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
82 #include "libavutil/float_dsp.h"
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;
490 ac->oc[1].m4ac.ps = 0;
493 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
494 uint8_t layout_map[MAX_ELEM_ID*4][3];
496 push_output_configuration(ac);
498 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
500 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
503 if (output_configure(ac, layout_map, layout_map_tags,
504 1, OC_TRIAL_FRAME) < 0)
507 ac->oc[1].m4ac.chan_config = 1;
508 if (ac->oc[1].m4ac.sbr)
509 ac->oc[1].m4ac.ps = -1;
511 // For indexed channel configurations map the channels solely based on position.
512 switch (ac->oc[1].m4ac.chan_config) {
514 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
516 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
519 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
520 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
521 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
522 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
524 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
527 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
529 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
532 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
534 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
538 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
540 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
541 } else if (ac->oc[1].m4ac.chan_config == 2) {
545 if (!ac->tags_mapped && type == TYPE_SCE) {
547 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
555 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
557 * @param type speaker type/position for these channels
559 static void decode_channel_map(uint8_t layout_map[][3],
560 enum ChannelPosition type,
561 GetBitContext *gb, int n)
564 enum RawDataBlockType syn_ele;
566 case AAC_CHANNEL_FRONT:
567 case AAC_CHANNEL_BACK:
568 case AAC_CHANNEL_SIDE:
569 syn_ele = get_bits1(gb);
575 case AAC_CHANNEL_LFE:
581 layout_map[0][0] = syn_ele;
582 layout_map[0][1] = get_bits(gb, 4);
583 layout_map[0][2] = type;
589 * Decode program configuration element; reference: table 4.2.
591 * @return Returns error status. 0 - OK, !0 - error
593 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
594 uint8_t (*layout_map)[3],
597 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
601 skip_bits(gb, 2); // object_type
603 sampling_index = get_bits(gb, 4);
604 if (m4ac->sampling_index != sampling_index)
605 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");
607 num_front = get_bits(gb, 4);
608 num_side = get_bits(gb, 4);
609 num_back = get_bits(gb, 4);
610 num_lfe = get_bits(gb, 2);
611 num_assoc_data = get_bits(gb, 3);
612 num_cc = get_bits(gb, 4);
615 skip_bits(gb, 4); // mono_mixdown_tag
617 skip_bits(gb, 4); // stereo_mixdown_tag
620 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
622 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
623 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
626 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
628 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
630 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
632 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
635 skip_bits_long(gb, 4 * num_assoc_data);
637 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
642 /* comment field, first byte is length */
643 comment_len = get_bits(gb, 8) * 8;
644 if (get_bits_left(gb) < comment_len) {
645 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
648 skip_bits_long(gb, comment_len);
653 * Decode GA "General Audio" specific configuration; reference: table 4.1.
655 * @param ac pointer to AACContext, may be null
656 * @param avctx pointer to AVCCodecContext, used for logging
658 * @return Returns error status. 0 - OK, !0 - error
660 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
662 MPEG4AudioConfig *m4ac,
665 int extension_flag, ret;
666 uint8_t layout_map[MAX_ELEM_ID*4][3];
669 if (get_bits1(gb)) { // frameLengthFlag
670 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
674 if (get_bits1(gb)) // dependsOnCoreCoder
675 skip_bits(gb, 14); // coreCoderDelay
676 extension_flag = get_bits1(gb);
678 if (m4ac->object_type == AOT_AAC_SCALABLE ||
679 m4ac->object_type == AOT_ER_AAC_SCALABLE)
680 skip_bits(gb, 3); // layerNr
682 if (channel_config == 0) {
683 skip_bits(gb, 4); // element_instance_tag
684 tags = decode_pce(avctx, m4ac, layout_map, gb);
688 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
692 if (count_channels(layout_map, tags) > 1) {
694 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
697 if (ac && (ret = output_configure(ac, layout_map, tags,
698 channel_config, OC_GLOBAL_HDR)))
701 if (extension_flag) {
702 switch (m4ac->object_type) {
704 skip_bits(gb, 5); // numOfSubFrame
705 skip_bits(gb, 11); // layer_length
709 case AOT_ER_AAC_SCALABLE:
711 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
712 * aacScalefactorDataResilienceFlag
713 * aacSpectralDataResilienceFlag
717 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
723 * Decode audio specific configuration; reference: table 1.13.
725 * @param ac pointer to AACContext, may be null
726 * @param avctx pointer to AVCCodecContext, used for logging
727 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
728 * @param data pointer to buffer holding an audio specific config
729 * @param bit_size size of audio specific config or data in bits
730 * @param sync_extension look for an appended sync extension
732 * @return Returns error status or number of consumed bits. <0 - error
734 static int decode_audio_specific_config(AACContext *ac,
735 AVCodecContext *avctx,
736 MPEG4AudioConfig *m4ac,
737 const uint8_t *data, int bit_size,
743 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
744 for (i = 0; i < bit_size >> 3; i++)
745 av_dlog(avctx, "%02x ", data[i]);
746 av_dlog(avctx, "\n");
748 init_get_bits(&gb, data, bit_size);
750 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
752 if (m4ac->sampling_index > 12) {
753 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
757 skip_bits_long(&gb, i);
759 switch (m4ac->object_type) {
763 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
767 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
768 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
772 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
773 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
774 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
776 return get_bits_count(&gb);
780 * linear congruential pseudorandom number generator
782 * @param previous_val pointer to the current state of the generator
784 * @return Returns a 32-bit pseudorandom integer
786 static av_always_inline int lcg_random(int previous_val)
788 return previous_val * 1664525 + 1013904223;
791 static av_always_inline void reset_predict_state(PredictorState *ps)
801 static void reset_all_predictors(PredictorState *ps)
804 for (i = 0; i < MAX_PREDICTORS; i++)
805 reset_predict_state(&ps[i]);
808 static int sample_rate_idx (int rate)
810 if (92017 <= rate) return 0;
811 else if (75132 <= rate) return 1;
812 else if (55426 <= rate) return 2;
813 else if (46009 <= rate) return 3;
814 else if (37566 <= rate) return 4;
815 else if (27713 <= rate) return 5;
816 else if (23004 <= rate) return 6;
817 else if (18783 <= rate) return 7;
818 else if (13856 <= rate) return 8;
819 else if (11502 <= rate) return 9;
820 else if (9391 <= rate) return 10;
824 static void reset_predictor_group(PredictorState *ps, int group_num)
827 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
828 reset_predict_state(&ps[i]);
831 #define AAC_INIT_VLC_STATIC(num, size) \
832 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
833 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
834 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
837 static av_cold int aac_decode_init(AVCodecContext *avctx)
839 AACContext *ac = avctx->priv_data;
840 float output_scale_factor;
843 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
845 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
846 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
847 output_scale_factor = 1.0 / 32768.0;
849 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
850 output_scale_factor = 1.0;
853 if (avctx->extradata_size > 0) {
854 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
856 avctx->extradata_size*8, 1) < 0)
860 uint8_t layout_map[MAX_ELEM_ID*4][3];
863 sr = sample_rate_idx(avctx->sample_rate);
864 ac->oc[1].m4ac.sampling_index = sr;
865 ac->oc[1].m4ac.channels = avctx->channels;
866 ac->oc[1].m4ac.sbr = -1;
867 ac->oc[1].m4ac.ps = -1;
869 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
870 if (ff_mpeg4audio_channels[i] == avctx->channels)
872 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
875 ac->oc[1].m4ac.chan_config = i;
877 if (ac->oc[1].m4ac.chan_config) {
878 int ret = set_default_channel_config(avctx, layout_map,
879 &layout_map_tags, ac->oc[1].m4ac.chan_config);
881 output_configure(ac, layout_map, layout_map_tags,
882 ac->oc[1].m4ac.chan_config, OC_GLOBAL_HDR);
883 else if (avctx->err_recognition & AV_EF_EXPLODE)
884 return AVERROR_INVALIDDATA;
888 AAC_INIT_VLC_STATIC( 0, 304);
889 AAC_INIT_VLC_STATIC( 1, 270);
890 AAC_INIT_VLC_STATIC( 2, 550);
891 AAC_INIT_VLC_STATIC( 3, 300);
892 AAC_INIT_VLC_STATIC( 4, 328);
893 AAC_INIT_VLC_STATIC( 5, 294);
894 AAC_INIT_VLC_STATIC( 6, 306);
895 AAC_INIT_VLC_STATIC( 7, 268);
896 AAC_INIT_VLC_STATIC( 8, 510);
897 AAC_INIT_VLC_STATIC( 9, 366);
898 AAC_INIT_VLC_STATIC(10, 462);
902 ff_dsputil_init(&ac->dsp, avctx);
903 ff_fmt_convert_init(&ac->fmt_conv, avctx);
904 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
906 ac->random_state = 0x1f2e3d4c;
910 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
911 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
912 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
915 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
916 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
917 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
918 // window initialization
919 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
920 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
921 ff_init_ff_sine_windows(10);
922 ff_init_ff_sine_windows( 7);
926 avcodec_get_frame_defaults(&ac->frame);
927 avctx->coded_frame = &ac->frame;
933 * Skip data_stream_element; reference: table 4.10.
935 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
937 int byte_align = get_bits1(gb);
938 int count = get_bits(gb, 8);
940 count += get_bits(gb, 8);
944 if (get_bits_left(gb) < 8 * count) {
945 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
948 skip_bits_long(gb, 8 * count);
952 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
957 ics->predictor_reset_group = get_bits(gb, 5);
958 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
959 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
963 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
964 ics->prediction_used[sfb] = get_bits1(gb);
970 * Decode Long Term Prediction data; reference: table 4.xx.
972 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
973 GetBitContext *gb, uint8_t max_sfb)
977 ltp->lag = get_bits(gb, 11);
978 ltp->coef = ltp_coef[get_bits(gb, 3)];
979 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
980 ltp->used[sfb] = get_bits1(gb);
984 * Decode Individual Channel Stream info; reference: table 4.6.
986 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
990 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
991 return AVERROR_INVALIDDATA;
993 ics->window_sequence[1] = ics->window_sequence[0];
994 ics->window_sequence[0] = get_bits(gb, 2);
995 ics->use_kb_window[1] = ics->use_kb_window[0];
996 ics->use_kb_window[0] = get_bits1(gb);
997 ics->num_window_groups = 1;
998 ics->group_len[0] = 1;
999 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1001 ics->max_sfb = get_bits(gb, 4);
1002 for (i = 0; i < 7; i++) {
1003 if (get_bits1(gb)) {
1004 ics->group_len[ics->num_window_groups - 1]++;
1006 ics->num_window_groups++;
1007 ics->group_len[ics->num_window_groups - 1] = 1;
1010 ics->num_windows = 8;
1011 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1012 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1013 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1014 ics->predictor_present = 0;
1016 ics->max_sfb = get_bits(gb, 6);
1017 ics->num_windows = 1;
1018 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1019 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1020 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1021 ics->predictor_present = get_bits1(gb);
1022 ics->predictor_reset_group = 0;
1023 if (ics->predictor_present) {
1024 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1025 if (decode_prediction(ac, ics, gb)) {
1028 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1029 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1032 if ((ics->ltp.present = get_bits(gb, 1)))
1033 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
1038 if (ics->max_sfb > ics->num_swb) {
1039 av_log(ac->avctx, AV_LOG_ERROR,
1040 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1041 ics->max_sfb, ics->num_swb);
1048 return AVERROR_INVALIDDATA;
1052 * Decode band types (section_data payload); reference: table 4.46.
1054 * @param band_type array of the used band type
1055 * @param band_type_run_end array of the last scalefactor band of a band type run
1057 * @return Returns error status. 0 - OK, !0 - error
1059 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1060 int band_type_run_end[120], GetBitContext *gb,
1061 IndividualChannelStream *ics)
1064 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1065 for (g = 0; g < ics->num_window_groups; g++) {
1067 while (k < ics->max_sfb) {
1068 uint8_t sect_end = k;
1070 int sect_band_type = get_bits(gb, 4);
1071 if (sect_band_type == 12) {
1072 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1076 sect_len_incr = get_bits(gb, bits);
1077 sect_end += sect_len_incr;
1078 if (get_bits_left(gb) < 0) {
1079 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1082 if (sect_end > ics->max_sfb) {
1083 av_log(ac->avctx, AV_LOG_ERROR,
1084 "Number of bands (%d) exceeds limit (%d).\n",
1085 sect_end, ics->max_sfb);
1088 } while (sect_len_incr == (1 << bits) - 1);
1089 for (; k < sect_end; k++) {
1090 band_type [idx] = sect_band_type;
1091 band_type_run_end[idx++] = sect_end;
1099 * Decode scalefactors; reference: table 4.47.
1101 * @param global_gain first scalefactor value as scalefactors are differentially coded
1102 * @param band_type array of the used band type
1103 * @param band_type_run_end array of the last scalefactor band of a band type run
1104 * @param sf array of scalefactors or intensity stereo positions
1106 * @return Returns error status. 0 - OK, !0 - error
1108 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1109 unsigned int global_gain,
1110 IndividualChannelStream *ics,
1111 enum BandType band_type[120],
1112 int band_type_run_end[120])
1115 int offset[3] = { global_gain, global_gain - 90, 0 };
1118 for (g = 0; g < ics->num_window_groups; g++) {
1119 for (i = 0; i < ics->max_sfb;) {
1120 int run_end = band_type_run_end[idx];
1121 if (band_type[idx] == ZERO_BT) {
1122 for (; i < run_end; i++, idx++)
1124 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1125 for (; i < run_end; i++, idx++) {
1126 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1127 clipped_offset = av_clip(offset[2], -155, 100);
1128 if (offset[2] != clipped_offset) {
1129 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1130 "position clipped (%d -> %d).\nIf you heard an "
1131 "audible artifact, there may be a bug in the "
1132 "decoder. ", offset[2], clipped_offset);
1134 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1136 } else if (band_type[idx] == NOISE_BT) {
1137 for (; i < run_end; i++, idx++) {
1138 if (noise_flag-- > 0)
1139 offset[1] += get_bits(gb, 9) - 256;
1141 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1142 clipped_offset = av_clip(offset[1], -100, 155);
1143 if (offset[1] != clipped_offset) {
1144 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1145 "(%d -> %d).\nIf you heard an audible "
1146 "artifact, there may be a bug in the decoder. ",
1147 offset[1], clipped_offset);
1149 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1152 for (; i < run_end; i++, idx++) {
1153 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1154 if (offset[0] > 255U) {
1155 av_log(ac->avctx, AV_LOG_ERROR,
1156 "Scalefactor (%d) out of range.\n", offset[0]);
1159 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1168 * Decode pulse data; reference: table 4.7.
1170 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1171 const uint16_t *swb_offset, int num_swb)
1174 pulse->num_pulse = get_bits(gb, 2) + 1;
1175 pulse_swb = get_bits(gb, 6);
1176 if (pulse_swb >= num_swb)
1178 pulse->pos[0] = swb_offset[pulse_swb];
1179 pulse->pos[0] += get_bits(gb, 5);
1180 if (pulse->pos[0] > 1023)
1182 pulse->amp[0] = get_bits(gb, 4);
1183 for (i = 1; i < pulse->num_pulse; i++) {
1184 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1185 if (pulse->pos[i] > 1023)
1187 pulse->amp[i] = get_bits(gb, 4);
1193 * Decode Temporal Noise Shaping data; reference: table 4.48.
1195 * @return Returns error status. 0 - OK, !0 - error
1197 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1198 GetBitContext *gb, const IndividualChannelStream *ics)
1200 int w, filt, i, coef_len, coef_res, coef_compress;
1201 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1202 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1203 for (w = 0; w < ics->num_windows; w++) {
1204 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1205 coef_res = get_bits1(gb);
1207 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1209 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1211 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1212 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1213 tns->order[w][filt], tns_max_order);
1214 tns->order[w][filt] = 0;
1217 if (tns->order[w][filt]) {
1218 tns->direction[w][filt] = get_bits1(gb);
1219 coef_compress = get_bits1(gb);
1220 coef_len = coef_res + 3 - coef_compress;
1221 tmp2_idx = 2 * coef_compress + coef_res;
1223 for (i = 0; i < tns->order[w][filt]; i++)
1224 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1233 * Decode Mid/Side data; reference: table 4.54.
1235 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1236 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1237 * [3] reserved for scalable AAC
1239 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1243 if (ms_present == 1) {
1244 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1245 cpe->ms_mask[idx] = get_bits1(gb);
1246 } else if (ms_present == 2) {
1247 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1252 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1256 *dst++ = v[idx & 15] * s;
1257 *dst++ = v[idx>>4 & 15] * s;
1263 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1267 *dst++ = v[idx & 3] * s;
1268 *dst++ = v[idx>>2 & 3] * s;
1269 *dst++ = v[idx>>4 & 3] * s;
1270 *dst++ = v[idx>>6 & 3] * s;
1276 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1277 unsigned sign, const float *scale)
1279 union av_intfloat32 s0, s1;
1281 s0.f = s1.f = *scale;
1282 s0.i ^= sign >> 1 << 31;
1285 *dst++ = v[idx & 15] * s0.f;
1286 *dst++ = v[idx>>4 & 15] * s1.f;
1293 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1294 unsigned sign, const float *scale)
1296 unsigned nz = idx >> 12;
1297 union av_intfloat32 s = { .f = *scale };
1298 union av_intfloat32 t;
1300 t.i = s.i ^ (sign & 1U<<31);
1301 *dst++ = v[idx & 3] * t.f;
1303 sign <<= nz & 1; nz >>= 1;
1304 t.i = s.i ^ (sign & 1U<<31);
1305 *dst++ = v[idx>>2 & 3] * t.f;
1307 sign <<= nz & 1; nz >>= 1;
1308 t.i = s.i ^ (sign & 1U<<31);
1309 *dst++ = v[idx>>4 & 3] * t.f;
1312 t.i = s.i ^ (sign & 1U<<31);
1313 *dst++ = v[idx>>6 & 3] * t.f;
1320 * Decode spectral data; reference: table 4.50.
1321 * Dequantize and scale spectral data; reference: 4.6.3.3.
1323 * @param coef array of dequantized, scaled spectral data
1324 * @param sf array of scalefactors or intensity stereo positions
1325 * @param pulse_present set if pulses are present
1326 * @param pulse pointer to pulse data struct
1327 * @param band_type array of the used band type
1329 * @return Returns error status. 0 - OK, !0 - error
1331 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1332 GetBitContext *gb, const float sf[120],
1333 int pulse_present, const Pulse *pulse,
1334 const IndividualChannelStream *ics,
1335 enum BandType band_type[120])
1337 int i, k, g, idx = 0;
1338 const int c = 1024 / ics->num_windows;
1339 const uint16_t *offsets = ics->swb_offset;
1340 float *coef_base = coef;
1342 for (g = 0; g < ics->num_windows; g++)
1343 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1345 for (g = 0; g < ics->num_window_groups; g++) {
1346 unsigned g_len = ics->group_len[g];
1348 for (i = 0; i < ics->max_sfb; i++, idx++) {
1349 const unsigned cbt_m1 = band_type[idx] - 1;
1350 float *cfo = coef + offsets[i];
1351 int off_len = offsets[i + 1] - offsets[i];
1354 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1355 for (group = 0; group < g_len; group++, cfo+=128) {
1356 memset(cfo, 0, off_len * sizeof(float));
1358 } else if (cbt_m1 == NOISE_BT - 1) {
1359 for (group = 0; group < g_len; group++, cfo+=128) {
1363 for (k = 0; k < off_len; k++) {
1364 ac->random_state = lcg_random(ac->random_state);
1365 cfo[k] = ac->random_state;
1368 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1369 scale = sf[idx] / sqrtf(band_energy);
1370 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1373 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1374 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1375 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1376 OPEN_READER(re, gb);
1378 switch (cbt_m1 >> 1) {
1380 for (group = 0; group < g_len; group++, cfo+=128) {
1388 UPDATE_CACHE(re, gb);
1389 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1390 cb_idx = cb_vector_idx[code];
1391 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1397 for (group = 0; group < g_len; group++, cfo+=128) {
1407 UPDATE_CACHE(re, gb);
1408 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1409 cb_idx = cb_vector_idx[code];
1410 nnz = cb_idx >> 8 & 15;
1411 bits = nnz ? GET_CACHE(re, gb) : 0;
1412 LAST_SKIP_BITS(re, gb, nnz);
1413 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1419 for (group = 0; group < g_len; group++, cfo+=128) {
1427 UPDATE_CACHE(re, gb);
1428 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1429 cb_idx = cb_vector_idx[code];
1430 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1437 for (group = 0; group < g_len; group++, cfo+=128) {
1447 UPDATE_CACHE(re, gb);
1448 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1449 cb_idx = cb_vector_idx[code];
1450 nnz = cb_idx >> 8 & 15;
1451 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1452 LAST_SKIP_BITS(re, gb, nnz);
1453 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1459 for (group = 0; group < g_len; group++, cfo+=128) {
1461 uint32_t *icf = (uint32_t *) cf;
1471 UPDATE_CACHE(re, gb);
1472 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1480 cb_idx = cb_vector_idx[code];
1483 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1484 LAST_SKIP_BITS(re, gb, nnz);
1486 for (j = 0; j < 2; j++) {
1490 /* The total length of escape_sequence must be < 22 bits according
1491 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1492 UPDATE_CACHE(re, gb);
1493 b = GET_CACHE(re, gb);
1494 b = 31 - av_log2(~b);
1497 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1501 SKIP_BITS(re, gb, b + 1);
1503 n = (1 << b) + SHOW_UBITS(re, gb, b);
1504 LAST_SKIP_BITS(re, gb, b);
1505 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1508 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1509 *icf++ = (bits & 1U<<31) | v;
1516 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1520 CLOSE_READER(re, gb);
1526 if (pulse_present) {
1528 for (i = 0; i < pulse->num_pulse; i++) {
1529 float co = coef_base[ pulse->pos[i] ];
1530 while (offsets[idx + 1] <= pulse->pos[i])
1532 if (band_type[idx] != NOISE_BT && sf[idx]) {
1533 float ico = -pulse->amp[i];
1536 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1538 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1545 static av_always_inline float flt16_round(float pf)
1547 union av_intfloat32 tmp;
1549 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1553 static av_always_inline float flt16_even(float pf)
1555 union av_intfloat32 tmp;
1557 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1561 static av_always_inline float flt16_trunc(float pf)
1563 union av_intfloat32 pun;
1565 pun.i &= 0xFFFF0000U;
1569 static av_always_inline void predict(PredictorState *ps, float *coef,
1572 const float a = 0.953125; // 61.0 / 64
1573 const float alpha = 0.90625; // 29.0 / 32
1577 float r0 = ps->r0, r1 = ps->r1;
1578 float cor0 = ps->cor0, cor1 = ps->cor1;
1579 float var0 = ps->var0, var1 = ps->var1;
1581 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1582 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1584 pv = flt16_round(k1 * r0 + k2 * r1);
1591 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1592 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1593 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1594 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1596 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1597 ps->r0 = flt16_trunc(a * e0);
1601 * Apply AAC-Main style frequency domain prediction.
1603 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1607 if (!sce->ics.predictor_initialized) {
1608 reset_all_predictors(sce->predictor_state);
1609 sce->ics.predictor_initialized = 1;
1612 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1613 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1614 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1615 predict(&sce->predictor_state[k], &sce->coeffs[k],
1616 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1619 if (sce->ics.predictor_reset_group)
1620 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1622 reset_all_predictors(sce->predictor_state);
1626 * Decode an individual_channel_stream payload; reference: table 4.44.
1628 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1629 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1631 * @return Returns error status. 0 - OK, !0 - error
1633 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1634 GetBitContext *gb, int common_window, int scale_flag)
1637 TemporalNoiseShaping *tns = &sce->tns;
1638 IndividualChannelStream *ics = &sce->ics;
1639 float *out = sce->coeffs;
1640 int global_gain, pulse_present = 0;
1642 /* This assignment is to silence a GCC warning about the variable being used
1643 * uninitialized when in fact it always is.
1645 pulse.num_pulse = 0;
1647 global_gain = get_bits(gb, 8);
1649 if (!common_window && !scale_flag) {
1650 if (decode_ics_info(ac, ics, gb) < 0)
1651 return AVERROR_INVALIDDATA;
1654 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1656 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1661 if ((pulse_present = get_bits1(gb))) {
1662 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1663 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1666 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1667 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1671 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1673 if (get_bits1(gb)) {
1674 av_log_missing_feature(ac->avctx, "SSR", 1);
1679 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1682 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1683 apply_prediction(ac, sce);
1689 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1691 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1693 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1694 float *ch0 = cpe->ch[0].coeffs;
1695 float *ch1 = cpe->ch[1].coeffs;
1696 int g, i, group, idx = 0;
1697 const uint16_t *offsets = ics->swb_offset;
1698 for (g = 0; g < ics->num_window_groups; g++) {
1699 for (i = 0; i < ics->max_sfb; i++, idx++) {
1700 if (cpe->ms_mask[idx] &&
1701 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1702 for (group = 0; group < ics->group_len[g]; group++) {
1703 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1704 ch1 + group * 128 + offsets[i],
1705 offsets[i+1] - offsets[i]);
1709 ch0 += ics->group_len[g] * 128;
1710 ch1 += ics->group_len[g] * 128;
1715 * intensity stereo decoding; reference: 4.6.8.2.3
1717 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1718 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1719 * [3] reserved for scalable AAC
1721 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1723 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1724 SingleChannelElement *sce1 = &cpe->ch[1];
1725 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1726 const uint16_t *offsets = ics->swb_offset;
1727 int g, group, i, idx = 0;
1730 for (g = 0; g < ics->num_window_groups; g++) {
1731 for (i = 0; i < ics->max_sfb;) {
1732 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1733 const int bt_run_end = sce1->band_type_run_end[idx];
1734 for (; i < bt_run_end; i++, idx++) {
1735 c = -1 + 2 * (sce1->band_type[idx] - 14);
1737 c *= 1 - 2 * cpe->ms_mask[idx];
1738 scale = c * sce1->sf[idx];
1739 for (group = 0; group < ics->group_len[g]; group++)
1740 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1741 coef0 + group * 128 + offsets[i],
1743 offsets[i + 1] - offsets[i]);
1746 int bt_run_end = sce1->band_type_run_end[idx];
1747 idx += bt_run_end - i;
1751 coef0 += ics->group_len[g] * 128;
1752 coef1 += ics->group_len[g] * 128;
1757 * Decode a channel_pair_element; reference: table 4.4.
1759 * @return Returns error status. 0 - OK, !0 - error
1761 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1763 int i, ret, common_window, ms_present = 0;
1765 common_window = get_bits1(gb);
1766 if (common_window) {
1767 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1768 return AVERROR_INVALIDDATA;
1769 i = cpe->ch[1].ics.use_kb_window[0];
1770 cpe->ch[1].ics = cpe->ch[0].ics;
1771 cpe->ch[1].ics.use_kb_window[1] = i;
1772 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1773 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1774 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1775 ms_present = get_bits(gb, 2);
1776 if (ms_present == 3) {
1777 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1779 } else if (ms_present)
1780 decode_mid_side_stereo(cpe, gb, ms_present);
1782 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1784 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1787 if (common_window) {
1789 apply_mid_side_stereo(ac, cpe);
1790 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1791 apply_prediction(ac, &cpe->ch[0]);
1792 apply_prediction(ac, &cpe->ch[1]);
1796 apply_intensity_stereo(ac, cpe, ms_present);
1800 static const float cce_scale[] = {
1801 1.09050773266525765921, //2^(1/8)
1802 1.18920711500272106672, //2^(1/4)
1808 * Decode coupling_channel_element; reference: table 4.8.
1810 * @return Returns error status. 0 - OK, !0 - error
1812 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1818 SingleChannelElement *sce = &che->ch[0];
1819 ChannelCoupling *coup = &che->coup;
1821 coup->coupling_point = 2 * get_bits1(gb);
1822 coup->num_coupled = get_bits(gb, 3);
1823 for (c = 0; c <= coup->num_coupled; c++) {
1825 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1826 coup->id_select[c] = get_bits(gb, 4);
1827 if (coup->type[c] == TYPE_CPE) {
1828 coup->ch_select[c] = get_bits(gb, 2);
1829 if (coup->ch_select[c] == 3)
1832 coup->ch_select[c] = 2;
1834 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1836 sign = get_bits(gb, 1);
1837 scale = cce_scale[get_bits(gb, 2)];
1839 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1842 for (c = 0; c < num_gain; c++) {
1846 float gain_cache = 1.;
1848 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1849 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1850 gain_cache = powf(scale, -gain);
1852 if (coup->coupling_point == AFTER_IMDCT) {
1853 coup->gain[c][0] = gain_cache;
1855 for (g = 0; g < sce->ics.num_window_groups; g++) {
1856 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1857 if (sce->band_type[idx] != ZERO_BT) {
1859 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1867 gain_cache = powf(scale, -t) * s;
1870 coup->gain[c][idx] = gain_cache;
1880 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1882 * @return Returns number of bytes consumed.
1884 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1888 int num_excl_chan = 0;
1891 for (i = 0; i < 7; i++)
1892 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1893 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1895 return num_excl_chan / 7;
1899 * Decode dynamic range information; reference: table 4.52.
1901 * @param cnt length of TYPE_FIL syntactic element in bytes
1903 * @return Returns number of bytes consumed.
1905 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1906 GetBitContext *gb, int cnt)
1909 int drc_num_bands = 1;
1912 /* pce_tag_present? */
1913 if (get_bits1(gb)) {
1914 che_drc->pce_instance_tag = get_bits(gb, 4);
1915 skip_bits(gb, 4); // tag_reserved_bits
1919 /* excluded_chns_present? */
1920 if (get_bits1(gb)) {
1921 n += decode_drc_channel_exclusions(che_drc, gb);
1924 /* drc_bands_present? */
1925 if (get_bits1(gb)) {
1926 che_drc->band_incr = get_bits(gb, 4);
1927 che_drc->interpolation_scheme = get_bits(gb, 4);
1929 drc_num_bands += che_drc->band_incr;
1930 for (i = 0; i < drc_num_bands; i++) {
1931 che_drc->band_top[i] = get_bits(gb, 8);
1936 /* prog_ref_level_present? */
1937 if (get_bits1(gb)) {
1938 che_drc->prog_ref_level = get_bits(gb, 7);
1939 skip_bits1(gb); // prog_ref_level_reserved_bits
1943 for (i = 0; i < drc_num_bands; i++) {
1944 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1945 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1953 * Decode extension data (incomplete); reference: table 4.51.
1955 * @param cnt length of TYPE_FIL syntactic element in bytes
1957 * @return Returns number of bytes consumed
1959 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1960 ChannelElement *che, enum RawDataBlockType elem_type)
1964 switch (get_bits(gb, 4)) { // extension type
1965 case EXT_SBR_DATA_CRC:
1969 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1971 } else if (!ac->oc[1].m4ac.sbr) {
1972 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1973 skip_bits_long(gb, 8 * cnt - 4);
1975 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
1976 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1977 skip_bits_long(gb, 8 * cnt - 4);
1979 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
1980 ac->oc[1].m4ac.sbr = 1;
1981 ac->oc[1].m4ac.ps = 1;
1982 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
1983 ac->oc[1].m4ac.chan_config, ac->oc[1].status);
1985 ac->oc[1].m4ac.sbr = 1;
1987 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1989 case EXT_DYNAMIC_RANGE:
1990 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1994 case EXT_DATA_ELEMENT:
1996 skip_bits_long(gb, 8 * cnt - 4);
2003 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2005 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2006 * @param coef spectral coefficients
2008 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2009 IndividualChannelStream *ics, int decode)
2011 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2013 int bottom, top, order, start, end, size, inc;
2014 float lpc[TNS_MAX_ORDER];
2015 float tmp[TNS_MAX_ORDER];
2017 for (w = 0; w < ics->num_windows; w++) {
2018 bottom = ics->num_swb;
2019 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2021 bottom = FFMAX(0, top - tns->length[w][filt]);
2022 order = tns->order[w][filt];
2027 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2029 start = ics->swb_offset[FFMIN(bottom, mmm)];
2030 end = ics->swb_offset[FFMIN( top, mmm)];
2031 if ((size = end - start) <= 0)
2033 if (tns->direction[w][filt]) {
2043 for (m = 0; m < size; m++, start += inc)
2044 for (i = 1; i <= FFMIN(m, order); i++)
2045 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2048 for (m = 0; m < size; m++, start += inc) {
2049 tmp[0] = coef[start];
2050 for (i = 1; i <= FFMIN(m, order); i++)
2051 coef[start] += tmp[i] * lpc[i - 1];
2052 for (i = order; i > 0; i--)
2053 tmp[i] = tmp[i - 1];
2061 * Apply windowing and MDCT to obtain the spectral
2062 * coefficient from the predicted sample by LTP.
2064 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2065 float *in, IndividualChannelStream *ics)
2067 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2068 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2069 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2070 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2072 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2073 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2075 memset(in, 0, 448 * sizeof(float));
2076 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2078 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2079 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2081 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2082 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2084 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2088 * Apply the long term prediction
2090 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2092 const LongTermPrediction *ltp = &sce->ics.ltp;
2093 const uint16_t *offsets = sce->ics.swb_offset;
2096 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2097 float *predTime = sce->ret;
2098 float *predFreq = ac->buf_mdct;
2099 int16_t num_samples = 2048;
2101 if (ltp->lag < 1024)
2102 num_samples = ltp->lag + 1024;
2103 for (i = 0; i < num_samples; i++)
2104 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2105 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2107 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2109 if (sce->tns.present)
2110 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2112 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2114 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2115 sce->coeffs[i] += predFreq[i];
2120 * Update the LTP buffer for next frame
2122 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2124 IndividualChannelStream *ics = &sce->ics;
2125 float *saved = sce->saved;
2126 float *saved_ltp = sce->coeffs;
2127 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2128 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2131 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2132 memcpy(saved_ltp, saved, 512 * sizeof(float));
2133 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2134 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2135 for (i = 0; i < 64; i++)
2136 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2137 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2138 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2139 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2140 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2141 for (i = 0; i < 64; i++)
2142 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2143 } else { // LONG_STOP or ONLY_LONG
2144 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2145 for (i = 0; i < 512; i++)
2146 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2149 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2150 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2151 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2155 * Conduct IMDCT and windowing.
2157 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2159 IndividualChannelStream *ics = &sce->ics;
2160 float *in = sce->coeffs;
2161 float *out = sce->ret;
2162 float *saved = sce->saved;
2163 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2164 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2165 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2166 float *buf = ac->buf_mdct;
2167 float *temp = ac->temp;
2171 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2172 for (i = 0; i < 1024; i += 128)
2173 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2175 ac->mdct.imdct_half(&ac->mdct, buf, in);
2177 /* window overlapping
2178 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2179 * and long to short transitions are considered to be short to short
2180 * transitions. This leaves just two cases (long to long and short to short)
2181 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2183 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2184 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2185 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2187 memcpy( out, saved, 448 * sizeof(float));
2189 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2190 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2191 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2192 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2193 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2194 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2195 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2197 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2198 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2203 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2204 memcpy( saved, temp + 64, 64 * sizeof(float));
2205 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2206 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2207 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2208 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2209 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2210 memcpy( saved, buf + 512, 448 * sizeof(float));
2211 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2212 } else { // LONG_STOP or ONLY_LONG
2213 memcpy( saved, buf + 512, 512 * sizeof(float));
2218 * Apply dependent channel coupling (applied before IMDCT).
2220 * @param index index into coupling gain array
2222 static void apply_dependent_coupling(AACContext *ac,
2223 SingleChannelElement *target,
2224 ChannelElement *cce, int index)
2226 IndividualChannelStream *ics = &cce->ch[0].ics;
2227 const uint16_t *offsets = ics->swb_offset;
2228 float *dest = target->coeffs;
2229 const float *src = cce->ch[0].coeffs;
2230 int g, i, group, k, idx = 0;
2231 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2232 av_log(ac->avctx, AV_LOG_ERROR,
2233 "Dependent coupling is not supported together with LTP\n");
2236 for (g = 0; g < ics->num_window_groups; g++) {
2237 for (i = 0; i < ics->max_sfb; i++, idx++) {
2238 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2239 const float gain = cce->coup.gain[index][idx];
2240 for (group = 0; group < ics->group_len[g]; group++) {
2241 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2243 dest[group * 128 + k] += gain * src[group * 128 + k];
2248 dest += ics->group_len[g] * 128;
2249 src += ics->group_len[g] * 128;
2254 * Apply independent channel coupling (applied after IMDCT).
2256 * @param index index into coupling gain array
2258 static void apply_independent_coupling(AACContext *ac,
2259 SingleChannelElement *target,
2260 ChannelElement *cce, int index)
2263 const float gain = cce->coup.gain[index][0];
2264 const float *src = cce->ch[0].ret;
2265 float *dest = target->ret;
2266 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2268 for (i = 0; i < len; i++)
2269 dest[i] += gain * src[i];
2273 * channel coupling transformation interface
2275 * @param apply_coupling_method pointer to (in)dependent coupling function
2277 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2278 enum RawDataBlockType type, int elem_id,
2279 enum CouplingPoint coupling_point,
2280 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2284 for (i = 0; i < MAX_ELEM_ID; i++) {
2285 ChannelElement *cce = ac->che[TYPE_CCE][i];
2288 if (cce && cce->coup.coupling_point == coupling_point) {
2289 ChannelCoupling *coup = &cce->coup;
2291 for (c = 0; c <= coup->num_coupled; c++) {
2292 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2293 if (coup->ch_select[c] != 1) {
2294 apply_coupling_method(ac, &cc->ch[0], cce, index);
2295 if (coup->ch_select[c] != 0)
2298 if (coup->ch_select[c] != 2)
2299 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2301 index += 1 + (coup->ch_select[c] == 3);
2308 * Convert spectral data to float samples, applying all supported tools as appropriate.
2310 static void spectral_to_sample(AACContext *ac)
2313 for (type = 3; type >= 0; type--) {
2314 for (i = 0; i < MAX_ELEM_ID; i++) {
2315 ChannelElement *che = ac->che[type][i];
2317 if (type <= TYPE_CPE)
2318 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2319 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2320 if (che->ch[0].ics.predictor_present) {
2321 if (che->ch[0].ics.ltp.present)
2322 apply_ltp(ac, &che->ch[0]);
2323 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2324 apply_ltp(ac, &che->ch[1]);
2327 if (che->ch[0].tns.present)
2328 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2329 if (che->ch[1].tns.present)
2330 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2331 if (type <= TYPE_CPE)
2332 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2333 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2334 imdct_and_windowing(ac, &che->ch[0]);
2335 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2336 update_ltp(ac, &che->ch[0]);
2337 if (type == TYPE_CPE) {
2338 imdct_and_windowing(ac, &che->ch[1]);
2339 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2340 update_ltp(ac, &che->ch[1]);
2342 if (ac->oc[1].m4ac.sbr > 0) {
2343 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2346 if (type <= TYPE_CCE)
2347 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2353 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2356 AACADTSHeaderInfo hdr_info;
2357 uint8_t layout_map[MAX_ELEM_ID*4][3];
2358 int layout_map_tags;
2360 size = avpriv_aac_parse_header(gb, &hdr_info);
2362 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2363 // This is 2 for "VLB " audio in NSV files.
2364 // See samples/nsv/vlb_audio.
2365 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2366 ac->warned_num_aac_frames = 1;
2368 push_output_configuration(ac);
2369 if (hdr_info.chan_config) {
2370 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2371 if (set_default_channel_config(ac->avctx, layout_map,
2372 &layout_map_tags, hdr_info.chan_config))
2374 if (output_configure(ac, layout_map, layout_map_tags,
2375 hdr_info.chan_config,
2376 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME)))
2379 ac->oc[1].m4ac.chan_config = 0;
2381 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2382 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2383 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2384 if (ac->oc[0].status != OC_LOCKED ||
2385 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2386 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2387 ac->oc[1].m4ac.sbr = -1;
2388 ac->oc[1].m4ac.ps = -1;
2390 if (!hdr_info.crc_absent)
2396 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2397 int *got_frame_ptr, GetBitContext *gb)
2399 AACContext *ac = avctx->priv_data;
2400 ChannelElement *che = NULL, *che_prev = NULL;
2401 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2403 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2405 if (show_bits(gb, 12) == 0xfff) {
2406 if (parse_adts_frame_header(ac, gb) < 0) {
2407 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2411 if (ac->oc[1].m4ac.sampling_index > 12) {
2412 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2418 ac->tags_mapped = 0;
2420 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2421 elem_id = get_bits(gb, 4);
2423 if (elem_type < TYPE_DSE) {
2424 if (!(che=get_che(ac, elem_type, elem_id))) {
2425 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2426 elem_type, elem_id);
2433 switch (elem_type) {
2436 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2441 err = decode_cpe(ac, gb, che);
2446 err = decode_cce(ac, gb, che);
2450 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2455 err = skip_data_stream_element(ac, gb);
2459 uint8_t layout_map[MAX_ELEM_ID*4][3];
2461 push_output_configuration(ac);
2462 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2468 av_log(avctx, AV_LOG_ERROR,
2469 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2470 pop_output_configuration(ac);
2472 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2474 ac->oc[1].m4ac.chan_config = 0;
2482 elem_id += get_bits(gb, 8) - 1;
2483 if (get_bits_left(gb) < 8 * elem_id) {
2484 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2489 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2490 err = 0; /* FIXME */
2494 err = -1; /* should not happen, but keeps compiler happy */
2499 elem_type_prev = elem_type;
2504 if (get_bits_left(gb) < 3) {
2505 av_log(avctx, AV_LOG_ERROR, overread_err);
2511 spectral_to_sample(ac);
2513 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2514 samples <<= multiplier;
2517 /* get output buffer */
2518 ac->frame.nb_samples = samples;
2519 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2520 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2525 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2526 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2527 (const float **)ac->output_data,
2528 samples, avctx->channels);
2530 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2531 (const float **)ac->output_data,
2532 samples, avctx->channels);
2534 *(AVFrame *)data = ac->frame;
2536 *got_frame_ptr = !!samples;
2538 if (ac->oc[1].status && audio_found) {
2539 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2540 avctx->frame_size = samples;
2541 ac->oc[1].status = OC_LOCKED;
2546 pop_output_configuration(ac);
2550 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2551 int *got_frame_ptr, AVPacket *avpkt)
2553 AACContext *ac = avctx->priv_data;
2554 const uint8_t *buf = avpkt->data;
2555 int buf_size = avpkt->size;
2560 int new_extradata_size;
2561 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2562 AV_PKT_DATA_NEW_EXTRADATA,
2563 &new_extradata_size);
2565 if (new_extradata && 0) {
2566 av_free(avctx->extradata);
2567 avctx->extradata = av_mallocz(new_extradata_size +
2568 FF_INPUT_BUFFER_PADDING_SIZE);
2569 if (!avctx->extradata)
2570 return AVERROR(ENOMEM);
2571 avctx->extradata_size = new_extradata_size;
2572 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2573 push_output_configuration(ac);
2574 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2576 avctx->extradata_size*8, 1) < 0) {
2577 pop_output_configuration(ac);
2578 return AVERROR_INVALIDDATA;
2582 init_get_bits(&gb, buf, buf_size * 8);
2584 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2587 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2588 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2589 if (buf[buf_offset])
2592 return buf_size > buf_offset ? buf_consumed : buf_size;
2595 static av_cold int aac_decode_close(AVCodecContext *avctx)
2597 AACContext *ac = avctx->priv_data;
2600 for (i = 0; i < MAX_ELEM_ID; i++) {
2601 for (type = 0; type < 4; type++) {
2602 if (ac->che[type][i])
2603 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2604 av_freep(&ac->che[type][i]);
2608 ff_mdct_end(&ac->mdct);
2609 ff_mdct_end(&ac->mdct_small);
2610 ff_mdct_end(&ac->mdct_ltp);
2615 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2617 struct LATMContext {
2618 AACContext aac_ctx; ///< containing AACContext
2619 int initialized; ///< initialized after a valid extradata was seen
2622 int audio_mux_version_A; ///< LATM syntax version
2623 int frame_length_type; ///< 0/1 variable/fixed frame length
2624 int frame_length; ///< frame length for fixed frame length
2627 static inline uint32_t latm_get_value(GetBitContext *b)
2629 int length = get_bits(b, 2);
2631 return get_bits_long(b, (length+1)*8);
2634 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2635 GetBitContext *gb, int asclen)
2637 AACContext *ac = &latmctx->aac_ctx;
2638 AVCodecContext *avctx = ac->avctx;
2639 MPEG4AudioConfig m4ac = { 0 };
2640 int config_start_bit = get_bits_count(gb);
2641 int sync_extension = 0;
2642 int bits_consumed, esize;
2646 asclen = FFMIN(asclen, get_bits_left(gb));
2648 asclen = get_bits_left(gb);
2650 if (config_start_bit % 8) {
2651 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2652 "config not byte aligned.\n", 1);
2653 return AVERROR_INVALIDDATA;
2656 return AVERROR_INVALIDDATA;
2657 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2658 gb->buffer + (config_start_bit / 8),
2659 asclen, sync_extension);
2661 if (bits_consumed < 0)
2662 return AVERROR_INVALIDDATA;
2664 if (!latmctx->initialized ||
2665 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2666 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2668 if(latmctx->initialized) {
2669 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2671 av_log(avctx, AV_LOG_INFO, "initializing latmctx\n");
2673 latmctx->initialized = 0;
2675 esize = (bits_consumed+7) / 8;
2677 if (avctx->extradata_size < esize) {
2678 av_free(avctx->extradata);
2679 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2680 if (!avctx->extradata)
2681 return AVERROR(ENOMEM);
2684 avctx->extradata_size = esize;
2685 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2686 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2688 skip_bits_long(gb, bits_consumed);
2690 return bits_consumed;
2693 static int read_stream_mux_config(struct LATMContext *latmctx,
2696 int ret, audio_mux_version = get_bits(gb, 1);
2698 latmctx->audio_mux_version_A = 0;
2699 if (audio_mux_version)
2700 latmctx->audio_mux_version_A = get_bits(gb, 1);
2702 if (!latmctx->audio_mux_version_A) {
2704 if (audio_mux_version)
2705 latm_get_value(gb); // taraFullness
2707 skip_bits(gb, 1); // allStreamSameTimeFraming
2708 skip_bits(gb, 6); // numSubFrames
2710 if (get_bits(gb, 4)) { // numPrograms
2711 av_log_missing_feature(latmctx->aac_ctx.avctx,
2712 "multiple programs are not supported\n", 1);
2713 return AVERROR_PATCHWELCOME;
2716 // for each program (which there is only on in DVB)
2718 // for each layer (which there is only on in DVB)
2719 if (get_bits(gb, 3)) { // numLayer
2720 av_log_missing_feature(latmctx->aac_ctx.avctx,
2721 "multiple layers are not supported\n", 1);
2722 return AVERROR_PATCHWELCOME;
2725 // for all but first stream: use_same_config = get_bits(gb, 1);
2726 if (!audio_mux_version) {
2727 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2730 int ascLen = latm_get_value(gb);
2731 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2734 skip_bits_long(gb, ascLen);
2737 latmctx->frame_length_type = get_bits(gb, 3);
2738 switch (latmctx->frame_length_type) {
2740 skip_bits(gb, 8); // latmBufferFullness
2743 latmctx->frame_length = get_bits(gb, 9);
2748 skip_bits(gb, 6); // CELP frame length table index
2752 skip_bits(gb, 1); // HVXC frame length table index
2756 if (get_bits(gb, 1)) { // other data
2757 if (audio_mux_version) {
2758 latm_get_value(gb); // other_data_bits
2762 esc = get_bits(gb, 1);
2768 if (get_bits(gb, 1)) // crc present
2769 skip_bits(gb, 8); // config_crc
2775 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2779 if (ctx->frame_length_type == 0) {
2780 int mux_slot_length = 0;
2782 tmp = get_bits(gb, 8);
2783 mux_slot_length += tmp;
2784 } while (tmp == 255);
2785 return mux_slot_length;
2786 } else if (ctx->frame_length_type == 1) {
2787 return ctx->frame_length;
2788 } else if (ctx->frame_length_type == 3 ||
2789 ctx->frame_length_type == 5 ||
2790 ctx->frame_length_type == 7) {
2791 skip_bits(gb, 2); // mux_slot_length_coded
2796 static int read_audio_mux_element(struct LATMContext *latmctx,
2800 uint8_t use_same_mux = get_bits(gb, 1);
2801 if (!use_same_mux) {
2802 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2804 } else if (!latmctx->aac_ctx.avctx->extradata) {
2805 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2806 "no decoder config found\n");
2807 return AVERROR(EAGAIN);
2809 if (latmctx->audio_mux_version_A == 0) {
2810 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2811 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2812 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2813 return AVERROR_INVALIDDATA;
2814 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2815 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2816 "frame length mismatch %d << %d\n",
2817 mux_slot_length_bytes * 8, get_bits_left(gb));
2818 return AVERROR_INVALIDDATA;
2825 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2826 int *got_frame_ptr, AVPacket *avpkt)
2828 struct LATMContext *latmctx = avctx->priv_data;
2832 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2834 // check for LOAS sync word
2835 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2836 return AVERROR_INVALIDDATA;
2838 muxlength = get_bits(&gb, 13) + 3;
2839 // not enough data, the parser should have sorted this
2840 if (muxlength > avpkt->size)
2841 return AVERROR_INVALIDDATA;
2843 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2846 if (!latmctx->initialized) {
2847 if (!avctx->extradata) {
2851 push_output_configuration(&latmctx->aac_ctx);
2852 if ((err = decode_audio_specific_config(
2853 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2854 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2855 pop_output_configuration(&latmctx->aac_ctx);
2858 latmctx->initialized = 1;
2862 if (show_bits(&gb, 12) == 0xfff) {
2863 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2864 "ADTS header detected, probably as result of configuration "
2866 return AVERROR_INVALIDDATA;
2869 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2875 static av_cold int latm_decode_init(AVCodecContext *avctx)
2877 struct LATMContext *latmctx = avctx->priv_data;
2878 int ret = aac_decode_init(avctx);
2880 if (avctx->extradata_size > 0)
2881 latmctx->initialized = !ret;
2887 AVCodec ff_aac_decoder = {
2889 .type = AVMEDIA_TYPE_AUDIO,
2890 .id = AV_CODEC_ID_AAC,
2891 .priv_data_size = sizeof(AACContext),
2892 .init = aac_decode_init,
2893 .close = aac_decode_close,
2894 .decode = aac_decode_frame,
2895 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
2896 .sample_fmts = (const enum AVSampleFormat[]) {
2897 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2899 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2900 .channel_layouts = aac_channel_layout,
2905 Note: This decoder filter is intended to decode LATM streams transferred
2906 in MPEG transport streams which only contain one program.
2907 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2909 AVCodec ff_aac_latm_decoder = {
2911 .type = AVMEDIA_TYPE_AUDIO,
2912 .id = AV_CODEC_ID_AAC_LATM,
2913 .priv_data_size = sizeof(struct LATMContext),
2914 .init = latm_decode_init,
2915 .close = aac_decode_close,
2916 .decode = latm_decode_frame,
2917 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
2918 .sample_fmts = (const enum AVSampleFormat[]) {
2919 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2921 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2922 .channel_layouts = aac_channel_layout,