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-libav@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
101 #include "libavutil/intfloat.h"
109 # include "arm/aac.h"
112 static VLC vlc_scalefactors;
113 static VLC vlc_spectral[11];
115 #define 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 FFmpeg 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 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
153 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
154 return AVERROR_INVALIDDATA;
156 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
157 if (type == TYPE_CPE ||
158 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
159 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
163 if (ac->che[type][id])
164 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
165 av_freep(&ac->che[type][id]);
170 struct elem_to_channel {
171 uint64_t av_position;
174 uint8_t aac_position;
177 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
178 uint8_t (*layout_map)[3], int offset, int tags, uint64_t left,
179 uint64_t right, int pos)
181 if (layout_map[offset][0] == TYPE_CPE) {
182 e2c_vec[offset] = (struct elem_to_channel) {
183 .av_position = left | right, .syn_ele = TYPE_CPE,
184 .elem_id = layout_map[offset ][1], .aac_position = pos };
187 e2c_vec[offset] = (struct elem_to_channel) {
188 .av_position = left, .syn_ele = TYPE_SCE,
189 .elem_id = layout_map[offset ][1], .aac_position = pos };
190 e2c_vec[offset + 1] = (struct elem_to_channel) {
191 .av_position = right, .syn_ele = TYPE_SCE,
192 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
197 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
198 int num_pos_channels = 0;
202 for (i = *current; i < tags; i++) {
203 if (layout_map[i][2] != pos)
205 if (layout_map[i][0] == TYPE_CPE) {
207 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
213 num_pos_channels += 2;
221 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
224 return num_pos_channels;
227 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
229 int i, n, total_non_cc_elements;
230 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
231 int num_front_channels, num_side_channels, num_back_channels;
234 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
239 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
240 if (num_front_channels < 0)
243 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
244 if (num_side_channels < 0)
247 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
248 if (num_back_channels < 0)
252 if (num_front_channels & 1) {
253 e2c_vec[i] = (struct elem_to_channel) {
254 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
255 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
257 num_front_channels--;
259 if (num_front_channels >= 4) {
260 i += assign_pair(e2c_vec, layout_map, i, tags,
261 AV_CH_FRONT_LEFT_OF_CENTER,
262 AV_CH_FRONT_RIGHT_OF_CENTER,
264 num_front_channels -= 2;
266 if (num_front_channels >= 2) {
267 i += assign_pair(e2c_vec, layout_map, i, tags,
271 num_front_channels -= 2;
273 while (num_front_channels >= 2) {
274 i += assign_pair(e2c_vec, layout_map, i, tags,
278 num_front_channels -= 2;
281 if (num_side_channels >= 2) {
282 i += assign_pair(e2c_vec, layout_map, i, tags,
286 num_side_channels -= 2;
288 while (num_side_channels >= 2) {
289 i += assign_pair(e2c_vec, layout_map, i, tags,
293 num_side_channels -= 2;
296 while (num_back_channels >= 4) {
297 i += assign_pair(e2c_vec, layout_map, i, tags,
301 num_back_channels -= 2;
303 if (num_back_channels >= 2) {
304 i += assign_pair(e2c_vec, layout_map, i, tags,
308 num_back_channels -= 2;
310 if (num_back_channels) {
311 e2c_vec[i] = (struct elem_to_channel) {
312 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
313 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
318 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
319 e2c_vec[i] = (struct elem_to_channel) {
320 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
321 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
324 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
325 e2c_vec[i] = (struct elem_to_channel) {
326 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
327 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
331 // Must choose a stable sort
332 total_non_cc_elements = n = i;
335 for (i = 1; i < n; i++) {
336 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
337 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
345 for (i = 0; i < total_non_cc_elements; i++) {
346 layout_map[i][0] = e2c_vec[i].syn_ele;
347 layout_map[i][1] = e2c_vec[i].elem_id;
348 layout_map[i][2] = e2c_vec[i].aac_position;
349 if (e2c_vec[i].av_position != UINT64_MAX) {
350 layout |= e2c_vec[i].av_position;
358 * Save current output configuration if and only if it has been locked.
360 static void push_output_configuration(AACContext *ac) {
361 if (ac->oc[1].status == OC_LOCKED) {
362 ac->oc[0] = ac->oc[1];
364 ac->oc[1].status = OC_NONE;
368 * Restore the previous output configuration if and only if the current
369 * configuration is unlocked.
371 static void pop_output_configuration(AACContext *ac) {
372 if (ac->oc[1].status != OC_LOCKED) {
373 if (ac->oc[0].status == OC_LOCKED) {
374 ac->oc[1] = ac->oc[0];
375 ac->avctx->channels = ac->oc[1].channels;
376 ac->avctx->channel_layout = ac->oc[1].channel_layout;
382 * Configure output channel order based on the current program configuration element.
384 * @return Returns error status. 0 - OK, !0 - error
386 static int output_configure(AACContext *ac,
387 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
388 int channel_config, enum OCStatus oc_type)
390 AVCodecContext *avctx = ac->avctx;
391 int i, channels = 0, ret;
394 if (ac->oc[1].layout_map != layout_map) {
395 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
396 ac->oc[1].layout_map_tags = tags;
399 // Try to sniff a reasonable channel order, otherwise output the
400 // channels in the order the PCE declared them.
401 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
402 layout = sniff_channel_order(layout_map, tags);
403 for (i = 0; i < tags; i++) {
404 int type = layout_map[i][0];
405 int id = layout_map[i][1];
406 int position = layout_map[i][2];
407 // Allocate or free elements depending on if they are in the
408 // current program configuration.
409 ret = che_configure(ac, position, type, id, &channels);
413 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
414 if (layout == AV_CH_FRONT_CENTER) {
415 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
421 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
422 if (layout) avctx->channel_layout = layout;
423 ac->oc[1].channel_layout = layout;
424 avctx->channels = ac->oc[1].channels = channels;
425 ac->oc[1].status = oc_type;
430 static void flush(AVCodecContext *avctx)
432 AACContext *ac= avctx->priv_data;
435 for (type = 3; type >= 0; type--) {
436 for (i = 0; i < MAX_ELEM_ID; i++) {
437 ChannelElement *che = ac->che[type][i];
439 for (j = 0; j <= 1; j++) {
440 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
448 * Set up channel positions based on a default channel configuration
449 * as specified in table 1.17.
451 * @return Returns error status. 0 - OK, !0 - error
453 static int set_default_channel_config(AVCodecContext *avctx,
454 uint8_t (*layout_map)[3],
458 if (channel_config < 1 || channel_config > 7) {
459 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
463 *tags = tags_per_config[channel_config];
464 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
468 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
470 // For PCE based channel configurations map the channels solely based on tags.
471 if (!ac->oc[1].m4ac.chan_config) {
472 return ac->tag_che_map[type][elem_id];
474 // Allow single CPE stereo files to be signalled with mono configuration.
475 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
476 uint8_t layout_map[MAX_ELEM_ID*4][3];
478 push_output_configuration(ac);
480 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
482 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
485 if (output_configure(ac, layout_map, layout_map_tags,
486 2, OC_TRIAL_FRAME) < 0)
489 ac->oc[1].m4ac.chan_config = 2;
492 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
493 uint8_t layout_map[MAX_ELEM_ID*4][3];
495 push_output_configuration(ac);
497 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
499 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
502 if (output_configure(ac, layout_map, layout_map_tags,
503 1, OC_TRIAL_FRAME) < 0)
506 ac->oc[1].m4ac.chan_config = 1;
508 // For indexed channel configurations map the channels solely based on position.
509 switch (ac->oc[1].m4ac.chan_config) {
511 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
513 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
516 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
517 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
518 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
519 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
521 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
524 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
526 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
529 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
531 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
535 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
537 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
538 } else if (ac->oc[1].m4ac.chan_config == 2) {
542 if (!ac->tags_mapped && type == TYPE_SCE) {
544 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
552 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
554 * @param type speaker type/position for these channels
556 static void decode_channel_map(uint8_t layout_map[][3],
557 enum ChannelPosition type,
558 GetBitContext *gb, int n)
561 enum RawDataBlockType syn_ele;
563 case AAC_CHANNEL_FRONT:
564 case AAC_CHANNEL_BACK:
565 case AAC_CHANNEL_SIDE:
566 syn_ele = get_bits1(gb);
572 case AAC_CHANNEL_LFE:
578 layout_map[0][0] = syn_ele;
579 layout_map[0][1] = get_bits(gb, 4);
580 layout_map[0][2] = type;
586 * Decode program configuration element; reference: table 4.2.
588 * @return Returns error status. 0 - OK, !0 - error
590 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
591 uint8_t (*layout_map)[3],
594 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
598 skip_bits(gb, 2); // object_type
600 sampling_index = get_bits(gb, 4);
601 if (m4ac->sampling_index != sampling_index)
602 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");
604 num_front = get_bits(gb, 4);
605 num_side = get_bits(gb, 4);
606 num_back = get_bits(gb, 4);
607 num_lfe = get_bits(gb, 2);
608 num_assoc_data = get_bits(gb, 3);
609 num_cc = get_bits(gb, 4);
612 skip_bits(gb, 4); // mono_mixdown_tag
614 skip_bits(gb, 4); // stereo_mixdown_tag
617 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
619 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
620 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
623 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
625 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
627 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
629 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
632 skip_bits_long(gb, 4 * num_assoc_data);
634 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
639 /* comment field, first byte is length */
640 comment_len = get_bits(gb, 8) * 8;
641 if (get_bits_left(gb) < comment_len) {
642 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
645 skip_bits_long(gb, comment_len);
650 * Decode GA "General Audio" specific configuration; reference: table 4.1.
652 * @param ac pointer to AACContext, may be null
653 * @param avctx pointer to AVCCodecContext, used for logging
655 * @return Returns error status. 0 - OK, !0 - error
657 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
659 MPEG4AudioConfig *m4ac,
662 int extension_flag, ret;
663 uint8_t layout_map[MAX_ELEM_ID*4][3];
666 if (get_bits1(gb)) { // frameLengthFlag
667 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
671 if (get_bits1(gb)) // dependsOnCoreCoder
672 skip_bits(gb, 14); // coreCoderDelay
673 extension_flag = get_bits1(gb);
675 if (m4ac->object_type == AOT_AAC_SCALABLE ||
676 m4ac->object_type == AOT_ER_AAC_SCALABLE)
677 skip_bits(gb, 3); // layerNr
679 if (channel_config == 0) {
680 skip_bits(gb, 4); // element_instance_tag
681 tags = decode_pce(avctx, m4ac, layout_map, gb);
685 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
689 if (count_channels(layout_map, tags) > 1) {
691 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
694 if (ac && (ret = output_configure(ac, layout_map, tags,
695 channel_config, OC_GLOBAL_HDR)))
698 if (extension_flag) {
699 switch (m4ac->object_type) {
701 skip_bits(gb, 5); // numOfSubFrame
702 skip_bits(gb, 11); // layer_length
706 case AOT_ER_AAC_SCALABLE:
708 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
709 * aacScalefactorDataResilienceFlag
710 * aacSpectralDataResilienceFlag
714 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
720 * Decode audio specific configuration; reference: table 1.13.
722 * @param ac pointer to AACContext, may be null
723 * @param avctx pointer to AVCCodecContext, used for logging
724 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
725 * @param data pointer to buffer holding an audio specific config
726 * @param bit_size size of audio specific config or data in bits
727 * @param sync_extension look for an appended sync extension
729 * @return Returns error status or number of consumed bits. <0 - error
731 static int decode_audio_specific_config(AACContext *ac,
732 AVCodecContext *avctx,
733 MPEG4AudioConfig *m4ac,
734 const uint8_t *data, int bit_size,
740 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
741 for (i = 0; i < bit_size >> 3; i++)
742 av_dlog(avctx, "%02x ", data[i]);
743 av_dlog(avctx, "\n");
745 init_get_bits(&gb, data, bit_size);
747 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
749 if (m4ac->sampling_index > 12) {
750 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
754 skip_bits_long(&gb, i);
756 switch (m4ac->object_type) {
760 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
764 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
765 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
769 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
770 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
771 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
773 return get_bits_count(&gb);
777 * linear congruential pseudorandom number generator
779 * @param previous_val pointer to the current state of the generator
781 * @return Returns a 32-bit pseudorandom integer
783 static av_always_inline int lcg_random(int previous_val)
785 return previous_val * 1664525 + 1013904223;
788 static av_always_inline void reset_predict_state(PredictorState *ps)
798 static void reset_all_predictors(PredictorState *ps)
801 for (i = 0; i < MAX_PREDICTORS; i++)
802 reset_predict_state(&ps[i]);
805 static int sample_rate_idx (int rate)
807 if (92017 <= rate) return 0;
808 else if (75132 <= rate) return 1;
809 else if (55426 <= rate) return 2;
810 else if (46009 <= rate) return 3;
811 else if (37566 <= rate) return 4;
812 else if (27713 <= rate) return 5;
813 else if (23004 <= rate) return 6;
814 else if (18783 <= rate) return 7;
815 else if (13856 <= rate) return 8;
816 else if (11502 <= rate) return 9;
817 else if (9391 <= rate) return 10;
821 static void reset_predictor_group(PredictorState *ps, int group_num)
824 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
825 reset_predict_state(&ps[i]);
828 #define AAC_INIT_VLC_STATIC(num, size) \
829 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
830 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
831 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
834 static av_cold int aac_decode_init(AVCodecContext *avctx)
836 AACContext *ac = avctx->priv_data;
837 float output_scale_factor;
840 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
842 if (avctx->extradata_size > 0) {
843 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
845 avctx->extradata_size*8, 1) < 0)
849 uint8_t layout_map[MAX_ELEM_ID*4][3];
852 sr = sample_rate_idx(avctx->sample_rate);
853 ac->oc[1].m4ac.sampling_index = sr;
854 ac->oc[1].m4ac.channels = avctx->channels;
855 ac->oc[1].m4ac.sbr = -1;
856 ac->oc[1].m4ac.ps = -1;
858 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
859 if (ff_mpeg4audio_channels[i] == avctx->channels)
861 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
864 ac->oc[1].m4ac.chan_config = i;
866 if (ac->oc[1].m4ac.chan_config) {
867 int ret = set_default_channel_config(avctx, layout_map,
868 &layout_map_tags, ac->oc[1].m4ac.chan_config);
870 output_configure(ac, layout_map, layout_map_tags,
871 ac->oc[1].m4ac.chan_config, OC_GLOBAL_HDR);
872 else if (avctx->err_recognition & AV_EF_EXPLODE)
873 return AVERROR_INVALIDDATA;
877 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
878 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
879 output_scale_factor = 1.0 / 32768.0;
881 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
882 output_scale_factor = 1.0;
885 AAC_INIT_VLC_STATIC( 0, 304);
886 AAC_INIT_VLC_STATIC( 1, 270);
887 AAC_INIT_VLC_STATIC( 2, 550);
888 AAC_INIT_VLC_STATIC( 3, 300);
889 AAC_INIT_VLC_STATIC( 4, 328);
890 AAC_INIT_VLC_STATIC( 5, 294);
891 AAC_INIT_VLC_STATIC( 6, 306);
892 AAC_INIT_VLC_STATIC( 7, 268);
893 AAC_INIT_VLC_STATIC( 8, 510);
894 AAC_INIT_VLC_STATIC( 9, 366);
895 AAC_INIT_VLC_STATIC(10, 462);
899 ff_dsputil_init(&ac->dsp, avctx);
900 ff_fmt_convert_init(&ac->fmt_conv, avctx);
902 ac->random_state = 0x1f2e3d4c;
906 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
907 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
908 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
911 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
912 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
913 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
914 // window initialization
915 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
916 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
917 ff_init_ff_sine_windows(10);
918 ff_init_ff_sine_windows( 7);
922 avcodec_get_frame_defaults(&ac->frame);
923 avctx->coded_frame = &ac->frame;
929 * Skip data_stream_element; reference: table 4.10.
931 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
933 int byte_align = get_bits1(gb);
934 int count = get_bits(gb, 8);
936 count += get_bits(gb, 8);
940 if (get_bits_left(gb) < 8 * count) {
941 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
944 skip_bits_long(gb, 8 * count);
948 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
953 ics->predictor_reset_group = get_bits(gb, 5);
954 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
955 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
959 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
960 ics->prediction_used[sfb] = get_bits1(gb);
966 * Decode Long Term Prediction data; reference: table 4.xx.
968 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
969 GetBitContext *gb, uint8_t max_sfb)
973 ltp->lag = get_bits(gb, 11);
974 ltp->coef = ltp_coef[get_bits(gb, 3)];
975 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
976 ltp->used[sfb] = get_bits1(gb);
980 * Decode Individual Channel Stream info; reference: table 4.6.
982 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
986 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
987 return AVERROR_INVALIDDATA;
989 ics->window_sequence[1] = ics->window_sequence[0];
990 ics->window_sequence[0] = get_bits(gb, 2);
991 ics->use_kb_window[1] = ics->use_kb_window[0];
992 ics->use_kb_window[0] = get_bits1(gb);
993 ics->num_window_groups = 1;
994 ics->group_len[0] = 1;
995 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
997 ics->max_sfb = get_bits(gb, 4);
998 for (i = 0; i < 7; i++) {
1000 ics->group_len[ics->num_window_groups - 1]++;
1002 ics->num_window_groups++;
1003 ics->group_len[ics->num_window_groups - 1] = 1;
1006 ics->num_windows = 8;
1007 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1008 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1009 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1010 ics->predictor_present = 0;
1012 ics->max_sfb = get_bits(gb, 6);
1013 ics->num_windows = 1;
1014 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1015 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1016 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1017 ics->predictor_present = get_bits1(gb);
1018 ics->predictor_reset_group = 0;
1019 if (ics->predictor_present) {
1020 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1021 if (decode_prediction(ac, ics, gb)) {
1024 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1025 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1028 if ((ics->ltp.present = get_bits(gb, 1)))
1029 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
1034 if (ics->max_sfb > ics->num_swb) {
1035 av_log(ac->avctx, AV_LOG_ERROR,
1036 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1037 ics->max_sfb, ics->num_swb);
1044 return AVERROR_INVALIDDATA;
1048 * Decode band types (section_data payload); reference: table 4.46.
1050 * @param band_type array of the used band type
1051 * @param band_type_run_end array of the last scalefactor band of a band type run
1053 * @return Returns error status. 0 - OK, !0 - error
1055 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1056 int band_type_run_end[120], GetBitContext *gb,
1057 IndividualChannelStream *ics)
1060 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1061 for (g = 0; g < ics->num_window_groups; g++) {
1063 while (k < ics->max_sfb) {
1064 uint8_t sect_end = k;
1066 int sect_band_type = get_bits(gb, 4);
1067 if (sect_band_type == 12) {
1068 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1072 sect_len_incr = get_bits(gb, bits);
1073 sect_end += sect_len_incr;
1074 if (get_bits_left(gb) < 0) {
1075 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1078 if (sect_end > ics->max_sfb) {
1079 av_log(ac->avctx, AV_LOG_ERROR,
1080 "Number of bands (%d) exceeds limit (%d).\n",
1081 sect_end, ics->max_sfb);
1084 } while (sect_len_incr == (1 << bits) - 1);
1085 for (; k < sect_end; k++) {
1086 band_type [idx] = sect_band_type;
1087 band_type_run_end[idx++] = sect_end;
1095 * Decode scalefactors; reference: table 4.47.
1097 * @param global_gain first scalefactor value as scalefactors are differentially coded
1098 * @param band_type array of the used band type
1099 * @param band_type_run_end array of the last scalefactor band of a band type run
1100 * @param sf array of scalefactors or intensity stereo positions
1102 * @return Returns error status. 0 - OK, !0 - error
1104 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1105 unsigned int global_gain,
1106 IndividualChannelStream *ics,
1107 enum BandType band_type[120],
1108 int band_type_run_end[120])
1111 int offset[3] = { global_gain, global_gain - 90, 0 };
1114 for (g = 0; g < ics->num_window_groups; g++) {
1115 for (i = 0; i < ics->max_sfb;) {
1116 int run_end = band_type_run_end[idx];
1117 if (band_type[idx] == ZERO_BT) {
1118 for (; i < run_end; i++, idx++)
1120 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1121 for (; i < run_end; i++, idx++) {
1122 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1123 clipped_offset = av_clip(offset[2], -155, 100);
1124 if (offset[2] != clipped_offset) {
1125 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1126 "position clipped (%d -> %d).\nIf you heard an "
1127 "audible artifact, there may be a bug in the "
1128 "decoder. ", offset[2], clipped_offset);
1130 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1132 } else if (band_type[idx] == NOISE_BT) {
1133 for (; i < run_end; i++, idx++) {
1134 if (noise_flag-- > 0)
1135 offset[1] += get_bits(gb, 9) - 256;
1137 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1138 clipped_offset = av_clip(offset[1], -100, 155);
1139 if (offset[1] != clipped_offset) {
1140 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1141 "(%d -> %d).\nIf you heard an audible "
1142 "artifact, there may be a bug in the decoder. ",
1143 offset[1], clipped_offset);
1145 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1148 for (; i < run_end; i++, idx++) {
1149 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1150 if (offset[0] > 255U) {
1151 av_log(ac->avctx, AV_LOG_ERROR,
1152 "Scalefactor (%d) out of range.\n", offset[0]);
1155 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1164 * Decode pulse data; reference: table 4.7.
1166 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1167 const uint16_t *swb_offset, int num_swb)
1170 pulse->num_pulse = get_bits(gb, 2) + 1;
1171 pulse_swb = get_bits(gb, 6);
1172 if (pulse_swb >= num_swb)
1174 pulse->pos[0] = swb_offset[pulse_swb];
1175 pulse->pos[0] += get_bits(gb, 5);
1176 if (pulse->pos[0] > 1023)
1178 pulse->amp[0] = get_bits(gb, 4);
1179 for (i = 1; i < pulse->num_pulse; i++) {
1180 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1181 if (pulse->pos[i] > 1023)
1183 pulse->amp[i] = get_bits(gb, 4);
1189 * Decode Temporal Noise Shaping data; reference: table 4.48.
1191 * @return Returns error status. 0 - OK, !0 - error
1193 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1194 GetBitContext *gb, const IndividualChannelStream *ics)
1196 int w, filt, i, coef_len, coef_res, coef_compress;
1197 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1198 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1199 for (w = 0; w < ics->num_windows; w++) {
1200 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1201 coef_res = get_bits1(gb);
1203 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1205 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1207 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1208 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1209 tns->order[w][filt], tns_max_order);
1210 tns->order[w][filt] = 0;
1213 if (tns->order[w][filt]) {
1214 tns->direction[w][filt] = get_bits1(gb);
1215 coef_compress = get_bits1(gb);
1216 coef_len = coef_res + 3 - coef_compress;
1217 tmp2_idx = 2 * coef_compress + coef_res;
1219 for (i = 0; i < tns->order[w][filt]; i++)
1220 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1229 * Decode Mid/Side data; reference: table 4.54.
1231 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1232 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1233 * [3] reserved for scalable AAC
1235 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1239 if (ms_present == 1) {
1240 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1241 cpe->ms_mask[idx] = get_bits1(gb);
1242 } else if (ms_present == 2) {
1243 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1248 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1252 *dst++ = v[idx & 15] * s;
1253 *dst++ = v[idx>>4 & 15] * s;
1259 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1263 *dst++ = v[idx & 3] * s;
1264 *dst++ = v[idx>>2 & 3] * s;
1265 *dst++ = v[idx>>4 & 3] * s;
1266 *dst++ = v[idx>>6 & 3] * s;
1272 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1273 unsigned sign, const float *scale)
1275 union av_intfloat32 s0, s1;
1277 s0.f = s1.f = *scale;
1278 s0.i ^= sign >> 1 << 31;
1281 *dst++ = v[idx & 15] * s0.f;
1282 *dst++ = v[idx>>4 & 15] * s1.f;
1289 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1290 unsigned sign, const float *scale)
1292 unsigned nz = idx >> 12;
1293 union av_intfloat32 s = { .f = *scale };
1294 union av_intfloat32 t;
1296 t.i = s.i ^ (sign & 1U<<31);
1297 *dst++ = v[idx & 3] * t.f;
1299 sign <<= nz & 1; nz >>= 1;
1300 t.i = s.i ^ (sign & 1U<<31);
1301 *dst++ = v[idx>>2 & 3] * t.f;
1303 sign <<= nz & 1; nz >>= 1;
1304 t.i = s.i ^ (sign & 1U<<31);
1305 *dst++ = v[idx>>4 & 3] * t.f;
1307 sign <<= nz & 1; nz >>= 1;
1308 t.i = s.i ^ (sign & 1U<<31);
1309 *dst++ = v[idx>>6 & 3] * t.f;
1316 * Decode spectral data; reference: table 4.50.
1317 * Dequantize and scale spectral data; reference: 4.6.3.3.
1319 * @param coef array of dequantized, scaled spectral data
1320 * @param sf array of scalefactors or intensity stereo positions
1321 * @param pulse_present set if pulses are present
1322 * @param pulse pointer to pulse data struct
1323 * @param band_type array of the used band type
1325 * @return Returns error status. 0 - OK, !0 - error
1327 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1328 GetBitContext *gb, const float sf[120],
1329 int pulse_present, const Pulse *pulse,
1330 const IndividualChannelStream *ics,
1331 enum BandType band_type[120])
1333 int i, k, g, idx = 0;
1334 const int c = 1024 / ics->num_windows;
1335 const uint16_t *offsets = ics->swb_offset;
1336 float *coef_base = coef;
1338 for (g = 0; g < ics->num_windows; g++)
1339 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1341 for (g = 0; g < ics->num_window_groups; g++) {
1342 unsigned g_len = ics->group_len[g];
1344 for (i = 0; i < ics->max_sfb; i++, idx++) {
1345 const unsigned cbt_m1 = band_type[idx] - 1;
1346 float *cfo = coef + offsets[i];
1347 int off_len = offsets[i + 1] - offsets[i];
1350 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1351 for (group = 0; group < g_len; group++, cfo+=128) {
1352 memset(cfo, 0, off_len * sizeof(float));
1354 } else if (cbt_m1 == NOISE_BT - 1) {
1355 for (group = 0; group < g_len; group++, cfo+=128) {
1359 for (k = 0; k < off_len; k++) {
1360 ac->random_state = lcg_random(ac->random_state);
1361 cfo[k] = ac->random_state;
1364 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1365 scale = sf[idx] / sqrtf(band_energy);
1366 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1369 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1370 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1371 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1372 OPEN_READER(re, gb);
1374 switch (cbt_m1 >> 1) {
1376 for (group = 0; group < g_len; group++, cfo+=128) {
1384 UPDATE_CACHE(re, gb);
1385 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1386 cb_idx = cb_vector_idx[code];
1387 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1393 for (group = 0; group < g_len; group++, cfo+=128) {
1403 UPDATE_CACHE(re, gb);
1404 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1405 cb_idx = cb_vector_idx[code];
1406 nnz = cb_idx >> 8 & 15;
1407 bits = nnz ? GET_CACHE(re, gb) : 0;
1408 LAST_SKIP_BITS(re, gb, nnz);
1409 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1415 for (group = 0; group < g_len; group++, cfo+=128) {
1423 UPDATE_CACHE(re, gb);
1424 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1425 cb_idx = cb_vector_idx[code];
1426 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1433 for (group = 0; group < g_len; group++, cfo+=128) {
1443 UPDATE_CACHE(re, gb);
1444 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1445 cb_idx = cb_vector_idx[code];
1446 nnz = cb_idx >> 8 & 15;
1447 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1448 LAST_SKIP_BITS(re, gb, nnz);
1449 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1455 for (group = 0; group < g_len; group++, cfo+=128) {
1457 uint32_t *icf = (uint32_t *) cf;
1467 UPDATE_CACHE(re, gb);
1468 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1476 cb_idx = cb_vector_idx[code];
1479 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1480 LAST_SKIP_BITS(re, gb, nnz);
1482 for (j = 0; j < 2; j++) {
1486 /* The total length of escape_sequence must be < 22 bits according
1487 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1488 UPDATE_CACHE(re, gb);
1489 b = GET_CACHE(re, gb);
1490 b = 31 - av_log2(~b);
1493 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1497 SKIP_BITS(re, gb, b + 1);
1499 n = (1 << b) + SHOW_UBITS(re, gb, b);
1500 LAST_SKIP_BITS(re, gb, b);
1501 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1504 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1505 *icf++ = (bits & 1U<<31) | v;
1512 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1516 CLOSE_READER(re, gb);
1522 if (pulse_present) {
1524 for (i = 0; i < pulse->num_pulse; i++) {
1525 float co = coef_base[ pulse->pos[i] ];
1526 while (offsets[idx + 1] <= pulse->pos[i])
1528 if (band_type[idx] != NOISE_BT && sf[idx]) {
1529 float ico = -pulse->amp[i];
1532 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1534 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1541 static av_always_inline float flt16_round(float pf)
1543 union av_intfloat32 tmp;
1545 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1549 static av_always_inline float flt16_even(float pf)
1551 union av_intfloat32 tmp;
1553 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1557 static av_always_inline float flt16_trunc(float pf)
1559 union av_intfloat32 pun;
1561 pun.i &= 0xFFFF0000U;
1565 static av_always_inline void predict(PredictorState *ps, float *coef,
1568 const float a = 0.953125; // 61.0 / 64
1569 const float alpha = 0.90625; // 29.0 / 32
1573 float r0 = ps->r0, r1 = ps->r1;
1574 float cor0 = ps->cor0, cor1 = ps->cor1;
1575 float var0 = ps->var0, var1 = ps->var1;
1577 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1578 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1580 pv = flt16_round(k1 * r0 + k2 * r1);
1587 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1588 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1589 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1590 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1592 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1593 ps->r0 = flt16_trunc(a * e0);
1597 * Apply AAC-Main style frequency domain prediction.
1599 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1603 if (!sce->ics.predictor_initialized) {
1604 reset_all_predictors(sce->predictor_state);
1605 sce->ics.predictor_initialized = 1;
1608 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1609 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1610 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1611 predict(&sce->predictor_state[k], &sce->coeffs[k],
1612 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1615 if (sce->ics.predictor_reset_group)
1616 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1618 reset_all_predictors(sce->predictor_state);
1622 * Decode an individual_channel_stream payload; reference: table 4.44.
1624 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1625 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1627 * @return Returns error status. 0 - OK, !0 - error
1629 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1630 GetBitContext *gb, int common_window, int scale_flag)
1633 TemporalNoiseShaping *tns = &sce->tns;
1634 IndividualChannelStream *ics = &sce->ics;
1635 float *out = sce->coeffs;
1636 int global_gain, pulse_present = 0;
1638 /* This assignment is to silence a GCC warning about the variable being used
1639 * uninitialized when in fact it always is.
1641 pulse.num_pulse = 0;
1643 global_gain = get_bits(gb, 8);
1645 if (!common_window && !scale_flag) {
1646 if (decode_ics_info(ac, ics, gb) < 0)
1647 return AVERROR_INVALIDDATA;
1650 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1652 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1657 if ((pulse_present = get_bits1(gb))) {
1658 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1659 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1662 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1663 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1667 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1669 if (get_bits1(gb)) {
1670 av_log_missing_feature(ac->avctx, "SSR", 1);
1675 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1678 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1679 apply_prediction(ac, sce);
1685 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1687 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1689 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1690 float *ch0 = cpe->ch[0].coeffs;
1691 float *ch1 = cpe->ch[1].coeffs;
1692 int g, i, group, idx = 0;
1693 const uint16_t *offsets = ics->swb_offset;
1694 for (g = 0; g < ics->num_window_groups; g++) {
1695 for (i = 0; i < ics->max_sfb; i++, idx++) {
1696 if (cpe->ms_mask[idx] &&
1697 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1698 for (group = 0; group < ics->group_len[g]; group++) {
1699 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1700 ch1 + group * 128 + offsets[i],
1701 offsets[i+1] - offsets[i]);
1705 ch0 += ics->group_len[g] * 128;
1706 ch1 += ics->group_len[g] * 128;
1711 * intensity stereo decoding; reference: 4.6.8.2.3
1713 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1714 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1715 * [3] reserved for scalable AAC
1717 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1719 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1720 SingleChannelElement *sce1 = &cpe->ch[1];
1721 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1722 const uint16_t *offsets = ics->swb_offset;
1723 int g, group, i, idx = 0;
1726 for (g = 0; g < ics->num_window_groups; g++) {
1727 for (i = 0; i < ics->max_sfb;) {
1728 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1729 const int bt_run_end = sce1->band_type_run_end[idx];
1730 for (; i < bt_run_end; i++, idx++) {
1731 c = -1 + 2 * (sce1->band_type[idx] - 14);
1733 c *= 1 - 2 * cpe->ms_mask[idx];
1734 scale = c * sce1->sf[idx];
1735 for (group = 0; group < ics->group_len[g]; group++)
1736 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1737 coef0 + group * 128 + offsets[i],
1739 offsets[i + 1] - offsets[i]);
1742 int bt_run_end = sce1->band_type_run_end[idx];
1743 idx += bt_run_end - i;
1747 coef0 += ics->group_len[g] * 128;
1748 coef1 += ics->group_len[g] * 128;
1753 * Decode a channel_pair_element; reference: table 4.4.
1755 * @return Returns error status. 0 - OK, !0 - error
1757 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1759 int i, ret, common_window, ms_present = 0;
1761 common_window = get_bits1(gb);
1762 if (common_window) {
1763 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1764 return AVERROR_INVALIDDATA;
1765 i = cpe->ch[1].ics.use_kb_window[0];
1766 cpe->ch[1].ics = cpe->ch[0].ics;
1767 cpe->ch[1].ics.use_kb_window[1] = i;
1768 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1769 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1770 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1771 ms_present = get_bits(gb, 2);
1772 if (ms_present == 3) {
1773 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1775 } else if (ms_present)
1776 decode_mid_side_stereo(cpe, gb, ms_present);
1778 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1780 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1783 if (common_window) {
1785 apply_mid_side_stereo(ac, cpe);
1786 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1787 apply_prediction(ac, &cpe->ch[0]);
1788 apply_prediction(ac, &cpe->ch[1]);
1792 apply_intensity_stereo(ac, cpe, ms_present);
1796 static const float cce_scale[] = {
1797 1.09050773266525765921, //2^(1/8)
1798 1.18920711500272106672, //2^(1/4)
1804 * Decode coupling_channel_element; reference: table 4.8.
1806 * @return Returns error status. 0 - OK, !0 - error
1808 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1814 SingleChannelElement *sce = &che->ch[0];
1815 ChannelCoupling *coup = &che->coup;
1817 coup->coupling_point = 2 * get_bits1(gb);
1818 coup->num_coupled = get_bits(gb, 3);
1819 for (c = 0; c <= coup->num_coupled; c++) {
1821 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1822 coup->id_select[c] = get_bits(gb, 4);
1823 if (coup->type[c] == TYPE_CPE) {
1824 coup->ch_select[c] = get_bits(gb, 2);
1825 if (coup->ch_select[c] == 3)
1828 coup->ch_select[c] = 2;
1830 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1832 sign = get_bits(gb, 1);
1833 scale = cce_scale[get_bits(gb, 2)];
1835 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1838 for (c = 0; c < num_gain; c++) {
1842 float gain_cache = 1.;
1844 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1845 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1846 gain_cache = powf(scale, -gain);
1848 if (coup->coupling_point == AFTER_IMDCT) {
1849 coup->gain[c][0] = gain_cache;
1851 for (g = 0; g < sce->ics.num_window_groups; g++) {
1852 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1853 if (sce->band_type[idx] != ZERO_BT) {
1855 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1863 gain_cache = powf(scale, -t) * s;
1866 coup->gain[c][idx] = gain_cache;
1876 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1878 * @return Returns number of bytes consumed.
1880 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1884 int num_excl_chan = 0;
1887 for (i = 0; i < 7; i++)
1888 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1889 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1891 return num_excl_chan / 7;
1895 * Decode dynamic range information; reference: table 4.52.
1897 * @param cnt length of TYPE_FIL syntactic element in bytes
1899 * @return Returns number of bytes consumed.
1901 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1902 GetBitContext *gb, int cnt)
1905 int drc_num_bands = 1;
1908 /* pce_tag_present? */
1909 if (get_bits1(gb)) {
1910 che_drc->pce_instance_tag = get_bits(gb, 4);
1911 skip_bits(gb, 4); // tag_reserved_bits
1915 /* excluded_chns_present? */
1916 if (get_bits1(gb)) {
1917 n += decode_drc_channel_exclusions(che_drc, gb);
1920 /* drc_bands_present? */
1921 if (get_bits1(gb)) {
1922 che_drc->band_incr = get_bits(gb, 4);
1923 che_drc->interpolation_scheme = get_bits(gb, 4);
1925 drc_num_bands += che_drc->band_incr;
1926 for (i = 0; i < drc_num_bands; i++) {
1927 che_drc->band_top[i] = get_bits(gb, 8);
1932 /* prog_ref_level_present? */
1933 if (get_bits1(gb)) {
1934 che_drc->prog_ref_level = get_bits(gb, 7);
1935 skip_bits1(gb); // prog_ref_level_reserved_bits
1939 for (i = 0; i < drc_num_bands; i++) {
1940 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1941 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1949 * Decode extension data (incomplete); reference: table 4.51.
1951 * @param cnt length of TYPE_FIL syntactic element in bytes
1953 * @return Returns number of bytes consumed
1955 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1956 ChannelElement *che, enum RawDataBlockType elem_type)
1960 switch (get_bits(gb, 4)) { // extension type
1961 case EXT_SBR_DATA_CRC:
1965 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1967 } else if (!ac->oc[1].m4ac.sbr) {
1968 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1969 skip_bits_long(gb, 8 * cnt - 4);
1971 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
1972 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1973 skip_bits_long(gb, 8 * cnt - 4);
1975 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
1976 ac->oc[1].m4ac.sbr = 1;
1977 ac->oc[1].m4ac.ps = 1;
1978 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
1979 ac->oc[1].m4ac.chan_config, ac->oc[1].status);
1981 ac->oc[1].m4ac.sbr = 1;
1983 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1985 case EXT_DYNAMIC_RANGE:
1986 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1990 case EXT_DATA_ELEMENT:
1992 skip_bits_long(gb, 8 * cnt - 4);
1999 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2001 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2002 * @param coef spectral coefficients
2004 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2005 IndividualChannelStream *ics, int decode)
2007 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2009 int bottom, top, order, start, end, size, inc;
2010 float lpc[TNS_MAX_ORDER];
2011 float tmp[TNS_MAX_ORDER];
2013 for (w = 0; w < ics->num_windows; w++) {
2014 bottom = ics->num_swb;
2015 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2017 bottom = FFMAX(0, top - tns->length[w][filt]);
2018 order = tns->order[w][filt];
2023 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2025 start = ics->swb_offset[FFMIN(bottom, mmm)];
2026 end = ics->swb_offset[FFMIN( top, mmm)];
2027 if ((size = end - start) <= 0)
2029 if (tns->direction[w][filt]) {
2039 for (m = 0; m < size; m++, start += inc)
2040 for (i = 1; i <= FFMIN(m, order); i++)
2041 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2044 for (m = 0; m < size; m++, start += inc) {
2045 tmp[0] = coef[start];
2046 for (i = 1; i <= FFMIN(m, order); i++)
2047 coef[start] += tmp[i] * lpc[i - 1];
2048 for (i = order; i > 0; i--)
2049 tmp[i] = tmp[i - 1];
2057 * Apply windowing and MDCT to obtain the spectral
2058 * coefficient from the predicted sample by LTP.
2060 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2061 float *in, IndividualChannelStream *ics)
2063 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2064 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2065 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2066 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2068 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2069 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
2071 memset(in, 0, 448 * sizeof(float));
2072 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2074 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2075 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2077 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2078 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2080 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2084 * Apply the long term prediction
2086 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2088 const LongTermPrediction *ltp = &sce->ics.ltp;
2089 const uint16_t *offsets = sce->ics.swb_offset;
2092 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2093 float *predTime = sce->ret;
2094 float *predFreq = ac->buf_mdct;
2095 int16_t num_samples = 2048;
2097 if (ltp->lag < 1024)
2098 num_samples = ltp->lag + 1024;
2099 for (i = 0; i < num_samples; i++)
2100 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2101 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2103 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2105 if (sce->tns.present)
2106 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2108 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2110 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2111 sce->coeffs[i] += predFreq[i];
2116 * Update the LTP buffer for next frame
2118 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2120 IndividualChannelStream *ics = &sce->ics;
2121 float *saved = sce->saved;
2122 float *saved_ltp = sce->coeffs;
2123 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2124 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2127 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2128 memcpy(saved_ltp, saved, 512 * sizeof(float));
2129 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2130 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2131 for (i = 0; i < 64; i++)
2132 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2133 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2134 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2135 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2136 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2137 for (i = 0; i < 64; i++)
2138 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2139 } else { // LONG_STOP or ONLY_LONG
2140 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2141 for (i = 0; i < 512; i++)
2142 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2145 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2146 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2147 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2151 * Conduct IMDCT and windowing.
2153 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2155 IndividualChannelStream *ics = &sce->ics;
2156 float *in = sce->coeffs;
2157 float *out = sce->ret;
2158 float *saved = sce->saved;
2159 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2160 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2161 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2162 float *buf = ac->buf_mdct;
2163 float *temp = ac->temp;
2167 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2168 for (i = 0; i < 1024; i += 128)
2169 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2171 ac->mdct.imdct_half(&ac->mdct, buf, in);
2173 /* window overlapping
2174 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2175 * and long to short transitions are considered to be short to short
2176 * transitions. This leaves just two cases (long to long and short to short)
2177 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2179 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2180 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2181 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2183 memcpy( out, saved, 448 * sizeof(float));
2185 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2186 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2187 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2188 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2189 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2190 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2191 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2193 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2194 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2199 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2200 memcpy( saved, temp + 64, 64 * sizeof(float));
2201 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2202 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2203 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2204 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2205 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2206 memcpy( saved, buf + 512, 448 * sizeof(float));
2207 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2208 } else { // LONG_STOP or ONLY_LONG
2209 memcpy( saved, buf + 512, 512 * sizeof(float));
2214 * Apply dependent channel coupling (applied before IMDCT).
2216 * @param index index into coupling gain array
2218 static void apply_dependent_coupling(AACContext *ac,
2219 SingleChannelElement *target,
2220 ChannelElement *cce, int index)
2222 IndividualChannelStream *ics = &cce->ch[0].ics;
2223 const uint16_t *offsets = ics->swb_offset;
2224 float *dest = target->coeffs;
2225 const float *src = cce->ch[0].coeffs;
2226 int g, i, group, k, idx = 0;
2227 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2228 av_log(ac->avctx, AV_LOG_ERROR,
2229 "Dependent coupling is not supported together with LTP\n");
2232 for (g = 0; g < ics->num_window_groups; g++) {
2233 for (i = 0; i < ics->max_sfb; i++, idx++) {
2234 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2235 const float gain = cce->coup.gain[index][idx];
2236 for (group = 0; group < ics->group_len[g]; group++) {
2237 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2239 dest[group * 128 + k] += gain * src[group * 128 + k];
2244 dest += ics->group_len[g] * 128;
2245 src += ics->group_len[g] * 128;
2250 * Apply independent channel coupling (applied after IMDCT).
2252 * @param index index into coupling gain array
2254 static void apply_independent_coupling(AACContext *ac,
2255 SingleChannelElement *target,
2256 ChannelElement *cce, int index)
2259 const float gain = cce->coup.gain[index][0];
2260 const float *src = cce->ch[0].ret;
2261 float *dest = target->ret;
2262 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2264 for (i = 0; i < len; i++)
2265 dest[i] += gain * src[i];
2269 * channel coupling transformation interface
2271 * @param apply_coupling_method pointer to (in)dependent coupling function
2273 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2274 enum RawDataBlockType type, int elem_id,
2275 enum CouplingPoint coupling_point,
2276 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2280 for (i = 0; i < MAX_ELEM_ID; i++) {
2281 ChannelElement *cce = ac->che[TYPE_CCE][i];
2284 if (cce && cce->coup.coupling_point == coupling_point) {
2285 ChannelCoupling *coup = &cce->coup;
2287 for (c = 0; c <= coup->num_coupled; c++) {
2288 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2289 if (coup->ch_select[c] != 1) {
2290 apply_coupling_method(ac, &cc->ch[0], cce, index);
2291 if (coup->ch_select[c] != 0)
2294 if (coup->ch_select[c] != 2)
2295 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2297 index += 1 + (coup->ch_select[c] == 3);
2304 * Convert spectral data to float samples, applying all supported tools as appropriate.
2306 static void spectral_to_sample(AACContext *ac)
2309 for (type = 3; type >= 0; type--) {
2310 for (i = 0; i < MAX_ELEM_ID; i++) {
2311 ChannelElement *che = ac->che[type][i];
2313 if (type <= TYPE_CPE)
2314 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2315 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2316 if (che->ch[0].ics.predictor_present) {
2317 if (che->ch[0].ics.ltp.present)
2318 apply_ltp(ac, &che->ch[0]);
2319 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2320 apply_ltp(ac, &che->ch[1]);
2323 if (che->ch[0].tns.present)
2324 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2325 if (che->ch[1].tns.present)
2326 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2327 if (type <= TYPE_CPE)
2328 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2329 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2330 imdct_and_windowing(ac, &che->ch[0]);
2331 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2332 update_ltp(ac, &che->ch[0]);
2333 if (type == TYPE_CPE) {
2334 imdct_and_windowing(ac, &che->ch[1]);
2335 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2336 update_ltp(ac, &che->ch[1]);
2338 if (ac->oc[1].m4ac.sbr > 0) {
2339 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2342 if (type <= TYPE_CCE)
2343 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2349 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2352 AACADTSHeaderInfo hdr_info;
2353 uint8_t layout_map[MAX_ELEM_ID*4][3];
2354 int layout_map_tags;
2356 size = avpriv_aac_parse_header(gb, &hdr_info);
2358 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2359 // This is 2 for "VLB " audio in NSV files.
2360 // See samples/nsv/vlb_audio.
2361 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2362 ac->warned_num_aac_frames = 1;
2364 push_output_configuration(ac);
2365 if (hdr_info.chan_config) {
2366 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2367 if (set_default_channel_config(ac->avctx, layout_map,
2368 &layout_map_tags, hdr_info.chan_config))
2370 if (output_configure(ac, layout_map, layout_map_tags,
2371 hdr_info.chan_config,
2372 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME)))
2375 ac->oc[1].m4ac.chan_config = 0;
2377 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2378 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2379 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2380 if (ac->oc[0].status != OC_LOCKED ||
2381 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2382 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2383 ac->oc[1].m4ac.sbr = -1;
2384 ac->oc[1].m4ac.ps = -1;
2386 if (!hdr_info.crc_absent)
2392 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2393 int *got_frame_ptr, GetBitContext *gb)
2395 AACContext *ac = avctx->priv_data;
2396 ChannelElement *che = NULL, *che_prev = NULL;
2397 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2399 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2401 if (show_bits(gb, 12) == 0xfff) {
2402 if (parse_adts_frame_header(ac, gb) < 0) {
2403 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2407 if (ac->oc[1].m4ac.sampling_index > 12) {
2408 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2414 ac->tags_mapped = 0;
2416 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2417 elem_id = get_bits(gb, 4);
2419 if (elem_type < TYPE_DSE) {
2420 if (!(che=get_che(ac, elem_type, elem_id))) {
2421 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2422 elem_type, elem_id);
2429 switch (elem_type) {
2432 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2437 err = decode_cpe(ac, gb, che);
2442 err = decode_cce(ac, gb, che);
2446 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2451 err = skip_data_stream_element(ac, gb);
2455 uint8_t layout_map[MAX_ELEM_ID*4][3];
2457 push_output_configuration(ac);
2458 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2464 av_log(avctx, AV_LOG_ERROR,
2465 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2466 pop_output_configuration(ac);
2468 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2470 ac->oc[1].m4ac.chan_config = 0;
2478 elem_id += get_bits(gb, 8) - 1;
2479 if (get_bits_left(gb) < 8 * elem_id) {
2480 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2485 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2486 err = 0; /* FIXME */
2490 err = -1; /* should not happen, but keeps compiler happy */
2495 elem_type_prev = elem_type;
2500 if (get_bits_left(gb) < 3) {
2501 av_log(avctx, AV_LOG_ERROR, overread_err);
2507 spectral_to_sample(ac);
2509 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2510 samples <<= multiplier;
2513 /* get output buffer */
2514 ac->frame.nb_samples = samples;
2515 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2516 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2521 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2522 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2523 (const float **)ac->output_data,
2524 samples, avctx->channels);
2526 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2527 (const float **)ac->output_data,
2528 samples, avctx->channels);
2530 *(AVFrame *)data = ac->frame;
2532 *got_frame_ptr = !!samples;
2534 if (ac->oc[1].status && audio_found) {
2535 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2536 avctx->frame_size = samples;
2537 ac->oc[1].status = OC_LOCKED;
2542 pop_output_configuration(ac);
2546 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2547 int *got_frame_ptr, AVPacket *avpkt)
2549 AACContext *ac = avctx->priv_data;
2550 const uint8_t *buf = avpkt->data;
2551 int buf_size = avpkt->size;
2556 int new_extradata_size;
2557 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2558 AV_PKT_DATA_NEW_EXTRADATA,
2559 &new_extradata_size);
2561 if (new_extradata && 0) {
2562 av_free(avctx->extradata);
2563 avctx->extradata = av_mallocz(new_extradata_size +
2564 FF_INPUT_BUFFER_PADDING_SIZE);
2565 if (!avctx->extradata)
2566 return AVERROR(ENOMEM);
2567 avctx->extradata_size = new_extradata_size;
2568 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2569 push_output_configuration(ac);
2570 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2572 avctx->extradata_size*8, 1) < 0) {
2573 pop_output_configuration(ac);
2574 return AVERROR_INVALIDDATA;
2578 init_get_bits(&gb, buf, buf_size * 8);
2580 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2583 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2584 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2585 if (buf[buf_offset])
2588 return buf_size > buf_offset ? buf_consumed : buf_size;
2591 static av_cold int aac_decode_close(AVCodecContext *avctx)
2593 AACContext *ac = avctx->priv_data;
2596 for (i = 0; i < MAX_ELEM_ID; i++) {
2597 for (type = 0; type < 4; type++) {
2598 if (ac->che[type][i])
2599 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2600 av_freep(&ac->che[type][i]);
2604 ff_mdct_end(&ac->mdct);
2605 ff_mdct_end(&ac->mdct_small);
2606 ff_mdct_end(&ac->mdct_ltp);
2611 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2613 struct LATMContext {
2614 AACContext aac_ctx; ///< containing AACContext
2615 int initialized; ///< initilized after a valid extradata was seen
2618 int audio_mux_version_A; ///< LATM syntax version
2619 int frame_length_type; ///< 0/1 variable/fixed frame length
2620 int frame_length; ///< frame length for fixed frame length
2623 static inline uint32_t latm_get_value(GetBitContext *b)
2625 int length = get_bits(b, 2);
2627 return get_bits_long(b, (length+1)*8);
2630 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2631 GetBitContext *gb, int asclen)
2633 AACContext *ac = &latmctx->aac_ctx;
2634 AVCodecContext *avctx = ac->avctx;
2635 MPEG4AudioConfig m4ac = { 0 };
2636 int config_start_bit = get_bits_count(gb);
2637 int sync_extension = 0;
2638 int bits_consumed, esize;
2642 asclen = FFMIN(asclen, get_bits_left(gb));
2644 asclen = get_bits_left(gb);
2646 if (config_start_bit % 8) {
2647 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2648 "config not byte aligned.\n", 1);
2649 return AVERROR_INVALIDDATA;
2652 return AVERROR_INVALIDDATA;
2653 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2654 gb->buffer + (config_start_bit / 8),
2655 asclen, sync_extension);
2657 if (bits_consumed < 0)
2658 return AVERROR_INVALIDDATA;
2660 if (ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2661 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2663 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2664 latmctx->initialized = 0;
2666 esize = (bits_consumed+7) / 8;
2668 if (avctx->extradata_size < esize) {
2669 av_free(avctx->extradata);
2670 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2671 if (!avctx->extradata)
2672 return AVERROR(ENOMEM);
2675 avctx->extradata_size = esize;
2676 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2677 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2679 skip_bits_long(gb, bits_consumed);
2681 return bits_consumed;
2684 static int read_stream_mux_config(struct LATMContext *latmctx,
2687 int ret, audio_mux_version = get_bits(gb, 1);
2689 latmctx->audio_mux_version_A = 0;
2690 if (audio_mux_version)
2691 latmctx->audio_mux_version_A = get_bits(gb, 1);
2693 if (!latmctx->audio_mux_version_A) {
2695 if (audio_mux_version)
2696 latm_get_value(gb); // taraFullness
2698 skip_bits(gb, 1); // allStreamSameTimeFraming
2699 skip_bits(gb, 6); // numSubFrames
2701 if (get_bits(gb, 4)) { // numPrograms
2702 av_log_missing_feature(latmctx->aac_ctx.avctx,
2703 "multiple programs are not supported\n", 1);
2704 return AVERROR_PATCHWELCOME;
2707 // for each program (which there is only on in DVB)
2709 // for each layer (which there is only on in DVB)
2710 if (get_bits(gb, 3)) { // numLayer
2711 av_log_missing_feature(latmctx->aac_ctx.avctx,
2712 "multiple layers are not supported\n", 1);
2713 return AVERROR_PATCHWELCOME;
2716 // for all but first stream: use_same_config = get_bits(gb, 1);
2717 if (!audio_mux_version) {
2718 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2721 int ascLen = latm_get_value(gb);
2722 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2725 skip_bits_long(gb, ascLen);
2728 latmctx->frame_length_type = get_bits(gb, 3);
2729 switch (latmctx->frame_length_type) {
2731 skip_bits(gb, 8); // latmBufferFullness
2734 latmctx->frame_length = get_bits(gb, 9);
2739 skip_bits(gb, 6); // CELP frame length table index
2743 skip_bits(gb, 1); // HVXC frame length table index
2747 if (get_bits(gb, 1)) { // other data
2748 if (audio_mux_version) {
2749 latm_get_value(gb); // other_data_bits
2753 esc = get_bits(gb, 1);
2759 if (get_bits(gb, 1)) // crc present
2760 skip_bits(gb, 8); // config_crc
2766 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2770 if (ctx->frame_length_type == 0) {
2771 int mux_slot_length = 0;
2773 tmp = get_bits(gb, 8);
2774 mux_slot_length += tmp;
2775 } while (tmp == 255);
2776 return mux_slot_length;
2777 } else if (ctx->frame_length_type == 1) {
2778 return ctx->frame_length;
2779 } else if (ctx->frame_length_type == 3 ||
2780 ctx->frame_length_type == 5 ||
2781 ctx->frame_length_type == 7) {
2782 skip_bits(gb, 2); // mux_slot_length_coded
2787 static int read_audio_mux_element(struct LATMContext *latmctx,
2791 uint8_t use_same_mux = get_bits(gb, 1);
2792 if (!use_same_mux) {
2793 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2795 } else if (!latmctx->aac_ctx.avctx->extradata) {
2796 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2797 "no decoder config found\n");
2798 return AVERROR(EAGAIN);
2800 if (latmctx->audio_mux_version_A == 0) {
2801 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2802 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2803 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2804 return AVERROR_INVALIDDATA;
2805 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2806 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2807 "frame length mismatch %d << %d\n",
2808 mux_slot_length_bytes * 8, get_bits_left(gb));
2809 return AVERROR_INVALIDDATA;
2816 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2817 int *got_frame_ptr, AVPacket *avpkt)
2819 struct LATMContext *latmctx = avctx->priv_data;
2823 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2825 // check for LOAS sync word
2826 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2827 return AVERROR_INVALIDDATA;
2829 muxlength = get_bits(&gb, 13) + 3;
2830 // not enough data, the parser should have sorted this
2831 if (muxlength > avpkt->size)
2832 return AVERROR_INVALIDDATA;
2834 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2837 if (!latmctx->initialized) {
2838 if (!avctx->extradata) {
2842 push_output_configuration(&latmctx->aac_ctx);
2843 if ((err = decode_audio_specific_config(
2844 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2845 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2846 pop_output_configuration(&latmctx->aac_ctx);
2849 latmctx->initialized = 1;
2853 if (show_bits(&gb, 12) == 0xfff) {
2854 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2855 "ADTS header detected, probably as result of configuration "
2857 return AVERROR_INVALIDDATA;
2860 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2866 static av_cold int latm_decode_init(AVCodecContext *avctx)
2868 struct LATMContext *latmctx = avctx->priv_data;
2869 int ret = aac_decode_init(avctx);
2871 if (avctx->extradata_size > 0)
2872 latmctx->initialized = !ret;
2878 AVCodec ff_aac_decoder = {
2880 .type = AVMEDIA_TYPE_AUDIO,
2882 .priv_data_size = sizeof(AACContext),
2883 .init = aac_decode_init,
2884 .close = aac_decode_close,
2885 .decode = aac_decode_frame,
2886 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2887 .sample_fmts = (const enum AVSampleFormat[]) {
2888 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2890 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2891 .channel_layouts = aac_channel_layout,
2896 Note: This decoder filter is intended to decode LATM streams transferred
2897 in MPEG transport streams which only contain one program.
2898 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2900 AVCodec ff_aac_latm_decoder = {
2902 .type = AVMEDIA_TYPE_AUDIO,
2903 .id = CODEC_ID_AAC_LATM,
2904 .priv_data_size = sizeof(struct LATMContext),
2905 .init = latm_decode_init,
2906 .close = aac_decode_close,
2907 .decode = latm_decode_frame,
2908 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2909 .sample_fmts = (const enum AVSampleFormat[]) {
2910 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2912 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2913 .channel_layouts = aac_channel_layout,