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"
91 #include "fmtconvert.h"
98 #include "aacdectab.h"
99 #include "cbrt_tablegen.h"
102 #include "mpeg4audio.h"
103 #include "aacadtsdec.h"
104 #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:
706 // AAC_CHANNEL_OFF has no channel map
709 layout_map[0][0] = syn_ele;
710 layout_map[0][1] = get_bits(gb, 4);
711 layout_map[0][2] = type;
717 * Decode program configuration element; reference: table 4.2.
719 * @return Returns error status. 0 - OK, !0 - error
721 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
722 uint8_t (*layout_map)[3],
725 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc;
730 skip_bits(gb, 2); // object_type
732 sampling_index = get_bits(gb, 4);
733 if (m4ac->sampling_index != sampling_index)
734 av_log(avctx, AV_LOG_WARNING,
735 "Sample rate index in program config element does not "
736 "match the sample rate index configured by the container.\n");
738 num_front = get_bits(gb, 4);
739 num_side = get_bits(gb, 4);
740 num_back = get_bits(gb, 4);
741 num_lfe = get_bits(gb, 2);
742 num_assoc_data = get_bits(gb, 3);
743 num_cc = get_bits(gb, 4);
746 skip_bits(gb, 4); // mono_mixdown_tag
748 skip_bits(gb, 4); // stereo_mixdown_tag
751 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
753 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
754 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
757 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
759 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
761 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
763 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
766 skip_bits_long(gb, 4 * num_assoc_data);
768 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
773 /* comment field, first byte is length */
774 comment_len = get_bits(gb, 8) * 8;
775 if (get_bits_left(gb) < comment_len) {
776 av_log(avctx, AV_LOG_ERROR, "decode_pce: " overread_err);
777 return AVERROR_INVALIDDATA;
779 skip_bits_long(gb, comment_len);
784 * Decode GA "General Audio" specific configuration; reference: table 4.1.
786 * @param ac pointer to AACContext, may be null
787 * @param avctx pointer to AVCCodecContext, used for logging
789 * @return Returns error status. 0 - OK, !0 - error
791 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
793 MPEG4AudioConfig *m4ac,
796 int extension_flag, ret, ep_config, res_flags;
797 uint8_t layout_map[MAX_ELEM_ID*4][3];
800 if (get_bits1(gb)) { // frameLengthFlag
801 avpriv_request_sample(avctx, "960/120 MDCT window");
802 return AVERROR_PATCHWELCOME;
804 m4ac->frame_length_short = 0;
806 if (get_bits1(gb)) // dependsOnCoreCoder
807 skip_bits(gb, 14); // coreCoderDelay
808 extension_flag = get_bits1(gb);
810 if (m4ac->object_type == AOT_AAC_SCALABLE ||
811 m4ac->object_type == AOT_ER_AAC_SCALABLE)
812 skip_bits(gb, 3); // layerNr
814 if (channel_config == 0) {
815 skip_bits(gb, 4); // element_instance_tag
816 tags = decode_pce(avctx, m4ac, layout_map, gb);
820 if ((ret = set_default_channel_config(avctx, layout_map,
821 &tags, channel_config)))
825 if (count_channels(layout_map, tags) > 1) {
827 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
830 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
833 if (extension_flag) {
834 switch (m4ac->object_type) {
836 skip_bits(gb, 5); // numOfSubFrame
837 skip_bits(gb, 11); // layer_length
841 case AOT_ER_AAC_SCALABLE:
843 res_flags = get_bits(gb, 3);
845 avpriv_report_missing_feature(avctx,
846 "AAC data resilience (flags %x)",
848 return AVERROR_PATCHWELCOME;
852 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
854 switch (m4ac->object_type) {
857 case AOT_ER_AAC_SCALABLE:
859 ep_config = get_bits(gb, 2);
861 avpriv_report_missing_feature(avctx,
862 "epConfig %d", ep_config);
863 return AVERROR_PATCHWELCOME;
869 static int decode_eld_specific_config(AACContext *ac, AVCodecContext *avctx,
871 MPEG4AudioConfig *m4ac,
874 int ret, ep_config, res_flags;
875 uint8_t layout_map[MAX_ELEM_ID*4][3];
877 const int ELDEXT_TERM = 0;
882 m4ac->frame_length_short = get_bits1(gb);
883 res_flags = get_bits(gb, 3);
885 avpriv_report_missing_feature(avctx,
886 "AAC data resilience (flags %x)",
888 return AVERROR_PATCHWELCOME;
891 if (get_bits1(gb)) { // ldSbrPresentFlag
892 avpriv_report_missing_feature(avctx,
894 return AVERROR_PATCHWELCOME;
897 while (get_bits(gb, 4) != ELDEXT_TERM) {
898 int len = get_bits(gb, 4);
900 len += get_bits(gb, 8);
902 len += get_bits(gb, 16);
903 if (get_bits_left(gb) < len * 8 + 4) {
904 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
905 return AVERROR_INVALIDDATA;
907 skip_bits_long(gb, 8 * len);
910 if ((ret = set_default_channel_config(avctx, layout_map,
911 &tags, channel_config)))
914 if (ac && (ret = output_configure(ac, layout_map, tags, OC_GLOBAL_HDR, 0)))
917 ep_config = get_bits(gb, 2);
919 avpriv_report_missing_feature(avctx,
920 "epConfig %d", ep_config);
921 return AVERROR_PATCHWELCOME;
927 * Decode audio specific configuration; reference: table 1.13.
929 * @param ac pointer to AACContext, may be null
930 * @param avctx pointer to AVCCodecContext, used for logging
931 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
932 * @param data pointer to buffer holding an audio specific config
933 * @param bit_size size of audio specific config or data in bits
934 * @param sync_extension look for an appended sync extension
936 * @return Returns error status or number of consumed bits. <0 - error
938 static int decode_audio_specific_config(AACContext *ac,
939 AVCodecContext *avctx,
940 MPEG4AudioConfig *m4ac,
941 const uint8_t *data, int bit_size,
947 av_dlog(avctx, "audio specific config size %d\n", bit_size >> 3);
948 for (i = 0; i < bit_size >> 3; i++)
949 av_dlog(avctx, "%02x ", data[i]);
950 av_dlog(avctx, "\n");
952 if ((ret = init_get_bits(&gb, data, bit_size)) < 0)
955 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size,
956 sync_extension)) < 0)
957 return AVERROR_INVALIDDATA;
958 if (m4ac->sampling_index > 12) {
959 av_log(avctx, AV_LOG_ERROR,
960 "invalid sampling rate index %d\n",
961 m4ac->sampling_index);
962 return AVERROR_INVALIDDATA;
964 if (m4ac->object_type == AOT_ER_AAC_LD &&
965 (m4ac->sampling_index < 3 || m4ac->sampling_index > 7)) {
966 av_log(avctx, AV_LOG_ERROR,
967 "invalid low delay sampling rate index %d\n",
968 m4ac->sampling_index);
969 return AVERROR_INVALIDDATA;
972 skip_bits_long(&gb, i);
974 switch (m4ac->object_type) {
980 if ((ret = decode_ga_specific_config(ac, avctx, &gb,
981 m4ac, m4ac->chan_config)) < 0)
985 if ((ret = decode_eld_specific_config(ac, avctx, &gb,
986 m4ac, m4ac->chan_config)) < 0)
990 avpriv_report_missing_feature(avctx,
991 "Audio object type %s%d",
992 m4ac->sbr == 1 ? "SBR+" : "",
994 return AVERROR(ENOSYS);
998 "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
999 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
1000 m4ac->sample_rate, m4ac->sbr,
1003 return get_bits_count(&gb);
1007 * linear congruential pseudorandom number generator
1009 * @param previous_val pointer to the current state of the generator
1011 * @return Returns a 32-bit pseudorandom integer
1013 static av_always_inline int lcg_random(unsigned previous_val)
1015 union { unsigned u; int s; } v = { previous_val * 1664525u + 1013904223 };
1019 static av_always_inline void reset_predict_state(PredictorState *ps)
1029 static void reset_all_predictors(PredictorState *ps)
1032 for (i = 0; i < MAX_PREDICTORS; i++)
1033 reset_predict_state(&ps[i]);
1036 static int sample_rate_idx (int rate)
1038 if (92017 <= rate) return 0;
1039 else if (75132 <= rate) return 1;
1040 else if (55426 <= rate) return 2;
1041 else if (46009 <= rate) return 3;
1042 else if (37566 <= rate) return 4;
1043 else if (27713 <= rate) return 5;
1044 else if (23004 <= rate) return 6;
1045 else if (18783 <= rate) return 7;
1046 else if (13856 <= rate) return 8;
1047 else if (11502 <= rate) return 9;
1048 else if (9391 <= rate) return 10;
1052 static void reset_predictor_group(PredictorState *ps, int group_num)
1055 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
1056 reset_predict_state(&ps[i]);
1059 #define AAC_INIT_VLC_STATIC(num, size) \
1060 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
1061 ff_aac_spectral_bits[num], sizeof(ff_aac_spectral_bits[num][0]), \
1062 sizeof(ff_aac_spectral_bits[num][0]), \
1063 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), \
1064 sizeof(ff_aac_spectral_codes[num][0]), \
1067 static void aacdec_init(AACContext *ac);
1069 static av_cold int aac_decode_init(AVCodecContext *avctx)
1071 AACContext *ac = avctx->priv_data;
1075 ac->oc[1].m4ac.sample_rate = avctx->sample_rate;
1079 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
1081 if (avctx->extradata_size > 0) {
1082 if ((ret = decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
1084 avctx->extradata_size * 8,
1089 uint8_t layout_map[MAX_ELEM_ID*4][3];
1090 int layout_map_tags;
1092 sr = sample_rate_idx(avctx->sample_rate);
1093 ac->oc[1].m4ac.sampling_index = sr;
1094 ac->oc[1].m4ac.channels = avctx->channels;
1095 ac->oc[1].m4ac.sbr = -1;
1096 ac->oc[1].m4ac.ps = -1;
1098 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
1099 if (ff_mpeg4audio_channels[i] == avctx->channels)
1101 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
1104 ac->oc[1].m4ac.chan_config = i;
1106 if (ac->oc[1].m4ac.chan_config) {
1107 int ret = set_default_channel_config(avctx, layout_map,
1108 &layout_map_tags, ac->oc[1].m4ac.chan_config);
1110 output_configure(ac, layout_map, layout_map_tags,
1112 else if (avctx->err_recognition & AV_EF_EXPLODE)
1113 return AVERROR_INVALIDDATA;
1117 if (avctx->channels > MAX_CHANNELS) {
1118 av_log(avctx, AV_LOG_ERROR, "Too many channels\n");
1119 return AVERROR_INVALIDDATA;
1122 AAC_INIT_VLC_STATIC( 0, 304);
1123 AAC_INIT_VLC_STATIC( 1, 270);
1124 AAC_INIT_VLC_STATIC( 2, 550);
1125 AAC_INIT_VLC_STATIC( 3, 300);
1126 AAC_INIT_VLC_STATIC( 4, 328);
1127 AAC_INIT_VLC_STATIC( 5, 294);
1128 AAC_INIT_VLC_STATIC( 6, 306);
1129 AAC_INIT_VLC_STATIC( 7, 268);
1130 AAC_INIT_VLC_STATIC( 8, 510);
1131 AAC_INIT_VLC_STATIC( 9, 366);
1132 AAC_INIT_VLC_STATIC(10, 462);
1136 ff_fmt_convert_init(&ac->fmt_conv, avctx);
1137 ac->fdsp = avpriv_float_dsp_alloc(avctx->flags & CODEC_FLAG_BITEXACT);
1139 return AVERROR(ENOMEM);
1142 ac->random_state = 0x1f2e3d4c;
1146 INIT_VLC_STATIC(&vlc_scalefactors, 7,
1147 FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
1148 ff_aac_scalefactor_bits,
1149 sizeof(ff_aac_scalefactor_bits[0]),
1150 sizeof(ff_aac_scalefactor_bits[0]),
1151 ff_aac_scalefactor_code,
1152 sizeof(ff_aac_scalefactor_code[0]),
1153 sizeof(ff_aac_scalefactor_code[0]),
1156 ff_mdct_init(&ac->mdct, 11, 1, 1.0 / (32768.0 * 1024.0));
1157 ff_mdct_init(&ac->mdct_ld, 10, 1, 1.0 / (32768.0 * 512.0));
1158 ff_mdct_init(&ac->mdct_small, 8, 1, 1.0 / (32768.0 * 128.0));
1159 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0 * 32768.0);
1160 ret = ff_imdct15_init(&ac->mdct480, 5);
1164 // window initialization
1165 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
1166 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
1167 ff_init_ff_sine_windows(10);
1168 ff_init_ff_sine_windows( 9);
1169 ff_init_ff_sine_windows( 7);
1177 * Skip data_stream_element; reference: table 4.10.
1179 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
1181 int byte_align = get_bits1(gb);
1182 int count = get_bits(gb, 8);
1184 count += get_bits(gb, 8);
1188 if (get_bits_left(gb) < 8 * count) {
1189 av_log(ac->avctx, AV_LOG_ERROR, "skip_data_stream_element: "overread_err);
1190 return AVERROR_INVALIDDATA;
1192 skip_bits_long(gb, 8 * count);
1196 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
1200 if (get_bits1(gb)) {
1201 ics->predictor_reset_group = get_bits(gb, 5);
1202 if (ics->predictor_reset_group == 0 ||
1203 ics->predictor_reset_group > 30) {
1204 av_log(ac->avctx, AV_LOG_ERROR,
1205 "Invalid Predictor Reset Group.\n");
1206 return AVERROR_INVALIDDATA;
1209 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index]); sfb++) {
1210 ics->prediction_used[sfb] = get_bits1(gb);
1216 * Decode Long Term Prediction data; reference: table 4.xx.
1218 static void decode_ltp(LongTermPrediction *ltp,
1219 GetBitContext *gb, uint8_t max_sfb)
1223 ltp->lag = get_bits(gb, 11);
1224 ltp->coef = ltp_coef[get_bits(gb, 3)];
1225 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
1226 ltp->used[sfb] = get_bits1(gb);
1230 * Decode Individual Channel Stream info; reference: table 4.6.
1232 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
1235 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
1236 const int aot = m4ac->object_type;
1237 const int sampling_index = m4ac->sampling_index;
1238 if (aot != AOT_ER_AAC_ELD) {
1239 if (get_bits1(gb)) {
1240 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
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, float 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 - 90, 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) - 60;
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);
1418 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1420 } else if (band_type[idx] == NOISE_BT) {
1421 for (; i < run_end; i++, idx++) {
1422 if (noise_flag-- > 0)
1423 offset[1] += get_bits(gb, 9) - 256;
1425 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1426 clipped_offset = av_clip(offset[1], -100, 155);
1427 if (offset[1] != clipped_offset) {
1428 avpriv_request_sample(ac->avctx,
1429 "If you heard an audible artifact, there may be a bug in the decoder. "
1430 "Clipped noise gain (%d -> %d)",
1431 offset[1], clipped_offset);
1433 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1436 for (; i < run_end; i++, idx++) {
1437 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1438 if (offset[0] > 255U) {
1439 av_log(ac->avctx, AV_LOG_ERROR,
1440 "Scalefactor (%d) out of range.\n", offset[0]);
1441 return AVERROR_INVALIDDATA;
1443 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1452 * Decode pulse data; reference: table 4.7.
1454 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1455 const uint16_t *swb_offset, int num_swb)
1458 pulse->num_pulse = get_bits(gb, 2) + 1;
1459 pulse_swb = get_bits(gb, 6);
1460 if (pulse_swb >= num_swb)
1462 pulse->pos[0] = swb_offset[pulse_swb];
1463 pulse->pos[0] += get_bits(gb, 5);
1464 if (pulse->pos[0] >= swb_offset[num_swb])
1466 pulse->amp[0] = get_bits(gb, 4);
1467 for (i = 1; i < pulse->num_pulse; i++) {
1468 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1469 if (pulse->pos[i] >= swb_offset[num_swb])
1471 pulse->amp[i] = get_bits(gb, 4);
1477 * Decode Temporal Noise Shaping data; reference: table 4.48.
1479 * @return Returns error status. 0 - OK, !0 - error
1481 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1482 GetBitContext *gb, const IndividualChannelStream *ics)
1484 int w, filt, i, coef_len, coef_res, coef_compress;
1485 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1486 const int tns_max_order = is8 ? 7 : ac->oc[1].m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1487 for (w = 0; w < ics->num_windows; w++) {
1488 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1489 coef_res = get_bits1(gb);
1491 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1493 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1495 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1496 av_log(ac->avctx, AV_LOG_ERROR,
1497 "TNS filter order %d is greater than maximum %d.\n",
1498 tns->order[w][filt], tns_max_order);
1499 tns->order[w][filt] = 0;
1500 return AVERROR_INVALIDDATA;
1502 if (tns->order[w][filt]) {
1503 tns->direction[w][filt] = get_bits1(gb);
1504 coef_compress = get_bits1(gb);
1505 coef_len = coef_res + 3 - coef_compress;
1506 tmp2_idx = 2 * coef_compress + coef_res;
1508 for (i = 0; i < tns->order[w][filt]; i++)
1509 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1518 * Decode Mid/Side data; reference: table 4.54.
1520 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1521 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1522 * [3] reserved for scalable AAC
1524 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1528 int max_idx = cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb;
1529 if (ms_present == 1) {
1530 for (idx = 0; idx < max_idx; idx++)
1531 cpe->ms_mask[idx] = get_bits1(gb);
1532 } else if (ms_present == 2) {
1533 memset(cpe->ms_mask, 1, max_idx * sizeof(cpe->ms_mask[0]));
1538 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1542 *dst++ = v[idx & 15] * s;
1543 *dst++ = v[idx>>4 & 15] * s;
1549 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1553 *dst++ = v[idx & 3] * s;
1554 *dst++ = v[idx>>2 & 3] * s;
1555 *dst++ = v[idx>>4 & 3] * s;
1556 *dst++ = v[idx>>6 & 3] * s;
1562 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1563 unsigned sign, const float *scale)
1565 union av_intfloat32 s0, s1;
1567 s0.f = s1.f = *scale;
1568 s0.i ^= sign >> 1 << 31;
1571 *dst++ = v[idx & 15] * s0.f;
1572 *dst++ = v[idx>>4 & 15] * s1.f;
1579 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1580 unsigned sign, const float *scale)
1582 unsigned nz = idx >> 12;
1583 union av_intfloat32 s = { .f = *scale };
1584 union av_intfloat32 t;
1586 t.i = s.i ^ (sign & 1U<<31);
1587 *dst++ = v[idx & 3] * t.f;
1589 sign <<= nz & 1; nz >>= 1;
1590 t.i = s.i ^ (sign & 1U<<31);
1591 *dst++ = v[idx>>2 & 3] * t.f;
1593 sign <<= nz & 1; nz >>= 1;
1594 t.i = s.i ^ (sign & 1U<<31);
1595 *dst++ = v[idx>>4 & 3] * t.f;
1598 t.i = s.i ^ (sign & 1U<<31);
1599 *dst++ = v[idx>>6 & 3] * t.f;
1606 * Decode spectral data; reference: table 4.50.
1607 * Dequantize and scale spectral data; reference: 4.6.3.3.
1609 * @param coef array of dequantized, scaled spectral data
1610 * @param sf array of scalefactors or intensity stereo positions
1611 * @param pulse_present set if pulses are present
1612 * @param pulse pointer to pulse data struct
1613 * @param band_type array of the used band type
1615 * @return Returns error status. 0 - OK, !0 - error
1617 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1618 GetBitContext *gb, const float sf[120],
1619 int pulse_present, const Pulse *pulse,
1620 const IndividualChannelStream *ics,
1621 enum BandType band_type[120])
1623 int i, k, g, idx = 0;
1624 const int c = 1024 / ics->num_windows;
1625 const uint16_t *offsets = ics->swb_offset;
1626 float *coef_base = coef;
1628 for (g = 0; g < ics->num_windows; g++)
1629 memset(coef + g * 128 + offsets[ics->max_sfb], 0,
1630 sizeof(float) * (c - offsets[ics->max_sfb]));
1632 for (g = 0; g < ics->num_window_groups; g++) {
1633 unsigned g_len = ics->group_len[g];
1635 for (i = 0; i < ics->max_sfb; i++, idx++) {
1636 const unsigned cbt_m1 = band_type[idx] - 1;
1637 float *cfo = coef + offsets[i];
1638 int off_len = offsets[i + 1] - offsets[i];
1641 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1642 for (group = 0; group < g_len; group++, cfo+=128) {
1643 memset(cfo, 0, off_len * sizeof(float));
1645 } else if (cbt_m1 == NOISE_BT - 1) {
1646 for (group = 0; group < g_len; group++, cfo+=128) {
1650 for (k = 0; k < off_len; k++) {
1651 ac->random_state = lcg_random(ac->random_state);
1652 cfo[k] = ac->random_state;
1655 band_energy = ac->fdsp->scalarproduct_float(cfo, cfo, off_len);
1656 scale = sf[idx] / sqrtf(band_energy);
1657 ac->fdsp->vector_fmul_scalar(cfo, cfo, scale, off_len);
1660 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1661 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1662 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1663 OPEN_READER(re, gb);
1665 switch (cbt_m1 >> 1) {
1667 for (group = 0; group < g_len; group++, cfo+=128) {
1675 UPDATE_CACHE(re, gb);
1676 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1677 cb_idx = cb_vector_idx[code];
1678 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1684 for (group = 0; group < g_len; group++, cfo+=128) {
1694 UPDATE_CACHE(re, gb);
1695 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1696 cb_idx = cb_vector_idx[code];
1697 nnz = cb_idx >> 8 & 15;
1698 bits = nnz ? GET_CACHE(re, gb) : 0;
1699 LAST_SKIP_BITS(re, gb, nnz);
1700 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1706 for (group = 0; group < g_len; group++, cfo+=128) {
1714 UPDATE_CACHE(re, gb);
1715 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1716 cb_idx = cb_vector_idx[code];
1717 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1724 for (group = 0; group < g_len; group++, cfo+=128) {
1734 UPDATE_CACHE(re, gb);
1735 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1736 cb_idx = cb_vector_idx[code];
1737 nnz = cb_idx >> 8 & 15;
1738 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1739 LAST_SKIP_BITS(re, gb, nnz);
1740 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1746 for (group = 0; group < g_len; group++, cfo+=128) {
1748 uint32_t *icf = (uint32_t *) cf;
1758 UPDATE_CACHE(re, gb);
1759 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1767 cb_idx = cb_vector_idx[code];
1770 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1771 LAST_SKIP_BITS(re, gb, nnz);
1773 for (j = 0; j < 2; j++) {
1777 /* The total length of escape_sequence must be < 22 bits according
1778 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1779 UPDATE_CACHE(re, gb);
1780 b = GET_CACHE(re, gb);
1781 b = 31 - av_log2(~b);
1784 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1785 return AVERROR_INVALIDDATA;
1788 SKIP_BITS(re, gb, b + 1);
1790 n = (1 << b) + SHOW_UBITS(re, gb, b);
1791 LAST_SKIP_BITS(re, gb, b);
1792 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1795 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1796 *icf++ = (bits & 1U<<31) | v;
1803 ac->fdsp->vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1807 CLOSE_READER(re, gb);
1813 if (pulse_present) {
1815 for (i = 0; i < pulse->num_pulse; i++) {
1816 float co = coef_base[ pulse->pos[i] ];
1817 while (offsets[idx + 1] <= pulse->pos[i])
1819 if (band_type[idx] != NOISE_BT && sf[idx]) {
1820 float ico = -pulse->amp[i];
1823 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1825 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1832 static av_always_inline float flt16_round(float pf)
1834 union av_intfloat32 tmp;
1836 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1840 static av_always_inline float flt16_even(float pf)
1842 union av_intfloat32 tmp;
1844 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1848 static av_always_inline float flt16_trunc(float pf)
1850 union av_intfloat32 pun;
1852 pun.i &= 0xFFFF0000U;
1856 static av_always_inline void predict(PredictorState *ps, float *coef,
1859 const float a = 0.953125; // 61.0 / 64
1860 const float alpha = 0.90625; // 29.0 / 32
1864 float r0 = ps->r0, r1 = ps->r1;
1865 float cor0 = ps->cor0, cor1 = ps->cor1;
1866 float var0 = ps->var0, var1 = ps->var1;
1868 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1869 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1871 pv = flt16_round(k1 * r0 + k2 * r1);
1878 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1879 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1880 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1881 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1883 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1884 ps->r0 = flt16_trunc(a * e0);
1888 * Apply AAC-Main style frequency domain prediction.
1890 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1894 if (!sce->ics.predictor_initialized) {
1895 reset_all_predictors(sce->predictor_state);
1896 sce->ics.predictor_initialized = 1;
1899 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1901 sfb < ff_aac_pred_sfb_max[ac->oc[1].m4ac.sampling_index];
1903 for (k = sce->ics.swb_offset[sfb];
1904 k < sce->ics.swb_offset[sfb + 1];
1906 predict(&sce->predictor_state[k], &sce->coeffs[k],
1907 sce->ics.predictor_present &&
1908 sce->ics.prediction_used[sfb]);
1911 if (sce->ics.predictor_reset_group)
1912 reset_predictor_group(sce->predictor_state,
1913 sce->ics.predictor_reset_group);
1915 reset_all_predictors(sce->predictor_state);
1919 * Decode an individual_channel_stream payload; reference: table 4.44.
1921 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1922 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1924 * @return Returns error status. 0 - OK, !0 - error
1926 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1927 GetBitContext *gb, int common_window, int scale_flag)
1930 TemporalNoiseShaping *tns = &sce->tns;
1931 IndividualChannelStream *ics = &sce->ics;
1932 float *out = sce->coeffs;
1933 int global_gain, eld_syntax, er_syntax, pulse_present = 0;
1936 eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1937 er_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_LC ||
1938 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LTP ||
1939 ac->oc[1].m4ac.object_type == AOT_ER_AAC_LD ||
1940 ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
1942 /* This assignment is to silence a GCC warning about the variable being used
1943 * uninitialized when in fact it always is.
1945 pulse.num_pulse = 0;
1947 global_gain = get_bits(gb, 8);
1949 if (!common_window && !scale_flag) {
1950 if (decode_ics_info(ac, ics, gb) < 0)
1951 return AVERROR_INVALIDDATA;
1954 if ((ret = decode_band_types(ac, sce->band_type,
1955 sce->band_type_run_end, gb, ics)) < 0)
1957 if ((ret = decode_scalefactors(ac, sce->sf, gb, global_gain, ics,
1958 sce->band_type, sce->band_type_run_end)) < 0)
1963 if (!eld_syntax && (pulse_present = get_bits1(gb))) {
1964 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1965 av_log(ac->avctx, AV_LOG_ERROR,
1966 "Pulse tool not allowed in eight short sequence.\n");
1967 return AVERROR_INVALIDDATA;
1969 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1970 av_log(ac->avctx, AV_LOG_ERROR,
1971 "Pulse data corrupt or invalid.\n");
1972 return AVERROR_INVALIDDATA;
1975 tns->present = get_bits1(gb);
1976 if (tns->present && !er_syntax)
1977 if (decode_tns(ac, tns, gb, ics) < 0)
1978 return AVERROR_INVALIDDATA;
1979 if (!eld_syntax && get_bits1(gb)) {
1980 avpriv_request_sample(ac->avctx, "SSR");
1981 return AVERROR_PATCHWELCOME;
1983 // I see no textual basis in the spec for this occurring after SSR gain
1984 // control, but this is what both reference and real implmentations do
1985 if (tns->present && er_syntax)
1986 if (decode_tns(ac, tns, gb, ics) < 0)
1987 return AVERROR_INVALIDDATA;
1990 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present,
1991 &pulse, ics, sce->band_type) < 0)
1992 return AVERROR_INVALIDDATA;
1994 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN && !common_window)
1995 apply_prediction(ac, sce);
2001 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
2003 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
2005 const IndividualChannelStream *ics = &cpe->ch[0].ics;
2006 float *ch0 = cpe->ch[0].coeffs;
2007 float *ch1 = cpe->ch[1].coeffs;
2008 int g, i, group, idx = 0;
2009 const uint16_t *offsets = ics->swb_offset;
2010 for (g = 0; g < ics->num_window_groups; g++) {
2011 for (i = 0; i < ics->max_sfb; i++, idx++) {
2012 if (cpe->ms_mask[idx] &&
2013 cpe->ch[0].band_type[idx] < NOISE_BT &&
2014 cpe->ch[1].band_type[idx] < NOISE_BT) {
2015 for (group = 0; group < ics->group_len[g]; group++) {
2016 ac->fdsp->butterflies_float(ch0 + group * 128 + offsets[i],
2017 ch1 + group * 128 + offsets[i],
2018 offsets[i+1] - offsets[i]);
2022 ch0 += ics->group_len[g] * 128;
2023 ch1 += ics->group_len[g] * 128;
2028 * intensity stereo decoding; reference: 4.6.8.2.3
2030 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
2031 * [1] mask is decoded from bitstream; [2] mask is all 1s;
2032 * [3] reserved for scalable AAC
2034 static void apply_intensity_stereo(AACContext *ac,
2035 ChannelElement *cpe, int ms_present)
2037 const IndividualChannelStream *ics = &cpe->ch[1].ics;
2038 SingleChannelElement *sce1 = &cpe->ch[1];
2039 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
2040 const uint16_t *offsets = ics->swb_offset;
2041 int g, group, i, idx = 0;
2044 for (g = 0; g < ics->num_window_groups; g++) {
2045 for (i = 0; i < ics->max_sfb;) {
2046 if (sce1->band_type[idx] == INTENSITY_BT ||
2047 sce1->band_type[idx] == INTENSITY_BT2) {
2048 const int bt_run_end = sce1->band_type_run_end[idx];
2049 for (; i < bt_run_end; i++, idx++) {
2050 c = -1 + 2 * (sce1->band_type[idx] - 14);
2052 c *= 1 - 2 * cpe->ms_mask[idx];
2053 scale = c * sce1->sf[idx];
2054 for (group = 0; group < ics->group_len[g]; group++)
2055 ac->fdsp->vector_fmul_scalar(coef1 + group * 128 + offsets[i],
2056 coef0 + group * 128 + offsets[i],
2058 offsets[i + 1] - offsets[i]);
2061 int bt_run_end = sce1->band_type_run_end[idx];
2062 idx += bt_run_end - i;
2066 coef0 += ics->group_len[g] * 128;
2067 coef1 += ics->group_len[g] * 128;
2072 * Decode a channel_pair_element; reference: table 4.4.
2074 * @return Returns error status. 0 - OK, !0 - error
2076 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
2078 int i, ret, common_window, ms_present = 0;
2079 int eld_syntax = ac->oc[1].m4ac.object_type == AOT_ER_AAC_ELD;
2081 common_window = eld_syntax || get_bits1(gb);
2082 if (common_window) {
2083 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
2084 return AVERROR_INVALIDDATA;
2085 i = cpe->ch[1].ics.use_kb_window[0];
2086 cpe->ch[1].ics = cpe->ch[0].ics;
2087 cpe->ch[1].ics.use_kb_window[1] = i;
2088 if (cpe->ch[1].ics.predictor_present &&
2089 (ac->oc[1].m4ac.object_type != AOT_AAC_MAIN))
2090 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
2091 decode_ltp(&cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
2092 ms_present = get_bits(gb, 2);
2093 if (ms_present == 3) {
2094 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
2095 return AVERROR_INVALIDDATA;
2096 } else if (ms_present)
2097 decode_mid_side_stereo(cpe, gb, ms_present);
2099 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
2101 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
2104 if (common_window) {
2106 apply_mid_side_stereo(ac, cpe);
2107 if (ac->oc[1].m4ac.object_type == AOT_AAC_MAIN) {
2108 apply_prediction(ac, &cpe->ch[0]);
2109 apply_prediction(ac, &cpe->ch[1]);
2113 apply_intensity_stereo(ac, cpe, ms_present);
2117 static const float cce_scale[] = {
2118 1.09050773266525765921, //2^(1/8)
2119 1.18920711500272106672, //2^(1/4)
2125 * Decode coupling_channel_element; reference: table 4.8.
2127 * @return Returns error status. 0 - OK, !0 - error
2129 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
2135 SingleChannelElement *sce = &che->ch[0];
2136 ChannelCoupling *coup = &che->coup;
2138 coup->coupling_point = 2 * get_bits1(gb);
2139 coup->num_coupled = get_bits(gb, 3);
2140 for (c = 0; c <= coup->num_coupled; c++) {
2142 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
2143 coup->id_select[c] = get_bits(gb, 4);
2144 if (coup->type[c] == TYPE_CPE) {
2145 coup->ch_select[c] = get_bits(gb, 2);
2146 if (coup->ch_select[c] == 3)
2149 coup->ch_select[c] = 2;
2151 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
2153 sign = get_bits(gb, 1);
2154 scale = cce_scale[get_bits(gb, 2)];
2156 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
2159 for (c = 0; c < num_gain; c++) {
2163 float gain_cache = 1.0;
2165 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
2166 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
2167 gain_cache = powf(scale, -gain);
2169 if (coup->coupling_point == AFTER_IMDCT) {
2170 coup->gain[c][0] = gain_cache;
2172 for (g = 0; g < sce->ics.num_window_groups; g++) {
2173 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
2174 if (sce->band_type[idx] != ZERO_BT) {
2176 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
2184 gain_cache = powf(scale, -t) * s;
2187 coup->gain[c][idx] = gain_cache;
2197 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
2199 * @return Returns number of bytes consumed.
2201 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
2205 int num_excl_chan = 0;
2208 for (i = 0; i < 7; i++)
2209 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
2210 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
2212 return num_excl_chan / 7;
2216 * Decode dynamic range information; reference: table 4.52.
2218 * @return Returns number of bytes consumed.
2220 static int decode_dynamic_range(DynamicRangeControl *che_drc,
2224 int drc_num_bands = 1;
2227 /* pce_tag_present? */
2228 if (get_bits1(gb)) {
2229 che_drc->pce_instance_tag = get_bits(gb, 4);
2230 skip_bits(gb, 4); // tag_reserved_bits
2234 /* excluded_chns_present? */
2235 if (get_bits1(gb)) {
2236 n += decode_drc_channel_exclusions(che_drc, gb);
2239 /* drc_bands_present? */
2240 if (get_bits1(gb)) {
2241 che_drc->band_incr = get_bits(gb, 4);
2242 che_drc->interpolation_scheme = get_bits(gb, 4);
2244 drc_num_bands += che_drc->band_incr;
2245 for (i = 0; i < drc_num_bands; i++) {
2246 che_drc->band_top[i] = get_bits(gb, 8);
2251 /* prog_ref_level_present? */
2252 if (get_bits1(gb)) {
2253 che_drc->prog_ref_level = get_bits(gb, 7);
2254 skip_bits1(gb); // prog_ref_level_reserved_bits
2258 for (i = 0; i < drc_num_bands; i++) {
2259 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
2260 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
2267 static int decode_fill(AACContext *ac, GetBitContext *gb, int len) {
2269 int i, major, minor;
2274 get_bits(gb, 13); len -= 13;
2276 for(i=0; i+1<sizeof(buf) && len>=8; i++, len-=8)
2277 buf[i] = get_bits(gb, 8);
2280 if (ac->avctx->debug & FF_DEBUG_PICT_INFO)
2281 av_log(ac->avctx, AV_LOG_DEBUG, "FILL:%s\n", buf);
2283 if (sscanf(buf, "libfaac %d.%d", &major, &minor) == 2){
2284 ac->avctx->internal->skip_samples = 1024;
2288 skip_bits_long(gb, len);
2294 * Decode extension data (incomplete); reference: table 4.51.
2296 * @param cnt length of TYPE_FIL syntactic element in bytes
2298 * @return Returns number of bytes consumed
2300 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
2301 ChannelElement *che, enum RawDataBlockType elem_type)
2305 int type = get_bits(gb, 4);
2307 if (ac->avctx->debug & FF_DEBUG_STARTCODE)
2308 av_log(ac->avctx, AV_LOG_DEBUG, "extension type: %d len:%d\n", type, cnt);
2310 switch (type) { // extension type
2311 case EXT_SBR_DATA_CRC:
2315 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
2317 } else if (!ac->oc[1].m4ac.sbr) {
2318 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
2319 skip_bits_long(gb, 8 * cnt - 4);
2321 } else if (ac->oc[1].m4ac.sbr == -1 && ac->oc[1].status == OC_LOCKED) {
2322 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
2323 skip_bits_long(gb, 8 * cnt - 4);
2325 } else if (ac->oc[1].m4ac.ps == -1 && ac->oc[1].status < OC_LOCKED && ac->avctx->channels == 1) {
2326 ac->oc[1].m4ac.sbr = 1;
2327 ac->oc[1].m4ac.ps = 1;
2328 ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
2329 output_configure(ac, ac->oc[1].layout_map, ac->oc[1].layout_map_tags,
2330 ac->oc[1].status, 1);
2332 ac->oc[1].m4ac.sbr = 1;
2333 ac->avctx->profile = FF_PROFILE_AAC_HE;
2335 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
2337 case EXT_DYNAMIC_RANGE:
2338 res = decode_dynamic_range(&ac->che_drc, gb);
2341 decode_fill(ac, gb, 8 * cnt - 4);
2344 case EXT_DATA_ELEMENT:
2346 skip_bits_long(gb, 8 * cnt - 4);
2353 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
2355 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
2356 * @param coef spectral coefficients
2358 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
2359 IndividualChannelStream *ics, int decode)
2361 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
2363 int bottom, top, order, start, end, size, inc;
2364 float lpc[TNS_MAX_ORDER];
2365 float tmp[TNS_MAX_ORDER+1];
2367 for (w = 0; w < ics->num_windows; w++) {
2368 bottom = ics->num_swb;
2369 for (filt = 0; filt < tns->n_filt[w]; filt++) {
2371 bottom = FFMAX(0, top - tns->length[w][filt]);
2372 order = tns->order[w][filt];
2377 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
2379 start = ics->swb_offset[FFMIN(bottom, mmm)];
2380 end = ics->swb_offset[FFMIN( top, mmm)];
2381 if ((size = end - start) <= 0)
2383 if (tns->direction[w][filt]) {
2393 for (m = 0; m < size; m++, start += inc)
2394 for (i = 1; i <= FFMIN(m, order); i++)
2395 coef[start] -= coef[start - i * inc] * lpc[i - 1];
2398 for (m = 0; m < size; m++, start += inc) {
2399 tmp[0] = coef[start];
2400 for (i = 1; i <= FFMIN(m, order); i++)
2401 coef[start] += tmp[i] * lpc[i - 1];
2402 for (i = order; i > 0; i--)
2403 tmp[i] = tmp[i - 1];
2411 * Apply windowing and MDCT to obtain the spectral
2412 * coefficient from the predicted sample by LTP.
2414 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2415 float *in, IndividualChannelStream *ics)
2417 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2418 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2419 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2420 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2422 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2423 ac->fdsp->vector_fmul(in, in, lwindow_prev, 1024);
2425 memset(in, 0, 448 * sizeof(float));
2426 ac->fdsp->vector_fmul(in + 448, in + 448, swindow_prev, 128);
2428 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2429 ac->fdsp->vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2431 ac->fdsp->vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2432 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2434 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2438 * Apply the long term prediction
2440 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2442 const LongTermPrediction *ltp = &sce->ics.ltp;
2443 const uint16_t *offsets = sce->ics.swb_offset;
2446 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2447 float *predTime = sce->ret;
2448 float *predFreq = ac->buf_mdct;
2449 int16_t num_samples = 2048;
2451 if (ltp->lag < 1024)
2452 num_samples = ltp->lag + 1024;
2453 for (i = 0; i < num_samples; i++)
2454 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2455 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2457 ac->windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2459 if (sce->tns.present)
2460 ac->apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2462 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2464 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2465 sce->coeffs[i] += predFreq[i];
2470 * Update the LTP buffer for next frame
2472 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2474 IndividualChannelStream *ics = &sce->ics;
2475 float *saved = sce->saved;
2476 float *saved_ltp = sce->coeffs;
2477 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2478 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2481 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2482 memcpy(saved_ltp, saved, 512 * sizeof(float));
2483 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2484 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2485 for (i = 0; i < 64; i++)
2486 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2487 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2488 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2489 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2490 ac->fdsp->vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2491 for (i = 0; i < 64; i++)
2492 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2493 } else { // LONG_STOP or ONLY_LONG
2494 ac->fdsp->vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2495 for (i = 0; i < 512; i++)
2496 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2499 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2500 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2501 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2505 * Conduct IMDCT and windowing.
2507 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2509 IndividualChannelStream *ics = &sce->ics;
2510 float *in = sce->coeffs;
2511 float *out = sce->ret;
2512 float *saved = sce->saved;
2513 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2514 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2515 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2516 float *buf = ac->buf_mdct;
2517 float *temp = ac->temp;
2521 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2522 for (i = 0; i < 1024; i += 128)
2523 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2525 ac->mdct.imdct_half(&ac->mdct, buf, in);
2527 /* window overlapping
2528 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2529 * and long to short transitions are considered to be short to short
2530 * transitions. This leaves just two cases (long to long and short to short)
2531 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2533 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2534 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2535 ac->fdsp->vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2537 memcpy( out, saved, 448 * sizeof(float));
2539 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2540 ac->fdsp->vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2541 ac->fdsp->vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2542 ac->fdsp->vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2543 ac->fdsp->vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2544 ac->fdsp->vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2545 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2547 ac->fdsp->vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2548 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2553 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2554 memcpy( saved, temp + 64, 64 * sizeof(float));
2555 ac->fdsp->vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2556 ac->fdsp->vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2557 ac->fdsp->vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2558 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2559 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2560 memcpy( saved, buf + 512, 448 * sizeof(float));
2561 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2562 } else { // LONG_STOP or ONLY_LONG
2563 memcpy( saved, buf + 512, 512 * sizeof(float));
2567 static void imdct_and_windowing_ld(AACContext *ac, SingleChannelElement *sce)
2569 IndividualChannelStream *ics = &sce->ics;
2570 float *in = sce->coeffs;
2571 float *out = sce->ret;
2572 float *saved = sce->saved;
2573 float *buf = ac->buf_mdct;
2576 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2578 // window overlapping
2579 if (ics->use_kb_window[1]) {
2580 // AAC LD uses a low overlap sine window instead of a KBD window
2581 memcpy(out, saved, 192 * sizeof(float));
2582 ac->fdsp->vector_fmul_window(out + 192, saved + 192, buf, ff_sine_128, 64);
2583 memcpy( out + 320, buf + 64, 192 * sizeof(float));
2585 ac->fdsp->vector_fmul_window(out, saved, buf, ff_sine_512, 256);
2589 memcpy(saved, buf + 256, 256 * sizeof(float));
2592 static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
2594 float *in = sce->coeffs;
2595 float *out = sce->ret;
2596 float *saved = sce->saved;
2597 float *buf = ac->buf_mdct;
2599 const int n = ac->oc[1].m4ac.frame_length_short ? 480 : 512;
2600 const int n2 = n >> 1;
2601 const int n4 = n >> 2;
2602 const float *const window = n == 480 ? ff_aac_eld_window_480 :
2603 ff_aac_eld_window_512;
2605 // Inverse transform, mapped to the conventional IMDCT by
2606 // Chivukula, R.K.; Reznik, Y.A.; Devarajan, V.,
2607 // "Efficient algorithms for MPEG-4 AAC-ELD, AAC-LD and AAC-LC filterbanks,"
2608 // International Conference on Audio, Language and Image Processing, ICALIP 2008.
2609 // URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4590245&isnumber=4589950
2610 for (i = 0; i < n2; i+=2) {
2612 temp = in[i ]; in[i ] = -in[n - 1 - i]; in[n - 1 - i] = temp;
2613 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));
2618 ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
2619 for (i = 0; i < n; i+=2) {
2622 // Like with the regular IMDCT at this point we still have the middle half
2623 // of a transform but with even symmetry on the left and odd symmetry on
2626 // window overlapping
2627 // The spec says to use samples [0..511] but the reference decoder uses
2628 // samples [128..639].
2629 for (i = n4; i < n2; i ++) {
2630 out[i - n4] = buf[n2 - 1 - i] * window[i - n4] +
2631 saved[ i + n2] * window[i + n - n4] +
2632 -saved[ n + n2 - 1 - i] * window[i + 2*n - n4] +
2633 -saved[2*n + n2 + i] * window[i + 3*n - n4];
2635 for (i = 0; i < n2; i ++) {
2636 out[n4 + i] = buf[i] * window[i + n2 - n4] +
2637 -saved[ n - 1 - i] * window[i + n2 + n - n4] +
2638 -saved[ n + i] * window[i + n2 + 2*n - n4] +
2639 saved[2*n + n - 1 - i] * window[i + n2 + 3*n - n4];
2641 for (i = 0; i < n4; i ++) {
2642 out[n2 + n4 + i] = buf[ i + n2] * window[i + n - n4] +
2643 -saved[ n2 - 1 - i] * window[i + 2*n - n4] +
2644 -saved[ n + n2 + i] * window[i + 3*n - n4];
2648 memmove(saved + n, saved, 2 * n * sizeof(float));
2649 memcpy( saved, buf, n * sizeof(float));
2653 * Apply dependent channel coupling (applied before IMDCT).
2655 * @param index index into coupling gain array
2657 static void apply_dependent_coupling(AACContext *ac,
2658 SingleChannelElement *target,
2659 ChannelElement *cce, int index)
2661 IndividualChannelStream *ics = &cce->ch[0].ics;
2662 const uint16_t *offsets = ics->swb_offset;
2663 float *dest = target->coeffs;
2664 const float *src = cce->ch[0].coeffs;
2665 int g, i, group, k, idx = 0;
2666 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2667 av_log(ac->avctx, AV_LOG_ERROR,
2668 "Dependent coupling is not supported together with LTP\n");
2671 for (g = 0; g < ics->num_window_groups; g++) {
2672 for (i = 0; i < ics->max_sfb; i++, idx++) {
2673 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2674 const float gain = cce->coup.gain[index][idx];
2675 for (group = 0; group < ics->group_len[g]; group++) {
2676 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2678 dest[group * 128 + k] += gain * src[group * 128 + k];
2683 dest += ics->group_len[g] * 128;
2684 src += ics->group_len[g] * 128;
2689 * Apply independent channel coupling (applied after IMDCT).
2691 * @param index index into coupling gain array
2693 static void apply_independent_coupling(AACContext *ac,
2694 SingleChannelElement *target,
2695 ChannelElement *cce, int index)
2698 const float gain = cce->coup.gain[index][0];
2699 const float *src = cce->ch[0].ret;
2700 float *dest = target->ret;
2701 const int len = 1024 << (ac->oc[1].m4ac.sbr == 1);
2703 for (i = 0; i < len; i++)
2704 dest[i] += gain * src[i];
2708 * channel coupling transformation interface
2710 * @param apply_coupling_method pointer to (in)dependent coupling function
2712 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2713 enum RawDataBlockType type, int elem_id,
2714 enum CouplingPoint coupling_point,
2715 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2719 for (i = 0; i < MAX_ELEM_ID; i++) {
2720 ChannelElement *cce = ac->che[TYPE_CCE][i];
2723 if (cce && cce->coup.coupling_point == coupling_point) {
2724 ChannelCoupling *coup = &cce->coup;
2726 for (c = 0; c <= coup->num_coupled; c++) {
2727 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2728 if (coup->ch_select[c] != 1) {
2729 apply_coupling_method(ac, &cc->ch[0], cce, index);
2730 if (coup->ch_select[c] != 0)
2733 if (coup->ch_select[c] != 2)
2734 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2736 index += 1 + (coup->ch_select[c] == 3);
2743 * Convert spectral data to float samples, applying all supported tools as appropriate.
2745 static void spectral_to_sample(AACContext *ac)
2748 void (*imdct_and_window)(AACContext *ac, SingleChannelElement *sce);
2749 switch (ac->oc[1].m4ac.object_type) {
2751 imdct_and_window = imdct_and_windowing_ld;
2753 case AOT_ER_AAC_ELD:
2754 imdct_and_window = imdct_and_windowing_eld;
2757 imdct_and_window = ac->imdct_and_windowing;
2759 for (type = 3; type >= 0; type--) {
2760 for (i = 0; i < MAX_ELEM_ID; i++) {
2761 ChannelElement *che = ac->che[type][i];
2762 if (che && che->present) {
2763 if (type <= TYPE_CPE)
2764 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2765 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP) {
2766 if (che->ch[0].ics.predictor_present) {
2767 if (che->ch[0].ics.ltp.present)
2768 ac->apply_ltp(ac, &che->ch[0]);
2769 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2770 ac->apply_ltp(ac, &che->ch[1]);
2773 if (che->ch[0].tns.present)
2774 ac->apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2775 if (che->ch[1].tns.present)
2776 ac->apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2777 if (type <= TYPE_CPE)
2778 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2779 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2780 imdct_and_window(ac, &che->ch[0]);
2781 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2782 ac->update_ltp(ac, &che->ch[0]);
2783 if (type == TYPE_CPE) {
2784 imdct_and_window(ac, &che->ch[1]);
2785 if (ac->oc[1].m4ac.object_type == AOT_AAC_LTP)
2786 ac->update_ltp(ac, &che->ch[1]);
2788 if (ac->oc[1].m4ac.sbr > 0) {
2789 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2792 if (type <= TYPE_CCE)
2793 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2796 av_log(ac->avctx, AV_LOG_VERBOSE, "ChannelElement %d.%d missing \n", type, i);
2802 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2805 AACADTSHeaderInfo hdr_info;
2806 uint8_t layout_map[MAX_ELEM_ID*4][3];
2807 int layout_map_tags, ret;
2809 size = avpriv_aac_parse_header(gb, &hdr_info);
2811 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2812 // This is 2 for "VLB " audio in NSV files.
2813 // See samples/nsv/vlb_audio.
2814 avpriv_report_missing_feature(ac->avctx,
2815 "More than one AAC RDB per ADTS frame");
2816 ac->warned_num_aac_frames = 1;
2818 push_output_configuration(ac);
2819 if (hdr_info.chan_config) {
2820 ac->oc[1].m4ac.chan_config = hdr_info.chan_config;
2821 if ((ret = set_default_channel_config(ac->avctx,
2824 hdr_info.chan_config)) < 0)
2826 if ((ret = output_configure(ac, layout_map, layout_map_tags,
2827 FFMAX(ac->oc[1].status,
2828 OC_TRIAL_FRAME), 0)) < 0)
2831 ac->oc[1].m4ac.chan_config = 0;
2833 * dual mono frames in Japanese DTV can have chan_config 0
2834 * WITHOUT specifying PCE.
2835 * thus, set dual mono as default.
2837 if (ac->dmono_mode && ac->oc[0].status == OC_NONE) {
2838 layout_map_tags = 2;
2839 layout_map[0][0] = layout_map[1][0] = TYPE_SCE;
2840 layout_map[0][2] = layout_map[1][2] = AAC_CHANNEL_FRONT;
2841 layout_map[0][1] = 0;
2842 layout_map[1][1] = 1;
2843 if (output_configure(ac, layout_map, layout_map_tags,
2848 ac->oc[1].m4ac.sample_rate = hdr_info.sample_rate;
2849 ac->oc[1].m4ac.sampling_index = hdr_info.sampling_index;
2850 ac->oc[1].m4ac.object_type = hdr_info.object_type;
2851 ac->oc[1].m4ac.frame_length_short = 0;
2852 if (ac->oc[0].status != OC_LOCKED ||
2853 ac->oc[0].m4ac.chan_config != hdr_info.chan_config ||
2854 ac->oc[0].m4ac.sample_rate != hdr_info.sample_rate) {
2855 ac->oc[1].m4ac.sbr = -1;
2856 ac->oc[1].m4ac.ps = -1;
2858 if (!hdr_info.crc_absent)
2864 static int aac_decode_er_frame(AVCodecContext *avctx, void *data,
2865 int *got_frame_ptr, GetBitContext *gb)
2867 AACContext *ac = avctx->priv_data;
2868 const MPEG4AudioConfig *const m4ac = &ac->oc[1].m4ac;
2869 ChannelElement *che;
2871 int samples = m4ac->frame_length_short ? 960 : 1024;
2872 int chan_config = m4ac->chan_config;
2873 int aot = m4ac->object_type;
2875 if (aot == AOT_ER_AAC_LD || aot == AOT_ER_AAC_ELD)
2880 if ((err = frame_configure_elements(avctx)) < 0)
2883 // The FF_PROFILE_AAC_* defines are all object_type - 1
2884 // This may lead to an undefined profile being signaled
2885 ac->avctx->profile = aot - 1;
2887 ac->tags_mapped = 0;
2889 if (chan_config < 0 || chan_config >= 8) {
2890 avpriv_request_sample(avctx, "Unknown ER channel configuration %d",
2892 return AVERROR_INVALIDDATA;
2894 for (i = 0; i < tags_per_config[chan_config]; i++) {
2895 const int elem_type = aac_channel_layout_map[chan_config-1][i][0];
2896 const int elem_id = aac_channel_layout_map[chan_config-1][i][1];
2897 if (!(che=get_che(ac, elem_type, elem_id))) {
2898 av_log(ac->avctx, AV_LOG_ERROR,
2899 "channel element %d.%d is not allocated\n",
2900 elem_type, elem_id);
2901 return AVERROR_INVALIDDATA;
2904 if (aot != AOT_ER_AAC_ELD)
2906 switch (elem_type) {
2908 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2911 err = decode_cpe(ac, gb, che);
2914 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2921 spectral_to_sample(ac);
2923 ac->frame->nb_samples = samples;
2924 ac->frame->sample_rate = avctx->sample_rate;
2927 skip_bits_long(gb, get_bits_left(gb));
2931 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2932 int *got_frame_ptr, GetBitContext *gb, AVPacket *avpkt)
2934 AACContext *ac = avctx->priv_data;
2935 ChannelElement *che = NULL, *che_prev = NULL;
2936 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2938 int samples = 0, multiplier, audio_found = 0, pce_found = 0;
2939 int is_dmono, sce_count = 0;
2943 if (show_bits(gb, 12) == 0xfff) {
2944 if ((err = parse_adts_frame_header(ac, gb)) < 0) {
2945 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2948 if (ac->oc[1].m4ac.sampling_index > 12) {
2949 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->oc[1].m4ac.sampling_index);
2950 err = AVERROR_INVALIDDATA;
2955 if ((err = frame_configure_elements(avctx)) < 0)
2958 // The FF_PROFILE_AAC_* defines are all object_type - 1
2959 // This may lead to an undefined profile being signaled
2960 ac->avctx->profile = ac->oc[1].m4ac.object_type - 1;
2962 ac->tags_mapped = 0;
2964 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2965 elem_id = get_bits(gb, 4);
2967 if (avctx->debug & FF_DEBUG_STARTCODE)
2968 av_log(avctx, AV_LOG_DEBUG, "Elem type:%x id:%x\n", elem_type, elem_id);
2970 if (elem_type < TYPE_DSE) {
2971 if (!(che=get_che(ac, elem_type, elem_id))) {
2972 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2973 elem_type, elem_id);
2974 err = AVERROR_INVALIDDATA;
2981 switch (elem_type) {
2984 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2990 err = decode_cpe(ac, gb, che);
2995 err = decode_cce(ac, gb, che);
2999 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
3004 err = skip_data_stream_element(ac, gb);
3008 uint8_t layout_map[MAX_ELEM_ID*4][3];
3010 push_output_configuration(ac);
3011 tags = decode_pce(avctx, &ac->oc[1].m4ac, layout_map, gb);
3017 av_log(avctx, AV_LOG_ERROR,
3018 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
3020 err = output_configure(ac, layout_map, tags, OC_TRIAL_PCE, 1);
3022 ac->oc[1].m4ac.chan_config = 0;
3030 elem_id += get_bits(gb, 8) - 1;
3031 if (get_bits_left(gb) < 8 * elem_id) {
3032 av_log(avctx, AV_LOG_ERROR, "TYPE_FIL: "overread_err);
3033 err = AVERROR_INVALIDDATA;
3037 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
3038 err = 0; /* FIXME */
3042 err = AVERROR_BUG; /* should not happen, but keeps compiler happy */
3047 elem_type_prev = elem_type;
3052 if (get_bits_left(gb) < 3) {
3053 av_log(avctx, AV_LOG_ERROR, overread_err);
3054 err = AVERROR_INVALIDDATA;
3059 spectral_to_sample(ac);
3061 multiplier = (ac->oc[1].m4ac.sbr == 1) ? ac->oc[1].m4ac.ext_sample_rate > ac->oc[1].m4ac.sample_rate : 0;
3062 samples <<= multiplier;
3064 if (ac->oc[1].status && audio_found) {
3065 avctx->sample_rate = ac->oc[1].m4ac.sample_rate << multiplier;
3066 avctx->frame_size = samples;
3067 ac->oc[1].status = OC_LOCKED;
3072 const uint8_t *side = av_packet_get_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, &side_size);
3073 if (side && side_size>=4)
3074 AV_WL32(side, 2*AV_RL32(side));
3077 *got_frame_ptr = !!samples;
3079 ac->frame->nb_samples = samples;
3080 ac->frame->sample_rate = avctx->sample_rate;
3082 av_frame_unref(ac->frame);
3083 *got_frame_ptr = !!samples;
3085 /* for dual-mono audio (SCE + SCE) */
3086 is_dmono = ac->dmono_mode && sce_count == 2 &&
3087 ac->oc[1].channel_layout == (AV_CH_FRONT_LEFT | AV_CH_FRONT_RIGHT);
3089 if (ac->dmono_mode == 1)
3090 ((AVFrame *)data)->data[1] =((AVFrame *)data)->data[0];
3091 else if (ac->dmono_mode == 2)
3092 ((AVFrame *)data)->data[0] =((AVFrame *)data)->data[1];
3097 pop_output_configuration(ac);
3101 static int aac_decode_frame(AVCodecContext *avctx, void *data,
3102 int *got_frame_ptr, AVPacket *avpkt)
3104 AACContext *ac = avctx->priv_data;
3105 const uint8_t *buf = avpkt->data;
3106 int buf_size = avpkt->size;
3111 int new_extradata_size;
3112 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
3113 AV_PKT_DATA_NEW_EXTRADATA,
3114 &new_extradata_size);
3115 int jp_dualmono_size;
3116 const uint8_t *jp_dualmono = av_packet_get_side_data(avpkt,
3117 AV_PKT_DATA_JP_DUALMONO,
3120 if (new_extradata && 0) {
3121 av_free(avctx->extradata);
3122 avctx->extradata = av_mallocz(new_extradata_size +
3123 FF_INPUT_BUFFER_PADDING_SIZE);
3124 if (!avctx->extradata)
3125 return AVERROR(ENOMEM);
3126 avctx->extradata_size = new_extradata_size;
3127 memcpy(avctx->extradata, new_extradata, new_extradata_size);
3128 push_output_configuration(ac);
3129 if (decode_audio_specific_config(ac, ac->avctx, &ac->oc[1].m4ac,
3131 avctx->extradata_size*8, 1) < 0) {
3132 pop_output_configuration(ac);
3133 return AVERROR_INVALIDDATA;
3138 if (jp_dualmono && jp_dualmono_size > 0)
3139 ac->dmono_mode = 1 + *jp_dualmono;
3140 if (ac->force_dmono_mode >= 0)
3141 ac->dmono_mode = ac->force_dmono_mode;
3143 if (INT_MAX / 8 <= buf_size)
3144 return AVERROR_INVALIDDATA;
3146 if ((err = init_get_bits(&gb, buf, buf_size * 8)) < 0)
3149 switch (ac->oc[1].m4ac.object_type) {
3151 case AOT_ER_AAC_LTP:
3153 case AOT_ER_AAC_ELD:
3154 err = aac_decode_er_frame(avctx, data, got_frame_ptr, &gb);
3157 err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb, avpkt);
3162 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
3163 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
3164 if (buf[buf_offset])
3167 return buf_size > buf_offset ? buf_consumed : buf_size;
3170 static av_cold int aac_decode_close(AVCodecContext *avctx)
3172 AACContext *ac = avctx->priv_data;
3175 for (i = 0; i < MAX_ELEM_ID; i++) {
3176 for (type = 0; type < 4; type++) {
3177 if (ac->che[type][i])
3178 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
3179 av_freep(&ac->che[type][i]);
3183 ff_mdct_end(&ac->mdct);
3184 ff_mdct_end(&ac->mdct_small);
3185 ff_mdct_end(&ac->mdct_ld);
3186 ff_mdct_end(&ac->mdct_ltp);
3187 ff_imdct15_uninit(&ac->mdct480);
3188 av_freep(&ac->fdsp);
3193 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
3195 struct LATMContext {
3196 AACContext aac_ctx; ///< containing AACContext
3197 int initialized; ///< initialized after a valid extradata was seen
3200 int audio_mux_version_A; ///< LATM syntax version
3201 int frame_length_type; ///< 0/1 variable/fixed frame length
3202 int frame_length; ///< frame length for fixed frame length
3205 static inline uint32_t latm_get_value(GetBitContext *b)
3207 int length = get_bits(b, 2);
3209 return get_bits_long(b, (length+1)*8);
3212 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
3213 GetBitContext *gb, int asclen)
3215 AACContext *ac = &latmctx->aac_ctx;
3216 AVCodecContext *avctx = ac->avctx;
3217 MPEG4AudioConfig m4ac = { 0 };
3218 int config_start_bit = get_bits_count(gb);
3219 int sync_extension = 0;
3220 int bits_consumed, esize;
3224 asclen = FFMIN(asclen, get_bits_left(gb));
3226 asclen = get_bits_left(gb);
3228 if (config_start_bit % 8) {
3229 avpriv_request_sample(latmctx->aac_ctx.avctx,
3230 "Non-byte-aligned audio-specific config");
3231 return AVERROR_PATCHWELCOME;
3234 return AVERROR_INVALIDDATA;
3235 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
3236 gb->buffer + (config_start_bit / 8),
3237 asclen, sync_extension);
3239 if (bits_consumed < 0)
3240 return AVERROR_INVALIDDATA;
3242 if (!latmctx->initialized ||
3243 ac->oc[1].m4ac.sample_rate != m4ac.sample_rate ||
3244 ac->oc[1].m4ac.chan_config != m4ac.chan_config) {
3246 if(latmctx->initialized) {
3247 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
3249 av_log(avctx, AV_LOG_DEBUG, "initializing latmctx\n");
3251 latmctx->initialized = 0;
3253 esize = (bits_consumed+7) / 8;
3255 if (avctx->extradata_size < esize) {
3256 av_free(avctx->extradata);
3257 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
3258 if (!avctx->extradata)
3259 return AVERROR(ENOMEM);
3262 avctx->extradata_size = esize;
3263 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
3264 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
3266 skip_bits_long(gb, bits_consumed);
3268 return bits_consumed;
3271 static int read_stream_mux_config(struct LATMContext *latmctx,
3274 int ret, audio_mux_version = get_bits(gb, 1);
3276 latmctx->audio_mux_version_A = 0;
3277 if (audio_mux_version)
3278 latmctx->audio_mux_version_A = get_bits(gb, 1);
3280 if (!latmctx->audio_mux_version_A) {
3282 if (audio_mux_version)
3283 latm_get_value(gb); // taraFullness
3285 skip_bits(gb, 1); // allStreamSameTimeFraming
3286 skip_bits(gb, 6); // numSubFrames
3288 if (get_bits(gb, 4)) { // numPrograms
3289 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple programs");
3290 return AVERROR_PATCHWELCOME;
3293 // for each program (which there is only one in DVB)
3295 // for each layer (which there is only one in DVB)
3296 if (get_bits(gb, 3)) { // numLayer
3297 avpriv_request_sample(latmctx->aac_ctx.avctx, "Multiple layers");
3298 return AVERROR_PATCHWELCOME;
3301 // for all but first stream: use_same_config = get_bits(gb, 1);
3302 if (!audio_mux_version) {
3303 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
3306 int ascLen = latm_get_value(gb);
3307 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
3310 skip_bits_long(gb, ascLen);
3313 latmctx->frame_length_type = get_bits(gb, 3);
3314 switch (latmctx->frame_length_type) {
3316 skip_bits(gb, 8); // latmBufferFullness
3319 latmctx->frame_length = get_bits(gb, 9);
3324 skip_bits(gb, 6); // CELP frame length table index
3328 skip_bits(gb, 1); // HVXC frame length table index
3332 if (get_bits(gb, 1)) { // other data
3333 if (audio_mux_version) {
3334 latm_get_value(gb); // other_data_bits
3338 esc = get_bits(gb, 1);
3344 if (get_bits(gb, 1)) // crc present
3345 skip_bits(gb, 8); // config_crc
3351 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
3355 if (ctx->frame_length_type == 0) {
3356 int mux_slot_length = 0;
3358 tmp = get_bits(gb, 8);
3359 mux_slot_length += tmp;
3360 } while (tmp == 255);
3361 return mux_slot_length;
3362 } else if (ctx->frame_length_type == 1) {
3363 return ctx->frame_length;
3364 } else if (ctx->frame_length_type == 3 ||
3365 ctx->frame_length_type == 5 ||
3366 ctx->frame_length_type == 7) {
3367 skip_bits(gb, 2); // mux_slot_length_coded
3372 static int read_audio_mux_element(struct LATMContext *latmctx,
3376 uint8_t use_same_mux = get_bits(gb, 1);
3377 if (!use_same_mux) {
3378 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
3380 } else if (!latmctx->aac_ctx.avctx->extradata) {
3381 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
3382 "no decoder config found\n");
3383 return AVERROR(EAGAIN);
3385 if (latmctx->audio_mux_version_A == 0) {
3386 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
3387 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
3388 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
3389 return AVERROR_INVALIDDATA;
3390 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
3391 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3392 "frame length mismatch %d << %d\n",
3393 mux_slot_length_bytes * 8, get_bits_left(gb));
3394 return AVERROR_INVALIDDATA;
3401 static int latm_decode_frame(AVCodecContext *avctx, void *out,
3402 int *got_frame_ptr, AVPacket *avpkt)
3404 struct LATMContext *latmctx = avctx->priv_data;
3408 if ((err = init_get_bits8(&gb, avpkt->data, avpkt->size)) < 0)
3411 // check for LOAS sync word
3412 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
3413 return AVERROR_INVALIDDATA;
3415 muxlength = get_bits(&gb, 13) + 3;
3416 // not enough data, the parser should have sorted this out
3417 if (muxlength > avpkt->size)
3418 return AVERROR_INVALIDDATA;
3420 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
3423 if (!latmctx->initialized) {
3424 if (!avctx->extradata) {
3428 push_output_configuration(&latmctx->aac_ctx);
3429 if ((err = decode_audio_specific_config(
3430 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.oc[1].m4ac,
3431 avctx->extradata, avctx->extradata_size*8, 1)) < 0) {
3432 pop_output_configuration(&latmctx->aac_ctx);
3435 latmctx->initialized = 1;
3439 if (show_bits(&gb, 12) == 0xfff) {
3440 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
3441 "ADTS header detected, probably as result of configuration "
3443 return AVERROR_INVALIDDATA;
3446 switch (latmctx->aac_ctx.oc[1].m4ac.object_type) {
3448 case AOT_ER_AAC_LTP:
3450 case AOT_ER_AAC_ELD:
3451 err = aac_decode_er_frame(avctx, out, got_frame_ptr, &gb);
3454 err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb, avpkt);
3462 static av_cold int latm_decode_init(AVCodecContext *avctx)
3464 struct LATMContext *latmctx = avctx->priv_data;
3465 int ret = aac_decode_init(avctx);
3467 if (avctx->extradata_size > 0)
3468 latmctx->initialized = !ret;
3473 static void aacdec_init(AACContext *c)
3475 c->imdct_and_windowing = imdct_and_windowing;
3476 c->apply_ltp = apply_ltp;
3477 c->apply_tns = apply_tns;
3478 c->windowing_and_mdct_ltp = windowing_and_mdct_ltp;
3479 c->update_ltp = update_ltp;
3482 ff_aacdec_init_mips(c);
3485 * AVOptions for Japanese DTV specific extensions (ADTS only)
3487 #define AACDEC_FLAGS AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
3488 static const AVOption options[] = {
3489 {"dual_mono_mode", "Select the channel to decode for dual mono",
3490 offsetof(AACContext, force_dmono_mode), AV_OPT_TYPE_INT, {.i64=-1}, -1, 2,
3491 AACDEC_FLAGS, "dual_mono_mode"},
3493 {"auto", "autoselection", 0, AV_OPT_TYPE_CONST, {.i64=-1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3494 {"main", "Select Main/Left channel", 0, AV_OPT_TYPE_CONST, {.i64= 1}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3495 {"sub" , "Select Sub/Right channel", 0, AV_OPT_TYPE_CONST, {.i64= 2}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3496 {"both", "Select both channels", 0, AV_OPT_TYPE_CONST, {.i64= 0}, INT_MIN, INT_MAX, AACDEC_FLAGS, "dual_mono_mode"},
3501 static const AVClass aac_decoder_class = {
3502 .class_name = "AAC decoder",
3503 .item_name = av_default_item_name,
3505 .version = LIBAVUTIL_VERSION_INT,
3508 static const AVProfile profiles[] = {
3509 { FF_PROFILE_AAC_MAIN, "Main" },
3510 { FF_PROFILE_AAC_LOW, "LC" },
3511 { FF_PROFILE_AAC_SSR, "SSR" },
3512 { FF_PROFILE_AAC_LTP, "LTP" },
3513 { FF_PROFILE_AAC_HE, "HE-AAC" },
3514 { FF_PROFILE_AAC_HE_V2, "HE-AACv2" },
3515 { FF_PROFILE_AAC_LD, "LD" },
3516 { FF_PROFILE_AAC_ELD, "ELD" },
3517 { FF_PROFILE_UNKNOWN },
3520 AVCodec ff_aac_decoder = {
3522 .long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
3523 .type = AVMEDIA_TYPE_AUDIO,
3524 .id = AV_CODEC_ID_AAC,
3525 .priv_data_size = sizeof(AACContext),
3526 .init = aac_decode_init,
3527 .close = aac_decode_close,
3528 .decode = aac_decode_frame,
3529 .sample_fmts = (const enum AVSampleFormat[]) {
3530 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3532 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3533 .channel_layouts = aac_channel_layout,
3535 .priv_class = &aac_decoder_class,
3536 .profiles = profiles,
3540 Note: This decoder filter is intended to decode LATM streams transferred
3541 in MPEG transport streams which only contain one program.
3542 To do a more complex LATM demuxing a separate LATM demuxer should be used.
3544 AVCodec ff_aac_latm_decoder = {
3546 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Coding LATM syntax)"),
3547 .type = AVMEDIA_TYPE_AUDIO,
3548 .id = AV_CODEC_ID_AAC_LATM,
3549 .priv_data_size = sizeof(struct LATMContext),
3550 .init = latm_decode_init,
3551 .close = aac_decode_close,
3552 .decode = latm_decode_frame,
3553 .sample_fmts = (const enum AVSampleFormat[]) {
3554 AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE
3556 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
3557 .channel_layouts = aac_channel_layout,
3559 .profiles = profiles,