3 * Copyright (c) 2005-2006 Oded Shimon ( ods15 ods15 dyndns org )
4 * Copyright (c) 2006-2007 Maxim Gavrilov ( maxim.gavrilov gmail com )
7 * Copyright (c) 2008-2010 Paul Kendall <paul@kcbbs.gen.nz>
8 * Copyright (c) 2010 Janne Grunau <janne-ffmpeg@jannau.net>
10 * This file is part of FFmpeg.
12 * FFmpeg is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public
14 * License as published by the Free Software Foundation; either
15 * version 2.1 of the License, or (at your option) any later version.
17 * FFmpeg is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with FFmpeg; if not, write to the Free Software
24 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
30 * @author Oded Shimon ( ods15 ods15 dyndns org )
31 * @author Maxim Gavrilov ( maxim.gavrilov gmail com )
38 * N (code in SoC repo) gain control
40 * Y window shapes - standard
41 * N window shapes - Low Delay
42 * Y filterbank - standard
43 * N (code in SoC repo) filterbank - Scalable Sample Rate
44 * Y Temporal Noise Shaping
45 * Y Long Term Prediction
48 * Y frequency domain prediction
49 * Y Perceptual Noise Substitution
51 * N Scalable Inverse AAC Quantization
52 * N Frequency Selective Switch
54 * Y quantization & coding - AAC
55 * N quantization & coding - TwinVQ
56 * N quantization & coding - BSAC
57 * N AAC Error Resilience tools
58 * N Error Resilience payload syntax
59 * N Error Protection tool
61 * N Silence Compression
64 * N Structured Audio tools
65 * N Structured Audio Sample Bank Format
67 * N Harmonic and Individual Lines plus Noise
68 * N Text-To-Speech Interface
69 * Y Spectral Band Replication
70 * Y (not in this code) Layer-1
71 * Y (not in this code) Layer-2
72 * Y (not in this code) Layer-3
73 * N SinuSoidal Coding (Transient, Sinusoid, Noise)
75 * N Direct Stream Transfer
77 * Note: - HE AAC v1 comprises LC AAC with Spectral Band Replication.
78 * - HE AAC v2 comprises LC AAC with Spectral Band Replication and
88 #include "fmtconvert.h"
95 #include "aacdectab.h"
96 #include "cbrt_tablegen.h"
99 #include "mpeg4audio.h"
100 #include "aacadtsdec.h"
101 #include "libavutil/intfloat.h"
109 # include "arm/aac.h"
112 static VLC vlc_scalefactors;
113 static VLC vlc_spectral[11];
115 static const char overread_err[] = "Input buffer exhausted before END element found\n";
117 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
119 // For PCE based channel configurations map the channels solely based on tags.
120 if (!ac->m4ac.chan_config) {
121 return ac->tag_che_map[type][elem_id];
123 // For indexed channel configurations map the channels solely based on position.
124 switch (ac->m4ac.chan_config) {
126 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
128 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
131 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
132 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
133 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
134 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
136 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
139 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
141 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
144 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
146 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
150 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
152 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
153 } else if (ac->m4ac.chan_config == 2) {
157 if (!ac->tags_mapped && type == TYPE_SCE) {
159 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
166 static int count_channels(uint8_t (*layout)[3], int tags)
169 for (i = 0; i < tags; i++) {
170 int syn_ele = layout[i][0];
171 int pos = layout[i][2];
172 sum += (1 + (syn_ele == TYPE_CPE)) *
173 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
179 * Check for the channel element in the current channel position configuration.
180 * If it exists, make sure the appropriate element is allocated and map the
181 * channel order to match the internal FFmpeg channel layout.
183 * @param che_pos current channel position configuration
184 * @param type channel element type
185 * @param id channel element id
186 * @param channels count of the number of channels in the configuration
188 * @return Returns error status. 0 - OK, !0 - error
190 static av_cold int che_configure(AACContext *ac,
191 enum ChannelPosition che_pos,
192 int type, int id, int *channels)
195 if (!ac->che[type][id]) {
196 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
197 return AVERROR(ENOMEM);
198 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
200 if (type != TYPE_CCE) {
201 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
202 if (type == TYPE_CPE ||
203 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
204 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
208 if (ac->che[type][id])
209 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
210 av_freep(&ac->che[type][id]);
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, int tags, 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, .syn_ele = TYPE_CPE,
229 .elem_id = layout_map[offset ][1], .aac_position = pos };
232 e2c_vec[offset] = (struct elem_to_channel) {
233 .av_position = left, .syn_ele = TYPE_SCE,
234 .elem_id = layout_map[offset ][1], .aac_position = pos };
235 e2c_vec[offset + 1] = (struct elem_to_channel) {
236 .av_position = right, .syn_ele = TYPE_SCE,
237 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
242 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
243 int num_pos_channels = 0;
247 for (i = *current; i < tags; i++) {
248 if (layout_map[i][2] != pos)
250 if (layout_map[i][0] == TYPE_CPE) {
252 if (pos == AAC_CHANNEL_FRONT || !first_cpe) {
258 num_pos_channels += 2;
266 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
269 return num_pos_channels;
272 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
274 int i, n, total_non_cc_elements;
275 struct elem_to_channel e2c_vec[MAX_ELEM_ID] = {{ 0 }};
276 int num_front_channels, num_side_channels, num_back_channels;
281 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
282 if (num_front_channels < 0)
285 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
286 if (num_side_channels < 0)
289 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
290 if (num_back_channels < 0)
294 if (num_front_channels & 1) {
295 e2c_vec[i] = (struct elem_to_channel) {
296 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
297 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
299 num_front_channels--;
301 if (num_front_channels >= 4) {
302 i += assign_pair(e2c_vec, layout_map, i, tags,
303 AV_CH_FRONT_LEFT_OF_CENTER,
304 AV_CH_FRONT_RIGHT_OF_CENTER,
306 num_front_channels -= 2;
308 if (num_front_channels >= 2) {
309 i += assign_pair(e2c_vec, layout_map, i, tags,
313 num_front_channels -= 2;
315 while (num_front_channels >= 2) {
316 i += assign_pair(e2c_vec, layout_map, i, tags,
320 num_front_channels -= 2;
323 if (num_side_channels >= 2) {
324 i += assign_pair(e2c_vec, layout_map, i, tags,
328 num_side_channels -= 2;
330 while (num_side_channels >= 2) {
331 i += assign_pair(e2c_vec, layout_map, i, tags,
335 num_side_channels -= 2;
338 while (num_back_channels >= 4) {
339 i += assign_pair(e2c_vec, layout_map, i, tags,
343 num_back_channels -= 2;
345 if (num_back_channels >= 2) {
346 i += assign_pair(e2c_vec, layout_map, i, tags,
350 num_back_channels -= 2;
352 if (num_back_channels) {
353 e2c_vec[i] = (struct elem_to_channel) {
354 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
355 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
360 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
361 e2c_vec[i] = (struct elem_to_channel) {
362 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
363 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
366 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
367 e2c_vec[i] = (struct elem_to_channel) {
368 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
369 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
373 // Must choose a stable sort
374 total_non_cc_elements = n = i;
377 for (i = 1; i < n; i++) {
378 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
379 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
387 for (i = 0; i < total_non_cc_elements; i++) {
388 layout_map[i][0] = e2c_vec[i].syn_ele;
389 layout_map[i][1] = e2c_vec[i].elem_id;
390 layout_map[i][2] = e2c_vec[i].aac_position;
391 if (e2c_vec[i].av_position != UINT64_MAX) {
392 layout |= e2c_vec[i].av_position;
400 * Configure output channel order based on the current program configuration element.
402 * @param che_pos current channel position configuration
404 * @return Returns error status. 0 - OK, !0 - error
406 static av_cold int output_configure(AACContext *ac,
407 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
408 int channel_config, enum OCStatus oc_type)
410 AVCodecContext *avctx = ac->avctx;
411 int i, channels = 0, ret;
414 if (ac->layout_map != layout_map) {
415 memcpy(ac->layout_map, layout_map, tags * sizeof(layout_map[0]));
416 ac->layout_map_tags = tags;
419 // Try to sniff a reasonable channel order, otherwise output the
420 // channels in the order the PCE declared them.
421 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
422 layout = sniff_channel_order(layout_map, tags);
423 for (i = 0; i < tags; i++) {
424 int type = layout_map[i][0];
425 int id = layout_map[i][1];
426 int position = layout_map[i][2];
427 // Allocate or free elements depending on if they are in the
428 // current program configuration.
429 ret = che_configure(ac, position, type, id, &channels);
434 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
435 if (layout) avctx->channel_layout = layout;
436 avctx->channels = channels;
437 ac->output_configured = oc_type;
442 static void flush(AVCodecContext *avctx)
444 AACContext *ac= avctx->priv_data;
447 for (type = 3; type >= 0; type--) {
448 for (i = 0; i < MAX_ELEM_ID; i++) {
449 ChannelElement *che = ac->che[type][i];
451 for (j = 0; j <= 1; j++) {
452 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
460 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
462 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
463 * @param sce_map mono (Single Channel Element) map
464 * @param type speaker type/position for these channels
466 static void decode_channel_map(uint8_t layout_map[][3],
467 enum ChannelPosition type,
468 GetBitContext *gb, int n)
471 enum RawDataBlockType syn_ele;
473 case AAC_CHANNEL_FRONT:
474 case AAC_CHANNEL_BACK:
475 case AAC_CHANNEL_SIDE:
476 syn_ele = get_bits1(gb);
482 case AAC_CHANNEL_LFE:
486 layout_map[0][0] = syn_ele;
487 layout_map[0][1] = get_bits(gb, 4);
488 layout_map[0][2] = type;
494 * Decode program configuration element; reference: table 4.2.
496 * @return Returns error status. 0 - OK, !0 - error
498 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
499 uint8_t (*layout_map)[3],
502 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
506 skip_bits(gb, 2); // object_type
508 sampling_index = get_bits(gb, 4);
509 if (m4ac->sampling_index != sampling_index)
510 av_log(avctx, AV_LOG_WARNING, "Sample rate index in program config element does not match the sample rate index configured by the container.\n");
512 num_front = get_bits(gb, 4);
513 num_side = get_bits(gb, 4);
514 num_back = get_bits(gb, 4);
515 num_lfe = get_bits(gb, 2);
516 num_assoc_data = get_bits(gb, 3);
517 num_cc = get_bits(gb, 4);
520 skip_bits(gb, 4); // mono_mixdown_tag
522 skip_bits(gb, 4); // stereo_mixdown_tag
525 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
527 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
528 av_log(avctx, AV_LOG_ERROR, overread_err);
531 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
533 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
535 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
537 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
540 skip_bits_long(gb, 4 * num_assoc_data);
542 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
547 /* comment field, first byte is length */
548 comment_len = get_bits(gb, 8) * 8;
549 if (get_bits_left(gb) < comment_len) {
550 av_log(avctx, AV_LOG_ERROR, overread_err);
553 skip_bits_long(gb, comment_len);
558 * Set up channel positions based on a default channel configuration
559 * as specified in table 1.17.
561 * @return Returns error status. 0 - OK, !0 - error
563 static av_cold int set_default_channel_config(AVCodecContext *avctx,
564 uint8_t (*layout_map)[3],
568 if (channel_config < 1 || channel_config > 7) {
569 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
573 *tags = tags_per_config[channel_config];
574 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
579 * Decode GA "General Audio" specific configuration; reference: table 4.1.
581 * @param ac pointer to AACContext, may be null
582 * @param avctx pointer to AVCCodecContext, used for logging
584 * @return Returns error status. 0 - OK, !0 - error
586 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
588 MPEG4AudioConfig *m4ac,
591 int extension_flag, ret;
592 uint8_t layout_map[MAX_ELEM_ID*4][3];
595 if (get_bits1(gb)) { // frameLengthFlag
596 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
600 if (get_bits1(gb)) // dependsOnCoreCoder
601 skip_bits(gb, 14); // coreCoderDelay
602 extension_flag = get_bits1(gb);
604 if (m4ac->object_type == AOT_AAC_SCALABLE ||
605 m4ac->object_type == AOT_ER_AAC_SCALABLE)
606 skip_bits(gb, 3); // layerNr
608 if (channel_config == 0) {
609 skip_bits(gb, 4); // element_instance_tag
610 tags = decode_pce(avctx, m4ac, layout_map, gb);
614 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
618 if (count_channels(layout_map, tags) > 1) {
620 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
623 if (ac && (ret = output_configure(ac, layout_map, tags,
624 channel_config, OC_GLOBAL_HDR)))
627 if (extension_flag) {
628 switch (m4ac->object_type) {
630 skip_bits(gb, 5); // numOfSubFrame
631 skip_bits(gb, 11); // layer_length
635 case AOT_ER_AAC_SCALABLE:
637 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
638 * aacScalefactorDataResilienceFlag
639 * aacSpectralDataResilienceFlag
643 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
649 * Decode audio specific configuration; reference: table 1.13.
651 * @param ac pointer to AACContext, may be null
652 * @param avctx pointer to AVCCodecContext, used for logging
653 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
654 * @param data pointer to buffer holding an audio specific config
655 * @param bit_size size of audio specific config or data in bits
656 * @param sync_extension look for an appended sync extension
658 * @return Returns error status or number of consumed bits. <0 - error
660 static int decode_audio_specific_config(AACContext *ac,
661 AVCodecContext *avctx,
662 MPEG4AudioConfig *m4ac,
663 const uint8_t *data, int bit_size,
669 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
670 for (i = 0; i < avctx->extradata_size; i++)
671 av_dlog(avctx, "%02x ", avctx->extradata[i]);
672 av_dlog(avctx, "\n");
674 init_get_bits(&gb, data, bit_size);
676 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
678 if (m4ac->sampling_index > 12) {
679 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
683 skip_bits_long(&gb, i);
685 switch (m4ac->object_type) {
689 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
693 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
694 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
698 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
699 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
700 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
702 return get_bits_count(&gb);
706 * linear congruential pseudorandom number generator
708 * @param previous_val pointer to the current state of the generator
710 * @return Returns a 32-bit pseudorandom integer
712 static av_always_inline int lcg_random(int previous_val)
714 return previous_val * 1664525 + 1013904223;
717 static av_always_inline void reset_predict_state(PredictorState *ps)
727 static void reset_all_predictors(PredictorState *ps)
730 for (i = 0; i < MAX_PREDICTORS; i++)
731 reset_predict_state(&ps[i]);
734 static int sample_rate_idx (int rate)
736 if (92017 <= rate) return 0;
737 else if (75132 <= rate) return 1;
738 else if (55426 <= rate) return 2;
739 else if (46009 <= rate) return 3;
740 else if (37566 <= rate) return 4;
741 else if (27713 <= rate) return 5;
742 else if (23004 <= rate) return 6;
743 else if (18783 <= rate) return 7;
744 else if (13856 <= rate) return 8;
745 else if (11502 <= rate) return 9;
746 else if (9391 <= rate) return 10;
750 static void reset_predictor_group(PredictorState *ps, int group_num)
753 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
754 reset_predict_state(&ps[i]);
757 #define AAC_INIT_VLC_STATIC(num, size) \
758 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
759 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
760 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
763 static av_cold int aac_decode_init(AVCodecContext *avctx)
765 AACContext *ac = avctx->priv_data;
766 float output_scale_factor;
769 ac->m4ac.sample_rate = avctx->sample_rate;
771 if (avctx->extradata_size > 0) {
772 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
774 avctx->extradata_size*8, 1) < 0)
778 uint8_t layout_map[MAX_ELEM_ID*4][3];
781 sr = sample_rate_idx(avctx->sample_rate);
782 ac->m4ac.sampling_index = sr;
783 ac->m4ac.channels = avctx->channels;
787 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
788 if (ff_mpeg4audio_channels[i] == avctx->channels)
790 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
793 ac->m4ac.chan_config = i;
795 if (ac->m4ac.chan_config) {
796 int ret = set_default_channel_config(avctx, layout_map,
797 &layout_map_tags, ac->m4ac.chan_config);
799 output_configure(ac, layout_map, layout_map_tags,
800 ac->m4ac.chan_config, OC_GLOBAL_HDR);
801 else if (avctx->err_recognition & AV_EF_EXPLODE)
802 return AVERROR_INVALIDDATA;
806 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
807 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
808 output_scale_factor = 1.0 / 32768.0;
810 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
811 output_scale_factor = 1.0;
814 AAC_INIT_VLC_STATIC( 0, 304);
815 AAC_INIT_VLC_STATIC( 1, 270);
816 AAC_INIT_VLC_STATIC( 2, 550);
817 AAC_INIT_VLC_STATIC( 3, 300);
818 AAC_INIT_VLC_STATIC( 4, 328);
819 AAC_INIT_VLC_STATIC( 5, 294);
820 AAC_INIT_VLC_STATIC( 6, 306);
821 AAC_INIT_VLC_STATIC( 7, 268);
822 AAC_INIT_VLC_STATIC( 8, 510);
823 AAC_INIT_VLC_STATIC( 9, 366);
824 AAC_INIT_VLC_STATIC(10, 462);
828 ff_dsputil_init(&ac->dsp, avctx);
829 ff_fmt_convert_init(&ac->fmt_conv, avctx);
831 ac->random_state = 0x1f2e3d4c;
835 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
836 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
837 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
840 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
841 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
842 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
843 // window initialization
844 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
845 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
846 ff_init_ff_sine_windows(10);
847 ff_init_ff_sine_windows( 7);
851 avcodec_get_frame_defaults(&ac->frame);
852 avctx->coded_frame = &ac->frame;
858 * Skip data_stream_element; reference: table 4.10.
860 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
862 int byte_align = get_bits1(gb);
863 int count = get_bits(gb, 8);
865 count += get_bits(gb, 8);
869 if (get_bits_left(gb) < 8 * count) {
870 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
873 skip_bits_long(gb, 8 * count);
877 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
882 ics->predictor_reset_group = get_bits(gb, 5);
883 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
884 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
888 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
889 ics->prediction_used[sfb] = get_bits1(gb);
895 * Decode Long Term Prediction data; reference: table 4.xx.
897 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
898 GetBitContext *gb, uint8_t max_sfb)
902 ltp->lag = get_bits(gb, 11);
903 ltp->coef = ltp_coef[get_bits(gb, 3)];
904 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
905 ltp->used[sfb] = get_bits1(gb);
909 * Decode Individual Channel Stream info; reference: table 4.6.
911 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
915 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
916 return AVERROR_INVALIDDATA;
918 ics->window_sequence[1] = ics->window_sequence[0];
919 ics->window_sequence[0] = get_bits(gb, 2);
920 ics->use_kb_window[1] = ics->use_kb_window[0];
921 ics->use_kb_window[0] = get_bits1(gb);
922 ics->num_window_groups = 1;
923 ics->group_len[0] = 1;
924 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
926 ics->max_sfb = get_bits(gb, 4);
927 for (i = 0; i < 7; i++) {
929 ics->group_len[ics->num_window_groups - 1]++;
931 ics->num_window_groups++;
932 ics->group_len[ics->num_window_groups - 1] = 1;
935 ics->num_windows = 8;
936 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
937 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
938 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
939 ics->predictor_present = 0;
941 ics->max_sfb = get_bits(gb, 6);
942 ics->num_windows = 1;
943 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
944 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
945 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
946 ics->predictor_present = get_bits1(gb);
947 ics->predictor_reset_group = 0;
948 if (ics->predictor_present) {
949 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
950 if (decode_prediction(ac, ics, gb)) {
951 return AVERROR_INVALIDDATA;
953 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
954 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
955 return AVERROR_INVALIDDATA;
957 if ((ics->ltp.present = get_bits(gb, 1)))
958 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
963 if (ics->max_sfb > ics->num_swb) {
964 av_log(ac->avctx, AV_LOG_ERROR,
965 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
966 ics->max_sfb, ics->num_swb);
967 return AVERROR_INVALIDDATA;
974 * Decode band types (section_data payload); reference: table 4.46.
976 * @param band_type array of the used band type
977 * @param band_type_run_end array of the last scalefactor band of a band type run
979 * @return Returns error status. 0 - OK, !0 - error
981 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
982 int band_type_run_end[120], GetBitContext *gb,
983 IndividualChannelStream *ics)
986 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
987 for (g = 0; g < ics->num_window_groups; g++) {
989 while (k < ics->max_sfb) {
990 uint8_t sect_end = k;
992 int sect_band_type = get_bits(gb, 4);
993 if (sect_band_type == 12) {
994 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
997 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1 && get_bits_left(gb) >= bits)
998 sect_end += sect_len_incr;
999 sect_end += sect_len_incr;
1000 if (get_bits_left(gb) < 0 || sect_len_incr == (1 << bits) - 1) {
1001 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1004 if (sect_end > ics->max_sfb) {
1005 av_log(ac->avctx, AV_LOG_ERROR,
1006 "Number of bands (%d) exceeds limit (%d).\n",
1007 sect_end, ics->max_sfb);
1010 for (; k < sect_end; k++) {
1011 band_type [idx] = sect_band_type;
1012 band_type_run_end[idx++] = sect_end;
1020 * Decode scalefactors; reference: table 4.47.
1022 * @param global_gain first scalefactor value as scalefactors are differentially coded
1023 * @param band_type array of the used band type
1024 * @param band_type_run_end array of the last scalefactor band of a band type run
1025 * @param sf array of scalefactors or intensity stereo positions
1027 * @return Returns error status. 0 - OK, !0 - error
1029 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1030 unsigned int global_gain,
1031 IndividualChannelStream *ics,
1032 enum BandType band_type[120],
1033 int band_type_run_end[120])
1036 int offset[3] = { global_gain, global_gain - 90, 0 };
1039 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
1040 for (g = 0; g < ics->num_window_groups; g++) {
1041 for (i = 0; i < ics->max_sfb;) {
1042 int run_end = band_type_run_end[idx];
1043 if (band_type[idx] == ZERO_BT) {
1044 for (; i < run_end; i++, idx++)
1046 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1047 for (; i < run_end; i++, idx++) {
1048 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1049 clipped_offset = av_clip(offset[2], -155, 100);
1050 if (offset[2] != clipped_offset) {
1051 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1052 "position clipped (%d -> %d).\nIf you heard an "
1053 "audible artifact, there may be a bug in the "
1054 "decoder. ", offset[2], clipped_offset);
1056 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1058 } else if (band_type[idx] == NOISE_BT) {
1059 for (; i < run_end; i++, idx++) {
1060 if (noise_flag-- > 0)
1061 offset[1] += get_bits(gb, 9) - 256;
1063 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1064 clipped_offset = av_clip(offset[1], -100, 155);
1065 if (offset[1] != clipped_offset) {
1066 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1067 "(%d -> %d).\nIf you heard an audible "
1068 "artifact, there may be a bug in the decoder. ",
1069 offset[1], clipped_offset);
1071 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1074 for (; i < run_end; i++, idx++) {
1075 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1076 if (offset[0] > 255U) {
1077 av_log(ac->avctx, AV_LOG_ERROR,
1078 "%s (%d) out of range.\n", sf_str[0], offset[0]);
1081 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1090 * Decode pulse data; reference: table 4.7.
1092 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1093 const uint16_t *swb_offset, int num_swb)
1096 pulse->num_pulse = get_bits(gb, 2) + 1;
1097 pulse_swb = get_bits(gb, 6);
1098 if (pulse_swb >= num_swb)
1100 pulse->pos[0] = swb_offset[pulse_swb];
1101 pulse->pos[0] += get_bits(gb, 5);
1102 if (pulse->pos[0] > 1023)
1104 pulse->amp[0] = get_bits(gb, 4);
1105 for (i = 1; i < pulse->num_pulse; i++) {
1106 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1107 if (pulse->pos[i] > 1023)
1109 pulse->amp[i] = get_bits(gb, 4);
1115 * Decode Temporal Noise Shaping data; reference: table 4.48.
1117 * @return Returns error status. 0 - OK, !0 - error
1119 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1120 GetBitContext *gb, const IndividualChannelStream *ics)
1122 int w, filt, i, coef_len, coef_res, coef_compress;
1123 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1124 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1125 for (w = 0; w < ics->num_windows; w++) {
1126 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1127 coef_res = get_bits1(gb);
1129 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1131 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1133 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1134 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1135 tns->order[w][filt], tns_max_order);
1136 tns->order[w][filt] = 0;
1139 if (tns->order[w][filt]) {
1140 tns->direction[w][filt] = get_bits1(gb);
1141 coef_compress = get_bits1(gb);
1142 coef_len = coef_res + 3 - coef_compress;
1143 tmp2_idx = 2 * coef_compress + coef_res;
1145 for (i = 0; i < tns->order[w][filt]; i++)
1146 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1155 * Decode Mid/Side data; reference: table 4.54.
1157 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1158 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1159 * [3] reserved for scalable AAC
1161 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1165 if (ms_present == 1) {
1166 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1167 cpe->ms_mask[idx] = get_bits1(gb);
1168 } else if (ms_present == 2) {
1169 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1174 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1178 *dst++ = v[idx & 15] * s;
1179 *dst++ = v[idx>>4 & 15] * s;
1185 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1189 *dst++ = v[idx & 3] * s;
1190 *dst++ = v[idx>>2 & 3] * s;
1191 *dst++ = v[idx>>4 & 3] * s;
1192 *dst++ = v[idx>>6 & 3] * s;
1198 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1199 unsigned sign, const float *scale)
1201 union av_intfloat32 s0, s1;
1203 s0.f = s1.f = *scale;
1204 s0.i ^= sign >> 1 << 31;
1207 *dst++ = v[idx & 15] * s0.f;
1208 *dst++ = v[idx>>4 & 15] * s1.f;
1215 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1216 unsigned sign, const float *scale)
1218 unsigned nz = idx >> 12;
1219 union av_intfloat32 s = { .f = *scale };
1220 union av_intfloat32 t;
1222 t.i = s.i ^ (sign & 1U<<31);
1223 *dst++ = v[idx & 3] * t.f;
1225 sign <<= nz & 1; nz >>= 1;
1226 t.i = s.i ^ (sign & 1U<<31);
1227 *dst++ = v[idx>>2 & 3] * t.f;
1229 sign <<= nz & 1; nz >>= 1;
1230 t.i = s.i ^ (sign & 1U<<31);
1231 *dst++ = v[idx>>4 & 3] * t.f;
1233 sign <<= nz & 1; nz >>= 1;
1234 t.i = s.i ^ (sign & 1U<<31);
1235 *dst++ = v[idx>>6 & 3] * t.f;
1242 * Decode spectral data; reference: table 4.50.
1243 * Dequantize and scale spectral data; reference: 4.6.3.3.
1245 * @param coef array of dequantized, scaled spectral data
1246 * @param sf array of scalefactors or intensity stereo positions
1247 * @param pulse_present set if pulses are present
1248 * @param pulse pointer to pulse data struct
1249 * @param band_type array of the used band type
1251 * @return Returns error status. 0 - OK, !0 - error
1253 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1254 GetBitContext *gb, const float sf[120],
1255 int pulse_present, const Pulse *pulse,
1256 const IndividualChannelStream *ics,
1257 enum BandType band_type[120])
1259 int i, k, g, idx = 0;
1260 const int c = 1024 / ics->num_windows;
1261 const uint16_t *offsets = ics->swb_offset;
1262 float *coef_base = coef;
1264 for (g = 0; g < ics->num_windows; g++)
1265 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1267 for (g = 0; g < ics->num_window_groups; g++) {
1268 unsigned g_len = ics->group_len[g];
1270 for (i = 0; i < ics->max_sfb; i++, idx++) {
1271 const unsigned cbt_m1 = band_type[idx] - 1;
1272 float *cfo = coef + offsets[i];
1273 int off_len = offsets[i + 1] - offsets[i];
1276 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1277 for (group = 0; group < g_len; group++, cfo+=128) {
1278 memset(cfo, 0, off_len * sizeof(float));
1280 } else if (cbt_m1 == NOISE_BT - 1) {
1281 for (group = 0; group < g_len; group++, cfo+=128) {
1285 for (k = 0; k < off_len; k++) {
1286 ac->random_state = lcg_random(ac->random_state);
1287 cfo[k] = ac->random_state;
1290 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1291 scale = sf[idx] / sqrtf(band_energy);
1292 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1295 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1296 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1297 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1298 OPEN_READER(re, gb);
1300 switch (cbt_m1 >> 1) {
1302 for (group = 0; group < g_len; group++, cfo+=128) {
1310 UPDATE_CACHE(re, gb);
1311 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1312 cb_idx = cb_vector_idx[code];
1313 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1319 for (group = 0; group < g_len; group++, cfo+=128) {
1329 UPDATE_CACHE(re, gb);
1330 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1331 cb_idx = cb_vector_idx[code];
1332 nnz = cb_idx >> 8 & 15;
1333 bits = nnz ? GET_CACHE(re, gb) : 0;
1334 LAST_SKIP_BITS(re, gb, nnz);
1335 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1341 for (group = 0; group < g_len; group++, cfo+=128) {
1349 UPDATE_CACHE(re, gb);
1350 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1351 cb_idx = cb_vector_idx[code];
1352 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1359 for (group = 0; group < g_len; group++, cfo+=128) {
1369 UPDATE_CACHE(re, gb);
1370 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1371 cb_idx = cb_vector_idx[code];
1372 nnz = cb_idx >> 8 & 15;
1373 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1374 LAST_SKIP_BITS(re, gb, nnz);
1375 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1381 for (group = 0; group < g_len; group++, cfo+=128) {
1383 uint32_t *icf = (uint32_t *) cf;
1393 UPDATE_CACHE(re, gb);
1394 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1402 cb_idx = cb_vector_idx[code];
1405 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1406 LAST_SKIP_BITS(re, gb, nnz);
1408 for (j = 0; j < 2; j++) {
1412 /* The total length of escape_sequence must be < 22 bits according
1413 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1414 UPDATE_CACHE(re, gb);
1415 b = GET_CACHE(re, gb);
1416 b = 31 - av_log2(~b);
1419 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1423 SKIP_BITS(re, gb, b + 1);
1425 n = (1 << b) + SHOW_UBITS(re, gb, b);
1426 LAST_SKIP_BITS(re, gb, b);
1427 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1430 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1431 *icf++ = (bits & 1U<<31) | v;
1438 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1442 CLOSE_READER(re, gb);
1448 if (pulse_present) {
1450 for (i = 0; i < pulse->num_pulse; i++) {
1451 float co = coef_base[ pulse->pos[i] ];
1452 while (offsets[idx + 1] <= pulse->pos[i])
1454 if (band_type[idx] != NOISE_BT && sf[idx]) {
1455 float ico = -pulse->amp[i];
1458 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1460 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1467 static av_always_inline float flt16_round(float pf)
1469 union av_intfloat32 tmp;
1471 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1475 static av_always_inline float flt16_even(float pf)
1477 union av_intfloat32 tmp;
1479 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1483 static av_always_inline float flt16_trunc(float pf)
1485 union av_intfloat32 pun;
1487 pun.i &= 0xFFFF0000U;
1491 static av_always_inline void predict(PredictorState *ps, float *coef,
1494 const float a = 0.953125; // 61.0 / 64
1495 const float alpha = 0.90625; // 29.0 / 32
1499 float r0 = ps->r0, r1 = ps->r1;
1500 float cor0 = ps->cor0, cor1 = ps->cor1;
1501 float var0 = ps->var0, var1 = ps->var1;
1503 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1504 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1506 pv = flt16_round(k1 * r0 + k2 * r1);
1513 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1514 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1515 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1516 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1518 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1519 ps->r0 = flt16_trunc(a * e0);
1523 * Apply AAC-Main style frequency domain prediction.
1525 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1529 if (!sce->ics.predictor_initialized) {
1530 reset_all_predictors(sce->predictor_state);
1531 sce->ics.predictor_initialized = 1;
1534 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1535 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1536 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1537 predict(&sce->predictor_state[k], &sce->coeffs[k],
1538 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1541 if (sce->ics.predictor_reset_group)
1542 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1544 reset_all_predictors(sce->predictor_state);
1548 * Decode an individual_channel_stream payload; reference: table 4.44.
1550 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1551 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1553 * @return Returns error status. 0 - OK, !0 - error
1555 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1556 GetBitContext *gb, int common_window, int scale_flag)
1559 TemporalNoiseShaping *tns = &sce->tns;
1560 IndividualChannelStream *ics = &sce->ics;
1561 float *out = sce->coeffs;
1562 int global_gain, pulse_present = 0;
1564 /* This assignment is to silence a GCC warning about the variable being used
1565 * uninitialized when in fact it always is.
1567 pulse.num_pulse = 0;
1569 global_gain = get_bits(gb, 8);
1571 if (!common_window && !scale_flag) {
1572 if (decode_ics_info(ac, ics, gb) < 0)
1573 return AVERROR_INVALIDDATA;
1576 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1578 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1583 if ((pulse_present = get_bits1(gb))) {
1584 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1585 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1588 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1589 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1593 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1595 if (get_bits1(gb)) {
1596 av_log_missing_feature(ac->avctx, "SSR", 1);
1601 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1604 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1605 apply_prediction(ac, sce);
1611 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1613 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1615 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1616 float *ch0 = cpe->ch[0].coeffs;
1617 float *ch1 = cpe->ch[1].coeffs;
1618 int g, i, group, idx = 0;
1619 const uint16_t *offsets = ics->swb_offset;
1620 for (g = 0; g < ics->num_window_groups; g++) {
1621 for (i = 0; i < ics->max_sfb; i++, idx++) {
1622 if (cpe->ms_mask[idx] &&
1623 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1624 for (group = 0; group < ics->group_len[g]; group++) {
1625 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1626 ch1 + group * 128 + offsets[i],
1627 offsets[i+1] - offsets[i]);
1631 ch0 += ics->group_len[g] * 128;
1632 ch1 += ics->group_len[g] * 128;
1637 * intensity stereo decoding; reference: 4.6.8.2.3
1639 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1640 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1641 * [3] reserved for scalable AAC
1643 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1645 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1646 SingleChannelElement *sce1 = &cpe->ch[1];
1647 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1648 const uint16_t *offsets = ics->swb_offset;
1649 int g, group, i, idx = 0;
1652 for (g = 0; g < ics->num_window_groups; g++) {
1653 for (i = 0; i < ics->max_sfb;) {
1654 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1655 const int bt_run_end = sce1->band_type_run_end[idx];
1656 for (; i < bt_run_end; i++, idx++) {
1657 c = -1 + 2 * (sce1->band_type[idx] - 14);
1659 c *= 1 - 2 * cpe->ms_mask[idx];
1660 scale = c * sce1->sf[idx];
1661 for (group = 0; group < ics->group_len[g]; group++)
1662 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1663 coef0 + group * 128 + offsets[i],
1665 offsets[i + 1] - offsets[i]);
1668 int bt_run_end = sce1->band_type_run_end[idx];
1669 idx += bt_run_end - i;
1673 coef0 += ics->group_len[g] * 128;
1674 coef1 += ics->group_len[g] * 128;
1679 * Decode a channel_pair_element; reference: table 4.4.
1681 * @return Returns error status. 0 - OK, !0 - error
1683 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1685 int i, ret, common_window, ms_present = 0;
1687 common_window = get_bits1(gb);
1688 if (common_window) {
1689 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1690 return AVERROR_INVALIDDATA;
1691 i = cpe->ch[1].ics.use_kb_window[0];
1692 cpe->ch[1].ics = cpe->ch[0].ics;
1693 cpe->ch[1].ics.use_kb_window[1] = i;
1694 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1695 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1696 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1697 ms_present = get_bits(gb, 2);
1698 if (ms_present == 3) {
1699 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1701 } else if (ms_present)
1702 decode_mid_side_stereo(cpe, gb, ms_present);
1704 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1706 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1709 if (common_window) {
1711 apply_mid_side_stereo(ac, cpe);
1712 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1713 apply_prediction(ac, &cpe->ch[0]);
1714 apply_prediction(ac, &cpe->ch[1]);
1718 apply_intensity_stereo(ac, cpe, ms_present);
1722 static const float cce_scale[] = {
1723 1.09050773266525765921, //2^(1/8)
1724 1.18920711500272106672, //2^(1/4)
1730 * Decode coupling_channel_element; reference: table 4.8.
1732 * @return Returns error status. 0 - OK, !0 - error
1734 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1740 SingleChannelElement *sce = &che->ch[0];
1741 ChannelCoupling *coup = &che->coup;
1743 coup->coupling_point = 2 * get_bits1(gb);
1744 coup->num_coupled = get_bits(gb, 3);
1745 for (c = 0; c <= coup->num_coupled; c++) {
1747 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1748 coup->id_select[c] = get_bits(gb, 4);
1749 if (coup->type[c] == TYPE_CPE) {
1750 coup->ch_select[c] = get_bits(gb, 2);
1751 if (coup->ch_select[c] == 3)
1754 coup->ch_select[c] = 2;
1756 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1758 sign = get_bits(gb, 1);
1759 scale = cce_scale[get_bits(gb, 2)];
1761 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1764 for (c = 0; c < num_gain; c++) {
1768 float gain_cache = 1.;
1770 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1771 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1772 gain_cache = powf(scale, -gain);
1774 if (coup->coupling_point == AFTER_IMDCT) {
1775 coup->gain[c][0] = gain_cache;
1777 for (g = 0; g < sce->ics.num_window_groups; g++) {
1778 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1779 if (sce->band_type[idx] != ZERO_BT) {
1781 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1789 gain_cache = powf(scale, -t) * s;
1792 coup->gain[c][idx] = gain_cache;
1802 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1804 * @return Returns number of bytes consumed.
1806 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1810 int num_excl_chan = 0;
1813 for (i = 0; i < 7; i++)
1814 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1815 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1817 return num_excl_chan / 7;
1821 * Decode dynamic range information; reference: table 4.52.
1823 * @param cnt length of TYPE_FIL syntactic element in bytes
1825 * @return Returns number of bytes consumed.
1827 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1828 GetBitContext *gb, int cnt)
1831 int drc_num_bands = 1;
1834 /* pce_tag_present? */
1835 if (get_bits1(gb)) {
1836 che_drc->pce_instance_tag = get_bits(gb, 4);
1837 skip_bits(gb, 4); // tag_reserved_bits
1841 /* excluded_chns_present? */
1842 if (get_bits1(gb)) {
1843 n += decode_drc_channel_exclusions(che_drc, gb);
1846 /* drc_bands_present? */
1847 if (get_bits1(gb)) {
1848 che_drc->band_incr = get_bits(gb, 4);
1849 che_drc->interpolation_scheme = get_bits(gb, 4);
1851 drc_num_bands += che_drc->band_incr;
1852 for (i = 0; i < drc_num_bands; i++) {
1853 che_drc->band_top[i] = get_bits(gb, 8);
1858 /* prog_ref_level_present? */
1859 if (get_bits1(gb)) {
1860 che_drc->prog_ref_level = get_bits(gb, 7);
1861 skip_bits1(gb); // prog_ref_level_reserved_bits
1865 for (i = 0; i < drc_num_bands; i++) {
1866 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1867 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1875 * Decode extension data (incomplete); reference: table 4.51.
1877 * @param cnt length of TYPE_FIL syntactic element in bytes
1879 * @return Returns number of bytes consumed
1881 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1882 ChannelElement *che, enum RawDataBlockType elem_type)
1886 switch (get_bits(gb, 4)) { // extension type
1887 case EXT_SBR_DATA_CRC:
1891 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1893 } else if (!ac->m4ac.sbr) {
1894 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1895 skip_bits_long(gb, 8 * cnt - 4);
1897 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1898 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1899 skip_bits_long(gb, 8 * cnt - 4);
1901 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1904 output_configure(ac, ac->layout_map, ac->layout_map_tags,
1905 ac->m4ac.chan_config, ac->output_configured);
1909 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1911 case EXT_DYNAMIC_RANGE:
1912 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1916 case EXT_DATA_ELEMENT:
1918 skip_bits_long(gb, 8 * cnt - 4);
1925 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1927 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1928 * @param coef spectral coefficients
1930 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1931 IndividualChannelStream *ics, int decode)
1933 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1935 int bottom, top, order, start, end, size, inc;
1936 float lpc[TNS_MAX_ORDER];
1937 float tmp[TNS_MAX_ORDER];
1939 for (w = 0; w < ics->num_windows; w++) {
1940 bottom = ics->num_swb;
1941 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1943 bottom = FFMAX(0, top - tns->length[w][filt]);
1944 order = tns->order[w][filt];
1949 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1951 start = ics->swb_offset[FFMIN(bottom, mmm)];
1952 end = ics->swb_offset[FFMIN( top, mmm)];
1953 if ((size = end - start) <= 0)
1955 if (tns->direction[w][filt]) {
1965 for (m = 0; m < size; m++, start += inc)
1966 for (i = 1; i <= FFMIN(m, order); i++)
1967 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1970 for (m = 0; m < size; m++, start += inc) {
1971 tmp[0] = coef[start];
1972 for (i = 1; i <= FFMIN(m, order); i++)
1973 coef[start] += tmp[i] * lpc[i - 1];
1974 for (i = order; i > 0; i--)
1975 tmp[i] = tmp[i - 1];
1983 * Apply windowing and MDCT to obtain the spectral
1984 * coefficient from the predicted sample by LTP.
1986 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1987 float *in, IndividualChannelStream *ics)
1989 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1990 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1991 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1992 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1994 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1995 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1997 memset(in, 0, 448 * sizeof(float));
1998 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2000 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2001 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2003 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2004 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2006 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2010 * Apply the long term prediction
2012 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2014 const LongTermPrediction *ltp = &sce->ics.ltp;
2015 const uint16_t *offsets = sce->ics.swb_offset;
2018 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2019 float *predTime = sce->ret;
2020 float *predFreq = ac->buf_mdct;
2021 int16_t num_samples = 2048;
2023 if (ltp->lag < 1024)
2024 num_samples = ltp->lag + 1024;
2025 for (i = 0; i < num_samples; i++)
2026 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2027 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2029 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2031 if (sce->tns.present)
2032 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2034 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2036 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2037 sce->coeffs[i] += predFreq[i];
2042 * Update the LTP buffer for next frame
2044 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2046 IndividualChannelStream *ics = &sce->ics;
2047 float *saved = sce->saved;
2048 float *saved_ltp = sce->coeffs;
2049 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2050 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2053 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2054 memcpy(saved_ltp, saved, 512 * sizeof(float));
2055 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2056 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2057 for (i = 0; i < 64; i++)
2058 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2059 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2060 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2061 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2062 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2063 for (i = 0; i < 64; i++)
2064 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2065 } else { // LONG_STOP or ONLY_LONG
2066 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2067 for (i = 0; i < 512; i++)
2068 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2071 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2072 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2073 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2077 * Conduct IMDCT and windowing.
2079 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2081 IndividualChannelStream *ics = &sce->ics;
2082 float *in = sce->coeffs;
2083 float *out = sce->ret;
2084 float *saved = sce->saved;
2085 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2086 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2087 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2088 float *buf = ac->buf_mdct;
2089 float *temp = ac->temp;
2093 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2094 for (i = 0; i < 1024; i += 128)
2095 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2097 ac->mdct.imdct_half(&ac->mdct, buf, in);
2099 /* window overlapping
2100 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2101 * and long to short transitions are considered to be short to short
2102 * transitions. This leaves just two cases (long to long and short to short)
2103 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2105 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2106 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2107 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2109 memcpy( out, saved, 448 * sizeof(float));
2111 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2112 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2113 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2114 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2115 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2116 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2117 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2119 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2120 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2125 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2126 memcpy( saved, temp + 64, 64 * sizeof(float));
2127 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2128 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2129 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2130 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2131 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2132 memcpy( saved, buf + 512, 448 * sizeof(float));
2133 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2134 } else { // LONG_STOP or ONLY_LONG
2135 memcpy( saved, buf + 512, 512 * sizeof(float));
2140 * Apply dependent channel coupling (applied before IMDCT).
2142 * @param index index into coupling gain array
2144 static void apply_dependent_coupling(AACContext *ac,
2145 SingleChannelElement *target,
2146 ChannelElement *cce, int index)
2148 IndividualChannelStream *ics = &cce->ch[0].ics;
2149 const uint16_t *offsets = ics->swb_offset;
2150 float *dest = target->coeffs;
2151 const float *src = cce->ch[0].coeffs;
2152 int g, i, group, k, idx = 0;
2153 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2154 av_log(ac->avctx, AV_LOG_ERROR,
2155 "Dependent coupling is not supported together with LTP\n");
2158 for (g = 0; g < ics->num_window_groups; g++) {
2159 for (i = 0; i < ics->max_sfb; i++, idx++) {
2160 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2161 const float gain = cce->coup.gain[index][idx];
2162 for (group = 0; group < ics->group_len[g]; group++) {
2163 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2165 dest[group * 128 + k] += gain * src[group * 128 + k];
2170 dest += ics->group_len[g] * 128;
2171 src += ics->group_len[g] * 128;
2176 * Apply independent channel coupling (applied after IMDCT).
2178 * @param index index into coupling gain array
2180 static void apply_independent_coupling(AACContext *ac,
2181 SingleChannelElement *target,
2182 ChannelElement *cce, int index)
2185 const float gain = cce->coup.gain[index][0];
2186 const float *src = cce->ch[0].ret;
2187 float *dest = target->ret;
2188 const int len = 1024 << (ac->m4ac.sbr == 1);
2190 for (i = 0; i < len; i++)
2191 dest[i] += gain * src[i];
2195 * channel coupling transformation interface
2197 * @param apply_coupling_method pointer to (in)dependent coupling function
2199 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2200 enum RawDataBlockType type, int elem_id,
2201 enum CouplingPoint coupling_point,
2202 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2206 for (i = 0; i < MAX_ELEM_ID; i++) {
2207 ChannelElement *cce = ac->che[TYPE_CCE][i];
2210 if (cce && cce->coup.coupling_point == coupling_point) {
2211 ChannelCoupling *coup = &cce->coup;
2213 for (c = 0; c <= coup->num_coupled; c++) {
2214 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2215 if (coup->ch_select[c] != 1) {
2216 apply_coupling_method(ac, &cc->ch[0], cce, index);
2217 if (coup->ch_select[c] != 0)
2220 if (coup->ch_select[c] != 2)
2221 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2223 index += 1 + (coup->ch_select[c] == 3);
2230 * Convert spectral data to float samples, applying all supported tools as appropriate.
2232 static void spectral_to_sample(AACContext *ac)
2235 for (type = 3; type >= 0; type--) {
2236 for (i = 0; i < MAX_ELEM_ID; i++) {
2237 ChannelElement *che = ac->che[type][i];
2239 if (type <= TYPE_CPE)
2240 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2241 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2242 if (che->ch[0].ics.predictor_present) {
2243 if (che->ch[0].ics.ltp.present)
2244 apply_ltp(ac, &che->ch[0]);
2245 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2246 apply_ltp(ac, &che->ch[1]);
2249 if (che->ch[0].tns.present)
2250 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2251 if (che->ch[1].tns.present)
2252 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2253 if (type <= TYPE_CPE)
2254 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2255 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2256 imdct_and_windowing(ac, &che->ch[0]);
2257 if (ac->m4ac.object_type == AOT_AAC_LTP)
2258 update_ltp(ac, &che->ch[0]);
2259 if (type == TYPE_CPE) {
2260 imdct_and_windowing(ac, &che->ch[1]);
2261 if (ac->m4ac.object_type == AOT_AAC_LTP)
2262 update_ltp(ac, &che->ch[1]);
2264 if (ac->m4ac.sbr > 0) {
2265 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2268 if (type <= TYPE_CCE)
2269 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2275 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2278 AACADTSHeaderInfo hdr_info;
2279 uint8_t layout_map[MAX_ELEM_ID*4][3];
2280 int layout_map_tags;
2282 size = avpriv_aac_parse_header(gb, &hdr_info);
2284 if (hdr_info.chan_config) {
2285 ac->m4ac.chan_config = hdr_info.chan_config;
2286 if (set_default_channel_config(ac->avctx, layout_map,
2287 &layout_map_tags, hdr_info.chan_config))
2289 if (output_configure(ac, layout_map, layout_map_tags,
2290 hdr_info.chan_config,
2291 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2293 } else if (ac->output_configured != OC_LOCKED) {
2294 ac->m4ac.chan_config = 0;
2295 ac->output_configured = OC_NONE;
2297 if (ac->output_configured != OC_LOCKED) {
2300 ac->m4ac.sample_rate = hdr_info.sample_rate;
2301 ac->m4ac.sampling_index = hdr_info.sampling_index;
2302 ac->m4ac.object_type = hdr_info.object_type;
2304 if (!ac->avctx->sample_rate)
2305 ac->avctx->sample_rate = hdr_info.sample_rate;
2306 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2307 // This is 2 for "VLB " audio in NSV files.
2308 // See samples/nsv/vlb_audio.
2309 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2310 ac->warned_num_aac_frames = 1;
2312 if (!hdr_info.crc_absent)
2318 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2319 int *got_frame_ptr, GetBitContext *gb)
2321 AACContext *ac = avctx->priv_data;
2322 ChannelElement *che = NULL, *che_prev = NULL;
2323 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2325 int samples = 0, multiplier, audio_found = 0;
2327 if (show_bits(gb, 12) == 0xfff) {
2328 if (parse_adts_frame_header(ac, gb) < 0) {
2329 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2332 if (ac->m4ac.sampling_index > 12) {
2333 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2338 ac->tags_mapped = 0;
2340 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2341 elem_id = get_bits(gb, 4);
2343 if (elem_type < TYPE_DSE) {
2344 if (!(che=get_che(ac, elem_type, elem_id))) {
2345 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2346 elem_type, elem_id);
2352 switch (elem_type) {
2355 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2360 err = decode_cpe(ac, gb, che);
2365 err = decode_cce(ac, gb, che);
2369 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2374 err = skip_data_stream_element(ac, gb);
2378 uint8_t layout_map[MAX_ELEM_ID*4][3];
2380 tags = decode_pce(avctx, &ac->m4ac, layout_map, gb);
2385 if (ac->output_configured > OC_TRIAL_PCE)
2386 av_log(avctx, AV_LOG_INFO,
2387 "Evaluating a further program_config_element.\n");
2388 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2390 ac->m4ac.chan_config = 0;
2396 elem_id += get_bits(gb, 8) - 1;
2397 if (get_bits_left(gb) < 8 * elem_id) {
2398 av_log(avctx, AV_LOG_ERROR, overread_err);
2402 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2403 err = 0; /* FIXME */
2407 err = -1; /* should not happen, but keeps compiler happy */
2412 elem_type_prev = elem_type;
2417 if (get_bits_left(gb) < 3) {
2418 av_log(avctx, AV_LOG_ERROR, overread_err);
2423 spectral_to_sample(ac);
2425 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2426 samples <<= multiplier;
2427 if (ac->output_configured < OC_LOCKED) {
2428 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2429 avctx->frame_size = samples;
2433 /* get output buffer */
2434 ac->frame.nb_samples = samples;
2435 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2436 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2440 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2441 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2442 (const float **)ac->output_data,
2443 samples, avctx->channels);
2445 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2446 (const float **)ac->output_data,
2447 samples, avctx->channels);
2449 *(AVFrame *)data = ac->frame;
2451 *got_frame_ptr = !!samples;
2453 if (ac->output_configured && audio_found)
2454 ac->output_configured = OC_LOCKED;
2459 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2460 int *got_frame_ptr, AVPacket *avpkt)
2462 AACContext *ac = avctx->priv_data;
2463 const uint8_t *buf = avpkt->data;
2464 int buf_size = avpkt->size;
2469 int new_extradata_size;
2470 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2471 AV_PKT_DATA_NEW_EXTRADATA,
2472 &new_extradata_size);
2474 if (new_extradata) {
2475 av_free(avctx->extradata);
2476 avctx->extradata = av_mallocz(new_extradata_size +
2477 FF_INPUT_BUFFER_PADDING_SIZE);
2478 if (!avctx->extradata)
2479 return AVERROR(ENOMEM);
2480 avctx->extradata_size = new_extradata_size;
2481 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2482 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2484 avctx->extradata_size*8, 1) < 0)
2485 return AVERROR_INVALIDDATA;
2488 init_get_bits(&gb, buf, buf_size * 8);
2490 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2493 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2494 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2495 if (buf[buf_offset])
2498 return buf_size > buf_offset ? buf_consumed : buf_size;
2501 static av_cold int aac_decode_close(AVCodecContext *avctx)
2503 AACContext *ac = avctx->priv_data;
2506 for (i = 0; i < MAX_ELEM_ID; i++) {
2507 for (type = 0; type < 4; type++) {
2508 if (ac->che[type][i])
2509 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2510 av_freep(&ac->che[type][i]);
2514 ff_mdct_end(&ac->mdct);
2515 ff_mdct_end(&ac->mdct_small);
2516 ff_mdct_end(&ac->mdct_ltp);
2521 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2523 struct LATMContext {
2524 AACContext aac_ctx; ///< containing AACContext
2525 int initialized; ///< initilized after a valid extradata was seen
2528 int audio_mux_version_A; ///< LATM syntax version
2529 int frame_length_type; ///< 0/1 variable/fixed frame length
2530 int frame_length; ///< frame length for fixed frame length
2533 static inline uint32_t latm_get_value(GetBitContext *b)
2535 int length = get_bits(b, 2);
2537 return get_bits_long(b, (length+1)*8);
2540 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2541 GetBitContext *gb, int asclen)
2543 AACContext *ac = &latmctx->aac_ctx;
2544 AVCodecContext *avctx = ac->avctx;
2545 MPEG4AudioConfig m4ac = {0};
2546 int config_start_bit = get_bits_count(gb);
2547 int sync_extension = 0;
2548 int bits_consumed, esize;
2552 asclen = FFMIN(asclen, get_bits_left(gb));
2554 asclen = get_bits_left(gb);
2556 if (config_start_bit % 8) {
2557 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2558 "config not byte aligned.\n", 1);
2559 return AVERROR_INVALIDDATA;
2562 return AVERROR_INVALIDDATA;
2563 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2564 gb->buffer + (config_start_bit / 8),
2565 asclen, sync_extension);
2567 if (bits_consumed < 0)
2568 return AVERROR_INVALIDDATA;
2570 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2571 ac->m4ac.chan_config != m4ac.chan_config) {
2573 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2574 latmctx->initialized = 0;
2576 esize = (bits_consumed+7) / 8;
2578 if (avctx->extradata_size < esize) {
2579 av_free(avctx->extradata);
2580 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2581 if (!avctx->extradata)
2582 return AVERROR(ENOMEM);
2585 avctx->extradata_size = esize;
2586 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2587 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2589 skip_bits_long(gb, bits_consumed);
2591 return bits_consumed;
2594 static int read_stream_mux_config(struct LATMContext *latmctx,
2597 int ret, audio_mux_version = get_bits(gb, 1);
2599 latmctx->audio_mux_version_A = 0;
2600 if (audio_mux_version)
2601 latmctx->audio_mux_version_A = get_bits(gb, 1);
2603 if (!latmctx->audio_mux_version_A) {
2605 if (audio_mux_version)
2606 latm_get_value(gb); // taraFullness
2608 skip_bits(gb, 1); // allStreamSameTimeFraming
2609 skip_bits(gb, 6); // numSubFrames
2611 if (get_bits(gb, 4)) { // numPrograms
2612 av_log_missing_feature(latmctx->aac_ctx.avctx,
2613 "multiple programs are not supported\n", 1);
2614 return AVERROR_PATCHWELCOME;
2617 // for each program (which there is only on in DVB)
2619 // for each layer (which there is only on in DVB)
2620 if (get_bits(gb, 3)) { // numLayer
2621 av_log_missing_feature(latmctx->aac_ctx.avctx,
2622 "multiple layers are not supported\n", 1);
2623 return AVERROR_PATCHWELCOME;
2626 // for all but first stream: use_same_config = get_bits(gb, 1);
2627 if (!audio_mux_version) {
2628 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2631 int ascLen = latm_get_value(gb);
2632 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2635 skip_bits_long(gb, ascLen);
2638 latmctx->frame_length_type = get_bits(gb, 3);
2639 switch (latmctx->frame_length_type) {
2641 skip_bits(gb, 8); // latmBufferFullness
2644 latmctx->frame_length = get_bits(gb, 9);
2649 skip_bits(gb, 6); // CELP frame length table index
2653 skip_bits(gb, 1); // HVXC frame length table index
2657 if (get_bits(gb, 1)) { // other data
2658 if (audio_mux_version) {
2659 latm_get_value(gb); // other_data_bits
2663 esc = get_bits(gb, 1);
2669 if (get_bits(gb, 1)) // crc present
2670 skip_bits(gb, 8); // config_crc
2676 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2680 if (ctx->frame_length_type == 0) {
2681 int mux_slot_length = 0;
2683 tmp = get_bits(gb, 8);
2684 mux_slot_length += tmp;
2685 } while (tmp == 255);
2686 return mux_slot_length;
2687 } else if (ctx->frame_length_type == 1) {
2688 return ctx->frame_length;
2689 } else if (ctx->frame_length_type == 3 ||
2690 ctx->frame_length_type == 5 ||
2691 ctx->frame_length_type == 7) {
2692 skip_bits(gb, 2); // mux_slot_length_coded
2697 static int read_audio_mux_element(struct LATMContext *latmctx,
2701 uint8_t use_same_mux = get_bits(gb, 1);
2702 if (!use_same_mux) {
2703 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2705 } else if (!latmctx->aac_ctx.avctx->extradata) {
2706 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2707 "no decoder config found\n");
2708 return AVERROR(EAGAIN);
2710 if (latmctx->audio_mux_version_A == 0) {
2711 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2712 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2713 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2714 return AVERROR_INVALIDDATA;
2715 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2716 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2717 "frame length mismatch %d << %d\n",
2718 mux_slot_length_bytes * 8, get_bits_left(gb));
2719 return AVERROR_INVALIDDATA;
2726 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2727 int *got_frame_ptr, AVPacket *avpkt)
2729 struct LATMContext *latmctx = avctx->priv_data;
2733 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2735 // check for LOAS sync word
2736 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2737 return AVERROR_INVALIDDATA;
2739 muxlength = get_bits(&gb, 13) + 3;
2740 // not enough data, the parser should have sorted this
2741 if (muxlength > avpkt->size)
2742 return AVERROR_INVALIDDATA;
2744 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2747 if (!latmctx->initialized) {
2748 if (!avctx->extradata) {
2752 if ((err = decode_audio_specific_config(
2753 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2754 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2756 latmctx->initialized = 1;
2760 if (show_bits(&gb, 12) == 0xfff) {
2761 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2762 "ADTS header detected, probably as result of configuration "
2764 return AVERROR_INVALIDDATA;
2767 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2773 av_cold static int latm_decode_init(AVCodecContext *avctx)
2775 struct LATMContext *latmctx = avctx->priv_data;
2776 int ret = aac_decode_init(avctx);
2778 if (avctx->extradata_size > 0)
2779 latmctx->initialized = !ret;
2785 AVCodec ff_aac_decoder = {
2787 .type = AVMEDIA_TYPE_AUDIO,
2789 .priv_data_size = sizeof(AACContext),
2790 .init = aac_decode_init,
2791 .close = aac_decode_close,
2792 .decode = aac_decode_frame,
2793 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2794 .sample_fmts = (const enum AVSampleFormat[]) {
2795 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2797 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2798 .channel_layouts = aac_channel_layout,
2802 Note: This decoder filter is intended to decode LATM streams transferred
2803 in MPEG transport streams which only contain one program.
2804 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2806 AVCodec ff_aac_latm_decoder = {
2808 .type = AVMEDIA_TYPE_AUDIO,
2809 .id = CODEC_ID_AAC_LATM,
2810 .priv_data_size = sizeof(struct LATMContext),
2811 .init = latm_decode_init,
2812 .close = aac_decode_close,
2813 .decode = latm_decode_frame,
2814 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2815 .sample_fmts = (const enum AVSampleFormat[]) {
2816 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2818 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2819 .channel_layouts = aac_channel_layout,