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 Libav.
12 * Libav 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 * Libav 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 Libav; 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 Libav 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 avctx->channel_layout = layout;
436 avctx->channels = channels;
437 ac->output_configured = oc_type;
443 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
445 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
446 * @param sce_map mono (Single Channel Element) map
447 * @param type speaker type/position for these channels
449 static void decode_channel_map(uint8_t layout_map[][3],
450 enum ChannelPosition type,
451 GetBitContext *gb, int n)
454 enum RawDataBlockType syn_ele;
456 case AAC_CHANNEL_FRONT:
457 case AAC_CHANNEL_BACK:
458 case AAC_CHANNEL_SIDE:
459 syn_ele = get_bits1(gb);
465 case AAC_CHANNEL_LFE:
469 layout_map[0][0] = syn_ele;
470 layout_map[0][1] = get_bits(gb, 4);
471 layout_map[0][2] = type;
477 * Decode program configuration element; reference: table 4.2.
479 * @return Returns error status. 0 - OK, !0 - error
481 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
482 uint8_t (*layout_map)[3],
485 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
489 skip_bits(gb, 2); // object_type
491 sampling_index = get_bits(gb, 4);
492 if (m4ac->sampling_index != sampling_index)
493 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");
495 num_front = get_bits(gb, 4);
496 num_side = get_bits(gb, 4);
497 num_back = get_bits(gb, 4);
498 num_lfe = get_bits(gb, 2);
499 num_assoc_data = get_bits(gb, 3);
500 num_cc = get_bits(gb, 4);
503 skip_bits(gb, 4); // mono_mixdown_tag
505 skip_bits(gb, 4); // stereo_mixdown_tag
508 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
510 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
512 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
514 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
516 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
519 skip_bits_long(gb, 4 * num_assoc_data);
521 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
526 /* comment field, first byte is length */
527 comment_len = get_bits(gb, 8) * 8;
528 if (get_bits_left(gb) < comment_len) {
529 av_log(avctx, AV_LOG_ERROR, overread_err);
532 skip_bits_long(gb, comment_len);
537 * Set up channel positions based on a default channel configuration
538 * as specified in table 1.17.
540 * @return Returns error status. 0 - OK, !0 - error
542 static av_cold int set_default_channel_config(AVCodecContext *avctx,
543 uint8_t (*layout_map)[3],
547 if (channel_config < 1 || channel_config > 7) {
548 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
552 *tags = tags_per_config[channel_config];
553 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
558 * Decode GA "General Audio" specific configuration; reference: table 4.1.
560 * @param ac pointer to AACContext, may be null
561 * @param avctx pointer to AVCCodecContext, used for logging
563 * @return Returns error status. 0 - OK, !0 - error
565 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
567 MPEG4AudioConfig *m4ac,
570 int extension_flag, ret;
571 uint8_t layout_map[MAX_ELEM_ID*4][3];
574 if (get_bits1(gb)) { // frameLengthFlag
575 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
579 if (get_bits1(gb)) // dependsOnCoreCoder
580 skip_bits(gb, 14); // coreCoderDelay
581 extension_flag = get_bits1(gb);
583 if (m4ac->object_type == AOT_AAC_SCALABLE ||
584 m4ac->object_type == AOT_ER_AAC_SCALABLE)
585 skip_bits(gb, 3); // layerNr
587 if (channel_config == 0) {
588 skip_bits(gb, 4); // element_instance_tag
589 tags = decode_pce(avctx, m4ac, layout_map, gb);
593 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
597 if (count_channels(layout_map, tags) > 1) {
599 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
602 if (ac && (ret = output_configure(ac, layout_map, tags,
603 channel_config, OC_GLOBAL_HDR)))
606 if (extension_flag) {
607 switch (m4ac->object_type) {
609 skip_bits(gb, 5); // numOfSubFrame
610 skip_bits(gb, 11); // layer_length
614 case AOT_ER_AAC_SCALABLE:
616 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
617 * aacScalefactorDataResilienceFlag
618 * aacSpectralDataResilienceFlag
622 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
628 * Decode audio specific configuration; reference: table 1.13.
630 * @param ac pointer to AACContext, may be null
631 * @param avctx pointer to AVCCodecContext, used for logging
632 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
633 * @param data pointer to buffer holding an audio specific config
634 * @param bit_size size of audio specific config or data in bits
635 * @param sync_extension look for an appended sync extension
637 * @return Returns error status or number of consumed bits. <0 - error
639 static int decode_audio_specific_config(AACContext *ac,
640 AVCodecContext *avctx,
641 MPEG4AudioConfig *m4ac,
642 const uint8_t *data, int bit_size,
648 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
649 for (i = 0; i < avctx->extradata_size; i++)
650 av_dlog(avctx, "%02x ", avctx->extradata[i]);
651 av_dlog(avctx, "\n");
653 init_get_bits(&gb, data, bit_size);
655 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
657 if (m4ac->sampling_index > 12) {
658 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
662 skip_bits_long(&gb, i);
664 switch (m4ac->object_type) {
668 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
672 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
673 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
677 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
678 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
679 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
681 return get_bits_count(&gb);
685 * linear congruential pseudorandom number generator
687 * @param previous_val pointer to the current state of the generator
689 * @return Returns a 32-bit pseudorandom integer
691 static av_always_inline int lcg_random(int previous_val)
693 return previous_val * 1664525 + 1013904223;
696 static av_always_inline void reset_predict_state(PredictorState *ps)
706 static void reset_all_predictors(PredictorState *ps)
709 for (i = 0; i < MAX_PREDICTORS; i++)
710 reset_predict_state(&ps[i]);
713 static int sample_rate_idx (int rate)
715 if (92017 <= rate) return 0;
716 else if (75132 <= rate) return 1;
717 else if (55426 <= rate) return 2;
718 else if (46009 <= rate) return 3;
719 else if (37566 <= rate) return 4;
720 else if (27713 <= rate) return 5;
721 else if (23004 <= rate) return 6;
722 else if (18783 <= rate) return 7;
723 else if (13856 <= rate) return 8;
724 else if (11502 <= rate) return 9;
725 else if (9391 <= rate) return 10;
729 static void reset_predictor_group(PredictorState *ps, int group_num)
732 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
733 reset_predict_state(&ps[i]);
736 #define AAC_INIT_VLC_STATIC(num, size) \
737 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
738 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
739 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
742 static av_cold int aac_decode_init(AVCodecContext *avctx)
744 AACContext *ac = avctx->priv_data;
745 float output_scale_factor;
748 ac->m4ac.sample_rate = avctx->sample_rate;
750 if (avctx->extradata_size > 0) {
751 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
753 avctx->extradata_size*8, 1) < 0)
757 uint8_t layout_map[MAX_ELEM_ID*4][3];
760 sr = sample_rate_idx(avctx->sample_rate);
761 ac->m4ac.sampling_index = sr;
762 ac->m4ac.channels = avctx->channels;
766 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
767 if (ff_mpeg4audio_channels[i] == avctx->channels)
769 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
772 ac->m4ac.chan_config = i;
774 if (ac->m4ac.chan_config) {
775 int ret = set_default_channel_config(avctx, layout_map,
776 &layout_map_tags, ac->m4ac.chan_config);
778 output_configure(ac, layout_map, layout_map_tags,
779 ac->m4ac.chan_config, OC_GLOBAL_HDR);
780 else if (avctx->err_recognition & AV_EF_EXPLODE)
781 return AVERROR_INVALIDDATA;
785 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
786 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
787 output_scale_factor = 1.0 / 32768.0;
789 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
790 output_scale_factor = 1.0;
793 AAC_INIT_VLC_STATIC( 0, 304);
794 AAC_INIT_VLC_STATIC( 1, 270);
795 AAC_INIT_VLC_STATIC( 2, 550);
796 AAC_INIT_VLC_STATIC( 3, 300);
797 AAC_INIT_VLC_STATIC( 4, 328);
798 AAC_INIT_VLC_STATIC( 5, 294);
799 AAC_INIT_VLC_STATIC( 6, 306);
800 AAC_INIT_VLC_STATIC( 7, 268);
801 AAC_INIT_VLC_STATIC( 8, 510);
802 AAC_INIT_VLC_STATIC( 9, 366);
803 AAC_INIT_VLC_STATIC(10, 462);
807 ff_dsputil_init(&ac->dsp, avctx);
808 ff_fmt_convert_init(&ac->fmt_conv, avctx);
810 ac->random_state = 0x1f2e3d4c;
814 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
815 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
816 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
819 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
820 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
821 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
822 // window initialization
823 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
824 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
825 ff_init_ff_sine_windows(10);
826 ff_init_ff_sine_windows( 7);
830 avcodec_get_frame_defaults(&ac->frame);
831 avctx->coded_frame = &ac->frame;
837 * Skip data_stream_element; reference: table 4.10.
839 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
841 int byte_align = get_bits1(gb);
842 int count = get_bits(gb, 8);
844 count += get_bits(gb, 8);
848 if (get_bits_left(gb) < 8 * count) {
849 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
852 skip_bits_long(gb, 8 * count);
856 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
861 ics->predictor_reset_group = get_bits(gb, 5);
862 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
863 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
867 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
868 ics->prediction_used[sfb] = get_bits1(gb);
874 * Decode Long Term Prediction data; reference: table 4.xx.
876 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
877 GetBitContext *gb, uint8_t max_sfb)
881 ltp->lag = get_bits(gb, 11);
882 ltp->coef = ltp_coef[get_bits(gb, 3)];
883 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
884 ltp->used[sfb] = get_bits1(gb);
888 * Decode Individual Channel Stream info; reference: table 4.6.
890 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
894 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
895 return AVERROR_INVALIDDATA;
897 ics->window_sequence[1] = ics->window_sequence[0];
898 ics->window_sequence[0] = get_bits(gb, 2);
899 ics->use_kb_window[1] = ics->use_kb_window[0];
900 ics->use_kb_window[0] = get_bits1(gb);
901 ics->num_window_groups = 1;
902 ics->group_len[0] = 1;
903 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
905 ics->max_sfb = get_bits(gb, 4);
906 for (i = 0; i < 7; i++) {
908 ics->group_len[ics->num_window_groups - 1]++;
910 ics->num_window_groups++;
911 ics->group_len[ics->num_window_groups - 1] = 1;
914 ics->num_windows = 8;
915 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
916 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
917 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
918 ics->predictor_present = 0;
920 ics->max_sfb = get_bits(gb, 6);
921 ics->num_windows = 1;
922 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
923 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
924 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
925 ics->predictor_present = get_bits1(gb);
926 ics->predictor_reset_group = 0;
927 if (ics->predictor_present) {
928 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
929 if (decode_prediction(ac, ics, gb)) {
930 return AVERROR_INVALIDDATA;
932 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
933 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
934 return AVERROR_INVALIDDATA;
936 if ((ics->ltp.present = get_bits(gb, 1)))
937 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
942 if (ics->max_sfb > ics->num_swb) {
943 av_log(ac->avctx, AV_LOG_ERROR,
944 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
945 ics->max_sfb, ics->num_swb);
946 return AVERROR_INVALIDDATA;
953 * Decode band types (section_data payload); reference: table 4.46.
955 * @param band_type array of the used band type
956 * @param band_type_run_end array of the last scalefactor band of a band type run
958 * @return Returns error status. 0 - OK, !0 - error
960 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
961 int band_type_run_end[120], GetBitContext *gb,
962 IndividualChannelStream *ics)
965 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
966 for (g = 0; g < ics->num_window_groups; g++) {
968 while (k < ics->max_sfb) {
969 uint8_t sect_end = k;
971 int sect_band_type = get_bits(gb, 4);
972 if (sect_band_type == 12) {
973 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
976 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
977 sect_end += sect_len_incr;
978 sect_end += sect_len_incr;
979 if (get_bits_left(gb) < 0) {
980 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
983 if (sect_end > ics->max_sfb) {
984 av_log(ac->avctx, AV_LOG_ERROR,
985 "Number of bands (%d) exceeds limit (%d).\n",
986 sect_end, ics->max_sfb);
989 for (; k < sect_end; k++) {
990 band_type [idx] = sect_band_type;
991 band_type_run_end[idx++] = sect_end;
999 * Decode scalefactors; reference: table 4.47.
1001 * @param global_gain first scalefactor value as scalefactors are differentially coded
1002 * @param band_type array of the used band type
1003 * @param band_type_run_end array of the last scalefactor band of a band type run
1004 * @param sf array of scalefactors or intensity stereo positions
1006 * @return Returns error status. 0 - OK, !0 - error
1008 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1009 unsigned int global_gain,
1010 IndividualChannelStream *ics,
1011 enum BandType band_type[120],
1012 int band_type_run_end[120])
1015 int offset[3] = { global_gain, global_gain - 90, 0 };
1018 static const char *const sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
1019 for (g = 0; g < ics->num_window_groups; g++) {
1020 for (i = 0; i < ics->max_sfb;) {
1021 int run_end = band_type_run_end[idx];
1022 if (band_type[idx] == ZERO_BT) {
1023 for (; i < run_end; i++, idx++)
1025 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1026 for (; i < run_end; i++, idx++) {
1027 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1028 clipped_offset = av_clip(offset[2], -155, 100);
1029 if (offset[2] != clipped_offset) {
1030 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1031 "position clipped (%d -> %d).\nIf you heard an "
1032 "audible artifact, there may be a bug in the "
1033 "decoder. ", offset[2], clipped_offset);
1035 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1037 } else if (band_type[idx] == NOISE_BT) {
1038 for (; i < run_end; i++, idx++) {
1039 if (noise_flag-- > 0)
1040 offset[1] += get_bits(gb, 9) - 256;
1042 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1043 clipped_offset = av_clip(offset[1], -100, 155);
1044 if (offset[1] != clipped_offset) {
1045 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1046 "(%d -> %d).\nIf you heard an audible "
1047 "artifact, there may be a bug in the decoder. ",
1048 offset[1], clipped_offset);
1050 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1053 for (; i < run_end; i++, idx++) {
1054 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1055 if (offset[0] > 255U) {
1056 av_log(ac->avctx, AV_LOG_ERROR,
1057 "%s (%d) out of range.\n", sf_str[0], offset[0]);
1060 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1069 * Decode pulse data; reference: table 4.7.
1071 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1072 const uint16_t *swb_offset, int num_swb)
1075 pulse->num_pulse = get_bits(gb, 2) + 1;
1076 pulse_swb = get_bits(gb, 6);
1077 if (pulse_swb >= num_swb)
1079 pulse->pos[0] = swb_offset[pulse_swb];
1080 pulse->pos[0] += get_bits(gb, 5);
1081 if (pulse->pos[0] > 1023)
1083 pulse->amp[0] = get_bits(gb, 4);
1084 for (i = 1; i < pulse->num_pulse; i++) {
1085 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1086 if (pulse->pos[i] > 1023)
1088 pulse->amp[i] = get_bits(gb, 4);
1094 * Decode Temporal Noise Shaping data; reference: table 4.48.
1096 * @return Returns error status. 0 - OK, !0 - error
1098 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1099 GetBitContext *gb, const IndividualChannelStream *ics)
1101 int w, filt, i, coef_len, coef_res, coef_compress;
1102 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1103 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1104 for (w = 0; w < ics->num_windows; w++) {
1105 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1106 coef_res = get_bits1(gb);
1108 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1110 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1112 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1113 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1114 tns->order[w][filt], tns_max_order);
1115 tns->order[w][filt] = 0;
1118 if (tns->order[w][filt]) {
1119 tns->direction[w][filt] = get_bits1(gb);
1120 coef_compress = get_bits1(gb);
1121 coef_len = coef_res + 3 - coef_compress;
1122 tmp2_idx = 2 * coef_compress + coef_res;
1124 for (i = 0; i < tns->order[w][filt]; i++)
1125 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1134 * Decode Mid/Side data; reference: table 4.54.
1136 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1137 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1138 * [3] reserved for scalable AAC
1140 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1144 if (ms_present == 1) {
1145 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1146 cpe->ms_mask[idx] = get_bits1(gb);
1147 } else if (ms_present == 2) {
1148 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1153 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1157 *dst++ = v[idx & 15] * s;
1158 *dst++ = v[idx>>4 & 15] * s;
1164 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1168 *dst++ = v[idx & 3] * s;
1169 *dst++ = v[idx>>2 & 3] * s;
1170 *dst++ = v[idx>>4 & 3] * s;
1171 *dst++ = v[idx>>6 & 3] * s;
1177 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1178 unsigned sign, const float *scale)
1180 union av_intfloat32 s0, s1;
1182 s0.f = s1.f = *scale;
1183 s0.i ^= sign >> 1 << 31;
1186 *dst++ = v[idx & 15] * s0.f;
1187 *dst++ = v[idx>>4 & 15] * s1.f;
1194 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1195 unsigned sign, const float *scale)
1197 unsigned nz = idx >> 12;
1198 union av_intfloat32 s = { .f = *scale };
1199 union av_intfloat32 t;
1201 t.i = s.i ^ (sign & 1U<<31);
1202 *dst++ = v[idx & 3] * t.f;
1204 sign <<= nz & 1; nz >>= 1;
1205 t.i = s.i ^ (sign & 1U<<31);
1206 *dst++ = v[idx>>2 & 3] * t.f;
1208 sign <<= nz & 1; nz >>= 1;
1209 t.i = s.i ^ (sign & 1U<<31);
1210 *dst++ = v[idx>>4 & 3] * t.f;
1212 sign <<= nz & 1; nz >>= 1;
1213 t.i = s.i ^ (sign & 1U<<31);
1214 *dst++ = v[idx>>6 & 3] * t.f;
1221 * Decode spectral data; reference: table 4.50.
1222 * Dequantize and scale spectral data; reference: 4.6.3.3.
1224 * @param coef array of dequantized, scaled spectral data
1225 * @param sf array of scalefactors or intensity stereo positions
1226 * @param pulse_present set if pulses are present
1227 * @param pulse pointer to pulse data struct
1228 * @param band_type array of the used band type
1230 * @return Returns error status. 0 - OK, !0 - error
1232 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1233 GetBitContext *gb, const float sf[120],
1234 int pulse_present, const Pulse *pulse,
1235 const IndividualChannelStream *ics,
1236 enum BandType band_type[120])
1238 int i, k, g, idx = 0;
1239 const int c = 1024 / ics->num_windows;
1240 const uint16_t *offsets = ics->swb_offset;
1241 float *coef_base = coef;
1243 for (g = 0; g < ics->num_windows; g++)
1244 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1246 for (g = 0; g < ics->num_window_groups; g++) {
1247 unsigned g_len = ics->group_len[g];
1249 for (i = 0; i < ics->max_sfb; i++, idx++) {
1250 const unsigned cbt_m1 = band_type[idx] - 1;
1251 float *cfo = coef + offsets[i];
1252 int off_len = offsets[i + 1] - offsets[i];
1255 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1256 for (group = 0; group < g_len; group++, cfo+=128) {
1257 memset(cfo, 0, off_len * sizeof(float));
1259 } else if (cbt_m1 == NOISE_BT - 1) {
1260 for (group = 0; group < g_len; group++, cfo+=128) {
1264 for (k = 0; k < off_len; k++) {
1265 ac->random_state = lcg_random(ac->random_state);
1266 cfo[k] = ac->random_state;
1269 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1270 scale = sf[idx] / sqrtf(band_energy);
1271 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1274 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1275 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1276 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1277 OPEN_READER(re, gb);
1279 switch (cbt_m1 >> 1) {
1281 for (group = 0; group < g_len; group++, cfo+=128) {
1289 UPDATE_CACHE(re, gb);
1290 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1291 cb_idx = cb_vector_idx[code];
1292 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1298 for (group = 0; group < g_len; group++, cfo+=128) {
1308 UPDATE_CACHE(re, gb);
1309 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1310 cb_idx = cb_vector_idx[code];
1311 nnz = cb_idx >> 8 & 15;
1312 bits = nnz ? GET_CACHE(re, gb) : 0;
1313 LAST_SKIP_BITS(re, gb, nnz);
1314 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1320 for (group = 0; group < g_len; group++, cfo+=128) {
1328 UPDATE_CACHE(re, gb);
1329 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1330 cb_idx = cb_vector_idx[code];
1331 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1338 for (group = 0; group < g_len; group++, cfo+=128) {
1348 UPDATE_CACHE(re, gb);
1349 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1350 cb_idx = cb_vector_idx[code];
1351 nnz = cb_idx >> 8 & 15;
1352 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1353 LAST_SKIP_BITS(re, gb, nnz);
1354 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1360 for (group = 0; group < g_len; group++, cfo+=128) {
1362 uint32_t *icf = (uint32_t *) cf;
1372 UPDATE_CACHE(re, gb);
1373 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1381 cb_idx = cb_vector_idx[code];
1384 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1385 LAST_SKIP_BITS(re, gb, nnz);
1387 for (j = 0; j < 2; j++) {
1391 /* The total length of escape_sequence must be < 22 bits according
1392 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1393 UPDATE_CACHE(re, gb);
1394 b = GET_CACHE(re, gb);
1395 b = 31 - av_log2(~b);
1398 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1402 SKIP_BITS(re, gb, b + 1);
1404 n = (1 << b) + SHOW_UBITS(re, gb, b);
1405 LAST_SKIP_BITS(re, gb, b);
1406 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1409 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1410 *icf++ = (bits & 1U<<31) | v;
1417 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1421 CLOSE_READER(re, gb);
1427 if (pulse_present) {
1429 for (i = 0; i < pulse->num_pulse; i++) {
1430 float co = coef_base[ pulse->pos[i] ];
1431 while (offsets[idx + 1] <= pulse->pos[i])
1433 if (band_type[idx] != NOISE_BT && sf[idx]) {
1434 float ico = -pulse->amp[i];
1437 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1439 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1446 static av_always_inline float flt16_round(float pf)
1448 union av_intfloat32 tmp;
1450 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1454 static av_always_inline float flt16_even(float pf)
1456 union av_intfloat32 tmp;
1458 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1462 static av_always_inline float flt16_trunc(float pf)
1464 union av_intfloat32 pun;
1466 pun.i &= 0xFFFF0000U;
1470 static av_always_inline void predict(PredictorState *ps, float *coef,
1473 const float a = 0.953125; // 61.0 / 64
1474 const float alpha = 0.90625; // 29.0 / 32
1478 float r0 = ps->r0, r1 = ps->r1;
1479 float cor0 = ps->cor0, cor1 = ps->cor1;
1480 float var0 = ps->var0, var1 = ps->var1;
1482 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1483 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1485 pv = flt16_round(k1 * r0 + k2 * r1);
1492 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1493 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1494 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1495 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1497 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1498 ps->r0 = flt16_trunc(a * e0);
1502 * Apply AAC-Main style frequency domain prediction.
1504 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1508 if (!sce->ics.predictor_initialized) {
1509 reset_all_predictors(sce->predictor_state);
1510 sce->ics.predictor_initialized = 1;
1513 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1514 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1515 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1516 predict(&sce->predictor_state[k], &sce->coeffs[k],
1517 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1520 if (sce->ics.predictor_reset_group)
1521 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1523 reset_all_predictors(sce->predictor_state);
1527 * Decode an individual_channel_stream payload; reference: table 4.44.
1529 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1530 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1532 * @return Returns error status. 0 - OK, !0 - error
1534 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1535 GetBitContext *gb, int common_window, int scale_flag)
1538 TemporalNoiseShaping *tns = &sce->tns;
1539 IndividualChannelStream *ics = &sce->ics;
1540 float *out = sce->coeffs;
1541 int global_gain, pulse_present = 0;
1543 /* This assignment is to silence a GCC warning about the variable being used
1544 * uninitialized when in fact it always is.
1546 pulse.num_pulse = 0;
1548 global_gain = get_bits(gb, 8);
1550 if (!common_window && !scale_flag) {
1551 if (decode_ics_info(ac, ics, gb) < 0)
1552 return AVERROR_INVALIDDATA;
1555 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1557 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1562 if ((pulse_present = get_bits1(gb))) {
1563 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1564 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1567 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1568 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1572 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1574 if (get_bits1(gb)) {
1575 av_log_missing_feature(ac->avctx, "SSR", 1);
1580 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1583 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1584 apply_prediction(ac, sce);
1590 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1592 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1594 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1595 float *ch0 = cpe->ch[0].coeffs;
1596 float *ch1 = cpe->ch[1].coeffs;
1597 int g, i, group, idx = 0;
1598 const uint16_t *offsets = ics->swb_offset;
1599 for (g = 0; g < ics->num_window_groups; g++) {
1600 for (i = 0; i < ics->max_sfb; i++, idx++) {
1601 if (cpe->ms_mask[idx] &&
1602 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1603 for (group = 0; group < ics->group_len[g]; group++) {
1604 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1605 ch1 + group * 128 + offsets[i],
1606 offsets[i+1] - offsets[i]);
1610 ch0 += ics->group_len[g] * 128;
1611 ch1 += ics->group_len[g] * 128;
1616 * intensity stereo decoding; reference: 4.6.8.2.3
1618 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1619 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1620 * [3] reserved for scalable AAC
1622 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1624 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1625 SingleChannelElement *sce1 = &cpe->ch[1];
1626 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1627 const uint16_t *offsets = ics->swb_offset;
1628 int g, group, i, idx = 0;
1631 for (g = 0; g < ics->num_window_groups; g++) {
1632 for (i = 0; i < ics->max_sfb;) {
1633 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1634 const int bt_run_end = sce1->band_type_run_end[idx];
1635 for (; i < bt_run_end; i++, idx++) {
1636 c = -1 + 2 * (sce1->band_type[idx] - 14);
1638 c *= 1 - 2 * cpe->ms_mask[idx];
1639 scale = c * sce1->sf[idx];
1640 for (group = 0; group < ics->group_len[g]; group++)
1641 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1642 coef0 + group * 128 + offsets[i],
1644 offsets[i + 1] - offsets[i]);
1647 int bt_run_end = sce1->band_type_run_end[idx];
1648 idx += bt_run_end - i;
1652 coef0 += ics->group_len[g] * 128;
1653 coef1 += ics->group_len[g] * 128;
1658 * Decode a channel_pair_element; reference: table 4.4.
1660 * @return Returns error status. 0 - OK, !0 - error
1662 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1664 int i, ret, common_window, ms_present = 0;
1666 common_window = get_bits1(gb);
1667 if (common_window) {
1668 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1669 return AVERROR_INVALIDDATA;
1670 i = cpe->ch[1].ics.use_kb_window[0];
1671 cpe->ch[1].ics = cpe->ch[0].ics;
1672 cpe->ch[1].ics.use_kb_window[1] = i;
1673 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1674 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1675 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1676 ms_present = get_bits(gb, 2);
1677 if (ms_present == 3) {
1678 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1680 } else if (ms_present)
1681 decode_mid_side_stereo(cpe, gb, ms_present);
1683 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1685 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1688 if (common_window) {
1690 apply_mid_side_stereo(ac, cpe);
1691 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1692 apply_prediction(ac, &cpe->ch[0]);
1693 apply_prediction(ac, &cpe->ch[1]);
1697 apply_intensity_stereo(ac, cpe, ms_present);
1701 static const float cce_scale[] = {
1702 1.09050773266525765921, //2^(1/8)
1703 1.18920711500272106672, //2^(1/4)
1709 * Decode coupling_channel_element; reference: table 4.8.
1711 * @return Returns error status. 0 - OK, !0 - error
1713 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1719 SingleChannelElement *sce = &che->ch[0];
1720 ChannelCoupling *coup = &che->coup;
1722 coup->coupling_point = 2 * get_bits1(gb);
1723 coup->num_coupled = get_bits(gb, 3);
1724 for (c = 0; c <= coup->num_coupled; c++) {
1726 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1727 coup->id_select[c] = get_bits(gb, 4);
1728 if (coup->type[c] == TYPE_CPE) {
1729 coup->ch_select[c] = get_bits(gb, 2);
1730 if (coup->ch_select[c] == 3)
1733 coup->ch_select[c] = 2;
1735 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1737 sign = get_bits(gb, 1);
1738 scale = cce_scale[get_bits(gb, 2)];
1740 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1743 for (c = 0; c < num_gain; c++) {
1747 float gain_cache = 1.;
1749 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1750 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1751 gain_cache = powf(scale, -gain);
1753 if (coup->coupling_point == AFTER_IMDCT) {
1754 coup->gain[c][0] = gain_cache;
1756 for (g = 0; g < sce->ics.num_window_groups; g++) {
1757 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1758 if (sce->band_type[idx] != ZERO_BT) {
1760 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1768 gain_cache = powf(scale, -t) * s;
1771 coup->gain[c][idx] = gain_cache;
1781 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1783 * @return Returns number of bytes consumed.
1785 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1789 int num_excl_chan = 0;
1792 for (i = 0; i < 7; i++)
1793 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1794 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1796 return num_excl_chan / 7;
1800 * Decode dynamic range information; reference: table 4.52.
1802 * @param cnt length of TYPE_FIL syntactic element in bytes
1804 * @return Returns number of bytes consumed.
1806 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1807 GetBitContext *gb, int cnt)
1810 int drc_num_bands = 1;
1813 /* pce_tag_present? */
1814 if (get_bits1(gb)) {
1815 che_drc->pce_instance_tag = get_bits(gb, 4);
1816 skip_bits(gb, 4); // tag_reserved_bits
1820 /* excluded_chns_present? */
1821 if (get_bits1(gb)) {
1822 n += decode_drc_channel_exclusions(che_drc, gb);
1825 /* drc_bands_present? */
1826 if (get_bits1(gb)) {
1827 che_drc->band_incr = get_bits(gb, 4);
1828 che_drc->interpolation_scheme = get_bits(gb, 4);
1830 drc_num_bands += che_drc->band_incr;
1831 for (i = 0; i < drc_num_bands; i++) {
1832 che_drc->band_top[i] = get_bits(gb, 8);
1837 /* prog_ref_level_present? */
1838 if (get_bits1(gb)) {
1839 che_drc->prog_ref_level = get_bits(gb, 7);
1840 skip_bits1(gb); // prog_ref_level_reserved_bits
1844 for (i = 0; i < drc_num_bands; i++) {
1845 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1846 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1854 * Decode extension data (incomplete); reference: table 4.51.
1856 * @param cnt length of TYPE_FIL syntactic element in bytes
1858 * @return Returns number of bytes consumed
1860 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1861 ChannelElement *che, enum RawDataBlockType elem_type)
1865 switch (get_bits(gb, 4)) { // extension type
1866 case EXT_SBR_DATA_CRC:
1870 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1872 } else if (!ac->m4ac.sbr) {
1873 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1874 skip_bits_long(gb, 8 * cnt - 4);
1876 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1877 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1878 skip_bits_long(gb, 8 * cnt - 4);
1880 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1883 output_configure(ac, ac->layout_map, ac->layout_map_tags,
1884 ac->m4ac.chan_config, ac->output_configured);
1888 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1890 case EXT_DYNAMIC_RANGE:
1891 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1895 case EXT_DATA_ELEMENT:
1897 skip_bits_long(gb, 8 * cnt - 4);
1904 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1906 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1907 * @param coef spectral coefficients
1909 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1910 IndividualChannelStream *ics, int decode)
1912 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1914 int bottom, top, order, start, end, size, inc;
1915 float lpc[TNS_MAX_ORDER];
1916 float tmp[TNS_MAX_ORDER];
1918 for (w = 0; w < ics->num_windows; w++) {
1919 bottom = ics->num_swb;
1920 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1922 bottom = FFMAX(0, top - tns->length[w][filt]);
1923 order = tns->order[w][filt];
1928 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1930 start = ics->swb_offset[FFMIN(bottom, mmm)];
1931 end = ics->swb_offset[FFMIN( top, mmm)];
1932 if ((size = end - start) <= 0)
1934 if (tns->direction[w][filt]) {
1944 for (m = 0; m < size; m++, start += inc)
1945 for (i = 1; i <= FFMIN(m, order); i++)
1946 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1949 for (m = 0; m < size; m++, start += inc) {
1950 tmp[0] = coef[start];
1951 for (i = 1; i <= FFMIN(m, order); i++)
1952 coef[start] += tmp[i] * lpc[i - 1];
1953 for (i = order; i > 0; i--)
1954 tmp[i] = tmp[i - 1];
1962 * Apply windowing and MDCT to obtain the spectral
1963 * coefficient from the predicted sample by LTP.
1965 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1966 float *in, IndividualChannelStream *ics)
1968 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1969 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1970 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1971 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1973 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1974 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1976 memset(in, 0, 448 * sizeof(float));
1977 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1979 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1980 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1982 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1983 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1985 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1989 * Apply the long term prediction
1991 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1993 const LongTermPrediction *ltp = &sce->ics.ltp;
1994 const uint16_t *offsets = sce->ics.swb_offset;
1997 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1998 float *predTime = sce->ret;
1999 float *predFreq = ac->buf_mdct;
2000 int16_t num_samples = 2048;
2002 if (ltp->lag < 1024)
2003 num_samples = ltp->lag + 1024;
2004 for (i = 0; i < num_samples; i++)
2005 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2006 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2008 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2010 if (sce->tns.present)
2011 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2013 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2015 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2016 sce->coeffs[i] += predFreq[i];
2021 * Update the LTP buffer for next frame
2023 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2025 IndividualChannelStream *ics = &sce->ics;
2026 float *saved = sce->saved;
2027 float *saved_ltp = sce->coeffs;
2028 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2029 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2032 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2033 memcpy(saved_ltp, saved, 512 * sizeof(float));
2034 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2035 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2036 for (i = 0; i < 64; i++)
2037 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2038 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2039 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2040 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2041 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2042 for (i = 0; i < 64; i++)
2043 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2044 } else { // LONG_STOP or ONLY_LONG
2045 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2046 for (i = 0; i < 512; i++)
2047 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2050 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2051 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2052 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2056 * Conduct IMDCT and windowing.
2058 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2060 IndividualChannelStream *ics = &sce->ics;
2061 float *in = sce->coeffs;
2062 float *out = sce->ret;
2063 float *saved = sce->saved;
2064 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2065 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2066 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2067 float *buf = ac->buf_mdct;
2068 float *temp = ac->temp;
2072 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2073 for (i = 0; i < 1024; i += 128)
2074 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2076 ac->mdct.imdct_half(&ac->mdct, buf, in);
2078 /* window overlapping
2079 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2080 * and long to short transitions are considered to be short to short
2081 * transitions. This leaves just two cases (long to long and short to short)
2082 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2084 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2085 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2086 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2088 memcpy( out, saved, 448 * sizeof(float));
2090 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2091 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2092 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2093 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2094 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2095 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2096 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2098 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2099 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2104 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2105 memcpy( saved, temp + 64, 64 * sizeof(float));
2106 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2107 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2108 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2109 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2110 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2111 memcpy( saved, buf + 512, 448 * sizeof(float));
2112 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2113 } else { // LONG_STOP or ONLY_LONG
2114 memcpy( saved, buf + 512, 512 * sizeof(float));
2119 * Apply dependent channel coupling (applied before IMDCT).
2121 * @param index index into coupling gain array
2123 static void apply_dependent_coupling(AACContext *ac,
2124 SingleChannelElement *target,
2125 ChannelElement *cce, int index)
2127 IndividualChannelStream *ics = &cce->ch[0].ics;
2128 const uint16_t *offsets = ics->swb_offset;
2129 float *dest = target->coeffs;
2130 const float *src = cce->ch[0].coeffs;
2131 int g, i, group, k, idx = 0;
2132 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2133 av_log(ac->avctx, AV_LOG_ERROR,
2134 "Dependent coupling is not supported together with LTP\n");
2137 for (g = 0; g < ics->num_window_groups; g++) {
2138 for (i = 0; i < ics->max_sfb; i++, idx++) {
2139 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2140 const float gain = cce->coup.gain[index][idx];
2141 for (group = 0; group < ics->group_len[g]; group++) {
2142 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2144 dest[group * 128 + k] += gain * src[group * 128 + k];
2149 dest += ics->group_len[g] * 128;
2150 src += ics->group_len[g] * 128;
2155 * Apply independent channel coupling (applied after IMDCT).
2157 * @param index index into coupling gain array
2159 static void apply_independent_coupling(AACContext *ac,
2160 SingleChannelElement *target,
2161 ChannelElement *cce, int index)
2164 const float gain = cce->coup.gain[index][0];
2165 const float *src = cce->ch[0].ret;
2166 float *dest = target->ret;
2167 const int len = 1024 << (ac->m4ac.sbr == 1);
2169 for (i = 0; i < len; i++)
2170 dest[i] += gain * src[i];
2174 * channel coupling transformation interface
2176 * @param apply_coupling_method pointer to (in)dependent coupling function
2178 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2179 enum RawDataBlockType type, int elem_id,
2180 enum CouplingPoint coupling_point,
2181 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2185 for (i = 0; i < MAX_ELEM_ID; i++) {
2186 ChannelElement *cce = ac->che[TYPE_CCE][i];
2189 if (cce && cce->coup.coupling_point == coupling_point) {
2190 ChannelCoupling *coup = &cce->coup;
2192 for (c = 0; c <= coup->num_coupled; c++) {
2193 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2194 if (coup->ch_select[c] != 1) {
2195 apply_coupling_method(ac, &cc->ch[0], cce, index);
2196 if (coup->ch_select[c] != 0)
2199 if (coup->ch_select[c] != 2)
2200 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2202 index += 1 + (coup->ch_select[c] == 3);
2209 * Convert spectral data to float samples, applying all supported tools as appropriate.
2211 static void spectral_to_sample(AACContext *ac)
2214 for (type = 3; type >= 0; type--) {
2215 for (i = 0; i < MAX_ELEM_ID; i++) {
2216 ChannelElement *che = ac->che[type][i];
2218 if (type <= TYPE_CPE)
2219 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2220 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2221 if (che->ch[0].ics.predictor_present) {
2222 if (che->ch[0].ics.ltp.present)
2223 apply_ltp(ac, &che->ch[0]);
2224 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2225 apply_ltp(ac, &che->ch[1]);
2228 if (che->ch[0].tns.present)
2229 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2230 if (che->ch[1].tns.present)
2231 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2232 if (type <= TYPE_CPE)
2233 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2234 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2235 imdct_and_windowing(ac, &che->ch[0]);
2236 if (ac->m4ac.object_type == AOT_AAC_LTP)
2237 update_ltp(ac, &che->ch[0]);
2238 if (type == TYPE_CPE) {
2239 imdct_and_windowing(ac, &che->ch[1]);
2240 if (ac->m4ac.object_type == AOT_AAC_LTP)
2241 update_ltp(ac, &che->ch[1]);
2243 if (ac->m4ac.sbr > 0) {
2244 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2247 if (type <= TYPE_CCE)
2248 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2254 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2257 AACADTSHeaderInfo hdr_info;
2258 uint8_t layout_map[MAX_ELEM_ID*4][3];
2259 int layout_map_tags;
2261 size = avpriv_aac_parse_header(gb, &hdr_info);
2263 if (hdr_info.chan_config) {
2264 ac->m4ac.chan_config = hdr_info.chan_config;
2265 if (set_default_channel_config(ac->avctx, layout_map,
2266 &layout_map_tags, hdr_info.chan_config))
2268 if (output_configure(ac, layout_map, layout_map_tags,
2269 hdr_info.chan_config,
2270 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2272 } else if (ac->output_configured != OC_LOCKED) {
2273 ac->m4ac.chan_config = 0;
2274 ac->output_configured = OC_NONE;
2276 if (ac->output_configured != OC_LOCKED) {
2279 ac->m4ac.sample_rate = hdr_info.sample_rate;
2280 ac->m4ac.sampling_index = hdr_info.sampling_index;
2281 ac->m4ac.object_type = hdr_info.object_type;
2283 if (!ac->avctx->sample_rate)
2284 ac->avctx->sample_rate = hdr_info.sample_rate;
2285 if (hdr_info.num_aac_frames == 1) {
2286 if (!hdr_info.crc_absent)
2289 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2296 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2297 int *got_frame_ptr, GetBitContext *gb)
2299 AACContext *ac = avctx->priv_data;
2300 ChannelElement *che = NULL, *che_prev = NULL;
2301 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2303 int samples = 0, multiplier, audio_found = 0;
2305 if (show_bits(gb, 12) == 0xfff) {
2306 if (parse_adts_frame_header(ac, gb) < 0) {
2307 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2310 if (ac->m4ac.sampling_index > 12) {
2311 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2316 ac->tags_mapped = 0;
2318 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2319 elem_id = get_bits(gb, 4);
2321 if (elem_type < TYPE_DSE) {
2322 if (!(che=get_che(ac, elem_type, elem_id))) {
2323 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2324 elem_type, elem_id);
2330 switch (elem_type) {
2333 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2338 err = decode_cpe(ac, gb, che);
2343 err = decode_cce(ac, gb, che);
2347 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2352 err = skip_data_stream_element(ac, gb);
2356 uint8_t layout_map[MAX_ELEM_ID*4][3];
2358 tags = decode_pce(avctx, &ac->m4ac, layout_map, gb);
2363 if (ac->output_configured > OC_TRIAL_PCE)
2364 av_log(avctx, AV_LOG_ERROR,
2365 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2367 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2373 elem_id += get_bits(gb, 8) - 1;
2374 if (get_bits_left(gb) < 8 * elem_id) {
2375 av_log(avctx, AV_LOG_ERROR, overread_err);
2379 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2380 err = 0; /* FIXME */
2384 err = -1; /* should not happen, but keeps compiler happy */
2389 elem_type_prev = elem_type;
2394 if (get_bits_left(gb) < 3) {
2395 av_log(avctx, AV_LOG_ERROR, overread_err);
2400 spectral_to_sample(ac);
2402 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2403 samples <<= multiplier;
2404 if (ac->output_configured < OC_LOCKED) {
2405 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2406 avctx->frame_size = samples;
2410 /* get output buffer */
2411 ac->frame.nb_samples = samples;
2412 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2413 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2417 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2418 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2419 (const float **)ac->output_data,
2420 samples, avctx->channels);
2422 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2423 (const float **)ac->output_data,
2424 samples, avctx->channels);
2426 *(AVFrame *)data = ac->frame;
2428 *got_frame_ptr = !!samples;
2430 if (ac->output_configured && audio_found)
2431 ac->output_configured = OC_LOCKED;
2436 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2437 int *got_frame_ptr, AVPacket *avpkt)
2439 AACContext *ac = avctx->priv_data;
2440 const uint8_t *buf = avpkt->data;
2441 int buf_size = avpkt->size;
2446 int new_extradata_size;
2447 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2448 AV_PKT_DATA_NEW_EXTRADATA,
2449 &new_extradata_size);
2451 if (new_extradata) {
2452 av_free(avctx->extradata);
2453 avctx->extradata = av_mallocz(new_extradata_size +
2454 FF_INPUT_BUFFER_PADDING_SIZE);
2455 if (!avctx->extradata)
2456 return AVERROR(ENOMEM);
2457 avctx->extradata_size = new_extradata_size;
2458 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2459 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2461 avctx->extradata_size*8, 1) < 0)
2462 return AVERROR_INVALIDDATA;
2465 init_get_bits(&gb, buf, buf_size * 8);
2467 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2470 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2471 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2472 if (buf[buf_offset])
2475 return buf_size > buf_offset ? buf_consumed : buf_size;
2478 static av_cold int aac_decode_close(AVCodecContext *avctx)
2480 AACContext *ac = avctx->priv_data;
2483 for (i = 0; i < MAX_ELEM_ID; i++) {
2484 for (type = 0; type < 4; type++) {
2485 if (ac->che[type][i])
2486 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2487 av_freep(&ac->che[type][i]);
2491 ff_mdct_end(&ac->mdct);
2492 ff_mdct_end(&ac->mdct_small);
2493 ff_mdct_end(&ac->mdct_ltp);
2498 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2500 struct LATMContext {
2501 AACContext aac_ctx; ///< containing AACContext
2502 int initialized; ///< initilized after a valid extradata was seen
2505 int audio_mux_version_A; ///< LATM syntax version
2506 int frame_length_type; ///< 0/1 variable/fixed frame length
2507 int frame_length; ///< frame length for fixed frame length
2510 static inline uint32_t latm_get_value(GetBitContext *b)
2512 int length = get_bits(b, 2);
2514 return get_bits_long(b, (length+1)*8);
2517 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2518 GetBitContext *gb, int asclen)
2520 AACContext *ac = &latmctx->aac_ctx;
2521 AVCodecContext *avctx = ac->avctx;
2522 MPEG4AudioConfig m4ac = {0};
2523 int config_start_bit = get_bits_count(gb);
2524 int sync_extension = 0;
2525 int bits_consumed, esize;
2529 asclen = FFMIN(asclen, get_bits_left(gb));
2531 asclen = get_bits_left(gb);
2533 if (config_start_bit % 8) {
2534 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2535 "config not byte aligned.\n", 1);
2536 return AVERROR_INVALIDDATA;
2539 return AVERROR_INVALIDDATA;
2540 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2541 gb->buffer + (config_start_bit / 8),
2542 asclen, sync_extension);
2544 if (bits_consumed < 0)
2545 return AVERROR_INVALIDDATA;
2547 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2548 ac->m4ac.chan_config != m4ac.chan_config) {
2550 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2551 latmctx->initialized = 0;
2553 esize = (bits_consumed+7) / 8;
2555 if (avctx->extradata_size < esize) {
2556 av_free(avctx->extradata);
2557 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2558 if (!avctx->extradata)
2559 return AVERROR(ENOMEM);
2562 avctx->extradata_size = esize;
2563 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2564 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2566 skip_bits_long(gb, bits_consumed);
2568 return bits_consumed;
2571 static int read_stream_mux_config(struct LATMContext *latmctx,
2574 int ret, audio_mux_version = get_bits(gb, 1);
2576 latmctx->audio_mux_version_A = 0;
2577 if (audio_mux_version)
2578 latmctx->audio_mux_version_A = get_bits(gb, 1);
2580 if (!latmctx->audio_mux_version_A) {
2582 if (audio_mux_version)
2583 latm_get_value(gb); // taraFullness
2585 skip_bits(gb, 1); // allStreamSameTimeFraming
2586 skip_bits(gb, 6); // numSubFrames
2588 if (get_bits(gb, 4)) { // numPrograms
2589 av_log_missing_feature(latmctx->aac_ctx.avctx,
2590 "multiple programs are not supported\n", 1);
2591 return AVERROR_PATCHWELCOME;
2594 // for each program (which there is only on in DVB)
2596 // for each layer (which there is only on in DVB)
2597 if (get_bits(gb, 3)) { // numLayer
2598 av_log_missing_feature(latmctx->aac_ctx.avctx,
2599 "multiple layers are not supported\n", 1);
2600 return AVERROR_PATCHWELCOME;
2603 // for all but first stream: use_same_config = get_bits(gb, 1);
2604 if (!audio_mux_version) {
2605 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2608 int ascLen = latm_get_value(gb);
2609 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2612 skip_bits_long(gb, ascLen);
2615 latmctx->frame_length_type = get_bits(gb, 3);
2616 switch (latmctx->frame_length_type) {
2618 skip_bits(gb, 8); // latmBufferFullness
2621 latmctx->frame_length = get_bits(gb, 9);
2626 skip_bits(gb, 6); // CELP frame length table index
2630 skip_bits(gb, 1); // HVXC frame length table index
2634 if (get_bits(gb, 1)) { // other data
2635 if (audio_mux_version) {
2636 latm_get_value(gb); // other_data_bits
2640 esc = get_bits(gb, 1);
2646 if (get_bits(gb, 1)) // crc present
2647 skip_bits(gb, 8); // config_crc
2653 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2657 if (ctx->frame_length_type == 0) {
2658 int mux_slot_length = 0;
2660 tmp = get_bits(gb, 8);
2661 mux_slot_length += tmp;
2662 } while (tmp == 255);
2663 return mux_slot_length;
2664 } else if (ctx->frame_length_type == 1) {
2665 return ctx->frame_length;
2666 } else if (ctx->frame_length_type == 3 ||
2667 ctx->frame_length_type == 5 ||
2668 ctx->frame_length_type == 7) {
2669 skip_bits(gb, 2); // mux_slot_length_coded
2674 static int read_audio_mux_element(struct LATMContext *latmctx,
2678 uint8_t use_same_mux = get_bits(gb, 1);
2679 if (!use_same_mux) {
2680 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2682 } else if (!latmctx->aac_ctx.avctx->extradata) {
2683 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2684 "no decoder config found\n");
2685 return AVERROR(EAGAIN);
2687 if (latmctx->audio_mux_version_A == 0) {
2688 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2689 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2690 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2691 return AVERROR_INVALIDDATA;
2692 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2693 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2694 "frame length mismatch %d << %d\n",
2695 mux_slot_length_bytes * 8, get_bits_left(gb));
2696 return AVERROR_INVALIDDATA;
2703 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2704 int *got_frame_ptr, AVPacket *avpkt)
2706 struct LATMContext *latmctx = avctx->priv_data;
2710 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2712 // check for LOAS sync word
2713 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2714 return AVERROR_INVALIDDATA;
2716 muxlength = get_bits(&gb, 13) + 3;
2717 // not enough data, the parser should have sorted this
2718 if (muxlength > avpkt->size)
2719 return AVERROR_INVALIDDATA;
2721 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2724 if (!latmctx->initialized) {
2725 if (!avctx->extradata) {
2729 if ((err = decode_audio_specific_config(
2730 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2731 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2733 latmctx->initialized = 1;
2737 if (show_bits(&gb, 12) == 0xfff) {
2738 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2739 "ADTS header detected, probably as result of configuration "
2741 return AVERROR_INVALIDDATA;
2744 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2750 av_cold static int latm_decode_init(AVCodecContext *avctx)
2752 struct LATMContext *latmctx = avctx->priv_data;
2753 int ret = aac_decode_init(avctx);
2755 if (avctx->extradata_size > 0)
2756 latmctx->initialized = !ret;
2762 AVCodec ff_aac_decoder = {
2764 .type = AVMEDIA_TYPE_AUDIO,
2766 .priv_data_size = sizeof(AACContext),
2767 .init = aac_decode_init,
2768 .close = aac_decode_close,
2769 .decode = aac_decode_frame,
2770 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2771 .sample_fmts = (const enum AVSampleFormat[]) {
2772 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2774 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2775 .channel_layouts = aac_channel_layout,
2779 Note: This decoder filter is intended to decode LATM streams transferred
2780 in MPEG transport streams which only contain one program.
2781 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2783 AVCodec ff_aac_latm_decoder = {
2785 .type = AVMEDIA_TYPE_AUDIO,
2786 .id = CODEC_ID_AAC_LATM,
2787 .priv_data_size = sizeof(struct LATMContext),
2788 .init = latm_decode_init,
2789 .close = aac_decode_close,
2790 .decode = latm_decode_frame,
2791 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2792 .sample_fmts = (const enum AVSampleFormat[]) {
2793 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2795 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2796 .channel_layouts = aac_channel_layout,