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, 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,
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,
271 num_front_channels -= 2;
273 while (num_front_channels >= 2) {
274 i += assign_pair(e2c_vec, layout_map, i,
278 num_front_channels -= 2;
281 if (num_side_channels >= 2) {
282 i += assign_pair(e2c_vec, layout_map, i,
286 num_side_channels -= 2;
288 while (num_side_channels >= 2) {
289 i += assign_pair(e2c_vec, layout_map, i,
293 num_side_channels -= 2;
296 while (num_back_channels >= 4) {
297 i += assign_pair(e2c_vec, layout_map, i,
301 num_back_channels -= 2;
303 if (num_back_channels >= 2) {
304 i += assign_pair(e2c_vec, layout_map, i,
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 && ac->oc[0].status != OC_NONE) {
373 ac->oc[1] = ac->oc[0];
374 ac->avctx->channels = ac->oc[1].channels;
375 ac->avctx->channel_layout = ac->oc[1].channel_layout;
380 * Configure output channel order based on the current program configuration element.
382 * @return Returns error status. 0 - OK, !0 - error
384 static int output_configure(AACContext *ac,
385 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
386 enum OCStatus oc_type)
388 AVCodecContext *avctx = ac->avctx;
389 int i, channels = 0, ret;
392 if (ac->oc[1].layout_map != layout_map) {
393 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
394 ac->oc[1].layout_map_tags = tags;
397 // Try to sniff a reasonable channel order, otherwise output the
398 // channels in the order the PCE declared them.
399 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
400 layout = sniff_channel_order(layout_map, tags);
401 for (i = 0; i < tags; i++) {
402 int type = layout_map[i][0];
403 int id = layout_map[i][1];
404 int position = layout_map[i][2];
405 // Allocate or free elements depending on if they are in the
406 // current program configuration.
407 ret = che_configure(ac, position, type, id, &channels);
411 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
412 if (layout == AV_CH_FRONT_CENTER) {
413 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
419 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
420 if (layout) avctx->channel_layout = layout;
421 ac->oc[1].channel_layout = layout;
422 avctx->channels = ac->oc[1].channels = channels;
423 ac->oc[1].status = oc_type;
428 static void flush(AVCodecContext *avctx)
430 AACContext *ac= avctx->priv_data;
433 for (type = 3; type >= 0; type--) {
434 for (i = 0; i < MAX_ELEM_ID; i++) {
435 ChannelElement *che = ac->che[type][i];
437 for (j = 0; j <= 1; j++) {
438 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
446 * Set up channel positions based on a default channel configuration
447 * as specified in table 1.17.
449 * @return Returns error status. 0 - OK, !0 - error
451 static int set_default_channel_config(AVCodecContext *avctx,
452 uint8_t (*layout_map)[3],
456 if (channel_config < 1 || channel_config > 7) {
457 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
461 *tags = tags_per_config[channel_config];
462 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
466 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
468 // For PCE based channel configurations map the channels solely based on tags.
469 if (!ac->oc[1].m4ac.chan_config) {
470 return ac->tag_che_map[type][elem_id];
472 // Allow single CPE stereo files to be signalled with mono configuration.
473 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
474 uint8_t layout_map[MAX_ELEM_ID*4][3];
476 push_output_configuration(ac);
478 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
480 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
483 if (output_configure(ac, layout_map, layout_map_tags,
487 ac->oc[1].m4ac.chan_config = 2;
488 ac->oc[1].m4ac.ps = 0;
491 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
492 uint8_t layout_map[MAX_ELEM_ID*4][3];
494 push_output_configuration(ac);
496 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
498 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
501 if (output_configure(ac, layout_map, layout_map_tags,
505 ac->oc[1].m4ac.chan_config = 1;
506 if (ac->oc[1].m4ac.sbr)
507 ac->oc[1].m4ac.ps = -1;
509 // For indexed channel configurations map the channels solely based on position.
510 switch (ac->oc[1].m4ac.chan_config) {
512 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
514 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
517 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
518 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
519 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
520 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
522 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
525 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
527 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
530 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
532 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
536 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
538 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
539 } else if (ac->oc[1].m4ac.chan_config == 2) {
543 if (!ac->tags_mapped && type == TYPE_SCE) {
545 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
553 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
555 * @param type speaker type/position for these channels
557 static void decode_channel_map(uint8_t layout_map[][3],
558 enum ChannelPosition type,
559 GetBitContext *gb, int n)
562 enum RawDataBlockType syn_ele;
564 case AAC_CHANNEL_FRONT:
565 case AAC_CHANNEL_BACK:
566 case AAC_CHANNEL_SIDE:
567 syn_ele = get_bits1(gb);
573 case AAC_CHANNEL_LFE:
579 layout_map[0][0] = syn_ele;
580 layout_map[0][1] = get_bits(gb, 4);
581 layout_map[0][2] = type;
587 * Decode program configuration element; reference: table 4.2.
589 * @return Returns error status. 0 - OK, !0 - error
591 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
592 uint8_t (*layout_map)[3],
595 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
599 skip_bits(gb, 2); // object_type
601 sampling_index = get_bits(gb, 4);
602 if (m4ac->sampling_index != sampling_index)
603 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");
605 num_front = get_bits(gb, 4);
606 num_side = get_bits(gb, 4);
607 num_back = get_bits(gb, 4);
608 num_lfe = get_bits(gb, 2);
609 num_assoc_data = get_bits(gb, 3);
610 num_cc = get_bits(gb, 4);
613 skip_bits(gb, 4); // mono_mixdown_tag
615 skip_bits(gb, 4); // stereo_mixdown_tag
618 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
620 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
621 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
624 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
626 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
628 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
630 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
633 skip_bits_long(gb, 4 * num_assoc_data);
635 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
640 /* comment field, first byte is length */
641 comment_len = get_bits(gb, 8) * 8;
642 if (get_bits_left(gb) < comment_len) {
643 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
646 skip_bits_long(gb, comment_len);
651 * Decode GA "General Audio" specific configuration; reference: table 4.1.
653 * @param ac pointer to AACContext, may be null
654 * @param avctx pointer to AVCCodecContext, used for logging
656 * @return Returns error status. 0 - OK, !0 - error
658 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
660 MPEG4AudioConfig *m4ac,
663 int extension_flag, ret;
664 uint8_t layout_map[MAX_ELEM_ID*4][3];
667 if (get_bits1(gb)) { // frameLengthFlag
668 av_log_missing_feature(avctx, "960/120 MDCT window", 1);
669 return AVERROR_PATCHWELCOME;
672 if (get_bits1(gb)) // dependsOnCoreCoder
673 skip_bits(gb, 14); // coreCoderDelay
674 extension_flag = get_bits1(gb);
676 if (m4ac->object_type == AOT_AAC_SCALABLE ||
677 m4ac->object_type == AOT_ER_AAC_SCALABLE)
678 skip_bits(gb, 3); // layerNr
680 if (channel_config == 0) {
681 skip_bits(gb, 4); // element_instance_tag
682 tags = decode_pce(avctx, m4ac, layout_map, gb);
686 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
690 if (count_channels(layout_map, tags) > 1) {
692 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
695 if (ac && (ret = output_configure(ac, layout_map, tags, 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(unsigned 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->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
843 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
844 output_scale_factor = 1.0 / 32768.0;
846 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
847 output_scale_factor = 1.0;
850 if (avctx->extradata_size > 0) {
851 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
853 avctx->extradata_size*8, 1) < 0)
857 uint8_t layout_map[MAX_ELEM_ID*4][3];
860 sr = sample_rate_idx(avctx->sample_rate);
861 ac->oc[1].m4ac.sampling_index = sr;
862 ac->oc[1].m4ac.channels = avctx->channels;
863 ac->oc[1].m4ac.sbr = -1;
864 ac->oc[1].m4ac.ps = -1;
866 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
867 if (ff_mpeg4audio_channels[i] == avctx->channels)
869 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
872 ac->oc[1].m4ac.chan_config = i;
874 if (ac->oc[1].m4ac.chan_config) {
875 int ret = set_default_channel_config(avctx, layout_map,
876 &layout_map_tags, ac->oc[1].m4ac.chan_config);
878 output_configure(ac, layout_map, layout_map_tags,
880 else if (avctx->err_recognition & AV_EF_EXPLODE)
881 return AVERROR_INVALIDDATA;
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);
901 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
903 ac->random_state = 0x1f2e3d4c;
907 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
908 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
909 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
912 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
913 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
914 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
915 // window initialization
916 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
917 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
918 ff_init_ff_sine_windows(10);
919 ff_init_ff_sine_windows( 7);
923 avcodec_get_frame_defaults(&ac->frame);
924 avctx->coded_frame = &ac->frame;
930 * Skip data_stream_element; reference: table 4.10.
932 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
934 int byte_align = get_bits1(gb);
935 int count = get_bits(gb, 8);
937 count += get_bits(gb, 8);
941 if (get_bits_left(gb) < 8 * count) {
942 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
945 skip_bits_long(gb, 8 * count);
949 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
954 ics->predictor_reset_group = get_bits(gb, 5);
955 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
956 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
960 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
961 ics->prediction_used[sfb] = get_bits1(gb);
967 * Decode Long Term Prediction data; reference: table 4.xx.
969 static void decode_ltp(LongTermPrediction *ltp,
970 GetBitContext *gb, uint8_t max_sfb)
974 ltp->lag = get_bits(gb, 11);
975 ltp->coef = ltp_coef[get_bits(gb, 3)];
976 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
977 ltp->used[sfb] = get_bits1(gb);
981 * Decode Individual Channel Stream info; reference: table 4.6.
983 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
987 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
988 return AVERROR_INVALIDDATA;
990 ics->window_sequence[1] = ics->window_sequence[0];
991 ics->window_sequence[0] = get_bits(gb, 2);
992 ics->use_kb_window[1] = ics->use_kb_window[0];
993 ics->use_kb_window[0] = get_bits1(gb);
994 ics->num_window_groups = 1;
995 ics->group_len[0] = 1;
996 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
998 ics->max_sfb = get_bits(gb, 4);
999 for (i = 0; i < 7; i++) {
1000 if (get_bits1(gb)) {
1001 ics->group_len[ics->num_window_groups - 1]++;
1003 ics->num_window_groups++;
1004 ics->group_len[ics->num_window_groups - 1] = 1;
1007 ics->num_windows = 8;
1008 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1009 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1010 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1011 ics->predictor_present = 0;
1013 ics->max_sfb = get_bits(gb, 6);
1014 ics->num_windows = 1;
1015 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1016 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1017 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1018 ics->predictor_present = get_bits1(gb);
1019 ics->predictor_reset_group = 0;
1020 if (ics->predictor_present) {
1021 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1022 if (decode_prediction(ac, ics, gb)) {
1025 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1026 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1029 if ((ics->ltp.present = get_bits(gb, 1)))
1030 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1035 if (ics->max_sfb > ics->num_swb) {
1036 av_log(ac->avctx, AV_LOG_ERROR,
1037 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1038 ics->max_sfb, ics->num_swb);
1045 return AVERROR_INVALIDDATA;
1049 * Decode band types (section_data payload); reference: table 4.46.
1051 * @param band_type array of the used band type
1052 * @param band_type_run_end array of the last scalefactor band of a band type run
1054 * @return Returns error status. 0 - OK, !0 - error
1056 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1057 int band_type_run_end[120], GetBitContext *gb,
1058 IndividualChannelStream *ics)
1061 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1062 for (g = 0; g < ics->num_window_groups; g++) {
1064 while (k < ics->max_sfb) {
1065 uint8_t sect_end = k;
1067 int sect_band_type = get_bits(gb, 4);
1068 if (sect_band_type == 12) {
1069 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1073 sect_len_incr = get_bits(gb, bits);
1074 sect_end += sect_len_incr;
1075 if (get_bits_left(gb) < 0) {
1076 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1079 if (sect_end > ics->max_sfb) {
1080 av_log(ac->avctx, AV_LOG_ERROR,
1081 "Number of bands (%d) exceeds limit (%d).\n",
1082 sect_end, ics->max_sfb);
1085 } while (sect_len_incr == (1 << bits) - 1);
1086 for (; k < sect_end; k++) {
1087 band_type [idx] = sect_band_type;
1088 band_type_run_end[idx++] = sect_end;
1096 * Decode scalefactors; reference: table 4.47.
1098 * @param global_gain first scalefactor value as scalefactors are differentially coded
1099 * @param band_type array of the used band type
1100 * @param band_type_run_end array of the last scalefactor band of a band type run
1101 * @param sf array of scalefactors or intensity stereo positions
1103 * @return Returns error status. 0 - OK, !0 - error
1105 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1106 unsigned int global_gain,
1107 IndividualChannelStream *ics,
1108 enum BandType band_type[120],
1109 int band_type_run_end[120])
1112 int offset[3] = { global_gain, global_gain - 90, 0 };
1115 for (g = 0; g < ics->num_window_groups; g++) {
1116 for (i = 0; i < ics->max_sfb;) {
1117 int run_end = band_type_run_end[idx];
1118 if (band_type[idx] == ZERO_BT) {
1119 for (; i < run_end; i++, idx++)
1121 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1122 for (; i < run_end; i++, idx++) {
1123 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1124 clipped_offset = av_clip(offset[2], -155, 100);
1125 if (offset[2] != clipped_offset) {
1126 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1127 "position clipped (%d -> %d).\nIf you heard an "
1128 "audible artifact, there may be a bug in the "
1129 "decoder. ", offset[2], clipped_offset);
1131 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1133 } else if (band_type[idx] == NOISE_BT) {
1134 for (; i < run_end; i++, idx++) {
1135 if (noise_flag-- > 0)
1136 offset[1] += get_bits(gb, 9) - 256;
1138 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1139 clipped_offset = av_clip(offset[1], -100, 155);
1140 if (offset[1] != clipped_offset) {
1141 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1142 "(%d -> %d).\nIf you heard an audible "
1143 "artifact, there may be a bug in the decoder. ",
1144 offset[1], clipped_offset);
1146 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1149 for (; i < run_end; i++, idx++) {
1150 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1151 if (offset[0] > 255U) {
1152 av_log(ac->avctx, AV_LOG_ERROR,
1153 "Scalefactor (%d) out of range.\n", offset[0]);
1156 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1165 * Decode pulse data; reference: table 4.7.
1167 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1168 const uint16_t *swb_offset, int num_swb)
1171 pulse->num_pulse = get_bits(gb, 2) + 1;
1172 pulse_swb = get_bits(gb, 6);
1173 if (pulse_swb >= num_swb)
1175 pulse->pos[0] = swb_offset[pulse_swb];
1176 pulse->pos[0] += get_bits(gb, 5);
1177 if (pulse->pos[0] > 1023)
1179 pulse->amp[0] = get_bits(gb, 4);
1180 for (i = 1; i < pulse->num_pulse; i++) {
1181 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1182 if (pulse->pos[i] > 1023)
1184 pulse->amp[i] = get_bits(gb, 4);
1190 * Decode Temporal Noise Shaping data; reference: table 4.48.
1192 * @return Returns error status. 0 - OK, !0 - error
1194 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1195 GetBitContext *gb, const IndividualChannelStream *ics)
1197 int w, filt, i, coef_len, coef_res, coef_compress;
1198 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1199 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1200 for (w = 0; w < ics->num_windows; w++) {
1201 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1202 coef_res = get_bits1(gb);
1204 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1206 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1208 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1209 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1210 tns->order[w][filt], tns_max_order);
1211 tns->order[w][filt] = 0;
1214 if (tns->order[w][filt]) {
1215 tns->direction[w][filt] = get_bits1(gb);
1216 coef_compress = get_bits1(gb);
1217 coef_len = coef_res + 3 - coef_compress;
1218 tmp2_idx = 2 * coef_compress + coef_res;
1220 for (i = 0; i < tns->order[w][filt]; i++)
1221 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1230 * Decode Mid/Side data; reference: table 4.54.
1232 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1233 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1234 * [3] reserved for scalable AAC
1236 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1240 if (ms_present == 1) {
1241 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1242 cpe->ms_mask[idx] = get_bits1(gb);
1243 } else if (ms_present == 2) {
1244 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1249 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1253 *dst++ = v[idx & 15] * s;
1254 *dst++ = v[idx>>4 & 15] * s;
1260 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1264 *dst++ = v[idx & 3] * s;
1265 *dst++ = v[idx>>2 & 3] * s;
1266 *dst++ = v[idx>>4 & 3] * s;
1267 *dst++ = v[idx>>6 & 3] * s;
1273 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1274 unsigned sign, const float *scale)
1276 union av_intfloat32 s0, s1;
1278 s0.f = s1.f = *scale;
1279 s0.i ^= sign >> 1 << 31;
1282 *dst++ = v[idx & 15] * s0.f;
1283 *dst++ = v[idx>>4 & 15] * s1.f;
1290 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1291 unsigned sign, const float *scale)
1293 unsigned nz = idx >> 12;
1294 union av_intfloat32 s = { .f = *scale };
1295 union av_intfloat32 t;
1297 t.i = s.i ^ (sign & 1U<<31);
1298 *dst++ = v[idx & 3] * t.f;
1300 sign <<= nz & 1; nz >>= 1;
1301 t.i = s.i ^ (sign & 1U<<31);
1302 *dst++ = v[idx>>2 & 3] * t.f;
1304 sign <<= nz & 1; nz >>= 1;
1305 t.i = s.i ^ (sign & 1U<<31);
1306 *dst++ = v[idx>>4 & 3] * t.f;
1309 t.i = s.i ^ (sign & 1U<<31);
1310 *dst++ = v[idx>>6 & 3] * t.f;
1317 * Decode spectral data; reference: table 4.50.
1318 * Dequantize and scale spectral data; reference: 4.6.3.3.
1320 * @param coef array of dequantized, scaled spectral data
1321 * @param sf array of scalefactors or intensity stereo positions
1322 * @param pulse_present set if pulses are present
1323 * @param pulse pointer to pulse data struct
1324 * @param band_type array of the used band type
1326 * @return Returns error status. 0 - OK, !0 - error
1328 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1329 GetBitContext *gb, const float sf[120],
1330 int pulse_present, const Pulse *pulse,
1331 const IndividualChannelStream *ics,
1332 enum BandType band_type[120])
1334 int i, k, g, idx = 0;
1335 const int c = 1024 / ics->num_windows;
1336 const uint16_t *offsets = ics->swb_offset;
1337 float *coef_base = coef;
1339 for (g = 0; g < ics->num_windows; g++)
1340 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1342 for (g = 0; g < ics->num_window_groups; g++) {
1343 unsigned g_len = ics->group_len[g];
1345 for (i = 0; i < ics->max_sfb; i++, idx++) {
1346 const unsigned cbt_m1 = band_type[idx] - 1;
1347 float *cfo = coef + offsets[i];
1348 int off_len = offsets[i + 1] - offsets[i];
1351 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1352 for (group = 0; group < g_len; group++, cfo+=128) {
1353 memset(cfo, 0, off_len * sizeof(float));
1355 } else if (cbt_m1 == NOISE_BT - 1) {
1356 for (group = 0; group < g_len; group++, cfo+=128) {
1360 for (k = 0; k < off_len; k++) {
1361 ac->random_state = lcg_random(ac->random_state);
1362 cfo[k] = ac->random_state;
1365 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1366 scale = sf[idx] / sqrtf(band_energy);
1367 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1370 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1371 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1372 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1373 OPEN_READER(re, gb);
1375 switch (cbt_m1 >> 1) {
1377 for (group = 0; group < g_len; group++, cfo+=128) {
1385 UPDATE_CACHE(re, gb);
1386 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1387 cb_idx = cb_vector_idx[code];
1388 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1394 for (group = 0; group < g_len; group++, cfo+=128) {
1404 UPDATE_CACHE(re, gb);
1405 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1406 cb_idx = cb_vector_idx[code];
1407 nnz = cb_idx >> 8 & 15;
1408 bits = nnz ? GET_CACHE(re, gb) : 0;
1409 LAST_SKIP_BITS(re, gb, nnz);
1410 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1416 for (group = 0; group < g_len; group++, cfo+=128) {
1424 UPDATE_CACHE(re, gb);
1425 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1426 cb_idx = cb_vector_idx[code];
1427 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1434 for (group = 0; group < g_len; group++, cfo+=128) {
1444 UPDATE_CACHE(re, gb);
1445 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1446 cb_idx = cb_vector_idx[code];
1447 nnz = cb_idx >> 8 & 15;
1448 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1449 LAST_SKIP_BITS(re, gb, nnz);
1450 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1456 for (group = 0; group < g_len; group++, cfo+=128) {
1458 uint32_t *icf = (uint32_t *) cf;
1468 UPDATE_CACHE(re, gb);
1469 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1477 cb_idx = cb_vector_idx[code];
1480 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1481 LAST_SKIP_BITS(re, gb, nnz);
1483 for (j = 0; j < 2; j++) {
1487 /* The total length of escape_sequence must be < 22 bits according
1488 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1489 UPDATE_CACHE(re, gb);
1490 b = GET_CACHE(re, gb);
1491 b = 31 - av_log2(~b);
1494 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1498 SKIP_BITS(re, gb, b + 1);
1500 n = (1 << b) + SHOW_UBITS(re, gb, b);
1501 LAST_SKIP_BITS(re, gb, b);
1502 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1505 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1506 *icf++ = (bits & 1U<<31) | v;
1513 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1517 CLOSE_READER(re, gb);
1523 if (pulse_present) {
1525 for (i = 0; i < pulse->num_pulse; i++) {
1526 float co = coef_base[ pulse->pos[i] ];
1527 while (offsets[idx + 1] <= pulse->pos[i])
1529 if (band_type[idx] != NOISE_BT && sf[idx]) {
1530 float ico = -pulse->amp[i];
1533 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1535 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1542 static av_always_inline float flt16_round(float pf)
1544 union av_intfloat32 tmp;
1546 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1550 static av_always_inline float flt16_even(float pf)
1552 union av_intfloat32 tmp;
1554 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1558 static av_always_inline float flt16_trunc(float pf)
1560 union av_intfloat32 pun;
1562 pun.i &= 0xFFFF0000U;
1566 static av_always_inline void predict(PredictorState *ps, float *coef,
1569 const float a = 0.953125; // 61.0 / 64
1570 const float alpha = 0.90625; // 29.0 / 32
1574 float r0 = ps->r0, r1 = ps->r1;
1575 float cor0 = ps->cor0, cor1 = ps->cor1;
1576 float var0 = ps->var0, var1 = ps->var1;
1578 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1579 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1581 pv = flt16_round(k1 * r0 + k2 * r1);
1588 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1589 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1590 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1591 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1593 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1594 ps->r0 = flt16_trunc(a * e0);
1598 * Apply AAC-Main style frequency domain prediction.
1600 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1604 if (!sce->ics.predictor_initialized) {
1605 reset_all_predictors(sce->predictor_state);
1606 sce->ics.predictor_initialized = 1;
1609 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1610 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1611 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1612 predict(&sce->predictor_state[k], &sce->coeffs[k],
1613 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1616 if (sce->ics.predictor_reset_group)
1617 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1619 reset_all_predictors(sce->predictor_state);
1623 * Decode an individual_channel_stream payload; reference: table 4.44.
1625 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1626 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1628 * @return Returns error status. 0 - OK, !0 - error
1630 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1631 GetBitContext *gb, int common_window, int scale_flag)
1634 TemporalNoiseShaping *tns = &sce->tns;
1635 IndividualChannelStream *ics = &sce->ics;
1636 float *out = sce->coeffs;
1637 int global_gain, pulse_present = 0;
1639 /* This assignment is to silence a GCC warning about the variable being used
1640 * uninitialized when in fact it always is.
1642 pulse.num_pulse = 0;
1644 global_gain = get_bits(gb, 8);
1646 if (!common_window && !scale_flag) {
1647 if (decode_ics_info(ac, ics, gb) < 0)
1648 return AVERROR_INVALIDDATA;
1651 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1653 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1658 if ((pulse_present = get_bits1(gb))) {
1659 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1660 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1663 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1664 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1668 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1670 if (get_bits1(gb)) {
1671 av_log_missing_feature(ac->avctx, "SSR", 1);
1672 return AVERROR_PATCHWELCOME;
1676 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1679 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1680 apply_prediction(ac, sce);
1686 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1688 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1690 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1691 float *ch0 = cpe->ch[0].coeffs;
1692 float *ch1 = cpe->ch[1].coeffs;
1693 int g, i, group, idx = 0;
1694 const uint16_t *offsets = ics->swb_offset;
1695 for (g = 0; g < ics->num_window_groups; g++) {
1696 for (i = 0; i < ics->max_sfb; i++, idx++) {
1697 if (cpe->ms_mask[idx] &&
1698 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1699 for (group = 0; group < ics->group_len[g]; group++) {
1700 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1701 ch1 + group * 128 + offsets[i],
1702 offsets[i+1] - offsets[i]);
1706 ch0 += ics->group_len[g] * 128;
1707 ch1 += ics->group_len[g] * 128;
1712 * intensity stereo decoding; reference: 4.6.8.2.3
1714 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1715 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1716 * [3] reserved for scalable AAC
1718 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1720 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1721 SingleChannelElement *sce1 = &cpe->ch[1];
1722 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1723 const uint16_t *offsets = ics->swb_offset;
1724 int g, group, i, idx = 0;
1727 for (g = 0; g < ics->num_window_groups; g++) {
1728 for (i = 0; i < ics->max_sfb;) {
1729 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1730 const int bt_run_end = sce1->band_type_run_end[idx];
1731 for (; i < bt_run_end; i++, idx++) {
1732 c = -1 + 2 * (sce1->band_type[idx] - 14);
1734 c *= 1 - 2 * cpe->ms_mask[idx];
1735 scale = c * sce1->sf[idx];
1736 for (group = 0; group < ics->group_len[g]; group++)
1737 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1738 coef0 + group * 128 + offsets[i],
1740 offsets[i + 1] - offsets[i]);
1743 int bt_run_end = sce1->band_type_run_end[idx];
1744 idx += bt_run_end - i;
1748 coef0 += ics->group_len[g] * 128;
1749 coef1 += ics->group_len[g] * 128;
1754 * Decode a channel_pair_element; reference: table 4.4.
1756 * @return Returns error status. 0 - OK, !0 - error
1758 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1760 int i, ret, common_window, ms_present = 0;
1762 common_window = get_bits1(gb);
1763 if (common_window) {
1764 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1765 return AVERROR_INVALIDDATA;
1766 i = cpe->ch[1].ics.use_kb_window[0];
1767 cpe->ch[1].ics = cpe->ch[0].ics;
1768 cpe->ch[1].ics.use_kb_window[1] = i;
1769 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1770 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1771 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1772 ms_present = get_bits(gb, 2);
1773 if (ms_present == 3) {
1774 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1776 } else if (ms_present)
1777 decode_mid_side_stereo(cpe, gb, ms_present);
1779 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1781 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1784 if (common_window) {
1786 apply_mid_side_stereo(ac, cpe);
1787 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1788 apply_prediction(ac, &cpe->ch[0]);
1789 apply_prediction(ac, &cpe->ch[1]);
1793 apply_intensity_stereo(ac, cpe, ms_present);
1797 static const float cce_scale[] = {
1798 1.09050773266525765921, //2^(1/8)
1799 1.18920711500272106672, //2^(1/4)
1805 * Decode coupling_channel_element; reference: table 4.8.
1807 * @return Returns error status. 0 - OK, !0 - error
1809 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1815 SingleChannelElement *sce = &che->ch[0];
1816 ChannelCoupling *coup = &che->coup;
1818 coup->coupling_point = 2 * get_bits1(gb);
1819 coup->num_coupled = get_bits(gb, 3);
1820 for (c = 0; c <= coup->num_coupled; c++) {
1822 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1823 coup->id_select[c] = get_bits(gb, 4);
1824 if (coup->type[c] == TYPE_CPE) {
1825 coup->ch_select[c] = get_bits(gb, 2);
1826 if (coup->ch_select[c] == 3)
1829 coup->ch_select[c] = 2;
1831 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1833 sign = get_bits(gb, 1);
1834 scale = cce_scale[get_bits(gb, 2)];
1836 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1839 for (c = 0; c < num_gain; c++) {
1843 float gain_cache = 1.;
1845 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1846 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1847 gain_cache = powf(scale, -gain);
1849 if (coup->coupling_point == AFTER_IMDCT) {
1850 coup->gain[c][0] = gain_cache;
1852 for (g = 0; g < sce->ics.num_window_groups; g++) {
1853 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1854 if (sce->band_type[idx] != ZERO_BT) {
1856 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1864 gain_cache = powf(scale, -t) * s;
1867 coup->gain[c][idx] = gain_cache;
1877 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1879 * @return Returns number of bytes consumed.
1881 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1885 int num_excl_chan = 0;
1888 for (i = 0; i < 7; i++)
1889 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1890 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1892 return num_excl_chan / 7;
1896 * Decode dynamic range information; reference: table 4.52.
1898 * @return Returns number of bytes consumed.
1900 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1904 int drc_num_bands = 1;
1907 /* pce_tag_present? */
1908 if (get_bits1(gb)) {
1909 che_drc->pce_instance_tag = get_bits(gb, 4);
1910 skip_bits(gb, 4); // tag_reserved_bits
1914 /* excluded_chns_present? */
1915 if (get_bits1(gb)) {
1916 n += decode_drc_channel_exclusions(che_drc, gb);
1919 /* drc_bands_present? */
1920 if (get_bits1(gb)) {
1921 che_drc->band_incr = get_bits(gb, 4);
1922 che_drc->interpolation_scheme = get_bits(gb, 4);
1924 drc_num_bands += che_drc->band_incr;
1925 for (i = 0; i < drc_num_bands; i++) {
1926 che_drc->band_top[i] = get_bits(gb, 8);
1931 /* prog_ref_level_present? */
1932 if (get_bits1(gb)) {
1933 che_drc->prog_ref_level = get_bits(gb, 7);
1934 skip_bits1(gb); // prog_ref_level_reserved_bits
1938 for (i = 0; i < drc_num_bands; i++) {
1939 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1940 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1947 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1949 int i, major, minor;
1954 get_bits(gb, 13); len -= 13;
1956 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
1957 buf[i] = get_bits(gb, 8);
1960 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
1961 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
1963 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
1964 ac->avctx->internal->skip_samples = 1024;
1968 skip_bits_long(gb, len);
1974 * Decode extension data (incomplete); reference: table 4.51.
1976 * @param cnt length of TYPE_FIL syntactic element in bytes
1978 * @return Returns number of bytes consumed
1980 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1981 ChannelElement *che, enum RawDataBlockType elem_type)
1985 switch (get_bits(gb, 4)) { // extension type
1986 case EXT_SBR_DATA_CRC:
1990 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1992 } else if (!ac->oc[1].m4ac.sbr) {
1993 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1994 skip_bits_long(gb, 8 * cnt - 4);
1996 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
1997 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1998 skip_bits_long(gb, 8 * cnt - 4);
2000 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2001 ac->oc[1].m4ac.sbr = 1;
2002 ac->oc[1].m4ac.ps = 1;
2003 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2006 ac->oc[1].m4ac.sbr = 1;
2008 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2010 case EXT_DYNAMIC_RANGE:
2011 res = decode_dynamic_range(&ac->che_drc, gb);
2014 decode_fill(ac, gb, 8 * cnt - 4);
2017 case EXT_DATA_ELEMENT:
2019 skip_bits_long(gb, 8 * cnt - 4);
2026 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2028 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2029 * @param coef spectral coefficients
2031 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2032 IndividualChannelStream *ics, int decode)
2034 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2036 int bottom, top, order, start, end, size, inc;
2037 float lpc[TNS_MAX_ORDER];
2038 float tmp[TNS_MAX_ORDER];
2040 for (w = 0; w < ics->num_windows; w++) {
2041 bottom = ics->num_swb;
2042 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2044 bottom = FFMAX(0, top - tns->length[w][filt]);
2045 order = tns->order[w][filt];
2050 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2052 start = ics->swb_offset[FFMIN(bottom, mmm)];
2053 end = ics->swb_offset[FFMIN( top, mmm)];
2054 if ((size = end - start) <= 0)
2056 if (tns->direction[w][filt]) {
2066 for (m = 0; m < size; m++, start += inc)
2067 for (i = 1; i <= FFMIN(m, order); i++)
2068 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2071 for (m = 0; m < size; m++, start += inc) {
2072 tmp[0] = coef[start];
2073 for (i = 1; i <= FFMIN(m, order); i++)
2074 coef[start] += tmp[i] * lpc[i - 1];
2075 for (i = order; i > 0; i--)
2076 tmp[i] = tmp[i - 1];
2084 * Apply windowing and MDCT to obtain the spectral
2085 * coefficient from the predicted sample by LTP.
2087 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2088 float *in, IndividualChannelStream *ics)
2090 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2091 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2092 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2093 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2095 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2096 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2098 memset(in, 0, 448 * sizeof(float));
2099 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2101 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2102 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2104 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2105 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2107 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2111 * Apply the long term prediction
2113 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2115 const LongTermPrediction *ltp = &sce->ics.ltp;
2116 const uint16_t *offsets = sce->ics.swb_offset;
2119 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2120 float *predTime = sce->ret;
2121 float *predFreq = ac->buf_mdct;
2122 int16_t num_samples = 2048;
2124 if (ltp->lag < 1024)
2125 num_samples = ltp->lag + 1024;
2126 for (i = 0; i < num_samples; i++)
2127 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2128 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2130 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2132 if (sce->tns.present)
2133 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2135 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2137 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2138 sce->coeffs[i] += predFreq[i];
2143 * Update the LTP buffer for next frame
2145 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2147 IndividualChannelStream *ics = &sce->ics;
2148 float *saved = sce->saved;
2149 float *saved_ltp = sce->coeffs;
2150 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2151 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2154 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2155 memcpy(saved_ltp, saved, 512 * sizeof(float));
2156 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2157 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2158 for (i = 0; i < 64; i++)
2159 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2160 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2161 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2162 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2163 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2164 for (i = 0; i < 64; i++)
2165 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2166 } else { // LONG_STOP or ONLY_LONG
2167 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2168 for (i = 0; i < 512; i++)
2169 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2172 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2173 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2174 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2178 * Conduct IMDCT and windowing.
2180 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2182 IndividualChannelStream *ics = &sce->ics;
2183 float *in = sce->coeffs;
2184 float *out = sce->ret;
2185 float *saved = sce->saved;
2186 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2187 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2188 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2189 float *buf = ac->buf_mdct;
2190 float *temp = ac->temp;
2194 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2195 for (i = 0; i < 1024; i += 128)
2196 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2198 ac->mdct.imdct_half(&ac->mdct, buf, in);
2200 /* window overlapping
2201 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2202 * and long to short transitions are considered to be short to short
2203 * transitions. This leaves just two cases (long to long and short to short)
2204 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2206 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2207 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2208 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2210 memcpy( out, saved, 448 * sizeof(float));
2212 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2213 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2214 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2215 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2216 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2217 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2218 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2220 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2221 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2226 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2227 memcpy( saved, temp + 64, 64 * sizeof(float));
2228 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2229 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2230 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2231 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2232 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2233 memcpy( saved, buf + 512, 448 * sizeof(float));
2234 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2235 } else { // LONG_STOP or ONLY_LONG
2236 memcpy( saved, buf + 512, 512 * sizeof(float));
2241 * Apply dependent channel coupling (applied before IMDCT).
2243 * @param index index into coupling gain array
2245 static void apply_dependent_coupling(AACContext *ac,
2246 SingleChannelElement *target,
2247 ChannelElement *cce, int index)
2249 IndividualChannelStream *ics = &cce->ch[0].ics;
2250 const uint16_t *offsets = ics->swb_offset;
2251 float *dest = target->coeffs;
2252 const float *src = cce->ch[0].coeffs;
2253 int g, i, group, k, idx = 0;
2254 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2255 av_log(ac->avctx, AV_LOG_ERROR,
2256 "Dependent coupling is not supported together with LTP\n");
2259 for (g = 0; g < ics->num_window_groups; g++) {
2260 for (i = 0; i < ics->max_sfb; i++, idx++) {
2261 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2262 const float gain = cce->coup.gain[index][idx];
2263 for (group = 0; group < ics->group_len[g]; group++) {
2264 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2266 dest[group * 128 + k] += gain * src[group * 128 + k];
2271 dest += ics->group_len[g] * 128;
2272 src += ics->group_len[g] * 128;
2277 * Apply independent channel coupling (applied after IMDCT).
2279 * @param index index into coupling gain array
2281 static void apply_independent_coupling(AACContext *ac,
2282 SingleChannelElement *target,
2283 ChannelElement *cce, int index)
2286 const float gain = cce->coup.gain[index][0];
2287 const float *src = cce->ch[0].ret;
2288 float *dest = target->ret;
2289 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2291 for (i = 0; i < len; i++)
2292 dest[i] += gain * src[i];
2296 * channel coupling transformation interface
2298 * @param apply_coupling_method pointer to (in)dependent coupling function
2300 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2301 enum RawDataBlockType type, int elem_id,
2302 enum CouplingPoint coupling_point,
2303 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2307 for (i = 0; i < MAX_ELEM_ID; i++) {
2308 ChannelElement *cce = ac->che[TYPE_CCE][i];
2311 if (cce && cce->coup.coupling_point == coupling_point) {
2312 ChannelCoupling *coup = &cce->coup;
2314 for (c = 0; c <= coup->num_coupled; c++) {
2315 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2316 if (coup->ch_select[c] != 1) {
2317 apply_coupling_method(ac, &cc->ch[0], cce, index);
2318 if (coup->ch_select[c] != 0)
2321 if (coup->ch_select[c] != 2)
2322 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2324 index += 1 + (coup->ch_select[c] == 3);
2331 * Convert spectral data to float samples, applying all supported tools as appropriate.
2333 static void spectral_to_sample(AACContext *ac)
2336 for (type = 3; type >= 0; type--) {
2337 for (i = 0; i < MAX_ELEM_ID; i++) {
2338 ChannelElement *che = ac->che[type][i];
2340 if (type <= TYPE_CPE)
2341 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2342 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2343 if (che->ch[0].ics.predictor_present) {
2344 if (che->ch[0].ics.ltp.present)
2345 apply_ltp(ac, &che->ch[0]);
2346 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2347 apply_ltp(ac, &che->ch[1]);
2350 if (che->ch[0].tns.present)
2351 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2352 if (che->ch[1].tns.present)
2353 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2354 if (type <= TYPE_CPE)
2355 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2356 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2357 imdct_and_windowing(ac, &che->ch[0]);
2358 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2359 update_ltp(ac, &che->ch[0]);
2360 if (type == TYPE_CPE) {
2361 imdct_and_windowing(ac, &che->ch[1]);
2362 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2363 update_ltp(ac, &che->ch[1]);
2365 if (ac->oc[1].m4ac.sbr > 0) {
2366 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2369 if (type <= TYPE_CCE)
2370 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2376 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2379 AACADTSHeaderInfo hdr_info;
2380 uint8_t layout_map[MAX_ELEM_ID*4][3];
2381 int layout_map_tags;
2383 size = avpriv_aac_parse_header(gb, &hdr_info);
2385 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2386 // This is 2 for "VLB " audio in NSV files.
2387 // See samples/nsv/vlb_audio.
2388 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2389 ac->warned_num_aac_frames = 1;
2391 push_output_configuration(ac);
2392 if (hdr_info.chan_config) {
2393 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2394 if (set_default_channel_config(ac->avctx, layout_map,
2395 &layout_map_tags, hdr_info.chan_config))
2397 if (output_configure(ac, layout_map, layout_map_tags,
2398 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME)))
2401 ac->oc[1].m4ac.chan_config = 0;
2403 * dual mono frames in Japanese DTV can have chan_config 0
2404 * WITHOUT specifying PCE.
2405 * thus, set dual mono as default.
2407 if (ac->enable_jp_dmono && ac->oc[0].status == OC_NONE) {
2408 layout_map_tags = 2;
2409 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2410 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2411 layout_map[0][1] = 0;
2412 layout_map[1][1] = 1;
2413 if (output_configure(ac, layout_map, layout_map_tags,
2418 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2419 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2420 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2421 if (ac->oc[0].status != OC_LOCKED ||
2422 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2423 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2424 ac->oc[1].m4ac.sbr = -1;
2425 ac->oc[1].m4ac.ps = -1;
2427 if (!hdr_info.crc_absent)
2433 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2434 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2436 AACContext *ac = avctx->priv_data;
2437 ChannelElement *che = NULL, *che_prev = NULL;
2438 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2440 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2441 int is_dmono, sce_count = 0;
2444 if (show_bits(gb, 12) == 0xfff) {
2445 if (parse_adts_frame_header(ac, gb) < 0) {
2446 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2450 if (ac->oc[1].m4ac.sampling_index > 12) {
2451 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2457 ac->tags_mapped = 0;
2459 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2460 elem_id = get_bits(gb, 4);
2462 if (elem_type < TYPE_DSE) {
2463 if (!(che=get_che(ac, elem_type, elem_id))) {
2464 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2465 elem_type, elem_id);
2472 switch (elem_type) {
2475 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2481 err = decode_cpe(ac, gb, che);
2486 err = decode_cce(ac, gb, che);
2490 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2495 err = skip_data_stream_element(ac, gb);
2499 uint8_t layout_map[MAX_ELEM_ID*4][3];
2501 push_output_configuration(ac);
2502 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2508 av_log(avctx, AV_LOG_ERROR,
2509 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2510 pop_output_configuration(ac);
2512 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE);
2514 ac->oc[1].m4ac.chan_config = 0;
2522 elem_id += get_bits(gb, 8) - 1;
2523 if (get_bits_left(gb) < 8 * elem_id) {
2524 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2529 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2530 err = 0; /* FIXME */
2534 err = -1; /* should not happen, but keeps compiler happy */
2539 elem_type_prev = elem_type;
2544 if (get_bits_left(gb) < 3) {
2545 av_log(avctx, AV_LOG_ERROR, overread_err);
2551 spectral_to_sample(ac);
2553 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2554 samples <<= multiplier;
2556 /* for dual-mono audio (SCE + SCE) */
2557 is_dmono = ac->enable_jp_dmono && sce_count == 2 &&
2558 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2561 if (ac->dmono_mode == 0) {
2562 tmp = ac->output_data[1];
2563 ac->output_data[1] = ac->output_data[0];
2564 } else if (ac->dmono_mode == 1) {
2565 tmp = ac->output_data[0];
2566 ac->output_data[0] = ac->output_data[1];
2571 /* get output buffer */
2572 ac->frame.nb_samples = samples;
2573 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2574 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2579 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2580 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2581 (const float **)ac->output_data,
2582 samples, avctx->channels);
2584 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2585 (const float **)ac->output_data,
2586 samples, avctx->channels);
2588 *(AVFrame *)data = ac->frame;
2590 *got_frame_ptr = !!samples;
2593 if (ac->dmono_mode == 0)
2594 ac->output_data[1] = tmp;
2595 else if (ac->dmono_mode == 1)
2596 ac->output_data[0] = tmp;
2599 if (ac->oc[1].status && audio_found) {
2600 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2601 avctx->frame_size = samples;
2602 ac->oc[1].status = OC_LOCKED;
2607 uint32_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2608 if (side && side_size>=4)
2609 AV_WL32(side, 2*AV_RL32(side));
2613 pop_output_configuration(ac);
2617 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2618 int *got_frame_ptr, AVPacket *avpkt)
2620 AACContext *ac = avctx->priv_data;
2621 const uint8_t *buf = avpkt->data;
2622 int buf_size = avpkt->size;
2627 int new_extradata_size;
2628 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2629 AV_PKT_DATA_NEW_EXTRADATA,
2630 &new_extradata_size);
2631 int jp_dualmono_size;
2632 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2633 AV_PKT_DATA_JP_DUALMONO,
2636 if (new_extradata && 0) {
2637 av_free(avctx->extradata);
2638 avctx->extradata = av_mallocz(new_extradata_size +
2639 FF_INPUT_BUFFER_PADDING_SIZE);
2640 if (!avctx->extradata)
2641 return AVERROR(ENOMEM);
2642 avctx->extradata_size = new_extradata_size;
2643 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2644 push_output_configuration(ac);
2645 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2647 avctx->extradata_size*8, 1) < 0) {
2648 pop_output_configuration(ac);
2649 return AVERROR_INVALIDDATA;
2653 ac->enable_jp_dmono = !!jp_dualmono;
2655 if (jp_dualmono && jp_dualmono_size > 0)
2656 ac->dmono_mode = *jp_dualmono;
2658 init_get_bits(&gb, buf, buf_size * 8);
2660 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2663 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2664 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2665 if (buf[buf_offset])
2668 return buf_size > buf_offset ? buf_consumed : buf_size;
2671 static av_cold int aac_decode_close(AVCodecContext *avctx)
2673 AACContext *ac = avctx->priv_data;
2676 for (i = 0; i < MAX_ELEM_ID; i++) {
2677 for (type = 0; type < 4; type++) {
2678 if (ac->che[type][i])
2679 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2680 av_freep(&ac->che[type][i]);
2684 ff_mdct_end(&ac->mdct);
2685 ff_mdct_end(&ac->mdct_small);
2686 ff_mdct_end(&ac->mdct_ltp);
2691 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2693 struct LATMContext {
2694 AACContext aac_ctx; ///< containing AACContext
2695 int initialized; ///< initialized after a valid extradata was seen
2698 int audio_mux_version_A; ///< LATM syntax version
2699 int frame_length_type; ///< 0/1 variable/fixed frame length
2700 int frame_length; ///< frame length for fixed frame length
2703 static inline uint32_t latm_get_value(GetBitContext *b)
2705 int length = get_bits(b, 2);
2707 return get_bits_long(b, (length+1)*8);
2710 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2711 GetBitContext *gb, int asclen)
2713 AACContext *ac = &latmctx->aac_ctx;
2714 AVCodecContext *avctx = ac->avctx;
2715 MPEG4AudioConfig m4ac = { 0 };
2716 int config_start_bit = get_bits_count(gb);
2717 int sync_extension = 0;
2718 int bits_consumed, esize;
2722 asclen = FFMIN(asclen, get_bits_left(gb));
2724 asclen = get_bits_left(gb);
2726 if (config_start_bit % 8) {
2727 av_log_missing_feature(latmctx->aac_ctx.avctx,
2728 "Non-byte-aligned audio-specific config", 1);
2729 return AVERROR_PATCHWELCOME;
2732 return AVERROR_INVALIDDATA;
2733 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2734 gb->buffer + (config_start_bit / 8),
2735 asclen, sync_extension);
2737 if (bits_consumed < 0)
2738 return AVERROR_INVALIDDATA;
2740 if (!latmctx->initialized ||
2741 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2742 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2744 if(latmctx->initialized) {
2745 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2747 av_log(avctx, AV_LOG_INFO, "initializing latmctx\n");
2749 latmctx->initialized = 0;
2751 esize = (bits_consumed+7) / 8;
2753 if (avctx->extradata_size < esize) {
2754 av_free(avctx->extradata);
2755 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2756 if (!avctx->extradata)
2757 return AVERROR(ENOMEM);
2760 avctx->extradata_size = esize;
2761 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2762 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2764 skip_bits_long(gb, bits_consumed);
2766 return bits_consumed;
2769 static int read_stream_mux_config(struct LATMContext *latmctx,
2772 int ret, audio_mux_version = get_bits(gb, 1);
2774 latmctx->audio_mux_version_A = 0;
2775 if (audio_mux_version)
2776 latmctx->audio_mux_version_A = get_bits(gb, 1);
2778 if (!latmctx->audio_mux_version_A) {
2780 if (audio_mux_version)
2781 latm_get_value(gb); // taraFullness
2783 skip_bits(gb, 1); // allStreamSameTimeFraming
2784 skip_bits(gb, 6); // numSubFrames
2786 if (get_bits(gb, 4)) { // numPrograms
2787 av_log_missing_feature(latmctx->aac_ctx.avctx,
2788 "Multiple programs", 1);
2789 return AVERROR_PATCHWELCOME;
2792 // for each program (which there is only one in DVB)
2794 // for each layer (which there is only one in DVB)
2795 if (get_bits(gb, 3)) { // numLayer
2796 av_log_missing_feature(latmctx->aac_ctx.avctx,
2797 "Multiple layers", 1);
2798 return AVERROR_PATCHWELCOME;
2801 // for all but first stream: use_same_config = get_bits(gb, 1);
2802 if (!audio_mux_version) {
2803 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2806 int ascLen = latm_get_value(gb);
2807 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2810 skip_bits_long(gb, ascLen);
2813 latmctx->frame_length_type = get_bits(gb, 3);
2814 switch (latmctx->frame_length_type) {
2816 skip_bits(gb, 8); // latmBufferFullness
2819 latmctx->frame_length = get_bits(gb, 9);
2824 skip_bits(gb, 6); // CELP frame length table index
2828 skip_bits(gb, 1); // HVXC frame length table index
2832 if (get_bits(gb, 1)) { // other data
2833 if (audio_mux_version) {
2834 latm_get_value(gb); // other_data_bits
2838 esc = get_bits(gb, 1);
2844 if (get_bits(gb, 1)) // crc present
2845 skip_bits(gb, 8); // config_crc
2851 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2855 if (ctx->frame_length_type == 0) {
2856 int mux_slot_length = 0;
2858 tmp = get_bits(gb, 8);
2859 mux_slot_length += tmp;
2860 } while (tmp == 255);
2861 return mux_slot_length;
2862 } else if (ctx->frame_length_type == 1) {
2863 return ctx->frame_length;
2864 } else if (ctx->frame_length_type == 3 ||
2865 ctx->frame_length_type == 5 ||
2866 ctx->frame_length_type == 7) {
2867 skip_bits(gb, 2); // mux_slot_length_coded
2872 static int read_audio_mux_element(struct LATMContext *latmctx,
2876 uint8_t use_same_mux = get_bits(gb, 1);
2877 if (!use_same_mux) {
2878 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2880 } else if (!latmctx->aac_ctx.avctx->extradata) {
2881 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2882 "no decoder config found\n");
2883 return AVERROR(EAGAIN);
2885 if (latmctx->audio_mux_version_A == 0) {
2886 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2887 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2888 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2889 return AVERROR_INVALIDDATA;
2890 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2891 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2892 "frame length mismatch %d << %d\n",
2893 mux_slot_length_bytes * 8, get_bits_left(gb));
2894 return AVERROR_INVALIDDATA;
2901 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2902 int *got_frame_ptr, AVPacket *avpkt)
2904 struct LATMContext *latmctx = avctx->priv_data;
2908 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2910 // check for LOAS sync word
2911 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2912 return AVERROR_INVALIDDATA;
2914 muxlength = get_bits(&gb, 13) + 3;
2915 // not enough data, the parser should have sorted this out
2916 if (muxlength > avpkt->size)
2917 return AVERROR_INVALIDDATA;
2919 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2922 if (!latmctx->initialized) {
2923 if (!avctx->extradata) {
2927 push_output_configuration(&latmctx->aac_ctx);
2928 if ((err = decode_audio_specific_config(
2929 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2930 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2931 pop_output_configuration(&latmctx->aac_ctx);
2934 latmctx->initialized = 1;
2938 if (show_bits(&gb, 12) == 0xfff) {
2939 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2940 "ADTS header detected, probably as result of configuration "
2942 return AVERROR_INVALIDDATA;
2945 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2951 static av_cold int latm_decode_init(AVCodecContext *avctx)
2953 struct LATMContext *latmctx = avctx->priv_data;
2954 int ret = aac_decode_init(avctx);
2956 if (avctx->extradata_size > 0)
2957 latmctx->initialized = !ret;
2963 AVCodec ff_aac_decoder = {
2965 .type = AVMEDIA_TYPE_AUDIO,
2966 .id = AV_CODEC_ID_AAC,
2967 .priv_data_size = sizeof(AACContext),
2968 .init = aac_decode_init,
2969 .close = aac_decode_close,
2970 .decode = aac_decode_frame,
2971 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
2972 .sample_fmts = (const enum AVSampleFormat[]) {
2973 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2975 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2976 .channel_layouts = aac_channel_layout,
2981 Note: This decoder filter is intended to decode LATM streams transferred
2982 in MPEG transport streams which only contain one program.
2983 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2985 AVCodec ff_aac_latm_decoder = {
2987 .type = AVMEDIA_TYPE_AUDIO,
2988 .id = AV_CODEC_ID_AAC_LATM,
2989 .priv_data_size = sizeof(struct LATMContext),
2990 .init = latm_decode_init,
2991 .close = aac_decode_close,
2992 .decode = latm_decode_frame,
2993 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
2994 .sample_fmts = (const enum AVSampleFormat[]) {
2995 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2997 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2998 .channel_layouts = aac_channel_layout,