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 * AAC decoder fixed-point implementation
13 * MIPS Technologies, Inc., California.
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU Lesser General Public
19 * License as published by the Free Software Foundation; either
20 * version 2.1 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * Lesser General Public License for more details.
27 * You should have received a copy of the GNU Lesser General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 * @author Oded Shimon ( ods15 ods15 dyndns org )
36 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * AAC decoder fixed-point implementation
39 * @author Stanislav Ocovaj ( stanislav.ocovaj imgtec com )
40 * @author Nedeljko Babic ( nedeljko.babic imgtec com )
47 * N (code in SoC repo) gain control
49 * Y window shapes - standard
50 * N window shapes - Low Delay
51 * Y filterbank - standard
52 * N (code in SoC repo) filterbank - Scalable Sample Rate
53 * Y Temporal Noise Shaping
54 * Y Long Term Prediction
57 * Y frequency domain prediction
58 * Y Perceptual Noise Substitution
60 * N Scalable Inverse AAC Quantization
61 * N Frequency Selective Switch
63 * Y quantization & coding - AAC
64 * N quantization & coding - TwinVQ
65 * N quantization & coding - BSAC
66 * N AAC Error Resilience tools
67 * N Error Resilience payload syntax
68 * N Error Protection tool
70 * N Silence Compression
73 * N Structured Audio tools
74 * N Structured Audio Sample Bank Format
76 * N Harmonic and Individual Lines plus Noise
77 * N Text-To-Speech Interface
78 * Y Spectral Band Replication
79 * Y (not in this code) Layer-1
80 * Y (not in this code) Layer-2
81 * Y (not in this code) Layer-3
82 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
84 * N Direct Stream Transfer
85 * Y (not in fixed point code) Enhanced AAC Low Delay (ER AAC ELD)
87 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
88 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
92 static VLC vlc_scalefactors;
93 static VLC vlc_spectral[11];
95 static int output_configure(AACContext *ac,
96 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
97 enum OCStatus oc_type, int get_new_frame);
99 #define overread_err "Input buffer exhausted before END element found\n"
101 static int count_channels(uint8_t (*layout)[3], int tags)
104 for (i = 0; i < tags; i++) {
105 int syn_ele = layout[i][0];
106 int pos = layout[i][2];
107 sum += (1 + (syn_ele == TYPE_CPE)) *
108 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
114 * Check for the channel element in the current channel position configuration.
115 * If it exists, make sure the appropriate element is allocated and map the
116 * channel order to match the internal FFmpeg channel layout.
118 * @param che_pos current channel position configuration
119 * @param type channel element type
120 * @param id channel element id
121 * @param channels count of the number of channels in the configuration
123 * @return Returns error status. 0 - OK, !0 - error
125 static av_cold int che_configure(AACContext *ac,
126 enum ChannelPosition che_pos,
127 int type, int id, int *channels)
129 if (*channels >= MAX_CHANNELS)
130 return AVERROR_INVALIDDATA;
132 if (!ac->che[type][id]) {
133 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
134 return AVERROR(ENOMEM);
135 AAC_RENAME(ff_aac_sbr_ctx_init)(ac, &ac->che[type][id]->sbr);
137 if (type != TYPE_CCE) {
138 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1))) {
139 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
140 return AVERROR_INVALIDDATA;
142 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[0];
143 if (type == TYPE_CPE ||
144 (type == TYPE_SCE && ac->oc[1].m4ac.ps == 1)) {
145 ac->output_element[(*channels)++] = &ac->che[type][id]->ch[1];
149 if (ac->che[type][id])
150 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][id]->sbr);
151 av_freep(&ac->che[type][id]);
156 static int frame_configure_elements(AVCodecContext *avctx)
158 AACContext *ac = avctx->priv_data;
159 int type, id, ch, ret;
161 /* set channel pointers to internal buffers by default */
162 for (type = 0; type < 4; type++) {
163 for (id = 0; id < MAX_ELEM_ID; id++) {
164 ChannelElement *che = ac->che[type][id];
166 che->ch[0].ret = che->ch[0].ret_buf;
167 che->ch[1].ret = che->ch[1].ret_buf;
172 /* get output buffer */
173 av_frame_unref(ac->frame);
174 if (!avctx->channels)
177 ac->frame->nb_samples = 2048;
178 if ((ret = ff_get_buffer(avctx, ac->frame, 0)) < 0)
181 /* map output channel pointers to AVFrame data */
182 for (ch = 0; ch < avctx->channels; ch++) {
183 if (ac->output_element[ch])
184 ac->output_element[ch]->ret = (INTFLOAT *)ac->frame->extended_data[ch];
190 struct elem_to_channel {
191 uint64_t av_position;
194 uint8_t aac_position;
197 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
198 uint8_t (*layout_map)[3], int offset, uint64_t left,
199 uint64_t right, int pos)
201 if (layout_map[offset][0] == TYPE_CPE) {
202 e2c_vec[offset] = (struct elem_to_channel) {
203 .av_position = left | right,
205 .elem_id = layout_map[offset][1],
210 e2c_vec[offset] = (struct elem_to_channel) {
213 .elem_id = layout_map[offset][1],
216 e2c_vec[offset + 1] = (struct elem_to_channel) {
217 .av_position = right,
219 .elem_id = layout_map[offset + 1][1],
226 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos,
229 int num_pos_channels = 0;
233 for (i = *current; i < tags; i++) {
234 if (layout_map[i][2] != pos)
236 if (layout_map[i][0] == TYPE_CPE) {
238 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
244 num_pos_channels += 2;
252 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
255 return num_pos_channels;
258 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
260 int i, n, total_non_cc_elements;
261 struct elem_to_channel e2c_vec[4 * MAX_ELEM_ID] = { { 0 } };
262 int num_front_channels, num_side_channels, num_back_channels;
265 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
270 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
271 if (num_front_channels < 0)
274 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
275 if (num_side_channels < 0)
278 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
279 if (num_back_channels < 0)
282 if (num_side_channels == 0 && num_back_channels >= 4) {
283 num_side_channels = 2;
284 num_back_channels -= 2;
288 if (num_front_channels & 1) {
289 e2c_vec[i] = (struct elem_to_channel) {
290 .av_position = AV_CH_FRONT_CENTER,
292 .elem_id = layout_map[i][1],
293 .aac_position = AAC_CHANNEL_FRONT
296 num_front_channels--;
298 if (num_front_channels >= 4) {
299 i += assign_pair(e2c_vec, layout_map, i,
300 AV_CH_FRONT_LEFT_OF_CENTER,
301 AV_CH_FRONT_RIGHT_OF_CENTER,
303 num_front_channels -= 2;
305 if (num_front_channels >= 2) {
306 i += assign_pair(e2c_vec, layout_map, i,
310 num_front_channels -= 2;
312 while (num_front_channels >= 2) {
313 i += assign_pair(e2c_vec, layout_map, i,
317 num_front_channels -= 2;
320 if (num_side_channels >= 2) {
321 i += assign_pair(e2c_vec, layout_map, i,
325 num_side_channels -= 2;
327 while (num_side_channels >= 2) {
328 i += assign_pair(e2c_vec, layout_map, i,
332 num_side_channels -= 2;
335 while (num_back_channels >= 4) {
336 i += assign_pair(e2c_vec, layout_map, i,
340 num_back_channels -= 2;
342 if (num_back_channels >= 2) {
343 i += assign_pair(e2c_vec, layout_map, i,
347 num_back_channels -= 2;
349 if (num_back_channels) {
350 e2c_vec[i] = (struct elem_to_channel) {
351 .av_position = AV_CH_BACK_CENTER,
353 .elem_id = layout_map[i][1],
354 .aac_position = AAC_CHANNEL_BACK
360 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
361 e2c_vec[i] = (struct elem_to_channel) {
362 .av_position = AV_CH_LOW_FREQUENCY,
364 .elem_id = layout_map[i][1],
365 .aac_position = AAC_CHANNEL_LFE
369 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
370 e2c_vec[i] = (struct elem_to_channel) {
371 .av_position = UINT64_MAX,
373 .elem_id = layout_map[i][1],
374 .aac_position = AAC_CHANNEL_LFE
379 // Must choose a stable sort
380 total_non_cc_elements = n = i;
383 for (i = 1; i < n; i++)
384 if (e2c_vec[i - 1].av_position > e2c_vec[i].av_position) {
385 FFSWAP(struct elem_to_channel, e2c_vec[i - 1], e2c_vec[i]);
392 for (i = 0; i < total_non_cc_elements; i++) {
393 layout_map[i][0] = e2c_vec[i].syn_ele;
394 layout_map[i][1] = e2c_vec[i].elem_id;
395 layout_map[i][2] = e2c_vec[i].aac_position;
396 if (e2c_vec[i].av_position != UINT64_MAX) {
397 layout |= e2c_vec[i].av_position;
405 * Save current output configuration if and only if it has been locked.
407 static void push_output_configuration(AACContext *ac) {
408 if (ac->oc[1].status == OC_LOCKED || ac->oc[0].status == OC_NONE) {
409 ac->oc[0] = ac->oc[1];
411 ac->oc[1].status = OC_NONE;
415 * Restore the previous output configuration if and only if the current
416 * configuration is unlocked.
418 static void pop_output_configuration(AACContext *ac) {
419 if (ac->oc[1].status != OC_LOCKED && ac->oc[0].status != OC_NONE) {
420 ac->oc[1] = ac->oc[0];
421 ac->avctx->channels = ac->oc[1].channels;
422 ac->avctx->channel_layout = ac->oc[1].channel_layout;
423 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
424 ac->oc[1].status, 0);
429 * Configure output channel order based on the current program
430 * configuration element.
432 * @return Returns error status. 0 - OK, !0 - error
434 static int output_configure(AACContext *ac,
435 uint8_t layout_map[MAX_ELEM_ID * 4][3], int tags,
436 enum OCStatus oc_type, int get_new_frame)
438 AVCodecContext *avctx = ac->avctx;
439 int i, channels = 0, ret;
441 uint8_t id_map[TYPE_END][MAX_ELEM_ID] = {{ 0 }};
442 uint8_t type_counts[TYPE_END] = { 0 };
444 if (ac->oc[1].layout_map != layout_map) {
445 memcpy(ac->oc[1].layout_map, layout_map, tags * sizeof(layout_map[0]));
446 ac->oc[1].layout_map_tags = tags;
448 for (i = 0; i < tags; i++) {
449 int type = layout_map[i][0];
450 int id = layout_map[i][1];
451 id_map[type][id] = type_counts[type]++;
453 // Try to sniff a reasonable channel order, otherwise output the
454 // channels in the order the PCE declared them.
455 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
456 layout = sniff_channel_order(layout_map, tags);
457 for (i = 0; i < tags; i++) {
458 int type = layout_map[i][0];
459 int id = layout_map[i][1];
460 int iid = id_map[type][id];
461 int position = layout_map[i][2];
462 // Allocate or free elements depending on if they are in the
463 // current program configuration.
464 ret = che_configure(ac, position, type, iid, &channels);
467 ac->tag_che_map[type][id] = ac->che[type][iid];
469 if (ac->oc[1].m4ac.ps == 1 && channels == 2) {
470 if (layout == AV_CH_FRONT_CENTER) {
471 layout = AV_CH_FRONT_LEFT|AV_CH_FRONT_RIGHT;
477 if (layout) avctx->channel_layout = layout;
478 ac->oc[1].channel_layout = layout;
479 avctx->channels = ac->oc[1].channels = channels;
480 ac->oc[1].status = oc_type;
483 if ((ret = frame_configure_elements(ac->avctx)) < 0)
490 static void flush(AVCodecContext *avctx)
492 AACContext *ac= avctx->priv_data;
495 for (type = 3; type >= 0; type--) {
496 for (i = 0; i < MAX_ELEM_ID; i++) {
497 ChannelElement *che = ac->che[type][i];
499 for (j = 0; j <= 1; j++) {
500 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
508 * Set up channel positions based on a default channel configuration
509 * as specified in table 1.17.
511 * @return Returns error status. 0 - OK, !0 - error
513 static int set_default_channel_config(AVCodecContext *avctx,
514 uint8_t (*layout_map)[3],
518 if (channel_config < 1 || (channel_config > 7 && channel_config < 11) ||
519 channel_config > 12) {
520 av_log(avctx, AV_LOG_ERROR,
521 "invalid default channel configuration (%d)\n",
523 return AVERROR_INVALIDDATA;
525 *tags = tags_per_config[channel_config];
526 memcpy(layout_map, aac_channel_layout_map[channel_config - 1],
527 *tags * sizeof(*layout_map));
530 * AAC specification has 7.1(wide) as a default layout for 8-channel streams.
531 * However, at least Nero AAC encoder encodes 7.1 streams using the default
532 * channel config 7, mapping the side channels of the original audio stream
533 * to the second AAC_CHANNEL_FRONT pair in the AAC stream. Similarly, e.g. FAAD
534 * decodes the second AAC_CHANNEL_FRONT pair as side channels, therefore decoding
535 * the incorrect streams as if they were correct (and as the encoder intended).
537 * As actual intended 7.1(wide) streams are very rare, default to assuming a
538 * 7.1 layout was intended.
540 if (channel_config == 7 && avctx->strict_std_compliance < FF_COMPLIANCE_STRICT) {
541 av_log(avctx, AV_LOG_INFO, "Assuming an incorrectly encoded 7.1 channel layout"
542 " instead of a spec-compliant 7.1(wide) layout, use -strict %d to decode"
543 " according to the specification instead.\n", FF_COMPLIANCE_STRICT);
544 layout_map[2][2] = AAC_CHANNEL_SIDE;
550 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
552 /* For PCE based channel configurations map the channels solely based
554 if (!ac->oc[1].m4ac.chan_config) {
555 return ac->tag_che_map[type][elem_id];
557 // Allow single CPE stereo files to be signalled with mono configuration.
558 if (!ac->tags_mapped && type == TYPE_CPE &&
559 ac->oc[1].m4ac.chan_config == 1) {
560 uint8_t layout_map[MAX_ELEM_ID*4][3];
562 push_output_configuration(ac);
564 av_log(ac->avctx, AV_LOG_DEBUG, "mono with CPE\n");
566 if (set_default_channel_config(ac->avctx, layout_map,
567 &layout_map_tags, 2) < 0)
569 if (output_configure(ac, layout_map, layout_map_tags,
570 OC_TRIAL_FRAME, 1) < 0)
573 ac->oc[1].m4ac.chan_config = 2;
574 ac->oc[1].m4ac.ps = 0;
577 if (!ac->tags_mapped && type == TYPE_SCE &&
578 ac->oc[1].m4ac.chan_config == 2) {
579 uint8_t layout_map[MAX_ELEM_ID * 4][3];
581 push_output_configuration(ac);
583 av_log(ac->avctx, AV_LOG_DEBUG, "stereo with SCE\n");
585 if (set_default_channel_config(ac->avctx, layout_map,
586 &layout_map_tags, 1) < 0)
588 if (output_configure(ac, layout_map, layout_map_tags,
589 OC_TRIAL_FRAME, 1) < 0)
592 ac->oc[1].m4ac.chan_config = 1;
593 if (ac->oc[1].m4ac.sbr)
594 ac->oc[1].m4ac.ps = -1;
596 /* For indexed channel configurations map the channels solely based
598 switch (ac->oc[1].m4ac.chan_config) {
601 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
603 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
606 if (ac->tags_mapped == 2 &&
607 ac->oc[1].m4ac.chan_config == 11 &&
610 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
613 /* Some streams incorrectly code 5.1 audio as
614 * SCE[0] CPE[0] CPE[1] SCE[1]
616 * SCE[0] CPE[0] CPE[1] LFE[0].
617 * If we seem to have encountered such a stream, transfer
618 * the LFE[0] element to the SCE[1]'s mapping */
619 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
620 if (!ac->warned_remapping_once && (type != TYPE_LFE || elem_id != 0)) {
621 av_log(ac->avctx, AV_LOG_WARNING,
622 "This stream seems to incorrectly report its last channel as %s[%d], mapping to LFE[0]\n",
623 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
624 ac->warned_remapping_once++;
627 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
630 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
632 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
635 /* Some streams incorrectly code 4.0 audio as
636 * SCE[0] CPE[0] LFE[0]
638 * SCE[0] CPE[0] SCE[1].
639 * If we seem to have encountered such a stream, transfer
640 * the SCE[1] element to the LFE[0]'s mapping */
641 if (ac->tags_mapped == tags_per_config[ac->oc[1].m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
642 if (!ac->warned_remapping_once && (type != TYPE_SCE || elem_id != 1)) {
643 av_log(ac->avctx, AV_LOG_WARNING,
644 "This stream seems to incorrectly report its last channel as %s[%d], mapping to SCE[1]\n",
645 type == TYPE_SCE ? "SCE" : "LFE", elem_id);
646 ac->warned_remapping_once++;
649 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_SCE][1];
651 if (ac->tags_mapped == 2 &&
652 ac->oc[1].m4ac.chan_config == 4 &&
655 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
659 if (ac->tags_mapped == (ac->oc[1].m4ac.chan_config != 2) &&
662 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
663 } else if (ac->oc[1].m4ac.chan_config == 2) {
667 if (!ac->tags_mapped && type == TYPE_SCE) {
669 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
677 * Decode an array of 4 bit element IDs, optionally interleaved with a
678 * stereo/mono switching bit.
680 * @param type speaker type/position for these channels
682 static void decode_channel_map(uint8_t layout_map[][3],
683 enum ChannelPosition type,
684 GetBitContext *gb, int n)
687 enum RawDataBlockType syn_ele;
689 case AAC_CHANNEL_FRONT:
690 case AAC_CHANNEL_BACK:
691 case AAC_CHANNEL_SIDE:
692 syn_ele = get_bits1(gb);
698 case AAC_CHANNEL_LFE:
702 // AAC_CHANNEL_OFF has no channel map
705 layout_map[0][0] = syn_ele;
706 layout_map[0][1] = get_bits(gb, 4);
707 layout_map[0][2] = type;
713 * Decode program configuration element; reference: table 4.2.
715 * @return Returns error status. 0 - OK, !0 - error
717 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
718 uint8_t (*layout_map)[3],
721 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
726 skip_bits(gb, 2); // object_type
728 sampling_index = get_bits(gb, 4);
729 if (m4ac->sampling_index != sampling_index)
730 av_log(avctx, AV_LOG_WARNING,
731 "Sample rate index in program config element does not "
732 "match the sample rate index configured by the container.\n");
734 num_front = get_bits(gb, 4);
735 num_side = get_bits(gb, 4);
736 num_back = get_bits(gb, 4);
737 num_lfe = get_bits(gb, 2);
738 num_assoc_data = get_bits(gb, 3);
739 num_cc = get_bits(gb, 4);
742 skip_bits(gb, 4); // mono_mixdown_tag
744 skip_bits(gb, 4); // stereo_mixdown_tag
747 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
749 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
750 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
753 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
755 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
757 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
759 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
762 skip_bits_long(gb, 4 * num_assoc_data);
764 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
769 /* comment field, first byte is length */
770 comment_len = get_bits(gb, 8) * 8;
771 if (get_bits_left(gb) < comment_len) {
772 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
773 return AVERROR_INVALIDDATA;
775 skip_bits_long(gb, comment_len);
780 * Decode GA "General Audio" specific configuration; reference: table 4.1.
782 * @param ac pointer to AACContext, may be null
783 * @param avctx pointer to AVCCodecContext, used for logging
785 * @return Returns error status. 0 - OK, !0 - error
787 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
789 MPEG4AudioConfig *m4ac,
792 int extension_flag, ret, ep_config, res_flags;
793 uint8_t layout_map[MAX_ELEM_ID*4][3];
796 if (get_bits1(gb)) { // frameLengthFlag
797 avpriv_request_sample(avctx, "960/120 MDCT window");
798 return AVERROR_PATCHWELCOME;
800 m4ac->frame_length_short = 0;
802 if (get_bits1(gb)) // dependsOnCoreCoder
803 skip_bits(gb, 14); // coreCoderDelay
804 extension_flag = get_bits1(gb);
806 if (m4ac->object_type == AOT_AAC_SCALABLE ||
807 m4ac->object_type == AOT_ER_AAC_SCALABLE)
808 skip_bits(gb, 3); // layerNr
810 if (channel_config == 0) {
811 skip_bits(gb, 4); // element_instance_tag
812 tags = decode_pce(avctx, m4ac, layout_map, gb);
816 if ((ret = set_default_channel_config(avctx, layout_map,
817 &tags, channel_config)))
821 if (count_channels(layout_map, tags) > 1) {
823 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
826 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
829 if (extension_flag) {
830 switch (m4ac->object_type) {
832 skip_bits(gb, 5); // numOfSubFrame
833 skip_bits(gb, 11); // layer_length
837 case AOT_ER_AAC_SCALABLE:
839 res_flags = get_bits(gb, 3);
841 avpriv_report_missing_feature(avctx,
842 "AAC data resilience (flags %x)",
844 return AVERROR_PATCHWELCOME;
848 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
850 switch (m4ac->object_type) {
853 case AOT_ER_AAC_SCALABLE:
855 ep_config = get_bits(gb, 2);
857 avpriv_report_missing_feature(avctx,
858 "epConfig %d", ep_config);
859 return AVERROR_PATCHWELCOME;
865 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
867 MPEG4AudioConfig *m4ac,
870 int ret, ep_config, res_flags;
871 uint8_t layout_map[MAX_ELEM_ID*4][3];
873 const int ELDEXT_TERM = 0;
878 if (get_bits1(gb)) { // frameLengthFlag
879 avpriv_request_sample(avctx, "960/120 MDCT window");
880 return AVERROR_PATCHWELCOME;
883 m4ac->frame_length_short = get_bits1(gb);
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(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 ff_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
950 for (i = 0; i < bit_size >> 3; i++)
951 ff_dlog(avctx, "%02x ", data[i]);
952 ff_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 void reset_all_predictors(PredictorState *ps)
1024 for (i = 0; i < MAX_PREDICTORS; i++)
1025 reset_predict_state(&ps[i]);
1028 static int sample_rate_idx (int rate)
1030 if (92017 <= rate) return 0;
1031 else if (75132 <= rate) return 1;
1032 else if (55426 <= rate) return 2;
1033 else if (46009 <= rate) return 3;
1034 else if (37566 <= rate) return 4;
1035 else if (27713 <= rate) return 5;
1036 else if (23004 <= rate) return 6;
1037 else if (18783 <= rate) return 7;
1038 else if (13856 <= rate) return 8;
1039 else if (11502 <= rate) return 9;
1040 else if (9391 <= rate) return 10;
1044 static void reset_predictor_group(PredictorState *ps, int group_num)
1047 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1048 reset_predict_state(&ps[i]);
1051 #define AAC_INIT_VLC_STATIC(num, size) \
1052 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1053 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1054 sizeof(ff_aac_spectral_bits[num][0]), \
1055 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1056 sizeof(ff_aac_spectral_codes[num][0]), \
1059 static void aacdec_init(AACContext *ac);
1061 static av_cold int aac_decode_init(AVCodecContext *avctx)
1063 AACContext *ac = avctx->priv_data;
1067 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1071 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
1073 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1074 #endif /* USE_FIXED */
1076 if (avctx->extradata_size > 0) {
1077 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1079 avctx->extradata_size * 8,
1084 uint8_t layout_map[MAX_ELEM_ID*4][3];
1085 int layout_map_tags;
1087 sr = sample_rate_idx(avctx->sample_rate);
1088 ac->oc[1].m4ac.sampling_index = sr;
1089 ac->oc[1].m4ac.channels = avctx->channels;
1090 ac->oc[1].m4ac.sbr = -1;
1091 ac->oc[1].m4ac.ps = -1;
1093 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1094 if (ff_mpeg4audio_channels[i] == avctx->channels)
1096 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1099 ac->oc[1].m4ac.chan_config = i;
1101 if (ac->oc[1].m4ac.chan_config) {
1102 int ret = set_default_channel_config(avctx, layout_map,
1103 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1105 output_configure(ac, layout_map, layout_map_tags,
1107 else if (avctx->err_recognition & AV_EF_EXPLODE)
1108 return AVERROR_INVALIDDATA;
1112 if (avctx->channels > MAX_CHANNELS) {
1113 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1114 return AVERROR_INVALIDDATA;
1117 AAC_INIT_VLC_STATIC( 0, 304);
1118 AAC_INIT_VLC_STATIC( 1, 270);
1119 AAC_INIT_VLC_STATIC( 2, 550);
1120 AAC_INIT_VLC_STATIC( 3, 300);
1121 AAC_INIT_VLC_STATIC( 4, 328);
1122 AAC_INIT_VLC_STATIC( 5, 294);
1123 AAC_INIT_VLC_STATIC( 6, 306);
1124 AAC_INIT_VLC_STATIC( 7, 268);
1125 AAC_INIT_VLC_STATIC( 8, 510);
1126 AAC_INIT_VLC_STATIC( 9, 366);
1127 AAC_INIT_VLC_STATIC(10, 462);
1129 AAC_RENAME(ff_aac_sbr_init)();
1132 ac->fdsp = avpriv_alloc_fixed_dsp(avctx->flags & CODEC_FLAG_BITEXACT);
1134 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
1135 #endif /* USE_FIXED */
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 AAC_RENAME_32(ff_mdct_init)(&ac->mdct, 11, 1, 1.0 / RANGE15(1024.0));
1155 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ld, 10, 1, 1.0 / RANGE15(512.0));
1156 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small, 8, 1, 1.0 / RANGE15(128.0));
1157 AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp, 11, 0, RANGE15(-2.0));
1159 ret = ff_imdct15_init(&ac->mdct480, 5);
1163 // window initialization
1164 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_long_1024), 4.0, 1024);
1165 AAC_RENAME(ff_kbd_window_init)(AAC_RENAME(ff_aac_kbd_short_128), 6.0, 128);
1166 AAC_RENAME(ff_init_ff_sine_windows)(10);
1167 AAC_RENAME(ff_init_ff_sine_windows)( 9);
1168 AAC_RENAME(ff_init_ff_sine_windows)( 7);
1170 AAC_RENAME(cbrt_tableinit)();
1176 * Skip data_stream_element; reference: table 4.10.
1178 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1180 int byte_align = get_bits1(gb);
1181 int count = get_bits(gb, 8);
1183 count += get_bits(gb, 8);
1187 if (get_bits_left(gb) < 8 * count) {
1188 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1189 return AVERROR_INVALIDDATA;
1191 skip_bits_long(gb, 8 * count);
1195 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1199 if (get_bits1(gb)) {
1200 ics->predictor_reset_group = get_bits(gb, 5);
1201 if (ics->predictor_reset_group == 0 ||
1202 ics->predictor_reset_group > 30) {
1203 av_log(ac->avctx, AV_LOG_ERROR,
1204 "Invalid Predictor Reset Group.\n");
1205 return AVERROR_INVALIDDATA;
1208 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1209 ics->prediction_used[sfb] = get_bits1(gb);
1215 * Decode Long Term Prediction data; reference: table 4.xx.
1217 static void decode_ltp(LongTermPrediction *ltp,
1218 GetBitContext *gb, uint8_t max_sfb)
1222 ltp->lag = get_bits(gb, 11);
1223 ltp->coef = ltp_coef[get_bits(gb, 3)];
1224 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1225 ltp->used[sfb] = get_bits1(gb);
1229 * Decode Individual Channel Stream info; reference: table 4.6.
1231 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1234 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1235 const int aot = m4ac->object_type;
1236 const int sampling_index = m4ac->sampling_index;
1237 if (aot != AOT_ER_AAC_ELD) {
1238 if (get_bits1(gb)) {
1239 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
1240 if (ac->avctx->err_recognition & AV_EF_BITSTREAM)
1241 return AVERROR_INVALIDDATA;
1243 ics->window_sequence[1] = ics->window_sequence[0];
1244 ics->window_sequence[0] = get_bits(gb, 2);
1245 if (aot == AOT_ER_AAC_LD &&
1246 ics->window_sequence[0] != ONLY_LONG_SEQUENCE) {
1247 av_log(ac->avctx, AV_LOG_ERROR,
1248 "AAC LD is only defined for ONLY_LONG_SEQUENCE but "
1249 "window sequence %d found.\n", ics->window_sequence[0]);
1250 ics->window_sequence[0] = ONLY_LONG_SEQUENCE;
1251 return AVERROR_INVALIDDATA;
1253 ics->use_kb_window[1] = ics->use_kb_window[0];
1254 ics->use_kb_window[0] = get_bits1(gb);
1256 ics->num_window_groups = 1;
1257 ics->group_len[0] = 1;
1258 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1260 ics->max_sfb = get_bits(gb, 4);
1261 for (i = 0; i < 7; i++) {
1262 if (get_bits1(gb)) {
1263 ics->group_len[ics->num_window_groups - 1]++;
1265 ics->num_window_groups++;
1266 ics->group_len[ics->num_window_groups - 1] = 1;
1269 ics->num_windows = 8;
1270 ics->swb_offset = ff_swb_offset_128[sampling_index];
1271 ics->num_swb = ff_aac_num_swb_128[sampling_index];
1272 ics->tns_max_bands = ff_tns_max_bands_128[sampling_index];
1273 ics->predictor_present = 0;
1275 ics->max_sfb = get_bits(gb, 6);
1276 ics->num_windows = 1;
1277 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD) {
1278 if (m4ac->frame_length_short) {
1279 ics->swb_offset = ff_swb_offset_480[sampling_index];
1280 ics->num_swb = ff_aac_num_swb_480[sampling_index];
1281 ics->tns_max_bands = ff_tns_max_bands_480[sampling_index];
1283 ics->swb_offset = ff_swb_offset_512[sampling_index];
1284 ics->num_swb = ff_aac_num_swb_512[sampling_index];
1285 ics->tns_max_bands = ff_tns_max_bands_512[sampling_index];
1287 if (!ics->num_swb || !ics->swb_offset)
1290 ics->swb_offset = ff_swb_offset_1024[sampling_index];
1291 ics->num_swb = ff_aac_num_swb_1024[sampling_index];
1292 ics->tns_max_bands = ff_tns_max_bands_1024[sampling_index];
1294 if (aot != AOT_ER_AAC_ELD) {
1295 ics->predictor_present = get_bits1(gb);
1296 ics->predictor_reset_group = 0;
1298 if (ics->predictor_present) {
1299 if (aot == AOT_AAC_MAIN) {
1300 if (decode_prediction(ac, ics, gb)) {
1303 } else if (aot == AOT_AAC_LC ||
1304 aot == AOT_ER_AAC_LC) {
1305 av_log(ac->avctx, AV_LOG_ERROR,
1306 "Prediction is not allowed in AAC-LC.\n");
1309 if (aot == AOT_ER_AAC_LD) {
1310 av_log(ac->avctx, AV_LOG_ERROR,
1311 "LTP in ER AAC LD not yet implemented.\n");
1312 return AVERROR_PATCHWELCOME;
1314 if ((ics->ltp.present = get_bits(gb, 1)))
1315 decode_ltp(&ics->ltp, gb, ics->max_sfb);
1320 if (ics->max_sfb > ics->num_swb) {
1321 av_log(ac->avctx, AV_LOG_ERROR,
1322 "Number of scalefactor bands in group (%d) "
1323 "exceeds limit (%d).\n",
1324 ics->max_sfb, ics->num_swb);
1331 return AVERROR_INVALIDDATA;
1335 * Decode band types (section_data payload); reference: table 4.46.
1337 * @param band_type array of the used band type
1338 * @param band_type_run_end array of the last scalefactor band of a band type run
1340 * @return Returns error status. 0 - OK, !0 - error
1342 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1343 int band_type_run_end[120], GetBitContext *gb,
1344 IndividualChannelStream *ics)
1347 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1348 for (g = 0; g < ics->num_window_groups; g++) {
1350 while (k < ics->max_sfb) {
1351 uint8_t sect_end = k;
1353 int sect_band_type = get_bits(gb, 4);
1354 if (sect_band_type == 12) {
1355 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1356 return AVERROR_INVALIDDATA;
1359 sect_len_incr = get_bits(gb, bits);
1360 sect_end += sect_len_incr;
1361 if (get_bits_left(gb) < 0) {
1362 av_log(ac->avctx, AV_LOG_ERROR, "decode_band_types: "overread_err);
1363 return AVERROR_INVALIDDATA;
1365 if (sect_end > ics->max_sfb) {
1366 av_log(ac->avctx, AV_LOG_ERROR,
1367 "Number of bands (%d) exceeds limit (%d).\n",
1368 sect_end, ics->max_sfb);
1369 return AVERROR_INVALIDDATA;
1371 } while (sect_len_incr == (1 << bits) - 1);
1372 for (; k < sect_end; k++) {
1373 band_type [idx] = sect_band_type;
1374 band_type_run_end[idx++] = sect_end;
1382 * Decode scalefactors; reference: table 4.47.
1384 * @param global_gain first scalefactor value as scalefactors are differentially coded
1385 * @param band_type array of the used band type
1386 * @param band_type_run_end array of the last scalefactor band of a band type run
1387 * @param sf array of scalefactors or intensity stereo positions
1389 * @return Returns error status. 0 - OK, !0 - error
1391 static int decode_scalefactors(AACContext *ac, INTFLOAT sf[120], GetBitContext *gb,
1392 unsigned int global_gain,
1393 IndividualChannelStream *ics,
1394 enum BandType band_type[120],
1395 int band_type_run_end[120])
1398 int offset[3] = { global_gain, global_gain - NOISE_OFFSET, 0 };
1401 for (g = 0; g < ics->num_window_groups; g++) {
1402 for (i = 0; i < ics->max_sfb;) {
1403 int run_end = band_type_run_end[idx];
1404 if (band_type[idx] == ZERO_BT) {
1405 for (; i < run_end; i++, idx++)
1407 } else if ((band_type[idx] == INTENSITY_BT) ||
1408 (band_type[idx] == INTENSITY_BT2)) {
1409 for (; i < run_end; i++, idx++) {
1410 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1411 clipped_offset = av_clip(offset[2], -155, 100);
1412 if (offset[2] != clipped_offset) {
1413 avpriv_request_sample(ac->avctx,
1414 "If you heard an audible artifact, there may be a bug in the decoder. "
1415 "Clipped intensity stereo position (%d -> %d)",
1416 offset[2], clipped_offset);
1419 sf[idx] = 100 - clipped_offset;
1421 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1422 #endif /* USE_FIXED */
1424 } else if (band_type[idx] == NOISE_BT) {
1425 for (; i < run_end; i++, idx++) {
1426 if (noise_flag-- > 0)
1427 offset[1] += get_bits(gb, NOISE_PRE_BITS) - NOISE_PRE;
1429 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1430 clipped_offset = av_clip(offset[1], -100, 155);
1431 if (offset[1] != clipped_offset) {
1432 avpriv_request_sample(ac->avctx,
1433 "If you heard an audible artifact, there may be a bug in the decoder. "
1434 "Clipped noise gain (%d -> %d)",
1435 offset[1], clipped_offset);
1438 sf[idx] = -(100 + clipped_offset);
1440 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1441 #endif /* USE_FIXED */
1444 for (; i < run_end; i++, idx++) {
1445 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - SCALE_DIFF_ZERO;
1446 if (offset[0] > 255U) {
1447 av_log(ac->avctx, AV_LOG_ERROR,
1448 "Scalefactor (%d) out of range.\n", offset[0]);
1449 return AVERROR_INVALIDDATA;
1452 sf[idx] = -offset[0];
1454 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1455 #endif /* USE_FIXED */
1464 * Decode pulse data; reference: table 4.7.
1466 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1467 const uint16_t *swb_offset, int num_swb)
1470 pulse->num_pulse = get_bits(gb, 2) + 1;
1471 pulse_swb = get_bits(gb, 6);
1472 if (pulse_swb >= num_swb)
1474 pulse->pos[0] = swb_offset[pulse_swb];
1475 pulse->pos[0] += get_bits(gb, 5);
1476 if (pulse->pos[0] >= swb_offset[num_swb])
1478 pulse->amp[0] = get_bits(gb, 4);
1479 for (i = 1; i < pulse->num_pulse; i++) {
1480 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1481 if (pulse->pos[i] >= swb_offset[num_swb])
1483 pulse->amp[i] = get_bits(gb, 4);
1489 * Decode Temporal Noise Shaping data; reference: table 4.48.
1491 * @return Returns error status. 0 - OK, !0 - error
1493 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1494 GetBitContext *gb, const IndividualChannelStream *ics)
1496 int w, filt, i, coef_len, coef_res, coef_compress;
1497 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1498 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1499 for (w = 0; w < ics->num_windows; w++) {
1500 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1501 coef_res = get_bits1(gb);
1503 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1505 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1507 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1508 av_log(ac->avctx, AV_LOG_ERROR,
1509 "TNS filter order %d is greater than maximum %d.\n",
1510 tns->order[w][filt], tns_max_order);
1511 tns->order[w][filt] = 0;
1512 return AVERROR_INVALIDDATA;
1514 if (tns->order[w][filt]) {
1515 tns->direction[w][filt] = get_bits1(gb);
1516 coef_compress = get_bits1(gb);
1517 coef_len = coef_res + 3 - coef_compress;
1518 tmp2_idx = 2 * coef_compress + coef_res;
1520 for (i = 0; i < tns->order[w][filt]; i++)
1521 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1530 * Decode Mid/Side data; reference: table 4.54.
1532 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1533 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1534 * [3] reserved for scalable AAC
1536 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1540 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1541 if (ms_present == 1) {
1542 for (idx = 0; idx < max_idx; idx++)
1543 cpe->ms_mask[idx] = get_bits1(gb);
1544 } else if (ms_present == 2) {
1545 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1550 * Decode spectral data; reference: table 4.50.
1551 * Dequantize and scale spectral data; reference: 4.6.3.3.
1553 * @param coef array of dequantized, scaled spectral data
1554 * @param sf array of scalefactors or intensity stereo positions
1555 * @param pulse_present set if pulses are present
1556 * @param pulse pointer to pulse data struct
1557 * @param band_type array of the used band type
1559 * @return Returns error status. 0 - OK, !0 - error
1561 static int decode_spectrum_and_dequant(AACContext *ac, INTFLOAT coef[1024],
1562 GetBitContext *gb, const INTFLOAT sf[120],
1563 int pulse_present, const Pulse *pulse,
1564 const IndividualChannelStream *ics,
1565 enum BandType band_type[120])
1567 int i, k, g, idx = 0;
1568 const int c = 1024 / ics->num_windows;
1569 const uint16_t *offsets = ics->swb_offset;
1570 INTFLOAT *coef_base = coef;
1572 for (g = 0; g < ics->num_windows; g++)
1573 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1574 sizeof(INTFLOAT) * (c - offsets[ics->max_sfb]));
1576 for (g = 0; g < ics->num_window_groups; g++) {
1577 unsigned g_len = ics->group_len[g];
1579 for (i = 0; i < ics->max_sfb; i++, idx++) {
1580 const unsigned cbt_m1 = band_type[idx] - 1;
1581 INTFLOAT *cfo = coef + offsets[i];
1582 int off_len = offsets[i + 1] - offsets[i];
1585 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1586 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1587 memset(cfo, 0, off_len * sizeof(*cfo));
1589 } else if (cbt_m1 == NOISE_BT - 1) {
1590 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1593 #endif /* !USE_FIXED */
1594 INTFLOAT band_energy;
1596 for (k = 0; k < off_len; k++) {
1597 ac->random_state = lcg_random(ac->random_state);
1599 cfo[k] = ac->random_state >> 3;
1601 cfo[k] = ac->random_state;
1602 #endif /* USE_FIXED */
1606 band_energy = ac->fdsp->scalarproduct_fixed(cfo, cfo, off_len);
1607 band_energy = fixed_sqrt(band_energy, 31);
1608 noise_scale(cfo, sf[idx], band_energy, off_len);
1610 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1611 scale = sf[idx] / sqrtf(band_energy);
1612 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1613 #endif /* USE_FIXED */
1617 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1618 #endif /* !USE_FIXED */
1619 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1620 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1621 OPEN_READER(re, gb);
1623 switch (cbt_m1 >> 1) {
1625 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1633 UPDATE_CACHE(re, gb);
1634 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1635 cb_idx = cb_vector_idx[code];
1637 cf = DEC_SQUAD(cf, cb_idx);
1639 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1640 #endif /* USE_FIXED */
1646 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1656 UPDATE_CACHE(re, gb);
1657 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1658 cb_idx = cb_vector_idx[code];
1659 nnz = cb_idx >> 8 & 15;
1660 bits = nnz ? GET_CACHE(re, gb) : 0;
1661 LAST_SKIP_BITS(re, gb, nnz);
1663 cf = DEC_UQUAD(cf, cb_idx, bits);
1665 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1666 #endif /* USE_FIXED */
1672 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1680 UPDATE_CACHE(re, gb);
1681 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1682 cb_idx = cb_vector_idx[code];
1684 cf = DEC_SPAIR(cf, cb_idx);
1686 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1687 #endif /* USE_FIXED */
1694 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1704 UPDATE_CACHE(re, gb);
1705 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1706 cb_idx = cb_vector_idx[code];
1707 nnz = cb_idx >> 8 & 15;
1708 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1709 LAST_SKIP_BITS(re, gb, nnz);
1711 cf = DEC_UPAIR(cf, cb_idx, sign);
1713 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1714 #endif /* USE_FIXED */
1720 for (group = 0; group < (AAC_SIGNE)g_len; group++, cfo+=128) {
1726 uint32_t *icf = (uint32_t *) cf;
1727 #endif /* USE_FIXED */
1737 UPDATE_CACHE(re, gb);
1738 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1746 cb_idx = cb_vector_idx[code];
1749 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1750 LAST_SKIP_BITS(re, gb, nnz);
1752 for (j = 0; j < 2; j++) {
1756 /* The total length of escape_sequence must be < 22 bits according
1757 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1758 UPDATE_CACHE(re, gb);
1759 b = GET_CACHE(re, gb);
1760 b = 31 - av_log2(~b);
1763 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1764 return AVERROR_INVALIDDATA;
1767 SKIP_BITS(re, gb, b + 1);
1769 n = (1 << b) + SHOW_UBITS(re, gb, b);
1770 LAST_SKIP_BITS(re, gb, b);
1777 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1778 #endif /* USE_FIXED */
1787 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1788 *icf++ = (bits & 1U<<31) | v;
1789 #endif /* USE_FIXED */
1796 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1797 #endif /* !USE_FIXED */
1801 CLOSE_READER(re, gb);
1807 if (pulse_present) {
1809 for (i = 0; i < pulse->num_pulse; i++) {
1810 INTFLOAT co = coef_base[ pulse->pos[i] ];
1811 while (offsets[idx + 1] <= pulse->pos[i])
1813 if (band_type[idx] != NOISE_BT && sf[idx]) {
1814 INTFLOAT ico = -pulse->amp[i];
1817 ico = co + (co > 0 ? -ico : ico);
1819 coef_base[ pulse->pos[i] ] = ico;
1823 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1825 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1826 #endif /* USE_FIXED */
1833 for (g = 0; g < ics->num_window_groups; g++) {
1834 unsigned g_len = ics->group_len[g];
1836 for (i = 0; i < ics->max_sfb; i++, idx++) {
1837 const unsigned cbt_m1 = band_type[idx] - 1;
1838 int *cfo = coef + offsets[i];
1839 int off_len = offsets[i + 1] - offsets[i];
1842 if (cbt_m1 < NOISE_BT - 1) {
1843 for (group = 0; group < (int)g_len; group++, cfo+=128) {
1844 ac->vector_pow43(cfo, off_len);
1845 ac->subband_scale(cfo, cfo, sf[idx], 34, off_len);
1851 #endif /* USE_FIXED */
1856 * Apply AAC-Main style frequency domain prediction.
1858 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1862 if (!sce->ics.predictor_initialized) {
1863 reset_all_predictors(sce->predictor_state);
1864 sce->ics.predictor_initialized = 1;
1867 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1869 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1871 for (k = sce->ics.swb_offset[sfb];
1872 k < sce->ics.swb_offset[sfb + 1];
1874 predict(&sce->predictor_state[k], &sce->coeffs[k],
1875 sce->ics.predictor_present &&
1876 sce->ics.prediction_used[sfb]);
1879 if (sce->ics.predictor_reset_group)
1880 reset_predictor_group(sce->predictor_state,
1881 sce->ics.predictor_reset_group);
1883 reset_all_predictors(sce->predictor_state);
1887 * Decode an individual_channel_stream payload; reference: table 4.44.
1889 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1890 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1892 * @return Returns error status. 0 - OK, !0 - error
1894 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1895 GetBitContext *gb, int common_window, int scale_flag)
1898 TemporalNoiseShaping *tns = &sce->tns;
1899 IndividualChannelStream *ics = &sce->ics;
1900 INTFLOAT *out = sce->coeffs;
1901 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1904 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1905 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1906 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1907 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1908 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1910 /* This assignment is to silence a GCC warning about the variable being used
1911 * uninitialized when in fact it always is.
1913 pulse.num_pulse = 0;
1915 global_gain = get_bits(gb, 8);
1917 if (!common_window && !scale_flag) {
1918 if (decode_ics_info(ac, ics, gb) < 0)
1919 return AVERROR_INVALIDDATA;
1922 if ((ret = decode_band_types(ac, sce->band_type,
1923 sce->band_type_run_end, gb, ics)) < 0)
1925 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1926 sce->band_type, sce->band_type_run_end)) < 0)
1931 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1932 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1933 av_log(ac->avctx, AV_LOG_ERROR,
1934 "Pulse tool not allowed in eight short sequence.\n");
1935 return AVERROR_INVALIDDATA;
1937 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1938 av_log(ac->avctx, AV_LOG_ERROR,
1939 "Pulse data corrupt or invalid.\n");
1940 return AVERROR_INVALIDDATA;
1943 tns->present = get_bits1(gb);
1944 if (tns->present && !er_syntax)
1945 if (decode_tns(ac, tns, gb, ics) < 0)
1946 return AVERROR_INVALIDDATA;
1947 if (!eld_syntax && get_bits1(gb)) {
1948 avpriv_request_sample(ac->avctx, "SSR");
1949 return AVERROR_PATCHWELCOME;
1951 // I see no textual basis in the spec for this occurring after SSR gain
1952 // control, but this is what both reference and real implmentations do
1953 if (tns->present && er_syntax)
1954 if (decode_tns(ac, tns, gb, ics) < 0)
1955 return AVERROR_INVALIDDATA;
1958 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1959 &pulse, ics, sce->band_type) < 0)
1960 return AVERROR_INVALIDDATA;
1962 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1963 apply_prediction(ac, sce);
1969 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1971 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1973 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1974 INTFLOAT *ch0 = cpe->ch[0].coeffs;
1975 INTFLOAT *ch1 = cpe->ch[1].coeffs;
1976 int g, i, group, idx = 0;
1977 const uint16_t *offsets = ics->swb_offset;
1978 for (g = 0; g < ics->num_window_groups; g++) {
1979 for (i = 0; i < ics->max_sfb; i++, idx++) {
1980 if (cpe->ms_mask[idx] &&
1981 cpe->ch[0].band_type[idx] < NOISE_BT &&
1982 cpe->ch[1].band_type[idx] < NOISE_BT) {
1984 for (group = 0; group < ics->group_len[g]; group++) {
1985 ac->fdsp->butterflies_fixed(ch0 + group * 128 + offsets[i],
1986 ch1 + group * 128 + offsets[i],
1987 offsets[i+1] - offsets[i]);
1989 for (group = 0; group < ics->group_len[g]; group++) {
1990 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
1991 ch1 + group * 128 + offsets[i],
1992 offsets[i+1] - offsets[i]);
1993 #endif /* USE_FIXED */
1997 ch0 += ics->group_len[g] * 128;
1998 ch1 += ics->group_len[g] * 128;
2003 * intensity stereo decoding; reference: 4.6.8.2.3
2005 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2006 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2007 * [3] reserved for scalable AAC
2009 static void apply_intensity_stereo(AACContext *ac,
2010 ChannelElement *cpe, int ms_present)
2012 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2013 SingleChannelElement *sce1 = &cpe->ch[1];
2014 INTFLOAT *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2015 const uint16_t *offsets = ics->swb_offset;
2016 int g, group, i, idx = 0;
2019 for (g = 0; g < ics->num_window_groups; g++) {
2020 for (i = 0; i < ics->max_sfb;) {
2021 if (sce1->band_type[idx] == INTENSITY_BT ||
2022 sce1->band_type[idx] == INTENSITY_BT2) {
2023 const int bt_run_end = sce1->band_type_run_end[idx];
2024 for (; i < bt_run_end; i++, idx++) {
2025 c = -1 + 2 * (sce1->band_type[idx] - 14);
2027 c *= 1 - 2 * cpe->ms_mask[idx];
2028 scale = c * sce1->sf[idx];
2029 for (group = 0; group < ics->group_len[g]; group++)
2031 ac->subband_scale(coef1 + group * 128 + offsets[i],
2032 coef0 + group * 128 + offsets[i],
2035 offsets[i + 1] - offsets[i]);
2037 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2038 coef0 + group * 128 + offsets[i],
2040 offsets[i + 1] - offsets[i]);
2041 #endif /* USE_FIXED */
2044 int bt_run_end = sce1->band_type_run_end[idx];
2045 idx += bt_run_end - i;
2049 coef0 += ics->group_len[g] * 128;
2050 coef1 += ics->group_len[g] * 128;
2055 * Decode a channel_pair_element; reference: table 4.4.
2057 * @return Returns error status. 0 - OK, !0 - error
2059 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2061 int i, ret, common_window, ms_present = 0;
2062 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2064 common_window = eld_syntax || get_bits1(gb);
2065 if (common_window) {
2066 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2067 return AVERROR_INVALIDDATA;
2068 i = cpe->ch[1].ics.use_kb_window[0];
2069 cpe->ch[1].ics = cpe->ch[0].ics;
2070 cpe->ch[1].ics.use_kb_window[1] = i;
2071 if (cpe->ch[1].ics.predictor_present &&
2072 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2073 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2074 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2075 ms_present = get_bits(gb, 2);
2076 if (ms_present == 3) {
2077 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2078 return AVERROR_INVALIDDATA;
2079 } else if (ms_present)
2080 decode_mid_side_stereo(cpe, gb, ms_present);
2082 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2084 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2087 if (common_window) {
2089 apply_mid_side_stereo(ac, cpe);
2090 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2091 apply_prediction(ac, &cpe->ch[0]);
2092 apply_prediction(ac, &cpe->ch[1]);
2096 apply_intensity_stereo(ac, cpe, ms_present);
2100 static const float cce_scale[] = {
2101 1.09050773266525765921, //2^(1/8)
2102 1.18920711500272106672, //2^(1/4)
2108 * Decode coupling_channel_element; reference: table 4.8.
2110 * @return Returns error status. 0 - OK, !0 - error
2112 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2118 SingleChannelElement *sce = &che->ch[0];
2119 ChannelCoupling *coup = &che->coup;
2121 coup->coupling_point = 2 * get_bits1(gb);
2122 coup->num_coupled = get_bits(gb, 3);
2123 for (c = 0; c <= coup->num_coupled; c++) {
2125 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2126 coup->id_select[c] = get_bits(gb, 4);
2127 if (coup->type[c] == TYPE_CPE) {
2128 coup->ch_select[c] = get_bits(gb, 2);
2129 if (coup->ch_select[c] == 3)
2132 coup->ch_select[c] = 2;
2134 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2136 sign = get_bits(gb, 1);
2137 scale = AAC_RENAME(cce_scale)[get_bits(gb, 2)];
2139 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2142 for (c = 0; c < num_gain; c++) {
2146 INTFLOAT gain_cache = FIXR10(1.);
2148 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2149 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2150 gain_cache = GET_GAIN(scale, gain);
2152 if (coup->coupling_point == AFTER_IMDCT) {
2153 coup->gain[c][0] = gain_cache;
2155 for (g = 0; g < sce->ics.num_window_groups; g++) {
2156 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2157 if (sce->band_type[idx] != ZERO_BT) {
2159 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2167 gain_cache = GET_GAIN(scale, t) * s;
2170 coup->gain[c][idx] = gain_cache;
2180 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2182 * @return Returns number of bytes consumed.
2184 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2188 int num_excl_chan = 0;
2191 for (i = 0; i < 7; i++)
2192 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2193 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2195 return num_excl_chan / 7;
2199 * Decode dynamic range information; reference: table 4.52.
2201 * @return Returns number of bytes consumed.
2203 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2207 int drc_num_bands = 1;
2210 /* pce_tag_present? */
2211 if (get_bits1(gb)) {
2212 che_drc->pce_instance_tag = get_bits(gb, 4);
2213 skip_bits(gb, 4); // tag_reserved_bits
2217 /* excluded_chns_present? */
2218 if (get_bits1(gb)) {
2219 n += decode_drc_channel_exclusions(che_drc, gb);
2222 /* drc_bands_present? */
2223 if (get_bits1(gb)) {
2224 che_drc->band_incr = get_bits(gb, 4);
2225 che_drc->interpolation_scheme = get_bits(gb, 4);
2227 drc_num_bands += che_drc->band_incr;
2228 for (i = 0; i < drc_num_bands; i++) {
2229 che_drc->band_top[i] = get_bits(gb, 8);
2234 /* prog_ref_level_present? */
2235 if (get_bits1(gb)) {
2236 che_drc->prog_ref_level = get_bits(gb, 7);
2237 skip_bits1(gb); // prog_ref_level_reserved_bits
2241 for (i = 0; i < drc_num_bands; i++) {
2242 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2243 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2250 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2252 int i, major, minor;
2257 get_bits(gb, 13); len -= 13;
2259 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2260 buf[i] = get_bits(gb, 8);
2263 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2264 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2266 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2267 ac->avctx->internal->skip_samples = 1024;
2271 skip_bits_long(gb, len);
2277 * Decode extension data (incomplete); reference: table 4.51.
2279 * @param cnt length of TYPE_FIL syntactic element in bytes
2281 * @return Returns number of bytes consumed
2283 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2284 ChannelElement *che, enum RawDataBlockType elem_type)
2288 int type = get_bits(gb, 4);
2290 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2291 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2293 switch (type) { // 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 = AAC_RENAME(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(INTFLOAT 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 INTFLOAT lpc[TNS_MAX_ORDER];
2348 INTFLOAT 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 AAC_RENAME(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] -= AAC_MUL26(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] += AAC_MUL26(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, INTFLOAT *out,
2398 INTFLOAT *in, IndividualChannelStream *ics)
2400 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2401 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2402 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2403 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(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(*in));
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(*in));
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 INTFLOAT *predTime = sce->ret;
2431 INTFLOAT *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] = AAC_MUL30(sce->ltp_state[i + 2048 - ltp->lag], ltp->coef);
2438 memset(&predTime[i], 0, (2048 - i) * sizeof(*predTime));
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 INTFLOAT *saved = sce->saved;
2459 INTFLOAT *saved_ltp = sce->coeffs;
2460 const INTFLOAT *lwindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2461 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2464 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2465 memcpy(saved_ltp, saved, 512 * sizeof(*saved_ltp));
2466 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2467 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2469 for (i = 0; i < 64; i++)
2470 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2471 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2472 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(*saved_ltp));
2473 memset(saved_ltp + 576, 0, 448 * sizeof(*saved_ltp));
2474 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2476 for (i = 0; i < 64; i++)
2477 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], swindow[63 - i]);
2478 } else { // LONG_STOP or ONLY_LONG
2479 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2481 for (i = 0; i < 512; i++)
2482 saved_ltp[i + 512] = AAC_MUL31(ac->buf_mdct[1023 - i], lwindow[511 - i]);
2485 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2486 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2487 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2491 * Conduct IMDCT and windowing.
2493 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2495 IndividualChannelStream *ics = &sce->ics;
2496 INTFLOAT *in = sce->coeffs;
2497 INTFLOAT *out = sce->ret;
2498 INTFLOAT *saved = sce->saved;
2499 const INTFLOAT *swindow = ics->use_kb_window[0] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2500 const INTFLOAT *lwindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_long_1024) : AAC_RENAME(ff_sine_1024);
2501 const INTFLOAT *swindow_prev = ics->use_kb_window[1] ? AAC_RENAME(ff_aac_kbd_short_128) : AAC_RENAME(ff_sine_128);
2502 INTFLOAT *buf = ac->buf_mdct;
2503 INTFLOAT *temp = ac->temp;
2507 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2508 for (i = 0; i < 1024; i += 128)
2509 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2511 ac->mdct.imdct_half(&ac->mdct, buf, in);
2513 for (i=0; i<1024; i++)
2514 buf[i] = (buf[i] + 4) >> 3;
2515 #endif /* USE_FIXED */
2518 /* window overlapping
2519 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2520 * and long to short transitions are considered to be short to short
2521 * transitions. This leaves just two cases (long to long and short to short)
2522 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2524 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2525 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2526 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2528 memcpy( out, saved, 448 * sizeof(*out));
2530 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2531 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2532 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2533 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2534 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2535 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2536 memcpy( out + 448 + 4*128, temp, 64 * sizeof(*out));
2538 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2539 memcpy( out + 576, buf + 64, 448 * sizeof(*out));
2544 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2545 memcpy( saved, temp + 64, 64 * sizeof(*saved));
2546 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2547 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2548 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2549 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2550 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2551 memcpy( saved, buf + 512, 448 * sizeof(*saved));
2552 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(*saved));
2553 } else { // LONG_STOP or ONLY_LONG
2554 memcpy( saved, buf + 512, 512 * sizeof(*saved));
2558 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2560 IndividualChannelStream *ics = &sce->ics;
2561 INTFLOAT *in = sce->coeffs;
2562 INTFLOAT *out = sce->ret;
2563 INTFLOAT *saved = sce->saved;
2564 INTFLOAT *buf = ac->buf_mdct;
2567 #endif /* USE_FIXED */
2570 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2573 for (i = 0; i < 1024; i++)
2574 buf[i] = (buf[i] + 2) >> 2;
2575 #endif /* USE_FIXED */
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(*out));
2581 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, AAC_RENAME(ff_sine_128), 64);
2582 memcpy( out + 320, buf + 64, 192 * sizeof(*out));
2584 ac->fdsp->vector_fmul_window(out, saved, buf, AAC_RENAME(ff_sine_512), 256);
2588 memcpy(saved, buf + 256, 256 * sizeof(*saved));
2591 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2593 INTFLOAT *in = sce->coeffs;
2594 INTFLOAT *out = sce->ret;
2595 INTFLOAT *saved = sce->saved;
2596 INTFLOAT *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 INTFLOAT *const window = n == 480 ? AAC_RENAME(ff_aac_eld_window_480) :
2602 AAC_RENAME(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;
2616 ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
2619 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2622 for (i = 0; i < 1024; i++)
2623 buf[i] = (buf[i] + 1) >> 1;
2624 #endif /* USE_FIXED */
2626 for (i = 0; i < n; i+=2) {
2629 // Like with the regular IMDCT at this point we still have the middle half
2630 // of a transform but with even symmetry on the left and odd symmetry on
2633 // window overlapping
2634 // The spec says to use samples [0..511] but the reference decoder uses
2635 // samples [128..639].
2636 for (i = n4; i < n2; i ++) {
2637 out[i - n4] = AAC_MUL31( buf[ n2 - 1 - i] , window[i - n4]) +
2638 AAC_MUL31( saved[ i + n2] , window[i + n - n4]) +
2639 AAC_MUL31(-saved[n + n2 - 1 - i] , window[i + 2*n - n4]) +
2640 AAC_MUL31(-saved[ 2*n + n2 + i] , window[i + 3*n - n4]);
2642 for (i = 0; i < n2; i ++) {
2643 out[n4 + i] = AAC_MUL31( buf[ i] , window[i + n2 - n4]) +
2644 AAC_MUL31(-saved[ n - 1 - i] , window[i + n2 + n - n4]) +
2645 AAC_MUL31(-saved[ n + i] , window[i + n2 + 2*n - n4]) +
2646 AAC_MUL31( saved[2*n + n - 1 - i] , window[i + n2 + 3*n - n4]);
2648 for (i = 0; i < n4; i ++) {
2649 out[n2 + n4 + i] = AAC_MUL31( buf[ i + n2] , window[i + n - n4]) +
2650 AAC_MUL31(-saved[n2 - 1 - i] , window[i + 2*n - n4]) +
2651 AAC_MUL31(-saved[n + n2 + i] , window[i + 3*n - n4]);
2655 memmove(saved + n, saved, 2 * n * sizeof(*saved));
2656 memcpy( saved, buf, n * sizeof(*saved));
2660 * channel coupling transformation interface
2662 * @param apply_coupling_method pointer to (in)dependent coupling function
2664 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2665 enum RawDataBlockType type, int elem_id,
2666 enum CouplingPoint coupling_point,
2667 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2671 for (i = 0; i < MAX_ELEM_ID; i++) {
2672 ChannelElement *cce = ac->che[TYPE_CCE][i];
2675 if (cce && cce->coup.coupling_point == coupling_point) {
2676 ChannelCoupling *coup = &cce->coup;
2678 for (c = 0; c <= coup->num_coupled; c++) {
2679 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2680 if (coup->ch_select[c] != 1) {
2681 apply_coupling_method(ac, &cc->ch[0], cce, index);
2682 if (coup->ch_select[c] != 0)
2685 if (coup->ch_select[c] != 2)
2686 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2688 index += 1 + (coup->ch_select[c] == 3);
2695 * Convert spectral data to samples, applying all supported tools as appropriate.
2697 static void spectral_to_sample(AACContext *ac)
2700 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2701 switch (ac->oc[1].m4ac.object_type) {
2703 imdct_and_window = imdct_and_windowing_ld;
2705 case AOT_ER_AAC_ELD:
2706 imdct_and_window = imdct_and_windowing_eld;
2709 imdct_and_window = ac->imdct_and_windowing;
2711 for (type = 3; type >= 0; type--) {
2712 for (i = 0; i < MAX_ELEM_ID; i++) {
2713 ChannelElement *che = ac->che[type][i];
2714 if (che && che->present) {
2715 if (type <= TYPE_CPE)
2716 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, AAC_RENAME(apply_dependent_coupling));
2717 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2718 if (che->ch[0].ics.predictor_present) {
2719 if (che->ch[0].ics.ltp.present)
2720 ac->apply_ltp(ac, &che->ch[0]);
2721 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2722 ac->apply_ltp(ac, &che->ch[1]);
2725 if (che->ch[0].tns.present)
2726 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2727 if (che->ch[1].tns.present)
2728 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2729 if (type <= TYPE_CPE)
2730 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, AAC_RENAME(apply_dependent_coupling));
2731 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2732 imdct_and_window(ac, &che->ch[0]);
2733 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2734 ac->update_ltp(ac, &che->ch[0]);
2735 if (type == TYPE_CPE) {
2736 imdct_and_window(ac, &che->ch[1]);
2737 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2738 ac->update_ltp(ac, &che->ch[1]);
2740 if (ac->oc[1].m4ac.sbr > 0) {
2741 AAC_RENAME(ff_sbr_apply)(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2744 if (type <= TYPE_CCE)
2745 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, AAC_RENAME(apply_independent_coupling));
2750 /* preparation for resampler */
2751 for(j = 0; j<2048; j++){
2752 che->ch[0].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[0].ret[j]<<7)+0x8000;
2753 che->ch[1].ret[j] = (int32_t)av_clipl_int32((int64_t)che->ch[1].ret[j]<<7)+0x8000;
2756 #endif /* USE_FIXED */
2759 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2765 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2768 AACADTSHeaderInfo hdr_info;
2769 uint8_t layout_map[MAX_ELEM_ID*4][3];
2770 int layout_map_tags, ret;
2772 size = avpriv_aac_parse_header(gb, &hdr_info);
2774 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2775 // This is 2 for "VLB " audio in NSV files.
2776 // See samples/nsv/vlb_audio.
2777 avpriv_report_missing_feature(ac->avctx,
2778 "More than one AAC RDB per ADTS frame");
2779 ac->warned_num_aac_frames = 1;
2781 push_output_configuration(ac);
2782 if (hdr_info.chan_config) {
2783 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2784 if ((ret = set_default_channel_config(ac->avctx,
2787 hdr_info.chan_config)) < 0)
2789 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2790 FFMAX(ac->oc[1].status,
2791 OC_TRIAL_FRAME), 0)) < 0)
2794 ac->oc[1].m4ac.chan_config = 0;
2796 * dual mono frames in Japanese DTV can have chan_config 0
2797 * WITHOUT specifying PCE.
2798 * thus, set dual mono as default.
2800 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2801 layout_map_tags = 2;
2802 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2803 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2804 layout_map[0][1] = 0;
2805 layout_map[1][1] = 1;
2806 if (output_configure(ac, layout_map, layout_map_tags,
2811 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2812 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2813 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2814 ac->oc[1].m4ac.frame_length_short = 0;
2815 if (ac->oc[0].status != OC_LOCKED ||
2816 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2817 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2818 ac->oc[1].m4ac.sbr = -1;
2819 ac->oc[1].m4ac.ps = -1;
2821 if (!hdr_info.crc_absent)
2827 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2828 int *got_frame_ptr, GetBitContext *gb)
2830 AACContext *ac = avctx->priv_data;
2831 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2832 ChannelElement *che;
2834 int samples = m4ac->frame_length_short ? 960 : 1024;
2835 int chan_config = m4ac->chan_config;
2836 int aot = m4ac->object_type;
2838 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2843 if ((err = frame_configure_elements(avctx)) < 0)
2846 // The FF_PROFILE_AAC_* defines are all object_type - 1
2847 // This may lead to an undefined profile being signaled
2848 ac->avctx->profile = aot - 1;
2850 ac->tags_mapped = 0;
2852 if (chan_config < 0 || (chan_config >= 8 && chan_config < 11) || chan_config >= 13) {
2853 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2855 return AVERROR_INVALIDDATA;
2857 for (i = 0; i < tags_per_config[chan_config]; i++) {
2858 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2859 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2860 if (!(che=get_che(ac, elem_type, elem_id))) {
2861 av_log(ac->avctx, AV_LOG_ERROR,
2862 "channel element %d.%d is not allocated\n",
2863 elem_type, elem_id);
2864 return AVERROR_INVALIDDATA;
2867 if (aot != AOT_ER_AAC_ELD)
2869 switch (elem_type) {
2871 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2874 err = decode_cpe(ac, gb, che);
2877 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2884 spectral_to_sample(ac);
2886 ac->frame->nb_samples = samples;
2887 ac->frame->sample_rate = avctx->sample_rate;
2890 skip_bits_long(gb, get_bits_left(gb));
2894 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2895 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2897 AACContext *ac = avctx->priv_data;
2898 ChannelElement *che = NULL, *che_prev = NULL;
2899 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2901 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2902 int is_dmono, sce_count = 0;
2906 if (show_bits(gb, 12) == 0xfff) {
2907 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2908 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2911 if (ac->oc[1].m4ac.sampling_index > 12) {
2912 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2913 err = AVERROR_INVALIDDATA;
2918 if ((err = frame_configure_elements(avctx)) < 0)
2921 // The FF_PROFILE_AAC_* defines are all object_type - 1
2922 // This may lead to an undefined profile being signaled
2923 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2925 ac->tags_mapped = 0;
2927 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2928 elem_id = get_bits(gb, 4);
2930 if (avctx->debug & FF_DEBUG_STARTCODE)
2931 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2933 if (!avctx->channels && elem_type != TYPE_PCE) {
2934 err = AVERROR_INVALIDDATA;
2938 if (elem_type < TYPE_DSE) {
2939 if (!(che=get_che(ac, elem_type, elem_id))) {
2940 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2941 elem_type, elem_id);
2942 err = AVERROR_INVALIDDATA;
2949 switch (elem_type) {
2952 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2958 err = decode_cpe(ac, gb, che);
2963 err = decode_cce(ac, gb, che);
2967 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2972 err = skip_data_stream_element(ac, gb);
2976 uint8_t layout_map[MAX_ELEM_ID*4][3];
2978 push_output_configuration(ac);
2979 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
2985 av_log(avctx, AV_LOG_ERROR,
2986 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2988 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
2990 ac->oc[1].m4ac.chan_config = 0;
2998 elem_id += get_bits(gb, 8) - 1;
2999 if (get_bits_left(gb) < 8 * elem_id) {
3000 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3001 err = AVERROR_INVALIDDATA;
3005 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3006 err = 0; /* FIXME */
3010 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3015 elem_type_prev = elem_type;
3020 if (get_bits_left(gb) < 3) {
3021 av_log(avctx, AV_LOG_ERROR, overread_err);
3022 err = AVERROR_INVALIDDATA;
3027 if (!avctx->channels) {
3032 spectral_to_sample(ac);
3034 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3035 samples <<= multiplier;
3037 if (ac->oc[1].status && audio_found) {
3038 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3039 avctx->frame_size = samples;
3040 ac->oc[1].status = OC_LOCKED;
3045 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3046 if (side && side_size>=4)
3047 AV_WL32(side, 2*AV_RL32(side));
3050 if (!ac->frame->data[0] && samples) {
3051 av_log(avctx, AV_LOG_ERROR, "no frame data found\n");
3052 err = AVERROR_INVALIDDATA;
3057 ac->frame->nb_samples = samples;
3058 ac->frame->sample_rate = avctx->sample_rate;
3060 av_frame_unref(ac->frame);
3061 *got_frame_ptr = !!samples;
3063 /* for dual-mono audio (SCE + SCE) */
3064 is_dmono = ac->dmono_mode && sce_count == 2 &&
3065 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3067 if (ac->dmono_mode == 1)
3068 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3069 else if (ac->dmono_mode == 2)
3070 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3075 pop_output_configuration(ac);
3079 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3080 int *got_frame_ptr, AVPacket *avpkt)
3082 AACContext *ac = avctx->priv_data;
3083 const uint8_t *buf = avpkt->data;
3084 int buf_size = avpkt->size;
3089 int new_extradata_size;
3090 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3091 AV_PKT_DATA_NEW_EXTRADATA,
3092 &new_extradata_size);
3093 int jp_dualmono_size;
3094 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3095 AV_PKT_DATA_JP_DUALMONO,
3098 if (new_extradata && 0) {
3099 av_free(avctx->extradata);
3100 avctx->extradata = av_mallocz(new_extradata_size +
3101 FF_INPUT_BUFFER_PADDING_SIZE);
3102 if (!avctx->extradata)
3103 return AVERROR(ENOMEM);
3104 avctx->extradata_size = new_extradata_size;
3105 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3106 push_output_configuration(ac);
3107 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3109 avctx->extradata_size*8, 1) < 0) {
3110 pop_output_configuration(ac);
3111 return AVERROR_INVALIDDATA;
3116 if (jp_dualmono && jp_dualmono_size > 0)
3117 ac->dmono_mode = 1 + *jp_dualmono;
3118 if (ac->force_dmono_mode >= 0)
3119 ac->dmono_mode = ac->force_dmono_mode;
3121 if (INT_MAX / 8 <= buf_size)
3122 return AVERROR_INVALIDDATA;
3124 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3127 switch (ac->oc[1].m4ac.object_type) {
3129 case AOT_ER_AAC_LTP:
3131 case AOT_ER_AAC_ELD:
3132 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3135 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3140 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3141 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3142 if (buf[buf_offset])
3145 return buf_size > buf_offset ? buf_consumed : buf_size;
3148 static av_cold int aac_decode_close(AVCodecContext *avctx)
3150 AACContext *ac = avctx->priv_data;
3153 for (i = 0; i < MAX_ELEM_ID; i++) {
3154 for (type = 0; type < 4; type++) {
3155 if (ac->che[type][i])
3156 AAC_RENAME(ff_aac_sbr_ctx_close)(&ac->che[type][i]->sbr);
3157 av_freep(&ac->che[type][i]);
3161 ff_mdct_end(&ac->mdct);
3162 ff_mdct_end(&ac->mdct_small);
3163 ff_mdct_end(&ac->mdct_ld);
3164 ff_mdct_end(&ac->mdct_ltp);
3166 ff_imdct15_uninit(&ac->mdct480);
3168 av_freep(&ac->fdsp);
3172 static void aacdec_init(AACContext *c)
3174 c->imdct_and_windowing = imdct_and_windowing;
3175 c->apply_ltp = apply_ltp;
3176 c->apply_tns = apply_tns;
3177 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3178 c->update_ltp = update_ltp;
3180 c->vector_pow43 = vector_pow43;
3181 c->subband_scale = subband_scale;
3186 ff_aacdec_init_mips(c);
3187 #endif /* !USE_FIXED */
3190 * AVOptions for Japanese DTV specific extensions (ADTS only)
3192 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3193 static const AVOption options[] = {
3194 {"dual_mono_mode", "Select the channel to decode for dual mono",
3195 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3196 AACDEC_FLAGS, "dual_mono_mode"},
3198 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3199 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3200 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3201 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3206 static const AVClass aac_decoder_class = {
3207 .class_name = "AAC decoder",
3208 .item_name = av_default_item_name,
3210 .version = LIBAVUTIL_VERSION_INT,
3213 static const AVProfile profiles[] = {
3214 { FF_PROFILE_AAC_MAIN, "Main" },
3215 { FF_PROFILE_AAC_LOW, "LC" },
3216 { FF_PROFILE_AAC_SSR, "SSR" },
3217 { FF_PROFILE_AAC_LTP, "LTP" },
3218 { FF_PROFILE_AAC_HE, "HE-AAC" },
3219 { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3220 { FF_PROFILE_AAC_LD, "LD" },
3221 { FF_PROFILE_AAC_ELD, "ELD" },
3222 { FF_PROFILE_UNKNOWN },