3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
5 * Copyright (c) 2008-2013 Alex Converse <alex.converse@gmail.com>
8 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
9 * Copyright (c) 2010 Janne Grunau <janne-libav@jannau.net>
11 * This file is part of FFmpeg.
13 * FFmpeg is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU Lesser General Public
15 * License as published by the Free Software Foundation; either
16 * version 2.1 of the License, or (at your option) any later version.
18 * FFmpeg is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
21 * Lesser General Public License for more details.
23 * You should have received a copy of the GNU Lesser General Public
24 * License along with FFmpeg; if not, write to the Free Software
25 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
31 * @author Oded Shimon ( ods15 ods15 dyndns org )
32 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
39 * N (code in SoC repo) gain control
41 * Y window shapes - standard
42 * N window shapes - Low Delay
43 * Y filterbank - standard
44 * N (code in SoC repo) filterbank - Scalable Sample Rate
45 * Y Temporal Noise Shaping
46 * Y Long Term Prediction
49 * Y frequency domain prediction
50 * Y Perceptual Noise Substitution
52 * N Scalable Inverse AAC Quantization
53 * N Frequency Selective Switch
55 * Y quantization & coding - AAC
56 * N quantization & coding - TwinVQ
57 * N quantization & coding - BSAC
58 * N AAC Error Resilience tools
59 * N Error Resilience payload syntax
60 * N Error Protection tool
62 * N Silence Compression
65 * N Structured Audio tools
66 * N Structured Audio Sample Bank Format
68 * N Harmonic and Individual Lines plus Noise
69 * N Text-To-Speech Interface
70 * Y Spectral Band Replication
71 * Y (not in this code) Layer-1
72 * Y (not in this code) Layer-2
73 * Y (not in this code) Layer-3
74 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
76 * N Direct Stream Transfer
77 * Y Enhanced AAC Low Delay (ER AAC ELD)
79 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
80 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
84 #include "libavutil/float_dsp.h"
85 #include "libavutil/opt.h"
90 #include "fmtconvert.h"
97 #include "aacdectab.h"
98 #include "cbrt_tablegen.h"
101 #include "mpeg4audio.h"
102 #include "aacadtsdec.h"
103 #include "libavutil/intfloat.h"
112 # include "arm/aac.h"
114 # include "mips/aacdec_mips.h"
117 static VLC vlc_scalefactors;
118 static VLC vlc_spectral[11];
120 static int output_configure(AACContext *ac,
121 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
122 enum OCStatus oc_type, int get_new_frame);
124 #define overread_err "Input buffer exhausted before END element found\n"
126 static int count_channels(uint8_t (*layout)[3], int tags)
129 for (i = 0; i < tags; i++) {
130 int syn_ele = layout[i][0];
131 int pos = layout[i][2];
132 sum += (1 + (syn_ele == TYPE_CPE)) *
133 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
139 * Check for the channel element in the current channel position configuration.
140 * If it exists, make sure the appropriate element is allocated and map the
141 * channel order to match the internal FFmpeg channel layout.
143 * @param che_pos current channel position configuration
144 * @param type channel element type
145 * @param id channel element id
146 * @param channels count of the number of channels in the configuration
148 * @return Returns error status. 0 - OK, !0 - error
150 static av_cold int che_configure(AACContext *ac,
151 enum ChannelPosition che_pos,
152 int type, int id, int *channels)
154 if (*channels >= MAX_CHANNELS)
155 return AVERROR_INVALIDDATA;
157 if (!ac->che[type][id]) {
158 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
159 return AVERROR(ENOMEM);
160 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
162 if (type != TYPE_CCE) {
163 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
164 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
165 return AVERROR_INVALIDDATA;
167 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
168 if (type == TYPE_CPE ||
169 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
170 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
174 if (ac->che[type][id])
175 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
176 av_freep(&ac->che[type][id]);
181 static int frame_configure_elements(AVCodecContext *avctx)
183 AACContext *ac = avctx->priv_data;
184 int type, id, ch, ret;
186 /* set channel pointers to internal buffers by default */
187 for (type = 0; type < 4; type++) {
188 for (id = 0; id < MAX_ELEM_ID; id++) {
189 ChannelElement *che = ac->che[type][id];
191 che->ch[0].ret = che->ch[0].ret_buf;
192 che->ch[1].ret = che->ch[1].ret_buf;
197 /* get output buffer */
198 av_frame_unref(ac->frame);
199 if (!avctx->channels)
202 ac->frame->nb_samples = 2048;
203 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
206 /* map output channel pointers to AVFrame data */
207 for (ch = 0; ch < avctx->channels; ch++) {
208 if (ac->output_element[ch])
209 ac->output_element[ch]->ret = (float *)ac->frame->extended_data[ch];
215 struct elem_to_channel {
216 uint64_t av_position;
219 uint8_t aac_position;
222 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
223 uint8_t (*layout_map)[3], int offset, uint64_t left,
224 uint64_t right, int pos)
226 if (layout_map[offset][0] == TYPE_CPE) {
227 e2c_vec[offset] = (struct elem_to_channel) {
228 .av_position = left | right,
230 .elem_id = layout_map[offset][1],
235 e2c_vec[offset] = (struct elem_to_channel) {
238 .elem_id = layout_map[offset][1],
241 e2c_vec[offset + 1] = (struct elem_to_channel) {
242 .av_position = right,
244 .elem_id = layout_map[offset + 1][1],
251 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
254 int num_pos_channels = 0;
258 for (i = *current; i < tags; i++) {
259 if (layout_map[i][2] != pos)
261 if (layout_map[i][0] == TYPE_CPE) {
263 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
269 num_pos_channels += 2;
277 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
280 return num_pos_channels;
283 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
285 int i, n, total_non_cc_elements;
286 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
287 int num_front_channels, num_side_channels, num_back_channels;
290 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
295 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
296 if (num_front_channels < 0)
299 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
300 if (num_side_channels < 0)
303 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
304 if (num_back_channels < 0)
308 if (num_front_channels & 1) {
309 e2c_vec[i] = (struct elem_to_channel) {
310 .av_position = AV_CH_FRONT_CENTER,
312 .elem_id = layout_map[i][1],
313 .aac_position = AAC_CHANNEL_FRONT
316 num_front_channels--;
318 if (num_front_channels >= 4) {
319 i += assign_pair(e2c_vec, layout_map, i,
320 AV_CH_FRONT_LEFT_OF_CENTER,
321 AV_CH_FRONT_RIGHT_OF_CENTER,
323 num_front_channels -= 2;
325 if (num_front_channels >= 2) {
326 i += assign_pair(e2c_vec, layout_map, i,
330 num_front_channels -= 2;
332 while (num_front_channels >= 2) {
333 i += assign_pair(e2c_vec, layout_map, i,
337 num_front_channels -= 2;
340 if (num_side_channels >= 2) {
341 i += assign_pair(e2c_vec, layout_map, i,
345 num_side_channels -= 2;
347 while (num_side_channels >= 2) {
348 i += assign_pair(e2c_vec, layout_map, i,
352 num_side_channels -= 2;
355 while (num_back_channels >= 4) {
356 i += assign_pair(e2c_vec, layout_map, i,
360 num_back_channels -= 2;
362 if (num_back_channels >= 2) {
363 i += assign_pair(e2c_vec, layout_map, i,
367 num_back_channels -= 2;
369 if (num_back_channels) {
370 e2c_vec[i] = (struct elem_to_channel) {
371 .av_position = AV_CH_BACK_CENTER,
373 .elem_id = layout_map[i][1],
374 .aac_position = AAC_CHANNEL_BACK
380 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
381 e2c_vec[i] = (struct elem_to_channel) {
382 .av_position = AV_CH_LOW_FREQUENCY,
384 .elem_id = layout_map[i][1],
385 .aac_position = AAC_CHANNEL_LFE
389 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
390 e2c_vec[i] = (struct elem_to_channel) {
391 .av_position = UINT64_MAX,
393 .elem_id = layout_map[i][1],
394 .aac_position = AAC_CHANNEL_LFE
399 // Must choose a stable sort
400 total_non_cc_elements = n = i;
403 for (i = 1; i < n; i++)
404 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
405 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
412 for (i = 0; i < total_non_cc_elements; i++) {
413 layout_map[i][0] = e2c_vec[i].syn_ele;
414 layout_map[i][1] = e2c_vec[i].elem_id;
415 layout_map[i][2] = e2c_vec[i].aac_position;
416 if (e2c_vec[i].av_position != UINT64_MAX) {
417 layout |= e2c_vec[i].av_position;
425 * Save current output configuration if and only if it has been locked.
427 static void push_output_configuration(AACContext *ac) {
428 if (ac->oc[1].status == OC_LOCKED) {
429 ac->oc[0] = ac->oc[1];
431 ac->oc[1].status = OC_NONE;
435 * Restore the previous output configuration if and only if the current
436 * configuration is unlocked.
438 static void pop_output_configuration(AACContext *ac) {
439 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
440 ac->oc[1] = ac->oc[0];
441 ac->avctx->channels = ac->oc[1].channels;
442 ac->avctx->channel_layout = ac->oc[1].channel_layout;
443 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
444 ac->oc[1].status, 0);
449 * Configure output channel order based on the current program
450 * configuration element.
452 * @return Returns error status. 0 - OK, !0 - error
454 static int output_configure(AACContext *ac,
455 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
456 enum OCStatus oc_type, int get_new_frame)
458 AVCodecContext *avctx = ac->avctx;
459 int i, channels = 0, ret;
462 if (ac->oc[1].layout_map != layout_map) {
463 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
464 ac->oc[1].layout_map_tags = tags;
467 // Try to sniff a reasonable channel order, otherwise output the
468 // channels in the order the PCE declared them.
469 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
470 layout = sniff_channel_order(layout_map, tags);
471 for (i = 0; i < tags; i++) {
472 int type = layout_map[i][0];
473 int id = layout_map[i][1];
474 int position = layout_map[i][2];
475 // Allocate or free elements depending on if they are in the
476 // current program configuration.
477 ret = che_configure(ac, position, type, id, &channels);
481 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
482 if (layout == AV_CH_FRONT_CENTER) {
483 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
489 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
490 if (layout) avctx->channel_layout = layout;
491 ac->oc[1].channel_layout = layout;
492 avctx->channels = ac->oc[1].channels = channels;
493 ac->oc[1].status = oc_type;
496 if ((ret = frame_configure_elements(ac->avctx)) < 0)
503 static void flush(AVCodecContext *avctx)
505 AACContext *ac= avctx->priv_data;
508 for (type = 3; type >= 0; type--) {
509 for (i = 0; i < MAX_ELEM_ID; i++) {
510 ChannelElement *che = ac->che[type][i];
512 for (j = 0; j <= 1; j++) {
513 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
521 * Set up channel positions based on a default channel configuration
522 * as specified in table 1.17.
524 * @return Returns error status. 0 - OK, !0 - error
526 static int set_default_channel_config(AVCodecContext *avctx,
527 uint8_t (*layout_map)[3],
531 if (channel_config < 1 || channel_config > 7) {
532 av_log(avctx, AV_LOG_ERROR,
533 "invalid default channel configuration (%d)\n",
535 return AVERROR_INVALIDDATA;
537 *tags = tags_per_config[channel_config];
538 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
539 *tags * sizeof(*layout_map));
542 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
543 * However, at least Nero AAC encoder encodes 7.1 streams using the default
544 * channel config 7, mapping the side channels of the original audio stream
545 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
546 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
547 * the incorrect streams as if they were correct (and as the encoder intended).
549 * As actual intended 7.1(wide) streams are very rare, default to assuming a
550 * 7.1 layout was intended.
552 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
553 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
554 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
555 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
556 layout_map[2][2] = AAC_CHANNEL_SIDE;
562 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
564 /* For PCE based channel configurations map the channels solely based
566 if (!ac->oc[1].m4ac.chan_config) {
567 return ac->tag_che_map[type][elem_id];
569 // Allow single CPE stereo files to be signalled with mono configuration.
570 if (!ac->tags_mapped && type == TYPE_CPE &&
571 ac->oc[1].m4ac.chan_config == 1) {
572 uint8_t layout_map[MAX_ELEM_ID*4][3];
574 push_output_configuration(ac);
576 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
578 if (set_default_channel_config(ac->avctx, layout_map,
579 &layout_map_tags, 2) < 0)
581 if (output_configure(ac, layout_map, layout_map_tags,
582 OC_TRIAL_FRAME, 1) < 0)
585 ac->oc[1].m4ac.chan_config = 2;
586 ac->oc[1].m4ac.ps = 0;
589 if (!ac->tags_mapped && type == TYPE_SCE &&
590 ac->oc[1].m4ac.chan_config == 2) {
591 uint8_t layout_map[MAX_ELEM_ID * 4][3];
593 push_output_configuration(ac);
595 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
597 if (set_default_channel_config(ac->avctx, layout_map,
598 &layout_map_tags, 1) < 0)
600 if (output_configure(ac, layout_map, layout_map_tags,
601 OC_TRIAL_FRAME, 1) < 0)
604 ac->oc[1].m4ac.chan_config = 1;
605 if (ac->oc[1].m4ac.sbr)
606 ac->oc[1].m4ac.ps = -1;
608 /* For indexed channel configurations map the channels solely based
610 switch (ac->oc[1].m4ac.chan_config) {
612 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
614 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
617 /* Some streams incorrectly code 5.1 audio as
618 * SCE[0] CPE[0] CPE[1] SCE[1]
620 * SCE[0] CPE[0] CPE[1] LFE[0].
621 * If we seem to have encountered such a stream, transfer
622 * the LFE[0] element to the SCE[1]'s mapping */
623 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
624 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
625 av_log(ac->avctx, AV_LOG_WARNING,
626 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
627 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
628 ac->warned_remapping_once++;
631 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
634 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
636 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
639 /* Some streams incorrectly code 4.0 audio as
640 * SCE[0] CPE[0] LFE[0]
642 * SCE[0] CPE[0] SCE[1].
643 * If we seem to have encountered such a stream, transfer
644 * the SCE[1] element to the LFE[0]'s mapping */
645 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
646 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
647 av_log(ac->avctx, AV_LOG_WARNING,
648 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
649 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
650 ac->warned_remapping_once++;
653 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
655 if (ac->tags_mapped == 2 &&
656 ac->oc[1].m4ac.chan_config == 4 &&
659 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
663 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
666 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
667 } else if (ac->oc[1].m4ac.chan_config == 2) {
671 if (!ac->tags_mapped && type == TYPE_SCE) {
673 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
681 * Decode an array of 4 bit element IDs, optionally interleaved with a
682 * stereo/mono switching bit.
684 * @param type speaker type/position for these channels
686 static void decode_channel_map(uint8_t layout_map[][3],
687 enum ChannelPosition type,
688 GetBitContext *gb, int n)
691 enum RawDataBlockType syn_ele;
693 case AAC_CHANNEL_FRONT:
694 case AAC_CHANNEL_BACK:
695 case AAC_CHANNEL_SIDE:
696 syn_ele = get_bits1(gb);
702 case AAC_CHANNEL_LFE:
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;
804 if (get_bits1(gb)) // dependsOnCoreCoder
805 skip_bits(gb, 14); // coreCoderDelay
806 extension_flag = get_bits1(gb);
808 if (m4ac->object_type == AOT_AAC_SCALABLE ||
809 m4ac->object_type == AOT_ER_AAC_SCALABLE)
810 skip_bits(gb, 3); // layerNr
812 if (channel_config == 0) {
813 skip_bits(gb, 4); // element_instance_tag
814 tags = decode_pce(avctx, m4ac, layout_map, gb);
818 if ((ret = set_default_channel_config(avctx, layout_map,
819 &tags, channel_config)))
823 if (count_channels(layout_map, tags) > 1) {
825 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
828 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
831 if (extension_flag) {
832 switch (m4ac->object_type) {
834 skip_bits(gb, 5); // numOfSubFrame
835 skip_bits(gb, 11); // layer_length
839 case AOT_ER_AAC_SCALABLE:
841 res_flags = get_bits(gb, 3);
843 avpriv_report_missing_feature(avctx,
844 "AAC data resilience (flags %x)",
846 return AVERROR_PATCHWELCOME;
850 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
852 switch (m4ac->object_type) {
855 case AOT_ER_AAC_SCALABLE:
857 ep_config = get_bits(gb, 2);
859 avpriv_report_missing_feature(avctx,
860 "epConfig %d", ep_config);
861 return AVERROR_PATCHWELCOME;
867 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
869 MPEG4AudioConfig *m4ac,
872 int ret, ep_config, res_flags;
873 uint8_t layout_map[MAX_ELEM_ID*4][3];
875 const int ELDEXT_TERM = 0;
880 if (get_bits1(gb)) { // frameLengthFlag
881 avpriv_request_sample(avctx, "960/120 MDCT window");
882 return AVERROR_PATCHWELCOME;
885 res_flags = get_bits(gb, 3);
887 avpriv_report_missing_feature(avctx,
888 "AAC data resilience (flags %x)",
890 return AVERROR_PATCHWELCOME;
893 if (get_bits1(gb)) { // ldSbrPresentFlag
894 avpriv_report_missing_feature(avctx,
896 return AVERROR_PATCHWELCOME;
899 while (get_bits(gb, 4) != ELDEXT_TERM) {
900 int len = get_bits(gb, 4);
902 len += get_bits(gb, 8);
904 len += get_bits(gb, 16);
905 if (get_bits_left(gb) < len * 8 + 4) {
906 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
907 return AVERROR_INVALIDDATA;
909 skip_bits_long(gb, 8 * len);
912 if ((ret = set_default_channel_config(avctx, layout_map,
913 &tags, channel_config)))
916 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
919 ep_config = get_bits(gb, 2);
921 avpriv_report_missing_feature(avctx,
922 "epConfig %d", ep_config);
923 return AVERROR_PATCHWELCOME;
929 * Decode audio specific configuration; reference: table 1.13.
931 * @param ac pointer to AACContext, may be null
932 * @param avctx pointer to AVCCodecContext, used for logging
933 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
934 * @param data pointer to buffer holding an audio specific config
935 * @param bit_size size of audio specific config or data in bits
936 * @param sync_extension look for an appended sync extension
938 * @return Returns error status or number of consumed bits. <0 - error
940 static int decode_audio_specific_config(AACContext *ac,
941 AVCodecContext *avctx,
942 MPEG4AudioConfig *m4ac,
943 const uint8_t *data, int bit_size,
949 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
950 for (i = 0; i < bit_size >> 3; i++)
951 av_dlog(avctx, "%02x ", data[i]);
952 av_dlog(avctx, "\n");
954 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
957 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
958 sync_extension)) < 0)
959 return AVERROR_INVALIDDATA;
960 if (m4ac->sampling_index > 12) {
961 av_log(avctx, AV_LOG_ERROR,
962 "invalid sampling rate index %d\n",
963 m4ac->sampling_index);
964 return AVERROR_INVALIDDATA;
966 if (m4ac->object_type == AOT_ER_AAC_LD &&
967 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
968 av_log(avctx, AV_LOG_ERROR,
969 "invalid low delay sampling rate index %d\n",
970 m4ac->sampling_index);
971 return AVERROR_INVALIDDATA;
974 skip_bits_long(&gb, i);
976 switch (m4ac->object_type) {
982 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
983 m4ac, m4ac->chan_config)) < 0)
987 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
988 m4ac, m4ac->chan_config)) < 0)
992 avpriv_report_missing_feature(avctx,
993 "Audio object type %s%d",
994 m4ac->sbr == 1 ? "SBR+" : "",
996 return AVERROR(ENOSYS);
1000 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
1001 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1002 m4ac->sample_rate, m4ac->sbr,
1005 return get_bits_count(&gb);
1009 * linear congruential pseudorandom number generator
1011 * @param previous_val pointer to the current state of the generator
1013 * @return Returns a 32-bit pseudorandom integer
1015 static av_always_inline int lcg_random(unsigned previous_val)
1017 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1021 static av_always_inline void reset_predict_state(PredictorState *ps)
1031 static void reset_all_predictors(PredictorState *ps)
1034 for (i = 0; i < MAX_PREDICTORS; i++)
1035 reset_predict_state(&ps[i]);
1038 static int sample_rate_idx (int rate)
1040 if (92017 <= rate) return 0;
1041 else if (75132 <= rate) return 1;
1042 else if (55426 <= rate) return 2;
1043 else if (46009 <= rate) return 3;
1044 else if (37566 <= rate) return 4;
1045 else if (27713 <= rate) return 5;
1046 else if (23004 <= rate) return 6;
1047 else if (18783 <= rate) return 7;
1048 else if (13856 <= rate) return 8;
1049 else if (11502 <= rate) return 9;
1050 else if (9391 <= rate) return 10;
1054 static void reset_predictor_group(PredictorState *ps, int group_num)
1057 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1058 reset_predict_state(&ps[i]);
1061 #define AAC_INIT_VLC_STATIC(num, size) \
1062 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1063 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1064 sizeof(ff_aac_spectral_bits[num][0]), \
1065 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1066 sizeof(ff_aac_spectral_codes[num][0]), \
1069 static void aacdec_init(AACContext *ac);
1071 static av_cold int aac_decode_init(AVCodecContext *avctx)
1073 AACContext *ac = avctx->priv_data;
1077 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1081 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1083 if (avctx->extradata_size > 0) {
1084 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1086 avctx->extradata_size * 8,
1091 uint8_t layout_map[MAX_ELEM_ID*4][3];
1092 int layout_map_tags;
1094 sr = sample_rate_idx(avctx->sample_rate);
1095 ac->oc[1].m4ac.sampling_index = sr;
1096 ac->oc[1].m4ac.channels = avctx->channels;
1097 ac->oc[1].m4ac.sbr = -1;
1098 ac->oc[1].m4ac.ps = -1;
1100 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1101 if (ff_mpeg4audio_channels[i] == avctx->channels)
1103 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1106 ac->oc[1].m4ac.chan_config = i;
1108 if (ac->oc[1].m4ac.chan_config) {
1109 int ret = set_default_channel_config(avctx, layout_map,
1110 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1112 output_configure(ac, layout_map, layout_map_tags,
1114 else if (avctx->err_recognition & AV_EF_EXPLODE)
1115 return AVERROR_INVALIDDATA;
1119 if (avctx->channels > MAX_CHANNELS) {
1120 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1121 return AVERROR_INVALIDDATA;
1124 AAC_INIT_VLC_STATIC( 0, 304);
1125 AAC_INIT_VLC_STATIC( 1, 270);
1126 AAC_INIT_VLC_STATIC( 2, 550);
1127 AAC_INIT_VLC_STATIC( 3, 300);
1128 AAC_INIT_VLC_STATIC( 4, 328);
1129 AAC_INIT_VLC_STATIC( 5, 294);
1130 AAC_INIT_VLC_STATIC( 6, 306);
1131 AAC_INIT_VLC_STATIC( 7, 268);
1132 AAC_INIT_VLC_STATIC( 8, 510);
1133 AAC_INIT_VLC_STATIC( 9, 366);
1134 AAC_INIT_VLC_STATIC(10, 462);
1138 ff_fmt_convert_init(&ac->fmt_conv, avctx);
1139 avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
1141 ac->random_state = 0x1f2e3d4c;
1145 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1146 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1147 ff_aac_scalefactor_bits,
1148 sizeof(ff_aac_scalefactor_bits[0]),
1149 sizeof(ff_aac_scalefactor_bits[0]),
1150 ff_aac_scalefactor_code,
1151 sizeof(ff_aac_scalefactor_code[0]),
1152 sizeof(ff_aac_scalefactor_code[0]),
1155 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1156 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1157 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1158 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1159 // window initialization
1160 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1161 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1162 ff_init_ff_sine_windows(10);
1163 ff_init_ff_sine_windows( 9);
1164 ff_init_ff_sine_windows( 7);
1172 * Skip data_stream_element; reference: table 4.10.
1174 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1176 int byte_align = get_bits1(gb);
1177 int count = get_bits(gb, 8);
1179 count += get_bits(gb, 8);
1183 if (get_bits_left(gb) < 8 * count) {
1184 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1185 return AVERROR_INVALIDDATA;
1187 skip_bits_long(gb, 8 * count);
1191 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1195 if (get_bits1(gb)) {
1196 ics->predictor_reset_group = get_bits(gb, 5);
1197 if (ics->predictor_reset_group == 0 ||
1198 ics->predictor_reset_group > 30) {
1199 av_log(ac->avctx, AV_LOG_ERROR,
1200 "Invalid Predictor Reset Group.\n");
1201 return AVERROR_INVALIDDATA;
1204 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1205 ics->prediction_used[sfb] = get_bits1(gb);
1211 * Decode Long Term Prediction data; reference: table 4.xx.
1213 static void decode_ltp(LongTermPrediction *ltp,
1214 GetBitContext *gb, uint8_t max_sfb)
1218 ltp->lag = get_bits(gb, 11);
1219 ltp->coef = ltp_coef[get_bits(gb, 3)];
1220 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1221 ltp->used[sfb] = get_bits1(gb);
1225 * Decode Individual Channel Stream info; reference: table 4.6.
1227 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1230 int aot = ac->oc[1].m4ac.object_type;
1231 if (aot != AOT_ER_AAC_ELD) {
1232 if (get_bits1(gb)) {
1233 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1234 return AVERROR_INVALIDDATA;
1236 ics->window_sequence[1] = ics->window_sequence[0];
1237 ics->window_sequence[0] = get_bits(gb, 2);
1238 if (aot == AOT_ER_AAC_LD &&
1239 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1240 av_log(ac->avctx, AV_LOG_ERROR,
1241 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1242 "window sequence %d found.\n", ics->window_sequence[0]);
1243 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1244 return AVERROR_INVALIDDATA;
1246 ics->use_kb_window[1] = ics->use_kb_window[0];
1247 ics->use_kb_window[0] = get_bits1(gb);
1249 ics->num_window_groups = 1;
1250 ics->group_len[0] = 1;
1251 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1253 ics->max_sfb = get_bits(gb, 4);
1254 for (i = 0; i < 7; i++) {
1255 if (get_bits1(gb)) {
1256 ics->group_len[ics->num_window_groups - 1]++;
1258 ics->num_window_groups++;
1259 ics->group_len[ics->num_window_groups - 1] = 1;
1262 ics->num_windows = 8;
1263 ics->swb_offset = ff_swb_offset_128[ac->oc[1].m4ac.sampling_index];
1264 ics->num_swb = ff_aac_num_swb_128[ac->oc[1].m4ac.sampling_index];
1265 ics->tns_max_bands = ff_tns_max_bands_128[ac->oc[1].m4ac.sampling_index];
1266 ics->predictor_present = 0;
1268 ics->max_sfb = get_bits(gb, 6);
1269 ics->num_windows = 1;
1270 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1271 ics->swb_offset = ff_swb_offset_512[ac->oc[1].m4ac.sampling_index];
1272 ics->num_swb = ff_aac_num_swb_512[ac->oc[1].m4ac.sampling_index];
1273 ics->tns_max_bands = ff_tns_max_bands_512[ac->oc[1].m4ac.sampling_index];
1274 if (!ics->num_swb || !ics->swb_offset)
1277 ics->swb_offset = ff_swb_offset_1024[ac->oc[1].m4ac.sampling_index];
1278 ics->num_swb = ff_aac_num_swb_1024[ac->oc[1].m4ac.sampling_index];
1279 ics->tns_max_bands = ff_tns_max_bands_1024[ac->oc[1].m4ac.sampling_index];
1281 if (aot != AOT_ER_AAC_ELD) {
1282 ics->predictor_present = get_bits1(gb);
1283 ics->predictor_reset_group = 0;
1285 if (ics->predictor_present) {
1286 if (aot == AOT_AAC_MAIN) {
1287 if (decode_prediction(ac, ics, gb)) {
1290 } else if (aot == AOT_AAC_LC ||
1291 aot == AOT_ER_AAC_LC) {
1292 av_log(ac->avctx, AV_LOG_ERROR,
1293 "Prediction is not allowed in AAC-LC.\n");
1296 if (aot == AOT_ER_AAC_LD) {
1297 av_log(ac->avctx, AV_LOG_ERROR,
1298 "LTP in ER AAC LD not yet implemented.\n");
1299 return AVERROR_PATCHWELCOME;
1301 if ((ics->ltp.present = get_bits(gb, 1)))
1302 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1307 if (ics->max_sfb > ics->num_swb) {
1308 av_log(ac->avctx, AV_LOG_ERROR,
1309 "Number of scalefactor bands in group (%d) "
1310 "exceeds limit (%d).\n",
1311 ics->max_sfb, ics->num_swb);
1318 return AVERROR_INVALIDDATA;
1322 * Decode band types (section_data payload); reference: table 4.46.
1324 * @param band_type array of the used band type
1325 * @param band_type_run_end array of the last scalefactor band of a band type run
1327 * @return Returns error status. 0 - OK, !0 - error
1329 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1330 int band_type_run_end[120], GetBitContext *gb,
1331 IndividualChannelStream *ics)
1334 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1335 for (g = 0; g < ics->num_window_groups; g++) {
1337 while (k < ics->max_sfb) {
1338 uint8_t sect_end = k;
1340 int sect_band_type = get_bits(gb, 4);
1341 if (sect_band_type == 12) {
1342 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1343 return AVERROR_INVALIDDATA;
1346 sect_len_incr = get_bits(gb, bits);
1347 sect_end += sect_len_incr;
1348 if (get_bits_left(gb) < 0) {
1349 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1350 return AVERROR_INVALIDDATA;
1352 if (sect_end > ics->max_sfb) {
1353 av_log(ac->avctx, AV_LOG_ERROR,
1354 "Number of bands (%d) exceeds limit (%d).\n",
1355 sect_end, ics->max_sfb);
1356 return AVERROR_INVALIDDATA;
1358 } while (sect_len_incr == (1 << bits) - 1);
1359 for (; k < sect_end; k++) {
1360 band_type [idx] = sect_band_type;
1361 band_type_run_end[idx++] = sect_end;
1369 * Decode scalefactors; reference: table 4.47.
1371 * @param global_gain first scalefactor value as scalefactors are differentially coded
1372 * @param band_type array of the used band type
1373 * @param band_type_run_end array of the last scalefactor band of a band type run
1374 * @param sf array of scalefactors or intensity stereo positions
1376 * @return Returns error status. 0 - OK, !0 - error
1378 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1379 unsigned int global_gain,
1380 IndividualChannelStream *ics,
1381 enum BandType band_type[120],
1382 int band_type_run_end[120])
1385 int offset[3] = { global_gain, global_gain - 90, 0 };
1388 for (g = 0; g < ics->num_window_groups; g++) {
1389 for (i = 0; i < ics->max_sfb;) {
1390 int run_end = band_type_run_end[idx];
1391 if (band_type[idx] == ZERO_BT) {
1392 for (; i < run_end; i++, idx++)
1394 } else if ((band_type[idx] == INTENSITY_BT) ||
1395 (band_type[idx] == INTENSITY_BT2)) {
1396 for (; i < run_end; i++, idx++) {
1397 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1398 clipped_offset = av_clip(offset[2], -155, 100);
1399 if (offset[2] != clipped_offset) {
1400 avpriv_request_sample(ac->avctx,
1401 "If you heard an audible artifact, there may be a bug in the decoder. "
1402 "Clipped intensity stereo position (%d -> %d)",
1403 offset[2], clipped_offset);
1405 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1407 } else if (band_type[idx] == NOISE_BT) {
1408 for (; i < run_end; i++, idx++) {
1409 if (noise_flag-- > 0)
1410 offset[1] += get_bits(gb, 9) - 256;
1412 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1413 clipped_offset = av_clip(offset[1], -100, 155);
1414 if (offset[1] != clipped_offset) {
1415 avpriv_request_sample(ac->avctx,
1416 "If you heard an audible artifact, there may be a bug in the decoder. "
1417 "Clipped noise gain (%d -> %d)",
1418 offset[1], clipped_offset);
1420 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1423 for (; i < run_end; i++, idx++) {
1424 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1425 if (offset[0] > 255U) {
1426 av_log(ac->avctx, AV_LOG_ERROR,
1427 "Scalefactor (%d) out of range.\n", offset[0]);
1428 return AVERROR_INVALIDDATA;
1430 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1439 * Decode pulse data; reference: table 4.7.
1441 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1442 const uint16_t *swb_offset, int num_swb)
1445 pulse->num_pulse = get_bits(gb, 2) + 1;
1446 pulse_swb = get_bits(gb, 6);
1447 if (pulse_swb >= num_swb)
1449 pulse->pos[0] = swb_offset[pulse_swb];
1450 pulse->pos[0] += get_bits(gb, 5);
1451 if (pulse->pos[0] >= swb_offset[num_swb])
1453 pulse->amp[0] = get_bits(gb, 4);
1454 for (i = 1; i < pulse->num_pulse; i++) {
1455 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1456 if (pulse->pos[i] >= swb_offset[num_swb])
1458 pulse->amp[i] = get_bits(gb, 4);
1464 * Decode Temporal Noise Shaping data; reference: table 4.48.
1466 * @return Returns error status. 0 - OK, !0 - error
1468 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1469 GetBitContext *gb, const IndividualChannelStream *ics)
1471 int w, filt, i, coef_len, coef_res, coef_compress;
1472 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1473 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1474 for (w = 0; w < ics->num_windows; w++) {
1475 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1476 coef_res = get_bits1(gb);
1478 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1480 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1482 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1483 av_log(ac->avctx, AV_LOG_ERROR,
1484 "TNS filter order %d is greater than maximum %d.\n",
1485 tns->order[w][filt], tns_max_order);
1486 tns->order[w][filt] = 0;
1487 return AVERROR_INVALIDDATA;
1489 if (tns->order[w][filt]) {
1490 tns->direction[w][filt] = get_bits1(gb);
1491 coef_compress = get_bits1(gb);
1492 coef_len = coef_res + 3 - coef_compress;
1493 tmp2_idx = 2 * coef_compress + coef_res;
1495 for (i = 0; i < tns->order[w][filt]; i++)
1496 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1505 * Decode Mid/Side data; reference: table 4.54.
1507 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1508 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1509 * [3] reserved for scalable AAC
1511 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1515 if (ms_present == 1) {
1517 idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1519 cpe->ms_mask[idx] = get_bits1(gb);
1520 } else if (ms_present == 2) {
1521 memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask[0]) * cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb);
1526 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1530 *dst++ = v[idx & 15] * s;
1531 *dst++ = v[idx>>4 & 15] * s;
1537 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1541 *dst++ = v[idx & 3] * s;
1542 *dst++ = v[idx>>2 & 3] * s;
1543 *dst++ = v[idx>>4 & 3] * s;
1544 *dst++ = v[idx>>6 & 3] * s;
1550 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1551 unsigned sign, const float *scale)
1553 union av_intfloat32 s0, s1;
1555 s0.f = s1.f = *scale;
1556 s0.i ^= sign >> 1 << 31;
1559 *dst++ = v[idx & 15] * s0.f;
1560 *dst++ = v[idx>>4 & 15] * s1.f;
1567 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1568 unsigned sign, const float *scale)
1570 unsigned nz = idx >> 12;
1571 union av_intfloat32 s = { .f = *scale };
1572 union av_intfloat32 t;
1574 t.i = s.i ^ (sign & 1U<<31);
1575 *dst++ = v[idx & 3] * t.f;
1577 sign <<= nz & 1; nz >>= 1;
1578 t.i = s.i ^ (sign & 1U<<31);
1579 *dst++ = v[idx>>2 & 3] * t.f;
1581 sign <<= nz & 1; nz >>= 1;
1582 t.i = s.i ^ (sign & 1U<<31);
1583 *dst++ = v[idx>>4 & 3] * t.f;
1586 t.i = s.i ^ (sign & 1U<<31);
1587 *dst++ = v[idx>>6 & 3] * t.f;
1594 * Decode spectral data; reference: table 4.50.
1595 * Dequantize and scale spectral data; reference: 4.6.3.3.
1597 * @param coef array of dequantized, scaled spectral data
1598 * @param sf array of scalefactors or intensity stereo positions
1599 * @param pulse_present set if pulses are present
1600 * @param pulse pointer to pulse data struct
1601 * @param band_type array of the used band type
1603 * @return Returns error status. 0 - OK, !0 - error
1605 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1606 GetBitContext *gb, const float sf[120],
1607 int pulse_present, const Pulse *pulse,
1608 const IndividualChannelStream *ics,
1609 enum BandType band_type[120])
1611 int i, k, g, idx = 0;
1612 const int c = 1024 / ics->num_windows;
1613 const uint16_t *offsets = ics->swb_offset;
1614 float *coef_base = coef;
1616 for (g = 0; g < ics->num_windows; g++)
1617 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1618 sizeof(float) * (c - offsets[ics->max_sfb]));
1620 for (g = 0; g < ics->num_window_groups; g++) {
1621 unsigned g_len = ics->group_len[g];
1623 for (i = 0; i < ics->max_sfb; i++, idx++) {
1624 const unsigned cbt_m1 = band_type[idx] - 1;
1625 float *cfo = coef + offsets[i];
1626 int off_len = offsets[i + 1] - offsets[i];
1629 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1630 for (group = 0; group < g_len; group++, cfo+=128) {
1631 memset(cfo, 0, off_len * sizeof(float));
1633 } else if (cbt_m1 == NOISE_BT - 1) {
1634 for (group = 0; group < g_len; group++, cfo+=128) {
1638 for (k = 0; k < off_len; k++) {
1639 ac->random_state = lcg_random(ac->random_state);
1640 cfo[k] = ac->random_state;
1643 band_energy = ac->fdsp.scalarproduct_float(cfo, cfo, off_len);
1644 scale = sf[idx] / sqrtf(band_energy);
1645 ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1648 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1649 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1650 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1651 OPEN_READER(re, gb);
1653 switch (cbt_m1 >> 1) {
1655 for (group = 0; group < g_len; group++, cfo+=128) {
1663 UPDATE_CACHE(re, gb);
1664 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1665 cb_idx = cb_vector_idx[code];
1666 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1672 for (group = 0; group < g_len; group++, cfo+=128) {
1682 UPDATE_CACHE(re, gb);
1683 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1684 cb_idx = cb_vector_idx[code];
1685 nnz = cb_idx >> 8 & 15;
1686 bits = nnz ? GET_CACHE(re, gb) : 0;
1687 LAST_SKIP_BITS(re, gb, nnz);
1688 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1694 for (group = 0; group < g_len; group++, cfo+=128) {
1702 UPDATE_CACHE(re, gb);
1703 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1704 cb_idx = cb_vector_idx[code];
1705 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1712 for (group = 0; group < g_len; group++, cfo+=128) {
1722 UPDATE_CACHE(re, gb);
1723 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1724 cb_idx = cb_vector_idx[code];
1725 nnz = cb_idx >> 8 & 15;
1726 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1727 LAST_SKIP_BITS(re, gb, nnz);
1728 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1734 for (group = 0; group < g_len; group++, cfo+=128) {
1736 uint32_t *icf = (uint32_t *) cf;
1746 UPDATE_CACHE(re, gb);
1747 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1755 cb_idx = cb_vector_idx[code];
1758 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1759 LAST_SKIP_BITS(re, gb, nnz);
1761 for (j = 0; j < 2; j++) {
1765 /* The total length of escape_sequence must be < 22 bits according
1766 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1767 UPDATE_CACHE(re, gb);
1768 b = GET_CACHE(re, gb);
1769 b = 31 - av_log2(~b);
1772 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1773 return AVERROR_INVALIDDATA;
1776 SKIP_BITS(re, gb, b + 1);
1778 n = (1 << b) + SHOW_UBITS(re, gb, b);
1779 LAST_SKIP_BITS(re, gb, b);
1780 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1783 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1784 *icf++ = (bits & 1U<<31) | v;
1791 ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1795 CLOSE_READER(re, gb);
1801 if (pulse_present) {
1803 for (i = 0; i < pulse->num_pulse; i++) {
1804 float co = coef_base[ pulse->pos[i] ];
1805 while (offsets[idx + 1] <= pulse->pos[i])
1807 if (band_type[idx] != NOISE_BT && sf[idx]) {
1808 float ico = -pulse->amp[i];
1811 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1813 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1820 static av_always_inline float flt16_round(float pf)
1822 union av_intfloat32 tmp;
1824 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1828 static av_always_inline float flt16_even(float pf)
1830 union av_intfloat32 tmp;
1832 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1836 static av_always_inline float flt16_trunc(float pf)
1838 union av_intfloat32 pun;
1840 pun.i &= 0xFFFF0000U;
1844 static av_always_inline void predict(PredictorState *ps, float *coef,
1847 const float a = 0.953125; // 61.0 / 64
1848 const float alpha = 0.90625; // 29.0 / 32
1852 float r0 = ps->r0, r1 = ps->r1;
1853 float cor0 = ps->cor0, cor1 = ps->cor1;
1854 float var0 = ps->var0, var1 = ps->var1;
1856 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1857 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1859 pv = flt16_round(k1 * r0 + k2 * r1);
1866 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1867 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1868 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1869 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1871 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1872 ps->r0 = flt16_trunc(a * e0);
1876 * Apply AAC-Main style frequency domain prediction.
1878 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1882 if (!sce->ics.predictor_initialized) {
1883 reset_all_predictors(sce->predictor_state);
1884 sce->ics.predictor_initialized = 1;
1887 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1889 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1891 for (k = sce->ics.swb_offset[sfb];
1892 k < sce->ics.swb_offset[sfb + 1];
1894 predict(&sce->predictor_state[k], &sce->coeffs[k],
1895 sce->ics.predictor_present &&
1896 sce->ics.prediction_used[sfb]);
1899 if (sce->ics.predictor_reset_group)
1900 reset_predictor_group(sce->predictor_state,
1901 sce->ics.predictor_reset_group);
1903 reset_all_predictors(sce->predictor_state);
1907 * Decode an individual_channel_stream payload; reference: table 4.44.
1909 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1910 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1912 * @return Returns error status. 0 - OK, !0 - error
1914 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1915 GetBitContext *gb, int common_window, int scale_flag)
1918 TemporalNoiseShaping *tns = &sce->tns;
1919 IndividualChannelStream *ics = &sce->ics;
1920 float *out = sce->coeffs;
1921 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1924 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1925 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1926 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1927 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1928 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1930 /* This assignment is to silence a GCC warning about the variable being used
1931 * uninitialized when in fact it always is.
1933 pulse.num_pulse = 0;
1935 global_gain = get_bits(gb, 8);
1937 if (!common_window && !scale_flag) {
1938 if (decode_ics_info(ac, ics, gb) < 0)
1939 return AVERROR_INVALIDDATA;
1942 if ((ret = decode_band_types(ac, sce->band_type,
1943 sce->band_type_run_end, gb, ics)) < 0)
1945 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1946 sce->band_type, sce->band_type_run_end)) < 0)
1951 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1952 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1953 av_log(ac->avctx, AV_LOG_ERROR,
1954 "Pulse tool not allowed in eight short sequence.\n");
1955 return AVERROR_INVALIDDATA;
1957 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1958 av_log(ac->avctx, AV_LOG_ERROR,
1959 "Pulse data corrupt or invalid.\n");
1960 return AVERROR_INVALIDDATA;
1963 tns->present = get_bits1(gb);
1964 if (tns->present && !er_syntax)
1965 if (decode_tns(ac, tns, gb, ics) < 0)
1966 return AVERROR_INVALIDDATA;
1967 if (!eld_syntax && get_bits1(gb)) {
1968 avpriv_request_sample(ac->avctx, "SSR");
1969 return AVERROR_PATCHWELCOME;
1971 // I see no textual basis in the spec for this occurring after SSR gain
1972 // control, but this is what both reference and real implmentations do
1973 if (tns->present && er_syntax)
1974 if (decode_tns(ac, tns, gb, ics) < 0)
1975 return AVERROR_INVALIDDATA;
1978 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1979 &pulse, ics, sce->band_type) < 0)
1980 return AVERROR_INVALIDDATA;
1982 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1983 apply_prediction(ac, sce);
1989 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1991 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1993 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1994 float *ch0 = cpe->ch[0].coeffs;
1995 float *ch1 = cpe->ch[1].coeffs;
1996 int g, i, group, idx = 0;
1997 const uint16_t *offsets = ics->swb_offset;
1998 for (g = 0; g < ics->num_window_groups; g++) {
1999 for (i = 0; i < ics->max_sfb; i++, idx++) {
2000 if (cpe->ms_mask[idx] &&
2001 cpe->ch[0].band_type[idx] < NOISE_BT &&
2002 cpe->ch[1].band_type[idx] < NOISE_BT) {
2003 for (group = 0; group < ics->group_len[g]; group++) {
2004 ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
2005 ch1 + group * 128 + offsets[i],
2006 offsets[i+1] - offsets[i]);
2010 ch0 += ics->group_len[g] * 128;
2011 ch1 += ics->group_len[g] * 128;
2016 * intensity stereo decoding; reference: 4.6.8.2.3
2018 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2019 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2020 * [3] reserved for scalable AAC
2022 static void apply_intensity_stereo(AACContext *ac,
2023 ChannelElement *cpe, int ms_present)
2025 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2026 SingleChannelElement *sce1 = &cpe->ch[1];
2027 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2028 const uint16_t *offsets = ics->swb_offset;
2029 int g, group, i, idx = 0;
2032 for (g = 0; g < ics->num_window_groups; g++) {
2033 for (i = 0; i < ics->max_sfb;) {
2034 if (sce1->band_type[idx] == INTENSITY_BT ||
2035 sce1->band_type[idx] == INTENSITY_BT2) {
2036 const int bt_run_end = sce1->band_type_run_end[idx];
2037 for (; i < bt_run_end; i++, idx++) {
2038 c = -1 + 2 * (sce1->band_type[idx] - 14);
2040 c *= 1 - 2 * cpe->ms_mask[idx];
2041 scale = c * sce1->sf[idx];
2042 for (group = 0; group < ics->group_len[g]; group++)
2043 ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2044 coef0 + group * 128 + offsets[i],
2046 offsets[i + 1] - offsets[i]);
2049 int bt_run_end = sce1->band_type_run_end[idx];
2050 idx += bt_run_end - i;
2054 coef0 += ics->group_len[g] * 128;
2055 coef1 += ics->group_len[g] * 128;
2060 * Decode a channel_pair_element; reference: table 4.4.
2062 * @return Returns error status. 0 - OK, !0 - error
2064 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2066 int i, ret, common_window, ms_present = 0;
2067 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2069 common_window = eld_syntax || get_bits1(gb);
2070 if (common_window) {
2071 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2072 return AVERROR_INVALIDDATA;
2073 i = cpe->ch[1].ics.use_kb_window[0];
2074 cpe->ch[1].ics = cpe->ch[0].ics;
2075 cpe->ch[1].ics.use_kb_window[1] = i;
2076 if (cpe->ch[1].ics.predictor_present &&
2077 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2078 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2079 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2080 ms_present = get_bits(gb, 2);
2081 if (ms_present == 3) {
2082 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2083 return AVERROR_INVALIDDATA;
2084 } else if (ms_present)
2085 decode_mid_side_stereo(cpe, gb, ms_present);
2087 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2089 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2092 if (common_window) {
2094 apply_mid_side_stereo(ac, cpe);
2095 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2096 apply_prediction(ac, &cpe->ch[0]);
2097 apply_prediction(ac, &cpe->ch[1]);
2101 apply_intensity_stereo(ac, cpe, ms_present);
2105 static const float cce_scale[] = {
2106 1.09050773266525765921, //2^(1/8)
2107 1.18920711500272106672, //2^(1/4)
2113 * Decode coupling_channel_element; reference: table 4.8.
2115 * @return Returns error status. 0 - OK, !0 - error
2117 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2123 SingleChannelElement *sce = &che->ch[0];
2124 ChannelCoupling *coup = &che->coup;
2126 coup->coupling_point = 2 * get_bits1(gb);
2127 coup->num_coupled = get_bits(gb, 3);
2128 for (c = 0; c <= coup->num_coupled; c++) {
2130 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2131 coup->id_select[c] = get_bits(gb, 4);
2132 if (coup->type[c] == TYPE_CPE) {
2133 coup->ch_select[c] = get_bits(gb, 2);
2134 if (coup->ch_select[c] == 3)
2137 coup->ch_select[c] = 2;
2139 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2141 sign = get_bits(gb, 1);
2142 scale = cce_scale[get_bits(gb, 2)];
2144 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2147 for (c = 0; c < num_gain; c++) {
2151 float gain_cache = 1.0;
2153 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2154 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2155 gain_cache = powf(scale, -gain);
2157 if (coup->coupling_point == AFTER_IMDCT) {
2158 coup->gain[c][0] = gain_cache;
2160 for (g = 0; g < sce->ics.num_window_groups; g++) {
2161 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2162 if (sce->band_type[idx] != ZERO_BT) {
2164 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2172 gain_cache = powf(scale, -t) * s;
2175 coup->gain[c][idx] = gain_cache;
2185 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2187 * @return Returns number of bytes consumed.
2189 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2193 int num_excl_chan = 0;
2196 for (i = 0; i < 7; i++)
2197 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2198 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2200 return num_excl_chan / 7;
2204 * Decode dynamic range information; reference: table 4.52.
2206 * @return Returns number of bytes consumed.
2208 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2212 int drc_num_bands = 1;
2215 /* pce_tag_present? */
2216 if (get_bits1(gb)) {
2217 che_drc->pce_instance_tag = get_bits(gb, 4);
2218 skip_bits(gb, 4); // tag_reserved_bits
2222 /* excluded_chns_present? */
2223 if (get_bits1(gb)) {
2224 n += decode_drc_channel_exclusions(che_drc, gb);
2227 /* drc_bands_present? */
2228 if (get_bits1(gb)) {
2229 che_drc->band_incr = get_bits(gb, 4);
2230 che_drc->interpolation_scheme = get_bits(gb, 4);
2232 drc_num_bands += che_drc->band_incr;
2233 for (i = 0; i < drc_num_bands; i++) {
2234 che_drc->band_top[i] = get_bits(gb, 8);
2239 /* prog_ref_level_present? */
2240 if (get_bits1(gb)) {
2241 che_drc->prog_ref_level = get_bits(gb, 7);
2242 skip_bits1(gb); // prog_ref_level_reserved_bits
2246 for (i = 0; i < drc_num_bands; i++) {
2247 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2248 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2255 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2257 int i, major, minor;
2262 get_bits(gb, 13); len -= 13;
2264 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2265 buf[i] = get_bits(gb, 8);
2268 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2269 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2271 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2272 ac->avctx->internal->skip_samples = 1024;
2276 skip_bits_long(gb, len);
2282 * Decode extension data (incomplete); reference: table 4.51.
2284 * @param cnt length of TYPE_FIL syntactic element in bytes
2286 * @return Returns number of bytes consumed
2288 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2289 ChannelElement *che, enum RawDataBlockType elem_type)
2293 switch (get_bits(gb, 4)) { // extension type
2294 case EXT_SBR_DATA_CRC:
2298 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2300 } else if (!ac->oc[1].m4ac.sbr) {
2301 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2302 skip_bits_long(gb, 8 * cnt - 4);
2304 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2305 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2306 skip_bits_long(gb, 8 * cnt - 4);
2308 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2309 ac->oc[1].m4ac.sbr = 1;
2310 ac->oc[1].m4ac.ps = 1;
2311 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2312 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2313 ac->oc[1].status, 1);
2315 ac->oc[1].m4ac.sbr = 1;
2316 ac->avctx->profile = FF_PROFILE_AAC_HE;
2318 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2320 case EXT_DYNAMIC_RANGE:
2321 res = decode_dynamic_range(&ac->che_drc, gb);
2324 decode_fill(ac, gb, 8 * cnt - 4);
2327 case EXT_DATA_ELEMENT:
2329 skip_bits_long(gb, 8 * cnt - 4);
2336 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2338 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2339 * @param coef spectral coefficients
2341 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2342 IndividualChannelStream *ics, int decode)
2344 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2346 int bottom, top, order, start, end, size, inc;
2347 float lpc[TNS_MAX_ORDER];
2348 float tmp[TNS_MAX_ORDER+1];
2350 for (w = 0; w < ics->num_windows; w++) {
2351 bottom = ics->num_swb;
2352 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2354 bottom = FFMAX(0, top - tns->length[w][filt]);
2355 order = tns->order[w][filt];
2360 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2362 start = ics->swb_offset[FFMIN(bottom, mmm)];
2363 end = ics->swb_offset[FFMIN( top, mmm)];
2364 if ((size = end - start) <= 0)
2366 if (tns->direction[w][filt]) {
2376 for (m = 0; m < size; m++, start += inc)
2377 for (i = 1; i <= FFMIN(m, order); i++)
2378 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2381 for (m = 0; m < size; m++, start += inc) {
2382 tmp[0] = coef[start];
2383 for (i = 1; i <= FFMIN(m, order); i++)
2384 coef[start] += tmp[i] * lpc[i - 1];
2385 for (i = order; i > 0; i--)
2386 tmp[i] = tmp[i - 1];
2394 * Apply windowing and MDCT to obtain the spectral
2395 * coefficient from the predicted sample by LTP.
2397 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2398 float *in, IndividualChannelStream *ics)
2400 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2401 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2402 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2403 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2405 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2406 ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
2408 memset(in, 0, 448 * sizeof(float));
2409 ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2411 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2412 ac->fdsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2414 ac->fdsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2415 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2417 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2421 * Apply the long term prediction
2423 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2425 const LongTermPrediction *ltp = &sce->ics.ltp;
2426 const uint16_t *offsets = sce->ics.swb_offset;
2429 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2430 float *predTime = sce->ret;
2431 float *predFreq = ac->buf_mdct;
2432 int16_t num_samples = 2048;
2434 if (ltp->lag < 1024)
2435 num_samples = ltp->lag + 1024;
2436 for (i = 0; i < num_samples; i++)
2437 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2438 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2440 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2442 if (sce->tns.present)
2443 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2445 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2447 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2448 sce->coeffs[i] += predFreq[i];
2453 * Update the LTP buffer for next frame
2455 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2457 IndividualChannelStream *ics = &sce->ics;
2458 float *saved = sce->saved;
2459 float *saved_ltp = sce->coeffs;
2460 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2461 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2464 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2465 memcpy(saved_ltp, saved, 512 * sizeof(float));
2466 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2467 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2468 for (i = 0; i < 64; i++)
2469 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2470 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2471 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2472 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2473 ac->fdsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2474 for (i = 0; i < 64; i++)
2475 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2476 } else { // LONG_STOP or ONLY_LONG
2477 ac->fdsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2478 for (i = 0; i < 512; i++)
2479 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2482 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2483 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2484 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2488 * Conduct IMDCT and windowing.
2490 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2492 IndividualChannelStream *ics = &sce->ics;
2493 float *in = sce->coeffs;
2494 float *out = sce->ret;
2495 float *saved = sce->saved;
2496 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2497 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2498 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2499 float *buf = ac->buf_mdct;
2500 float *temp = ac->temp;
2504 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2505 for (i = 0; i < 1024; i += 128)
2506 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2508 ac->mdct.imdct_half(&ac->mdct, buf, in);
2510 /* window overlapping
2511 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2512 * and long to short transitions are considered to be short to short
2513 * transitions. This leaves just two cases (long to long and short to short)
2514 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2516 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2517 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2518 ac->fdsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2520 memcpy( out, saved, 448 * sizeof(float));
2522 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2523 ac->fdsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2524 ac->fdsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2525 ac->fdsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2526 ac->fdsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2527 ac->fdsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2528 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2530 ac->fdsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2531 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2536 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2537 memcpy( saved, temp + 64, 64 * sizeof(float));
2538 ac->fdsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2539 ac->fdsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2540 ac->fdsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2541 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2542 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2543 memcpy( saved, buf + 512, 448 * sizeof(float));
2544 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2545 } else { // LONG_STOP or ONLY_LONG
2546 memcpy( saved, buf + 512, 512 * sizeof(float));
2550 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2552 IndividualChannelStream *ics = &sce->ics;
2553 float *in = sce->coeffs;
2554 float *out = sce->ret;
2555 float *saved = sce->saved;
2556 float *buf = ac->buf_mdct;
2559 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2561 // window overlapping
2562 if (ics->use_kb_window[1]) {
2563 // AAC LD uses a low overlap sine window instead of a KBD window
2564 memcpy(out, saved, 192 * sizeof(float));
2565 ac->fdsp.vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2566 memcpy( out + 320, buf + 64, 192 * sizeof(float));
2568 ac->fdsp.vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2572 memcpy(saved, buf + 256, 256 * sizeof(float));
2575 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2577 float *in = sce->coeffs;
2578 float *out = sce->ret;
2579 float *saved = sce->saved;
2580 const float *const window = ff_aac_eld_window;
2581 float *buf = ac->buf_mdct;
2584 const int n2 = n >> 1;
2585 const int n4 = n >> 2;
2587 // Inverse transform, mapped to the conventional IMDCT by
2588 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2589 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2590 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2591 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2592 for (i = 0; i < n2; i+=2) {
2594 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2595 temp = -in[i + 1]; in[i + 1] = in[n - 2 - i]; in[n - 2 - i] = temp;
2597 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2598 for (i = 0; i < n; i+=2) {
2601 // Like with the regular IMDCT at this point we still have the middle half
2602 // of a transform but with even symmetry on the left and odd symmetry on
2605 // window overlapping
2606 // The spec says to use samples [0..511] but the reference decoder uses
2607 // samples [128..639].
2608 for (i = n4; i < n2; i ++) {
2609 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2610 saved[ i + n2] * window[i + n - n4] +
2611 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2612 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2614 for (i = 0; i < n2; i ++) {
2615 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2616 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2617 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2618 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2620 for (i = 0; i < n4; i ++) {
2621 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2622 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2623 -saved[ n + n2 + i] * window[i + 3*n - n4];
2627 memmove(saved + n, saved, 2 * n * sizeof(float));
2628 memcpy( saved, buf, n * sizeof(float));
2632 * Apply dependent channel coupling (applied before IMDCT).
2634 * @param index index into coupling gain array
2636 static void apply_dependent_coupling(AACContext *ac,
2637 SingleChannelElement *target,
2638 ChannelElement *cce, int index)
2640 IndividualChannelStream *ics = &cce->ch[0].ics;
2641 const uint16_t *offsets = ics->swb_offset;
2642 float *dest = target->coeffs;
2643 const float *src = cce->ch[0].coeffs;
2644 int g, i, group, k, idx = 0;
2645 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2646 av_log(ac->avctx, AV_LOG_ERROR,
2647 "Dependent coupling is not supported together with LTP\n");
2650 for (g = 0; g < ics->num_window_groups; g++) {
2651 for (i = 0; i < ics->max_sfb; i++, idx++) {
2652 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2653 const float gain = cce->coup.gain[index][idx];
2654 for (group = 0; group < ics->group_len[g]; group++) {
2655 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2657 dest[group * 128 + k] += gain * src[group * 128 + k];
2662 dest += ics->group_len[g] * 128;
2663 src += ics->group_len[g] * 128;
2668 * Apply independent channel coupling (applied after IMDCT).
2670 * @param index index into coupling gain array
2672 static void apply_independent_coupling(AACContext *ac,
2673 SingleChannelElement *target,
2674 ChannelElement *cce, int index)
2677 const float gain = cce->coup.gain[index][0];
2678 const float *src = cce->ch[0].ret;
2679 float *dest = target->ret;
2680 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2682 for (i = 0; i < len; i++)
2683 dest[i] += gain * src[i];
2687 * channel coupling transformation interface
2689 * @param apply_coupling_method pointer to (in)dependent coupling function
2691 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2692 enum RawDataBlockType type, int elem_id,
2693 enum CouplingPoint coupling_point,
2694 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2698 for (i = 0; i < MAX_ELEM_ID; i++) {
2699 ChannelElement *cce = ac->che[TYPE_CCE][i];
2702 if (cce && cce->coup.coupling_point == coupling_point) {
2703 ChannelCoupling *coup = &cce->coup;
2705 for (c = 0; c <= coup->num_coupled; c++) {
2706 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2707 if (coup->ch_select[c] != 1) {
2708 apply_coupling_method(ac, &cc->ch[0], cce, index);
2709 if (coup->ch_select[c] != 0)
2712 if (coup->ch_select[c] != 2)
2713 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2715 index += 1 + (coup->ch_select[c] == 3);
2722 * Convert spectral data to float samples, applying all supported tools as appropriate.
2724 static void spectral_to_sample(AACContext *ac)
2727 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2728 switch (ac->oc[1].m4ac.object_type) {
2730 imdct_and_window = imdct_and_windowing_ld;
2732 case AOT_ER_AAC_ELD:
2733 imdct_and_window = imdct_and_windowing_eld;
2736 imdct_and_window = ac->imdct_and_windowing;
2738 for (type = 3; type >= 0; type--) {
2739 for (i = 0; i < MAX_ELEM_ID; i++) {
2740 ChannelElement *che = ac->che[type][i];
2742 if (type <= TYPE_CPE)
2743 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2744 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2745 if (che->ch[0].ics.predictor_present) {
2746 if (che->ch[0].ics.ltp.present)
2747 ac->apply_ltp(ac, &che->ch[0]);
2748 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2749 ac->apply_ltp(ac, &che->ch[1]);
2752 if (che->ch[0].tns.present)
2753 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2754 if (che->ch[1].tns.present)
2755 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2756 if (type <= TYPE_CPE)
2757 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2758 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2759 imdct_and_window(ac, &che->ch[0]);
2760 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2761 ac->update_ltp(ac, &che->ch[0]);
2762 if (type == TYPE_CPE) {
2763 imdct_and_window(ac, &che->ch[1]);
2764 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2765 ac->update_ltp(ac, &che->ch[1]);
2767 if (ac->oc[1].m4ac.sbr > 0) {
2768 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2771 if (type <= TYPE_CCE)
2772 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2778 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2781 AACADTSHeaderInfo hdr_info;
2782 uint8_t layout_map[MAX_ELEM_ID*4][3];
2783 int layout_map_tags, ret;
2785 size = avpriv_aac_parse_header(gb, &hdr_info);
2787 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2788 // This is 2 for "VLB " audio in NSV files.
2789 // See samples/nsv/vlb_audio.
2790 avpriv_report_missing_feature(ac->avctx,
2791 "More than one AAC RDB per ADTS frame");
2792 ac->warned_num_aac_frames = 1;
2794 push_output_configuration(ac);
2795 if (hdr_info.chan_config) {
2796 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2797 if ((ret = set_default_channel_config(ac->avctx,
2800 hdr_info.chan_config)) < 0)
2802 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2803 FFMAX(ac->oc[1].status,
2804 OC_TRIAL_FRAME), 0)) < 0)
2807 ac->oc[1].m4ac.chan_config = 0;
2809 * dual mono frames in Japanese DTV can have chan_config 0
2810 * WITHOUT specifying PCE.
2811 * thus, set dual mono as default.
2813 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2814 layout_map_tags = 2;
2815 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2816 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2817 layout_map[0][1] = 0;
2818 layout_map[1][1] = 1;
2819 if (output_configure(ac, layout_map, layout_map_tags,
2824 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2825 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2826 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2827 if (ac->oc[0].status != OC_LOCKED ||
2828 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2829 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2830 ac->oc[1].m4ac.sbr = -1;
2831 ac->oc[1].m4ac.ps = -1;
2833 if (!hdr_info.crc_absent)
2839 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2840 int *got_frame_ptr, GetBitContext *gb)
2842 AACContext *ac = avctx->priv_data;
2843 ChannelElement *che;
2846 int chan_config = ac->oc[1].m4ac.chan_config;
2847 int aot = ac->oc[1].m4ac.object_type;
2849 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2854 if ((err = frame_configure_elements(avctx)) < 0)
2857 // The FF_PROFILE_AAC_* defines are all object_type - 1
2858 // This may lead to an undefined profile being signaled
2859 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2861 ac->tags_mapped = 0;
2863 if (chan_config < 0 || chan_config >= 8) {
2864 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2865 ac->oc[1].m4ac.chan_config);
2866 return AVERROR_INVALIDDATA;
2868 for (i = 0; i < tags_per_config[chan_config]; i++) {
2869 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2870 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2871 if (!(che=get_che(ac, elem_type, elem_id))) {
2872 av_log(ac->avctx, AV_LOG_ERROR,
2873 "channel element %d.%d is not allocated\n",
2874 elem_type, elem_id);
2875 return AVERROR_INVALIDDATA;
2877 if (aot != AOT_ER_AAC_ELD)
2879 switch (elem_type) {
2881 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2884 err = decode_cpe(ac, gb, che);
2887 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2894 spectral_to_sample(ac);
2896 ac->frame->nb_samples = samples;
2897 ac->frame->sample_rate = avctx->sample_rate;
2900 skip_bits_long(gb, get_bits_left(gb));
2904 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2905 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2907 AACContext *ac = avctx->priv_data;
2908 ChannelElement *che = NULL, *che_prev = NULL;
2909 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2911 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2912 int is_dmono, sce_count = 0;
2916 if (show_bits(gb, 12) == 0xfff) {
2917 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2918 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2921 if (ac->oc[1].m4ac.sampling_index > 12) {
2922 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2923 err = AVERROR_INVALIDDATA;
2928 if ((err = frame_configure_elements(avctx)) < 0)
2931 // The FF_PROFILE_AAC_* defines are all object_type - 1
2932 // This may lead to an undefined profile being signaled
2933 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2935 ac->tags_mapped = 0;
2937 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2938 elem_id = get_bits(gb, 4);
2940 if (avctx->debug & FF_DEBUG_STARTCODE)
2941 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2943 if (elem_type < TYPE_DSE) {
2944 if (!(che=get_che(ac, elem_type, elem_id))) {
2945 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2946 elem_type, elem_id);
2947 err = AVERROR_INVALIDDATA;
2953 switch (elem_type) {
2956 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2962 err = decode_cpe(ac, gb, che);
2967 err = decode_cce(ac, gb, che);
2971 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2976 err = skip_data_stream_element(ac, gb);
2980 uint8_t layout_map[MAX_ELEM_ID*4][3];
2982 push_output_configuration(ac);
2983 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2989 av_log(avctx, AV_LOG_ERROR,
2990 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2992 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2994 ac->oc[1].m4ac.chan_config = 0;
3002 elem_id += get_bits(gb, 8) - 1;
3003 if (get_bits_left(gb) < 8 * elem_id) {
3004 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3005 err = AVERROR_INVALIDDATA;
3009 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3010 err = 0; /* FIXME */
3014 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3019 elem_type_prev = elem_type;
3024 if (get_bits_left(gb) < 3) {
3025 av_log(avctx, AV_LOG_ERROR, overread_err);
3026 err = AVERROR_INVALIDDATA;
3031 spectral_to_sample(ac);
3033 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3034 samples <<= multiplier;
3036 if (ac->oc[1].status && audio_found) {
3037 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3038 avctx->frame_size = samples;
3039 ac->oc[1].status = OC_LOCKED;
3044 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3045 if (side && side_size>=4)
3046 AV_WL32(side, 2*AV_RL32(side));
3049 *got_frame_ptr = !!samples;
3051 ac->frame->nb_samples = samples;
3052 ac->frame->sample_rate = avctx->sample_rate;
3054 av_frame_unref(ac->frame);
3055 *got_frame_ptr = !!samples;
3057 /* for dual-mono audio (SCE + SCE) */
3058 is_dmono = ac->dmono_mode && sce_count == 2 &&
3059 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3061 if (ac->dmono_mode == 1)
3062 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3063 else if (ac->dmono_mode == 2)
3064 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3069 pop_output_configuration(ac);
3073 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3074 int *got_frame_ptr, AVPacket *avpkt)
3076 AACContext *ac = avctx->priv_data;
3077 const uint8_t *buf = avpkt->data;
3078 int buf_size = avpkt->size;
3083 int new_extradata_size;
3084 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3085 AV_PKT_DATA_NEW_EXTRADATA,
3086 &new_extradata_size);
3087 int jp_dualmono_size;
3088 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3089 AV_PKT_DATA_JP_DUALMONO,
3092 if (new_extradata && 0) {
3093 av_free(avctx->extradata);
3094 avctx->extradata = av_mallocz(new_extradata_size +
3095 FF_INPUT_BUFFER_PADDING_SIZE);
3096 if (!avctx->extradata)
3097 return AVERROR(ENOMEM);
3098 avctx->extradata_size = new_extradata_size;
3099 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3100 push_output_configuration(ac);
3101 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3103 avctx->extradata_size*8, 1) < 0) {
3104 pop_output_configuration(ac);
3105 return AVERROR_INVALIDDATA;
3110 if (jp_dualmono && jp_dualmono_size > 0)
3111 ac->dmono_mode = 1 + *jp_dualmono;
3112 if (ac->force_dmono_mode >= 0)
3113 ac->dmono_mode = ac->force_dmono_mode;
3115 if (INT_MAX / 8 <= buf_size)
3116 return AVERROR_INVALIDDATA;
3118 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3121 switch (ac->oc[1].m4ac.object_type) {
3123 case AOT_ER_AAC_LTP:
3125 case AOT_ER_AAC_ELD:
3126 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3129 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3134 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3135 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3136 if (buf[buf_offset])
3139 return buf_size > buf_offset ? buf_consumed : buf_size;
3142 static av_cold int aac_decode_close(AVCodecContext *avctx)
3144 AACContext *ac = avctx->priv_data;
3147 for (i = 0; i < MAX_ELEM_ID; i++) {
3148 for (type = 0; type < 4; type++) {
3149 if (ac->che[type][i])
3150 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3151 av_freep(&ac->che[type][i]);
3155 ff_mdct_end(&ac->mdct);
3156 ff_mdct_end(&ac->mdct_small);
3157 ff_mdct_end(&ac->mdct_ld);
3158 ff_mdct_end(&ac->mdct_ltp);
3163 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3165 struct LATMContext {
3166 AACContext aac_ctx; ///< containing AACContext
3167 int initialized; ///< initialized after a valid extradata was seen
3170 int audio_mux_version_A; ///< LATM syntax version
3171 int frame_length_type; ///< 0/1 variable/fixed frame length
3172 int frame_length; ///< frame length for fixed frame length
3175 static inline uint32_t latm_get_value(GetBitContext *b)
3177 int length = get_bits(b, 2);
3179 return get_bits_long(b, (length+1)*8);
3182 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3183 GetBitContext *gb, int asclen)
3185 AACContext *ac = &latmctx->aac_ctx;
3186 AVCodecContext *avctx = ac->avctx;
3187 MPEG4AudioConfig m4ac = { 0 };
3188 int config_start_bit = get_bits_count(gb);
3189 int sync_extension = 0;
3190 int bits_consumed, esize;
3194 asclen = FFMIN(asclen, get_bits_left(gb));
3196 asclen = get_bits_left(gb);
3198 if (config_start_bit % 8) {
3199 avpriv_request_sample(latmctx->aac_ctx.avctx,
3200 "Non-byte-aligned audio-specific config");
3201 return AVERROR_PATCHWELCOME;
3204 return AVERROR_INVALIDDATA;
3205 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3206 gb->buffer + (config_start_bit / 8),
3207 asclen, sync_extension);
3209 if (bits_consumed < 0)
3210 return AVERROR_INVALIDDATA;
3212 if (!latmctx->initialized ||
3213 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3214 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3216 if(latmctx->initialized) {
3217 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3219 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3221 latmctx->initialized = 0;
3223 esize = (bits_consumed+7) / 8;
3225 if (avctx->extradata_size < esize) {
3226 av_free(avctx->extradata);
3227 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3228 if (!avctx->extradata)
3229 return AVERROR(ENOMEM);
3232 avctx->extradata_size = esize;
3233 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3234 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3236 skip_bits_long(gb, bits_consumed);
3238 return bits_consumed;
3241 static int read_stream_mux_config(struct LATMContext *latmctx,
3244 int ret, audio_mux_version = get_bits(gb, 1);
3246 latmctx->audio_mux_version_A = 0;
3247 if (audio_mux_version)
3248 latmctx->audio_mux_version_A = get_bits(gb, 1);
3250 if (!latmctx->audio_mux_version_A) {
3252 if (audio_mux_version)
3253 latm_get_value(gb); // taraFullness
3255 skip_bits(gb, 1); // allStreamSameTimeFraming
3256 skip_bits(gb, 6); // numSubFrames
3258 if (get_bits(gb, 4)) { // numPrograms
3259 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3260 return AVERROR_PATCHWELCOME;
3263 // for each program (which there is only one in DVB)
3265 // for each layer (which there is only one in DVB)
3266 if (get_bits(gb, 3)) { // numLayer
3267 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3268 return AVERROR_PATCHWELCOME;
3271 // for all but first stream: use_same_config = get_bits(gb, 1);
3272 if (!audio_mux_version) {
3273 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3276 int ascLen = latm_get_value(gb);
3277 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3280 skip_bits_long(gb, ascLen);
3283 latmctx->frame_length_type = get_bits(gb, 3);
3284 switch (latmctx->frame_length_type) {
3286 skip_bits(gb, 8); // latmBufferFullness
3289 latmctx->frame_length = get_bits(gb, 9);
3294 skip_bits(gb, 6); // CELP frame length table index
3298 skip_bits(gb, 1); // HVXC frame length table index
3302 if (get_bits(gb, 1)) { // other data
3303 if (audio_mux_version) {
3304 latm_get_value(gb); // other_data_bits
3308 esc = get_bits(gb, 1);
3314 if (get_bits(gb, 1)) // crc present
3315 skip_bits(gb, 8); // config_crc
3321 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3325 if (ctx->frame_length_type == 0) {
3326 int mux_slot_length = 0;
3328 tmp = get_bits(gb, 8);
3329 mux_slot_length += tmp;
3330 } while (tmp == 255);
3331 return mux_slot_length;
3332 } else if (ctx->frame_length_type == 1) {
3333 return ctx->frame_length;
3334 } else if (ctx->frame_length_type == 3 ||
3335 ctx->frame_length_type == 5 ||
3336 ctx->frame_length_type == 7) {
3337 skip_bits(gb, 2); // mux_slot_length_coded
3342 static int read_audio_mux_element(struct LATMContext *latmctx,
3346 uint8_t use_same_mux = get_bits(gb, 1);
3347 if (!use_same_mux) {
3348 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3350 } else if (!latmctx->aac_ctx.avctx->extradata) {
3351 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3352 "no decoder config found\n");
3353 return AVERROR(EAGAIN);
3355 if (latmctx->audio_mux_version_A == 0) {
3356 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3357 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3358 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3359 return AVERROR_INVALIDDATA;
3360 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3361 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3362 "frame length mismatch %d << %d\n",
3363 mux_slot_length_bytes * 8, get_bits_left(gb));
3364 return AVERROR_INVALIDDATA;
3371 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3372 int *got_frame_ptr, AVPacket *avpkt)
3374 struct LATMContext *latmctx = avctx->priv_data;
3378 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3381 // check for LOAS sync word
3382 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3383 return AVERROR_INVALIDDATA;
3385 muxlength = get_bits(&gb, 13) + 3;
3386 // not enough data, the parser should have sorted this out
3387 if (muxlength > avpkt->size)
3388 return AVERROR_INVALIDDATA;
3390 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3393 if (!latmctx->initialized) {
3394 if (!avctx->extradata) {
3398 push_output_configuration(&latmctx->aac_ctx);
3399 if ((err = decode_audio_specific_config(
3400 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3401 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3402 pop_output_configuration(&latmctx->aac_ctx);
3405 latmctx->initialized = 1;
3409 if (show_bits(&gb, 12) == 0xfff) {
3410 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3411 "ADTS header detected, probably as result of configuration "
3413 return AVERROR_INVALIDDATA;
3416 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt)) < 0)
3422 static av_cold int latm_decode_init(AVCodecContext *avctx)
3424 struct LATMContext *latmctx = avctx->priv_data;
3425 int ret = aac_decode_init(avctx);
3427 if (avctx->extradata_size > 0)
3428 latmctx->initialized = !ret;
3433 static void aacdec_init(AACContext *c)
3435 c->imdct_and_windowing = imdct_and_windowing;
3436 c->apply_ltp = apply_ltp;
3437 c->apply_tns = apply_tns;
3438 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3439 c->update_ltp = update_ltp;
3442 ff_aacdec_init_mips(c);
3445 * AVOptions for Japanese DTV specific extensions (ADTS only)
3447 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3448 static const AVOption options[] = {
3449 {"dual_mono_mode", "Select the channel to decode for dual mono",
3450 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3451 AACDEC_FLAGS, "dual_mono_mode"},
3453 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3454 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3455 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3456 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3461 static const AVClass aac_decoder_class = {
3462 .class_name = "AAC decoder",
3463 .item_name = av_default_item_name,
3465 .version = LIBAVUTIL_VERSION_INT,
3468 AVCodec ff_aac_decoder = {
3470 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3471 .type = AVMEDIA_TYPE_AUDIO,
3472 .id = AV_CODEC_ID_AAC,
3473 .priv_data_size = sizeof(AACContext),
3474 .init = aac_decode_init,
3475 .close = aac_decode_close,
3476 .decode = aac_decode_frame,
3477 .sample_fmts = (const enum AVSampleFormat[]) {
3478 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3480 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3481 .channel_layouts = aac_channel_layout,
3483 .priv_class = &aac_decoder_class,
3487 Note: This decoder filter is intended to decode LATM streams transferred
3488 in MPEG transport streams which only contain one program.
3489 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3491 AVCodec ff_aac_latm_decoder = {
3493 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3494 .type = AVMEDIA_TYPE_AUDIO,
3495 .id = AV_CODEC_ID_AAC_LATM,
3496 .priv_data_size = sizeof(struct LATMContext),
3497 .init = latm_decode_init,
3498 .close = aac_decode_close,
3499 .decode = latm_decode_frame,
3500 .sample_fmts = (const enum AVSampleFormat[]) {
3501 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3503 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3504 .channel_layouts = aac_channel_layout,