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 FFmpeg.
13 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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
77 * Y Enhanced AAC Low Delay (ER AAC ELD)
79 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
80 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
84 #include "libavutil/float_dsp.h"
85 #include "libavutil/opt.h"
90 #include "fmtconvert.h"
97 #include "aacdectab.h"
98 #include "cbrt_tablegen.h"
101 #include "mpeg4audio.h"
102 #include "aacadtsdec.h"
103 #include "libavutil/intfloat.h"
112 # include "arm/aac.h"
114 # include "mips/aacdec_mips.h"
117 static VLC vlc_scalefactors;
118 static VLC vlc_spectral[11];
120 static int output_configure(AACContext *ac,
121 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
122 enum OCStatus oc_type, int get_new_frame);
124 #define overread_err "Input buffer exhausted before END element found\n"
126 static int count_channels(uint8_t (*layout)[3], int tags)
129 for (i = 0; i < tags; i++) {
130 int syn_ele = layout[i][0];
131 int pos = layout[i][2];
132 sum += (1 + (syn_ele == TYPE_CPE)) *
133 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
139 * Check for the channel element in the current channel position configuration.
140 * If it exists, make sure the appropriate element is allocated and map the
141 * channel order to match the internal FFmpeg channel layout.
143 * @param che_pos current channel position configuration
144 * @param type channel element type
145 * @param id channel element id
146 * @param channels count of the number of channels in the configuration
148 * @return Returns error status. 0 - OK, !0 - error
150 static av_cold int che_configure(AACContext *ac,
151 enum ChannelPosition che_pos,
152 int type, int id, int *channels)
154 if (*channels >= MAX_CHANNELS)
155 return AVERROR_INVALIDDATA;
157 if (!ac->che[type][id]) {
158 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
159 return AVERROR(ENOMEM);
160 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
162 if (type != TYPE_CCE) {
163 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
164 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
165 return AVERROR_INVALIDDATA;
167 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
168 if (type == TYPE_CPE ||
169 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
170 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
174 if (ac->che[type][id])
175 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
176 av_freep(&ac->che[type][id]);
181 static int frame_configure_elements(AVCodecContext *avctx)
183 AACContext *ac = avctx->priv_data;
184 int type, id, ch, ret;
186 /* set channel pointers to internal buffers by default */
187 for (type = 0; type < 4; type++) {
188 for (id = 0; id < MAX_ELEM_ID; id++) {
189 ChannelElement *che = ac->che[type][id];
191 che->ch[0].ret = che->ch[0].ret_buf;
192 che->ch[1].ret = che->ch[1].ret_buf;
197 /* get output buffer */
198 av_frame_unref(ac->frame);
199 ac->frame->nb_samples = 2048;
200 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
203 /* map output channel pointers to AVFrame data */
204 for (ch = 0; ch < avctx->channels; ch++) {
205 if (ac->output_element[ch])
206 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
212 struct elem_to_channel {
213 uint64_t av_position;
216 uint8_t aac_position;
219 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
220 uint8_t (*layout_map)[3], int offset, uint64_t left,
221 uint64_t right, int pos)
223 if (layout_map[offset][0] == TYPE_CPE) {
224 e2c_vec[offset] = (struct elem_to_channel) {
225 .av_position = left | right,
227 .elem_id = layout_map[offset][1],
232 e2c_vec[offset] = (struct elem_to_channel) {
235 .elem_id = layout_map[offset][1],
238 e2c_vec[offset + 1] = (struct elem_to_channel) {
239 .av_position = right,
241 .elem_id = layout_map[offset + 1][1],
248 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
251 int num_pos_channels = 0;
255 for (i = *current; i < tags; i++) {
256 if (layout_map[i][2] != pos)
258 if (layout_map[i][0] == TYPE_CPE) {
260 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
266 num_pos_channels += 2;
274 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
277 return num_pos_channels;
280 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
282 int i, n, total_non_cc_elements;
283 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
284 int num_front_channels, num_side_channels, num_back_channels;
287 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
292 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
293 if (num_front_channels < 0)
296 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
297 if (num_side_channels < 0)
300 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
301 if (num_back_channels < 0)
305 if (num_front_channels & 1) {
306 e2c_vec[i] = (struct elem_to_channel) {
307 .av_position = AV_CH_FRONT_CENTER,
309 .elem_id = layout_map[i][1],
310 .aac_position = AAC_CHANNEL_FRONT
313 num_front_channels--;
315 if (num_front_channels >= 4) {
316 i += assign_pair(e2c_vec, layout_map, i,
317 AV_CH_FRONT_LEFT_OF_CENTER,
318 AV_CH_FRONT_RIGHT_OF_CENTER,
320 num_front_channels -= 2;
322 if (num_front_channels >= 2) {
323 i += assign_pair(e2c_vec, layout_map, i,
327 num_front_channels -= 2;
329 while (num_front_channels >= 2) {
330 i += assign_pair(e2c_vec, layout_map, i,
334 num_front_channels -= 2;
337 if (num_side_channels >= 2) {
338 i += assign_pair(e2c_vec, layout_map, i,
342 num_side_channels -= 2;
344 while (num_side_channels >= 2) {
345 i += assign_pair(e2c_vec, layout_map, i,
349 num_side_channels -= 2;
352 while (num_back_channels >= 4) {
353 i += assign_pair(e2c_vec, layout_map, i,
357 num_back_channels -= 2;
359 if (num_back_channels >= 2) {
360 i += assign_pair(e2c_vec, layout_map, i,
364 num_back_channels -= 2;
366 if (num_back_channels) {
367 e2c_vec[i] = (struct elem_to_channel) {
368 .av_position = AV_CH_BACK_CENTER,
370 .elem_id = layout_map[i][1],
371 .aac_position = AAC_CHANNEL_BACK
377 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
378 e2c_vec[i] = (struct elem_to_channel) {
379 .av_position = AV_CH_LOW_FREQUENCY,
381 .elem_id = layout_map[i][1],
382 .aac_position = AAC_CHANNEL_LFE
386 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
387 e2c_vec[i] = (struct elem_to_channel) {
388 .av_position = UINT64_MAX,
390 .elem_id = layout_map[i][1],
391 .aac_position = AAC_CHANNEL_LFE
396 // Must choose a stable sort
397 total_non_cc_elements = n = i;
400 for (i = 1; i < n; i++)
401 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
402 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
409 for (i = 0; i < total_non_cc_elements; i++) {
410 layout_map[i][0] = e2c_vec[i].syn_ele;
411 layout_map[i][1] = e2c_vec[i].elem_id;
412 layout_map[i][2] = e2c_vec[i].aac_position;
413 if (e2c_vec[i].av_position != UINT64_MAX) {
414 layout |= e2c_vec[i].av_position;
422 * Save current output configuration if and only if it has been locked.
424 static void push_output_configuration(AACContext *ac) {
425 if (ac->oc[1].status == OC_LOCKED) {
426 ac->oc[0] = ac->oc[1];
428 ac->oc[1].status = OC_NONE;
432 * Restore the previous output configuration if and only if the current
433 * configuration is unlocked.
435 static void pop_output_configuration(AACContext *ac) {
436 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
437 ac->oc[1] = ac->oc[0];
438 ac->avctx->channels = ac->oc[1].channels;
439 ac->avctx->channel_layout = ac->oc[1].channel_layout;
440 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
441 ac->oc[1].status, 0);
446 * Configure output channel order based on the current program
447 * configuration element.
449 * @return Returns error status. 0 - OK, !0 - error
451 static int output_configure(AACContext *ac,
452 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
453 enum OCStatus oc_type, int get_new_frame)
455 AVCodecContext *avctx = ac->avctx;
456 int i, channels = 0, ret;
459 if (ac->oc[1].layout_map != layout_map) {
460 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
461 ac->oc[1].layout_map_tags = tags;
464 // Try to sniff a reasonable channel order, otherwise output the
465 // channels in the order the PCE declared them.
466 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
467 layout = sniff_channel_order(layout_map, tags);
468 for (i = 0; i < tags; i++) {
469 int type = layout_map[i][0];
470 int id = layout_map[i][1];
471 int position = layout_map[i][2];
472 // Allocate or free elements depending on if they are in the
473 // current program configuration.
474 ret = che_configure(ac, position, type, id, &channels);
478 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
479 if (layout == AV_CH_FRONT_CENTER) {
480 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
486 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
487 if (layout) avctx->channel_layout = layout;
488 ac->oc[1].channel_layout = layout;
489 avctx->channels = ac->oc[1].channels = channels;
490 ac->oc[1].status = oc_type;
493 if ((ret = frame_configure_elements(ac->avctx)) < 0)
500 static void flush(AVCodecContext *avctx)
502 AACContext *ac= avctx->priv_data;
505 for (type = 3; type >= 0; type--) {
506 for (i = 0; i < MAX_ELEM_ID; i++) {
507 ChannelElement *che = ac->che[type][i];
509 for (j = 0; j <= 1; j++) {
510 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
518 * Set up channel positions based on a default channel configuration
519 * as specified in table 1.17.
521 * @return Returns error status. 0 - OK, !0 - error
523 static int set_default_channel_config(AVCodecContext *avctx,
524 uint8_t (*layout_map)[3],
528 if (channel_config < 1 || channel_config > 7) {
529 av_log(avctx, AV_LOG_ERROR,
530 "invalid default channel configuration (%d)\n",
532 return AVERROR_INVALIDDATA;
534 *tags = tags_per_config[channel_config];
535 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
536 *tags * sizeof(*layout_map));
540 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
542 /* For PCE based channel configurations map the channels solely based
544 if (!ac->oc[1].m4ac.chan_config) {
545 return ac->tag_che_map[type][elem_id];
547 // Allow single CPE stereo files to be signalled with mono configuration.
548 if (!ac->tags_mapped && type == TYPE_CPE &&
549 ac->oc[1].m4ac.chan_config == 1) {
550 uint8_t layout_map[MAX_ELEM_ID*4][3];
552 push_output_configuration(ac);
554 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
556 if (set_default_channel_config(ac->avctx, layout_map,
557 &layout_map_tags, 2) < 0)
559 if (output_configure(ac, layout_map, layout_map_tags,
560 OC_TRIAL_FRAME, 1) < 0)
563 ac->oc[1].m4ac.chan_config = 2;
564 ac->oc[1].m4ac.ps = 0;
567 if (!ac->tags_mapped && type == TYPE_SCE &&
568 ac->oc[1].m4ac.chan_config == 2) {
569 uint8_t layout_map[MAX_ELEM_ID * 4][3];
571 push_output_configuration(ac);
573 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
575 if (set_default_channel_config(ac->avctx, layout_map,
576 &layout_map_tags, 1) < 0)
578 if (output_configure(ac, layout_map, layout_map_tags,
579 OC_TRIAL_FRAME, 1) < 0)
582 ac->oc[1].m4ac.chan_config = 1;
583 if (ac->oc[1].m4ac.sbr)
584 ac->oc[1].m4ac.ps = -1;
586 /* For indexed channel configurations map the channels solely based
588 switch (ac->oc[1].m4ac.chan_config) {
590 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
592 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
595 /* Some streams incorrectly code 5.1 audio as
596 * SCE[0] CPE[0] CPE[1] SCE[1]
598 * SCE[0] CPE[0] CPE[1] LFE[0].
599 * If we seem to have encountered such a stream, transfer
600 * the LFE[0] element to the SCE[1]'s mapping */
601 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
603 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
606 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
608 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
611 if (ac->tags_mapped == 2 &&
612 ac->oc[1].m4ac.chan_config == 4 &&
615 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
619 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
622 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
623 } else if (ac->oc[1].m4ac.chan_config == 2) {
627 if (!ac->tags_mapped && type == TYPE_SCE) {
629 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
637 * Decode an array of 4 bit element IDs, optionally interleaved with a
638 * stereo/mono switching bit.
640 * @param type speaker type/position for these channels
642 static void decode_channel_map(uint8_t layout_map[][3],
643 enum ChannelPosition type,
644 GetBitContext *gb, int n)
647 enum RawDataBlockType syn_ele;
649 case AAC_CHANNEL_FRONT:
650 case AAC_CHANNEL_BACK:
651 case AAC_CHANNEL_SIDE:
652 syn_ele = get_bits1(gb);
658 case AAC_CHANNEL_LFE:
664 layout_map[0][0] = syn_ele;
665 layout_map[0][1] = get_bits(gb, 4);
666 layout_map[0][2] = type;
672 * Decode program configuration element; reference: table 4.2.
674 * @return Returns error status. 0 - OK, !0 - error
676 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
677 uint8_t (*layout_map)[3],
680 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
685 skip_bits(gb, 2); // object_type
687 sampling_index = get_bits(gb, 4);
688 if (m4ac->sampling_index != sampling_index)
689 av_log(avctx, AV_LOG_WARNING,
690 "Sample rate index in program config element does not "
691 "match the sample rate index configured by the container.\n");
693 num_front = get_bits(gb, 4);
694 num_side = get_bits(gb, 4);
695 num_back = get_bits(gb, 4);
696 num_lfe = get_bits(gb, 2);
697 num_assoc_data = get_bits(gb, 3);
698 num_cc = get_bits(gb, 4);
701 skip_bits(gb, 4); // mono_mixdown_tag
703 skip_bits(gb, 4); // stereo_mixdown_tag
706 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
708 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
709 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
712 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
714 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
716 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
718 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
721 skip_bits_long(gb, 4 * num_assoc_data);
723 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
728 /* comment field, first byte is length */
729 comment_len = get_bits(gb, 8) * 8;
730 if (get_bits_left(gb) < comment_len) {
731 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
732 return AVERROR_INVALIDDATA;
734 skip_bits_long(gb, comment_len);
739 * Decode GA "General Audio" specific configuration; reference: table 4.1.
741 * @param ac pointer to AACContext, may be null
742 * @param avctx pointer to AVCCodecContext, used for logging
744 * @return Returns error status. 0 - OK, !0 - error
746 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
748 MPEG4AudioConfig *m4ac,
751 int extension_flag, ret, ep_config, res_flags;
752 uint8_t layout_map[MAX_ELEM_ID*4][3];
755 if (get_bits1(gb)) { // frameLengthFlag
756 avpriv_request_sample(avctx, "960/120 MDCT window");
757 return AVERROR_PATCHWELCOME;
760 if (get_bits1(gb)) // dependsOnCoreCoder
761 skip_bits(gb, 14); // coreCoderDelay
762 extension_flag = get_bits1(gb);
764 if (m4ac->object_type == AOT_AAC_SCALABLE ||
765 m4ac->object_type == AOT_ER_AAC_SCALABLE)
766 skip_bits(gb, 3); // layerNr
768 if (channel_config == 0) {
769 skip_bits(gb, 4); // element_instance_tag
770 tags = decode_pce(avctx, m4ac, layout_map, gb);
774 if ((ret = set_default_channel_config(avctx, layout_map,
775 &tags, channel_config)))
779 if (count_channels(layout_map, tags) > 1) {
781 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
784 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
787 if (extension_flag) {
788 switch (m4ac->object_type) {
790 skip_bits(gb, 5); // numOfSubFrame
791 skip_bits(gb, 11); // layer_length
795 case AOT_ER_AAC_SCALABLE:
797 res_flags = get_bits(gb, 3);
799 avpriv_report_missing_feature(avctx,
800 "AAC data resilience (flags %x)",
802 return AVERROR_PATCHWELCOME;
806 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
808 switch (m4ac->object_type) {
811 case AOT_ER_AAC_SCALABLE:
813 ep_config = get_bits(gb, 2);
815 avpriv_report_missing_feature(avctx,
816 "epConfig %d", ep_config);
817 return AVERROR_PATCHWELCOME;
823 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
825 MPEG4AudioConfig *m4ac,
828 int ret, ep_config, res_flags;
829 uint8_t layout_map[MAX_ELEM_ID*4][3];
831 const int ELDEXT_TERM = 0;
836 if (get_bits1(gb)) { // frameLengthFlag
837 avpriv_request_sample(avctx, "960/120 MDCT window");
838 return AVERROR_PATCHWELCOME;
841 res_flags = get_bits(gb, 3);
843 avpriv_report_missing_feature(avctx,
844 "AAC data resilience (flags %x)",
846 return AVERROR_PATCHWELCOME;
849 if (get_bits1(gb)) { // ldSbrPresentFlag
850 avpriv_report_missing_feature(avctx,
852 return AVERROR_PATCHWELCOME;
855 while (get_bits(gb, 4) != ELDEXT_TERM) {
856 int len = get_bits(gb, 4);
858 len += get_bits(gb, 8);
860 len += get_bits(gb, 16);
861 if (get_bits_left(gb) < len * 8 + 4) {
862 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
863 return AVERROR_INVALIDDATA;
865 skip_bits_long(gb, 8 * len);
868 if ((ret = set_default_channel_config(avctx, layout_map,
869 &tags, channel_config)))
872 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
875 ep_config = get_bits(gb, 2);
877 avpriv_report_missing_feature(avctx,
878 "epConfig %d", ep_config);
879 return AVERROR_PATCHWELCOME;
885 * Decode audio specific configuration; reference: table 1.13.
887 * @param ac pointer to AACContext, may be null
888 * @param avctx pointer to AVCCodecContext, used for logging
889 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
890 * @param data pointer to buffer holding an audio specific config
891 * @param bit_size size of audio specific config or data in bits
892 * @param sync_extension look for an appended sync extension
894 * @return Returns error status or number of consumed bits. <0 - error
896 static int decode_audio_specific_config(AACContext *ac,
897 AVCodecContext *avctx,
898 MPEG4AudioConfig *m4ac,
899 const uint8_t *data, int bit_size,
905 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
906 for (i = 0; i < bit_size >> 3; i++)
907 av_dlog(avctx, "%02x ", data[i]);
908 av_dlog(avctx, "\n");
910 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
913 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
914 sync_extension)) < 0)
915 return AVERROR_INVALIDDATA;
916 if (m4ac->sampling_index > 12) {
917 av_log(avctx, AV_LOG_ERROR,
918 "invalid sampling rate index %d\n",
919 m4ac->sampling_index);
920 return AVERROR_INVALIDDATA;
922 if (m4ac->object_type == AOT_ER_AAC_LD &&
923 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
924 av_log(avctx, AV_LOG_ERROR,
925 "invalid low delay sampling rate index %d\n",
926 m4ac->sampling_index);
927 return AVERROR_INVALIDDATA;
930 skip_bits_long(&gb, i);
932 switch (m4ac->object_type) {
938 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
939 m4ac, m4ac->chan_config)) < 0)
943 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
944 m4ac, m4ac->chan_config)) < 0)
948 avpriv_report_missing_feature(avctx,
949 "Audio object type %s%d",
950 m4ac->sbr == 1 ? "SBR+" : "",
952 return AVERROR(ENOSYS);
956 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
957 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
958 m4ac->sample_rate, m4ac->sbr,
961 return get_bits_count(&gb);
965 * linear congruential pseudorandom number generator
967 * @param previous_val pointer to the current state of the generator
969 * @return Returns a 32-bit pseudorandom integer
971 static av_always_inline int lcg_random(unsigned previous_val)
973 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
977 static av_always_inline void reset_predict_state(PredictorState *ps)
987 static void reset_all_predictors(PredictorState *ps)
990 for (i = 0; i < MAX_PREDICTORS; i++)
991 reset_predict_state(&ps[i]);
994 static int sample_rate_idx (int rate)
996 if (92017 <= rate) return 0;
997 else if (75132 <= rate) return 1;
998 else if (55426 <= rate) return 2;
999 else if (46009 <= rate) return 3;
1000 else if (37566 <= rate) return 4;
1001 else if (27713 <= rate) return 5;
1002 else if (23004 <= rate) return 6;
1003 else if (18783 <= rate) return 7;
1004 else if (13856 <= rate) return 8;
1005 else if (11502 <= rate) return 9;
1006 else if (9391 <= rate) return 10;
1010 static void reset_predictor_group(PredictorState *ps, int group_num)
1013 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1014 reset_predict_state(&ps[i]);
1017 #define AAC_INIT_VLC_STATIC(num, size) \
1018 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1019 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1020 sizeof(ff_aac_spectral_bits[num][0]), \
1021 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1022 sizeof(ff_aac_spectral_codes[num][0]), \
1025 static void aacdec_init(AACContext *ac);
1027 static av_cold int aac_decode_init(AVCodecContext *avctx)
1029 AACContext *ac = avctx->priv_data;
1033 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1037 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1039 if (avctx->extradata_size > 0) {
1040 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1042 avctx->extradata_size * 8,
1047 uint8_t layout_map[MAX_ELEM_ID*4][3];
1048 int layout_map_tags;
1050 sr = sample_rate_idx(avctx->sample_rate);
1051 ac->oc[1].m4ac.sampling_index = sr;
1052 ac->oc[1].m4ac.channels = avctx->channels;
1053 ac->oc[1].m4ac.sbr = -1;
1054 ac->oc[1].m4ac.ps = -1;
1056 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1057 if (ff_mpeg4audio_channels[i] == avctx->channels)
1059 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1062 ac->oc[1].m4ac.chan_config = i;
1064 if (ac->oc[1].m4ac.chan_config) {
1065 int ret = set_default_channel_config(avctx, layout_map,
1066 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1068 output_configure(ac, layout_map, layout_map_tags,
1070 else if (avctx->err_recognition & AV_EF_EXPLODE)
1071 return AVERROR_INVALIDDATA;
1075 if (avctx->channels > MAX_CHANNELS) {
1076 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1077 return AVERROR_INVALIDDATA;
1080 AAC_INIT_VLC_STATIC( 0, 304);
1081 AAC_INIT_VLC_STATIC( 1, 270);
1082 AAC_INIT_VLC_STATIC( 2, 550);
1083 AAC_INIT_VLC_STATIC( 3, 300);
1084 AAC_INIT_VLC_STATIC( 4, 328);
1085 AAC_INIT_VLC_STATIC( 5, 294);
1086 AAC_INIT_VLC_STATIC( 6, 306);
1087 AAC_INIT_VLC_STATIC( 7, 268);
1088 AAC_INIT_VLC_STATIC( 8, 510);
1089 AAC_INIT_VLC_STATIC( 9, 366);
1090 AAC_INIT_VLC_STATIC(10, 462);
1094 ff_fmt_convert_init(&ac->fmt_conv, avctx);
1095 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1097 ac->random_state = 0x1f2e3d4c;
1101 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1102 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1103 ff_aac_scalefactor_bits,
1104 sizeof(ff_aac_scalefactor_bits[0]),
1105 sizeof(ff_aac_scalefactor_bits[0]),
1106 ff_aac_scalefactor_code,
1107 sizeof(ff_aac_scalefactor_code[0]),
1108 sizeof(ff_aac_scalefactor_code[0]),
1111 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1112 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1113 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1114 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1115 // window initialization
1116 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1117 ff_kbd_window_init(ff_aac_kbd_long_512, 4.0, 512);
1118 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1119 ff_init_ff_sine_windows(10);
1120 ff_init_ff_sine_windows( 9);
1121 ff_init_ff_sine_windows( 7);
1129 * Skip data_stream_element; reference: table 4.10.
1131 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1133 int byte_align = get_bits1(gb);
1134 int count = get_bits(gb, 8);
1136 count += get_bits(gb, 8);
1140 if (get_bits_left(gb) < 8 * count) {
1141 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1142 return AVERROR_INVALIDDATA;
1144 skip_bits_long(gb, 8 * count);
1148 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1152 if (get_bits1(gb)) {
1153 ics->predictor_reset_group = get_bits(gb, 5);
1154 if (ics->predictor_reset_group == 0 ||
1155 ics->predictor_reset_group > 30) {
1156 av_log(ac->avctx, AV_LOG_ERROR,
1157 "Invalid Predictor Reset Group.\n");
1158 return AVERROR_INVALIDDATA;
1161 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1162 ics->prediction_used[sfb] = get_bits1(gb);
1168 * Decode Long Term Prediction data; reference: table 4.xx.
1170 static void decode_ltp(LongTermPrediction *ltp,
1171 GetBitContext *gb, uint8_t max_sfb)
1175 ltp->lag = get_bits(gb, 11);
1176 ltp->coef = ltp_coef[get_bits(gb, 3)];
1177 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1178 ltp->used[sfb] = get_bits1(gb);
1182 * Decode Individual Channel Stream info; reference: table 4.6.
1184 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1187 int aot = ac->oc[1].m4ac.object_type;
1188 if (aot != AOT_ER_AAC_ELD) {
1189 if (get_bits1(gb)) {
1190 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1191 return AVERROR_INVALIDDATA;
1193 ics->window_sequence[1] = ics->window_sequence[0];
1194 ics->window_sequence[0] = get_bits(gb, 2);
1195 if (aot == AOT_ER_AAC_LD &&
1196 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1197 av_log(ac->avctx, AV_LOG_ERROR,
1198 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1199 "window sequence %d found.\n", ics->window_sequence[0]);
1200 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1201 return AVERROR_INVALIDDATA;
1203 ics->use_kb_window[1] = ics->use_kb_window[0];
1204 ics->use_kb_window[0] = get_bits1(gb);
1206 ics->num_window_groups = 1;
1207 ics->group_len[0] = 1;
1208 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1210 ics->max_sfb = get_bits(gb, 4);
1211 for (i = 0; i < 7; i++) {
1212 if (get_bits1(gb)) {
1213 ics->group_len[ics->num_window_groups - 1]++;
1215 ics->num_window_groups++;
1216 ics->group_len[ics->num_window_groups - 1] = 1;
1219 ics->num_windows = 8;
1220 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1221 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1222 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1223 ics->predictor_present = 0;
1225 ics->max_sfb = get_bits(gb, 6);
1226 ics->num_windows = 1;
1227 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1228 ics->swb_offset = ff_swb_offset_512[ac->oc[1].m4ac.sampling_index];
1229 ics->num_swb = ff_aac_num_swb_512[ac->oc[1].m4ac.sampling_index];
1230 if (!ics->num_swb || !ics->swb_offset)
1233 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1234 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1236 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1237 if (aot != AOT_ER_AAC_ELD) {
1238 ics->predictor_present = get_bits1(gb);
1239 ics->predictor_reset_group = 0;
1241 if (ics->predictor_present) {
1242 if (aot == AOT_AAC_MAIN) {
1243 if (decode_prediction(ac, ics, gb)) {
1246 } else if (aot == AOT_AAC_LC ||
1247 aot == AOT_ER_AAC_LC) {
1248 av_log(ac->avctx, AV_LOG_ERROR,
1249 "Prediction is not allowed in AAC-LC.\n");
1252 if (aot == AOT_ER_AAC_LD) {
1253 av_log(ac->avctx, AV_LOG_ERROR,
1254 "LTP in ER AAC LD not yet implemented.\n");
1255 return AVERROR_PATCHWELCOME;
1257 if ((ics->ltp.present = get_bits(gb, 1)))
1258 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1263 if (ics->max_sfb > ics->num_swb) {
1264 av_log(ac->avctx, AV_LOG_ERROR,
1265 "Number of scalefactor bands in group (%d) "
1266 "exceeds limit (%d).\n",
1267 ics->max_sfb, ics->num_swb);
1274 return AVERROR_INVALIDDATA;
1278 * Decode band types (section_data payload); reference: table 4.46.
1280 * @param band_type array of the used band type
1281 * @param band_type_run_end array of the last scalefactor band of a band type run
1283 * @return Returns error status. 0 - OK, !0 - error
1285 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1286 int band_type_run_end[120], GetBitContext *gb,
1287 IndividualChannelStream *ics)
1290 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1291 for (g = 0; g < ics->num_window_groups; g++) {
1293 while (k < ics->max_sfb) {
1294 uint8_t sect_end = k;
1296 int sect_band_type = get_bits(gb, 4);
1297 if (sect_band_type == 12) {
1298 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1299 return AVERROR_INVALIDDATA;
1302 sect_len_incr = get_bits(gb, bits);
1303 sect_end += sect_len_incr;
1304 if (get_bits_left(gb) < 0) {
1305 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1306 return AVERROR_INVALIDDATA;
1308 if (sect_end > ics->max_sfb) {
1309 av_log(ac->avctx, AV_LOG_ERROR,
1310 "Number of bands (%d) exceeds limit (%d).\n",
1311 sect_end, ics->max_sfb);
1312 return AVERROR_INVALIDDATA;
1314 } while (sect_len_incr == (1 << bits) - 1);
1315 for (; k < sect_end; k++) {
1316 band_type [idx] = sect_band_type;
1317 band_type_run_end[idx++] = sect_end;
1325 * Decode scalefactors; reference: table 4.47.
1327 * @param global_gain first scalefactor value as scalefactors are differentially coded
1328 * @param band_type array of the used band type
1329 * @param band_type_run_end array of the last scalefactor band of a band type run
1330 * @param sf array of scalefactors or intensity stereo positions
1332 * @return Returns error status. 0 - OK, !0 - error
1334 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1335 unsigned int global_gain,
1336 IndividualChannelStream *ics,
1337 enum BandType band_type[120],
1338 int band_type_run_end[120])
1341 int offset[3] = { global_gain, global_gain - 90, 0 };
1344 for (g = 0; g < ics->num_window_groups; g++) {
1345 for (i = 0; i < ics->max_sfb;) {
1346 int run_end = band_type_run_end[idx];
1347 if (band_type[idx] == ZERO_BT) {
1348 for (; i < run_end; i++, idx++)
1350 } else if ((band_type[idx] == INTENSITY_BT) ||
1351 (band_type[idx] == INTENSITY_BT2)) {
1352 for (; i < run_end; i++, idx++) {
1353 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1354 clipped_offset = av_clip(offset[2], -155, 100);
1355 if (offset[2] != clipped_offset) {
1356 avpriv_request_sample(ac->avctx,
1357 "If you heard an audible artifact, there may be a bug in the decoder. "
1358 "Clipped intensity stereo position (%d -> %d)",
1359 offset[2], clipped_offset);
1361 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1363 } else if (band_type[idx] == NOISE_BT) {
1364 for (; i < run_end; i++, idx++) {
1365 if (noise_flag-- > 0)
1366 offset[1] += get_bits(gb, 9) - 256;
1368 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1369 clipped_offset = av_clip(offset[1], -100, 155);
1370 if (offset[1] != clipped_offset) {
1371 avpriv_request_sample(ac->avctx,
1372 "If you heard an audible artifact, there may be a bug in the decoder. "
1373 "Clipped noise gain (%d -> %d)",
1374 offset[1], clipped_offset);
1376 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1379 for (; i < run_end; i++, idx++) {
1380 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1381 if (offset[0] > 255U) {
1382 av_log(ac->avctx, AV_LOG_ERROR,
1383 "Scalefactor (%d) out of range.\n", offset[0]);
1384 return AVERROR_INVALIDDATA;
1386 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1395 * Decode pulse data; reference: table 4.7.
1397 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1398 const uint16_t *swb_offset, int num_swb)
1401 pulse->num_pulse = get_bits(gb, 2) + 1;
1402 pulse_swb = get_bits(gb, 6);
1403 if (pulse_swb >= num_swb)
1405 pulse->pos[0] = swb_offset[pulse_swb];
1406 pulse->pos[0] += get_bits(gb, 5);
1407 if (pulse->pos[0] > 1023)
1409 pulse->amp[0] = get_bits(gb, 4);
1410 for (i = 1; i < pulse->num_pulse; i++) {
1411 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1412 if (pulse->pos[i] > 1023)
1414 pulse->amp[i] = get_bits(gb, 4);
1420 * Decode Temporal Noise Shaping data; reference: table 4.48.
1422 * @return Returns error status. 0 - OK, !0 - error
1424 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1425 GetBitContext *gb, const IndividualChannelStream *ics)
1427 int w, filt, i, coef_len, coef_res, coef_compress;
1428 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1429 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1430 for (w = 0; w < ics->num_windows; w++) {
1431 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1432 coef_res = get_bits1(gb);
1434 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1436 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1438 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1439 av_log(ac->avctx, AV_LOG_ERROR,
1440 "TNS filter order %d is greater than maximum %d.\n",
1441 tns->order[w][filt], tns_max_order);
1442 tns->order[w][filt] = 0;
1443 return AVERROR_INVALIDDATA;
1445 if (tns->order[w][filt]) {
1446 tns->direction[w][filt] = get_bits1(gb);
1447 coef_compress = get_bits1(gb);
1448 coef_len = coef_res + 3 - coef_compress;
1449 tmp2_idx = 2 * coef_compress + coef_res;
1451 for (i = 0; i < tns->order[w][filt]; i++)
1452 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1461 * Decode Mid/Side data; reference: table 4.54.
1463 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1464 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1465 * [3] reserved for scalable AAC
1467 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1471 if (ms_present == 1) {
1473 idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1475 cpe->ms_mask[idx] = get_bits1(gb);
1476 } else if (ms_present == 2) {
1477 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1482 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1486 *dst++ = v[idx & 15] * s;
1487 *dst++ = v[idx>>4 & 15] * s;
1493 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1497 *dst++ = v[idx & 3] * s;
1498 *dst++ = v[idx>>2 & 3] * s;
1499 *dst++ = v[idx>>4 & 3] * s;
1500 *dst++ = v[idx>>6 & 3] * s;
1506 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1507 unsigned sign, const float *scale)
1509 union av_intfloat32 s0, s1;
1511 s0.f = s1.f = *scale;
1512 s0.i ^= sign >> 1 << 31;
1515 *dst++ = v[idx & 15] * s0.f;
1516 *dst++ = v[idx>>4 & 15] * s1.f;
1523 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1524 unsigned sign, const float *scale)
1526 unsigned nz = idx >> 12;
1527 union av_intfloat32 s = { .f = *scale };
1528 union av_intfloat32 t;
1530 t.i = s.i ^ (sign & 1U<<31);
1531 *dst++ = v[idx & 3] * t.f;
1533 sign <<= nz & 1; nz >>= 1;
1534 t.i = s.i ^ (sign & 1U<<31);
1535 *dst++ = v[idx>>2 & 3] * t.f;
1537 sign <<= nz & 1; nz >>= 1;
1538 t.i = s.i ^ (sign & 1U<<31);
1539 *dst++ = v[idx>>4 & 3] * t.f;
1542 t.i = s.i ^ (sign & 1U<<31);
1543 *dst++ = v[idx>>6 & 3] * t.f;
1550 * Decode spectral data; reference: table 4.50.
1551 * Dequantize and scale spectral data; reference: 4.6.3.3.
1553 * @param coef array of dequantized, scaled spectral data
1554 * @param sf array of scalefactors or intensity stereo positions
1555 * @param pulse_present set if pulses are present
1556 * @param pulse pointer to pulse data struct
1557 * @param band_type array of the used band type
1559 * @return Returns error status. 0 - OK, !0 - error
1561 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1562 GetBitContext *gb, const float sf[120],
1563 int pulse_present, const Pulse *pulse,
1564 const IndividualChannelStream *ics,
1565 enum BandType band_type[120])
1567 int i, k, g, idx = 0;
1568 const int c = 1024 / ics->num_windows;
1569 const uint16_t *offsets = ics->swb_offset;
1570 float *coef_base = coef;
1572 for (g = 0; g < ics->num_windows; g++)
1573 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1574 sizeof(float) * (c - offsets[ics->max_sfb]));
1576 for (g = 0; g < ics->num_window_groups; g++) {
1577 unsigned g_len = ics->group_len[g];
1579 for (i = 0; i < ics->max_sfb; i++, idx++) {
1580 const unsigned cbt_m1 = band_type[idx] - 1;
1581 float *cfo = coef + offsets[i];
1582 int off_len = offsets[i + 1] - offsets[i];
1585 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1586 for (group = 0; group < g_len; group++, cfo+=128) {
1587 memset(cfo, 0, off_len * sizeof(float));
1589 } else if (cbt_m1 == NOISE_BT - 1) {
1590 for (group = 0; group < g_len; group++, cfo+=128) {
1594 for (k = 0; k < off_len; k++) {
1595 ac->random_state = lcg_random(ac->random_state);
1596 cfo[k] = ac->random_state;
1599 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1600 scale = sf[idx] / sqrtf(band_energy);
1601 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1604 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1605 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1606 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1607 OPEN_READER(re, gb);
1609 switch (cbt_m1 >> 1) {
1611 for (group = 0; group < g_len; group++, cfo+=128) {
1619 UPDATE_CACHE(re, gb);
1620 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1621 cb_idx = cb_vector_idx[code];
1622 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1628 for (group = 0; group < g_len; group++, cfo+=128) {
1638 UPDATE_CACHE(re, gb);
1639 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1640 cb_idx = cb_vector_idx[code];
1641 nnz = cb_idx >> 8 & 15;
1642 bits = nnz ? GET_CACHE(re, gb) : 0;
1643 LAST_SKIP_BITS(re, gb, nnz);
1644 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1650 for (group = 0; group < g_len; group++, cfo+=128) {
1658 UPDATE_CACHE(re, gb);
1659 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1660 cb_idx = cb_vector_idx[code];
1661 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1668 for (group = 0; group < g_len; group++, cfo+=128) {
1678 UPDATE_CACHE(re, gb);
1679 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1680 cb_idx = cb_vector_idx[code];
1681 nnz = cb_idx >> 8 & 15;
1682 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1683 LAST_SKIP_BITS(re, gb, nnz);
1684 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1690 for (group = 0; group < g_len; group++, cfo+=128) {
1692 uint32_t *icf = (uint32_t *) cf;
1702 UPDATE_CACHE(re, gb);
1703 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1711 cb_idx = cb_vector_idx[code];
1714 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1715 LAST_SKIP_BITS(re, gb, nnz);
1717 for (j = 0; j < 2; j++) {
1721 /* The total length of escape_sequence must be < 22 bits according
1722 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1723 UPDATE_CACHE(re, gb);
1724 b = GET_CACHE(re, gb);
1725 b = 31 - av_log2(~b);
1728 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1729 return AVERROR_INVALIDDATA;
1732 SKIP_BITS(re, gb, b + 1);
1734 n = (1 << b) + SHOW_UBITS(re, gb, b);
1735 LAST_SKIP_BITS(re, gb, b);
1736 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1739 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1740 *icf++ = (bits & 1U<<31) | v;
1747 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1751 CLOSE_READER(re, gb);
1757 if (pulse_present) {
1759 for (i = 0; i < pulse->num_pulse; i++) {
1760 float co = coef_base[ pulse->pos[i] ];
1761 while (offsets[idx + 1] <= pulse->pos[i])
1763 if (band_type[idx] != NOISE_BT && sf[idx]) {
1764 float ico = -pulse->amp[i];
1767 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1769 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1776 static av_always_inline float flt16_round(float pf)
1778 union av_intfloat32 tmp;
1780 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1784 static av_always_inline float flt16_even(float pf)
1786 union av_intfloat32 tmp;
1788 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1792 static av_always_inline float flt16_trunc(float pf)
1794 union av_intfloat32 pun;
1796 pun.i &= 0xFFFF0000U;
1800 static av_always_inline void predict(PredictorState *ps, float *coef,
1803 const float a = 0.953125; // 61.0 / 64
1804 const float alpha = 0.90625; // 29.0 / 32
1808 float r0 = ps->r0, r1 = ps->r1;
1809 float cor0 = ps->cor0, cor1 = ps->cor1;
1810 float var0 = ps->var0, var1 = ps->var1;
1812 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1813 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1815 pv = flt16_round(k1 * r0 + k2 * r1);
1822 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1823 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1824 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1825 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1827 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1828 ps->r0 = flt16_trunc(a * e0);
1832 * Apply AAC-Main style frequency domain prediction.
1834 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1838 if (!sce->ics.predictor_initialized) {
1839 reset_all_predictors(sce->predictor_state);
1840 sce->ics.predictor_initialized = 1;
1843 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1845 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1847 for (k = sce->ics.swb_offset[sfb];
1848 k < sce->ics.swb_offset[sfb + 1];
1850 predict(&sce->predictor_state[k], &sce->coeffs[k],
1851 sce->ics.predictor_present &&
1852 sce->ics.prediction_used[sfb]);
1855 if (sce->ics.predictor_reset_group)
1856 reset_predictor_group(sce->predictor_state,
1857 sce->ics.predictor_reset_group);
1859 reset_all_predictors(sce->predictor_state);
1863 * Decode an individual_channel_stream payload; reference: table 4.44.
1865 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1866 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1868 * @return Returns error status. 0 - OK, !0 - error
1870 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1871 GetBitContext *gb, int common_window, int scale_flag)
1874 TemporalNoiseShaping *tns = &sce->tns;
1875 IndividualChannelStream *ics = &sce->ics;
1876 float *out = sce->coeffs;
1877 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1880 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1881 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1882 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1883 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1884 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1886 /* This assignment is to silence a GCC warning about the variable being used
1887 * uninitialized when in fact it always is.
1889 pulse.num_pulse = 0;
1891 global_gain = get_bits(gb, 8);
1893 if (!common_window && !scale_flag) {
1894 if (decode_ics_info(ac, ics, gb) < 0)
1895 return AVERROR_INVALIDDATA;
1898 if ((ret = decode_band_types(ac, sce->band_type,
1899 sce->band_type_run_end, gb, ics)) < 0)
1901 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1902 sce->band_type, sce->band_type_run_end)) < 0)
1907 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1908 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1909 av_log(ac->avctx, AV_LOG_ERROR,
1910 "Pulse tool not allowed in eight short sequence.\n");
1911 return AVERROR_INVALIDDATA;
1913 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1914 av_log(ac->avctx, AV_LOG_ERROR,
1915 "Pulse data corrupt or invalid.\n");
1916 return AVERROR_INVALIDDATA;
1919 tns->present = get_bits1(gb);
1920 if (tns->present && !er_syntax)
1921 if (decode_tns(ac, tns, gb, ics) < 0)
1922 return AVERROR_INVALIDDATA;
1923 if (!eld_syntax && get_bits1(gb)) {
1924 avpriv_request_sample(ac->avctx, "SSR");
1925 return AVERROR_PATCHWELCOME;
1927 // I see no textual basis in the spec for this occuring after SSR gain
1928 // control, but this is what both reference and real implmentations do
1929 if (tns->present && er_syntax)
1930 if (decode_tns(ac, tns, gb, ics) < 0)
1931 return AVERROR_INVALIDDATA;
1934 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1935 &pulse, ics, sce->band_type) < 0)
1936 return AVERROR_INVALIDDATA;
1938 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1939 apply_prediction(ac, sce);
1945 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1947 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1949 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1950 float *ch0 = cpe->ch[0].coeffs;
1951 float *ch1 = cpe->ch[1].coeffs;
1952 int g, i, group, idx = 0;
1953 const uint16_t *offsets = ics->swb_offset;
1954 for (g = 0; g < ics->num_window_groups; g++) {
1955 for (i = 0; i < ics->max_sfb; i++, idx++) {
1956 if (cpe->ms_mask[idx] &&
1957 cpe->ch[0].band_type[idx] < NOISE_BT &&
1958 cpe->ch[1].band_type[idx] < NOISE_BT) {
1959 for (group = 0; group < ics->group_len[g]; group++) {
1960 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1961 ch1 + group * 128 + offsets[i],
1962 offsets[i+1] - offsets[i]);
1966 ch0 += ics->group_len[g] * 128;
1967 ch1 += ics->group_len[g] * 128;
1972 * intensity stereo decoding; reference: 4.6.8.2.3
1974 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1975 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1976 * [3] reserved for scalable AAC
1978 static void apply_intensity_stereo(AACContext *ac,
1979 ChannelElement *cpe, int ms_present)
1981 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1982 SingleChannelElement *sce1 = &cpe->ch[1];
1983 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1984 const uint16_t *offsets = ics->swb_offset;
1985 int g, group, i, idx = 0;
1988 for (g = 0; g < ics->num_window_groups; g++) {
1989 for (i = 0; i < ics->max_sfb;) {
1990 if (sce1->band_type[idx] == INTENSITY_BT ||
1991 sce1->band_type[idx] == INTENSITY_BT2) {
1992 const int bt_run_end = sce1->band_type_run_end[idx];
1993 for (; i < bt_run_end; i++, idx++) {
1994 c = -1 + 2 * (sce1->band_type[idx] - 14);
1996 c *= 1 - 2 * cpe->ms_mask[idx];
1997 scale = c * sce1->sf[idx];
1998 for (group = 0; group < ics->group_len[g]; group++)
1999 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2000 coef0 + group * 128 + offsets[i],
2002 offsets[i + 1] - offsets[i]);
2005 int bt_run_end = sce1->band_type_run_end[idx];
2006 idx += bt_run_end - i;
2010 coef0 += ics->group_len[g] * 128;
2011 coef1 += ics->group_len[g] * 128;
2016 * Decode a channel_pair_element; reference: table 4.4.
2018 * @return Returns error status. 0 - OK, !0 - error
2020 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2022 int i, ret, common_window, ms_present = 0;
2023 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2025 common_window = eld_syntax || get_bits1(gb);
2026 if (common_window) {
2027 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2028 return AVERROR_INVALIDDATA;
2029 i = cpe->ch[1].ics.use_kb_window[0];
2030 cpe->ch[1].ics = cpe->ch[0].ics;
2031 cpe->ch[1].ics.use_kb_window[1] = i;
2032 if (cpe->ch[1].ics.predictor_present &&
2033 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2034 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2035 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2036 ms_present = get_bits(gb, 2);
2037 if (ms_present == 3) {
2038 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2039 return AVERROR_INVALIDDATA;
2040 } else if (ms_present)
2041 decode_mid_side_stereo(cpe, gb, ms_present);
2043 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2045 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2048 if (common_window) {
2050 apply_mid_side_stereo(ac, cpe);
2051 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2052 apply_prediction(ac, &cpe->ch[0]);
2053 apply_prediction(ac, &cpe->ch[1]);
2057 apply_intensity_stereo(ac, cpe, ms_present);
2061 static const float cce_scale[] = {
2062 1.09050773266525765921, //2^(1/8)
2063 1.18920711500272106672, //2^(1/4)
2069 * Decode coupling_channel_element; reference: table 4.8.
2071 * @return Returns error status. 0 - OK, !0 - error
2073 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2079 SingleChannelElement *sce = &che->ch[0];
2080 ChannelCoupling *coup = &che->coup;
2082 coup->coupling_point = 2 * get_bits1(gb);
2083 coup->num_coupled = get_bits(gb, 3);
2084 for (c = 0; c <= coup->num_coupled; c++) {
2086 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2087 coup->id_select[c] = get_bits(gb, 4);
2088 if (coup->type[c] == TYPE_CPE) {
2089 coup->ch_select[c] = get_bits(gb, 2);
2090 if (coup->ch_select[c] == 3)
2093 coup->ch_select[c] = 2;
2095 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2097 sign = get_bits(gb, 1);
2098 scale = cce_scale[get_bits(gb, 2)];
2100 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2103 for (c = 0; c < num_gain; c++) {
2107 float gain_cache = 1.0;
2109 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2110 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2111 gain_cache = powf(scale, -gain);
2113 if (coup->coupling_point == AFTER_IMDCT) {
2114 coup->gain[c][0] = gain_cache;
2116 for (g = 0; g < sce->ics.num_window_groups; g++) {
2117 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2118 if (sce->band_type[idx] != ZERO_BT) {
2120 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2128 gain_cache = powf(scale, -t) * s;
2131 coup->gain[c][idx] = gain_cache;
2141 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2143 * @return Returns number of bytes consumed.
2145 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2149 int num_excl_chan = 0;
2152 for (i = 0; i < 7; i++)
2153 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2154 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2156 return num_excl_chan / 7;
2160 * Decode dynamic range information; reference: table 4.52.
2162 * @return Returns number of bytes consumed.
2164 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2168 int drc_num_bands = 1;
2171 /* pce_tag_present? */
2172 if (get_bits1(gb)) {
2173 che_drc->pce_instance_tag = get_bits(gb, 4);
2174 skip_bits(gb, 4); // tag_reserved_bits
2178 /* excluded_chns_present? */
2179 if (get_bits1(gb)) {
2180 n += decode_drc_channel_exclusions(che_drc, gb);
2183 /* drc_bands_present? */
2184 if (get_bits1(gb)) {
2185 che_drc->band_incr = get_bits(gb, 4);
2186 che_drc->interpolation_scheme = get_bits(gb, 4);
2188 drc_num_bands += che_drc->band_incr;
2189 for (i = 0; i < drc_num_bands; i++) {
2190 che_drc->band_top[i] = get_bits(gb, 8);
2195 /* prog_ref_level_present? */
2196 if (get_bits1(gb)) {
2197 che_drc->prog_ref_level = get_bits(gb, 7);
2198 skip_bits1(gb); // prog_ref_level_reserved_bits
2202 for (i = 0; i < drc_num_bands; i++) {
2203 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2204 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2211 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2213 int i, major, minor;
2218 get_bits(gb, 13); len -= 13;
2220 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2221 buf[i] = get_bits(gb, 8);
2224 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2225 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2227 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2228 ac->avctx->internal->skip_samples = 1024;
2232 skip_bits_long(gb, len);
2238 * Decode extension data (incomplete); reference: table 4.51.
2240 * @param cnt length of TYPE_FIL syntactic element in bytes
2242 * @return Returns number of bytes consumed
2244 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2245 ChannelElement *che, enum RawDataBlockType elem_type)
2249 switch (get_bits(gb, 4)) { // extension type
2250 case EXT_SBR_DATA_CRC:
2254 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2256 } else if (!ac->oc[1].m4ac.sbr) {
2257 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2258 skip_bits_long(gb, 8 * cnt - 4);
2260 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2261 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2262 skip_bits_long(gb, 8 * cnt - 4);
2264 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2265 ac->oc[1].m4ac.sbr = 1;
2266 ac->oc[1].m4ac.ps = 1;
2267 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2268 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2269 ac->oc[1].status, 1);
2271 ac->oc[1].m4ac.sbr = 1;
2272 ac->avctx->profile = FF_PROFILE_AAC_HE;
2274 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2276 case EXT_DYNAMIC_RANGE:
2277 res = decode_dynamic_range(&ac->che_drc, gb);
2280 decode_fill(ac, gb, 8 * cnt - 4);
2283 case EXT_DATA_ELEMENT:
2285 skip_bits_long(gb, 8 * cnt - 4);
2292 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2294 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2295 * @param coef spectral coefficients
2297 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2298 IndividualChannelStream *ics, int decode)
2300 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2302 int bottom, top, order, start, end, size, inc;
2303 float lpc[TNS_MAX_ORDER];
2304 float tmp[TNS_MAX_ORDER+1];
2306 for (w = 0; w < ics->num_windows; w++) {
2307 bottom = ics->num_swb;
2308 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2310 bottom = FFMAX(0, top - tns->length[w][filt]);
2311 order = tns->order[w][filt];
2316 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2318 start = ics->swb_offset[FFMIN(bottom, mmm)];
2319 end = ics->swb_offset[FFMIN( top, mmm)];
2320 if ((size = end - start) <= 0)
2322 if (tns->direction[w][filt]) {
2332 for (m = 0; m < size; m++, start += inc)
2333 for (i = 1; i <= FFMIN(m, order); i++)
2334 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2337 for (m = 0; m < size; m++, start += inc) {
2338 tmp[0] = coef[start];
2339 for (i = 1; i <= FFMIN(m, order); i++)
2340 coef[start] += tmp[i] * lpc[i - 1];
2341 for (i = order; i > 0; i--)
2342 tmp[i] = tmp[i - 1];
2350 * Apply windowing and MDCT to obtain the spectral
2351 * coefficient from the predicted sample by LTP.
2353 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2354 float *in, IndividualChannelStream *ics)
2356 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2357 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2358 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2359 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2361 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2362 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2364 memset(in, 0, 448 * sizeof(float));
2365 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2367 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2368 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2370 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2371 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2373 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2377 * Apply the long term prediction
2379 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2381 const LongTermPrediction *ltp = &sce->ics.ltp;
2382 const uint16_t *offsets = sce->ics.swb_offset;
2385 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2386 float *predTime = sce->ret;
2387 float *predFreq = ac->buf_mdct;
2388 int16_t num_samples = 2048;
2390 if (ltp->lag < 1024)
2391 num_samples = ltp->lag + 1024;
2392 for (i = 0; i < num_samples; i++)
2393 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2394 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2396 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2398 if (sce->tns.present)
2399 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2401 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2403 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2404 sce->coeffs[i] += predFreq[i];
2409 * Update the LTP buffer for next frame
2411 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2413 IndividualChannelStream *ics = &sce->ics;
2414 float *saved = sce->saved;
2415 float *saved_ltp = sce->coeffs;
2416 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2417 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2420 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2421 memcpy(saved_ltp, saved, 512 * sizeof(float));
2422 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2423 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2424 for (i = 0; i < 64; i++)
2425 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2426 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2427 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2428 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2429 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2430 for (i = 0; i < 64; i++)
2431 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2432 } else { // LONG_STOP or ONLY_LONG
2433 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2434 for (i = 0; i < 512; i++)
2435 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2438 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2439 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2440 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2444 * Conduct IMDCT and windowing.
2446 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2448 IndividualChannelStream *ics = &sce->ics;
2449 float *in = sce->coeffs;
2450 float *out = sce->ret;
2451 float *saved = sce->saved;
2452 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2453 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2454 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2455 float *buf = ac->buf_mdct;
2456 float *temp = ac->temp;
2460 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2461 for (i = 0; i < 1024; i += 128)
2462 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2464 ac->mdct.imdct_half(&ac->mdct, buf, in);
2466 /* window overlapping
2467 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2468 * and long to short transitions are considered to be short to short
2469 * transitions. This leaves just two cases (long to long and short to short)
2470 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2472 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2473 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2474 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2476 memcpy( out, saved, 448 * sizeof(float));
2478 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2479 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2480 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2481 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2482 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2483 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2484 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2486 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2487 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2492 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2493 memcpy( saved, temp + 64, 64 * sizeof(float));
2494 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2495 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2496 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2497 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2498 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2499 memcpy( saved, buf + 512, 448 * sizeof(float));
2500 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2501 } else { // LONG_STOP or ONLY_LONG
2502 memcpy( saved, buf + 512, 512 * sizeof(float));
2506 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2508 IndividualChannelStream *ics = &sce->ics;
2509 float *in = sce->coeffs;
2510 float *out = sce->ret;
2511 float *saved = sce->saved;
2512 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_512 : ff_sine_512;
2513 float *buf = ac->buf_mdct;
2516 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2518 // window overlapping
2519 ac->fdsp.vector_fmul_window(out, saved, buf, lwindow_prev, 256);
2522 memcpy(saved, buf + 256, 256 * sizeof(float));
2525 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2527 float *in = sce->coeffs;
2528 float *out = sce->ret;
2529 float *saved = sce->saved;
2530 const float *const window = ff_aac_eld_window;
2531 float *buf = ac->buf_mdct;
2534 const int n2 = n >> 1;
2535 const int n4 = n >> 2;
2537 // Inverse transform, mapped to the conventional IMDCT by
2538 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2539 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2540 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2541 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2542 for (i = 0; i < n2; i+=2) {
2544 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2545 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2547 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2548 for (i = 0; i < n; i+=2) {
2551 // Like with the regular IMDCT at this point we still have the middle half
2552 // of a transform but with even symmetry on the left and odd symmetry on
2555 // window overlapping
2556 // The spec says to use samples [0..511] but the reference decoder uses
2557 // samples [128..639].
2558 for (i = n4; i < n2; i ++) {
2559 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2560 saved[ i + n2] * window[i + n - n4] +
2561 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2562 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2564 for (i = 0; i < n2; i ++) {
2565 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2566 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2567 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2568 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2570 for (i = 0; i < n4; i ++) {
2571 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2572 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2573 -saved[ n + n2 + i] * window[i + 3*n - n4];
2577 memmove(saved + n, saved, 2 * n * sizeof(float));
2578 memcpy( saved, buf, n * sizeof(float));
2582 * Apply dependent channel coupling (applied before IMDCT).
2584 * @param index index into coupling gain array
2586 static void apply_dependent_coupling(AACContext *ac,
2587 SingleChannelElement *target,
2588 ChannelElement *cce, int index)
2590 IndividualChannelStream *ics = &cce->ch[0].ics;
2591 const uint16_t *offsets = ics->swb_offset;
2592 float *dest = target->coeffs;
2593 const float *src = cce->ch[0].coeffs;
2594 int g, i, group, k, idx = 0;
2595 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2596 av_log(ac->avctx, AV_LOG_ERROR,
2597 "Dependent coupling is not supported together with LTP\n");
2600 for (g = 0; g < ics->num_window_groups; g++) {
2601 for (i = 0; i < ics->max_sfb; i++, idx++) {
2602 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2603 const float gain = cce->coup.gain[index][idx];
2604 for (group = 0; group < ics->group_len[g]; group++) {
2605 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2607 dest[group * 128 + k] += gain * src[group * 128 + k];
2612 dest += ics->group_len[g] * 128;
2613 src += ics->group_len[g] * 128;
2618 * Apply independent channel coupling (applied after IMDCT).
2620 * @param index index into coupling gain array
2622 static void apply_independent_coupling(AACContext *ac,
2623 SingleChannelElement *target,
2624 ChannelElement *cce, int index)
2627 const float gain = cce->coup.gain[index][0];
2628 const float *src = cce->ch[0].ret;
2629 float *dest = target->ret;
2630 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2632 for (i = 0; i < len; i++)
2633 dest[i] += gain * src[i];
2637 * channel coupling transformation interface
2639 * @param apply_coupling_method pointer to (in)dependent coupling function
2641 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2642 enum RawDataBlockType type, int elem_id,
2643 enum CouplingPoint coupling_point,
2644 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2648 for (i = 0; i < MAX_ELEM_ID; i++) {
2649 ChannelElement *cce = ac->che[TYPE_CCE][i];
2652 if (cce && cce->coup.coupling_point == coupling_point) {
2653 ChannelCoupling *coup = &cce->coup;
2655 for (c = 0; c <= coup->num_coupled; c++) {
2656 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2657 if (coup->ch_select[c] != 1) {
2658 apply_coupling_method(ac, &cc->ch[0], cce, index);
2659 if (coup->ch_select[c] != 0)
2662 if (coup->ch_select[c] != 2)
2663 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2665 index += 1 + (coup->ch_select[c] == 3);
2672 * Convert spectral data to float samples, applying all supported tools as appropriate.
2674 static void spectral_to_sample(AACContext *ac)
2677 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2678 switch (ac->oc[1].m4ac.object_type) {
2680 imdct_and_window = imdct_and_windowing_ld;
2682 case AOT_ER_AAC_ELD:
2683 imdct_and_window = imdct_and_windowing_eld;
2686 imdct_and_window = ac->imdct_and_windowing;
2688 for (type = 3; type >= 0; type--) {
2689 for (i = 0; i < MAX_ELEM_ID; i++) {
2690 ChannelElement *che = ac->che[type][i];
2692 if (type <= TYPE_CPE)
2693 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2694 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2695 if (che->ch[0].ics.predictor_present) {
2696 if (che->ch[0].ics.ltp.present)
2697 ac->apply_ltp(ac, &che->ch[0]);
2698 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2699 ac->apply_ltp(ac, &che->ch[1]);
2702 if (che->ch[0].tns.present)
2703 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2704 if (che->ch[1].tns.present)
2705 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2706 if (type <= TYPE_CPE)
2707 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2708 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2709 imdct_and_window(ac, &che->ch[0]);
2710 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2711 ac->update_ltp(ac, &che->ch[0]);
2712 if (type == TYPE_CPE) {
2713 imdct_and_window(ac, &che->ch[1]);
2714 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2715 ac->update_ltp(ac, &che->ch[1]);
2717 if (ac->oc[1].m4ac.sbr > 0) {
2718 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2721 if (type <= TYPE_CCE)
2722 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2728 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2731 AACADTSHeaderInfo hdr_info;
2732 uint8_t layout_map[MAX_ELEM_ID*4][3];
2733 int layout_map_tags, ret;
2735 size = avpriv_aac_parse_header(gb, &hdr_info);
2737 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2738 // This is 2 for "VLB " audio in NSV files.
2739 // See samples/nsv/vlb_audio.
2740 avpriv_report_missing_feature(ac->avctx,
2741 "More than one AAC RDB per ADTS frame");
2742 ac->warned_num_aac_frames = 1;
2744 push_output_configuration(ac);
2745 if (hdr_info.chan_config) {
2746 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2747 if ((ret = set_default_channel_config(ac->avctx,
2750 hdr_info.chan_config)) < 0)
2752 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2753 FFMAX(ac->oc[1].status,
2754 OC_TRIAL_FRAME), 0)) < 0)
2757 ac->oc[1].m4ac.chan_config = 0;
2759 * dual mono frames in Japanese DTV can have chan_config 0
2760 * WITHOUT specifying PCE.
2761 * thus, set dual mono as default.
2763 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2764 layout_map_tags = 2;
2765 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2766 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2767 layout_map[0][1] = 0;
2768 layout_map[1][1] = 1;
2769 if (output_configure(ac, layout_map, layout_map_tags,
2774 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2775 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2776 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2777 if (ac->oc[0].status != OC_LOCKED ||
2778 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2779 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2780 ac->oc[1].m4ac.sbr = -1;
2781 ac->oc[1].m4ac.ps = -1;
2783 if (!hdr_info.crc_absent)
2789 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2790 int *got_frame_ptr, GetBitContext *gb)
2792 AACContext *ac = avctx->priv_data;
2793 ChannelElement *che;
2796 int chan_config = ac->oc[1].m4ac.chan_config;
2797 int aot = ac->oc[1].m4ac.object_type;
2799 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2804 if ((err = frame_configure_elements(avctx)) < 0)
2807 // The FF_PROFILE_AAC_* defines are all object_type - 1
2808 // This may lead to an undefined profile being signaled
2809 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2811 ac->tags_mapped = 0;
2813 if (chan_config < 0 || chan_config >= 8) {
2814 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2815 ac->oc[1].m4ac.chan_config);
2816 return AVERROR_INVALIDDATA;
2818 for (i = 0; i < tags_per_config[chan_config]; i++) {
2819 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2820 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2821 if (!(che=get_che(ac, elem_type, elem_id))) {
2822 av_log(ac->avctx, AV_LOG_ERROR,
2823 "channel element %d.%d is not allocated\n",
2824 elem_type, elem_id);
2825 return AVERROR_INVALIDDATA;
2827 if (aot != AOT_ER_AAC_ELD)
2829 switch (elem_type) {
2831 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2834 err = decode_cpe(ac, gb, che);
2837 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2844 spectral_to_sample(ac);
2846 ac->frame->nb_samples = samples;
2849 skip_bits_long(gb, get_bits_left(gb));
2853 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2854 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2856 AACContext *ac = avctx->priv_data;
2857 ChannelElement *che = NULL, *che_prev = NULL;
2858 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2860 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2861 int is_dmono, sce_count = 0;
2865 if (show_bits(gb, 12) == 0xfff) {
2866 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2867 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2870 if (ac->oc[1].m4ac.sampling_index > 12) {
2871 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2872 err = AVERROR_INVALIDDATA;
2877 if ((err = frame_configure_elements(avctx)) < 0)
2880 // The FF_PROFILE_AAC_* defines are all object_type - 1
2881 // This may lead to an undefined profile being signaled
2882 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2884 ac->tags_mapped = 0;
2886 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2887 elem_id = get_bits(gb, 4);
2889 if (elem_type < TYPE_DSE) {
2890 if (!(che=get_che(ac, elem_type, elem_id))) {
2891 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2892 elem_type, elem_id);
2893 err = AVERROR_INVALIDDATA;
2899 switch (elem_type) {
2902 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2908 err = decode_cpe(ac, gb, che);
2913 err = decode_cce(ac, gb, che);
2917 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2922 err = skip_data_stream_element(ac, gb);
2926 uint8_t layout_map[MAX_ELEM_ID*4][3];
2928 push_output_configuration(ac);
2929 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2935 av_log(avctx, AV_LOG_ERROR,
2936 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2938 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2940 ac->oc[1].m4ac.chan_config = 0;
2948 elem_id += get_bits(gb, 8) - 1;
2949 if (get_bits_left(gb) < 8 * elem_id) {
2950 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
2951 err = AVERROR_INVALIDDATA;
2955 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2956 err = 0; /* FIXME */
2960 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
2965 elem_type_prev = elem_type;
2970 if (get_bits_left(gb) < 3) {
2971 av_log(avctx, AV_LOG_ERROR, overread_err);
2972 err = AVERROR_INVALIDDATA;
2977 spectral_to_sample(ac);
2979 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2980 samples <<= multiplier;
2981 /* for dual-mono audio (SCE + SCE) */
2982 is_dmono = ac->dmono_mode && sce_count == 2 &&
2983 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
2986 ac->frame->nb_samples = samples;
2988 av_frame_unref(ac->frame);
2989 *got_frame_ptr = !!samples;
2992 if (ac->dmono_mode == 1)
2993 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
2994 else if (ac->dmono_mode == 2)
2995 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
2998 if (ac->oc[1].status && audio_found) {
2999 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3000 avctx->frame_size = samples;
3001 ac->oc[1].status = OC_LOCKED;
3006 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3007 if (side && side_size>=4)
3008 AV_WL32(side, 2*AV_RL32(side));
3012 pop_output_configuration(ac);
3016 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3017 int *got_frame_ptr, AVPacket *avpkt)
3019 AACContext *ac = avctx->priv_data;
3020 const uint8_t *buf = avpkt->data;
3021 int buf_size = avpkt->size;
3026 int new_extradata_size;
3027 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3028 AV_PKT_DATA_NEW_EXTRADATA,
3029 &new_extradata_size);
3030 int jp_dualmono_size;
3031 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3032 AV_PKT_DATA_JP_DUALMONO,
3035 if (new_extradata && 0) {
3036 av_free(avctx->extradata);
3037 avctx->extradata = av_mallocz(new_extradata_size +
3038 FF_INPUT_BUFFER_PADDING_SIZE);
3039 if (!avctx->extradata)
3040 return AVERROR(ENOMEM);
3041 avctx->extradata_size = new_extradata_size;
3042 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3043 push_output_configuration(ac);
3044 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3046 avctx->extradata_size*8, 1) < 0) {
3047 pop_output_configuration(ac);
3048 return AVERROR_INVALIDDATA;
3053 if (jp_dualmono && jp_dualmono_size > 0)
3054 ac->dmono_mode = 1 + *jp_dualmono;
3055 if (ac->force_dmono_mode >= 0)
3056 ac->dmono_mode = ac->force_dmono_mode;
3058 if (INT_MAX / 8 <= buf_size)
3059 return AVERROR_INVALIDDATA;
3061 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3064 switch (ac->oc[1].m4ac.object_type) {
3066 case AOT_ER_AAC_LTP:
3068 case AOT_ER_AAC_ELD:
3069 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3072 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3077 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3078 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3079 if (buf[buf_offset])
3082 return buf_size > buf_offset ? buf_consumed : buf_size;
3085 static av_cold int aac_decode_close(AVCodecContext *avctx)
3087 AACContext *ac = avctx->priv_data;
3090 for (i = 0; i < MAX_ELEM_ID; i++) {
3091 for (type = 0; type < 4; type++) {
3092 if (ac->che[type][i])
3093 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3094 av_freep(&ac->che[type][i]);
3098 ff_mdct_end(&ac->mdct);
3099 ff_mdct_end(&ac->mdct_small);
3100 ff_mdct_end(&ac->mdct_ld);
3101 ff_mdct_end(&ac->mdct_ltp);
3106 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3108 struct LATMContext {
3109 AACContext aac_ctx; ///< containing AACContext
3110 int initialized; ///< initialized after a valid extradata was seen
3113 int audio_mux_version_A; ///< LATM syntax version
3114 int frame_length_type; ///< 0/1 variable/fixed frame length
3115 int frame_length; ///< frame length for fixed frame length
3118 static inline uint32_t latm_get_value(GetBitContext *b)
3120 int length = get_bits(b, 2);
3122 return get_bits_long(b, (length+1)*8);
3125 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3126 GetBitContext *gb, int asclen)
3128 AACContext *ac = &latmctx->aac_ctx;
3129 AVCodecContext *avctx = ac->avctx;
3130 MPEG4AudioConfig m4ac = { 0 };
3131 int config_start_bit = get_bits_count(gb);
3132 int sync_extension = 0;
3133 int bits_consumed, esize;
3137 asclen = FFMIN(asclen, get_bits_left(gb));
3139 asclen = get_bits_left(gb);
3141 if (config_start_bit % 8) {
3142 avpriv_request_sample(latmctx->aac_ctx.avctx,
3143 "Non-byte-aligned audio-specific config");
3144 return AVERROR_PATCHWELCOME;
3147 return AVERROR_INVALIDDATA;
3148 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3149 gb->buffer + (config_start_bit / 8),
3150 asclen, sync_extension);
3152 if (bits_consumed < 0)
3153 return AVERROR_INVALIDDATA;
3155 if (!latmctx->initialized ||
3156 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3157 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3159 if(latmctx->initialized) {
3160 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3162 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3164 latmctx->initialized = 0;
3166 esize = (bits_consumed+7) / 8;
3168 if (avctx->extradata_size < esize) {
3169 av_free(avctx->extradata);
3170 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3171 if (!avctx->extradata)
3172 return AVERROR(ENOMEM);
3175 avctx->extradata_size = esize;
3176 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3177 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3179 skip_bits_long(gb, bits_consumed);
3181 return bits_consumed;
3184 static int read_stream_mux_config(struct LATMContext *latmctx,
3187 int ret, audio_mux_version = get_bits(gb, 1);
3189 latmctx->audio_mux_version_A = 0;
3190 if (audio_mux_version)
3191 latmctx->audio_mux_version_A = get_bits(gb, 1);
3193 if (!latmctx->audio_mux_version_A) {
3195 if (audio_mux_version)
3196 latm_get_value(gb); // taraFullness
3198 skip_bits(gb, 1); // allStreamSameTimeFraming
3199 skip_bits(gb, 6); // numSubFrames
3201 if (get_bits(gb, 4)) { // numPrograms
3202 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3203 return AVERROR_PATCHWELCOME;
3206 // for each program (which there is only one in DVB)
3208 // for each layer (which there is only one in DVB)
3209 if (get_bits(gb, 3)) { // numLayer
3210 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3211 return AVERROR_PATCHWELCOME;
3214 // for all but first stream: use_same_config = get_bits(gb, 1);
3215 if (!audio_mux_version) {
3216 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3219 int ascLen = latm_get_value(gb);
3220 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3223 skip_bits_long(gb, ascLen);
3226 latmctx->frame_length_type = get_bits(gb, 3);
3227 switch (latmctx->frame_length_type) {
3229 skip_bits(gb, 8); // latmBufferFullness
3232 latmctx->frame_length = get_bits(gb, 9);
3237 skip_bits(gb, 6); // CELP frame length table index
3241 skip_bits(gb, 1); // HVXC frame length table index
3245 if (get_bits(gb, 1)) { // other data
3246 if (audio_mux_version) {
3247 latm_get_value(gb); // other_data_bits
3251 esc = get_bits(gb, 1);
3257 if (get_bits(gb, 1)) // crc present
3258 skip_bits(gb, 8); // config_crc
3264 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3268 if (ctx->frame_length_type == 0) {
3269 int mux_slot_length = 0;
3271 tmp = get_bits(gb, 8);
3272 mux_slot_length += tmp;
3273 } while (tmp == 255);
3274 return mux_slot_length;
3275 } else if (ctx->frame_length_type == 1) {
3276 return ctx->frame_length;
3277 } else if (ctx->frame_length_type == 3 ||
3278 ctx->frame_length_type == 5 ||
3279 ctx->frame_length_type == 7) {
3280 skip_bits(gb, 2); // mux_slot_length_coded
3285 static int read_audio_mux_element(struct LATMContext *latmctx,
3289 uint8_t use_same_mux = get_bits(gb, 1);
3290 if (!use_same_mux) {
3291 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3293 } else if (!latmctx->aac_ctx.avctx->extradata) {
3294 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3295 "no decoder config found\n");
3296 return AVERROR(EAGAIN);
3298 if (latmctx->audio_mux_version_A == 0) {
3299 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3300 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3301 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3302 return AVERROR_INVALIDDATA;
3303 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3304 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3305 "frame length mismatch %d << %d\n",
3306 mux_slot_length_bytes * 8, get_bits_left(gb));
3307 return AVERROR_INVALIDDATA;
3314 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3315 int *got_frame_ptr, AVPacket *avpkt)
3317 struct LATMContext *latmctx = avctx->priv_data;
3321 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3324 // check for LOAS sync word
3325 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3326 return AVERROR_INVALIDDATA;
3328 muxlength = get_bits(&gb, 13) + 3;
3329 // not enough data, the parser should have sorted this out
3330 if (muxlength > avpkt->size)
3331 return AVERROR_INVALIDDATA;
3333 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3336 if (!latmctx->initialized) {
3337 if (!avctx->extradata) {
3341 push_output_configuration(&latmctx->aac_ctx);
3342 if ((err = decode_audio_specific_config(
3343 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3344 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3345 pop_output_configuration(&latmctx->aac_ctx);
3348 latmctx->initialized = 1;
3352 if (show_bits(&gb, 12) == 0xfff) {
3353 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3354 "ADTS header detected, probably as result of configuration "
3356 return AVERROR_INVALIDDATA;
3359 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
3365 static av_cold int latm_decode_init(AVCodecContext *avctx)
3367 struct LATMContext *latmctx = avctx->priv_data;
3368 int ret = aac_decode_init(avctx);
3370 if (avctx->extradata_size > 0)
3371 latmctx->initialized = !ret;
3376 static void aacdec_init(AACContext *c)
3378 c->imdct_and_windowing = imdct_and_windowing;
3379 c->apply_ltp = apply_ltp;
3380 c->apply_tns = apply_tns;
3381 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3382 c->update_ltp = update_ltp;
3385 ff_aacdec_init_mips(c);
3388 * AVOptions for Japanese DTV specific extensions (ADTS only)
3390 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3391 static const AVOption options[] = {
3392 {"dual_mono_mode", "Select the channel to decode for dual mono",
3393 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3394 AACDEC_FLAGS, "dual_mono_mode"},
3396 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3397 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3398 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3399 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3404 static const AVClass aac_decoder_class = {
3405 .class_name = "AAC decoder",
3406 .item_name = av_default_item_name,
3408 .version = LIBAVUTIL_VERSION_INT,
3411 AVCodec ff_aac_decoder = {
3413 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3414 .type = AVMEDIA_TYPE_AUDIO,
3415 .id = AV_CODEC_ID_AAC,
3416 .priv_data_size = sizeof(AACContext),
3417 .init = aac_decode_init,
3418 .close = aac_decode_close,
3419 .decode = aac_decode_frame,
3420 .sample_fmts = (const enum AVSampleFormat[]) {
3421 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3423 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3424 .channel_layouts = aac_channel_layout,
3426 .priv_class = &aac_decoder_class,
3430 Note: This decoder filter is intended to decode LATM streams transferred
3431 in MPEG transport streams which only contain one program.
3432 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3434 AVCodec ff_aac_latm_decoder = {
3436 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3437 .type = AVMEDIA_TYPE_AUDIO,
3438 .id = AV_CODEC_ID_AAC_LATM,
3439 .priv_data_size = sizeof(struct LATMContext),
3440 .init = latm_decode_init,
3441 .close = aac_decode_close,
3442 .decode = latm_decode_frame,
3443 .sample_fmts = (const enum AVSampleFormat[]) {
3444 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3446 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3447 .channel_layouts = aac_channel_layout,