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"
109 # include "arm/aac.h"
112 static VLC vlc_scalefactors;
113 static VLC vlc_spectral[11];
115 static const char overread_err[] = "Input buffer exhausted before END element found\n";
117 static int count_channels(uint8_t (*layout)[3], int tags)
120 for (i = 0; i < tags; i++) {
121 int syn_ele = layout[i][0];
122 int pos = layout[i][2];
123 sum += (1 + (syn_ele == TYPE_CPE)) *
124 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
130 * Check for the channel element in the current channel position configuration.
131 * If it exists, make sure the appropriate element is allocated and map the
132 * channel order to match the internal Libav channel layout.
134 * @param che_pos current channel position configuration
135 * @param type channel element type
136 * @param id channel element id
137 * @param channels count of the number of channels in the configuration
139 * @return Returns error status. 0 - OK, !0 - error
141 static av_cold int che_configure(AACContext *ac,
142 enum ChannelPosition che_pos,
143 int type, int id, int *channels)
145 if (*channels >= MAX_CHANNELS)
146 return AVERROR_INVALIDDATA;
148 if (!ac->che[type][id]) {
149 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
150 return AVERROR(ENOMEM);
151 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
153 if (type != TYPE_CCE) {
154 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
155 if (type == TYPE_CPE ||
156 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
157 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
161 if (ac->che[type][id])
162 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
163 av_freep(&ac->che[type][id]);
168 static int frame_configure_elements(AVCodecContext *avctx)
170 AACContext *ac = avctx->priv_data;
171 int type, id, ch, ret;
173 /* set channel pointers to internal buffers by default */
174 for (type = 0; type < 4; type++) {
175 for (id = 0; id < MAX_ELEM_ID; id++) {
176 ChannelElement *che = ac->che[type][id];
178 che->ch[0].ret = che->ch[0].ret_buf;
179 che->ch[1].ret = che->ch[1].ret_buf;
184 /* get output buffer */
185 av_frame_unref(ac->frame);
186 ac->frame->nb_samples = 2048;
187 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0) {
188 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
192 /* map output channel pointers to AVFrame data */
193 for (ch = 0; ch < avctx->channels; ch++) {
194 if (ac->output_element[ch])
195 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
201 struct elem_to_channel {
202 uint64_t av_position;
205 uint8_t aac_position;
208 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
209 uint8_t (*layout_map)[3], int offset, uint64_t left,
210 uint64_t right, int pos)
212 if (layout_map[offset][0] == TYPE_CPE) {
213 e2c_vec[offset] = (struct elem_to_channel) {
214 .av_position = left | right,
216 .elem_id = layout_map[offset][1],
221 e2c_vec[offset] = (struct elem_to_channel) {
224 .elem_id = layout_map[offset][1],
227 e2c_vec[offset + 1] = (struct elem_to_channel) {
228 .av_position = right,
230 .elem_id = layout_map[offset + 1][1],
237 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
240 int num_pos_channels = 0;
244 for (i = *current; i < tags; i++) {
245 if (layout_map[i][2] != pos)
247 if (layout_map[i][0] == TYPE_CPE) {
249 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
255 num_pos_channels += 2;
263 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
266 return num_pos_channels;
269 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
271 int i, n, total_non_cc_elements;
272 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
273 int num_front_channels, num_side_channels, num_back_channels;
276 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
281 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
282 if (num_front_channels < 0)
285 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
286 if (num_side_channels < 0)
289 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
290 if (num_back_channels < 0)
294 if (num_front_channels & 1) {
295 e2c_vec[i] = (struct elem_to_channel) {
296 .av_position = AV_CH_FRONT_CENTER,
298 .elem_id = layout_map[i][1],
299 .aac_position = AAC_CHANNEL_FRONT
302 num_front_channels--;
304 if (num_front_channels >= 4) {
305 i += assign_pair(e2c_vec, layout_map, i,
306 AV_CH_FRONT_LEFT_OF_CENTER,
307 AV_CH_FRONT_RIGHT_OF_CENTER,
309 num_front_channels -= 2;
311 if (num_front_channels >= 2) {
312 i += assign_pair(e2c_vec, layout_map, i,
316 num_front_channels -= 2;
318 while (num_front_channels >= 2) {
319 i += assign_pair(e2c_vec, layout_map, i,
323 num_front_channels -= 2;
326 if (num_side_channels >= 2) {
327 i += assign_pair(e2c_vec, layout_map, i,
331 num_side_channels -= 2;
333 while (num_side_channels >= 2) {
334 i += assign_pair(e2c_vec, layout_map, i,
338 num_side_channels -= 2;
341 while (num_back_channels >= 4) {
342 i += assign_pair(e2c_vec, layout_map, i,
346 num_back_channels -= 2;
348 if (num_back_channels >= 2) {
349 i += assign_pair(e2c_vec, layout_map, i,
353 num_back_channels -= 2;
355 if (num_back_channels) {
356 e2c_vec[i] = (struct elem_to_channel) {
357 .av_position = AV_CH_BACK_CENTER,
359 .elem_id = layout_map[i][1],
360 .aac_position = AAC_CHANNEL_BACK
366 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
367 e2c_vec[i] = (struct elem_to_channel) {
368 .av_position = AV_CH_LOW_FREQUENCY,
370 .elem_id = layout_map[i][1],
371 .aac_position = AAC_CHANNEL_LFE
375 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
376 e2c_vec[i] = (struct elem_to_channel) {
377 .av_position = UINT64_MAX,
379 .elem_id = layout_map[i][1],
380 .aac_position = AAC_CHANNEL_LFE
385 // Must choose a stable sort
386 total_non_cc_elements = n = i;
389 for (i = 1; i < n; i++)
390 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
391 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
398 for (i = 0; i < total_non_cc_elements; i++) {
399 layout_map[i][0] = e2c_vec[i].syn_ele;
400 layout_map[i][1] = e2c_vec[i].elem_id;
401 layout_map[i][2] = e2c_vec[i].aac_position;
402 if (e2c_vec[i].av_position != UINT64_MAX) {
403 layout |= e2c_vec[i].av_position;
411 * Save current output configuration if and only if it has been locked.
413 static void push_output_configuration(AACContext *ac) {
414 if (ac->oc[1].status == OC_LOCKED) {
415 ac->oc[0] = ac->oc[1];
417 ac->oc[1].status = OC_NONE;
421 * Restore the previous output configuration if and only if the current
422 * configuration is unlocked.
424 static void pop_output_configuration(AACContext *ac) {
425 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
426 ac->oc[1] = ac->oc[0];
427 ac->avctx->channels = ac->oc[1].channels;
428 ac->avctx->channel_layout = ac->oc[1].channel_layout;
433 * Configure output channel order based on the current program
434 * configuration element.
436 * @return Returns error status. 0 - OK, !0 - error
438 static int output_configure(AACContext *ac,
439 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
440 enum OCStatus oc_type, int get_new_frame)
442 AVCodecContext *avctx = ac->avctx;
443 int i, channels = 0, ret;
446 if (ac->oc[1].layout_map != layout_map) {
447 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
448 ac->oc[1].layout_map_tags = tags;
451 // Try to sniff a reasonable channel order, otherwise output the
452 // channels in the order the PCE declared them.
453 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
454 layout = sniff_channel_order(layout_map, tags);
455 for (i = 0; i < tags; i++) {
456 int type = layout_map[i][0];
457 int id = layout_map[i][1];
458 int position = layout_map[i][2];
459 // Allocate or free elements depending on if they are in the
460 // current program configuration.
461 ret = che_configure(ac, position, type, id, &channels);
465 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
466 if (layout == AV_CH_FRONT_CENTER) {
467 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
473 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
474 avctx->channel_layout = ac->oc[1].channel_layout = layout;
475 avctx->channels = ac->oc[1].channels = channels;
476 ac->oc[1].status = oc_type;
479 if ((ret = frame_configure_elements(ac->avctx)) < 0)
487 * Set up channel positions based on a default channel configuration
488 * as specified in table 1.17.
490 * @return Returns error status. 0 - OK, !0 - error
492 static int set_default_channel_config(AVCodecContext *avctx,
493 uint8_t (*layout_map)[3],
497 if (channel_config < 1 || channel_config > 7) {
498 av_log(avctx, AV_LOG_ERROR,
499 "invalid default channel configuration (%d)\n",
501 return AVERROR_INVALIDDATA;
503 *tags = tags_per_config[channel_config];
504 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
505 *tags * sizeof(*layout_map));
509 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
511 /* For PCE based channel configurations map the channels solely based
513 if (!ac->oc[1].m4ac.chan_config) {
514 return ac->tag_che_map[type][elem_id];
516 // Allow single CPE stereo files to be signalled with mono configuration.
517 if (!ac->tags_mapped && type == TYPE_CPE &&
518 ac->oc[1].m4ac.chan_config == 1) {
519 uint8_t layout_map[MAX_ELEM_ID*4][3];
521 push_output_configuration(ac);
523 if (set_default_channel_config(ac->avctx, layout_map,
524 &layout_map_tags, 2) < 0)
526 if (output_configure(ac, layout_map, layout_map_tags,
527 OC_TRIAL_FRAME, 1) < 0)
530 ac->oc[1].m4ac.chan_config = 2;
531 ac->oc[1].m4ac.ps = 0;
534 if (!ac->tags_mapped && type == TYPE_SCE &&
535 ac->oc[1].m4ac.chan_config == 2) {
536 uint8_t layout_map[MAX_ELEM_ID * 4][3];
538 push_output_configuration(ac);
540 if (set_default_channel_config(ac->avctx, layout_map,
541 &layout_map_tags, 1) < 0)
543 if (output_configure(ac, layout_map, layout_map_tags,
544 OC_TRIAL_FRAME, 1) < 0)
547 ac->oc[1].m4ac.chan_config = 1;
548 if (ac->oc[1].m4ac.sbr)
549 ac->oc[1].m4ac.ps = -1;
551 /* For indexed channel configurations map the channels solely based
553 switch (ac->oc[1].m4ac.chan_config) {
555 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
557 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
560 /* Some streams incorrectly code 5.1 audio as
561 * SCE[0] CPE[0] CPE[1] SCE[1]
563 * SCE[0] CPE[0] CPE[1] LFE[0].
564 * If we seem to have encountered such a stream, transfer
565 * the LFE[0] element to the SCE[1]'s mapping */
566 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
568 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
571 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
573 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
576 if (ac->tags_mapped == 2 &&
577 ac->oc[1].m4ac.chan_config == 4 &&
580 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
584 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
587 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
588 } else if (ac->oc[1].m4ac.chan_config == 2) {
592 if (!ac->tags_mapped && type == TYPE_SCE) {
594 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
602 * Decode an array of 4 bit element IDs, optionally interleaved with a
603 * stereo/mono switching bit.
605 * @param type speaker type/position for these channels
607 static void decode_channel_map(uint8_t layout_map[][3],
608 enum ChannelPosition type,
609 GetBitContext *gb, int n)
612 enum RawDataBlockType syn_ele;
614 case AAC_CHANNEL_FRONT:
615 case AAC_CHANNEL_BACK:
616 case AAC_CHANNEL_SIDE:
617 syn_ele = get_bits1(gb);
623 case AAC_CHANNEL_LFE:
627 layout_map[0][0] = syn_ele;
628 layout_map[0][1] = get_bits(gb, 4);
629 layout_map[0][2] = type;
635 * Decode program configuration element; reference: table 4.2.
637 * @return Returns error status. 0 - OK, !0 - error
639 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
640 uint8_t (*layout_map)[3],
643 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
648 skip_bits(gb, 2); // object_type
650 sampling_index = get_bits(gb, 4);
651 if (m4ac->sampling_index != sampling_index)
652 av_log(avctx, AV_LOG_WARNING,
653 "Sample rate index in program config element does not "
654 "match the sample rate index configured by the container.\n");
656 num_front = get_bits(gb, 4);
657 num_side = get_bits(gb, 4);
658 num_back = get_bits(gb, 4);
659 num_lfe = get_bits(gb, 2);
660 num_assoc_data = get_bits(gb, 3);
661 num_cc = get_bits(gb, 4);
664 skip_bits(gb, 4); // mono_mixdown_tag
666 skip_bits(gb, 4); // stereo_mixdown_tag
669 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
671 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
673 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
675 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
677 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
680 skip_bits_long(gb, 4 * num_assoc_data);
682 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
687 /* comment field, first byte is length */
688 comment_len = get_bits(gb, 8) * 8;
689 if (get_bits_left(gb) < comment_len) {
690 av_log(avctx, AV_LOG_ERROR, overread_err);
691 return AVERROR_INVALIDDATA;
693 skip_bits_long(gb, comment_len);
698 * Decode GA "General Audio" specific configuration; reference: table 4.1.
700 * @param ac pointer to AACContext, may be null
701 * @param avctx pointer to AVCCodecContext, used for logging
703 * @return Returns error status. 0 - OK, !0 - error
705 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
707 MPEG4AudioConfig *m4ac,
710 int extension_flag, ret, ep_config, res_flags;
711 uint8_t layout_map[MAX_ELEM_ID*4][3];
714 if (get_bits1(gb)) { // frameLengthFlag
715 avpriv_request_sample(avctx, "960/120 MDCT window");
716 return AVERROR_PATCHWELCOME;
719 if (get_bits1(gb)) // dependsOnCoreCoder
720 skip_bits(gb, 14); // coreCoderDelay
721 extension_flag = get_bits1(gb);
723 if (m4ac->object_type == AOT_AAC_SCALABLE ||
724 m4ac->object_type == AOT_ER_AAC_SCALABLE)
725 skip_bits(gb, 3); // layerNr
727 if (channel_config == 0) {
728 skip_bits(gb, 4); // element_instance_tag
729 tags = decode_pce(avctx, m4ac, layout_map, gb);
733 if ((ret = set_default_channel_config(avctx, layout_map,
734 &tags, channel_config)))
738 if (count_channels(layout_map, tags) > 1) {
740 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
743 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
746 if (extension_flag) {
747 switch (m4ac->object_type) {
749 skip_bits(gb, 5); // numOfSubFrame
750 skip_bits(gb, 11); // layer_length
754 case AOT_ER_AAC_SCALABLE:
756 res_flags = get_bits(gb, 3);
758 avpriv_report_missing_feature(avctx,
759 "AAC data resilience (flags %x)",
761 return AVERROR_PATCHWELCOME;
765 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
767 switch (m4ac->object_type) {
770 case AOT_ER_AAC_SCALABLE:
772 ep_config = get_bits(gb, 2);
774 avpriv_report_missing_feature(avctx,
775 "epConfig %d", ep_config);
776 return AVERROR_PATCHWELCOME;
782 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
784 MPEG4AudioConfig *m4ac,
787 int ret, ep_config, res_flags;
788 uint8_t layout_map[MAX_ELEM_ID*4][3];
790 const int ELDEXT_TERM = 0;
795 if (get_bits1(gb)) { // frameLengthFlag
796 avpriv_request_sample(avctx, "960/120 MDCT window");
797 return AVERROR_PATCHWELCOME;
800 res_flags = get_bits(gb, 3);
802 avpriv_report_missing_feature(avctx,
803 "AAC data resilience (flags %x)",
805 return AVERROR_PATCHWELCOME;
808 if (get_bits1(gb)) { // ldSbrPresentFlag
809 avpriv_report_missing_feature(avctx,
811 return AVERROR_PATCHWELCOME;
814 while (get_bits(gb, 4) != ELDEXT_TERM) {
815 int len = get_bits(gb, 4);
817 len += get_bits(gb, 8);
819 len += get_bits(gb, 16);
820 if (get_bits_left(gb) < len * 8 + 4) {
821 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
822 return AVERROR_INVALIDDATA;
824 skip_bits_long(gb, 8 * len);
827 if ((ret = set_default_channel_config(avctx, layout_map,
828 &tags, channel_config)))
831 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
834 ep_config = get_bits(gb, 2);
836 avpriv_report_missing_feature(avctx,
837 "epConfig %d", ep_config);
838 return AVERROR_PATCHWELCOME;
844 * Decode audio specific configuration; reference: table 1.13.
846 * @param ac pointer to AACContext, may be null
847 * @param avctx pointer to AVCCodecContext, used for logging
848 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
849 * @param data pointer to buffer holding an audio specific config
850 * @param bit_size size of audio specific config or data in bits
851 * @param sync_extension look for an appended sync extension
853 * @return Returns error status or number of consumed bits. <0 - error
855 static int decode_audio_specific_config(AACContext *ac,
856 AVCodecContext *avctx,
857 MPEG4AudioConfig *m4ac,
858 const uint8_t *data, int bit_size,
864 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
865 for (i = 0; i < avctx->extradata_size; i++)
866 av_dlog(avctx, "%02x ", avctx->extradata[i]);
867 av_dlog(avctx, "\n");
869 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
872 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
873 sync_extension)) < 0)
874 return AVERROR_INVALIDDATA;
875 if (m4ac->sampling_index > 12) {
876 av_log(avctx, AV_LOG_ERROR,
877 "invalid sampling rate index %d\n",
878 m4ac->sampling_index);
879 return AVERROR_INVALIDDATA;
881 if (m4ac->object_type == AOT_ER_AAC_LD &&
882 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
883 av_log(avctx, AV_LOG_ERROR,
884 "invalid low delay sampling rate index %d\n",
885 m4ac->sampling_index);
886 return AVERROR_INVALIDDATA;
889 skip_bits_long(&gb, i);
891 switch (m4ac->object_type) {
897 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
898 m4ac, m4ac->chan_config)) < 0)
902 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
903 m4ac, m4ac->chan_config)) < 0)
907 avpriv_report_missing_feature(avctx,
908 "Audio object type %s%d",
909 m4ac->sbr == 1 ? "SBR+" : "",
911 return AVERROR(ENOSYS);
915 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
916 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
917 m4ac->sample_rate, m4ac->sbr,
920 return get_bits_count(&gb);
924 * linear congruential pseudorandom number generator
926 * @param previous_val pointer to the current state of the generator
928 * @return Returns a 32-bit pseudorandom integer
930 static av_always_inline int lcg_random(int previous_val)
932 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
936 static av_always_inline void reset_predict_state(PredictorState *ps)
946 static void reset_all_predictors(PredictorState *ps)
949 for (i = 0; i < MAX_PREDICTORS; i++)
950 reset_predict_state(&ps[i]);
953 static int sample_rate_idx (int rate)
955 if (92017 <= rate) return 0;
956 else if (75132 <= rate) return 1;
957 else if (55426 <= rate) return 2;
958 else if (46009 <= rate) return 3;
959 else if (37566 <= rate) return 4;
960 else if (27713 <= rate) return 5;
961 else if (23004 <= rate) return 6;
962 else if (18783 <= rate) return 7;
963 else if (13856 <= rate) return 8;
964 else if (11502 <= rate) return 9;
965 else if (9391 <= rate) return 10;
969 static void reset_predictor_group(PredictorState *ps, int group_num)
972 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
973 reset_predict_state(&ps[i]);
976 #define AAC_INIT_VLC_STATIC(num, size) \
977 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
978 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
979 sizeof(ff_aac_spectral_bits[num][0]), \
980 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
981 sizeof(ff_aac_spectral_codes[num][0]), \
984 static av_cold int aac_decode_init(AVCodecContext *avctx)
986 AACContext *ac = avctx->priv_data;
990 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
992 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
994 if (avctx->extradata_size > 0) {
995 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
997 avctx->extradata_size * 8,
1002 uint8_t layout_map[MAX_ELEM_ID*4][3];
1003 int layout_map_tags;
1005 sr = sample_rate_idx(avctx->sample_rate);
1006 ac->oc[1].m4ac.sampling_index = sr;
1007 ac->oc[1].m4ac.channels = avctx->channels;
1008 ac->oc[1].m4ac.sbr = -1;
1009 ac->oc[1].m4ac.ps = -1;
1011 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1012 if (ff_mpeg4audio_channels[i] == avctx->channels)
1014 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1017 ac->oc[1].m4ac.chan_config = i;
1019 if (ac->oc[1].m4ac.chan_config) {
1020 int ret = set_default_channel_config(avctx, layout_map,
1021 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1023 output_configure(ac, layout_map, layout_map_tags,
1025 else if (avctx->err_recognition & AV_EF_EXPLODE)
1026 return AVERROR_INVALIDDATA;
1030 AAC_INIT_VLC_STATIC( 0, 304);
1031 AAC_INIT_VLC_STATIC( 1, 270);
1032 AAC_INIT_VLC_STATIC( 2, 550);
1033 AAC_INIT_VLC_STATIC( 3, 300);
1034 AAC_INIT_VLC_STATIC( 4, 328);
1035 AAC_INIT_VLC_STATIC( 5, 294);
1036 AAC_INIT_VLC_STATIC( 6, 306);
1037 AAC_INIT_VLC_STATIC( 7, 268);
1038 AAC_INIT_VLC_STATIC( 8, 510);
1039 AAC_INIT_VLC_STATIC( 9, 366);
1040 AAC_INIT_VLC_STATIC(10, 462);
1044 ff_fmt_convert_init(&ac->fmt_conv, avctx);
1045 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1047 ac->random_state = 0x1f2e3d4c;
1051 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1052 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1053 ff_aac_scalefactor_bits,
1054 sizeof(ff_aac_scalefactor_bits[0]),
1055 sizeof(ff_aac_scalefactor_bits[0]),
1056 ff_aac_scalefactor_code,
1057 sizeof(ff_aac_scalefactor_code[0]),
1058 sizeof(ff_aac_scalefactor_code[0]),
1061 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1062 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1063 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1064 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1065 // window initialization
1066 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1067 ff_kbd_window_init(ff_aac_kbd_long_512, 4.0, 512);
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 if (!ics->num_swb || !ics->swb_offset)
1183 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1184 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];
1187 if (aot != AOT_ER_AAC_ELD) {
1188 ics->predictor_present = get_bits1(gb);
1189 ics->predictor_reset_group = 0;
1191 if (ics->predictor_present) {
1192 if (aot == AOT_AAC_MAIN) {
1193 if (decode_prediction(ac, ics, gb)) {
1194 return AVERROR_INVALIDDATA;
1196 } else if (aot == AOT_AAC_LC ||
1197 aot == AOT_ER_AAC_LC) {
1198 av_log(ac->avctx, AV_LOG_ERROR,
1199 "Prediction is not allowed in AAC-LC.\n");
1200 return AVERROR_INVALIDDATA;
1202 if (aot == AOT_ER_AAC_LD) {
1203 av_log(ac->avctx, AV_LOG_ERROR,
1204 "LTP in ER AAC LD not yet implemented.\n");
1205 return AVERROR_PATCHWELCOME;
1207 if ((ics->ltp.present = get_bits(gb, 1)))
1208 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1213 if (ics->max_sfb > ics->num_swb) {
1214 av_log(ac->avctx, AV_LOG_ERROR,
1215 "Number of scalefactor bands in group (%d) "
1216 "exceeds limit (%d).\n",
1217 ics->max_sfb, ics->num_swb);
1218 return AVERROR_INVALIDDATA;
1225 * Decode band types (section_data payload); reference: table 4.46.
1227 * @param band_type array of the used band type
1228 * @param band_type_run_end array of the last scalefactor band of a band type run
1230 * @return Returns error status. 0 - OK, !0 - error
1232 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1233 int band_type_run_end[120], GetBitContext *gb,
1234 IndividualChannelStream *ics)
1237 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1238 for (g = 0; g < ics->num_window_groups; g++) {
1240 while (k < ics->max_sfb) {
1241 uint8_t sect_end = k;
1243 int sect_band_type = get_bits(gb, 4);
1244 if (sect_band_type == 12) {
1245 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1246 return AVERROR_INVALIDDATA;
1249 sect_len_incr = get_bits(gb, bits);
1250 sect_end += sect_len_incr;
1251 if (get_bits_left(gb) < 0) {
1252 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1253 return AVERROR_INVALIDDATA;
1255 if (sect_end > ics->max_sfb) {
1256 av_log(ac->avctx, AV_LOG_ERROR,
1257 "Number of bands (%d) exceeds limit (%d).\n",
1258 sect_end, ics->max_sfb);
1259 return AVERROR_INVALIDDATA;
1261 } while (sect_len_incr == (1 << bits) - 1);
1262 for (; k < sect_end; k++) {
1263 band_type [idx] = sect_band_type;
1264 band_type_run_end[idx++] = sect_end;
1272 * Decode scalefactors; reference: table 4.47.
1274 * @param global_gain first scalefactor value as scalefactors are differentially coded
1275 * @param band_type array of the used band type
1276 * @param band_type_run_end array of the last scalefactor band of a band type run
1277 * @param sf array of scalefactors or intensity stereo positions
1279 * @return Returns error status. 0 - OK, !0 - error
1281 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1282 unsigned int global_gain,
1283 IndividualChannelStream *ics,
1284 enum BandType band_type[120],
1285 int band_type_run_end[120])
1288 int offset[3] = { global_gain, global_gain - 90, 0 };
1291 for (g = 0; g < ics->num_window_groups; g++) {
1292 for (i = 0; i < ics->max_sfb;) {
1293 int run_end = band_type_run_end[idx];
1294 if (band_type[idx] == ZERO_BT) {
1295 for (; i < run_end; i++, idx++)
1297 } else if ((band_type[idx] == INTENSITY_BT) ||
1298 (band_type[idx] == INTENSITY_BT2)) {
1299 for (; i < run_end; i++, idx++) {
1300 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1301 clipped_offset = av_clip(offset[2], -155, 100);
1302 if (offset[2] != clipped_offset) {
1303 avpriv_request_sample(ac->avctx,
1304 "If you heard an audible artifact, there may be a bug in the decoder. "
1305 "Clipped intensity stereo position (%d -> %d)",
1306 offset[2], clipped_offset);
1308 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1310 } else if (band_type[idx] == NOISE_BT) {
1311 for (; i < run_end; i++, idx++) {
1312 if (noise_flag-- > 0)
1313 offset[1] += get_bits(gb, 9) - 256;
1315 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1316 clipped_offset = av_clip(offset[1], -100, 155);
1317 if (offset[1] != clipped_offset) {
1318 avpriv_request_sample(ac->avctx,
1319 "If you heard an audible artifact, there may be a bug in the decoder. "
1320 "Clipped noise gain (%d -> %d)",
1321 offset[1], clipped_offset);
1323 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1326 for (; i < run_end; i++, idx++) {
1327 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1328 if (offset[0] > 255U) {
1329 av_log(ac->avctx, AV_LOG_ERROR,
1330 "Scalefactor (%d) out of range.\n", offset[0]);
1331 return AVERROR_INVALIDDATA;
1333 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1342 * Decode pulse data; reference: table 4.7.
1344 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1345 const uint16_t *swb_offset, int num_swb)
1348 pulse->num_pulse = get_bits(gb, 2) + 1;
1349 pulse_swb = get_bits(gb, 6);
1350 if (pulse_swb >= num_swb)
1352 pulse->pos[0] = swb_offset[pulse_swb];
1353 pulse->pos[0] += get_bits(gb, 5);
1354 if (pulse->pos[0] > 1023)
1356 pulse->amp[0] = get_bits(gb, 4);
1357 for (i = 1; i < pulse->num_pulse; i++) {
1358 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1359 if (pulse->pos[i] > 1023)
1361 pulse->amp[i] = get_bits(gb, 4);
1367 * Decode Temporal Noise Shaping data; reference: table 4.48.
1369 * @return Returns error status. 0 - OK, !0 - error
1371 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1372 GetBitContext *gb, const IndividualChannelStream *ics)
1374 int w, filt, i, coef_len, coef_res, coef_compress;
1375 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1376 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1377 for (w = 0; w < ics->num_windows; w++) {
1378 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1379 coef_res = get_bits1(gb);
1381 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1383 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1385 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1386 av_log(ac->avctx, AV_LOG_ERROR,
1387 "TNS filter order %d is greater than maximum %d.\n",
1388 tns->order[w][filt], tns_max_order);
1389 tns->order[w][filt] = 0;
1390 return AVERROR_INVALIDDATA;
1392 if (tns->order[w][filt]) {
1393 tns->direction[w][filt] = get_bits1(gb);
1394 coef_compress = get_bits1(gb);
1395 coef_len = coef_res + 3 - coef_compress;
1396 tmp2_idx = 2 * coef_compress + coef_res;
1398 for (i = 0; i < tns->order[w][filt]; i++)
1399 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1408 * Decode Mid/Side data; reference: table 4.54.
1410 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1411 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1412 * [3] reserved for scalable AAC
1414 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1418 if (ms_present == 1) {
1420 idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1422 cpe->ms_mask[idx] = get_bits1(gb);
1423 } else if (ms_present == 2) {
1424 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1429 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1433 *dst++ = v[idx & 15] * s;
1434 *dst++ = v[idx>>4 & 15] * s;
1440 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1444 *dst++ = v[idx & 3] * s;
1445 *dst++ = v[idx>>2 & 3] * s;
1446 *dst++ = v[idx>>4 & 3] * s;
1447 *dst++ = v[idx>>6 & 3] * s;
1453 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1454 unsigned sign, const float *scale)
1456 union av_intfloat32 s0, s1;
1458 s0.f = s1.f = *scale;
1459 s0.i ^= sign >> 1 << 31;
1462 *dst++ = v[idx & 15] * s0.f;
1463 *dst++ = v[idx>>4 & 15] * s1.f;
1470 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1471 unsigned sign, const float *scale)
1473 unsigned nz = idx >> 12;
1474 union av_intfloat32 s = { .f = *scale };
1475 union av_intfloat32 t;
1477 t.i = s.i ^ (sign & 1U<<31);
1478 *dst++ = v[idx & 3] * t.f;
1480 sign <<= nz & 1; nz >>= 1;
1481 t.i = s.i ^ (sign & 1U<<31);
1482 *dst++ = v[idx>>2 & 3] * t.f;
1484 sign <<= nz & 1; nz >>= 1;
1485 t.i = s.i ^ (sign & 1U<<31);
1486 *dst++ = v[idx>>4 & 3] * t.f;
1489 t.i = s.i ^ (sign & 1U<<31);
1490 *dst++ = v[idx>>6 & 3] * t.f;
1497 * Decode spectral data; reference: table 4.50.
1498 * Dequantize and scale spectral data; reference: 4.6.3.3.
1500 * @param coef array of dequantized, scaled spectral data
1501 * @param sf array of scalefactors or intensity stereo positions
1502 * @param pulse_present set if pulses are present
1503 * @param pulse pointer to pulse data struct
1504 * @param band_type array of the used band type
1506 * @return Returns error status. 0 - OK, !0 - error
1508 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1509 GetBitContext *gb, const float sf[120],
1510 int pulse_present, const Pulse *pulse,
1511 const IndividualChannelStream *ics,
1512 enum BandType band_type[120])
1514 int i, k, g, idx = 0;
1515 const int c = 1024 / ics->num_windows;
1516 const uint16_t *offsets = ics->swb_offset;
1517 float *coef_base = coef;
1519 for (g = 0; g < ics->num_windows; g++)
1520 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1521 sizeof(float) * (c - offsets[ics->max_sfb]));
1523 for (g = 0; g < ics->num_window_groups; g++) {
1524 unsigned g_len = ics->group_len[g];
1526 for (i = 0; i < ics->max_sfb; i++, idx++) {
1527 const unsigned cbt_m1 = band_type[idx] - 1;
1528 float *cfo = coef + offsets[i];
1529 int off_len = offsets[i + 1] - offsets[i];
1532 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1533 for (group = 0; group < g_len; group++, cfo+=128) {
1534 memset(cfo, 0, off_len * sizeof(float));
1536 } else if (cbt_m1 == NOISE_BT - 1) {
1537 for (group = 0; group < g_len; group++, cfo+=128) {
1541 for (k = 0; k < off_len; k++) {
1542 ac->random_state = lcg_random(ac->random_state);
1543 cfo[k] = ac->random_state;
1546 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1547 scale = sf[idx] / sqrtf(band_energy);
1548 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1551 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1552 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1553 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1554 OPEN_READER(re, gb);
1556 switch (cbt_m1 >> 1) {
1558 for (group = 0; group < g_len; group++, cfo+=128) {
1566 UPDATE_CACHE(re, gb);
1567 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1568 cb_idx = cb_vector_idx[code];
1569 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1575 for (group = 0; group < g_len; group++, cfo+=128) {
1585 UPDATE_CACHE(re, gb);
1586 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1587 cb_idx = cb_vector_idx[code];
1588 nnz = cb_idx >> 8 & 15;
1589 bits = nnz ? GET_CACHE(re, gb) : 0;
1590 LAST_SKIP_BITS(re, gb, nnz);
1591 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1597 for (group = 0; group < g_len; group++, cfo+=128) {
1605 UPDATE_CACHE(re, gb);
1606 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1607 cb_idx = cb_vector_idx[code];
1608 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1615 for (group = 0; group < g_len; group++, cfo+=128) {
1625 UPDATE_CACHE(re, gb);
1626 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1627 cb_idx = cb_vector_idx[code];
1628 nnz = cb_idx >> 8 & 15;
1629 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1630 LAST_SKIP_BITS(re, gb, nnz);
1631 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1637 for (group = 0; group < g_len; group++, cfo+=128) {
1639 uint32_t *icf = (uint32_t *) cf;
1649 UPDATE_CACHE(re, gb);
1650 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1658 cb_idx = cb_vector_idx[code];
1661 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1662 LAST_SKIP_BITS(re, gb, nnz);
1664 for (j = 0; j < 2; j++) {
1668 /* The total length of escape_sequence must be < 22 bits according
1669 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1670 UPDATE_CACHE(re, gb);
1671 b = GET_CACHE(re, gb);
1672 b = 31 - av_log2(~b);
1675 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1676 return AVERROR_INVALIDDATA;
1679 SKIP_BITS(re, gb, b + 1);
1681 n = (1 << b) + SHOW_UBITS(re, gb, b);
1682 LAST_SKIP_BITS(re, gb, b);
1683 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1686 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1687 *icf++ = (bits & 1U<<31) | v;
1694 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1698 CLOSE_READER(re, gb);
1704 if (pulse_present) {
1706 for (i = 0; i < pulse->num_pulse; i++) {
1707 float co = coef_base[ pulse->pos[i] ];
1708 while (offsets[idx + 1] <= pulse->pos[i])
1710 if (band_type[idx] != NOISE_BT && sf[idx]) {
1711 float ico = -pulse->amp[i];
1714 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1716 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1723 static av_always_inline float flt16_round(float pf)
1725 union av_intfloat32 tmp;
1727 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1731 static av_always_inline float flt16_even(float pf)
1733 union av_intfloat32 tmp;
1735 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1739 static av_always_inline float flt16_trunc(float pf)
1741 union av_intfloat32 pun;
1743 pun.i &= 0xFFFF0000U;
1747 static av_always_inline void predict(PredictorState *ps, float *coef,
1750 const float a = 0.953125; // 61.0 / 64
1751 const float alpha = 0.90625; // 29.0 / 32
1755 float r0 = ps->r0, r1 = ps->r1;
1756 float cor0 = ps->cor0, cor1 = ps->cor1;
1757 float var0 = ps->var0, var1 = ps->var1;
1759 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1760 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1762 pv = flt16_round(k1 * r0 + k2 * r1);
1769 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1770 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1771 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1772 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1774 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1775 ps->r0 = flt16_trunc(a * e0);
1779 * Apply AAC-Main style frequency domain prediction.
1781 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1785 if (!sce->ics.predictor_initialized) {
1786 reset_all_predictors(sce->predictor_state);
1787 sce->ics.predictor_initialized = 1;
1790 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1792 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1794 for (k = sce->ics.swb_offset[sfb];
1795 k < sce->ics.swb_offset[sfb + 1];
1797 predict(&sce->predictor_state[k], &sce->coeffs[k],
1798 sce->ics.predictor_present &&
1799 sce->ics.prediction_used[sfb]);
1802 if (sce->ics.predictor_reset_group)
1803 reset_predictor_group(sce->predictor_state,
1804 sce->ics.predictor_reset_group);
1806 reset_all_predictors(sce->predictor_state);
1810 * Decode an individual_channel_stream payload; reference: table 4.44.
1812 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1813 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1815 * @return Returns error status. 0 - OK, !0 - error
1817 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1818 GetBitContext *gb, int common_window, int scale_flag)
1821 TemporalNoiseShaping *tns = &sce->tns;
1822 IndividualChannelStream *ics = &sce->ics;
1823 float *out = sce->coeffs;
1824 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1827 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1828 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1829 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1830 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1831 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1833 /* This assignment is to silence a GCC warning about the variable being used
1834 * uninitialized when in fact it always is.
1836 pulse.num_pulse = 0;
1838 global_gain = get_bits(gb, 8);
1840 if (!common_window && !scale_flag) {
1841 if (decode_ics_info(ac, ics, gb) < 0)
1842 return AVERROR_INVALIDDATA;
1845 if ((ret = decode_band_types(ac, sce->band_type,
1846 sce->band_type_run_end, gb, ics)) < 0)
1848 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1849 sce->band_type, sce->band_type_run_end)) < 0)
1854 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1855 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1856 av_log(ac->avctx, AV_LOG_ERROR,
1857 "Pulse tool not allowed in eight short sequence.\n");
1858 return AVERROR_INVALIDDATA;
1860 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1861 av_log(ac->avctx, AV_LOG_ERROR,
1862 "Pulse data corrupt or invalid.\n");
1863 return AVERROR_INVALIDDATA;
1866 tns->present = get_bits1(gb);
1867 if (tns->present && !er_syntax)
1868 if (decode_tns(ac, tns, gb, ics) < 0)
1869 return AVERROR_INVALIDDATA;
1870 if (!eld_syntax && get_bits1(gb)) {
1871 avpriv_request_sample(ac->avctx, "SSR");
1872 return AVERROR_PATCHWELCOME;
1874 // I see no textual basis in the spec for this occuring after SSR gain
1875 // control, but this is what both reference and real implmentations do
1876 if (tns->present && er_syntax)
1877 if (decode_tns(ac, tns, gb, ics) < 0)
1878 return AVERROR_INVALIDDATA;
1881 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1882 &pulse, ics, sce->band_type) < 0)
1883 return AVERROR_INVALIDDATA;
1885 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1886 apply_prediction(ac, sce);
1892 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1894 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1896 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1897 float *ch0 = cpe->ch[0].coeffs;
1898 float *ch1 = cpe->ch[1].coeffs;
1899 int g, i, group, idx = 0;
1900 const uint16_t *offsets = ics->swb_offset;
1901 for (g = 0; g < ics->num_window_groups; g++) {
1902 for (i = 0; i < ics->max_sfb; i++, idx++) {
1903 if (cpe->ms_mask[idx] &&
1904 cpe->ch[0].band_type[idx] < NOISE_BT &&
1905 cpe->ch[1].band_type[idx] < NOISE_BT) {
1906 for (group = 0; group < ics->group_len[g]; group++) {
1907 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
1908 ch1 + group * 128 + offsets[i],
1909 offsets[i+1] - offsets[i]);
1913 ch0 += ics->group_len[g] * 128;
1914 ch1 += ics->group_len[g] * 128;
1919 * intensity stereo decoding; reference: 4.6.8.2.3
1921 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1922 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1923 * [3] reserved for scalable AAC
1925 static void apply_intensity_stereo(AACContext *ac,
1926 ChannelElement *cpe, int ms_present)
1928 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1929 SingleChannelElement *sce1 = &cpe->ch[1];
1930 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1931 const uint16_t *offsets = ics->swb_offset;
1932 int g, group, i, idx = 0;
1935 for (g = 0; g < ics->num_window_groups; g++) {
1936 for (i = 0; i < ics->max_sfb;) {
1937 if (sce1->band_type[idx] == INTENSITY_BT ||
1938 sce1->band_type[idx] == INTENSITY_BT2) {
1939 const int bt_run_end = sce1->band_type_run_end[idx];
1940 for (; i < bt_run_end; i++, idx++) {
1941 c = -1 + 2 * (sce1->band_type[idx] - 14);
1943 c *= 1 - 2 * cpe->ms_mask[idx];
1944 scale = c * sce1->sf[idx];
1945 for (group = 0; group < ics->group_len[g]; group++)
1946 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1947 coef0 + group * 128 + offsets[i],
1949 offsets[i + 1] - offsets[i]);
1952 int bt_run_end = sce1->band_type_run_end[idx];
1953 idx += bt_run_end - i;
1957 coef0 += ics->group_len[g] * 128;
1958 coef1 += ics->group_len[g] * 128;
1963 * Decode a channel_pair_element; reference: table 4.4.
1965 * @return Returns error status. 0 - OK, !0 - error
1967 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1969 int i, ret, common_window, ms_present = 0;
1970 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1972 common_window = eld_syntax || get_bits1(gb);
1973 if (common_window) {
1974 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1975 return AVERROR_INVALIDDATA;
1976 i = cpe->ch[1].ics.use_kb_window[0];
1977 cpe->ch[1].ics = cpe->ch[0].ics;
1978 cpe->ch[1].ics.use_kb_window[1] = i;
1979 if (cpe->ch[1].ics.predictor_present &&
1980 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
1981 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1982 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1983 ms_present = get_bits(gb, 2);
1984 if (ms_present == 3) {
1985 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1986 return AVERROR_INVALIDDATA;
1987 } else if (ms_present)
1988 decode_mid_side_stereo(cpe, gb, ms_present);
1990 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1992 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1995 if (common_window) {
1997 apply_mid_side_stereo(ac, cpe);
1998 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
1999 apply_prediction(ac, &cpe->ch[0]);
2000 apply_prediction(ac, &cpe->ch[1]);
2004 apply_intensity_stereo(ac, cpe, ms_present);
2008 static const float cce_scale[] = {
2009 1.09050773266525765921, //2^(1/8)
2010 1.18920711500272106672, //2^(1/4)
2016 * Decode coupling_channel_element; reference: table 4.8.
2018 * @return Returns error status. 0 - OK, !0 - error
2020 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2026 SingleChannelElement *sce = &che->ch[0];
2027 ChannelCoupling *coup = &che->coup;
2029 coup->coupling_point = 2 * get_bits1(gb);
2030 coup->num_coupled = get_bits(gb, 3);
2031 for (c = 0; c <= coup->num_coupled; c++) {
2033 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2034 coup->id_select[c] = get_bits(gb, 4);
2035 if (coup->type[c] == TYPE_CPE) {
2036 coup->ch_select[c] = get_bits(gb, 2);
2037 if (coup->ch_select[c] == 3)
2040 coup->ch_select[c] = 2;
2042 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2044 sign = get_bits(gb, 1);
2045 scale = cce_scale[get_bits(gb, 2)];
2047 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2050 for (c = 0; c < num_gain; c++) {
2054 float gain_cache = 1.0;
2056 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2057 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2058 gain_cache = powf(scale, -gain);
2060 if (coup->coupling_point == AFTER_IMDCT) {
2061 coup->gain[c][0] = gain_cache;
2063 for (g = 0; g < sce->ics.num_window_groups; g++) {
2064 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2065 if (sce->band_type[idx] != ZERO_BT) {
2067 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2075 gain_cache = powf(scale, -t) * s;
2078 coup->gain[c][idx] = gain_cache;
2088 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2090 * @return Returns number of bytes consumed.
2092 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2096 int num_excl_chan = 0;
2099 for (i = 0; i < 7; i++)
2100 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2101 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2103 return num_excl_chan / 7;
2107 * Decode dynamic range information; reference: table 4.52.
2109 * @return Returns number of bytes consumed.
2111 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2115 int drc_num_bands = 1;
2118 /* pce_tag_present? */
2119 if (get_bits1(gb)) {
2120 che_drc->pce_instance_tag = get_bits(gb, 4);
2121 skip_bits(gb, 4); // tag_reserved_bits
2125 /* excluded_chns_present? */
2126 if (get_bits1(gb)) {
2127 n += decode_drc_channel_exclusions(che_drc, gb);
2130 /* drc_bands_present? */
2131 if (get_bits1(gb)) {
2132 che_drc->band_incr = get_bits(gb, 4);
2133 che_drc->interpolation_scheme = get_bits(gb, 4);
2135 drc_num_bands += che_drc->band_incr;
2136 for (i = 0; i < drc_num_bands; i++) {
2137 che_drc->band_top[i] = get_bits(gb, 8);
2142 /* prog_ref_level_present? */
2143 if (get_bits1(gb)) {
2144 che_drc->prog_ref_level = get_bits(gb, 7);
2145 skip_bits1(gb); // prog_ref_level_reserved_bits
2149 for (i = 0; i < drc_num_bands; i++) {
2150 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2151 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2159 * Decode extension data (incomplete); reference: table 4.51.
2161 * @param cnt length of TYPE_FIL syntactic element in bytes
2163 * @return Returns number of bytes consumed
2165 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2166 ChannelElement *che, enum RawDataBlockType elem_type)
2170 switch (get_bits(gb, 4)) { // extension type
2171 case EXT_SBR_DATA_CRC:
2175 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2177 } else if (!ac->oc[1].m4ac.sbr) {
2178 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2179 skip_bits_long(gb, 8 * cnt - 4);
2181 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2182 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2183 skip_bits_long(gb, 8 * cnt - 4);
2185 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2186 ac->oc[1].m4ac.sbr = 1;
2187 ac->oc[1].m4ac.ps = 1;
2188 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2189 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2190 ac->oc[1].status, 1);
2192 ac->oc[1].m4ac.sbr = 1;
2193 ac->avctx->profile = FF_PROFILE_AAC_HE;
2195 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2197 case EXT_DYNAMIC_RANGE:
2198 res = decode_dynamic_range(&ac->che_drc, gb);
2202 case EXT_DATA_ELEMENT:
2204 skip_bits_long(gb, 8 * cnt - 4);
2211 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2213 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2214 * @param coef spectral coefficients
2216 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2217 IndividualChannelStream *ics, int decode)
2219 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2221 int bottom, top, order, start, end, size, inc;
2222 float lpc[TNS_MAX_ORDER];
2223 float tmp[TNS_MAX_ORDER + 1];
2225 for (w = 0; w < ics->num_windows; w++) {
2226 bottom = ics->num_swb;
2227 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2229 bottom = FFMAX(0, top - tns->length[w][filt]);
2230 order = tns->order[w][filt];
2235 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2237 start = ics->swb_offset[FFMIN(bottom, mmm)];
2238 end = ics->swb_offset[FFMIN( top, mmm)];
2239 if ((size = end - start) <= 0)
2241 if (tns->direction[w][filt]) {
2251 for (m = 0; m < size; m++, start += inc)
2252 for (i = 1; i <= FFMIN(m, order); i++)
2253 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2256 for (m = 0; m < size; m++, start += inc) {
2257 tmp[0] = coef[start];
2258 for (i = 1; i <= FFMIN(m, order); i++)
2259 coef[start] += tmp[i] * lpc[i - 1];
2260 for (i = order; i > 0; i--)
2261 tmp[i] = tmp[i - 1];
2269 * Apply windowing and MDCT to obtain the spectral
2270 * coefficient from the predicted sample by LTP.
2272 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2273 float *in, IndividualChannelStream *ics)
2275 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2276 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2277 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2278 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2280 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2281 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2283 memset(in, 0, 448 * sizeof(float));
2284 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2286 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2287 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2289 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2290 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2292 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2296 * Apply the long term prediction
2298 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2300 const LongTermPrediction *ltp = &sce->ics.ltp;
2301 const uint16_t *offsets = sce->ics.swb_offset;
2304 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2305 float *predTime = sce->ret;
2306 float *predFreq = ac->buf_mdct;
2307 int16_t num_samples = 2048;
2309 if (ltp->lag < 1024)
2310 num_samples = ltp->lag + 1024;
2311 for (i = 0; i < num_samples; i++)
2312 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2313 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2315 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2317 if (sce->tns.present)
2318 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2320 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2322 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2323 sce->coeffs[i] += predFreq[i];
2328 * Update the LTP buffer for next frame
2330 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2332 IndividualChannelStream *ics = &sce->ics;
2333 float *saved = sce->saved;
2334 float *saved_ltp = sce->coeffs;
2335 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2336 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2339 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2340 memcpy(saved_ltp, saved, 512 * sizeof(float));
2341 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2342 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2343 for (i = 0; i < 64; i++)
2344 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2345 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2346 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2347 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2348 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2349 for (i = 0; i < 64; i++)
2350 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2351 } else { // LONG_STOP or ONLY_LONG
2352 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2353 for (i = 0; i < 512; i++)
2354 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2357 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2358 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2359 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2363 * Conduct IMDCT and windowing.
2365 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2367 IndividualChannelStream *ics = &sce->ics;
2368 float *in = sce->coeffs;
2369 float *out = sce->ret;
2370 float *saved = sce->saved;
2371 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2372 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2373 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2374 float *buf = ac->buf_mdct;
2375 float *temp = ac->temp;
2379 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2380 for (i = 0; i < 1024; i += 128)
2381 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2383 ac->mdct.imdct_half(&ac->mdct, buf, in);
2385 /* window overlapping
2386 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2387 * and long to short transitions are considered to be short to short
2388 * transitions. This leaves just two cases (long to long and short to short)
2389 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2391 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2392 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2393 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2395 memcpy( out, saved, 448 * sizeof(float));
2397 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2398 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2399 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2400 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2401 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2402 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2403 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2405 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2406 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2411 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2412 memcpy( saved, temp + 64, 64 * sizeof(float));
2413 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2414 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2415 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2416 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2417 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2418 memcpy( saved, buf + 512, 448 * sizeof(float));
2419 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2420 } else { // LONG_STOP or ONLY_LONG
2421 memcpy( saved, buf + 512, 512 * sizeof(float));
2425 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2427 IndividualChannelStream *ics = &sce->ics;
2428 float *in = sce->coeffs;
2429 float *out = sce->ret;
2430 float *saved = sce->saved;
2431 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_512 : ff_sine_512;
2432 float *buf = ac->buf_mdct;
2435 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2437 // window overlapping
2438 ac->fdsp.vector_fmul_window(out, saved, buf, lwindow_prev, 256);
2441 memcpy(saved, buf + 256, 256 * sizeof(float));
2444 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2446 float *in = sce->coeffs;
2447 float *out = sce->ret;
2448 float *saved = sce->saved;
2449 const float *const window = ff_aac_eld_window;
2450 float *buf = ac->buf_mdct;
2453 const int n2 = n >> 1;
2454 const int n4 = n >> 2;
2456 // Inverse transform, mapped to the conventional IMDCT by
2457 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2458 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2459 // Audio, Language and Image Processing, 2008. ICALIP 2008. International Conference on
2460 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2461 for (i = 0; i < n2; i+=2) {
2463 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2464 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2466 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2467 for (i = 0; i < n; i+=2) {
2470 // Like with the regular IMDCT at this point we still have the middle half
2471 // of a transform but with even symmetry on the left and odd symmetry on
2474 // window overlapping
2475 // The spec says to use samples [0..511] but the reference decoder uses
2476 // samples [128..639].
2477 for (i = n4; i < n2; i ++) {
2478 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2479 saved[ i + n2] * window[i + n - n4] +
2480 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2481 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2483 for (i = 0; i < n2; i ++) {
2484 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2485 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2486 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2487 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2489 for (i = 0; i < n4; i ++) {
2490 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2491 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2492 -saved[ n + n2 + i] * window[i + 3*n - n4];
2496 memmove(saved + n, saved, 2 * n * sizeof(float));
2497 memcpy( saved, buf, n * sizeof(float));
2501 * Apply dependent channel coupling (applied before IMDCT).
2503 * @param index index into coupling gain array
2505 static void apply_dependent_coupling(AACContext *ac,
2506 SingleChannelElement *target,
2507 ChannelElement *cce, int index)
2509 IndividualChannelStream *ics = &cce->ch[0].ics;
2510 const uint16_t *offsets = ics->swb_offset;
2511 float *dest = target->coeffs;
2512 const float *src = cce->ch[0].coeffs;
2513 int g, i, group, k, idx = 0;
2514 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2515 av_log(ac->avctx, AV_LOG_ERROR,
2516 "Dependent coupling is not supported together with LTP\n");
2519 for (g = 0; g < ics->num_window_groups; g++) {
2520 for (i = 0; i < ics->max_sfb; i++, idx++) {
2521 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2522 const float gain = cce->coup.gain[index][idx];
2523 for (group = 0; group < ics->group_len[g]; group++) {
2524 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2526 dest[group * 128 + k] += gain * src[group * 128 + k];
2531 dest += ics->group_len[g] * 128;
2532 src += ics->group_len[g] * 128;
2537 * Apply independent channel coupling (applied after IMDCT).
2539 * @param index index into coupling gain array
2541 static void apply_independent_coupling(AACContext *ac,
2542 SingleChannelElement *target,
2543 ChannelElement *cce, int index)
2546 const float gain = cce->coup.gain[index][0];
2547 const float *src = cce->ch[0].ret;
2548 float *dest = target->ret;
2549 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2551 for (i = 0; i < len; i++)
2552 dest[i] += gain * src[i];
2556 * channel coupling transformation interface
2558 * @param apply_coupling_method pointer to (in)dependent coupling function
2560 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2561 enum RawDataBlockType type, int elem_id,
2562 enum CouplingPoint coupling_point,
2563 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2567 for (i = 0; i < MAX_ELEM_ID; i++) {
2568 ChannelElement *cce = ac->che[TYPE_CCE][i];
2571 if (cce && cce->coup.coupling_point == coupling_point) {
2572 ChannelCoupling *coup = &cce->coup;
2574 for (c = 0; c <= coup->num_coupled; c++) {
2575 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2576 if (coup->ch_select[c] != 1) {
2577 apply_coupling_method(ac, &cc->ch[0], cce, index);
2578 if (coup->ch_select[c] != 0)
2581 if (coup->ch_select[c] != 2)
2582 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2584 index += 1 + (coup->ch_select[c] == 3);
2591 * Convert spectral data to float samples, applying all supported tools as appropriate.
2593 static void spectral_to_sample(AACContext *ac)
2596 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2597 switch (ac->oc[1].m4ac.object_type) {
2599 imdct_and_window = imdct_and_windowing_ld;
2601 case AOT_ER_AAC_ELD:
2602 imdct_and_window = imdct_and_windowing_eld;
2605 imdct_and_window = imdct_and_windowing;
2607 for (type = 3; type >= 0; type--) {
2608 for (i = 0; i < MAX_ELEM_ID; i++) {
2609 ChannelElement *che = ac->che[type][i];
2611 if (type <= TYPE_CPE)
2612 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2613 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2614 if (che->ch[0].ics.predictor_present) {
2615 if (che->ch[0].ics.ltp.present)
2616 apply_ltp(ac, &che->ch[0]);
2617 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2618 apply_ltp(ac, &che->ch[1]);
2621 if (che->ch[0].tns.present)
2622 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2623 if (che->ch[1].tns.present)
2624 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2625 if (type <= TYPE_CPE)
2626 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2627 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2628 imdct_and_window(ac, &che->ch[0]);
2629 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2630 update_ltp(ac, &che->ch[0]);
2631 if (type == TYPE_CPE) {
2632 imdct_and_window(ac, &che->ch[1]);
2633 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2634 update_ltp(ac, &che->ch[1]);
2636 if (ac->oc[1].m4ac.sbr > 0) {
2637 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2640 if (type <= TYPE_CCE)
2641 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2647 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2650 AACADTSHeaderInfo hdr_info;
2651 uint8_t layout_map[MAX_ELEM_ID*4][3];
2652 int layout_map_tags, ret;
2654 size = avpriv_aac_parse_header(gb, &hdr_info);
2656 if (hdr_info.num_aac_frames != 1) {
2657 avpriv_report_missing_feature(ac->avctx,
2658 "More than one AAC RDB per ADTS frame");
2659 return AVERROR_PATCHWELCOME;
2661 push_output_configuration(ac);
2662 if (hdr_info.chan_config) {
2663 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2664 if ((ret = set_default_channel_config(ac->avctx,
2667 hdr_info.chan_config)) < 0)
2669 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2670 FFMAX(ac->oc[1].status,
2671 OC_TRIAL_FRAME), 0)) < 0)
2674 ac->oc[1].m4ac.chan_config = 0;
2676 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2677 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2678 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2679 if (ac->oc[0].status != OC_LOCKED ||
2680 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2681 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2682 ac->oc[1].m4ac.sbr = -1;
2683 ac->oc[1].m4ac.ps = -1;
2685 if (!hdr_info.crc_absent)
2691 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2692 int *got_frame_ptr, GetBitContext *gb)
2694 AACContext *ac = avctx->priv_data;
2695 ChannelElement *che;
2698 int chan_config = ac->oc[1].m4ac.chan_config;
2699 int aot = ac->oc[1].m4ac.object_type;
2701 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2706 if ((err = frame_configure_elements(avctx)) < 0)
2709 // The FF_PROFILE_AAC_* defines are all object_type - 1
2710 // This may lead to an undefined profile being signaled
2711 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2713 ac->tags_mapped = 0;
2715 if (chan_config < 0 || chan_config >= 8) {
2716 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2717 ac->oc[1].m4ac.chan_config);
2718 return AVERROR_INVALIDDATA;
2720 for (i = 0; i < tags_per_config[chan_config]; i++) {
2721 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2722 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2723 if (!(che=get_che(ac, elem_type, elem_id))) {
2724 av_log(ac->avctx, AV_LOG_ERROR,
2725 "channel element %d.%d is not allocated\n",
2726 elem_type, elem_id);
2727 return AVERROR_INVALIDDATA;
2729 if (aot != AOT_ER_AAC_ELD)
2731 switch (elem_type) {
2733 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2736 err = decode_cpe(ac, gb, che);
2739 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2746 spectral_to_sample(ac);
2748 ac->frame->nb_samples = samples;
2751 skip_bits_long(gb, get_bits_left(gb));
2755 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2756 int *got_frame_ptr, GetBitContext *gb)
2758 AACContext *ac = avctx->priv_data;
2759 ChannelElement *che = NULL, *che_prev = NULL;
2760 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2762 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2766 if (show_bits(gb, 12) == 0xfff) {
2767 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2768 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2771 if (ac->oc[1].m4ac.sampling_index > 12) {
2772 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2773 err = AVERROR_INVALIDDATA;
2778 if ((err = frame_configure_elements(avctx)) < 0)
2781 // The FF_PROFILE_AAC_* defines are all object_type - 1
2782 // This may lead to an undefined profile being signaled
2783 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2785 ac->tags_mapped = 0;
2787 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2788 elem_id = get_bits(gb, 4);
2790 if (elem_type < TYPE_DSE) {
2791 if (!(che=get_che(ac, elem_type, elem_id))) {
2792 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2793 elem_type, elem_id);
2794 err = AVERROR_INVALIDDATA;
2800 switch (elem_type) {
2803 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2808 err = decode_cpe(ac, gb, che);
2813 err = decode_cce(ac, gb, che);
2817 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2822 err = skip_data_stream_element(ac, gb);
2826 uint8_t layout_map[MAX_ELEM_ID*4][3];
2828 push_output_configuration(ac);
2829 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2835 av_log(avctx, AV_LOG_ERROR,
2836 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2837 pop_output_configuration(ac);
2839 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2847 elem_id += get_bits(gb, 8) - 1;
2848 if (get_bits_left(gb) < 8 * elem_id) {
2849 av_log(avctx, AV_LOG_ERROR, overread_err);
2850 err = AVERROR_INVALIDDATA;
2854 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2855 err = 0; /* FIXME */
2859 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
2864 elem_type_prev = elem_type;
2869 if (get_bits_left(gb) < 3) {
2870 av_log(avctx, AV_LOG_ERROR, overread_err);
2871 err = AVERROR_INVALIDDATA;
2876 spectral_to_sample(ac);
2878 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
2879 samples <<= multiplier;
2882 ac->frame->nb_samples = samples;
2883 *got_frame_ptr = !!samples;
2885 if (ac->oc[1].status && audio_found) {
2886 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
2887 avctx->frame_size = samples;
2888 ac->oc[1].status = OC_LOCKED;
2893 pop_output_configuration(ac);
2897 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2898 int *got_frame_ptr, AVPacket *avpkt)
2900 AACContext *ac = avctx->priv_data;
2901 const uint8_t *buf = avpkt->data;
2902 int buf_size = avpkt->size;
2907 int new_extradata_size;
2908 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2909 AV_PKT_DATA_NEW_EXTRADATA,
2910 &new_extradata_size);
2912 if (new_extradata) {
2913 av_free(avctx->extradata);
2914 avctx->extradata = av_mallocz(new_extradata_size +
2915 FF_INPUT_BUFFER_PADDING_SIZE);
2916 if (!avctx->extradata)
2917 return AVERROR(ENOMEM);
2918 avctx->extradata_size = new_extradata_size;
2919 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2920 push_output_configuration(ac);
2921 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
2923 avctx->extradata_size*8, 1) < 0) {
2924 pop_output_configuration(ac);
2925 return AVERROR_INVALIDDATA;
2929 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
2932 switch (ac->oc[1].m4ac.object_type) {
2934 case AOT_ER_AAC_LTP:
2936 case AOT_ER_AAC_ELD:
2937 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
2940 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb);
2945 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2946 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2947 if (buf[buf_offset])
2950 return buf_size > buf_offset ? buf_consumed : buf_size;
2953 static av_cold int aac_decode_close(AVCodecContext *avctx)
2955 AACContext *ac = avctx->priv_data;
2958 for (i = 0; i < MAX_ELEM_ID; i++) {
2959 for (type = 0; type < 4; type++) {
2960 if (ac->che[type][i])
2961 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2962 av_freep(&ac->che[type][i]);
2966 ff_mdct_end(&ac->mdct);
2967 ff_mdct_end(&ac->mdct_small);
2968 ff_mdct_end(&ac->mdct_ld);
2969 ff_mdct_end(&ac->mdct_ltp);
2974 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2976 struct LATMContext {
2977 AACContext aac_ctx; ///< containing AACContext
2978 int initialized; ///< initilized after a valid extradata was seen
2981 int audio_mux_version_A; ///< LATM syntax version
2982 int frame_length_type; ///< 0/1 variable/fixed frame length
2983 int frame_length; ///< frame length for fixed frame length
2986 static inline uint32_t latm_get_value(GetBitContext *b)
2988 int length = get_bits(b, 2);
2990 return get_bits_long(b, (length+1)*8);
2993 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2994 GetBitContext *gb, int asclen)
2996 AACContext *ac = &latmctx->aac_ctx;
2997 AVCodecContext *avctx = ac->avctx;
2998 MPEG4AudioConfig m4ac = { 0 };
2999 int config_start_bit = get_bits_count(gb);
3000 int sync_extension = 0;
3001 int bits_consumed, esize;
3005 asclen = FFMIN(asclen, get_bits_left(gb));
3007 asclen = get_bits_left(gb);
3009 if (config_start_bit % 8) {
3010 avpriv_request_sample(latmctx->aac_ctx.avctx,
3011 "Non-byte-aligned audio-specific config");
3012 return AVERROR_PATCHWELCOME;
3015 return AVERROR_INVALIDDATA;
3016 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3017 gb->buffer + (config_start_bit / 8),
3018 asclen, sync_extension);
3020 if (bits_consumed < 0)
3021 return AVERROR_INVALIDDATA;
3023 if (ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3024 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3026 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3027 latmctx->initialized = 0;
3029 esize = (bits_consumed+7) / 8;
3031 if (avctx->extradata_size < esize) {
3032 av_free(avctx->extradata);
3033 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3034 if (!avctx->extradata)
3035 return AVERROR(ENOMEM);
3038 avctx->extradata_size = esize;
3039 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3040 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3042 skip_bits_long(gb, bits_consumed);
3044 return bits_consumed;
3047 static int read_stream_mux_config(struct LATMContext *latmctx,
3050 int ret, audio_mux_version = get_bits(gb, 1);
3052 latmctx->audio_mux_version_A = 0;
3053 if (audio_mux_version)
3054 latmctx->audio_mux_version_A = get_bits(gb, 1);
3056 if (!latmctx->audio_mux_version_A) {
3058 if (audio_mux_version)
3059 latm_get_value(gb); // taraFullness
3061 skip_bits(gb, 1); // allStreamSameTimeFraming
3062 skip_bits(gb, 6); // numSubFrames
3064 if (get_bits(gb, 4)) { // numPrograms
3065 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3066 return AVERROR_PATCHWELCOME;
3069 // for each program (which there is only on in DVB)
3071 // for each layer (which there is only on in DVB)
3072 if (get_bits(gb, 3)) { // numLayer
3073 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3074 return AVERROR_PATCHWELCOME;
3077 // for all but first stream: use_same_config = get_bits(gb, 1);
3078 if (!audio_mux_version) {
3079 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3082 int ascLen = latm_get_value(gb);
3083 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3086 skip_bits_long(gb, ascLen);
3089 latmctx->frame_length_type = get_bits(gb, 3);
3090 switch (latmctx->frame_length_type) {
3092 skip_bits(gb, 8); // latmBufferFullness
3095 latmctx->frame_length = get_bits(gb, 9);
3100 skip_bits(gb, 6); // CELP frame length table index
3104 skip_bits(gb, 1); // HVXC frame length table index
3108 if (get_bits(gb, 1)) { // other data
3109 if (audio_mux_version) {
3110 latm_get_value(gb); // other_data_bits
3114 esc = get_bits(gb, 1);
3120 if (get_bits(gb, 1)) // crc present
3121 skip_bits(gb, 8); // config_crc
3127 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3131 if (ctx->frame_length_type == 0) {
3132 int mux_slot_length = 0;
3134 tmp = get_bits(gb, 8);
3135 mux_slot_length += tmp;
3136 } while (tmp == 255);
3137 return mux_slot_length;
3138 } else if (ctx->frame_length_type == 1) {
3139 return ctx->frame_length;
3140 } else if (ctx->frame_length_type == 3 ||
3141 ctx->frame_length_type == 5 ||
3142 ctx->frame_length_type == 7) {
3143 skip_bits(gb, 2); // mux_slot_length_coded
3148 static int read_audio_mux_element(struct LATMContext *latmctx,
3152 uint8_t use_same_mux = get_bits(gb, 1);
3153 if (!use_same_mux) {
3154 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3156 } else if (!latmctx->aac_ctx.avctx->extradata) {
3157 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3158 "no decoder config found\n");
3159 return AVERROR(EAGAIN);
3161 if (latmctx->audio_mux_version_A == 0) {
3162 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3163 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3164 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3165 return AVERROR_INVALIDDATA;
3166 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3167 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3168 "frame length mismatch %d << %d\n",
3169 mux_slot_length_bytes * 8, get_bits_left(gb));
3170 return AVERROR_INVALIDDATA;
3177 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3178 int *got_frame_ptr, AVPacket *avpkt)
3180 struct LATMContext *latmctx = avctx->priv_data;
3184 if ((err = init_get_bits(&gb, avpkt->data, avpkt->size * 8)) < 0)
3187 // check for LOAS sync word
3188 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3189 return AVERROR_INVALIDDATA;
3191 muxlength = get_bits(&gb, 13) + 3;
3192 // not enough data, the parser should have sorted this
3193 if (muxlength > avpkt->size)
3194 return AVERROR_INVALIDDATA;
3196 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3199 if (!latmctx->initialized) {
3200 if (!avctx->extradata) {
3204 push_output_configuration(&latmctx->aac_ctx);
3205 if ((err = decode_audio_specific_config(
3206 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3207 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3208 pop_output_configuration(&latmctx->aac_ctx);
3211 latmctx->initialized = 1;
3215 if (show_bits(&gb, 12) == 0xfff) {
3216 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3217 "ADTS header detected, probably as result of configuration "
3219 return AVERROR_INVALIDDATA;
3222 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
3228 static av_cold int latm_decode_init(AVCodecContext *avctx)
3230 struct LATMContext *latmctx = avctx->priv_data;
3231 int ret = aac_decode_init(avctx);
3233 if (avctx->extradata_size > 0)
3234 latmctx->initialized = !ret;
3240 AVCodec ff_aac_decoder = {
3242 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3243 .type = AVMEDIA_TYPE_AUDIO,
3244 .id = AV_CODEC_ID_AAC,
3245 .priv_data_size = sizeof(AACContext),
3246 .init = aac_decode_init,
3247 .close = aac_decode_close,
3248 .decode = aac_decode_frame,
3249 .sample_fmts = (const enum AVSampleFormat[]) {
3250 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3252 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3253 .channel_layouts = aac_channel_layout,
3257 Note: This decoder filter is intended to decode LATM streams transferred
3258 in MPEG transport streams which only contain one program.
3259 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3261 AVCodec ff_aac_latm_decoder = {
3263 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3264 .type = AVMEDIA_TYPE_AUDIO,
3265 .id = AV_CODEC_ID_AAC_LATM,
3266 .priv_data_size = sizeof(struct LATMContext),
3267 .init = latm_decode_init,
3268 .close = aac_decode_close,
3269 .decode = latm_decode_frame,
3270 .sample_fmts = (const enum AVSampleFormat[]) {
3271 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3273 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3274 .channel_layouts = aac_channel_layout,