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"
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"
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)
146 if (*channels >= MAX_CHANNELS)
147 return AVERROR_INVALIDDATA;
149 if (!ac->che[type][id]) {
150 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
151 return AVERROR(ENOMEM);
152 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
154 if (type != TYPE_CCE) {
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 layout_map[0][0] = syn_ele;
629 layout_map[0][1] = get_bits(gb, 4);
630 layout_map[0][2] = type;
636 * Decode program configuration element; reference: table 4.2.
638 * @return Returns error status. 0 - OK, !0 - error
640 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
641 uint8_t (*layout_map)[3],
644 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
649 skip_bits(gb, 2); // object_type
651 sampling_index = get_bits(gb, 4);
652 if (m4ac->sampling_index != sampling_index)
653 av_log(avctx, AV_LOG_WARNING,
654 "Sample rate index in program config element does not "
655 "match the sample rate index configured by the container.\n");
657 num_front = get_bits(gb, 4);
658 num_side = get_bits(gb, 4);
659 num_back = get_bits(gb, 4);
660 num_lfe = get_bits(gb, 2);
661 num_assoc_data = get_bits(gb, 3);
662 num_cc = get_bits(gb, 4);
665 skip_bits(gb, 4); // mono_mixdown_tag
667 skip_bits(gb, 4); // stereo_mixdown_tag
670 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
672 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
674 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
676 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
678 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
681 skip_bits_long(gb, 4 * num_assoc_data);
683 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
688 /* comment field, first byte is length */
689 comment_len = get_bits(gb, 8) * 8;
690 if (get_bits_left(gb) < comment_len) {
691 av_log(avctx, AV_LOG_ERROR, overread_err);
692 return AVERROR_INVALIDDATA;
694 skip_bits_long(gb, comment_len);
699 * Decode GA "General Audio" specific configuration; reference: table 4.1.
701 * @param ac pointer to AACContext, may be null
702 * @param avctx pointer to AVCCodecContext, used for logging
704 * @return Returns error status. 0 - OK, !0 - error
706 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
708 MPEG4AudioConfig *m4ac,
711 int extension_flag, ret, ep_config, res_flags;
712 uint8_t layout_map[MAX_ELEM_ID*4][3];
715 if (get_bits1(gb)) { // frameLengthFlag
716 avpriv_request_sample(avctx, "960/120 MDCT window");
717 return AVERROR_PATCHWELCOME;
720 if (get_bits1(gb)) // dependsOnCoreCoder
721 skip_bits(gb, 14); // coreCoderDelay
722 extension_flag = get_bits1(gb);
724 if (m4ac->object_type == AOT_AAC_SCALABLE ||
725 m4ac->object_type == AOT_ER_AAC_SCALABLE)
726 skip_bits(gb, 3); // layerNr
728 if (channel_config == 0) {
729 skip_bits(gb, 4); // element_instance_tag
730 tags = decode_pce(avctx, m4ac, layout_map, gb);
734 if ((ret = set_default_channel_config(avctx, layout_map,
735 &tags, channel_config)))
739 if (count_channels(layout_map, tags) > 1) {
741 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
744 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
747 if (extension_flag) {
748 switch (m4ac->object_type) {
750 skip_bits(gb, 5); // numOfSubFrame
751 skip_bits(gb, 11); // layer_length
755 case AOT_ER_AAC_SCALABLE:
757 res_flags = get_bits(gb, 3);
759 avpriv_report_missing_feature(avctx,
760 "AAC data resilience (flags %x)",
762 return AVERROR_PATCHWELCOME;
766 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
768 switch (m4ac->object_type) {
771 case AOT_ER_AAC_SCALABLE:
773 ep_config = get_bits(gb, 2);
775 avpriv_report_missing_feature(avctx,
776 "epConfig %d", ep_config);
777 return AVERROR_PATCHWELCOME;
783 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
785 MPEG4AudioConfig *m4ac,
788 int ret, ep_config, res_flags;
789 uint8_t layout_map[MAX_ELEM_ID*4][3];
791 const int ELDEXT_TERM = 0;
796 if (get_bits1(gb)) { // frameLengthFlag
797 avpriv_request_sample(avctx, "960/120 MDCT window");
798 return AVERROR_PATCHWELCOME;
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 ff_fmt_convert_init(&ac->fmt_conv, avctx);
1046 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1048 ac->random_state = 0x1f2e3d4c;
1052 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1053 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1054 ff_aac_scalefactor_bits,
1055 sizeof(ff_aac_scalefactor_bits[0]),
1056 sizeof(ff_aac_scalefactor_bits[0]),
1057 ff_aac_scalefactor_code,
1058 sizeof(ff_aac_scalefactor_code[0]),
1059 sizeof(ff_aac_scalefactor_code[0]),
1062 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1063 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1064 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1065 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1066 // window initialization
1067 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1068 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1069 ff_init_ff_sine_windows(10);
1070 ff_init_ff_sine_windows( 9);
1071 ff_init_ff_sine_windows( 7);
1079 * Skip data_stream_element; reference: table 4.10.
1081 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1083 int byte_align = get_bits1(gb);
1084 int count = get_bits(gb, 8);
1086 count += get_bits(gb, 8);
1090 if (get_bits_left(gb) < 8 * count) {
1091 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1092 return AVERROR_INVALIDDATA;
1094 skip_bits_long(gb, 8 * count);
1098 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1102 if (get_bits1(gb)) {
1103 ics->predictor_reset_group = get_bits(gb, 5);
1104 if (ics->predictor_reset_group == 0 ||
1105 ics->predictor_reset_group > 30) {
1106 av_log(ac->avctx, AV_LOG_ERROR,
1107 "Invalid Predictor Reset Group.\n");
1108 return AVERROR_INVALIDDATA;
1111 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1112 ics->prediction_used[sfb] = get_bits1(gb);
1118 * Decode Long Term Prediction data; reference: table 4.xx.
1120 static void decode_ltp(LongTermPrediction *ltp,
1121 GetBitContext *gb, uint8_t max_sfb)
1125 ltp->lag = get_bits(gb, 11);
1126 ltp->coef = ltp_coef[get_bits(gb, 3)];
1127 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1128 ltp->used[sfb] = get_bits1(gb);
1132 * Decode Individual Channel Stream info; reference: table 4.6.
1134 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1137 int aot = ac->oc[1].m4ac.object_type;
1138 if (aot != AOT_ER_AAC_ELD) {
1139 if (get_bits1(gb)) {
1140 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1141 return AVERROR_INVALIDDATA;
1143 ics->window_sequence[1] = ics->window_sequence[0];
1144 ics->window_sequence[0] = get_bits(gb, 2);
1145 if (aot == AOT_ER_AAC_LD &&
1146 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1147 av_log(ac->avctx, AV_LOG_ERROR,
1148 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1149 "window sequence %d found.\n", ics->window_sequence[0]);
1150 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1151 return AVERROR_INVALIDDATA;
1153 ics->use_kb_window[1] = ics->use_kb_window[0];
1154 ics->use_kb_window[0] = get_bits1(gb);
1156 ics->num_window_groups = 1;
1157 ics->group_len[0] = 1;
1158 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1160 ics->max_sfb = get_bits(gb, 4);
1161 for (i = 0; i < 7; i++) {
1162 if (get_bits1(gb)) {
1163 ics->group_len[ics->num_window_groups - 1]++;
1165 ics->num_window_groups++;
1166 ics->group_len[ics->num_window_groups - 1] = 1;
1169 ics->num_windows = 8;
1170 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1171 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1172 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1173 ics->predictor_present = 0;
1175 ics->max_sfb = get_bits(gb, 6);
1176 ics->num_windows = 1;
1177 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1178 ics->swb_offset = ff_swb_offset_512[ac->oc[1].m4ac.sampling_index];
1179 ics->num_swb = ff_aac_num_swb_512[ac->oc[1].m4ac.sampling_index];
1180 ics->tns_max_bands = ff_tns_max_bands_512[ac->oc[1].m4ac.sampling_index];
1181 if (!ics->num_swb || !ics->swb_offset)
1184 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1185 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1186 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1188 if (aot != AOT_ER_AAC_ELD) {
1189 ics->predictor_present = get_bits1(gb);
1190 ics->predictor_reset_group = 0;
1192 if (ics->predictor_present) {
1193 if (aot == AOT_AAC_MAIN) {
1194 if (decode_prediction(ac, ics, gb)) {
1195 return AVERROR_INVALIDDATA;
1197 } else if (aot == AOT_AAC_LC ||
1198 aot == AOT_ER_AAC_LC) {
1199 av_log(ac->avctx, AV_LOG_ERROR,
1200 "Prediction is not allowed in AAC-LC.\n");
1201 return AVERROR_INVALIDDATA;
1203 if (aot == AOT_ER_AAC_LD) {
1204 av_log(ac->avctx, AV_LOG_ERROR,
1205 "LTP in ER AAC LD not yet implemented.\n");
1206 return AVERROR_PATCHWELCOME;
1208 if ((ics->ltp.present = get_bits(gb, 1)))
1209 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1214 if (ics->max_sfb > ics->num_swb) {
1215 av_log(ac->avctx, AV_LOG_ERROR,
1216 "Number of scalefactor bands in group (%d) "
1217 "exceeds limit (%d).\n",
1218 ics->max_sfb, ics->num_swb);
1219 return AVERROR_INVALIDDATA;
1226 * Decode band types (section_data payload); reference: table 4.46.
1228 * @param band_type array of the used band type
1229 * @param band_type_run_end array of the last scalefactor band of a band type run
1231 * @return Returns error status. 0 - OK, !0 - error
1233 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1234 int band_type_run_end[120], GetBitContext *gb,
1235 IndividualChannelStream *ics)
1238 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1239 for (g = 0; g < ics->num_window_groups; g++) {
1241 while (k < ics->max_sfb) {
1242 uint8_t sect_end = k;
1244 int sect_band_type = get_bits(gb, 4);
1245 if (sect_band_type == 12) {
1246 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1247 return AVERROR_INVALIDDATA;
1250 sect_len_incr = get_bits(gb, bits);
1251 sect_end += sect_len_incr;
1252 if (get_bits_left(gb) < 0) {
1253 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1254 return AVERROR_INVALIDDATA;
1256 if (sect_end > ics->max_sfb) {
1257 av_log(ac->avctx, AV_LOG_ERROR,
1258 "Number of bands (%d) exceeds limit (%d).\n",
1259 sect_end, ics->max_sfb);
1260 return AVERROR_INVALIDDATA;
1262 } while (sect_len_incr == (1 << bits) - 1);
1263 for (; k < sect_end; k++) {
1264 band_type [idx] = sect_band_type;
1265 band_type_run_end[idx++] = sect_end;
1273 * Decode scalefactors; reference: table 4.47.
1275 * @param global_gain first scalefactor value as scalefactors are differentially coded
1276 * @param band_type array of the used band type
1277 * @param band_type_run_end array of the last scalefactor band of a band type run
1278 * @param sf array of scalefactors or intensity stereo positions
1280 * @return Returns error status. 0 - OK, !0 - error
1282 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1283 unsigned int global_gain,
1284 IndividualChannelStream *ics,
1285 enum BandType band_type[120],
1286 int band_type_run_end[120])
1289 int offset[3] = { global_gain, global_gain - 90, 0 };
1292 for (g = 0; g < ics->num_window_groups; g++) {
1293 for (i = 0; i < ics->max_sfb;) {
1294 int run_end = band_type_run_end[idx];
1295 if (band_type[idx] == ZERO_BT) {
1296 for (; i < run_end; i++, idx++)
1298 } else if ((band_type[idx] == INTENSITY_BT) ||
1299 (band_type[idx] == INTENSITY_BT2)) {
1300 for (; i < run_end; i++, idx++) {
1301 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1302 clipped_offset = av_clip(offset[2], -155, 100);
1303 if (offset[2] != clipped_offset) {
1304 avpriv_request_sample(ac->avctx,
1305 "If you heard an audible artifact, there may be a bug in the decoder. "
1306 "Clipped intensity stereo position (%d -> %d)",
1307 offset[2], clipped_offset);
1309 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1311 } else if (band_type[idx] == NOISE_BT) {
1312 for (; i < run_end; i++, idx++) {
1313 if (noise_flag-- > 0)
1314 offset[1] += get_bits(gb, 9) - 256;
1316 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1317 clipped_offset = av_clip(offset[1], -100, 155);
1318 if (offset[1] != clipped_offset) {
1319 avpriv_request_sample(ac->avctx,
1320 "If you heard an audible artifact, there may be a bug in the decoder. "
1321 "Clipped noise gain (%d -> %d)",
1322 offset[1], clipped_offset);
1324 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1327 for (; i < run_end; i++, idx++) {
1328 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1329 if (offset[0] > 255U) {
1330 av_log(ac->avctx, AV_LOG_ERROR,
1331 "Scalefactor (%d) out of range.\n", offset[0]);
1332 return AVERROR_INVALIDDATA;
1334 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1343 * Decode pulse data; reference: table 4.7.
1345 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1346 const uint16_t *swb_offset, int num_swb)
1349 pulse->num_pulse = get_bits(gb, 2) + 1;
1350 pulse_swb = get_bits(gb, 6);
1351 if (pulse_swb >= num_swb)
1353 pulse->pos[0] = swb_offset[pulse_swb];
1354 pulse->pos[0] += get_bits(gb, 5);
1355 if (pulse->pos[0] > 1023)
1357 pulse->amp[0] = get_bits(gb, 4);
1358 for (i = 1; i < pulse->num_pulse; i++) {
1359 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1360 if (pulse->pos[i] > 1023)
1362 pulse->amp[i] = get_bits(gb, 4);
1368 * Decode Temporal Noise Shaping data; reference: table 4.48.
1370 * @return Returns error status. 0 - OK, !0 - error
1372 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1373 GetBitContext *gb, const IndividualChannelStream *ics)
1375 int w, filt, i, coef_len, coef_res, coef_compress;
1376 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1377 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1378 for (w = 0; w < ics->num_windows; w++) {
1379 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1380 coef_res = get_bits1(gb);
1382 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1384 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1386 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1387 av_log(ac->avctx, AV_LOG_ERROR,
1388 "TNS filter order %d is greater than maximum %d.\n",
1389 tns->order[w][filt], tns_max_order);
1390 tns->order[w][filt] = 0;
1391 return AVERROR_INVALIDDATA;
1393 if (tns->order[w][filt]) {
1394 tns->direction[w][filt] = get_bits1(gb);
1395 coef_compress = get_bits1(gb);
1396 coef_len = coef_res + 3 - coef_compress;
1397 tmp2_idx = 2 * coef_compress + coef_res;
1399 for (i = 0; i < tns->order[w][filt]; i++)
1400 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1409 * Decode Mid/Side data; reference: table 4.54.
1411 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1412 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1413 * [3] reserved for scalable AAC
1415 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1419 if (ms_present == 1) {
1421 idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1423 cpe->ms_mask[idx] = get_bits1(gb);
1424 } else if (ms_present == 2) {
1425 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1430 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1434 *dst++ = v[idx & 15] * s;
1435 *dst++ = v[idx>>4 & 15] * s;
1441 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1445 *dst++ = v[idx & 3] * s;
1446 *dst++ = v[idx>>2 & 3] * s;
1447 *dst++ = v[idx>>4 & 3] * s;
1448 *dst++ = v[idx>>6 & 3] * s;
1454 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1455 unsigned sign, const float *scale)
1457 union av_intfloat32 s0, s1;
1459 s0.f = s1.f = *scale;
1460 s0.i ^= sign >> 1 << 31;
1463 *dst++ = v[idx & 15] * s0.f;
1464 *dst++ = v[idx>>4 & 15] * s1.f;
1471 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1472 unsigned sign, const float *scale)
1474 unsigned nz = idx >> 12;
1475 union av_intfloat32 s = { .f = *scale };
1476 union av_intfloat32 t;
1478 t.i = s.i ^ (sign & 1U<<31);
1479 *dst++ = v[idx & 3] * t.f;
1481 sign <<= nz & 1; nz >>= 1;
1482 t.i = s.i ^ (sign & 1U<<31);
1483 *dst++ = v[idx>>2 & 3] * t.f;
1485 sign <<= nz & 1; nz >>= 1;
1486 t.i = s.i ^ (sign & 1U<<31);
1487 *dst++ = v[idx>>4 & 3] * t.f;
1490 t.i = s.i ^ (sign & 1U<<31);
1491 *dst++ = v[idx>>6 & 3] * t.f;
1498 * Decode spectral data; reference: table 4.50.
1499 * Dequantize and scale spectral data; reference: 4.6.3.3.
1501 * @param coef array of dequantized, scaled spectral data
1502 * @param sf array of scalefactors or intensity stereo positions
1503 * @param pulse_present set if pulses are present
1504 * @param pulse pointer to pulse data struct
1505 * @param band_type array of the used band type
1507 * @return Returns error status. 0 - OK, !0 - error
1509 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1510 GetBitContext *gb, const float sf[120],
1511 int pulse_present, const Pulse *pulse,
1512 const IndividualChannelStream *ics,
1513 enum BandType band_type[120])
1515 int i, k, g, idx = 0;
1516 const int c = 1024 / ics->num_windows;
1517 const uint16_t *offsets = ics->swb_offset;
1518 float *coef_base = coef;
1520 for (g = 0; g < ics->num_windows; g++)
1521 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1522 sizeof(float) * (c - offsets[ics->max_sfb]));
1524 for (g = 0; g < ics->num_window_groups; g++) {
1525 unsigned g_len = ics->group_len[g];
1527 for (i = 0; i < ics->max_sfb; i++, idx++) {
1528 const unsigned cbt_m1 = band_type[idx] - 1;
1529 float *cfo = coef + offsets[i];
1530 int off_len = offsets[i + 1] - offsets[i];
1533 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1534 for (group = 0; group < g_len; group++, cfo+=128) {
1535 memset(cfo, 0, off_len * sizeof(float));
1537 } else if (cbt_m1 == NOISE_BT - 1) {
1538 for (group = 0; group < g_len; group++, cfo+=128) {
1542 for (k = 0; k < off_len; k++) {
1543 ac->random_state = lcg_random(ac->random_state);
1544 cfo[k] = ac->random_state;
1547 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1548 scale = sf[idx] / sqrtf(band_energy);
1549 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1552 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1553 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1554 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1555 OPEN_READER(re, gb);
1557 switch (cbt_m1 >> 1) {
1559 for (group = 0; group < g_len; group++, cfo+=128) {
1567 UPDATE_CACHE(re, gb);
1568 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1569 cb_idx = cb_vector_idx[code];
1570 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1576 for (group = 0; group < g_len; group++, cfo+=128) {
1586 UPDATE_CACHE(re, gb);
1587 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1588 cb_idx = cb_vector_idx[code];
1589 nnz = cb_idx >> 8 & 15;
1590 bits = nnz ? GET_CACHE(re, gb) : 0;
1591 LAST_SKIP_BITS(re, gb, nnz);
1592 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1598 for (group = 0; group < g_len; group++, cfo+=128) {
1606 UPDATE_CACHE(re, gb);
1607 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1608 cb_idx = cb_vector_idx[code];
1609 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1616 for (group = 0; group < g_len; group++, cfo+=128) {
1626 UPDATE_CACHE(re, gb);
1627 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1628 cb_idx = cb_vector_idx[code];
1629 nnz = cb_idx >> 8 & 15;
1630 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1631 LAST_SKIP_BITS(re, gb, nnz);
1632 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1638 for (group = 0; group < g_len; group++, cfo+=128) {
1640 uint32_t *icf = (uint32_t *) cf;
1650 UPDATE_CACHE(re, gb);
1651 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1659 cb_idx = cb_vector_idx[code];
1662 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1663 LAST_SKIP_BITS(re, gb, nnz);
1665 for (j = 0; j < 2; j++) {
1669 /* The total length of escape_sequence must be < 22 bits according
1670 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1671 UPDATE_CACHE(re, gb);
1672 b = GET_CACHE(re, gb);
1673 b = 31 - av_log2(~b);
1676 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1677 return AVERROR_INVALIDDATA;
1680 SKIP_BITS(re, gb, b + 1);
1682 n = (1 << b) + SHOW_UBITS(re, gb, b);
1683 LAST_SKIP_BITS(re, gb, b);
1684 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1687 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1688 *icf++ = (bits & 1U<<31) | v;
1695 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1699 CLOSE_READER(re, gb);
1705 if (pulse_present) {
1707 for (i = 0; i < pulse->num_pulse; i++) {
1708 float co = coef_base[ pulse->pos[i] ];
1709 while (offsets[idx + 1] <= pulse->pos[i])
1711 if (band_type[idx] != NOISE_BT && sf[idx]) {
1712 float ico = -pulse->amp[i];
1715 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1717 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1724 static av_always_inline float flt16_round(float pf)
1726 union av_intfloat32 tmp;
1728 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1732 static av_always_inline float flt16_even(float pf)
1734 union av_intfloat32 tmp;
1736 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1740 static av_always_inline float flt16_trunc(float pf)
1742 union av_intfloat32 pun;
1744 pun.i &= 0xFFFF0000U;
1748 static av_always_inline void predict(PredictorState *ps, float *coef,
1751 const float a = 0.953125; // 61.0 / 64
1752 const float alpha = 0.90625; // 29.0 / 32
1756 float r0 = ps->r0, r1 = ps->r1;
1757 float cor0 = ps->cor0, cor1 = ps->cor1;
1758 float var0 = ps->var0, var1 = ps->var1;
1760 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1761 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1763 pv = flt16_round(k1 * r0 + k2 * r1);
1770 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1771 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1772 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1773 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1775 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1776 ps->r0 = flt16_trunc(a * e0);
1780 * Apply AAC-Main style frequency domain prediction.
1782 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1786 if (!sce->ics.predictor_initialized) {
1787 reset_all_predictors(sce->predictor_state);
1788 sce->ics.predictor_initialized = 1;
1791 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1793 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1795 for (k = sce->ics.swb_offset[sfb];
1796 k < sce->ics.swb_offset[sfb + 1];
1798 predict(&sce->predictor_state[k], &sce->coeffs[k],
1799 sce->ics.predictor_present &&
1800 sce->ics.prediction_used[sfb]);
1803 if (sce->ics.predictor_reset_group)
1804 reset_predictor_group(sce->predictor_state,
1805 sce->ics.predictor_reset_group);
1807 reset_all_predictors(sce->predictor_state);
1811 * Decode an individual_channel_stream payload; reference: table 4.44.
1813 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1814 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1816 * @return Returns error status. 0 - OK, !0 - error
1818 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1819 GetBitContext *gb, int common_window, int scale_flag)
1822 TemporalNoiseShaping *tns = &sce->tns;
1823 IndividualChannelStream *ics = &sce->ics;
1824 float *out = sce->coeffs;
1825 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1828 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1829 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1830 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1831 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1832 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1834 /* This assignment is to silence a GCC warning about the variable being used
1835 * uninitialized when in fact it always is.
1837 pulse.num_pulse = 0;
1839 global_gain = get_bits(gb, 8);
1841 if (!common_window && !scale_flag) {
1842 if (decode_ics_info(ac, ics, gb) < 0)
1843 return AVERROR_INVALIDDATA;
1846 if ((ret = decode_band_types(ac, sce->band_type,
1847 sce->band_type_run_end, gb, ics)) < 0)
1849 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1850 sce->band_type, sce->band_type_run_end)) < 0)
1855 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1856 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1857 av_log(ac->avctx, AV_LOG_ERROR,
1858 "Pulse tool not allowed in eight short sequence.\n");
1859 return AVERROR_INVALIDDATA;
1861 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1862 av_log(ac->avctx, AV_LOG_ERROR,
1863 "Pulse data corrupt or invalid.\n");
1864 return AVERROR_INVALIDDATA;
1867 tns->present = get_bits1(gb);
1868 if (tns->present && !er_syntax)
1869 if (decode_tns(ac, tns, gb, ics) < 0)
1870 return AVERROR_INVALIDDATA;
1871 if (!eld_syntax && get_bits1(gb)) {
1872 avpriv_request_sample(ac->avctx, "SSR");
1873 return AVERROR_PATCHWELCOME;
1875 // I see no textual basis in the spec for this occuring after SSR gain
1876 // control, but this is what both reference and real implmentations do
1877 if (tns->present && er_syntax)
1878 if (decode_tns(ac, tns, gb, ics) < 0)
1879 return AVERROR_INVALIDDATA;
1882 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1883 &pulse, ics, sce->band_type) < 0)
1884 return AVERROR_INVALIDDATA;
1886 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1887 apply_prediction(ac, sce);
1893 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1895 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1897 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1898 float *ch0 = cpe->ch[0].coeffs;
1899 float *ch1 = cpe->ch[1].coeffs;
1900 int g, i, group, idx = 0;
1901 const uint16_t *offsets = ics->swb_offset;
1902 for (g = 0; g < ics->num_window_groups; g++) {
1903 for (i = 0; i < ics->max_sfb; i++, idx++) {
1904 if (cpe->ms_mask[idx] &&
1905 cpe->ch[0].band_type[idx] < NOISE_BT &&
1906 cpe->ch[1].band_type[idx] < NOISE_BT) {
1907 for (group = 0; group < ics->group_len[g]; group++) {
1908 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1909 ch1 + group * 128 + offsets[i],
1910 offsets[i+1] - offsets[i]);
1914 ch0 += ics->group_len[g] * 128;
1915 ch1 += ics->group_len[g] * 128;
1920 * intensity stereo decoding; reference: 4.6.8.2.3
1922 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1923 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1924 * [3] reserved for scalable AAC
1926 static void apply_intensity_stereo(AACContext *ac,
1927 ChannelElement *cpe, int ms_present)
1929 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1930 SingleChannelElement *sce1 = &cpe->ch[1];
1931 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1932 const uint16_t *offsets = ics->swb_offset;
1933 int g, group, i, idx = 0;
1936 for (g = 0; g < ics->num_window_groups; g++) {
1937 for (i = 0; i < ics->max_sfb;) {
1938 if (sce1->band_type[idx] == INTENSITY_BT ||
1939 sce1->band_type[idx] == INTENSITY_BT2) {
1940 const int bt_run_end = sce1->band_type_run_end[idx];
1941 for (; i < bt_run_end; i++, idx++) {
1942 c = -1 + 2 * (sce1->band_type[idx] - 14);
1944 c *= 1 - 2 * cpe->ms_mask[idx];
1945 scale = c * sce1->sf[idx];
1946 for (group = 0; group < ics->group_len[g]; group++)
1947 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1948 coef0 + group * 128 + offsets[i],
1950 offsets[i + 1] - offsets[i]);
1953 int bt_run_end = sce1->band_type_run_end[idx];
1954 idx += bt_run_end - i;
1958 coef0 += ics->group_len[g] * 128;
1959 coef1 += ics->group_len[g] * 128;
1964 * Decode a channel_pair_element; reference: table 4.4.
1966 * @return Returns error status. 0 - OK, !0 - error
1968 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1970 int i, ret, common_window, ms_present = 0;
1971 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1973 common_window = eld_syntax || get_bits1(gb);
1974 if (common_window) {
1975 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1976 return AVERROR_INVALIDDATA;
1977 i = cpe->ch[1].ics.use_kb_window[0];
1978 cpe->ch[1].ics = cpe->ch[0].ics;
1979 cpe->ch[1].ics.use_kb_window[1] = i;
1980 if (cpe->ch[1].ics.predictor_present &&
1981 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1982 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1983 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1984 ms_present = get_bits(gb, 2);
1985 if (ms_present == 3) {
1986 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1987 return AVERROR_INVALIDDATA;
1988 } else if (ms_present)
1989 decode_mid_side_stereo(cpe, gb, ms_present);
1991 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1993 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1996 if (common_window) {
1998 apply_mid_side_stereo(ac, cpe);
1999 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2000 apply_prediction(ac, &cpe->ch[0]);
2001 apply_prediction(ac, &cpe->ch[1]);
2005 apply_intensity_stereo(ac, cpe, ms_present);
2009 static const float cce_scale[] = {
2010 1.09050773266525765921, //2^(1/8)
2011 1.18920711500272106672, //2^(1/4)
2017 * Decode coupling_channel_element; reference: table 4.8.
2019 * @return Returns error status. 0 - OK, !0 - error
2021 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2027 SingleChannelElement *sce = &che->ch[0];
2028 ChannelCoupling *coup = &che->coup;
2030 coup->coupling_point = 2 * get_bits1(gb);
2031 coup->num_coupled = get_bits(gb, 3);
2032 for (c = 0; c <= coup->num_coupled; c++) {
2034 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2035 coup->id_select[c] = get_bits(gb, 4);
2036 if (coup->type[c] == TYPE_CPE) {
2037 coup->ch_select[c] = get_bits(gb, 2);
2038 if (coup->ch_select[c] == 3)
2041 coup->ch_select[c] = 2;
2043 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2045 sign = get_bits(gb, 1);
2046 scale = cce_scale[get_bits(gb, 2)];
2048 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2051 for (c = 0; c < num_gain; c++) {
2055 float gain_cache = 1.0;
2057 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2058 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2059 gain_cache = powf(scale, -gain);
2061 if (coup->coupling_point == AFTER_IMDCT) {
2062 coup->gain[c][0] = gain_cache;
2064 for (g = 0; g < sce->ics.num_window_groups; g++) {
2065 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2066 if (sce->band_type[idx] != ZERO_BT) {
2068 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2076 gain_cache = powf(scale, -t) * s;
2079 coup->gain[c][idx] = gain_cache;
2089 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2091 * @return Returns number of bytes consumed.
2093 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2097 int num_excl_chan = 0;
2100 for (i = 0; i < 7; i++)
2101 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2102 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2104 return num_excl_chan / 7;
2108 * Decode dynamic range information; reference: table 4.52.
2110 * @return Returns number of bytes consumed.
2112 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2116 int drc_num_bands = 1;
2119 /* pce_tag_present? */
2120 if (get_bits1(gb)) {
2121 che_drc->pce_instance_tag = get_bits(gb, 4);
2122 skip_bits(gb, 4); // tag_reserved_bits
2126 /* excluded_chns_present? */
2127 if (get_bits1(gb)) {
2128 n += decode_drc_channel_exclusions(che_drc, gb);
2131 /* drc_bands_present? */
2132 if (get_bits1(gb)) {
2133 che_drc->band_incr = get_bits(gb, 4);
2134 che_drc->interpolation_scheme = get_bits(gb, 4);
2136 drc_num_bands += che_drc->band_incr;
2137 for (i = 0; i < drc_num_bands; i++) {
2138 che_drc->band_top[i] = get_bits(gb, 8);
2143 /* prog_ref_level_present? */
2144 if (get_bits1(gb)) {
2145 che_drc->prog_ref_level = get_bits(gb, 7);
2146 skip_bits1(gb); // prog_ref_level_reserved_bits
2150 for (i = 0; i < drc_num_bands; i++) {
2151 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2152 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2160 * Decode extension data (incomplete); reference: table 4.51.
2162 * @param cnt length of TYPE_FIL syntactic element in bytes
2164 * @return Returns number of bytes consumed
2166 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2167 ChannelElement *che, enum RawDataBlockType elem_type)
2171 switch (get_bits(gb, 4)) { // extension type
2172 case EXT_SBR_DATA_CRC:
2176 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2178 } else if (!ac->oc[1].m4ac.sbr) {
2179 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2180 skip_bits_long(gb, 8 * cnt - 4);
2182 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2183 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2184 skip_bits_long(gb, 8 * cnt - 4);
2186 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2187 ac->oc[1].m4ac.sbr = 1;
2188 ac->oc[1].m4ac.ps = 1;
2189 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2190 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2191 ac->oc[1].status, 1);
2193 ac->oc[1].m4ac.sbr = 1;
2194 ac->avctx->profile = FF_PROFILE_AAC_HE;
2196 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2198 case EXT_DYNAMIC_RANGE:
2199 res = decode_dynamic_range(&ac->che_drc, gb);
2203 case EXT_DATA_ELEMENT:
2205 skip_bits_long(gb, 8 * cnt - 4);
2212 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2214 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2215 * @param coef spectral coefficients
2217 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2218 IndividualChannelStream *ics, int decode)
2220 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2222 int bottom, top, order, start, end, size, inc;
2223 float lpc[TNS_MAX_ORDER];
2224 float tmp[TNS_MAX_ORDER + 1];
2226 for (w = 0; w < ics->num_windows; w++) {
2227 bottom = ics->num_swb;
2228 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2230 bottom = FFMAX(0, top - tns->length[w][filt]);
2231 order = tns->order[w][filt];
2236 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2238 start = ics->swb_offset[FFMIN(bottom, mmm)];
2239 end = ics->swb_offset[FFMIN( top, mmm)];
2240 if ((size = end - start) <= 0)
2242 if (tns->direction[w][filt]) {
2252 for (m = 0; m < size; m++, start += inc)
2253 for (i = 1; i <= FFMIN(m, order); i++)
2254 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2257 for (m = 0; m < size; m++, start += inc) {
2258 tmp[0] = coef[start];
2259 for (i = 1; i <= FFMIN(m, order); i++)
2260 coef[start] += tmp[i] * lpc[i - 1];
2261 for (i = order; i > 0; i--)
2262 tmp[i] = tmp[i - 1];
2270 * Apply windowing and MDCT to obtain the spectral
2271 * coefficient from the predicted sample by LTP.
2273 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2274 float *in, IndividualChannelStream *ics)
2276 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2277 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2278 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2279 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2281 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2282 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2284 memset(in, 0, 448 * sizeof(float));
2285 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2287 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2288 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2290 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2291 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2293 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2297 * Apply the long term prediction
2299 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2301 const LongTermPrediction *ltp = &sce->ics.ltp;
2302 const uint16_t *offsets = sce->ics.swb_offset;
2305 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2306 float *predTime = sce->ret;
2307 float *predFreq = ac->buf_mdct;
2308 int16_t num_samples = 2048;
2310 if (ltp->lag < 1024)
2311 num_samples = ltp->lag + 1024;
2312 for (i = 0; i < num_samples; i++)
2313 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2314 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2316 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2318 if (sce->tns.present)
2319 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2321 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2323 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2324 sce->coeffs[i] += predFreq[i];
2329 * Update the LTP buffer for next frame
2331 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2333 IndividualChannelStream *ics = &sce->ics;
2334 float *saved = sce->saved;
2335 float *saved_ltp = sce->coeffs;
2336 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2337 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2340 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2341 memcpy(saved_ltp, saved, 512 * sizeof(float));
2342 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2343 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2344 for (i = 0; i < 64; i++)
2345 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2346 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2347 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2348 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2349 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2350 for (i = 0; i < 64; i++)
2351 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2352 } else { // LONG_STOP or ONLY_LONG
2353 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2354 for (i = 0; i < 512; i++)
2355 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2358 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2359 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2360 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2364 * Conduct IMDCT and windowing.
2366 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2368 IndividualChannelStream *ics = &sce->ics;
2369 float *in = sce->coeffs;
2370 float *out = sce->ret;
2371 float *saved = sce->saved;
2372 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2373 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2374 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2375 float *buf = ac->buf_mdct;
2376 float *temp = ac->temp;
2380 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2381 for (i = 0; i < 1024; i += 128)
2382 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2384 ac->mdct.imdct_half(&ac->mdct, buf, in);
2386 /* window overlapping
2387 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2388 * and long to short transitions are considered to be short to short
2389 * transitions. This leaves just two cases (long to long and short to short)
2390 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2392 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2393 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2394 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2396 memcpy( out, saved, 448 * sizeof(float));
2398 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2399 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2400 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2401 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2402 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2403 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2404 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2406 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2407 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2412 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2413 memcpy( saved, temp + 64, 64 * sizeof(float));
2414 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2415 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2416 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2417 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2418 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2419 memcpy( saved, buf + 512, 448 * sizeof(float));
2420 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2421 } else { // LONG_STOP or ONLY_LONG
2422 memcpy( saved, buf + 512, 512 * sizeof(float));
2426 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2428 IndividualChannelStream *ics = &sce->ics;
2429 float *in = sce->coeffs;
2430 float *out = sce->ret;
2431 float *saved = sce->saved;
2432 float *buf = ac->buf_mdct;
2435 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2437 // window overlapping
2438 if (ics->use_kb_window[1]) {
2439 // AAC LD uses a low overlap sine window instead of a KBD window
2440 memcpy(out, saved, 192 * sizeof(float));
2441 ac->fdsp.vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2442 memcpy( out + 320, buf + 64, 192 * sizeof(float));
2444 ac->fdsp.vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2448 memcpy(saved, buf + 256, 256 * sizeof(float));
2451 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2453 float *in = sce->coeffs;
2454 float *out = sce->ret;
2455 float *saved = sce->saved;
2456 const float *const window = ff_aac_eld_window;
2457 float *buf = ac->buf_mdct;
2460 const int n2 = n >> 1;
2461 const int n4 = n >> 2;
2463 // Inverse transform, mapped to the conventional IMDCT by
2464 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2465 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2466 // Audio, Language and Image Processing, 2008. ICALIP 2008. International Conference on
2467 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2468 for (i = 0; i < n2; i+=2) {
2470 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2471 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2473 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2474 for (i = 0; i < n; i+=2) {
2477 // Like with the regular IMDCT at this point we still have the middle half
2478 // of a transform but with even symmetry on the left and odd symmetry on
2481 // window overlapping
2482 // The spec says to use samples [0..511] but the reference decoder uses
2483 // samples [128..639].
2484 for (i = n4; i < n2; i ++) {
2485 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2486 saved[ i + n2] * window[i + n - n4] +
2487 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2488 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2490 for (i = 0; i < n2; i ++) {
2491 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2492 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2493 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2494 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2496 for (i = 0; i < n4; i ++) {
2497 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2498 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2499 -saved[ n + n2 + i] * window[i + 3*n - n4];
2503 memmove(saved + n, saved, 2 * n * sizeof(float));
2504 memcpy( saved, buf, n * sizeof(float));
2508 * Apply dependent channel coupling (applied before IMDCT).
2510 * @param index index into coupling gain array
2512 static void apply_dependent_coupling(AACContext *ac,
2513 SingleChannelElement *target,
2514 ChannelElement *cce, int index)
2516 IndividualChannelStream *ics = &cce->ch[0].ics;
2517 const uint16_t *offsets = ics->swb_offset;
2518 float *dest = target->coeffs;
2519 const float *src = cce->ch[0].coeffs;
2520 int g, i, group, k, idx = 0;
2521 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2522 av_log(ac->avctx, AV_LOG_ERROR,
2523 "Dependent coupling is not supported together with LTP\n");
2526 for (g = 0; g < ics->num_window_groups; g++) {
2527 for (i = 0; i < ics->max_sfb; i++, idx++) {
2528 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2529 const float gain = cce->coup.gain[index][idx];
2530 for (group = 0; group < ics->group_len[g]; group++) {
2531 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2533 dest[group * 128 + k] += gain * src[group * 128 + k];
2538 dest += ics->group_len[g] * 128;
2539 src += ics->group_len[g] * 128;
2544 * Apply independent channel coupling (applied after IMDCT).
2546 * @param index index into coupling gain array
2548 static void apply_independent_coupling(AACContext *ac,
2549 SingleChannelElement *target,
2550 ChannelElement *cce, int index)
2553 const float gain = cce->coup.gain[index][0];
2554 const float *src = cce->ch[0].ret;
2555 float *dest = target->ret;
2556 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2558 for (i = 0; i < len; i++)
2559 dest[i] += gain * src[i];
2563 * channel coupling transformation interface
2565 * @param apply_coupling_method pointer to (in)dependent coupling function
2567 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2568 enum RawDataBlockType type, int elem_id,
2569 enum CouplingPoint coupling_point,
2570 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2574 for (i = 0; i < MAX_ELEM_ID; i++) {
2575 ChannelElement *cce = ac->che[TYPE_CCE][i];
2578 if (cce && cce->coup.coupling_point == coupling_point) {
2579 ChannelCoupling *coup = &cce->coup;
2581 for (c = 0; c <= coup->num_coupled; c++) {
2582 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2583 if (coup->ch_select[c] != 1) {
2584 apply_coupling_method(ac, &cc->ch[0], cce, index);
2585 if (coup->ch_select[c] != 0)
2588 if (coup->ch_select[c] != 2)
2589 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2591 index += 1 + (coup->ch_select[c] == 3);
2598 * Convert spectral data to float samples, applying all supported tools as appropriate.
2600 static void spectral_to_sample(AACContext *ac)
2603 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2604 switch (ac->oc[1].m4ac.object_type) {
2606 imdct_and_window = imdct_and_windowing_ld;
2608 case AOT_ER_AAC_ELD:
2609 imdct_and_window = imdct_and_windowing_eld;
2612 imdct_and_window = imdct_and_windowing;
2614 for (type = 3; type >= 0; type--) {
2615 for (i = 0; i < MAX_ELEM_ID; i++) {
2616 ChannelElement *che = ac->che[type][i];
2618 if (type <= TYPE_CPE)
2619 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2620 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2621 if (che->ch[0].ics.predictor_present) {
2622 if (che->ch[0].ics.ltp.present)
2623 apply_ltp(ac, &che->ch[0]);
2624 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2625 apply_ltp(ac, &che->ch[1]);
2628 if (che->ch[0].tns.present)
2629 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2630 if (che->ch[1].tns.present)
2631 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2632 if (type <= TYPE_CPE)
2633 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2634 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2635 imdct_and_window(ac, &che->ch[0]);
2636 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2637 update_ltp(ac, &che->ch[0]);
2638 if (type == TYPE_CPE) {
2639 imdct_and_window(ac, &che->ch[1]);
2640 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2641 update_ltp(ac, &che->ch[1]);
2643 if (ac->oc[1].m4ac.sbr > 0) {
2644 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2647 if (type <= TYPE_CCE)
2648 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2654 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2657 AACADTSHeaderInfo hdr_info;
2658 uint8_t layout_map[MAX_ELEM_ID*4][3];
2659 int layout_map_tags, ret;
2661 size = avpriv_aac_parse_header(gb, &hdr_info);
2663 if (hdr_info.num_aac_frames != 1) {
2664 avpriv_report_missing_feature(ac->avctx,
2665 "More than one AAC RDB per ADTS frame");
2666 return AVERROR_PATCHWELCOME;
2668 push_output_configuration(ac);
2669 if (hdr_info.chan_config) {
2670 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2671 if ((ret = set_default_channel_config(ac->avctx,
2674 hdr_info.chan_config)) < 0)
2676 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2677 FFMAX(ac->oc[1].status,
2678 OC_TRIAL_FRAME), 0)) < 0)
2681 ac->oc[1].m4ac.chan_config = 0;
2683 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2684 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2685 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2686 if (ac->oc[0].status != OC_LOCKED ||
2687 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2688 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2689 ac->oc[1].m4ac.sbr = -1;
2690 ac->oc[1].m4ac.ps = -1;
2692 if (!hdr_info.crc_absent)
2698 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2699 int *got_frame_ptr, GetBitContext *gb)
2701 AACContext *ac = avctx->priv_data;
2702 ChannelElement *che;
2705 int chan_config = ac->oc[1].m4ac.chan_config;
2706 int aot = ac->oc[1].m4ac.object_type;
2708 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2713 if ((err = frame_configure_elements(avctx)) < 0)
2716 // The FF_PROFILE_AAC_* defines are all object_type - 1
2717 // This may lead to an undefined profile being signaled
2718 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2720 ac->tags_mapped = 0;
2722 if (chan_config < 0 || chan_config >= 8) {
2723 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2724 ac->oc[1].m4ac.chan_config);
2725 return AVERROR_INVALIDDATA;
2727 for (i = 0; i < tags_per_config[chan_config]; i++) {
2728 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2729 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2730 if (!(che=get_che(ac, elem_type, elem_id))) {
2731 av_log(ac->avctx, AV_LOG_ERROR,
2732 "channel element %d.%d is not allocated\n",
2733 elem_type, elem_id);
2734 return AVERROR_INVALIDDATA;
2736 if (aot != AOT_ER_AAC_ELD)
2738 switch (elem_type) {
2740 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2743 err = decode_cpe(ac, gb, che);
2746 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2753 spectral_to_sample(ac);
2755 ac->frame->nb_samples = samples;
2756 ac->frame->sample_rate = avctx->sample_rate;
2759 skip_bits_long(gb, get_bits_left(gb));
2763 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2764 int *got_frame_ptr, GetBitContext *gb)
2766 AACContext *ac = avctx->priv_data;
2767 ChannelElement *che = NULL, *che_prev = NULL;
2768 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2770 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2774 if (show_bits(gb, 12) == 0xfff) {
2775 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2776 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2779 if (ac->oc[1].m4ac.sampling_index > 12) {
2780 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2781 err = AVERROR_INVALIDDATA;
2786 if ((err = frame_configure_elements(avctx)) < 0)
2789 // The FF_PROFILE_AAC_* defines are all object_type - 1
2790 // This may lead to an undefined profile being signaled
2791 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2793 ac->tags_mapped = 0;
2795 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2796 elem_id = get_bits(gb, 4);
2798 if (elem_type < TYPE_DSE) {
2799 if (!(che=get_che(ac, elem_type, elem_id))) {
2800 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2801 elem_type, elem_id);
2802 err = AVERROR_INVALIDDATA;
2808 switch (elem_type) {
2811 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2816 err = decode_cpe(ac, gb, che);
2821 err = decode_cce(ac, gb, che);
2825 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2830 err = skip_data_stream_element(ac, gb);
2834 uint8_t layout_map[MAX_ELEM_ID*4][3];
2836 push_output_configuration(ac);
2837 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2843 av_log(avctx, AV_LOG_ERROR,
2844 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2845 pop_output_configuration(ac);
2847 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2855 elem_id += get_bits(gb, 8) - 1;
2856 if (get_bits_left(gb) < 8 * elem_id) {
2857 av_log(avctx, AV_LOG_ERROR, overread_err);
2858 err = AVERROR_INVALIDDATA;
2862 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2863 err = 0; /* FIXME */
2867 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
2872 elem_type_prev = elem_type;
2877 if (get_bits_left(gb) < 3) {
2878 av_log(avctx, AV_LOG_ERROR, overread_err);
2879 err = AVERROR_INVALIDDATA;
2884 spectral_to_sample(ac);
2886 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2887 samples <<= multiplier;
2889 if (ac->oc[1].status && audio_found) {
2890 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2891 avctx->frame_size = samples;
2892 ac->oc[1].status = OC_LOCKED;
2896 ac->frame->nb_samples = samples;
2897 ac->frame->sample_rate = avctx->sample_rate;
2899 *got_frame_ptr = !!samples;
2903 pop_output_configuration(ac);
2907 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2908 int *got_frame_ptr, AVPacket *avpkt)
2910 AACContext *ac = avctx->priv_data;
2911 const uint8_t *buf = avpkt->data;
2912 int buf_size = avpkt->size;
2917 int new_extradata_size;
2918 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2919 AV_PKT_DATA_NEW_EXTRADATA,
2920 &new_extradata_size);
2922 if (new_extradata) {
2923 av_free(avctx->extradata);
2924 avctx->extradata = av_mallocz(new_extradata_size +
2925 FF_INPUT_BUFFER_PADDING_SIZE);
2926 if (!avctx->extradata)
2927 return AVERROR(ENOMEM);
2928 avctx->extradata_size = new_extradata_size;
2929 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2930 push_output_configuration(ac);
2931 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2933 avctx->extradata_size*8, 1) < 0) {
2934 pop_output_configuration(ac);
2935 return AVERROR_INVALIDDATA;
2939 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
2942 switch (ac->oc[1].m4ac.object_type) {
2944 case AOT_ER_AAC_LTP:
2946 case AOT_ER_AAC_ELD:
2947 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
2950 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb);
2955 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2956 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2957 if (buf[buf_offset])
2960 return buf_size > buf_offset ? buf_consumed : buf_size;
2963 static av_cold int aac_decode_close(AVCodecContext *avctx)
2965 AACContext *ac = avctx->priv_data;
2968 for (i = 0; i < MAX_ELEM_ID; i++) {
2969 for (type = 0; type < 4; type++) {
2970 if (ac->che[type][i])
2971 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2972 av_freep(&ac->che[type][i]);
2976 ff_mdct_end(&ac->mdct);
2977 ff_mdct_end(&ac->mdct_small);
2978 ff_mdct_end(&ac->mdct_ld);
2979 ff_mdct_end(&ac->mdct_ltp);
2984 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2986 struct LATMContext {
2987 AACContext aac_ctx; ///< containing AACContext
2988 int initialized; ///< initilized after a valid extradata was seen
2991 int audio_mux_version_A; ///< LATM syntax version
2992 int frame_length_type; ///< 0/1 variable/fixed frame length
2993 int frame_length; ///< frame length for fixed frame length
2996 static inline uint32_t latm_get_value(GetBitContext *b)
2998 int length = get_bits(b, 2);
3000 return get_bits_long(b, (length+1)*8);
3003 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3004 GetBitContext *gb, int asclen)
3006 AACContext *ac = &latmctx->aac_ctx;
3007 AVCodecContext *avctx = ac->avctx;
3008 MPEG4AudioConfig m4ac = { 0 };
3009 int config_start_bit = get_bits_count(gb);
3010 int sync_extension = 0;
3011 int bits_consumed, esize;
3015 asclen = FFMIN(asclen, get_bits_left(gb));
3017 asclen = get_bits_left(gb);
3019 if (config_start_bit % 8) {
3020 avpriv_request_sample(latmctx->aac_ctx.avctx,
3021 "Non-byte-aligned audio-specific config");
3022 return AVERROR_PATCHWELCOME;
3025 return AVERROR_INVALIDDATA;
3026 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3027 gb->buffer + (config_start_bit / 8),
3028 asclen, sync_extension);
3030 if (bits_consumed < 0)
3031 return AVERROR_INVALIDDATA;
3033 if (ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3034 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3036 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3037 latmctx->initialized = 0;
3039 esize = (bits_consumed+7) / 8;
3041 if (avctx->extradata_size < esize) {
3042 av_free(avctx->extradata);
3043 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3044 if (!avctx->extradata)
3045 return AVERROR(ENOMEM);
3048 avctx->extradata_size = esize;
3049 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3050 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3052 skip_bits_long(gb, bits_consumed);
3054 return bits_consumed;
3057 static int read_stream_mux_config(struct LATMContext *latmctx,
3060 int ret, audio_mux_version = get_bits(gb, 1);
3062 latmctx->audio_mux_version_A = 0;
3063 if (audio_mux_version)
3064 latmctx->audio_mux_version_A = get_bits(gb, 1);
3066 if (!latmctx->audio_mux_version_A) {
3068 if (audio_mux_version)
3069 latm_get_value(gb); // taraFullness
3071 skip_bits(gb, 1); // allStreamSameTimeFraming
3072 skip_bits(gb, 6); // numSubFrames
3074 if (get_bits(gb, 4)) { // numPrograms
3075 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3076 return AVERROR_PATCHWELCOME;
3079 // for each program (which there is only on in DVB)
3081 // for each layer (which there is only on in DVB)
3082 if (get_bits(gb, 3)) { // numLayer
3083 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3084 return AVERROR_PATCHWELCOME;
3087 // for all but first stream: use_same_config = get_bits(gb, 1);
3088 if (!audio_mux_version) {
3089 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3092 int ascLen = latm_get_value(gb);
3093 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3096 skip_bits_long(gb, ascLen);
3099 latmctx->frame_length_type = get_bits(gb, 3);
3100 switch (latmctx->frame_length_type) {
3102 skip_bits(gb, 8); // latmBufferFullness
3105 latmctx->frame_length = get_bits(gb, 9);
3110 skip_bits(gb, 6); // CELP frame length table index
3114 skip_bits(gb, 1); // HVXC frame length table index
3118 if (get_bits(gb, 1)) { // other data
3119 if (audio_mux_version) {
3120 latm_get_value(gb); // other_data_bits
3124 esc = get_bits(gb, 1);
3130 if (get_bits(gb, 1)) // crc present
3131 skip_bits(gb, 8); // config_crc
3137 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3141 if (ctx->frame_length_type == 0) {
3142 int mux_slot_length = 0;
3144 tmp = get_bits(gb, 8);
3145 mux_slot_length += tmp;
3146 } while (tmp == 255);
3147 return mux_slot_length;
3148 } else if (ctx->frame_length_type == 1) {
3149 return ctx->frame_length;
3150 } else if (ctx->frame_length_type == 3 ||
3151 ctx->frame_length_type == 5 ||
3152 ctx->frame_length_type == 7) {
3153 skip_bits(gb, 2); // mux_slot_length_coded
3158 static int read_audio_mux_element(struct LATMContext *latmctx,
3162 uint8_t use_same_mux = get_bits(gb, 1);
3163 if (!use_same_mux) {
3164 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3166 } else if (!latmctx->aac_ctx.avctx->extradata) {
3167 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3168 "no decoder config found\n");
3169 return AVERROR(EAGAIN);
3171 if (latmctx->audio_mux_version_A == 0) {
3172 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3173 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3174 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3175 return AVERROR_INVALIDDATA;
3176 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3177 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3178 "frame length mismatch %d << %d\n",
3179 mux_slot_length_bytes * 8, get_bits_left(gb));
3180 return AVERROR_INVALIDDATA;
3187 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3188 int *got_frame_ptr, AVPacket *avpkt)
3190 struct LATMContext *latmctx = avctx->priv_data;
3194 if ((err = init_get_bits(&gb, avpkt->data, avpkt->size * 8)) < 0)
3197 // check for LOAS sync word
3198 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3199 return AVERROR_INVALIDDATA;
3201 muxlength = get_bits(&gb, 13) + 3;
3202 // not enough data, the parser should have sorted this
3203 if (muxlength > avpkt->size)
3204 return AVERROR_INVALIDDATA;
3206 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3209 if (!latmctx->initialized) {
3210 if (!avctx->extradata) {
3214 push_output_configuration(&latmctx->aac_ctx);
3215 if ((err = decode_audio_specific_config(
3216 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3217 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3218 pop_output_configuration(&latmctx->aac_ctx);
3221 latmctx->initialized = 1;
3225 if (show_bits(&gb, 12) == 0xfff) {
3226 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3227 "ADTS header detected, probably as result of configuration "
3229 return AVERROR_INVALIDDATA;
3232 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
3238 static av_cold int latm_decode_init(AVCodecContext *avctx)
3240 struct LATMContext *latmctx = avctx->priv_data;
3241 int ret = aac_decode_init(avctx);
3243 if (avctx->extradata_size > 0)
3244 latmctx->initialized = !ret;
3250 AVCodec ff_aac_decoder = {
3252 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3253 .type = AVMEDIA_TYPE_AUDIO,
3254 .id = AV_CODEC_ID_AAC,
3255 .priv_data_size = sizeof(AACContext),
3256 .init = aac_decode_init,
3257 .close = aac_decode_close,
3258 .decode = aac_decode_frame,
3259 .sample_fmts = (const enum AVSampleFormat[]) {
3260 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3262 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3263 .channel_layouts = aac_channel_layout,
3267 Note: This decoder filter is intended to decode LATM streams transferred
3268 in MPEG transport streams which only contain one program.
3269 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3271 AVCodec ff_aac_latm_decoder = {
3273 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3274 .type = AVMEDIA_TYPE_AUDIO,
3275 .id = AV_CODEC_ID_AAC_LATM,
3276 .priv_data_size = sizeof(struct LATMContext),
3277 .init = latm_decode_init,
3278 .close = aac_decode_close,
3279 .decode = latm_decode_frame,
3280 .sample_fmts = (const enum AVSampleFormat[]) {
3281 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3283 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3284 .channel_layouts = aac_channel_layout,