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 && 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 int channel_config, 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,
484 2, OC_TRIAL_FRAME) < 0)
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,
502 1, OC_TRIAL_FRAME) < 0)
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,
696 channel_config, OC_GLOBAL_HDR)))
699 if (extension_flag) {
700 switch (m4ac->object_type) {
702 skip_bits(gb, 5); // numOfSubFrame
703 skip_bits(gb, 11); // layer_length
707 case AOT_ER_AAC_SCALABLE:
709 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
710 * aacScalefactorDataResilienceFlag
711 * aacSpectralDataResilienceFlag
715 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
721 * Decode audio specific configuration; reference: table 1.13.
723 * @param ac pointer to AACContext, may be null
724 * @param avctx pointer to AVCCodecContext, used for logging
725 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
726 * @param data pointer to buffer holding an audio specific config
727 * @param bit_size size of audio specific config or data in bits
728 * @param sync_extension look for an appended sync extension
730 * @return Returns error status or number of consumed bits. <0 - error
732 static int decode_audio_specific_config(AACContext *ac,
733 AVCodecContext *avctx,
734 MPEG4AudioConfig *m4ac,
735 const uint8_t *data, int bit_size,
741 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
742 for (i = 0; i < bit_size >> 3; i++)
743 av_dlog(avctx, "%02x ", data[i]);
744 av_dlog(avctx, "\n");
746 init_get_bits(&gb, data, bit_size);
748 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
750 if (m4ac->sampling_index > 12) {
751 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
755 skip_bits_long(&gb, i);
757 switch (m4ac->object_type) {
761 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
765 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
766 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
770 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
771 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
772 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
774 return get_bits_count(&gb);
778 * linear congruential pseudorandom number generator
780 * @param previous_val pointer to the current state of the generator
782 * @return Returns a 32-bit pseudorandom integer
784 static av_always_inline int lcg_random(unsigned previous_val)
786 return previous_val * 1664525 + 1013904223;
789 static av_always_inline void reset_predict_state(PredictorState *ps)
799 static void reset_all_predictors(PredictorState *ps)
802 for (i = 0; i < MAX_PREDICTORS; i++)
803 reset_predict_state(&ps[i]);
806 static int sample_rate_idx (int rate)
808 if (92017 <= rate) return 0;
809 else if (75132 <= rate) return 1;
810 else if (55426 <= rate) return 2;
811 else if (46009 <= rate) return 3;
812 else if (37566 <= rate) return 4;
813 else if (27713 <= rate) return 5;
814 else if (23004 <= rate) return 6;
815 else if (18783 <= rate) return 7;
816 else if (13856 <= rate) return 8;
817 else if (11502 <= rate) return 9;
818 else if (9391 <= rate) return 10;
822 static void reset_predictor_group(PredictorState *ps, int group_num)
825 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
826 reset_predict_state(&ps[i]);
829 #define AAC_INIT_VLC_STATIC(num, size) \
830 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
831 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
832 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
835 static av_cold int aac_decode_init(AVCodecContext *avctx)
837 AACContext *ac = avctx->priv_data;
838 float output_scale_factor;
841 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
843 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
844 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
845 output_scale_factor = 1.0 / 32768.0;
847 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
848 output_scale_factor = 1.0;
851 if (avctx->extradata_size > 0) {
852 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
854 avctx->extradata_size*8, 1) < 0)
858 uint8_t layout_map[MAX_ELEM_ID*4][3];
861 sr = sample_rate_idx(avctx->sample_rate);
862 ac->oc[1].m4ac.sampling_index = sr;
863 ac->oc[1].m4ac.channels = avctx->channels;
864 ac->oc[1].m4ac.sbr = -1;
865 ac->oc[1].m4ac.ps = -1;
867 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
868 if (ff_mpeg4audio_channels[i] == avctx->channels)
870 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
873 ac->oc[1].m4ac.chan_config = i;
875 if (ac->oc[1].m4ac.chan_config) {
876 int ret = set_default_channel_config(avctx, layout_map,
877 &layout_map_tags, ac->oc[1].m4ac.chan_config);
879 output_configure(ac, layout_map, layout_map_tags,
880 ac->oc[1].m4ac.chan_config, OC_GLOBAL_HDR);
881 else if (avctx->err_recognition & AV_EF_EXPLODE)
882 return AVERROR_INVALIDDATA;
886 AAC_INIT_VLC_STATIC( 0, 304);
887 AAC_INIT_VLC_STATIC( 1, 270);
888 AAC_INIT_VLC_STATIC( 2, 550);
889 AAC_INIT_VLC_STATIC( 3, 300);
890 AAC_INIT_VLC_STATIC( 4, 328);
891 AAC_INIT_VLC_STATIC( 5, 294);
892 AAC_INIT_VLC_STATIC( 6, 306);
893 AAC_INIT_VLC_STATIC( 7, 268);
894 AAC_INIT_VLC_STATIC( 8, 510);
895 AAC_INIT_VLC_STATIC( 9, 366);
896 AAC_INIT_VLC_STATIC(10, 462);
900 ff_dsputil_init(&ac->dsp, avctx);
901 ff_fmt_convert_init(&ac->fmt_conv, avctx);
902 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
904 ac->random_state = 0x1f2e3d4c;
908 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
909 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
910 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
913 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
914 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
915 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
916 // window initialization
917 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
918 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
919 ff_init_ff_sine_windows(10);
920 ff_init_ff_sine_windows( 7);
924 avcodec_get_frame_defaults(&ac->frame);
925 avctx->coded_frame = &ac->frame;
931 * Skip data_stream_element; reference: table 4.10.
933 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
935 int byte_align = get_bits1(gb);
936 int count = get_bits(gb, 8);
938 count += get_bits(gb, 8);
942 if (get_bits_left(gb) < 8 * count) {
943 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
946 skip_bits_long(gb, 8 * count);
950 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
955 ics->predictor_reset_group = get_bits(gb, 5);
956 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
957 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
961 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
962 ics->prediction_used[sfb] = get_bits1(gb);
968 * Decode Long Term Prediction data; reference: table 4.xx.
970 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
971 GetBitContext *gb, uint8_t max_sfb)
975 ltp->lag = get_bits(gb, 11);
976 ltp->coef = ltp_coef[get_bits(gb, 3)];
977 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
978 ltp->used[sfb] = get_bits1(gb);
982 * Decode Individual Channel Stream info; reference: table 4.6.
984 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
988 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
989 return AVERROR_INVALIDDATA;
991 ics->window_sequence[1] = ics->window_sequence[0];
992 ics->window_sequence[0] = get_bits(gb, 2);
993 ics->use_kb_window[1] = ics->use_kb_window[0];
994 ics->use_kb_window[0] = get_bits1(gb);
995 ics->num_window_groups = 1;
996 ics->group_len[0] = 1;
997 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
999 ics->max_sfb = get_bits(gb, 4);
1000 for (i = 0; i < 7; i++) {
1001 if (get_bits1(gb)) {
1002 ics->group_len[ics->num_window_groups - 1]++;
1004 ics->num_window_groups++;
1005 ics->group_len[ics->num_window_groups - 1] = 1;
1008 ics->num_windows = 8;
1009 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1010 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1011 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1012 ics->predictor_present = 0;
1014 ics->max_sfb = get_bits(gb, 6);
1015 ics->num_windows = 1;
1016 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1017 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1018 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1019 ics->predictor_present = get_bits1(gb);
1020 ics->predictor_reset_group = 0;
1021 if (ics->predictor_present) {
1022 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1023 if (decode_prediction(ac, ics, gb)) {
1026 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1027 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1030 if ((ics->ltp.present = get_bits(gb, 1)))
1031 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
1036 if (ics->max_sfb > ics->num_swb) {
1037 av_log(ac->avctx, AV_LOG_ERROR,
1038 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1039 ics->max_sfb, ics->num_swb);
1046 return AVERROR_INVALIDDATA;
1050 * Decode band types (section_data payload); reference: table 4.46.
1052 * @param band_type array of the used band type
1053 * @param band_type_run_end array of the last scalefactor band of a band type run
1055 * @return Returns error status. 0 - OK, !0 - error
1057 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1058 int band_type_run_end[120], GetBitContext *gb,
1059 IndividualChannelStream *ics)
1062 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1063 for (g = 0; g < ics->num_window_groups; g++) {
1065 while (k < ics->max_sfb) {
1066 uint8_t sect_end = k;
1068 int sect_band_type = get_bits(gb, 4);
1069 if (sect_band_type == 12) {
1070 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1074 sect_len_incr = get_bits(gb, bits);
1075 sect_end += sect_len_incr;
1076 if (get_bits_left(gb) < 0) {
1077 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1080 if (sect_end > ics->max_sfb) {
1081 av_log(ac->avctx, AV_LOG_ERROR,
1082 "Number of bands (%d) exceeds limit (%d).\n",
1083 sect_end, ics->max_sfb);
1086 } while (sect_len_incr == (1 << bits) - 1);
1087 for (; k < sect_end; k++) {
1088 band_type [idx] = sect_band_type;
1089 band_type_run_end[idx++] = sect_end;
1097 * Decode scalefactors; reference: table 4.47.
1099 * @param global_gain first scalefactor value as scalefactors are differentially coded
1100 * @param band_type array of the used band type
1101 * @param band_type_run_end array of the last scalefactor band of a band type run
1102 * @param sf array of scalefactors or intensity stereo positions
1104 * @return Returns error status. 0 - OK, !0 - error
1106 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1107 unsigned int global_gain,
1108 IndividualChannelStream *ics,
1109 enum BandType band_type[120],
1110 int band_type_run_end[120])
1113 int offset[3] = { global_gain, global_gain - 90, 0 };
1116 for (g = 0; g < ics->num_window_groups; g++) {
1117 for (i = 0; i < ics->max_sfb;) {
1118 int run_end = band_type_run_end[idx];
1119 if (band_type[idx] == ZERO_BT) {
1120 for (; i < run_end; i++, idx++)
1122 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1123 for (; i < run_end; i++, idx++) {
1124 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1125 clipped_offset = av_clip(offset[2], -155, 100);
1126 if (offset[2] != clipped_offset) {
1127 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1128 "position clipped (%d -> %d).\nIf you heard an "
1129 "audible artifact, there may be a bug in the "
1130 "decoder. ", offset[2], clipped_offset);
1132 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1134 } else if (band_type[idx] == NOISE_BT) {
1135 for (; i < run_end; i++, idx++) {
1136 if (noise_flag-- > 0)
1137 offset[1] += get_bits(gb, 9) - 256;
1139 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1140 clipped_offset = av_clip(offset[1], -100, 155);
1141 if (offset[1] != clipped_offset) {
1142 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1143 "(%d -> %d).\nIf you heard an audible "
1144 "artifact, there may be a bug in the decoder. ",
1145 offset[1], clipped_offset);
1147 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1150 for (; i < run_end; i++, idx++) {
1151 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1152 if (offset[0] > 255U) {
1153 av_log(ac->avctx, AV_LOG_ERROR,
1154 "Scalefactor (%d) out of range.\n", offset[0]);
1157 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1166 * Decode pulse data; reference: table 4.7.
1168 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1169 const uint16_t *swb_offset, int num_swb)
1172 pulse->num_pulse = get_bits(gb, 2) + 1;
1173 pulse_swb = get_bits(gb, 6);
1174 if (pulse_swb >= num_swb)
1176 pulse->pos[0] = swb_offset[pulse_swb];
1177 pulse->pos[0] += get_bits(gb, 5);
1178 if (pulse->pos[0] > 1023)
1180 pulse->amp[0] = get_bits(gb, 4);
1181 for (i = 1; i < pulse->num_pulse; i++) {
1182 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1183 if (pulse->pos[i] > 1023)
1185 pulse->amp[i] = get_bits(gb, 4);
1191 * Decode Temporal Noise Shaping data; reference: table 4.48.
1193 * @return Returns error status. 0 - OK, !0 - error
1195 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1196 GetBitContext *gb, const IndividualChannelStream *ics)
1198 int w, filt, i, coef_len, coef_res, coef_compress;
1199 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1200 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1201 for (w = 0; w < ics->num_windows; w++) {
1202 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1203 coef_res = get_bits1(gb);
1205 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1207 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1209 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1210 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1211 tns->order[w][filt], tns_max_order);
1212 tns->order[w][filt] = 0;
1215 if (tns->order[w][filt]) {
1216 tns->direction[w][filt] = get_bits1(gb);
1217 coef_compress = get_bits1(gb);
1218 coef_len = coef_res + 3 - coef_compress;
1219 tmp2_idx = 2 * coef_compress + coef_res;
1221 for (i = 0; i < tns->order[w][filt]; i++)
1222 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1231 * Decode Mid/Side data; reference: table 4.54.
1233 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1234 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1235 * [3] reserved for scalable AAC
1237 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1241 if (ms_present == 1) {
1242 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1243 cpe->ms_mask[idx] = get_bits1(gb);
1244 } else if (ms_present == 2) {
1245 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1250 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1254 *dst++ = v[idx & 15] * s;
1255 *dst++ = v[idx>>4 & 15] * s;
1261 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1265 *dst++ = v[idx & 3] * s;
1266 *dst++ = v[idx>>2 & 3] * s;
1267 *dst++ = v[idx>>4 & 3] * s;
1268 *dst++ = v[idx>>6 & 3] * s;
1274 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1275 unsigned sign, const float *scale)
1277 union av_intfloat32 s0, s1;
1279 s0.f = s1.f = *scale;
1280 s0.i ^= sign >> 1 << 31;
1283 *dst++ = v[idx & 15] * s0.f;
1284 *dst++ = v[idx>>4 & 15] * s1.f;
1291 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1292 unsigned sign, const float *scale)
1294 unsigned nz = idx >> 12;
1295 union av_intfloat32 s = { .f = *scale };
1296 union av_intfloat32 t;
1298 t.i = s.i ^ (sign & 1U<<31);
1299 *dst++ = v[idx & 3] * t.f;
1301 sign <<= nz & 1; nz >>= 1;
1302 t.i = s.i ^ (sign & 1U<<31);
1303 *dst++ = v[idx>>2 & 3] * t.f;
1305 sign <<= nz & 1; nz >>= 1;
1306 t.i = s.i ^ (sign & 1U<<31);
1307 *dst++ = v[idx>>4 & 3] * t.f;
1310 t.i = s.i ^ (sign & 1U<<31);
1311 *dst++ = v[idx>>6 & 3] * t.f;
1318 * Decode spectral data; reference: table 4.50.
1319 * Dequantize and scale spectral data; reference: 4.6.3.3.
1321 * @param coef array of dequantized, scaled spectral data
1322 * @param sf array of scalefactors or intensity stereo positions
1323 * @param pulse_present set if pulses are present
1324 * @param pulse pointer to pulse data struct
1325 * @param band_type array of the used band type
1327 * @return Returns error status. 0 - OK, !0 - error
1329 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1330 GetBitContext *gb, const float sf[120],
1331 int pulse_present, const Pulse *pulse,
1332 const IndividualChannelStream *ics,
1333 enum BandType band_type[120])
1335 int i, k, g, idx = 0;
1336 const int c = 1024 / ics->num_windows;
1337 const uint16_t *offsets = ics->swb_offset;
1338 float *coef_base = coef;
1340 for (g = 0; g < ics->num_windows; g++)
1341 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1343 for (g = 0; g < ics->num_window_groups; g++) {
1344 unsigned g_len = ics->group_len[g];
1346 for (i = 0; i < ics->max_sfb; i++, idx++) {
1347 const unsigned cbt_m1 = band_type[idx] - 1;
1348 float *cfo = coef + offsets[i];
1349 int off_len = offsets[i + 1] - offsets[i];
1352 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1353 for (group = 0; group < g_len; group++, cfo+=128) {
1354 memset(cfo, 0, off_len * sizeof(float));
1356 } else if (cbt_m1 == NOISE_BT - 1) {
1357 for (group = 0; group < g_len; group++, cfo+=128) {
1361 for (k = 0; k < off_len; k++) {
1362 ac->random_state = lcg_random(ac->random_state);
1363 cfo[k] = ac->random_state;
1366 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1367 scale = sf[idx] / sqrtf(band_energy);
1368 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1371 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1372 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1373 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1374 OPEN_READER(re, gb);
1376 switch (cbt_m1 >> 1) {
1378 for (group = 0; group < g_len; group++, cfo+=128) {
1386 UPDATE_CACHE(re, gb);
1387 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1388 cb_idx = cb_vector_idx[code];
1389 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1395 for (group = 0; group < g_len; group++, cfo+=128) {
1405 UPDATE_CACHE(re, gb);
1406 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1407 cb_idx = cb_vector_idx[code];
1408 nnz = cb_idx >> 8 & 15;
1409 bits = nnz ? GET_CACHE(re, gb) : 0;
1410 LAST_SKIP_BITS(re, gb, nnz);
1411 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1417 for (group = 0; group < g_len; group++, cfo+=128) {
1425 UPDATE_CACHE(re, gb);
1426 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1427 cb_idx = cb_vector_idx[code];
1428 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1435 for (group = 0; group < g_len; group++, cfo+=128) {
1445 UPDATE_CACHE(re, gb);
1446 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1447 cb_idx = cb_vector_idx[code];
1448 nnz = cb_idx >> 8 & 15;
1449 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1450 LAST_SKIP_BITS(re, gb, nnz);
1451 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1457 for (group = 0; group < g_len; group++, cfo+=128) {
1459 uint32_t *icf = (uint32_t *) cf;
1469 UPDATE_CACHE(re, gb);
1470 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1478 cb_idx = cb_vector_idx[code];
1481 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1482 LAST_SKIP_BITS(re, gb, nnz);
1484 for (j = 0; j < 2; j++) {
1488 /* The total length of escape_sequence must be < 22 bits according
1489 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1490 UPDATE_CACHE(re, gb);
1491 b = GET_CACHE(re, gb);
1492 b = 31 - av_log2(~b);
1495 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1499 SKIP_BITS(re, gb, b + 1);
1501 n = (1 << b) + SHOW_UBITS(re, gb, b);
1502 LAST_SKIP_BITS(re, gb, b);
1503 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1506 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1507 *icf++ = (bits & 1U<<31) | v;
1514 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1518 CLOSE_READER(re, gb);
1524 if (pulse_present) {
1526 for (i = 0; i < pulse->num_pulse; i++) {
1527 float co = coef_base[ pulse->pos[i] ];
1528 while (offsets[idx + 1] <= pulse->pos[i])
1530 if (band_type[idx] != NOISE_BT && sf[idx]) {
1531 float ico = -pulse->amp[i];
1534 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1536 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1543 static av_always_inline float flt16_round(float pf)
1545 union av_intfloat32 tmp;
1547 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1551 static av_always_inline float flt16_even(float pf)
1553 union av_intfloat32 tmp;
1555 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1559 static av_always_inline float flt16_trunc(float pf)
1561 union av_intfloat32 pun;
1563 pun.i &= 0xFFFF0000U;
1567 static av_always_inline void predict(PredictorState *ps, float *coef,
1570 const float a = 0.953125; // 61.0 / 64
1571 const float alpha = 0.90625; // 29.0 / 32
1575 float r0 = ps->r0, r1 = ps->r1;
1576 float cor0 = ps->cor0, cor1 = ps->cor1;
1577 float var0 = ps->var0, var1 = ps->var1;
1579 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1580 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1582 pv = flt16_round(k1 * r0 + k2 * r1);
1589 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1590 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1591 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1592 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1594 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1595 ps->r0 = flt16_trunc(a * e0);
1599 * Apply AAC-Main style frequency domain prediction.
1601 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1605 if (!sce->ics.predictor_initialized) {
1606 reset_all_predictors(sce->predictor_state);
1607 sce->ics.predictor_initialized = 1;
1610 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1611 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1612 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1613 predict(&sce->predictor_state[k], &sce->coeffs[k],
1614 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1617 if (sce->ics.predictor_reset_group)
1618 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1620 reset_all_predictors(sce->predictor_state);
1624 * Decode an individual_channel_stream payload; reference: table 4.44.
1626 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1627 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1629 * @return Returns error status. 0 - OK, !0 - error
1631 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1632 GetBitContext *gb, int common_window, int scale_flag)
1635 TemporalNoiseShaping *tns = &sce->tns;
1636 IndividualChannelStream *ics = &sce->ics;
1637 float *out = sce->coeffs;
1638 int global_gain, pulse_present = 0;
1640 /* This assignment is to silence a GCC warning about the variable being used
1641 * uninitialized when in fact it always is.
1643 pulse.num_pulse = 0;
1645 global_gain = get_bits(gb, 8);
1647 if (!common_window && !scale_flag) {
1648 if (decode_ics_info(ac, ics, gb) < 0)
1649 return AVERROR_INVALIDDATA;
1652 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1654 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1659 if ((pulse_present = get_bits1(gb))) {
1660 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1661 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1664 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1665 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1669 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1671 if (get_bits1(gb)) {
1672 av_log_missing_feature(ac->avctx, "SSR", 1);
1673 return AVERROR_PATCHWELCOME;
1677 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1680 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1681 apply_prediction(ac, sce);
1687 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1689 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1691 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1692 float *ch0 = cpe->ch[0].coeffs;
1693 float *ch1 = cpe->ch[1].coeffs;
1694 int g, i, group, idx = 0;
1695 const uint16_t *offsets = ics->swb_offset;
1696 for (g = 0; g < ics->num_window_groups; g++) {
1697 for (i = 0; i < ics->max_sfb; i++, idx++) {
1698 if (cpe->ms_mask[idx] &&
1699 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1700 for (group = 0; group < ics->group_len[g]; group++) {
1701 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1702 ch1 + group * 128 + offsets[i],
1703 offsets[i+1] - offsets[i]);
1707 ch0 += ics->group_len[g] * 128;
1708 ch1 += ics->group_len[g] * 128;
1713 * intensity stereo decoding; reference: 4.6.8.2.3
1715 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1716 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1717 * [3] reserved for scalable AAC
1719 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1721 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1722 SingleChannelElement *sce1 = &cpe->ch[1];
1723 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1724 const uint16_t *offsets = ics->swb_offset;
1725 int g, group, i, idx = 0;
1728 for (g = 0; g < ics->num_window_groups; g++) {
1729 for (i = 0; i < ics->max_sfb;) {
1730 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1731 const int bt_run_end = sce1->band_type_run_end[idx];
1732 for (; i < bt_run_end; i++, idx++) {
1733 c = -1 + 2 * (sce1->band_type[idx] - 14);
1735 c *= 1 - 2 * cpe->ms_mask[idx];
1736 scale = c * sce1->sf[idx];
1737 for (group = 0; group < ics->group_len[g]; group++)
1738 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1739 coef0 + group * 128 + offsets[i],
1741 offsets[i + 1] - offsets[i]);
1744 int bt_run_end = sce1->band_type_run_end[idx];
1745 idx += bt_run_end - i;
1749 coef0 += ics->group_len[g] * 128;
1750 coef1 += ics->group_len[g] * 128;
1755 * Decode a channel_pair_element; reference: table 4.4.
1757 * @return Returns error status. 0 - OK, !0 - error
1759 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1761 int i, ret, common_window, ms_present = 0;
1763 common_window = get_bits1(gb);
1764 if (common_window) {
1765 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1766 return AVERROR_INVALIDDATA;
1767 i = cpe->ch[1].ics.use_kb_window[0];
1768 cpe->ch[1].ics = cpe->ch[0].ics;
1769 cpe->ch[1].ics.use_kb_window[1] = i;
1770 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1771 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1772 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1773 ms_present = get_bits(gb, 2);
1774 if (ms_present == 3) {
1775 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1777 } else if (ms_present)
1778 decode_mid_side_stereo(cpe, gb, ms_present);
1780 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1782 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1785 if (common_window) {
1787 apply_mid_side_stereo(ac, cpe);
1788 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1789 apply_prediction(ac, &cpe->ch[0]);
1790 apply_prediction(ac, &cpe->ch[1]);
1794 apply_intensity_stereo(ac, cpe, ms_present);
1798 static const float cce_scale[] = {
1799 1.09050773266525765921, //2^(1/8)
1800 1.18920711500272106672, //2^(1/4)
1806 * Decode coupling_channel_element; reference: table 4.8.
1808 * @return Returns error status. 0 - OK, !0 - error
1810 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1816 SingleChannelElement *sce = &che->ch[0];
1817 ChannelCoupling *coup = &che->coup;
1819 coup->coupling_point = 2 * get_bits1(gb);
1820 coup->num_coupled = get_bits(gb, 3);
1821 for (c = 0; c <= coup->num_coupled; c++) {
1823 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1824 coup->id_select[c] = get_bits(gb, 4);
1825 if (coup->type[c] == TYPE_CPE) {
1826 coup->ch_select[c] = get_bits(gb, 2);
1827 if (coup->ch_select[c] == 3)
1830 coup->ch_select[c] = 2;
1832 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1834 sign = get_bits(gb, 1);
1835 scale = cce_scale[get_bits(gb, 2)];
1837 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1840 for (c = 0; c < num_gain; c++) {
1844 float gain_cache = 1.;
1846 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1847 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1848 gain_cache = powf(scale, -gain);
1850 if (coup->coupling_point == AFTER_IMDCT) {
1851 coup->gain[c][0] = gain_cache;
1853 for (g = 0; g < sce->ics.num_window_groups; g++) {
1854 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1855 if (sce->band_type[idx] != ZERO_BT) {
1857 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1865 gain_cache = powf(scale, -t) * s;
1868 coup->gain[c][idx] = gain_cache;
1878 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1880 * @return Returns number of bytes consumed.
1882 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1886 int num_excl_chan = 0;
1889 for (i = 0; i < 7; i++)
1890 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1891 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1893 return num_excl_chan / 7;
1897 * Decode dynamic range information; reference: table 4.52.
1899 * @param cnt length of TYPE_FIL syntactic element in bytes
1901 * @return Returns number of bytes consumed.
1903 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1904 GetBitContext *gb, int cnt)
1907 int drc_num_bands = 1;
1910 /* pce_tag_present? */
1911 if (get_bits1(gb)) {
1912 che_drc->pce_instance_tag = get_bits(gb, 4);
1913 skip_bits(gb, 4); // tag_reserved_bits
1917 /* excluded_chns_present? */
1918 if (get_bits1(gb)) {
1919 n += decode_drc_channel_exclusions(che_drc, gb);
1922 /* drc_bands_present? */
1923 if (get_bits1(gb)) {
1924 che_drc->band_incr = get_bits(gb, 4);
1925 che_drc->interpolation_scheme = get_bits(gb, 4);
1927 drc_num_bands += che_drc->band_incr;
1928 for (i = 0; i < drc_num_bands; i++) {
1929 che_drc->band_top[i] = get_bits(gb, 8);
1934 /* prog_ref_level_present? */
1935 if (get_bits1(gb)) {
1936 che_drc->prog_ref_level = get_bits(gb, 7);
1937 skip_bits1(gb); // prog_ref_level_reserved_bits
1941 for (i = 0; i < drc_num_bands; i++) {
1942 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1943 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1950 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1952 int i, major, minor;
1957 get_bits(gb, 13); len -= 13;
1959 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
1960 buf[i] = get_bits(gb, 8);
1963 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
1964 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
1966 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
1967 ac->avctx->internal->skip_samples = 1024;
1971 skip_bits_long(gb, len);
1977 * Decode extension data (incomplete); reference: table 4.51.
1979 * @param cnt length of TYPE_FIL syntactic element in bytes
1981 * @return Returns number of bytes consumed
1983 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1984 ChannelElement *che, enum RawDataBlockType elem_type)
1988 switch (get_bits(gb, 4)) { // extension type
1989 case EXT_SBR_DATA_CRC:
1993 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1995 } else if (!ac->oc[1].m4ac.sbr) {
1996 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1997 skip_bits_long(gb, 8 * cnt - 4);
1999 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2000 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2001 skip_bits_long(gb, 8 * cnt - 4);
2003 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2004 ac->oc[1].m4ac.sbr = 1;
2005 ac->oc[1].m4ac.ps = 1;
2006 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2007 ac->oc[1].m4ac.chan_config, ac->oc[1].status);
2009 ac->oc[1].m4ac.sbr = 1;
2011 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2013 case EXT_DYNAMIC_RANGE:
2014 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
2017 decode_fill(ac, gb, 8 * cnt - 4);
2020 case EXT_DATA_ELEMENT:
2022 skip_bits_long(gb, 8 * cnt - 4);
2029 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2031 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2032 * @param coef spectral coefficients
2034 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2035 IndividualChannelStream *ics, int decode)
2037 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2039 int bottom, top, order, start, end, size, inc;
2040 float lpc[TNS_MAX_ORDER];
2041 float tmp[TNS_MAX_ORDER];
2043 for (w = 0; w < ics->num_windows; w++) {
2044 bottom = ics->num_swb;
2045 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2047 bottom = FFMAX(0, top - tns->length[w][filt]);
2048 order = tns->order[w][filt];
2053 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2055 start = ics->swb_offset[FFMIN(bottom, mmm)];
2056 end = ics->swb_offset[FFMIN( top, mmm)];
2057 if ((size = end - start) <= 0)
2059 if (tns->direction[w][filt]) {
2069 for (m = 0; m < size; m++, start += inc)
2070 for (i = 1; i <= FFMIN(m, order); i++)
2071 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2074 for (m = 0; m < size; m++, start += inc) {
2075 tmp[0] = coef[start];
2076 for (i = 1; i <= FFMIN(m, order); i++)
2077 coef[start] += tmp[i] * lpc[i - 1];
2078 for (i = order; i > 0; i--)
2079 tmp[i] = tmp[i - 1];
2087 * Apply windowing and MDCT to obtain the spectral
2088 * coefficient from the predicted sample by LTP.
2090 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2091 float *in, IndividualChannelStream *ics)
2093 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2094 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2095 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2096 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2098 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2099 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2101 memset(in, 0, 448 * sizeof(float));
2102 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2104 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2105 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2107 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2108 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2110 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2114 * Apply the long term prediction
2116 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2118 const LongTermPrediction *ltp = &sce->ics.ltp;
2119 const uint16_t *offsets = sce->ics.swb_offset;
2122 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2123 float *predTime = sce->ret;
2124 float *predFreq = ac->buf_mdct;
2125 int16_t num_samples = 2048;
2127 if (ltp->lag < 1024)
2128 num_samples = ltp->lag + 1024;
2129 for (i = 0; i < num_samples; i++)
2130 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2131 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2133 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2135 if (sce->tns.present)
2136 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2138 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2140 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2141 sce->coeffs[i] += predFreq[i];
2146 * Update the LTP buffer for next frame
2148 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2150 IndividualChannelStream *ics = &sce->ics;
2151 float *saved = sce->saved;
2152 float *saved_ltp = sce->coeffs;
2153 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2154 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2157 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2158 memcpy(saved_ltp, saved, 512 * sizeof(float));
2159 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2160 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2161 for (i = 0; i < 64; i++)
2162 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2163 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2164 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2165 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2166 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2167 for (i = 0; i < 64; i++)
2168 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2169 } else { // LONG_STOP or ONLY_LONG
2170 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2171 for (i = 0; i < 512; i++)
2172 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2175 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2176 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2177 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2181 * Conduct IMDCT and windowing.
2183 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2185 IndividualChannelStream *ics = &sce->ics;
2186 float *in = sce->coeffs;
2187 float *out = sce->ret;
2188 float *saved = sce->saved;
2189 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2190 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2191 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2192 float *buf = ac->buf_mdct;
2193 float *temp = ac->temp;
2197 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2198 for (i = 0; i < 1024; i += 128)
2199 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2201 ac->mdct.imdct_half(&ac->mdct, buf, in);
2203 /* window overlapping
2204 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2205 * and long to short transitions are considered to be short to short
2206 * transitions. This leaves just two cases (long to long and short to short)
2207 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2209 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2210 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2211 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2213 memcpy( out, saved, 448 * sizeof(float));
2215 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2216 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2217 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2218 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2219 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2220 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2221 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2223 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2224 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2229 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2230 memcpy( saved, temp + 64, 64 * sizeof(float));
2231 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2232 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2233 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2234 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2235 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2236 memcpy( saved, buf + 512, 448 * sizeof(float));
2237 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2238 } else { // LONG_STOP or ONLY_LONG
2239 memcpy( saved, buf + 512, 512 * sizeof(float));
2244 * Apply dependent channel coupling (applied before IMDCT).
2246 * @param index index into coupling gain array
2248 static void apply_dependent_coupling(AACContext *ac,
2249 SingleChannelElement *target,
2250 ChannelElement *cce, int index)
2252 IndividualChannelStream *ics = &cce->ch[0].ics;
2253 const uint16_t *offsets = ics->swb_offset;
2254 float *dest = target->coeffs;
2255 const float *src = cce->ch[0].coeffs;
2256 int g, i, group, k, idx = 0;
2257 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2258 av_log(ac->avctx, AV_LOG_ERROR,
2259 "Dependent coupling is not supported together with LTP\n");
2262 for (g = 0; g < ics->num_window_groups; g++) {
2263 for (i = 0; i < ics->max_sfb; i++, idx++) {
2264 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2265 const float gain = cce->coup.gain[index][idx];
2266 for (group = 0; group < ics->group_len[g]; group++) {
2267 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2269 dest[group * 128 + k] += gain * src[group * 128 + k];
2274 dest += ics->group_len[g] * 128;
2275 src += ics->group_len[g] * 128;
2280 * Apply independent channel coupling (applied after IMDCT).
2282 * @param index index into coupling gain array
2284 static void apply_independent_coupling(AACContext *ac,
2285 SingleChannelElement *target,
2286 ChannelElement *cce, int index)
2289 const float gain = cce->coup.gain[index][0];
2290 const float *src = cce->ch[0].ret;
2291 float *dest = target->ret;
2292 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2294 for (i = 0; i < len; i++)
2295 dest[i] += gain * src[i];
2299 * channel coupling transformation interface
2301 * @param apply_coupling_method pointer to (in)dependent coupling function
2303 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2304 enum RawDataBlockType type, int elem_id,
2305 enum CouplingPoint coupling_point,
2306 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2310 for (i = 0; i < MAX_ELEM_ID; i++) {
2311 ChannelElement *cce = ac->che[TYPE_CCE][i];
2314 if (cce && cce->coup.coupling_point == coupling_point) {
2315 ChannelCoupling *coup = &cce->coup;
2317 for (c = 0; c <= coup->num_coupled; c++) {
2318 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2319 if (coup->ch_select[c] != 1) {
2320 apply_coupling_method(ac, &cc->ch[0], cce, index);
2321 if (coup->ch_select[c] != 0)
2324 if (coup->ch_select[c] != 2)
2325 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2327 index += 1 + (coup->ch_select[c] == 3);
2334 * Convert spectral data to float samples, applying all supported tools as appropriate.
2336 static void spectral_to_sample(AACContext *ac)
2339 for (type = 3; type >= 0; type--) {
2340 for (i = 0; i < MAX_ELEM_ID; i++) {
2341 ChannelElement *che = ac->che[type][i];
2343 if (type <= TYPE_CPE)
2344 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2345 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2346 if (che->ch[0].ics.predictor_present) {
2347 if (che->ch[0].ics.ltp.present)
2348 apply_ltp(ac, &che->ch[0]);
2349 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2350 apply_ltp(ac, &che->ch[1]);
2353 if (che->ch[0].tns.present)
2354 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2355 if (che->ch[1].tns.present)
2356 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2357 if (type <= TYPE_CPE)
2358 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2359 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2360 imdct_and_windowing(ac, &che->ch[0]);
2361 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2362 update_ltp(ac, &che->ch[0]);
2363 if (type == TYPE_CPE) {
2364 imdct_and_windowing(ac, &che->ch[1]);
2365 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2366 update_ltp(ac, &che->ch[1]);
2368 if (ac->oc[1].m4ac.sbr > 0) {
2369 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2372 if (type <= TYPE_CCE)
2373 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2379 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2382 AACADTSHeaderInfo hdr_info;
2383 uint8_t layout_map[MAX_ELEM_ID*4][3];
2384 int layout_map_tags;
2386 size = avpriv_aac_parse_header(gb, &hdr_info);
2388 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2389 // This is 2 for "VLB " audio in NSV files.
2390 // See samples/nsv/vlb_audio.
2391 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2392 ac->warned_num_aac_frames = 1;
2394 push_output_configuration(ac);
2395 if (hdr_info.chan_config) {
2396 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2397 if (set_default_channel_config(ac->avctx, layout_map,
2398 &layout_map_tags, hdr_info.chan_config))
2400 if (output_configure(ac, layout_map, layout_map_tags,
2401 hdr_info.chan_config,
2402 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME)))
2405 ac->oc[1].m4ac.chan_config = 0;
2407 * dual mono frames in Japanese DTV can have chan_config 0
2408 * WITHOUT specifying PCE.
2409 * thus, set dual mono as default.
2411 if (ac->enable_jp_dmono && ac->oc[0].status == OC_NONE) {
2412 layout_map_tags = 2;
2413 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2414 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2415 layout_map[0][1] = 0;
2416 layout_map[1][1] = 1;
2417 if (output_configure(ac, layout_map, layout_map_tags,
2422 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2423 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2424 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2425 if (ac->oc[0].status != OC_LOCKED ||
2426 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2427 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2428 ac->oc[1].m4ac.sbr = -1;
2429 ac->oc[1].m4ac.ps = -1;
2431 if (!hdr_info.crc_absent)
2437 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2438 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2440 AACContext *ac = avctx->priv_data;
2441 ChannelElement *che = NULL, *che_prev = NULL;
2442 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2444 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2445 int is_dmono, sce_count = 0;
2448 if (show_bits(gb, 12) == 0xfff) {
2449 if (parse_adts_frame_header(ac, gb) < 0) {
2450 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2454 if (ac->oc[1].m4ac.sampling_index > 12) {
2455 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2461 ac->tags_mapped = 0;
2463 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2464 elem_id = get_bits(gb, 4);
2466 if (elem_type < TYPE_DSE) {
2467 if (!(che=get_che(ac, elem_type, elem_id))) {
2468 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2469 elem_type, elem_id);
2476 switch (elem_type) {
2479 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2485 err = decode_cpe(ac, gb, che);
2490 err = decode_cce(ac, gb, che);
2494 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2499 err = skip_data_stream_element(ac, gb);
2503 uint8_t layout_map[MAX_ELEM_ID*4][3];
2505 push_output_configuration(ac);
2506 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2512 av_log(avctx, AV_LOG_ERROR,
2513 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2514 pop_output_configuration(ac);
2516 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2518 ac->oc[1].m4ac.chan_config = 0;
2526 elem_id += get_bits(gb, 8) - 1;
2527 if (get_bits_left(gb) < 8 * elem_id) {
2528 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2533 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2534 err = 0; /* FIXME */
2538 err = -1; /* should not happen, but keeps compiler happy */
2543 elem_type_prev = elem_type;
2548 if (get_bits_left(gb) < 3) {
2549 av_log(avctx, AV_LOG_ERROR, overread_err);
2555 spectral_to_sample(ac);
2557 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2558 samples <<= multiplier;
2560 /* for dual-mono audio (SCE + SCE) */
2561 is_dmono = ac->enable_jp_dmono && sce_count == 2 &&
2562 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2565 if (ac->dmono_mode == 0) {
2566 tmp = ac->output_data[1];
2567 ac->output_data[1] = ac->output_data[0];
2568 } else if (ac->dmono_mode == 1) {
2569 tmp = ac->output_data[0];
2570 ac->output_data[0] = ac->output_data[1];
2575 /* get output buffer */
2576 ac->frame.nb_samples = samples;
2577 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2578 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2583 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2584 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2585 (const float **)ac->output_data,
2586 samples, avctx->channels);
2588 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2589 (const float **)ac->output_data,
2590 samples, avctx->channels);
2592 *(AVFrame *)data = ac->frame;
2594 *got_frame_ptr = !!samples;
2597 if (ac->dmono_mode == 0)
2598 ac->output_data[1] = tmp;
2599 else if (ac->dmono_mode == 1)
2600 ac->output_data[0] = tmp;
2603 if (ac->oc[1].status && audio_found) {
2604 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2605 avctx->frame_size = samples;
2606 ac->oc[1].status = OC_LOCKED;
2611 uint32_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2612 if (side && side_size>=4)
2613 AV_WL32(side, 2*AV_RL32(side));
2617 pop_output_configuration(ac);
2621 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2622 int *got_frame_ptr, AVPacket *avpkt)
2624 AACContext *ac = avctx->priv_data;
2625 const uint8_t *buf = avpkt->data;
2626 int buf_size = avpkt->size;
2631 int new_extradata_size;
2632 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2633 AV_PKT_DATA_NEW_EXTRADATA,
2634 &new_extradata_size);
2635 int jp_dualmono_size;
2636 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2637 AV_PKT_DATA_JP_DUALMONO,
2640 if (new_extradata && 0) {
2641 av_free(avctx->extradata);
2642 avctx->extradata = av_mallocz(new_extradata_size +
2643 FF_INPUT_BUFFER_PADDING_SIZE);
2644 if (!avctx->extradata)
2645 return AVERROR(ENOMEM);
2646 avctx->extradata_size = new_extradata_size;
2647 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2648 push_output_configuration(ac);
2649 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2651 avctx->extradata_size*8, 1) < 0) {
2652 pop_output_configuration(ac);
2653 return AVERROR_INVALIDDATA;
2657 ac->enable_jp_dmono = !!jp_dualmono;
2659 if (jp_dualmono && jp_dualmono_size > 0)
2660 ac->dmono_mode = *jp_dualmono;
2662 init_get_bits(&gb, buf, buf_size * 8);
2664 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2667 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2668 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2669 if (buf[buf_offset])
2672 return buf_size > buf_offset ? buf_consumed : buf_size;
2675 static av_cold int aac_decode_close(AVCodecContext *avctx)
2677 AACContext *ac = avctx->priv_data;
2680 for (i = 0; i < MAX_ELEM_ID; i++) {
2681 for (type = 0; type < 4; type++) {
2682 if (ac->che[type][i])
2683 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2684 av_freep(&ac->che[type][i]);
2688 ff_mdct_end(&ac->mdct);
2689 ff_mdct_end(&ac->mdct_small);
2690 ff_mdct_end(&ac->mdct_ltp);
2695 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2697 struct LATMContext {
2698 AACContext aac_ctx; ///< containing AACContext
2699 int initialized; ///< initialized after a valid extradata was seen
2702 int audio_mux_version_A; ///< LATM syntax version
2703 int frame_length_type; ///< 0/1 variable/fixed frame length
2704 int frame_length; ///< frame length for fixed frame length
2707 static inline uint32_t latm_get_value(GetBitContext *b)
2709 int length = get_bits(b, 2);
2711 return get_bits_long(b, (length+1)*8);
2714 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2715 GetBitContext *gb, int asclen)
2717 AACContext *ac = &latmctx->aac_ctx;
2718 AVCodecContext *avctx = ac->avctx;
2719 MPEG4AudioConfig m4ac = { 0 };
2720 int config_start_bit = get_bits_count(gb);
2721 int sync_extension = 0;
2722 int bits_consumed, esize;
2726 asclen = FFMIN(asclen, get_bits_left(gb));
2728 asclen = get_bits_left(gb);
2730 if (config_start_bit % 8) {
2731 av_log_missing_feature(latmctx->aac_ctx.avctx,
2732 "Non-byte-aligned audio-specific config", 1);
2733 return AVERROR_PATCHWELCOME;
2736 return AVERROR_INVALIDDATA;
2737 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2738 gb->buffer + (config_start_bit / 8),
2739 asclen, sync_extension);
2741 if (bits_consumed < 0)
2742 return AVERROR_INVALIDDATA;
2744 if (!latmctx->initialized ||
2745 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2746 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2748 if(latmctx->initialized) {
2749 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2751 av_log(avctx, AV_LOG_INFO, "initializing latmctx\n");
2753 latmctx->initialized = 0;
2755 esize = (bits_consumed+7) / 8;
2757 if (avctx->extradata_size < esize) {
2758 av_free(avctx->extradata);
2759 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2760 if (!avctx->extradata)
2761 return AVERROR(ENOMEM);
2764 avctx->extradata_size = esize;
2765 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2766 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2768 skip_bits_long(gb, bits_consumed);
2770 return bits_consumed;
2773 static int read_stream_mux_config(struct LATMContext *latmctx,
2776 int ret, audio_mux_version = get_bits(gb, 1);
2778 latmctx->audio_mux_version_A = 0;
2779 if (audio_mux_version)
2780 latmctx->audio_mux_version_A = get_bits(gb, 1);
2782 if (!latmctx->audio_mux_version_A) {
2784 if (audio_mux_version)
2785 latm_get_value(gb); // taraFullness
2787 skip_bits(gb, 1); // allStreamSameTimeFraming
2788 skip_bits(gb, 6); // numSubFrames
2790 if (get_bits(gb, 4)) { // numPrograms
2791 av_log_missing_feature(latmctx->aac_ctx.avctx,
2792 "Multiple programs", 1);
2793 return AVERROR_PATCHWELCOME;
2796 // for each program (which there is only one in DVB)
2798 // for each layer (which there is only one in DVB)
2799 if (get_bits(gb, 3)) { // numLayer
2800 av_log_missing_feature(latmctx->aac_ctx.avctx,
2801 "Multiple layers", 1);
2802 return AVERROR_PATCHWELCOME;
2805 // for all but first stream: use_same_config = get_bits(gb, 1);
2806 if (!audio_mux_version) {
2807 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2810 int ascLen = latm_get_value(gb);
2811 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2814 skip_bits_long(gb, ascLen);
2817 latmctx->frame_length_type = get_bits(gb, 3);
2818 switch (latmctx->frame_length_type) {
2820 skip_bits(gb, 8); // latmBufferFullness
2823 latmctx->frame_length = get_bits(gb, 9);
2828 skip_bits(gb, 6); // CELP frame length table index
2832 skip_bits(gb, 1); // HVXC frame length table index
2836 if (get_bits(gb, 1)) { // other data
2837 if (audio_mux_version) {
2838 latm_get_value(gb); // other_data_bits
2842 esc = get_bits(gb, 1);
2848 if (get_bits(gb, 1)) // crc present
2849 skip_bits(gb, 8); // config_crc
2855 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2859 if (ctx->frame_length_type == 0) {
2860 int mux_slot_length = 0;
2862 tmp = get_bits(gb, 8);
2863 mux_slot_length += tmp;
2864 } while (tmp == 255);
2865 return mux_slot_length;
2866 } else if (ctx->frame_length_type == 1) {
2867 return ctx->frame_length;
2868 } else if (ctx->frame_length_type == 3 ||
2869 ctx->frame_length_type == 5 ||
2870 ctx->frame_length_type == 7) {
2871 skip_bits(gb, 2); // mux_slot_length_coded
2876 static int read_audio_mux_element(struct LATMContext *latmctx,
2880 uint8_t use_same_mux = get_bits(gb, 1);
2881 if (!use_same_mux) {
2882 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2884 } else if (!latmctx->aac_ctx.avctx->extradata) {
2885 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2886 "no decoder config found\n");
2887 return AVERROR(EAGAIN);
2889 if (latmctx->audio_mux_version_A == 0) {
2890 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2891 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2892 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2893 return AVERROR_INVALIDDATA;
2894 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2895 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2896 "frame length mismatch %d << %d\n",
2897 mux_slot_length_bytes * 8, get_bits_left(gb));
2898 return AVERROR_INVALIDDATA;
2905 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2906 int *got_frame_ptr, AVPacket *avpkt)
2908 struct LATMContext *latmctx = avctx->priv_data;
2912 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2914 // check for LOAS sync word
2915 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2916 return AVERROR_INVALIDDATA;
2918 muxlength = get_bits(&gb, 13) + 3;
2919 // not enough data, the parser should have sorted this out
2920 if (muxlength > avpkt->size)
2921 return AVERROR_INVALIDDATA;
2923 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2926 if (!latmctx->initialized) {
2927 if (!avctx->extradata) {
2931 push_output_configuration(&latmctx->aac_ctx);
2932 if ((err = decode_audio_specific_config(
2933 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2934 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2935 pop_output_configuration(&latmctx->aac_ctx);
2938 latmctx->initialized = 1;
2942 if (show_bits(&gb, 12) == 0xfff) {
2943 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2944 "ADTS header detected, probably as result of configuration "
2946 return AVERROR_INVALIDDATA;
2949 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2955 static av_cold int latm_decode_init(AVCodecContext *avctx)
2957 struct LATMContext *latmctx = avctx->priv_data;
2958 int ret = aac_decode_init(avctx);
2960 if (avctx->extradata_size > 0)
2961 latmctx->initialized = !ret;
2967 AVCodec ff_aac_decoder = {
2969 .type = AVMEDIA_TYPE_AUDIO,
2970 .id = AV_CODEC_ID_AAC,
2971 .priv_data_size = sizeof(AACContext),
2972 .init = aac_decode_init,
2973 .close = aac_decode_close,
2974 .decode = aac_decode_frame,
2975 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
2976 .sample_fmts = (const enum AVSampleFormat[]) {
2977 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2979 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2980 .channel_layouts = aac_channel_layout,
2985 Note: This decoder filter is intended to decode LATM streams transferred
2986 in MPEG transport streams which only contain one program.
2987 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2989 AVCodec ff_aac_latm_decoder = {
2991 .type = AVMEDIA_TYPE_AUDIO,
2992 .id = AV_CODEC_ID_AAC_LATM,
2993 .priv_data_size = sizeof(struct LATMContext),
2994 .init = latm_decode_init,
2995 .close = aac_decode_close,
2996 .decode = latm_decode_frame,
2997 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
2998 .sample_fmts = (const enum AVSampleFormat[]) {
2999 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
3001 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3002 .channel_layouts = aac_channel_layout,