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(enum ChannelPosition che_pos[4][MAX_ELEM_ID])
168 int i, type, sum = 0;
169 for (i = 0; i < MAX_ELEM_ID; i++) {
170 for (type = 0; type < 4; type++) {
171 sum += (1 + (type == TYPE_CPE)) *
172 (che_pos[type][i] != AAC_CHANNEL_OFF &&
173 che_pos[type][i] != AAC_CHANNEL_CC);
180 * Check for the channel element in the current channel position configuration.
181 * If it exists, make sure the appropriate element is allocated and map the
182 * channel order to match the internal Libav channel layout.
184 * @param che_pos current channel position configuration
185 * @param type channel element type
186 * @param id channel element id
187 * @param channels count of the number of channels in the configuration
189 * @return Returns error status. 0 - OK, !0 - error
191 static av_cold int che_configure(AACContext *ac,
192 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
193 int type, int id, int *channels)
195 if (che_pos[type][id]) {
196 if (!ac->che[type][id]) {
197 if (!(ac->che[type][id] = av_mallocz(sizeof(ChannelElement))))
198 return AVERROR(ENOMEM);
199 ff_aac_sbr_ctx_init(ac, &ac->che[type][id]->sbr);
201 if (type != TYPE_CCE) {
202 ac->output_data[(*channels)++] = ac->che[type][id]->ch[0].ret;
203 if (type == TYPE_CPE ||
204 (type == TYPE_SCE && ac->m4ac.ps == 1)) {
205 ac->output_data[(*channels)++] = ac->che[type][id]->ch[1].ret;
209 if (ac->che[type][id])
210 ff_aac_sbr_ctx_close(&ac->che[type][id]->sbr);
211 av_freep(&ac->che[type][id]);
217 * Configure output channel order based on the current program configuration element.
219 * @param che_pos current channel position configuration
220 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
222 * @return Returns error status. 0 - OK, !0 - error
224 static av_cold int output_configure(AACContext *ac,
225 enum ChannelPosition che_pos[4][MAX_ELEM_ID],
226 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
227 int channel_config, enum OCStatus oc_type)
229 AVCodecContext *avctx = ac->avctx;
230 int i, type, channels = 0, ret;
232 if (new_che_pos != che_pos)
233 memcpy(che_pos, new_che_pos, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
235 if (channel_config) {
236 for (i = 0; i < tags_per_config[channel_config]; i++) {
237 if ((ret = che_configure(ac, che_pos,
238 aac_channel_layout_map[channel_config - 1][i][0],
239 aac_channel_layout_map[channel_config - 1][i][1],
244 memset(ac->tag_che_map, 0, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
246 avctx->channel_layout = aac_channel_layout[channel_config - 1];
248 /* Allocate or free elements depending on if they are in the
249 * current program configuration.
251 * Set up default 1:1 output mapping.
253 * For a 5.1 stream the output order will be:
254 * [ Center ] [ Front Left ] [ Front Right ] [ LFE ] [ Surround Left ] [ Surround Right ]
257 for (i = 0; i < MAX_ELEM_ID; i++) {
258 for (type = 0; type < 4; type++) {
259 if ((ret = che_configure(ac, che_pos, type, i, &channels)))
264 memcpy(ac->tag_che_map, ac->che, 4 * MAX_ELEM_ID * sizeof(ac->che[0][0]));
266 avctx->channel_layout = 0;
269 avctx->channels = channels;
271 ac->output_configured = oc_type;
277 * Decode an array of 4 bit element IDs, optionally interleaved with a stereo/mono switching bit.
279 * @param cpe_map Stereo (Channel Pair Element) map, NULL if stereo bit is not present.
280 * @param sce_map mono (Single Channel Element) map
281 * @param type speaker type/position for these channels
283 static void decode_channel_map(enum ChannelPosition *cpe_map,
284 enum ChannelPosition *sce_map,
285 enum ChannelPosition type,
286 GetBitContext *gb, int n)
289 enum ChannelPosition *map = cpe_map && get_bits1(gb) ? cpe_map : sce_map; // stereo or mono map
290 map[get_bits(gb, 4)] = type;
295 * Decode program configuration element; reference: table 4.2.
297 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
299 * @return Returns error status. 0 - OK, !0 - error
301 static int decode_pce(AVCodecContext *avctx, MPEG4AudioConfig *m4ac,
302 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
305 int num_front, num_side, num_back, num_lfe, num_assoc_data, num_cc, sampling_index;
308 skip_bits(gb, 2); // object_type
310 sampling_index = get_bits(gb, 4);
311 if (m4ac->sampling_index != sampling_index)
312 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");
314 num_front = get_bits(gb, 4);
315 num_side = get_bits(gb, 4);
316 num_back = get_bits(gb, 4);
317 num_lfe = get_bits(gb, 2);
318 num_assoc_data = get_bits(gb, 3);
319 num_cc = get_bits(gb, 4);
322 skip_bits(gb, 4); // mono_mixdown_tag
324 skip_bits(gb, 4); // stereo_mixdown_tag
327 skip_bits(gb, 3); // mixdown_coeff_index and pseudo_surround
329 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_FRONT, gb, num_front);
330 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_SIDE, gb, num_side );
331 decode_channel_map(new_che_pos[TYPE_CPE], new_che_pos[TYPE_SCE], AAC_CHANNEL_BACK, gb, num_back );
332 decode_channel_map(NULL, new_che_pos[TYPE_LFE], AAC_CHANNEL_LFE, gb, num_lfe );
334 skip_bits_long(gb, 4 * num_assoc_data);
336 decode_channel_map(new_che_pos[TYPE_CCE], new_che_pos[TYPE_CCE], AAC_CHANNEL_CC, gb, num_cc );
340 /* comment field, first byte is length */
341 comment_len = get_bits(gb, 8) * 8;
342 if (get_bits_left(gb) < comment_len) {
343 av_log(avctx, AV_LOG_ERROR, overread_err);
346 skip_bits_long(gb, comment_len);
351 * Set up channel positions based on a default channel configuration
352 * as specified in table 1.17.
354 * @param new_che_pos New channel position configuration - we only do something if it differs from the current one.
356 * @return Returns error status. 0 - OK, !0 - error
358 static av_cold int set_default_channel_config(AVCodecContext *avctx,
359 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID],
362 if (channel_config < 1 || channel_config > 7) {
363 av_log(avctx, AV_LOG_ERROR, "invalid default channel configuration (%d)\n",
368 /* default channel configurations:
370 * 1ch : front center (mono)
371 * 2ch : L + R (stereo)
372 * 3ch : front center + L + R
373 * 4ch : front center + L + R + back center
374 * 5ch : front center + L + R + back stereo
375 * 6ch : front center + L + R + back stereo + LFE
376 * 7ch : front center + L + R + outer front left + outer front right + back stereo + LFE
379 if (channel_config != 2)
380 new_che_pos[TYPE_SCE][0] = AAC_CHANNEL_FRONT; // front center (or mono)
381 if (channel_config > 1)
382 new_che_pos[TYPE_CPE][0] = AAC_CHANNEL_FRONT; // L + R (or stereo)
383 if (channel_config == 4)
384 new_che_pos[TYPE_SCE][1] = AAC_CHANNEL_BACK; // back center
385 if (channel_config > 4)
386 new_che_pos[TYPE_CPE][(channel_config == 7) + 1]
387 = AAC_CHANNEL_BACK; // back stereo
388 if (channel_config > 5)
389 new_che_pos[TYPE_LFE][0] = AAC_CHANNEL_LFE; // LFE
390 if (channel_config == 7)
391 new_che_pos[TYPE_CPE][1] = AAC_CHANNEL_FRONT; // outer front left + outer front right
397 * Decode GA "General Audio" specific configuration; reference: table 4.1.
399 * @param ac pointer to AACContext, may be null
400 * @param avctx pointer to AVCCodecContext, used for logging
402 * @return Returns error status. 0 - OK, !0 - error
404 static int decode_ga_specific_config(AACContext *ac, AVCodecContext *avctx,
406 MPEG4AudioConfig *m4ac,
409 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
410 int extension_flag, ret;
412 if (get_bits1(gb)) { // frameLengthFlag
413 av_log_missing_feature(avctx, "960/120 MDCT window is", 1);
417 if (get_bits1(gb)) // dependsOnCoreCoder
418 skip_bits(gb, 14); // coreCoderDelay
419 extension_flag = get_bits1(gb);
421 if (m4ac->object_type == AOT_AAC_SCALABLE ||
422 m4ac->object_type == AOT_ER_AAC_SCALABLE)
423 skip_bits(gb, 3); // layerNr
425 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
426 if (channel_config == 0) {
427 skip_bits(gb, 4); // element_instance_tag
428 if ((ret = decode_pce(avctx, m4ac, new_che_pos, gb)))
431 if ((ret = set_default_channel_config(avctx, new_che_pos, channel_config)))
435 if (count_channels(new_che_pos) > 1) {
437 } else if (m4ac->sbr == 1 && m4ac->ps == -1)
440 if (ac && (ret = output_configure(ac, ac->che_pos, new_che_pos, channel_config, OC_GLOBAL_HDR)))
443 if (extension_flag) {
444 switch (m4ac->object_type) {
446 skip_bits(gb, 5); // numOfSubFrame
447 skip_bits(gb, 11); // layer_length
451 case AOT_ER_AAC_SCALABLE:
453 skip_bits(gb, 3); /* aacSectionDataResilienceFlag
454 * aacScalefactorDataResilienceFlag
455 * aacSpectralDataResilienceFlag
459 skip_bits1(gb); // extensionFlag3 (TBD in version 3)
465 * Decode audio specific configuration; reference: table 1.13.
467 * @param ac pointer to AACContext, may be null
468 * @param avctx pointer to AVCCodecContext, used for logging
469 * @param m4ac pointer to MPEG4AudioConfig, used for parsing
470 * @param data pointer to buffer holding an audio specific config
471 * @param bit_size size of audio specific config or data in bits
472 * @param sync_extension look for an appended sync extension
474 * @return Returns error status or number of consumed bits. <0 - error
476 static int decode_audio_specific_config(AACContext *ac,
477 AVCodecContext *avctx,
478 MPEG4AudioConfig *m4ac,
479 const uint8_t *data, int bit_size,
485 av_dlog(avctx, "extradata size %d\n", avctx->extradata_size);
486 for (i = 0; i < avctx->extradata_size; i++)
487 av_dlog(avctx, "%02x ", avctx->extradata[i]);
488 av_dlog(avctx, "\n");
490 init_get_bits(&gb, data, bit_size);
492 if ((i = avpriv_mpeg4audio_get_config(m4ac, data, bit_size, sync_extension)) < 0)
494 if (m4ac->sampling_index > 12) {
495 av_log(avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", m4ac->sampling_index);
499 skip_bits_long(&gb, i);
501 switch (m4ac->object_type) {
505 if (decode_ga_specific_config(ac, avctx, &gb, m4ac, m4ac->chan_config))
509 av_log(avctx, AV_LOG_ERROR, "Audio object type %s%d is not supported.\n",
510 m4ac->sbr == 1? "SBR+" : "", m4ac->object_type);
514 av_dlog(avctx, "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
515 m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
516 m4ac->sample_rate, m4ac->sbr, m4ac->ps);
518 return get_bits_count(&gb);
522 * linear congruential pseudorandom number generator
524 * @param previous_val pointer to the current state of the generator
526 * @return Returns a 32-bit pseudorandom integer
528 static av_always_inline int lcg_random(int previous_val)
530 return previous_val * 1664525 + 1013904223;
533 static av_always_inline void reset_predict_state(PredictorState *ps)
543 static void reset_all_predictors(PredictorState *ps)
546 for (i = 0; i < MAX_PREDICTORS; i++)
547 reset_predict_state(&ps[i]);
550 static int sample_rate_idx (int rate)
552 if (92017 <= rate) return 0;
553 else if (75132 <= rate) return 1;
554 else if (55426 <= rate) return 2;
555 else if (46009 <= rate) return 3;
556 else if (37566 <= rate) return 4;
557 else if (27713 <= rate) return 5;
558 else if (23004 <= rate) return 6;
559 else if (18783 <= rate) return 7;
560 else if (13856 <= rate) return 8;
561 else if (11502 <= rate) return 9;
562 else if (9391 <= rate) return 10;
566 static void reset_predictor_group(PredictorState *ps, int group_num)
569 for (i = group_num - 1; i < MAX_PREDICTORS; i += 30)
570 reset_predict_state(&ps[i]);
573 #define AAC_INIT_VLC_STATIC(num, size) \
574 INIT_VLC_STATIC(&vlc_spectral[num], 8, ff_aac_spectral_sizes[num], \
575 ff_aac_spectral_bits[num], sizeof( ff_aac_spectral_bits[num][0]), sizeof( ff_aac_spectral_bits[num][0]), \
576 ff_aac_spectral_codes[num], sizeof(ff_aac_spectral_codes[num][0]), sizeof(ff_aac_spectral_codes[num][0]), \
579 static av_cold int aac_decode_init(AVCodecContext *avctx)
581 AACContext *ac = avctx->priv_data;
582 float output_scale_factor;
585 ac->m4ac.sample_rate = avctx->sample_rate;
587 if (avctx->extradata_size > 0) {
588 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
590 avctx->extradata_size*8, 1) < 0)
594 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
596 sr = sample_rate_idx(avctx->sample_rate);
597 ac->m4ac.sampling_index = sr;
598 ac->m4ac.channels = avctx->channels;
602 for (i = 0; i < FF_ARRAY_ELEMS(ff_mpeg4audio_channels); i++)
603 if (ff_mpeg4audio_channels[i] == avctx->channels)
605 if (i == FF_ARRAY_ELEMS(ff_mpeg4audio_channels)) {
608 ac->m4ac.chan_config = i;
610 if (ac->m4ac.chan_config) {
611 int ret = set_default_channel_config(avctx, new_che_pos, ac->m4ac.chan_config);
613 output_configure(ac, ac->che_pos, new_che_pos, ac->m4ac.chan_config, OC_GLOBAL_HDR);
614 else if (avctx->err_recognition & AV_EF_EXPLODE)
615 return AVERROR_INVALIDDATA;
619 if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
620 avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
621 output_scale_factor = 1.0 / 32768.0;
623 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
624 output_scale_factor = 1.0;
627 AAC_INIT_VLC_STATIC( 0, 304);
628 AAC_INIT_VLC_STATIC( 1, 270);
629 AAC_INIT_VLC_STATIC( 2, 550);
630 AAC_INIT_VLC_STATIC( 3, 300);
631 AAC_INIT_VLC_STATIC( 4, 328);
632 AAC_INIT_VLC_STATIC( 5, 294);
633 AAC_INIT_VLC_STATIC( 6, 306);
634 AAC_INIT_VLC_STATIC( 7, 268);
635 AAC_INIT_VLC_STATIC( 8, 510);
636 AAC_INIT_VLC_STATIC( 9, 366);
637 AAC_INIT_VLC_STATIC(10, 462);
641 dsputil_init(&ac->dsp, avctx);
642 ff_fmt_convert_init(&ac->fmt_conv, avctx);
644 ac->random_state = 0x1f2e3d4c;
648 INIT_VLC_STATIC(&vlc_scalefactors,7,FF_ARRAY_ELEMS(ff_aac_scalefactor_code),
649 ff_aac_scalefactor_bits, sizeof(ff_aac_scalefactor_bits[0]), sizeof(ff_aac_scalefactor_bits[0]),
650 ff_aac_scalefactor_code, sizeof(ff_aac_scalefactor_code[0]), sizeof(ff_aac_scalefactor_code[0]),
653 ff_mdct_init(&ac->mdct, 11, 1, output_scale_factor/1024.0);
654 ff_mdct_init(&ac->mdct_small, 8, 1, output_scale_factor/128.0);
655 ff_mdct_init(&ac->mdct_ltp, 11, 0, -2.0/output_scale_factor);
656 // window initialization
657 ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
658 ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
659 ff_init_ff_sine_windows(10);
660 ff_init_ff_sine_windows( 7);
664 avcodec_get_frame_defaults(&ac->frame);
665 avctx->coded_frame = &ac->frame;
671 * Skip data_stream_element; reference: table 4.10.
673 static int skip_data_stream_element(AACContext *ac, GetBitContext *gb)
675 int byte_align = get_bits1(gb);
676 int count = get_bits(gb, 8);
678 count += get_bits(gb, 8);
682 if (get_bits_left(gb) < 8 * count) {
683 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
686 skip_bits_long(gb, 8 * count);
690 static int decode_prediction(AACContext *ac, IndividualChannelStream *ics,
695 ics->predictor_reset_group = get_bits(gb, 5);
696 if (ics->predictor_reset_group == 0 || ics->predictor_reset_group > 30) {
697 av_log(ac->avctx, AV_LOG_ERROR, "Invalid Predictor Reset Group.\n");
701 for (sfb = 0; sfb < FFMIN(ics->max_sfb, ff_aac_pred_sfb_max[ac->m4ac.sampling_index]); sfb++) {
702 ics->prediction_used[sfb] = get_bits1(gb);
708 * Decode Long Term Prediction data; reference: table 4.xx.
710 static void decode_ltp(AACContext *ac, LongTermPrediction *ltp,
711 GetBitContext *gb, uint8_t max_sfb)
715 ltp->lag = get_bits(gb, 11);
716 ltp->coef = ltp_coef[get_bits(gb, 3)];
717 for (sfb = 0; sfb < FFMIN(max_sfb, MAX_LTP_LONG_SFB); sfb++)
718 ltp->used[sfb] = get_bits1(gb);
722 * Decode Individual Channel Stream info; reference: table 4.6.
724 static int decode_ics_info(AACContext *ac, IndividualChannelStream *ics,
728 av_log(ac->avctx, AV_LOG_ERROR, "Reserved bit set.\n");
729 return AVERROR_INVALIDDATA;
731 ics->window_sequence[1] = ics->window_sequence[0];
732 ics->window_sequence[0] = get_bits(gb, 2);
733 ics->use_kb_window[1] = ics->use_kb_window[0];
734 ics->use_kb_window[0] = get_bits1(gb);
735 ics->num_window_groups = 1;
736 ics->group_len[0] = 1;
737 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
739 ics->max_sfb = get_bits(gb, 4);
740 for (i = 0; i < 7; i++) {
742 ics->group_len[ics->num_window_groups - 1]++;
744 ics->num_window_groups++;
745 ics->group_len[ics->num_window_groups - 1] = 1;
748 ics->num_windows = 8;
749 ics->swb_offset = ff_swb_offset_128[ac->m4ac.sampling_index];
750 ics->num_swb = ff_aac_num_swb_128[ac->m4ac.sampling_index];
751 ics->tns_max_bands = ff_tns_max_bands_128[ac->m4ac.sampling_index];
752 ics->predictor_present = 0;
754 ics->max_sfb = get_bits(gb, 6);
755 ics->num_windows = 1;
756 ics->swb_offset = ff_swb_offset_1024[ac->m4ac.sampling_index];
757 ics->num_swb = ff_aac_num_swb_1024[ac->m4ac.sampling_index];
758 ics->tns_max_bands = ff_tns_max_bands_1024[ac->m4ac.sampling_index];
759 ics->predictor_present = get_bits1(gb);
760 ics->predictor_reset_group = 0;
761 if (ics->predictor_present) {
762 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
763 if (decode_prediction(ac, ics, gb)) {
764 return AVERROR_INVALIDDATA;
766 } else if (ac->m4ac.object_type == AOT_AAC_LC) {
767 av_log(ac->avctx, AV_LOG_ERROR, "Prediction is not allowed in AAC-LC.\n");
768 return AVERROR_INVALIDDATA;
770 if ((ics->ltp.present = get_bits(gb, 1)))
771 decode_ltp(ac, &ics->ltp, gb, ics->max_sfb);
776 if (ics->max_sfb > ics->num_swb) {
777 av_log(ac->avctx, AV_LOG_ERROR,
778 "Number of scalefactor bands in group (%d) exceeds limit (%d).\n",
779 ics->max_sfb, ics->num_swb);
780 return AVERROR_INVALIDDATA;
787 * Decode band types (section_data payload); reference: table 4.46.
789 * @param band_type array of the used band type
790 * @param band_type_run_end array of the last scalefactor band of a band type run
792 * @return Returns error status. 0 - OK, !0 - error
794 static int decode_band_types(AACContext *ac, enum BandType band_type[120],
795 int band_type_run_end[120], GetBitContext *gb,
796 IndividualChannelStream *ics)
799 const int bits = (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) ? 3 : 5;
800 for (g = 0; g < ics->num_window_groups; g++) {
802 while (k < ics->max_sfb) {
803 uint8_t sect_end = k;
805 int sect_band_type = get_bits(gb, 4);
806 if (sect_band_type == 12) {
807 av_log(ac->avctx, AV_LOG_ERROR, "invalid band type\n");
810 while ((sect_len_incr = get_bits(gb, bits)) == (1 << bits) - 1)
811 sect_end += sect_len_incr;
812 sect_end += sect_len_incr;
813 if (get_bits_left(gb) < 0) {
814 av_log(ac->avctx, AV_LOG_ERROR, overread_err);
817 if (sect_end > ics->max_sfb) {
818 av_log(ac->avctx, AV_LOG_ERROR,
819 "Number of bands (%d) exceeds limit (%d).\n",
820 sect_end, ics->max_sfb);
823 for (; k < sect_end; k++) {
824 band_type [idx] = sect_band_type;
825 band_type_run_end[idx++] = sect_end;
833 * Decode scalefactors; reference: table 4.47.
835 * @param global_gain first scalefactor value as scalefactors are differentially coded
836 * @param band_type array of the used band type
837 * @param band_type_run_end array of the last scalefactor band of a band type run
838 * @param sf array of scalefactors or intensity stereo positions
840 * @return Returns error status. 0 - OK, !0 - error
842 static int decode_scalefactors(AACContext *ac, float sf[120], GetBitContext *gb,
843 unsigned int global_gain,
844 IndividualChannelStream *ics,
845 enum BandType band_type[120],
846 int band_type_run_end[120])
849 int offset[3] = { global_gain, global_gain - 90, 0 };
852 static const char *sf_str[3] = { "Global gain", "Noise gain", "Intensity stereo position" };
853 for (g = 0; g < ics->num_window_groups; g++) {
854 for (i = 0; i < ics->max_sfb;) {
855 int run_end = band_type_run_end[idx];
856 if (band_type[idx] == ZERO_BT) {
857 for (; i < run_end; i++, idx++)
859 } else if ((band_type[idx] == INTENSITY_BT) || (band_type[idx] == INTENSITY_BT2)) {
860 for (; i < run_end; i++, idx++) {
861 offset[2] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
862 clipped_offset = av_clip(offset[2], -155, 100);
863 if (offset[2] != clipped_offset) {
864 av_log_ask_for_sample(ac->avctx, "Intensity stereo "
865 "position clipped (%d -> %d).\nIf you heard an "
866 "audible artifact, there may be a bug in the "
867 "decoder. ", offset[2], clipped_offset);
869 sf[idx] = ff_aac_pow2sf_tab[-clipped_offset + POW_SF2_ZERO];
871 } else if (band_type[idx] == NOISE_BT) {
872 for (; i < run_end; i++, idx++) {
873 if (noise_flag-- > 0)
874 offset[1] += get_bits(gb, 9) - 256;
876 offset[1] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
877 clipped_offset = av_clip(offset[1], -100, 155);
878 if (offset[1] != clipped_offset) {
879 av_log_ask_for_sample(ac->avctx, "Noise gain clipped "
880 "(%d -> %d).\nIf you heard an audible "
881 "artifact, there may be a bug in the decoder. ",
882 offset[1], clipped_offset);
884 sf[idx] = -ff_aac_pow2sf_tab[clipped_offset + POW_SF2_ZERO];
887 for (; i < run_end; i++, idx++) {
888 offset[0] += get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
889 if (offset[0] > 255U) {
890 av_log(ac->avctx, AV_LOG_ERROR,
891 "%s (%d) out of range.\n", sf_str[0], offset[0]);
894 sf[idx] = -ff_aac_pow2sf_tab[offset[0] - 100 + POW_SF2_ZERO];
903 * Decode pulse data; reference: table 4.7.
905 static int decode_pulses(Pulse *pulse, GetBitContext *gb,
906 const uint16_t *swb_offset, int num_swb)
909 pulse->num_pulse = get_bits(gb, 2) + 1;
910 pulse_swb = get_bits(gb, 6);
911 if (pulse_swb >= num_swb)
913 pulse->pos[0] = swb_offset[pulse_swb];
914 pulse->pos[0] += get_bits(gb, 5);
915 if (pulse->pos[0] > 1023)
917 pulse->amp[0] = get_bits(gb, 4);
918 for (i = 1; i < pulse->num_pulse; i++) {
919 pulse->pos[i] = get_bits(gb, 5) + pulse->pos[i - 1];
920 if (pulse->pos[i] > 1023)
922 pulse->amp[i] = get_bits(gb, 4);
928 * Decode Temporal Noise Shaping data; reference: table 4.48.
930 * @return Returns error status. 0 - OK, !0 - error
932 static int decode_tns(AACContext *ac, TemporalNoiseShaping *tns,
933 GetBitContext *gb, const IndividualChannelStream *ics)
935 int w, filt, i, coef_len, coef_res, coef_compress;
936 const int is8 = ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE;
937 const int tns_max_order = is8 ? 7 : ac->m4ac.object_type == AOT_AAC_MAIN ? 20 : 12;
938 for (w = 0; w < ics->num_windows; w++) {
939 if ((tns->n_filt[w] = get_bits(gb, 2 - is8))) {
940 coef_res = get_bits1(gb);
942 for (filt = 0; filt < tns->n_filt[w]; filt++) {
944 tns->length[w][filt] = get_bits(gb, 6 - 2 * is8);
946 if ((tns->order[w][filt] = get_bits(gb, 5 - 2 * is8)) > tns_max_order) {
947 av_log(ac->avctx, AV_LOG_ERROR, "TNS filter order %d is greater than maximum %d.\n",
948 tns->order[w][filt], tns_max_order);
949 tns->order[w][filt] = 0;
952 if (tns->order[w][filt]) {
953 tns->direction[w][filt] = get_bits1(gb);
954 coef_compress = get_bits1(gb);
955 coef_len = coef_res + 3 - coef_compress;
956 tmp2_idx = 2 * coef_compress + coef_res;
958 for (i = 0; i < tns->order[w][filt]; i++)
959 tns->coef[w][filt][i] = tns_tmp2_map[tmp2_idx][get_bits(gb, coef_len)];
968 * Decode Mid/Side data; reference: table 4.54.
970 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
971 * [1] mask is decoded from bitstream; [2] mask is all 1s;
972 * [3] reserved for scalable AAC
974 static void decode_mid_side_stereo(ChannelElement *cpe, GetBitContext *gb,
978 if (ms_present == 1) {
979 for (idx = 0; idx < cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb; idx++)
980 cpe->ms_mask[idx] = get_bits1(gb);
981 } else if (ms_present == 2) {
982 memset(cpe->ms_mask, 1, cpe->ch[0].ics.num_window_groups * cpe->ch[0].ics.max_sfb * sizeof(cpe->ms_mask[0]));
987 static inline float *VMUL2(float *dst, const float *v, unsigned idx,
991 *dst++ = v[idx & 15] * s;
992 *dst++ = v[idx>>4 & 15] * s;
998 static inline float *VMUL4(float *dst, const float *v, unsigned idx,
1002 *dst++ = v[idx & 3] * s;
1003 *dst++ = v[idx>>2 & 3] * s;
1004 *dst++ = v[idx>>4 & 3] * s;
1005 *dst++ = v[idx>>6 & 3] * s;
1011 static inline float *VMUL2S(float *dst, const float *v, unsigned idx,
1012 unsigned sign, const float *scale)
1014 union av_intfloat32 s0, s1;
1016 s0.f = s1.f = *scale;
1017 s0.i ^= sign >> 1 << 31;
1020 *dst++ = v[idx & 15] * s0.f;
1021 *dst++ = v[idx>>4 & 15] * s1.f;
1028 static inline float *VMUL4S(float *dst, const float *v, unsigned idx,
1029 unsigned sign, const float *scale)
1031 unsigned nz = idx >> 12;
1032 union av_intfloat32 s = { .f = *scale };
1033 union av_intfloat32 t;
1035 t.i = s.i ^ (sign & 1U<<31);
1036 *dst++ = v[idx & 3] * t.f;
1038 sign <<= nz & 1; nz >>= 1;
1039 t.i = s.i ^ (sign & 1U<<31);
1040 *dst++ = v[idx>>2 & 3] * t.f;
1042 sign <<= nz & 1; nz >>= 1;
1043 t.i = s.i ^ (sign & 1U<<31);
1044 *dst++ = v[idx>>4 & 3] * t.f;
1046 sign <<= nz & 1; nz >>= 1;
1047 t.i = s.i ^ (sign & 1U<<31);
1048 *dst++ = v[idx>>6 & 3] * t.f;
1055 * Decode spectral data; reference: table 4.50.
1056 * Dequantize and scale spectral data; reference: 4.6.3.3.
1058 * @param coef array of dequantized, scaled spectral data
1059 * @param sf array of scalefactors or intensity stereo positions
1060 * @param pulse_present set if pulses are present
1061 * @param pulse pointer to pulse data struct
1062 * @param band_type array of the used band type
1064 * @return Returns error status. 0 - OK, !0 - error
1066 static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
1067 GetBitContext *gb, const float sf[120],
1068 int pulse_present, const Pulse *pulse,
1069 const IndividualChannelStream *ics,
1070 enum BandType band_type[120])
1072 int i, k, g, idx = 0;
1073 const int c = 1024 / ics->num_windows;
1074 const uint16_t *offsets = ics->swb_offset;
1075 float *coef_base = coef;
1077 for (g = 0; g < ics->num_windows; g++)
1078 memset(coef + g * 128 + offsets[ics->max_sfb], 0, sizeof(float) * (c - offsets[ics->max_sfb]));
1080 for (g = 0; g < ics->num_window_groups; g++) {
1081 unsigned g_len = ics->group_len[g];
1083 for (i = 0; i < ics->max_sfb; i++, idx++) {
1084 const unsigned cbt_m1 = band_type[idx] - 1;
1085 float *cfo = coef + offsets[i];
1086 int off_len = offsets[i + 1] - offsets[i];
1089 if (cbt_m1 >= INTENSITY_BT2 - 1) {
1090 for (group = 0; group < g_len; group++, cfo+=128) {
1091 memset(cfo, 0, off_len * sizeof(float));
1093 } else if (cbt_m1 == NOISE_BT - 1) {
1094 for (group = 0; group < g_len; group++, cfo+=128) {
1098 for (k = 0; k < off_len; k++) {
1099 ac->random_state = lcg_random(ac->random_state);
1100 cfo[k] = ac->random_state;
1103 band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
1104 scale = sf[idx] / sqrtf(band_energy);
1105 ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
1108 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
1109 const uint16_t *cb_vector_idx = ff_aac_codebook_vector_idx[cbt_m1];
1110 VLC_TYPE (*vlc_tab)[2] = vlc_spectral[cbt_m1].table;
1111 OPEN_READER(re, gb);
1113 switch (cbt_m1 >> 1) {
1115 for (group = 0; group < g_len; group++, cfo+=128) {
1123 UPDATE_CACHE(re, gb);
1124 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1125 cb_idx = cb_vector_idx[code];
1126 cf = VMUL4(cf, vq, cb_idx, sf + idx);
1132 for (group = 0; group < g_len; group++, cfo+=128) {
1142 UPDATE_CACHE(re, gb);
1143 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1144 cb_idx = cb_vector_idx[code];
1145 nnz = cb_idx >> 8 & 15;
1146 bits = nnz ? GET_CACHE(re, gb) : 0;
1147 LAST_SKIP_BITS(re, gb, nnz);
1148 cf = VMUL4S(cf, vq, cb_idx, bits, sf + idx);
1154 for (group = 0; group < g_len; group++, cfo+=128) {
1162 UPDATE_CACHE(re, gb);
1163 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1164 cb_idx = cb_vector_idx[code];
1165 cf = VMUL2(cf, vq, cb_idx, sf + idx);
1172 for (group = 0; group < g_len; group++, cfo+=128) {
1182 UPDATE_CACHE(re, gb);
1183 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1184 cb_idx = cb_vector_idx[code];
1185 nnz = cb_idx >> 8 & 15;
1186 sign = nnz ? SHOW_UBITS(re, gb, nnz) << (cb_idx >> 12) : 0;
1187 LAST_SKIP_BITS(re, gb, nnz);
1188 cf = VMUL2S(cf, vq, cb_idx, sign, sf + idx);
1194 for (group = 0; group < g_len; group++, cfo+=128) {
1196 uint32_t *icf = (uint32_t *) cf;
1206 UPDATE_CACHE(re, gb);
1207 GET_VLC(code, re, gb, vlc_tab, 8, 2);
1215 cb_idx = cb_vector_idx[code];
1218 bits = SHOW_UBITS(re, gb, nnz) << (32-nnz);
1219 LAST_SKIP_BITS(re, gb, nnz);
1221 for (j = 0; j < 2; j++) {
1225 /* The total length of escape_sequence must be < 22 bits according
1226 to the specification (i.e. max is 111111110xxxxxxxxxxxx). */
1227 UPDATE_CACHE(re, gb);
1228 b = GET_CACHE(re, gb);
1229 b = 31 - av_log2(~b);
1232 av_log(ac->avctx, AV_LOG_ERROR, "error in spectral data, ESC overflow\n");
1236 SKIP_BITS(re, gb, b + 1);
1238 n = (1 << b) + SHOW_UBITS(re, gb, b);
1239 LAST_SKIP_BITS(re, gb, b);
1240 *icf++ = cbrt_tab[n] | (bits & 1U<<31);
1243 unsigned v = ((const uint32_t*)vq)[cb_idx & 15];
1244 *icf++ = (bits & 1U<<31) | v;
1251 ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
1255 CLOSE_READER(re, gb);
1261 if (pulse_present) {
1263 for (i = 0; i < pulse->num_pulse; i++) {
1264 float co = coef_base[ pulse->pos[i] ];
1265 while (offsets[idx + 1] <= pulse->pos[i])
1267 if (band_type[idx] != NOISE_BT && sf[idx]) {
1268 float ico = -pulse->amp[i];
1271 ico = co / sqrtf(sqrtf(fabsf(co))) + (co > 0 ? -ico : ico);
1273 coef_base[ pulse->pos[i] ] = cbrtf(fabsf(ico)) * ico * sf[idx];
1280 static av_always_inline float flt16_round(float pf)
1282 union av_intfloat32 tmp;
1284 tmp.i = (tmp.i + 0x00008000U) & 0xFFFF0000U;
1288 static av_always_inline float flt16_even(float pf)
1290 union av_intfloat32 tmp;
1292 tmp.i = (tmp.i + 0x00007FFFU + (tmp.i & 0x00010000U >> 16)) & 0xFFFF0000U;
1296 static av_always_inline float flt16_trunc(float pf)
1298 union av_intfloat32 pun;
1300 pun.i &= 0xFFFF0000U;
1304 static av_always_inline void predict(PredictorState *ps, float *coef,
1307 const float a = 0.953125; // 61.0 / 64
1308 const float alpha = 0.90625; // 29.0 / 32
1312 float r0 = ps->r0, r1 = ps->r1;
1313 float cor0 = ps->cor0, cor1 = ps->cor1;
1314 float var0 = ps->var0, var1 = ps->var1;
1316 k1 = var0 > 1 ? cor0 * flt16_even(a / var0) : 0;
1317 k2 = var1 > 1 ? cor1 * flt16_even(a / var1) : 0;
1319 pv = flt16_round(k1 * r0 + k2 * r1);
1326 ps->cor1 = flt16_trunc(alpha * cor1 + r1 * e1);
1327 ps->var1 = flt16_trunc(alpha * var1 + 0.5f * (r1 * r1 + e1 * e1));
1328 ps->cor0 = flt16_trunc(alpha * cor0 + r0 * e0);
1329 ps->var0 = flt16_trunc(alpha * var0 + 0.5f * (r0 * r0 + e0 * e0));
1331 ps->r1 = flt16_trunc(a * (r0 - k1 * e0));
1332 ps->r0 = flt16_trunc(a * e0);
1336 * Apply AAC-Main style frequency domain prediction.
1338 static void apply_prediction(AACContext *ac, SingleChannelElement *sce)
1342 if (!sce->ics.predictor_initialized) {
1343 reset_all_predictors(sce->predictor_state);
1344 sce->ics.predictor_initialized = 1;
1347 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1348 for (sfb = 0; sfb < ff_aac_pred_sfb_max[ac->m4ac.sampling_index]; sfb++) {
1349 for (k = sce->ics.swb_offset[sfb]; k < sce->ics.swb_offset[sfb + 1]; k++) {
1350 predict(&sce->predictor_state[k], &sce->coeffs[k],
1351 sce->ics.predictor_present && sce->ics.prediction_used[sfb]);
1354 if (sce->ics.predictor_reset_group)
1355 reset_predictor_group(sce->predictor_state, sce->ics.predictor_reset_group);
1357 reset_all_predictors(sce->predictor_state);
1361 * Decode an individual_channel_stream payload; reference: table 4.44.
1363 * @param common_window Channels have independent [0], or shared [1], Individual Channel Stream information.
1364 * @param scale_flag scalable [1] or non-scalable [0] AAC (Unused until scalable AAC is implemented.)
1366 * @return Returns error status. 0 - OK, !0 - error
1368 static int decode_ics(AACContext *ac, SingleChannelElement *sce,
1369 GetBitContext *gb, int common_window, int scale_flag)
1372 TemporalNoiseShaping *tns = &sce->tns;
1373 IndividualChannelStream *ics = &sce->ics;
1374 float *out = sce->coeffs;
1375 int global_gain, pulse_present = 0;
1377 /* This assignment is to silence a GCC warning about the variable being used
1378 * uninitialized when in fact it always is.
1380 pulse.num_pulse = 0;
1382 global_gain = get_bits(gb, 8);
1384 if (!common_window && !scale_flag) {
1385 if (decode_ics_info(ac, ics, gb) < 0)
1386 return AVERROR_INVALIDDATA;
1389 if (decode_band_types(ac, sce->band_type, sce->band_type_run_end, gb, ics) < 0)
1391 if (decode_scalefactors(ac, sce->sf, gb, global_gain, ics, sce->band_type, sce->band_type_run_end) < 0)
1396 if ((pulse_present = get_bits1(gb))) {
1397 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1398 av_log(ac->avctx, AV_LOG_ERROR, "Pulse tool not allowed in eight short sequence.\n");
1401 if (decode_pulses(&pulse, gb, ics->swb_offset, ics->num_swb)) {
1402 av_log(ac->avctx, AV_LOG_ERROR, "Pulse data corrupt or invalid.\n");
1406 if ((tns->present = get_bits1(gb)) && decode_tns(ac, tns, gb, ics))
1408 if (get_bits1(gb)) {
1409 av_log_missing_feature(ac->avctx, "SSR", 1);
1414 if (decode_spectrum_and_dequant(ac, out, gb, sce->sf, pulse_present, &pulse, ics, sce->band_type) < 0)
1417 if (ac->m4ac.object_type == AOT_AAC_MAIN && !common_window)
1418 apply_prediction(ac, sce);
1424 * Mid/Side stereo decoding; reference: 4.6.8.1.3.
1426 static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
1428 const IndividualChannelStream *ics = &cpe->ch[0].ics;
1429 float *ch0 = cpe->ch[0].coeffs;
1430 float *ch1 = cpe->ch[1].coeffs;
1431 int g, i, group, idx = 0;
1432 const uint16_t *offsets = ics->swb_offset;
1433 for (g = 0; g < ics->num_window_groups; g++) {
1434 for (i = 0; i < ics->max_sfb; i++, idx++) {
1435 if (cpe->ms_mask[idx] &&
1436 cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
1437 for (group = 0; group < ics->group_len[g]; group++) {
1438 ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
1439 ch1 + group * 128 + offsets[i],
1440 offsets[i+1] - offsets[i]);
1444 ch0 += ics->group_len[g] * 128;
1445 ch1 += ics->group_len[g] * 128;
1450 * intensity stereo decoding; reference: 4.6.8.2.3
1452 * @param ms_present Indicates mid/side stereo presence. [0] mask is all 0s;
1453 * [1] mask is decoded from bitstream; [2] mask is all 1s;
1454 * [3] reserved for scalable AAC
1456 static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_present)
1458 const IndividualChannelStream *ics = &cpe->ch[1].ics;
1459 SingleChannelElement *sce1 = &cpe->ch[1];
1460 float *coef0 = cpe->ch[0].coeffs, *coef1 = cpe->ch[1].coeffs;
1461 const uint16_t *offsets = ics->swb_offset;
1462 int g, group, i, idx = 0;
1465 for (g = 0; g < ics->num_window_groups; g++) {
1466 for (i = 0; i < ics->max_sfb;) {
1467 if (sce1->band_type[idx] == INTENSITY_BT || sce1->band_type[idx] == INTENSITY_BT2) {
1468 const int bt_run_end = sce1->band_type_run_end[idx];
1469 for (; i < bt_run_end; i++, idx++) {
1470 c = -1 + 2 * (sce1->band_type[idx] - 14);
1472 c *= 1 - 2 * cpe->ms_mask[idx];
1473 scale = c * sce1->sf[idx];
1474 for (group = 0; group < ics->group_len[g]; group++)
1475 ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
1476 coef0 + group * 128 + offsets[i],
1478 offsets[i + 1] - offsets[i]);
1481 int bt_run_end = sce1->band_type_run_end[idx];
1482 idx += bt_run_end - i;
1486 coef0 += ics->group_len[g] * 128;
1487 coef1 += ics->group_len[g] * 128;
1492 * Decode a channel_pair_element; reference: table 4.4.
1494 * @return Returns error status. 0 - OK, !0 - error
1496 static int decode_cpe(AACContext *ac, GetBitContext *gb, ChannelElement *cpe)
1498 int i, ret, common_window, ms_present = 0;
1500 common_window = get_bits1(gb);
1501 if (common_window) {
1502 if (decode_ics_info(ac, &cpe->ch[0].ics, gb))
1503 return AVERROR_INVALIDDATA;
1504 i = cpe->ch[1].ics.use_kb_window[0];
1505 cpe->ch[1].ics = cpe->ch[0].ics;
1506 cpe->ch[1].ics.use_kb_window[1] = i;
1507 if (cpe->ch[1].ics.predictor_present && (ac->m4ac.object_type != AOT_AAC_MAIN))
1508 if ((cpe->ch[1].ics.ltp.present = get_bits(gb, 1)))
1509 decode_ltp(ac, &cpe->ch[1].ics.ltp, gb, cpe->ch[1].ics.max_sfb);
1510 ms_present = get_bits(gb, 2);
1511 if (ms_present == 3) {
1512 av_log(ac->avctx, AV_LOG_ERROR, "ms_present = 3 is reserved.\n");
1514 } else if (ms_present)
1515 decode_mid_side_stereo(cpe, gb, ms_present);
1517 if ((ret = decode_ics(ac, &cpe->ch[0], gb, common_window, 0)))
1519 if ((ret = decode_ics(ac, &cpe->ch[1], gb, common_window, 0)))
1522 if (common_window) {
1524 apply_mid_side_stereo(ac, cpe);
1525 if (ac->m4ac.object_type == AOT_AAC_MAIN) {
1526 apply_prediction(ac, &cpe->ch[0]);
1527 apply_prediction(ac, &cpe->ch[1]);
1531 apply_intensity_stereo(ac, cpe, ms_present);
1535 static const float cce_scale[] = {
1536 1.09050773266525765921, //2^(1/8)
1537 1.18920711500272106672, //2^(1/4)
1543 * Decode coupling_channel_element; reference: table 4.8.
1545 * @return Returns error status. 0 - OK, !0 - error
1547 static int decode_cce(AACContext *ac, GetBitContext *gb, ChannelElement *che)
1553 SingleChannelElement *sce = &che->ch[0];
1554 ChannelCoupling *coup = &che->coup;
1556 coup->coupling_point = 2 * get_bits1(gb);
1557 coup->num_coupled = get_bits(gb, 3);
1558 for (c = 0; c <= coup->num_coupled; c++) {
1560 coup->type[c] = get_bits1(gb) ? TYPE_CPE : TYPE_SCE;
1561 coup->id_select[c] = get_bits(gb, 4);
1562 if (coup->type[c] == TYPE_CPE) {
1563 coup->ch_select[c] = get_bits(gb, 2);
1564 if (coup->ch_select[c] == 3)
1567 coup->ch_select[c] = 2;
1569 coup->coupling_point += get_bits1(gb) || (coup->coupling_point >> 1);
1571 sign = get_bits(gb, 1);
1572 scale = cce_scale[get_bits(gb, 2)];
1574 if ((ret = decode_ics(ac, sce, gb, 0, 0)))
1577 for (c = 0; c < num_gain; c++) {
1581 float gain_cache = 1.;
1583 cge = coup->coupling_point == AFTER_IMDCT ? 1 : get_bits1(gb);
1584 gain = cge ? get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60: 0;
1585 gain_cache = powf(scale, -gain);
1587 if (coup->coupling_point == AFTER_IMDCT) {
1588 coup->gain[c][0] = gain_cache;
1590 for (g = 0; g < sce->ics.num_window_groups; g++) {
1591 for (sfb = 0; sfb < sce->ics.max_sfb; sfb++, idx++) {
1592 if (sce->band_type[idx] != ZERO_BT) {
1594 int t = get_vlc2(gb, vlc_scalefactors.table, 7, 3) - 60;
1602 gain_cache = powf(scale, -t) * s;
1605 coup->gain[c][idx] = gain_cache;
1615 * Parse whether channels are to be excluded from Dynamic Range Compression; reference: table 4.53.
1617 * @return Returns number of bytes consumed.
1619 static int decode_drc_channel_exclusions(DynamicRangeControl *che_drc,
1623 int num_excl_chan = 0;
1626 for (i = 0; i < 7; i++)
1627 che_drc->exclude_mask[num_excl_chan++] = get_bits1(gb);
1628 } while (num_excl_chan < MAX_CHANNELS - 7 && get_bits1(gb));
1630 return num_excl_chan / 7;
1634 * Decode dynamic range information; reference: table 4.52.
1636 * @param cnt length of TYPE_FIL syntactic element in bytes
1638 * @return Returns number of bytes consumed.
1640 static int decode_dynamic_range(DynamicRangeControl *che_drc,
1641 GetBitContext *gb, int cnt)
1644 int drc_num_bands = 1;
1647 /* pce_tag_present? */
1648 if (get_bits1(gb)) {
1649 che_drc->pce_instance_tag = get_bits(gb, 4);
1650 skip_bits(gb, 4); // tag_reserved_bits
1654 /* excluded_chns_present? */
1655 if (get_bits1(gb)) {
1656 n += decode_drc_channel_exclusions(che_drc, gb);
1659 /* drc_bands_present? */
1660 if (get_bits1(gb)) {
1661 che_drc->band_incr = get_bits(gb, 4);
1662 che_drc->interpolation_scheme = get_bits(gb, 4);
1664 drc_num_bands += che_drc->band_incr;
1665 for (i = 0; i < drc_num_bands; i++) {
1666 che_drc->band_top[i] = get_bits(gb, 8);
1671 /* prog_ref_level_present? */
1672 if (get_bits1(gb)) {
1673 che_drc->prog_ref_level = get_bits(gb, 7);
1674 skip_bits1(gb); // prog_ref_level_reserved_bits
1678 for (i = 0; i < drc_num_bands; i++) {
1679 che_drc->dyn_rng_sgn[i] = get_bits1(gb);
1680 che_drc->dyn_rng_ctl[i] = get_bits(gb, 7);
1688 * Decode extension data (incomplete); reference: table 4.51.
1690 * @param cnt length of TYPE_FIL syntactic element in bytes
1692 * @return Returns number of bytes consumed
1694 static int decode_extension_payload(AACContext *ac, GetBitContext *gb, int cnt,
1695 ChannelElement *che, enum RawDataBlockType elem_type)
1699 switch (get_bits(gb, 4)) { // extension type
1700 case EXT_SBR_DATA_CRC:
1704 av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the first channel element.\n");
1706 } else if (!ac->m4ac.sbr) {
1707 av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be not-present but was found in the bitstream.\n");
1708 skip_bits_long(gb, 8 * cnt - 4);
1710 } else if (ac->m4ac.sbr == -1 && ac->output_configured == OC_LOCKED) {
1711 av_log(ac->avctx, AV_LOG_ERROR, "Implicit SBR was found with a first occurrence after the first frame.\n");
1712 skip_bits_long(gb, 8 * cnt - 4);
1714 } else if (ac->m4ac.ps == -1 && ac->output_configured < OC_LOCKED && ac->avctx->channels == 1) {
1717 output_configure(ac, ac->che_pos, ac->che_pos, ac->m4ac.chan_config, ac->output_configured);
1721 res = ff_decode_sbr_extension(ac, &che->sbr, gb, crc_flag, cnt, elem_type);
1723 case EXT_DYNAMIC_RANGE:
1724 res = decode_dynamic_range(&ac->che_drc, gb, cnt);
1728 case EXT_DATA_ELEMENT:
1730 skip_bits_long(gb, 8 * cnt - 4);
1737 * Decode Temporal Noise Shaping filter coefficients and apply all-pole filters; reference: 4.6.9.3.
1739 * @param decode 1 if tool is used normally, 0 if tool is used in LTP.
1740 * @param coef spectral coefficients
1742 static void apply_tns(float coef[1024], TemporalNoiseShaping *tns,
1743 IndividualChannelStream *ics, int decode)
1745 const int mmm = FFMIN(ics->tns_max_bands, ics->max_sfb);
1747 int bottom, top, order, start, end, size, inc;
1748 float lpc[TNS_MAX_ORDER];
1749 float tmp[TNS_MAX_ORDER];
1751 for (w = 0; w < ics->num_windows; w++) {
1752 bottom = ics->num_swb;
1753 for (filt = 0; filt < tns->n_filt[w]; filt++) {
1755 bottom = FFMAX(0, top - tns->length[w][filt]);
1756 order = tns->order[w][filt];
1761 compute_lpc_coefs(tns->coef[w][filt], order, lpc, 0, 0, 0);
1763 start = ics->swb_offset[FFMIN(bottom, mmm)];
1764 end = ics->swb_offset[FFMIN( top, mmm)];
1765 if ((size = end - start) <= 0)
1767 if (tns->direction[w][filt]) {
1777 for (m = 0; m < size; m++, start += inc)
1778 for (i = 1; i <= FFMIN(m, order); i++)
1779 coef[start] -= coef[start - i * inc] * lpc[i - 1];
1782 for (m = 0; m < size; m++, start += inc) {
1783 tmp[0] = coef[start];
1784 for (i = 1; i <= FFMIN(m, order); i++)
1785 coef[start] += tmp[i] * lpc[i - 1];
1786 for (i = order; i > 0; i--)
1787 tmp[i] = tmp[i - 1];
1795 * Apply windowing and MDCT to obtain the spectral
1796 * coefficient from the predicted sample by LTP.
1798 static void windowing_and_mdct_ltp(AACContext *ac, float *out,
1799 float *in, IndividualChannelStream *ics)
1801 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1802 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1803 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1804 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1806 if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
1807 ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
1809 memset(in, 0, 448 * sizeof(float));
1810 ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
1812 if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
1813 ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
1815 ac->dsp.vector_fmul_reverse(in + 1024 + 448, in + 1024 + 448, swindow, 128);
1816 memset(in + 1024 + 576, 0, 448 * sizeof(float));
1818 ac->mdct_ltp.mdct_calc(&ac->mdct_ltp, out, in);
1822 * Apply the long term prediction
1824 static void apply_ltp(AACContext *ac, SingleChannelElement *sce)
1826 const LongTermPrediction *ltp = &sce->ics.ltp;
1827 const uint16_t *offsets = sce->ics.swb_offset;
1830 if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
1831 float *predTime = sce->ret;
1832 float *predFreq = ac->buf_mdct;
1833 int16_t num_samples = 2048;
1835 if (ltp->lag < 1024)
1836 num_samples = ltp->lag + 1024;
1837 for (i = 0; i < num_samples; i++)
1838 predTime[i] = sce->ltp_state[i + 2048 - ltp->lag] * ltp->coef;
1839 memset(&predTime[i], 0, (2048 - i) * sizeof(float));
1841 windowing_and_mdct_ltp(ac, predFreq, predTime, &sce->ics);
1843 if (sce->tns.present)
1844 apply_tns(predFreq, &sce->tns, &sce->ics, 0);
1846 for (sfb = 0; sfb < FFMIN(sce->ics.max_sfb, MAX_LTP_LONG_SFB); sfb++)
1848 for (i = offsets[sfb]; i < offsets[sfb + 1]; i++)
1849 sce->coeffs[i] += predFreq[i];
1854 * Update the LTP buffer for next frame
1856 static void update_ltp(AACContext *ac, SingleChannelElement *sce)
1858 IndividualChannelStream *ics = &sce->ics;
1859 float *saved = sce->saved;
1860 float *saved_ltp = sce->coeffs;
1861 const float *lwindow = ics->use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1862 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1865 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1866 memcpy(saved_ltp, saved, 512 * sizeof(float));
1867 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1868 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1869 for (i = 0; i < 64; i++)
1870 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1871 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1872 memcpy(saved_ltp, ac->buf_mdct + 512, 448 * sizeof(float));
1873 memset(saved_ltp + 576, 0, 448 * sizeof(float));
1874 ac->dsp.vector_fmul_reverse(saved_ltp + 448, ac->buf_mdct + 960, &swindow[64], 64);
1875 for (i = 0; i < 64; i++)
1876 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * swindow[63 - i];
1877 } else { // LONG_STOP or ONLY_LONG
1878 ac->dsp.vector_fmul_reverse(saved_ltp, ac->buf_mdct + 512, &lwindow[512], 512);
1879 for (i = 0; i < 512; i++)
1880 saved_ltp[i + 512] = ac->buf_mdct[1023 - i] * lwindow[511 - i];
1883 memcpy(sce->ltp_state, sce->ltp_state+1024, 1024 * sizeof(*sce->ltp_state));
1884 memcpy(sce->ltp_state+1024, sce->ret, 1024 * sizeof(*sce->ltp_state));
1885 memcpy(sce->ltp_state+2048, saved_ltp, 1024 * sizeof(*sce->ltp_state));
1889 * Conduct IMDCT and windowing.
1891 static void imdct_and_windowing(AACContext *ac, SingleChannelElement *sce)
1893 IndividualChannelStream *ics = &sce->ics;
1894 float *in = sce->coeffs;
1895 float *out = sce->ret;
1896 float *saved = sce->saved;
1897 const float *swindow = ics->use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
1898 const float *lwindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
1899 const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
1900 float *buf = ac->buf_mdct;
1901 float *temp = ac->temp;
1905 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1906 for (i = 0; i < 1024; i += 128)
1907 ac->mdct_small.imdct_half(&ac->mdct_small, buf + i, in + i);
1909 ac->mdct.imdct_half(&ac->mdct, buf, in);
1911 /* window overlapping
1912 * NOTE: To simplify the overlapping code, all 'meaningless' short to long
1913 * and long to short transitions are considered to be short to short
1914 * transitions. This leaves just two cases (long to long and short to short)
1915 * with a little special sauce for EIGHT_SHORT_SEQUENCE.
1917 if ((ics->window_sequence[1] == ONLY_LONG_SEQUENCE || ics->window_sequence[1] == LONG_STOP_SEQUENCE) &&
1918 (ics->window_sequence[0] == ONLY_LONG_SEQUENCE || ics->window_sequence[0] == LONG_START_SEQUENCE)) {
1919 ac->dsp.vector_fmul_window( out, saved, buf, lwindow_prev, 512);
1921 memcpy( out, saved, 448 * sizeof(float));
1923 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1924 ac->dsp.vector_fmul_window(out + 448 + 0*128, saved + 448, buf + 0*128, swindow_prev, 64);
1925 ac->dsp.vector_fmul_window(out + 448 + 1*128, buf + 0*128 + 64, buf + 1*128, swindow, 64);
1926 ac->dsp.vector_fmul_window(out + 448 + 2*128, buf + 1*128 + 64, buf + 2*128, swindow, 64);
1927 ac->dsp.vector_fmul_window(out + 448 + 3*128, buf + 2*128 + 64, buf + 3*128, swindow, 64);
1928 ac->dsp.vector_fmul_window(temp, buf + 3*128 + 64, buf + 4*128, swindow, 64);
1929 memcpy( out + 448 + 4*128, temp, 64 * sizeof(float));
1931 ac->dsp.vector_fmul_window(out + 448, saved + 448, buf, swindow_prev, 64);
1932 memcpy( out + 576, buf + 64, 448 * sizeof(float));
1937 if (ics->window_sequence[0] == EIGHT_SHORT_SEQUENCE) {
1938 memcpy( saved, temp + 64, 64 * sizeof(float));
1939 ac->dsp.vector_fmul_window(saved + 64, buf + 4*128 + 64, buf + 5*128, swindow, 64);
1940 ac->dsp.vector_fmul_window(saved + 192, buf + 5*128 + 64, buf + 6*128, swindow, 64);
1941 ac->dsp.vector_fmul_window(saved + 320, buf + 6*128 + 64, buf + 7*128, swindow, 64);
1942 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1943 } else if (ics->window_sequence[0] == LONG_START_SEQUENCE) {
1944 memcpy( saved, buf + 512, 448 * sizeof(float));
1945 memcpy( saved + 448, buf + 7*128 + 64, 64 * sizeof(float));
1946 } else { // LONG_STOP or ONLY_LONG
1947 memcpy( saved, buf + 512, 512 * sizeof(float));
1952 * Apply dependent channel coupling (applied before IMDCT).
1954 * @param index index into coupling gain array
1956 static void apply_dependent_coupling(AACContext *ac,
1957 SingleChannelElement *target,
1958 ChannelElement *cce, int index)
1960 IndividualChannelStream *ics = &cce->ch[0].ics;
1961 const uint16_t *offsets = ics->swb_offset;
1962 float *dest = target->coeffs;
1963 const float *src = cce->ch[0].coeffs;
1964 int g, i, group, k, idx = 0;
1965 if (ac->m4ac.object_type == AOT_AAC_LTP) {
1966 av_log(ac->avctx, AV_LOG_ERROR,
1967 "Dependent coupling is not supported together with LTP\n");
1970 for (g = 0; g < ics->num_window_groups; g++) {
1971 for (i = 0; i < ics->max_sfb; i++, idx++) {
1972 if (cce->ch[0].band_type[idx] != ZERO_BT) {
1973 const float gain = cce->coup.gain[index][idx];
1974 for (group = 0; group < ics->group_len[g]; group++) {
1975 for (k = offsets[i]; k < offsets[i + 1]; k++) {
1977 dest[group * 128 + k] += gain * src[group * 128 + k];
1982 dest += ics->group_len[g] * 128;
1983 src += ics->group_len[g] * 128;
1988 * Apply independent channel coupling (applied after IMDCT).
1990 * @param index index into coupling gain array
1992 static void apply_independent_coupling(AACContext *ac,
1993 SingleChannelElement *target,
1994 ChannelElement *cce, int index)
1997 const float gain = cce->coup.gain[index][0];
1998 const float *src = cce->ch[0].ret;
1999 float *dest = target->ret;
2000 const int len = 1024 << (ac->m4ac.sbr == 1);
2002 for (i = 0; i < len; i++)
2003 dest[i] += gain * src[i];
2007 * channel coupling transformation interface
2009 * @param apply_coupling_method pointer to (in)dependent coupling function
2011 static void apply_channel_coupling(AACContext *ac, ChannelElement *cc,
2012 enum RawDataBlockType type, int elem_id,
2013 enum CouplingPoint coupling_point,
2014 void (*apply_coupling_method)(AACContext *ac, SingleChannelElement *target, ChannelElement *cce, int index))
2018 for (i = 0; i < MAX_ELEM_ID; i++) {
2019 ChannelElement *cce = ac->che[TYPE_CCE][i];
2022 if (cce && cce->coup.coupling_point == coupling_point) {
2023 ChannelCoupling *coup = &cce->coup;
2025 for (c = 0; c <= coup->num_coupled; c++) {
2026 if (coup->type[c] == type && coup->id_select[c] == elem_id) {
2027 if (coup->ch_select[c] != 1) {
2028 apply_coupling_method(ac, &cc->ch[0], cce, index);
2029 if (coup->ch_select[c] != 0)
2032 if (coup->ch_select[c] != 2)
2033 apply_coupling_method(ac, &cc->ch[1], cce, index++);
2035 index += 1 + (coup->ch_select[c] == 3);
2042 * Convert spectral data to float samples, applying all supported tools as appropriate.
2044 static void spectral_to_sample(AACContext *ac)
2047 for (type = 3; type >= 0; type--) {
2048 for (i = 0; i < MAX_ELEM_ID; i++) {
2049 ChannelElement *che = ac->che[type][i];
2051 if (type <= TYPE_CPE)
2052 apply_channel_coupling(ac, che, type, i, BEFORE_TNS, apply_dependent_coupling);
2053 if (ac->m4ac.object_type == AOT_AAC_LTP) {
2054 if (che->ch[0].ics.predictor_present) {
2055 if (che->ch[0].ics.ltp.present)
2056 apply_ltp(ac, &che->ch[0]);
2057 if (che->ch[1].ics.ltp.present && type == TYPE_CPE)
2058 apply_ltp(ac, &che->ch[1]);
2061 if (che->ch[0].tns.present)
2062 apply_tns(che->ch[0].coeffs, &che->ch[0].tns, &che->ch[0].ics, 1);
2063 if (che->ch[1].tns.present)
2064 apply_tns(che->ch[1].coeffs, &che->ch[1].tns, &che->ch[1].ics, 1);
2065 if (type <= TYPE_CPE)
2066 apply_channel_coupling(ac, che, type, i, BETWEEN_TNS_AND_IMDCT, apply_dependent_coupling);
2067 if (type != TYPE_CCE || che->coup.coupling_point == AFTER_IMDCT) {
2068 imdct_and_windowing(ac, &che->ch[0]);
2069 if (ac->m4ac.object_type == AOT_AAC_LTP)
2070 update_ltp(ac, &che->ch[0]);
2071 if (type == TYPE_CPE) {
2072 imdct_and_windowing(ac, &che->ch[1]);
2073 if (ac->m4ac.object_type == AOT_AAC_LTP)
2074 update_ltp(ac, &che->ch[1]);
2076 if (ac->m4ac.sbr > 0) {
2077 ff_sbr_apply(ac, &che->sbr, type, che->ch[0].ret, che->ch[1].ret);
2080 if (type <= TYPE_CCE)
2081 apply_channel_coupling(ac, che, type, i, AFTER_IMDCT, apply_independent_coupling);
2087 static int parse_adts_frame_header(AACContext *ac, GetBitContext *gb)
2090 AACADTSHeaderInfo hdr_info;
2092 size = avpriv_aac_parse_header(gb, &hdr_info);
2094 if (hdr_info.chan_config) {
2095 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2096 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2097 ac->m4ac.chan_config = hdr_info.chan_config;
2098 if (set_default_channel_config(ac->avctx, new_che_pos, hdr_info.chan_config))
2100 if (output_configure(ac, ac->che_pos, new_che_pos, hdr_info.chan_config,
2101 FFMAX(ac->output_configured, OC_TRIAL_FRAME)))
2103 } else if (ac->output_configured != OC_LOCKED) {
2104 ac->m4ac.chan_config = 0;
2105 ac->output_configured = OC_NONE;
2107 if (ac->output_configured != OC_LOCKED) {
2110 ac->m4ac.sample_rate = hdr_info.sample_rate;
2111 ac->m4ac.sampling_index = hdr_info.sampling_index;
2112 ac->m4ac.object_type = hdr_info.object_type;
2114 if (!ac->avctx->sample_rate)
2115 ac->avctx->sample_rate = hdr_info.sample_rate;
2116 if (hdr_info.num_aac_frames == 1) {
2117 if (!hdr_info.crc_absent)
2120 av_log_missing_feature(ac->avctx, "More than one AAC RDB per ADTS frame is", 0);
2127 static int aac_decode_frame_int(AVCodecContext *avctx, void *data,
2128 int *got_frame_ptr, GetBitContext *gb)
2130 AACContext *ac = avctx->priv_data;
2131 ChannelElement *che = NULL, *che_prev = NULL;
2132 enum RawDataBlockType elem_type, elem_type_prev = TYPE_END;
2134 int samples = 0, multiplier, audio_found = 0;
2136 if (show_bits(gb, 12) == 0xfff) {
2137 if (parse_adts_frame_header(ac, gb) < 0) {
2138 av_log(avctx, AV_LOG_ERROR, "Error decoding AAC frame header.\n");
2141 if (ac->m4ac.sampling_index > 12) {
2142 av_log(ac->avctx, AV_LOG_ERROR, "invalid sampling rate index %d\n", ac->m4ac.sampling_index);
2147 ac->tags_mapped = 0;
2149 while ((elem_type = get_bits(gb, 3)) != TYPE_END) {
2150 elem_id = get_bits(gb, 4);
2152 if (elem_type < TYPE_DSE) {
2153 if (!(che=get_che(ac, elem_type, elem_id))) {
2154 av_log(ac->avctx, AV_LOG_ERROR, "channel element %d.%d is not allocated\n",
2155 elem_type, elem_id);
2161 switch (elem_type) {
2164 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2169 err = decode_cpe(ac, gb, che);
2174 err = decode_cce(ac, gb, che);
2178 err = decode_ics(ac, &che->ch[0], gb, 0, 0);
2183 err = skip_data_stream_element(ac, gb);
2187 enum ChannelPosition new_che_pos[4][MAX_ELEM_ID];
2188 memset(new_che_pos, 0, 4 * MAX_ELEM_ID * sizeof(new_che_pos[0][0]));
2189 if ((err = decode_pce(avctx, &ac->m4ac, new_che_pos, gb)))
2191 if (ac->output_configured > OC_TRIAL_PCE)
2192 av_log(avctx, AV_LOG_ERROR,
2193 "Not evaluating a further program_config_element as this construct is dubious at best.\n");
2195 err = output_configure(ac, ac->che_pos, new_che_pos, 0, OC_TRIAL_PCE);
2201 elem_id += get_bits(gb, 8) - 1;
2202 if (get_bits_left(gb) < 8 * elem_id) {
2203 av_log(avctx, AV_LOG_ERROR, overread_err);
2207 elem_id -= decode_extension_payload(ac, gb, elem_id, che_prev, elem_type_prev);
2208 err = 0; /* FIXME */
2212 err = -1; /* should not happen, but keeps compiler happy */
2217 elem_type_prev = elem_type;
2222 if (get_bits_left(gb) < 3) {
2223 av_log(avctx, AV_LOG_ERROR, overread_err);
2228 spectral_to_sample(ac);
2230 multiplier = (ac->m4ac.sbr == 1) ? ac->m4ac.ext_sample_rate > ac->m4ac.sample_rate : 0;
2231 samples <<= multiplier;
2232 if (ac->output_configured < OC_LOCKED) {
2233 avctx->sample_rate = ac->m4ac.sample_rate << multiplier;
2234 avctx->frame_size = samples;
2238 /* get output buffer */
2239 ac->frame.nb_samples = samples;
2240 if ((err = avctx->get_buffer(avctx, &ac->frame)) < 0) {
2241 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
2245 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
2246 ac->fmt_conv.float_interleave((float *)ac->frame.data[0],
2247 (const float **)ac->output_data,
2248 samples, avctx->channels);
2250 ac->fmt_conv.float_to_int16_interleave((int16_t *)ac->frame.data[0],
2251 (const float **)ac->output_data,
2252 samples, avctx->channels);
2254 *(AVFrame *)data = ac->frame;
2256 *got_frame_ptr = !!samples;
2258 if (ac->output_configured && audio_found)
2259 ac->output_configured = OC_LOCKED;
2264 static int aac_decode_frame(AVCodecContext *avctx, void *data,
2265 int *got_frame_ptr, AVPacket *avpkt)
2267 AACContext *ac = avctx->priv_data;
2268 const uint8_t *buf = avpkt->data;
2269 int buf_size = avpkt->size;
2274 int new_extradata_size;
2275 const uint8_t *new_extradata = av_packet_get_side_data(avpkt,
2276 AV_PKT_DATA_NEW_EXTRADATA,
2277 &new_extradata_size);
2279 if (new_extradata) {
2280 av_free(avctx->extradata);
2281 avctx->extradata = av_mallocz(new_extradata_size +
2282 FF_INPUT_BUFFER_PADDING_SIZE);
2283 if (!avctx->extradata)
2284 return AVERROR(ENOMEM);
2285 avctx->extradata_size = new_extradata_size;
2286 memcpy(avctx->extradata, new_extradata, new_extradata_size);
2287 if (decode_audio_specific_config(ac, ac->avctx, &ac->m4ac,
2289 avctx->extradata_size*8, 1) < 0)
2290 return AVERROR_INVALIDDATA;
2293 init_get_bits(&gb, buf, buf_size * 8);
2295 if ((err = aac_decode_frame_int(avctx, data, got_frame_ptr, &gb)) < 0)
2298 buf_consumed = (get_bits_count(&gb) + 7) >> 3;
2299 for (buf_offset = buf_consumed; buf_offset < buf_size; buf_offset++)
2300 if (buf[buf_offset])
2303 return buf_size > buf_offset ? buf_consumed : buf_size;
2306 static av_cold int aac_decode_close(AVCodecContext *avctx)
2308 AACContext *ac = avctx->priv_data;
2311 for (i = 0; i < MAX_ELEM_ID; i++) {
2312 for (type = 0; type < 4; type++) {
2313 if (ac->che[type][i])
2314 ff_aac_sbr_ctx_close(&ac->che[type][i]->sbr);
2315 av_freep(&ac->che[type][i]);
2319 ff_mdct_end(&ac->mdct);
2320 ff_mdct_end(&ac->mdct_small);
2321 ff_mdct_end(&ac->mdct_ltp);
2326 #define LOAS_SYNC_WORD 0x2b7 ///< 11 bits LOAS sync word
2328 struct LATMContext {
2329 AACContext aac_ctx; ///< containing AACContext
2330 int initialized; ///< initilized after a valid extradata was seen
2333 int audio_mux_version_A; ///< LATM syntax version
2334 int frame_length_type; ///< 0/1 variable/fixed frame length
2335 int frame_length; ///< frame length for fixed frame length
2338 static inline uint32_t latm_get_value(GetBitContext *b)
2340 int length = get_bits(b, 2);
2342 return get_bits_long(b, (length+1)*8);
2345 static int latm_decode_audio_specific_config(struct LATMContext *latmctx,
2346 GetBitContext *gb, int asclen)
2348 AACContext *ac = &latmctx->aac_ctx;
2349 AVCodecContext *avctx = ac->avctx;
2350 MPEG4AudioConfig m4ac = {0};
2351 int config_start_bit = get_bits_count(gb);
2352 int sync_extension = 0;
2353 int bits_consumed, esize;
2357 asclen = FFMIN(asclen, get_bits_left(gb));
2359 asclen = get_bits_left(gb);
2361 if (config_start_bit % 8) {
2362 av_log_missing_feature(latmctx->aac_ctx.avctx, "audio specific "
2363 "config not byte aligned.\n", 1);
2364 return AVERROR_INVALIDDATA;
2367 return AVERROR_INVALIDDATA;
2368 bits_consumed = decode_audio_specific_config(NULL, avctx, &m4ac,
2369 gb->buffer + (config_start_bit / 8),
2370 asclen, sync_extension);
2372 if (bits_consumed < 0)
2373 return AVERROR_INVALIDDATA;
2375 if (ac->m4ac.sample_rate != m4ac.sample_rate ||
2376 ac->m4ac.chan_config != m4ac.chan_config) {
2378 av_log(avctx, AV_LOG_INFO, "audio config changed\n");
2379 latmctx->initialized = 0;
2381 esize = (bits_consumed+7) / 8;
2383 if (avctx->extradata_size < esize) {
2384 av_free(avctx->extradata);
2385 avctx->extradata = av_malloc(esize + FF_INPUT_BUFFER_PADDING_SIZE);
2386 if (!avctx->extradata)
2387 return AVERROR(ENOMEM);
2390 avctx->extradata_size = esize;
2391 memcpy(avctx->extradata, gb->buffer + (config_start_bit/8), esize);
2392 memset(avctx->extradata+esize, 0, FF_INPUT_BUFFER_PADDING_SIZE);
2394 skip_bits_long(gb, bits_consumed);
2396 return bits_consumed;
2399 static int read_stream_mux_config(struct LATMContext *latmctx,
2402 int ret, audio_mux_version = get_bits(gb, 1);
2404 latmctx->audio_mux_version_A = 0;
2405 if (audio_mux_version)
2406 latmctx->audio_mux_version_A = get_bits(gb, 1);
2408 if (!latmctx->audio_mux_version_A) {
2410 if (audio_mux_version)
2411 latm_get_value(gb); // taraFullness
2413 skip_bits(gb, 1); // allStreamSameTimeFraming
2414 skip_bits(gb, 6); // numSubFrames
2416 if (get_bits(gb, 4)) { // numPrograms
2417 av_log_missing_feature(latmctx->aac_ctx.avctx,
2418 "multiple programs are not supported\n", 1);
2419 return AVERROR_PATCHWELCOME;
2422 // for each program (which there is only on in DVB)
2424 // for each layer (which there is only on in DVB)
2425 if (get_bits(gb, 3)) { // numLayer
2426 av_log_missing_feature(latmctx->aac_ctx.avctx,
2427 "multiple layers are not supported\n", 1);
2428 return AVERROR_PATCHWELCOME;
2431 // for all but first stream: use_same_config = get_bits(gb, 1);
2432 if (!audio_mux_version) {
2433 if ((ret = latm_decode_audio_specific_config(latmctx, gb, 0)) < 0)
2436 int ascLen = latm_get_value(gb);
2437 if ((ret = latm_decode_audio_specific_config(latmctx, gb, ascLen)) < 0)
2440 skip_bits_long(gb, ascLen);
2443 latmctx->frame_length_type = get_bits(gb, 3);
2444 switch (latmctx->frame_length_type) {
2446 skip_bits(gb, 8); // latmBufferFullness
2449 latmctx->frame_length = get_bits(gb, 9);
2454 skip_bits(gb, 6); // CELP frame length table index
2458 skip_bits(gb, 1); // HVXC frame length table index
2462 if (get_bits(gb, 1)) { // other data
2463 if (audio_mux_version) {
2464 latm_get_value(gb); // other_data_bits
2468 esc = get_bits(gb, 1);
2474 if (get_bits(gb, 1)) // crc present
2475 skip_bits(gb, 8); // config_crc
2481 static int read_payload_length_info(struct LATMContext *ctx, GetBitContext *gb)
2485 if (ctx->frame_length_type == 0) {
2486 int mux_slot_length = 0;
2488 tmp = get_bits(gb, 8);
2489 mux_slot_length += tmp;
2490 } while (tmp == 255);
2491 return mux_slot_length;
2492 } else if (ctx->frame_length_type == 1) {
2493 return ctx->frame_length;
2494 } else if (ctx->frame_length_type == 3 ||
2495 ctx->frame_length_type == 5 ||
2496 ctx->frame_length_type == 7) {
2497 skip_bits(gb, 2); // mux_slot_length_coded
2502 static int read_audio_mux_element(struct LATMContext *latmctx,
2506 uint8_t use_same_mux = get_bits(gb, 1);
2507 if (!use_same_mux) {
2508 if ((err = read_stream_mux_config(latmctx, gb)) < 0)
2510 } else if (!latmctx->aac_ctx.avctx->extradata) {
2511 av_log(latmctx->aac_ctx.avctx, AV_LOG_DEBUG,
2512 "no decoder config found\n");
2513 return AVERROR(EAGAIN);
2515 if (latmctx->audio_mux_version_A == 0) {
2516 int mux_slot_length_bytes = read_payload_length_info(latmctx, gb);
2517 if (mux_slot_length_bytes * 8 > get_bits_left(gb)) {
2518 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR, "incomplete frame\n");
2519 return AVERROR_INVALIDDATA;
2520 } else if (mux_slot_length_bytes * 8 + 256 < get_bits_left(gb)) {
2521 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2522 "frame length mismatch %d << %d\n",
2523 mux_slot_length_bytes * 8, get_bits_left(gb));
2524 return AVERROR_INVALIDDATA;
2531 static int latm_decode_frame(AVCodecContext *avctx, void *out,
2532 int *got_frame_ptr, AVPacket *avpkt)
2534 struct LATMContext *latmctx = avctx->priv_data;
2538 init_get_bits(&gb, avpkt->data, avpkt->size * 8);
2540 // check for LOAS sync word
2541 if (get_bits(&gb, 11) != LOAS_SYNC_WORD)
2542 return AVERROR_INVALIDDATA;
2544 muxlength = get_bits(&gb, 13) + 3;
2545 // not enough data, the parser should have sorted this
2546 if (muxlength > avpkt->size)
2547 return AVERROR_INVALIDDATA;
2549 if ((err = read_audio_mux_element(latmctx, &gb)) < 0)
2552 if (!latmctx->initialized) {
2553 if (!avctx->extradata) {
2557 if ((err = decode_audio_specific_config(
2558 &latmctx->aac_ctx, avctx, &latmctx->aac_ctx.m4ac,
2559 avctx->extradata, avctx->extradata_size*8, 1)) < 0)
2561 latmctx->initialized = 1;
2565 if (show_bits(&gb, 12) == 0xfff) {
2566 av_log(latmctx->aac_ctx.avctx, AV_LOG_ERROR,
2567 "ADTS header detected, probably as result of configuration "
2569 return AVERROR_INVALIDDATA;
2572 if ((err = aac_decode_frame_int(avctx, out, got_frame_ptr, &gb)) < 0)
2578 av_cold static int latm_decode_init(AVCodecContext *avctx)
2580 struct LATMContext *latmctx = avctx->priv_data;
2581 int ret = aac_decode_init(avctx);
2583 if (avctx->extradata_size > 0)
2584 latmctx->initialized = !ret;
2590 AVCodec ff_aac_decoder = {
2592 .type = AVMEDIA_TYPE_AUDIO,
2594 .priv_data_size = sizeof(AACContext),
2595 .init = aac_decode_init,
2596 .close = aac_decode_close,
2597 .decode = aac_decode_frame,
2598 .long_name = NULL_IF_CONFIG_SMALL("Advanced Audio Coding"),
2599 .sample_fmts = (const enum AVSampleFormat[]) {
2600 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2602 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2603 .channel_layouts = aac_channel_layout,
2607 Note: This decoder filter is intended to decode LATM streams transferred
2608 in MPEG transport streams which only contain one program.
2609 To do a more complex LATM demuxing a separate LATM demuxer should be used.
2611 AVCodec ff_aac_latm_decoder = {
2613 .type = AVMEDIA_TYPE_AUDIO,
2614 .id = CODEC_ID_AAC_LATM,
2615 .priv_data_size = sizeof(struct LATMContext),
2616 .init = latm_decode_init,
2617 .close = aac_decode_close,
2618 .decode = latm_decode_frame,
2619 .long_name = NULL_IF_CONFIG_SMALL("AAC LATM (Advanced Audio Codec LATM syntax)"),
2620 .sample_fmts = (const enum AVSampleFormat[]) {
2621 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
2623 .capabilities = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
2624 .channel_layouts = aac_channel_layout,