_decode_init,
2799 .close = aac_decode_close,
2800 .decode = latm_decode_frame,
2801 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2802 .sample_fmts = (const enum AVSampleFormat[]) {
2803 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2805 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2806 .channel_layouts = aac_channel_layout,