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-libav@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 int count_channels(uint8_t (*layout)[3], int tags)
120 for (i = 0; i < tags; i++) {
121 int syn_ele = layout[i][0];
122 int pos = layout[i][2];
123 sum += (1 + (syn_ele == TYPE_CPE)) *
124 (pos != AAC_CHANNEL_OFF && pos != AAC_CHANNEL_CC);
130 * Check for the channel element in the current channel position configuration.
131 * If it exists, make sure the appropriate element is allocated and map the
132 * channel order to match the internal FFmpeg channel layout.
134 * @param che_pos current channel position configuration
135 * @param type channel element type
136 * @param id channel element id
137 * @param channels count of the number of channels in the configuration
139 * @return Returns error status. 0 - OK, !0 - error
141 static av_cold int che_configure(AACContext *ac,
142 enum ChannelPosition che_pos,
143 int type, int id, int *channels)
146 if (!ac->che[type][id]) {
147 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
148 return AVERROR(ENOMEM);
149 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
151 if (type != TYPE_CCE) {
152 if (*channels >= MAX_CHANNELS - (type == TYPE_CPE || (type == TYPE_SCE && ac->m4ac.ps == 1))) {
153 av_log(ac->avctx, AV_LOG_ERROR, "Too many channels\n");
154 return AVERROR_INVALIDDATA;
156 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
157 if (type == TYPE_CPE ||
158 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
159 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
163 if (ac->che[type][id])
164 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
165 av_freep(&ac->che[type][id]);
170 struct elem_to_channel {
171 uint64_t av_position;
174 uint8_t aac_position;
177 static int assign_pair(struct elem_to_channel e2c_vec[MAX_ELEM_ID],
178 uint8_t (*layout_map)[3], int offset, int tags, uint64_t left,
179 uint64_t right, int pos)
181 if (layout_map[offset][0] == TYPE_CPE) {
182 e2c_vec[offset] = (struct elem_to_channel) {
183 .av_position = left | right, .syn_ele = TYPE_CPE,
184 .elem_id = layout_map[offset ][1], .aac_position = pos };
187 e2c_vec[offset] = (struct elem_to_channel) {
188 .av_position = left, .syn_ele = TYPE_SCE,
189 .elem_id = layout_map[offset ][1], .aac_position = pos };
190 e2c_vec[offset + 1] = (struct elem_to_channel) {
191 .av_position = right, .syn_ele = TYPE_SCE,
192 .elem_id = layout_map[offset + 1][1], .aac_position = pos };
197 static int count_paired_channels(uint8_t (*layout_map)[3], int tags, int pos, int *current) {
198 int num_pos_channels = 0;
202 for (i = *current; i < tags; i++) {
203 if (layout_map[i][2] != pos)
205 if (layout_map[i][0] == TYPE_CPE) {
207 if (pos == AAC_CHANNEL_FRONT && !first_cpe) {
213 num_pos_channels += 2;
221 ((pos == AAC_CHANNEL_FRONT && first_cpe) || pos == AAC_CHANNEL_SIDE))
224 return num_pos_channels;
227 static uint64_t sniff_channel_order(uint8_t (*layout_map)[3], int tags)
229 int i, n, total_non_cc_elements;
230 struct elem_to_channel e2c_vec[4*MAX_ELEM_ID] = {{ 0 }};
231 int num_front_channels, num_side_channels, num_back_channels;
234 if (FF_ARRAY_ELEMS(e2c_vec) < tags)
239 count_paired_channels(layout_map, tags, AAC_CHANNEL_FRONT, &i);
240 if (num_front_channels < 0)
243 count_paired_channels(layout_map, tags, AAC_CHANNEL_SIDE, &i);
244 if (num_side_channels < 0)
247 count_paired_channels(layout_map, tags, AAC_CHANNEL_BACK, &i);
248 if (num_back_channels < 0)
252 if (num_front_channels & 1) {
253 e2c_vec[i] = (struct elem_to_channel) {
254 .av_position = AV_CH_FRONT_CENTER, .syn_ele = TYPE_SCE,
255 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_FRONT };
257 num_front_channels--;
259 if (num_front_channels >= 4) {
260 i += assign_pair(e2c_vec, layout_map, i, tags,
261 AV_CH_FRONT_LEFT_OF_CENTER,
262 AV_CH_FRONT_RIGHT_OF_CENTER,
264 num_front_channels -= 2;
266 if (num_front_channels >= 2) {
267 i += assign_pair(e2c_vec, layout_map, i, tags,
271 num_front_channels -= 2;
273 while (num_front_channels >= 2) {
274 i += assign_pair(e2c_vec, layout_map, i, tags,
278 num_front_channels -= 2;
281 if (num_side_channels >= 2) {
282 i += assign_pair(e2c_vec, layout_map, i, tags,
286 num_side_channels -= 2;
288 while (num_side_channels >= 2) {
289 i += assign_pair(e2c_vec, layout_map, i, tags,
293 num_side_channels -= 2;
296 while (num_back_channels >= 4) {
297 i += assign_pair(e2c_vec, layout_map, i, tags,
301 num_back_channels -= 2;
303 if (num_back_channels >= 2) {
304 i += assign_pair(e2c_vec, layout_map, i, tags,
308 num_back_channels -= 2;
310 if (num_back_channels) {
311 e2c_vec[i] = (struct elem_to_channel) {
312 .av_position = AV_CH_BACK_CENTER, .syn_ele = TYPE_SCE,
313 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_BACK };
318 if (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
319 e2c_vec[i] = (struct elem_to_channel) {
320 .av_position = AV_CH_LOW_FREQUENCY, .syn_ele = TYPE_LFE,
321 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
324 while (i < tags && layout_map[i][2] == AAC_CHANNEL_LFE) {
325 e2c_vec[i] = (struct elem_to_channel) {
326 .av_position = UINT64_MAX, .syn_ele = TYPE_LFE,
327 .elem_id = layout_map[i][1], .aac_position = AAC_CHANNEL_LFE };
331 // Must choose a stable sort
332 total_non_cc_elements = n = i;
335 for (i = 1; i < n; i++) {
336 if (e2c_vec[i-1].av_position > e2c_vec[i].av_position) {
337 FFSWAP(struct elem_to_channel, e2c_vec[i-1], e2c_vec[i]);
345 for (i = 0; i < total_non_cc_elements; i++) {
346 layout_map[i][0] = e2c_vec[i].syn_ele;
347 layout_map[i][1] = e2c_vec[i].elem_id;
348 layout_map[i][2] = e2c_vec[i].aac_position;
349 if (e2c_vec[i].av_position != UINT64_MAX) {
350 layout |= e2c_vec[i].av_position;
358 * Configure output channel order based on the current program configuration element.
360 * @return Returns error status. 0 - OK, !0 - error
362 static av_cold int output_configure(AACContext *ac,
363 uint8_t layout_map[MAX_ELEM_ID*4][3], int tags,
364 int channel_config, enum OCStatus oc_type)
366 AVCodecContext *avctx = ac->avctx;
367 int i, channels = 0, ret;
370 if (ac->layout_map != layout_map) {
371 memcpy(ac->layout_map, layout_map, tags * sizeof(layout_map[0]));
372 ac->layout_map_tags = tags;
375 // Try to sniff a reasonable channel order, otherwise output the
376 // channels in the order the PCE declared them.
377 if (avctx->request_channel_layout != AV_CH_LAYOUT_NATIVE)
378 layout = sniff_channel_order(layout_map, tags);
379 for (i = 0; i < tags; i++) {
380 int type = layout_map[i][0];
381 int id = layout_map[i][1];
382 int position = layout_map[i][2];
383 // Allocate or free elements depending on if they are in the
384 // current program configuration.
385 ret = che_configure(ac, position, type, id, &channels);
390 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
391 if (layout) avctx->channel_layout = layout;
392 avctx->channels = channels;
393 ac->output_configured = oc_type;
398 static void flush(AVCodecContext *avctx)
400 AACContext *ac= avctx->priv_data;
403 for (type = 3; type >= 0; type--) {
404 for (i = 0; i < MAX_ELEM_ID; i++) {
405 ChannelElement *che = ac->che[type][i];
407 for (j = 0; j <= 1; j++) {
408 memset(che->ch[j].saved, 0, sizeof(che->ch[j].saved));
416 * Set up channel positions based on a default channel configuration
417 * as specified in table 1.17.
419 * @return Returns error status. 0 - OK, !0 - error
421 static av_cold int set_default_channel_config(AVCodecContext *avctx,
422 uint8_t (*layout_map)[3],
426 if (channel_config < 1 || channel_config > 7) {
427 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
431 *tags = tags_per_config[channel_config];
432 memcpy(layout_map, aac_channel_layout_map[channel_config-1], *tags * sizeof(*layout_map));
436 static ChannelElement *get_che(AACContext *ac, int type, int elem_id)
438 // For PCE based channel configurations map the channels solely based on tags.
439 if (!ac->m4ac.chan_config) {
440 return ac->tag_che_map[type][elem_id];
442 // Allow single CPE stereo files to be signalled with mono configuration.
443 if (!ac->tags_mapped && type == TYPE_CPE && ac->m4ac.chan_config == 1) {
444 uint8_t layout_map[MAX_ELEM_ID*4][3];
447 if (set_default_channel_config(ac->avctx, layout_map, &layout_map_tags,
450 if (output_configure(ac, layout_map, layout_map_tags,
451 2, OC_TRIAL_FRAME) < 0)
454 ac->m4ac.chan_config = 2;
456 // For indexed channel configurations map the channels solely based on position.
457 switch (ac->m4ac.chan_config) {
459 if (ac->tags_mapped == 3 && type == TYPE_CPE) {
461 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][2];
464 /* Some streams incorrectly code 5.1 audio as SCE[0] CPE[0] CPE[1] SCE[1]
465 instead of SCE[0] CPE[0] CPE[1] LFE[0]. If we seem to have
466 encountered such a stream, transfer the LFE[0] element to the SCE[1]'s mapping */
467 if (ac->tags_mapped == tags_per_config[ac->m4ac.chan_config] - 1 && (type == TYPE_LFE || type == TYPE_SCE)) {
469 return ac->tag_che_map[type][elem_id] = ac->che[TYPE_LFE][0];
472 if (ac->tags_mapped == 2 && type == TYPE_CPE) {
474 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][1];
477 if (ac->tags_mapped == 2 && ac->m4ac.chan_config == 4 && type == TYPE_SCE) {
479 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][1];
483 if (ac->tags_mapped == (ac->m4ac.chan_config != 2) && type == TYPE_CPE) {
485 return ac->tag_che_map[TYPE_CPE][elem_id] = ac->che[TYPE_CPE][0];
486 } else if (ac->m4ac.chan_config == 2) {
490 if (!ac->tags_mapped && type == TYPE_SCE) {
492 return ac->tag_che_map[TYPE_SCE][elem_id] = ac->che[TYPE_SCE][0];
500 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
502 * @param type speaker type/position for these channels
504 static void decode_channel_map(uint8_t layout_map[][3],
505 enum ChannelPosition type,
506 GetBitContext *gb, int n)
509 enum RawDataBlockType syn_ele;
511 case AAC_CHANNEL_FRONT:
512 case AAC_CHANNEL_BACK:
513 case AAC_CHANNEL_SIDE:
514 syn_ele = get_bits1(gb);
520 case AAC_CHANNEL_LFE:
524 layout_map[0][0] = syn_ele;
525 layout_map[0][1] = get_bits(gb, 4);
526 layout_map[0][2] = type;
532 * Decode program configuration element; reference: table 4.2.
534 * @return Returns error status. 0 - OK, !0 - error
536 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
537 uint8_t (*layout_map)[3],
540 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
544 skip_bits(gb, 2); // object_type
546 sampling_index = get_bits(gb, 4);
547 if (m4ac->sampling_index != sampling_index)
548 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");
550 num_front = get_bits(gb, 4);
551 num_side = get_bits(gb, 4);
552 num_back = get_bits(gb, 4);
553 num_lfe = get_bits(gb, 2);
554 num_assoc_data = get_bits(gb, 3);
555 num_cc = get_bits(gb, 4);
558 skip_bits(gb, 4); // mono_mixdown_tag
560 skip_bits(gb, 4); // stereo_mixdown_tag
563 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
565 if (get_bits_left(gb) < 4 * (num_front + num_side + num_back + num_lfe + num_assoc_data + num_cc)) {
566 av_log(avctx, AV_LOG_ERROR, overread_err);
569 decode_channel_map(layout_map , AAC_CHANNEL_FRONT, gb, num_front);
571 decode_channel_map(layout_map + tags, AAC_CHANNEL_SIDE, gb, num_side);
573 decode_channel_map(layout_map + tags, AAC_CHANNEL_BACK, gb, num_back);
575 decode_channel_map(layout_map + tags, AAC_CHANNEL_LFE, gb, num_lfe);
578 skip_bits_long(gb, 4 * num_assoc_data);
580 decode_channel_map(layout_map + tags, AAC_CHANNEL_CC, gb, num_cc);
585 /* comment field, first byte is length */
586 comment_len = get_bits(gb, 8) * 8;
587 if (get_bits_left(gb) < comment_len) {
588 av_log(avctx, AV_LOG_ERROR, overread_err);
591 skip_bits_long(gb, comment_len);
596 * Decode GA "General Audio" specific configuration; reference: table 4.1.
598 * @param ac pointer to AACContext, may be null
599 * @param avctx pointer to AVCCodecContext, used for logging
601 * @return Returns error status. 0 - OK, !0 - error
603 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
605 MPEG4AudioConfig *m4ac,
608 int extension_flag, ret;
609 uint8_t layout_map[MAX_ELEM_ID*4][3];
612 if (get_bits1(gb)) { // frameLengthFlag
613 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
617 if (get_bits1(gb)) // dependsOnCoreCoder
618 skip_bits(gb, 14); // coreCoderDelay
619 extension_flag = get_bits1(gb);
621 if (m4ac->object_type == AOT_AAC_SCALABLE ||
622 m4ac->object_type == AOT_ER_AAC_SCALABLE)
623 skip_bits(gb, 3); // layerNr
625 if (channel_config == 0) {
626 skip_bits(gb, 4); // element_instance_tag
627 tags = decode_pce(avctx, m4ac, layout_map, gb);
631 if ((ret = set_default_channel_config(avctx, layout_map, &tags, channel_config)))
635 if (count_channels(layout_map, tags) > 1) {
637 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
640 if (ac && (ret = output_configure(ac, layout_map, tags,
641 channel_config, OC_GLOBAL_HDR)))
644 if (extension_flag) {
645 switch (m4ac->object_type) {
647 skip_bits(gb, 5); // numOfSubFrame
648 skip_bits(gb, 11); // layer_length
652 case AOT_ER_AAC_SCALABLE:
654 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
655 * aacScalefactorDataResilienceFlag
656 * aacSpectralDataResilienceFlag
660 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
666 * Decode audio specific configuration; reference: table 1.13.
668 * @param ac pointer to AACContext, may be null
669 * @param avctx pointer to AVCCodecContext, used for logging
670 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
671 * @param data pointer to buffer holding an audio specific config
672 * @param bit_size size of audio specific config or data in bits
673 * @param sync_extension look for an appended sync extension
675 * @return Returns error status or number of consumed bits. <0 - error
677 static int decode_audio_specific_config(AACContext *ac,
678 AVCodecContext *avctx,
679 MPEG4AudioConfig *m4ac,
680 const uint8_t *data, int bit_size,
686 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
687 for (i = 0; i < avctx->extradata_size; i++)
688 av_dlog(avctx, "%02x ", avctx->extradata[i]);
689 av_dlog(avctx, "\n");
691 init_get_bits(&gb, data, bit_size);
693 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
695 if (m4ac->sampling_index > 12) {
696 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
700 skip_bits_long(&gb, i);
702 switch (m4ac->object_type) {
706 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
710 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
711 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
715 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
716 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
717 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
719 return get_bits_count(&gb);
723 * linear congruential pseudorandom number generator
725 * @param previous_val pointer to the current state of the generator
727 * @return Returns a 32-bit pseudorandom integer
729 static av_always_inline int lcg_random(int previous_val)
731 return previous_val * 1664525 + 1013904223;
734 static av_always_inline void reset_predict_state(PredictorState *ps)
744 static void reset_all_predictors(PredictorState *ps)
747 for (i = 0; i < MAX_PREDICTORS; i++)
748 reset_predict_state(&ps[i]);
751 static int sample_rate_idx (int rate)
753 if (92017 <= rate) return 0;
754 else if (75132 <= rate) return 1;
755 else if (55426 <= rate) return 2;
756 else if (46009 <= rate) return 3;
757 else if (37566 <= rate) return 4;
758 else if (27713 <= rate) return 5;
759 else if (23004 <= rate) return 6;
760 else if (18783 <= rate) return 7;
761 else if (13856 <= rate) return 8;
762 else if (11502 <= rate) return 9;
763 else if (9391 <= rate) return 10;
767 static void reset_predictor_group(PredictorState *ps, int group_num)
770 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
771 reset_predict_state(&ps[i]);
774 #define AAC_INIT_VLC_STATIC(num, size) \
775 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
776 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
777 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
780 static av_cold int aac_decode_init(AVCodecContext *avctx)
782 AACContext *ac = avctx->priv_data;
783 float output_scale_factor;
786 ac->m4ac.sample_rate = avctx->sample_rate;
788 if (avctx->extradata_size > 0) {
789 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
791 avctx->extradata_size*8, 1) < 0)
795 uint8_t layout_map[MAX_ELEM_ID*4][3];
798 sr = sample_rate_idx(avctx->sample_rate);
799 ac->m4ac.sampling_index = sr;
800 ac->m4ac.channels = avctx->channels;
804 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
805 if (ff_mpeg4audio_channels[i] == avctx->channels)
807 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
810 ac->m4ac.chan_config = i;
812 if (ac->m4ac.chan_config) {
813 int ret = set_default_channel_config(avctx, layout_map,
814 &layout_map_tags, ac->m4ac.chan_config);
816 output_configure(ac, layout_map, layout_map_tags,
817 ac->m4ac.chan_config, OC_GLOBAL_HDR);
818 else if (avctx->err_recognition & AV_EF_EXPLODE)
819 return AVERROR_INVALIDDATA;
823 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
824 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
825 output_scale_factor = 1.0 / 32768.0;
827 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
828 output_scale_factor = 1.0;
831 AAC_INIT_VLC_STATIC( 0, 304);
832 AAC_INIT_VLC_STATIC( 1, 270);
833 AAC_INIT_VLC_STATIC( 2, 550);
834 AAC_INIT_VLC_STATIC( 3, 300);
835 AAC_INIT_VLC_STATIC( 4, 328);
836 AAC_INIT_VLC_STATIC( 5, 294);
837 AAC_INIT_VLC_STATIC( 6, 306);
838 AAC_INIT_VLC_STATIC( 7, 268);
839 AAC_INIT_VLC_STATIC( 8, 510);
840 AAC_INIT_VLC_STATIC( 9, 366);
841 AAC_INIT_VLC_STATIC(10, 462);
845 ff_dsputil_init(&ac->dsp, avctx);
846 ff_fmt_convert_init(&ac->fmt_conv, avctx);
848 ac->random_state = 0x1f2e3d4c;
852 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
853 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
854 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
857 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
858 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
859 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
860 // window initialization
861 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
862 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
863 ff_init_ff_sine_windows(10);
864 ff_init_ff_sine_windows( 7);
868 avcodec_get_frame_defaults(&ac->frame);
869 avctx->coded_frame = &ac->frame;
875 * Skip data_stream_element; reference: table 4.10.
877 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
879 int byte_align = get_bits1(gb);
880 int count = get_bits(gb, 8);
882 count += get_bits(gb, 8);
886 if (get_bits_left(gb) < 8 * count) {
887 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
890 skip_bits_long(gb, 8 * count);
894 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
899 ics->predictor_reset_group = get_bits(gb, 5);
900 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
901 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
905 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
906 ics->prediction_used[sfb] = get_bits1(gb);
912 * Decode Long Term Prediction data; reference: table 4.xx.
914 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
915 GetBitContext *gb, uint8_t max_sfb)
919 ltp->lag = get_bits(gb, 11);
920 ltp->coef = ltp_coef[get_bits(gb, 3)];
921 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
922 ltp->used[sfb] = get_bits1(gb);
926 * Decode Individual Channel Stream info; reference: table 4.6.
928 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
932 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
933 return AVERROR_INVALIDDATA;
935 ics->window_sequence[1] = ics->window_sequence[0];
936 ics->window_sequence[0] = get_bits(gb, 2);
937 ics->use_kb_window[1] = ics->use_kb_window[0];
938 ics->use_kb_window[0] = get_bits1(gb);
939 ics->num_window_groups = 1;
940 ics->group_len[0] = 1;
941 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
943 ics->max_sfb = get_bits(gb, 4);
944 for (i = 0; i < 7; i++) {
946 ics->group_len[ics->num_window_groups - 1]++;
948 ics->num_window_groups++;
949 ics->group_len[ics->num_window_groups - 1] = 1;
952 ics->num_windows = 8;
953 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
954 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
955 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
956 ics->predictor_present = 0;
958 ics->max_sfb = get_bits(gb, 6);
959 ics->num_windows = 1;
960 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
961 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
962 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
963 ics->predictor_present = get_bits1(gb);
964 ics->predictor_reset_group = 0;
965 if (ics->predictor_present) {
966 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
967 if (decode_prediction(ac, ics, gb)) {
970 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
971 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
974 if ((ics->ltp.present = get_bits(gb, 1)))
975 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
980 if (ics->max_sfb > ics->num_swb) {
981 av_log(ac->avctx, AV_LOG_ERROR,
982 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
983 ics->max_sfb, ics->num_swb);
990 return AVERROR_INVALIDDATA;
994 * Decode band types (section_data payload); reference: table 4.46.
996 * @param band_type array of the used band type
997 * @param band_type_run_end array of the last scalefactor band of a band type run
999 * @return Returns error status. 0 - OK, !0 - error
1001 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
1002 int band_type_run_end[120], GetBitContext *gb,
1003 IndividualChannelStream *ics)
1006 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
1007 for (g = 0; g < ics->num_window_groups; g++) {
1009 while (k < ics->max_sfb) {
1010 uint8_t sect_end = k;
1012 int sect_band_type = get_bits(gb, 4);
1013 if (sect_band_type == 12) {
1014 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
1018 sect_len_incr = get_bits(gb, bits);
1019 sect_end += sect_len_incr;
1020 if (get_bits_left(gb) < 0) {
1021 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
1024 if (sect_end > ics->max_sfb) {
1025 av_log(ac->avctx, AV_LOG_ERROR,
1026 "Number of bands (%d) exceeds limit (%d).\n",
1027 sect_end, ics->max_sfb);
1030 } while (sect_len_incr == (1 << bits) - 1);
1031 for (; k < sect_end; k++) {
1032 band_type [idx] = sect_band_type;
1033 band_type_run_end[idx++] = sect_end;
1041 * Decode scalefactors; reference: table 4.47.
1043 * @param global_gain first scalefactor value as scalefactors are differentially coded
1044 * @param band_type array of the used band type
1045 * @param band_type_run_end array of the last scalefactor band of a band type run
1046 * @param sf array of scalefactors or intensity stereo positions
1048 * @return Returns error status. 0 - OK, !0 - error
1050 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
1051 unsigned int global_gain,
1052 IndividualChannelStream *ics,
1053 enum BandType band_type[120],
1054 int band_type_run_end[120])
1057 int offset[3] = { global_gain, global_gain - 90, 0 };
1060 for (g = 0; g < ics->num_window_groups; g++) {
1061 for (i = 0; i < ics->max_sfb;) {
1062 int run_end = band_type_run_end[idx];
1063 if (band_type[idx] == ZERO_BT) {
1064 for (; i < run_end; i++, idx++)
1066 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
1067 for (; i < run_end; i++, idx++) {
1068 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1069 clipped_offset = av_clip(offset[2], -155, 100);
1070 if (offset[2] != clipped_offset) {
1071 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
1072 "position clipped (%d -> %d).\nIf you heard an "
1073 "audible artifact, there may be a bug in the "
1074 "decoder. ", offset[2], clipped_offset);
1076 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
1078 } else if (band_type[idx] == NOISE_BT) {
1079 for (; i < run_end; i++, idx++) {
1080 if (noise_flag-- > 0)
1081 offset[1] += get_bits(gb, 9) - 256;
1083 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1084 clipped_offset = av_clip(offset[1], -100, 155);
1085 if (offset[1] != clipped_offset) {
1086 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
1087 "(%d -> %d).\nIf you heard an audible "
1088 "artifact, there may be a bug in the decoder. ",
1089 offset[1], clipped_offset);
1091 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
1094 for (; i < run_end; i++, idx++) {
1095 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1096 if (offset[0] > 255U) {
1097 av_log(ac->avctx, AV_LOG_ERROR,
1098 "Scalefactor (%d) out of range.\n", offset[0]);
1101 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
1110 * Decode pulse data; reference: table 4.7.
1112 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
1113 const uint16_t *swb_offset, int num_swb)
1116 pulse->num_pulse = get_bits(gb, 2) + 1;
1117 pulse_swb = get_bits(gb, 6);
1118 if (pulse_swb >= num_swb)
1120 pulse->pos[0] = swb_offset[pulse_swb];
1121 pulse->pos[0] += get_bits(gb, 5);
1122 if (pulse->pos[0] > 1023)
1124 pulse->amp[0] = get_bits(gb, 4);
1125 for (i = 1; i < pulse->num_pulse; i++) {
1126 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
1127 if (pulse->pos[i] > 1023)
1129 pulse->amp[i] = get_bits(gb, 4);
1135 * Decode Temporal Noise Shaping data; reference: table 4.48.
1137 * @return Returns error status. 0 - OK, !0 - error
1139 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
1140 GetBitContext *gb, const IndividualChannelStream *ics)
1142 int w, filt, i, coef_len, coef_res, coef_compress;
1143 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
1144 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
1145 for (w = 0; w < ics->num_windows; w++) {
1146 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
1147 coef_res = get_bits1(gb);
1149 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1151 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
1153 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
1154 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
1155 tns->order[w][filt], tns_max_order);
1156 tns->order[w][filt] = 0;
1159 if (tns->order[w][filt]) {
1160 tns->direction[w][filt] = get_bits1(gb);
1161 coef_compress = get_bits1(gb);
1162 coef_len = coef_res + 3 - coef_compress;
1163 tmp2_idx = 2 * coef_compress + coef_res;
1165 for (i = 0; i < tns->order[w][filt]; i++)
1166 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
1175 * Decode Mid/Side data; reference: table 4.54.
1177 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1178 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1179 * [3] reserved for scalable AAC
1181 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
1185 if (ms_present == 1) {
1186 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
1187 cpe->ms_mask[idx] = get_bits1(gb);
1188 } else if (ms_present == 2) {
1189 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
1194 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
1198 *dst++ = v[idx & 15] * s;
1199 *dst++ = v[idx>>4 & 15] * s;
1205 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1209 *dst++ = v[idx & 3] * s;
1210 *dst++ = v[idx>>2 & 3] * s;
1211 *dst++ = v[idx>>4 & 3] * s;
1212 *dst++ = v[idx>>6 & 3] * s;
1218 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1219 unsigned sign, const float *scale)
1221 union av_intfloat32 s0, s1;
1223 s0.f = s1.f = *scale;
1224 s0.i ^= sign >> 1 << 31;
1227 *dst++ = v[idx & 15] * s0.f;
1228 *dst++ = v[idx>>4 & 15] * s1.f;
1235 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1236 unsigned sign, const float *scale)
1238 unsigned nz = idx >> 12;
1239 union av_intfloat32 s = { .f = *scale };
1240 union av_intfloat32 t;
1242 t.i = s.i ^ (sign & 1U<<31);
1243 *dst++ = v[idx & 3] * t.f;
1245 sign <<= nz & 1; nz >>= 1;
1246 t.i = s.i ^ (sign & 1U<<31);
1247 *dst++ = v[idx>>2 & 3] * t.f;
1249 sign <<= nz & 1; nz >>= 1;
1250 t.i = s.i ^ (sign & 1U<<31);
1251 *dst++ = v[idx>>4 & 3] * t.f;
1253 sign <<= nz & 1; nz >>= 1;
1254 t.i = s.i ^ (sign & 1U<<31);
1255 *dst++ = v[idx>>6 & 3] * t.f;
1262 * Decode spectral data; reference: table 4.50.
1263 * Dequantize and scale spectral data; reference: 4.6.3.3.
1265 * @param coef array of dequantized, scaled spectral data
1266 * @param sf array of scalefactors or intensity stereo positions
1267 * @param pulse_present set if pulses are present
1268 * @param pulse pointer to pulse data struct
1269 * @param band_type array of the used band type
1271 * @return Returns error status. 0 - OK, !0 - error
1273 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1274 GetBitContext *gb, const float sf[120],
1275 int pulse_present, const Pulse *pulse,
1276 const IndividualChannelStream *ics,
1277 enum BandType band_type[120])
1279 int i, k, g, idx = 0;
1280 const int c = 1024 / ics->num_windows;
1281 const uint16_t *offsets = ics->swb_offset;
1282 float *coef_base = coef;
1284 for (g = 0; g < ics->num_windows; g++)
1285 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1287 for (g = 0; g < ics->num_window_groups; g++) {
1288 unsigned g_len = ics->group_len[g];
1290 for (i = 0; i < ics->max_sfb; i++, idx++) {
1291 const unsigned cbt_m1 = band_type[idx] - 1;
1292 float *cfo = coef + offsets[i];
1293 int off_len = offsets[i + 1] - offsets[i];
1296 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1297 for (group = 0; group < g_len; group++, cfo+=128) {
1298 memset(cfo, 0, off_len * sizeof(float));
1300 } else if (cbt_m1 == NOISE_BT - 1) {
1301 for (group = 0; group < g_len; group++, cfo+=128) {
1305 for (k = 0; k < off_len; k++) {
1306 ac->random_state = lcg_random(ac->random_state);
1307 cfo[k] = ac->random_state;
1310 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1311 scale = sf[idx] / sqrtf(band_energy);
1312 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1315 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1316 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1317 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1318 OPEN_READER(re, gb);
1320 switch (cbt_m1 >> 1) {
1322 for (group = 0; group < g_len; group++, cfo+=128) {
1330 UPDATE_CACHE(re, gb);
1331 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1332 cb_idx = cb_vector_idx[code];
1333 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1339 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 nnz = cb_idx >> 8 & 15;
1353 bits = nnz ? GET_CACHE(re, gb) : 0;
1354 LAST_SKIP_BITS(re, gb, nnz);
1355 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1361 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 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1379 for (group = 0; group < g_len; group++, cfo+=128) {
1389 UPDATE_CACHE(re, gb);
1390 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1391 cb_idx = cb_vector_idx[code];
1392 nnz = cb_idx >> 8 & 15;
1393 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1394 LAST_SKIP_BITS(re, gb, nnz);
1395 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1401 for (group = 0; group < g_len; group++, cfo+=128) {
1403 uint32_t *icf = (uint32_t *) cf;
1413 UPDATE_CACHE(re, gb);
1414 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1422 cb_idx = cb_vector_idx[code];
1425 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1426 LAST_SKIP_BITS(re, gb, nnz);
1428 for (j = 0; j < 2; j++) {
1432 /* The total length of escape_sequence must be < 22 bits according
1433 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1434 UPDATE_CACHE(re, gb);
1435 b = GET_CACHE(re, gb);
1436 b = 31 - av_log2(~b);
1439 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1443 SKIP_BITS(re, gb, b + 1);
1445 n = (1 << b) + SHOW_UBITS(re, gb, b);
1446 LAST_SKIP_BITS(re, gb, b);
1447 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1450 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1451 *icf++ = (bits & 1U<<31) | v;
1458 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1462 CLOSE_READER(re, gb);
1468 if (pulse_present) {
1470 for (i = 0; i < pulse->num_pulse; i++) {
1471 float co = coef_base[ pulse->pos[i] ];
1472 while (offsets[idx + 1] <= pulse->pos[i])
1474 if (band_type[idx] != NOISE_BT && sf[idx]) {
1475 float ico = -pulse->amp[i];
1478 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1480 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1487 static av_always_inline float flt16_round(float pf)
1489 union av_intfloat32 tmp;
1491 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1495 static av_always_inline float flt16_even(float pf)
1497 union av_intfloat32 tmp;
1499 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1503 static av_always_inline float flt16_trunc(float pf)
1505 union av_intfloat32 pun;
1507 pun.i &= 0xFFFF0000U;
1511 static av_always_inline void predict(PredictorState *ps, float *coef,
1514 const float a = 0.953125; // 61.0 / 64
1515 const float alpha = 0.90625; // 29.0 / 32
1519 float r0 = ps->r0, r1 = ps->r1;
1520 float cor0 = ps->cor0, cor1 = ps->cor1;
1521 float var0 = ps->var0, var1 = ps->var1;
1523 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1524 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1526 pv = flt16_round(k1 * r0 + k2 * r1);
1533 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1534 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1535 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1536 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1538 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1539 ps->r0 = flt16_trunc(a * e0);
1543 * Apply AAC-Main style frequency domain prediction.
1545 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1549 if (!sce->ics.predictor_initialized) {
1550 reset_all_predictors(sce->predictor_state);
1551 sce->ics.predictor_initialized = 1;
1554 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1555 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1556 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1557 predict(&sce->predictor_state[k], &sce->coeffs[k],
1558 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1561 if (sce->ics.predictor_reset_group)
1562 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1564 reset_all_predictors(sce->predictor_state);
1568 * Decode an individual_channel_stream payload; reference: table 4.44.
1570 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1571 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1573 * @return Returns error status. 0 - OK, !0 - error
1575 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1576 GetBitContext *gb, int common_window, int scale_flag)
1579 TemporalNoiseShaping *tns = &sce->tns;
1580 IndividualChannelStream *ics = &sce->ics;
1581 float *out = sce->coeffs;
1582 int global_gain, pulse_present = 0;
1584 /* This assignment is to silence a GCC warning about the variable being used
1585 * uninitialized when in fact it always is.
1587 pulse.num_pulse = 0;
1589 global_gain = get_bits(gb, 8);
1591 if (!common_window && !scale_flag) {
1592 if (decode_ics_info(ac, ics, gb) < 0)
1593 return AVERROR_INVALIDDATA;
1596 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1598 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1603 if ((pulse_present = get_bits1(gb))) {
1604 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1605 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1608 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1609 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1613 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1615 if (get_bits1(gb)) {
1616 av_log_missing_feature(ac->avctx, "SSR", 1);
1621 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1624 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1625 apply_prediction(ac, sce);
1631 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1633 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1635 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1636 float *ch0 = cpe->ch[0].coeffs;
1637 float *ch1 = cpe->ch[1].coeffs;
1638 int g, i, group, idx = 0;
1639 const uint16_t *offsets = ics->swb_offset;
1640 for (g = 0; g < ics->num_window_groups; g++) {
1641 for (i = 0; i < ics->max_sfb; i++, idx++) {
1642 if (cpe->ms_mask[idx] &&
1643 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1644 for (group = 0; group < ics->group_len[g]; group++) {
1645 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1646 ch1 + group * 128 + offsets[i],
1647 offsets[i+1] - offsets[i]);
1651 ch0 += ics->group_len[g] * 128;
1652 ch1 += ics->group_len[g] * 128;
1657 * intensity stereo decoding; reference: 4.6.8.2.3
1659 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1660 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1661 * [3] reserved for scalable AAC
1663 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1665 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1666 SingleChannelElement *sce1 = &cpe->ch[1];
1667 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1668 const uint16_t *offsets = ics->swb_offset;
1669 int g, group, i, idx = 0;
1672 for (g = 0; g < ics->num_window_groups; g++) {
1673 for (i = 0; i < ics->max_sfb;) {
1674 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1675 const int bt_run_end = sce1->band_type_run_end[idx];
1676 for (; i < bt_run_end; i++, idx++) {
1677 c = -1 + 2 * (sce1->band_type[idx] - 14);
1679 c *= 1 - 2 * cpe->ms_mask[idx];
1680 scale = c * sce1->sf[idx];
1681 for (group = 0; group < ics->group_len[g]; group++)
1682 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1683 coef0 + group * 128 + offsets[i],
1685 offsets[i + 1] - offsets[i]);
1688 int bt_run_end = sce1->band_type_run_end[idx];
1689 idx += bt_run_end - i;
1693 coef0 += ics->group_len[g] * 128;
1694 coef1 += ics->group_len[g] * 128;
1699 * Decode a channel_pair_element; reference: table 4.4.
1701 * @return Returns error status. 0 - OK, !0 - error
1703 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1705 int i, ret, common_window, ms_present = 0;
1707 common_window = get_bits1(gb);
1708 if (common_window) {
1709 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1710 return AVERROR_INVALIDDATA;
1711 i = cpe->ch[1].ics.use_kb_window[0];
1712 cpe->ch[1].ics = cpe->ch[0].ics;
1713 cpe->ch[1].ics.use_kb_window[1] = i;
1714 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1715 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1716 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1717 ms_present = get_bits(gb, 2);
1718 if (ms_present == 3) {
1719 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1721 } else if (ms_present)
1722 decode_mid_side_stereo(cpe, gb, ms_present);
1724 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1726 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1729 if (common_window) {
1731 apply_mid_side_stereo(ac, cpe);
1732 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1733 apply_prediction(ac, &cpe->ch[0]);
1734 apply_prediction(ac, &cpe->ch[1]);
1738 apply_intensity_stereo(ac, cpe, ms_present);
1742 static const float cce_scale[] = {
1743 1.09050773266525765921, //2^(1/8)
1744 1.18920711500272106672, //2^(1/4)
1750 * Decode coupling_channel_element; reference: table 4.8.
1752 * @return Returns error status. 0 - OK, !0 - error
1754 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1760 SingleChannelElement *sce = &che->ch[0];
1761 ChannelCoupling *coup = &che->coup;
1763 coup->coupling_point = 2 * get_bits1(gb);
1764 coup->num_coupled = get_bits(gb, 3);
1765 for (c = 0; c <= coup->num_coupled; c++) {
1767 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1768 coup->id_select[c] = get_bits(gb, 4);
1769 if (coup->type[c] == TYPE_CPE) {
1770 coup->ch_select[c] = get_bits(gb, 2);
1771 if (coup->ch_select[c] == 3)
1774 coup->ch_select[c] = 2;
1776 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1778 sign = get_bits(gb, 1);
1779 scale = cce_scale[get_bits(gb, 2)];
1781 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1784 for (c = 0; c < num_gain; c++) {
1788 float gain_cache = 1.;
1790 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1791 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1792 gain_cache = powf(scale, -gain);
1794 if (coup->coupling_point == AFTER_IMDCT) {
1795 coup->gain[c][0] = gain_cache;
1797 for (g = 0; g < sce->ics.num_window_groups; g++) {
1798 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1799 if (sce->band_type[idx] != ZERO_BT) {
1801 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1809 gain_cache = powf(scale, -t) * s;
1812 coup->gain[c][idx] = gain_cache;
1822 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1824 * @return Returns number of bytes consumed.
1826 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1830 int num_excl_chan = 0;
1833 for (i = 0; i < 7; i++)
1834 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1835 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1837 return num_excl_chan / 7;
1841 * Decode dynamic range information; reference: table 4.52.
1843 * @param cnt length of TYPE_FIL syntactic element in bytes
1845 * @return Returns number of bytes consumed.
1847 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1848 GetBitContext *gb, int cnt)
1851 int drc_num_bands = 1;
1854 /* pce_tag_present? */
1855 if (get_bits1(gb)) {
1856 che_drc->pce_instance_tag = get_bits(gb, 4);
1857 skip_bits(gb, 4); // tag_reserved_bits
1861 /* excluded_chns_present? */
1862 if (get_bits1(gb)) {
1863 n += decode_drc_channel_exclusions(che_drc, gb);
1866 /* drc_bands_present? */
1867 if (get_bits1(gb)) {
1868 che_drc->band_incr = get_bits(gb, 4);
1869 che_drc->interpolation_scheme = get_bits(gb, 4);
1871 drc_num_bands += che_drc->band_incr;
1872 for (i = 0; i < drc_num_bands; i++) {
1873 che_drc->band_top[i] = get_bits(gb, 8);
1878 /* prog_ref_level_present? */
1879 if (get_bits1(gb)) {
1880 che_drc->prog_ref_level = get_bits(gb, 7);
1881 skip_bits1(gb); // prog_ref_level_reserved_bits
1885 for (i = 0; i < drc_num_bands; i++) {
1886 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1887 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1895 * Decode extension data (incomplete); reference: table 4.51.
1897 * @param cnt length of TYPE_FIL syntactic element in bytes
1899 * @return Returns number of bytes consumed
1901 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1902 ChannelElement *che, enum RawDataBlockType elem_type)
1906 switch (get_bits(gb, 4)) { // extension type
1907 case EXT_SBR_DATA_CRC:
1911 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1913 } else if (!ac->m4ac.sbr) {
1914 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1915 skip_bits_long(gb, 8 * cnt - 4);
1917 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1918 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1919 skip_bits_long(gb, 8 * cnt - 4);
1921 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1924 output_configure(ac, ac->layout_map, ac->layout_map_tags,
1925 ac->m4ac.chan_config, ac->output_configured);
1929 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1931 case EXT_DYNAMIC_RANGE:
1932 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1936 case EXT_DATA_ELEMENT:
1938 skip_bits_long(gb, 8 * cnt - 4);
1945 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1947 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1948 * @param coef spectral coefficients
1950 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1951 IndividualChannelStream *ics, int decode)
1953 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1955 int bottom, top, order, start, end, size, inc;
1956 float lpc[TNS_MAX_ORDER];
1957 float tmp[TNS_MAX_ORDER];
1959 for (w = 0; w < ics->num_windows; w++) {
1960 bottom = ics->num_swb;
1961 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1963 bottom = FFMAX(0, top - tns->length[w][filt]);
1964 order = tns->order[w][filt];
1969 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1971 start = ics->swb_offset[FFMIN(bottom, mmm)];
1972 end = ics->swb_offset[FFMIN( top, mmm)];
1973 if ((size = end - start) <= 0)
1975 if (tns->direction[w][filt]) {
1985 for (m = 0; m < size; m++, start += inc)
1986 for (i = 1; i <= FFMIN(m, order); i++)
1987 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1990 for (m = 0; m < size; m++, start += inc) {
1991 tmp[0] = coef[start];
1992 for (i = 1; i <= FFMIN(m, order); i++)
1993 coef[start] += tmp[i] * lpc[i - 1];
1994 for (i = order; i > 0; i--)
1995 tmp[i] = tmp[i - 1];
2003 * Apply windowing and MDCT to obtain the spectral
2004 * coefficient from the predicted sample by LTP.
2006 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
2007 float *in, IndividualChannelStream *ics)
2009 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2010 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2011 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2012 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2014 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
2015 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
2017 memset(in, 0, 448 * sizeof(float));
2018 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
2020 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
2021 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
2023 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
2024 memset(in + 1024 + 576, 0, 448 * sizeof(float));
2026 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
2030 * Apply the long term prediction
2032 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
2034 const LongTermPrediction *ltp = &sce->ics.ltp;
2035 const uint16_t *offsets = sce->ics.swb_offset;
2038 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
2039 float *predTime = sce->ret;
2040 float *predFreq = ac->buf_mdct;
2041 int16_t num_samples = 2048;
2043 if (ltp->lag < 1024)
2044 num_samples = ltp->lag + 1024;
2045 for (i = 0; i < num_samples; i++)
2046 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
2047 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
2049 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
2051 if (sce->tns.present)
2052 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
2054 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
2056 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
2057 sce->coeffs[i] += predFreq[i];
2062 * Update the LTP buffer for next frame
2064 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
2066 IndividualChannelStream *ics = &sce->ics;
2067 float *saved = sce->saved;
2068 float *saved_ltp = sce->coeffs;
2069 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2070 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2073 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2074 memcpy(saved_ltp, saved, 512 * sizeof(float));
2075 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2076 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2077 for (i = 0; i < 64; i++)
2078 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2079 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2080 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
2081 memset(saved_ltp + 576, 0, 448 * sizeof(float));
2082 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
2083 for (i = 0; i < 64; i++)
2084 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
2085 } else { // LONG_STOP or ONLY_LONG
2086 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
2087 for (i = 0; i < 512; i++)
2088 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
2091 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
2092 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
2093 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
2097 * Conduct IMDCT and windowing.
2099 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
2101 IndividualChannelStream *ics = &sce->ics;
2102 float *in = sce->coeffs;
2103 float *out = sce->ret;
2104 float *saved = sce->saved;
2105 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
2106 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
2107 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
2108 float *buf = ac->buf_mdct;
2109 float *temp = ac->temp;
2113 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2114 for (i = 0; i < 1024; i += 128)
2115 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
2117 ac->mdct.imdct_half(&ac->mdct, buf, in);
2119 /* window overlapping
2120 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
2121 * and long to short transitions are considered to be short to short
2122 * transitions. This leaves just two cases (long to long and short to short)
2123 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
2125 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
2126 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
2127 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
2129 memcpy( out, saved, 448 * sizeof(float));
2131 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2132 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
2133 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
2134 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
2135 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
2136 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
2137 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
2139 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
2140 memcpy( out + 576, buf + 64, 448 * sizeof(float));
2145 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
2146 memcpy( saved, temp + 64, 64 * sizeof(float));
2147 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
2148 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
2149 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
2150 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2151 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
2152 memcpy( saved, buf + 512, 448 * sizeof(float));
2153 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
2154 } else { // LONG_STOP or ONLY_LONG
2155 memcpy( saved, buf + 512, 512 * sizeof(float));
2160 * Apply dependent channel coupling (applied before IMDCT).
2162 * @param index index into coupling gain array
2164 static void apply_dependent_coupling(AACContext *ac,
2165 SingleChannelElement *target,
2166 ChannelElement *cce, int index)
2168 IndividualChannelStream *ics = &cce->ch[0].ics;
2169 const uint16_t *offsets = ics->swb_offset;
2170 float *dest = target->coeffs;
2171 const float *src = cce->ch[0].coeffs;
2172 int g, i, group, k, idx = 0;
2173 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2174 av_log(ac->avctx, AV_LOG_ERROR,
2175 "Dependent coupling is not supported together with LTP\n");
2178 for (g = 0; g < ics->num_window_groups; g++) {
2179 for (i = 0; i < ics->max_sfb; i++, idx++) {
2180 if (cce->ch[0].band_type[idx] != ZERO_BT) {
2181 const float gain = cce->coup.gain[index][idx];
2182 for (group = 0; group < ics->group_len[g]; group++) {
2183 for (k = offsets[i]; k < offsets[i + 1]; k++) {
2185 dest[group * 128 + k] += gain * src[group * 128 + k];
2190 dest += ics->group_len[g] * 128;
2191 src += ics->group_len[g] * 128;
2196 * Apply independent channel coupling (applied after IMDCT).
2198 * @param index index into coupling gain array
2200 static void apply_independent_coupling(AACContext *ac,
2201 SingleChannelElement *target,
2202 ChannelElement *cce, int index)
2205 const float gain = cce->coup.gain[index][0];
2206 const float *src = cce->ch[0].ret;
2207 float *dest = target->ret;
2208 const int len = 1024 << (ac->m4ac.sbr == 1);
2210 for (i = 0; i < len; i++)
2211 dest[i] += gain * src[i];
2215 * channel coupling transformation interface
2217 * @param apply_coupling_method pointer to (in)dependent coupling function
2219 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2220 enum RawDataBlockType type, int elem_id,
2221 enum CouplingPoint coupling_point,
2222 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2226 for (i = 0; i < MAX_ELEM_ID; i++) {
2227 ChannelElement *cce = ac->che[TYPE_CCE][i];
2230 if (cce && cce->coup.coupling_point == coupling_point) {
2231 ChannelCoupling *coup = &cce->coup;
2233 for (c = 0; c <= coup->num_coupled; c++) {
2234 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2235 if (coup->ch_select[c] != 1) {
2236 apply_coupling_method(ac, &cc->ch[0], cce, index);
2237 if (coup->ch_select[c] != 0)
2240 if (coup->ch_select[c] != 2)
2241 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2243 index += 1 + (coup->ch_select[c] == 3);
2250 * Convert spectral data to float samples, applying all supported tools as appropriate.
2252 static void spectral_to_sample(AACContext *ac)
2255 for (type = 3; type >= 0; type--) {
2256 for (i = 0; i < MAX_ELEM_ID; i++) {
2257 ChannelElement *che = ac->che[type][i];
2259 if (type <= TYPE_CPE)
2260 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2261 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2262 if (che->ch[0].ics.predictor_present) {
2263 if (che->ch[0].ics.ltp.present)
2264 apply_ltp(ac, &che->ch[0]);
2265 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2266 apply_ltp(ac, &che->ch[1]);
2269 if (che->ch[0].tns.present)
2270 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2271 if (che->ch[1].tns.present)
2272 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2273 if (type <= TYPE_CPE)
2274 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2275 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2276 imdct_and_windowing(ac, &che->ch[0]);
2277 if (ac->m4ac.object_type == AOT_AAC_LTP)
2278 update_ltp(ac, &che->ch[0]);
2279 if (type == TYPE_CPE) {
2280 imdct_and_windowing(ac, &che->ch[1]);
2281 if (ac->m4ac.object_type == AOT_AAC_LTP)
2282 update_ltp(ac, &che->ch[1]);
2284 if (ac->m4ac.sbr > 0) {
2285 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2288 if (type <= TYPE_CCE)
2289 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2295 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2298 AACADTSHeaderInfo hdr_info;
2299 uint8_t layout_map[MAX_ELEM_ID*4][3];
2300 int layout_map_tags;
2302 size = avpriv_aac_parse_header(gb, &hdr_info);
2304 if (hdr_info.chan_config) {
2305 ac->m4ac.chan_config = hdr_info.chan_config;
2306 if (set_default_channel_config(ac->avctx, layout_map,
2307 &layout_map_tags, hdr_info.chan_config))
2309 if (output_configure(ac, layout_map, layout_map_tags,
2310 hdr_info.chan_config,
2311 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2313 } else if (ac->output_configured != OC_LOCKED) {
2314 ac->m4ac.chan_config = 0;
2315 ac->output_configured = OC_NONE;
2317 if (ac->output_configured != OC_LOCKED) {
2320 ac->m4ac.sample_rate = hdr_info.sample_rate;
2321 ac->m4ac.sampling_index = hdr_info.sampling_index;
2322 ac->m4ac.object_type = hdr_info.object_type;
2324 if (!ac->avctx->sample_rate)
2325 ac->avctx->sample_rate = hdr_info.sample_rate;
2326 if (!ac->warned_num_aac_frames && hdr_info.num_aac_frames != 1) {
2327 // This is 2 for "VLB " audio in NSV files.
2328 // See samples/nsv/vlb_audio.
2329 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2330 ac->warned_num_aac_frames = 1;
2332 if (!hdr_info.crc_absent)
2338 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2339 int *got_frame_ptr, GetBitContext *gb)
2341 AACContext *ac = avctx->priv_data;
2342 ChannelElement *che = NULL, *che_prev = NULL;
2343 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2345 int samples = 0, multiplier, audio_found = 0;
2347 if (show_bits(gb, 12) == 0xfff) {
2348 if (parse_adts_frame_header(ac, gb) < 0) {
2349 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2352 if (ac->m4ac.sampling_index > 12) {
2353 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2358 ac->tags_mapped = 0;
2360 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2361 elem_id = get_bits(gb, 4);
2363 if (elem_type < TYPE_DSE) {
2364 if (!(che=get_che(ac, elem_type, elem_id))) {
2365 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2366 elem_type, elem_id);
2372 switch (elem_type) {
2375 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2380 err = decode_cpe(ac, gb, che);
2385 err = decode_cce(ac, gb, che);
2389 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2394 err = skip_data_stream_element(ac, gb);
2398 uint8_t layout_map[MAX_ELEM_ID*4][3];
2400 tags = decode_pce(avctx, &ac->m4ac, layout_map, gb);
2405 if (ac->output_configured > OC_TRIAL_PCE)
2406 av_log(avctx, AV_LOG_INFO,
2407 "Evaluating a further program_config_element.\n");
2408 err = output_configure(ac, layout_map, tags, 0, OC_TRIAL_PCE);
2410 ac->m4ac.chan_config = 0;
2416 elem_id += get_bits(gb, 8) - 1;
2417 if (get_bits_left(gb) < 8 * elem_id) {
2418 av_log(avctx, AV_LOG_ERROR, overread_err);
2422 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2423 err = 0; /* FIXME */
2427 err = -1; /* should not happen, but keeps compiler happy */
2432 elem_type_prev = elem_type;
2437 if (get_bits_left(gb) < 3) {
2438 av_log(avctx, AV_LOG_ERROR, overread_err);
2443 spectral_to_sample(ac);
2445 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2446 samples <<= multiplier;
2447 if (ac->output_configured < OC_LOCKED) {
2448 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2449 avctx->frame_size = samples;
2453 /* get output buffer */
2454 ac->frame.nb_samples = samples;
2455 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2456 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2460 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2461 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2462 (const float **)ac->output_data,
2463 samples, avctx->channels);
2465 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2466 (const float **)ac->output_data,
2467 samples, avctx->channels);
2469 *(AVFrame *)data = ac->frame;
2471 *got_frame_ptr = !!samples;
2473 if (ac->output_configured && audio_found)
2474 ac->output_configured = OC_LOCKED;
2479 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2480 int *got_frame_ptr, AVPacket *avpkt)
2482 AACContext *ac = avctx->priv_data;
2483 const uint8_t *buf = avpkt->data;
2484 int buf_size = avpkt->size;
2489 int new_extradata_size;
2490 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2491 AV_PKT_DATA_NEW_EXTRADATA,
2492 &new_extradata_size);
2494 if (new_extradata) {
2495 av_free(avctx->extradata);
2496 avctx->extradata = av_mallocz(new_extradata_size +
2497 FF_INPUT_BUFFER_PADDING_SIZE);
2498 if (!avctx->extradata)
2499 return AVERROR(ENOMEM);
2500 avctx->extradata_size = new_extradata_size;
2501 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2502 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2504 avctx->extradata_size*8, 1) < 0)
2505 return AVERROR_INVALIDDATA;
2508 init_get_bits(&gb, buf, buf_size * 8);
2510 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2513 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2514 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2515 if (buf[buf_offset])
2518 return buf_size > buf_offset ? buf_consumed : buf_size;
2521 static av_cold int aac_decode_close(AVCodecContext *avctx)
2523 AACContext *ac = avctx->priv_data;
2526 for (i = 0; i < MAX_ELEM_ID; i++) {
2527 for (type = 0; type < 4; type++) {
2528 if (ac->che[type][i])
2529 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2530 av_freep(&ac->che[type][i]);
2534 ff_mdct_end(&ac->mdct);
2535 ff_mdct_end(&ac->mdct_small);
2536 ff_mdct_end(&ac->mdct_ltp);
2541 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2543 struct LATMContext {
2544 AACContext aac_ctx; ///< containing AACContext
2545 int initialized; ///< initilized after a valid extradata was seen
2548 int audio_mux_version_A; ///< LATM syntax version
2549 int frame_length_type; ///< 0/1 variable/fixed frame length
2550 int frame_length; ///< frame length for fixed frame length
2553 static inline uint32_t latm_get_value(GetBitContext *b)
2555 int length = get_bits(b, 2);
2557 return get_bits_long(b, (length+1)*8);
2560 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2561 GetBitContext *gb, int asclen)
2563 AACContext *ac = &latmctx->aac_ctx;
2564 AVCodecContext *avctx = ac->avctx;
2565 MPEG4AudioConfig m4ac = {0};
2566 int config_start_bit = get_bits_count(gb);
2567 int sync_extension = 0;
2568 int bits_consumed, esize;
2572 asclen = FFMIN(asclen, get_bits_left(gb));
2574 asclen = get_bits_left(gb);
2576 if (config_start_bit % 8) {
2577 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2578 "config not byte aligned.\n", 1);
2579 return AVERROR_INVALIDDATA;
2582 return AVERROR_INVALIDDATA;
2583 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2584 gb->buffer + (config_start_bit / 8),
2585 asclen, sync_extension);
2587 if (bits_consumed < 0)
2588 return AVERROR_INVALIDDATA;
2590 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2591 ac->m4ac.chan_config != m4ac.chan_config) {
2593 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2594 latmctx->initialized = 0;
2596 esize = (bits_consumed+7) / 8;
2598 if (avctx->extradata_size < esize) {
2599 av_free(avctx->extradata);
2600 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2601 if (!avctx->extradata)
2602 return AVERROR(ENOMEM);
2605 avctx->extradata_size = esize;
2606 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2607 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2609 skip_bits_long(gb, bits_consumed);
2611 return bits_consumed;
2614 static int read_stream_mux_config(struct LATMContext *latmctx,
2617 int ret, audio_mux_version = get_bits(gb, 1);
2619 latmctx->audio_mux_version_A = 0;
2620 if (audio_mux_version)
2621 latmctx->audio_mux_version_A = get_bits(gb, 1);
2623 if (!latmctx->audio_mux_version_A) {
2625 if (audio_mux_version)
2626 latm_get_value(gb); // taraFullness
2628 skip_bits(gb, 1); // allStreamSameTimeFraming
2629 skip_bits(gb, 6); // numSubFrames
2631 if (get_bits(gb, 4)) { // numPrograms
2632 av_log_missing_feature(latmctx->aac_ctx.avctx,
2633 "multiple programs are not supported\n", 1);
2634 return AVERROR_PATCHWELCOME;
2637 // for each program (which there is only on in DVB)
2639 // for each layer (which there is only on in DVB)
2640 if (get_bits(gb, 3)) { // numLayer
2641 av_log_missing_feature(latmctx->aac_ctx.avctx,
2642 "multiple layers are not supported\n", 1);
2643 return AVERROR_PATCHWELCOME;
2646 // for all but first stream: use_same_config = get_bits(gb, 1);
2647 if (!audio_mux_version) {
2648 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2651 int ascLen = latm_get_value(gb);
2652 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2655 skip_bits_long(gb, ascLen);
2658 latmctx->frame_length_type = get_bits(gb, 3);
2659 switch (latmctx->frame_length_type) {
2661 skip_bits(gb, 8); // latmBufferFullness
2664 latmctx->frame_length = get_bits(gb, 9);
2669 skip_bits(gb, 6); // CELP frame length table index
2673 skip_bits(gb, 1); // HVXC frame length table index
2677 if (get_bits(gb, 1)) { // other data
2678 if (audio_mux_version) {
2679 latm_get_value(gb); // other_data_bits
2683 esc = get_bits(gb, 1);
2689 if (get_bits(gb, 1)) // crc present
2690 skip_bits(gb, 8); // config_crc
2696 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2700 if (ctx->frame_length_type == 0) {
2701 int mux_slot_length = 0;
2703 tmp = get_bits(gb, 8);
2704 mux_slot_length += tmp;
2705 } while (tmp == 255);
2706 return mux_slot_length;
2707 } else if (ctx->frame_length_type == 1) {
2708 return ctx->frame_length;
2709 } else if (ctx->frame_length_type == 3 ||
2710 ctx->frame_length_type == 5 ||
2711 ctx->frame_length_type == 7) {
2712 skip_bits(gb, 2); // mux_slot_length_coded
2717 static int read_audio_mux_element(struct LATMContext *latmctx,
2721 uint8_t use_same_mux = get_bits(gb, 1);
2722 if (!use_same_mux) {
2723 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2725 } else if (!latmctx->aac_ctx.avctx->extradata) {
2726 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2727 "no decoder config found\n");
2728 return AVERROR(EAGAIN);
2730 if (latmctx->audio_mux_version_A == 0) {
2731 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2732 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2733 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2734 return AVERROR_INVALIDDATA;
2735 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2736 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2737 "frame length mismatch %d << %d\n",
2738 mux_slot_length_bytes * 8, get_bits_left(gb));
2739 return AVERROR_INVALIDDATA;
2746 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2747 int *got_frame_ptr, AVPacket *avpkt)
2749 struct LATMContext *latmctx = avctx->priv_data;
2753 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2755 // check for LOAS sync word
2756 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2757 return AVERROR_INVALIDDATA;
2759 muxlength = get_bits(&gb, 13) + 3;
2760 // not enough data, the parser should have sorted this
2761 if (muxlength > avpkt->size)
2762 return AVERROR_INVALIDDATA;
2764 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2767 if (!latmctx->initialized) {
2768 if (!avctx->extradata) {
2772 if ((err = decode_audio_specific_config(
2773 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2774 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2776 latmctx->initialized = 1;
2780 if (show_bits(&gb, 12) == 0xfff) {
2781 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2782 "ADTS header detected, probably as result of configuration "
2784 return AVERROR_INVALIDDATA;
2787 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2793 static av_cold int latm_decode_init(AVCodecContext *avctx)
2795 struct LATMContext *latmctx = avctx->priv_data;
2796 int ret = aac_decode_init(avctx);
2798 if (avctx->extradata_size > 0)
2799 latmctx->initialized = !ret;
2805 AVCodec ff_aac_decoder = {
2807 .type = AVMEDIA_TYPE_AUDIO,
2809 .priv_data_size = sizeof(AACContext),
2810 .init = aac_decode_init,
2811 .close = aac_decode_close,
2812 .decode = aac_decode_frame,
2813 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2814 .sample_fmts = (const enum AVSampleFormat[]) {
2815 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2817 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2818 .channel_layouts = aac_channel_layout,
2822 Note: This decoder filter is intended to decode LATM streams transferred
2823 in MPEG transport streams which only contain one program.
2824 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2826 AVCodec ff_aac_latm_decoder = {
2828 .type = AVMEDIA_TYPE_AUDIO,
2829 .id = CODEC_ID_AAC_LATM,
2830 .priv_data_size = sizeof(struct LATMContext),
2831 .init = latm_decode_init,
2832 .close = aac_decode_close,
2833 .decode = latm_decode_frame,
2834 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2835 .sample_fmts = (const enum AVSampleFormat[]) {
2836 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2838 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2839 .channel_layouts = aac_channel_layout,