3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5 * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
11 * This file is part of Libav.
13 * Libav is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU Lesser General Public
15 * License as published by the Free Software Foundation; either
16 * version 2.1 of the License, or (at your option) any later version.
18 * Libav is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * Lesser General Public License for more details.
23 * You should have received a copy of the GNU Lesser General Public
24 * License along with Libav; if not, write to the Free Software
25 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
31 * @author Oded Shimon ( ods15 ods15 dyndns org )
32 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
39 * N (code in SoC repo) gain control
41 * Y window shapes - standard
42 * N window shapes - Low Delay
43 * Y filterbank - standard
44 * N (code in SoC repo) filterbank - Scalable Sample Rate
45 * Y Temporal Noise Shaping
46 * Y Long Term Prediction
49 * Y frequency domain prediction
50 * Y Perceptual Noise Substitution
52 * N Scalable Inverse AAC Quantization
53 * N Frequency Selective Switch
55 * Y quantization & coding - AAC
56 * N quantization & coding - TwinVQ
57 * N quantization & coding - BSAC
58 * N AAC Error Resilience tools
59 * N Error Resilience payload syntax
60 * N Error Protection tool
62 * N Silence Compression
65 * N Structured Audio tools
66 * N Structured Audio Sample Bank Format
68 * N Harmonic and Individual Lines plus Noise
69 * N Text-To-Speech Interface
70 * Y Spectral Band Replication
71 * Y (not in this code) Layer-1
72 * Y (not in this code) Layer-2
73 * Y (not in this code) Layer-3
74 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
76 * N Direct Stream Transfer
78 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
79 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
83 #include "libavutil/float_dsp.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
101 #include "libavutil/intfloat.h"
110 # include "arm/aac.h"
113 static VLC vlc_scalefactors;
114 static VLC vlc_spectral[11];
116 static const char overread_err[] = "Input buffer exhausted before END element found\n";
118 static int count_channels(uint8_t (*layout)[3], int tags)
121 for (i = 0; i < tags; i++) {
122 int syn_ele = layout[i][0];
123 int pos = layout[i][2];
124 sum += (1 + (syn_ele == TYPE_CPE)) *
125 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
131 * Check for the channel element in the current channel position configuration.
132 * If it exists, make sure the appropriate element is allocated and map the
133 * channel order to match the internal Libav channel layout.
135 * @param che_pos current channel position configuration
136 * @param type channel element type
137 * @param id channel element id
138 * @param channels count of the number of channels in the configuration
140 * @return Returns error status. 0 - OK, !0 - error
142 static av_cold int che_configure(AACContext *ac,
143 enum ChannelPosition che_pos,
144 int type, int id, int *channels)
147 if (!ac->che[type][id]) {
148 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
149 return AVERROR(ENOMEM);
150 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
152 if (type != TYPE_CCE) {
153 if (*channels >= MAX_CHANNELS - 2)
154 return AVERROR_INVALIDDATA;
155 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
156 if (type == TYPE_CPE ||
157 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
158 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
162 if (ac->che[type][id])
163 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
164 av_freep(&ac->che[type][id]);
169 static int frame_configure_elements(AVCodecContext *avctx)
171 AACContext *ac = avctx->priv_data;
172 int type, id, ch, ret;
174 /* set channel pointers to internal buffers by default */
175 for (type = 0; type < 4; type++) {
176 for (id = 0; id < MAX_ELEM_ID; id++) {
177 ChannelElement *che = ac->che[type][id];
179 che->ch[0].ret = che->ch[0].ret_buf;
180 che->ch[1].ret = che->ch[1].ret_buf;
185 /* get output buffer */
186 av_frame_unref(ac->frame);
187 ac->frame->nb_samples = 2048;
188 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 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,
217 .elem_id = layout_map[offset][1],
222 e2c_vec[offset] = (struct elem_to_channel) {
225 .elem_id = layout_map[offset][1],
228 e2c_vec[offset + 1] = (struct elem_to_channel) {
229 .av_position = right,
231 .elem_id = layout_map[offset + 1][1],
238 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
241 int num_pos_channels = 0;
245 for (i = *current; i < tags; i++) {
246 if (layout_map[i][2] != pos)
248 if (layout_map[i][0] == TYPE_CPE) {
250 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
256 num_pos_channels += 2;
264 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
267 return num_pos_channels;
270 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
272 int i, n, total_non_cc_elements;
273 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
274 int num_front_channels, num_side_channels, num_back_channels;
277 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
282 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
283 if (num_front_channels < 0)
286 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
287 if (num_side_channels < 0)
290 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
291 if (num_back_channels < 0)
295 if (num_front_channels & 1) {
296 e2c_vec[i] = (struct elem_to_channel) {
297 .av_position = AV_CH_FRONT_CENTER,
299 .elem_id = layout_map[i][1],
300 .aac_position = AAC_CHANNEL_FRONT
303 num_front_channels--;
305 if (num_front_channels >= 4) {
306 i += assign_pair(e2c_vec, layout_map, i,
307 AV_CH_FRONT_LEFT_OF_CENTER,
308 AV_CH_FRONT_RIGHT_OF_CENTER,
310 num_front_channels -= 2;
312 if (num_front_channels >= 2) {
313 i += assign_pair(e2c_vec, layout_map, i,
317 num_front_channels -= 2;
319 while (num_front_channels >= 2) {
320 i += assign_pair(e2c_vec, layout_map, i,
324 num_front_channels -= 2;
327 if (num_side_channels >= 2) {
328 i += assign_pair(e2c_vec, layout_map, i,
332 num_side_channels -= 2;
334 while (num_side_channels >= 2) {
335 i += assign_pair(e2c_vec, layout_map, i,
339 num_side_channels -= 2;
342 while (num_back_channels >= 4) {
343 i += assign_pair(e2c_vec, layout_map, i,
347 num_back_channels -= 2;
349 if (num_back_channels >= 2) {
350 i += assign_pair(e2c_vec, layout_map, i,
354 num_back_channels -= 2;
356 if (num_back_channels) {
357 e2c_vec[i] = (struct elem_to_channel) {
358 .av_position = AV_CH_BACK_CENTER,
360 .elem_id = layout_map[i][1],
361 .aac_position = AAC_CHANNEL_BACK
367 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
368 e2c_vec[i] = (struct elem_to_channel) {
369 .av_position = AV_CH_LOW_FREQUENCY,
371 .elem_id = layout_map[i][1],
372 .aac_position = AAC_CHANNEL_LFE
376 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
377 e2c_vec[i] = (struct elem_to_channel) {
378 .av_position = UINT64_MAX,
380 .elem_id = layout_map[i][1],
381 .aac_position = AAC_CHANNEL_LFE
386 // Must choose a stable sort
387 total_non_cc_elements = n = i;
390 for (i = 1; i < n; i++)
391 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
392 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
399 for (i = 0; i < total_non_cc_elements; i++) {
400 layout_map[i][0] = e2c_vec[i].syn_ele;
401 layout_map[i][1] = e2c_vec[i].elem_id;
402 layout_map[i][2] = e2c_vec[i].aac_position;
403 if (e2c_vec[i].av_position != UINT64_MAX) {
404 layout |= e2c_vec[i].av_position;
412 * Save current output configuration if and only if it has been locked.
414 static void push_output_configuration(AACContext *ac) {
415 if (ac->oc[1].status == OC_LOCKED) {
416 ac->oc[0] = ac->oc[1];
418 ac->oc[1].status = OC_NONE;
422 * Restore the previous output configuration if and only if the current
423 * configuration is unlocked.
425 static void pop_output_configuration(AACContext *ac) {
426 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
427 ac->oc[1] = ac->oc[0];
428 ac->avctx->channels = ac->oc[1].channels;
429 ac->avctx->channel_layout = ac->oc[1].channel_layout;
434 * Configure output channel order based on the current program
435 * configuration element.
437 * @return Returns error status. 0 - OK, !0 - error
439 static int output_configure(AACContext *ac,
440 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
441 enum OCStatus oc_type, int get_new_frame)
443 AVCodecContext *avctx = ac->avctx;
444 int i, channels = 0, ret;
447 if (ac->oc[1].layout_map != layout_map) {
448 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
449 ac->oc[1].layout_map_tags = tags;
452 // Try to sniff a reasonable channel order, otherwise output the
453 // channels in the order the PCE declared them.
454 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
455 layout = sniff_channel_order(layout_map, tags);
456 for (i = 0; i < tags; i++) {
457 int type = layout_map[i][0];
458 int id = layout_map[i][1];
459 int position = layout_map[i][2];
460 // Allocate or free elements depending on if they are in the
461 // current program configuration.
462 ret = che_configure(ac, position, type, id, &channels);
466 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
467 if (layout == AV_CH_FRONT_CENTER) {
468 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
474 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
475 avctx->channel_layout = ac->oc[1].channel_layout = layout;
476 avctx->channels = ac->oc[1].channels = channels;
477 ac->oc[1].status = oc_type;
480 if ((ret = frame_configure_elements(ac->avctx)) < 0)
488 * Set up channel positions based on a default channel configuration
489 * as specified in table 1.17.
491 * @return Returns error status. 0 - OK, !0 - error
493 static int set_default_channel_config(AVCodecContext *avctx,
494 uint8_t (*layout_map)[3],
498 if (channel_config < 1 || channel_config > 7) {
499 av_log(avctx, AV_LOG_ERROR,
500 "invalid default channel configuration (%d)\n",
502 return AVERROR_INVALIDDATA;
504 *tags = tags_per_config[channel_config];
505 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
506 *tags * sizeof(*layout_map));
510 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
512 /* For PCE based channel configurations map the channels solely based
514 if (!ac->oc[1].m4ac.chan_config) {
515 return ac->tag_che_map[type][elem_id];
517 // Allow single CPE stereo files to be signalled with mono configuration.
518 if (!ac->tags_mapped && type == TYPE_CPE &&
519 ac->oc[1].m4ac.chan_config == 1) {
520 uint8_t layout_map[MAX_ELEM_ID*4][3];
522 push_output_configuration(ac);
524 if (set_default_channel_config(ac->avctx, layout_map,
525 &layout_map_tags, 2) < 0)
527 if (output_configure(ac, layout_map, layout_map_tags,
528 OC_TRIAL_FRAME, 1) < 0)
531 ac->oc[1].m4ac.chan_config = 2;
532 ac->oc[1].m4ac.ps = 0;
535 if (!ac->tags_mapped && type == TYPE_SCE &&
536 ac->oc[1].m4ac.chan_config == 2) {
537 uint8_t layout_map[MAX_ELEM_ID * 4][3];
539 push_output_configuration(ac);
541 if (set_default_channel_config(ac->avctx, layout_map,
542 &layout_map_tags, 1) < 0)
544 if (output_configure(ac, layout_map, layout_map_tags,
545 OC_TRIAL_FRAME, 1) < 0)
548 ac->oc[1].m4ac.chan_config = 1;
549 if (ac->oc[1].m4ac.sbr)
550 ac->oc[1].m4ac.ps = -1;
552 /* For indexed channel configurations map the channels solely based
554 switch (ac->oc[1].m4ac.chan_config) {
556 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
558 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
561 /* Some streams incorrectly code 5.1 audio as
562 * SCE[0] CPE[0] CPE[1] SCE[1]
564 * SCE[0] CPE[0] CPE[1] LFE[0].
565 * If we seem to have encountered such a stream, transfer
566 * the LFE[0] element to the SCE[1]'s mapping */
567 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
569 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
572 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
574 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
577 if (ac->tags_mapped == 2 &&
578 ac->oc[1].m4ac.chan_config == 4 &&
581 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
585 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
588 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
589 } else if (ac->oc[1].m4ac.chan_config == 2) {
593 if (!ac->tags_mapped && type == TYPE_SCE) {
595 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
603 * Decode an array of 4 bit element IDs, optionally interleaved with a
604 * stereo/mono switching bit.
606 * @param type speaker type/position for these channels
608 static void decode_channel_map(uint8_t layout_map[][3],
609 enum ChannelPosition type,
610 GetBitContext *gb, int n)
613 enum RawDataBlockType syn_ele;
615 case AAC_CHANNEL_FRONT:
616 case AAC_CHANNEL_BACK:
617 case AAC_CHANNEL_SIDE:
618 syn_ele = get_bits1(gb);
624 case AAC_CHANNEL_LFE:
628 // AAC_CHANNEL_OFF has no channel map
631 layout_map[0][0] = syn_ele;
632 layout_map[0][1] = get_bits(gb, 4);
633 layout_map[0][2] = type;
639 * Decode program configuration element; reference: table 4.2.
641 * @return Returns error status. 0 - OK, !0 - error
643 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
644 uint8_t (*layout_map)[3],
647 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
652 skip_bits(gb, 2); // object_type
654 sampling_index = get_bits(gb, 4);
655 if (m4ac->sampling_index != sampling_index)
656 av_log(avctx, AV_LOG_WARNING,
657 "Sample rate index in program config element does not "
658 "match the sample rate index configured by the container.\n");
660 num_front = get_bits(gb, 4);
661 num_side = get_bits(gb, 4);
662 num_back = get_bits(gb, 4);
663 num_lfe = get_bits(gb, 2);
664 num_assoc_data = get_bits(gb, 3);
665 num_cc = get_bits(gb, 4);
668 skip_bits(gb, 4); // mono_mixdown_tag
670 skip_bits(gb, 4); // stereo_mixdown_tag
673 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
675 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
677 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
679 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
681 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
684 skip_bits_long(gb, 4 * num_assoc_data);
686 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
691 /* comment field, first byte is length */
692 comment_len = get_bits(gb, 8) * 8;
693 if (get_bits_left(gb) < comment_len) {
694 av_log(avctx, AV_LOG_ERROR, overread_err);
695 return AVERROR_INVALIDDATA;
697 skip_bits_long(gb, comment_len);
702 * Decode GA "General Audio" specific configuration; reference: table 4.1.
704 * @param ac pointer to AACContext, may be null
705 * @param avctx pointer to AVCCodecContext, used for logging
707 * @return Returns error status. 0 - OK, !0 - error
709 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
711 MPEG4AudioConfig *m4ac,
714 int extension_flag, ret, ep_config, res_flags;
715 uint8_t layout_map[MAX_ELEM_ID*4][3];
718 if (get_bits1(gb)) { // frameLengthFlag
719 avpriv_request_sample(avctx, "960/120 MDCT window");
720 return AVERROR_PATCHWELCOME;
722 m4ac->frame_length_short = 0;
724 if (get_bits1(gb)) // dependsOnCoreCoder
725 skip_bits(gb, 14); // coreCoderDelay
726 extension_flag = get_bits1(gb);
728 if (m4ac->object_type == AOT_AAC_SCALABLE ||
729 m4ac->object_type == AOT_ER_AAC_SCALABLE)
730 skip_bits(gb, 3); // layerNr
732 if (channel_config == 0) {
733 skip_bits(gb, 4); // element_instance_tag
734 tags = decode_pce(avctx, m4ac, layout_map, gb);
738 if ((ret = set_default_channel_config(avctx, layout_map,
739 &tags, channel_config)))
743 if (count_channels(layout_map, tags) > 1) {
745 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
748 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
751 if (extension_flag) {
752 switch (m4ac->object_type) {
754 skip_bits(gb, 5); // numOfSubFrame
755 skip_bits(gb, 11); // layer_length
759 case AOT_ER_AAC_SCALABLE:
761 res_flags = get_bits(gb, 3);
763 avpriv_report_missing_feature(avctx,
764 "AAC data resilience (flags %x)",
766 return AVERROR_PATCHWELCOME;
770 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
772 switch (m4ac->object_type) {
775 case AOT_ER_AAC_SCALABLE:
777 ep_config = get_bits(gb, 2);
779 avpriv_report_missing_feature(avctx,
780 "epConfig %d", ep_config);
781 return AVERROR_PATCHWELCOME;
787 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
789 MPEG4AudioConfig *m4ac,
792 int ret, ep_config, res_flags;
793 uint8_t layout_map[MAX_ELEM_ID*4][3];
795 const int ELDEXT_TERM = 0;
800 m4ac->frame_length_short = get_bits1(gb);
801 res_flags = get_bits(gb, 3);
803 avpriv_report_missing_feature(avctx,
804 "AAC data resilience (flags %x)",
806 return AVERROR_PATCHWELCOME;
809 if (get_bits1(gb)) { // ldSbrPresentFlag
810 avpriv_report_missing_feature(avctx,
812 return AVERROR_PATCHWELCOME;
815 while (get_bits(gb, 4) != ELDEXT_TERM) {
816 int len = get_bits(gb, 4);
818 len += get_bits(gb, 8);
820 len += get_bits(gb, 16);
821 if (get_bits_left(gb) < len * 8 + 4) {
822 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
823 return AVERROR_INVALIDDATA;
825 skip_bits_long(gb, 8 * len);
828 if ((ret = set_default_channel_config(avctx, layout_map,
829 &tags, channel_config)))
832 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
835 ep_config = get_bits(gb, 2);
837 avpriv_report_missing_feature(avctx,
838 "epConfig %d", ep_config);
839 return AVERROR_PATCHWELCOME;
845 * Decode audio specific configuration; reference: table 1.13.
847 * @param ac pointer to AACContext, may be null
848 * @param avctx pointer to AVCCodecContext, used for logging
849 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
850 * @param data pointer to buffer holding an audio specific config
851 * @param bit_size size of audio specific config or data in bits
852 * @param sync_extension look for an appended sync extension
854 * @return Returns error status or number of consumed bits. <0 - error
856 static int decode_audio_specific_config(AACContext *ac,
857 AVCodecContext *avctx,
858 MPEG4AudioConfig *m4ac,
859 const uint8_t *data, int bit_size,
865 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
866 for (i = 0; i < avctx->extradata_size; i++)
867 av_dlog(avctx, "%02x ", avctx->extradata[i]);
868 av_dlog(avctx, "\n");
870 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
873 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
874 sync_extension)) < 0)
875 return AVERROR_INVALIDDATA;
876 if (m4ac->sampling_index > 12) {
877 av_log(avctx, AV_LOG_ERROR,
878 "invalid sampling rate index %d\n",
879 m4ac->sampling_index);
880 return AVERROR_INVALIDDATA;
882 if (m4ac->object_type == AOT_ER_AAC_LD &&
883 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
884 av_log(avctx, AV_LOG_ERROR,
885 "invalid low delay sampling rate index %d\n",
886 m4ac->sampling_index);
887 return AVERROR_INVALIDDATA;
890 skip_bits_long(&gb, i);
892 switch (m4ac->object_type) {
898 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
899 m4ac, m4ac->chan_config)) < 0)
903 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
904 m4ac, m4ac->chan_config)) < 0)
908 avpriv_report_missing_feature(avctx,
909 "Audio object type %s%d",
910 m4ac->sbr == 1 ? "SBR+" : "",
912 return AVERROR(ENOSYS);
916 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
917 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
918 m4ac->sample_rate, m4ac->sbr,
921 return get_bits_count(&gb);
925 * linear congruential pseudorandom number generator
927 * @param previous_val pointer to the current state of the generator
929 * @return Returns a 32-bit pseudorandom integer
931 static av_always_inline int lcg_random(int previous_val)
933 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
937 static av_always_inline void reset_predict_state(PredictorState *ps)
947 static void reset_all_predictors(PredictorState *ps)
950 for (i = 0; i < MAX_PREDICTORS; i++)
951 reset_predict_state(&ps[i]);
954 static int sample_rate_idx (int rate)
956 if (92017 <= rate) return 0;
957 else if (75132 <= rate) return 1;
958 else if (55426 <= rate) return 2;
959 else if (46009 <= rate) return 3;
960 else if (37566 <= rate) return 4;
961 else if (27713 <= rate) return 5;
962 else if (23004 <= rate) return 6;
963 else if (18783 <= rate) return 7;
964 else if (13856 <= rate) return 8;
965 else if (11502 <= rate) return 9;
966 else if (9391 <= rate) return 10;
970 static void reset_predictor_group(PredictorState *ps, int group_num)
973 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
974 reset_predict_state(&ps[i]);
977 #define AAC_INIT_VLC_STATIC(num, size) \
978 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
979 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
980 sizeof(ff_aac_spectral_bits[num][0]), \
981 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
982 sizeof(ff_aac_spectral_codes[num][0]), \
985 static av_cold int aac_decode_init(AVCodecContext *avctx)
987 AACContext *ac = avctx->priv_data;
991 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
993 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
995 if (avctx->extradata_size > 0) {
996 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
998 avctx->extradata_size * 8,
1003 uint8_t layout_map[MAX_ELEM_ID*4][3];
1004 int layout_map_tags;
1006 sr = sample_rate_idx(avctx->sample_rate);
1007 ac->oc[1].m4ac.sampling_index = sr;
1008 ac->oc[1].m4ac.channels = avctx->channels;
1009 ac->oc[1].m4ac.sbr = -1;
1010 ac->oc[1].m4ac.ps = -1;
1012 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1013 if (ff_mpeg4audio_channels[i] == avctx->channels)
1015 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1018 ac->oc[1].m4ac.chan_config = i;
1020 if (ac->oc[1].m4ac.chan_config) {
1021 int ret = set_default_channel_config(avctx, layout_map,
1022 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1024 output_configure(ac, layout_map, layout_map_tags,
1026 else if (avctx->err_recognition & AV_EF_EXPLODE)
1027 return AVERROR_INVALIDDATA;
1031 AAC_INIT_VLC_STATIC( 0, 304);
1032 AAC_INIT_VLC_STATIC( 1, 270);
1033 AAC_INIT_VLC_STATIC( 2, 550);
1034 AAC_INIT_VLC_STATIC( 3, 300);
1035 AAC_INIT_VLC_STATIC( 4, 328);
1036 AAC_INIT_VLC_STATIC( 5, 294);
1037 AAC_INIT_VLC_STATIC( 6, 306);
1038 AAC_INIT_VLC_STATIC( 7, 268);
1039 AAC_INIT_VLC_STATIC( 8, 510);
1040 AAC_INIT_VLC_STATIC( 9, 366);
1041 AAC_INIT_VLC_STATIC(10, 462);
1045 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1047 ac->random_state = 0x1f2e3d4c;
1051 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1052 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1053 ff_aac_scalefactor_bits,
1054 sizeof(ff_aac_scalefactor_bits[0]),
1055 sizeof(ff_aac_scalefactor_bits[0]),
1056 ff_aac_scalefactor_code,
1057 sizeof(ff_aac_scalefactor_code[0]),
1058 sizeof(ff_aac_scalefactor_code[0]),
1061 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1062 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1063 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1064 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1065 ret = ff_imdct15_init(&ac->mdct480, 5);
1069 // window initialization
1070 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1071 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1072 ff_init_ff_sine_windows(10);
1073 ff_init_ff_sine_windows( 9);
1074 ff_init_ff_sine_windows( 7);
1082 * Skip data_stream_element; reference: table 4.10.
1084 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1086 int byte_align = get_bits1(gb);
1087 int count = get_bits(gb, 8);
1089 count += get_bits(gb, 8);
1093 if (get_bits_left(gb) < 8 * count) {
1094 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1095 return AVERROR_INVALIDDATA;
1097 skip_bits_long(gb, 8 * count);
1101 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1105 if (get_bits1(gb)) {
1106 ics->predictor_reset_group = get_bits(gb, 5);
1107 if (ics->predictor_reset_group == 0 ||
1108 ics->predictor_reset_group > 30) {
1109 av_log(ac->avctx, AV_LOG_ERROR,
1110 "Invalid Predictor Reset Group.\n");
1111 return AVERROR_INVALIDDATA;
1114 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1115 ics->prediction_used[sfb] = get_bits1(gb);
1121 * Decode Long Term Prediction data; reference: table 4.xx.
1123 static void decode_ltp(LongTermPrediction *ltp,
1124 GetBitContext *gb, uint8_t max_sfb)
1128 ltp->lag = get_bits(gb, 11);
1129 ltp->coef = ltp_coef[get_bits(gb, 3)];
1130 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1131 ltp->used[sfb] = get_bits1(gb);
1135 * Decode Individual Channel Stream info; reference: table 4.6.
1137 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1140 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1141 const int aot = m4ac->object_type;
1142 const int sampling_index = m4ac->sampling_index;
1143 if (aot != AOT_ER_AAC_ELD) {
1144 if (get_bits1(gb)) {
1145 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1146 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1147 return AVERROR_INVALIDDATA;
1149 ics->window_sequence[1] = ics->window_sequence[0];
1150 ics->window_sequence[0] = get_bits(gb, 2);
1151 if (aot == AOT_ER_AAC_LD &&
1152 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1153 av_log(ac->avctx, AV_LOG_ERROR,
1154 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1155 "window sequence %d found.\n", ics->window_sequence[0]);
1156 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1157 return AVERROR_INVALIDDATA;
1159 ics->use_kb_window[1] = ics->use_kb_window[0];
1160 ics->use_kb_window[0] = get_bits1(gb);
1162 ics->num_window_groups = 1;
1163 ics->group_len[0] = 1;
1164 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1166 ics->max_sfb = get_bits(gb, 4);
1167 for (i = 0; i < 7; i++) {
1168 if (get_bits1(gb)) {
1169 ics->group_len[ics->num_window_groups - 1]++;
1171 ics->num_window_groups++;
1172 ics->group_len[ics->num_window_groups - 1] = 1;
1175 ics->num_windows = 8;
1176 ics->swb_offset = ff_swb_offset_128[sampling_index];
1177 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1178 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1179 ics->predictor_present = 0;
1181 ics->max_sfb = get_bits(gb, 6);
1182 ics->num_windows = 1;
1183 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1184 if (m4ac->frame_length_short) {
1185 ics->swb_offset = ff_swb_offset_480[sampling_index];
1186 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1187 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1189 ics->swb_offset = ff_swb_offset_512[sampling_index];
1190 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1191 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1193 if (!ics->num_swb || !ics->swb_offset)
1196 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1197 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1198 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1200 if (aot != AOT_ER_AAC_ELD) {
1201 ics->predictor_present = get_bits1(gb);
1202 ics->predictor_reset_group = 0;
1204 if (ics->predictor_present) {
1205 if (aot == AOT_AAC_MAIN) {
1206 if (decode_prediction(ac, ics, gb)) {
1207 return AVERROR_INVALIDDATA;
1209 } else if (aot == AOT_AAC_LC ||
1210 aot == AOT_ER_AAC_LC) {
1211 av_log(ac->avctx, AV_LOG_ERROR,
1212 "Prediction is not allowed in AAC-LC.\n");
1213 return AVERROR_INVALIDDATA;
1215 if (aot == AOT_ER_AAC_LD) {
1216 av_log(ac->avctx, AV_LOG_ERROR,
1217 "LTP in ER AAC LD not yet implemented.\n");
1218 return AVERROR_PATCHWELCOME;
1220 if ((ics->ltp.present = get_bits(gb, 1)))
1221 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1226 if (ics->max_sfb > ics->num_swb) {
1227 av_log(ac->avctx, AV_LOG_ERROR,
1228 "Number of scalefactor bands in group (%d) "
1229 "exceeds limit (%d).\n",
1230 ics->max_sfb, ics->num_swb);
1231 return AVERROR_INVALIDDATA;
1238 * Decode band types (section_data payload); reference: table 4.46.
1240 * @param band_type array of the used band type
1241 * @param band_type_run_end array of the last scalefactor band of a band type run
1243 * @return Returns error status. 0 - OK, !0 - error
1245 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1246 int band_type_run_end[120], GetBitContext *gb,
1247 IndividualChannelStream *ics)
1250 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1251 for (g = 0; g < ics->num_window_groups; g++) {
1253 while (k < ics->max_sfb) {
1254 uint8_t sect_end = k;
1256 int sect_band_type = get_bits(gb, 4);
1257 if (sect_band_type == 12) {
1258 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1259 return AVERROR_INVALIDDATA;
1262 sect_len_incr = get_bits(gb, bits);
1263 sect_end += sect_len_incr;
1264 if (get_bits_left(gb) < 0) {
1265 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1266 return AVERROR_INVALIDDATA;
1268 if (sect_end > ics->max_sfb) {
1269 av_log(ac->avctx, AV_LOG_ERROR,
1270 "Number of bands (%d) exceeds limit (%d).\n",
1271 sect_end, ics->max_sfb);
1272 return AVERROR_INVALIDDATA;
1274 } while (sect_len_incr == (1 << bits) - 1);
1275 for (; k < sect_end; k++) {
1276 band_type [idx] = sect_band_type;
1277 band_type_run_end[idx++] = sect_end;
1285 * Decode scalefactors; reference: table 4.47.
1287 * @param global_gain first scalefactor value as scalefactors are differentially coded
1288 * @param band_type array of the used band type
1289 * @param band_type_run_end array of the last scalefactor band of a band type run
1290 * @param sf array of scalefactors or intensity stereo positions
1292 * @return Returns error status. 0 - OK, !0 - error
1294 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1295 unsigned int global_gain,
1296 IndividualChannelStream *ics,
1297 enum BandType band_type[120],
1298 int band_type_run_end[120])
1301 int offset[3] = { global_gain, global_gain - 90, 0 };
1304 for (g = 0; g < ics->num_window_groups; g++) {
1305 for (i = 0; i < ics->max_sfb;) {
1306 int run_end = band_type_run_end[idx];
1307 if (band_type[idx] == ZERO_BT) {
1308 for (; i < run_end; i++, idx++)
1310 } else if ((band_type[idx] == INTENSITY_BT) ||
1311 (band_type[idx] == INTENSITY_BT2)) {
1312 for (; i < run_end; i++, idx++) {
1313 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1314 clipped_offset = av_clip(offset[2], -155, 100);
1315 if (offset[2] != clipped_offset) {
1316 avpriv_request_sample(ac->avctx,
1317 "If you heard an audible artifact, there may be a bug in the decoder. "
1318 "Clipped intensity stereo position (%d -> %d)",
1319 offset[2], clipped_offset);
1321 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1323 } else if (band_type[idx] == NOISE_BT) {
1324 for (; i < run_end; i++, idx++) {
1325 if (noise_flag-- > 0)
1326 offset[1] += get_bits(gb, 9) - 256;
1328 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1329 clipped_offset = av_clip(offset[1], -100, 155);
1330 if (offset[1] != clipped_offset) {
1331 avpriv_request_sample(ac->avctx,
1332 "If you heard an audible artifact, there may be a bug in the decoder. "
1333 "Clipped noise gain (%d -> %d)",
1334 offset[1], clipped_offset);
1336 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1339 for (; i < run_end; i++, idx++) {
1340 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1341 if (offset[0] > 255U) {
1342 av_log(ac->avctx, AV_LOG_ERROR,
1343 "Scalefactor (%d) out of range.\n", offset[0]);
1344 return AVERROR_INVALIDDATA;
1346 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1355 * Decode pulse data; reference: table 4.7.
1357 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1358 const uint16_t *swb_offset, int num_swb)
1361 pulse->num_pulse = get_bits(gb, 2) + 1;
1362 pulse_swb = get_bits(gb, 6);
1363 if (pulse_swb >= num_swb)
1365 pulse->pos[0] = swb_offset[pulse_swb];
1366 pulse->pos[0] += get_bits(gb, 5);
1367 if (pulse->pos[0] > 1023)
1369 pulse->amp[0] = get_bits(gb, 4);
1370 for (i = 1; i < pulse->num_pulse; i++) {
1371 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1372 if (pulse->pos[i] > 1023)
1374 pulse->amp[i] = get_bits(gb, 4);
1380 * Decode Temporal Noise Shaping data; reference: table 4.48.
1382 * @return Returns error status. 0 - OK, !0 - error
1384 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1385 GetBitContext *gb, const IndividualChannelStream *ics)
1387 int w, filt, i, coef_len, coef_res, coef_compress;
1388 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1389 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1390 for (w = 0; w < ics->num_windows; w++) {
1391 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1392 coef_res = get_bits1(gb);
1394 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1396 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1398 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1399 av_log(ac->avctx, AV_LOG_ERROR,
1400 "TNS filter order %d is greater than maximum %d.\n",
1401 tns->order[w][filt], tns_max_order);
1402 tns->order[w][filt] = 0;
1403 return AVERROR_INVALIDDATA;
1405 if (tns->order[w][filt]) {
1406 tns->direction[w][filt] = get_bits1(gb);
1407 coef_compress = get_bits1(gb);
1408 coef_len = coef_res + 3 - coef_compress;
1409 tmp2_idx = 2 * coef_compress + coef_res;
1411 for (i = 0; i < tns->order[w][filt]; i++)
1412 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1421 * Decode Mid/Side data; reference: table 4.54.
1423 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1424 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1425 * [3] reserved for scalable AAC
1427 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1431 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1432 if (ms_present == 1) {
1433 for (idx = 0; idx < max_idx; idx++)
1434 cpe->ms_mask[idx] = get_bits1(gb);
1435 } else if (ms_present == 2) {
1436 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1441 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1445 *dst++ = v[idx & 15] * s;
1446 *dst++ = v[idx>>4 & 15] * s;
1452 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1456 *dst++ = v[idx & 3] * s;
1457 *dst++ = v[idx>>2 & 3] * s;
1458 *dst++ = v[idx>>4 & 3] * s;
1459 *dst++ = v[idx>>6 & 3] * s;
1465 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1466 unsigned sign, const float *scale)
1468 union av_intfloat32 s0, s1;
1470 s0.f = s1.f = *scale;
1471 s0.i ^= sign >> 1 << 31;
1474 *dst++ = v[idx & 15] * s0.f;
1475 *dst++ = v[idx>>4 & 15] * s1.f;
1482 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1483 unsigned sign, const float *scale)
1485 unsigned nz = idx >> 12;
1486 union av_intfloat32 s = { .f = *scale };
1487 union av_intfloat32 t;
1489 t.i = s.i ^ (sign & 1U<<31);
1490 *dst++ = v[idx & 3] * t.f;
1492 sign <<= nz & 1; nz >>= 1;
1493 t.i = s.i ^ (sign & 1U<<31);
1494 *dst++ = v[idx>>2 & 3] * t.f;
1496 sign <<= nz & 1; nz >>= 1;
1497 t.i = s.i ^ (sign & 1U<<31);
1498 *dst++ = v[idx>>4 & 3] * t.f;
1501 t.i = s.i ^ (sign & 1U<<31);
1502 *dst++ = v[idx>>6 & 3] * t.f;
1509 * Decode spectral data; reference: table 4.50.
1510 * Dequantize and scale spectral data; reference: 4.6.3.3.
1512 * @param coef array of dequantized, scaled spectral data
1513 * @param sf array of scalefactors or intensity stereo positions
1514 * @param pulse_present set if pulses are present
1515 * @param pulse pointer to pulse data struct
1516 * @param band_type array of the used band type
1518 * @return Returns error status. 0 - OK, !0 - error
1520 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1521 GetBitContext *gb, const float sf[120],
1522 int pulse_present, const Pulse *pulse,
1523 const IndividualChannelStream *ics,
1524 enum BandType band_type[120])
1526 int i, k, g, idx = 0;
1527 const int c = 1024 / ics->num_windows;
1528 const uint16_t *offsets = ics->swb_offset;
1529 float *coef_base = coef;
1531 for (g = 0; g < ics->num_windows; g++)
1532 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1533 sizeof(float) * (c - offsets[ics->max_sfb]));
1535 for (g = 0; g < ics->num_window_groups; g++) {
1536 unsigned g_len = ics->group_len[g];
1538 for (i = 0; i < ics->max_sfb; i++, idx++) {
1539 const unsigned cbt_m1 = band_type[idx] - 1;
1540 float *cfo = coef + offsets[i];
1541 int off_len = offsets[i + 1] - offsets[i];
1544 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1545 for (group = 0; group < g_len; group++, cfo+=128) {
1546 memset(cfo, 0, off_len * sizeof(float));
1548 } else if (cbt_m1 == NOISE_BT - 1) {
1549 for (group = 0; group < g_len; group++, cfo+=128) {
1553 for (k = 0; k < off_len; k++) {
1554 ac->random_state = lcg_random(ac->random_state);
1555 cfo[k] = ac->random_state;
1558 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1559 scale = sf[idx] / sqrtf(band_energy);
1560 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1563 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1564 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1565 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1566 OPEN_READER(re, gb);
1568 switch (cbt_m1 >> 1) {
1570 for (group = 0; group < g_len; group++, cfo+=128) {
1578 UPDATE_CACHE(re, gb);
1579 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1580 cb_idx = cb_vector_idx[code];
1581 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1587 for (group = 0; group < g_len; group++, cfo+=128) {
1597 UPDATE_CACHE(re, gb);
1598 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1599 cb_idx = cb_vector_idx[code];
1600 nnz = cb_idx >> 8 & 15;
1601 bits = nnz ? GET_CACHE(re, gb) : 0;
1602 LAST_SKIP_BITS(re, gb, nnz);
1603 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1609 for (group = 0; group < g_len; group++, cfo+=128) {
1617 UPDATE_CACHE(re, gb);
1618 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1619 cb_idx = cb_vector_idx[code];
1620 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1627 for (group = 0; group < g_len; group++, cfo+=128) {
1637 UPDATE_CACHE(re, gb);
1638 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1639 cb_idx = cb_vector_idx[code];
1640 nnz = cb_idx >> 8 & 15;
1641 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1642 LAST_SKIP_BITS(re, gb, nnz);
1643 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1649 for (group = 0; group < g_len; group++, cfo+=128) {
1651 uint32_t *icf = (uint32_t *) cf;
1661 UPDATE_CACHE(re, gb);
1662 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1670 cb_idx = cb_vector_idx[code];
1673 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1674 LAST_SKIP_BITS(re, gb, nnz);
1676 for (j = 0; j < 2; j++) {
1680 /* The total length of escape_sequence must be < 22 bits according
1681 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1682 UPDATE_CACHE(re, gb);
1683 b = GET_CACHE(re, gb);
1684 b = 31 - av_log2(~b);
1687 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1688 return AVERROR_INVALIDDATA;
1691 SKIP_BITS(re, gb, b + 1);
1693 n = (1 << b) + SHOW_UBITS(re, gb, b);
1694 LAST_SKIP_BITS(re, gb, b);
1695 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1698 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1699 *icf++ = (bits & 1U<<31) | v;
1706 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1710 CLOSE_READER(re, gb);
1716 if (pulse_present) {
1718 for (i = 0; i < pulse->num_pulse; i++) {
1719 float co = coef_base[ pulse->pos[i] ];
1720 while (offsets[idx + 1] <= pulse->pos[i])
1722 if (band_type[idx] != NOISE_BT && sf[idx]) {
1723 float ico = -pulse->amp[i];
1726 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1728 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1735 static av_always_inline float flt16_round(float pf)
1737 union av_intfloat32 tmp;
1739 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1743 static av_always_inline float flt16_even(float pf)
1745 union av_intfloat32 tmp;
1747 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1751 static av_always_inline float flt16_trunc(float pf)
1753 union av_intfloat32 pun;
1755 pun.i &= 0xFFFF0000U;
1759 static av_always_inline void predict(PredictorState *ps, float *coef,
1762 const float a = 0.953125; // 61.0 / 64
1763 const float alpha = 0.90625; // 29.0 / 32
1767 float r0 = ps->r0, r1 = ps->r1;
1768 float cor0 = ps->cor0, cor1 = ps->cor1;
1769 float var0 = ps->var0, var1 = ps->var1;
1771 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1772 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1774 pv = flt16_round(k1 * r0 + k2 * r1);
1781 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1782 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1783 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1784 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1786 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1787 ps->r0 = flt16_trunc(a * e0);
1791 * Apply AAC-Main style frequency domain prediction.
1793 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1797 if (!sce->ics.predictor_initialized) {
1798 reset_all_predictors(sce->predictor_state);
1799 sce->ics.predictor_initialized = 1;
1802 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1804 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1806 for (k = sce->ics.swb_offset[sfb];
1807 k < sce->ics.swb_offset[sfb + 1];
1809 predict(&sce->predictor_state[k], &sce->coeffs[k],
1810 sce->ics.predictor_present &&
1811 sce->ics.prediction_used[sfb]);
1814 if (sce->ics.predictor_reset_group)
1815 reset_predictor_group(sce->predictor_state,
1816 sce->ics.predictor_reset_group);
1818 reset_all_predictors(sce->predictor_state);
1822 * Decode an individual_channel_stream payload; reference: table 4.44.
1824 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1825 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1827 * @return Returns error status. 0 - OK, !0 - error
1829 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1830 GetBitContext *gb, int common_window, int scale_flag)
1833 TemporalNoiseShaping *tns = &sce->tns;
1834 IndividualChannelStream *ics = &sce->ics;
1835 float *out = sce->coeffs;
1836 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1839 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1840 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1841 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1842 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1843 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1845 /* This assignment is to silence a GCC warning about the variable being used
1846 * uninitialized when in fact it always is.
1848 pulse.num_pulse = 0;
1850 global_gain = get_bits(gb, 8);
1852 if (!common_window && !scale_flag) {
1853 if (decode_ics_info(ac, ics, gb) < 0)
1854 return AVERROR_INVALIDDATA;
1857 if ((ret = decode_band_types(ac, sce->band_type,
1858 sce->band_type_run_end, gb, ics)) < 0)
1860 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1861 sce->band_type, sce->band_type_run_end)) < 0)
1866 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1867 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1868 av_log(ac->avctx, AV_LOG_ERROR,
1869 "Pulse tool not allowed in eight short sequence.\n");
1870 return AVERROR_INVALIDDATA;
1872 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1873 av_log(ac->avctx, AV_LOG_ERROR,
1874 "Pulse data corrupt or invalid.\n");
1875 return AVERROR_INVALIDDATA;
1878 tns->present = get_bits1(gb);
1879 if (tns->present && !er_syntax)
1880 if (decode_tns(ac, tns, gb, ics) < 0)
1881 return AVERROR_INVALIDDATA;
1882 if (!eld_syntax && get_bits1(gb)) {
1883 avpriv_request_sample(ac->avctx, "SSR");
1884 return AVERROR_PATCHWELCOME;
1886 // I see no textual basis in the spec for this occuring after SSR gain
1887 // control, but this is what both reference and real implmentations do
1888 if (tns->present && er_syntax)
1889 if (decode_tns(ac, tns, gb, ics) < 0)
1890 return AVERROR_INVALIDDATA;
1893 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1894 &pulse, ics, sce->band_type) < 0)
1895 return AVERROR_INVALIDDATA;
1897 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1898 apply_prediction(ac, sce);
1904 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1906 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1908 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1909 float *ch0 = cpe->ch[0].coeffs;
1910 float *ch1 = cpe->ch[1].coeffs;
1911 int g, i, group, idx = 0;
1912 const uint16_t *offsets = ics->swb_offset;
1913 for (g = 0; g < ics->num_window_groups; g++) {
1914 for (i = 0; i < ics->max_sfb; i++, idx++) {
1915 if (cpe->ms_mask[idx] &&
1916 cpe->ch[0].band_type[idx] < NOISE_BT &&
1917 cpe->ch[1].band_type[idx] < NOISE_BT) {
1918 for (group = 0; group < ics->group_len[g]; group++) {
1919 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1920 ch1 + group * 128 + offsets[i],
1921 offsets[i+1] - offsets[i]);
1925 ch0 += ics->group_len[g] * 128;
1926 ch1 += ics->group_len[g] * 128;
1931 * intensity stereo decoding; reference: 4.6.8.2.3
1933 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1934 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1935 * [3] reserved for scalable AAC
1937 static void apply_intensity_stereo(AACContext *ac,
1938 ChannelElement *cpe, int ms_present)
1940 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1941 SingleChannelElement *sce1 = &cpe->ch[1];
1942 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1943 const uint16_t *offsets = ics->swb_offset;
1944 int g, group, i, idx = 0;
1947 for (g = 0; g < ics->num_window_groups; g++) {
1948 for (i = 0; i < ics->max_sfb;) {
1949 if (sce1->band_type[idx] == INTENSITY_BT ||
1950 sce1->band_type[idx] == INTENSITY_BT2) {
1951 const int bt_run_end = sce1->band_type_run_end[idx];
1952 for (; i < bt_run_end; i++, idx++) {
1953 c = -1 + 2 * (sce1->band_type[idx] - 14);
1955 c *= 1 - 2 * cpe->ms_mask[idx];
1956 scale = c * sce1->sf[idx];
1957 for (group = 0; group < ics->group_len[g]; group++)
1958 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1959 coef0 + group * 128 + offsets[i],
1961 offsets[i + 1] - offsets[i]);
1964 int bt_run_end = sce1->band_type_run_end[idx];
1965 idx += bt_run_end - i;
1969 coef0 += ics->group_len[g] * 128;
1970 coef1 += ics->group_len[g] * 128;
1975 * Decode a channel_pair_element; reference: table 4.4.
1977 * @return Returns error status. 0 - OK, !0 - error
1979 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1981 int i, ret, common_window, ms_present = 0;
1982 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1984 common_window = eld_syntax || get_bits1(gb);
1985 if (common_window) {
1986 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1987 return AVERROR_INVALIDDATA;
1988 i = cpe->ch[1].ics.use_kb_window[0];
1989 cpe->ch[1].ics = cpe->ch[0].ics;
1990 cpe->ch[1].ics.use_kb_window[1] = i;
1991 if (cpe->ch[1].ics.predictor_present &&
1992 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1993 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1994 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1995 ms_present = get_bits(gb, 2);
1996 if (ms_present == 3) {
1997 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1998 return AVERROR_INVALIDDATA;
1999 } else if (ms_present)
2000 decode_mid_side_stereo(cpe, gb, ms_present);
2002 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2004 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2007 if (common_window) {
2009 apply_mid_side_stereo(ac, cpe);
2010 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2011 apply_prediction(ac, &cpe->ch[0]);
2012 apply_prediction(ac, &cpe->ch[1]);
2016 apply_intensity_stereo(ac, cpe, ms_present);
2020 static const float cce_scale[] = {
2021 1.09050773266525765921, //2^(1/8)
2022 1.18920711500272106672, //2^(1/4)
2028 * Decode coupling_channel_element; reference: table 4.8.
2030 * @return Returns error status. 0 - OK, !0 - error
2032 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2038 SingleChannelElement *sce = &che->ch[0];
2039 ChannelCoupling *coup = &che->coup;
2041 coup->coupling_point = 2 * get_bits1(gb);
2042 coup->num_coupled = get_bits(gb, 3);
2043 for (c = 0; c <= coup->num_coupled; c++) {
2045 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2046 coup->id_select[c] = get_bits(gb, 4);
2047 if (coup->type[c] == TYPE_CPE) {
2048 coup->ch_select[c] = get_bits(gb, 2);
2049 if (coup->ch_select[c] == 3)
2052 coup->ch_select[c] = 2;
2054 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2056 sign = get_bits(gb, 1);
2057 scale = cce_scale[get_bits(gb, 2)];
2059 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2062 for (c = 0; c < num_gain; c++) {
2066 float gain_cache = 1.0;
2068 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2069 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2070 gain_cache = powf(scale, -gain);
2072 if (coup->coupling_point == AFTER_IMDCT) {
2073 coup->gain[c][0] = gain_cache;
2075 for (g = 0; g < sce->ics.num_window_groups; g++) {
2076 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2077 if (sce->band_type[idx] != ZERO_BT) {
2079 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2087 gain_cache = powf(scale, -t) * s;
2090 coup->gain[c][idx] = gain_cache;
2100 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2102 * @return Returns number of bytes consumed.
2104 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2108 int num_excl_chan = 0;
2111 for (i = 0; i < 7; i++)
2112 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2113 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2115 return num_excl_chan / 7;
2119 * Decode dynamic range information; reference: table 4.52.
2121 * @return Returns number of bytes consumed.
2123 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2127 int drc_num_bands = 1;
2130 /* pce_tag_present? */
2131 if (get_bits1(gb)) {
2132 che_drc->pce_instance_tag = get_bits(gb, 4);
2133 skip_bits(gb, 4); // tag_reserved_bits
2137 /* excluded_chns_present? */
2138 if (get_bits1(gb)) {
2139 n += decode_drc_channel_exclusions(che_drc, gb);
2142 /* drc_bands_present? */
2143 if (get_bits1(gb)) {
2144 che_drc->band_incr = get_bits(gb, 4);
2145 che_drc->interpolation_scheme = get_bits(gb, 4);
2147 drc_num_bands += che_drc->band_incr;
2148 for (i = 0; i < drc_num_bands; i++) {
2149 che_drc->band_top[i] = get_bits(gb, 8);
2154 /* prog_ref_level_present? */
2155 if (get_bits1(gb)) {
2156 che_drc->prog_ref_level = get_bits(gb, 7);
2157 skip_bits1(gb); // prog_ref_level_reserved_bits
2161 for (i = 0; i < drc_num_bands; i++) {
2162 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2163 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2171 * Decode extension data (incomplete); reference: table 4.51.
2173 * @param cnt length of TYPE_FIL syntactic element in bytes
2175 * @return Returns number of bytes consumed
2177 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2178 ChannelElement *che, enum RawDataBlockType elem_type)
2182 switch (get_bits(gb, 4)) { // extension type
2183 case EXT_SBR_DATA_CRC:
2187 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2189 } else if (!ac->oc[1].m4ac.sbr) {
2190 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2191 skip_bits_long(gb, 8 * cnt - 4);
2193 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2194 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2195 skip_bits_long(gb, 8 * cnt - 4);
2197 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2198 ac->oc[1].m4ac.sbr = 1;
2199 ac->oc[1].m4ac.ps = 1;
2200 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2201 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2202 ac->oc[1].status, 1);
2204 ac->oc[1].m4ac.sbr = 1;
2205 ac->avctx->profile = FF_PROFILE_AAC_HE;
2207 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2209 case EXT_DYNAMIC_RANGE:
2210 res = decode_dynamic_range(&ac->che_drc, gb);
2214 case EXT_DATA_ELEMENT:
2216 skip_bits_long(gb, 8 * cnt - 4);
2223 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2225 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2226 * @param coef spectral coefficients
2228 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2229 IndividualChannelStream *ics, int decode)
2231 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2233 int bottom, top, order, start, end, size, inc;
2234 float lpc[TNS_MAX_ORDER];
2235 float tmp[TNS_MAX_ORDER + 1];
2237 for (w = 0; w < ics->num_windows; w++) {
2238 bottom = ics->num_swb;
2239 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2241 bottom = FFMAX(0, top - tns->length[w][filt]);
2242 order = tns->order[w][filt];
2247 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2249 start = ics->swb_offset[FFMIN(bottom, mmm)];
2250 end = ics->swb_offset[FFMIN( top, mmm)];
2251 if ((size = end - start) <= 0)
2253 if (tns->direction[w][filt]) {
2263 for (m = 0; m < size; m++, start += inc)
2264 for (i = 1; i <= FFMIN(m, order); i++)
2265 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2268 for (m = 0; m < size; m++, start += inc) {
2269 tmp[0] = coef[start];
2270 for (i = 1; i <= FFMIN(m, order); i++)
2271 coef[start] += tmp[i] * lpc[i - 1];
2272 for (i = order; i > 0; i--)
2273 tmp[i] = tmp[i - 1];
2281 * Apply windowing and MDCT to obtain the spectral
2282 * coefficient from the predicted sample by LTP.
2284 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2285 float *in, IndividualChannelStream *ics)
2287 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2288 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2289 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2290 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2292 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2293 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2295 memset(in, 0, 448 * sizeof(float));
2296 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2298 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2299 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2301 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2302 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2304 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2308 * Apply the long term prediction
2310 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2312 const LongTermPrediction *ltp = &sce->ics.ltp;
2313 const uint16_t *offsets = sce->ics.swb_offset;
2316 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2317 float *predTime = sce->ret;
2318 float *predFreq = ac->buf_mdct;
2319 int16_t num_samples = 2048;
2321 if (ltp->lag < 1024)
2322 num_samples = ltp->lag + 1024;
2323 for (i = 0; i < num_samples; i++)
2324 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2325 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2327 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2329 if (sce->tns.present)
2330 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2332 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2334 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2335 sce->coeffs[i] += predFreq[i];
2340 * Update the LTP buffer for next frame
2342 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2344 IndividualChannelStream *ics = &sce->ics;
2345 float *saved = sce->saved;
2346 float *saved_ltp = sce->coeffs;
2347 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2348 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2351 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2352 memcpy(saved_ltp, saved, 512 * sizeof(float));
2353 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2354 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2355 for (i = 0; i < 64; i++)
2356 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2357 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2358 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2359 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2360 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2361 for (i = 0; i < 64; i++)
2362 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2363 } else { // LONG_STOP or ONLY_LONG
2364 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2365 for (i = 0; i < 512; i++)
2366 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2369 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2370 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2371 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2375 * Conduct IMDCT and windowing.
2377 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2379 IndividualChannelStream *ics = &sce->ics;
2380 float *in = sce->coeffs;
2381 float *out = sce->ret;
2382 float *saved = sce->saved;
2383 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2384 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2385 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2386 float *buf = ac->buf_mdct;
2387 float *temp = ac->temp;
2391 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2392 for (i = 0; i < 1024; i += 128)
2393 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2395 ac->mdct.imdct_half(&ac->mdct, buf, in);
2397 /* window overlapping
2398 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2399 * and long to short transitions are considered to be short to short
2400 * transitions. This leaves just two cases (long to long and short to short)
2401 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2403 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2404 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2405 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2407 memcpy( out, saved, 448 * sizeof(float));
2409 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2410 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2411 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2412 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2413 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2414 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2415 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2417 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2418 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2423 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2424 memcpy( saved, temp + 64, 64 * sizeof(float));
2425 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2426 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2427 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2428 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2429 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2430 memcpy( saved, buf + 512, 448 * sizeof(float));
2431 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2432 } else { // LONG_STOP or ONLY_LONG
2433 memcpy( saved, buf + 512, 512 * sizeof(float));
2437 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2439 IndividualChannelStream *ics = &sce->ics;
2440 float *in = sce->coeffs;
2441 float *out = sce->ret;
2442 float *saved = sce->saved;
2443 float *buf = ac->buf_mdct;
2446 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2448 // window overlapping
2449 if (ics->use_kb_window[1]) {
2450 // AAC LD uses a low overlap sine window instead of a KBD window
2451 memcpy(out, saved, 192 * sizeof(float));
2452 ac->fdsp.vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2453 memcpy( out + 320, buf + 64, 192 * sizeof(float));
2455 ac->fdsp.vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2459 memcpy(saved, buf + 256, 256 * sizeof(float));
2462 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2464 float *in = sce->coeffs;
2465 float *out = sce->ret;
2466 float *saved = sce->saved;
2467 float *buf = ac->buf_mdct;
2469 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2470 const int n2 = n >> 1;
2471 const int n4 = n >> 2;
2472 const float *const window = n == 480 ? ff_aac_eld_window_480 :
2473 ff_aac_eld_window_512;
2475 // Inverse transform, mapped to the conventional IMDCT by
2476 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2477 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2478 // Audio, Language and Image Processing, 2008. ICALIP 2008. International Conference on
2479 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2480 for (i = 0; i < n2; i+=2) {
2482 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2483 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2486 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2488 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2489 for (i = 0; i < n; i+=2) {
2492 // Like with the regular IMDCT at this point we still have the middle half
2493 // of a transform but with even symmetry on the left and odd symmetry on
2496 // window overlapping
2497 // The spec says to use samples [0..511] but the reference decoder uses
2498 // samples [128..639].
2499 for (i = n4; i < n2; i ++) {
2500 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2501 saved[ i + n2] * window[i + n - n4] +
2502 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2503 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2505 for (i = 0; i < n2; i ++) {
2506 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2507 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2508 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2509 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2511 for (i = 0; i < n4; i ++) {
2512 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2513 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2514 -saved[ n + n2 + i] * window[i + 3*n - n4];
2518 memmove(saved + n, saved, 2 * n * sizeof(float));
2519 memcpy( saved, buf, n * sizeof(float));
2523 * Apply dependent channel coupling (applied before IMDCT).
2525 * @param index index into coupling gain array
2527 static void apply_dependent_coupling(AACContext *ac,
2528 SingleChannelElement *target,
2529 ChannelElement *cce, int index)
2531 IndividualChannelStream *ics = &cce->ch[0].ics;
2532 const uint16_t *offsets = ics->swb_offset;
2533 float *dest = target->coeffs;
2534 const float *src = cce->ch[0].coeffs;
2535 int g, i, group, k, idx = 0;
2536 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2537 av_log(ac->avctx, AV_LOG_ERROR,
2538 "Dependent coupling is not supported together with LTP\n");
2541 for (g = 0; g < ics->num_window_groups; g++) {
2542 for (i = 0; i < ics->max_sfb; i++, idx++) {
2543 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2544 const float gain = cce->coup.gain[index][idx];
2545 for (group = 0; group < ics->group_len[g]; group++) {
2546 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2548 dest[group * 128 + k] += gain * src[group * 128 + k];
2553 dest += ics->group_len[g] * 128;
2554 src += ics->group_len[g] * 128;
2559 * Apply independent channel coupling (applied after IMDCT).
2561 * @param index index into coupling gain array
2563 static void apply_independent_coupling(AACContext *ac,
2564 SingleChannelElement *target,
2565 ChannelElement *cce, int index)
2568 const float gain = cce->coup.gain[index][0];
2569 const float *src = cce->ch[0].ret;
2570 float *dest = target->ret;
2571 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2573 for (i = 0; i < len; i++)
2574 dest[i] += gain * src[i];
2578 * channel coupling transformation interface
2580 * @param apply_coupling_method pointer to (in)dependent coupling function
2582 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2583 enum RawDataBlockType type, int elem_id,
2584 enum CouplingPoint coupling_point,
2585 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2589 for (i = 0; i < MAX_ELEM_ID; i++) {
2590 ChannelElement *cce = ac->che[TYPE_CCE][i];
2593 if (cce && cce->coup.coupling_point == coupling_point) {
2594 ChannelCoupling *coup = &cce->coup;
2596 for (c = 0; c <= coup->num_coupled; c++) {
2597 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2598 if (coup->ch_select[c] != 1) {
2599 apply_coupling_method(ac, &cc->ch[0], cce, index);
2600 if (coup->ch_select[c] != 0)
2603 if (coup->ch_select[c] != 2)
2604 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2606 index += 1 + (coup->ch_select[c] == 3);
2613 * Convert spectral data to float samples, applying all supported tools as appropriate.
2615 static void spectral_to_sample(AACContext *ac)
2618 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2619 switch (ac->oc[1].m4ac.object_type) {
2621 imdct_and_window = imdct_and_windowing_ld;
2623 case AOT_ER_AAC_ELD:
2624 imdct_and_window = imdct_and_windowing_eld;
2627 imdct_and_window = imdct_and_windowing;
2629 for (type = 3; type >= 0; type--) {
2630 for (i = 0; i < MAX_ELEM_ID; i++) {
2631 ChannelElement *che = ac->che[type][i];
2633 if (type <= TYPE_CPE)
2634 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2635 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2636 if (che->ch[0].ics.predictor_present) {
2637 if (che->ch[0].ics.ltp.present)
2638 apply_ltp(ac, &che->ch[0]);
2639 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2640 apply_ltp(ac, &che->ch[1]);
2643 if (che->ch[0].tns.present)
2644 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2645 if (che->ch[1].tns.present)
2646 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2647 if (type <= TYPE_CPE)
2648 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2649 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2650 imdct_and_window(ac, &che->ch[0]);
2651 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2652 update_ltp(ac, &che->ch[0]);
2653 if (type == TYPE_CPE) {
2654 imdct_and_window(ac, &che->ch[1]);
2655 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2656 update_ltp(ac, &che->ch[1]);
2658 if (ac->oc[1].m4ac.sbr > 0) {
2659 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2662 if (type <= TYPE_CCE)
2663 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2669 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2672 AACADTSHeaderInfo hdr_info;
2673 uint8_t layout_map[MAX_ELEM_ID*4][3];
2674 int layout_map_tags, ret;
2676 size = avpriv_aac_parse_header(gb, &hdr_info);
2678 if (hdr_info.num_aac_frames != 1) {
2679 avpriv_report_missing_feature(ac->avctx,
2680 "More than one AAC RDB per ADTS frame");
2681 return AVERROR_PATCHWELCOME;
2683 push_output_configuration(ac);
2684 if (hdr_info.chan_config) {
2685 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2686 if ((ret = set_default_channel_config(ac->avctx,
2689 hdr_info.chan_config)) < 0)
2691 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2692 FFMAX(ac->oc[1].status,
2693 OC_TRIAL_FRAME), 0)) < 0)
2696 ac->oc[1].m4ac.chan_config = 0;
2698 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2699 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2700 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2701 ac->oc[1].m4ac.frame_length_short = 0;
2702 if (ac->oc[0].status != OC_LOCKED ||
2703 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2704 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2705 ac->oc[1].m4ac.sbr = -1;
2706 ac->oc[1].m4ac.ps = -1;
2708 if (!hdr_info.crc_absent)
2714 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2715 int *got_frame_ptr, GetBitContext *gb)
2717 AACContext *ac = avctx->priv_data;
2718 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2719 ChannelElement *che;
2721 int samples = m4ac->frame_length_short ? 960 : 1024;
2722 int chan_config = m4ac->chan_config;
2723 int aot = m4ac->object_type;
2725 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2730 if ((err = frame_configure_elements(avctx)) < 0)
2733 // The FF_PROFILE_AAC_* defines are all object_type - 1
2734 // This may lead to an undefined profile being signaled
2735 ac->avctx->profile = aot - 1;
2737 ac->tags_mapped = 0;
2739 if (chan_config < 0 || chan_config >= 8) {
2740 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2742 return AVERROR_INVALIDDATA;
2744 for (i = 0; i < tags_per_config[chan_config]; i++) {
2745 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2746 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2747 if (!(che=get_che(ac, elem_type, elem_id))) {
2748 av_log(ac->avctx, AV_LOG_ERROR,
2749 "channel element %d.%d is not allocated\n",
2750 elem_type, elem_id);
2751 return AVERROR_INVALIDDATA;
2753 if (aot != AOT_ER_AAC_ELD)
2755 switch (elem_type) {
2757 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2760 err = decode_cpe(ac, gb, che);
2763 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2770 spectral_to_sample(ac);
2772 ac->frame->nb_samples = samples;
2773 ac->frame->sample_rate = avctx->sample_rate;
2776 skip_bits_long(gb, get_bits_left(gb));
2780 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2781 int *got_frame_ptr, GetBitContext *gb)
2783 AACContext *ac = avctx->priv_data;
2784 ChannelElement *che = NULL, *che_prev = NULL;
2785 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2787 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2791 if (show_bits(gb, 12) == 0xfff) {
2792 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2793 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2796 if (ac->oc[1].m4ac.sampling_index > 12) {
2797 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2798 err = AVERROR_INVALIDDATA;
2803 if ((err = frame_configure_elements(avctx)) < 0)
2806 // The FF_PROFILE_AAC_* defines are all object_type - 1
2807 // This may lead to an undefined profile being signaled
2808 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2810 ac->tags_mapped = 0;
2812 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2813 elem_id = get_bits(gb, 4);
2815 if (elem_type < TYPE_DSE) {
2816 if (!(che=get_che(ac, elem_type, elem_id))) {
2817 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2818 elem_type, elem_id);
2819 err = AVERROR_INVALIDDATA;
2825 switch (elem_type) {
2828 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2833 err = decode_cpe(ac, gb, che);
2838 err = decode_cce(ac, gb, che);
2842 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2847 err = skip_data_stream_element(ac, gb);
2851 uint8_t layout_map[MAX_ELEM_ID*4][3];
2853 push_output_configuration(ac);
2854 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2860 av_log(avctx, AV_LOG_ERROR,
2861 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2862 pop_output_configuration(ac);
2864 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2872 elem_id += get_bits(gb, 8) - 1;
2873 if (get_bits_left(gb) < 8 * elem_id) {
2874 av_log(avctx, AV_LOG_ERROR, overread_err);
2875 err = AVERROR_INVALIDDATA;
2879 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2880 err = 0; /* FIXME */
2884 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
2889 elem_type_prev = elem_type;
2894 if (get_bits_left(gb) < 3) {
2895 av_log(avctx, AV_LOG_ERROR, overread_err);
2896 err = AVERROR_INVALIDDATA;
2901 spectral_to_sample(ac);
2903 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2904 samples <<= multiplier;
2906 if (ac->oc[1].status && audio_found) {
2907 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2908 avctx->frame_size = samples;
2909 ac->oc[1].status = OC_LOCKED;
2913 ac->frame->nb_samples = samples;
2914 ac->frame->sample_rate = avctx->sample_rate;
2916 *got_frame_ptr = !!samples;
2920 pop_output_configuration(ac);
2924 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2925 int *got_frame_ptr, AVPacket *avpkt)
2927 AACContext *ac = avctx->priv_data;
2928 const uint8_t *buf = avpkt->data;
2929 int buf_size = avpkt->size;
2934 int new_extradata_size;
2935 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2936 AV_PKT_DATA_NEW_EXTRADATA,
2937 &new_extradata_size);
2939 if (new_extradata) {
2940 av_free(avctx->extradata);
2941 avctx->extradata = av_mallocz(new_extradata_size +
2942 FF_INPUT_BUFFER_PADDING_SIZE);
2943 if (!avctx->extradata)
2944 return AVERROR(ENOMEM);
2945 avctx->extradata_size = new_extradata_size;
2946 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2947 push_output_configuration(ac);
2948 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2950 avctx->extradata_size*8, 1) < 0) {
2951 pop_output_configuration(ac);
2952 return AVERROR_INVALIDDATA;
2956 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
2959 switch (ac->oc[1].m4ac.object_type) {
2961 case AOT_ER_AAC_LTP:
2963 case AOT_ER_AAC_ELD:
2964 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
2967 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb);
2972 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2973 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2974 if (buf[buf_offset])
2977 return buf_size > buf_offset ? buf_consumed : buf_size;
2980 static av_cold int aac_decode_close(AVCodecContext *avctx)
2982 AACContext *ac = avctx->priv_data;
2985 for (i = 0; i < MAX_ELEM_ID; i++) {
2986 for (type = 0; type < 4; type++) {
2987 if (ac->che[type][i])
2988 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2989 av_freep(&ac->che[type][i]);
2993 ff_mdct_end(&ac->mdct);
2994 ff_mdct_end(&ac->mdct_small);
2995 ff_mdct_end(&ac->mdct_ld);
2996 ff_mdct_end(&ac->mdct_ltp);
2997 ff_imdct15_uninit(&ac->mdct480);
3002 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3004 struct LATMContext {
3005 AACContext aac_ctx; ///< containing AACContext
3006 int initialized; ///< initilized after a valid extradata was seen
3009 int audio_mux_version_A; ///< LATM syntax version
3010 int frame_length_type; ///< 0/1 variable/fixed frame length
3011 int frame_length; ///< frame length for fixed frame length
3014 static inline uint32_t latm_get_value(GetBitContext *b)
3016 int length = get_bits(b, 2);
3018 return get_bits_long(b, (length+1)*8);
3021 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3022 GetBitContext *gb, int asclen)
3024 AACContext *ac = &latmctx->aac_ctx;
3025 AVCodecContext *avctx = ac->avctx;
3026 MPEG4AudioConfig m4ac = { 0 };
3027 int config_start_bit = get_bits_count(gb);
3028 int sync_extension = 0;
3029 int bits_consumed, esize;
3033 asclen = FFMIN(asclen, get_bits_left(gb));
3035 asclen = get_bits_left(gb);
3037 if (config_start_bit % 8) {
3038 avpriv_request_sample(latmctx->aac_ctx.avctx,
3039 "Non-byte-aligned audio-specific config");
3040 return AVERROR_PATCHWELCOME;
3043 return AVERROR_INVALIDDATA;
3044 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3045 gb->buffer + (config_start_bit / 8),
3046 asclen, sync_extension);
3048 if (bits_consumed < 0)
3049 return AVERROR_INVALIDDATA;
3051 if (!latmctx->initialized ||
3052 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3053 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3055 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3056 latmctx->initialized = 0;
3058 esize = (bits_consumed+7) / 8;
3060 if (avctx->extradata_size < esize) {
3061 av_free(avctx->extradata);
3062 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3063 if (!avctx->extradata)
3064 return AVERROR(ENOMEM);
3067 avctx->extradata_size = esize;
3068 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3069 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3071 skip_bits_long(gb, bits_consumed);
3073 return bits_consumed;
3076 static int read_stream_mux_config(struct LATMContext *latmctx,
3079 int ret, audio_mux_version = get_bits(gb, 1);
3081 latmctx->audio_mux_version_A = 0;
3082 if (audio_mux_version)
3083 latmctx->audio_mux_version_A = get_bits(gb, 1);
3085 if (!latmctx->audio_mux_version_A) {
3087 if (audio_mux_version)
3088 latm_get_value(gb); // taraFullness
3090 skip_bits(gb, 1); // allStreamSameTimeFraming
3091 skip_bits(gb, 6); // numSubFrames
3093 if (get_bits(gb, 4)) { // numPrograms
3094 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3095 return AVERROR_PATCHWELCOME;
3098 // for each program (which there is only on in DVB)
3100 // for each layer (which there is only on in DVB)
3101 if (get_bits(gb, 3)) { // numLayer
3102 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3103 return AVERROR_PATCHWELCOME;
3106 // for all but first stream: use_same_config = get_bits(gb, 1);
3107 if (!audio_mux_version) {
3108 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3111 int ascLen = latm_get_value(gb);
3112 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3115 skip_bits_long(gb, ascLen);
3118 latmctx->frame_length_type = get_bits(gb, 3);
3119 switch (latmctx->frame_length_type) {
3121 skip_bits(gb, 8); // latmBufferFullness
3124 latmctx->frame_length = get_bits(gb, 9);
3129 skip_bits(gb, 6); // CELP frame length table index
3133 skip_bits(gb, 1); // HVXC frame length table index
3137 if (get_bits(gb, 1)) { // other data
3138 if (audio_mux_version) {
3139 latm_get_value(gb); // other_data_bits
3143 esc = get_bits(gb, 1);
3149 if (get_bits(gb, 1)) // crc present
3150 skip_bits(gb, 8); // config_crc
3156 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3160 if (ctx->frame_length_type == 0) {
3161 int mux_slot_length = 0;
3163 tmp = get_bits(gb, 8);
3164 mux_slot_length += tmp;
3165 } while (tmp == 255);
3166 return mux_slot_length;
3167 } else if (ctx->frame_length_type == 1) {
3168 return ctx->frame_length;
3169 } else if (ctx->frame_length_type == 3 ||
3170 ctx->frame_length_type == 5 ||
3171 ctx->frame_length_type == 7) {
3172 skip_bits(gb, 2); // mux_slot_length_coded
3177 static int read_audio_mux_element(struct LATMContext *latmctx,
3181 uint8_t use_same_mux = get_bits(gb, 1);
3182 if (!use_same_mux) {
3183 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3185 } else if (!latmctx->aac_ctx.avctx->extradata) {
3186 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3187 "no decoder config found\n");
3188 return AVERROR(EAGAIN);
3190 if (latmctx->audio_mux_version_A == 0) {
3191 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3192 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3193 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3194 return AVERROR_INVALIDDATA;
3195 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3196 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3197 "frame length mismatch %d << %d\n",
3198 mux_slot_length_bytes * 8, get_bits_left(gb));
3199 return AVERROR_INVALIDDATA;
3206 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3207 int *got_frame_ptr, AVPacket *avpkt)
3209 struct LATMContext *latmctx = avctx->priv_data;
3213 if ((err = init_get_bits(&gb, avpkt->data, avpkt->size * 8)) < 0)
3216 // check for LOAS sync word
3217 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3218 return AVERROR_INVALIDDATA;
3220 muxlength = get_bits(&gb, 13) + 3;
3221 // not enough data, the parser should have sorted this
3222 if (muxlength > avpkt->size)
3223 return AVERROR_INVALIDDATA;
3225 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3228 if (!latmctx->initialized) {
3229 if (!avctx->extradata) {
3233 push_output_configuration(&latmctx->aac_ctx);
3234 if ((err = decode_audio_specific_config(
3235 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3236 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3237 pop_output_configuration(&latmctx->aac_ctx);
3240 latmctx->initialized = 1;
3244 if (show_bits(&gb, 12) == 0xfff) {
3245 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3246 "ADTS header detected, probably as result of configuration "
3248 return AVERROR_INVALIDDATA;
3251 switch (latmctx->aac_ctx.oc[1].m4ac.object_type) {
3253 case AOT_ER_AAC_LTP:
3255 case AOT_ER_AAC_ELD:
3256 err = aac_decode_er_frame(avctx, out, got_frame_ptr, &gb);
3259 err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb);
3267 static av_cold int latm_decode_init(AVCodecContext *avctx)
3269 struct LATMContext *latmctx = avctx->priv_data;
3270 int ret = aac_decode_init(avctx);
3272 if (avctx->extradata_size > 0)
3273 latmctx->initialized = !ret;
3279 AVCodec ff_aac_decoder = {
3281 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3282 .type = AVMEDIA_TYPE_AUDIO,
3283 .id = AV_CODEC_ID_AAC,
3284 .priv_data_size = sizeof(AACContext),
3285 .init = aac_decode_init,
3286 .close = aac_decode_close,
3287 .decode = aac_decode_frame,
3288 .sample_fmts = (const enum AVSampleFormat[]) {
3289 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3291 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3292 .channel_layouts = aac_channel_layout,
3296 Note: This decoder filter is intended to decode LATM streams transferred
3297 in MPEG transport streams which only contain one program.
3298 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3300 AVCodec ff_aac_latm_decoder = {
3302 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3303 .type = AVMEDIA_TYPE_AUDIO,
3304 .id = AV_CODEC_ID_AAC_LATM,
3305 .priv_data_size = sizeof(struct LATMContext),
3306 .init = latm_decode_init,
3307 .close = aac_decode_close,
3308 .decode = latm_decode_frame,
3309 .sample_fmts = (const enum AVSampleFormat[]) {
3310 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3312 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3313 .channel_layouts = aac_channel_layout,