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"
97 #include "aacdectab.h"
98 #include "cbrt_tablegen.h"
101 #include "mpeg4audio.h"
102 #include "aacadtsdec.h"
103 #include "libavutil/intfloat.h"
111 # include "arm/aac.h"
113 # include "mips/aacdec_mips.h"
116 static VLC vlc_scalefactors;
117 static VLC vlc_spectral[11];
119 static int output_configure(AACContext *ac,
120 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
121 enum OCStatus oc_type, int get_new_frame);
123 #define overread_err "Input buffer exhausted before END element found\n"
125 static int count_channels(uint8_t (*layout)[3], int tags)
128 for (i = 0; i < tags; i++) {
129 int syn_ele = layout[i][0];
130 int pos = layout[i][2];
131 sum += (1 + (syn_ele == TYPE_CPE)) *
132 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
138 * Check for the channel element in the current channel position configuration.
139 * If it exists, make sure the appropriate element is allocated and map the
140 * channel order to match the internal FFmpeg channel layout.
142 * @param che_pos current channel position configuration
143 * @param type channel element type
144 * @param id channel element id
145 * @param channels count of the number of channels in the configuration
147 * @return Returns error status. 0 - OK, !0 - error
149 static av_cold int che_configure(AACContext *ac,
150 enum ChannelPosition che_pos,
151 int type, int id, int *channels)
153 if (*channels >= MAX_CHANNELS)
154 return AVERROR_INVALIDDATA;
156 if (!ac->che[type][id]) {
157 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
158 return AVERROR(ENOMEM);
159 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
161 if (type != TYPE_CCE) {
162 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
163 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
164 return AVERROR_INVALIDDATA;
166 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
167 if (type == TYPE_CPE ||
168 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
169 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
173 if (ac->che[type][id])
174 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
175 av_freep(&ac->che[type][id]);
180 static int frame_configure_elements(AVCodecContext *avctx)
182 AACContext *ac = avctx->priv_data;
183 int type, id, ch, ret;
185 /* set channel pointers to internal buffers by default */
186 for (type = 0; type < 4; type++) {
187 for (id = 0; id < MAX_ELEM_ID; id++) {
188 ChannelElement *che = ac->che[type][id];
190 che->ch[0].ret = che->ch[0].ret_buf;
191 che->ch[1].ret = che->ch[1].ret_buf;
196 /* get output buffer */
197 av_frame_unref(ac->frame);
198 if (!avctx->channels)
201 ac->frame->nb_samples = 2048;
202 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
205 /* map output channel pointers to AVFrame data */
206 for (ch = 0; ch < avctx->channels; ch++) {
207 if (ac->output_element[ch])
208 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
214 struct elem_to_channel {
215 uint64_t av_position;
218 uint8_t aac_position;
221 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
222 uint8_t (*layout_map)[3], int offset, uint64_t left,
223 uint64_t right, int pos)
225 if (layout_map[offset][0] == TYPE_CPE) {
226 e2c_vec[offset] = (struct elem_to_channel) {
227 .av_position = left | right,
229 .elem_id = layout_map[offset][1],
234 e2c_vec[offset] = (struct elem_to_channel) {
237 .elem_id = layout_map[offset][1],
240 e2c_vec[offset + 1] = (struct elem_to_channel) {
241 .av_position = right,
243 .elem_id = layout_map[offset + 1][1],
250 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
253 int num_pos_channels = 0;
257 for (i = *current; i < tags; i++) {
258 if (layout_map[i][2] != pos)
260 if (layout_map[i][0] == TYPE_CPE) {
262 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
268 num_pos_channels += 2;
276 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
279 return num_pos_channels;
282 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
284 int i, n, total_non_cc_elements;
285 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
286 int num_front_channels, num_side_channels, num_back_channels;
289 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
294 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
295 if (num_front_channels < 0)
298 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
299 if (num_side_channels < 0)
302 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
303 if (num_back_channels < 0)
307 if (num_front_channels & 1) {
308 e2c_vec[i] = (struct elem_to_channel) {
309 .av_position = AV_CH_FRONT_CENTER,
311 .elem_id = layout_map[i][1],
312 .aac_position = AAC_CHANNEL_FRONT
315 num_front_channels--;
317 if (num_front_channels >= 4) {
318 i += assign_pair(e2c_vec, layout_map, i,
319 AV_CH_FRONT_LEFT_OF_CENTER,
320 AV_CH_FRONT_RIGHT_OF_CENTER,
322 num_front_channels -= 2;
324 if (num_front_channels >= 2) {
325 i += assign_pair(e2c_vec, layout_map, i,
329 num_front_channels -= 2;
331 while (num_front_channels >= 2) {
332 i += assign_pair(e2c_vec, layout_map, i,
336 num_front_channels -= 2;
339 if (num_side_channels >= 2) {
340 i += assign_pair(e2c_vec, layout_map, i,
344 num_side_channels -= 2;
346 while (num_side_channels >= 2) {
347 i += assign_pair(e2c_vec, layout_map, i,
351 num_side_channels -= 2;
354 while (num_back_channels >= 4) {
355 i += assign_pair(e2c_vec, layout_map, i,
359 num_back_channels -= 2;
361 if (num_back_channels >= 2) {
362 i += assign_pair(e2c_vec, layout_map, i,
366 num_back_channels -= 2;
368 if (num_back_channels) {
369 e2c_vec[i] = (struct elem_to_channel) {
370 .av_position = AV_CH_BACK_CENTER,
372 .elem_id = layout_map[i][1],
373 .aac_position = AAC_CHANNEL_BACK
379 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
380 e2c_vec[i] = (struct elem_to_channel) {
381 .av_position = AV_CH_LOW_FREQUENCY,
383 .elem_id = layout_map[i][1],
384 .aac_position = AAC_CHANNEL_LFE
388 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
389 e2c_vec[i] = (struct elem_to_channel) {
390 .av_position = UINT64_MAX,
392 .elem_id = layout_map[i][1],
393 .aac_position = AAC_CHANNEL_LFE
398 // Must choose a stable sort
399 total_non_cc_elements = n = i;
402 for (i = 1; i < n; i++)
403 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
404 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
411 for (i = 0; i < total_non_cc_elements; i++) {
412 layout_map[i][0] = e2c_vec[i].syn_ele;
413 layout_map[i][1] = e2c_vec[i].elem_id;
414 layout_map[i][2] = e2c_vec[i].aac_position;
415 if (e2c_vec[i].av_position != UINT64_MAX) {
416 layout |= e2c_vec[i].av_position;
424 * Save current output configuration if and only if it has been locked.
426 static void push_output_configuration(AACContext *ac) {
427 if (ac->oc[1].status == OC_LOCKED || ac->oc[0].status == OC_NONE) {
428 ac->oc[0] = ac->oc[1];
430 ac->oc[1].status = OC_NONE;
434 * Restore the previous output configuration if and only if the current
435 * configuration is unlocked.
437 static void pop_output_configuration(AACContext *ac) {
438 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
439 ac->oc[1] = ac->oc[0];
440 ac->avctx->channels = ac->oc[1].channels;
441 ac->avctx->channel_layout = ac->oc[1].channel_layout;
442 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
443 ac->oc[1].status, 0);
448 * Configure output channel order based on the current program
449 * configuration element.
451 * @return Returns error status. 0 - OK, !0 - error
453 static int output_configure(AACContext *ac,
454 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
455 enum OCStatus oc_type, int get_new_frame)
457 AVCodecContext *avctx = ac->avctx;
458 int i, channels = 0, ret;
461 if (ac->oc[1].layout_map != layout_map) {
462 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
463 ac->oc[1].layout_map_tags = tags;
466 // Try to sniff a reasonable channel order, otherwise output the
467 // channels in the order the PCE declared them.
468 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
469 layout = sniff_channel_order(layout_map, tags);
470 for (i = 0; i < tags; i++) {
471 int type = layout_map[i][0];
472 int id = layout_map[i][1];
473 int position = layout_map[i][2];
474 // Allocate or free elements depending on if they are in the
475 // current program configuration.
476 ret = che_configure(ac, position, type, id, &channels);
480 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
481 if (layout == AV_CH_FRONT_CENTER) {
482 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
488 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
489 if (layout) avctx->channel_layout = layout;
490 ac->oc[1].channel_layout = layout;
491 avctx->channels = ac->oc[1].channels = channels;
492 ac->oc[1].status = oc_type;
495 if ((ret = frame_configure_elements(ac->avctx)) < 0)
502 static void flush(AVCodecContext *avctx)
504 AACContext *ac= avctx->priv_data;
507 for (type = 3; type >= 0; type--) {
508 for (i = 0; i < MAX_ELEM_ID; i++) {
509 ChannelElement *che = ac->che[type][i];
511 for (j = 0; j <= 1; j++) {
512 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
520 * Set up channel positions based on a default channel configuration
521 * as specified in table 1.17.
523 * @return Returns error status. 0 - OK, !0 - error
525 static int set_default_channel_config(AVCodecContext *avctx,
526 uint8_t (*layout_map)[3],
530 if (channel_config < 1 || channel_config > 7) {
531 av_log(avctx, AV_LOG_ERROR,
532 "invalid default channel configuration (%d)\n",
534 return AVERROR_INVALIDDATA;
536 *tags = tags_per_config[channel_config];
537 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
538 *tags * sizeof(*layout_map));
541 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
542 * However, at least Nero AAC encoder encodes 7.1 streams using the default
543 * channel config 7, mapping the side channels of the original audio stream
544 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
545 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
546 * the incorrect streams as if they were correct (and as the encoder intended).
548 * As actual intended 7.1(wide) streams are very rare, default to assuming a
549 * 7.1 layout was intended.
551 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
552 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
553 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
554 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
555 layout_map[2][2] = AAC_CHANNEL_SIDE;
561 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
563 /* For PCE based channel configurations map the channels solely based
565 if (!ac->oc[1].m4ac.chan_config) {
566 return ac->tag_che_map[type][elem_id];
568 // Allow single CPE stereo files to be signalled with mono configuration.
569 if (!ac->tags_mapped && type == TYPE_CPE &&
570 ac->oc[1].m4ac.chan_config == 1) {
571 uint8_t layout_map[MAX_ELEM_ID*4][3];
573 push_output_configuration(ac);
575 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
577 if (set_default_channel_config(ac->avctx, layout_map,
578 &layout_map_tags, 2) < 0)
580 if (output_configure(ac, layout_map, layout_map_tags,
581 OC_TRIAL_FRAME, 1) < 0)
584 ac->oc[1].m4ac.chan_config = 2;
585 ac->oc[1].m4ac.ps = 0;
588 if (!ac->tags_mapped && type == TYPE_SCE &&
589 ac->oc[1].m4ac.chan_config == 2) {
590 uint8_t layout_map[MAX_ELEM_ID * 4][3];
592 push_output_configuration(ac);
594 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
596 if (set_default_channel_config(ac->avctx, layout_map,
597 &layout_map_tags, 1) < 0)
599 if (output_configure(ac, layout_map, layout_map_tags,
600 OC_TRIAL_FRAME, 1) < 0)
603 ac->oc[1].m4ac.chan_config = 1;
604 if (ac->oc[1].m4ac.sbr)
605 ac->oc[1].m4ac.ps = -1;
607 /* For indexed channel configurations map the channels solely based
609 switch (ac->oc[1].m4ac.chan_config) {
611 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
613 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
616 /* Some streams incorrectly code 5.1 audio as
617 * SCE[0] CPE[0] CPE[1] SCE[1]
619 * SCE[0] CPE[0] CPE[1] LFE[0].
620 * If we seem to have encountered such a stream, transfer
621 * the LFE[0] element to the SCE[1]'s mapping */
622 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
623 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
624 av_log(ac->avctx, AV_LOG_WARNING,
625 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
626 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
627 ac->warned_remapping_once++;
630 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
633 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
635 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
638 /* Some streams incorrectly code 4.0 audio as
639 * SCE[0] CPE[0] LFE[0]
641 * SCE[0] CPE[0] SCE[1].
642 * If we seem to have encountered such a stream, transfer
643 * the SCE[1] element to the LFE[0]'s mapping */
644 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
645 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
646 av_log(ac->avctx, AV_LOG_WARNING,
647 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
648 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
649 ac->warned_remapping_once++;
652 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
654 if (ac->tags_mapped == 2 &&
655 ac->oc[1].m4ac.chan_config == 4 &&
658 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
662 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
665 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
666 } else if (ac->oc[1].m4ac.chan_config == 2) {
670 if (!ac->tags_mapped && type == TYPE_SCE) {
672 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
680 * Decode an array of 4 bit element IDs, optionally interleaved with a
681 * stereo/mono switching bit.
683 * @param type speaker type/position for these channels
685 static void decode_channel_map(uint8_t layout_map[][3],
686 enum ChannelPosition type,
687 GetBitContext *gb, int n)
690 enum RawDataBlockType syn_ele;
692 case AAC_CHANNEL_FRONT:
693 case AAC_CHANNEL_BACK:
694 case AAC_CHANNEL_SIDE:
695 syn_ele = get_bits1(gb);
701 case AAC_CHANNEL_LFE:
705 // AAC_CHANNEL_OFF has no channel map
708 layout_map[0][0] = syn_ele;
709 layout_map[0][1] = get_bits(gb, 4);
710 layout_map[0][2] = type;
716 * Decode program configuration element; reference: table 4.2.
718 * @return Returns error status. 0 - OK, !0 - error
720 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
721 uint8_t (*layout_map)[3],
724 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
729 skip_bits(gb, 2); // object_type
731 sampling_index = get_bits(gb, 4);
732 if (m4ac->sampling_index != sampling_index)
733 av_log(avctx, AV_LOG_WARNING,
734 "Sample rate index in program config element does not "
735 "match the sample rate index configured by the container.\n");
737 num_front = get_bits(gb, 4);
738 num_side = get_bits(gb, 4);
739 num_back = get_bits(gb, 4);
740 num_lfe = get_bits(gb, 2);
741 num_assoc_data = get_bits(gb, 3);
742 num_cc = get_bits(gb, 4);
745 skip_bits(gb, 4); // mono_mixdown_tag
747 skip_bits(gb, 4); // stereo_mixdown_tag
750 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
752 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
753 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
756 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
758 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
760 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
762 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
765 skip_bits_long(gb, 4 * num_assoc_data);
767 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
772 /* comment field, first byte is length */
773 comment_len = get_bits(gb, 8) * 8;
774 if (get_bits_left(gb) < comment_len) {
775 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
776 return AVERROR_INVALIDDATA;
778 skip_bits_long(gb, comment_len);
783 * Decode GA "General Audio" specific configuration; reference: table 4.1.
785 * @param ac pointer to AACContext, may be null
786 * @param avctx pointer to AVCCodecContext, used for logging
788 * @return Returns error status. 0 - OK, !0 - error
790 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
792 MPEG4AudioConfig *m4ac,
795 int extension_flag, ret, ep_config, res_flags;
796 uint8_t layout_map[MAX_ELEM_ID*4][3];
799 if (get_bits1(gb)) { // frameLengthFlag
800 avpriv_request_sample(avctx, "960/120 MDCT window");
801 return AVERROR_PATCHWELCOME;
803 m4ac->frame_length_short = 0;
805 if (get_bits1(gb)) // dependsOnCoreCoder
806 skip_bits(gb, 14); // coreCoderDelay
807 extension_flag = get_bits1(gb);
809 if (m4ac->object_type == AOT_AAC_SCALABLE ||
810 m4ac->object_type == AOT_ER_AAC_SCALABLE)
811 skip_bits(gb, 3); // layerNr
813 if (channel_config == 0) {
814 skip_bits(gb, 4); // element_instance_tag
815 tags = decode_pce(avctx, m4ac, layout_map, gb);
819 if ((ret = set_default_channel_config(avctx, layout_map,
820 &tags, channel_config)))
824 if (count_channels(layout_map, tags) > 1) {
826 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
829 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
832 if (extension_flag) {
833 switch (m4ac->object_type) {
835 skip_bits(gb, 5); // numOfSubFrame
836 skip_bits(gb, 11); // layer_length
840 case AOT_ER_AAC_SCALABLE:
842 res_flags = get_bits(gb, 3);
844 avpriv_report_missing_feature(avctx,
845 "AAC data resilience (flags %x)",
847 return AVERROR_PATCHWELCOME;
851 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
853 switch (m4ac->object_type) {
856 case AOT_ER_AAC_SCALABLE:
858 ep_config = get_bits(gb, 2);
860 avpriv_report_missing_feature(avctx,
861 "epConfig %d", ep_config);
862 return AVERROR_PATCHWELCOME;
868 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
870 MPEG4AudioConfig *m4ac,
873 int ret, ep_config, res_flags;
874 uint8_t layout_map[MAX_ELEM_ID*4][3];
876 const int ELDEXT_TERM = 0;
881 m4ac->frame_length_short = get_bits1(gb);
882 res_flags = get_bits(gb, 3);
884 avpriv_report_missing_feature(avctx,
885 "AAC data resilience (flags %x)",
887 return AVERROR_PATCHWELCOME;
890 if (get_bits1(gb)) { // ldSbrPresentFlag
891 avpriv_report_missing_feature(avctx,
893 return AVERROR_PATCHWELCOME;
896 while (get_bits(gb, 4) != ELDEXT_TERM) {
897 int len = get_bits(gb, 4);
899 len += get_bits(gb, 8);
901 len += get_bits(gb, 16);
902 if (get_bits_left(gb) < len * 8 + 4) {
903 av_log(avctx, AV_LOG_ERROR, overread_err);
904 return AVERROR_INVALIDDATA;
906 skip_bits_long(gb, 8 * len);
909 if ((ret = set_default_channel_config(avctx, layout_map,
910 &tags, channel_config)))
913 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
916 ep_config = get_bits(gb, 2);
918 avpriv_report_missing_feature(avctx,
919 "epConfig %d", ep_config);
920 return AVERROR_PATCHWELCOME;
926 * Decode audio specific configuration; reference: table 1.13.
928 * @param ac pointer to AACContext, may be null
929 * @param avctx pointer to AVCCodecContext, used for logging
930 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
931 * @param data pointer to buffer holding an audio specific config
932 * @param bit_size size of audio specific config or data in bits
933 * @param sync_extension look for an appended sync extension
935 * @return Returns error status or number of consumed bits. <0 - error
937 static int decode_audio_specific_config(AACContext *ac,
938 AVCodecContext *avctx,
939 MPEG4AudioConfig *m4ac,
940 const uint8_t *data, int bit_size,
946 ff_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
947 for (i = 0; i < bit_size >> 3; i++)
948 ff_dlog(avctx, "%02x ", data[i]);
949 ff_dlog(avctx, "\n");
951 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
954 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
955 sync_extension)) < 0)
956 return AVERROR_INVALIDDATA;
957 if (m4ac->sampling_index > 12) {
958 av_log(avctx, AV_LOG_ERROR,
959 "invalid sampling rate index %d\n",
960 m4ac->sampling_index);
961 return AVERROR_INVALIDDATA;
963 if (m4ac->object_type == AOT_ER_AAC_LD &&
964 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
965 av_log(avctx, AV_LOG_ERROR,
966 "invalid low delay sampling rate index %d\n",
967 m4ac->sampling_index);
968 return AVERROR_INVALIDDATA;
971 skip_bits_long(&gb, i);
973 switch (m4ac->object_type) {
979 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
980 m4ac, m4ac->chan_config)) < 0)
984 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
985 m4ac, m4ac->chan_config)) < 0)
989 avpriv_report_missing_feature(avctx,
990 "Audio object type %s%d",
991 m4ac->sbr == 1 ? "SBR+" : "",
993 return AVERROR(ENOSYS);
997 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
998 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
999 m4ac->sample_rate, m4ac->sbr,
1002 return get_bits_count(&gb);
1006 * linear congruential pseudorandom number generator
1008 * @param previous_val pointer to the current state of the generator
1010 * @return Returns a 32-bit pseudorandom integer
1012 static av_always_inline int lcg_random(unsigned previous_val)
1014 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1018 static av_always_inline void reset_predict_state(PredictorState *ps)
1028 static void reset_all_predictors(PredictorState *ps)
1031 for (i = 0; i < MAX_PREDICTORS; i++)
1032 reset_predict_state(&ps[i]);
1035 static int sample_rate_idx (int rate)
1037 if (92017 <= rate) return 0;
1038 else if (75132 <= rate) return 1;
1039 else if (55426 <= rate) return 2;
1040 else if (46009 <= rate) return 3;
1041 else if (37566 <= rate) return 4;
1042 else if (27713 <= rate) return 5;
1043 else if (23004 <= rate) return 6;
1044 else if (18783 <= rate) return 7;
1045 else if (13856 <= rate) return 8;
1046 else if (11502 <= rate) return 9;
1047 else if (9391 <= rate) return 10;
1051 static void reset_predictor_group(PredictorState *ps, int group_num)
1054 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1055 reset_predict_state(&ps[i]);
1058 #define AAC_INIT_VLC_STATIC(num, size) \
1059 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1060 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1061 sizeof(ff_aac_spectral_bits[num][0]), \
1062 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1063 sizeof(ff_aac_spectral_codes[num][0]), \
1066 static void aacdec_init(AACContext *ac);
1068 static av_cold int aac_decode_init(AVCodecContext *avctx)
1070 AACContext *ac = avctx->priv_data;
1074 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1078 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1080 if (avctx->extradata_size > 0) {
1081 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1083 avctx->extradata_size * 8,
1088 uint8_t layout_map[MAX_ELEM_ID*4][3];
1089 int layout_map_tags;
1091 sr = sample_rate_idx(avctx->sample_rate);
1092 ac->oc[1].m4ac.sampling_index = sr;
1093 ac->oc[1].m4ac.channels = avctx->channels;
1094 ac->oc[1].m4ac.sbr = -1;
1095 ac->oc[1].m4ac.ps = -1;
1097 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1098 if (ff_mpeg4audio_channels[i] == avctx->channels)
1100 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1103 ac->oc[1].m4ac.chan_config = i;
1105 if (ac->oc[1].m4ac.chan_config) {
1106 int ret = set_default_channel_config(avctx, layout_map,
1107 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1109 output_configure(ac, layout_map, layout_map_tags,
1111 else if (avctx->err_recognition & AV_EF_EXPLODE)
1112 return AVERROR_INVALIDDATA;
1116 if (avctx->channels > MAX_CHANNELS) {
1117 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1118 return AVERROR_INVALIDDATA;
1121 AAC_INIT_VLC_STATIC( 0, 304);
1122 AAC_INIT_VLC_STATIC( 1, 270);
1123 AAC_INIT_VLC_STATIC( 2, 550);
1124 AAC_INIT_VLC_STATIC( 3, 300);
1125 AAC_INIT_VLC_STATIC( 4, 328);
1126 AAC_INIT_VLC_STATIC( 5, 294);
1127 AAC_INIT_VLC_STATIC( 6, 306);
1128 AAC_INIT_VLC_STATIC( 7, 268);
1129 AAC_INIT_VLC_STATIC( 8, 510);
1130 AAC_INIT_VLC_STATIC( 9, 366);
1131 AAC_INIT_VLC_STATIC(10, 462);
1135 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
1137 return AVERROR(ENOMEM);
1140 ac->random_state = 0x1f2e3d4c;
1144 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1145 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1146 ff_aac_scalefactor_bits,
1147 sizeof(ff_aac_scalefactor_bits[0]),
1148 sizeof(ff_aac_scalefactor_bits[0]),
1149 ff_aac_scalefactor_code,
1150 sizeof(ff_aac_scalefactor_code[0]),
1151 sizeof(ff_aac_scalefactor_code[0]),
1154 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1155 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1156 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1157 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1158 ret = ff_imdct15_init(&ac->mdct480, 5);
1162 // window initialization
1163 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1164 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1165 ff_init_ff_sine_windows(10);
1166 ff_init_ff_sine_windows( 9);
1167 ff_init_ff_sine_windows( 7);
1175 * Skip data_stream_element; reference: table 4.10.
1177 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1179 int byte_align = get_bits1(gb);
1180 int count = get_bits(gb, 8);
1182 count += get_bits(gb, 8);
1186 if (get_bits_left(gb) < 8 * count) {
1187 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1188 return AVERROR_INVALIDDATA;
1190 skip_bits_long(gb, 8 * count);
1194 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1198 if (get_bits1(gb)) {
1199 ics->predictor_reset_group = get_bits(gb, 5);
1200 if (ics->predictor_reset_group == 0 ||
1201 ics->predictor_reset_group > 30) {
1202 av_log(ac->avctx, AV_LOG_ERROR,
1203 "Invalid Predictor Reset Group.\n");
1204 return AVERROR_INVALIDDATA;
1207 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1208 ics->prediction_used[sfb] = get_bits1(gb);
1214 * Decode Long Term Prediction data; reference: table 4.xx.
1216 static void decode_ltp(LongTermPrediction *ltp,
1217 GetBitContext *gb, uint8_t max_sfb)
1221 ltp->lag = get_bits(gb, 11);
1222 ltp->coef = ltp_coef[get_bits(gb, 3)];
1223 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1224 ltp->used[sfb] = get_bits1(gb);
1228 * Decode Individual Channel Stream info; reference: table 4.6.
1230 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1233 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1234 const int aot = m4ac->object_type;
1235 const int sampling_index = m4ac->sampling_index;
1236 if (aot != AOT_ER_AAC_ELD) {
1237 if (get_bits1(gb)) {
1238 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1239 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1240 return AVERROR_INVALIDDATA;
1242 ics->window_sequence[1] = ics->window_sequence[0];
1243 ics->window_sequence[0] = get_bits(gb, 2);
1244 if (aot == AOT_ER_AAC_LD &&
1245 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1246 av_log(ac->avctx, AV_LOG_ERROR,
1247 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1248 "window sequence %d found.\n", ics->window_sequence[0]);
1249 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1250 return AVERROR_INVALIDDATA;
1252 ics->use_kb_window[1] = ics->use_kb_window[0];
1253 ics->use_kb_window[0] = get_bits1(gb);
1255 ics->num_window_groups = 1;
1256 ics->group_len[0] = 1;
1257 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1259 ics->max_sfb = get_bits(gb, 4);
1260 for (i = 0; i < 7; i++) {
1261 if (get_bits1(gb)) {
1262 ics->group_len[ics->num_window_groups - 1]++;
1264 ics->num_window_groups++;
1265 ics->group_len[ics->num_window_groups - 1] = 1;
1268 ics->num_windows = 8;
1269 ics->swb_offset = ff_swb_offset_128[sampling_index];
1270 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1271 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1272 ics->predictor_present = 0;
1274 ics->max_sfb = get_bits(gb, 6);
1275 ics->num_windows = 1;
1276 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1277 if (m4ac->frame_length_short) {
1278 ics->swb_offset = ff_swb_offset_480[sampling_index];
1279 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1280 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1282 ics->swb_offset = ff_swb_offset_512[sampling_index];
1283 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1284 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1286 if (!ics->num_swb || !ics->swb_offset)
1289 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1290 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1291 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1293 if (aot != AOT_ER_AAC_ELD) {
1294 ics->predictor_present = get_bits1(gb);
1295 ics->predictor_reset_group = 0;
1297 if (ics->predictor_present) {
1298 if (aot == AOT_AAC_MAIN) {
1299 if (decode_prediction(ac, ics, gb)) {
1302 } else if (aot == AOT_AAC_LC ||
1303 aot == AOT_ER_AAC_LC) {
1304 av_log(ac->avctx, AV_LOG_ERROR,
1305 "Prediction is not allowed in AAC-LC.\n");
1308 if (aot == AOT_ER_AAC_LD) {
1309 av_log(ac->avctx, AV_LOG_ERROR,
1310 "LTP in ER AAC LD not yet implemented.\n");
1311 return AVERROR_PATCHWELCOME;
1313 if ((ics->ltp.present = get_bits(gb, 1)))
1314 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1319 if (ics->max_sfb > ics->num_swb) {
1320 av_log(ac->avctx, AV_LOG_ERROR,
1321 "Number of scalefactor bands in group (%d) "
1322 "exceeds limit (%d).\n",
1323 ics->max_sfb, ics->num_swb);
1330 return AVERROR_INVALIDDATA;
1334 * Decode band types (section_data payload); reference: table 4.46.
1336 * @param band_type array of the used band type
1337 * @param band_type_run_end array of the last scalefactor band of a band type run
1339 * @return Returns error status. 0 - OK, !0 - error
1341 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1342 int band_type_run_end[120], GetBitContext *gb,
1343 IndividualChannelStream *ics)
1346 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1347 for (g = 0; g < ics->num_window_groups; g++) {
1349 while (k < ics->max_sfb) {
1350 uint8_t sect_end = k;
1352 int sect_band_type = get_bits(gb, 4);
1353 if (sect_band_type == 12) {
1354 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1355 return AVERROR_INVALIDDATA;
1358 sect_len_incr = get_bits(gb, bits);
1359 sect_end += sect_len_incr;
1360 if (get_bits_left(gb) < 0) {
1361 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1362 return AVERROR_INVALIDDATA;
1364 if (sect_end > ics->max_sfb) {
1365 av_log(ac->avctx, AV_LOG_ERROR,
1366 "Number of bands (%d) exceeds limit (%d).\n",
1367 sect_end, ics->max_sfb);
1368 return AVERROR_INVALIDDATA;
1370 } while (sect_len_incr == (1 << bits) - 1);
1371 for (; k < sect_end; k++) {
1372 band_type [idx] = sect_band_type;
1373 band_type_run_end[idx++] = sect_end;
1381 * Decode scalefactors; reference: table 4.47.
1383 * @param global_gain first scalefactor value as scalefactors are differentially coded
1384 * @param band_type array of the used band type
1385 * @param band_type_run_end array of the last scalefactor band of a band type run
1386 * @param sf array of scalefactors or intensity stereo positions
1388 * @return Returns error status. 0 - OK, !0 - error
1390 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1391 unsigned int global_gain,
1392 IndividualChannelStream *ics,
1393 enum BandType band_type[120],
1394 int band_type_run_end[120])
1397 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1400 for (g = 0; g < ics->num_window_groups; g++) {
1401 for (i = 0; i < ics->max_sfb;) {
1402 int run_end = band_type_run_end[idx];
1403 if (band_type[idx] == ZERO_BT) {
1404 for (; i < run_end; i++, idx++)
1406 } else if ((band_type[idx] == INTENSITY_BT) ||
1407 (band_type[idx] == INTENSITY_BT2)) {
1408 for (; i < run_end; i++, idx++) {
1409 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1410 clipped_offset = av_clip(offset[2], -155, 100);
1411 if (offset[2] != clipped_offset) {
1412 avpriv_request_sample(ac->avctx,
1413 "If you heard an audible artifact, there may be a bug in the decoder. "
1414 "Clipped intensity stereo position (%d -> %d)",
1415 offset[2], clipped_offset);
1417 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1419 } else if (band_type[idx] == NOISE_BT) {
1420 for (; i < run_end; i++, idx++) {
1421 if (noise_flag-- > 0)
1422 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1424 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1425 clipped_offset = av_clip(offset[1], -100, 155);
1426 if (offset[1] != clipped_offset) {
1427 avpriv_request_sample(ac->avctx,
1428 "If you heard an audible artifact, there may be a bug in the decoder. "
1429 "Clipped noise gain (%d -> %d)",
1430 offset[1], clipped_offset);
1432 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1435 for (; i < run_end; i++, idx++) {
1436 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1437 if (offset[0] > 255U) {
1438 av_log(ac->avctx, AV_LOG_ERROR,
1439 "Scalefactor (%d) out of range.\n", offset[0]);
1440 return AVERROR_INVALIDDATA;
1442 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1451 * Decode pulse data; reference: table 4.7.
1453 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1454 const uint16_t *swb_offset, int num_swb)
1457 pulse->num_pulse = get_bits(gb, 2) + 1;
1458 pulse_swb = get_bits(gb, 6);
1459 if (pulse_swb >= num_swb)
1461 pulse->pos[0] = swb_offset[pulse_swb];
1462 pulse->pos[0] += get_bits(gb, 5);
1463 if (pulse->pos[0] >= swb_offset[num_swb])
1465 pulse->amp[0] = get_bits(gb, 4);
1466 for (i = 1; i < pulse->num_pulse; i++) {
1467 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1468 if (pulse->pos[i] >= swb_offset[num_swb])
1470 pulse->amp[i] = get_bits(gb, 4);
1476 * Decode Temporal Noise Shaping data; reference: table 4.48.
1478 * @return Returns error status. 0 - OK, !0 - error
1480 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1481 GetBitContext *gb, const IndividualChannelStream *ics)
1483 int w, filt, i, coef_len, coef_res, coef_compress;
1484 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1485 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1486 for (w = 0; w < ics->num_windows; w++) {
1487 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1488 coef_res = get_bits1(gb);
1490 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1492 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1494 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1495 av_log(ac->avctx, AV_LOG_ERROR,
1496 "TNS filter order %d is greater than maximum %d.\n",
1497 tns->order[w][filt], tns_max_order);
1498 tns->order[w][filt] = 0;
1499 return AVERROR_INVALIDDATA;
1501 if (tns->order[w][filt]) {
1502 tns->direction[w][filt] = get_bits1(gb);
1503 coef_compress = get_bits1(gb);
1504 coef_len = coef_res + 3 - coef_compress;
1505 tmp2_idx = 2 * coef_compress + coef_res;
1507 for (i = 0; i < tns->order[w][filt]; i++)
1508 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1517 * Decode Mid/Side data; reference: table 4.54.
1519 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1520 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1521 * [3] reserved for scalable AAC
1523 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1527 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1528 if (ms_present == 1) {
1529 for (idx = 0; idx < max_idx; idx++)
1530 cpe->ms_mask[idx] = get_bits1(gb);
1531 } else if (ms_present == 2) {
1532 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1537 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1541 *dst++ = v[idx & 15] * s;
1542 *dst++ = v[idx>>4 & 15] * s;
1548 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1552 *dst++ = v[idx & 3] * s;
1553 *dst++ = v[idx>>2 & 3] * s;
1554 *dst++ = v[idx>>4 & 3] * s;
1555 *dst++ = v[idx>>6 & 3] * s;
1561 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1562 unsigned sign, const float *scale)
1564 union av_intfloat32 s0, s1;
1566 s0.f = s1.f = *scale;
1567 s0.i ^= sign >> 1 << 31;
1570 *dst++ = v[idx & 15] * s0.f;
1571 *dst++ = v[idx>>4 & 15] * s1.f;
1578 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1579 unsigned sign, const float *scale)
1581 unsigned nz = idx >> 12;
1582 union av_intfloat32 s = { .f = *scale };
1583 union av_intfloat32 t;
1585 t.i = s.i ^ (sign & 1U<<31);
1586 *dst++ = v[idx & 3] * t.f;
1588 sign <<= nz & 1; nz >>= 1;
1589 t.i = s.i ^ (sign & 1U<<31);
1590 *dst++ = v[idx>>2 & 3] * t.f;
1592 sign <<= nz & 1; nz >>= 1;
1593 t.i = s.i ^ (sign & 1U<<31);
1594 *dst++ = v[idx>>4 & 3] * t.f;
1597 t.i = s.i ^ (sign & 1U<<31);
1598 *dst++ = v[idx>>6 & 3] * t.f;
1605 * Decode spectral data; reference: table 4.50.
1606 * Dequantize and scale spectral data; reference: 4.6.3.3.
1608 * @param coef array of dequantized, scaled spectral data
1609 * @param sf array of scalefactors or intensity stereo positions
1610 * @param pulse_present set if pulses are present
1611 * @param pulse pointer to pulse data struct
1612 * @param band_type array of the used band type
1614 * @return Returns error status. 0 - OK, !0 - error
1616 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1617 GetBitContext *gb, const float sf[120],
1618 int pulse_present, const Pulse *pulse,
1619 const IndividualChannelStream *ics,
1620 enum BandType band_type[120])
1622 int i, k, g, idx = 0;
1623 const int c = 1024 / ics->num_windows;
1624 const uint16_t *offsets = ics->swb_offset;
1625 float *coef_base = coef;
1627 for (g = 0; g < ics->num_windows; g++)
1628 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1629 sizeof(float) * (c - offsets[ics->max_sfb]));
1631 for (g = 0; g < ics->num_window_groups; g++) {
1632 unsigned g_len = ics->group_len[g];
1634 for (i = 0; i < ics->max_sfb; i++, idx++) {
1635 const unsigned cbt_m1 = band_type[idx] - 1;
1636 float *cfo = coef + offsets[i];
1637 int off_len = offsets[i + 1] - offsets[i];
1640 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1641 for (group = 0; group < g_len; group++, cfo+=128) {
1642 memset(cfo, 0, off_len * sizeof(float));
1644 } else if (cbt_m1 == NOISE_BT - 1) {
1645 for (group = 0; group < g_len; group++, cfo+=128) {
1649 for (k = 0; k < off_len; k++) {
1650 ac->random_state = lcg_random(ac->random_state);
1651 cfo[k] = ac->random_state;
1654 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1655 scale = sf[idx] / sqrtf(band_energy);
1656 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1659 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1660 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1661 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1662 OPEN_READER(re, gb);
1664 switch (cbt_m1 >> 1) {
1666 for (group = 0; group < g_len; group++, cfo+=128) {
1674 UPDATE_CACHE(re, gb);
1675 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1676 cb_idx = cb_vector_idx[code];
1677 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1683 for (group = 0; group < g_len; group++, cfo+=128) {
1693 UPDATE_CACHE(re, gb);
1694 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1695 cb_idx = cb_vector_idx[code];
1696 nnz = cb_idx >> 8 & 15;
1697 bits = nnz ? GET_CACHE(re, gb) : 0;
1698 LAST_SKIP_BITS(re, gb, nnz);
1699 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1705 for (group = 0; group < g_len; group++, cfo+=128) {
1713 UPDATE_CACHE(re, gb);
1714 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1715 cb_idx = cb_vector_idx[code];
1716 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1723 for (group = 0; group < g_len; group++, cfo+=128) {
1733 UPDATE_CACHE(re, gb);
1734 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1735 cb_idx = cb_vector_idx[code];
1736 nnz = cb_idx >> 8 & 15;
1737 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1738 LAST_SKIP_BITS(re, gb, nnz);
1739 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1745 for (group = 0; group < g_len; group++, cfo+=128) {
1747 uint32_t *icf = (uint32_t *) cf;
1757 UPDATE_CACHE(re, gb);
1758 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1766 cb_idx = cb_vector_idx[code];
1769 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1770 LAST_SKIP_BITS(re, gb, nnz);
1772 for (j = 0; j < 2; j++) {
1776 /* The total length of escape_sequence must be < 22 bits according
1777 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1778 UPDATE_CACHE(re, gb);
1779 b = GET_CACHE(re, gb);
1780 b = 31 - av_log2(~b);
1783 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1784 return AVERROR_INVALIDDATA;
1787 SKIP_BITS(re, gb, b + 1);
1789 n = (1 << b) + SHOW_UBITS(re, gb, b);
1790 LAST_SKIP_BITS(re, gb, b);
1791 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1794 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1795 *icf++ = (bits & 1U<<31) | v;
1802 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1806 CLOSE_READER(re, gb);
1812 if (pulse_present) {
1814 for (i = 0; i < pulse->num_pulse; i++) {
1815 float co = coef_base[ pulse->pos[i] ];
1816 while (offsets[idx + 1] <= pulse->pos[i])
1818 if (band_type[idx] != NOISE_BT && sf[idx]) {
1819 float ico = -pulse->amp[i];
1822 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1824 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1831 static av_always_inline float flt16_round(float pf)
1833 union av_intfloat32 tmp;
1835 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1839 static av_always_inline float flt16_even(float pf)
1841 union av_intfloat32 tmp;
1843 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1847 static av_always_inline float flt16_trunc(float pf)
1849 union av_intfloat32 pun;
1851 pun.i &= 0xFFFF0000U;
1855 static av_always_inline void predict(PredictorState *ps, float *coef,
1858 const float a = 0.953125; // 61.0 / 64
1859 const float alpha = 0.90625; // 29.0 / 32
1863 float r0 = ps->r0, r1 = ps->r1;
1864 float cor0 = ps->cor0, cor1 = ps->cor1;
1865 float var0 = ps->var0, var1 = ps->var1;
1867 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1868 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1870 pv = flt16_round(k1 * r0 + k2 * r1);
1877 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1878 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1879 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1880 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1882 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1883 ps->r0 = flt16_trunc(a * e0);
1887 * Apply AAC-Main style frequency domain prediction.
1889 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1893 if (!sce->ics.predictor_initialized) {
1894 reset_all_predictors(sce->predictor_state);
1895 sce->ics.predictor_initialized = 1;
1898 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1900 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1902 for (k = sce->ics.swb_offset[sfb];
1903 k < sce->ics.swb_offset[sfb + 1];
1905 predict(&sce->predictor_state[k], &sce->coeffs[k],
1906 sce->ics.predictor_present &&
1907 sce->ics.prediction_used[sfb]);
1910 if (sce->ics.predictor_reset_group)
1911 reset_predictor_group(sce->predictor_state,
1912 sce->ics.predictor_reset_group);
1914 reset_all_predictors(sce->predictor_state);
1918 * Decode an individual_channel_stream payload; reference: table 4.44.
1920 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1921 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1923 * @return Returns error status. 0 - OK, !0 - error
1925 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1926 GetBitContext *gb, int common_window, int scale_flag)
1929 TemporalNoiseShaping *tns = &sce->tns;
1930 IndividualChannelStream *ics = &sce->ics;
1931 float *out = sce->coeffs;
1932 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1935 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1936 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1937 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1938 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1939 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1941 /* This assignment is to silence a GCC warning about the variable being used
1942 * uninitialized when in fact it always is.
1944 pulse.num_pulse = 0;
1946 global_gain = get_bits(gb, 8);
1948 if (!common_window && !scale_flag) {
1949 if (decode_ics_info(ac, ics, gb) < 0)
1950 return AVERROR_INVALIDDATA;
1953 if ((ret = decode_band_types(ac, sce->band_type,
1954 sce->band_type_run_end, gb, ics)) < 0)
1956 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1957 sce->band_type, sce->band_type_run_end)) < 0)
1962 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1963 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1964 av_log(ac->avctx, AV_LOG_ERROR,
1965 "Pulse tool not allowed in eight short sequence.\n");
1966 return AVERROR_INVALIDDATA;
1968 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1969 av_log(ac->avctx, AV_LOG_ERROR,
1970 "Pulse data corrupt or invalid.\n");
1971 return AVERROR_INVALIDDATA;
1974 tns->present = get_bits1(gb);
1975 if (tns->present && !er_syntax)
1976 if (decode_tns(ac, tns, gb, ics) < 0)
1977 return AVERROR_INVALIDDATA;
1978 if (!eld_syntax && get_bits1(gb)) {
1979 avpriv_request_sample(ac->avctx, "SSR");
1980 return AVERROR_PATCHWELCOME;
1982 // I see no textual basis in the spec for this occurring after SSR gain
1983 // control, but this is what both reference and real implmentations do
1984 if (tns->present && er_syntax)
1985 if (decode_tns(ac, tns, gb, ics) < 0)
1986 return AVERROR_INVALIDDATA;
1989 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1990 &pulse, ics, sce->band_type) < 0)
1991 return AVERROR_INVALIDDATA;
1993 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1994 apply_prediction(ac, sce);
2000 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
2002 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
2004 const IndividualChannelStream *ics = &cpe->ch[0].ics;
2005 float *ch0 = cpe->ch[0].coeffs;
2006 float *ch1 = cpe->ch[1].coeffs;
2007 int g, i, group, idx = 0;
2008 const uint16_t *offsets = ics->swb_offset;
2009 for (g = 0; g < ics->num_window_groups; g++) {
2010 for (i = 0; i < ics->max_sfb; i++, idx++) {
2011 if (cpe->ms_mask[idx] &&
2012 cpe->ch[0].band_type[idx] < NOISE_BT &&
2013 cpe->ch[1].band_type[idx] < NOISE_BT) {
2014 for (group = 0; group < ics->group_len[g]; group++) {
2015 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2016 ch1 + group * 128 + offsets[i],
2017 offsets[i+1] - offsets[i]);
2021 ch0 += ics->group_len[g] * 128;
2022 ch1 += ics->group_len[g] * 128;
2027 * intensity stereo decoding; reference: 4.6.8.2.3
2029 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2030 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2031 * [3] reserved for scalable AAC
2033 static void apply_intensity_stereo(AACContext *ac,
2034 ChannelElement *cpe, int ms_present)
2036 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2037 SingleChannelElement *sce1 = &cpe->ch[1];
2038 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2039 const uint16_t *offsets = ics->swb_offset;
2040 int g, group, i, idx = 0;
2043 for (g = 0; g < ics->num_window_groups; g++) {
2044 for (i = 0; i < ics->max_sfb;) {
2045 if (sce1->band_type[idx] == INTENSITY_BT ||
2046 sce1->band_type[idx] == INTENSITY_BT2) {
2047 const int bt_run_end = sce1->band_type_run_end[idx];
2048 for (; i < bt_run_end; i++, idx++) {
2049 c = -1 + 2 * (sce1->band_type[idx] - 14);
2051 c *= 1 - 2 * cpe->ms_mask[idx];
2052 scale = c * sce1->sf[idx];
2053 for (group = 0; group < ics->group_len[g]; group++)
2054 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2055 coef0 + group * 128 + offsets[i],
2057 offsets[i + 1] - offsets[i]);
2060 int bt_run_end = sce1->band_type_run_end[idx];
2061 idx += bt_run_end - i;
2065 coef0 += ics->group_len[g] * 128;
2066 coef1 += ics->group_len[g] * 128;
2071 * Decode a channel_pair_element; reference: table 4.4.
2073 * @return Returns error status. 0 - OK, !0 - error
2075 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2077 int i, ret, common_window, ms_present = 0;
2078 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2080 common_window = eld_syntax || get_bits1(gb);
2081 if (common_window) {
2082 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2083 return AVERROR_INVALIDDATA;
2084 i = cpe->ch[1].ics.use_kb_window[0];
2085 cpe->ch[1].ics = cpe->ch[0].ics;
2086 cpe->ch[1].ics.use_kb_window[1] = i;
2087 if (cpe->ch[1].ics.predictor_present &&
2088 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2089 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2090 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2091 ms_present = get_bits(gb, 2);
2092 if (ms_present == 3) {
2093 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2094 return AVERROR_INVALIDDATA;
2095 } else if (ms_present)
2096 decode_mid_side_stereo(cpe, gb, ms_present);
2098 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2100 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2103 if (common_window) {
2105 apply_mid_side_stereo(ac, cpe);
2106 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2107 apply_prediction(ac, &cpe->ch[0]);
2108 apply_prediction(ac, &cpe->ch[1]);
2112 apply_intensity_stereo(ac, cpe, ms_present);
2116 static const float cce_scale[] = {
2117 1.09050773266525765921, //2^(1/8)
2118 1.18920711500272106672, //2^(1/4)
2124 * Decode coupling_channel_element; reference: table 4.8.
2126 * @return Returns error status. 0 - OK, !0 - error
2128 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2134 SingleChannelElement *sce = &che->ch[0];
2135 ChannelCoupling *coup = &che->coup;
2137 coup->coupling_point = 2 * get_bits1(gb);
2138 coup->num_coupled = get_bits(gb, 3);
2139 for (c = 0; c <= coup->num_coupled; c++) {
2141 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2142 coup->id_select[c] = get_bits(gb, 4);
2143 if (coup->type[c] == TYPE_CPE) {
2144 coup->ch_select[c] = get_bits(gb, 2);
2145 if (coup->ch_select[c] == 3)
2148 coup->ch_select[c] = 2;
2150 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2152 sign = get_bits(gb, 1);
2153 scale = cce_scale[get_bits(gb, 2)];
2155 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2158 for (c = 0; c < num_gain; c++) {
2162 float gain_cache = 1.0;
2164 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2165 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2166 gain_cache = powf(scale, -gain);
2168 if (coup->coupling_point == AFTER_IMDCT) {
2169 coup->gain[c][0] = gain_cache;
2171 for (g = 0; g < sce->ics.num_window_groups; g++) {
2172 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2173 if (sce->band_type[idx] != ZERO_BT) {
2175 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2183 gain_cache = powf(scale, -t) * s;
2186 coup->gain[c][idx] = gain_cache;
2196 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2198 * @return Returns number of bytes consumed.
2200 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2204 int num_excl_chan = 0;
2207 for (i = 0; i < 7; i++)
2208 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2209 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2211 return num_excl_chan / 7;
2215 * Decode dynamic range information; reference: table 4.52.
2217 * @return Returns number of bytes consumed.
2219 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2223 int drc_num_bands = 1;
2226 /* pce_tag_present? */
2227 if (get_bits1(gb)) {
2228 che_drc->pce_instance_tag = get_bits(gb, 4);
2229 skip_bits(gb, 4); // tag_reserved_bits
2233 /* excluded_chns_present? */
2234 if (get_bits1(gb)) {
2235 n += decode_drc_channel_exclusions(che_drc, gb);
2238 /* drc_bands_present? */
2239 if (get_bits1(gb)) {
2240 che_drc->band_incr = get_bits(gb, 4);
2241 che_drc->interpolation_scheme = get_bits(gb, 4);
2243 drc_num_bands += che_drc->band_incr;
2244 for (i = 0; i < drc_num_bands; i++) {
2245 che_drc->band_top[i] = get_bits(gb, 8);
2250 /* prog_ref_level_present? */
2251 if (get_bits1(gb)) {
2252 che_drc->prog_ref_level = get_bits(gb, 7);
2253 skip_bits1(gb); // prog_ref_level_reserved_bits
2257 for (i = 0; i < drc_num_bands; i++) {
2258 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2259 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2266 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2268 int i, major, minor;
2273 get_bits(gb, 13); len -= 13;
2275 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2276 buf[i] = get_bits(gb, 8);
2279 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2280 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2282 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2283 ac->avctx->internal->skip_samples = 1024;
2287 skip_bits_long(gb, len);
2293 * Decode extension data (incomplete); reference: table 4.51.
2295 * @param cnt length of TYPE_FIL syntactic element in bytes
2297 * @return Returns number of bytes consumed
2299 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2300 ChannelElement *che, enum RawDataBlockType elem_type)
2304 int type = get_bits(gb, 4);
2306 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2307 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2309 switch (type) { // extension type
2310 case EXT_SBR_DATA_CRC:
2314 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2316 } else if (!ac->oc[1].m4ac.sbr) {
2317 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2318 skip_bits_long(gb, 8 * cnt - 4);
2320 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2321 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2322 skip_bits_long(gb, 8 * cnt - 4);
2324 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2325 ac->oc[1].m4ac.sbr = 1;
2326 ac->oc[1].m4ac.ps = 1;
2327 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2328 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2329 ac->oc[1].status, 1);
2331 ac->oc[1].m4ac.sbr = 1;
2332 ac->avctx->profile = FF_PROFILE_AAC_HE;
2334 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2336 case EXT_DYNAMIC_RANGE:
2337 res = decode_dynamic_range(&ac->che_drc, gb);
2340 decode_fill(ac, gb, 8 * cnt - 4);
2343 case EXT_DATA_ELEMENT:
2345 skip_bits_long(gb, 8 * cnt - 4);
2352 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2354 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2355 * @param coef spectral coefficients
2357 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2358 IndividualChannelStream *ics, int decode)
2360 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2362 int bottom, top, order, start, end, size, inc;
2363 float lpc[TNS_MAX_ORDER];
2364 float tmp[TNS_MAX_ORDER+1];
2366 for (w = 0; w < ics->num_windows; w++) {
2367 bottom = ics->num_swb;
2368 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2370 bottom = FFMAX(0, top - tns->length[w][filt]);
2371 order = tns->order[w][filt];
2376 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2378 start = ics->swb_offset[FFMIN(bottom, mmm)];
2379 end = ics->swb_offset[FFMIN( top, mmm)];
2380 if ((size = end - start) <= 0)
2382 if (tns->direction[w][filt]) {
2392 for (m = 0; m < size; m++, start += inc)
2393 for (i = 1; i <= FFMIN(m, order); i++)
2394 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2397 for (m = 0; m < size; m++, start += inc) {
2398 tmp[0] = coef[start];
2399 for (i = 1; i <= FFMIN(m, order); i++)
2400 coef[start] += tmp[i] * lpc[i - 1];
2401 for (i = order; i > 0; i--)
2402 tmp[i] = tmp[i - 1];
2410 * Apply windowing and MDCT to obtain the spectral
2411 * coefficient from the predicted sample by LTP.
2413 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2414 float *in, IndividualChannelStream *ics)
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;
2418 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2419 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2421 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2422 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2424 memset(in, 0, 448 * sizeof(float));
2425 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2427 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2428 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2430 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2431 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2433 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2437 * Apply the long term prediction
2439 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2441 const LongTermPrediction *ltp = &sce->ics.ltp;
2442 const uint16_t *offsets = sce->ics.swb_offset;
2445 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2446 float *predTime = sce->ret;
2447 float *predFreq = ac->buf_mdct;
2448 int16_t num_samples = 2048;
2450 if (ltp->lag < 1024)
2451 num_samples = ltp->lag + 1024;
2452 for (i = 0; i < num_samples; i++)
2453 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2454 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2456 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2458 if (sce->tns.present)
2459 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2461 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2463 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2464 sce->coeffs[i] += predFreq[i];
2469 * Update the LTP buffer for next frame
2471 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2473 IndividualChannelStream *ics = &sce->ics;
2474 float *saved = sce->saved;
2475 float *saved_ltp = sce->coeffs;
2476 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2477 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2480 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2481 memcpy(saved_ltp, saved, 512 * sizeof(float));
2482 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2483 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2484 for (i = 0; i < 64; i++)
2485 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2486 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2487 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2488 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2489 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2490 for (i = 0; i < 64; i++)
2491 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2492 } else { // LONG_STOP or ONLY_LONG
2493 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2494 for (i = 0; i < 512; i++)
2495 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2498 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2499 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2500 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2504 * Conduct IMDCT and windowing.
2506 static void imdct_and_windowing(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 *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2513 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2514 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2515 float *buf = ac->buf_mdct;
2516 float *temp = ac->temp;
2520 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2521 for (i = 0; i < 1024; i += 128)
2522 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2524 ac->mdct.imdct_half(&ac->mdct, buf, in);
2526 /* window overlapping
2527 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2528 * and long to short transitions are considered to be short to short
2529 * transitions. This leaves just two cases (long to long and short to short)
2530 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2532 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2533 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2534 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2536 memcpy( out, saved, 448 * sizeof(float));
2538 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2539 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2540 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2541 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2542 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2543 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2544 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2546 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2547 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2552 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2553 memcpy( saved, temp + 64, 64 * sizeof(float));
2554 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2555 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2556 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2557 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2558 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2559 memcpy( saved, buf + 512, 448 * sizeof(float));
2560 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2561 } else { // LONG_STOP or ONLY_LONG
2562 memcpy( saved, buf + 512, 512 * sizeof(float));
2566 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2568 IndividualChannelStream *ics = &sce->ics;
2569 float *in = sce->coeffs;
2570 float *out = sce->ret;
2571 float *saved = sce->saved;
2572 float *buf = ac->buf_mdct;
2575 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2577 // window overlapping
2578 if (ics->use_kb_window[1]) {
2579 // AAC LD uses a low overlap sine window instead of a KBD window
2580 memcpy(out, saved, 192 * sizeof(float));
2581 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2582 memcpy( out + 320, buf + 64, 192 * sizeof(float));
2584 ac->fdsp->vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2588 memcpy(saved, buf + 256, 256 * sizeof(float));
2591 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2593 float *in = sce->coeffs;
2594 float *out = sce->ret;
2595 float *saved = sce->saved;
2596 float *buf = ac->buf_mdct;
2598 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2599 const int n2 = n >> 1;
2600 const int n4 = n >> 2;
2601 const float *const window = n == 480 ? ff_aac_eld_window_480 :
2602 ff_aac_eld_window_512;
2604 // Inverse transform, mapped to the conventional IMDCT by
2605 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2606 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2607 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2608 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2609 for (i = 0; i < n2; i+=2) {
2611 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2612 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2615 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2617 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2618 for (i = 0; i < n; i+=2) {
2621 // Like with the regular IMDCT at this point we still have the middle half
2622 // of a transform but with even symmetry on the left and odd symmetry on
2625 // window overlapping
2626 // The spec says to use samples [0..511] but the reference decoder uses
2627 // samples [128..639].
2628 for (i = n4; i < n2; i ++) {
2629 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2630 saved[ i + n2] * window[i + n - n4] +
2631 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2632 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2634 for (i = 0; i < n2; i ++) {
2635 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2636 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2637 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2638 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2640 for (i = 0; i < n4; i ++) {
2641 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2642 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2643 -saved[ n + n2 + i] * window[i + 3*n - n4];
2647 memmove(saved + n, saved, 2 * n * sizeof(float));
2648 memcpy( saved, buf, n * sizeof(float));
2652 * Apply dependent channel coupling (applied before IMDCT).
2654 * @param index index into coupling gain array
2656 static void apply_dependent_coupling(AACContext *ac,
2657 SingleChannelElement *target,
2658 ChannelElement *cce, int index)
2660 IndividualChannelStream *ics = &cce->ch[0].ics;
2661 const uint16_t *offsets = ics->swb_offset;
2662 float *dest = target->coeffs;
2663 const float *src = cce->ch[0].coeffs;
2664 int g, i, group, k, idx = 0;
2665 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2666 av_log(ac->avctx, AV_LOG_ERROR,
2667 "Dependent coupling is not supported together with LTP\n");
2670 for (g = 0; g < ics->num_window_groups; g++) {
2671 for (i = 0; i < ics->max_sfb; i++, idx++) {
2672 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2673 const float gain = cce->coup.gain[index][idx];
2674 for (group = 0; group < ics->group_len[g]; group++) {
2675 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2677 dest[group * 128 + k] += gain * src[group * 128 + k];
2682 dest += ics->group_len[g] * 128;
2683 src += ics->group_len[g] * 128;
2688 * Apply independent channel coupling (applied after IMDCT).
2690 * @param index index into coupling gain array
2692 static void apply_independent_coupling(AACContext *ac,
2693 SingleChannelElement *target,
2694 ChannelElement *cce, int index)
2697 const float gain = cce->coup.gain[index][0];
2698 const float *src = cce->ch[0].ret;
2699 float *dest = target->ret;
2700 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2702 for (i = 0; i < len; i++)
2703 dest[i] += gain * src[i];
2707 * channel coupling transformation interface
2709 * @param apply_coupling_method pointer to (in)dependent coupling function
2711 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2712 enum RawDataBlockType type, int elem_id,
2713 enum CouplingPoint coupling_point,
2714 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2718 for (i = 0; i < MAX_ELEM_ID; i++) {
2719 ChannelElement *cce = ac->che[TYPE_CCE][i];
2722 if (cce && cce->coup.coupling_point == coupling_point) {
2723 ChannelCoupling *coup = &cce->coup;
2725 for (c = 0; c <= coup->num_coupled; c++) {
2726 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2727 if (coup->ch_select[c] != 1) {
2728 apply_coupling_method(ac, &cc->ch[0], cce, index);
2729 if (coup->ch_select[c] != 0)
2732 if (coup->ch_select[c] != 2)
2733 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2735 index += 1 + (coup->ch_select[c] == 3);
2742 * Convert spectral data to float samples, applying all supported tools as appropriate.
2744 static void spectral_to_sample(AACContext *ac)
2747 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2748 switch (ac->oc[1].m4ac.object_type) {
2750 imdct_and_window = imdct_and_windowing_ld;
2752 case AOT_ER_AAC_ELD:
2753 imdct_and_window = imdct_and_windowing_eld;
2756 imdct_and_window = ac->imdct_and_windowing;
2758 for (type = 3; type >= 0; type--) {
2759 for (i = 0; i < MAX_ELEM_ID; i++) {
2760 ChannelElement *che = ac->che[type][i];
2761 if (che && che->present) {
2762 if (type <= TYPE_CPE)
2763 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2764 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2765 if (che->ch[0].ics.predictor_present) {
2766 if (che->ch[0].ics.ltp.present)
2767 ac->apply_ltp(ac, &che->ch[0]);
2768 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2769 ac->apply_ltp(ac, &che->ch[1]);
2772 if (che->ch[0].tns.present)
2773 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2774 if (che->ch[1].tns.present)
2775 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2776 if (type <= TYPE_CPE)
2777 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2778 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2779 imdct_and_window(ac, &che->ch[0]);
2780 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2781 ac->update_ltp(ac, &che->ch[0]);
2782 if (type == TYPE_CPE) {
2783 imdct_and_window(ac, &che->ch[1]);
2784 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2785 ac->update_ltp(ac, &che->ch[1]);
2787 if (ac->oc[1].m4ac.sbr > 0) {
2788 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2791 if (type <= TYPE_CCE)
2792 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2795 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2801 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2804 AACADTSHeaderInfo hdr_info;
2805 uint8_t layout_map[MAX_ELEM_ID*4][3];
2806 int layout_map_tags, ret;
2808 size = avpriv_aac_parse_header(gb, &hdr_info);
2810 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2811 // This is 2 for "VLB " audio in NSV files.
2812 // See samples/nsv/vlb_audio.
2813 avpriv_report_missing_feature(ac->avctx,
2814 "More than one AAC RDB per ADTS frame");
2815 ac->warned_num_aac_frames = 1;
2817 push_output_configuration(ac);
2818 if (hdr_info.chan_config) {
2819 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2820 if ((ret = set_default_channel_config(ac->avctx,
2823 hdr_info.chan_config)) < 0)
2825 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2826 FFMAX(ac->oc[1].status,
2827 OC_TRIAL_FRAME), 0)) < 0)
2830 ac->oc[1].m4ac.chan_config = 0;
2832 * dual mono frames in Japanese DTV can have chan_config 0
2833 * WITHOUT specifying PCE.
2834 * thus, set dual mono as default.
2836 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2837 layout_map_tags = 2;
2838 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2839 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2840 layout_map[0][1] = 0;
2841 layout_map[1][1] = 1;
2842 if (output_configure(ac, layout_map, layout_map_tags,
2847 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2848 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2849 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2850 ac->oc[1].m4ac.frame_length_short = 0;
2851 if (ac->oc[0].status != OC_LOCKED ||
2852 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2853 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2854 ac->oc[1].m4ac.sbr = -1;
2855 ac->oc[1].m4ac.ps = -1;
2857 if (!hdr_info.crc_absent)
2863 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2864 int *got_frame_ptr, GetBitContext *gb)
2866 AACContext *ac = avctx->priv_data;
2867 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2868 ChannelElement *che;
2870 int samples = m4ac->frame_length_short ? 960 : 1024;
2871 int chan_config = m4ac->chan_config;
2872 int aot = m4ac->object_type;
2874 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2879 if ((err = frame_configure_elements(avctx)) < 0)
2882 // The FF_PROFILE_AAC_* defines are all object_type - 1
2883 // This may lead to an undefined profile being signaled
2884 ac->avctx->profile = aot - 1;
2886 ac->tags_mapped = 0;
2888 if (chan_config < 0 || chan_config >= 8) {
2889 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2891 return AVERROR_INVALIDDATA;
2893 for (i = 0; i < tags_per_config[chan_config]; i++) {
2894 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2895 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2896 if (!(che=get_che(ac, elem_type, elem_id))) {
2897 av_log(ac->avctx, AV_LOG_ERROR,
2898 "channel element %d.%d is not allocated\n",
2899 elem_type, elem_id);
2900 return AVERROR_INVALIDDATA;
2903 if (aot != AOT_ER_AAC_ELD)
2905 switch (elem_type) {
2907 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2910 err = decode_cpe(ac, gb, che);
2913 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2920 spectral_to_sample(ac);
2922 ac->frame->nb_samples = samples;
2923 ac->frame->sample_rate = avctx->sample_rate;
2926 skip_bits_long(gb, get_bits_left(gb));
2930 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2931 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2933 AACContext *ac = avctx->priv_data;
2934 ChannelElement *che = NULL, *che_prev = NULL;
2935 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2937 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2938 int is_dmono, sce_count = 0;
2942 if (show_bits(gb, 12) == 0xfff) {
2943 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2944 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2947 if (ac->oc[1].m4ac.sampling_index > 12) {
2948 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2949 err = AVERROR_INVALIDDATA;
2954 if ((err = frame_configure_elements(avctx)) < 0)
2957 // The FF_PROFILE_AAC_* defines are all object_type - 1
2958 // This may lead to an undefined profile being signaled
2959 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2961 ac->tags_mapped = 0;
2963 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2964 elem_id = get_bits(gb, 4);
2966 if (avctx->debug & FF_DEBUG_STARTCODE)
2967 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2969 if (elem_type < TYPE_DSE) {
2970 if (!(che=get_che(ac, elem_type, elem_id))) {
2971 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2972 elem_type, elem_id);
2973 err = AVERROR_INVALIDDATA;
2980 switch (elem_type) {
2983 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2989 err = decode_cpe(ac, gb, che);
2994 err = decode_cce(ac, gb, che);
2998 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3003 err = skip_data_stream_element(ac, gb);
3007 uint8_t layout_map[MAX_ELEM_ID*4][3];
3009 push_output_configuration(ac);
3010 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
3016 av_log(avctx, AV_LOG_ERROR,
3017 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3019 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3021 ac->oc[1].m4ac.chan_config = 0;
3029 elem_id += get_bits(gb, 8) - 1;
3030 if (get_bits_left(gb) < 8 * elem_id) {
3031 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3032 err = AVERROR_INVALIDDATA;
3036 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3037 err = 0; /* FIXME */
3041 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3046 elem_type_prev = elem_type;
3051 if (get_bits_left(gb) < 3) {
3052 av_log(avctx, AV_LOG_ERROR, overread_err);
3053 err = AVERROR_INVALIDDATA;
3058 spectral_to_sample(ac);
3060 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3061 samples <<= multiplier;
3063 if (ac->oc[1].status && audio_found) {
3064 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3065 avctx->frame_size = samples;
3066 ac->oc[1].status = OC_LOCKED;
3071 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3072 if (side && side_size>=4)
3073 AV_WL32(side, 2*AV_RL32(side));
3076 *got_frame_ptr = !!samples;
3078 ac->frame->nb_samples = samples;
3079 ac->frame->sample_rate = avctx->sample_rate;
3081 av_frame_unref(ac->frame);
3082 *got_frame_ptr = !!samples;
3084 /* for dual-mono audio (SCE + SCE) */
3085 is_dmono = ac->dmono_mode && sce_count == 2 &&
3086 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3088 if (ac->dmono_mode == 1)
3089 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3090 else if (ac->dmono_mode == 2)
3091 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3096 pop_output_configuration(ac);
3100 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3101 int *got_frame_ptr, AVPacket *avpkt)
3103 AACContext *ac = avctx->priv_data;
3104 const uint8_t *buf = avpkt->data;
3105 int buf_size = avpkt->size;
3110 int new_extradata_size;
3111 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3112 AV_PKT_DATA_NEW_EXTRADATA,
3113 &new_extradata_size);
3114 int jp_dualmono_size;
3115 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3116 AV_PKT_DATA_JP_DUALMONO,
3119 if (new_extradata && 0) {
3120 av_free(avctx->extradata);
3121 avctx->extradata = av_mallocz(new_extradata_size +
3122 FF_INPUT_BUFFER_PADDING_SIZE);
3123 if (!avctx->extradata)
3124 return AVERROR(ENOMEM);
3125 avctx->extradata_size = new_extradata_size;
3126 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3127 push_output_configuration(ac);
3128 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3130 avctx->extradata_size*8, 1) < 0) {
3131 pop_output_configuration(ac);
3132 return AVERROR_INVALIDDATA;
3137 if (jp_dualmono && jp_dualmono_size > 0)
3138 ac->dmono_mode = 1 + *jp_dualmono;
3139 if (ac->force_dmono_mode >= 0)
3140 ac->dmono_mode = ac->force_dmono_mode;
3142 if (INT_MAX / 8 <= buf_size)
3143 return AVERROR_INVALIDDATA;
3145 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3148 switch (ac->oc[1].m4ac.object_type) {
3150 case AOT_ER_AAC_LTP:
3152 case AOT_ER_AAC_ELD:
3153 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3156 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3161 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3162 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3163 if (buf[buf_offset])
3166 return buf_size > buf_offset ? buf_consumed : buf_size;
3169 static av_cold int aac_decode_close(AVCodecContext *avctx)
3171 AACContext *ac = avctx->priv_data;
3174 for (i = 0; i < MAX_ELEM_ID; i++) {
3175 for (type = 0; type < 4; type++) {
3176 if (ac->che[type][i])
3177 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3178 av_freep(&ac->che[type][i]);
3182 ff_mdct_end(&ac->mdct);
3183 ff_mdct_end(&ac->mdct_small);
3184 ff_mdct_end(&ac->mdct_ld);
3185 ff_mdct_end(&ac->mdct_ltp);
3186 ff_imdct15_uninit(&ac->mdct480);
3187 av_freep(&ac->fdsp);
3192 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3194 struct LATMContext {
3195 AACContext aac_ctx; ///< containing AACContext
3196 int initialized; ///< initialized after a valid extradata was seen
3199 int audio_mux_version_A; ///< LATM syntax version
3200 int frame_length_type; ///< 0/1 variable/fixed frame length
3201 int frame_length; ///< frame length for fixed frame length
3204 static inline uint32_t latm_get_value(GetBitContext *b)
3206 int length = get_bits(b, 2);
3208 return get_bits_long(b, (length+1)*8);
3211 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3212 GetBitContext *gb, int asclen)
3214 AACContext *ac = &latmctx->aac_ctx;
3215 AVCodecContext *avctx = ac->avctx;
3216 MPEG4AudioConfig m4ac = { 0 };
3217 int config_start_bit = get_bits_count(gb);
3218 int sync_extension = 0;
3219 int bits_consumed, esize;
3223 asclen = FFMIN(asclen, get_bits_left(gb));
3225 asclen = get_bits_left(gb);
3227 if (config_start_bit % 8) {
3228 avpriv_request_sample(latmctx->aac_ctx.avctx,
3229 "Non-byte-aligned audio-specific config");
3230 return AVERROR_PATCHWELCOME;
3233 return AVERROR_INVALIDDATA;
3234 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3235 gb->buffer + (config_start_bit / 8),
3236 asclen, sync_extension);
3238 if (bits_consumed < 0)
3239 return AVERROR_INVALIDDATA;
3241 if (!latmctx->initialized ||
3242 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3243 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3245 if(latmctx->initialized) {
3246 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3248 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3250 latmctx->initialized = 0;
3252 esize = (bits_consumed+7) / 8;
3254 if (avctx->extradata_size < esize) {
3255 av_free(avctx->extradata);
3256 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3257 if (!avctx->extradata)
3258 return AVERROR(ENOMEM);
3261 avctx->extradata_size = esize;
3262 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3263 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3265 skip_bits_long(gb, bits_consumed);
3267 return bits_consumed;
3270 static int read_stream_mux_config(struct LATMContext *latmctx,
3273 int ret, audio_mux_version = get_bits(gb, 1);
3275 latmctx->audio_mux_version_A = 0;
3276 if (audio_mux_version)
3277 latmctx->audio_mux_version_A = get_bits(gb, 1);
3279 if (!latmctx->audio_mux_version_A) {
3281 if (audio_mux_version)
3282 latm_get_value(gb); // taraFullness
3284 skip_bits(gb, 1); // allStreamSameTimeFraming
3285 skip_bits(gb, 6); // numSubFrames
3287 if (get_bits(gb, 4)) { // numPrograms
3288 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3289 return AVERROR_PATCHWELCOME;
3292 // for each program (which there is only one in DVB)
3294 // for each layer (which there is only one in DVB)
3295 if (get_bits(gb, 3)) { // numLayer
3296 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3297 return AVERROR_PATCHWELCOME;
3300 // for all but first stream: use_same_config = get_bits(gb, 1);
3301 if (!audio_mux_version) {
3302 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3305 int ascLen = latm_get_value(gb);
3306 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3309 skip_bits_long(gb, ascLen);
3312 latmctx->frame_length_type = get_bits(gb, 3);
3313 switch (latmctx->frame_length_type) {
3315 skip_bits(gb, 8); // latmBufferFullness
3318 latmctx->frame_length = get_bits(gb, 9);
3323 skip_bits(gb, 6); // CELP frame length table index
3327 skip_bits(gb, 1); // HVXC frame length table index
3331 if (get_bits(gb, 1)) { // other data
3332 if (audio_mux_version) {
3333 latm_get_value(gb); // other_data_bits
3337 esc = get_bits(gb, 1);
3343 if (get_bits(gb, 1)) // crc present
3344 skip_bits(gb, 8); // config_crc
3350 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3354 if (ctx->frame_length_type == 0) {
3355 int mux_slot_length = 0;
3357 tmp = get_bits(gb, 8);
3358 mux_slot_length += tmp;
3359 } while (tmp == 255);
3360 return mux_slot_length;
3361 } else if (ctx->frame_length_type == 1) {
3362 return ctx->frame_length;
3363 } else if (ctx->frame_length_type == 3 ||
3364 ctx->frame_length_type == 5 ||
3365 ctx->frame_length_type == 7) {
3366 skip_bits(gb, 2); // mux_slot_length_coded
3371 static int read_audio_mux_element(struct LATMContext *latmctx,
3375 uint8_t use_same_mux = get_bits(gb, 1);
3376 if (!use_same_mux) {
3377 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3379 } else if (!latmctx->aac_ctx.avctx->extradata) {
3380 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3381 "no decoder config found\n");
3382 return AVERROR(EAGAIN);
3384 if (latmctx->audio_mux_version_A == 0) {
3385 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3386 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3387 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3388 return AVERROR_INVALIDDATA;
3389 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3390 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3391 "frame length mismatch %d << %d\n",
3392 mux_slot_length_bytes * 8, get_bits_left(gb));
3393 return AVERROR_INVALIDDATA;
3400 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3401 int *got_frame_ptr, AVPacket *avpkt)
3403 struct LATMContext *latmctx = avctx->priv_data;
3407 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3410 // check for LOAS sync word
3411 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3412 return AVERROR_INVALIDDATA;
3414 muxlength = get_bits(&gb, 13) + 3;
3415 // not enough data, the parser should have sorted this out
3416 if (muxlength > avpkt->size)
3417 return AVERROR_INVALIDDATA;
3419 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3422 if (!latmctx->initialized) {
3423 if (!avctx->extradata) {
3427 push_output_configuration(&latmctx->aac_ctx);
3428 if ((err = decode_audio_specific_config(
3429 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3430 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3431 pop_output_configuration(&latmctx->aac_ctx);
3434 latmctx->initialized = 1;
3438 if (show_bits(&gb, 12) == 0xfff) {
3439 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3440 "ADTS header detected, probably as result of configuration "
3442 return AVERROR_INVALIDDATA;
3445 switch (latmctx->aac_ctx.oc[1].m4ac.object_type) {
3447 case AOT_ER_AAC_LTP:
3449 case AOT_ER_AAC_ELD:
3450 err = aac_decode_er_frame(avctx, out, got_frame_ptr, &gb);
3453 err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt);
3461 static av_cold int latm_decode_init(AVCodecContext *avctx)
3463 struct LATMContext *latmctx = avctx->priv_data;
3464 int ret = aac_decode_init(avctx);
3466 if (avctx->extradata_size > 0)
3467 latmctx->initialized = !ret;
3472 static void aacdec_init(AACContext *c)
3474 c->imdct_and_windowing = imdct_and_windowing;
3475 c->apply_ltp = apply_ltp;
3476 c->apply_tns = apply_tns;
3477 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3478 c->update_ltp = update_ltp;
3481 ff_aacdec_init_mips(c);
3484 * AVOptions for Japanese DTV specific extensions (ADTS only)
3486 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3487 static const AVOption options[] = {
3488 {"dual_mono_mode", "Select the channel to decode for dual mono",
3489 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3490 AACDEC_FLAGS, "dual_mono_mode"},
3492 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3493 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3494 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3495 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3500 static const AVClass aac_decoder_class = {
3501 .class_name = "AAC decoder",
3502 .item_name = av_default_item_name,
3504 .version = LIBAVUTIL_VERSION_INT,
3507 static const AVProfile profiles[] = {
3508 { FF_PROFILE_AAC_MAIN, "Main" },
3509 { FF_PROFILE_AAC_LOW, "LC" },
3510 { FF_PROFILE_AAC_SSR, "SSR" },
3511 { FF_PROFILE_AAC_LTP, "LTP" },
3512 { FF_PROFILE_AAC_HE, "HE-AAC" },
3513 { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3514 { FF_PROFILE_AAC_LD, "LD" },
3515 { FF_PROFILE_AAC_ELD, "ELD" },
3516 { FF_PROFILE_UNKNOWN },
3519 AVCodec ff_aac_decoder = {
3521 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3522 .type = AVMEDIA_TYPE_AUDIO,
3523 .id = AV_CODEC_ID_AAC,
3524 .priv_data_size = sizeof(AACContext),
3525 .init = aac_decode_init,
3526 .close = aac_decode_close,
3527 .decode = aac_decode_frame,
3528 .sample_fmts = (const enum AVSampleFormat[]) {
3529 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3531 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3532 .channel_layouts = aac_channel_layout,
3534 .priv_class = &aac_decoder_class,
3535 .profiles = profiles,
3539 Note: This decoder filter is intended to decode LATM streams transferred
3540 in MPEG transport streams which only contain one program.
3541 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3543 AVCodec ff_aac_latm_decoder = {
3545 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3546 .type = AVMEDIA_TYPE_AUDIO,
3547 .id = AV_CODEC_ID_AAC_LATM,
3548 .priv_data_size = sizeof(struct LATMContext),
3549 .init = latm_decode_init,
3550 .close = aac_decode_close,
3551 .decode = latm_decode_frame,
3552 .sample_fmts = (const enum AVSampleFormat[]) {
3553 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3555 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3556 .channel_layouts = aac_channel_layout,
3558 .profiles = profiles,