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_element[(*channels)++] = &ac->che[type][id]->ch[0];
157 if (type == TYPE_CPE ||
158 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
159 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
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 static int frame_configure_elements(AVCodecContext *avctx)
172 AACContext *ac = avctx->priv_data;
173 int type, id, ch, ret;
175 /* set channel pointers to internal buffers by default */
176 for (type = 0; type < 4; type++) {
177 for (id = 0; id < MAX_ELEM_ID; id++) {
178 ChannelElement *che = ac->che[type][id];
180 che->ch[0].ret = che->ch[0].ret_buf;
181 che->ch[1].ret = che->ch[1].ret_buf;
186 /* get output buffer */
187 ac->frame.nb_samples = 2048;
188 if ((ret = avctx->get_buffer(avctx, &ac->frame)) < 0) {
189 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
193 /* map output channel pointers to AVFrame data */
194 for (ch = 0; ch < avctx->channels; ch++) {
195 if (ac->output_element[ch])
196 ac->output_element[ch]->ret = (float *)ac->frame.extended_data[ch];
202 struct elem_to_channel {
203 uint64_t av_position;
206 uint8_t aac_position;
209 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
210 uint8_t (*layout_map)[3], int offset, uint64_t left,
211 uint64_t right, int pos)
213 if (layout_map[offset][0] == TYPE_CPE) {
214 e2c_vec[offset] = (struct elem_to_channel) {
215 .av_position = left | right, .syn_ele = TYPE_CPE,
216 .elem_id = layout_map[offset ][1], .aac_position = pos };
219 e2c_vec[offset] = (struct elem_to_channel) {
220 .av_position = left, .syn_ele = TYPE_SCE,
221 .elem_id = layout_map[offset ][1], .aac_position = pos };
222 e2c_vec[offset + 1] = (struct elem_to_channel) {
223 .av_position = right, .syn_ele = TYPE_SCE,
224 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
229 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
230 int num_pos_channels = 0;
234 for (i = *current; i < tags; i++) {
235 if (layout_map[i][2] != pos)
237 if (layout_map[i][0] == TYPE_CPE) {
239 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
245 num_pos_channels += 2;
253 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
256 return num_pos_channels;
259 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
261 int i, n, total_non_cc_elements;
262 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
263 int num_front_channels, num_side_channels, num_back_channels;
266 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
271 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
272 if (num_front_channels < 0)
275 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
276 if (num_side_channels < 0)
279 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
280 if (num_back_channels < 0)
284 if (num_front_channels & 1) {
285 e2c_vec[i] = (struct elem_to_channel) {
286 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
287 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
289 num_front_channels--;
291 if (num_front_channels >= 4) {
292 i += assign_pair(e2c_vec, layout_map, i,
293 AV_CH_FRONT_LEFT_OF_CENTER,
294 AV_CH_FRONT_RIGHT_OF_CENTER,
296 num_front_channels -= 2;
298 if (num_front_channels >= 2) {
299 i += assign_pair(e2c_vec, layout_map, i,
303 num_front_channels -= 2;
305 while (num_front_channels >= 2) {
306 i += assign_pair(e2c_vec, layout_map, i,
310 num_front_channels -= 2;
313 if (num_side_channels >= 2) {
314 i += assign_pair(e2c_vec, layout_map, i,
318 num_side_channels -= 2;
320 while (num_side_channels >= 2) {
321 i += assign_pair(e2c_vec, layout_map, i,
325 num_side_channels -= 2;
328 while (num_back_channels >= 4) {
329 i += assign_pair(e2c_vec, layout_map, i,
333 num_back_channels -= 2;
335 if (num_back_channels >= 2) {
336 i += assign_pair(e2c_vec, layout_map, i,
340 num_back_channels -= 2;
342 if (num_back_channels) {
343 e2c_vec[i] = (struct elem_to_channel) {
344 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
345 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
350 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
351 e2c_vec[i] = (struct elem_to_channel) {
352 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
353 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
356 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
357 e2c_vec[i] = (struct elem_to_channel) {
358 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
359 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
363 // Must choose a stable sort
364 total_non_cc_elements = n = i;
367 for (i = 1; i < n; i++) {
368 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
369 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
377 for (i = 0; i < total_non_cc_elements; i++) {
378 layout_map[i][0] = e2c_vec[i].syn_ele;
379 layout_map[i][1] = e2c_vec[i].elem_id;
380 layout_map[i][2] = e2c_vec[i].aac_position;
381 if (e2c_vec[i].av_position != UINT64_MAX) {
382 layout |= e2c_vec[i].av_position;
390 * Save current output configuration if and only if it has been locked.
392 static void push_output_configuration(AACContext *ac) {
393 if (ac->oc[1].status == OC_LOCKED) {
394 ac->oc[0] = ac->oc[1];
396 ac->oc[1].status = OC_NONE;
400 * Restore the previous output configuration if and only if the current
401 * configuration is unlocked.
403 static void pop_output_configuration(AACContext *ac) {
404 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
405 ac->oc[1] = ac->oc[0];
406 ac->avctx->channels = ac->oc[1].channels;
407 ac->avctx->channel_layout = ac->oc[1].channel_layout;
412 * Configure output channel order based on the current program configuration element.
414 * @return Returns error status. 0 - OK, !0 - error
416 static int output_configure(AACContext *ac,
417 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
418 enum OCStatus oc_type, int get_new_frame)
420 AVCodecContext *avctx = ac->avctx;
421 int i, channels = 0, ret;
424 if (ac->oc[1].layout_map != layout_map) {
425 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
426 ac->oc[1].layout_map_tags = tags;
429 // Try to sniff a reasonable channel order, otherwise output the
430 // channels in the order the PCE declared them.
431 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
432 layout = sniff_channel_order(layout_map, tags);
433 for (i = 0; i < tags; i++) {
434 int type = layout_map[i][0];
435 int id = layout_map[i][1];
436 int position = layout_map[i][2];
437 // Allocate or free elements depending on if they are in the
438 // current program configuration.
439 ret = che_configure(ac, position, type, id, &channels);
443 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
444 if (layout == AV_CH_FRONT_CENTER) {
445 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
451 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
452 if (layout) avctx->channel_layout = layout;
453 ac->oc[1].channel_layout = layout;
454 avctx->channels = ac->oc[1].channels = channels;
455 ac->oc[1].status = oc_type;
458 if ((ret = frame_configure_elements(ac->avctx)) < 0)
465 static void flush(AVCodecContext *avctx)
467 AACContext *ac= avctx->priv_data;
470 for (type = 3; type >= 0; type--) {
471 for (i = 0; i < MAX_ELEM_ID; i++) {
472 ChannelElement *che = ac->che[type][i];
474 for (j = 0; j <= 1; j++) {
475 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
483 * Set up channel positions based on a default channel configuration
484 * as specified in table 1.17.
486 * @return Returns error status. 0 - OK, !0 - error
488 static int set_default_channel_config(AVCodecContext *avctx,
489 uint8_t (*layout_map)[3],
493 if (channel_config < 1 || channel_config > 7) {
494 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
498 *tags = tags_per_config[channel_config];
499 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
503 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
505 // For PCE based channel configurations map the channels solely based on tags.
506 if (!ac->oc[1].m4ac.chan_config) {
507 return ac->tag_che_map[type][elem_id];
509 // Allow single CPE stereo files to be signalled with mono configuration.
510 if (!ac->tags_mapped && type == TYPE_CPE && ac->oc[1].m4ac.chan_config == 1) {
511 uint8_t layout_map[MAX_ELEM_ID*4][3];
513 push_output_configuration(ac);
515 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
517 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
520 if (output_configure(ac, layout_map, layout_map_tags,
521 OC_TRIAL_FRAME, 1) < 0)
524 ac->oc[1].m4ac.chan_config = 2;
525 ac->oc[1].m4ac.ps = 0;
528 if (!ac->tags_mapped && type == TYPE_SCE && ac->oc[1].m4ac.chan_config == 2) {
529 uint8_t layout_map[MAX_ELEM_ID*4][3];
531 push_output_configuration(ac);
533 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
535 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
538 if (output_configure(ac, layout_map, layout_map_tags,
539 OC_TRIAL_FRAME, 1) < 0)
542 ac->oc[1].m4ac.chan_config = 1;
543 if (ac->oc[1].m4ac.sbr)
544 ac->oc[1].m4ac.ps = -1;
546 // For indexed channel configurations map the channels solely based on position.
547 switch (ac->oc[1].m4ac.chan_config) {
549 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
551 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
554 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
555 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
556 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
557 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
559 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
562 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
564 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
567 if (ac->tags_mapped == 2 && ac->oc[1].m4ac.chan_config == 4 && type == TYPE_SCE) {
569 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
573 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) && type == TYPE_CPE) {
575 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
576 } else if (ac->oc[1].m4ac.chan_config == 2) {
580 if (!ac->tags_mapped && type == TYPE_SCE) {
582 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
590 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
592 * @param type speaker type/position for these channels
594 static void decode_channel_map(uint8_t layout_map[][3],
595 enum ChannelPosition type,
596 GetBitContext *gb, int n)
599 enum RawDataBlockType syn_ele;
601 case AAC_CHANNEL_FRONT:
602 case AAC_CHANNEL_BACK:
603 case AAC_CHANNEL_SIDE:
604 syn_ele = get_bits1(gb);
610 case AAC_CHANNEL_LFE:
616 layout_map[0][0] = syn_ele;
617 layout_map[0][1] = get_bits(gb, 4);
618 layout_map[0][2] = type;
624 * Decode program configuration element; reference: table 4.2.
626 * @return Returns error status. 0 - OK, !0 - error
628 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
629 uint8_t (*layout_map)[3],
632 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
636 skip_bits(gb, 2); // object_type
638 sampling_index = get_bits(gb, 4);
639 if (m4ac->sampling_index != sampling_index)
640 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");
642 num_front = get_bits(gb, 4);
643 num_side = get_bits(gb, 4);
644 num_back = get_bits(gb, 4);
645 num_lfe = get_bits(gb, 2);
646 num_assoc_data = get_bits(gb, 3);
647 num_cc = get_bits(gb, 4);
650 skip_bits(gb, 4); // mono_mixdown_tag
652 skip_bits(gb, 4); // stereo_mixdown_tag
655 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
657 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
658 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
661 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
663 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
665 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
667 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
670 skip_bits_long(gb, 4 * num_assoc_data);
672 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
677 /* comment field, first byte is length */
678 comment_len = get_bits(gb, 8) * 8;
679 if (get_bits_left(gb) < comment_len) {
680 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
683 skip_bits_long(gb, comment_len);
688 * Decode GA "General Audio" specific configuration; reference: table 4.1.
690 * @param ac pointer to AACContext, may be null
691 * @param avctx pointer to AVCCodecContext, used for logging
693 * @return Returns error status. 0 - OK, !0 - error
695 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
697 MPEG4AudioConfig *m4ac,
700 int extension_flag, ret;
701 uint8_t layout_map[MAX_ELEM_ID*4][3];
704 if (get_bits1(gb)) { // frameLengthFlag
705 av_log_missing_feature(avctx, "960/120 MDCT window", 1);
706 return AVERROR_PATCHWELCOME;
709 if (get_bits1(gb)) // dependsOnCoreCoder
710 skip_bits(gb, 14); // coreCoderDelay
711 extension_flag = get_bits1(gb);
713 if (m4ac->object_type == AOT_AAC_SCALABLE ||
714 m4ac->object_type == AOT_ER_AAC_SCALABLE)
715 skip_bits(gb, 3); // layerNr
717 if (channel_config == 0) {
718 skip_bits(gb, 4); // element_instance_tag
719 tags = decode_pce(avctx, m4ac, layout_map, gb);
723 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
727 if (count_channels(layout_map, tags) > 1) {
729 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
732 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
735 if (extension_flag) {
736 switch (m4ac->object_type) {
738 skip_bits(gb, 5); // numOfSubFrame
739 skip_bits(gb, 11); // layer_length
743 case AOT_ER_AAC_SCALABLE:
745 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
746 * aacScalefactorDataResilienceFlag
747 * aacSpectralDataResilienceFlag
751 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
757 * Decode audio specific configuration; reference: table 1.13.
759 * @param ac pointer to AACContext, may be null
760 * @param avctx pointer to AVCCodecContext, used for logging
761 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
762 * @param data pointer to buffer holding an audio specific config
763 * @param bit_size size of audio specific config or data in bits
764 * @param sync_extension look for an appended sync extension
766 * @return Returns error status or number of consumed bits. <0 - error
768 static int decode_audio_specific_config(AACContext *ac,
769 AVCodecContext *avctx,
770 MPEG4AudioConfig *m4ac,
771 const uint8_t *data, int bit_size,
777 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
778 for (i = 0; i < bit_size >> 3; i++)
779 av_dlog(avctx, "%02x ", data[i]);
780 av_dlog(avctx, "\n");
782 init_get_bits(&gb, data, bit_size);
784 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
786 if (m4ac->sampling_index > 12) {
787 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
791 skip_bits_long(&gb, i);
793 switch (m4ac->object_type) {
797 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
801 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
802 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
806 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
807 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
808 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
810 return get_bits_count(&gb);
814 * linear congruential pseudorandom number generator
816 * @param previous_val pointer to the current state of the generator
818 * @return Returns a 32-bit pseudorandom integer
820 static av_always_inline int lcg_random(unsigned previous_val)
822 return previous_val * 1664525 + 1013904223;
825 static av_always_inline void reset_predict_state(PredictorState *ps)
835 static void reset_all_predictors(PredictorState *ps)
838 for (i = 0; i < MAX_PREDICTORS; i++)
839 reset_predict_state(&ps[i]);
842 static int sample_rate_idx (int rate)
844 if (92017 <= rate) return 0;
845 else if (75132 <= rate) return 1;
846 else if (55426 <= rate) return 2;
847 else if (46009 <= rate) return 3;
848 else if (37566 <= rate) return 4;
849 else if (27713 <= rate) return 5;
850 else if (23004 <= rate) return 6;
851 else if (18783 <= rate) return 7;
852 else if (13856 <= rate) return 8;
853 else if (11502 <= rate) return 9;
854 else if (9391 <= rate) return 10;
858 static void reset_predictor_group(PredictorState *ps, int group_num)
861 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
862 reset_predict_state(&ps[i]);
865 #define AAC_INIT_VLC_STATIC(num, size) \
866 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
867 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
868 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
871 static av_cold int aac_decode_init(AVCodecContext *avctx)
873 AACContext *ac = avctx->priv_data;
876 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
878 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
880 if (avctx->extradata_size > 0) {
881 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
883 avctx->extradata_size*8, 1) < 0)
887 uint8_t layout_map[MAX_ELEM_ID*4][3];
890 sr = sample_rate_idx(avctx->sample_rate);
891 ac->oc[1].m4ac.sampling_index = sr;
892 ac->oc[1].m4ac.channels = avctx->channels;
893 ac->oc[1].m4ac.sbr = -1;
894 ac->oc[1].m4ac.ps = -1;
896 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
897 if (ff_mpeg4audio_channels[i] == avctx->channels)
899 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
902 ac->oc[1].m4ac.chan_config = i;
904 if (ac->oc[1].m4ac.chan_config) {
905 int ret = set_default_channel_config(avctx, layout_map,
906 &layout_map_tags, ac->oc[1].m4ac.chan_config);
908 output_configure(ac, layout_map, layout_map_tags,
910 else if (avctx->err_recognition & AV_EF_EXPLODE)
911 return AVERROR_INVALIDDATA;
915 AAC_INIT_VLC_STATIC( 0, 304);
916 AAC_INIT_VLC_STATIC( 1, 270);
917 AAC_INIT_VLC_STATIC( 2, 550);
918 AAC_INIT_VLC_STATIC( 3, 300);
919 AAC_INIT_VLC_STATIC( 4, 328);
920 AAC_INIT_VLC_STATIC( 5, 294);
921 AAC_INIT_VLC_STATIC( 6, 306);
922 AAC_INIT_VLC_STATIC( 7, 268);
923 AAC_INIT_VLC_STATIC( 8, 510);
924 AAC_INIT_VLC_STATIC( 9, 366);
925 AAC_INIT_VLC_STATIC(10, 462);
929 ff_dsputil_init(&ac->dsp, avctx);
930 ff_fmt_convert_init(&ac->fmt_conv, avctx);
931 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
933 ac->random_state = 0x1f2e3d4c;
937 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
938 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
939 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
942 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
943 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
944 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
945 // window initialization
946 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
947 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
948 ff_init_ff_sine_windows(10);
949 ff_init_ff_sine_windows( 7);
953 avcodec_get_frame_defaults(&ac->frame);
954 avctx->coded_frame = &ac->frame;
960 * Skip data_stream_element; reference: table 4.10.
962 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
964 int byte_align = get_bits1(gb);
965 int count = get_bits(gb, 8);
967 count += get_bits(gb, 8);
971 if (get_bits_left(gb) < 8 * count) {
972 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
975 skip_bits_long(gb, 8 * count);
979 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
984 ics->predictor_reset_group = get_bits(gb, 5);
985 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
986 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
990 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
991 ics->prediction_used[sfb] = get_bits1(gb);
997 * Decode Long Term Prediction data; reference: table 4.xx.
999 static void decode_ltp(LongTermPrediction *ltp,
1000 GetBitContext *gb, uint8_t max_sfb)
1004 ltp->lag = get_bits(gb, 11);
1005 ltp->coef = ltp_coef[get_bits(gb, 3)];
1006 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1007 ltp->used[sfb] = get_bits1(gb);
1011 * Decode Individual Channel Stream info; reference: table 4.6.
1013 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1016 if (get_bits1(gb)) {
1017 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1018 return AVERROR_INVALIDDATA;
1020 ics->window_sequence[1] = ics->window_sequence[0];
1021 ics->window_sequence[0] = get_bits(gb, 2);
1022 ics->use_kb_window[1] = ics->use_kb_window[0];
1023 ics->use_kb_window[0] = get_bits1(gb);
1024 ics->num_window_groups = 1;
1025 ics->group_len[0] = 1;
1026 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1028 ics->max_sfb = get_bits(gb, 4);
1029 for (i = 0; i < 7; i++) {
1030 if (get_bits1(gb)) {
1031 ics->group_len[ics->num_window_groups - 1]++;
1033 ics->num_window_groups++;
1034 ics->group_len[ics->num_window_groups - 1] = 1;
1037 ics->num_windows = 8;
1038 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1039 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1040 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1041 ics->predictor_present = 0;
1043 ics->max_sfb = get_bits(gb, 6);
1044 ics->num_windows = 1;
1045 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1046 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1047 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1048 ics->predictor_present = get_bits1(gb);
1049 ics->predictor_reset_group = 0;
1050 if (ics->predictor_present) {
1051 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1052 if (decode_prediction(ac, ics, gb)) {
1055 } else if (ac->oc[1].m4ac.object_type == AOT_AAC_LC) {
1056 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
1059 if ((ics->ltp.present = get_bits(gb, 1)))
1060 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1065 if (ics->max_sfb > ics->num_swb) {
1066 av_log(ac->avctx, AV_LOG_ERROR,
1067 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
1068 ics->max_sfb, ics->num_swb);
1075 return AVERROR_INVALIDDATA;
1079 * Decode band types (section_data payload); reference: table 4.46.
1081 * @param band_type array of the used band type
1082 * @param band_type_run_end array of the last scalefactor band of a band type run
1084 * @return Returns error status. 0 - OK, !0 - error
1086 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1087 int band_type_run_end[120], GetBitContext *gb,
1088 IndividualChannelStream *ics)
1091 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1092 for (g = 0; g < ics->num_window_groups; g++) {
1094 while (k < ics->max_sfb) {
1095 uint8_t sect_end = k;
1097 int sect_band_type = get_bits(gb, 4);
1098 if (sect_band_type == 12) {
1099 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1103 sect_len_incr = get_bits(gb, bits);
1104 sect_end += sect_len_incr;
1105 if (get_bits_left(gb) < 0) {
1106 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1109 if (sect_end > ics->max_sfb) {
1110 av_log(ac->avctx, AV_LOG_ERROR,
1111 "Number of bands (%d) exceeds limit (%d).\n",
1112 sect_end, ics->max_sfb);
1115 } while (sect_len_incr == (1 << bits) - 1);
1116 for (; k < sect_end; k++) {
1117 band_type [idx] = sect_band_type;
1118 band_type_run_end[idx++] = sect_end;
1126 * Decode scalefactors; reference: table 4.47.
1128 * @param global_gain first scalefactor value as scalefactors are differentially coded
1129 * @param band_type array of the used band type
1130 * @param band_type_run_end array of the last scalefactor band of a band type run
1131 * @param sf array of scalefactors or intensity stereo positions
1133 * @return Returns error status. 0 - OK, !0 - error
1135 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1136 unsigned int global_gain,
1137 IndividualChannelStream *ics,
1138 enum BandType band_type[120],
1139 int band_type_run_end[120])
1142 int offset[3] = { global_gain, global_gain - 90, 0 };
1145 for (g = 0; g < ics->num_window_groups; g++) {
1146 for (i = 0; i < ics->max_sfb;) {
1147 int run_end = band_type_run_end[idx];
1148 if (band_type[idx] == ZERO_BT) {
1149 for (; i < run_end; i++, idx++)
1151 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1152 for (; i < run_end; i++, idx++) {
1153 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1154 clipped_offset = av_clip(offset[2], -155, 100);
1155 if (offset[2] != clipped_offset) {
1156 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1157 "position clipped (%d -> %d).\nIf you heard an "
1158 "audible artifact, there may be a bug in the "
1159 "decoder. ", offset[2], clipped_offset);
1161 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1163 } else if (band_type[idx] == NOISE_BT) {
1164 for (; i < run_end; i++, idx++) {
1165 if (noise_flag-- > 0)
1166 offset[1] += get_bits(gb, 9) - 256;
1168 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1169 clipped_offset = av_clip(offset[1], -100, 155);
1170 if (offset[1] != clipped_offset) {
1171 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1172 "(%d -> %d).\nIf you heard an audible "
1173 "artifact, there may be a bug in the decoder. ",
1174 offset[1], clipped_offset);
1176 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1179 for (; i < run_end; i++, idx++) {
1180 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1181 if (offset[0] > 255U) {
1182 av_log(ac->avctx, AV_LOG_ERROR,
1183 "Scalefactor (%d) out of range.\n", offset[0]);
1186 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1195 * Decode pulse data; reference: table 4.7.
1197 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1198 const uint16_t *swb_offset, int num_swb)
1201 pulse->num_pulse = get_bits(gb, 2) + 1;
1202 pulse_swb = get_bits(gb, 6);
1203 if (pulse_swb >= num_swb)
1205 pulse->pos[0] = swb_offset[pulse_swb];
1206 pulse->pos[0] += get_bits(gb, 5);
1207 if (pulse->pos[0] > 1023)
1209 pulse->amp[0] = get_bits(gb, 4);
1210 for (i = 1; i < pulse->num_pulse; i++) {
1211 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1212 if (pulse->pos[i] > 1023)
1214 pulse->amp[i] = get_bits(gb, 4);
1220 * Decode Temporal Noise Shaping data; reference: table 4.48.
1222 * @return Returns error status. 0 - OK, !0 - error
1224 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1225 GetBitContext *gb, const IndividualChannelStream *ics)
1227 int w, filt, i, coef_len, coef_res, coef_compress;
1228 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1229 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1230 for (w = 0; w < ics->num_windows; w++) {
1231 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1232 coef_res = get_bits1(gb);
1234 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1236 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1238 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1239 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1240 tns->order[w][filt], tns_max_order);
1241 tns->order[w][filt] = 0;
1244 if (tns->order[w][filt]) {
1245 tns->direction[w][filt] = get_bits1(gb);
1246 coef_compress = get_bits1(gb);
1247 coef_len = coef_res + 3 - coef_compress;
1248 tmp2_idx = 2 * coef_compress + coef_res;
1250 for (i = 0; i < tns->order[w][filt]; i++)
1251 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1260 * Decode Mid/Side data; reference: table 4.54.
1262 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1263 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1264 * [3] reserved for scalable AAC
1266 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1270 if (ms_present == 1) {
1271 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1272 cpe->ms_mask[idx] = get_bits1(gb);
1273 } else if (ms_present == 2) {
1274 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1279 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1283 *dst++ = v[idx & 15] * s;
1284 *dst++ = v[idx>>4 & 15] * s;
1290 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1294 *dst++ = v[idx & 3] * s;
1295 *dst++ = v[idx>>2 & 3] * s;
1296 *dst++ = v[idx>>4 & 3] * s;
1297 *dst++ = v[idx>>6 & 3] * s;
1303 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1304 unsigned sign, const float *scale)
1306 union av_intfloat32 s0, s1;
1308 s0.f = s1.f = *scale;
1309 s0.i ^= sign >> 1 << 31;
1312 *dst++ = v[idx & 15] * s0.f;
1313 *dst++ = v[idx>>4 & 15] * s1.f;
1320 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1321 unsigned sign, const float *scale)
1323 unsigned nz = idx >> 12;
1324 union av_intfloat32 s = { .f = *scale };
1325 union av_intfloat32 t;
1327 t.i = s.i ^ (sign & 1U<<31);
1328 *dst++ = v[idx & 3] * t.f;
1330 sign <<= nz & 1; nz >>= 1;
1331 t.i = s.i ^ (sign & 1U<<31);
1332 *dst++ = v[idx>>2 & 3] * t.f;
1334 sign <<= nz & 1; nz >>= 1;
1335 t.i = s.i ^ (sign & 1U<<31);
1336 *dst++ = v[idx>>4 & 3] * t.f;
1339 t.i = s.i ^ (sign & 1U<<31);
1340 *dst++ = v[idx>>6 & 3] * t.f;
1347 * Decode spectral data; reference: table 4.50.
1348 * Dequantize and scale spectral data; reference: 4.6.3.3.
1350 * @param coef array of dequantized, scaled spectral data
1351 * @param sf array of scalefactors or intensity stereo positions
1352 * @param pulse_present set if pulses are present
1353 * @param pulse pointer to pulse data struct
1354 * @param band_type array of the used band type
1356 * @return Returns error status. 0 - OK, !0 - error
1358 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1359 GetBitContext *gb, const float sf[120],
1360 int pulse_present, const Pulse *pulse,
1361 const IndividualChannelStream *ics,
1362 enum BandType band_type[120])
1364 int i, k, g, idx = 0;
1365 const int c = 1024 / ics->num_windows;
1366 const uint16_t *offsets = ics->swb_offset;
1367 float *coef_base = coef;
1369 for (g = 0; g < ics->num_windows; g++)
1370 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1372 for (g = 0; g < ics->num_window_groups; g++) {
1373 unsigned g_len = ics->group_len[g];
1375 for (i = 0; i < ics->max_sfb; i++, idx++) {
1376 const unsigned cbt_m1 = band_type[idx] - 1;
1377 float *cfo = coef + offsets[i];
1378 int off_len = offsets[i + 1] - offsets[i];
1381 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1382 for (group = 0; group < g_len; group++, cfo+=128) {
1383 memset(cfo, 0, off_len * sizeof(float));
1385 } else if (cbt_m1 == NOISE_BT - 1) {
1386 for (group = 0; group < g_len; group++, cfo+=128) {
1390 for (k = 0; k < off_len; k++) {
1391 ac->random_state = lcg_random(ac->random_state);
1392 cfo[k] = ac->random_state;
1395 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1396 scale = sf[idx] / sqrtf(band_energy);
1397 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1400 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1401 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1402 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1403 OPEN_READER(re, gb);
1405 switch (cbt_m1 >> 1) {
1407 for (group = 0; group < g_len; group++, cfo+=128) {
1415 UPDATE_CACHE(re, gb);
1416 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1417 cb_idx = cb_vector_idx[code];
1418 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1424 for (group = 0; group < g_len; group++, cfo+=128) {
1434 UPDATE_CACHE(re, gb);
1435 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1436 cb_idx = cb_vector_idx[code];
1437 nnz = cb_idx >> 8 & 15;
1438 bits = nnz ? GET_CACHE(re, gb) : 0;
1439 LAST_SKIP_BITS(re, gb, nnz);
1440 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1446 for (group = 0; group < g_len; group++, cfo+=128) {
1454 UPDATE_CACHE(re, gb);
1455 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1456 cb_idx = cb_vector_idx[code];
1457 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1464 for (group = 0; group < g_len; group++, cfo+=128) {
1474 UPDATE_CACHE(re, gb);
1475 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1476 cb_idx = cb_vector_idx[code];
1477 nnz = cb_idx >> 8 & 15;
1478 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1479 LAST_SKIP_BITS(re, gb, nnz);
1480 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1486 for (group = 0; group < g_len; group++, cfo+=128) {
1488 uint32_t *icf = (uint32_t *) cf;
1498 UPDATE_CACHE(re, gb);
1499 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1507 cb_idx = cb_vector_idx[code];
1510 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1511 LAST_SKIP_BITS(re, gb, nnz);
1513 for (j = 0; j < 2; j++) {
1517 /* The total length of escape_sequence must be < 22 bits according
1518 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1519 UPDATE_CACHE(re, gb);
1520 b = GET_CACHE(re, gb);
1521 b = 31 - av_log2(~b);
1524 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1528 SKIP_BITS(re, gb, b + 1);
1530 n = (1 << b) + SHOW_UBITS(re, gb, b);
1531 LAST_SKIP_BITS(re, gb, b);
1532 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1535 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1536 *icf++ = (bits & 1U<<31) | v;
1543 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1547 CLOSE_READER(re, gb);
1553 if (pulse_present) {
1555 for (i = 0; i < pulse->num_pulse; i++) {
1556 float co = coef_base[ pulse->pos[i] ];
1557 while (offsets[idx + 1] <= pulse->pos[i])
1559 if (band_type[idx] != NOISE_BT && sf[idx]) {
1560 float ico = -pulse->amp[i];
1563 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1565 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1572 static av_always_inline float flt16_round(float pf)
1574 union av_intfloat32 tmp;
1576 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1580 static av_always_inline float flt16_even(float pf)
1582 union av_intfloat32 tmp;
1584 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1588 static av_always_inline float flt16_trunc(float pf)
1590 union av_intfloat32 pun;
1592 pun.i &= 0xFFFF0000U;
1596 static av_always_inline void predict(PredictorState *ps, float *coef,
1599 const float a = 0.953125; // 61.0 / 64
1600 const float alpha = 0.90625; // 29.0 / 32
1604 float r0 = ps->r0, r1 = ps->r1;
1605 float cor0 = ps->cor0, cor1 = ps->cor1;
1606 float var0 = ps->var0, var1 = ps->var1;
1608 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1609 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1611 pv = flt16_round(k1 * r0 + k2 * r1);
1618 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1619 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1620 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1621 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1623 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1624 ps->r0 = flt16_trunc(a * e0);
1628 * Apply AAC-Main style frequency domain prediction.
1630 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1634 if (!sce->ics.predictor_initialized) {
1635 reset_all_predictors(sce->predictor_state);
1636 sce->ics.predictor_initialized = 1;
1639 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1640 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]; sfb++) {
1641 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1642 predict(&sce->predictor_state[k], &sce->coeffs[k],
1643 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1646 if (sce->ics.predictor_reset_group)
1647 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1649 reset_all_predictors(sce->predictor_state);
1653 * Decode an individual_channel_stream payload; reference: table 4.44.
1655 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1656 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1658 * @return Returns error status. 0 - OK, !0 - error
1660 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1661 GetBitContext *gb, int common_window, int scale_flag)
1664 TemporalNoiseShaping *tns = &sce->tns;
1665 IndividualChannelStream *ics = &sce->ics;
1666 float *out = sce->coeffs;
1667 int global_gain, pulse_present = 0;
1669 /* This assignment is to silence a GCC warning about the variable being used
1670 * uninitialized when in fact it always is.
1672 pulse.num_pulse = 0;
1674 global_gain = get_bits(gb, 8);
1676 if (!common_window && !scale_flag) {
1677 if (decode_ics_info(ac, ics, gb) < 0)
1678 return AVERROR_INVALIDDATA;
1681 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1683 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1688 if ((pulse_present = get_bits1(gb))) {
1689 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1690 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1693 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1694 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1698 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1700 if (get_bits1(gb)) {
1701 av_log_missing_feature(ac->avctx, "SSR", 1);
1702 return AVERROR_PATCHWELCOME;
1706 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1709 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1710 apply_prediction(ac, sce);
1716 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1718 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1720 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1721 float *ch0 = cpe->ch[0].coeffs;
1722 float *ch1 = cpe->ch[1].coeffs;
1723 int g, i, group, idx = 0;
1724 const uint16_t *offsets = ics->swb_offset;
1725 for (g = 0; g < ics->num_window_groups; g++) {
1726 for (i = 0; i < ics->max_sfb; i++, idx++) {
1727 if (cpe->ms_mask[idx] &&
1728 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1729 for (group = 0; group < ics->group_len[g]; group++) {
1730 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1731 ch1 + group * 128 + offsets[i],
1732 offsets[i+1] - offsets[i]);
1736 ch0 += ics->group_len[g] * 128;
1737 ch1 += ics->group_len[g] * 128;
1742 * intensity stereo decoding; reference: 4.6.8.2.3
1744 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1745 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1746 * [3] reserved for scalable AAC
1748 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1750 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1751 SingleChannelElement *sce1 = &cpe->ch[1];
1752 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1753 const uint16_t *offsets = ics->swb_offset;
1754 int g, group, i, idx = 0;
1757 for (g = 0; g < ics->num_window_groups; g++) {
1758 for (i = 0; i < ics->max_sfb;) {
1759 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1760 const int bt_run_end = sce1->band_type_run_end[idx];
1761 for (; i < bt_run_end; i++, idx++) {
1762 c = -1 + 2 * (sce1->band_type[idx] - 14);
1764 c *= 1 - 2 * cpe->ms_mask[idx];
1765 scale = c * sce1->sf[idx];
1766 for (group = 0; group < ics->group_len[g]; group++)
1767 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1768 coef0 + group * 128 + offsets[i],
1770 offsets[i + 1] - offsets[i]);
1773 int bt_run_end = sce1->band_type_run_end[idx];
1774 idx += bt_run_end - i;
1778 coef0 += ics->group_len[g] * 128;
1779 coef1 += ics->group_len[g] * 128;
1784 * Decode a channel_pair_element; reference: table 4.4.
1786 * @return Returns error status. 0 - OK, !0 - error
1788 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1790 int i, ret, common_window, ms_present = 0;
1792 common_window = get_bits1(gb);
1793 if (common_window) {
1794 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1795 return AVERROR_INVALIDDATA;
1796 i = cpe->ch[1].ics.use_kb_window[0];
1797 cpe->ch[1].ics = cpe->ch[0].ics;
1798 cpe->ch[1].ics.use_kb_window[1] = i;
1799 if (cpe->ch[1].ics.predictor_present && (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1800 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1801 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1802 ms_present = get_bits(gb, 2);
1803 if (ms_present == 3) {
1804 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1806 } else if (ms_present)
1807 decode_mid_side_stereo(cpe, gb, ms_present);
1809 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1811 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1814 if (common_window) {
1816 apply_mid_side_stereo(ac, cpe);
1817 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1818 apply_prediction(ac, &cpe->ch[0]);
1819 apply_prediction(ac, &cpe->ch[1]);
1823 apply_intensity_stereo(ac, cpe, ms_present);
1827 static const float cce_scale[] = {
1828 1.09050773266525765921, //2^(1/8)
1829 1.18920711500272106672, //2^(1/4)
1835 * Decode coupling_channel_element; reference: table 4.8.
1837 * @return Returns error status. 0 - OK, !0 - error
1839 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1845 SingleChannelElement *sce = &che->ch[0];
1846 ChannelCoupling *coup = &che->coup;
1848 coup->coupling_point = 2 * get_bits1(gb);
1849 coup->num_coupled = get_bits(gb, 3);
1850 for (c = 0; c <= coup->num_coupled; c++) {
1852 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1853 coup->id_select[c] = get_bits(gb, 4);
1854 if (coup->type[c] == TYPE_CPE) {
1855 coup->ch_select[c] = get_bits(gb, 2);
1856 if (coup->ch_select[c] == 3)
1859 coup->ch_select[c] = 2;
1861 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1863 sign = get_bits(gb, 1);
1864 scale = cce_scale[get_bits(gb, 2)];
1866 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1869 for (c = 0; c < num_gain; c++) {
1873 float gain_cache = 1.;
1875 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1876 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1877 gain_cache = powf(scale, -gain);
1879 if (coup->coupling_point == AFTER_IMDCT) {
1880 coup->gain[c][0] = gain_cache;
1882 for (g = 0; g < sce->ics.num_window_groups; g++) {
1883 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1884 if (sce->band_type[idx] != ZERO_BT) {
1886 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1894 gain_cache = powf(scale, -t) * s;
1897 coup->gain[c][idx] = gain_cache;
1907 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1909 * @return Returns number of bytes consumed.
1911 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1915 int num_excl_chan = 0;
1918 for (i = 0; i < 7; i++)
1919 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1920 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1922 return num_excl_chan / 7;
1926 * Decode dynamic range information; reference: table 4.52.
1928 * @return Returns number of bytes consumed.
1930 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1934 int drc_num_bands = 1;
1937 /* pce_tag_present? */
1938 if (get_bits1(gb)) {
1939 che_drc->pce_instance_tag = get_bits(gb, 4);
1940 skip_bits(gb, 4); // tag_reserved_bits
1944 /* excluded_chns_present? */
1945 if (get_bits1(gb)) {
1946 n += decode_drc_channel_exclusions(che_drc, gb);
1949 /* drc_bands_present? */
1950 if (get_bits1(gb)) {
1951 che_drc->band_incr = get_bits(gb, 4);
1952 che_drc->interpolation_scheme = get_bits(gb, 4);
1954 drc_num_bands += che_drc->band_incr;
1955 for (i = 0; i < drc_num_bands; i++) {
1956 che_drc->band_top[i] = get_bits(gb, 8);
1961 /* prog_ref_level_present? */
1962 if (get_bits1(gb)) {
1963 che_drc->prog_ref_level = get_bits(gb, 7);
1964 skip_bits1(gb); // prog_ref_level_reserved_bits
1968 for (i = 0; i < drc_num_bands; i++) {
1969 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1970 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1977 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
1979 int i, major, minor;
1984 get_bits(gb, 13); len -= 13;
1986 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
1987 buf[i] = get_bits(gb, 8);
1990 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
1991 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
1993 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
1994 ac->avctx->internal->skip_samples = 1024;
1998 skip_bits_long(gb, len);
2004 * Decode extension data (incomplete); reference: table 4.51.
2006 * @param cnt length of TYPE_FIL syntactic element in bytes
2008 * @return Returns number of bytes consumed
2010 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2011 ChannelElement *che, enum RawDataBlockType elem_type)
2015 switch (get_bits(gb, 4)) { // extension type
2016 case EXT_SBR_DATA_CRC:
2020 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2022 } else if (!ac->oc[1].m4ac.sbr) {
2023 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2024 skip_bits_long(gb, 8 * cnt - 4);
2026 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2027 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2028 skip_bits_long(gb, 8 * cnt - 4);
2030 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2031 ac->oc[1].m4ac.sbr = 1;
2032 ac->oc[1].m4ac.ps = 1;
2033 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2034 ac->oc[1].status, 1);
2036 ac->oc[1].m4ac.sbr = 1;
2038 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2040 case EXT_DYNAMIC_RANGE:
2041 res = decode_dynamic_range(&ac->che_drc, gb);
2044 decode_fill(ac, gb, 8 * cnt - 4);
2047 case EXT_DATA_ELEMENT:
2049 skip_bits_long(gb, 8 * cnt - 4);
2056 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2058 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2059 * @param coef spectral coefficients
2061 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2062 IndividualChannelStream *ics, int decode)
2064 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2066 int bottom, top, order, start, end, size, inc;
2067 float lpc[TNS_MAX_ORDER];
2068 float tmp[TNS_MAX_ORDER+1];
2070 for (w = 0; w < ics->num_windows; w++) {
2071 bottom = ics->num_swb;
2072 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2074 bottom = FFMAX(0, top - tns->length[w][filt]);
2075 order = tns->order[w][filt];
2080 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2082 start = ics->swb_offset[FFMIN(bottom, mmm)];
2083 end = ics->swb_offset[FFMIN( top, mmm)];
2084 if ((size = end - start) <= 0)
2086 if (tns->direction[w][filt]) {
2096 for (m = 0; m < size; m++, start += inc)
2097 for (i = 1; i <= FFMIN(m, order); i++)
2098 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2101 for (m = 0; m < size; m++, start += inc) {
2102 tmp[0] = coef[start];
2103 for (i = 1; i <= FFMIN(m, order); i++)
2104 coef[start] += tmp[i] * lpc[i - 1];
2105 for (i = order; i > 0; i--)
2106 tmp[i] = tmp[i - 1];
2114 * Apply windowing and MDCT to obtain the spectral
2115 * coefficient from the predicted sample by LTP.
2117 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2118 float *in, IndividualChannelStream *ics)
2120 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2121 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2122 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2123 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2125 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2126 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2128 memset(in, 0, 448 * sizeof(float));
2129 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2131 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2132 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2134 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2135 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2137 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2141 * Apply the long term prediction
2143 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2145 const LongTermPrediction *ltp = &sce->ics.ltp;
2146 const uint16_t *offsets = sce->ics.swb_offset;
2149 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2150 float *predTime = sce->ret;
2151 float *predFreq = ac->buf_mdct;
2152 int16_t num_samples = 2048;
2154 if (ltp->lag < 1024)
2155 num_samples = ltp->lag + 1024;
2156 for (i = 0; i < num_samples; i++)
2157 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2158 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2160 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2162 if (sce->tns.present)
2163 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2165 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2167 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2168 sce->coeffs[i] += predFreq[i];
2173 * Update the LTP buffer for next frame
2175 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2177 IndividualChannelStream *ics = &sce->ics;
2178 float *saved = sce->saved;
2179 float *saved_ltp = sce->coeffs;
2180 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2181 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2184 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2185 memcpy(saved_ltp, saved, 512 * sizeof(float));
2186 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2187 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2188 for (i = 0; i < 64; i++)
2189 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2190 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2191 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2192 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2193 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2194 for (i = 0; i < 64; i++)
2195 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2196 } else { // LONG_STOP or ONLY_LONG
2197 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2198 for (i = 0; i < 512; i++)
2199 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2202 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2203 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2204 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2208 * Conduct IMDCT and windowing.
2210 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2212 IndividualChannelStream *ics = &sce->ics;
2213 float *in = sce->coeffs;
2214 float *out = sce->ret;
2215 float *saved = sce->saved;
2216 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2217 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2218 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2219 float *buf = ac->buf_mdct;
2220 float *temp = ac->temp;
2224 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2225 for (i = 0; i < 1024; i += 128)
2226 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2228 ac->mdct.imdct_half(&ac->mdct, buf, in);
2230 /* window overlapping
2231 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2232 * and long to short transitions are considered to be short to short
2233 * transitions. This leaves just two cases (long to long and short to short)
2234 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2236 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2237 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2238 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2240 memcpy( out, saved, 448 * sizeof(float));
2242 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2243 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2244 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2245 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2246 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2247 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2248 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2250 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2251 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2256 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2257 memcpy( saved, temp + 64, 64 * sizeof(float));
2258 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2259 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2260 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2261 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2262 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2263 memcpy( saved, buf + 512, 448 * sizeof(float));
2264 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2265 } else { // LONG_STOP or ONLY_LONG
2266 memcpy( saved, buf + 512, 512 * sizeof(float));
2271 * Apply dependent channel coupling (applied before IMDCT).
2273 * @param index index into coupling gain array
2275 static void apply_dependent_coupling(AACContext *ac,
2276 SingleChannelElement *target,
2277 ChannelElement *cce, int index)
2279 IndividualChannelStream *ics = &cce->ch[0].ics;
2280 const uint16_t *offsets = ics->swb_offset;
2281 float *dest = target->coeffs;
2282 const float *src = cce->ch[0].coeffs;
2283 int g, i, group, k, idx = 0;
2284 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2285 av_log(ac->avctx, AV_LOG_ERROR,
2286 "Dependent coupling is not supported together with LTP\n");
2289 for (g = 0; g < ics->num_window_groups; g++) {
2290 for (i = 0; i < ics->max_sfb; i++, idx++) {
2291 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2292 const float gain = cce->coup.gain[index][idx];
2293 for (group = 0; group < ics->group_len[g]; group++) {
2294 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2296 dest[group * 128 + k] += gain * src[group * 128 + k];
2301 dest += ics->group_len[g] * 128;
2302 src += ics->group_len[g] * 128;
2307 * Apply independent channel coupling (applied after IMDCT).
2309 * @param index index into coupling gain array
2311 static void apply_independent_coupling(AACContext *ac,
2312 SingleChannelElement *target,
2313 ChannelElement *cce, int index)
2316 const float gain = cce->coup.gain[index][0];
2317 const float *src = cce->ch[0].ret;
2318 float *dest = target->ret;
2319 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2321 for (i = 0; i < len; i++)
2322 dest[i] += gain * src[i];
2326 * channel coupling transformation interface
2328 * @param apply_coupling_method pointer to (in)dependent coupling function
2330 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2331 enum RawDataBlockType type, int elem_id,
2332 enum CouplingPoint coupling_point,
2333 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2337 for (i = 0; i < MAX_ELEM_ID; i++) {
2338 ChannelElement *cce = ac->che[TYPE_CCE][i];
2341 if (cce && cce->coup.coupling_point == coupling_point) {
2342 ChannelCoupling *coup = &cce->coup;
2344 for (c = 0; c <= coup->num_coupled; c++) {
2345 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2346 if (coup->ch_select[c] != 1) {
2347 apply_coupling_method(ac, &cc->ch[0], cce, index);
2348 if (coup->ch_select[c] != 0)
2351 if (coup->ch_select[c] != 2)
2352 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2354 index += 1 + (coup->ch_select[c] == 3);
2361 * Convert spectral data to float samples, applying all supported tools as appropriate.
2363 static void spectral_to_sample(AACContext *ac)
2366 for (type = 3; type >= 0; type--) {
2367 for (i = 0; i < MAX_ELEM_ID; i++) {
2368 ChannelElement *che = ac->che[type][i];
2370 if (type <= TYPE_CPE)
2371 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2372 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2373 if (che->ch[0].ics.predictor_present) {
2374 if (che->ch[0].ics.ltp.present)
2375 apply_ltp(ac, &che->ch[0]);
2376 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2377 apply_ltp(ac, &che->ch[1]);
2380 if (che->ch[0].tns.present)
2381 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2382 if (che->ch[1].tns.present)
2383 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2384 if (type <= TYPE_CPE)
2385 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2386 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2387 imdct_and_windowing(ac, &che->ch[0]);
2388 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2389 update_ltp(ac, &che->ch[0]);
2390 if (type == TYPE_CPE) {
2391 imdct_and_windowing(ac, &che->ch[1]);
2392 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2393 update_ltp(ac, &che->ch[1]);
2395 if (ac->oc[1].m4ac.sbr > 0) {
2396 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2399 if (type <= TYPE_CCE)
2400 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2406 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2409 AACADTSHeaderInfo hdr_info;
2410 uint8_t layout_map[MAX_ELEM_ID*4][3];
2411 int layout_map_tags;
2413 size = avpriv_aac_parse_header(gb, &hdr_info);
2415 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2416 // This is 2 for "VLB " audio in NSV files.
2417 // See samples/nsv/vlb_audio.
2418 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame", 0);
2419 ac->warned_num_aac_frames = 1;
2421 push_output_configuration(ac);
2422 if (hdr_info.chan_config) {
2423 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2424 if (set_default_channel_config(ac->avctx, layout_map,
2425 &layout_map_tags, hdr_info.chan_config))
2427 if (output_configure(ac, layout_map, layout_map_tags,
2428 FFMAX(ac->oc[1].status, OC_TRIAL_FRAME), 0))
2431 ac->oc[1].m4ac.chan_config = 0;
2433 * dual mono frames in Japanese DTV can have chan_config 0
2434 * WITHOUT specifying PCE.
2435 * thus, set dual mono as default.
2438 if (ac->enable_jp_dmono && ac->oc[0].status == OC_NONE) {
2439 layout_map_tags = 2;
2440 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2441 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2442 layout_map[0][1] = 0;
2443 layout_map[1][1] = 1;
2444 if (output_configure(ac, layout_map, layout_map_tags,
2450 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2451 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2452 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2453 if (ac->oc[0].status != OC_LOCKED ||
2454 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2455 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2456 ac->oc[1].m4ac.sbr = -1;
2457 ac->oc[1].m4ac.ps = -1;
2459 if (!hdr_info.crc_absent)
2465 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2466 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2468 AACContext *ac = avctx->priv_data;
2469 ChannelElement *che = NULL, *che_prev = NULL;
2470 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2472 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2473 int is_dmono, sce_count = 0;
2476 if (show_bits(gb, 12) == 0xfff) {
2477 if (parse_adts_frame_header(ac, gb) < 0) {
2478 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2482 if (ac->oc[1].m4ac.sampling_index > 12) {
2483 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2489 if (frame_configure_elements(avctx) < 0) {
2494 ac->tags_mapped = 0;
2496 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2497 elem_id = get_bits(gb, 4);
2499 if (elem_type < TYPE_DSE) {
2500 if (!(che=get_che(ac, elem_type, elem_id))) {
2501 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2502 elem_type, elem_id);
2509 switch (elem_type) {
2512 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2518 err = decode_cpe(ac, gb, che);
2523 err = decode_cce(ac, gb, che);
2527 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2532 err = skip_data_stream_element(ac, gb);
2536 uint8_t layout_map[MAX_ELEM_ID*4][3];
2538 push_output_configuration(ac);
2539 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2545 av_log(avctx, AV_LOG_ERROR,
2546 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2547 pop_output_configuration(ac);
2549 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2551 ac->oc[1].m4ac.chan_config = 0;
2559 elem_id += get_bits(gb, 8) - 1;
2560 if (get_bits_left(gb) < 8 * elem_id) {
2561 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2566 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2567 err = 0; /* FIXME */
2571 err = -1; /* should not happen, but keeps compiler happy */
2576 elem_type_prev = elem_type;
2581 if (get_bits_left(gb) < 3) {
2582 av_log(avctx, AV_LOG_ERROR, overread_err);
2588 spectral_to_sample(ac);
2590 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2591 samples <<= multiplier;
2593 /* for dual-mono audio (SCE + SCE) */
2594 is_dmono = ac->enable_jp_dmono && sce_count == 2 &&
2595 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2598 if (ac->dmono_mode == 0) {
2599 tmp = ac->output_data[1];
2600 ac->output_data[1] = ac->output_data[0];
2601 } else if (ac->dmono_mode == 1) {
2602 tmp = ac->output_data[0];
2603 ac->output_data[0] = ac->output_data[1];
2608 ac->frame.nb_samples = samples;
2609 *(AVFrame *)data = ac->frame;
2611 *got_frame_ptr = !!samples;
2614 if (ac->dmono_mode == 0)
2615 ac->output_data[1] = tmp;
2616 else if (ac->dmono_mode == 1)
2617 ac->output_data[0] = tmp;
2620 if (ac->oc[1].status && audio_found) {
2621 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2622 avctx->frame_size = samples;
2623 ac->oc[1].status = OC_LOCKED;
2628 uint32_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
2629 if (side && side_size>=4)
2630 AV_WL32(side, 2*AV_RL32(side));
2634 pop_output_configuration(ac);
2638 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2639 int *got_frame_ptr, AVPacket *avpkt)
2641 AACContext *ac = avctx->priv_data;
2642 const uint8_t *buf = avpkt->data;
2643 int buf_size = avpkt->size;
2648 int new_extradata_size;
2649 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2650 AV_PKT_DATA_NEW_EXTRADATA,
2651 &new_extradata_size);
2652 int jp_dualmono_size;
2653 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
2654 AV_PKT_DATA_JP_DUALMONO,
2657 if (new_extradata && 0) {
2658 av_free(avctx->extradata);
2659 avctx->extradata = av_mallocz(new_extradata_size +
2660 FF_INPUT_BUFFER_PADDING_SIZE);
2661 if (!avctx->extradata)
2662 return AVERROR(ENOMEM);
2663 avctx->extradata_size = new_extradata_size;
2664 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2665 push_output_configuration(ac);
2666 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2668 avctx->extradata_size*8, 1) < 0) {
2669 pop_output_configuration(ac);
2670 return AVERROR_INVALIDDATA;
2674 ac->enable_jp_dmono = !!jp_dualmono;
2676 if (jp_dualmono && jp_dualmono_size > 0)
2677 ac->dmono_mode = *jp_dualmono;
2679 init_get_bits(&gb, buf, buf_size * 8);
2681 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt)) < 0)
2684 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2685 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2686 if (buf[buf_offset])
2689 return buf_size > buf_offset ? buf_consumed : buf_size;
2692 static av_cold int aac_decode_close(AVCodecContext *avctx)
2694 AACContext *ac = avctx->priv_data;
2697 for (i = 0; i < MAX_ELEM_ID; i++) {
2698 for (type = 0; type < 4; type++) {
2699 if (ac->che[type][i])
2700 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2701 av_freep(&ac->che[type][i]);
2705 ff_mdct_end(&ac->mdct);
2706 ff_mdct_end(&ac->mdct_small);
2707 ff_mdct_end(&ac->mdct_ltp);
2712 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2714 struct LATMContext {
2715 AACContext aac_ctx; ///< containing AACContext
2716 int initialized; ///< initialized after a valid extradata was seen
2719 int audio_mux_version_A; ///< LATM syntax version
2720 int frame_length_type; ///< 0/1 variable/fixed frame length
2721 int frame_length; ///< frame length for fixed frame length
2724 static inline uint32_t latm_get_value(GetBitContext *b)
2726 int length = get_bits(b, 2);
2728 return get_bits_long(b, (length+1)*8);
2731 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2732 GetBitContext *gb, int asclen)
2734 AACContext *ac = &latmctx->aac_ctx;
2735 AVCodecContext *avctx = ac->avctx;
2736 MPEG4AudioConfig m4ac = { 0 };
2737 int config_start_bit = get_bits_count(gb);
2738 int sync_extension = 0;
2739 int bits_consumed, esize;
2743 asclen = FFMIN(asclen, get_bits_left(gb));
2745 asclen = get_bits_left(gb);
2747 if (config_start_bit % 8) {
2748 av_log_missing_feature(latmctx->aac_ctx.avctx,
2749 "Non-byte-aligned audio-specific config", 1);
2750 return AVERROR_PATCHWELCOME;
2753 return AVERROR_INVALIDDATA;
2754 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2755 gb->buffer + (config_start_bit / 8),
2756 asclen, sync_extension);
2758 if (bits_consumed < 0)
2759 return AVERROR_INVALIDDATA;
2761 if (!latmctx->initialized ||
2762 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
2763 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
2765 if(latmctx->initialized) {
2766 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2768 av_log(avctx, AV_LOG_INFO, "initializing latmctx\n");
2770 latmctx->initialized = 0;
2772 esize = (bits_consumed+7) / 8;
2774 if (avctx->extradata_size < esize) {
2775 av_free(avctx->extradata);
2776 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2777 if (!avctx->extradata)
2778 return AVERROR(ENOMEM);
2781 avctx->extradata_size = esize;
2782 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2783 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2785 skip_bits_long(gb, bits_consumed);
2787 return bits_consumed;
2790 static int read_stream_mux_config(struct LATMContext *latmctx,
2793 int ret, audio_mux_version = get_bits(gb, 1);
2795 latmctx->audio_mux_version_A = 0;
2796 if (audio_mux_version)
2797 latmctx->audio_mux_version_A = get_bits(gb, 1);
2799 if (!latmctx->audio_mux_version_A) {
2801 if (audio_mux_version)
2802 latm_get_value(gb); // taraFullness
2804 skip_bits(gb, 1); // allStreamSameTimeFraming
2805 skip_bits(gb, 6); // numSubFrames
2807 if (get_bits(gb, 4)) { // numPrograms
2808 av_log_missing_feature(latmctx->aac_ctx.avctx,
2809 "Multiple programs", 1);
2810 return AVERROR_PATCHWELCOME;
2813 // for each program (which there is only one in DVB)
2815 // for each layer (which there is only one in DVB)
2816 if (get_bits(gb, 3)) { // numLayer
2817 av_log_missing_feature(latmctx->aac_ctx.avctx,
2818 "Multiple layers", 1);
2819 return AVERROR_PATCHWELCOME;
2822 // for all but first stream: use_same_config = get_bits(gb, 1);
2823 if (!audio_mux_version) {
2824 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2827 int ascLen = latm_get_value(gb);
2828 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2831 skip_bits_long(gb, ascLen);
2834 latmctx->frame_length_type = get_bits(gb, 3);
2835 switch (latmctx->frame_length_type) {
2837 skip_bits(gb, 8); // latmBufferFullness
2840 latmctx->frame_length = get_bits(gb, 9);
2845 skip_bits(gb, 6); // CELP frame length table index
2849 skip_bits(gb, 1); // HVXC frame length table index
2853 if (get_bits(gb, 1)) { // other data
2854 if (audio_mux_version) {
2855 latm_get_value(gb); // other_data_bits
2859 esc = get_bits(gb, 1);
2865 if (get_bits(gb, 1)) // crc present
2866 skip_bits(gb, 8); // config_crc
2872 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2876 if (ctx->frame_length_type == 0) {
2877 int mux_slot_length = 0;
2879 tmp = get_bits(gb, 8);
2880 mux_slot_length += tmp;
2881 } while (tmp == 255);
2882 return mux_slot_length;
2883 } else if (ctx->frame_length_type == 1) {
2884 return ctx->frame_length;
2885 } else if (ctx->frame_length_type == 3 ||
2886 ctx->frame_length_type == 5 ||
2887 ctx->frame_length_type == 7) {
2888 skip_bits(gb, 2); // mux_slot_length_coded
2893 static int read_audio_mux_element(struct LATMContext *latmctx,
2897 uint8_t use_same_mux = get_bits(gb, 1);
2898 if (!use_same_mux) {
2899 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2901 } else if (!latmctx->aac_ctx.avctx->extradata) {
2902 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2903 "no decoder config found\n");
2904 return AVERROR(EAGAIN);
2906 if (latmctx->audio_mux_version_A == 0) {
2907 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2908 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2909 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2910 return AVERROR_INVALIDDATA;
2911 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2912 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2913 "frame length mismatch %d << %d\n",
2914 mux_slot_length_bytes * 8, get_bits_left(gb));
2915 return AVERROR_INVALIDDATA;
2922 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2923 int *got_frame_ptr, AVPacket *avpkt)
2925 struct LATMContext *latmctx = avctx->priv_data;
2929 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2931 // check for LOAS sync word
2932 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2933 return AVERROR_INVALIDDATA;
2935 muxlength = get_bits(&gb, 13) + 3;
2936 // not enough data, the parser should have sorted this out
2937 if (muxlength > avpkt->size)
2938 return AVERROR_INVALIDDATA;
2940 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2943 if (!latmctx->initialized) {
2944 if (!avctx->extradata) {
2948 push_output_configuration(&latmctx->aac_ctx);
2949 if ((err = decode_audio_specific_config(
2950 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
2951 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
2952 pop_output_configuration(&latmctx->aac_ctx);
2955 latmctx->initialized = 1;
2959 if (show_bits(&gb, 12) == 0xfff) {
2960 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2961 "ADTS header detected, probably as result of configuration "
2963 return AVERROR_INVALIDDATA;
2966 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
2972 static av_cold int latm_decode_init(AVCodecContext *avctx)
2974 struct LATMContext *latmctx = avctx->priv_data;
2975 int ret = aac_decode_init(avctx);
2977 if (avctx->extradata_size > 0)
2978 latmctx->initialized = !ret;
2984 AVCodec ff_aac_decoder = {
2986 .type = AVMEDIA_TYPE_AUDIO,
2987 .id = AV_CODEC_ID_AAC,
2988 .priv_data_size = sizeof(AACContext),
2989 .init = aac_decode_init,
2990 .close = aac_decode_close,
2991 .decode = aac_decode_frame,
2992 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
2993 .sample_fmts = (const enum AVSampleFormat[]) {
2994 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
2996 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2997 .channel_layouts = aac_channel_layout,
3002 Note: This decoder filter is intended to decode LATM streams transferred
3003 in MPEG transport streams which only contain one program.
3004 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3006 AVCodec ff_aac_latm_decoder = {
3008 .type = AVMEDIA_TYPE_AUDIO,
3009 .id = AV_CODEC_ID_AAC_LATM,
3010 .priv_data_size = sizeof(struct LATMContext),
3011 .init = latm_decode_init,
3012 .close = aac_decode_close,
3013 .decode = latm_decode_frame,
3014 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3015 .sample_fmts = (const enum AVSampleFormat[]) {
3016 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3018 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3019 .channel_layouts = aac_channel_layout,